Statistics
| Branch: | Revision:

ffmpeg / libavcodec / apedec.c @ 2912e87a

History | View | Annotate | Download (27.2 KB)

1
/*
2
 * Monkey's Audio lossless audio decoder
3
 * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org>
4
 *  based upon libdemac from Dave Chapman.
5
 *
6
 * This file is part of Libav.
7
 *
8
 * Libav is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * Libav is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with Libav; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22

    
23
#define ALT_BITSTREAM_READER_LE
24
#include "avcodec.h"
25
#include "dsputil.h"
26
#include "get_bits.h"
27
#include "bytestream.h"
28
#include "libavutil/audioconvert.h"
29

    
30
/**
31
 * @file
32
 * Monkey's Audio lossless audio decoder
33
 */
34

    
35
#define BLOCKS_PER_LOOP     4608
36
#define MAX_CHANNELS        2
37
#define MAX_BYTESPERSAMPLE  3
38

    
39
#define APE_FRAMECODE_MONO_SILENCE    1
40
#define APE_FRAMECODE_STEREO_SILENCE  3
41
#define APE_FRAMECODE_PSEUDO_STEREO   4
42

    
43
#define HISTORY_SIZE 512
44
#define PREDICTOR_ORDER 8
45
/** Total size of all predictor histories */
46
#define PREDICTOR_SIZE 50
47

    
48
#define YDELAYA (18 + PREDICTOR_ORDER*4)
49
#define YDELAYB (18 + PREDICTOR_ORDER*3)
50
#define XDELAYA (18 + PREDICTOR_ORDER*2)
51
#define XDELAYB (18 + PREDICTOR_ORDER)
52

    
53
#define YADAPTCOEFFSA 18
54
#define XADAPTCOEFFSA 14
55
#define YADAPTCOEFFSB 10
56
#define XADAPTCOEFFSB 5
57

    
58
/**
59
 * Possible compression levels
60
 * @{
61
 */
62
enum APECompressionLevel {
63
    COMPRESSION_LEVEL_FAST       = 1000,
64
    COMPRESSION_LEVEL_NORMAL     = 2000,
65
    COMPRESSION_LEVEL_HIGH       = 3000,
66
    COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
67
    COMPRESSION_LEVEL_INSANE     = 5000
68
};
69
/** @} */
70

    
71
#define APE_FILTER_LEVELS 3
72

    
73
/** Filter orders depending on compression level */
74
static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
75
    {  0,   0,    0 },
76
    { 16,   0,    0 },
77
    { 64,   0,    0 },
78
    { 32, 256,    0 },
79
    { 16, 256, 1280 }
80
};
81

    
82
/** Filter fraction bits depending on compression level */
83
static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
84
    {  0,  0,  0 },
85
    { 11,  0,  0 },
86
    { 11,  0,  0 },
87
    { 10, 13,  0 },
88
    { 11, 13, 15 }
89
};
90

    
91

    
92
/** Filters applied to the decoded data */
93
typedef struct APEFilter {
94
    int16_t *coeffs;        ///< actual coefficients used in filtering
95
    int16_t *adaptcoeffs;   ///< adaptive filter coefficients used for correcting of actual filter coefficients
96
    int16_t *historybuffer; ///< filter memory
97
    int16_t *delay;         ///< filtered values
98

    
99
    int avg;
100
} APEFilter;
101

    
102
typedef struct APERice {
103
    uint32_t k;
104
    uint32_t ksum;
105
} APERice;
106

    
107
typedef struct APERangecoder {
108
    uint32_t low;           ///< low end of interval
109
    uint32_t range;         ///< length of interval
110
    uint32_t help;          ///< bytes_to_follow resp. intermediate value
111
    unsigned int buffer;    ///< buffer for input/output
112
} APERangecoder;
113

    
114
/** Filter histories */
115
typedef struct APEPredictor {
116
    int32_t *buf;
117

    
118
    int32_t lastA[2];
119

    
120
    int32_t filterA[2];
121
    int32_t filterB[2];
122

    
123
    int32_t coeffsA[2][4];  ///< adaption coefficients
124
    int32_t coeffsB[2][5];  ///< adaption coefficients
125
    int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
126
} APEPredictor;
127

    
128
/** Decoder context */
129
typedef struct APEContext {
130
    AVCodecContext *avctx;
131
    DSPContext dsp;
132
    int channels;
133
    int samples;                             ///< samples left to decode in current frame
134

    
135
    int fileversion;                         ///< codec version, very important in decoding process
136
    int compression_level;                   ///< compression levels
137
    int fset;                                ///< which filter set to use (calculated from compression level)
138
    int flags;                               ///< global decoder flags
139

    
140
    uint32_t CRC;                            ///< frame CRC
141
    int frameflags;                          ///< frame flags
142
    int currentframeblocks;                  ///< samples (per channel) in current frame
143
    int blocksdecoded;                       ///< count of decoded samples in current frame
144
    APEPredictor predictor;                  ///< predictor used for final reconstruction
145

    
146
    int32_t decoded0[BLOCKS_PER_LOOP];       ///< decoded data for the first channel
147
    int32_t decoded1[BLOCKS_PER_LOOP];       ///< decoded data for the second channel
148

    
149
    int16_t* filterbuf[APE_FILTER_LEVELS];   ///< filter memory
150

    
151
    APERangecoder rc;                        ///< rangecoder used to decode actual values
152
    APERice riceX;                           ///< rice code parameters for the second channel
153
    APERice riceY;                           ///< rice code parameters for the first channel
154
    APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
155

    
156
    uint8_t *data;                           ///< current frame data
157
    uint8_t *data_end;                       ///< frame data end
158
    const uint8_t *ptr;                      ///< current position in frame data
159
    const uint8_t *last_ptr;                 ///< position where last 4608-sample block ended
160

    
161
    int error;
162
} APEContext;
163

    
164
// TODO: dsputilize
165

    
166
static av_cold int ape_decode_init(AVCodecContext * avctx)
167
{
168
    APEContext *s = avctx->priv_data;
169
    int i;
170

    
171
    if (avctx->extradata_size != 6) {
172
        av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
173
        return -1;
174
    }
175
    if (avctx->bits_per_coded_sample != 16) {
176
        av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n");
177
        return -1;
178
    }
179
    if (avctx->channels > 2) {
180
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
181
        return -1;
182
    }
183
    s->avctx             = avctx;
184
    s->channels          = avctx->channels;
185
    s->fileversion       = AV_RL16(avctx->extradata);
186
    s->compression_level = AV_RL16(avctx->extradata + 2);
187
    s->flags             = AV_RL16(avctx->extradata + 4);
188

    
189
    av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n", s->compression_level, s->flags);
190
    if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
191
        av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n", s->compression_level);
192
        return -1;
193
    }
194
    s->fset = s->compression_level / 1000 - 1;
195
    for (i = 0; i < APE_FILTER_LEVELS; i++) {
196
        if (!ape_filter_orders[s->fset][i])
197
            break;
198
        s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4);
199
    }
200

    
201
    dsputil_init(&s->dsp, avctx);
202
    avctx->sample_fmt = AV_SAMPLE_FMT_S16;
203
    avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
204
    return 0;
205
}
206

    
207
static av_cold int ape_decode_close(AVCodecContext * avctx)
208
{
209
    APEContext *s = avctx->priv_data;
210
    int i;
211

    
212
    for (i = 0; i < APE_FILTER_LEVELS; i++)
213
        av_freep(&s->filterbuf[i]);
214

    
215
    av_freep(&s->data);
216
    return 0;
217
}
218

    
219
/**
220
 * @defgroup rangecoder APE range decoder
221
 * @{
222
 */
223

    
224
#define CODE_BITS    32
225
#define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
226
#define SHIFT_BITS   (CODE_BITS - 9)
227
#define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
228
#define BOTTOM_VALUE (TOP_VALUE >> 8)
229

    
230
/** Start the decoder */
231
static inline void range_start_decoding(APEContext * ctx)
232
{
233
    ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
234
    ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
235
    ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
236
}
237

    
238
/** Perform normalization */
239
static inline void range_dec_normalize(APEContext * ctx)
240
{
241
    while (ctx->rc.range <= BOTTOM_VALUE) {
242
        ctx->rc.buffer <<= 8;
243
        if(ctx->ptr < ctx->data_end)
244
            ctx->rc.buffer += *ctx->ptr;
245
        ctx->ptr++;
246
        ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
247
        ctx->rc.range  <<= 8;
248
    }
249
}
250

    
251
/**
252
 * Calculate culmulative frequency for next symbol. Does NO update!
253
 * @param ctx decoder context
254
 * @param tot_f is the total frequency or (code_value)1<<shift
255
 * @return the culmulative frequency
256
 */
257
static inline int range_decode_culfreq(APEContext * ctx, int tot_f)
258
{
259
    range_dec_normalize(ctx);
260
    ctx->rc.help = ctx->rc.range / tot_f;
261
    return ctx->rc.low / ctx->rc.help;
262
}
263

    
264
/**
265
 * Decode value with given size in bits
266
 * @param ctx decoder context
267
 * @param shift number of bits to decode
268
 */
269
static inline int range_decode_culshift(APEContext * ctx, int shift)
270
{
271
    range_dec_normalize(ctx);
272
    ctx->rc.help = ctx->rc.range >> shift;
273
    return ctx->rc.low / ctx->rc.help;
274
}
275

    
276

    
277
/**
278
 * Update decoding state
279
 * @param ctx decoder context
280
 * @param sy_f the interval length (frequency of the symbol)
281
 * @param lt_f the lower end (frequency sum of < symbols)
282
 */
283
static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f)
284
{
285
    ctx->rc.low  -= ctx->rc.help * lt_f;
286
    ctx->rc.range = ctx->rc.help * sy_f;
287
}
288

    
289
/** Decode n bits (n <= 16) without modelling */
290
static inline int range_decode_bits(APEContext * ctx, int n)
291
{
292
    int sym = range_decode_culshift(ctx, n);
293
    range_decode_update(ctx, 1, sym);
294
    return sym;
295
}
296

    
297

    
298
#define MODEL_ELEMENTS 64
299

    
300
/**
301
 * Fixed probabilities for symbols in Monkey Audio version 3.97
302
 */
303
static const uint16_t counts_3970[22] = {
304
        0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
305
    62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
306
    65450, 65469, 65480, 65487, 65491, 65493,
307
};
308

    
309
/**
310
 * Probability ranges for symbols in Monkey Audio version 3.97
311
 */
312
static const uint16_t counts_diff_3970[21] = {
313
    14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
314
    1104, 677, 415, 248, 150, 89, 54, 31,
315
    19, 11, 7, 4, 2,
316
};
317

    
318
/**
319
 * Fixed probabilities for symbols in Monkey Audio version 3.98
320
 */
321
static const uint16_t counts_3980[22] = {
322
        0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
323
    64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
324
    65485, 65488, 65490, 65491, 65492, 65493,
325
};
326

    
327
/**
328
 * Probability ranges for symbols in Monkey Audio version 3.98
329
 */
330
static const uint16_t counts_diff_3980[21] = {
331
    19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
332
    261, 119, 65, 31, 19, 10, 6, 3,
333
    3, 2, 1, 1, 1,
334
};
335

    
336
/**
337
 * Decode symbol
338
 * @param ctx decoder context
339
 * @param counts probability range start position
340
 * @param counts_diff probability range widths
341
 */
342
static inline int range_get_symbol(APEContext * ctx,
343
                                   const uint16_t counts[],
344
                                   const uint16_t counts_diff[])
345
{
346
    int symbol, cf;
347

    
348
    cf = range_decode_culshift(ctx, 16);
349

    
350
    if(cf > 65492){
351
        symbol= cf - 65535 + 63;
352
        range_decode_update(ctx, 1, cf);
353
        if(cf > 65535)
354
            ctx->error=1;
355
        return symbol;
356
    }
357
    /* figure out the symbol inefficiently; a binary search would be much better */
358
    for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
359

    
360
    range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
361

    
362
    return symbol;
363
}
364
/** @} */ // group rangecoder
365

    
366
static inline void update_rice(APERice *rice, int x)
367
{
368
    int lim = rice->k ? (1 << (rice->k + 4)) : 0;
369
    rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
370

    
371
    if (rice->ksum < lim)
372
        rice->k--;
373
    else if (rice->ksum >= (1 << (rice->k + 5)))
374
        rice->k++;
375
}
376

    
377
static inline int ape_decode_value(APEContext * ctx, APERice *rice)
378
{
379
    int x, overflow;
380

    
381
    if (ctx->fileversion < 3990) {
382
        int tmpk;
383

    
384
        overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
385

    
386
        if (overflow == (MODEL_ELEMENTS - 1)) {
387
            tmpk = range_decode_bits(ctx, 5);
388
            overflow = 0;
389
        } else
390
            tmpk = (rice->k < 1) ? 0 : rice->k - 1;
391

    
392
        if (tmpk <= 16)
393
            x = range_decode_bits(ctx, tmpk);
394
        else {
395
            x = range_decode_bits(ctx, 16);
396
            x |= (range_decode_bits(ctx, tmpk - 16) << 16);
397
        }
398
        x += overflow << tmpk;
399
    } else {
400
        int base, pivot;
401

    
402
        pivot = rice->ksum >> 5;
403
        if (pivot == 0)
404
            pivot = 1;
405

    
406
        overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
407

    
408
        if (overflow == (MODEL_ELEMENTS - 1)) {
409
            overflow  = range_decode_bits(ctx, 16) << 16;
410
            overflow |= range_decode_bits(ctx, 16);
411
        }
412

    
413
        if (pivot < 0x10000) {
414
            base = range_decode_culfreq(ctx, pivot);
415
            range_decode_update(ctx, 1, base);
416
        } else {
417
            int base_hi = pivot, base_lo;
418
            int bbits = 0;
419

    
420
            while (base_hi & ~0xFFFF) {
421
                base_hi >>= 1;
422
                bbits++;
423
            }
424
            base_hi = range_decode_culfreq(ctx, base_hi + 1);
425
            range_decode_update(ctx, 1, base_hi);
426
            base_lo = range_decode_culfreq(ctx, 1 << bbits);
427
            range_decode_update(ctx, 1, base_lo);
428

    
429
            base = (base_hi << bbits) + base_lo;
430
        }
431

    
432
        x = base + overflow * pivot;
433
    }
434

    
435
    update_rice(rice, x);
436

    
437
    /* Convert to signed */
438
    if (x & 1)
439
        return (x >> 1) + 1;
440
    else
441
        return -(x >> 1);
442
}
443

    
444
static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo)
445
{
446
    int32_t *decoded0 = ctx->decoded0;
447
    int32_t *decoded1 = ctx->decoded1;
448

    
449
    ctx->blocksdecoded = blockstodecode;
450

    
451
    if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
452
        /* We are pure silence, just memset the output buffer. */
453
        memset(decoded0, 0, blockstodecode * sizeof(int32_t));
454
        memset(decoded1, 0, blockstodecode * sizeof(int32_t));
455
    } else {
456
        while (blockstodecode--) {
457
            *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
458
            if (stereo)
459
                *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
460
        }
461
    }
462

    
463
    if (ctx->blocksdecoded == ctx->currentframeblocks)
464
        range_dec_normalize(ctx);   /* normalize to use up all bytes */
465
}
466

    
467
static void init_entropy_decoder(APEContext * ctx)
468
{
469
    /* Read the CRC */
470
    ctx->CRC = bytestream_get_be32(&ctx->ptr);
471

    
472
    /* Read the frame flags if they exist */
473
    ctx->frameflags = 0;
474
    if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
475
        ctx->CRC &= ~0x80000000;
476

    
477
        ctx->frameflags = bytestream_get_be32(&ctx->ptr);
478
    }
479

    
480
    /* Keep a count of the blocks decoded in this frame */
481
    ctx->blocksdecoded = 0;
482

    
483
    /* Initialize the rice structs */
484
    ctx->riceX.k = 10;
485
    ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
486
    ctx->riceY.k = 10;
487
    ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
488

    
489
    /* The first 8 bits of input are ignored. */
490
    ctx->ptr++;
491

    
492
    range_start_decoding(ctx);
493
}
494

    
495
static const int32_t initial_coeffs[4] = {
496
    360, 317, -109, 98
497
};
498

    
499
static void init_predictor_decoder(APEContext * ctx)
500
{
501
    APEPredictor *p = &ctx->predictor;
502

    
503
    /* Zero the history buffers */
504
    memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
505
    p->buf = p->historybuffer;
506

    
507
    /* Initialize and zero the coefficients */
508
    memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
509
    memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
510
    memset(p->coeffsB, 0, sizeof(p->coeffsB));
511

    
512
    p->filterA[0] = p->filterA[1] = 0;
513
    p->filterB[0] = p->filterB[1] = 0;
514
    p->lastA[0]   = p->lastA[1]   = 0;
515
}
516

    
517
/** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
518
static inline int APESIGN(int32_t x) {
519
    return (x < 0) - (x > 0);
520
}
521

    
522
static av_always_inline int predictor_update_filter(APEPredictor *p, const int decoded, const int filter, const int delayA, const int delayB, const int adaptA, const int adaptB)
523
{
524
    int32_t predictionA, predictionB, sign;
525

    
526
    p->buf[delayA]     = p->lastA[filter];
527
    p->buf[adaptA]     = APESIGN(p->buf[delayA]);
528
    p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
529
    p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
530

    
531
    predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
532
                  p->buf[delayA - 1] * p->coeffsA[filter][1] +
533
                  p->buf[delayA - 2] * p->coeffsA[filter][2] +
534
                  p->buf[delayA - 3] * p->coeffsA[filter][3];
535

    
536
    /*  Apply a scaled first-order filter compression */
537
    p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
538
    p->buf[adaptB]     = APESIGN(p->buf[delayB]);
539
    p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
540
    p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
541
    p->filterB[filter] = p->filterA[filter ^ 1];
542

    
543
    predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
544
                  p->buf[delayB - 1] * p->coeffsB[filter][1] +
545
                  p->buf[delayB - 2] * p->coeffsB[filter][2] +
546
                  p->buf[delayB - 3] * p->coeffsB[filter][3] +
547
                  p->buf[delayB - 4] * p->coeffsB[filter][4];
548

    
549
    p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
550
    p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
551

    
552
    sign = APESIGN(decoded);
553
    p->coeffsA[filter][0] += p->buf[adaptA    ] * sign;
554
    p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
555
    p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
556
    p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
557
    p->coeffsB[filter][0] += p->buf[adaptB    ] * sign;
558
    p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
559
    p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
560
    p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
561
    p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
562

    
563
    return p->filterA[filter];
564
}
565

    
566
static void predictor_decode_stereo(APEContext * ctx, int count)
567
{
568
    APEPredictor *p = &ctx->predictor;
569
    int32_t *decoded0 = ctx->decoded0;
570
    int32_t *decoded1 = ctx->decoded1;
571

    
572
    while (count--) {
573
        /* Predictor Y */
574
        *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB);
575
        decoded0++;
576
        *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB);
577
        decoded1++;
578

    
579
        /* Combined */
580
        p->buf++;
581

    
582
        /* Have we filled the history buffer? */
583
        if (p->buf == p->historybuffer + HISTORY_SIZE) {
584
            memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
585
            p->buf = p->historybuffer;
586
        }
587
    }
588
}
589

    
590
static void predictor_decode_mono(APEContext * ctx, int count)
591
{
592
    APEPredictor *p = &ctx->predictor;
593
    int32_t *decoded0 = ctx->decoded0;
594
    int32_t predictionA, currentA, A, sign;
595

    
596
    currentA = p->lastA[0];
597

    
598
    while (count--) {
599
        A = *decoded0;
600

    
601
        p->buf[YDELAYA] = currentA;
602
        p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
603

    
604
        predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
605
                      p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
606
                      p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
607
                      p->buf[YDELAYA - 3] * p->coeffsA[0][3];
608

    
609
        currentA = A + (predictionA >> 10);
610

    
611
        p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
612
        p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
613

    
614
        sign = APESIGN(A);
615
        p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ] * sign;
616
        p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
617
        p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
618
        p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
619

    
620
        p->buf++;
621

    
622
        /* Have we filled the history buffer? */
623
        if (p->buf == p->historybuffer + HISTORY_SIZE) {
624
            memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
625
            p->buf = p->historybuffer;
626
        }
627

    
628
        p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
629
        *(decoded0++) = p->filterA[0];
630
    }
631

    
632
    p->lastA[0] = currentA;
633
}
634

    
635
static void do_init_filter(APEFilter *f, int16_t * buf, int order)
636
{
637
    f->coeffs = buf;
638
    f->historybuffer = buf + order;
639
    f->delay       = f->historybuffer + order * 2;
640
    f->adaptcoeffs = f->historybuffer + order;
641

    
642
    memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
643
    memset(f->coeffs, 0, order * sizeof(int16_t));
644
    f->avg = 0;
645
}
646

    
647
static void init_filter(APEContext * ctx, APEFilter *f, int16_t * buf, int order)
648
{
649
    do_init_filter(&f[0], buf, order);
650
    do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
651
}
652

    
653
static void do_apply_filter(APEContext * ctx, int version, APEFilter *f, int32_t *data, int count, int order, int fracbits)
654
{
655
    int res;
656
    int absres;
657

    
658
    while (count--) {
659
        /* round fixedpoint scalar product */
660
        res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order, f->adaptcoeffs - order, order, APESIGN(*data));
661
        res = (res + (1 << (fracbits - 1))) >> fracbits;
662
        res += *data;
663
        *data++ = res;
664

    
665
        /* Update the output history */
666
        *f->delay++ = av_clip_int16(res);
667

    
668
        if (version < 3980) {
669
            /* Version ??? to < 3.98 files (untested) */
670
            f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
671
            f->adaptcoeffs[-4] >>= 1;
672
            f->adaptcoeffs[-8] >>= 1;
673
        } else {
674
            /* Version 3.98 and later files */
675

    
676
            /* Update the adaption coefficients */
677
            absres = FFABS(res);
678
            if (absres)
679
                *f->adaptcoeffs = ((res & (1<<31)) - (1<<30)) >> (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
680
            else
681
                *f->adaptcoeffs = 0;
682

    
683
            f->avg += (absres - f->avg) / 16;
684

    
685
            f->adaptcoeffs[-1] >>= 1;
686
            f->adaptcoeffs[-2] >>= 1;
687
            f->adaptcoeffs[-8] >>= 1;
688
        }
689

    
690
        f->adaptcoeffs++;
691

    
692
        /* Have we filled the history buffer? */
693
        if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
694
            memmove(f->historybuffer, f->delay - (order * 2),
695
                    (order * 2) * sizeof(int16_t));
696
            f->delay = f->historybuffer + order * 2;
697
            f->adaptcoeffs = f->historybuffer + order;
698
        }
699
    }
700
}
701

    
702
static void apply_filter(APEContext * ctx, APEFilter *f,
703
                         int32_t * data0, int32_t * data1,
704
                         int count, int order, int fracbits)
705
{
706
    do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
707
    if (data1)
708
        do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
709
}
710

    
711
static void ape_apply_filters(APEContext * ctx, int32_t * decoded0,
712
                              int32_t * decoded1, int count)
713
{
714
    int i;
715

    
716
    for (i = 0; i < APE_FILTER_LEVELS; i++) {
717
        if (!ape_filter_orders[ctx->fset][i])
718
            break;
719
        apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]);
720
    }
721
}
722

    
723
static void init_frame_decoder(APEContext * ctx)
724
{
725
    int i;
726
    init_entropy_decoder(ctx);
727
    init_predictor_decoder(ctx);
728

    
729
    for (i = 0; i < APE_FILTER_LEVELS; i++) {
730
        if (!ape_filter_orders[ctx->fset][i])
731
            break;
732
        init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]);
733
    }
734
}
735

    
736
static void ape_unpack_mono(APEContext * ctx, int count)
737
{
738
    int32_t left;
739
    int32_t *decoded0 = ctx->decoded0;
740
    int32_t *decoded1 = ctx->decoded1;
741

    
742
    if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
743
        entropy_decode(ctx, count, 0);
744
        /* We are pure silence, so we're done. */
745
        av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
746
        return;
747
    }
748

    
749
    entropy_decode(ctx, count, 0);
750
    ape_apply_filters(ctx, decoded0, NULL, count);
751

    
752
    /* Now apply the predictor decoding */
753
    predictor_decode_mono(ctx, count);
754

    
755
    /* Pseudo-stereo - just copy left channel to right channel */
756
    if (ctx->channels == 2) {
757
        while (count--) {
758
            left = *decoded0;
759
            *(decoded1++) = *(decoded0++) = left;
760
        }
761
    }
762
}
763

    
764
static void ape_unpack_stereo(APEContext * ctx, int count)
765
{
766
    int32_t left, right;
767
    int32_t *decoded0 = ctx->decoded0;
768
    int32_t *decoded1 = ctx->decoded1;
769

    
770
    if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
771
        /* We are pure silence, so we're done. */
772
        av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
773
        return;
774
    }
775

    
776
    entropy_decode(ctx, count, 1);
777
    ape_apply_filters(ctx, decoded0, decoded1, count);
778

    
779
    /* Now apply the predictor decoding */
780
    predictor_decode_stereo(ctx, count);
781

    
782
    /* Decorrelate and scale to output depth */
783
    while (count--) {
784
        left = *decoded1 - (*decoded0 / 2);
785
        right = left + *decoded0;
786

    
787
        *(decoded0++) = left;
788
        *(decoded1++) = right;
789
    }
790
}
791

    
792
static int ape_decode_frame(AVCodecContext * avctx,
793
                            void *data, int *data_size,
794
                            AVPacket *avpkt)
795
{
796
    const uint8_t *buf = avpkt->data;
797
    int buf_size = avpkt->size;
798
    APEContext *s = avctx->priv_data;
799
    int16_t *samples = data;
800
    int nblocks;
801
    int i, n;
802
    int blockstodecode;
803
    int bytes_used;
804

    
805
    if (buf_size == 0 && !s->samples) {
806
        *data_size = 0;
807
        return 0;
808
    }
809

    
810
    /* should not happen but who knows */
811
    if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) {
812
        av_log (avctx, AV_LOG_ERROR, "Packet size is too big to be handled in lavc! (max is %d where you have %d)\n", *data_size, s->samples * 2 * avctx->channels);
813
        return -1;
814
    }
815

    
816
    if(!s->samples){
817
        s->data = av_realloc(s->data, (buf_size + 3) & ~3);
818
        s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
819
        s->ptr = s->last_ptr = s->data;
820
        s->data_end = s->data + buf_size;
821

    
822
        nblocks = s->samples = bytestream_get_be32(&s->ptr);
823
        n =  bytestream_get_be32(&s->ptr);
824
        if(n < 0 || n > 3){
825
            av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
826
            s->data = NULL;
827
            return -1;
828
        }
829
        s->ptr += n;
830

    
831
        s->currentframeblocks = nblocks;
832
        buf += 4;
833
        if (s->samples <= 0) {
834
            *data_size = 0;
835
            return buf_size;
836
        }
837

    
838
        memset(s->decoded0,  0, sizeof(s->decoded0));
839
        memset(s->decoded1,  0, sizeof(s->decoded1));
840

    
841
        /* Initialize the frame decoder */
842
        init_frame_decoder(s);
843
    }
844

    
845
    if (!s->data) {
846
        *data_size = 0;
847
        return buf_size;
848
    }
849

    
850
    nblocks = s->samples;
851
    blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks);
852

    
853
    s->error=0;
854

    
855
    if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
856
        ape_unpack_mono(s, blockstodecode);
857
    else
858
        ape_unpack_stereo(s, blockstodecode);
859
    emms_c();
860

    
861
    if(s->error || s->ptr > s->data_end){
862
        s->samples=0;
863
        av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
864
        return -1;
865
    }
866

    
867
    for (i = 0; i < blockstodecode; i++) {
868
        *samples++ = s->decoded0[i];
869
        if(s->channels == 2)
870
            *samples++ = s->decoded1[i];
871
    }
872

    
873
    s->samples -= blockstodecode;
874

    
875
    *data_size = blockstodecode * 2 * s->channels;
876
    bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size;
877
    s->last_ptr = s->ptr;
878
    return bytes_used;
879
}
880

    
881
static void ape_flush(AVCodecContext *avctx)
882
{
883
    APEContext *s = avctx->priv_data;
884
    s->samples= 0;
885
}
886

    
887
AVCodec ff_ape_decoder = {
888
    "ape",
889
    AVMEDIA_TYPE_AUDIO,
890
    CODEC_ID_APE,
891
    sizeof(APEContext),
892
    ape_decode_init,
893
    NULL,
894
    ape_decode_close,
895
    ape_decode_frame,
896
    .capabilities = CODEC_CAP_SUBFRAMES,
897
    .flush = ape_flush,
898
    .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
899
};