FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
shorten.c
Go to the documentation of this file.
1 /*
2  * Shorten decoder
3  * Copyright (c) 2005 Jeff Muizelaar
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
24  * Shorten decoder
25  * @author Jeff Muizelaar
26  *
27  */
28 
29 #include <limits.h>
30 #include "avcodec.h"
31 #include "bytestream.h"
32 #include "get_bits.h"
33 #include "golomb.h"
34 #include "internal.h"
35 
36 #define MAX_CHANNELS 8
37 #define MAX_BLOCKSIZE 65535
38 
39 #define OUT_BUFFER_SIZE 16384
40 
41 #define ULONGSIZE 2
42 
43 #define WAVE_FORMAT_PCM 0x0001
44 
45 #define DEFAULT_BLOCK_SIZE 256
46 
47 #define TYPESIZE 4
48 #define CHANSIZE 0
49 #define LPCQSIZE 2
50 #define ENERGYSIZE 3
51 #define BITSHIFTSIZE 2
52 
53 #define TYPE_S8 1
54 #define TYPE_U8 2
55 #define TYPE_S16HL 3
56 #define TYPE_U16HL 4
57 #define TYPE_S16LH 5
58 #define TYPE_U16LH 6
59 
60 #define NWRAP 3
61 #define NSKIPSIZE 1
62 
63 #define LPCQUANT 5
64 #define V2LPCQOFFSET (1 << LPCQUANT)
65 
66 #define FNSIZE 2
67 #define FN_DIFF0 0
68 #define FN_DIFF1 1
69 #define FN_DIFF2 2
70 #define FN_DIFF3 3
71 #define FN_QUIT 4
72 #define FN_BLOCKSIZE 5
73 #define FN_BITSHIFT 6
74 #define FN_QLPC 7
75 #define FN_ZERO 8
76 #define FN_VERBATIM 9
77 
78 /** indicates if the FN_* command is audio or non-audio */
79 static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
80 
81 #define VERBATIM_CKSIZE_SIZE 5
82 #define VERBATIM_BYTE_SIZE 8
83 #define CANONICAL_HEADER_SIZE 44
84 
85 typedef struct ShortenContext {
88 
90  unsigned channels;
91 
95  int *coeffs;
102  int version;
103  int cur_chan;
104  int bitshift;
105  int nmean;
107  int nwrap;
109  int bitindex;
114 
116 {
117  ShortenContext *s = avctx->priv_data;
118  s->avctx = avctx;
119 
120  return 0;
121 }
122 
124 {
125  int i, chan;
126  int *coeffs;
127  void *tmp_ptr;
128 
129  for (chan = 0; chan < s->channels; chan++) {
130  if (FFMAX(1, s->nmean) >= UINT_MAX / sizeof(int32_t)) {
131  av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
132  return AVERROR_INVALIDDATA;
133  }
134  if (s->blocksize + s->nwrap >= UINT_MAX / sizeof(int32_t) ||
135  s->blocksize + s->nwrap <= (unsigned)s->nwrap) {
137  "s->blocksize + s->nwrap too large\n");
138  return AVERROR_INVALIDDATA;
139  }
140 
141  tmp_ptr =
142  av_realloc(s->offset[chan], sizeof(int32_t) * FFMAX(1, s->nmean));
143  if (!tmp_ptr)
144  return AVERROR(ENOMEM);
145  s->offset[chan] = tmp_ptr;
146 
147  tmp_ptr = av_realloc(s->decoded_base[chan], (s->blocksize + s->nwrap) *
148  sizeof(s->decoded_base[0][0]));
149  if (!tmp_ptr)
150  return AVERROR(ENOMEM);
151  s->decoded_base[chan] = tmp_ptr;
152  for (i = 0; i < s->nwrap; i++)
153  s->decoded_base[chan][i] = 0;
154  s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
155  }
156 
157  coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
158  if (!coeffs)
159  return AVERROR(ENOMEM);
160  s->coeffs = coeffs;
161 
162  return 0;
163 }
164 
165 static inline unsigned int get_uint(ShortenContext *s, int k)
166 {
167  if (s->version != 0)
169  return get_ur_golomb_shorten(&s->gb, k);
170 }
171 
173 {
174  int i;
175 
176  if (s->bitshift != 0)
177  for (i = 0; i < s->blocksize; i++)
178  buffer[i] <<= s->bitshift;
179 }
180 
182 {
183  int32_t mean = 0;
184  int chan, i;
185  int nblock = FFMAX(1, s->nmean);
186  /* initialise offset */
187  switch (s->internal_ftype) {
188  case TYPE_U8:
190  mean = 0x80;
191  break;
192  case TYPE_S16HL:
193  case TYPE_S16LH:
195  break;
196  default:
197  av_log(s->avctx, AV_LOG_ERROR, "unknown audio type\n");
198  return AVERROR_PATCHWELCOME;
199  }
200 
201  for (chan = 0; chan < s->channels; chan++)
202  for (i = 0; i < nblock; i++)
203  s->offset[chan][i] = mean;
204  return 0;
205 }
206 
207 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
208  int header_size)
209 {
210  int len, bps;
211  short wave_format;
212  const uint8_t *end= header + header_size;
213 
214  if (bytestream_get_le32(&header) != MKTAG('R', 'I', 'F', 'F')) {
215  av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
216  return AVERROR_INVALIDDATA;
217  }
218 
219  header += 4; /* chunk size */
220 
221  if (bytestream_get_le32(&header) != MKTAG('W', 'A', 'V', 'E')) {
222  av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
223  return AVERROR_INVALIDDATA;
224  }
225 
226  while (bytestream_get_le32(&header) != MKTAG('f', 'm', 't', ' ')) {
227  len = bytestream_get_le32(&header);
228  if (len<0 || end - header - 8 < len)
229  return AVERROR_INVALIDDATA;
230  header += len;
231  }
232  len = bytestream_get_le32(&header);
233 
234  if (len < 16) {
235  av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
236  return AVERROR_INVALIDDATA;
237  }
238 
239  wave_format = bytestream_get_le16(&header);
240 
241  switch (wave_format) {
242  case WAVE_FORMAT_PCM:
243  break;
244  default:
245  av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
246  return AVERROR(ENOSYS);
247  }
248 
249  header += 2; // skip channels (already got from shorten header)
250  avctx->sample_rate = bytestream_get_le32(&header);
251  header += 4; // skip bit rate (represents original uncompressed bit rate)
252  header += 2; // skip block align (not needed)
253  bps = bytestream_get_le16(&header);
254  avctx->bits_per_coded_sample = bps;
255 
256  if (bps != 16 && bps != 8) {
257  av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample: %d\n", bps);
258  return AVERROR(ENOSYS);
259  }
260 
261  len -= 16;
262  if (len > 0)
263  av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
264 
265  return 0;
266 }
267 
268 static const int fixed_coeffs[3][3] = {
269  { 1, 0, 0 },
270  { 2, -1, 0 },
271  { 3, -3, 1 }
272 };
273 
274 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
275  int residual_size, int32_t coffset)
276 {
277  int pred_order, sum, qshift, init_sum, i, j;
278  const int *coeffs;
279 
280  if (command == FN_QLPC) {
281  /* read/validate prediction order */
282  pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
283  if (pred_order > s->nwrap) {
284  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
285  pred_order);
286  return AVERROR(EINVAL);
287  }
288  /* read LPC coefficients */
289  for (i = 0; i < pred_order; i++)
290  s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
291  coeffs = s->coeffs;
292 
293  qshift = LPCQUANT;
294  } else {
295  /* fixed LPC coeffs */
296  pred_order = command;
297  coeffs = fixed_coeffs[pred_order - 1];
298  qshift = 0;
299  }
300 
301  /* subtract offset from previous samples to use in prediction */
302  if (command == FN_QLPC && coffset)
303  for (i = -pred_order; i < 0; i++)
304  s->decoded[channel][i] -= coffset;
305 
306  /* decode residual and do LPC prediction */
307  init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
308  for (i = 0; i < s->blocksize; i++) {
309  sum = init_sum;
310  for (j = 0; j < pred_order; j++)
311  sum += coeffs[j] * s->decoded[channel][i - j - 1];
312  s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
313  (sum >> qshift);
314  }
315 
316  /* add offset to current samples */
317  if (command == FN_QLPC && coffset)
318  for (i = 0; i < s->blocksize; i++)
319  s->decoded[channel][i] += coffset;
320 
321  return 0;
322 }
323 
325 {
326  int i, ret;
327  int maxnlpc = 0;
328  /* shorten signature */
329  if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
330  av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
331  return AVERROR_INVALIDDATA;
332  }
333 
334  s->lpcqoffset = 0;
336  s->nmean = -1;
337  s->version = get_bits(&s->gb, 8);
339 
340  s->channels = get_uint(s, CHANSIZE);
341  if (!s->channels) {
342  av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
343  return AVERROR_INVALIDDATA;
344  }
345  if (s->channels > MAX_CHANNELS) {
346  av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
347  s->channels = 0;
348  return AVERROR_INVALIDDATA;
349  }
350  s->avctx->channels = s->channels;
351 
352  /* get blocksize if version > 0 */
353  if (s->version > 0) {
354  int skip_bytes;
355  unsigned blocksize;
356 
357  blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
358  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
360  "invalid or unsupported block size: %d\n",
361  blocksize);
362  return AVERROR(EINVAL);
363  }
364  s->blocksize = blocksize;
365 
366  maxnlpc = get_uint(s, LPCQSIZE);
367  s->nmean = get_uint(s, 0);
368 
369  skip_bytes = get_uint(s, NSKIPSIZE);
370  for (i = 0; i < skip_bytes; i++)
371  skip_bits(&s->gb, 8);
372  }
373  s->nwrap = FFMAX(NWRAP, maxnlpc);
374 
375  if ((ret = allocate_buffers(s)) < 0)
376  return ret;
377 
378  if ((ret = init_offset(s)) < 0)
379  return ret;
380 
381  if (s->version > 1)
383 
386  "missing verbatim section at beginning of stream\n");
387  return AVERROR_INVALIDDATA;
388  }
389 
391  if (s->header_size >= OUT_BUFFER_SIZE ||
393  av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
394  s->header_size);
395  return AVERROR_INVALIDDATA;
396  }
397 
398  for (i = 0; i < s->header_size; i++)
400 
401  if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
402  return ret;
403 
404  s->cur_chan = 0;
405  s->bitshift = 0;
406 
407  s->got_header = 1;
408 
409  return 0;
410 }
411 
412 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
413  int *got_frame_ptr, AVPacket *avpkt)
414 {
415  AVFrame *frame = data;
416  const uint8_t *buf = avpkt->data;
417  int buf_size = avpkt->size;
418  ShortenContext *s = avctx->priv_data;
419  int i, input_buf_size = 0;
420  int ret;
421 
422  /* allocate internal bitstream buffer */
423  if (s->max_framesize == 0) {
424  void *tmp_ptr;
425  s->max_framesize = 8192; // should hopefully be enough for the first header
427  s->max_framesize);
428  if (!tmp_ptr) {
429  av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
430  return AVERROR(ENOMEM);
431  }
432  s->bitstream = tmp_ptr;
433  }
434 
435  /* append current packet data to bitstream buffer */
436  if (1 && s->max_framesize) { //FIXME truncated
437  buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
438  input_buf_size = buf_size;
439 
440  if (s->bitstream_index + s->bitstream_size + buf_size >
442  memmove(s->bitstream, &s->bitstream[s->bitstream_index],
443  s->bitstream_size);
444  s->bitstream_index = 0;
445  }
446  if (buf)
447  memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
448  buf_size);
449  buf = &s->bitstream[s->bitstream_index];
450  buf_size += s->bitstream_size;
451  s->bitstream_size = buf_size;
452 
453  /* do not decode until buffer has at least max_framesize bytes or
454  * the end of the file has been reached */
455  if (buf_size < s->max_framesize && avpkt->data) {
456  *got_frame_ptr = 0;
457  return input_buf_size;
458  }
459  }
460  /* init and position bitstream reader */
461  init_get_bits(&s->gb, buf, buf_size * 8);
462  skip_bits(&s->gb, s->bitindex);
463 
464  /* process header or next subblock */
465  if (!s->got_header) {
466  if ((ret = read_header(s)) < 0)
467  return ret;
468  *got_frame_ptr = 0;
469  goto finish_frame;
470  }
471 
472  /* if quit command was read previously, don't decode anything */
473  if (s->got_quit_command) {
474  *got_frame_ptr = 0;
475  return avpkt->size;
476  }
477 
478  s->cur_chan = 0;
479  while (s->cur_chan < s->channels) {
480  unsigned cmd;
481  int len;
482 
483  if (get_bits_left(&s->gb) < 3 + FNSIZE) {
484  *got_frame_ptr = 0;
485  break;
486  }
487 
488  cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
489 
490  if (cmd > FN_VERBATIM) {
491  av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
492  *got_frame_ptr = 0;
493  break;
494  }
495 
496  if (!is_audio_command[cmd]) {
497  /* process non-audio command */
498  switch (cmd) {
499  case FN_VERBATIM:
501  while (len--)
503  break;
504  case FN_BITSHIFT:
506  break;
507  case FN_BLOCKSIZE: {
508  unsigned blocksize = get_uint(s, av_log2(s->blocksize));
509  if (blocksize > s->blocksize) {
510  av_log(avctx, AV_LOG_ERROR,
511  "Increasing block size is not supported\n");
512  return AVERROR_PATCHWELCOME;
513  }
514  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
515  av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
516  "block size: %d\n", blocksize);
517  return AVERROR(EINVAL);
518  }
519  s->blocksize = blocksize;
520  break;
521  }
522  case FN_QUIT:
523  s->got_quit_command = 1;
524  break;
525  }
526  if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
527  *got_frame_ptr = 0;
528  break;
529  }
530  } else {
531  /* process audio command */
532  int residual_size = 0;
533  int channel = s->cur_chan;
534  int32_t coffset;
535 
536  /* get Rice code for residual decoding */
537  if (cmd != FN_ZERO) {
538  residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
539  /* This is a hack as version 0 differed in the definition
540  * of get_sr_golomb_shorten(). */
541  if (s->version == 0)
542  residual_size--;
543  }
544 
545  /* calculate sample offset using means from previous blocks */
546  if (s->nmean == 0)
547  coffset = s->offset[channel][0];
548  else {
549  int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
550  for (i = 0; i < s->nmean; i++)
551  sum += s->offset[channel][i];
552  coffset = sum / s->nmean;
553  if (s->version >= 2)
554  coffset = s->bitshift == 0 ? coffset : coffset >> s->bitshift - 1 >> 1;
555  }
556 
557  /* decode samples for this channel */
558  if (cmd == FN_ZERO) {
559  for (i = 0; i < s->blocksize; i++)
560  s->decoded[channel][i] = 0;
561  } else {
562  if ((ret = decode_subframe_lpc(s, cmd, channel,
563  residual_size, coffset)) < 0)
564  return ret;
565  }
566 
567  /* update means with info from the current block */
568  if (s->nmean > 0) {
569  int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
570  for (i = 0; i < s->blocksize; i++)
571  sum += s->decoded[channel][i];
572 
573  for (i = 1; i < s->nmean; i++)
574  s->offset[channel][i - 1] = s->offset[channel][i];
575 
576  if (s->version < 2)
577  s->offset[channel][s->nmean - 1] = sum / s->blocksize;
578  else
579  s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
580  }
581 
582  /* copy wrap samples for use with next block */
583  for (i = -s->nwrap; i < 0; i++)
584  s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
585 
586  /* shift samples to add in unused zero bits which were removed
587  * during encoding */
588  fix_bitshift(s, s->decoded[channel]);
589 
590  /* if this is the last channel in the block, output the samples */
591  s->cur_chan++;
592  if (s->cur_chan == s->channels) {
593  uint8_t *samples_u8;
594  int16_t *samples_s16;
595  int chan;
596 
597  /* get output buffer */
598  frame->nb_samples = s->blocksize;
599  if ((ret = ff_get_buffer(avctx, frame)) < 0) {
600  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
601  return ret;
602  }
603 
604  for (chan = 0; chan < s->channels; chan++) {
605  samples_u8 = ((uint8_t **)frame->extended_data)[chan];
606  samples_s16 = ((int16_t **)frame->extended_data)[chan];
607  for (i = 0; i < s->blocksize; i++) {
608  switch (s->internal_ftype) {
609  case TYPE_U8:
610  *samples_u8++ = av_clip_uint8(s->decoded[chan][i]);
611  break;
612  case TYPE_S16HL:
613  case TYPE_S16LH:
614  *samples_s16++ = av_clip_int16(s->decoded[chan][i]);
615  break;
616  }
617  }
618  }
619 
620  *got_frame_ptr = 1;
621  }
622  }
623  }
624  if (s->cur_chan < s->channels)
625  *got_frame_ptr = 0;
626 
628  s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
629  i = get_bits_count(&s->gb) / 8;
630  if (i > buf_size) {
631  av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
632  s->bitstream_size = 0;
633  s->bitstream_index = 0;
634  return AVERROR_INVALIDDATA;
635  }
636  if (s->bitstream_size) {
637  s->bitstream_index += i;
638  s->bitstream_size -= i;
639  return input_buf_size;
640  } else
641  return i;
642 }
643 
645 {
646  ShortenContext *s = avctx->priv_data;
647  int i;
648 
649  for (i = 0; i < s->channels; i++) {
650  s->decoded[i] = NULL;
651  av_freep(&s->decoded_base[i]);
652  av_freep(&s->offset[i]);
653  }
654  av_freep(&s->bitstream);
655  av_freep(&s->coeffs);
656 
657  return 0;
658 }
659 
661  .name = "shorten",
662  .type = AVMEDIA_TYPE_AUDIO,
663  .id = AV_CODEC_ID_SHORTEN,
664  .priv_data_size = sizeof(ShortenContext),
668  .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
669  .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
670  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
673 };