FFmpeg  2.6.9
 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, err;
126 
127  for (chan = 0; chan < s->channels; chan++) {
128  if (FFMAX(1, s->nmean) >= UINT_MAX / sizeof(int32_t)) {
129  av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
130  return AVERROR_INVALIDDATA;
131  }
132  if (s->blocksize + (uint64_t)s->nwrap >= UINT_MAX / sizeof(int32_t)) {
134  "s->blocksize + s->nwrap too large\n");
135  return AVERROR_INVALIDDATA;
136  }
137 
138  if ((err = av_reallocp_array(&s->offset[chan],
139  sizeof(int32_t),
140  FFMAX(1, s->nmean))) < 0)
141  return err;
142 
143  if ((err = av_reallocp_array(&s->decoded_base[chan], (s->blocksize + s->nwrap),
144  sizeof(s->decoded_base[0][0]))) < 0)
145  return err;
146  for (i = 0; i < s->nwrap; i++)
147  s->decoded_base[chan][i] = 0;
148  s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
149  }
150 
151  if ((err = av_reallocp_array(&s->coeffs, s->nwrap, sizeof(*s->coeffs))) < 0)
152  return err;
153 
154  return 0;
155 }
156 
157 static inline unsigned int get_uint(ShortenContext *s, int k)
158 {
159  if (s->version != 0)
161  return get_ur_golomb_shorten(&s->gb, k);
162 }
163 
165 {
166  int i;
167 
168  if (s->bitshift != 0)
169  for (i = 0; i < s->blocksize; i++)
170  buffer[i] <<= s->bitshift;
171 }
172 
174 {
175  int32_t mean = 0;
176  int chan, i;
177  int nblock = FFMAX(1, s->nmean);
178  /* initialise offset */
179  switch (s->internal_ftype) {
180  case TYPE_U8:
182  mean = 0x80;
183  break;
184  case TYPE_S16HL:
185  case TYPE_S16LH:
187  break;
188  default:
189  av_log(s->avctx, AV_LOG_ERROR, "unknown audio type\n");
190  return AVERROR_PATCHWELCOME;
191  }
192 
193  for (chan = 0; chan < s->channels; chan++)
194  for (i = 0; i < nblock; i++)
195  s->offset[chan][i] = mean;
196  return 0;
197 }
198 
200  int header_size)
201 {
202  int len, bps;
203  short wave_format;
204  GetByteContext gb;
205 
206  bytestream2_init(&gb, header, header_size);
207 
208  if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
209  av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
210  return AVERROR_INVALIDDATA;
211  }
212 
213  bytestream2_skip(&gb, 4); /* chunk size */
214 
215  if (bytestream2_get_le32(&gb) != MKTAG('W', 'A', 'V', 'E')) {
216  av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
217  return AVERROR_INVALIDDATA;
218  }
219 
220  while (bytestream2_get_le32(&gb) != MKTAG('f', 'm', 't', ' ')) {
221  len = bytestream2_get_le32(&gb);
222  bytestream2_skip(&gb, len);
223  if (len < 0 || bytestream2_get_bytes_left(&gb) < 16) {
224  av_log(avctx, AV_LOG_ERROR, "no fmt chunk found\n");
225  return AVERROR_INVALIDDATA;
226  }
227  }
228  len = bytestream2_get_le32(&gb);
229 
230  if (len < 16) {
231  av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
232  return AVERROR_INVALIDDATA;
233  }
234 
235  wave_format = bytestream2_get_le16(&gb);
236 
237  switch (wave_format) {
238  case WAVE_FORMAT_PCM:
239  break;
240  default:
241  av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
242  return AVERROR(ENOSYS);
243  }
244 
245  bytestream2_skip(&gb, 2); // skip channels (already got from shorten header)
246  avctx->sample_rate = bytestream2_get_le32(&gb);
247  bytestream2_skip(&gb, 4); // skip bit rate (represents original uncompressed bit rate)
248  bytestream2_skip(&gb, 2); // skip block align (not needed)
249  bps = bytestream2_get_le16(&gb);
250  avctx->bits_per_coded_sample = bps;
251 
252  if (bps != 16 && bps != 8) {
253  av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample: %d\n", bps);
254  return AVERROR(ENOSYS);
255  }
256 
257  len -= 16;
258  if (len > 0)
259  av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
260 
261  return 0;
262 }
263 
264 static const int fixed_coeffs[][3] = {
265  { 0, 0, 0 },
266  { 1, 0, 0 },
267  { 2, -1, 0 },
268  { 3, -3, 1 }
269 };
270 
271 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
272  int residual_size, int32_t coffset)
273 {
274  int pred_order, sum, qshift, init_sum, i, j;
275  const int *coeffs;
276 
277  if (command == FN_QLPC) {
278  /* read/validate prediction order */
279  pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
280  if ((unsigned)pred_order > s->nwrap) {
281  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
282  pred_order);
283  return AVERROR(EINVAL);
284  }
285  /* read LPC coefficients */
286  for (i = 0; i < pred_order; i++)
287  s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
288  coeffs = s->coeffs;
289 
290  qshift = LPCQUANT;
291  } else {
292  /* fixed LPC coeffs */
293  pred_order = command;
294  if (pred_order >= FF_ARRAY_ELEMS(fixed_coeffs)) {
295  av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
296  pred_order);
297  return AVERROR_INVALIDDATA;
298  }
299  coeffs = fixed_coeffs[pred_order];
300  qshift = 0;
301  }
302 
303  /* subtract offset from previous samples to use in prediction */
304  if (command == FN_QLPC && coffset)
305  for (i = -pred_order; i < 0; i++)
306  s->decoded[channel][i] -= coffset;
307 
308  /* decode residual and do LPC prediction */
309  init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
310  for (i = 0; i < s->blocksize; i++) {
311  sum = init_sum;
312  for (j = 0; j < pred_order; j++)
313  sum += coeffs[j] * s->decoded[channel][i - j - 1];
314  s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
315  (sum >> qshift);
316  }
317 
318  /* add offset to current samples */
319  if (command == FN_QLPC && coffset)
320  for (i = 0; i < s->blocksize; i++)
321  s->decoded[channel][i] += coffset;
322 
323  return 0;
324 }
325 
327 {
328  int i, ret;
329  int maxnlpc = 0;
330  /* shorten signature */
331  if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
332  av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
333  return AVERROR_INVALIDDATA;
334  }
335 
336  s->lpcqoffset = 0;
338  s->nmean = -1;
339  s->version = get_bits(&s->gb, 8);
341 
342  s->channels = get_uint(s, CHANSIZE);
343  if (!s->channels) {
344  av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
345  return AVERROR_INVALIDDATA;
346  }
347  if (s->channels > MAX_CHANNELS) {
348  av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
349  s->channels = 0;
350  return AVERROR_INVALIDDATA;
351  }
352  s->avctx->channels = s->channels;
353 
354  /* get blocksize if version > 0 */
355  if (s->version > 0) {
356  int skip_bytes;
357  unsigned blocksize;
358 
359  blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
360  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
362  "invalid or unsupported block size: %d\n",
363  blocksize);
364  return AVERROR(EINVAL);
365  }
366  s->blocksize = blocksize;
367 
368  maxnlpc = get_uint(s, LPCQSIZE);
369  s->nmean = get_uint(s, 0);
370 
371  skip_bytes = get_uint(s, NSKIPSIZE);
372  if ((unsigned)skip_bytes > get_bits_left(&s->gb)/8) {
373  av_log(s->avctx, AV_LOG_ERROR, "invalid skip_bytes: %d\n", skip_bytes);
374  return AVERROR_INVALIDDATA;
375  }
376 
377  for (i = 0; i < skip_bytes; i++)
378  skip_bits(&s->gb, 8);
379  }
380  s->nwrap = FFMAX(NWRAP, maxnlpc);
381 
382  if ((ret = allocate_buffers(s)) < 0)
383  return ret;
384 
385  if ((ret = init_offset(s)) < 0)
386  return ret;
387 
388  if (s->version > 1)
390 
393  "missing verbatim section at beginning of stream\n");
394  return AVERROR_INVALIDDATA;
395  }
396 
398  if (s->header_size >= OUT_BUFFER_SIZE ||
400  av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
401  s->header_size);
402  return AVERROR_INVALIDDATA;
403  }
404 
405  for (i = 0; i < s->header_size; i++)
407 
408  if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
409  return ret;
410 
411  s->cur_chan = 0;
412  s->bitshift = 0;
413 
414  s->got_header = 1;
415 
416  return 0;
417 }
418 
419 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
420  int *got_frame_ptr, AVPacket *avpkt)
421 {
422  AVFrame *frame = data;
423  const uint8_t *buf = avpkt->data;
424  int buf_size = avpkt->size;
425  ShortenContext *s = avctx->priv_data;
426  int i, input_buf_size = 0;
427  int ret;
428 
429  /* allocate internal bitstream buffer */
430  if (s->max_framesize == 0) {
431  void *tmp_ptr;
432  s->max_framesize = 8192; // should hopefully be enough for the first header
435  if (!tmp_ptr) {
436  av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
437  return AVERROR(ENOMEM);
438  }
439  memset(tmp_ptr, 0, s->allocated_bitstream_size);
440  s->bitstream = tmp_ptr;
441  }
442 
443  /* append current packet data to bitstream buffer */
444  if (1 && s->max_framesize) { //FIXME truncated
445  buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
446  input_buf_size = buf_size;
447 
450  memmove(s->bitstream, &s->bitstream[s->bitstream_index],
451  s->bitstream_size);
452  s->bitstream_index = 0;
453  }
454  if (buf)
455  memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
456  buf_size);
457  buf = &s->bitstream[s->bitstream_index];
458  buf_size += s->bitstream_size;
459  s->bitstream_size = buf_size;
460 
461  /* do not decode until buffer has at least max_framesize bytes or
462  * the end of the file has been reached */
463  if (buf_size < s->max_framesize && avpkt->data) {
464  *got_frame_ptr = 0;
465  return input_buf_size;
466  }
467  }
468  /* init and position bitstream reader */
469  init_get_bits(&s->gb, buf, buf_size * 8);
470  skip_bits(&s->gb, s->bitindex);
471 
472  /* process header or next subblock */
473  if (!s->got_header) {
474  if ((ret = read_header(s)) < 0)
475  return ret;
476  *got_frame_ptr = 0;
477  goto finish_frame;
478  }
479 
480  /* if quit command was read previously, don't decode anything */
481  if (s->got_quit_command) {
482  *got_frame_ptr = 0;
483  return avpkt->size;
484  }
485 
486  s->cur_chan = 0;
487  while (s->cur_chan < s->channels) {
488  unsigned cmd;
489  int len;
490 
491  if (get_bits_left(&s->gb) < 3 + FNSIZE) {
492  *got_frame_ptr = 0;
493  break;
494  }
495 
496  cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
497 
498  if (cmd > FN_VERBATIM) {
499  av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
500  *got_frame_ptr = 0;
501  break;
502  }
503 
504  if (!is_audio_command[cmd]) {
505  /* process non-audio command */
506  switch (cmd) {
507  case FN_VERBATIM:
509  while (len--)
511  break;
512  case FN_BITSHIFT: {
513  unsigned bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
514  if (bitshift > 31) {
515  av_log(avctx, AV_LOG_ERROR, "bitshift %d is invalid\n",
516  bitshift);
517  return AVERROR_INVALIDDATA;
518  }
519  s->bitshift = bitshift;
520  break;
521  }
522  case FN_BLOCKSIZE: {
523  unsigned blocksize = get_uint(s, av_log2(s->blocksize));
524  if (blocksize > s->blocksize) {
525  av_log(avctx, AV_LOG_ERROR,
526  "Increasing block size is not supported\n");
527  return AVERROR_PATCHWELCOME;
528  }
529  if (!blocksize || blocksize > MAX_BLOCKSIZE) {
530  av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
531  "block size: %d\n", blocksize);
532  return AVERROR(EINVAL);
533  }
534  s->blocksize = blocksize;
535  break;
536  }
537  case FN_QUIT:
538  s->got_quit_command = 1;
539  break;
540  }
541  if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
542  *got_frame_ptr = 0;
543  break;
544  }
545  } else {
546  /* process audio command */
547  int residual_size = 0;
548  int channel = s->cur_chan;
549  int32_t coffset;
550 
551  /* get Rice code for residual decoding */
552  if (cmd != FN_ZERO) {
553  residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
554  /* This is a hack as version 0 differed in the definition
555  * of get_sr_golomb_shorten(). */
556  if (s->version == 0)
557  residual_size--;
558  }
559 
560  /* calculate sample offset using means from previous blocks */
561  if (s->nmean == 0)
562  coffset = s->offset[channel][0];
563  else {
564  int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
565  for (i = 0; i < s->nmean; i++)
566  sum += s->offset[channel][i];
567  coffset = sum / s->nmean;
568  if (s->version >= 2)
569  coffset = s->bitshift == 0 ? coffset : coffset >> s->bitshift - 1 >> 1;
570  }
571 
572  /* decode samples for this channel */
573  if (cmd == FN_ZERO) {
574  for (i = 0; i < s->blocksize; i++)
575  s->decoded[channel][i] = 0;
576  } else {
577  if ((ret = decode_subframe_lpc(s, cmd, channel,
578  residual_size, coffset)) < 0)
579  return ret;
580  }
581 
582  /* update means with info from the current block */
583  if (s->nmean > 0) {
584  int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
585  for (i = 0; i < s->blocksize; i++)
586  sum += s->decoded[channel][i];
587 
588  for (i = 1; i < s->nmean; i++)
589  s->offset[channel][i - 1] = s->offset[channel][i];
590 
591  if (s->version < 2)
592  s->offset[channel][s->nmean - 1] = sum / s->blocksize;
593  else
594  s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
595  }
596 
597  /* copy wrap samples for use with next block */
598  for (i = -s->nwrap; i < 0; i++)
599  s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
600 
601  /* shift samples to add in unused zero bits which were removed
602  * during encoding */
603  fix_bitshift(s, s->decoded[channel]);
604 
605  /* if this is the last channel in the block, output the samples */
606  s->cur_chan++;
607  if (s->cur_chan == s->channels) {
608  uint8_t *samples_u8;
609  int16_t *samples_s16;
610  int chan;
611 
612  /* get output buffer */
613  frame->nb_samples = s->blocksize;
614  if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
615  return ret;
616 
617  for (chan = 0; chan < s->channels; chan++) {
618  samples_u8 = ((uint8_t **)frame->extended_data)[chan];
619  samples_s16 = ((int16_t **)frame->extended_data)[chan];
620  for (i = 0; i < s->blocksize; i++) {
621  switch (s->internal_ftype) {
622  case TYPE_U8:
623  *samples_u8++ = av_clip_uint8(s->decoded[chan][i]);
624  break;
625  case TYPE_S16HL:
626  case TYPE_S16LH:
627  *samples_s16++ = av_clip_int16(s->decoded[chan][i]);
628  break;
629  }
630  }
631  }
632 
633  *got_frame_ptr = 1;
634  }
635  }
636  }
637  if (s->cur_chan < s->channels)
638  *got_frame_ptr = 0;
639 
641  s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
642  i = get_bits_count(&s->gb) / 8;
643  if (i > buf_size) {
644  av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
645  s->bitstream_size = 0;
646  s->bitstream_index = 0;
647  return AVERROR_INVALIDDATA;
648  }
649  if (s->bitstream_size) {
650  s->bitstream_index += i;
651  s->bitstream_size -= i;
652  return input_buf_size;
653  } else
654  return i;
655 }
656 
658 {
659  ShortenContext *s = avctx->priv_data;
660  int i;
661 
662  for (i = 0; i < s->channels; i++) {
663  s->decoded[i] = NULL;
664  av_freep(&s->decoded_base[i]);
665  av_freep(&s->offset[i]);
666  }
667  av_freep(&s->bitstream);
668  av_freep(&s->coeffs);
669 
670  return 0;
671 }
672 
674  .name = "shorten",
675  .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
676  .type = AVMEDIA_TYPE_AUDIO,
677  .id = AV_CODEC_ID_SHORTEN,
678  .priv_data_size = sizeof(ShortenContext),
680  .close = shorten_decode_close,
682  .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
683  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
686 };