libavcodec/alac.c

   1 /*
   2  * ALAC (Apple Lossless Audio Codec) decoder
   3  * Copyright (c) 2005 David Hammerton
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file libavcodec/alac.c
  24  * ALAC (Apple Lossless Audio Codec) decoder
  25  * @author 2005 David Hammerton
  26  *
  27  * For more information on the ALAC format, visit:
  28  *  http://crazney.net/programs/itunes/alac.html
  29  *
  30  * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
  31  * passed through the extradata[_size] fields. This atom is tacked onto
  32  * the end of an 'alac' stsd atom and has the following format:
  33  *  bytes 0-3   atom size (0x24), big-endian
  34  *  bytes 4-7   atom type ('alac', not the 'alac' tag from start of stsd)
  35  *  bytes 8-35  data bytes needed by decoder
  36  *
  37  * Extradata:
  38  * 32bit  size
  39  * 32bit  tag (=alac)
  40  * 32bit  zero?
  41  * 32bit  max sample per frame
  42  *  8bit  ?? (zero?)
  43  *  8bit  sample size
  44  *  8bit  history mult
  45  *  8bit  initial history
  46  *  8bit  kmodifier
  47  *  8bit  channels?
  48  * 16bit  ??
  49  * 32bit  max coded frame size
  50  * 32bit  bitrate?
  51  * 32bit  samplerate
  52  */
  53
  54
  55 #include "avcodec.h"
  56 #include "bitstream.h"
  57 #include "bytestream.h"
  58 #include "unary.h"
  59 #include "mathops.h"
  60
  61 #define ALAC_EXTRADATA_SIZE 36
  62 #define MAX_CHANNELS 2
  63
  64 typedef struct {
  65
  66     AVCodecContext *avctx;
  67     GetBitContext gb;
  68     /* init to 0; first frame decode should initialize from extradata and
  69      * set this to 1 */
  70     int context_initialized;
  71
  72     int numchannels;
  73     int bytespersample;
  74
  75     /* buffers */
  76     int32_t *predicterror_buffer[MAX_CHANNELS];
  77
  78     int32_t *outputsamples_buffer[MAX_CHANNELS];
  79
  80     /* stuff from setinfo */
  81     uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */    /* max samples per frame? */
  82     uint8_t setinfo_sample_size; /* 0x10 */
  83     uint8_t setinfo_rice_historymult; /* 0x28 */
  84     uint8_t setinfo_rice_initialhistory; /* 0x0a */
  85     uint8_t setinfo_rice_kmodifier; /* 0x0e */
  86     /* end setinfo stuff */
  87
  88 } ALACContext;
  89
  90 static void allocate_buffers(ALACContext *alac)
  91 {
  92     int chan;
  93     for (chan = 0; chan < MAX_CHANNELS; chan++) {
  94         alac->predicterror_buffer[chan] =
  95             av_malloc(alac->setinfo_max_samples_per_frame * 4);
  96
  97         alac->outputsamples_buffer[chan] =
  98             av_malloc(alac->setinfo_max_samples_per_frame * 4);
  99     }
 100 }
 101
 102 static int alac_set_info(ALACContext *alac)
 103 {
 104     const unsigned char *ptr = alac->avctx->extradata;
 105
 106     ptr += 4; /* size */
 107     ptr += 4; /* alac */
 108     ptr += 4; /* 0 ? */
 109
 110     if(AV_RB32(ptr) >= UINT_MAX/4){
 111         av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
 112         return -1;
 113     }
 114
 115     /* buffer size / 2 ? */
 116     alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
 117     ptr++;                          /* ??? */
 118     alac->setinfo_sample_size           = *ptr++;
 119     if (alac->setinfo_sample_size > 32) {
 120         av_log(alac->avctx, AV_LOG_ERROR, "setinfo_sample_size too large\n");
 121         return -1;
 122     }
 123     alac->setinfo_rice_historymult      = *ptr++;
 124     alac->setinfo_rice_initialhistory   = *ptr++;
 125     alac->setinfo_rice_kmodifier        = *ptr++;
 126     ptr++;                         /* channels? */
 127     bytestream_get_be16(&ptr);      /* ??? */
 128     bytestream_get_be32(&ptr);      /* max coded frame size */
 129     bytestream_get_be32(&ptr);      /* bitrate ? */
 130     bytestream_get_be32(&ptr);      /* samplerate */
 131
 132     allocate_buffers(alac);
 133
 134     return 0;
 135 }
 136
 137 static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){
 138     /* read x - number of 1s before 0 represent the rice */
 139     int x = get_unary_0_9(gb);
 140
 141     if (x > 8) { /* RICE THRESHOLD */
 142         /* use alternative encoding */
 143         x = get_bits(gb, readsamplesize);
 144     } else {
 145         if (k >= limit)
 146             k = limit;
 147
 148         if (k != 1) {
 149             int extrabits = show_bits(gb, k);
 150
 151             /* multiply x by 2^k - 1, as part of their strange algorithm */
 152             x = (x << k) - x;
 153
 154             if (extrabits > 1) {
 155                 x += extrabits - 1;
 156                 skip_bits(gb, k);
 157             } else
 158                 skip_bits(gb, k - 1);
 159         }
 160     }
 161     return x;
 162 }
 163
 164 static void bastardized_rice_decompress(ALACContext *alac,
 165                                  int32_t *output_buffer,
 166                                  int output_size,
 167                                  int readsamplesize, /* arg_10 */
 168                                  int rice_initialhistory, /* arg424->b */
 169                                  int rice_kmodifier, /* arg424->d */
 170                                  int rice_historymult, /* arg424->c */
 171                                  int rice_kmodifier_mask /* arg424->e */
 172         )
 173 {
 174     int output_count;
 175     unsigned int history = rice_initialhistory;
 176     int sign_modifier = 0;
 177
 178     for (output_count = 0; output_count < output_size; output_count++) {
 179         int32_t x;
 180         int32_t x_modified;
 181         int32_t final_val;
 182
 183         /* standard rice encoding */
 184         int k; /* size of extra bits */
 185
 186         /* read k, that is bits as is */
 187         k = av_log2((history >> 9) + 3);
 188         x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize);
 189
 190         x_modified = sign_modifier + x;
 191         final_val = (x_modified + 1) / 2;
 192         if (x_modified & 1) final_val *= -1;
 193
 194         output_buffer[output_count] = final_val;
 195
 196         sign_modifier = 0;
 197
 198         /* now update the history */
 199         history += x_modified * rice_historymult
 200                    - ((history * rice_historymult) >> 9);
 201
 202         if (x_modified > 0xffff)
 203             history = 0xffff;
 204
 205         /* special case: there may be compressed blocks of 0 */
 206         if ((history < 128) && (output_count+1 < output_size)) {
 207             int k;
 208             unsigned int block_size;
 209
 210             sign_modifier = 1;
 211
 212             k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
 213
 214             block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16);
 215
 216             if (block_size > 0) {
 217                 if(block_size >= output_size - output_count){
 218                     av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count);
 219                     block_size= output_size - output_count - 1;
 220                 }
 221                 memset(&output_buffer[output_count+1], 0, block_size * 4);
 222                 output_count += block_size;
 223             }
 224
 225             if (block_size > 0xffff)
 226                 sign_modifier = 0;
 227
 228             history = 0;
 229         }
 230     }
 231 }
 232
 233 static inline int sign_only(int v)
 234 {
 235     return v ? FFSIGN(v) : 0;
 236 }
 237
 238 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 239                                            int32_t *buffer_out,
 240                                            int output_size,
 241                                            int readsamplesize,
 242                                            int16_t *predictor_coef_table,
 243                                            int predictor_coef_num,
 244                                            int predictor_quantitization)
 245 {
 246     int i;
 247
 248     /* first sample always copies */
 249     *buffer_out = *error_buffer;
 250
 251     if (!predictor_coef_num) {
 252         if (output_size <= 1)
 253             return;
 254
 255         memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
 256         return;
 257     }
 258
 259     if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
 260       /* second-best case scenario for fir decompression,
 261        * error describes a small difference from the previous sample only
 262        */
 263         if (output_size <= 1)
 264             return;
 265         for (i = 0; i < output_size - 1; i++) {
 266             int32_t prev_value;
 267             int32_t error_value;
 268
 269             prev_value = buffer_out[i];
 270             error_value = error_buffer[i+1];
 271             buffer_out[i+1] =
 272                 sign_extend((prev_value + error_value), readsamplesize);
 273         }
 274         return;
 275     }
 276
 277     /* read warm-up samples */
 278     if (predictor_coef_num > 0)
 279         for (i = 0; i < predictor_coef_num; i++) {
 280             int32_t val;
 281
 282             val = buffer_out[i] + error_buffer[i+1];
 283             val = sign_extend(val, readsamplesize);
 284             buffer_out[i+1] = val;
 285         }
 286
 287 #if 0
 288     /* 4 and 8 are very common cases (the only ones i've seen). these
 289      * should be unrolled and optimized
 290      */
 291     if (predictor_coef_num == 4) {
 292         /* FIXME: optimized general case */
 293         return;
 294     }
 295
 296     if (predictor_coef_table == 8) {
 297         /* FIXME: optimized general case */
 298         return;
 299     }
 300 #endif
 301
 302     /* general case */
 303     if (predictor_coef_num > 0) {
 304         for (i = predictor_coef_num + 1; i < output_size; i++) {
 305             int j;
 306             int sum = 0;
 307             int outval;
 308             int error_val = error_buffer[i];
 309
 310             for (j = 0; j < predictor_coef_num; j++) {
 311                 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
 312                        predictor_coef_table[j];
 313             }
 314
 315             outval = (1 << (predictor_quantitization-1)) + sum;
 316             outval = outval >> predictor_quantitization;
 317             outval = outval + buffer_out[0] + error_val;
 318             outval = sign_extend(outval, readsamplesize);
 319
 320             buffer_out[predictor_coef_num+1] = outval;
 321
 322             if (error_val > 0) {
 323                 int predictor_num = predictor_coef_num - 1;
 324
 325                 while (predictor_num >= 0 && error_val > 0) {
 326                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
 327                     int sign = sign_only(val);
 328
 329                     predictor_coef_table[predictor_num] -= sign;
 330
 331                     val *= sign; /* absolute value */
 332
 333                     error_val -= ((val >> predictor_quantitization) *
 334                                   (predictor_coef_num - predictor_num));
 335
 336                     predictor_num--;
 337                 }
 338             } else if (error_val < 0) {
 339                 int predictor_num = predictor_coef_num - 1;
 340
 341                 while (predictor_num >= 0 && error_val < 0) {
 342                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
 343                     int sign = - sign_only(val);
 344
 345                     predictor_coef_table[predictor_num] -= sign;
 346
 347                     val *= sign; /* neg value */
 348
 349                     error_val -= ((val >> predictor_quantitization) *
 350                                   (predictor_coef_num - predictor_num));
 351
 352                     predictor_num--;
 353                 }
 354             }
 355
 356             buffer_out++;
 357         }
 358     }
 359 }
 360
 361 static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
 362                                   int16_t *buffer_out,
 363                                   int numchannels, int numsamples,
 364                                   uint8_t interlacing_shift,
 365                                   uint8_t interlacing_leftweight)
 366 {
 367     int i;
 368     if (numsamples <= 0)
 369         return;
 370
 371     /* weighted interlacing */
 372     if (interlacing_leftweight) {
 373         for (i = 0; i < numsamples; i++) {
 374             int32_t a, b;
 375
 376             a = buffer[0][i];
 377             b = buffer[1][i];
 378
 379             a -= (b * interlacing_leftweight) >> interlacing_shift;
 380             b += a;
 381
 382             buffer_out[i*numchannels] = b;
 383             buffer_out[i*numchannels + 1] = a;
 384         }
 385
 386         return;
 387     }
 388
 389     /* otherwise basic interlacing took place */
 390     for (i = 0; i < numsamples; i++) {
 391         int16_t left, right;
 392
 393         left = buffer[0][i];
 394         right = buffer[1][i];
 395
 396         buffer_out[i*numchannels] = left;
 397         buffer_out[i*numchannels + 1] = right;
 398     }
 399 }
 400
 401 static int alac_decode_frame(AVCodecContext *avctx,
 402                              void *outbuffer, int *outputsize,
 403                              const uint8_t *inbuffer, int input_buffer_size)
 404 {
 405     ALACContext *alac = avctx->priv_data;
 406
 407     int channels;
 408     unsigned int outputsamples;
 409     int hassize;
 410     unsigned int readsamplesize;
 411     int wasted_bytes;
 412     int isnotcompressed;
 413     uint8_t interlacing_shift;
 414     uint8_t interlacing_leftweight;
 415
 416     /* short-circuit null buffers */
 417     if (!inbuffer || !input_buffer_size)
 418         return input_buffer_size;
 419
 420     /* initialize from the extradata */
 421     if (!alac->context_initialized) {
 422         if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
 423             av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
 424                 ALAC_EXTRADATA_SIZE);
 425             return input_buffer_size;
 426         }
 427         if (alac_set_info(alac)) {
 428             av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
 429             return input_buffer_size;
 430         }
 431         alac->context_initialized = 1;
 432     }
 433
 434     init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
 435
 436     channels = get_bits(&alac->gb, 3) + 1;
 437     if (channels > MAX_CHANNELS) {
 438         av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
 439                MAX_CHANNELS);
 440         return input_buffer_size;
 441     }
 442
 443     /* 2^result = something to do with output waiting.
 444      * perhaps matters if we read > 1 frame in a pass?
 445      */
 446     skip_bits(&alac->gb, 4);
 447
 448     skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
 449
 450     /* the output sample size is stored soon */
 451     hassize = get_bits1(&alac->gb);
 452
 453     wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
 454
 455     /* whether the frame is compressed */
 456     isnotcompressed = get_bits1(&alac->gb);
 457
 458     if (hassize) {
 459         /* now read the number of samples as a 32bit integer */
 460         outputsamples = get_bits_long(&alac->gb, 32);
 461         if(outputsamples > alac->setinfo_max_samples_per_frame){
 462             av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame);
 463             return -1;
 464         }
 465     } else
 466         outputsamples = alac->setinfo_max_samples_per_frame;
 467
 468     if(outputsamples > *outputsize / alac->bytespersample){
 469         av_log(avctx, AV_LOG_ERROR, "sample buffer too small\n");
 470         return -1;
 471     }
 472
 473     *outputsize = outputsamples * alac->bytespersample;
 474     readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
 475     if (readsamplesize > MIN_CACHE_BITS) {
 476         av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize);
 477         return -1;
 478     }
 479
 480     if (!isnotcompressed) {
 481         /* so it is compressed */
 482         int16_t predictor_coef_table[channels][32];
 483         int predictor_coef_num[channels];
 484         int prediction_type[channels];
 485         int prediction_quantitization[channels];
 486         int ricemodifier[channels];
 487         int i, chan;
 488
 489         interlacing_shift = get_bits(&alac->gb, 8);
 490         interlacing_leftweight = get_bits(&alac->gb, 8);
 491
 492         for (chan = 0; chan < channels; chan++) {
 493             prediction_type[chan] = get_bits(&alac->gb, 4);
 494             prediction_quantitization[chan] = get_bits(&alac->gb, 4);
 495
 496             ricemodifier[chan] = get_bits(&alac->gb, 3);
 497             predictor_coef_num[chan] = get_bits(&alac->gb, 5);
 498
 499             /* read the predictor table */
 500             for (i = 0; i < predictor_coef_num[chan]; i++)
 501                 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
 502         }
 503
 504         if (wasted_bytes)
 505             av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
 506
 507         for (chan = 0; chan < channels; chan++) {
 508             bastardized_rice_decompress(alac,
 509                                         alac->predicterror_buffer[chan],
 510                                         outputsamples,
 511                                         readsamplesize,
 512                                         alac->setinfo_rice_initialhistory,
 513                                         alac->setinfo_rice_kmodifier,
 514                                         ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
 515                                         (1 << alac->setinfo_rice_kmodifier) - 1);
 516
 517             if (prediction_type[chan] == 0) {
 518                 /* adaptive fir */
 519                 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
 520                                                alac->outputsamples_buffer[chan],
 521                                                outputsamples,
 522                                                readsamplesize,
 523                                                predictor_coef_table[chan],
 524                                                predictor_coef_num[chan],
 525                                                prediction_quantitization[chan]);
 526             } else {
 527                 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
 528                 /* I think the only other prediction type (or perhaps this is
 529                  * just a boolean?) runs adaptive fir twice.. like:
 530                  * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
 531                  * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
 532                  * little strange..
 533                  */
 534             }
 535         }
 536     } else {
 537         /* not compressed, easy case */
 538         int i, chan;
 539         for (i = 0; i < outputsamples; i++)
 540             for (chan = 0; chan < channels; chan++) {
 541                 int32_t audiobits;
 542
 543                 audiobits = get_sbits_long(&alac->gb, alac->setinfo_sample_size);
 544
 545                 alac->outputsamples_buffer[chan][i] = audiobits;
 546             }
 547         /* wasted_bytes = 0; */
 548         interlacing_shift = 0;
 549         interlacing_leftweight = 0;
 550     }
 551     if (get_bits(&alac->gb, 3) != 7)
 552         av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n");
 553
 554     switch(alac->setinfo_sample_size) {
 555     case 16:
 556         if (channels == 2) {
 557             reconstruct_stereo_16(alac->outputsamples_buffer,
 558                                   (int16_t*)outbuffer,
 559                                   alac->numchannels,
 560                                   outputsamples,
 561                                   interlacing_shift,
 562                                   interlacing_leftweight);
 563         } else {
 564             int i;
 565             for (i = 0; i < outputsamples; i++) {
 566                 int16_t sample = alac->outputsamples_buffer[0][i];
 567                 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
 568             }
 569         }
 570         break;
 571     case 20:
 572     case 24:
 573         // It is not clear if there exist any encoder that creates 24 bit ALAC
 574         // files. iTunes convert 24 bit raw files to 16 bit before encoding.
 575     case 32:
 576         av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
 577         break;
 578     default:
 579         break;
 580     }
 581
 582     if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8)
 583         av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb));
 584
 585     return input_buffer_size;
 586 }
 587
 588 static av_cold int alac_decode_init(AVCodecContext * avctx)
 589 {
 590     ALACContext *alac = avctx->priv_data;
 591     alac->avctx = avctx;
 592     alac->context_initialized = 0;
 593
 594     alac->numchannels = alac->avctx->channels;
 595     alac->bytespersample = 2 * alac->numchannels;
 596     avctx->sample_fmt = SAMPLE_FMT_S16;
 597
 598     return 0;
 599 }
 600
 601 static av_cold int alac_decode_close(AVCodecContext *avctx)
 602 {
 603     ALACContext *alac = avctx->priv_data;
 604
 605     int chan;
 606     for (chan = 0; chan < MAX_CHANNELS; chan++) {
 607         av_free(alac->predicterror_buffer[chan]);
 608         av_free(alac->outputsamples_buffer[chan]);
 609     }
 610
 611     return 0;
 612 }
 613
 614 AVCodec alac_decoder = {
 615     "alac",
 616     CODEC_TYPE_AUDIO,
 617     CODEC_ID_ALAC,
 618     sizeof(ALACContext),
 619     alac_decode_init,
 620     NULL,
 621     alac_decode_close,
 622     alac_decode_frame,
 623     .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 624 };