libavcodec/apedec.c

   1 /*
   2  * Monkey's Audio lossless audio decoder
   3  * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org>
   4  *  based upon libdemac from Dave Chapman.
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #define ALT_BITSTREAM_READER_LE
  24 #include "avcodec.h"
  25 #include "dsputil.h"
  26 #include "bitstream.h"
  27 #include "bytestream.h"
  28
  29 /**
  30  * @file apedec.c
  31  * Monkey's Audio lossless audio decoder
  32  */
  33
  34 #define BLOCKS_PER_LOOP     4608
  35 #define MAX_CHANNELS        2
  36 #define MAX_BYTESPERSAMPLE  3
  37
  38 #define APE_FRAMECODE_MONO_SILENCE    1
  39 #define APE_FRAMECODE_STEREO_SILENCE  3
  40 #define APE_FRAMECODE_PSEUDO_STEREO   4
  41
  42 #define HISTORY_SIZE 512
  43 #define PREDICTOR_ORDER 8
  44 /** Total size of all predictor histories */
  45 #define PREDICTOR_SIZE 50
  46
  47 #define YDELAYA (18 + PREDICTOR_ORDER*4)
  48 #define YDELAYB (18 + PREDICTOR_ORDER*3)
  49 #define XDELAYA (18 + PREDICTOR_ORDER*2)
  50 #define XDELAYB (18 + PREDICTOR_ORDER)
  51
  52 #define YADAPTCOEFFSA 18
  53 #define XADAPTCOEFFSA 14
  54 #define YADAPTCOEFFSB 10
  55 #define XADAPTCOEFFSB 5
  56
  57 /**
  58  * Possible compression levels
  59  * @{
  60  */
  61 enum APECompressionLevel {
  62     COMPRESSION_LEVEL_FAST       = 1000,
  63     COMPRESSION_LEVEL_NORMAL     = 2000,
  64     COMPRESSION_LEVEL_HIGH       = 3000,
  65     COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
  66     COMPRESSION_LEVEL_INSANE     = 5000
  67 };
  68 /** @} */
  69
  70 #define APE_FILTER_LEVELS 3
  71
  72 /** Filter orders depending on compression level */
  73 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
  74     {  0,   0,    0 },
  75     { 16,   0,    0 },
  76     { 64,   0,    0 },
  77     { 32, 256,    0 },
  78     { 16, 256, 1280 }
  79 };
  80
  81 /** Filter fraction bits depending on compression level */
  82 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
  83     {  0,  0,  0 },
  84     { 11,  0,  0 },
  85     { 11,  0,  0 },
  86     { 10, 13,  0 },
  87     { 11, 13, 15 }
  88 };
  89
  90
  91 /** Filters applied to the decoded data */
  92 typedef struct APEFilter {
  93     int16_t *coeffs;        ///< actual coefficients used in filtering
  94     int16_t *adaptcoeffs;   ///< adaptive filter coefficients used for correcting of actual filter coefficients
  95     int16_t *historybuffer; ///< filter memory
  96     int16_t *delay;         ///< filtered values
  97
  98     int avg;
  99 } APEFilter;
 100
 101 typedef struct APERice {
 102     uint32_t k;
 103     uint32_t ksum;
 104 } APERice;
 105
 106 typedef struct APERangecoder {
 107     uint32_t low;           ///< low end of interval
 108     uint32_t range;         ///< length of interval
 109     uint32_t help;          ///< bytes_to_follow resp. intermediate value
 110     unsigned int buffer;    ///< buffer for input/output
 111 } APERangecoder;
 112
 113 /** Filter histories */
 114 typedef struct APEPredictor {
 115     int32_t *buf;
 116
 117     int32_t lastA[2];
 118
 119     int32_t filterA[2];
 120     int32_t filterB[2];
 121
 122     int32_t coeffsA[2][4];  ///< adaption coefficients
 123     int32_t coeffsB[2][5];  ///< adaption coefficients
 124     int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
 125 } APEPredictor;
 126
 127 /** Decoder context */
 128 typedef struct APEContext {
 129     AVCodecContext *avctx;
 130     DSPContext dsp;
 131     int channels;
 132     int samples;                             ///< samples left to decode in current frame
 133
 134     int fileversion;                         ///< codec version, very important in decoding process
 135     int compression_level;                   ///< compression levels
 136     int fset;                                ///< which filter set to use (calculated from compression level)
 137     int flags;                               ///< global decoder flags
 138
 139     uint32_t CRC;                            ///< frame CRC
 140     int frameflags;                          ///< frame flags
 141     int currentframeblocks;                  ///< samples (per channel) in current frame
 142     int blocksdecoded;                       ///< count of decoded samples in current frame
 143     APEPredictor predictor;                  ///< predictor used for final reconstruction
 144
 145     int32_t decoded0[BLOCKS_PER_LOOP];       ///< decoded data for the first channel
 146     int32_t decoded1[BLOCKS_PER_LOOP];       ///< decoded data for the second channel
 147
 148     int16_t* filterbuf[APE_FILTER_LEVELS];   ///< filter memory
 149
 150     APERangecoder rc;                        ///< rangecoder used to decode actual values
 151     APERice riceX;                           ///< rice code parameters for the second channel
 152     APERice riceY;                           ///< rice code parameters for the first channel
 153     APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
 154
 155     uint8_t *data;                           ///< current frame data
 156     uint8_t *data_end;                       ///< frame data end
 157     const uint8_t *ptr;                      ///< current position in frame data
 158     const uint8_t *last_ptr;                 ///< position where last 4608-sample block ended
 159
 160     int error;
 161 } APEContext;
 162
 163 // TODO: dsputilize
 164 static inline void vector_add(int16_t * v1, int16_t * v2, int order)
 165 {
 166     while (order--)
 167        *v1++ += *v2++;
 168 }
 169
 170 // TODO: dsputilize
 171 static inline void vector_sub(int16_t * v1, int16_t * v2, int order)
 172 {
 173     while (order--)
 174         *v1++ -= *v2++;
 175 }
 176
 177 // TODO: dsputilize
 178 static inline int32_t scalarproduct(int16_t * v1, int16_t * v2, int order)
 179 {
 180     int res = 0;
 181
 182     while (order--)
 183         res += *v1++ * *v2++;
 184
 185     return res;
 186 }
 187
 188 static av_cold int ape_decode_init(AVCodecContext * avctx)
 189 {
 190     APEContext *s = avctx->priv_data;
 191     int i;
 192
 193     if (avctx->extradata_size != 6) {
 194         av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
 195         return -1;
 196     }
 197     if (avctx->bits_per_sample != 16) {
 198         av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n");
 199         return -1;
 200     }
 201     if (avctx->channels > 2) {
 202         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
 203         return -1;
 204     }
 205     s->avctx             = avctx;
 206     s->channels          = avctx->channels;
 207     s->fileversion       = AV_RL16(avctx->extradata);
 208     s->compression_level = AV_RL16(avctx->extradata + 2);
 209     s->flags             = AV_RL16(avctx->extradata + 4);
 210
 211     av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n", s->compression_level, s->flags);
 212     if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
 213         av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n", s->compression_level);
 214         return -1;
 215     }
 216     s->fset = s->compression_level / 1000 - 1;
 217     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 218         if (!ape_filter_orders[s->fset][i])
 219             break;
 220         s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4);
 221     }
 222
 223     dsputil_init(&s->dsp, avctx);
 224     return 0;
 225 }
 226
 227 static av_cold int ape_decode_close(AVCodecContext * avctx)
 228 {
 229     APEContext *s = avctx->priv_data;
 230     int i;
 231
 232     for (i = 0; i < APE_FILTER_LEVELS; i++)
 233         av_freep(&s->filterbuf[i]);
 234
 235     return 0;
 236 }
 237
 238 /**
 239  * @defgroup rangecoder APE range decoder
 240  * @{
 241  */
 242
 243 #define CODE_BITS    32
 244 #define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
 245 #define SHIFT_BITS   (CODE_BITS - 9)
 246 #define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
 247 #define BOTTOM_VALUE (TOP_VALUE >> 8)
 248
 249 /** Start the decoder */
 250 static inline void range_start_decoding(APEContext * ctx)
 251 {
 252     ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
 253     ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
 254     ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
 255 }
 256
 257 /** Perform normalization */
 258 static inline void range_dec_normalize(APEContext * ctx)
 259 {
 260     while (ctx->rc.range <= BOTTOM_VALUE) {
 261         ctx->rc.buffer <<= 8;
 262         if(ctx->ptr < ctx->data_end)
 263             ctx->rc.buffer += *ctx->ptr;
 264         ctx->ptr++;
 265         ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
 266         ctx->rc.range  <<= 8;
 267     }
 268 }
 269
 270 /**
 271  * Calculate culmulative frequency for next symbol. Does NO update!
 272  * @param tot_f is the total frequency or (code_value)1<<shift
 273  * @return the culmulative frequency
 274  */
 275 static inline int range_decode_culfreq(APEContext * ctx, int tot_f)
 276 {
 277     range_dec_normalize(ctx);
 278     ctx->rc.help = ctx->rc.range / tot_f;
 279     return ctx->rc.low / ctx->rc.help;
 280 }
 281
 282 /**
 283  * Decode value with given size in bits
 284  * @param shift number of bits to decode
 285  */
 286 static inline int range_decode_culshift(APEContext * ctx, int shift)
 287 {
 288     range_dec_normalize(ctx);
 289     ctx->rc.help = ctx->rc.range >> shift;
 290     return ctx->rc.low / ctx->rc.help;
 291 }
 292
 293
 294 /**
 295  * Update decoding state
 296  * @param sy_f the interval length (frequency of the symbol)
 297  * @param lt_f the lower end (frequency sum of < symbols)
 298  */
 299 static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f)
 300 {
 301     ctx->rc.low  -= ctx->rc.help * lt_f;
 302     ctx->rc.range = ctx->rc.help * sy_f;
 303 }
 304
 305 /** Decode n bits (n <= 16) without modelling */
 306 static inline int range_decode_bits(APEContext * ctx, int n)
 307 {
 308     int sym = range_decode_culshift(ctx, n);
 309     range_decode_update(ctx, 1, sym);
 310     return sym;
 311 }
 312
 313
 314 #define MODEL_ELEMENTS 64
 315
 316 /**
 317  * Fixed probabilities for symbols in Monkey Audio version 3.97
 318  */
 319 static const uint16_t counts_3970[22] = {
 320         0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
 321     62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
 322     65450, 65469, 65480, 65487, 65491, 65493,
 323 };
 324
 325 /**
 326  * Probability ranges for symbols in Monkey Audio version 3.97
 327  */
 328 static const uint16_t counts_diff_3970[21] = {
 329     14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
 330     1104, 677, 415, 248, 150, 89, 54, 31,
 331     19, 11, 7, 4, 2,
 332 };
 333
 334 /**
 335  * Fixed probabilities for symbols in Monkey Audio version 3.98
 336  */
 337 static const uint16_t counts_3980[22] = {
 338         0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
 339     64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
 340     65485, 65488, 65490, 65491, 65492, 65493,
 341 };
 342
 343 /**
 344  * Probability ranges for symbols in Monkey Audio version 3.98
 345  */
 346 static const uint16_t counts_diff_3980[21] = {
 347     19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
 348     261, 119, 65, 31, 19, 10, 6, 3,
 349     3, 2, 1, 1, 1,
 350 };
 351
 352 /**
 353  * Decode symbol
 354  * @param counts probability range start position
 355  * @param count_diffs probability range widths
 356  */
 357 static inline int range_get_symbol(APEContext * ctx,
 358                                    const uint16_t counts[],
 359                                    const uint16_t counts_diff[])
 360 {
 361     int symbol, cf;
 362
 363     cf = range_decode_culshift(ctx, 16);
 364
 365     if(cf > 65492){
 366         symbol= cf - 65535 + 63;
 367         range_decode_update(ctx, 1, cf);
 368         if(cf > 65535)
 369             ctx->error=1;
 370         return symbol;
 371     }
 372     /* figure out the symbol inefficiently; a binary search would be much better */
 373     for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
 374
 375     range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
 376
 377     return symbol;
 378 }
 379 /** @} */ // group rangecoder
 380
 381 static inline void update_rice(APERice *rice, int x)
 382 {
 383     rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
 384
 385     if (rice->k == 0)
 386         rice->k = 1;
 387     else if (rice->ksum < (1 << (rice->k + 4)))
 388         rice->k--;
 389     else if (rice->ksum >= (1 << (rice->k + 5)))
 390         rice->k++;
 391 }
 392
 393 static inline int ape_decode_value(APEContext * ctx, APERice *rice)
 394 {
 395     int x, overflow;
 396
 397     if (ctx->fileversion < 3980) {
 398         int tmpk;
 399
 400         overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
 401
 402         if (overflow == (MODEL_ELEMENTS - 1)) {
 403             tmpk = range_decode_bits(ctx, 5);
 404             overflow = 0;
 405         } else
 406             tmpk = (rice->k < 1) ? 0 : rice->k - 1;
 407
 408         if (tmpk <= 16)
 409             x = range_decode_bits(ctx, tmpk);
 410         else {
 411             x = range_decode_bits(ctx, 16);
 412             x |= (range_decode_bits(ctx, tmpk - 16) << 16);
 413         }
 414         x += overflow << tmpk;
 415     } else {
 416         int base, pivot;
 417
 418         pivot = rice->ksum >> 5;
 419         if (pivot == 0)
 420             pivot = 1;
 421
 422         overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
 423
 424         if (overflow == (MODEL_ELEMENTS - 1)) {
 425             overflow  = range_decode_bits(ctx, 16) << 16;
 426             overflow |= range_decode_bits(ctx, 16);
 427         }
 428
 429         base = range_decode_culfreq(ctx, pivot);
 430         range_decode_update(ctx, 1, base);
 431
 432         x = base + overflow * pivot;
 433     }
 434
 435     update_rice(rice, x);
 436
 437     /* Convert to signed */
 438     if (x & 1)
 439         return (x >> 1) + 1;
 440     else
 441         return -(x >> 1);
 442 }
 443
 444 static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo)
 445 {
 446     int32_t *decoded0 = ctx->decoded0;
 447     int32_t *decoded1 = ctx->decoded1;
 448
 449     ctx->blocksdecoded = blockstodecode;
 450
 451     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 452         /* We are pure silence, just memset the output buffer. */
 453         memset(decoded0, 0, blockstodecode * sizeof(int32_t));
 454         memset(decoded1, 0, blockstodecode * sizeof(int32_t));
 455     } else {
 456         while (blockstodecode--) {
 457             *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
 458             if (stereo)
 459                 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
 460         }
 461     }
 462
 463     if (ctx->blocksdecoded == ctx->currentframeblocks)
 464         range_dec_normalize(ctx);   /* normalize to use up all bytes */
 465 }
 466
 467 static void init_entropy_decoder(APEContext * ctx)
 468 {
 469     /* Read the CRC */
 470     ctx->CRC = bytestream_get_be32(&ctx->ptr);
 471
 472     /* Read the frame flags if they exist */
 473     ctx->frameflags = 0;
 474     if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
 475         ctx->CRC &= ~0x80000000;
 476
 477         ctx->frameflags = bytestream_get_be32(&ctx->ptr);
 478     }
 479
 480     /* Keep a count of the blocks decoded in this frame */
 481     ctx->blocksdecoded = 0;
 482
 483     /* Initialize the rice structs */
 484     ctx->riceX.k = 10;
 485     ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
 486     ctx->riceY.k = 10;
 487     ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
 488
 489     /* The first 8 bits of input are ignored. */
 490     ctx->ptr++;
 491
 492     range_start_decoding(ctx);
 493 }
 494
 495 static const int32_t initial_coeffs[4] = {
 496     360, 317, -109, 98
 497 };
 498
 499 static void init_predictor_decoder(APEContext * ctx)
 500 {
 501     APEPredictor *p = &ctx->predictor;
 502
 503     /* Zero the history buffers */
 504     memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
 505     p->buf = p->historybuffer;
 506
 507     /* Initialize and zero the coefficients */
 508     memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
 509     memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
 510     memset(p->coeffsB, 0, sizeof(p->coeffsB));
 511
 512     p->filterA[0] = p->filterA[1] = 0;
 513     p->filterB[0] = p->filterB[1] = 0;
 514     p->lastA[0]   = p->lastA[1]   = 0;
 515 }
 516
 517 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
 518 static inline int APESIGN(int32_t x) {
 519     return (x < 0) - (x > 0);
 520 }
 521
 522 static int predictor_update_filter(APEPredictor *p, const int decoded, const int filter, const int delayA, const int delayB, const int adaptA, const int adaptB)
 523 {
 524     int32_t predictionA, predictionB;
 525
 526     p->buf[delayA]     = p->lastA[filter];
 527     p->buf[adaptA]     = APESIGN(p->buf[delayA]);
 528     p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
 529     p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
 530
 531     predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
 532                   p->buf[delayA - 1] * p->coeffsA[filter][1] +
 533                   p->buf[delayA - 2] * p->coeffsA[filter][2] +
 534                   p->buf[delayA - 3] * p->coeffsA[filter][3];
 535
 536     /*  Apply a scaled first-order filter compression */
 537     p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
 538     p->buf[adaptB]     = APESIGN(p->buf[delayB]);
 539     p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
 540     p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
 541     p->filterB[filter] = p->filterA[filter ^ 1];
 542
 543     predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
 544                   p->buf[delayB - 1] * p->coeffsB[filter][1] +
 545                   p->buf[delayB - 2] * p->coeffsB[filter][2] +
 546                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
 547                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 548
 549     p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
 550     p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
 551
 552     if (!decoded) // no need updating filter coefficients
 553         return p->filterA[filter];
 554
 555     if (decoded > 0) {
 556         p->coeffsA[filter][0] -= p->buf[adaptA    ];
 557         p->coeffsA[filter][1] -= p->buf[adaptA - 1];
 558         p->coeffsA[filter][2] -= p->buf[adaptA - 2];
 559         p->coeffsA[filter][3] -= p->buf[adaptA - 3];
 560
 561         p->coeffsB[filter][0] -= p->buf[adaptB    ];
 562         p->coeffsB[filter][1] -= p->buf[adaptB - 1];
 563         p->coeffsB[filter][2] -= p->buf[adaptB - 2];
 564         p->coeffsB[filter][3] -= p->buf[adaptB - 3];
 565         p->coeffsB[filter][4] -= p->buf[adaptB - 4];
 566     } else {
 567         p->coeffsA[filter][0] += p->buf[adaptA    ];
 568         p->coeffsA[filter][1] += p->buf[adaptA - 1];
 569         p->coeffsA[filter][2] += p->buf[adaptA - 2];
 570         p->coeffsA[filter][3] += p->buf[adaptA - 3];
 571
 572         p->coeffsB[filter][0] += p->buf[adaptB    ];
 573         p->coeffsB[filter][1] += p->buf[adaptB - 1];
 574         p->coeffsB[filter][2] += p->buf[adaptB - 2];
 575         p->coeffsB[filter][3] += p->buf[adaptB - 3];
 576         p->coeffsB[filter][4] += p->buf[adaptB - 4];
 577     }
 578     return p->filterA[filter];
 579 }
 580
 581 static void predictor_decode_stereo(APEContext * ctx, int count)
 582 {
 583     int32_t predictionA, predictionB;
 584     APEPredictor *p = &ctx->predictor;
 585     int32_t *decoded0 = ctx->decoded0;
 586     int32_t *decoded1 = ctx->decoded1;
 587
 588     while (count--) {
 589         /* Predictor Y */
 590         predictionA = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB);
 591         predictionB = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB);
 592         *(decoded0++) = predictionA;
 593         *(decoded1++) = predictionB;
 594
 595         /* Combined */
 596         p->buf++;
 597
 598         /* Have we filled the history buffer? */
 599         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 600             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 601             p->buf = p->historybuffer;
 602         }
 603     }
 604 }
 605
 606 static void predictor_decode_mono(APEContext * ctx, int count)
 607 {
 608     APEPredictor *p = &ctx->predictor;
 609     int32_t *decoded0 = ctx->decoded0;
 610     int32_t predictionA, currentA, A;
 611
 612     currentA = p->lastA[0];
 613
 614     while (count--) {
 615         A = *decoded0;
 616
 617         p->buf[YDELAYA] = currentA;
 618         p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
 619
 620         predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
 621                       p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
 622                       p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
 623                       p->buf[YDELAYA - 3] * p->coeffsA[0][3];
 624
 625         currentA = A + (predictionA >> 10);
 626
 627         p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
 628         p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
 629
 630         if (A > 0) {
 631             p->coeffsA[0][0] -= p->buf[YADAPTCOEFFSA    ];
 632             p->coeffsA[0][1] -= p->buf[YADAPTCOEFFSA - 1];
 633             p->coeffsA[0][2] -= p->buf[YADAPTCOEFFSA - 2];
 634             p->coeffsA[0][3] -= p->buf[YADAPTCOEFFSA - 3];
 635         } else if (A < 0) {
 636             p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ];
 637             p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1];
 638             p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2];
 639             p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3];
 640         }
 641
 642         p->buf++;
 643
 644         /* Have we filled the history buffer? */
 645         if (p->buf == p->historybuffer + HISTORY_SIZE) {
 646             memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
 647             p->buf = p->historybuffer;
 648         }
 649
 650         p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
 651         *(decoded0++) = p->filterA[0];
 652     }
 653
 654     p->lastA[0] = currentA;
 655 }
 656
 657 static void do_init_filter(APEFilter *f, int16_t * buf, int order)
 658 {
 659     f->coeffs = buf;
 660     f->historybuffer = buf + order;
 661     f->delay       = f->historybuffer + order * 2;
 662     f->adaptcoeffs = f->historybuffer + order;
 663
 664     memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
 665     memset(f->coeffs, 0, order * sizeof(int16_t));
 666     f->avg = 0;
 667 }
 668
 669 static void init_filter(APEContext * ctx, APEFilter *f, int16_t * buf, int order)
 670 {
 671     do_init_filter(&f[0], buf, order);
 672     do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
 673 }
 674
 675 static inline void do_apply_filter(int version, APEFilter *f, int32_t *data, int count, int order, int fracbits)
 676 {
 677     int res;
 678     int absres;
 679
 680     while (count--) {
 681         /* round fixedpoint scalar product */
 682         res = (scalarproduct(f->delay - order, f->coeffs, order) + (1 << (fracbits - 1))) >> fracbits;
 683
 684         if (*data < 0)
 685             vector_add(f->coeffs, f->adaptcoeffs - order, order);
 686         else if (*data > 0)
 687             vector_sub(f->coeffs, f->adaptcoeffs - order, order);
 688
 689         res += *data;
 690
 691         *data++ = res;
 692
 693         /* Update the output history */
 694         *f->delay++ = av_clip_int16(res);
 695
 696         if (version < 3980) {
 697             /* Version ??? to < 3.98 files (untested) */
 698             f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
 699             f->adaptcoeffs[-4] >>= 1;
 700             f->adaptcoeffs[-8] >>= 1;
 701         } else {
 702             /* Version 3.98 and later files */
 703
 704             /* Update the adaption coefficients */
 705             absres = (res < 0 ? -res : res);
 706
 707             if (absres > (f->avg * 3))
 708                 *f->adaptcoeffs = ((res >> 25) & 64) - 32;
 709             else if (absres > (f->avg * 4) / 3)
 710                 *f->adaptcoeffs = ((res >> 26) & 32) - 16;
 711             else if (absres > 0)
 712                 *f->adaptcoeffs = ((res >> 27) & 16) - 8;
 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->fileversion, &f[0], data0, count, order, fracbits);
 740     if (data1)
 741         do_apply_filter(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, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]);
 753     }
 754 }
 755
 756 static void init_frame_decoder(APEContext * ctx)
 757 {
 758     int i;
 759     init_entropy_decoder(ctx);
 760     init_predictor_decoder(ctx);
 761
 762     for (i = 0; i < APE_FILTER_LEVELS; i++) {
 763         if (!ape_filter_orders[ctx->fset][i])
 764             break;
 765         init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]);
 766     }
 767 }
 768
 769 static void ape_unpack_mono(APEContext * ctx, int count)
 770 {
 771     int32_t left;
 772     int32_t *decoded0 = ctx->decoded0;
 773     int32_t *decoded1 = ctx->decoded1;
 774
 775     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 776         entropy_decode(ctx, count, 0);
 777         /* We are pure silence, so we're done. */
 778         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
 779         return;
 780     }
 781
 782     entropy_decode(ctx, count, 0);
 783     ape_apply_filters(ctx, decoded0, NULL, count);
 784
 785     /* Now apply the predictor decoding */
 786     predictor_decode_mono(ctx, count);
 787
 788     /* Pseudo-stereo - just copy left channel to right channel */
 789     if (ctx->channels == 2) {
 790         while (count--) {
 791             left = *decoded0;
 792             *(decoded1++) = *(decoded0++) = left;
 793         }
 794     }
 795 }
 796
 797 static void ape_unpack_stereo(APEContext * ctx, int count)
 798 {
 799     int32_t left, right;
 800     int32_t *decoded0 = ctx->decoded0;
 801     int32_t *decoded1 = ctx->decoded1;
 802
 803     if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
 804         /* We are pure silence, so we're done. */
 805         av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
 806         return;
 807     }
 808
 809     entropy_decode(ctx, count, 1);
 810     ape_apply_filters(ctx, decoded0, decoded1, count);
 811
 812     /* Now apply the predictor decoding */
 813     predictor_decode_stereo(ctx, count);
 814
 815     /* Decorrelate and scale to output depth */
 816     while (count--) {
 817         left = *decoded1 - (*decoded0 / 2);
 818         right = left + *decoded0;
 819
 820         *(decoded0++) = left;
 821         *(decoded1++) = right;
 822     }
 823 }
 824
 825 static int ape_decode_frame(AVCodecContext * avctx,
 826                             void *data, int *data_size,
 827                             const uint8_t * buf, int buf_size)
 828 {
 829     APEContext *s = avctx->priv_data;
 830     int16_t *samples = data;
 831     int nblocks;
 832     int i, n;
 833     int blockstodecode;
 834     int bytes_used;
 835
 836     if (buf_size == 0 && !s->samples) {
 837         *data_size = 0;
 838         return 0;
 839     }
 840
 841     /* should not happen but who knows */
 842     if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) {
 843         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);
 844         return -1;
 845     }
 846
 847     if(!s->samples){
 848         s->data = av_realloc(s->data, (buf_size + 3) & ~3);
 849         s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
 850         s->ptr = s->last_ptr = s->data;
 851         s->data_end = s->data + buf_size;
 852
 853         nblocks = s->samples = bytestream_get_be32(&s->ptr);
 854         n =  bytestream_get_be32(&s->ptr);
 855         if(n < 0 || n > 3){
 856             av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
 857             s->data = NULL;
 858             return -1;
 859         }
 860         s->ptr += n;
 861
 862         s->currentframeblocks = nblocks;
 863         buf += 4;
 864         if (s->samples <= 0) {
 865             *data_size = 0;
 866             return buf_size;
 867         }
 868
 869         memset(s->decoded0,  0, sizeof(s->decoded0));
 870         memset(s->decoded1,  0, sizeof(s->decoded1));
 871
 872         /* Initialize the frame decoder */
 873         init_frame_decoder(s);
 874     }
 875
 876     if (!s->data) {
 877         *data_size = 0;
 878         return buf_size;
 879     }
 880
 881     nblocks = s->samples;
 882     blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks);
 883
 884     s->error=0;
 885
 886     if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
 887         ape_unpack_mono(s, blockstodecode);
 888     else
 889         ape_unpack_stereo(s, blockstodecode);
 890
 891     if(s->error || s->ptr > s->data_end){
 892         s->samples=0;
 893         av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
 894         return -1;
 895     }
 896
 897     for (i = 0; i < blockstodecode; i++) {
 898         *samples++ = s->decoded0[i];
 899         if(s->channels == 2)
 900             *samples++ = s->decoded1[i];
 901     }
 902
 903     s->samples -= blockstodecode;
 904
 905     *data_size = blockstodecode * 2 * s->channels;
 906     bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size;
 907     s->last_ptr = s->ptr;
 908     return bytes_used;
 909 }
 910
 911 AVCodec ape_decoder = {
 912     "ape",
 913     CODEC_TYPE_AUDIO,
 914     CODEC_ID_APE,
 915     sizeof(APEContext),
 916     ape_decode_init,
 917     NULL,
 918     ape_decode_close,
 919     ape_decode_frame,
 920     .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
 921 };