git.sesse.net Git - ffmpeg/blob - libavcodec/shorten.c

   1 /*
   2  * Shorten decoder
   3  * Copyright (c) 2005 Jeff Muizelaar
   4  *
   5  * This library is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU Lesser General Public
   7  * License as published by the Free Software Foundation; either
   8  * version 2 of the License, or (at your option) any later version.
   9  *
  10  * This library is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * Lesser General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU Lesser General Public
  16  * License along with this library; if not, write to the Free Software
  17  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  18  */
  19
  20 /**
  21  * @file shorten.c
  22  * Shorten decoder
  23  * @author Jeff Muizelaar
  24  *
  25  */
  26
  27 #define DEBUG
  28 #include <limits.h>
  29 #include "avcodec.h"
  30 #include "bitstream.h"
  31 #include "golomb.h"
  32
  33 #define MAX_CHANNELS 8
  34 #define MAX_BLOCKSIZE 65535
  35
  36 #define OUT_BUFFER_SIZE 16384
  37
  38 #define ULONGSIZE 2
  39
  40 #define WAVE_FORMAT_PCM 0x0001
  41
  42 #define DEFAULT_BLOCK_SIZE 256
  43
  44 #define TYPESIZE 4
  45 #define CHANSIZE 0
  46 #define LPCQSIZE 2
  47 #define ENERGYSIZE 3
  48 #define BITSHIFTSIZE 2
  49
  50 #define TYPE_S16HL 3
  51 #define TYPE_S16LH 5
  52
  53 #define NWRAP 3
  54 #define NSKIPSIZE 1
  55
  56 #define LPCQUANT 5
  57 #define V2LPCQOFFSET (1 << LPCQUANT)
  58
  59 #define FNSIZE 2
  60 #define FN_DIFF0        0
  61 #define FN_DIFF1        1
  62 #define FN_DIFF2        2
  63 #define FN_DIFF3        3
  64 #define FN_QUIT         4
  65 #define FN_BLOCKSIZE    5
  66 #define FN_BITSHIFT     6
  67 #define FN_QLPC         7
  68 #define FN_ZERO         8
  69 #define FN_VERBATIM     9
  70
  71 #define VERBATIM_CKSIZE_SIZE 5
  72 #define VERBATIM_BYTE_SIZE 8
  73 #define CANONICAL_HEADER_SIZE 44
  74
  75 typedef struct ShortenContext {
  76     AVCodecContext *avctx;
  77     GetBitContext gb;
  78
  79     int min_framesize, max_framesize;
  80     int channels;
  81
  82     int32_t *decoded[MAX_CHANNELS];
  83     int32_t *offset[MAX_CHANNELS];
  84     uint8_t *bitstream;
  85     int bitstream_size;
  86     int bitstream_index;
  87     unsigned int allocated_bitstream_size;
  88     int header_size;
  89     uint8_t header[OUT_BUFFER_SIZE];
  90     int version;
  91     int cur_chan;
  92     int bitshift;
  93     int nmean;
  94     int internal_ftype;
  95     int nwrap;
  96     int blocksize;
  97     int bitindex;
  98     int32_t lpcqoffset;
  99 } ShortenContext;
 100
 101 static int shorten_decode_init(AVCodecContext * avctx)
 102 {
 103     ShortenContext *s = avctx->priv_data;
 104     s->avctx = avctx;
 105
 106     return 0;
 107 }
 108
 109 static int allocate_buffers(ShortenContext *s)
 110 {
 111     int i, chan;
 112     for (chan=0; chan<s->channels; chan++) {
 113         if(FFMAX(1, s->nmean) >= UINT_MAX/sizeof(int32_t)){
 114             av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
 115             return -1;
 116         }
 117         if(s->blocksize + s->nwrap >= UINT_MAX/sizeof(int32_t) || s->blocksize + s->nwrap <= (unsigned)s->nwrap){
 118             av_log(s->avctx, AV_LOG_ERROR, "s->blocksize + s->nwrap too large\n");
 119             return -1;
 120         }
 121
 122         s->offset[chan] = av_realloc(s->offset[chan], sizeof(int32_t)*FFMAX(1, s->nmean));
 123
 124         s->decoded[chan] = av_realloc(s->decoded[chan], sizeof(int32_t)*(s->blocksize + s->nwrap));
 125         for (i=0; i<s->nwrap; i++)
 126             s->decoded[chan][i] = 0;
 127         s->decoded[chan] += s->nwrap;
 128     }
 129     return 0;
 130 }
 131
 132
 133 static inline unsigned int get_uint(ShortenContext *s, int k)
 134 {
 135     if (s->version != 0)
 136         k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
 137     return get_ur_golomb_shorten(&s->gb, k);
 138 }
 139
 140
 141 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
 142 {
 143     int i;
 144
 145     if (s->bitshift != 0)
 146         for (i = 0; i < s->blocksize; i++)
 147             buffer[s->nwrap + i] <<= s->bitshift;
 148 }
 149
 150
 151 static void init_offset(ShortenContext *s)
 152 {
 153     int32_t mean = 0;
 154     int  chan, i;
 155     int nblock = FFMAX(1, s->nmean);
 156     /* initialise offset */
 157     switch (s->internal_ftype)
 158     {
 159         case TYPE_S16HL:
 160         case TYPE_S16LH:
 161             mean = 0;
 162             break;
 163         default:
 164             av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
 165             abort();
 166     }
 167
 168     for (chan = 0; chan < s->channels; chan++)
 169         for (i = 0; i < nblock; i++)
 170             s->offset[chan][i] = mean;
 171 }
 172
 173 static int inline get_le32(GetBitContext *gb)
 174 {
 175     return bswap_32(get_bits_long(gb, 32));
 176 }
 177
 178 static short inline get_le16(GetBitContext *gb)
 179 {
 180     return bswap_16(get_bits_long(gb, 16));
 181 }
 182
 183 static int decode_wave_header(AVCodecContext *avctx, uint8_t *header, int header_size)
 184 {
 185     GetBitContext hb;
 186     int len;
 187     int chunk_size;
 188     short wave_format;
 189
 190     init_get_bits(&hb, header, header_size*8);
 191     if (get_le32(&hb) != MKTAG('R','I','F','F')) {
 192         av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
 193         return -1;
 194     }
 195
 196     chunk_size = get_le32(&hb);
 197
 198     if (get_le32(&hb) != MKTAG('W','A','V','E')) {
 199         av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
 200         return -1;
 201     }
 202
 203     while (get_le32(&hb) != MKTAG('f','m','t',' ')) {
 204         len = get_le32(&hb);
 205         skip_bits(&hb, 8*len);
 206     }
 207     len = get_le32(&hb);
 208
 209     if (len < 16) {
 210         av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
 211         return -1;
 212     }
 213
 214     wave_format = get_le16(&hb);
 215
 216     switch (wave_format) {
 217         case WAVE_FORMAT_PCM:
 218             break;
 219         default:
 220             av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
 221             return -1;
 222     }
 223
 224     avctx->channels = get_le16(&hb);
 225     avctx->sample_rate = get_le32(&hb);
 226     avctx->bit_rate = get_le32(&hb) * 8;
 227     avctx->block_align = get_le16(&hb);
 228     avctx->bits_per_sample = get_le16(&hb);
 229
 230     if (avctx->bits_per_sample != 16) {
 231         av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
 232         return -1;
 233     }
 234
 235     len -= 16;
 236     if (len > 0)
 237         av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
 238
 239     return 0;
 240 }
 241
 242 static int16_t * interleave_buffer(int16_t *samples, int nchan, int blocksize, int32_t **buffer) {
 243     int i, chan;
 244     for (i=0; i<blocksize; i++)
 245         for (chan=0; chan < nchan; chan++)
 246             *samples++ = FFMIN(buffer[chan][i], 32768);
 247     return samples;
 248 }
 249
 250 static void decode_subframe_lpc(ShortenContext *s, int channel, int residual_size, int pred_order)
 251 {
 252     int sum, i, j;
 253     int coeffs[pred_order];
 254
 255     for (i=0; i<pred_order; i++)
 256         coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
 257
 258     for (i=0; i < s->blocksize; i++) {
 259         sum = s->lpcqoffset;
 260         for (j=0; j<pred_order; j++)
 261             sum += coeffs[j] * s->decoded[channel][i-j-1];
 262         s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> LPCQUANT);
 263     }
 264 }
 265
 266
 267 static int shorten_decode_frame(AVCodecContext *avctx,
 268         void *data, int *data_size,
 269         uint8_t *buf, int buf_size)
 270 {
 271     ShortenContext *s = avctx->priv_data;
 272     int i, input_buf_size = 0;
 273     int16_t *samples = data;
 274     if(s->max_framesize == 0){
 275         s->max_framesize= 1024; // should hopefully be enough for the first header
 276         s->bitstream= av_fast_realloc(s->bitstream, &s->allocated_bitstream_size, s->max_framesize);
 277     }
 278
 279     if(1 && s->max_framesize){//FIXME truncated
 280         buf_size= FFMIN(buf_size, s->max_framesize - s->bitstream_size);
 281         input_buf_size= buf_size;
 282
 283         if(s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size){
 284             //                printf("memmove\n");
 285             memmove(s->bitstream, &s->bitstream[s->bitstream_index], s->bitstream_size);
 286             s->bitstream_index=0;
 287         }
 288         memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
 289         buf= &s->bitstream[s->bitstream_index];
 290         buf_size += s->bitstream_size;
 291         s->bitstream_size= buf_size;
 292
 293         if(buf_size < s->max_framesize){
 294             //dprintf("wanna more data ... %d\n", buf_size);
 295             return input_buf_size;
 296         }
 297     }
 298     init_get_bits(&s->gb, buf, buf_size*8);
 299     get_bits(&s->gb, s->bitindex);
 300     if (!s->blocksize)
 301     {
 302         int maxnlpc = 0;
 303         /* shorten signature */
 304         if (get_bits_long(&s->gb, 32) != bswap_32(ff_get_fourcc("ajkg"))) {
 305             av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
 306             return -1;
 307         }
 308
 309         s->lpcqoffset = 0;
 310         s->blocksize = DEFAULT_BLOCK_SIZE;
 311         s->channels = 1;
 312         s->nmean = -1;
 313         s->version = get_bits(&s->gb, 8);
 314         s->internal_ftype = get_uint(s, TYPESIZE);
 315
 316         s->channels = get_uint(s, CHANSIZE);
 317         if (s->channels > MAX_CHANNELS) {
 318             av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
 319             return -1;
 320         }
 321
 322         /* get blocksize if version > 0 */
 323         if (s->version > 0) {
 324             int skip_bytes;
 325             s->blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
 326             maxnlpc = get_uint(s, LPCQSIZE);
 327             s->nmean = get_uint(s, 0);
 328
 329             skip_bytes = get_uint(s, NSKIPSIZE);
 330             for (i=0; i<skip_bytes; i++) {
 331                 skip_bits(&s->gb, 8);
 332             }
 333         }
 334         s->nwrap = FFMAX(NWRAP, maxnlpc);
 335
 336         allocate_buffers(s);
 337
 338         init_offset(s);
 339
 340         if (s->version > 1)
 341             s->lpcqoffset = V2LPCQOFFSET;
 342
 343         if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
 344             av_log(s->avctx, AV_LOG_ERROR, "missing verbatim section at begining of stream\n");
 345             return -1;
 346         }
 347
 348         s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
 349         if (s->header_size >= OUT_BUFFER_SIZE || s->header_size < CANONICAL_HEADER_SIZE) {
 350             av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n", s->header_size);
 351             return -1;
 352         }
 353
 354         for (i=0; i<s->header_size; i++)
 355             s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
 356
 357         if (decode_wave_header(avctx, s->header, s->header_size) < 0)
 358             return -1;
 359
 360         s->cur_chan = 0;
 361         s->bitshift = 0;
 362     }
 363     else
 364     {
 365         int cmd;
 366         int len;
 367         cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
 368         switch (cmd) {
 369             case FN_ZERO:
 370             case FN_DIFF0:
 371             case FN_DIFF1:
 372             case FN_DIFF2:
 373             case FN_DIFF3:
 374             case FN_QLPC:
 375                 {
 376                     int residual_size = 0;
 377                     int channel = s->cur_chan;
 378                     int32_t coffset;
 379                     if (cmd != FN_ZERO) {
 380                         residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
 381                         /* this is a hack as version 0 differed in defintion of get_sr_golomb_shorten */
 382                         if (s->version == 0)
 383                             residual_size--;
 384                     }
 385
 386                     if (s->nmean == 0)
 387                         coffset = s->offset[channel][0];
 388                     else {
 389                         int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
 390                         for (i=0; i<s->nmean; i++)
 391                             sum += s->offset[channel][i];
 392                         coffset = sum / s->nmean;
 393                         if (s->version >= 2)
 394                             coffset >>= FFMIN(1, s->bitshift);
 395                     }
 396                     switch (cmd) {
 397                         case FN_ZERO:
 398                             for (i=0; i<s->blocksize; i++)
 399                                 s->decoded[channel][i] = 0;
 400                             break;
 401                         case FN_DIFF0:
 402                             for (i=0; i<s->blocksize; i++)
 403                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + coffset;
 404                             break;
 405                         case FN_DIFF1:
 406                             for (i=0; i<s->blocksize; i++)
 407                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + s->decoded[channel][i - 1];
 408                             break;
 409                         case FN_DIFF2:
 410                             for (i=0; i<s->blocksize; i++)
 411                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 2*s->decoded[channel][i-1]
 412                                                                                                       -   s->decoded[channel][i-2];
 413                             break;
 414                         case FN_DIFF3:
 415                             for (i=0; i<s->blocksize; i++)
 416                                 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + 3*s->decoded[channel][i-1]
 417                                                                                                       - 3*s->decoded[channel][i-2]
 418                                                                                                       +   s->decoded[channel][i-3];
 419                             break;
 420                         case FN_QLPC:
 421                             {
 422                                 int pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
 423                                 for (i=0; i<pred_order; i++)
 424                                     s->decoded[channel][i - pred_order] -= coffset;
 425                                 decode_subframe_lpc(s, channel, residual_size, pred_order);
 426                                 if (coffset != 0)
 427                                     for (i=0; i < s->blocksize; i++)
 428                                         s->decoded[channel][i] += coffset;
 429                             }
 430                     }
 431                     if (s->nmean > 0) {
 432                         int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
 433                         for (i=0; i<s->blocksize; i++)
 434                             sum += s->decoded[channel][i];
 435
 436                         for (i=1; i<s->nmean; i++)
 437                             s->offset[channel][i-1] = s->offset[channel][i];
 438
 439                         if (s->version < 2)
 440                             s->offset[channel][s->nmean - 1] = sum / s->blocksize;
 441                         else
 442                             s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
 443                     }
 444                     for (i=-s->nwrap; i<0; i++)
 445                         s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
 446
 447                     fix_bitshift(s, s->decoded[channel]);
 448
 449                     s->cur_chan++;
 450                     if (s->cur_chan == s->channels) {
 451                         samples = interleave_buffer(samples, s->channels, s->blocksize, s->decoded);
 452                         s->cur_chan = 0;
 453                         goto frame_done;
 454                     }
 455                     break;
 456                 }
 457                 break;
 458             case FN_VERBATIM:
 459                 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
 460                 while (len--) {
 461                     get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
 462                 }
 463                 break;
 464             case FN_BITSHIFT:
 465                 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
 466                 break;
 467             case FN_BLOCKSIZE:
 468                 s->blocksize = get_uint(s, av_log2(s->blocksize));
 469                 break;
 470             case FN_QUIT:
 471                 return buf_size;
 472                 break;
 473             default:
 474                 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
 475                 return -1;
 476                 break;
 477         }
 478     }
 479 frame_done:
 480     *data_size = (int8_t *)samples - (int8_t *)data;
 481
 482     //    s->last_blocksize = s->blocksize;
 483     s->bitindex = get_bits_count(&s->gb) - 8*((get_bits_count(&s->gb))/8);
 484     i= (get_bits_count(&s->gb))/8;
 485     if (i > buf_size) {
 486         av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
 487         s->bitstream_size=0;
 488         s->bitstream_index=0;
 489         return -1;
 490     }
 491     if (s->bitstream_size) {
 492         s->bitstream_index += i;
 493         s->bitstream_size  -= i;
 494         return input_buf_size;
 495     } else
 496         return i;
 497 }
 498
 499 static int shorten_decode_close(AVCodecContext *avctx)
 500 {
 501     ShortenContext *s = avctx->priv_data;
 502     int i;
 503
 504     for (i = 0; i < s->channels; i++) {
 505         s->decoded[i] -= s->nwrap;
 506         av_freep(&s->decoded[i]);
 507         av_freep(&s->offset[i]);
 508     }
 509     av_freep(&s->bitstream);
 510     return 0;
 511 }
 512
 513 static void shorten_flush(AVCodecContext *avctx){
 514     ShortenContext *s = avctx->priv_data;
 515
 516     s->bitstream_size=
 517         s->bitstream_index= 0;
 518 }
 519
 520 AVCodec shorten_decoder = {
 521     "shorten",
 522     CODEC_TYPE_AUDIO,
 523     CODEC_ID_SHORTEN,
 524     sizeof(ShortenContext),
 525     shorten_decode_init,
 526     NULL,
 527     shorten_decode_close,
 528     shorten_decode_frame,
 529     .flush= shorten_flush,
 530 };