libavcodec/apedec.c

   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 FFmpeg.
   7  *
   8  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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 "bitstream.h"
  27 #include "bytestream.h"
  28
  29 /**
  30  * @file apedec.c
  31  * Monkey's Audio lossless audio decoder
  32  */
  33
  34 #define BLOCKS_PER_LOOP     4608
  35 #define MAX_CHANNELS        2
  36 #define MAX_BYTESPERSAMPLE  3
  37
  38 #define APE_FRAMECODE_MONO_SILENCE    1
  39 #define APE_FRAMECODE_STEREO_SILENCE  3
  40 #define APE_FRAMECODE_PSEUDO_STEREO   4
  41
  42 #define HISTORY_SIZE 512
  43 #define PREDICTOR_ORDER 8
  44 /** Total size of all predictor histories */
  45 #define PREDICTOR_SIZE 50
  46
  47 #define YDELAYA (18 + PREDICTOR_ORDER*4)
  48 #define YDELAYB (18 + PREDICTOR_ORDER*3)
  49 #define XDELAYA (18 + PREDICTOR_ORDER*2)
  50 #define XDELAYB (18 + PREDICTOR_ORDER)
  51
  52 #define YADAPTCOEFFSA 18
  53 #define XADAPTCOEFFSA 14
  54 #define YADAPTCOEFFSB 10
  55 #define XADAPTCOEFFSB 5
  56
  57 /**
  58  * Possible compression levels
  59  * @{
  60  */
  61 enum APECompressionLevel {
  62     COMPRESSION_LEVEL_FAST       = 1000,
  63     COMPRESSION_LEVEL_NORMAL     = 2000,
  64     COMPRESSION_LEVEL_HIGH       = 3000,
  65     COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
  66     COMPRESSION_LEVEL_INSANE     = 5000
  67 };
  68 /** @} */
  69
  70 #define APE_FILTER_LEVELS 3
  71
  72 /** Filter orders depending on compression level */
  73 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
  74     {  0,   0,    0 },
  75     { 16,   0,    0 },
  76     { 64,   0,    0 },
  77     { 32, 256,    0 },
  78     { 16, 256, 1280 }
  79 };
  80
  81 /** Filter fraction bits depending on compression level */
  82 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
  83     {  0,  0,  0 },
  84     { 11,  0,  0 },
  85     { 11,  0,  0 },
  86     { 10, 13,  0 },
  87     { 11, 13, 15 }
  88 };
  89
  90
  91 /** Filters applied to the decoded data */
  92 typedef struct APEFilter {
  93     int16_t *coeffs;        ///< actual coefficients used in filtering
  94     int16_t *adaptcoeffs;   ///< adaptive filter coefficients used for correcting of actual filter coefficients
  95     int16_t *historybuffer; ///< filter memory
  96     int16_t *delay;         ///< filtered values
  97
  98     int avg;
  99 } APEFilter;
 100
 101 typedef struct APERice {
 102     uint32_t k;
 103     uint32_t ksum;
 104 } APERice;
 105
 106 typedef struct APERangecoder {
 107     uint32_t low;           ///< low end of interval
 108     uint32_t range;         ///< length of interval
 109     uint32_t help;          ///< bytes_to_follow resp. intermediate value
 110     unsigned int buffer;    ///< buffer for input/output
 111 } APERangecoder;
 112
 113 /** Filter histories */
 114 typedef struct APEPredictor {
 115     int32_t *buf;
 116
 117     int32_t lastA[2];
 118
 119     int32_t filterA[2];
 120     int32_t filterB[2];
 121
 122     int32_t coeffsA[2][4];  ///< adaption coefficients
 123     int32_t coeffsB[2][5];  ///< adaption coefficients
 124     int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
 125 } APEPredictor;
 126
 127 /** Decoder context */
 128 typedef struct APEContext {
 129     AVCodecContext *avctx;
 130     DSPContext dsp;
 131     int channels;
 132     int samples;                             ///< samples left to decode in current frame
 133
 134     int fileversion;                         ///< codec version, very important in decoding process
 135     int compression_level;                   ///< compression levels
 136     int fset;                                ///< which filter set to use (calculated from compression level)
 137     int flags;                               ///< global decoder flags
 138
 139     uint32_t CRC;                            ///< frame CRC
 140     int frameflags;                          ///< frame flags
 141     int currentframeblocks;                  ///< samples (per channel) in current frame
 142     int blocksdecoded;                       ///< count of decoded samples in current frame
 143     APEPredictor predictor;                  ///< predictor used for final reconstruction
 144
 145     int32_t decoded0[BLOCKS_PER_LOOP];       ///< decoded data for the first channel
 146     int32_t decoded1[BLOCKS_PER_LOOP];       ///< decoded data for the second channel
 147
 148     int16_t* filterbuf[APE_FILTER_LEVELS];   ///< filter memory
 149
 150     APERangecoder rc;                        ///< rangecoder used to decode actual values
 151     APERice riceX;                           ///< rice code parameters for the second channel
 152     APERice riceY;                           ///< rice code parameters for the first channel
 153     APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
 154
 155     uint8_t *data;                           ///< current frame data
 156     uint8_t *data_end;                       ///< frame data end
 157     const uint8_t *ptr;                      ///< current position in frame data
 158     const uint8_t *last_ptr;                 ///< position where last 4608-sample block ended
 159
 160     int error;
 161 } APEContext;
 162
 163 // TODO: dsputilize
 164
 165 static av_cold int ape_decode_init(AVCodecContext * avctx)
 166 {
 167     APEContext *s = avctx->priv_data;
 168     int i;
 169
 170     if (avctx->extradata_size != 6) {
 171         av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
 172         return -1;
 173     }
 174     if (avctx->bits_per_sample != 16) {
 175         av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n");
 176         return -1;
 177     }
 178     if (avctx->channels > 2) {
 179         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
 180         return -1;
 181     }
 182     s->avctx             = avctx;
 183     s->channels          = avctx->channels;
 184     s->fileversion       = AV_RL16(avctx->extradata);
 185     s->compression_level = AV_RL16(avctx->extradata + 2);
 186     s->flags             = AV_RL16(avctx->extradata + 4);
 187
 188     av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n", s->compression_level, s->flags);
 189     if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
 190         av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n", s->compression_level);
 191         return -1;
 192     }
 193     s->fset = s->compression_level / 1000 - 1;
 194     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 195         if (!ape_filter_orders[s->fset][i])
 196             break;
 197         s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4);
 198     }
 199
 200     dsputil_init(&s->dsp, avctx);
 201     avctx->sample_fmt = SAMPLE_FMT_S16;
 202     return 0;
 203 }
 204
 205 static av_cold int ape_decode_close(AVCodecContext * avctx)
 206 {
 207     APEContext *s = avctx->priv_data;
 208     int i;
 209
 210     for (i = 0; i < APE_FILTER_LEVELS; i++)
 211         av_freep(&s->filterbuf[i]);
 212
 213     return 0;
 214 }
 215
 216 /**
 217  * @defgroup rangecoder APE range decoder
 218  * @{
 219  */
 220
 221 #define CODE_BITS    32
 222 #define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
 223 #define SHIFT_BITS   (CODE_BITS - 9)
 224 #define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
 225 #define BOTTOM_VALUE (TOP_VALUE >> 8)
 226
 227 /** Start the decoder */
 228 static inline void range_start_decoding(APEContext * ctx)
 229 {
 230     ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
 231     ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
 232     ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
 233 }
 234
 235 /** Perform normalization */
 236 static inline void range_dec_normalize(APEContext * ctx)
 237 {
 238     while (ctx->rc.range <= BOTTOM_VALUE) {
 239         ctx->rc.buffer <<= 8;
 240         if(ctx->ptr < ctx->data_end)
 241             ctx->rc.buffer += *ctx->ptr;
 242         ctx->ptr++;
 243         ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
 244         ctx->rc.range  <<= 8;
 245     }
 246 }
 247
 248 /**
 249  * Calculate culmulative frequency for next symbol. Does NO update!
 250  * @param tot_f is the total frequency or (code_value)1<<shift
 251  * @return the culmulative frequency
 252  */
 253 static inline int range_decode_culfreq(APEContext * ctx, int tot_f)
 254 {
 255     range_dec_normalize(ctx);
 256     ctx->rc.help = ctx->rc.range / tot_f;
 257     return ctx->rc.low / ctx->rc.help;
 258 }
 259
 260 /**
 261  * Decode value with given size in bits
 262  * @param shift number of bits to decode
 263  */
 264 static inline int range_decode_culshift(APEContext * ctx, int shift)
 265 {
 266     range_dec_normalize(ctx);
 267     ctx->rc.help = ctx->rc.range >> shift;
 268     return ctx->rc.low / ctx->rc.help;
 269 }
 270
 271
 272 /**
 273  * Update decoding state
 274  * @param sy_f the interval length (frequency of the symbol)
 275  * @param lt_f the lower end (frequency sum of < symbols)
 276  */
 277 static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f)
 278 {
 279     ctx->rc.low  -= ctx->rc.help * lt_f;
 280     ctx->rc.range = ctx->rc.help * sy_f;
 281 }
 282
 283 /** Decode n bits (n <= 16) without modelling */
 284 static inline int range_decode_bits(APEContext * ctx, int n)
 285 {
 286     int sym = range_decode_culshift(ctx, n);
 287     range_decode_update(ctx, 1, sym);
 288     return sym;
 289 }
 290
 291
 292 #define MODEL_ELEMENTS 64
 293
 294 /**
 295  * Fixed probabilities for symbols in Monkey Audio version 3.97
 296  */
 297 static const uint16_t counts_3970[22] = {
 298         0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
 299     62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
 300     65450, 65469, 65480, 65487, 65491, 65493,
 301 };
 302
 303 /**
 304  * Probability ranges for symbols in Monkey Audio version 3.97
 305  */
 306 static const uint16_t counts_diff_3970[21] = {
 307     14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
 308     1104, 677, 415, 248, 150, 89, 54, 31,
 309     19, 11, 7, 4, 2,
 310 };
 311
 312 /**
 313  * Fixed probabilities for symbols in Monkey Audio version 3.98
 314  */
 315 static const uint16_t counts_3980[22] = {
 316         0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
 317     64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
 318     65485, 65488, 65490, 65491, 65492, 65493,
 319 };
 320
 321 /**
 322  * Probability ranges for symbols in Monkey Audio version 3.98
 323  */
 324 static const uint16_t counts_diff_3980[21] = {
 325     19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
 326     261, 119, 65, 31, 19, 10, 6, 3,
 327     3, 2, 1, 1, 1,
 328 };
 329
 330 /**
 331  * Decode symbol
 332  * @param counts probability range start position
 333  * @param count_diffs probability range widths
 334  */
 335 static inline int range_get_symbol(APEContext * ctx,
 336                                    const uint16_t counts[],
 337                                    const uint16_t counts_diff[])
 338 {
 339     int symbol, cf;
 340
 341     cf = range_decode_culshift(ctx, 16);
 342
 343     if(cf > 65492){
 344         symbol= cf - 65535 + 63;
 345         range_decode_update(ctx, 1, cf);
 346         if(cf > 65535)
 347             ctx->error=1;
 348         return symbol;
 349     }
 350     /* figure out the symbol inefficiently; a binary search would be much better */
 351     for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
 352
 353     range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
 354
 355     return symbol;
 356 }
 357 /** @} */ // group rangecoder
 358
 359 static inline void update_rice(APERice *rice, int x)
 360 {
 361     rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
 362
 363     if (rice->k == 0)
 364         rice->k = 1;
 365     else if (rice->ksum < (1 << (rice->k + 4)))
 366         rice->k--;
 367     else if (rice->ksum >= (1 << (rice->k + 5)))
 368         rice->k++;
 369 }
 370
 371 static inline int ape_decode_value(APEContext * ctx, APERice *rice)
 372 {
 373     int x, overflow;
 374
 375     if (ctx->fileversion < 3990) {
 376         int tmpk;
 377
 378         overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
 379
 380         if (overflow == (MODEL_ELEMENTS - 1)) {
 381             tmpk = range_decode_bits(ctx, 5);
 382             overflow = 0;
 383         } else
 384             tmpk = (rice->k < 1) ? 0 : rice->k - 1;
 385
 386         if (tmpk <= 16)
 387             x = range_decode_bits(ctx, tmpk);
 388         else {
 389             x = range_decode_bits(ctx, 16);
 390             x |= (range_decode_bits(ctx, tmpk - 16) << 16);
 391         }
 392         x += overflow << tmpk;
 393     } else {
 394         int base, pivot;
 395
 396         pivot = rice->ksum >> 5;
 397         if (pivot == 0)
 398             pivot = 1;
 399
 400         overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
 401
 402         if (overflow == (MODEL_ELEMENTS - 1)) {
 403             overflow  = range_decode_bits(ctx, 16) << 16;
 404             overflow |= range_decode_bits(ctx, 16);
 405         }
 406
 407         base = range_decode_culfreq(ctx, pivot);
 408         range_decode_update(ctx, 1, base);
 409
 410         x = base + overflow * pivot;
 411     }
 412
 413     update_rice(rice, x);
 414
 415     /* Convert to signed */
 416     if (x & 1)
 417         return (x >> 1) + 1;
 418     else
 419         return -(x >> 1);
 420 }
 421
 422 static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo)
 423 {
 424     int32_t *decoded0 = ctx->decoded0;
 425     int32_t *decoded1 = ctx->decoded1;
 426
 427     ctx->blocksdecoded = blockstodecode;
 428
 429     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 430         /* We are pure silence, just memset the output buffer. */
 431         memset(decoded0, 0, blockstodecode * sizeof(int32_t));
 432         memset(decoded1, 0, blockstodecode * sizeof(int32_t));
 433     } else {
 434         while (blockstodecode--) {
 435             *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
 436             if (stereo)
 437                 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
 438         }
 439     }
 440
 441     if (ctx->blocksdecoded == ctx->currentframeblocks)
 442         range_dec_normalize(ctx);   /* normalize to use up all bytes */
 443 }
 444
 445 static void init_entropy_decoder(APEContext * ctx)
 446 {
 447     /* Read the CRC */
 448     ctx->CRC = bytestream_get_be32(&ctx->ptr);
 449
 450     /* Read the frame flags if they exist */
 451     ctx->frameflags = 0;
 452     if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
 453         ctx->CRC &= ~0x80000000;
 454
 455         ctx->frameflags = bytestream_get_be32(&ctx->ptr);
 456     }
 457
 458     /* Keep a count of the blocks decoded in this frame */
 459     ctx->blocksdecoded = 0;
 460
 461     /* Initialize the rice structs */
 462     ctx->riceX.k = 10;
 463     ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
 464     ctx->riceY.k = 10;
 465     ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
 466
 467     /* The first 8 bits of input are ignored. */
 468     ctx->ptr++;
 469
 470     range_start_decoding(ctx);
 471 }
 472
 473 static const int32_t initial_coeffs[4] = {
 474     360, 317, -109, 98
 475 };
 476
 477 static void init_predictor_decoder(APEContext * ctx)
 478 {
 479     APEPredictor *p = &ctx->predictor;
 480
 481     /* Zero the history buffers */
 482     memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
 483     p->buf = p->historybuffer;
 484
 485     /* Initialize and zero the coefficients */
 486     memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
 487     memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
 488     memset(p->coeffsB, 0, sizeof(p->coeffsB));
 489
 490     p->filterA[0] = p->filterA[1] = 0;
 491     p->filterB[0] = p->filterB[1] = 0;
 492     p->lastA[0]   = p->lastA[1]   = 0;
 493 }
 494
 495 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
 496 static inline int APESIGN(int32_t x) {
 497     return (x < 0) - (x > 0);
 498 }
 499
 500 static int predictor_update_filter(APEPredictor *p, const int decoded, const int filter, const int delayA, const int delayB, const int adaptA, const int adaptB)
 501 {
 502     int32_t predictionA, predictionB;
 503
 504     p->buf[delayA]     = p->lastA[filter];
 505     p->buf[adaptA]     = APESIGN(p->buf[delayA]);
 506     p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
 507     p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
 508
 509     predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
 510                   p->buf[delayA - 1] * p->coeffsA[filter][1] +
 511                   p->buf[delayA - 2] * p->coeffsA[filter][2] +
 512                   p->buf[delayA - 3] * p->coeffsA[filter][3];
 513
 514     /*  Apply a scaled first-order filter compression */
 515     p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
 516     p->buf[adaptB]     = APESIGN(p->buf[delayB]);
 517     p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
 518     p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
 519     p->filterB[filter] = p->filterA[filter ^ 1];
 520
 521     predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
 522                   p->buf[delayB - 1] * p->coeffsB[filter][1] +
 523                   p->buf[delayB - 2] * p->coeffsB[filter][2] +
 524                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
 525                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 526
 527     p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
 528     p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
 529
 530     if (!decoded) // no need updating filter coefficients
 531         return p->filterA[filter];
 532
 533     if (decoded > 0) {
 534         p->coeffsA[filter][0] -= p->buf[adaptA    ];
 535         p->coeffsA[filter][1] -= p->buf[adaptA - 1];
 536         p->coeffsA[filter][2] -= p->buf[adaptA - 2];
 537         p->coeffsA[filter][3] -= p->buf[adaptA - 3];
 538
 539         p->coeffsB[filter][0] -= p->buf[adaptB    ];
 540         p->coeffsB[filter][1] -= p->buf[adaptB - 1];
 541         p->coeffsB[filter][2] -= p->buf[adaptB - 2];
 542         p->coeffsB[filter][3] -= p->buf[adaptB - 3];
 543         p->coeffsB[filter][4] -= p->buf[adaptB - 4];
 544     } else {
 545         p->coeffsA[filter][0] += p->buf[adaptA    ];
 546         p->coeffsA[filter][1] += p->buf[adaptA - 1];
 547         p->coeffsA[filter][2] += p->buf[adaptA - 2];
 548         p->coeffsA[filter][3] += p->buf[adaptA - 3];
 549
 550         p->coeffsB[filter][0] += p->buf[adaptB    ];
 551         p->coeffsB[filter][1] += p->buf[adaptB - 1];
 552         p->coeffsB[filter][2] += p->buf[adaptB - 2];
 553         p->coeffsB[filter][3] += p->buf[adaptB - 3];
 554         p->coeffsB[filter][4] += p->buf[adaptB - 4];
 555     }
 556     return p->filterA[filter];
 557 }
 558
 559 static void predictor_decode_stereo(APEContext * ctx, int count)
 560 {
 561     int32_t predictionA, predictionB;
 562     APEPredictor *p = &ctx->predictor;
 563     int32_t *decoded0 = ctx->decoded0;
 564     int32_t *decoded1 = ctx->decoded1;
 565
 566     while (count--) {
 567         /* Predictor Y */
 568         predictionA = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB);
 569         predictionB = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB);
 570         *(decoded0++) = predictionA;
 571         *(decoded1++) = predictionB;
 572
 573         /* Combined */
 574         p->buf++;
 575
 576         /* Have we filled the history buffer? */
 577         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 578             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 579             p->buf = p->historybuffer;
 580         }
 581     }
 582 }
 583
 584 static void predictor_decode_mono(APEContext * ctx, int count)
 585 {
 586     APEPredictor *p = &ctx->predictor;
 587     int32_t *decoded0 = ctx->decoded0;
 588     int32_t predictionA, currentA, A;
 589
 590     currentA = p->lastA[0];
 591
 592     while (count--) {
 593         A = *decoded0;
 594
 595         p->buf[YDELAYA] = currentA;
 596         p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
 597
 598         predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
 599                       p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
 600                       p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
 601                       p->buf[YDELAYA - 3] * p->coeffsA[0][3];
 602
 603         currentA = A + (predictionA >> 10);
 604
 605         p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
 606         p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
 607
 608         if (A > 0) {
 609             p->coeffsA[0][0] -= p->buf[YADAPTCOEFFSA    ];
 610             p->coeffsA[0][1] -= p->buf[YADAPTCOEFFSA - 1];
 611             p->coeffsA[0][2] -= p->buf[YADAPTCOEFFSA - 2];
 612             p->coeffsA[0][3] -= p->buf[YADAPTCOEFFSA - 3];
 613         } else if (A < 0) {
 614             p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ];
 615             p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1];
 616             p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2];
 617             p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3];
 618         }
 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 inline 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_int16(f->delay - order, f->coeffs, order, 0) + (1 << (fracbits - 1))) >> fracbits;
 661
 662         if (*data < 0)
 663             ctx->dsp.add_int16(f->coeffs, f->adaptcoeffs - order, order);
 664         else if (*data > 0)
 665             ctx->dsp.sub_int16(f->coeffs, f->adaptcoeffs - order, order);
 666
 667         res += *data;
 668
 669         *data++ = res;
 670
 671         /* Update the output history */
 672         *f->delay++ = av_clip_int16(res);
 673
 674         if (version < 3980) {
 675             /* Version ??? to < 3.98 files (untested) */
 676             f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
 677             f->adaptcoeffs[-4] >>= 1;
 678             f->adaptcoeffs[-8] >>= 1;
 679         } else {
 680             /* Version 3.98 and later files */
 681
 682             /* Update the adaption coefficients */
 683             absres = (res < 0 ? -res : res);
 684
 685             if (absres > (f->avg * 3))
 686                 *f->adaptcoeffs = ((res >> 25) & 64) - 32;
 687             else if (absres > (f->avg * 4) / 3)
 688                 *f->adaptcoeffs = ((res >> 26) & 32) - 16;
 689             else if (absres > 0)
 690                 *f->adaptcoeffs = ((res >> 27) & 16) - 8;
 691             else
 692                 *f->adaptcoeffs = 0;
 693
 694             f->avg += (absres - f->avg) / 16;
 695
 696             f->adaptcoeffs[-1] >>= 1;
 697             f->adaptcoeffs[-2] >>= 1;
 698             f->adaptcoeffs[-8] >>= 1;
 699         }
 700
 701         f->adaptcoeffs++;
 702
 703         /* Have we filled the history buffer? */
 704         if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
 705             memmove(f->historybuffer, f->delay - (order * 2),
 706                     (order * 2) * sizeof(int16_t));
 707             f->delay = f->historybuffer + order * 2;
 708             f->adaptcoeffs = f->historybuffer + order;
 709         }
 710     }
 711 }
 712
 713 static void apply_filter(APEContext * ctx, APEFilter *f,
 714                          int32_t * data0, int32_t * data1,
 715                          int count, int order, int fracbits)
 716 {
 717     do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
 718     if (data1)
 719         do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
 720 }
 721
 722 static void ape_apply_filters(APEContext * ctx, int32_t * decoded0,
 723                               int32_t * decoded1, int count)
 724 {
 725     int i;
 726
 727     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 728         if (!ape_filter_orders[ctx->fset][i])
 729             break;
 730         apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]);
 731     }
 732 }
 733
 734 static void init_frame_decoder(APEContext * ctx)
 735 {
 736     int i;
 737     init_entropy_decoder(ctx);
 738     init_predictor_decoder(ctx);
 739
 740     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 741         if (!ape_filter_orders[ctx->fset][i])
 742             break;
 743         init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]);
 744     }
 745 }
 746
 747 static void ape_unpack_mono(APEContext * ctx, int count)
 748 {
 749     int32_t left;
 750     int32_t *decoded0 = ctx->decoded0;
 751     int32_t *decoded1 = ctx->decoded1;
 752
 753     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 754         entropy_decode(ctx, count, 0);
 755         /* We are pure silence, so we're done. */
 756         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
 757         return;
 758     }
 759
 760     entropy_decode(ctx, count, 0);
 761     ape_apply_filters(ctx, decoded0, NULL, count);
 762
 763     /* Now apply the predictor decoding */
 764     predictor_decode_mono(ctx, count);
 765
 766     /* Pseudo-stereo - just copy left channel to right channel */
 767     if (ctx->channels == 2) {
 768         while (count--) {
 769             left = *decoded0;
 770             *(decoded1++) = *(decoded0++) = left;
 771         }
 772     }
 773 }
 774
 775 static void ape_unpack_stereo(APEContext * ctx, int count)
 776 {
 777     int32_t left, right;
 778     int32_t *decoded0 = ctx->decoded0;
 779     int32_t *decoded1 = ctx->decoded1;
 780
 781     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 782         /* We are pure silence, so we're done. */
 783         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
 784         return;
 785     }
 786
 787     entropy_decode(ctx, count, 1);
 788     ape_apply_filters(ctx, decoded0, decoded1, count);
 789
 790     /* Now apply the predictor decoding */
 791     predictor_decode_stereo(ctx, count);
 792
 793     /* Decorrelate and scale to output depth */
 794     while (count--) {
 795         left = *decoded1 - (*decoded0 / 2);
 796         right = left + *decoded0;
 797
 798         *(decoded0++) = left;
 799         *(decoded1++) = right;
 800     }
 801 }
 802
 803 static int ape_decode_frame(AVCodecContext * avctx,
 804                             void *data, int *data_size,
 805                             const uint8_t * buf, int buf_size)
 806 {
 807     APEContext *s = avctx->priv_data;
 808     int16_t *samples = data;
 809     int nblocks;
 810     int i, n;
 811     int blockstodecode;
 812     int bytes_used;
 813
 814     if (buf_size == 0 && !s->samples) {
 815         *data_size = 0;
 816         return 0;
 817     }
 818
 819     /* should not happen but who knows */
 820     if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) {
 821         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);
 822         return -1;
 823     }
 824
 825     if(!s->samples){
 826         s->data = av_realloc(s->data, (buf_size + 3) & ~3);
 827         s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
 828         s->ptr = s->last_ptr = s->data;
 829         s->data_end = s->data + buf_size;
 830
 831         nblocks = s->samples = bytestream_get_be32(&s->ptr);
 832         n =  bytestream_get_be32(&s->ptr);
 833         if(n < 0 || n > 3){
 834             av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
 835             s->data = NULL;
 836             return -1;
 837         }
 838         s->ptr += n;
 839
 840         s->currentframeblocks = nblocks;
 841         buf += 4;
 842         if (s->samples <= 0) {
 843             *data_size = 0;
 844             return buf_size;
 845         }
 846
 847         memset(s->decoded0,  0, sizeof(s->decoded0));
 848         memset(s->decoded1,  0, sizeof(s->decoded1));
 849
 850         /* Initialize the frame decoder */
 851         init_frame_decoder(s);
 852     }
 853
 854     if (!s->data) {
 855         *data_size = 0;
 856         return buf_size;
 857     }
 858
 859     nblocks = s->samples;
 860     blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks);
 861
 862     s->error=0;
 863
 864     if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
 865         ape_unpack_mono(s, blockstodecode);
 866     else
 867         ape_unpack_stereo(s, blockstodecode);
 868
 869     if(s->error || s->ptr > s->data_end){
 870         s->samples=0;
 871         av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
 872         return -1;
 873     }
 874
 875     for (i = 0; i < blockstodecode; i++) {
 876         *samples++ = s->decoded0[i];
 877         if(s->channels == 2)
 878             *samples++ = s->decoded1[i];
 879     }
 880
 881     s->samples -= blockstodecode;
 882
 883     *data_size = blockstodecode * 2 * s->channels;
 884     bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size;
 885     s->last_ptr = s->ptr;
 886     return bytes_used;
 887 }
 888
 889 AVCodec ape_decoder = {
 890     "ape",
 891     CODEC_TYPE_AUDIO,
 892     CODEC_ID_APE,
 893     sizeof(APEContext),
 894     ape_decode_init,
 895     NULL,
 896     ape_decode_close,
 897     ape_decode_frame,
 898     .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
 899 };