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 "get_bits.h"
  27 #include "bytestream.h"
  28
  29 /**
  30  * @file libavcodec/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_coded_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     avctx->channel_layout = (avctx->channels==2) ? CH_LAYOUT_STEREO : CH_LAYOUT_MONO;
 203     return 0;
 204 }
 205
 206 static av_cold int ape_decode_close(AVCodecContext * avctx)
 207 {
 208     APEContext *s = avctx->priv_data;
 209     int i;
 210
 211     for (i = 0; i < APE_FILTER_LEVELS; i++)
 212         av_freep(&s->filterbuf[i]);
 213
 214     return 0;
 215 }
 216
 217 /**
 218  * @defgroup rangecoder APE range decoder
 219  * @{
 220  */
 221
 222 #define CODE_BITS    32
 223 #define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
 224 #define SHIFT_BITS   (CODE_BITS - 9)
 225 #define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
 226 #define BOTTOM_VALUE (TOP_VALUE >> 8)
 227
 228 /** Start the decoder */
 229 static inline void range_start_decoding(APEContext * ctx)
 230 {
 231     ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
 232     ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
 233     ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
 234 }
 235
 236 /** Perform normalization */
 237 static inline void range_dec_normalize(APEContext * ctx)
 238 {
 239     while (ctx->rc.range <= BOTTOM_VALUE) {
 240         ctx->rc.buffer <<= 8;
 241         if(ctx->ptr < ctx->data_end)
 242             ctx->rc.buffer += *ctx->ptr;
 243         ctx->ptr++;
 244         ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
 245         ctx->rc.range  <<= 8;
 246     }
 247 }
 248
 249 /**
 250  * Calculate culmulative frequency for next symbol. Does NO update!
 251  * @param ctx decoder context
 252  * @param tot_f is the total frequency or (code_value)1<<shift
 253  * @return the culmulative frequency
 254  */
 255 static inline int range_decode_culfreq(APEContext * ctx, int tot_f)
 256 {
 257     range_dec_normalize(ctx);
 258     ctx->rc.help = ctx->rc.range / tot_f;
 259     return ctx->rc.low / ctx->rc.help;
 260 }
 261
 262 /**
 263  * Decode value with given size in bits
 264  * @param ctx decoder context
 265  * @param shift number of bits to decode
 266  */
 267 static inline int range_decode_culshift(APEContext * ctx, int shift)
 268 {
 269     range_dec_normalize(ctx);
 270     ctx->rc.help = ctx->rc.range >> shift;
 271     return ctx->rc.low / ctx->rc.help;
 272 }
 273
 274
 275 /**
 276  * Update decoding state
 277  * @param ctx decoder context
 278  * @param sy_f the interval length (frequency of the symbol)
 279  * @param lt_f the lower end (frequency sum of < symbols)
 280  */
 281 static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f)
 282 {
 283     ctx->rc.low  -= ctx->rc.help * lt_f;
 284     ctx->rc.range = ctx->rc.help * sy_f;
 285 }
 286
 287 /** Decode n bits (n <= 16) without modelling */
 288 static inline int range_decode_bits(APEContext * ctx, int n)
 289 {
 290     int sym = range_decode_culshift(ctx, n);
 291     range_decode_update(ctx, 1, sym);
 292     return sym;
 293 }
 294
 295
 296 #define MODEL_ELEMENTS 64
 297
 298 /**
 299  * Fixed probabilities for symbols in Monkey Audio version 3.97
 300  */
 301 static const uint16_t counts_3970[22] = {
 302         0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
 303     62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
 304     65450, 65469, 65480, 65487, 65491, 65493,
 305 };
 306
 307 /**
 308  * Probability ranges for symbols in Monkey Audio version 3.97
 309  */
 310 static const uint16_t counts_diff_3970[21] = {
 311     14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
 312     1104, 677, 415, 248, 150, 89, 54, 31,
 313     19, 11, 7, 4, 2,
 314 };
 315
 316 /**
 317  * Fixed probabilities for symbols in Monkey Audio version 3.98
 318  */
 319 static const uint16_t counts_3980[22] = {
 320         0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
 321     64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
 322     65485, 65488, 65490, 65491, 65492, 65493,
 323 };
 324
 325 /**
 326  * Probability ranges for symbols in Monkey Audio version 3.98
 327  */
 328 static const uint16_t counts_diff_3980[21] = {
 329     19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
 330     261, 119, 65, 31, 19, 10, 6, 3,
 331     3, 2, 1, 1, 1,
 332 };
 333
 334 /**
 335  * Decode symbol
 336  * @param ctx decoder context
 337  * @param counts probability range start position
 338  * @param counts_diff probability range widths
 339  */
 340 static inline int range_get_symbol(APEContext * ctx,
 341                                    const uint16_t counts[],
 342                                    const uint16_t counts_diff[])
 343 {
 344     int symbol, cf;
 345
 346     cf = range_decode_culshift(ctx, 16);
 347
 348     if(cf > 65492){
 349         symbol= cf - 65535 + 63;
 350         range_decode_update(ctx, 1, cf);
 351         if(cf > 65535)
 352             ctx->error=1;
 353         return symbol;
 354     }
 355     /* figure out the symbol inefficiently; a binary search would be much better */
 356     for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
 357
 358     range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
 359
 360     return symbol;
 361 }
 362 /** @} */ // group rangecoder
 363
 364 static inline void update_rice(APERice *rice, int x)
 365 {
 366     int lim = rice->k ? (1 << (rice->k + 4)) : 0;
 367     rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
 368
 369     if (rice->ksum < lim)
 370         rice->k--;
 371     else if (rice->ksum >= (1 << (rice->k + 5)))
 372         rice->k++;
 373 }
 374
 375 static inline int ape_decode_value(APEContext * ctx, APERice *rice)
 376 {
 377     int x, overflow;
 378
 379     if (ctx->fileversion < 3990) {
 380         int tmpk;
 381
 382         overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
 383
 384         if (overflow == (MODEL_ELEMENTS - 1)) {
 385             tmpk = range_decode_bits(ctx, 5);
 386             overflow = 0;
 387         } else
 388             tmpk = (rice->k < 1) ? 0 : rice->k - 1;
 389
 390         if (tmpk <= 16)
 391             x = range_decode_bits(ctx, tmpk);
 392         else {
 393             x = range_decode_bits(ctx, 16);
 394             x |= (range_decode_bits(ctx, tmpk - 16) << 16);
 395         }
 396         x += overflow << tmpk;
 397     } else {
 398         int base, pivot;
 399
 400         pivot = rice->ksum >> 5;
 401         if (pivot == 0)
 402             pivot = 1;
 403
 404         overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
 405
 406         if (overflow == (MODEL_ELEMENTS - 1)) {
 407             overflow  = range_decode_bits(ctx, 16) << 16;
 408             overflow |= range_decode_bits(ctx, 16);
 409         }
 410
 411         base = range_decode_culfreq(ctx, pivot);
 412         range_decode_update(ctx, 1, base);
 413
 414         x = base + overflow * pivot;
 415     }
 416
 417     update_rice(rice, x);
 418
 419     /* Convert to signed */
 420     if (x & 1)
 421         return (x >> 1) + 1;
 422     else
 423         return -(x >> 1);
 424 }
 425
 426 static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo)
 427 {
 428     int32_t *decoded0 = ctx->decoded0;
 429     int32_t *decoded1 = ctx->decoded1;
 430
 431     ctx->blocksdecoded = blockstodecode;
 432
 433     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 434         /* We are pure silence, just memset the output buffer. */
 435         memset(decoded0, 0, blockstodecode * sizeof(int32_t));
 436         memset(decoded1, 0, blockstodecode * sizeof(int32_t));
 437     } else {
 438         while (blockstodecode--) {
 439             *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
 440             if (stereo)
 441                 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
 442         }
 443     }
 444
 445     if (ctx->blocksdecoded == ctx->currentframeblocks)
 446         range_dec_normalize(ctx);   /* normalize to use up all bytes */
 447 }
 448
 449 static void init_entropy_decoder(APEContext * ctx)
 450 {
 451     /* Read the CRC */
 452     ctx->CRC = bytestream_get_be32(&ctx->ptr);
 453
 454     /* Read the frame flags if they exist */
 455     ctx->frameflags = 0;
 456     if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
 457         ctx->CRC &= ~0x80000000;
 458
 459         ctx->frameflags = bytestream_get_be32(&ctx->ptr);
 460     }
 461
 462     /* Keep a count of the blocks decoded in this frame */
 463     ctx->blocksdecoded = 0;
 464
 465     /* Initialize the rice structs */
 466     ctx->riceX.k = 10;
 467     ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
 468     ctx->riceY.k = 10;
 469     ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
 470
 471     /* The first 8 bits of input are ignored. */
 472     ctx->ptr++;
 473
 474     range_start_decoding(ctx);
 475 }
 476
 477 static const int32_t initial_coeffs[4] = {
 478     360, 317, -109, 98
 479 };
 480
 481 static void init_predictor_decoder(APEContext * ctx)
 482 {
 483     APEPredictor *p = &ctx->predictor;
 484
 485     /* Zero the history buffers */
 486     memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
 487     p->buf = p->historybuffer;
 488
 489     /* Initialize and zero the coefficients */
 490     memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
 491     memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
 492     memset(p->coeffsB, 0, sizeof(p->coeffsB));
 493
 494     p->filterA[0] = p->filterA[1] = 0;
 495     p->filterB[0] = p->filterB[1] = 0;
 496     p->lastA[0]   = p->lastA[1]   = 0;
 497 }
 498
 499 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
 500 static inline int APESIGN(int32_t x) {
 501     return (x < 0) - (x > 0);
 502 }
 503
 504 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)
 505 {
 506     int32_t predictionA, predictionB;
 507
 508     p->buf[delayA]     = p->lastA[filter];
 509     p->buf[adaptA]     = APESIGN(p->buf[delayA]);
 510     p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
 511     p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
 512
 513     predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
 514                   p->buf[delayA - 1] * p->coeffsA[filter][1] +
 515                   p->buf[delayA - 2] * p->coeffsA[filter][2] +
 516                   p->buf[delayA - 3] * p->coeffsA[filter][3];
 517
 518     /*  Apply a scaled first-order filter compression */
 519     p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
 520     p->buf[adaptB]     = APESIGN(p->buf[delayB]);
 521     p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
 522     p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
 523     p->filterB[filter] = p->filterA[filter ^ 1];
 524
 525     predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
 526                   p->buf[delayB - 1] * p->coeffsB[filter][1] +
 527                   p->buf[delayB - 2] * p->coeffsB[filter][2] +
 528                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
 529                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 530
 531     p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
 532     p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
 533
 534     if (!decoded) // no need updating filter coefficients
 535         return p->filterA[filter];
 536
 537     if (decoded > 0) {
 538         p->coeffsA[filter][0] -= p->buf[adaptA    ];
 539         p->coeffsA[filter][1] -= p->buf[adaptA - 1];
 540         p->coeffsA[filter][2] -= p->buf[adaptA - 2];
 541         p->coeffsA[filter][3] -= p->buf[adaptA - 3];
 542
 543         p->coeffsB[filter][0] -= p->buf[adaptB    ];
 544         p->coeffsB[filter][1] -= p->buf[adaptB - 1];
 545         p->coeffsB[filter][2] -= p->buf[adaptB - 2];
 546         p->coeffsB[filter][3] -= p->buf[adaptB - 3];
 547         p->coeffsB[filter][4] -= p->buf[adaptB - 4];
 548     } else {
 549         p->coeffsA[filter][0] += p->buf[adaptA    ];
 550         p->coeffsA[filter][1] += p->buf[adaptA - 1];
 551         p->coeffsA[filter][2] += p->buf[adaptA - 2];
 552         p->coeffsA[filter][3] += p->buf[adaptA - 3];
 553
 554         p->coeffsB[filter][0] += p->buf[adaptB    ];
 555         p->coeffsB[filter][1] += p->buf[adaptB - 1];
 556         p->coeffsB[filter][2] += p->buf[adaptB - 2];
 557         p->coeffsB[filter][3] += p->buf[adaptB - 3];
 558         p->coeffsB[filter][4] += p->buf[adaptB - 4];
 559     }
 560     return p->filterA[filter];
 561 }
 562
 563 static void predictor_decode_stereo(APEContext * ctx, int count)
 564 {
 565     int32_t predictionA, predictionB;
 566     APEPredictor *p = &ctx->predictor;
 567     int32_t *decoded0 = ctx->decoded0;
 568     int32_t *decoded1 = ctx->decoded1;
 569
 570     while (count--) {
 571         /* Predictor Y */
 572         predictionA = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB);
 573         predictionB = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB);
 574         *(decoded0++) = predictionA;
 575         *(decoded1++) = predictionB;
 576
 577         /* Combined */
 578         p->buf++;
 579
 580         /* Have we filled the history buffer? */
 581         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 582             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 583             p->buf = p->historybuffer;
 584         }
 585     }
 586 }
 587
 588 static void predictor_decode_mono(APEContext * ctx, int count)
 589 {
 590     APEPredictor *p = &ctx->predictor;
 591     int32_t *decoded0 = ctx->decoded0;
 592     int32_t predictionA, currentA, A;
 593
 594     currentA = p->lastA[0];
 595
 596     while (count--) {
 597         A = *decoded0;
 598
 599         p->buf[YDELAYA] = currentA;
 600         p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
 601
 602         predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
 603                       p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
 604                       p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
 605                       p->buf[YDELAYA - 3] * p->coeffsA[0][3];
 606
 607         currentA = A + (predictionA >> 10);
 608
 609         p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
 610         p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
 611
 612         if (A > 0) {
 613             p->coeffsA[0][0] -= p->buf[YADAPTCOEFFSA    ];
 614             p->coeffsA[0][1] -= p->buf[YADAPTCOEFFSA - 1];
 615             p->coeffsA[0][2] -= p->buf[YADAPTCOEFFSA - 2];
 616             p->coeffsA[0][3] -= p->buf[YADAPTCOEFFSA - 3];
 617         } else if (A < 0) {
 618             p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ];
 619             p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1];
 620             p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2];
 621             p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3];
 622         }
 623
 624         p->buf++;
 625
 626         /* Have we filled the history buffer? */
 627         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 628             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 629             p->buf = p->historybuffer;
 630         }
 631
 632         p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
 633         *(decoded0++) = p->filterA[0];
 634     }
 635
 636     p->lastA[0] = currentA;
 637 }
 638
 639 static void do_init_filter(APEFilter *f, int16_t * buf, int order)
 640 {
 641     f->coeffs = buf;
 642     f->historybuffer = buf + order;
 643     f->delay       = f->historybuffer + order * 2;
 644     f->adaptcoeffs = f->historybuffer + order;
 645
 646     memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
 647     memset(f->coeffs, 0, order * sizeof(int16_t));
 648     f->avg = 0;
 649 }
 650
 651 static void init_filter(APEContext * ctx, APEFilter *f, int16_t * buf, int order)
 652 {
 653     do_init_filter(&f[0], buf, order);
 654     do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
 655 }
 656
 657 static inline void do_apply_filter(APEContext * ctx, int version, APEFilter *f, int32_t *data, int count, int order, int fracbits)
 658 {
 659     int res;
 660     int absres;
 661
 662     while (count--) {
 663         /* round fixedpoint scalar product */
 664         res = (ctx->dsp.scalarproduct_int16(f->delay - order, f->coeffs, order, 0) + (1 << (fracbits - 1))) >> fracbits;
 665
 666         if (*data < 0)
 667             ctx->dsp.add_int16(f->coeffs, f->adaptcoeffs - order, order);
 668         else if (*data > 0)
 669             ctx->dsp.sub_int16(f->coeffs, f->adaptcoeffs - order, order);
 670
 671         res += *data;
 672
 673         *data++ = res;
 674
 675         /* Update the output history */
 676         *f->delay++ = av_clip_int16(res);
 677
 678         if (version < 3980) {
 679             /* Version ??? to < 3.98 files (untested) */
 680             f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
 681             f->adaptcoeffs[-4] >>= 1;
 682             f->adaptcoeffs[-8] >>= 1;
 683         } else {
 684             /* Version 3.98 and later files */
 685
 686             /* Update the adaption coefficients */
 687             absres = (res < 0 ? -res : res);
 688
 689             if (absres > (f->avg * 3))
 690                 *f->adaptcoeffs = ((res >> 25) & 64) - 32;
 691             else if (absres > (f->avg * 4) / 3)
 692                 *f->adaptcoeffs = ((res >> 26) & 32) - 16;
 693             else if (absres > 0)
 694                 *f->adaptcoeffs = ((res >> 27) & 16) - 8;
 695             else
 696                 *f->adaptcoeffs = 0;
 697
 698             f->avg += (absres - f->avg) / 16;
 699
 700             f->adaptcoeffs[-1] >>= 1;
 701             f->adaptcoeffs[-2] >>= 1;
 702             f->adaptcoeffs[-8] >>= 1;
 703         }
 704
 705         f->adaptcoeffs++;
 706
 707         /* Have we filled the history buffer? */
 708         if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
 709             memmove(f->historybuffer, f->delay - (order * 2),
 710                     (order * 2) * sizeof(int16_t));
 711             f->delay = f->historybuffer + order * 2;
 712             f->adaptcoeffs = f->historybuffer + order;
 713         }
 714     }
 715 }
 716
 717 static void apply_filter(APEContext * ctx, APEFilter *f,
 718                          int32_t * data0, int32_t * data1,
 719                          int count, int order, int fracbits)
 720 {
 721     do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
 722     if (data1)
 723         do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
 724 }
 725
 726 static void ape_apply_filters(APEContext * ctx, int32_t * decoded0,
 727                               int32_t * decoded1, int count)
 728 {
 729     int i;
 730
 731     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 732         if (!ape_filter_orders[ctx->fset][i])
 733             break;
 734         apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]);
 735     }
 736 }
 737
 738 static void init_frame_decoder(APEContext * ctx)
 739 {
 740     int i;
 741     init_entropy_decoder(ctx);
 742     init_predictor_decoder(ctx);
 743
 744     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 745         if (!ape_filter_orders[ctx->fset][i])
 746             break;
 747         init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]);
 748     }
 749 }
 750
 751 static void ape_unpack_mono(APEContext * ctx, int count)
 752 {
 753     int32_t left;
 754     int32_t *decoded0 = ctx->decoded0;
 755     int32_t *decoded1 = ctx->decoded1;
 756
 757     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 758         entropy_decode(ctx, count, 0);
 759         /* We are pure silence, so we're done. */
 760         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
 761         return;
 762     }
 763
 764     entropy_decode(ctx, count, 0);
 765     ape_apply_filters(ctx, decoded0, NULL, count);
 766
 767     /* Now apply the predictor decoding */
 768     predictor_decode_mono(ctx, count);
 769
 770     /* Pseudo-stereo - just copy left channel to right channel */
 771     if (ctx->channels == 2) {
 772         while (count--) {
 773             left = *decoded0;
 774             *(decoded1++) = *(decoded0++) = left;
 775         }
 776     }
 777 }
 778
 779 static void ape_unpack_stereo(APEContext * ctx, int count)
 780 {
 781     int32_t left, right;
 782     int32_t *decoded0 = ctx->decoded0;
 783     int32_t *decoded1 = ctx->decoded1;
 784
 785     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 786         /* We are pure silence, so we're done. */
 787         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
 788         return;
 789     }
 790
 791     entropy_decode(ctx, count, 1);
 792     ape_apply_filters(ctx, decoded0, decoded1, count);
 793
 794     /* Now apply the predictor decoding */
 795     predictor_decode_stereo(ctx, count);
 796
 797     /* Decorrelate and scale to output depth */
 798     while (count--) {
 799         left = *decoded1 - (*decoded0 / 2);
 800         right = left + *decoded0;
 801
 802         *(decoded0++) = left;
 803         *(decoded1++) = right;
 804     }
 805 }
 806
 807 static int ape_decode_frame(AVCodecContext * avctx,
 808                             void *data, int *data_size,
 809                             AVPacket *avpkt)
 810 {
 811     const uint8_t *buf = avpkt->data;
 812     int buf_size = avpkt->size;
 813     APEContext *s = avctx->priv_data;
 814     int16_t *samples = data;
 815     int nblocks;
 816     int i, n;
 817     int blockstodecode;
 818     int bytes_used;
 819
 820     if (buf_size == 0 && !s->samples) {
 821         *data_size = 0;
 822         return 0;
 823     }
 824
 825     /* should not happen but who knows */
 826     if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) {
 827         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);
 828         return -1;
 829     }
 830
 831     if(!s->samples){
 832         s->data = av_realloc(s->data, (buf_size + 3) & ~3);
 833         s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
 834         s->ptr = s->last_ptr = s->data;
 835         s->data_end = s->data + buf_size;
 836
 837         nblocks = s->samples = bytestream_get_be32(&s->ptr);
 838         n =  bytestream_get_be32(&s->ptr);
 839         if(n < 0 || n > 3){
 840             av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
 841             s->data = NULL;
 842             return -1;
 843         }
 844         s->ptr += n;
 845
 846         s->currentframeblocks = nblocks;
 847         buf += 4;
 848         if (s->samples <= 0) {
 849             *data_size = 0;
 850             return buf_size;
 851         }
 852
 853         memset(s->decoded0,  0, sizeof(s->decoded0));
 854         memset(s->decoded1,  0, sizeof(s->decoded1));
 855
 856         /* Initialize the frame decoder */
 857         init_frame_decoder(s);
 858     }
 859
 860     if (!s->data) {
 861         *data_size = 0;
 862         return buf_size;
 863     }
 864
 865     nblocks = s->samples;
 866     blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks);
 867
 868     s->error=0;
 869
 870     if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
 871         ape_unpack_mono(s, blockstodecode);
 872     else
 873         ape_unpack_stereo(s, blockstodecode);
 874
 875     if(s->error || s->ptr > s->data_end){
 876         s->samples=0;
 877         av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
 878         return -1;
 879     }
 880
 881     for (i = 0; i < blockstodecode; i++) {
 882         *samples++ = s->decoded0[i];
 883         if(s->channels == 2)
 884             *samples++ = s->decoded1[i];
 885     }
 886
 887     s->samples -= blockstodecode;
 888
 889     *data_size = blockstodecode * 2 * s->channels;
 890     bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size;
 891     s->last_ptr = s->ptr;
 892     return bytes_used;
 893 }
 894
 895 AVCodec ape_decoder = {
 896     "ape",
 897     CODEC_TYPE_AUDIO,
 898     CODEC_ID_APE,
 899     sizeof(APEContext),
 900     ape_decode_init,
 901     NULL,
 902     ape_decode_close,
 903     ape_decode_frame,
 904     .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
 905 };