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
 135     int fileversion;                         ///< codec version, very important in decoding process
 136     int compression_level;                   ///< compression levels
 137     int fset;                                ///< which filter set to use (calculated from compression level)
 138     int flags;                               ///< global decoder flags
 139
 140     uint32_t CRC;                            ///< frame CRC
 141     int frameflags;                          ///< frame flags
 142     APEPredictor predictor;                  ///< predictor used for final reconstruction
 143
 144     int32_t decoded0[BLOCKS_PER_LOOP];       ///< decoded data for the first channel
 145     int32_t decoded1[BLOCKS_PER_LOOP];       ///< decoded data for the second channel
 146
 147     int16_t* filterbuf[APE_FILTER_LEVELS];   ///< filter memory
 148
 149     APERangecoder rc;                        ///< rangecoder used to decode actual values
 150     APERice riceX;                           ///< rice code parameters for the second channel
 151     APERice riceY;                           ///< rice code parameters for the first channel
 152     APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
 153
 154     uint8_t *data;                           ///< current frame data
 155     uint8_t *data_end;                       ///< frame data end
 156     int data_size;                           ///< frame data allocated size
 157     const uint8_t *ptr;                      ///< current position in frame data
 158
 159     int error;
 160 } APEContext;
 161
 162 // TODO: dsputilize
 163
 164 static av_cold int ape_decode_close(AVCodecContext *avctx)
 165 {
 166     APEContext *s = avctx->priv_data;
 167     int i;
 168
 169     for (i = 0; i < APE_FILTER_LEVELS; i++)
 170         av_freep(&s->filterbuf[i]);
 171
 172     av_freep(&s->data);
 173     s->data_size = 0;
 174
 175     return 0;
 176 }
 177
 178 static av_cold int ape_decode_init(AVCodecContext *avctx)
 179 {
 180     APEContext *s = avctx->priv_data;
 181     int i;
 182
 183     if (avctx->extradata_size != 6) {
 184         av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
 185         return AVERROR(EINVAL);
 186     }
 187     if (avctx->bits_per_coded_sample != 16) {
 188         av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n");
 189         return AVERROR(EINVAL);
 190     }
 191     if (avctx->channels > 2) {
 192         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
 193         return AVERROR(EINVAL);
 194     }
 195     s->avctx             = avctx;
 196     s->channels          = avctx->channels;
 197     s->fileversion       = AV_RL16(avctx->extradata);
 198     s->compression_level = AV_RL16(avctx->extradata + 2);
 199     s->flags             = AV_RL16(avctx->extradata + 4);
 200
 201     av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
 202            s->compression_level, s->flags);
 203     if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
 204         av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
 205                s->compression_level);
 206         return AVERROR_INVALIDDATA;
 207     }
 208     s->fset = s->compression_level / 1000 - 1;
 209     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 210         if (!ape_filter_orders[s->fset][i])
 211             break;
 212         FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
 213                          (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
 214                          filter_alloc_fail);
 215     }
 216
 217     dsputil_init(&s->dsp, avctx);
 218     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 219     avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
 220
 221     avcodec_get_frame_defaults(&s->frame);
 222     avctx->coded_frame = &s->frame;
 223
 224     return 0;
 225 filter_alloc_fail:
 226     ape_decode_close(avctx);
 227     return AVERROR(ENOMEM);
 228 }
 229
 230 /**
 231  * @name APE range decoding functions
 232  * @{
 233  */
 234
 235 #define CODE_BITS    32
 236 #define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
 237 #define SHIFT_BITS   (CODE_BITS - 9)
 238 #define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
 239 #define BOTTOM_VALUE (TOP_VALUE >> 8)
 240
 241 /** Start the decoder */
 242 static inline void range_start_decoding(APEContext *ctx)
 243 {
 244     ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
 245     ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
 246     ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
 247 }
 248
 249 /** Perform normalization */
 250 static inline void range_dec_normalize(APEContext *ctx)
 251 {
 252     while (ctx->rc.range <= BOTTOM_VALUE) {
 253         ctx->rc.buffer <<= 8;
 254         if(ctx->ptr < ctx->data_end) {
 255             ctx->rc.buffer += *ctx->ptr;
 256             ctx->ptr++;
 257         } else {
 258             ctx->error = 1;
 259         }
 260         ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
 261         ctx->rc.range  <<= 8;
 262     }
 263 }
 264
 265 /**
 266  * Calculate culmulative frequency for next symbol. Does NO update!
 267  * @param ctx decoder context
 268  * @param tot_f is the total frequency or (code_value)1<<shift
 269  * @return the culmulative frequency
 270  */
 271 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
 272 {
 273     range_dec_normalize(ctx);
 274     ctx->rc.help = ctx->rc.range / tot_f;
 275     return ctx->rc.low / ctx->rc.help;
 276 }
 277
 278 /**
 279  * Decode value with given size in bits
 280  * @param ctx decoder context
 281  * @param shift number of bits to decode
 282  */
 283 static inline int range_decode_culshift(APEContext *ctx, int shift)
 284 {
 285     range_dec_normalize(ctx);
 286     ctx->rc.help = ctx->rc.range >> shift;
 287     return ctx->rc.low / ctx->rc.help;
 288 }
 289
 290
 291 /**
 292  * Update decoding state
 293  * @param ctx decoder context
 294  * @param sy_f the interval length (frequency of the symbol)
 295  * @param lt_f the lower end (frequency sum of < symbols)
 296  */
 297 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
 298 {
 299     ctx->rc.low  -= ctx->rc.help * lt_f;
 300     ctx->rc.range = ctx->rc.help * sy_f;
 301 }
 302
 303 /** Decode n bits (n <= 16) without modelling */
 304 static inline int range_decode_bits(APEContext *ctx, int n)
 305 {
 306     int sym = range_decode_culshift(ctx, n);
 307     range_decode_update(ctx, 1, sym);
 308     return sym;
 309 }
 310
 311
 312 #define MODEL_ELEMENTS 64
 313
 314 /**
 315  * Fixed probabilities for symbols in Monkey Audio version 3.97
 316  */
 317 static const uint16_t counts_3970[22] = {
 318         0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
 319     62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
 320     65450, 65469, 65480, 65487, 65491, 65493,
 321 };
 322
 323 /**
 324  * Probability ranges for symbols in Monkey Audio version 3.97
 325  */
 326 static const uint16_t counts_diff_3970[21] = {
 327     14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
 328     1104, 677, 415, 248, 150, 89, 54, 31,
 329     19, 11, 7, 4, 2,
 330 };
 331
 332 /**
 333  * Fixed probabilities for symbols in Monkey Audio version 3.98
 334  */
 335 static const uint16_t counts_3980[22] = {
 336         0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
 337     64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
 338     65485, 65488, 65490, 65491, 65492, 65493,
 339 };
 340
 341 /**
 342  * Probability ranges for symbols in Monkey Audio version 3.98
 343  */
 344 static const uint16_t counts_diff_3980[21] = {
 345     19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
 346     261, 119, 65, 31, 19, 10, 6, 3,
 347     3, 2, 1, 1, 1,
 348 };
 349
 350 /**
 351  * Decode symbol
 352  * @param ctx decoder context
 353  * @param counts probability range start position
 354  * @param counts_diff probability range widths
 355  */
 356 static inline int range_get_symbol(APEContext *ctx,
 357                                    const uint16_t counts[],
 358                                    const uint16_t counts_diff[])
 359 {
 360     int symbol, cf;
 361
 362     cf = range_decode_culshift(ctx, 16);
 363
 364     if(cf > 65492){
 365         symbol= cf - 65535 + 63;
 366         range_decode_update(ctx, 1, cf);
 367         if(cf > 65535)
 368             ctx->error=1;
 369         return symbol;
 370     }
 371     /* figure out the symbol inefficiently; a binary search would be much better */
 372     for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
 373
 374     range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
 375
 376     return symbol;
 377 }
 378 /** @} */ // group rangecoder
 379
 380 static inline void update_rice(APERice *rice, int x)
 381 {
 382     int lim = rice->k ? (1 << (rice->k + 4)) : 0;
 383     rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
 384
 385     if (rice->ksum < lim)
 386         rice->k--;
 387     else if (rice->ksum >= (1 << (rice->k + 5)))
 388         rice->k++;
 389 }
 390
 391 static inline int ape_decode_value(APEContext *ctx, APERice *rice)
 392 {
 393     int x, overflow;
 394
 395     if (ctx->fileversion < 3990) {
 396         int tmpk;
 397
 398         overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
 399
 400         if (overflow == (MODEL_ELEMENTS - 1)) {
 401             tmpk = range_decode_bits(ctx, 5);
 402             overflow = 0;
 403         } else
 404             tmpk = (rice->k < 1) ? 0 : rice->k - 1;
 405
 406         if (tmpk <= 16)
 407             x = range_decode_bits(ctx, tmpk);
 408         else {
 409             x = range_decode_bits(ctx, 16);
 410             x |= (range_decode_bits(ctx, tmpk - 16) << 16);
 411         }
 412         x += overflow << tmpk;
 413     } else {
 414         int base, pivot;
 415
 416         pivot = rice->ksum >> 5;
 417         if (pivot == 0)
 418             pivot = 1;
 419
 420         overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
 421
 422         if (overflow == (MODEL_ELEMENTS - 1)) {
 423             overflow  = range_decode_bits(ctx, 16) << 16;
 424             overflow |= range_decode_bits(ctx, 16);
 425         }
 426
 427         if (pivot < 0x10000) {
 428             base = range_decode_culfreq(ctx, pivot);
 429             range_decode_update(ctx, 1, base);
 430         } else {
 431             int base_hi = pivot, base_lo;
 432             int bbits = 0;
 433
 434             while (base_hi & ~0xFFFF) {
 435                 base_hi >>= 1;
 436                 bbits++;
 437             }
 438             base_hi = range_decode_culfreq(ctx, base_hi + 1);
 439             range_decode_update(ctx, 1, base_hi);
 440             base_lo = range_decode_culfreq(ctx, 1 << bbits);
 441             range_decode_update(ctx, 1, base_lo);
 442
 443             base = (base_hi << bbits) + base_lo;
 444         }
 445
 446         x = base + overflow * pivot;
 447     }
 448
 449     update_rice(rice, x);
 450
 451     /* Convert to signed */
 452     if (x & 1)
 453         return (x >> 1) + 1;
 454     else
 455         return -(x >> 1);
 456 }
 457
 458 static void entropy_decode(APEContext *ctx, int blockstodecode, int stereo)
 459 {
 460     int32_t *decoded0 = ctx->decoded0;
 461     int32_t *decoded1 = ctx->decoded1;
 462
 463     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 464         /* We are pure silence, just memset the output buffer. */
 465         memset(decoded0, 0, blockstodecode * sizeof(int32_t));
 466         memset(decoded1, 0, blockstodecode * sizeof(int32_t));
 467     } else {
 468         while (blockstodecode--) {
 469             *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
 470             if (stereo)
 471                 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
 472         }
 473     }
 474 }
 475
 476 static int init_entropy_decoder(APEContext *ctx)
 477 {
 478     /* Read the CRC */
 479     if (ctx->data_end - ctx->ptr < 6)
 480         return AVERROR_INVALIDDATA;
 481     ctx->CRC = bytestream_get_be32(&ctx->ptr);
 482
 483     /* Read the frame flags if they exist */
 484     ctx->frameflags = 0;
 485     if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
 486         ctx->CRC &= ~0x80000000;
 487
 488         if (ctx->data_end - ctx->ptr < 6)
 489             return AVERROR_INVALIDDATA;
 490         ctx->frameflags = bytestream_get_be32(&ctx->ptr);
 491     }
 492
 493     /* Initialize the rice structs */
 494     ctx->riceX.k = 10;
 495     ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
 496     ctx->riceY.k = 10;
 497     ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
 498
 499     /* The first 8 bits of input are ignored. */
 500     ctx->ptr++;
 501
 502     range_start_decoding(ctx);
 503
 504     return 0;
 505 }
 506
 507 static const int32_t initial_coeffs[4] = {
 508     360, 317, -109, 98
 509 };
 510
 511 static void init_predictor_decoder(APEContext *ctx)
 512 {
 513     APEPredictor *p = &ctx->predictor;
 514
 515     /* Zero the history buffers */
 516     memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
 517     p->buf = p->historybuffer;
 518
 519     /* Initialize and zero the coefficients */
 520     memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
 521     memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
 522     memset(p->coeffsB, 0, sizeof(p->coeffsB));
 523
 524     p->filterA[0] = p->filterA[1] = 0;
 525     p->filterB[0] = p->filterB[1] = 0;
 526     p->lastA[0]   = p->lastA[1]   = 0;
 527 }
 528
 529 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
 530 static inline int APESIGN(int32_t x) {
 531     return (x < 0) - (x > 0);
 532 }
 533
 534 static av_always_inline int predictor_update_filter(APEPredictor *p,
 535                                                     const int decoded, const int filter,
 536                                                     const int delayA,  const int delayB,
 537                                                     const int adaptA,  const int adaptB)
 538 {
 539     int32_t predictionA, predictionB, sign;
 540
 541     p->buf[delayA]     = p->lastA[filter];
 542     p->buf[adaptA]     = APESIGN(p->buf[delayA]);
 543     p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
 544     p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
 545
 546     predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
 547                   p->buf[delayA - 1] * p->coeffsA[filter][1] +
 548                   p->buf[delayA - 2] * p->coeffsA[filter][2] +
 549                   p->buf[delayA - 3] * p->coeffsA[filter][3];
 550
 551     /*  Apply a scaled first-order filter compression */
 552     p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
 553     p->buf[adaptB]     = APESIGN(p->buf[delayB]);
 554     p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
 555     p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
 556     p->filterB[filter] = p->filterA[filter ^ 1];
 557
 558     predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
 559                   p->buf[delayB - 1] * p->coeffsB[filter][1] +
 560                   p->buf[delayB - 2] * p->coeffsB[filter][2] +
 561                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
 562                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 563
 564     p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
 565     p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
 566
 567     sign = APESIGN(decoded);
 568     p->coeffsA[filter][0] += p->buf[adaptA    ] * sign;
 569     p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
 570     p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
 571     p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
 572     p->coeffsB[filter][0] += p->buf[adaptB    ] * sign;
 573     p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
 574     p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
 575     p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
 576     p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
 577
 578     return p->filterA[filter];
 579 }
 580
 581 static void predictor_decode_stereo(APEContext *ctx, int count)
 582 {
 583     APEPredictor *p = &ctx->predictor;
 584     int32_t *decoded0 = ctx->decoded0;
 585     int32_t *decoded1 = ctx->decoded1;
 586
 587     while (count--) {
 588         /* Predictor Y */
 589         *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
 590                                             YADAPTCOEFFSA, YADAPTCOEFFSB);
 591         decoded0++;
 592         *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
 593                                             XADAPTCOEFFSA, XADAPTCOEFFSB);
 594         decoded1++;
 595
 596         /* Combined */
 597         p->buf++;
 598
 599         /* Have we filled the history buffer? */
 600         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 601             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 602             p->buf = p->historybuffer;
 603         }
 604     }
 605 }
 606
 607 static void predictor_decode_mono(APEContext *ctx, int count)
 608 {
 609     APEPredictor *p = &ctx->predictor;
 610     int32_t *decoded0 = ctx->decoded0;
 611     int32_t predictionA, currentA, A, sign;
 612
 613     currentA = p->lastA[0];
 614
 615     while (count--) {
 616         A = *decoded0;
 617
 618         p->buf[YDELAYA] = currentA;
 619         p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
 620
 621         predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
 622                       p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
 623                       p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
 624                       p->buf[YDELAYA - 3] * p->coeffsA[0][3];
 625
 626         currentA = A + (predictionA >> 10);
 627
 628         p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
 629         p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
 630
 631         sign = APESIGN(A);
 632         p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ] * sign;
 633         p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
 634         p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
 635         p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
 636
 637         p->buf++;
 638
 639         /* Have we filled the history buffer? */
 640         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 641             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 642             p->buf = p->historybuffer;
 643         }
 644
 645         p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
 646         *(decoded0++) = p->filterA[0];
 647     }
 648
 649     p->lastA[0] = currentA;
 650 }
 651
 652 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
 653 {
 654     f->coeffs = buf;
 655     f->historybuffer = buf + order;
 656     f->delay       = f->historybuffer + order * 2;
 657     f->adaptcoeffs = f->historybuffer + order;
 658
 659     memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
 660     memset(f->coeffs, 0, order * sizeof(int16_t));
 661     f->avg = 0;
 662 }
 663
 664 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
 665 {
 666     do_init_filter(&f[0], buf, order);
 667     do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
 668 }
 669
 670 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
 671                             int32_t *data, int count, int order, int fracbits)
 672 {
 673     int res;
 674     int absres;
 675
 676     while (count--) {
 677         /* round fixedpoint scalar product */
 678         res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order,
 679                                                     f->adaptcoeffs - order,
 680                                                     order, APESIGN(*data));
 681         res = (res + (1 << (fracbits - 1))) >> fracbits;
 682         res += *data;
 683         *data++ = res;
 684
 685         /* Update the output history */
 686         *f->delay++ = av_clip_int16(res);
 687
 688         if (version < 3980) {
 689             /* Version ??? to < 3.98 files (untested) */
 690             f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
 691             f->adaptcoeffs[-4] >>= 1;
 692             f->adaptcoeffs[-8] >>= 1;
 693         } else {
 694             /* Version 3.98 and later files */
 695
 696             /* Update the adaption coefficients */
 697             absres = FFABS(res);
 698             if (absres)
 699                 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
 700                                   (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
 701             else
 702                 *f->adaptcoeffs = 0;
 703
 704             f->avg += (absres - f->avg) / 16;
 705
 706             f->adaptcoeffs[-1] >>= 1;
 707             f->adaptcoeffs[-2] >>= 1;
 708             f->adaptcoeffs[-8] >>= 1;
 709         }
 710
 711         f->adaptcoeffs++;
 712
 713         /* Have we filled the history buffer? */
 714         if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
 715             memmove(f->historybuffer, f->delay - (order * 2),
 716                     (order * 2) * sizeof(int16_t));
 717             f->delay = f->historybuffer + order * 2;
 718             f->adaptcoeffs = f->historybuffer + order;
 719         }
 720     }
 721 }
 722
 723 static void apply_filter(APEContext *ctx, APEFilter *f,
 724                          int32_t *data0, int32_t *data1,
 725                          int count, int order, int fracbits)
 726 {
 727     do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
 728     if (data1)
 729         do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
 730 }
 731
 732 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
 733                               int32_t *decoded1, int count)
 734 {
 735     int i;
 736
 737     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 738         if (!ape_filter_orders[ctx->fset][i])
 739             break;
 740         apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
 741                      ape_filter_orders[ctx->fset][i],
 742                      ape_filter_fracbits[ctx->fset][i]);
 743     }
 744 }
 745
 746 static int init_frame_decoder(APEContext *ctx)
 747 {
 748     int i, ret;
 749     if ((ret = init_entropy_decoder(ctx)) < 0)
 750         return ret;
 751     init_predictor_decoder(ctx);
 752
 753     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 754         if (!ape_filter_orders[ctx->fset][i])
 755             break;
 756         init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
 757                     ape_filter_orders[ctx->fset][i]);
 758     }
 759     return 0;
 760 }
 761
 762 static void ape_unpack_mono(APEContext *ctx, int count)
 763 {
 764     int32_t *decoded0 = ctx->decoded0;
 765     int32_t *decoded1 = ctx->decoded1;
 766
 767     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 768         entropy_decode(ctx, count, 0);
 769         /* We are pure silence, so we're done. */
 770         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
 771         return;
 772     }
 773
 774     entropy_decode(ctx, count, 0);
 775     ape_apply_filters(ctx, decoded0, NULL, count);
 776
 777     /* Now apply the predictor decoding */
 778     predictor_decode_mono(ctx, count);
 779
 780     /* Pseudo-stereo - just copy left channel to right channel */
 781     if (ctx->channels == 2) {
 782         memcpy(decoded1, decoded0, count * sizeof(*decoded1));
 783     }
 784 }
 785
 786 static void ape_unpack_stereo(APEContext *ctx, int count)
 787 {
 788     int32_t left, right;
 789     int32_t *decoded0 = ctx->decoded0;
 790     int32_t *decoded1 = ctx->decoded1;
 791
 792     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 793         /* We are pure silence, so we're done. */
 794         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
 795         return;
 796     }
 797
 798     entropy_decode(ctx, count, 1);
 799     ape_apply_filters(ctx, decoded0, decoded1, count);
 800
 801     /* Now apply the predictor decoding */
 802     predictor_decode_stereo(ctx, count);
 803
 804     /* Decorrelate and scale to output depth */
 805     while (count--) {
 806         left = *decoded1 - (*decoded0 / 2);
 807         right = left + *decoded0;
 808
 809         *(decoded0++) = left;
 810         *(decoded1++) = right;
 811     }
 812 }
 813
 814 static int ape_decode_frame(AVCodecContext *avctx, void *data,
 815                             int *got_frame_ptr, AVPacket *avpkt)
 816 {
 817     const uint8_t *buf = avpkt->data;
 818     APEContext *s = avctx->priv_data;
 819     int16_t *samples;
 820     int i, ret;
 821     int blockstodecode;
 822     int bytes_used = 0;
 823
 824     /* this should never be negative, but bad things will happen if it is, so
 825        check it just to make sure. */
 826     av_assert0(s->samples >= 0);
 827
 828     if(!s->samples){
 829         uint32_t nblocks, offset;
 830         int buf_size;
 831
 832         if (!avpkt->size) {
 833             *got_frame_ptr = 0;
 834             return 0;
 835         }
 836         if (avpkt->size < 8) {
 837             av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
 838             return AVERROR_INVALIDDATA;
 839         }
 840         buf_size = avpkt->size & ~3;
 841         if (buf_size != avpkt->size) {
 842             av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
 843                    "extra bytes at the end will be skipped.\n");
 844         }
 845
 846         av_fast_malloc(&s->data, &s->data_size, buf_size);
 847         if (!s->data)
 848             return AVERROR(ENOMEM);
 849         s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
 850         s->ptr = s->data;
 851         s->data_end = s->data + buf_size;
 852
 853         nblocks = bytestream_get_be32(&s->ptr);
 854         offset  = bytestream_get_be32(&s->ptr);
 855         if (offset > 3) {
 856             av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
 857             s->data = NULL;
 858             return AVERROR_INVALIDDATA;
 859         }
 860         if (s->data_end - s->ptr < offset) {
 861             av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
 862             return AVERROR_INVALIDDATA;
 863         }
 864         s->ptr += offset;
 865
 866         if (!nblocks || nblocks > INT_MAX) {
 867             av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %u.\n", nblocks);
 868             return AVERROR_INVALIDDATA;
 869         }
 870         s->samples = nblocks;
 871
 872         memset(s->decoded0,  0, sizeof(s->decoded0));
 873         memset(s->decoded1,  0, sizeof(s->decoded1));
 874
 875         /* Initialize the frame decoder */
 876         if (init_frame_decoder(s) < 0) {
 877             av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
 878             return AVERROR_INVALIDDATA;
 879         }
 880
 881         bytes_used = avpkt->size;
 882     }
 883
 884     if (!s->data) {
 885         *got_frame_ptr = 0;
 886         return avpkt->size;
 887     }
 888
 889     blockstodecode = FFMIN(BLOCKS_PER_LOOP, s->samples);
 890
 891     /* get output buffer */
 892     s->frame.nb_samples = blockstodecode;
 893     if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
 894         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 895         return ret;
 896     }
 897     samples = (int16_t *)s->frame.data[0];
 898
 899     s->error=0;
 900
 901     if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
 902         ape_unpack_mono(s, blockstodecode);
 903     else
 904         ape_unpack_stereo(s, blockstodecode);
 905     emms_c();
 906
 907     if (s->error) {
 908         s->samples=0;
 909         av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
 910         return AVERROR_INVALIDDATA;
 911     }
 912
 913     for (i = 0; i < blockstodecode; i++) {
 914         *samples++ = s->decoded0[i];
 915         if(s->channels == 2)
 916             *samples++ = s->decoded1[i];
 917     }
 918
 919     s->samples -= blockstodecode;
 920
 921     *got_frame_ptr   = 1;
 922     *(AVFrame *)data = s->frame;
 923
 924     return bytes_used;
 925 }
 926
 927 static void ape_flush(AVCodecContext *avctx)
 928 {
 929     APEContext *s = avctx->priv_data;
 930     s->samples= 0;
 931 }
 932
 933 AVCodec ff_ape_decoder = {
 934     .name           = "ape",
 935     .type           = AVMEDIA_TYPE_AUDIO,
 936     .id             = CODEC_ID_APE,
 937     .priv_data_size = sizeof(APEContext),
 938     .init           = ape_decode_init,
 939     .close          = ape_decode_close,
 940     .decode         = ape_decode_frame,
 941     .capabilities   = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
 942     .flush = ape_flush,
 943     .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
 944 };