FFmpeg
opus.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012 Andrew D'Addesio
3  * Copyright (c) 2013-2014 Mozilla Corporation
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  * Opus decoder/parser shared code
25  */
26 
27 #include <stdint.h>
28 
29 #include "libavutil/error.h"
30 #include "libavutil/ffmath.h"
31 
32 #include "opus_celt.h"
33 #include "opustab.h"
34 #include "internal.h"
35 #include "vorbis.h"
36 
37 static const uint16_t opus_frame_duration[32] = {
38  480, 960, 1920, 2880,
39  480, 960, 1920, 2880,
40  480, 960, 1920, 2880,
41  480, 960,
42  480, 960,
43  120, 240, 480, 960,
44  120, 240, 480, 960,
45  120, 240, 480, 960,
46  120, 240, 480, 960,
47 };
48 
49 /**
50  * Read a 1- or 2-byte frame length
51  */
52 static inline int xiph_lacing_16bit(const uint8_t **ptr, const uint8_t *end)
53 {
54  int val;
55 
56  if (*ptr >= end)
57  return AVERROR_INVALIDDATA;
58  val = *(*ptr)++;
59  if (val >= 252) {
60  if (*ptr >= end)
61  return AVERROR_INVALIDDATA;
62  val += 4 * *(*ptr)++;
63  }
64  return val;
65 }
66 
67 /**
68  * Read a multi-byte length (used for code 3 packet padding size)
69  */
70 static inline int xiph_lacing_full(const uint8_t **ptr, const uint8_t *end)
71 {
72  int val = 0;
73  int next;
74 
75  while (1) {
76  if (*ptr >= end || val > INT_MAX - 254)
77  return AVERROR_INVALIDDATA;
78  next = *(*ptr)++;
79  val += next;
80  if (next < 255)
81  break;
82  else
83  val--;
84  }
85  return val;
86 }
87 
88 /**
89  * Parse Opus packet info from raw packet data
90  */
91 int ff_opus_parse_packet(OpusPacket *pkt, const uint8_t *buf, int buf_size,
92  int self_delimiting)
93 {
94  const uint8_t *ptr = buf;
95  const uint8_t *end = buf + buf_size;
96  int padding = 0;
97  int frame_bytes, i;
98 
99  if (buf_size < 1)
100  goto fail;
101 
102  /* TOC byte */
103  i = *ptr++;
104  pkt->code = (i ) & 0x3;
105  pkt->stereo = (i >> 2) & 0x1;
106  pkt->config = (i >> 3) & 0x1F;
107 
108  /* code 2 and code 3 packets have at least 1 byte after the TOC */
109  if (pkt->code >= 2 && buf_size < 2)
110  goto fail;
111 
112  switch (pkt->code) {
113  case 0:
114  /* 1 frame */
115  pkt->frame_count = 1;
116  pkt->vbr = 0;
117 
118  if (self_delimiting) {
119  int len = xiph_lacing_16bit(&ptr, end);
120  if (len < 0 || len > end - ptr)
121  goto fail;
122  end = ptr + len;
123  buf_size = end - buf;
124  }
125 
126  frame_bytes = end - ptr;
127  if (frame_bytes > MAX_FRAME_SIZE)
128  goto fail;
129  pkt->frame_offset[0] = ptr - buf;
130  pkt->frame_size[0] = frame_bytes;
131  break;
132  case 1:
133  /* 2 frames, equal size */
134  pkt->frame_count = 2;
135  pkt->vbr = 0;
136 
137  if (self_delimiting) {
138  int len = xiph_lacing_16bit(&ptr, end);
139  if (len < 0 || 2 * len > end - ptr)
140  goto fail;
141  end = ptr + 2 * len;
142  buf_size = end - buf;
143  }
144 
145  frame_bytes = end - ptr;
146  if (frame_bytes & 1 || frame_bytes >> 1 > MAX_FRAME_SIZE)
147  goto fail;
148  pkt->frame_offset[0] = ptr - buf;
149  pkt->frame_size[0] = frame_bytes >> 1;
150  pkt->frame_offset[1] = pkt->frame_offset[0] + pkt->frame_size[0];
151  pkt->frame_size[1] = frame_bytes >> 1;
152  break;
153  case 2:
154  /* 2 frames, different sizes */
155  pkt->frame_count = 2;
156  pkt->vbr = 1;
157 
158  /* read 1st frame size */
159  frame_bytes = xiph_lacing_16bit(&ptr, end);
160  if (frame_bytes < 0)
161  goto fail;
162 
163  if (self_delimiting) {
164  int len = xiph_lacing_16bit(&ptr, end);
165  if (len < 0 || len + frame_bytes > end - ptr)
166  goto fail;
167  end = ptr + frame_bytes + len;
168  buf_size = end - buf;
169  }
170 
171  pkt->frame_offset[0] = ptr - buf;
172  pkt->frame_size[0] = frame_bytes;
173 
174  /* calculate 2nd frame size */
175  frame_bytes = end - ptr - pkt->frame_size[0];
176  if (frame_bytes < 0 || frame_bytes > MAX_FRAME_SIZE)
177  goto fail;
178  pkt->frame_offset[1] = pkt->frame_offset[0] + pkt->frame_size[0];
179  pkt->frame_size[1] = frame_bytes;
180  break;
181  case 3:
182  /* 1 to 48 frames, can be different sizes */
183  i = *ptr++;
184  pkt->frame_count = (i ) & 0x3F;
185  padding = (i >> 6) & 0x01;
186  pkt->vbr = (i >> 7) & 0x01;
187 
188  if (pkt->frame_count == 0 || pkt->frame_count > MAX_FRAMES)
189  goto fail;
190 
191  /* read padding size */
192  if (padding) {
193  padding = xiph_lacing_full(&ptr, end);
194  if (padding < 0)
195  goto fail;
196  }
197 
198  /* read frame sizes */
199  if (pkt->vbr) {
200  /* for VBR, all frames except the final one have their size coded
201  in the bitstream. the last frame size is implicit. */
202  int total_bytes = 0;
203  for (i = 0; i < pkt->frame_count - 1; i++) {
204  frame_bytes = xiph_lacing_16bit(&ptr, end);
205  if (frame_bytes < 0)
206  goto fail;
207  pkt->frame_size[i] = frame_bytes;
208  total_bytes += frame_bytes;
209  }
210 
211  if (self_delimiting) {
212  int len = xiph_lacing_16bit(&ptr, end);
213  if (len < 0 || len + total_bytes + padding > end - ptr)
214  goto fail;
215  end = ptr + total_bytes + len + padding;
216  buf_size = end - buf;
217  }
218 
219  frame_bytes = end - ptr - padding;
220  if (total_bytes > frame_bytes)
221  goto fail;
222  pkt->frame_offset[0] = ptr - buf;
223  for (i = 1; i < pkt->frame_count; i++)
224  pkt->frame_offset[i] = pkt->frame_offset[i-1] + pkt->frame_size[i-1];
225  pkt->frame_size[pkt->frame_count-1] = frame_bytes - total_bytes;
226  } else {
227  /* for CBR, the remaining packet bytes are divided evenly between
228  the frames */
229  if (self_delimiting) {
230  frame_bytes = xiph_lacing_16bit(&ptr, end);
231  if (frame_bytes < 0 || pkt->frame_count * frame_bytes + padding > end - ptr)
232  goto fail;
233  end = ptr + pkt->frame_count * frame_bytes + padding;
234  buf_size = end - buf;
235  } else {
236  frame_bytes = end - ptr - padding;
237  if (frame_bytes % pkt->frame_count ||
238  frame_bytes / pkt->frame_count > MAX_FRAME_SIZE)
239  goto fail;
240  frame_bytes /= pkt->frame_count;
241  }
242 
243  pkt->frame_offset[0] = ptr - buf;
244  pkt->frame_size[0] = frame_bytes;
245  for (i = 1; i < pkt->frame_count; i++) {
246  pkt->frame_offset[i] = pkt->frame_offset[i-1] + pkt->frame_size[i-1];
247  pkt->frame_size[i] = frame_bytes;
248  }
249  }
250  }
251 
252  pkt->packet_size = buf_size;
253  pkt->data_size = pkt->packet_size - padding;
254 
255  /* total packet duration cannot be larger than 120ms */
256  pkt->frame_duration = opus_frame_duration[pkt->config];
257  if (pkt->frame_duration * pkt->frame_count > MAX_PACKET_DUR)
258  goto fail;
259 
260  /* set mode and bandwidth */
261  if (pkt->config < 12) {
262  pkt->mode = OPUS_MODE_SILK;
263  pkt->bandwidth = pkt->config >> 2;
264  } else if (pkt->config < 16) {
265  pkt->mode = OPUS_MODE_HYBRID;
266  pkt->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND + (pkt->config >= 14);
267  } else {
268  pkt->mode = OPUS_MODE_CELT;
269  pkt->bandwidth = (pkt->config - 16) >> 2;
270  /* skip medium band */
271  if (pkt->bandwidth)
272  pkt->bandwidth++;
273  }
274 
275  return 0;
276 
277 fail:
278  memset(pkt, 0, sizeof(*pkt));
279  return AVERROR_INVALIDDATA;
280 }
281 
282 static int channel_reorder_vorbis(int nb_channels, int channel_idx)
283 {
284  return ff_vorbis_channel_layout_offsets[nb_channels - 1][channel_idx];
285 }
286 
287 static int channel_reorder_unknown(int nb_channels, int channel_idx)
288 {
289  return channel_idx;
290 }
291 
293  OpusContext *s)
294 {
295  static const uint8_t default_channel_map[2] = { 0, 1 };
296 
297  int (*channel_reorder)(int, int) = channel_reorder_unknown;
298 
299  const uint8_t *extradata, *channel_map;
300  int extradata_size;
301  int version, channels, map_type, streams, stereo_streams, i, j;
302  uint64_t layout;
303 
304  if (!avctx->extradata) {
305  if (avctx->channels > 2) {
306  av_log(avctx, AV_LOG_ERROR,
307  "Multichannel configuration without extradata.\n");
308  return AVERROR(EINVAL);
309  }
310  extradata = opus_default_extradata;
311  extradata_size = sizeof(opus_default_extradata);
312  } else {
313  extradata = avctx->extradata;
314  extradata_size = avctx->extradata_size;
315  }
316 
317  if (extradata_size < 19) {
318  av_log(avctx, AV_LOG_ERROR, "Invalid extradata size: %d\n",
319  extradata_size);
320  return AVERROR_INVALIDDATA;
321  }
322 
323  version = extradata[8];
324  if (version > 15) {
325  avpriv_request_sample(avctx, "Extradata version %d", version);
326  return AVERROR_PATCHWELCOME;
327  }
328 
329  avctx->delay = AV_RL16(extradata + 10);
330  if (avctx->internal)
331  avctx->internal->skip_samples = avctx->delay;
332 
333  channels = avctx->extradata ? extradata[9] : (avctx->channels == 1) ? 1 : 2;
334  if (!channels) {
335  av_log(avctx, AV_LOG_ERROR, "Zero channel count specified in the extradata\n");
336  return AVERROR_INVALIDDATA;
337  }
338 
339  s->gain_i = AV_RL16(extradata + 16);
340  if (s->gain_i)
341  s->gain = ff_exp10(s->gain_i / (20.0 * 256));
342 
343  map_type = extradata[18];
344  if (!map_type) {
345  if (channels > 2) {
346  av_log(avctx, AV_LOG_ERROR,
347  "Channel mapping 0 is only specified for up to 2 channels\n");
348  return AVERROR_INVALIDDATA;
349  }
351  streams = 1;
352  stereo_streams = channels - 1;
353  channel_map = default_channel_map;
354  } else if (map_type == 1 || map_type == 2 || map_type == 255) {
355  if (extradata_size < 21 + channels) {
356  av_log(avctx, AV_LOG_ERROR, "Invalid extradata size: %d\n",
357  extradata_size);
358  return AVERROR_INVALIDDATA;
359  }
360 
361  streams = extradata[19];
362  stereo_streams = extradata[20];
363  if (!streams || stereo_streams > streams ||
364  streams + stereo_streams > 255) {
365  av_log(avctx, AV_LOG_ERROR,
366  "Invalid stream/stereo stream count: %d/%d\n", streams, stereo_streams);
367  return AVERROR_INVALIDDATA;
368  }
369 
370  if (map_type == 1) {
371  if (channels > 8) {
372  av_log(avctx, AV_LOG_ERROR,
373  "Channel mapping 1 is only specified for up to 8 channels\n");
374  return AVERROR_INVALIDDATA;
375  }
377  channel_reorder = channel_reorder_vorbis;
378  } else if (map_type == 2) {
379  int ambisonic_order = ff_sqrt(channels) - 1;
380  if (channels != ((ambisonic_order + 1) * (ambisonic_order + 1)) &&
381  channels != ((ambisonic_order + 1) * (ambisonic_order + 1) + 2)) {
382  av_log(avctx, AV_LOG_ERROR,
383  "Channel mapping 2 is only specified for channel counts"
384  " which can be written as (n + 1)^2 or (n + 1)^2 + 2"
385  " for nonnegative integer n\n");
386  return AVERROR_INVALIDDATA;
387  }
388  if (channels > 227) {
389  av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
390  return AVERROR_INVALIDDATA;
391  }
392  layout = 0;
393  } else
394  layout = 0;
395 
396  channel_map = extradata + 21;
397  } else {
398  avpriv_request_sample(avctx, "Mapping type %d", map_type);
399  return AVERROR_PATCHWELCOME;
400  }
401 
402  s->channel_maps = av_mallocz_array(channels, sizeof(*s->channel_maps));
403  if (!s->channel_maps)
404  return AVERROR(ENOMEM);
405 
406  for (i = 0; i < channels; i++) {
407  ChannelMap *map = &s->channel_maps[i];
408  uint8_t idx = channel_map[channel_reorder(channels, i)];
409 
410  if (idx == 255) {
411  map->silence = 1;
412  continue;
413  } else if (idx >= streams + stereo_streams) {
414  av_log(avctx, AV_LOG_ERROR,
415  "Invalid channel map for output channel %d: %d\n", i, idx);
416  av_freep(&s->channel_maps);
417  return AVERROR_INVALIDDATA;
418  }
419 
420  /* check that we did not see this index yet */
421  map->copy = 0;
422  for (j = 0; j < i; j++)
423  if (channel_map[channel_reorder(channels, j)] == idx) {
424  map->copy = 1;
425  map->copy_idx = j;
426  break;
427  }
428 
429  if (idx < 2 * stereo_streams) {
430  map->stream_idx = idx / 2;
431  map->channel_idx = idx & 1;
432  } else {
433  map->stream_idx = idx - stereo_streams;
434  map->channel_idx = 0;
435  }
436  }
437 
438  avctx->channels = channels;
439  avctx->channel_layout = layout;
440  s->nb_streams = streams;
441  s->nb_stereo_streams = stereo_streams;
442 
443  return 0;
444 }
445 
447 {
448  float lowband_scratch[8 * 22];
449  float norm1[2 * 8 * 100];
450  float *norm2 = norm1 + 8 * 100;
451 
452  int totalbits = (f->framebits << 3) - f->anticollapse_needed;
453 
454  int update_lowband = 1;
455  int lowband_offset = 0;
456 
457  int i, j;
458 
459  for (i = f->start_band; i < f->end_band; i++) {
460  uint32_t cm[2] = { (1 << f->blocks) - 1, (1 << f->blocks) - 1 };
461  int band_offset = ff_celt_freq_bands[i] << f->size;
462  int band_size = ff_celt_freq_range[i] << f->size;
463  float *X = f->block[0].coeffs + band_offset;
464  float *Y = (f->channels == 2) ? f->block[1].coeffs + band_offset : NULL;
465  float *norm_loc1, *norm_loc2;
466 
467  int consumed = opus_rc_tell_frac(rc);
468  int effective_lowband = -1;
469  int b = 0;
470 
471  /* Compute how many bits we want to allocate to this band */
472  if (i != f->start_band)
473  f->remaining -= consumed;
474  f->remaining2 = totalbits - consumed - 1;
475  if (i <= f->coded_bands - 1) {
476  int curr_balance = f->remaining / FFMIN(3, f->coded_bands-i);
477  b = av_clip_uintp2(FFMIN(f->remaining2 + 1, f->pulses[i] + curr_balance), 14);
478  }
479 
480  if ((ff_celt_freq_bands[i] - ff_celt_freq_range[i] >= ff_celt_freq_bands[f->start_band] ||
481  i == f->start_band + 1) && (update_lowband || lowband_offset == 0))
482  lowband_offset = i;
483 
484  if (i == f->start_band + 1) {
485  /* Special Hybrid Folding (RFC 8251 section 9). Copy the first band into
486  the second to ensure the second band never has to use the LCG. */
487  int count = (ff_celt_freq_range[i] - ff_celt_freq_range[i-1]) << f->size;
488 
489  memcpy(&norm1[band_offset], &norm1[band_offset - count], count * sizeof(float));
490 
491  if (f->channels == 2)
492  memcpy(&norm2[band_offset], &norm2[band_offset - count], count * sizeof(float));
493  }
494 
495  /* Get a conservative estimate of the collapse_mask's for the bands we're
496  going to be folding from. */
497  if (lowband_offset != 0 && (f->spread != CELT_SPREAD_AGGRESSIVE ||
498  f->blocks > 1 || f->tf_change[i] < 0)) {
499  int foldstart, foldend;
500 
501  /* This ensures we never repeat spectral content within one band */
502  effective_lowband = FFMAX(ff_celt_freq_bands[f->start_band],
503  ff_celt_freq_bands[lowband_offset] - ff_celt_freq_range[i]);
504  foldstart = lowband_offset;
505  while (ff_celt_freq_bands[--foldstart] > effective_lowband);
506  foldend = lowband_offset - 1;
507  while (++foldend < i && ff_celt_freq_bands[foldend] < effective_lowband + ff_celt_freq_range[i]);
508 
509  cm[0] = cm[1] = 0;
510  for (j = foldstart; j < foldend; j++) {
511  cm[0] |= f->block[0].collapse_masks[j];
512  cm[1] |= f->block[f->channels - 1].collapse_masks[j];
513  }
514  }
515 
516  if (f->dual_stereo && i == f->intensity_stereo) {
517  /* Switch off dual stereo to do intensity */
518  f->dual_stereo = 0;
519  for (j = ff_celt_freq_bands[f->start_band] << f->size; j < band_offset; j++)
520  norm1[j] = (norm1[j] + norm2[j]) / 2;
521  }
522 
523  norm_loc1 = effective_lowband != -1 ? norm1 + (effective_lowband << f->size) : NULL;
524  norm_loc2 = effective_lowband != -1 ? norm2 + (effective_lowband << f->size) : NULL;
525 
526  if (f->dual_stereo) {
527  cm[0] = f->pvq->quant_band(f->pvq, f, rc, i, X, NULL, band_size, b >> 1,
528  f->blocks, norm_loc1, f->size,
529  norm1 + band_offset, 0, 1.0f,
530  lowband_scratch, cm[0]);
531 
532  cm[1] = f->pvq->quant_band(f->pvq, f, rc, i, Y, NULL, band_size, b >> 1,
533  f->blocks, norm_loc2, f->size,
534  norm2 + band_offset, 0, 1.0f,
535  lowband_scratch, cm[1]);
536  } else {
537  cm[0] = f->pvq->quant_band(f->pvq, f, rc, i, X, Y, band_size, b >> 0,
538  f->blocks, norm_loc1, f->size,
539  norm1 + band_offset, 0, 1.0f,
540  lowband_scratch, cm[0] | cm[1]);
541  cm[1] = cm[0];
542  }
543 
544  f->block[0].collapse_masks[i] = (uint8_t)cm[0];
545  f->block[f->channels - 1].collapse_masks[i] = (uint8_t)cm[1];
546  f->remaining += f->pulses[i] + consumed;
547 
548  /* Update the folding position only as long as we have 1 bit/sample depth */
549  update_lowband = (b > band_size << 3);
550  }
551 }
552 
553 #define NORMC(bits) ((bits) << (f->channels - 1) << f->size >> 2)
554 
556 {
557  int i, j, low, high, total, done, bandbits, remaining, tbits_8ths;
558  int skip_startband = f->start_band;
559  int skip_bit = 0;
560  int intensitystereo_bit = 0;
561  int dualstereo_bit = 0;
562  int dynalloc = 6;
563  int extrabits = 0;
564 
565  int boost[CELT_MAX_BANDS] = { 0 };
566  int trim_offset[CELT_MAX_BANDS];
567  int threshold[CELT_MAX_BANDS];
568  int bits1[CELT_MAX_BANDS];
569  int bits2[CELT_MAX_BANDS];
570 
571  /* Spread */
572  if (opus_rc_tell(rc) + 4 <= f->framebits) {
573  if (encode)
575  else
577  } else {
578  f->spread = CELT_SPREAD_NORMAL;
579  }
580 
581  /* Initialize static allocation caps */
582  for (i = 0; i < CELT_MAX_BANDS; i++)
583  f->caps[i] = NORMC((ff_celt_static_caps[f->size][f->channels - 1][i] + 64) * ff_celt_freq_range[i]);
584 
585  /* Band boosts */
586  tbits_8ths = f->framebits << 3;
587  for (i = f->start_band; i < f->end_band; i++) {
588  int quanta = ff_celt_freq_range[i] << (f->channels - 1) << f->size;
589  int b_dynalloc = dynalloc;
590  int boost_amount = f->alloc_boost[i];
591  quanta = FFMIN(quanta << 3, FFMAX(6 << 3, quanta));
592 
593  while (opus_rc_tell_frac(rc) + (b_dynalloc << 3) < tbits_8ths && boost[i] < f->caps[i]) {
594  int is_boost;
595  if (encode) {
596  is_boost = boost_amount--;
597  ff_opus_rc_enc_log(rc, is_boost, b_dynalloc);
598  } else {
599  is_boost = ff_opus_rc_dec_log(rc, b_dynalloc);
600  }
601 
602  if (!is_boost)
603  break;
604 
605  boost[i] += quanta;
606  tbits_8ths -= quanta;
607 
608  b_dynalloc = 1;
609  }
610 
611  if (boost[i])
612  dynalloc = FFMAX(dynalloc - 1, 2);
613  }
614 
615  /* Allocation trim */
616  if (!encode)
617  f->alloc_trim = 5;
618  if (opus_rc_tell_frac(rc) + (6 << 3) <= tbits_8ths)
619  if (encode)
621  else
622  f->alloc_trim = ff_opus_rc_dec_cdf(rc, ff_celt_model_alloc_trim);
623 
624  /* Anti-collapse bit reservation */
625  tbits_8ths = (f->framebits << 3) - opus_rc_tell_frac(rc) - 1;
626  f->anticollapse_needed = 0;
627  if (f->transient && f->size >= 2 && tbits_8ths >= ((f->size + 2) << 3))
628  f->anticollapse_needed = 1 << 3;
629  tbits_8ths -= f->anticollapse_needed;
630 
631  /* Band skip bit reservation */
632  if (tbits_8ths >= 1 << 3)
633  skip_bit = 1 << 3;
634  tbits_8ths -= skip_bit;
635 
636  /* Intensity/dual stereo bit reservation */
637  if (f->channels == 2) {
638  intensitystereo_bit = ff_celt_log2_frac[f->end_band - f->start_band];
639  if (intensitystereo_bit <= tbits_8ths) {
640  tbits_8ths -= intensitystereo_bit;
641  if (tbits_8ths >= 1 << 3) {
642  dualstereo_bit = 1 << 3;
643  tbits_8ths -= 1 << 3;
644  }
645  } else {
646  intensitystereo_bit = 0;
647  }
648  }
649 
650  /* Trim offsets */
651  for (i = f->start_band; i < f->end_band; i++) {
652  int trim = f->alloc_trim - 5 - f->size;
653  int band = ff_celt_freq_range[i] * (f->end_band - i - 1);
654  int duration = f->size + 3;
655  int scale = duration + f->channels - 1;
656 
657  /* PVQ minimum allocation threshold, below this value the band is
658  * skipped */
659  threshold[i] = FFMAX(3 * ff_celt_freq_range[i] << duration >> 4,
660  f->channels << 3);
661 
662  trim_offset[i] = trim * (band << scale) >> 6;
663 
664  if (ff_celt_freq_range[i] << f->size == 1)
665  trim_offset[i] -= f->channels << 3;
666  }
667 
668  /* Bisection */
669  low = 1;
670  high = CELT_VECTORS - 1;
671  while (low <= high) {
672  int center = (low + high) >> 1;
673  done = total = 0;
674 
675  for (i = f->end_band - 1; i >= f->start_band; i--) {
676  bandbits = NORMC(ff_celt_freq_range[i] * ff_celt_static_alloc[center][i]);
677 
678  if (bandbits)
679  bandbits = FFMAX(bandbits + trim_offset[i], 0);
680  bandbits += boost[i];
681 
682  if (bandbits >= threshold[i] || done) {
683  done = 1;
684  total += FFMIN(bandbits, f->caps[i]);
685  } else if (bandbits >= f->channels << 3) {
686  total += f->channels << 3;
687  }
688  }
689 
690  if (total > tbits_8ths)
691  high = center - 1;
692  else
693  low = center + 1;
694  }
695  high = low--;
696 
697  /* Bisection */
698  for (i = f->start_band; i < f->end_band; i++) {
700  bits2[i] = high >= CELT_VECTORS ? f->caps[i] :
702 
703  if (bits1[i])
704  bits1[i] = FFMAX(bits1[i] + trim_offset[i], 0);
705  if (bits2[i])
706  bits2[i] = FFMAX(bits2[i] + trim_offset[i], 0);
707 
708  if (low)
709  bits1[i] += boost[i];
710  bits2[i] += boost[i];
711 
712  if (boost[i])
713  skip_startband = i;
714  bits2[i] = FFMAX(bits2[i] - bits1[i], 0);
715  }
716 
717  /* Bisection */
718  low = 0;
719  high = 1 << CELT_ALLOC_STEPS;
720  for (i = 0; i < CELT_ALLOC_STEPS; i++) {
721  int center = (low + high) >> 1;
722  done = total = 0;
723 
724  for (j = f->end_band - 1; j >= f->start_band; j--) {
725  bandbits = bits1[j] + (center * bits2[j] >> CELT_ALLOC_STEPS);
726 
727  if (bandbits >= threshold[j] || done) {
728  done = 1;
729  total += FFMIN(bandbits, f->caps[j]);
730  } else if (bandbits >= f->channels << 3)
731  total += f->channels << 3;
732  }
733  if (total > tbits_8ths)
734  high = center;
735  else
736  low = center;
737  }
738 
739  /* Bisection */
740  done = total = 0;
741  for (i = f->end_band - 1; i >= f->start_band; i--) {
742  bandbits = bits1[i] + (low * bits2[i] >> CELT_ALLOC_STEPS);
743 
744  if (bandbits >= threshold[i] || done)
745  done = 1;
746  else
747  bandbits = (bandbits >= f->channels << 3) ?
748  f->channels << 3 : 0;
749 
750  bandbits = FFMIN(bandbits, f->caps[i]);
751  f->pulses[i] = bandbits;
752  total += bandbits;
753  }
754 
755  /* Band skipping */
756  for (f->coded_bands = f->end_band; ; f->coded_bands--) {
757  int allocation;
758  j = f->coded_bands - 1;
759 
760  if (j == skip_startband) {
761  /* all remaining bands are not skipped */
762  tbits_8ths += skip_bit;
763  break;
764  }
765 
766  /* determine the number of bits available for coding "do not skip" markers */
767  remaining = tbits_8ths - total;
768  bandbits = remaining / (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]);
769  remaining -= bandbits * (ff_celt_freq_bands[j+1] - ff_celt_freq_bands[f->start_band]);
770  allocation = f->pulses[j] + bandbits * ff_celt_freq_range[j];
771  allocation += FFMAX(remaining - (ff_celt_freq_bands[j] - ff_celt_freq_bands[f->start_band]), 0);
772 
773  /* a "do not skip" marker is only coded if the allocation is
774  * above the chosen threshold */
775  if (allocation >= FFMAX(threshold[j], (f->channels + 1) << 3)) {
776  int do_not_skip;
777  if (encode) {
778  do_not_skip = f->coded_bands <= f->skip_band_floor;
779  ff_opus_rc_enc_log(rc, do_not_skip, 1);
780  } else {
781  do_not_skip = ff_opus_rc_dec_log(rc, 1);
782  }
783 
784  if (do_not_skip)
785  break;
786 
787  total += 1 << 3;
788  allocation -= 1 << 3;
789  }
790 
791  /* the band is skipped, so reclaim its bits */
792  total -= f->pulses[j];
793  if (intensitystereo_bit) {
794  total -= intensitystereo_bit;
795  intensitystereo_bit = ff_celt_log2_frac[j - f->start_band];
796  total += intensitystereo_bit;
797  }
798 
799  total += f->pulses[j] = (allocation >= f->channels << 3) ? f->channels << 3 : 0;
800  }
801 
802  /* IS start band */
803  if (encode) {
804  if (intensitystereo_bit) {
805  f->intensity_stereo = FFMIN(f->intensity_stereo, f->coded_bands);
806  ff_opus_rc_enc_uint(rc, f->intensity_stereo, f->coded_bands + 1 - f->start_band);
807  }
808  } else {
809  f->intensity_stereo = f->dual_stereo = 0;
810  if (intensitystereo_bit)
811  f->intensity_stereo = f->start_band + ff_opus_rc_dec_uint(rc, f->coded_bands + 1 - f->start_band);
812  }
813 
814  /* DS flag */
815  if (f->intensity_stereo <= f->start_band)
816  tbits_8ths += dualstereo_bit; /* no intensity stereo means no dual stereo */
817  else if (dualstereo_bit)
818  if (encode)
819  ff_opus_rc_enc_log(rc, f->dual_stereo, 1);
820  else
821  f->dual_stereo = ff_opus_rc_dec_log(rc, 1);
822 
823  /* Supply the remaining bits in this frame to lower bands */
824  remaining = tbits_8ths - total;
825  bandbits = remaining / (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]);
826  remaining -= bandbits * (ff_celt_freq_bands[f->coded_bands] - ff_celt_freq_bands[f->start_band]);
827  for (i = f->start_band; i < f->coded_bands; i++) {
828  const int bits = FFMIN(remaining, ff_celt_freq_range[i]);
829  f->pulses[i] += bits + bandbits * ff_celt_freq_range[i];
830  remaining -= bits;
831  }
832 
833  /* Finally determine the allocation */
834  for (i = f->start_band; i < f->coded_bands; i++) {
835  int N = ff_celt_freq_range[i] << f->size;
836  int prev_extra = extrabits;
837  f->pulses[i] += extrabits;
838 
839  if (N > 1) {
840  int dof; /* degrees of freedom */
841  int temp; /* dof * channels * log(dof) */
842  int fine_bits;
843  int max_bits;
844  int offset; /* fine energy quantization offset, i.e.
845  * extra bits assigned over the standard
846  * totalbits/dof */
847 
848  extrabits = FFMAX(f->pulses[i] - f->caps[i], 0);
849  f->pulses[i] -= extrabits;
850 
851  /* intensity stereo makes use of an extra degree of freedom */
852  dof = N * f->channels + (f->channels == 2 && N > 2 && !f->dual_stereo && i < f->intensity_stereo);
853  temp = dof * (ff_celt_log_freq_range[i] + (f->size << 3));
854  offset = (temp >> 1) - dof * CELT_FINE_OFFSET;
855  if (N == 2) /* dof=2 is the only case that doesn't fit the model */
856  offset += dof << 1;
857 
858  /* grant an additional bias for the first and second pulses */
859  if (f->pulses[i] + offset < 2 * (dof << 3))
860  offset += temp >> 2;
861  else if (f->pulses[i] + offset < 3 * (dof << 3))
862  offset += temp >> 3;
863 
864  fine_bits = (f->pulses[i] + offset + (dof << 2)) / (dof << 3);
865  max_bits = FFMIN((f->pulses[i] >> 3) >> (f->channels - 1), CELT_MAX_FINE_BITS);
866  max_bits = FFMAX(max_bits, 0);
867  f->fine_bits[i] = av_clip(fine_bits, 0, max_bits);
868 
869  /* If fine_bits was rounded down or capped,
870  * give priority for the final fine energy pass */
871  f->fine_priority[i] = (f->fine_bits[i] * (dof << 3) >= f->pulses[i] + offset);
872 
873  /* the remaining bits are assigned to PVQ */
874  f->pulses[i] -= f->fine_bits[i] << (f->channels - 1) << 3;
875  } else {
876  /* all bits go to fine energy except for the sign bit */
877  extrabits = FFMAX(f->pulses[i] - (f->channels << 3), 0);
878  f->pulses[i] -= extrabits;
879  f->fine_bits[i] = 0;
880  f->fine_priority[i] = 1;
881  }
882 
883  /* hand back a limited number of extra fine energy bits to this band */
884  if (extrabits > 0) {
885  int fineextra = FFMIN(extrabits >> (f->channels + 2),
886  CELT_MAX_FINE_BITS - f->fine_bits[i]);
887  f->fine_bits[i] += fineextra;
888 
889  fineextra <<= f->channels + 2;
890  f->fine_priority[i] = (fineextra >= extrabits - prev_extra);
891  extrabits -= fineextra;
892  }
893  }
894  f->remaining = extrabits;
895 
896  /* skipped bands dedicate all of their bits for fine energy */
897  for (; i < f->end_band; i++) {
898  f->fine_bits[i] = f->pulses[i] >> (f->channels - 1) >> 3;
899  f->pulses[i] = 0;
900  f->fine_priority[i] = f->fine_bits[i] < 1;
901  }
902 }
ff_opus_rc_enc_cdf
void ff_opus_rc_enc_cdf(OpusRangeCoder *rc, int val, const uint16_t *cdf)
Definition: opus_rc.c:109
ff_exp10
static av_always_inline double ff_exp10(double x)
Compute 10^x for floating point values.
Definition: ffmath.h:42
av_clip
#define av_clip
Definition: common.h:122
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
ff_celt_freq_bands
const uint8_t ff_celt_freq_bands[]
Definition: opustab.c:768
AVCodecContext::channel_layout
uint64_t channel_layout
Audio channel layout.
Definition: avcodec.h:1247
ff_opus_parse_extradata
av_cold int ff_opus_parse_extradata(AVCodecContext *avctx, OpusContext *s)
Definition: opus.c:292
opus_default_extradata
static const uint8_t opus_default_extradata[30]
Definition: opus.h:57
av_clip_uintp2
#define av_clip_uintp2
Definition: common.h:146
AV_CH_LAYOUT_MONO
#define AV_CH_LAYOUT_MONO
Definition: channel_layout.h:90
ff_opus_rc_enc_uint
void ff_opus_rc_enc_uint(OpusRangeCoder *rc, uint32_t val, uint32_t size)
CELT: write a uniformly distributed integer.
Definition: opus_rc.c:204
AVCodecInternal::skip_samples
int skip_samples
Number of audio samples to skip at the start of the next decoded frame.
Definition: internal.h:175
X
@ X
Definition: vf_addroi.c:26
internal.h
b
#define b
Definition: input.c:41
ff_celt_log_freq_range
const uint8_t ff_celt_log_freq_range[]
Definition: opustab.c:776
av_mallocz_array
void * av_mallocz_array(size_t nmemb, size_t size)
Definition: mem.c:190
ff_opus_parse_packet
int ff_opus_parse_packet(OpusPacket *pkt, const uint8_t *buf, int buf_size, int self_delimiting)
Parse Opus packet info from raw packet data.
Definition: opus.c:91
AVCodecContext::delay
int delay
Codec delay.
Definition: avcodec.h:692
MAX_FRAME_SIZE
#define MAX_FRAME_SIZE
Definition: 8svx.c:60
opus_rc_tell
static av_always_inline uint32_t opus_rc_tell(const OpusRangeCoder *rc)
CELT: estimate bits of entropy that have thus far been consumed for the current CELT frame,...
Definition: opus_rc.h:61
opus_frame_duration
static const uint16_t opus_frame_duration[32]
Definition: opus.c:37
ff_celt_log2_frac
const uint8_t ff_celt_log2_frac[]
Definition: opustab.c:930
fail
#define fail()
Definition: checkasm.h:133
val
static double val(void *priv, double ch)
Definition: aeval.c:76
ff_sqrt
#define ff_sqrt
Definition: mathops.h:206
AV_CH_LAYOUT_STEREO
#define AV_CH_LAYOUT_STEREO
Definition: channel_layout.h:91
channel_reorder_vorbis
static int channel_reorder_vorbis(int nb_channels, int channel_idx)
Definition: opus.c:282
pkt
AVPacket * pkt
Definition: movenc.c:59
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:194
av_cold
#define av_cold
Definition: attributes.h:90
CELT_VECTORS
#define CELT_VECTORS
Definition: opus_celt.h:38
duration
int64_t duration
Definition: movenc.c:64
xiph_lacing_16bit
static int xiph_lacing_16bit(const uint8_t **ptr, const uint8_t *end)
Read a 1- or 2-byte frame length.
Definition: opus.c:52
AVCodecContext::extradata_size
int extradata_size
Definition: avcodec.h:638
ff_opus_rc_dec_uint
uint32_t ff_opus_rc_dec_uint(OpusRangeCoder *rc, uint32_t size)
CELT: read a uniform distribution.
Definition: opus_rc.c:182
s
#define s(width, name)
Definition: cbs_vp9.c:257
bits1
static const uint8_t bits1[81]
Definition: aactab.c:136
CELT_SPREAD_NORMAL
@ CELT_SPREAD_NORMAL
Definition: opus_celt.h:53
CELT_MAX_FINE_BITS
#define CELT_MAX_FINE_BITS
Definition: opus_celt.h:41
bits
uint8_t bits
Definition: vp3data.h:141
channels
channels
Definition: aptx.h:33
AV_RL16
uint64_t_TMPL AV_WL64 unsigned int_TMPL AV_WL32 unsigned int_TMPL AV_WL24 unsigned int_TMPL AV_RL16
Definition: bytestream.h:94
SwrContext::channel_map
const int * channel_map
channel index (or -1 if muted channel) map
Definition: swresample_internal.h:113
f
#define f(width, name)
Definition: cbs_vp9.c:255
NORMC
#define NORMC(bits)
Definition: opus.c:553
if
if(ret)
Definition: filter_design.txt:179
ff_celt_bitalloc
void ff_celt_bitalloc(CeltFrame *f, OpusRangeCoder *rc, int encode)
Definition: opus.c:555
OPUS_MODE_CELT
@ OPUS_MODE_CELT
Definition: opus.h:66
ff_celt_static_caps
const uint8_t ff_celt_static_caps[4][2][21]
Definition: opustab.c:866
NULL
#define NULL
Definition: coverity.c:32
AVERROR_PATCHWELCOME
#define AVERROR_PATCHWELCOME
Not yet implemented in FFmpeg, patches welcome.
Definition: error.h:62
channel_reorder_unknown
static int channel_reorder_unknown(int nb_channels, int channel_idx)
Definition: opus.c:287
AVCodecContext::internal
struct AVCodecInternal * internal
Private context used for internal data.
Definition: avcodec.h:571
CELT_MAX_BANDS
#define CELT_MAX_BANDS
Definition: opus.h:46
ff_celt_model_spread
const uint16_t ff_celt_model_spread[]
Definition: opustab.c:760
ff_vorbis_channel_layout_offsets
const uint8_t ff_vorbis_channel_layout_offsets[8][8]
Definition: vorbis_data.c:26
CELT_ALLOC_STEPS
#define CELT_ALLOC_STEPS
Definition: opus_celt.h:39
OPUS_BANDWIDTH_SUPERWIDEBAND
@ OPUS_BANDWIDTH_SUPERWIDEBAND
Definition: opus.h:75
ff_celt_freq_range
const uint8_t ff_celt_freq_range[]
Definition: opustab.c:772
error.h
for
for(j=16;j >0;--j)
Definition: h264pred_template.c:469
opustab.h
FFMAX
#define FFMAX(a, b)
Definition: common.h:103
OpusRangeCoder
Definition: opus_rc.h:40
encode
static void encode(AVCodecContext *ctx, AVFrame *frame, AVPacket *pkt, FILE *output)
Definition: encode_audio.c:95
FFMIN
#define FFMIN(a, b)
Definition: common.h:105
offset
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf offset
Definition: writing_filters.txt:86
N
#define N
Definition: af_mcompand.c:54
version
version
Definition: libkvazaar.c:326
ff_celt_static_alloc
const uint8_t ff_celt_static_alloc[11][21]
Definition: opustab.c:852
Y
#define Y
Definition: boxblur.h:38
AVCodecContext::channels
int channels
number of audio channels
Definition: avcodec.h:1197
ff_opus_rc_dec_cdf
uint32_t ff_opus_rc_dec_cdf(OpusRangeCoder *rc, const uint16_t *cdf)
Definition: opus_rc.c:90
layout
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel layout
Definition: filter_design.txt:18
i
int i
Definition: input.c:407
vorbis.h
AVCodecContext::extradata
uint8_t * extradata
some codecs need / can use extradata like Huffman tables.
Definition: avcodec.h:637
opus_celt.h
OPUS_MODE_HYBRID
@ OPUS_MODE_HYBRID
Definition: opus.h:65
uint8_t
uint8_t
Definition: audio_convert.c:194
ff_celt_model_alloc_trim
const uint16_t ff_celt_model_alloc_trim[]
Definition: opustab.c:762
len
int len
Definition: vorbis_enc_data.h:452
opus_rc_tell_frac
static av_always_inline uint32_t opus_rc_tell_frac(const OpusRangeCoder *rc)
Definition: opus_rc.h:66
CELT_SPREAD_AGGRESSIVE
@ CELT_SPREAD_AGGRESSIVE
Definition: opus_celt.h:54
MAX_PACKET_DUR
#define MAX_PACKET_DUR
Definition: opus.h:40
AVCodecContext
main external API structure.
Definition: avcodec.h:536
OPUS_MODE_SILK
@ OPUS_MODE_SILK
Definition: opus.h:64
ff_celt_quant_bands
void ff_celt_quant_bands(CeltFrame *f, OpusRangeCoder *rc)
Definition: opus.c:446
cm
#define cm
Definition: dvbsubdec.c:37
OpusPacket
Definition: opus.h:85
ff_opus_rc_enc_log
void ff_opus_rc_enc_log(OpusRangeCoder *rc, int val, uint32_t bits)
Definition: opus_rc.c:131
temp
else temp
Definition: vf_mcdeint.c:259
ffmath.h
ChannelMap
Definition: opus.h:147
avpriv_request_sample
#define avpriv_request_sample(...)
Definition: tableprint_vlc.h:39
map
const VDPAUPixFmtMap * map
Definition: hwcontext_vdpau.c:71
CELT_FINE_OFFSET
#define CELT_FINE_OFFSET
Definition: opus_celt.h:40
av_freep
#define av_freep(p)
Definition: tableprint_vlc.h:35
xiph_lacing_full
static int xiph_lacing_full(const uint8_t **ptr, const uint8_t *end)
Read a multi-byte length (used for code 3 packet padding size)
Definition: opus.c:70
ff_vorbis_channel_layouts
const uint64_t ff_vorbis_channel_layouts[9]
Definition: vorbis_data.c:37
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:28
AVERROR_INVALIDDATA
#define AVERROR_INVALIDDATA
Invalid data found when processing input.
Definition: error.h:59
ff_opus_rc_dec_log
uint32_t ff_opus_rc_dec_log(OpusRangeCoder *rc, uint32_t bits)
Definition: opus_rc.c:114
int
int
Definition: ffmpeg_filter.c:170
bits2
static const uint8_t bits2[81]
Definition: aactab.c:159
CeltFrame
Definition: opus_celt.h:93
MAX_FRAMES
#define MAX_FRAMES
Definition: diracdec.c:53
nb_channels
int nb_channels
Definition: channel_layout.c:81
OpusContext
Definition: opus.h:162