Statistics
| Branch: | Revision:

ffmpeg / libavcodec / flacenc.c @ 3abe5fbd

History | View | Annotate | Download (43.7 KB)

1
/**
2
 * FLAC audio encoder
3
 * Copyright (c) 2006  Justin Ruggles <jruggle@earthlink.net>
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 "avcodec.h"
23
#include "bitstream.h"
24
#include "crc.h"
25
#include "dsputil.h"
26
#include "golomb.h"
27
#include "lls.h"
28

    
29
#define FLAC_MAX_CH  8
30
#define FLAC_MIN_BLOCKSIZE  16
31
#define FLAC_MAX_BLOCKSIZE  65535
32

    
33
#define FLAC_SUBFRAME_CONSTANT  0
34
#define FLAC_SUBFRAME_VERBATIM  1
35
#define FLAC_SUBFRAME_FIXED     8
36
#define FLAC_SUBFRAME_LPC      32
37

    
38
#define FLAC_CHMODE_NOT_STEREO      0
39
#define FLAC_CHMODE_LEFT_RIGHT      1
40
#define FLAC_CHMODE_LEFT_SIDE       8
41
#define FLAC_CHMODE_RIGHT_SIDE      9
42
#define FLAC_CHMODE_MID_SIDE       10
43

    
44
#define ORDER_METHOD_EST     0
45
#define ORDER_METHOD_2LEVEL  1
46
#define ORDER_METHOD_4LEVEL  2
47
#define ORDER_METHOD_8LEVEL  3
48
#define ORDER_METHOD_SEARCH  4
49
#define ORDER_METHOD_LOG     5
50

    
51
#define FLAC_STREAMINFO_SIZE  34
52

    
53
#define MIN_LPC_ORDER       1
54
#define MAX_LPC_ORDER      32
55
#define MAX_FIXED_ORDER     4
56
#define MAX_PARTITION_ORDER 8
57
#define MAX_PARTITIONS     (1 << MAX_PARTITION_ORDER)
58
#define MAX_LPC_PRECISION  15
59
#define MAX_LPC_SHIFT      15
60
#define MAX_RICE_PARAM     14
61

    
62
typedef struct CompressionOptions {
63
    int compression_level;
64
    int block_time_ms;
65
    int use_lpc;
66
    int lpc_coeff_precision;
67
    int min_prediction_order;
68
    int max_prediction_order;
69
    int prediction_order_method;
70
    int min_partition_order;
71
    int max_partition_order;
72
} CompressionOptions;
73

    
74
typedef struct RiceContext {
75
    int porder;
76
    int params[MAX_PARTITIONS];
77
} RiceContext;
78

    
79
typedef struct FlacSubframe {
80
    int type;
81
    int type_code;
82
    int obits;
83
    int order;
84
    int32_t coefs[MAX_LPC_ORDER];
85
    int shift;
86
    RiceContext rc;
87
    int32_t samples[FLAC_MAX_BLOCKSIZE];
88
    int32_t residual[FLAC_MAX_BLOCKSIZE+1];
89
} FlacSubframe;
90

    
91
typedef struct FlacFrame {
92
    FlacSubframe subframes[FLAC_MAX_CH];
93
    int blocksize;
94
    int bs_code[2];
95
    uint8_t crc8;
96
    int ch_mode;
97
} FlacFrame;
98

    
99
typedef struct FlacEncodeContext {
100
    PutBitContext pb;
101
    int channels;
102
    int ch_code;
103
    int samplerate;
104
    int sr_code[2];
105
    int blocksize;
106
    int max_framesize;
107
    uint32_t frame_count;
108
    FlacFrame frame;
109
    CompressionOptions options;
110
    AVCodecContext *avctx;
111
    DSPContext dsp;
112
} FlacEncodeContext;
113

    
114
static const int flac_samplerates[16] = {
115
    0, 0, 0, 0,
116
    8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
117
    0, 0, 0, 0
118
};
119

    
120
static const int flac_blocksizes[16] = {
121
    0,
122
    192,
123
    576, 1152, 2304, 4608,
124
    0, 0,
125
    256, 512, 1024, 2048, 4096, 8192, 16384, 32768
126
};
127

    
128
/**
129
 * Writes streaminfo metadata block to byte array
130
 */
131
static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
132
{
133
    PutBitContext pb;
134

    
135
    memset(header, 0, FLAC_STREAMINFO_SIZE);
136
    init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
137

    
138
    /* streaminfo metadata block */
139
    put_bits(&pb, 16, s->blocksize);
140
    put_bits(&pb, 16, s->blocksize);
141
    put_bits(&pb, 24, 0);
142
    put_bits(&pb, 24, s->max_framesize);
143
    put_bits(&pb, 20, s->samplerate);
144
    put_bits(&pb, 3, s->channels-1);
145
    put_bits(&pb, 5, 15);       /* bits per sample - 1 */
146
    flush_put_bits(&pb);
147
    /* total samples = 0 */
148
    /* MD5 signature = 0 */
149
}
150

    
151
/**
152
 * Sets blocksize based on samplerate
153
 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
154
 */
155
static int select_blocksize(int samplerate, int block_time_ms)
156
{
157
    int i;
158
    int target;
159
    int blocksize;
160

    
161
    assert(samplerate > 0);
162
    blocksize = flac_blocksizes[1];
163
    target = (samplerate * block_time_ms) / 1000;
164
    for(i=0; i<16; i++) {
165
        if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
166
            blocksize = flac_blocksizes[i];
167
        }
168
    }
169
    return blocksize;
170
}
171

    
172
static int flac_encode_init(AVCodecContext *avctx)
173
{
174
    int freq = avctx->sample_rate;
175
    int channels = avctx->channels;
176
    FlacEncodeContext *s = avctx->priv_data;
177
    int i, level;
178
    uint8_t *streaminfo;
179

    
180
    s->avctx = avctx;
181

    
182
    dsputil_init(&s->dsp, avctx);
183

    
184
    if(avctx->sample_fmt != SAMPLE_FMT_S16) {
185
        return -1;
186
    }
187

    
188
    if(channels < 1 || channels > FLAC_MAX_CH) {
189
        return -1;
190
    }
191
    s->channels = channels;
192
    s->ch_code = s->channels-1;
193

    
194
    /* find samplerate in table */
195
    if(freq < 1)
196
        return -1;
197
    for(i=4; i<12; i++) {
198
        if(freq == flac_samplerates[i]) {
199
            s->samplerate = flac_samplerates[i];
200
            s->sr_code[0] = i;
201
            s->sr_code[1] = 0;
202
            break;
203
        }
204
    }
205
    /* if not in table, samplerate is non-standard */
206
    if(i == 12) {
207
        if(freq % 1000 == 0 && freq < 255000) {
208
            s->sr_code[0] = 12;
209
            s->sr_code[1] = freq / 1000;
210
        } else if(freq % 10 == 0 && freq < 655350) {
211
            s->sr_code[0] = 14;
212
            s->sr_code[1] = freq / 10;
213
        } else if(freq < 65535) {
214
            s->sr_code[0] = 13;
215
            s->sr_code[1] = freq;
216
        } else {
217
            return -1;
218
        }
219
        s->samplerate = freq;
220
    }
221

    
222
    /* set compression option defaults based on avctx->compression_level */
223
    if(avctx->compression_level < 0) {
224
        s->options.compression_level = 5;
225
    } else {
226
        s->options.compression_level = avctx->compression_level;
227
    }
228
    av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
229

    
230
    level= s->options.compression_level;
231
    if(level > 12) {
232
        av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
233
               s->options.compression_level);
234
        return -1;
235
    }
236

    
237
    s->options.block_time_ms       = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
238
    s->options.use_lpc             = ((int[]){  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
239
    s->options.min_prediction_order= ((int[]){  2,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
240
    s->options.max_prediction_order= ((int[]){  3,  4,  4,  6,  8,  8,  8,  8, 12, 12, 12, 32, 32})[level];
241
    s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
242
                                                   ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
243
                                                   ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG,    ORDER_METHOD_4LEVEL,
244
                                                   ORDER_METHOD_LOG,    ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
245
                                                   ORDER_METHOD_SEARCH})[level];
246
    s->options.min_partition_order = ((int[]){  2,  2,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0})[level];
247
    s->options.max_partition_order = ((int[]){  2,  2,  3,  3,  3,  8,  8,  8,  8,  8,  8,  8,  8})[level];
248

    
249
    /* set compression option overrides from AVCodecContext */
250
    if(avctx->use_lpc >= 0) {
251
        s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
252
    }
253
    if(s->options.use_lpc == 1)
254
        av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
255
    else if(s->options.use_lpc > 1)
256
        av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
257

    
258
    if(avctx->min_prediction_order >= 0) {
259
        if(s->options.use_lpc) {
260
            if(avctx->min_prediction_order < MIN_LPC_ORDER ||
261
                    avctx->min_prediction_order > MAX_LPC_ORDER) {
262
                av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
263
                       avctx->min_prediction_order);
264
                return -1;
265
            }
266
        } else {
267
            if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
268
                av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
269
                       avctx->min_prediction_order);
270
                return -1;
271
            }
272
        }
273
        s->options.min_prediction_order = avctx->min_prediction_order;
274
    }
275
    if(avctx->max_prediction_order >= 0) {
276
        if(s->options.use_lpc) {
277
            if(avctx->max_prediction_order < MIN_LPC_ORDER ||
278
                    avctx->max_prediction_order > MAX_LPC_ORDER) {
279
                av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
280
                       avctx->max_prediction_order);
281
                return -1;
282
            }
283
        } else {
284
            if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
285
                av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
286
                       avctx->max_prediction_order);
287
                return -1;
288
            }
289
        }
290
        s->options.max_prediction_order = avctx->max_prediction_order;
291
    }
292
    if(s->options.max_prediction_order < s->options.min_prediction_order) {
293
        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
294
               s->options.min_prediction_order, s->options.max_prediction_order);
295
        return -1;
296
    }
297
    av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
298
           s->options.min_prediction_order, s->options.max_prediction_order);
299

    
300
    if(avctx->prediction_order_method >= 0) {
301
        if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
302
            av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
303
                   avctx->prediction_order_method);
304
            return -1;
305
        }
306
        s->options.prediction_order_method = avctx->prediction_order_method;
307
    }
308
    switch(s->options.prediction_order_method) {
309
        case ORDER_METHOD_EST:    av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
310
                                         "estimate"); break;
311
        case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
312
                                         "2-level"); break;
313
        case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
314
                                         "4-level"); break;
315
        case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
316
                                         "8-level"); break;
317
        case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
318
                                         "full search"); break;
319
        case ORDER_METHOD_LOG:    av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
320
                                         "log search"); break;
321
    }
322

    
323
    if(avctx->min_partition_order >= 0) {
324
        if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
325
            av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
326
                   avctx->min_partition_order);
327
            return -1;
328
        }
329
        s->options.min_partition_order = avctx->min_partition_order;
330
    }
331
    if(avctx->max_partition_order >= 0) {
332
        if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
333
            av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
334
                   avctx->max_partition_order);
335
            return -1;
336
        }
337
        s->options.max_partition_order = avctx->max_partition_order;
338
    }
339
    if(s->options.max_partition_order < s->options.min_partition_order) {
340
        av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
341
               s->options.min_partition_order, s->options.max_partition_order);
342
        return -1;
343
    }
344
    av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
345
           s->options.min_partition_order, s->options.max_partition_order);
346

    
347
    if(avctx->frame_size > 0) {
348
        if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
349
                avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
350
            av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
351
                   avctx->frame_size);
352
            return -1;
353
        }
354
        s->blocksize = avctx->frame_size;
355
    } else {
356
        s->blocksize = select_blocksize(s->samplerate, s->options.block_time_ms);
357
        avctx->frame_size = s->blocksize;
358
    }
359
    av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->blocksize);
360

    
361
    /* set LPC precision */
362
    if(avctx->lpc_coeff_precision > 0) {
363
        if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
364
            av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
365
                   avctx->lpc_coeff_precision);
366
            return -1;
367
        }
368
        s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
369
    } else {
370
        /* select LPC precision based on block size */
371
        if(     s->blocksize <=   192) s->options.lpc_coeff_precision =  7;
372
        else if(s->blocksize <=   384) s->options.lpc_coeff_precision =  8;
373
        else if(s->blocksize <=   576) s->options.lpc_coeff_precision =  9;
374
        else if(s->blocksize <=  1152) s->options.lpc_coeff_precision = 10;
375
        else if(s->blocksize <=  2304) s->options.lpc_coeff_precision = 11;
376
        else if(s->blocksize <=  4608) s->options.lpc_coeff_precision = 12;
377
        else if(s->blocksize <=  8192) s->options.lpc_coeff_precision = 13;
378
        else if(s->blocksize <= 16384) s->options.lpc_coeff_precision = 14;
379
        else                           s->options.lpc_coeff_precision = 15;
380
    }
381
    av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
382
           s->options.lpc_coeff_precision);
383

    
384
    /* set maximum encoded frame size in verbatim mode */
385
    if(s->channels == 2) {
386
        s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3);
387
    } else {
388
        s->max_framesize = 14 + (s->blocksize * s->channels * 2);
389
    }
390

    
391
    streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
392
    write_streaminfo(s, streaminfo);
393
    avctx->extradata = streaminfo;
394
    avctx->extradata_size = FLAC_STREAMINFO_SIZE;
395

    
396
    s->frame_count = 0;
397

    
398
    avctx->coded_frame = avcodec_alloc_frame();
399
    avctx->coded_frame->key_frame = 1;
400

    
401
    return 0;
402
}
403

    
404
static void init_frame(FlacEncodeContext *s)
405
{
406
    int i, ch;
407
    FlacFrame *frame;
408

    
409
    frame = &s->frame;
410

    
411
    for(i=0; i<16; i++) {
412
        if(s->blocksize == flac_blocksizes[i]) {
413
            frame->blocksize = flac_blocksizes[i];
414
            frame->bs_code[0] = i;
415
            frame->bs_code[1] = 0;
416
            break;
417
        }
418
    }
419
    if(i == 16) {
420
        frame->blocksize = s->blocksize;
421
        if(frame->blocksize <= 256) {
422
            frame->bs_code[0] = 6;
423
            frame->bs_code[1] = frame->blocksize-1;
424
        } else {
425
            frame->bs_code[0] = 7;
426
            frame->bs_code[1] = frame->blocksize-1;
427
        }
428
    }
429

    
430
    for(ch=0; ch<s->channels; ch++) {
431
        frame->subframes[ch].obits = 16;
432
    }
433
}
434

    
435
/**
436
 * Copy channel-interleaved input samples into separate subframes
437
 */
438
static void copy_samples(FlacEncodeContext *s, int16_t *samples)
439
{
440
    int i, j, ch;
441
    FlacFrame *frame;
442

    
443
    frame = &s->frame;
444
    for(i=0,j=0; i<frame->blocksize; i++) {
445
        for(ch=0; ch<s->channels; ch++,j++) {
446
            frame->subframes[ch].samples[i] = samples[j];
447
        }
448
    }
449
}
450

    
451

    
452
#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
453

    
454
/**
455
 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
456
 */
457
static int find_optimal_param(uint32_t sum, int n)
458
{
459
    int k;
460
    uint32_t sum2;
461

    
462
    if(sum <= n>>1)
463
        return 0;
464
    sum2 = sum-(n>>1);
465
    k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
466
    return FFMIN(k, MAX_RICE_PARAM);
467
}
468

    
469
static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
470
                                         uint32_t *sums, int n, int pred_order)
471
{
472
    int i;
473
    int k, cnt, part;
474
    uint32_t all_bits;
475

    
476
    part = (1 << porder);
477
    all_bits = 4 * part;
478

    
479
    cnt = (n >> porder) - pred_order;
480
    for(i=0; i<part; i++) {
481
        k = find_optimal_param(sums[i], cnt);
482
        rc->params[i] = k;
483
        all_bits += rice_encode_count(sums[i], cnt, k);
484
        cnt = n >> porder;
485
    }
486

    
487
    rc->porder = porder;
488

    
489
    return all_bits;
490
}
491

    
492
static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
493
                      uint32_t sums[][MAX_PARTITIONS])
494
{
495
    int i, j;
496
    int parts;
497
    uint32_t *res, *res_end;
498

    
499
    /* sums for highest level */
500
    parts = (1 << pmax);
501
    res = &data[pred_order];
502
    res_end = &data[n >> pmax];
503
    for(i=0; i<parts; i++) {
504
        uint32_t sum = 0;
505
        while(res < res_end){
506
            sum += *(res++);
507
        }
508
        sums[pmax][i] = sum;
509
        res_end+= n >> pmax;
510
    }
511
    /* sums for lower levels */
512
    for(i=pmax-1; i>=pmin; i--) {
513
        parts = (1 << i);
514
        for(j=0; j<parts; j++) {
515
            sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
516
        }
517
    }
518
}
519

    
520
static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
521
                                 int32_t *data, int n, int pred_order)
522
{
523
    int i;
524
    uint32_t bits[MAX_PARTITION_ORDER+1];
525
    int opt_porder;
526
    RiceContext tmp_rc;
527
    uint32_t *udata;
528
    uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
529

    
530
    assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
531
    assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
532
    assert(pmin <= pmax);
533

    
534
    udata = av_malloc(n * sizeof(uint32_t));
535
    for(i=0; i<n; i++) {
536
        udata[i] = (2*data[i]) ^ (data[i]>>31);
537
    }
538

    
539
    calc_sums(pmin, pmax, udata, n, pred_order, sums);
540

    
541
    opt_porder = pmin;
542
    bits[pmin] = UINT32_MAX;
543
    for(i=pmin; i<=pmax; i++) {
544
        bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
545
        if(bits[i] <= bits[opt_porder]) {
546
            opt_porder = i;
547
            *rc= tmp_rc;
548
        }
549
    }
550

    
551
    av_freep(&udata);
552
    return bits[opt_porder];
553
}
554

    
555
static int get_max_p_order(int max_porder, int n, int order)
556
{
557
    int porder = FFMIN(max_porder, av_log2(n^(n-1)));
558
    if(order > 0)
559
        porder = FFMIN(porder, av_log2(n/order));
560
    return porder;
561
}
562

    
563
static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
564
                                       int32_t *data, int n, int pred_order,
565
                                       int bps)
566
{
567
    uint32_t bits;
568
    pmin = get_max_p_order(pmin, n, pred_order);
569
    pmax = get_max_p_order(pmax, n, pred_order);
570
    bits = pred_order*bps + 6;
571
    bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
572
    return bits;
573
}
574

    
575
static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
576
                                     int32_t *data, int n, int pred_order,
577
                                     int bps, int precision)
578
{
579
    uint32_t bits;
580
    pmin = get_max_p_order(pmin, n, pred_order);
581
    pmax = get_max_p_order(pmax, n, pred_order);
582
    bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
583
    bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
584
    return bits;
585
}
586

    
587
/**
588
 * Apply Welch window function to audio block
589
 */
590
static void apply_welch_window(const int32_t *data, int len, double *w_data)
591
{
592
    int i, n2;
593
    double w;
594
    double c;
595

    
596
    n2 = (len >> 1);
597
    c = 2.0 / (len - 1.0);
598
    for(i=0; i<n2; i++) {
599
        w = c - i - 1.0;
600
        w = 1.0 - (w * w);
601
        w_data[i] = data[i] * w;
602
        w_data[len-1-i] = data[len-1-i] * w;
603
    }
604
}
605

    
606
/**
607
 * Calculates autocorrelation data from audio samples
608
 * A Welch window function is applied before calculation.
609
 */
610
void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
611
                              double *autoc)
612
{
613
    int i, j;
614
    double tmp[len + lag + 1];
615
    double *data1= tmp + lag;
616

    
617
    apply_welch_window(data, len, data1);
618

    
619
    for(j=0; j<lag; j++)
620
        data1[j-lag]= 0.0;
621
    data1[len] = 0.0;
622

    
623
    for(j=0; j<lag; j+=2){
624
        double sum0 = 1.0, sum1 = 1.0;
625
        for(i=0; i<len; i++){
626
            sum0 += data1[i] * data1[i-j];
627
            sum1 += data1[i] * data1[i-j-1];
628
        }
629
        autoc[j  ] = sum0;
630
        autoc[j+1] = sum1;
631
    }
632

    
633
    if(j==lag){
634
        double sum = 1.0;
635
        for(i=0; i<len; i+=2){
636
            sum += data1[i  ] * data1[i-j  ]
637
                 + data1[i+1] * data1[i-j+1];
638
        }
639
        autoc[j] = sum;
640
    }
641
}
642

    
643
/**
644
 * Levinson-Durbin recursion.
645
 * Produces LPC coefficients from autocorrelation data.
646
 */
647
static void compute_lpc_coefs(const double *autoc, int max_order,
648
                              double lpc[][MAX_LPC_ORDER], double *ref)
649
{
650
   int i, j, i2;
651
   double r, err, tmp;
652
   double lpc_tmp[MAX_LPC_ORDER];
653

    
654
   for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
655
   err = autoc[0];
656

    
657
   for(i=0; i<max_order; i++) {
658
      r = -autoc[i+1];
659
      for(j=0; j<i; j++) {
660
          r -= lpc_tmp[j] * autoc[i-j];
661
      }
662
      r /= err;
663
      ref[i] = fabs(r);
664

    
665
      err *= 1.0 - (r * r);
666

    
667
      i2 = (i >> 1);
668
      lpc_tmp[i] = r;
669
      for(j=0; j<i2; j++) {
670
         tmp = lpc_tmp[j];
671
         lpc_tmp[j] += r * lpc_tmp[i-1-j];
672
         lpc_tmp[i-1-j] += r * tmp;
673
      }
674
      if(i & 1) {
675
          lpc_tmp[j] += lpc_tmp[j] * r;
676
      }
677

    
678
      for(j=0; j<=i; j++) {
679
          lpc[i][j] = -lpc_tmp[j];
680
      }
681
   }
682
}
683

    
684
/**
685
 * Quantize LPC coefficients
686
 */
687
static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
688
                               int32_t *lpc_out, int *shift)
689
{
690
    int i;
691
    double cmax, error;
692
    int32_t qmax;
693
    int sh;
694

    
695
    /* define maximum levels */
696
    qmax = (1 << (precision - 1)) - 1;
697

    
698
    /* find maximum coefficient value */
699
    cmax = 0.0;
700
    for(i=0; i<order; i++) {
701
        cmax= FFMAX(cmax, fabs(lpc_in[i]));
702
    }
703

    
704
    /* if maximum value quantizes to zero, return all zeros */
705
    if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) {
706
        *shift = 0;
707
        memset(lpc_out, 0, sizeof(int32_t) * order);
708
        return;
709
    }
710

    
711
    /* calculate level shift which scales max coeff to available bits */
712
    sh = MAX_LPC_SHIFT;
713
    while((cmax * (1 << sh) > qmax) && (sh > 0)) {
714
        sh--;
715
    }
716

    
717
    /* since negative shift values are unsupported in decoder, scale down
718
       coefficients instead */
719
    if(sh == 0 && cmax > qmax) {
720
        double scale = ((double)qmax) / cmax;
721
        for(i=0; i<order; i++) {
722
            lpc_in[i] *= scale;
723
        }
724
    }
725

    
726
    /* output quantized coefficients and level shift */
727
    error=0;
728
    for(i=0; i<order; i++) {
729
        error += lpc_in[i] * (1 << sh);
730
        lpc_out[i] = av_clip(lrintf(error), -qmax, qmax);
731
        error -= lpc_out[i];
732
    }
733
    *shift = sh;
734
}
735

    
736
static int estimate_best_order(double *ref, int max_order)
737
{
738
    int i, est;
739

    
740
    est = 1;
741
    for(i=max_order-1; i>=0; i--) {
742
        if(ref[i] > 0.10) {
743
            est = i+1;
744
            break;
745
        }
746
    }
747
    return est;
748
}
749

    
750
/**
751
 * Calculate LPC coefficients for multiple orders
752
 */
753
static int lpc_calc_coefs(FlacEncodeContext *s,
754
                          const int32_t *samples, int blocksize, int max_order,
755
                          int precision, int32_t coefs[][MAX_LPC_ORDER],
756
                          int *shift, int use_lpc, int omethod)
757
{
758
    double autoc[MAX_LPC_ORDER+1];
759
    double ref[MAX_LPC_ORDER];
760
    double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
761
    int i, j, pass;
762
    int opt_order;
763

    
764
    assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);
765

    
766
    if(use_lpc == 1){
767
        s->dsp.flac_compute_autocorr(samples, blocksize, max_order, autoc);
768

    
769
        compute_lpc_coefs(autoc, max_order, lpc, ref);
770
    }else{
771
        LLSModel m[2];
772
        double var[MAX_LPC_ORDER+1], weight;
773

    
774
        for(pass=0; pass<use_lpc-1; pass++){
775
            av_init_lls(&m[pass&1], max_order);
776

    
777
            weight=0;
778
            for(i=max_order; i<blocksize; i++){
779
                for(j=0; j<=max_order; j++)
780
                    var[j]= samples[i-j];
781

    
782
                if(pass){
783
                    double eval, inv, rinv;
784
                    eval= av_evaluate_lls(&m[(pass-1)&1], var+1, max_order-1);
785
                    eval= (512>>pass) + fabs(eval - var[0]);
786
                    inv = 1/eval;
787
                    rinv = sqrt(inv);
788
                    for(j=0; j<=max_order; j++)
789
                        var[j] *= rinv;
790
                    weight += inv;
791
                }else
792
                    weight++;
793

    
794
                av_update_lls(&m[pass&1], var, 1.0);
795
            }
796
            av_solve_lls(&m[pass&1], 0.001, 0);
797
        }
798

    
799
        for(i=0; i<max_order; i++){
800
            for(j=0; j<max_order; j++)
801
                lpc[i][j]= m[(pass-1)&1].coeff[i][j];
802
            ref[i]= sqrt(m[(pass-1)&1].variance[i] / weight) * (blocksize - max_order) / 4000;
803
        }
804
        for(i=max_order-1; i>0; i--)
805
            ref[i] = ref[i-1] - ref[i];
806
    }
807
    opt_order = max_order;
808

    
809
    if(omethod == ORDER_METHOD_EST) {
810
        opt_order = estimate_best_order(ref, max_order);
811
        i = opt_order-1;
812
        quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
813
    } else {
814
        for(i=0; i<max_order; i++) {
815
            quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
816
        }
817
    }
818

    
819
    return opt_order;
820
}
821

    
822

    
823
static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
824
{
825
    assert(n > 0);
826
    memcpy(res, smp, n * sizeof(int32_t));
827
}
828

    
829
static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
830
                                  int order)
831
{
832
    int i;
833

    
834
    for(i=0; i<order; i++) {
835
        res[i] = smp[i];
836
    }
837

    
838
    if(order==0){
839
        for(i=order; i<n; i++)
840
            res[i]= smp[i];
841
    }else if(order==1){
842
        for(i=order; i<n; i++)
843
            res[i]= smp[i] - smp[i-1];
844
    }else if(order==2){
845
        int a = smp[order-1] - smp[order-2];
846
        for(i=order; i<n; i+=2) {
847
            int b = smp[i] - smp[i-1];
848
            res[i]= b - a;
849
            a = smp[i+1] - smp[i];
850
            res[i+1]= a - b;
851
        }
852
    }else if(order==3){
853
        int a = smp[order-1] - smp[order-2];
854
        int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
855
        for(i=order; i<n; i+=2) {
856
            int b = smp[i] - smp[i-1];
857
            int d = b - a;
858
            res[i]= d - c;
859
            a = smp[i+1] - smp[i];
860
            c = a - b;
861
            res[i+1]= c - d;
862
        }
863
    }else{
864
        int a = smp[order-1] - smp[order-2];
865
        int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
866
        int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
867
        for(i=order; i<n; i+=2) {
868
            int b = smp[i] - smp[i-1];
869
            int d = b - a;
870
            int f = d - c;
871
            res[i]= f - e;
872
            a = smp[i+1] - smp[i];
873
            c = a - b;
874
            e = c - d;
875
            res[i+1]= e - f;
876
        }
877
    }
878
}
879

    
880
#define LPC1(x) {\
881
    int c = coefs[(x)-1];\
882
    p0 += c*s;\
883
    s = smp[i-(x)+1];\
884
    p1 += c*s;\
885
}
886

    
887
static av_always_inline void encode_residual_lpc_unrolled(
888
    int32_t *res, const int32_t *smp, int n,
889
    int order, const int32_t *coefs, int shift, int big)
890
{
891
    int i;
892
    for(i=order; i<n; i+=2) {
893
        int s = smp[i-order];
894
        int p0 = 0, p1 = 0;
895
        if(big) {
896
            switch(order) {
897
                case 32: LPC1(32)
898
                case 31: LPC1(31)
899
                case 30: LPC1(30)
900
                case 29: LPC1(29)
901
                case 28: LPC1(28)
902
                case 27: LPC1(27)
903
                case 26: LPC1(26)
904
                case 25: LPC1(25)
905
                case 24: LPC1(24)
906
                case 23: LPC1(23)
907
                case 22: LPC1(22)
908
                case 21: LPC1(21)
909
                case 20: LPC1(20)
910
                case 19: LPC1(19)
911
                case 18: LPC1(18)
912
                case 17: LPC1(17)
913
                case 16: LPC1(16)
914
                case 15: LPC1(15)
915
                case 14: LPC1(14)
916
                case 13: LPC1(13)
917
                case 12: LPC1(12)
918
                case 11: LPC1(11)
919
                case 10: LPC1(10)
920
                case  9: LPC1( 9)
921
                         LPC1( 8)
922
                         LPC1( 7)
923
                         LPC1( 6)
924
                         LPC1( 5)
925
                         LPC1( 4)
926
                         LPC1( 3)
927
                         LPC1( 2)
928
                         LPC1( 1)
929
            }
930
        } else {
931
            switch(order) {
932
                case  8: LPC1( 8)
933
                case  7: LPC1( 7)
934
                case  6: LPC1( 6)
935
                case  5: LPC1( 5)
936
                case  4: LPC1( 4)
937
                case  3: LPC1( 3)
938
                case  2: LPC1( 2)
939
                case  1: LPC1( 1)
940
            }
941
        }
942
        res[i  ] = smp[i  ] - (p0 >> shift);
943
        res[i+1] = smp[i+1] - (p1 >> shift);
944
    }
945
}
946

    
947
static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
948
                                int order, const int32_t *coefs, int shift)
949
{
950
    int i;
951
    for(i=0; i<order; i++) {
952
        res[i] = smp[i];
953
    }
954
#ifdef CONFIG_SMALL
955
    for(i=order; i<n; i+=2) {
956
        int j;
957
        int s = smp[i];
958
        int p0 = 0, p1 = 0;
959
        for(j=0; j<order; j++) {
960
            int c = coefs[j];
961
            p1 += c*s;
962
            s = smp[i-j-1];
963
            p0 += c*s;
964
        }
965
        res[i  ] = smp[i  ] - (p0 >> shift);
966
        res[i+1] = smp[i+1] - (p1 >> shift);
967
    }
968
#else
969
    switch(order) {
970
        case  1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
971
        case  2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
972
        case  3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
973
        case  4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
974
        case  5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
975
        case  6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
976
        case  7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
977
        case  8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
978
        default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
979
    }
980
#endif
981
}
982

    
983
static int encode_residual(FlacEncodeContext *ctx, int ch)
984
{
985
    int i, n;
986
    int min_order, max_order, opt_order, precision, omethod;
987
    int min_porder, max_porder;
988
    FlacFrame *frame;
989
    FlacSubframe *sub;
990
    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
991
    int shift[MAX_LPC_ORDER];
992
    int32_t *res, *smp;
993

    
994
    frame = &ctx->frame;
995
    sub = &frame->subframes[ch];
996
    res = sub->residual;
997
    smp = sub->samples;
998
    n = frame->blocksize;
999

    
1000
    /* CONSTANT */
1001
    for(i=1; i<n; i++) {
1002
        if(smp[i] != smp[0]) break;
1003
    }
1004
    if(i == n) {
1005
        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
1006
        res[0] = smp[0];
1007
        return sub->obits;
1008
    }
1009

    
1010
    /* VERBATIM */
1011
    if(n < 5) {
1012
        sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
1013
        encode_residual_verbatim(res, smp, n);
1014
        return sub->obits * n;
1015
    }
1016

    
1017
    min_order = ctx->options.min_prediction_order;
1018
    max_order = ctx->options.max_prediction_order;
1019
    min_porder = ctx->options.min_partition_order;
1020
    max_porder = ctx->options.max_partition_order;
1021
    precision = ctx->options.lpc_coeff_precision;
1022
    omethod = ctx->options.prediction_order_method;
1023

    
1024
    /* FIXED */
1025
    if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
1026
        uint32_t bits[MAX_FIXED_ORDER+1];
1027
        if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
1028
        opt_order = 0;
1029
        bits[0] = UINT32_MAX;
1030
        for(i=min_order; i<=max_order; i++) {
1031
            encode_residual_fixed(res, smp, n, i);
1032
            bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
1033
                                             n, i, sub->obits);
1034
            if(bits[i] < bits[opt_order]) {
1035
                opt_order = i;
1036
            }
1037
        }
1038
        sub->order = opt_order;
1039
        sub->type = FLAC_SUBFRAME_FIXED;
1040
        sub->type_code = sub->type | sub->order;
1041
        if(sub->order != max_order) {
1042
            encode_residual_fixed(res, smp, n, sub->order);
1043
            return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
1044
                                          sub->order, sub->obits);
1045
        }
1046
        return bits[sub->order];
1047
    }
1048

    
1049
    /* LPC */
1050
    opt_order = lpc_calc_coefs(ctx, smp, n, max_order, precision, coefs, shift, ctx->options.use_lpc, omethod);
1051

    
1052
    if(omethod == ORDER_METHOD_2LEVEL ||
1053
       omethod == ORDER_METHOD_4LEVEL ||
1054
       omethod == ORDER_METHOD_8LEVEL) {
1055
        int levels = 1 << omethod;
1056
        uint32_t bits[levels];
1057
        int order;
1058
        int opt_index = levels-1;
1059
        opt_order = max_order-1;
1060
        bits[opt_index] = UINT32_MAX;
1061
        for(i=levels-1; i>=0; i--) {
1062
            order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
1063
            if(order < 0) order = 0;
1064
            encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
1065
            bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1066
                                           res, n, order+1, sub->obits, precision);
1067
            if(bits[i] < bits[opt_index]) {
1068
                opt_index = i;
1069
                opt_order = order;
1070
            }
1071
        }
1072
        opt_order++;
1073
    } else if(omethod == ORDER_METHOD_SEARCH) {
1074
        // brute-force optimal order search
1075
        uint32_t bits[MAX_LPC_ORDER];
1076
        opt_order = 0;
1077
        bits[0] = UINT32_MAX;
1078
        for(i=min_order-1; i<max_order; i++) {
1079
            encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
1080
            bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1081
                                           res, n, i+1, sub->obits, precision);
1082
            if(bits[i] < bits[opt_order]) {
1083
                opt_order = i;
1084
            }
1085
        }
1086
        opt_order++;
1087
    } else if(omethod == ORDER_METHOD_LOG) {
1088
        uint32_t bits[MAX_LPC_ORDER];
1089
        int step;
1090

    
1091
        opt_order= min_order - 1 + (max_order-min_order)/3;
1092
        memset(bits, -1, sizeof(bits));
1093

    
1094
        for(step=16 ;step; step>>=1){
1095
            int last= opt_order;
1096
            for(i=last-step; i<=last+step; i+= step){
1097
                if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
1098
                    continue;
1099
                encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
1100
                bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
1101
                                            res, n, i+1, sub->obits, precision);
1102
                if(bits[i] < bits[opt_order])
1103
                    opt_order= i;
1104
            }
1105
        }
1106
        opt_order++;
1107
    }
1108

    
1109
    sub->order = opt_order;
1110
    sub->type = FLAC_SUBFRAME_LPC;
1111
    sub->type_code = sub->type | (sub->order-1);
1112
    sub->shift = shift[sub->order-1];
1113
    for(i=0; i<sub->order; i++) {
1114
        sub->coefs[i] = coefs[sub->order-1][i];
1115
    }
1116
    encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
1117
    return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
1118
                                sub->obits, precision);
1119
}
1120

    
1121
static int encode_residual_v(FlacEncodeContext *ctx, int ch)
1122
{
1123
    int i, n;
1124
    FlacFrame *frame;
1125
    FlacSubframe *sub;
1126
    int32_t *res, *smp;
1127

    
1128
    frame = &ctx->frame;
1129
    sub = &frame->subframes[ch];
1130
    res = sub->residual;
1131
    smp = sub->samples;
1132
    n = frame->blocksize;
1133

    
1134
    /* CONSTANT */
1135
    for(i=1; i<n; i++) {
1136
        if(smp[i] != smp[0]) break;
1137
    }
1138
    if(i == n) {
1139
        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
1140
        res[0] = smp[0];
1141
        return sub->obits;
1142
    }
1143

    
1144
    /* VERBATIM */
1145
    sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
1146
    encode_residual_verbatim(res, smp, n);
1147
    return sub->obits * n;
1148
}
1149

    
1150
static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
1151
{
1152
    int i, best;
1153
    int32_t lt, rt;
1154
    uint64_t sum[4];
1155
    uint64_t score[4];
1156
    int k;
1157

    
1158
    /* calculate sum of 2nd order residual for each channel */
1159
    sum[0] = sum[1] = sum[2] = sum[3] = 0;
1160
    for(i=2; i<n; i++) {
1161
        lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
1162
        rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1163
        sum[2] += FFABS((lt + rt) >> 1);
1164
        sum[3] += FFABS(lt - rt);
1165
        sum[0] += FFABS(lt);
1166
        sum[1] += FFABS(rt);
1167
    }
1168
    /* estimate bit counts */
1169
    for(i=0; i<4; i++) {
1170
        k = find_optimal_param(2*sum[i], n);
1171
        sum[i] = rice_encode_count(2*sum[i], n, k);
1172
    }
1173

    
1174
    /* calculate score for each mode */
1175
    score[0] = sum[0] + sum[1];
1176
    score[1] = sum[0] + sum[3];
1177
    score[2] = sum[1] + sum[3];
1178
    score[3] = sum[2] + sum[3];
1179

    
1180
    /* return mode with lowest score */
1181
    best = 0;
1182
    for(i=1; i<4; i++) {
1183
        if(score[i] < score[best]) {
1184
            best = i;
1185
        }
1186
    }
1187
    if(best == 0) {
1188
        return FLAC_CHMODE_LEFT_RIGHT;
1189
    } else if(best == 1) {
1190
        return FLAC_CHMODE_LEFT_SIDE;
1191
    } else if(best == 2) {
1192
        return FLAC_CHMODE_RIGHT_SIDE;
1193
    } else {
1194
        return FLAC_CHMODE_MID_SIDE;
1195
    }
1196
}
1197

    
1198
/**
1199
 * Perform stereo channel decorrelation
1200
 */
1201
static void channel_decorrelation(FlacEncodeContext *ctx)
1202
{
1203
    FlacFrame *frame;
1204
    int32_t *left, *right;
1205
    int i, n;
1206

    
1207
    frame = &ctx->frame;
1208
    n = frame->blocksize;
1209
    left  = frame->subframes[0].samples;
1210
    right = frame->subframes[1].samples;
1211

    
1212
    if(ctx->channels != 2) {
1213
        frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
1214
        return;
1215
    }
1216

    
1217
    frame->ch_mode = estimate_stereo_mode(left, right, n);
1218

    
1219
    /* perform decorrelation and adjust bits-per-sample */
1220
    if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
1221
        return;
1222
    }
1223
    if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
1224
        int32_t tmp;
1225
        for(i=0; i<n; i++) {
1226
            tmp = left[i];
1227
            left[i] = (tmp + right[i]) >> 1;
1228
            right[i] = tmp - right[i];
1229
        }
1230
        frame->subframes[1].obits++;
1231
    } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
1232
        for(i=0; i<n; i++) {
1233
            right[i] = left[i] - right[i];
1234
        }
1235
        frame->subframes[1].obits++;
1236
    } else {
1237
        for(i=0; i<n; i++) {
1238
            left[i] -= right[i];
1239
        }
1240
        frame->subframes[0].obits++;
1241
    }
1242
}
1243

    
1244
static void put_sbits(PutBitContext *pb, int bits, int32_t val)
1245
{
1246
    assert(bits >= 0 && bits <= 31);
1247

    
1248
    put_bits(pb, bits, val & ((1<<bits)-1));
1249
}
1250

    
1251
static void write_utf8(PutBitContext *pb, uint32_t val)
1252
{
1253
    uint8_t tmp;
1254
    PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1255
}
1256

    
1257
static void output_frame_header(FlacEncodeContext *s)
1258
{
1259
    FlacFrame *frame;
1260
    int crc;
1261

    
1262
    frame = &s->frame;
1263

    
1264
    put_bits(&s->pb, 16, 0xFFF8);
1265
    put_bits(&s->pb, 4, frame->bs_code[0]);
1266
    put_bits(&s->pb, 4, s->sr_code[0]);
1267
    if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) {
1268
        put_bits(&s->pb, 4, s->ch_code);
1269
    } else {
1270
        put_bits(&s->pb, 4, frame->ch_mode);
1271
    }
1272
    put_bits(&s->pb, 3, 4); /* bits-per-sample code */
1273
    put_bits(&s->pb, 1, 0);
1274
    write_utf8(&s->pb, s->frame_count);
1275
    if(frame->bs_code[0] == 6) {
1276
        put_bits(&s->pb, 8, frame->bs_code[1]);
1277
    } else if(frame->bs_code[0] == 7) {
1278
        put_bits(&s->pb, 16, frame->bs_code[1]);
1279
    }
1280
    if(s->sr_code[0] == 12) {
1281
        put_bits(&s->pb, 8, s->sr_code[1]);
1282
    } else if(s->sr_code[0] > 12) {
1283
        put_bits(&s->pb, 16, s->sr_code[1]);
1284
    }
1285
    flush_put_bits(&s->pb);
1286
    crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
1287
                 s->pb.buf, put_bits_count(&s->pb)>>3);
1288
    put_bits(&s->pb, 8, crc);
1289
}
1290

    
1291
static void output_subframe_constant(FlacEncodeContext *s, int ch)
1292
{
1293
    FlacSubframe *sub;
1294
    int32_t res;
1295

    
1296
    sub = &s->frame.subframes[ch];
1297
    res = sub->residual[0];
1298
    put_sbits(&s->pb, sub->obits, res);
1299
}
1300

    
1301
static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
1302
{
1303
    int i;
1304
    FlacFrame *frame;
1305
    FlacSubframe *sub;
1306
    int32_t res;
1307

    
1308
    frame = &s->frame;
1309
    sub = &frame->subframes[ch];
1310

    
1311
    for(i=0; i<frame->blocksize; i++) {
1312
        res = sub->residual[i];
1313
        put_sbits(&s->pb, sub->obits, res);
1314
    }
1315
}
1316

    
1317
static void output_residual(FlacEncodeContext *ctx, int ch)
1318
{
1319
    int i, j, p, n, parts;
1320
    int k, porder, psize, res_cnt;
1321
    FlacFrame *frame;
1322
    FlacSubframe *sub;
1323
    int32_t *res;
1324

    
1325
    frame = &ctx->frame;
1326
    sub = &frame->subframes[ch];
1327
    res = sub->residual;
1328
    n = frame->blocksize;
1329

    
1330
    /* rice-encoded block */
1331
    put_bits(&ctx->pb, 2, 0);
1332

    
1333
    /* partition order */
1334
    porder = sub->rc.porder;
1335
    psize = n >> porder;
1336
    parts = (1 << porder);
1337
    put_bits(&ctx->pb, 4, porder);
1338
    res_cnt = psize - sub->order;
1339

    
1340
    /* residual */
1341
    j = sub->order;
1342
    for(p=0; p<parts; p++) {
1343
        k = sub->rc.params[p];
1344
        put_bits(&ctx->pb, 4, k);
1345
        if(p == 1) res_cnt = psize;
1346
        for(i=0; i<res_cnt && j<n; i++, j++) {
1347
            set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1348
        }
1349
    }
1350
}
1351

    
1352
static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1353
{
1354
    int i;
1355
    FlacFrame *frame;
1356
    FlacSubframe *sub;
1357

    
1358
    frame = &ctx->frame;
1359
    sub = &frame->subframes[ch];
1360

    
1361
    /* warm-up samples */
1362
    for(i=0; i<sub->order; i++) {
1363
        put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1364
    }
1365

    
1366
    /* residual */
1367
    output_residual(ctx, ch);
1368
}
1369

    
1370
static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
1371
{
1372
    int i, cbits;
1373
    FlacFrame *frame;
1374
    FlacSubframe *sub;
1375

    
1376
    frame = &ctx->frame;
1377
    sub = &frame->subframes[ch];
1378

    
1379
    /* warm-up samples */
1380
    for(i=0; i<sub->order; i++) {
1381
        put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
1382
    }
1383

    
1384
    /* LPC coefficients */
1385
    cbits = ctx->options.lpc_coeff_precision;
1386
    put_bits(&ctx->pb, 4, cbits-1);
1387
    put_sbits(&ctx->pb, 5, sub->shift);
1388
    for(i=0; i<sub->order; i++) {
1389
        put_sbits(&ctx->pb, cbits, sub->coefs[i]);
1390
    }
1391

    
1392
    /* residual */
1393
    output_residual(ctx, ch);
1394
}
1395

    
1396
static void output_subframes(FlacEncodeContext *s)
1397
{
1398
    FlacFrame *frame;
1399
    FlacSubframe *sub;
1400
    int ch;
1401

    
1402
    frame = &s->frame;
1403

    
1404
    for(ch=0; ch<s->channels; ch++) {
1405
        sub = &frame->subframes[ch];
1406

    
1407
        /* subframe header */
1408
        put_bits(&s->pb, 1, 0);
1409
        put_bits(&s->pb, 6, sub->type_code);
1410
        put_bits(&s->pb, 1, 0); /* no wasted bits */
1411

    
1412
        /* subframe */
1413
        if(sub->type == FLAC_SUBFRAME_CONSTANT) {
1414
            output_subframe_constant(s, ch);
1415
        } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1416
            output_subframe_verbatim(s, ch);
1417
        } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1418
            output_subframe_fixed(s, ch);
1419
        } else if(sub->type == FLAC_SUBFRAME_LPC) {
1420
            output_subframe_lpc(s, ch);
1421
        }
1422
    }
1423
}
1424

    
1425
static void output_frame_footer(FlacEncodeContext *s)
1426
{
1427
    int crc;
1428
    flush_put_bits(&s->pb);
1429
    crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
1430
                          s->pb.buf, put_bits_count(&s->pb)>>3));
1431
    put_bits(&s->pb, 16, crc);
1432
    flush_put_bits(&s->pb);
1433
}
1434

    
1435
static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
1436
                             int buf_size, void *data)
1437
{
1438
    int ch;
1439
    FlacEncodeContext *s;
1440
    int16_t *samples = data;
1441
    int out_bytes;
1442

    
1443
    s = avctx->priv_data;
1444

    
1445
    s->blocksize = avctx->frame_size;
1446
    init_frame(s);
1447

    
1448
    copy_samples(s, samples);
1449

    
1450
    channel_decorrelation(s);
1451

    
1452
    for(ch=0; ch<s->channels; ch++) {
1453
        encode_residual(s, ch);
1454
    }
1455
    init_put_bits(&s->pb, frame, buf_size);
1456
    output_frame_header(s);
1457
    output_subframes(s);
1458
    output_frame_footer(s);
1459
    out_bytes = put_bits_count(&s->pb) >> 3;
1460

    
1461
    if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1462
        /* frame too large. use verbatim mode */
1463
        for(ch=0; ch<s->channels; ch++) {
1464
            encode_residual_v(s, ch);
1465
        }
1466
        init_put_bits(&s->pb, frame, buf_size);
1467
        output_frame_header(s);
1468
        output_subframes(s);
1469
        output_frame_footer(s);
1470
        out_bytes = put_bits_count(&s->pb) >> 3;
1471

    
1472
        if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
1473
            /* still too large. must be an error. */
1474
            av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
1475
            return -1;
1476
        }
1477
    }
1478

    
1479
    s->frame_count++;
1480
    return out_bytes;
1481
}
1482

    
1483
static int flac_encode_close(AVCodecContext *avctx)
1484
{
1485
    av_freep(&avctx->extradata);
1486
    avctx->extradata_size = 0;
1487
    av_freep(&avctx->coded_frame);
1488
    return 0;
1489
}
1490

    
1491
AVCodec flac_encoder = {
1492
    "flac",
1493
    CODEC_TYPE_AUDIO,
1494
    CODEC_ID_FLAC,
1495
    sizeof(FlacEncodeContext),
1496
    flac_encode_init,
1497
    flac_encode_frame,
1498
    flac_encode_close,
1499
    NULL,
1500
    .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
1501
};