Statistics
| Branch: | Revision:

ffmpeg / libavcodec / alacenc.c @ b2755007

History | View | Annotate | Download (15.4 KB)

1
/**
2
 * ALAC audio encoder
3
 * Copyright (c) 2008  Jaikrishnan Menon <realityman@gmx.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 "put_bits.h"
25
#include "dsputil.h"
26
#include "lpc.h"
27
#include "mathops.h"
28

    
29
#define DEFAULT_FRAME_SIZE        4096
30
#define DEFAULT_SAMPLE_SIZE       16
31
#define MAX_CHANNELS              8
32
#define ALAC_EXTRADATA_SIZE       36
33
#define ALAC_FRAME_HEADER_SIZE    55
34
#define ALAC_FRAME_FOOTER_SIZE    3
35

    
36
#define ALAC_ESCAPE_CODE          0x1FF
37
#define ALAC_MAX_LPC_ORDER        30
38
#define DEFAULT_MAX_PRED_ORDER    6
39
#define DEFAULT_MIN_PRED_ORDER    4
40
#define ALAC_MAX_LPC_PRECISION    9
41
#define ALAC_MAX_LPC_SHIFT        9
42

    
43
#define ALAC_CHMODE_LEFT_RIGHT    0
44
#define ALAC_CHMODE_LEFT_SIDE     1
45
#define ALAC_CHMODE_RIGHT_SIDE    2
46
#define ALAC_CHMODE_MID_SIDE      3
47

    
48
typedef struct RiceContext {
49
    int history_mult;
50
    int initial_history;
51
    int k_modifier;
52
    int rice_modifier;
53
} RiceContext;
54

    
55
typedef struct LPCContext {
56
    int lpc_order;
57
    int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
58
    int lpc_quant;
59
} LPCContext;
60

    
61
typedef struct AlacEncodeContext {
62
    int compression_level;
63
    int min_prediction_order;
64
    int max_prediction_order;
65
    int max_coded_frame_size;
66
    int write_sample_size;
67
    int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
68
    int32_t predictor_buf[DEFAULT_FRAME_SIZE];
69
    int interlacing_shift;
70
    int interlacing_leftweight;
71
    PutBitContext pbctx;
72
    RiceContext rc;
73
    LPCContext lpc[MAX_CHANNELS];
74
    DSPContext dspctx;
75
    AVCodecContext *avctx;
76
} AlacEncodeContext;
77

    
78

    
79
static void init_sample_buffers(AlacEncodeContext *s, int16_t *input_samples)
80
{
81
    int ch, i;
82

    
83
    for(ch=0;ch<s->avctx->channels;ch++) {
84
        int16_t *sptr = input_samples + ch;
85
        for(i=0;i<s->avctx->frame_size;i++) {
86
            s->sample_buf[ch][i] = *sptr;
87
            sptr += s->avctx->channels;
88
        }
89
    }
90
}
91

    
92
static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_size)
93
{
94
    int divisor, q, r;
95

    
96
    k = FFMIN(k, s->rc.k_modifier);
97
    divisor = (1<<k) - 1;
98
    q = x / divisor;
99
    r = x % divisor;
100

    
101
    if(q > 8) {
102
        // write escape code and sample value directly
103
        put_bits(&s->pbctx, 9, ALAC_ESCAPE_CODE);
104
        put_bits(&s->pbctx, write_sample_size, x);
105
    } else {
106
        if(q)
107
            put_bits(&s->pbctx, q, (1<<q) - 1);
108
        put_bits(&s->pbctx, 1, 0);
109

    
110
        if(k != 1) {
111
            if(r > 0)
112
                put_bits(&s->pbctx, k, r+1);
113
            else
114
                put_bits(&s->pbctx, k-1, 0);
115
        }
116
    }
117
}
118

    
119
static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
120
{
121
    put_bits(&s->pbctx, 3,  s->avctx->channels-1);          // No. of channels -1
122
    put_bits(&s->pbctx, 16, 0);                             // Seems to be zero
123
    put_bits(&s->pbctx, 1,  1);                             // Sample count is in the header
124
    put_bits(&s->pbctx, 2,  0);                             // FIXME: Wasted bytes field
125
    put_bits(&s->pbctx, 1,  is_verbatim);                   // Audio block is verbatim
126
    put_bits(&s->pbctx, 32, s->avctx->frame_size);          // No. of samples in the frame
127
}
128

    
129
static void calc_predictor_params(AlacEncodeContext *s, int ch)
130
{
131
    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
132
    int shift[MAX_LPC_ORDER];
133
    int opt_order;
134

    
135
    opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch], s->avctx->frame_size, s->min_prediction_order, s->max_prediction_order,
136
                                   ALAC_MAX_LPC_PRECISION, coefs, shift, 1, ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
137

    
138
    s->lpc[ch].lpc_order = opt_order;
139
    s->lpc[ch].lpc_quant = shift[opt_order-1];
140
    memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
141
}
142

    
143
static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
144
{
145
    int i, best;
146
    int32_t lt, rt;
147
    uint64_t sum[4];
148
    uint64_t score[4];
149

    
150
    /* calculate sum of 2nd order residual for each channel */
151
    sum[0] = sum[1] = sum[2] = sum[3] = 0;
152
    for(i=2; i<n; i++) {
153
        lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
154
        rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
155
        sum[2] += FFABS((lt + rt) >> 1);
156
        sum[3] += FFABS(lt - rt);
157
        sum[0] += FFABS(lt);
158
        sum[1] += FFABS(rt);
159
    }
160

    
161
    /* calculate score for each mode */
162
    score[0] = sum[0] + sum[1];
163
    score[1] = sum[0] + sum[3];
164
    score[2] = sum[1] + sum[3];
165
    score[3] = sum[2] + sum[3];
166

    
167
    /* return mode with lowest score */
168
    best = 0;
169
    for(i=1; i<4; i++) {
170
        if(score[i] < score[best]) {
171
            best = i;
172
        }
173
    }
174
    return best;
175
}
176

    
177
static void alac_stereo_decorrelation(AlacEncodeContext *s)
178
{
179
    int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
180
    int i, mode, n = s->avctx->frame_size;
181
    int32_t tmp;
182

    
183
    mode = estimate_stereo_mode(left, right, n);
184

    
185
    switch(mode)
186
    {
187
        case ALAC_CHMODE_LEFT_RIGHT:
188
            s->interlacing_leftweight = 0;
189
            s->interlacing_shift = 0;
190
            break;
191

    
192
        case ALAC_CHMODE_LEFT_SIDE:
193
            for(i=0; i<n; i++) {
194
                right[i] = left[i] - right[i];
195
            }
196
            s->interlacing_leftweight = 1;
197
            s->interlacing_shift = 0;
198
            break;
199

    
200
        case ALAC_CHMODE_RIGHT_SIDE:
201
            for(i=0; i<n; i++) {
202
                tmp = right[i];
203
                right[i] = left[i] - right[i];
204
                left[i] = tmp + (right[i] >> 31);
205
            }
206
            s->interlacing_leftweight = 1;
207
            s->interlacing_shift = 31;
208
            break;
209

    
210
        default:
211
            for(i=0; i<n; i++) {
212
                tmp = left[i];
213
                left[i] = (tmp + right[i]) >> 1;
214
                right[i] = tmp - right[i];
215
            }
216
            s->interlacing_leftweight = 1;
217
            s->interlacing_shift = 1;
218
            break;
219
    }
220
}
221

    
222
static void alac_linear_predictor(AlacEncodeContext *s, int ch)
223
{
224
    int i;
225
    LPCContext lpc = s->lpc[ch];
226

    
227
    if(lpc.lpc_order == 31) {
228
        s->predictor_buf[0] = s->sample_buf[ch][0];
229

    
230
        for(i=1; i<s->avctx->frame_size; i++)
231
            s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
232

    
233
        return;
234
    }
235

    
236
    // generalised linear predictor
237

    
238
    if(lpc.lpc_order > 0) {
239
        int32_t *samples  = s->sample_buf[ch];
240
        int32_t *residual = s->predictor_buf;
241

    
242
        // generate warm-up samples
243
        residual[0] = samples[0];
244
        for(i=1;i<=lpc.lpc_order;i++)
245
            residual[i] = samples[i] - samples[i-1];
246

    
247
        // perform lpc on remaining samples
248
        for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
249
            int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
250

    
251
            for (j = 0; j < lpc.lpc_order; j++) {
252
                sum += (samples[lpc.lpc_order-j] - samples[0]) *
253
                        lpc.lpc_coeff[j];
254
            }
255

    
256
            sum >>= lpc.lpc_quant;
257
            sum += samples[0];
258
            residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum,
259
                                      s->write_sample_size);
260
            res_val = residual[i];
261

    
262
            if(res_val) {
263
                int index = lpc.lpc_order - 1;
264
                int neg = (res_val < 0);
265

    
266
                while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
267
                    int val = samples[0] - samples[lpc.lpc_order - index];
268
                    int sign = (val ? FFSIGN(val) : 0);
269

    
270
                    if(neg)
271
                        sign*=-1;
272

    
273
                    lpc.lpc_coeff[index] -= sign;
274
                    val *= sign;
275
                    res_val -= ((val >> lpc.lpc_quant) *
276
                            (lpc.lpc_order - index));
277
                    index--;
278
                }
279
            }
280
            samples++;
281
        }
282
    }
283
}
284

    
285
static void alac_entropy_coder(AlacEncodeContext *s)
286
{
287
    unsigned int history = s->rc.initial_history;
288
    int sign_modifier = 0, i, k;
289
    int32_t *samples = s->predictor_buf;
290

    
291
    for(i=0;i < s->avctx->frame_size;) {
292
        int x;
293

    
294
        k = av_log2((history >> 9) + 3);
295

    
296
        x = -2*(*samples)-1;
297
        x ^= (x>>31);
298

    
299
        samples++;
300
        i++;
301

    
302
        encode_scalar(s, x - sign_modifier, k, s->write_sample_size);
303

    
304
        history += x * s->rc.history_mult
305
                   - ((history * s->rc.history_mult) >> 9);
306

    
307
        sign_modifier = 0;
308
        if(x > 0xFFFF)
309
            history = 0xFFFF;
310

    
311
        if((history < 128) && (i < s->avctx->frame_size)) {
312
            unsigned int block_size = 0;
313

    
314
            k = 7 - av_log2(history) + ((history + 16) >> 6);
315

    
316
            while((*samples == 0) && (i < s->avctx->frame_size)) {
317
                samples++;
318
                i++;
319
                block_size++;
320
            }
321
            encode_scalar(s, block_size, k, 16);
322

    
323
            sign_modifier = (block_size <= 0xFFFF);
324

    
325
            history = 0;
326
        }
327

    
328
    }
329
}
330

    
331
static void write_compressed_frame(AlacEncodeContext *s)
332
{
333
    int i, j;
334

    
335
    /* only simple mid/side decorrelation supported as of now */
336
    if(s->avctx->channels == 2)
337
        alac_stereo_decorrelation(s);
338
    put_bits(&s->pbctx, 8, s->interlacing_shift);
339
    put_bits(&s->pbctx, 8, s->interlacing_leftweight);
340

    
341
    for(i=0;i<s->avctx->channels;i++) {
342

    
343
        calc_predictor_params(s, i);
344

    
345
        put_bits(&s->pbctx, 4, 0);  // prediction type : currently only type 0 has been RE'd
346
        put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant);
347

    
348
        put_bits(&s->pbctx, 3, s->rc.rice_modifier);
349
        put_bits(&s->pbctx, 5, s->lpc[i].lpc_order);
350
        // predictor coeff. table
351
        for(j=0;j<s->lpc[i].lpc_order;j++) {
352
            put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]);
353
        }
354
    }
355

    
356
    // apply lpc and entropy coding to audio samples
357

    
358
    for(i=0;i<s->avctx->channels;i++) {
359
        alac_linear_predictor(s, i);
360
        alac_entropy_coder(s);
361
    }
362
}
363

    
364
static av_cold int alac_encode_init(AVCodecContext *avctx)
365
{
366
    AlacEncodeContext *s    = avctx->priv_data;
367
    uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1);
368

    
369
    avctx->frame_size      = DEFAULT_FRAME_SIZE;
370
    avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE;
371

    
372
    if(avctx->sample_fmt != SAMPLE_FMT_S16) {
373
        av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
374
        return -1;
375
    }
376

    
377
    // Set default compression level
378
    if(avctx->compression_level == FF_COMPRESSION_DEFAULT)
379
        s->compression_level = 1;
380
    else
381
        s->compression_level = av_clip(avctx->compression_level, 0, 1);
382

    
383
    // Initialize default Rice parameters
384
    s->rc.history_mult    = 40;
385
    s->rc.initial_history = 10;
386
    s->rc.k_modifier      = 14;
387
    s->rc.rice_modifier   = 4;
388

    
389
    s->max_coded_frame_size = (ALAC_FRAME_HEADER_SIZE + ALAC_FRAME_FOOTER_SIZE +
390
                               avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample)>>3;
391

    
392
    s->write_sample_size  = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
393

    
394
    AV_WB32(alac_extradata,    ALAC_EXTRADATA_SIZE);
395
    AV_WB32(alac_extradata+4,  MKBETAG('a','l','a','c'));
396
    AV_WB32(alac_extradata+12, avctx->frame_size);
397
    AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample);
398
    AV_WB8 (alac_extradata+21, avctx->channels);
399
    AV_WB32(alac_extradata+24, s->max_coded_frame_size);
400
    AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate
401
    AV_WB32(alac_extradata+32, avctx->sample_rate);
402

    
403
    // Set relevant extradata fields
404
    if(s->compression_level > 0) {
405
        AV_WB8(alac_extradata+18, s->rc.history_mult);
406
        AV_WB8(alac_extradata+19, s->rc.initial_history);
407
        AV_WB8(alac_extradata+20, s->rc.k_modifier);
408
    }
409

    
410
    s->min_prediction_order = DEFAULT_MIN_PRED_ORDER;
411
    if(avctx->min_prediction_order >= 0) {
412
        if(avctx->min_prediction_order < MIN_LPC_ORDER ||
413
           avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) {
414
            av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
415
                return -1;
416
        }
417

    
418
        s->min_prediction_order = avctx->min_prediction_order;
419
    }
420

    
421
    s->max_prediction_order = DEFAULT_MAX_PRED_ORDER;
422
    if(avctx->max_prediction_order >= 0) {
423
        if(avctx->max_prediction_order < MIN_LPC_ORDER ||
424
           avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) {
425
            av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
426
                return -1;
427
        }
428

    
429
        s->max_prediction_order = avctx->max_prediction_order;
430
    }
431

    
432
    if(s->max_prediction_order < s->min_prediction_order) {
433
        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
434
               s->min_prediction_order, s->max_prediction_order);
435
        return -1;
436
    }
437

    
438
    avctx->extradata = alac_extradata;
439
    avctx->extradata_size = ALAC_EXTRADATA_SIZE;
440

    
441
    avctx->coded_frame = avcodec_alloc_frame();
442
    avctx->coded_frame->key_frame = 1;
443

    
444
    s->avctx = avctx;
445
    dsputil_init(&s->dspctx, avctx);
446

    
447
    return 0;
448
}
449

    
450
static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
451
                             int buf_size, void *data)
452
{
453
    AlacEncodeContext *s = avctx->priv_data;
454
    PutBitContext *pb = &s->pbctx;
455
    int i, out_bytes, verbatim_flag = 0;
456

    
457
    if(avctx->frame_size > DEFAULT_FRAME_SIZE) {
458
        av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n");
459
        return -1;
460
    }
461

    
462
    if(buf_size < 2*s->max_coded_frame_size) {
463
        av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
464
        return -1;
465
    }
466

    
467
verbatim:
468
    init_put_bits(pb, frame, buf_size);
469

    
470
    if((s->compression_level == 0) || verbatim_flag) {
471
        // Verbatim mode
472
        int16_t *samples = data;
473
        write_frame_header(s, 1);
474
        for(i=0; i<avctx->frame_size*avctx->channels; i++) {
475
            put_sbits(pb, 16, *samples++);
476
        }
477
    } else {
478
        init_sample_buffers(s, data);
479
        write_frame_header(s, 0);
480
        write_compressed_frame(s);
481
    }
482

    
483
    put_bits(pb, 3, 7);
484
    flush_put_bits(pb);
485
    out_bytes = put_bits_count(pb) >> 3;
486

    
487
    if(out_bytes > s->max_coded_frame_size) {
488
        /* frame too large. use verbatim mode */
489
        if(verbatim_flag || (s->compression_level == 0)) {
490
            /* still too large. must be an error. */
491
            av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
492
            return -1;
493
        }
494
        verbatim_flag = 1;
495
        goto verbatim;
496
    }
497

    
498
    return out_bytes;
499
}
500

    
501
static av_cold int alac_encode_close(AVCodecContext *avctx)
502
{
503
    av_freep(&avctx->extradata);
504
    avctx->extradata_size = 0;
505
    av_freep(&avctx->coded_frame);
506
    return 0;
507
}
508

    
509
AVCodec alac_encoder = {
510
    "alac",
511
    CODEC_TYPE_AUDIO,
512
    CODEC_ID_ALAC,
513
    sizeof(AlacEncodeContext),
514
    alac_encode_init,
515
    alac_encode_frame,
516
    alac_encode_close,
517
    .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
518
    .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
519
};