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 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
  60 #define ALAC_EXTRADATA_SIZE 36
  61 #define MAX_CHANNELS 2
  62
  63 typedef struct {
  64
  65     AVCodecContext *avctx;
  66     GetBitContext gb;
  67     /* init to 0; first frame decode should initialize from extradata and
  68      * set this to 1 */
  69     int context_initialized;
  70
  71     int numchannels;
  72     int bytespersample;
  73
  74     /* buffers */
  75     int32_t *predicterror_buffer[MAX_CHANNELS];
  76
  77     int32_t *outputsamples_buffer[MAX_CHANNELS];
  78
  79     /* stuff from setinfo */
  80     uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */    /* max samples per frame? */
  81     uint8_t setinfo_sample_size; /* 0x10 */
  82     uint8_t setinfo_rice_historymult; /* 0x28 */
  83     uint8_t setinfo_rice_initialhistory; /* 0x0a */
  84     uint8_t setinfo_rice_kmodifier; /* 0x0e */
  85     /* end setinfo stuff */
  86
  87 } ALACContext;
  88
  89 static void allocate_buffers(ALACContext *alac)
  90 {
  91     int chan;
  92     for (chan = 0; chan < MAX_CHANNELS; chan++) {
  93         alac->predicterror_buffer[chan] =
  94             av_malloc(alac->setinfo_max_samples_per_frame * 4);
  95
  96         alac->outputsamples_buffer[chan] =
  97             av_malloc(alac->setinfo_max_samples_per_frame * 4);
  98     }
  99 }
 100
 101 static int alac_set_info(ALACContext *alac)
 102 {
 103     const unsigned char *ptr = alac->avctx->extradata;
 104
 105     ptr += 4; /* size */
 106     ptr += 4; /* alac */
 107     ptr += 4; /* 0 ? */
 108
 109     if(AV_RB32(ptr) >= UINT_MAX/4){
 110         av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
 111         return -1;
 112     }
 113
 114     /* buffer size / 2 ? */
 115     alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
 116     ptr++;                          /* ??? */
 117     alac->setinfo_sample_size           = *ptr++;
 118     if (alac->setinfo_sample_size > 32) {
 119         av_log(alac->avctx, AV_LOG_ERROR, "setinfo_sample_size too large\n");
 120         return -1;
 121     }
 122     alac->setinfo_rice_historymult      = *ptr++;
 123     alac->setinfo_rice_initialhistory   = *ptr++;
 124     alac->setinfo_rice_kmodifier        = *ptr++;
 125     ptr++;                         /* channels? */
 126     bytestream_get_be16(&ptr);      /* ??? */
 127     bytestream_get_be32(&ptr);      /* max coded frame size */
 128     bytestream_get_be32(&ptr);      /* bitrate ? */
 129     bytestream_get_be32(&ptr);      /* samplerate */
 130
 131     allocate_buffers(alac);
 132
 133     return 0;
 134 }
 135
 136 static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){
 137     /* read x - number of 1s before 0 represent the rice */
 138     int x = get_unary_0_9(gb);
 139
 140     if (x > 8) { /* RICE THRESHOLD */
 141         /* use alternative encoding */
 142         x = get_bits(gb, readsamplesize);
 143     } else {
 144         if (k >= limit)
 145             k = limit;
 146
 147         if (k != 1) {
 148             int extrabits = show_bits(gb, k);
 149
 150             /* multiply x by 2^k - 1, as part of their strange algorithm */
 151             x = (x << k) - x;
 152
 153             if (extrabits > 1) {
 154                 x += extrabits - 1;
 155                 skip_bits(gb, k);
 156             } else
 157                 skip_bits(gb, k - 1);
 158         }
 159     }
 160     return x;
 161 }
 162
 163 static void bastardized_rice_decompress(ALACContext *alac,
 164                                  int32_t *output_buffer,
 165                                  int output_size,
 166                                  int readsamplesize, /* arg_10 */
 167                                  int rice_initialhistory, /* arg424->b */
 168                                  int rice_kmodifier, /* arg424->d */
 169                                  int rice_historymult, /* arg424->c */
 170                                  int rice_kmodifier_mask /* arg424->e */
 171         )
 172 {
 173     int output_count;
 174     unsigned int history = rice_initialhistory;
 175     int sign_modifier = 0;
 176
 177     for (output_count = 0; output_count < output_size; output_count++) {
 178         int32_t x;
 179         int32_t x_modified;
 180         int32_t final_val;
 181
 182         /* standard rice encoding */
 183         int k; /* size of extra bits */
 184
 185         /* read k, that is bits as is */
 186         k = av_log2((history >> 9) + 3);
 187         x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize);
 188
 189         x_modified = sign_modifier + x;
 190         final_val = (x_modified + 1) / 2;
 191         if (x_modified & 1) final_val *= -1;
 192
 193         output_buffer[output_count] = final_val;
 194
 195         sign_modifier = 0;
 196
 197         /* now update the history */
 198         history += x_modified * rice_historymult
 199                    - ((history * rice_historymult) >> 9);
 200
 201         if (x_modified > 0xffff)
 202             history = 0xffff;
 203
 204         /* special case: there may be compressed blocks of 0 */
 205         if ((history < 128) && (output_count+1 < output_size)) {
 206             int k;
 207             unsigned int block_size;
 208
 209             sign_modifier = 1;
 210
 211             k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
 212
 213             block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16);
 214
 215             if (block_size > 0) {
 216                 if(block_size >= output_size - output_count){
 217                     av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count);
 218                     block_size= output_size - output_count - 1;
 219                 }
 220                 memset(&output_buffer[output_count+1], 0, block_size * 4);
 221                 output_count += block_size;
 222             }
 223
 224             if (block_size > 0xffff)
 225                 sign_modifier = 0;
 226
 227             history = 0;
 228         }
 229     }
 230 }
 231
 232 static inline int32_t extend_sign32(int32_t val, int bits)
 233 {
 234     return (val << (32 - bits)) >> (32 - bits);
 235 }
 236
 237 static inline int sign_only(int v)
 238 {
 239     return v ? FFSIGN(v) : 0;
 240 }
 241
 242 static void predictor_decompress_fir_adapt(int32_t *error_buffer,
 243                                            int32_t *buffer_out,
 244                                            int output_size,
 245                                            int readsamplesize,
 246                                            int16_t *predictor_coef_table,
 247                                            int predictor_coef_num,
 248                                            int predictor_quantitization)
 249 {
 250     int i;
 251
 252     /* first sample always copies */
 253     *buffer_out = *error_buffer;
 254
 255     if (!predictor_coef_num) {
 256         if (output_size <= 1)
 257             return;
 258
 259         memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
 260         return;
 261     }
 262
 263     if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
 264       /* second-best case scenario for fir decompression,
 265        * error describes a small difference from the previous sample only
 266        */
 267         if (output_size <= 1)
 268             return;
 269         for (i = 0; i < output_size - 1; i++) {
 270             int32_t prev_value;
 271             int32_t error_value;
 272
 273             prev_value = buffer_out[i];
 274             error_value = error_buffer[i+1];
 275             buffer_out[i+1] =
 276                 extend_sign32((prev_value + error_value), readsamplesize);
 277         }
 278         return;
 279     }
 280
 281     /* read warm-up samples */
 282     if (predictor_coef_num > 0)
 283         for (i = 0; i < predictor_coef_num; i++) {
 284             int32_t val;
 285
 286             val = buffer_out[i] + error_buffer[i+1];
 287             val = extend_sign32(val, readsamplesize);
 288             buffer_out[i+1] = val;
 289         }
 290
 291 #if 0
 292     /* 4 and 8 are very common cases (the only ones i've seen). these
 293      * should be unrolled and optimized
 294      */
 295     if (predictor_coef_num == 4) {
 296         /* FIXME: optimized general case */
 297         return;
 298     }
 299
 300     if (predictor_coef_table == 8) {
 301         /* FIXME: optimized general case */
 302         return;
 303     }
 304 #endif
 305
 306     /* general case */
 307     if (predictor_coef_num > 0) {
 308         for (i = predictor_coef_num + 1; i < output_size; i++) {
 309             int j;
 310             int sum = 0;
 311             int outval;
 312             int error_val = error_buffer[i];
 313
 314             for (j = 0; j < predictor_coef_num; j++) {
 315                 sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
 316                        predictor_coef_table[j];
 317             }
 318
 319             outval = (1 << (predictor_quantitization-1)) + sum;
 320             outval = outval >> predictor_quantitization;
 321             outval = outval + buffer_out[0] + error_val;
 322             outval = extend_sign32(outval, readsamplesize);
 323
 324             buffer_out[predictor_coef_num+1] = outval;
 325
 326             if (error_val > 0) {
 327                 int predictor_num = predictor_coef_num - 1;
 328
 329                 while (predictor_num >= 0 && error_val > 0) {
 330                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
 331                     int sign = sign_only(val);
 332
 333                     predictor_coef_table[predictor_num] -= sign;
 334
 335                     val *= sign; /* absolute value */
 336
 337                     error_val -= ((val >> predictor_quantitization) *
 338                                   (predictor_coef_num - predictor_num));
 339
 340                     predictor_num--;
 341                 }
 342             } else if (error_val < 0) {
 343                 int predictor_num = predictor_coef_num - 1;
 344
 345                 while (predictor_num >= 0 && error_val < 0) {
 346                     int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
 347                     int sign = - sign_only(val);
 348
 349                     predictor_coef_table[predictor_num] -= sign;
 350
 351                     val *= sign; /* neg value */
 352
 353                     error_val -= ((val >> predictor_quantitization) *
 354                                   (predictor_coef_num - predictor_num));
 355
 356                     predictor_num--;
 357                 }
 358             }
 359
 360             buffer_out++;
 361         }
 362     }
 363 }
 364
 365 static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
 366                                   int16_t *buffer_out,
 367                                   int numchannels, int numsamples,
 368                                   uint8_t interlacing_shift,
 369                                   uint8_t interlacing_leftweight)
 370 {
 371     int i;
 372     if (numsamples <= 0)
 373         return;
 374
 375     /* weighted interlacing */
 376     if (interlacing_leftweight) {
 377         for (i = 0; i < numsamples; i++) {
 378             int32_t a, b;
 379
 380             a = buffer[0][i];
 381             b = buffer[1][i];
 382
 383             a -= (b * interlacing_leftweight) >> interlacing_shift;
 384             b += a;
 385
 386             buffer_out[i*numchannels] = b;
 387             buffer_out[i*numchannels + 1] = a;
 388         }
 389
 390         return;
 391     }
 392
 393     /* otherwise basic interlacing took place */
 394     for (i = 0; i < numsamples; i++) {
 395         int16_t left, right;
 396
 397         left = buffer[0][i];
 398         right = buffer[1][i];
 399
 400         buffer_out[i*numchannels] = left;
 401         buffer_out[i*numchannels + 1] = right;
 402     }
 403 }
 404
 405 static int alac_decode_frame(AVCodecContext *avctx,
 406                              void *outbuffer, int *outputsize,
 407                              const uint8_t *inbuffer, int input_buffer_size)
 408 {
 409     ALACContext *alac = avctx->priv_data;
 410
 411     int channels;
 412     unsigned int outputsamples;
 413     int hassize;
 414     unsigned int readsamplesize;
 415     int wasted_bytes;
 416     int isnotcompressed;
 417     uint8_t interlacing_shift;
 418     uint8_t interlacing_leftweight;
 419
 420     /* short-circuit null buffers */
 421     if (!inbuffer || !input_buffer_size)
 422         return input_buffer_size;
 423
 424     /* initialize from the extradata */
 425     if (!alac->context_initialized) {
 426         if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
 427             av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
 428                 ALAC_EXTRADATA_SIZE);
 429             return input_buffer_size;
 430         }
 431         if (alac_set_info(alac)) {
 432             av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
 433             return input_buffer_size;
 434         }
 435         alac->context_initialized = 1;
 436     }
 437
 438     init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
 439
 440     channels = get_bits(&alac->gb, 3) + 1;
 441     if (channels > MAX_CHANNELS) {
 442         av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
 443                MAX_CHANNELS);
 444         return input_buffer_size;
 445     }
 446
 447     /* 2^result = something to do with output waiting.
 448      * perhaps matters if we read > 1 frame in a pass?
 449      */
 450     skip_bits(&alac->gb, 4);
 451
 452     skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
 453
 454     /* the output sample size is stored soon */
 455     hassize = get_bits1(&alac->gb);
 456
 457     wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
 458
 459     /* whether the frame is compressed */
 460     isnotcompressed = get_bits1(&alac->gb);
 461
 462     if (hassize) {
 463         /* now read the number of samples as a 32bit integer */
 464         outputsamples = get_bits_long(&alac->gb, 32);
 465         if(outputsamples > alac->setinfo_max_samples_per_frame){
 466             av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame);
 467             return -1;
 468         }
 469     } else
 470         outputsamples = alac->setinfo_max_samples_per_frame;
 471
 472     if(outputsamples > *outputsize / alac->bytespersample){
 473         av_log(avctx, AV_LOG_ERROR, "sample buffer too small\n");
 474         return -1;
 475     }
 476
 477     *outputsize = outputsamples * alac->bytespersample;
 478     readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
 479     if (readsamplesize > MIN_CACHE_BITS) {
 480         av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize);
 481         return -1;
 482     }
 483
 484     if (!isnotcompressed) {
 485         /* so it is compressed */
 486         int16_t predictor_coef_table[channels][32];
 487         int predictor_coef_num[channels];
 488         int prediction_type[channels];
 489         int prediction_quantitization[channels];
 490         int ricemodifier[channels];
 491         int i, chan;
 492
 493         interlacing_shift = get_bits(&alac->gb, 8);
 494         interlacing_leftweight = get_bits(&alac->gb, 8);
 495
 496         for (chan = 0; chan < channels; chan++) {
 497             prediction_type[chan] = get_bits(&alac->gb, 4);
 498             prediction_quantitization[chan] = get_bits(&alac->gb, 4);
 499
 500             ricemodifier[chan] = get_bits(&alac->gb, 3);
 501             predictor_coef_num[chan] = get_bits(&alac->gb, 5);
 502
 503             /* read the predictor table */
 504             for (i = 0; i < predictor_coef_num[chan]; i++)
 505                 predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
 506         }
 507
 508         if (wasted_bytes)
 509             av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
 510
 511         for (chan = 0; chan < channels; chan++) {
 512             bastardized_rice_decompress(alac,
 513                                         alac->predicterror_buffer[chan],
 514                                         outputsamples,
 515                                         readsamplesize,
 516                                         alac->setinfo_rice_initialhistory,
 517                                         alac->setinfo_rice_kmodifier,
 518                                         ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
 519                                         (1 << alac->setinfo_rice_kmodifier) - 1);
 520
 521             if (prediction_type[chan] == 0) {
 522                 /* adaptive fir */
 523                 predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
 524                                                alac->outputsamples_buffer[chan],
 525                                                outputsamples,
 526                                                readsamplesize,
 527                                                predictor_coef_table[chan],
 528                                                predictor_coef_num[chan],
 529                                                prediction_quantitization[chan]);
 530             } else {
 531                 av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
 532                 /* I think the only other prediction type (or perhaps this is
 533                  * just a boolean?) runs adaptive fir twice.. like:
 534                  * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
 535                  * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
 536                  * little strange..
 537                  */
 538             }
 539         }
 540     } else {
 541         /* not compressed, easy case */
 542         int i, chan;
 543         for (i = 0; i < outputsamples; i++)
 544             for (chan = 0; chan < channels; chan++) {
 545                 int32_t audiobits;
 546
 547                 audiobits = get_bits_long(&alac->gb, alac->setinfo_sample_size);
 548                 audiobits = extend_sign32(audiobits, alac->setinfo_sample_size);
 549
 550                 alac->outputsamples_buffer[chan][i] = audiobits;
 551             }
 552         /* wasted_bytes = 0; */
 553         interlacing_shift = 0;
 554         interlacing_leftweight = 0;
 555     }
 556     if (get_bits(&alac->gb, 3) != 7)
 557         av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n");
 558
 559     switch(alac->setinfo_sample_size) {
 560     case 16:
 561         if (channels == 2) {
 562             reconstruct_stereo_16(alac->outputsamples_buffer,
 563                                   (int16_t*)outbuffer,
 564                                   alac->numchannels,
 565                                   outputsamples,
 566                                   interlacing_shift,
 567                                   interlacing_leftweight);
 568         } else {
 569             int i;
 570             for (i = 0; i < outputsamples; i++) {
 571                 int16_t sample = alac->outputsamples_buffer[0][i];
 572                 ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
 573             }
 574         }
 575         break;
 576     case 20:
 577     case 24:
 578         // It is not clear if there exist any encoder that creates 24 bit ALAC
 579         // files. iTunes convert 24 bit raw files to 16 bit before encoding.
 580     case 32:
 581         av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
 582         break;
 583     default:
 584         break;
 585     }
 586
 587     if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8)
 588         av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb));
 589
 590     return input_buffer_size;
 591 }
 592
 593 static av_cold int alac_decode_init(AVCodecContext * avctx)
 594 {
 595     ALACContext *alac = avctx->priv_data;
 596     alac->avctx = avctx;
 597     alac->context_initialized = 0;
 598
 599     alac->numchannels = alac->avctx->channels;
 600     alac->bytespersample = (avctx->bits_per_sample / 8) * alac->numchannels;
 601     avctx->sample_fmt = SAMPLE_FMT_S16;
 602
 603     return 0;
 604 }
 605
 606 static av_cold int alac_decode_close(AVCodecContext *avctx)
 607 {
 608     ALACContext *alac = avctx->priv_data;
 609
 610     int chan;
 611     for (chan = 0; chan < MAX_CHANNELS; chan++) {
 612         av_free(alac->predicterror_buffer[chan]);
 613         av_free(alac->outputsamples_buffer[chan]);
 614     }
 615
 616     return 0;
 617 }
 618
 619 AVCodec alac_decoder = {
 620     "alac",
 621     CODEC_TYPE_AUDIO,
 622     CODEC_ID_ALAC,
 623     sizeof(ALACContext),
 624     alac_decode_init,
 625     NULL,
 626     alac_decode_close,
 627     alac_decode_frame,
 628     .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 629 };