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 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 #include "avcodec.h"
  24 #include "dsputil.h"
  25 #include "bytestream.h"
  26 #include "libavutil/audioconvert.h"
  27 #include "libavutil/avassert.h"
  28
  29 /**
  30  * @file
  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     AVFrame frame;
 131     DSPContext dsp;
 132     int channels;
 133     int samples;                             ///< samples left to decode in current frame
 134     int bps;
 135
 136     int fileversion;                         ///< codec version, very important in decoding process
 137     int compression_level;                   ///< compression levels
 138     int fset;                                ///< which filter set to use (calculated from compression level)
 139     int flags;                               ///< global decoder flags
 140
 141     uint32_t CRC;                            ///< frame CRC
 142     int frameflags;                          ///< frame flags
 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     int data_size;                           ///< frame data allocated size
 158     const uint8_t *ptr;                      ///< current position in frame data
 159
 160     int error;
 161 } APEContext;
 162
 163 // TODO: dsputilize
 164
 165 static av_cold int ape_decode_close(AVCodecContext *avctx)
 166 {
 167     APEContext *s = avctx->priv_data;
 168     int i;
 169
 170     for (i = 0; i < APE_FILTER_LEVELS; i++)
 171         av_freep(&s->filterbuf[i]);
 172
 173     av_freep(&s->data);
 174     s->data_size = 0;
 175
 176     return 0;
 177 }
 178
 179 static av_cold int ape_decode_init(AVCodecContext *avctx)
 180 {
 181     APEContext *s = avctx->priv_data;
 182     int i;
 183
 184     if (avctx->extradata_size != 6) {
 185         av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
 186         return AVERROR(EINVAL);
 187     }
 188     if (avctx->channels > 2) {
 189         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
 190         return AVERROR(EINVAL);
 191     }
 192     s->bps = avctx->bits_per_coded_sample;
 193     switch (s->bps) {
 194     case 8:
 195         avctx->sample_fmt = AV_SAMPLE_FMT_U8;
 196         break;
 197     case 16:
 198         avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 199         break;
 200     case 24:
 201         avctx->sample_fmt = AV_SAMPLE_FMT_S32;
 202         break;
 203     default:
 204         av_log_ask_for_sample(avctx, "Unsupported bits per coded sample %d\n",
 205                               s->bps);
 206         return AVERROR_PATCHWELCOME;
 207     }
 208     s->avctx             = avctx;
 209     s->channels          = avctx->channels;
 210     s->fileversion       = AV_RL16(avctx->extradata);
 211     s->compression_level = AV_RL16(avctx->extradata + 2);
 212     s->flags             = AV_RL16(avctx->extradata + 4);
 213
 214     av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
 215            s->compression_level, s->flags);
 216     if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
 217         av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
 218                s->compression_level);
 219         return AVERROR_INVALIDDATA;
 220     }
 221     s->fset = s->compression_level / 1000 - 1;
 222     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 223         if (!ape_filter_orders[s->fset][i])
 224             break;
 225         FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
 226                          (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
 227                          filter_alloc_fail);
 228     }
 229
 230     dsputil_init(&s->dsp, avctx);
 231     avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
 232
 233     avcodec_get_frame_defaults(&s->frame);
 234     avctx->coded_frame = &s->frame;
 235
 236     return 0;
 237 filter_alloc_fail:
 238     ape_decode_close(avctx);
 239     return AVERROR(ENOMEM);
 240 }
 241
 242 /**
 243  * @name APE range decoding functions
 244  * @{
 245  */
 246
 247 #define CODE_BITS    32
 248 #define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
 249 #define SHIFT_BITS   (CODE_BITS - 9)
 250 #define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
 251 #define BOTTOM_VALUE (TOP_VALUE >> 8)
 252
 253 /** Start the decoder */
 254 static inline void range_start_decoding(APEContext *ctx)
 255 {
 256     ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
 257     ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
 258     ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
 259 }
 260
 261 /** Perform normalization */
 262 static inline void range_dec_normalize(APEContext *ctx)
 263 {
 264     while (ctx->rc.range <= BOTTOM_VALUE) {
 265         ctx->rc.buffer <<= 8;
 266         if(ctx->ptr < ctx->data_end) {
 267             ctx->rc.buffer += *ctx->ptr;
 268             ctx->ptr++;
 269         } else {
 270             ctx->error = 1;
 271         }
 272         ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
 273         ctx->rc.range  <<= 8;
 274     }
 275 }
 276
 277 /**
 278  * Calculate culmulative frequency for next symbol. Does NO update!
 279  * @param ctx decoder context
 280  * @param tot_f is the total frequency or (code_value)1<<shift
 281  * @return the culmulative frequency
 282  */
 283 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
 284 {
 285     range_dec_normalize(ctx);
 286     ctx->rc.help = ctx->rc.range / tot_f;
 287     return ctx->rc.low / ctx->rc.help;
 288 }
 289
 290 /**
 291  * Decode value with given size in bits
 292  * @param ctx decoder context
 293  * @param shift number of bits to decode
 294  */
 295 static inline int range_decode_culshift(APEContext *ctx, int shift)
 296 {
 297     range_dec_normalize(ctx);
 298     ctx->rc.help = ctx->rc.range >> shift;
 299     return ctx->rc.low / ctx->rc.help;
 300 }
 301
 302
 303 /**
 304  * Update decoding state
 305  * @param ctx decoder context
 306  * @param sy_f the interval length (frequency of the symbol)
 307  * @param lt_f the lower end (frequency sum of < symbols)
 308  */
 309 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
 310 {
 311     ctx->rc.low  -= ctx->rc.help * lt_f;
 312     ctx->rc.range = ctx->rc.help * sy_f;
 313 }
 314
 315 /** Decode n bits (n <= 16) without modelling */
 316 static inline int range_decode_bits(APEContext *ctx, int n)
 317 {
 318     int sym = range_decode_culshift(ctx, n);
 319     range_decode_update(ctx, 1, sym);
 320     return sym;
 321 }
 322
 323
 324 #define MODEL_ELEMENTS 64
 325
 326 /**
 327  * Fixed probabilities for symbols in Monkey Audio version 3.97
 328  */
 329 static const uint16_t counts_3970[22] = {
 330         0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
 331     62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
 332     65450, 65469, 65480, 65487, 65491, 65493,
 333 };
 334
 335 /**
 336  * Probability ranges for symbols in Monkey Audio version 3.97
 337  */
 338 static const uint16_t counts_diff_3970[21] = {
 339     14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
 340     1104, 677, 415, 248, 150, 89, 54, 31,
 341     19, 11, 7, 4, 2,
 342 };
 343
 344 /**
 345  * Fixed probabilities for symbols in Monkey Audio version 3.98
 346  */
 347 static const uint16_t counts_3980[22] = {
 348         0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
 349     64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
 350     65485, 65488, 65490, 65491, 65492, 65493,
 351 };
 352
 353 /**
 354  * Probability ranges for symbols in Monkey Audio version 3.98
 355  */
 356 static const uint16_t counts_diff_3980[21] = {
 357     19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
 358     261, 119, 65, 31, 19, 10, 6, 3,
 359     3, 2, 1, 1, 1,
 360 };
 361
 362 /**
 363  * Decode symbol
 364  * @param ctx decoder context
 365  * @param counts probability range start position
 366  * @param counts_diff probability range widths
 367  */
 368 static inline int range_get_symbol(APEContext *ctx,
 369                                    const uint16_t counts[],
 370                                    const uint16_t counts_diff[])
 371 {
 372     int symbol, cf;
 373
 374     cf = range_decode_culshift(ctx, 16);
 375
 376     if(cf > 65492){
 377         symbol= cf - 65535 + 63;
 378         range_decode_update(ctx, 1, cf);
 379         if(cf > 65535)
 380             ctx->error=1;
 381         return symbol;
 382     }
 383     /* figure out the symbol inefficiently; a binary search would be much better */
 384     for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
 385
 386     range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
 387
 388     return symbol;
 389 }
 390 /** @} */ // group rangecoder
 391
 392 static inline void update_rice(APERice *rice, int x)
 393 {
 394     int lim = rice->k ? (1 << (rice->k + 4)) : 0;
 395     rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
 396
 397     if (rice->ksum < lim)
 398         rice->k--;
 399     else if (rice->ksum >= (1 << (rice->k + 5)))
 400         rice->k++;
 401 }
 402
 403 static inline int ape_decode_value(APEContext *ctx, APERice *rice)
 404 {
 405     int x, overflow;
 406
 407     if (ctx->fileversion < 3990) {
 408         int tmpk;
 409
 410         overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
 411
 412         if (overflow == (MODEL_ELEMENTS - 1)) {
 413             tmpk = range_decode_bits(ctx, 5);
 414             overflow = 0;
 415         } else
 416             tmpk = (rice->k < 1) ? 0 : rice->k - 1;
 417
 418         if (tmpk <= 16)
 419             x = range_decode_bits(ctx, tmpk);
 420         else {
 421             x = range_decode_bits(ctx, 16);
 422             x |= (range_decode_bits(ctx, tmpk - 16) << 16);
 423         }
 424         x += overflow << tmpk;
 425     } else {
 426         int base, pivot;
 427
 428         pivot = rice->ksum >> 5;
 429         if (pivot == 0)
 430             pivot = 1;
 431
 432         overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
 433
 434         if (overflow == (MODEL_ELEMENTS - 1)) {
 435             overflow  = range_decode_bits(ctx, 16) << 16;
 436             overflow |= range_decode_bits(ctx, 16);
 437         }
 438
 439         if (pivot < 0x10000) {
 440             base = range_decode_culfreq(ctx, pivot);
 441             range_decode_update(ctx, 1, base);
 442         } else {
 443             int base_hi = pivot, base_lo;
 444             int bbits = 0;
 445
 446             while (base_hi & ~0xFFFF) {
 447                 base_hi >>= 1;
 448                 bbits++;
 449             }
 450             base_hi = range_decode_culfreq(ctx, base_hi + 1);
 451             range_decode_update(ctx, 1, base_hi);
 452             base_lo = range_decode_culfreq(ctx, 1 << bbits);
 453             range_decode_update(ctx, 1, base_lo);
 454
 455             base = (base_hi << bbits) + base_lo;
 456         }
 457
 458         x = base + overflow * pivot;
 459     }
 460
 461     update_rice(rice, x);
 462
 463     /* Convert to signed */
 464     if (x & 1)
 465         return (x >> 1) + 1;
 466     else
 467         return -(x >> 1);
 468 }
 469
 470 static void entropy_decode(APEContext *ctx, int blockstodecode, int stereo)
 471 {
 472     int32_t *decoded0 = ctx->decoded0;
 473     int32_t *decoded1 = ctx->decoded1;
 474
 475     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 476         /* We are pure silence, just memset the output buffer. */
 477         memset(decoded0, 0, blockstodecode * sizeof(int32_t));
 478         memset(decoded1, 0, blockstodecode * sizeof(int32_t));
 479     } else {
 480         while (blockstodecode--) {
 481             *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
 482             if (stereo)
 483                 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
 484         }
 485     }
 486 }
 487
 488 static int init_entropy_decoder(APEContext *ctx)
 489 {
 490     /* Read the CRC */
 491     if (ctx->data_end - ctx->ptr < 6)
 492         return AVERROR_INVALIDDATA;
 493     ctx->CRC = bytestream_get_be32(&ctx->ptr);
 494
 495     /* Read the frame flags if they exist */
 496     ctx->frameflags = 0;
 497     if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
 498         ctx->CRC &= ~0x80000000;
 499
 500         if (ctx->data_end - ctx->ptr < 6)
 501             return AVERROR_INVALIDDATA;
 502         ctx->frameflags = bytestream_get_be32(&ctx->ptr);
 503     }
 504
 505     /* Initialize the rice structs */
 506     ctx->riceX.k = 10;
 507     ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
 508     ctx->riceY.k = 10;
 509     ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
 510
 511     /* The first 8 bits of input are ignored. */
 512     ctx->ptr++;
 513
 514     range_start_decoding(ctx);
 515
 516     return 0;
 517 }
 518
 519 static const int32_t initial_coeffs[4] = {
 520     360, 317, -109, 98
 521 };
 522
 523 static void init_predictor_decoder(APEContext *ctx)
 524 {
 525     APEPredictor *p = &ctx->predictor;
 526
 527     /* Zero the history buffers */
 528     memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
 529     p->buf = p->historybuffer;
 530
 531     /* Initialize and zero the coefficients */
 532     memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
 533     memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
 534     memset(p->coeffsB, 0, sizeof(p->coeffsB));
 535
 536     p->filterA[0] = p->filterA[1] = 0;
 537     p->filterB[0] = p->filterB[1] = 0;
 538     p->lastA[0]   = p->lastA[1]   = 0;
 539 }
 540
 541 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
 542 static inline int APESIGN(int32_t x) {
 543     return (x < 0) - (x > 0);
 544 }
 545
 546 static av_always_inline int predictor_update_filter(APEPredictor *p,
 547                                                     const int decoded, const int filter,
 548                                                     const int delayA,  const int delayB,
 549                                                     const int adaptA,  const int adaptB)
 550 {
 551     int32_t predictionA, predictionB, sign;
 552
 553     p->buf[delayA]     = p->lastA[filter];
 554     p->buf[adaptA]     = APESIGN(p->buf[delayA]);
 555     p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
 556     p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
 557
 558     predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
 559                   p->buf[delayA - 1] * p->coeffsA[filter][1] +
 560                   p->buf[delayA - 2] * p->coeffsA[filter][2] +
 561                   p->buf[delayA - 3] * p->coeffsA[filter][3];
 562
 563     /*  Apply a scaled first-order filter compression */
 564     p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
 565     p->buf[adaptB]     = APESIGN(p->buf[delayB]);
 566     p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
 567     p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
 568     p->filterB[filter] = p->filterA[filter ^ 1];
 569
 570     predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
 571                   p->buf[delayB - 1] * p->coeffsB[filter][1] +
 572                   p->buf[delayB - 2] * p->coeffsB[filter][2] +
 573                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
 574                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 575
 576     p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
 577     p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
 578
 579     sign = APESIGN(decoded);
 580     p->coeffsA[filter][0] += p->buf[adaptA    ] * sign;
 581     p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
 582     p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
 583     p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
 584     p->coeffsB[filter][0] += p->buf[adaptB    ] * sign;
 585     p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
 586     p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
 587     p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
 588     p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
 589
 590     return p->filterA[filter];
 591 }
 592
 593 static void predictor_decode_stereo(APEContext *ctx, int count)
 594 {
 595     APEPredictor *p = &ctx->predictor;
 596     int32_t *decoded0 = ctx->decoded0;
 597     int32_t *decoded1 = ctx->decoded1;
 598
 599     while (count--) {
 600         /* Predictor Y */
 601         *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
 602                                             YADAPTCOEFFSA, YADAPTCOEFFSB);
 603         decoded0++;
 604         *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
 605                                             XADAPTCOEFFSA, XADAPTCOEFFSB);
 606         decoded1++;
 607
 608         /* Combined */
 609         p->buf++;
 610
 611         /* Have we filled the history buffer? */
 612         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 613             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 614             p->buf = p->historybuffer;
 615         }
 616     }
 617 }
 618
 619 static void predictor_decode_mono(APEContext *ctx, int count)
 620 {
 621     APEPredictor *p = &ctx->predictor;
 622     int32_t *decoded0 = ctx->decoded0;
 623     int32_t predictionA, currentA, A, sign;
 624
 625     currentA = p->lastA[0];
 626
 627     while (count--) {
 628         A = *decoded0;
 629
 630         p->buf[YDELAYA] = currentA;
 631         p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
 632
 633         predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
 634                       p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
 635                       p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
 636                       p->buf[YDELAYA - 3] * p->coeffsA[0][3];
 637
 638         currentA = A + (predictionA >> 10);
 639
 640         p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
 641         p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
 642
 643         sign = APESIGN(A);
 644         p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ] * sign;
 645         p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
 646         p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
 647         p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
 648
 649         p->buf++;
 650
 651         /* Have we filled the history buffer? */
 652         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 653             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 654             p->buf = p->historybuffer;
 655         }
 656
 657         p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
 658         *(decoded0++) = p->filterA[0];
 659     }
 660
 661     p->lastA[0] = currentA;
 662 }
 663
 664 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
 665 {
 666     f->coeffs = buf;
 667     f->historybuffer = buf + order;
 668     f->delay       = f->historybuffer + order * 2;
 669     f->adaptcoeffs = f->historybuffer + order;
 670
 671     memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
 672     memset(f->coeffs, 0, order * sizeof(int16_t));
 673     f->avg = 0;
 674 }
 675
 676 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
 677 {
 678     do_init_filter(&f[0], buf, order);
 679     do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
 680 }
 681
 682 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
 683                             int32_t *data, int count, int order, int fracbits)
 684 {
 685     int res;
 686     int absres;
 687
 688     while (count--) {
 689         /* round fixedpoint scalar product */
 690         res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order,
 691                                                     f->adaptcoeffs - order,
 692                                                     order, APESIGN(*data));
 693         res = (res + (1 << (fracbits - 1))) >> fracbits;
 694         res += *data;
 695         *data++ = res;
 696
 697         /* Update the output history */
 698         *f->delay++ = av_clip_int16(res);
 699
 700         if (version < 3980) {
 701             /* Version ??? to < 3.98 files (untested) */
 702             f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
 703             f->adaptcoeffs[-4] >>= 1;
 704             f->adaptcoeffs[-8] >>= 1;
 705         } else {
 706             /* Version 3.98 and later files */
 707
 708             /* Update the adaption coefficients */
 709             absres = FFABS(res);
 710             if (absres)
 711                 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
 712                                   (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
 713             else
 714                 *f->adaptcoeffs = 0;
 715
 716             f->avg += (absres - f->avg) / 16;
 717
 718             f->adaptcoeffs[-1] >>= 1;
 719             f->adaptcoeffs[-2] >>= 1;
 720             f->adaptcoeffs[-8] >>= 1;
 721         }
 722
 723         f->adaptcoeffs++;
 724
 725         /* Have we filled the history buffer? */
 726         if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
 727             memmove(f->historybuffer, f->delay - (order * 2),
 728                     (order * 2) * sizeof(int16_t));
 729             f->delay = f->historybuffer + order * 2;
 730             f->adaptcoeffs = f->historybuffer + order;
 731         }
 732     }
 733 }
 734
 735 static void apply_filter(APEContext *ctx, APEFilter *f,
 736                          int32_t *data0, int32_t *data1,
 737                          int count, int order, int fracbits)
 738 {
 739     do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
 740     if (data1)
 741         do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
 742 }
 743
 744 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
 745                               int32_t *decoded1, int count)
 746 {
 747     int i;
 748
 749     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 750         if (!ape_filter_orders[ctx->fset][i])
 751             break;
 752         apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
 753                      ape_filter_orders[ctx->fset][i],
 754                      ape_filter_fracbits[ctx->fset][i]);
 755     }
 756 }
 757
 758 static int init_frame_decoder(APEContext *ctx)
 759 {
 760     int i, ret;
 761     if ((ret = init_entropy_decoder(ctx)) < 0)
 762         return ret;
 763     init_predictor_decoder(ctx);
 764
 765     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 766         if (!ape_filter_orders[ctx->fset][i])
 767             break;
 768         init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
 769                     ape_filter_orders[ctx->fset][i]);
 770     }
 771     return 0;
 772 }
 773
 774 static void ape_unpack_mono(APEContext *ctx, int count)
 775 {
 776     int32_t *decoded0 = ctx->decoded0;
 777     int32_t *decoded1 = ctx->decoded1;
 778
 779     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 780         entropy_decode(ctx, count, 0);
 781         /* We are pure silence, so we're done. */
 782         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
 783         return;
 784     }
 785
 786     entropy_decode(ctx, count, 0);
 787     ape_apply_filters(ctx, decoded0, NULL, count);
 788
 789     /* Now apply the predictor decoding */
 790     predictor_decode_mono(ctx, count);
 791
 792     /* Pseudo-stereo - just copy left channel to right channel */
 793     if (ctx->channels == 2) {
 794         memcpy(decoded1, decoded0, count * sizeof(*decoded1));
 795     }
 796 }
 797
 798 static void ape_unpack_stereo(APEContext *ctx, int count)
 799 {
 800     int32_t left, right;
 801     int32_t *decoded0 = ctx->decoded0;
 802     int32_t *decoded1 = ctx->decoded1;
 803
 804     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 805         /* We are pure silence, so we're done. */
 806         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
 807         return;
 808     }
 809
 810     entropy_decode(ctx, count, 1);
 811     ape_apply_filters(ctx, decoded0, decoded1, count);
 812
 813     /* Now apply the predictor decoding */
 814     predictor_decode_stereo(ctx, count);
 815
 816     /* Decorrelate and scale to output depth */
 817     while (count--) {
 818         left = *decoded1 - (*decoded0 / 2);
 819         right = left + *decoded0;
 820
 821         *(decoded0++) = left;
 822         *(decoded1++) = right;
 823     }
 824 }
 825
 826 static int ape_decode_frame(AVCodecContext *avctx, void *data,
 827                             int *got_frame_ptr, AVPacket *avpkt)
 828 {
 829     const uint8_t *buf = avpkt->data;
 830     APEContext *s = avctx->priv_data;
 831     uint8_t *sample8;
 832     int16_t *sample16;
 833     int32_t *sample24;
 834     int i, ret;
 835     int blockstodecode;
 836     int bytes_used = 0;
 837
 838     /* this should never be negative, but bad things will happen if it is, so
 839        check it just to make sure. */
 840     av_assert0(s->samples >= 0);
 841
 842     if(!s->samples){
 843         uint32_t nblocks, offset;
 844         int buf_size;
 845
 846         if (!avpkt->size) {
 847             *got_frame_ptr = 0;
 848             return 0;
 849         }
 850         if (avpkt->size < 8) {
 851             av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
 852             return AVERROR_INVALIDDATA;
 853         }
 854         buf_size = avpkt->size & ~3;
 855         if (buf_size != avpkt->size) {
 856             av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
 857                    "extra bytes at the end will be skipped.\n");
 858         }
 859
 860         av_fast_malloc(&s->data, &s->data_size, buf_size);
 861         if (!s->data)
 862             return AVERROR(ENOMEM);
 863         s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
 864         s->ptr = s->data;
 865         s->data_end = s->data + buf_size;
 866
 867         nblocks = bytestream_get_be32(&s->ptr);
 868         offset  = bytestream_get_be32(&s->ptr);
 869         if (offset > 3) {
 870             av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
 871             s->data = NULL;
 872             return AVERROR_INVALIDDATA;
 873         }
 874         if (s->data_end - s->ptr < offset) {
 875             av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
 876             return AVERROR_INVALIDDATA;
 877         }
 878         s->ptr += offset;
 879
 880         if (!nblocks || nblocks > INT_MAX) {
 881             av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %u.\n", nblocks);
 882             return AVERROR_INVALIDDATA;
 883         }
 884         s->samples = nblocks;
 885
 886         memset(s->decoded0,  0, sizeof(s->decoded0));
 887         memset(s->decoded1,  0, sizeof(s->decoded1));
 888
 889         /* Initialize the frame decoder */
 890         if (init_frame_decoder(s) < 0) {
 891             av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
 892             return AVERROR_INVALIDDATA;
 893         }
 894
 895         bytes_used = avpkt->size;
 896     }
 897
 898     if (!s->data) {
 899         *got_frame_ptr = 0;
 900         return avpkt->size;
 901     }
 902
 903     blockstodecode = FFMIN(BLOCKS_PER_LOOP, s->samples);
 904
 905     /* get output buffer */
 906     s->frame.nb_samples = blockstodecode;
 907     if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
 908         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 909         return ret;
 910     }
 911
 912     s->error=0;
 913
 914     if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
 915         ape_unpack_mono(s, blockstodecode);
 916     else
 917         ape_unpack_stereo(s, blockstodecode);
 918     emms_c();
 919
 920     if (s->error) {
 921         s->samples=0;
 922         av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
 923         return AVERROR_INVALIDDATA;
 924     }
 925
 926     switch (s->bps) {
 927     case 8:
 928         sample8 = (uint8_t *)s->frame.data[0];
 929         for (i = 0; i < blockstodecode; i++) {
 930             *sample8++ = (s->decoded0[i] + 0x80) & 0xff;
 931             if (s->channels == 2)
 932                 *sample8++ = (s->decoded1[i] + 0x80) & 0xff;
 933         }
 934         break;
 935     case 16:
 936         sample16 = (int16_t *)s->frame.data[0];
 937         for (i = 0; i < blockstodecode; i++) {
 938             *sample16++ = s->decoded0[i];
 939             if (s->channels == 2)
 940                 *sample16++ = s->decoded1[i];
 941         }
 942         break;
 943     case 24:
 944         sample24 = (int32_t *)s->frame.data[0];
 945         for (i = 0; i < blockstodecode; i++) {
 946             *sample24++ = s->decoded0[i] << 8;
 947             if (s->channels == 2)
 948                 *sample24++ = s->decoded1[i] << 8;
 949         }
 950         break;
 951     }
 952
 953     s->samples -= blockstodecode;
 954
 955     *got_frame_ptr   = 1;
 956     *(AVFrame *)data = s->frame;
 957
 958     return bytes_used;
 959 }
 960
 961 static void ape_flush(AVCodecContext *avctx)
 962 {
 963     APEContext *s = avctx->priv_data;
 964     s->samples= 0;
 965 }
 966
 967 AVCodec ff_ape_decoder = {
 968     .name           = "ape",
 969     .type           = AVMEDIA_TYPE_AUDIO,
 970     .id             = CODEC_ID_APE,
 971     .priv_data_size = sizeof(APEContext),
 972     .init           = ape_decode_init,
 973     .close          = ape_decode_close,
 974     .decode         = ape_decode_frame,
 975     .capabilities   = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
 976     .flush = ape_flush,
 977     .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
 978 };