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 Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
29 #include <limits.h>
30 #include "avcodec.h"
31 #include "bytestream.h"
32 #include "get_bits.h"
33 #include "golomb.h"
34 
35 #define MAX_CHANNELS 8
36 #define MAX_BLOCKSIZE 65535
37 
38 #define OUT_BUFFER_SIZE 16384
39 
40 #define ULONGSIZE 2
41 
42 #define WAVE_FORMAT_PCM 0x0001
43 
44 #define DEFAULT_BLOCK_SIZE 256
45 
46 #define TYPESIZE 4
47 #define CHANSIZE 0
48 #define LPCQSIZE 2
49 #define ENERGYSIZE 3
50 #define BITSHIFTSIZE 2
51 
52 #define TYPE_S16HL 3
53 #define TYPE_S16LH 5
54 
55 #define NWRAP 3
56 #define NSKIPSIZE 1
57 
58 #define LPCQUANT 5
59 #define V2LPCQOFFSET (1 << LPCQUANT)
60 
61 #define FNSIZE 2
62 #define FN_DIFF0 0
63 #define FN_DIFF1 1
64 #define FN_DIFF2 2
65 #define FN_DIFF3 3
66 #define FN_QUIT 4
67 #define FN_BLOCKSIZE 5
68 #define FN_BITSHIFT 6
69 #define FN_QLPC 7
70 #define FN_ZERO 8
71 #define FN_VERBATIM 9
72 
74 static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
75 
76 #define VERBATIM_CKSIZE_SIZE 5
77 #define VERBATIM_BYTE_SIZE 8
78 #define CANONICAL_HEADER_SIZE 44
79 
80 typedef struct ShortenContext {
84 
86  unsigned channels;
87 
88  int32_t *decoded[MAX_CHANNELS];
90  int32_t *offset[MAX_CHANNELS];
91  int *coeffs;
92  uint8_t *bitstream;
98  int version;
99  int cur_chan;
100  int bitshift;
101  int nmean;
103  int nwrap;
105  int bitindex;
106  int32_t lpcqoffset;
110 
112 {
113  ShortenContext *s = avctx->priv_data;
114  s->avctx = avctx;
115  avctx->sample_fmt = AV_SAMPLE_FMT_S16;
116 
118  avctx->coded_frame = &s->frame;
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 
172 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
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_S16HL:
189  case TYPE_S16LH:
190  mean = 0;
191  break;
192  default:
193  av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
194  return AVERROR_INVALIDDATA;
195  }
196 
197  for (chan = 0; chan < s->channels; chan++)
198  for (i = 0; i < nblock; i++)
199  s->offset[chan][i] = mean;
200  return 0;
201 }
202 
203 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
204  int header_size)
205 {
206  int len;
207  short wave_format;
208 
209  if (bytestream_get_le32(&header) != MKTAG('R', 'I', 'F', 'F')) {
210  av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
211  return AVERROR_INVALIDDATA;
212  }
213 
214  header += 4; /* chunk size */
215 
216  if (bytestream_get_le32(&header) != MKTAG('W', 'A', 'V', 'E')) {
217  av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
218  return AVERROR_INVALIDDATA;
219  }
220 
221  while (bytestream_get_le32(&header) != MKTAG('f', 'm', 't', ' ')) {
222  len = bytestream_get_le32(&header);
223  header += len;
224  }
225  len = bytestream_get_le32(&header);
226 
227  if (len < 16) {
228  av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
229  return AVERROR_INVALIDDATA;
230  }
231 
232  wave_format = bytestream_get_le16(&header);
233 
234  switch (wave_format) {
235  case WAVE_FORMAT_PCM:
236  break;
237  default:
238  av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
239  return AVERROR(ENOSYS);
240  }
241 
242  header += 2; // skip channels (already got from shorten header)
243  avctx->sample_rate = bytestream_get_le32(&header);
244  header += 4; // skip bit rate (represents original uncompressed bit rate)
245  header += 2; // skip block align (not needed)
246  avctx->bits_per_coded_sample = bytestream_get_le16(&header);
247 
248  if (avctx->bits_per_coded_sample != 16) {
249  av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
250  return AVERROR(ENOSYS);
251  }
252 
253  len -= 16;
254  if (len > 0)
255  av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
256 
257  return 0;
258 }
259 
260 static void interleave_buffer(int16_t *samples, int nchan, int blocksize,
261  int32_t **buffer)
262 {
263  int i, chan;
264  for (i=0; i<blocksize; i++)
265  for (chan=0; chan < nchan; chan++)
266  *samples++ = av_clip_int16(buffer[chan][i]);
267 }
268 
269 static const int fixed_coeffs[3][3] = {
270  { 1, 0, 0 },
271  { 2, -1, 0 },
272  { 3, -3, 1 }
273 };
274 
275 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
276  int residual_size, int32_t coffset)
277 {
278  int pred_order, sum, qshift, init_sum, i, j;
279  const int *coeffs;
280 
281  if (command == FN_QLPC) {
282  /* read/validate prediction order */
283  pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
284  if (pred_order > s->nwrap) {
285  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
286  pred_order);
287  return AVERROR(EINVAL);
288  }
289  /* read LPC coefficients */
290  for (i = 0; i < pred_order; i++)
291  s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
292  coeffs = s->coeffs;
293 
294  qshift = LPCQUANT;
295  } else {
296  /* fixed LPC coeffs */
297  pred_order = command;
298  coeffs = fixed_coeffs[pred_order - 1];
299  qshift = 0;
300  }
301 
302  /* subtract offset from previous samples to use in prediction */
303  if (command == FN_QLPC && coffset)
304  for (i = -pred_order; i < 0; i++)
305  s->decoded[channel][i] -= coffset;
306 
307  /* decode residual and do LPC prediction */
308  init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
309  for (i = 0; i < s->blocksize; i++) {
310  sum = init_sum;
311  for (j = 0; j < pred_order; j++)
312  sum += coeffs[j] * s->decoded[channel][i - j - 1];
313  s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
314  (sum >> qshift);
315  }
316 
317  /* add offset to current samples */
318  if (command == FN_QLPC && coffset)
319  for (i = 0; i < s->blocksize; i++)
320  s->decoded[channel][i] += coffset;
321 
322  return 0;
323 }
324 
326 {
327  int i, ret;
328  int maxnlpc = 0;
329  /* shorten signature */
330  if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
331  av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
332  return AVERROR_INVALIDDATA;
333  }
334 
335  s->lpcqoffset = 0;
337  s->nmean = -1;
338  s->version = get_bits(&s->gb, 8);
340 
341  s->channels = get_uint(s, CHANSIZE);
342  if (!s->channels) {
343  av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
344  return AVERROR_INVALIDDATA;
345  }
346  if (s->channels > MAX_CHANNELS) {
347  av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
348  s->channels = 0;
349  return AVERROR_INVALIDDATA;
350  }
351  s->avctx->channels = s->channels;
352 
353  /* get blocksize if version > 0 */
354  if (s->version > 0) {
355  int skip_bytes;
356  unsigned blocksize;
357 
358  blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
359  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
361  "invalid or unsupported block size: %d\n",
362  blocksize);
363  return AVERROR(EINVAL);
364  }
365  s->blocksize = blocksize;
366 
367  maxnlpc = get_uint(s, LPCQSIZE);
368  s->nmean = get_uint(s, 0);
369 
370  skip_bytes = get_uint(s, NSKIPSIZE);
371  for (i = 0; i < skip_bytes; i++)
372  skip_bits(&s->gb, 8);
373  }
374  s->nwrap = FFMAX(NWRAP, maxnlpc);
375 
376  if ((ret = allocate_buffers(s)) < 0)
377  return ret;
378 
379  if ((ret = init_offset(s)) < 0)
380  return ret;
381 
382  if (s->version > 1)
384 
387  "missing verbatim section at beginning of stream\n");
388  return AVERROR_INVALIDDATA;
389  }
390 
392  if (s->header_size >= OUT_BUFFER_SIZE ||
394  av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
395  s->header_size);
396  return AVERROR_INVALIDDATA;
397  }
398 
399  for (i = 0; i < s->header_size; i++)
401 
402  if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
403  return ret;
404 
405  s->cur_chan = 0;
406  s->bitshift = 0;
407 
408  s->got_header = 1;
409 
410  return 0;
411 }
412 
413 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
414  int *got_frame_ptr, AVPacket *avpkt)
415 {
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 = 1024; // 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  int 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 defintion of get_sr_golomb_shorten */
540  if (s->version == 0)
541  residual_size--;
542  }
543 
544  /* calculate sample offset using means from previous blocks */
545  if (s->nmean == 0)
546  coffset = s->offset[channel][0];
547  else {
548  int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
549  for (i = 0; i < s->nmean; i++)
550  sum += s->offset[channel][i];
551  coffset = sum / s->nmean;
552  if (s->version >= 2)
553  coffset >>= FFMIN(1, s->bitshift);
554  }
555 
556  /* decode samples for this channel */
557  if (cmd == FN_ZERO) {
558  for (i = 0; i < s->blocksize; i++)
559  s->decoded[channel][i] = 0;
560  } else {
561  if ((ret = decode_subframe_lpc(s, cmd, channel,
562  residual_size, coffset)) < 0)
563  return ret;
564  }
565 
566  /* update means with info from the current block */
567  if (s->nmean > 0) {
568  int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
569  for (i = 0; i < s->blocksize; i++)
570  sum += s->decoded[channel][i];
571 
572  for (i = 1; i < s->nmean; i++)
573  s->offset[channel][i - 1] = s->offset[channel][i];
574 
575  if (s->version < 2)
576  s->offset[channel][s->nmean - 1] = sum / s->blocksize;
577  else
578  s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
579  }
580 
581  /* copy wrap samples for use with next block */
582  for (i = -s->nwrap; i < 0; i++)
583  s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
584 
585  /* shift samples to add in unused zero bits which were removed
586  * during encoding */
587  fix_bitshift(s, s->decoded[channel]);
588 
589  /* if this is the last channel in the block, output the samples */
590  s->cur_chan++;
591  if (s->cur_chan == s->channels) {
592  /* get output buffer */
593  s->frame.nb_samples = s->blocksize;
594  if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
595  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
596  return ret;
597  }
598  /* interleave output */
599  interleave_buffer((int16_t *)s->frame.data[0], s->channels,
600  s->blocksize, s->decoded);
601 
602  *got_frame_ptr = 1;
603  *(AVFrame *)data = s->frame;
604  }
605  }
606  }
607  if (s->cur_chan < s->channels)
608  *got_frame_ptr = 0;
609 
611  s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
612  i = get_bits_count(&s->gb) / 8;
613  if (i > buf_size) {
614  av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
615  s->bitstream_size = 0;
616  s->bitstream_index = 0;
617  return AVERROR_INVALIDDATA;
618  }
619  if (s->bitstream_size) {
620  s->bitstream_index += i;
621  s->bitstream_size -= i;
622  return input_buf_size;
623  } else
624  return i;
625 }
626 
628 {
629  ShortenContext *s = avctx->priv_data;
630  int i;
631 
632  for (i = 0; i < s->channels; i++) {
633  s->decoded[i] = NULL;
634  av_freep(&s->decoded_base[i]);
635  av_freep(&s->offset[i]);
636  }
637  av_freep(&s->bitstream);
638  av_freep(&s->coeffs);
639 
640  return 0;
641 }
642 
644  .name = "shorten",
645  .type = AVMEDIA_TYPE_AUDIO,
646  .id = CODEC_ID_SHORTEN,
647  .priv_data_size = sizeof(ShortenContext),
651  .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
652  .long_name= NULL_IF_CONFIG_SMALL("Shorten"),
653 };