FFmpeg
 All Data Structures Namespaces 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  GetByteContext gb;
213 
214  bytestream2_init(&gb, header, header_size);
215 
216  if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
217  av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
218  return AVERROR_INVALIDDATA;
219  }
220 
221  bytestream2_skip(&gb, 4); /* chunk size */
222 
223  if (bytestream2_get_le32(&gb) != MKTAG('W', 'A', 'V', 'E')) {
224  av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
225  return AVERROR_INVALIDDATA;
226  }
227 
228  while (bytestream2_get_le32(&gb) != MKTAG('f', 'm', 't', ' ')) {
229  len = bytestream2_get_le32(&gb);
230  bytestream2_skip(&gb, len);
231  if (len < 0 || bytestream2_get_bytes_left(&gb) < 16) {
232  av_log(avctx, AV_LOG_ERROR, "no fmt chunk found\n");
233  return AVERROR_INVALIDDATA;
234  }
235  }
236  len = bytestream2_get_le32(&gb);
237 
238  if (len < 16) {
239  av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
240  return AVERROR_INVALIDDATA;
241  }
242 
243  wave_format = bytestream2_get_le16(&gb);
244 
245  switch (wave_format) {
246  case WAVE_FORMAT_PCM:
247  break;
248  default:
249  av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
250  return AVERROR(ENOSYS);
251  }
252 
253  bytestream2_skip(&gb, 2); // skip channels (already got from shorten header)
254  avctx->sample_rate = bytestream2_get_le32(&gb);
255  bytestream2_skip(&gb, 4); // skip bit rate (represents original uncompressed bit rate)
256  bytestream2_skip(&gb, 2); // skip block align (not needed)
257  bps = bytestream2_get_le16(&gb);
258  avctx->bits_per_coded_sample = bps;
259 
260  if (bps != 16 && bps != 8) {
261  av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample: %d\n", bps);
262  return AVERROR(ENOSYS);
263  }
264 
265  len -= 16;
266  if (len > 0)
267  av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
268 
269  return 0;
270 }
271 
272 static const int fixed_coeffs[][3] = {
273  { 0, 0, 0 },
274  { 1, 0, 0 },
275  { 2, -1, 0 },
276  { 3, -3, 1 }
277 };
278 
279 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
280  int residual_size, int32_t coffset)
281 {
282  int pred_order, sum, qshift, init_sum, i, j;
283  const int *coeffs;
284 
285  if (command == FN_QLPC) {
286  /* read/validate prediction order */
287  pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
288  if (pred_order > s->nwrap) {
289  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
290  pred_order);
291  return AVERROR(EINVAL);
292  }
293  /* read LPC coefficients */
294  for (i = 0; i < pred_order; i++)
295  s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
296  coeffs = s->coeffs;
297 
298  qshift = LPCQUANT;
299  } else {
300  /* fixed LPC coeffs */
301  pred_order = command;
302  if (pred_order >= FF_ARRAY_ELEMS(fixed_coeffs)) {
303  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
304  pred_order);
305  return AVERROR_INVALIDDATA;
306  }
307  coeffs = fixed_coeffs[pred_order];
308  qshift = 0;
309  }
310 
311  /* subtract offset from previous samples to use in prediction */
312  if (command == FN_QLPC && coffset)
313  for (i = -pred_order; i < 0; i++)
314  s->decoded[channel][i] -= coffset;
315 
316  /* decode residual and do LPC prediction */
317  init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
318  for (i = 0; i < s->blocksize; i++) {
319  sum = init_sum;
320  for (j = 0; j < pred_order; j++)
321  sum += coeffs[j] * s->decoded[channel][i - j - 1];
322  s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
323  (sum >> qshift);
324  }
325 
326  /* add offset to current samples */
327  if (command == FN_QLPC && coffset)
328  for (i = 0; i < s->blocksize; i++)
329  s->decoded[channel][i] += coffset;
330 
331  return 0;
332 }
333 
335 {
336  int i, ret;
337  int maxnlpc = 0;
338  /* shorten signature */
339  if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
340  av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
341  return AVERROR_INVALIDDATA;
342  }
343 
344  s->lpcqoffset = 0;
346  s->nmean = -1;
347  s->version = get_bits(&s->gb, 8);
349 
350  s->channels = get_uint(s, CHANSIZE);
351  if (!s->channels) {
352  av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
353  return AVERROR_INVALIDDATA;
354  }
355  if (s->channels > MAX_CHANNELS) {
356  av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
357  s->channels = 0;
358  return AVERROR_INVALIDDATA;
359  }
360  s->avctx->channels = s->channels;
361 
362  /* get blocksize if version > 0 */
363  if (s->version > 0) {
364  int skip_bytes;
365  unsigned blocksize;
366 
367  blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
368  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
370  "invalid or unsupported block size: %d\n",
371  blocksize);
372  return AVERROR(EINVAL);
373  }
374  s->blocksize = blocksize;
375 
376  maxnlpc = get_uint(s, LPCQSIZE);
377  s->nmean = get_uint(s, 0);
378 
379  skip_bytes = get_uint(s, NSKIPSIZE);
380  for (i = 0; i < skip_bytes; i++)
381  skip_bits(&s->gb, 8);
382  }
383  s->nwrap = FFMAX(NWRAP, maxnlpc);
384 
385  if ((ret = allocate_buffers(s)) < 0)
386  return ret;
387 
388  if ((ret = init_offset(s)) < 0)
389  return ret;
390 
391  if (s->version > 1)
393 
396  "missing verbatim section at beginning of stream\n");
397  return AVERROR_INVALIDDATA;
398  }
399 
401  if (s->header_size >= OUT_BUFFER_SIZE ||
403  av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
404  s->header_size);
405  return AVERROR_INVALIDDATA;
406  }
407 
408  for (i = 0; i < s->header_size; i++)
410 
411  if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
412  return ret;
413 
414  s->cur_chan = 0;
415  s->bitshift = 0;
416 
417  s->got_header = 1;
418 
419  return 0;
420 }
421 
422 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
423  int *got_frame_ptr, AVPacket *avpkt)
424 {
425  AVFrame *frame = data;
426  const uint8_t *buf = avpkt->data;
427  int buf_size = avpkt->size;
428  ShortenContext *s = avctx->priv_data;
429  int i, input_buf_size = 0;
430  int ret;
431 
432  /* allocate internal bitstream buffer */
433  if (s->max_framesize == 0) {
434  void *tmp_ptr;
435  s->max_framesize = 8192; // should hopefully be enough for the first header
438  if (!tmp_ptr) {
439  av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
440  return AVERROR(ENOMEM);
441  }
442  s->bitstream = tmp_ptr;
443  }
444 
445  /* append current packet data to bitstream buffer */
446  if (1 && s->max_framesize) { //FIXME truncated
447  buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
448  input_buf_size = buf_size;
449 
452  memmove(s->bitstream, &s->bitstream[s->bitstream_index],
453  s->bitstream_size);
454  s->bitstream_index = 0;
455  }
456  if (buf)
457  memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
458  buf_size);
459  buf = &s->bitstream[s->bitstream_index];
460  buf_size += s->bitstream_size;
461  s->bitstream_size = buf_size;
462 
463  /* do not decode until buffer has at least max_framesize bytes or
464  * the end of the file has been reached */
465  if (buf_size < s->max_framesize && avpkt->data) {
466  *got_frame_ptr = 0;
467  return input_buf_size;
468  }
469  }
470  /* init and position bitstream reader */
471  init_get_bits(&s->gb, buf, buf_size * 8);
472  skip_bits(&s->gb, s->bitindex);
473 
474  /* process header or next subblock */
475  if (!s->got_header) {
476  if ((ret = read_header(s)) < 0)
477  return ret;
478  *got_frame_ptr = 0;
479  goto finish_frame;
480  }
481 
482  /* if quit command was read previously, don't decode anything */
483  if (s->got_quit_command) {
484  *got_frame_ptr = 0;
485  return avpkt->size;
486  }
487 
488  s->cur_chan = 0;
489  while (s->cur_chan < s->channels) {
490  unsigned cmd;
491  int len;
492 
493  if (get_bits_left(&s->gb) < 3 + FNSIZE) {
494  *got_frame_ptr = 0;
495  break;
496  }
497 
498  cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
499 
500  if (cmd > FN_VERBATIM) {
501  av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
502  *got_frame_ptr = 0;
503  break;
504  }
505 
506  if (!is_audio_command[cmd]) {
507  /* process non-audio command */
508  switch (cmd) {
509  case FN_VERBATIM:
511  while (len--)
513  break;
514  case FN_BITSHIFT:
516  break;
517  case FN_BLOCKSIZE: {
518  unsigned blocksize = get_uint(s, av_log2(s->blocksize));
519  if (blocksize > s->blocksize) {
520  av_log(avctx, AV_LOG_ERROR,
521  "Increasing block size is not supported\n");
522  return AVERROR_PATCHWELCOME;
523  }
524  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
525  av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
526  "block size: %d\n", blocksize);
527  return AVERROR(EINVAL);
528  }
529  s->blocksize = blocksize;
530  break;
531  }
532  case FN_QUIT:
533  s->got_quit_command = 1;
534  break;
535  }
536  if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
537  *got_frame_ptr = 0;
538  break;
539  }
540  } else {
541  /* process audio command */
542  int residual_size = 0;
543  int channel = s->cur_chan;
544  int32_t coffset;
545 
546  /* get Rice code for residual decoding */
547  if (cmd != FN_ZERO) {
548  residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
549  /* This is a hack as version 0 differed in the definition
550  * of get_sr_golomb_shorten(). */
551  if (s->version == 0)
552  residual_size--;
553  }
554 
555  /* calculate sample offset using means from previous blocks */
556  if (s->nmean == 0)
557  coffset = s->offset[channel][0];
558  else {
559  int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
560  for (i = 0; i < s->nmean; i++)
561  sum += s->offset[channel][i];
562  coffset = sum / s->nmean;
563  if (s->version >= 2)
564  coffset = s->bitshift == 0 ? coffset : coffset >> s->bitshift - 1 >> 1;
565  }
566 
567  /* decode samples for this channel */
568  if (cmd == FN_ZERO) {
569  for (i = 0; i < s->blocksize; i++)
570  s->decoded[channel][i] = 0;
571  } else {
572  if ((ret = decode_subframe_lpc(s, cmd, channel,
573  residual_size, coffset)) < 0)
574  return ret;
575  }
576 
577  /* update means with info from the current block */
578  if (s->nmean > 0) {
579  int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
580  for (i = 0; i < s->blocksize; i++)
581  sum += s->decoded[channel][i];
582 
583  for (i = 1; i < s->nmean; i++)
584  s->offset[channel][i - 1] = s->offset[channel][i];
585 
586  if (s->version < 2)
587  s->offset[channel][s->nmean - 1] = sum / s->blocksize;
588  else
589  s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
590  }
591 
592  /* copy wrap samples for use with next block */
593  for (i = -s->nwrap; i < 0; i++)
594  s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
595 
596  /* shift samples to add in unused zero bits which were removed
597  * during encoding */
598  fix_bitshift(s, s->decoded[channel]);
599 
600  /* if this is the last channel in the block, output the samples */
601  s->cur_chan++;
602  if (s->cur_chan == s->channels) {
603  uint8_t *samples_u8;
604  int16_t *samples_s16;
605  int chan;
606 
607  /* get output buffer */
608  frame->nb_samples = s->blocksize;
609  if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
610  return ret;
611 
612  for (chan = 0; chan < s->channels; chan++) {
613  samples_u8 = ((uint8_t **)frame->extended_data)[chan];
614  samples_s16 = ((int16_t **)frame->extended_data)[chan];
615  for (i = 0; i < s->blocksize; i++) {
616  switch (s->internal_ftype) {
617  case TYPE_U8:
618  *samples_u8++ = av_clip_uint8(s->decoded[chan][i]);
619  break;
620  case TYPE_S16HL:
621  case TYPE_S16LH:
622  *samples_s16++ = av_clip_int16(s->decoded[chan][i]);
623  break;
624  }
625  }
626  }
627 
628  *got_frame_ptr = 1;
629  }
630  }
631  }
632  if (s->cur_chan < s->channels)
633  *got_frame_ptr = 0;
634 
636  s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
637  i = get_bits_count(&s->gb) / 8;
638  if (i > buf_size) {
639  av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
640  s->bitstream_size = 0;
641  s->bitstream_index = 0;
642  return AVERROR_INVALIDDATA;
643  }
644  if (s->bitstream_size) {
645  s->bitstream_index += i;
646  s->bitstream_size -= i;
647  return input_buf_size;
648  } else
649  return i;
650 }
651 
653 {
654  ShortenContext *s = avctx->priv_data;
655  int i;
656 
657  for (i = 0; i < s->channels; i++) {
658  s->decoded[i] = NULL;
659  av_freep(&s->decoded_base[i]);
660  av_freep(&s->offset[i]);
661  }
662  av_freep(&s->bitstream);
663  av_freep(&s->coeffs);
664 
665  return 0;
666 }
667 
669  .name = "shorten",
670  .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
671  .type = AVMEDIA_TYPE_AUDIO,
672  .id = AV_CODEC_ID_SHORTEN,
673  .priv_data_size = sizeof(ShortenContext),
677  .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
678  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
681 };