FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
aacdec.c
Go to the documentation of this file.
1 /*
2  * AAC decoder
3  * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4  * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5  * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
6  *
7  * AAC LATM decoder
8  * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9  * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
10  *
11  * This file is part of FFmpeg.
12  *
13  * FFmpeg is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU Lesser General Public
15  * License as published by the Free Software Foundation; either
16  * version 2.1 of the License, or (at your option) any later version.
17  *
18  * FFmpeg is distributed in the hope that it will be useful,
19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21  * Lesser General Public License for more details.
22  *
23  * You should have received a copy of the GNU Lesser General Public
24  * License along with FFmpeg; if not, write to the Free Software
25  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26  */
27 
28 /**
29  * @file
30  * AAC decoder
31  * @author Oded Shimon ( ods15 ods15 dyndns org )
32  * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
33  */
34 
35 /*
36  * supported tools
37  *
38  * Support? Name
39  * N (code in SoC repo) gain control
40  * Y block switching
41  * Y window shapes - standard
42  * N window shapes - Low Delay
43  * Y filterbank - standard
44  * N (code in SoC repo) filterbank - Scalable Sample Rate
45  * Y Temporal Noise Shaping
46  * Y Long Term Prediction
47  * Y intensity stereo
48  * Y channel coupling
49  * Y frequency domain prediction
50  * Y Perceptual Noise Substitution
51  * Y Mid/Side stereo
52  * N Scalable Inverse AAC Quantization
53  * N Frequency Selective Switch
54  * N upsampling filter
55  * Y quantization & coding - AAC
56  * N quantization & coding - TwinVQ
57  * N quantization & coding - BSAC
58  * N AAC Error Resilience tools
59  * N Error Resilience payload syntax
60  * N Error Protection tool
61  * N CELP
62  * N Silence Compression
63  * N HVXC
64  * N HVXC 4kbits/s VR
65  * N Structured Audio tools
66  * N Structured Audio Sample Bank Format
67  * N MIDI
68  * N Harmonic and Individual Lines plus Noise
69  * N Text-To-Speech Interface
70  * Y Spectral Band Replication
71  * Y (not in this code) Layer-1
72  * Y (not in this code) Layer-2
73  * Y (not in this code) Layer-3
74  * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75  * Y Parametric Stereo
76  * N Direct Stream Transfer
77  * Y Enhanced AAC Low Delay (ER AAC ELD)
78  *
79  * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
80  * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
81  Parametric Stereo.
82  */
83 
84 #include "libavutil/float_dsp.h"
85 #include "libavutil/opt.h"
86 #include "avcodec.h"
87 #include "internal.h"
88 #include "get_bits.h"
89 #include "fft.h"
90 #include "fmtconvert.h"
91 #include "lpc.h"
92 #include "kbdwin.h"
93 #include "sinewin.h"
94 
95 #include "aac.h"
96 #include "aactab.h"
97 #include "aacdectab.h"
98 #include "cbrt_tablegen.h"
99 #include "sbr.h"
100 #include "aacsbr.h"
101 #include "mpeg4audio.h"
102 #include "aacadtsdec.h"
103 #include "libavutil/intfloat.h"
104 
105 #include <assert.h>
106 #include <errno.h>
107 #include <math.h>
108 #include <stdint.h>
109 #include <string.h>
110 
111 #if ARCH_ARM
112 # include "arm/aac.h"
113 #elif ARCH_MIPS
114 # include "mips/aacdec_mips.h"
115 #endif
116 
118 static VLC vlc_spectral[11];
119 
120 static int output_configure(AACContext *ac,
121  uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
122  enum OCStatus oc_type, int get_new_frame);
123 
124 #define overread_err "Input buffer exhausted before END element found\n"
125 
126 static int count_channels(uint8_t (*layout)[3], int tags)
127 {
128  int i, sum = 0;
129  for (i = 0; i < tags; i++) {
130  int syn_ele = layout[i][0];
131  int pos = layout[i][2];
132  sum += (1 + (syn_ele == TYPE_CPE)) *
133  (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
134  }
135  return sum;
136 }
137 
138 /**
139  * Check for the channel element in the current channel position configuration.
140  * If it exists, make sure the appropriate element is allocated and map the
141  * channel order to match the internal FFmpeg channel layout.
142  *
143  * @param che_pos current channel position configuration
144  * @param type channel element type
145  * @param id channel element id
146  * @param channels count of the number of channels in the configuration
147  *
148  * @return Returns error status. 0 - OK, !0 - error
149  */
151  enum ChannelPosition che_pos,
152  int type, int id, int *channels)
153 {
154  if (*channels >= MAX_CHANNELS)
155  return AVERROR_INVALIDDATA;
156  if (che_pos) {
157  if (!ac->che[type][id]) {
158  if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
159  return AVERROR(ENOMEM);
160  ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
161  }
162  if (type != TYPE_CCE) {
163  if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
164  av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
165  return AVERROR_INVALIDDATA;
166  }
167  ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
168  if (type == TYPE_CPE ||
169  (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
170  ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
171  }
172  }
173  } else {
174  if (ac->che[type][id])
175  ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
176  av_freep(&ac->che[type][id]);
177  }
178  return 0;
179 }
180 
182 {
183  AACContext *ac = avctx->priv_data;
184  int type, id, ch, ret;
185 
186  /* set channel pointers to internal buffers by default */
187  for (type = 0; type < 4; type++) {
188  for (id = 0; id < MAX_ELEM_ID; id++) {
189  ChannelElement *che = ac->che[type][id];
190  if (che) {
191  che->ch[0].ret = che->ch[0].ret_buf;
192  che->ch[1].ret = che->ch[1].ret_buf;
193  }
194  }
195  }
196 
197  /* get output buffer */
198  av_frame_unref(ac->frame);
199  if (!avctx->channels)
200  return 1;
201 
202  ac->frame->nb_samples = 2048;
203  if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
204  return ret;
205 
206  /* map output channel pointers to AVFrame data */
207  for (ch = 0; ch < avctx->channels; ch++) {
208  if (ac->output_element[ch])
209  ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
210  }
211 
212  return 0;
213 }
214 
216  uint64_t av_position;
220 };
221 
222 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
223  uint8_t (*layout_map)[3], int offset, uint64_t left,
224  uint64_t right, int pos)
225 {
226  if (layout_map[offset][0] == TYPE_CPE) {
227  e2c_vec[offset] = (struct elem_to_channel) {
228  .av_position = left | right,
229  .syn_ele = TYPE_CPE,
230  .elem_id = layout_map[offset][1],
231  .aac_position = pos
232  };
233  return 1;
234  } else {
235  e2c_vec[offset] = (struct elem_to_channel) {
236  .av_position = left,
237  .syn_ele = TYPE_SCE,
238  .elem_id = layout_map[offset][1],
239  .aac_position = pos
240  };
241  e2c_vec[offset + 1] = (struct elem_to_channel) {
242  .av_position = right,
243  .syn_ele = TYPE_SCE,
244  .elem_id = layout_map[offset + 1][1],
245  .aac_position = pos
246  };
247  return 2;
248  }
249 }
250 
251 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
252  int *current)
253 {
254  int num_pos_channels = 0;
255  int first_cpe = 0;
256  int sce_parity = 0;
257  int i;
258  for (i = *current; i < tags; i++) {
259  if (layout_map[i][2] != pos)
260  break;
261  if (layout_map[i][0] == TYPE_CPE) {
262  if (sce_parity) {
263  if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
264  sce_parity = 0;
265  } else {
266  return -1;
267  }
268  }
269  num_pos_channels += 2;
270  first_cpe = 1;
271  } else {
272  num_pos_channels++;
273  sce_parity ^= 1;
274  }
275  }
276  if (sce_parity &&
277  ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
278  return -1;
279  *current = i;
280  return num_pos_channels;
281 }
282 
283 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
284 {
285  int i, n, total_non_cc_elements;
286  struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
287  int num_front_channels, num_side_channels, num_back_channels;
288  uint64_t layout;
289 
290  if (FF_ARRAY_ELEMS(e2c_vec) < tags)
291  return 0;
292 
293  i = 0;
294  num_front_channels =
295  count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
296  if (num_front_channels < 0)
297  return 0;
298  num_side_channels =
299  count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
300  if (num_side_channels < 0)
301  return 0;
302  num_back_channels =
303  count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
304  if (num_back_channels < 0)
305  return 0;
306 
307  i = 0;
308  if (num_front_channels & 1) {
309  e2c_vec[i] = (struct elem_to_channel) {
311  .syn_ele = TYPE_SCE,
312  .elem_id = layout_map[i][1],
313  .aac_position = AAC_CHANNEL_FRONT
314  };
315  i++;
316  num_front_channels--;
317  }
318  if (num_front_channels >= 4) {
319  i += assign_pair(e2c_vec, layout_map, i,
323  num_front_channels -= 2;
324  }
325  if (num_front_channels >= 2) {
326  i += assign_pair(e2c_vec, layout_map, i,
330  num_front_channels -= 2;
331  }
332  while (num_front_channels >= 2) {
333  i += assign_pair(e2c_vec, layout_map, i,
334  UINT64_MAX,
335  UINT64_MAX,
337  num_front_channels -= 2;
338  }
339 
340  if (num_side_channels >= 2) {
341  i += assign_pair(e2c_vec, layout_map, i,
345  num_side_channels -= 2;
346  }
347  while (num_side_channels >= 2) {
348  i += assign_pair(e2c_vec, layout_map, i,
349  UINT64_MAX,
350  UINT64_MAX,
352  num_side_channels -= 2;
353  }
354 
355  while (num_back_channels >= 4) {
356  i += assign_pair(e2c_vec, layout_map, i,
357  UINT64_MAX,
358  UINT64_MAX,
360  num_back_channels -= 2;
361  }
362  if (num_back_channels >= 2) {
363  i += assign_pair(e2c_vec, layout_map, i,
367  num_back_channels -= 2;
368  }
369  if (num_back_channels) {
370  e2c_vec[i] = (struct elem_to_channel) {
372  .syn_ele = TYPE_SCE,
373  .elem_id = layout_map[i][1],
374  .aac_position = AAC_CHANNEL_BACK
375  };
376  i++;
377  num_back_channels--;
378  }
379 
380  if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
381  e2c_vec[i] = (struct elem_to_channel) {
383  .syn_ele = TYPE_LFE,
384  .elem_id = layout_map[i][1],
385  .aac_position = AAC_CHANNEL_LFE
386  };
387  i++;
388  }
389  while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
390  e2c_vec[i] = (struct elem_to_channel) {
391  .av_position = UINT64_MAX,
392  .syn_ele = TYPE_LFE,
393  .elem_id = layout_map[i][1],
394  .aac_position = AAC_CHANNEL_LFE
395  };
396  i++;
397  }
398 
399  // Must choose a stable sort
400  total_non_cc_elements = n = i;
401  do {
402  int next_n = 0;
403  for (i = 1; i < n; i++)
404  if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
405  FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
406  next_n = i;
407  }
408  n = next_n;
409  } while (n > 0);
410 
411  layout = 0;
412  for (i = 0; i < total_non_cc_elements; i++) {
413  layout_map[i][0] = e2c_vec[i].syn_ele;
414  layout_map[i][1] = e2c_vec[i].elem_id;
415  layout_map[i][2] = e2c_vec[i].aac_position;
416  if (e2c_vec[i].av_position != UINT64_MAX) {
417  layout |= e2c_vec[i].av_position;
418  }
419  }
420 
421  return layout;
422 }
423 
424 /**
425  * Save current output configuration if and only if it has been locked.
426  */
428  if (ac->oc[1].status == OC_LOCKED) {
429  ac->oc[0] = ac->oc[1];
430  }
431  ac->oc[1].status = OC_NONE;
432 }
433 
434 /**
435  * Restore the previous output configuration if and only if the current
436  * configuration is unlocked.
437  */
439  if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
440  ac->oc[1] = ac->oc[0];
441  ac->avctx->channels = ac->oc[1].channels;
442  ac->avctx->channel_layout = ac->oc[1].channel_layout;
443  output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
444  ac->oc[1].status, 0);
445  }
446 }
447 
448 /**
449  * Configure output channel order based on the current program
450  * configuration element.
451  *
452  * @return Returns error status. 0 - OK, !0 - error
453  */
455  uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
456  enum OCStatus oc_type, int get_new_frame)
457 {
458  AVCodecContext *avctx = ac->avctx;
459  int i, channels = 0, ret;
460  uint64_t layout = 0;
461 
462  if (ac->oc[1].layout_map != layout_map) {
463  memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
464  ac->oc[1].layout_map_tags = tags;
465  }
466 
467  // Try to sniff a reasonable channel order, otherwise output the
468  // channels in the order the PCE declared them.
470  layout = sniff_channel_order(layout_map, tags);
471  for (i = 0; i < tags; i++) {
472  int type = layout_map[i][0];
473  int id = layout_map[i][1];
474  int position = layout_map[i][2];
475  // Allocate or free elements depending on if they are in the
476  // current program configuration.
477  ret = che_configure(ac, position, type, id, &channels);
478  if (ret < 0)
479  return ret;
480  }
481  if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
482  if (layout == AV_CH_FRONT_CENTER) {
484  } else {
485  layout = 0;
486  }
487  }
488 
489  memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
490  if (layout) avctx->channel_layout = layout;
491  ac->oc[1].channel_layout = layout;
492  avctx->channels = ac->oc[1].channels = channels;
493  ac->oc[1].status = oc_type;
494 
495  if (get_new_frame) {
496  if ((ret = frame_configure_elements(ac->avctx)) < 0)
497  return ret;
498  }
499 
500  return 0;
501 }
502 
503 static void flush(AVCodecContext *avctx)
504 {
505  AACContext *ac= avctx->priv_data;
506  int type, i, j;
507 
508  for (type = 3; type >= 0; type--) {
509  for (i = 0; i < MAX_ELEM_ID; i++) {
510  ChannelElement *che = ac->che[type][i];
511  if (che) {
512  for (j = 0; j <= 1; j++) {
513  memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
514  }
515  }
516  }
517  }
518 }
519 
520 /**
521  * Set up channel positions based on a default channel configuration
522  * as specified in table 1.17.
523  *
524  * @return Returns error status. 0 - OK, !0 - error
525  */
527  uint8_t (*layout_map)[3],
528  int *tags,
529  int channel_config)
530 {
531  if (channel_config < 1 || channel_config > 7) {
532  av_log(avctx, AV_LOG_ERROR,
533  "invalid default channel configuration (%d)\n",
534  channel_config);
535  return AVERROR_INVALIDDATA;
536  }
537  *tags = tags_per_config[channel_config];
538  memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
539  *tags * sizeof(*layout_map));
540 
541  /*
542  * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
543  * However, at least Nero AAC encoder encodes 7.1 streams using the default
544  * channel config 7, mapping the side channels of the original audio stream
545  * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
546  * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
547  * the incorrect streams as if they were correct (and as the encoder intended).
548  *
549  * As actual intended 7.1(wide) streams are very rare, default to assuming a
550  * 7.1 layout was intended.
551  */
552  if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
553  av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
554  " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
555  " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
556  layout_map[2][2] = AAC_CHANNEL_SIDE;
557  }
558 
559  return 0;
560 }
561 
562 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
563 {
564  /* For PCE based channel configurations map the channels solely based
565  * on tags. */
566  if (!ac->oc[1].m4ac.chan_config) {
567  return ac->tag_che_map[type][elem_id];
568  }
569  // Allow single CPE stereo files to be signalled with mono configuration.
570  if (!ac->tags_mapped && type == TYPE_CPE &&
571  ac->oc[1].m4ac.chan_config == 1) {
572  uint8_t layout_map[MAX_ELEM_ID*4][3];
573  int layout_map_tags;
575 
576  av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
577 
578  if (set_default_channel_config(ac->avctx, layout_map,
579  &layout_map_tags, 2) < 0)
580  return NULL;
581  if (output_configure(ac, layout_map, layout_map_tags,
582  OC_TRIAL_FRAME, 1) < 0)
583  return NULL;
584 
585  ac->oc[1].m4ac.chan_config = 2;
586  ac->oc[1].m4ac.ps = 0;
587  }
588  // And vice-versa
589  if (!ac->tags_mapped && type == TYPE_SCE &&
590  ac->oc[1].m4ac.chan_config == 2) {
591  uint8_t layout_map[MAX_ELEM_ID * 4][3];
592  int layout_map_tags;
594 
595  av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
596 
597  if (set_default_channel_config(ac->avctx, layout_map,
598  &layout_map_tags, 1) < 0)
599  return NULL;
600  if (output_configure(ac, layout_map, layout_map_tags,
601  OC_TRIAL_FRAME, 1) < 0)
602  return NULL;
603 
604  ac->oc[1].m4ac.chan_config = 1;
605  if (ac->oc[1].m4ac.sbr)
606  ac->oc[1].m4ac.ps = -1;
607  }
608  /* For indexed channel configurations map the channels solely based
609  * on position. */
610  switch (ac->oc[1].m4ac.chan_config) {
611  case 7:
612  if (ac->tags_mapped == 3 && type == TYPE_CPE) {
613  ac->tags_mapped++;
614  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
615  }
616  case 6:
617  /* Some streams incorrectly code 5.1 audio as
618  * SCE[0] CPE[0] CPE[1] SCE[1]
619  * instead of
620  * SCE[0] CPE[0] CPE[1] LFE[0].
621  * If we seem to have encountered such a stream, transfer
622  * the LFE[0] element to the SCE[1]'s mapping */
623  if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
624  if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
626  "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
627  type == TYPE_SCE ? "SCE" : "LFE", elem_id);
628  ac->warned_remapping_once++;
629  }
630  ac->tags_mapped++;
631  return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
632  }
633  case 5:
634  if (ac->tags_mapped == 2 && type == TYPE_CPE) {
635  ac->tags_mapped++;
636  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
637  }
638  case 4:
639  /* Some streams incorrectly code 4.0 audio as
640  * SCE[0] CPE[0] LFE[0]
641  * instead of
642  * SCE[0] CPE[0] SCE[1].
643  * If we seem to have encountered such a stream, transfer
644  * the SCE[1] element to the LFE[0]'s mapping */
645  if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
646  if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
648  "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
649  type == TYPE_SCE ? "SCE" : "LFE", elem_id);
650  ac->warned_remapping_once++;
651  }
652  ac->tags_mapped++;
653  return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
654  }
655  if (ac->tags_mapped == 2 &&
656  ac->oc[1].m4ac.chan_config == 4 &&
657  type == TYPE_SCE) {
658  ac->tags_mapped++;
659  return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
660  }
661  case 3:
662  case 2:
663  if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
664  type == TYPE_CPE) {
665  ac->tags_mapped++;
666  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
667  } else if (ac->oc[1].m4ac.chan_config == 2) {
668  return NULL;
669  }
670  case 1:
671  if (!ac->tags_mapped && type == TYPE_SCE) {
672  ac->tags_mapped++;
673  return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
674  }
675  default:
676  return NULL;
677  }
678 }
679 
680 /**
681  * Decode an array of 4 bit element IDs, optionally interleaved with a
682  * stereo/mono switching bit.
683  *
684  * @param type speaker type/position for these channels
685  */
686 static void decode_channel_map(uint8_t layout_map[][3],
687  enum ChannelPosition type,
688  GetBitContext *gb, int n)
689 {
690  while (n--) {
691  enum RawDataBlockType syn_ele;
692  switch (type) {
693  case AAC_CHANNEL_FRONT:
694  case AAC_CHANNEL_BACK:
695  case AAC_CHANNEL_SIDE:
696  syn_ele = get_bits1(gb);
697  break;
698  case AAC_CHANNEL_CC:
699  skip_bits1(gb);
700  syn_ele = TYPE_CCE;
701  break;
702  case AAC_CHANNEL_LFE:
703  syn_ele = TYPE_LFE;
704  break;
705  default:
706  // AAC_CHANNEL_OFF has no channel map
707  av_assert0(0);
708  }
709  layout_map[0][0] = syn_ele;
710  layout_map[0][1] = get_bits(gb, 4);
711  layout_map[0][2] = type;
712  layout_map++;
713  }
714 }
715 
716 /**
717  * Decode program configuration element; reference: table 4.2.
718  *
719  * @return Returns error status. 0 - OK, !0 - error
720  */
721 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
722  uint8_t (*layout_map)[3],
723  GetBitContext *gb)
724 {
725  int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
726  int sampling_index;
727  int comment_len;
728  int tags;
729 
730  skip_bits(gb, 2); // object_type
731 
732  sampling_index = get_bits(gb, 4);
733  if (m4ac->sampling_index != sampling_index)
734  av_log(avctx, AV_LOG_WARNING,
735  "Sample rate index in program config element does not "
736  "match the sample rate index configured by the container.\n");
737 
738  num_front = get_bits(gb, 4);
739  num_side = get_bits(gb, 4);
740  num_back = get_bits(gb, 4);
741  num_lfe = get_bits(gb, 2);
742  num_assoc_data = get_bits(gb, 3);
743  num_cc = get_bits(gb, 4);
744 
745  if (get_bits1(gb))
746  skip_bits(gb, 4); // mono_mixdown_tag
747  if (get_bits1(gb))
748  skip_bits(gb, 4); // stereo_mixdown_tag
749 
750  if (get_bits1(gb))
751  skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
752 
753  if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
754  av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
755  return -1;
756  }
757  decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
758  tags = num_front;
759  decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
760  tags += num_side;
761  decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
762  tags += num_back;
763  decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
764  tags += num_lfe;
765 
766  skip_bits_long(gb, 4 * num_assoc_data);
767 
768  decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
769  tags += num_cc;
770 
771  align_get_bits(gb);
772 
773  /* comment field, first byte is length */
774  comment_len = get_bits(gb, 8) * 8;
775  if (get_bits_left(gb) < comment_len) {
776  av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
777  return AVERROR_INVALIDDATA;
778  }
779  skip_bits_long(gb, comment_len);
780  return tags;
781 }
782 
783 /**
784  * Decode GA "General Audio" specific configuration; reference: table 4.1.
785  *
786  * @param ac pointer to AACContext, may be null
787  * @param avctx pointer to AVCCodecContext, used for logging
788  *
789  * @return Returns error status. 0 - OK, !0 - error
790  */
792  GetBitContext *gb,
793  MPEG4AudioConfig *m4ac,
794  int channel_config)
795 {
796  int extension_flag, ret, ep_config, res_flags;
797  uint8_t layout_map[MAX_ELEM_ID*4][3];
798  int tags = 0;
799 
800  if (get_bits1(gb)) { // frameLengthFlag
801  avpriv_request_sample(avctx, "960/120 MDCT window");
802  return AVERROR_PATCHWELCOME;
803  }
804 
805  if (get_bits1(gb)) // dependsOnCoreCoder
806  skip_bits(gb, 14); // coreCoderDelay
807  extension_flag = get_bits1(gb);
808 
809  if (m4ac->object_type == AOT_AAC_SCALABLE ||
811  skip_bits(gb, 3); // layerNr
812 
813  if (channel_config == 0) {
814  skip_bits(gb, 4); // element_instance_tag
815  tags = decode_pce(avctx, m4ac, layout_map, gb);
816  if (tags < 0)
817  return tags;
818  } else {
819  if ((ret = set_default_channel_config(avctx, layout_map,
820  &tags, channel_config)))
821  return ret;
822  }
823 
824  if (count_channels(layout_map, tags) > 1) {
825  m4ac->ps = 0;
826  } else if (m4ac->sbr == 1 && m4ac->ps == -1)
827  m4ac->ps = 1;
828 
829  if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
830  return ret;
831 
832  if (extension_flag) {
833  switch (m4ac->object_type) {
834  case AOT_ER_BSAC:
835  skip_bits(gb, 5); // numOfSubFrame
836  skip_bits(gb, 11); // layer_length
837  break;
838  case AOT_ER_AAC_LC:
839  case AOT_ER_AAC_LTP:
840  case AOT_ER_AAC_SCALABLE:
841  case AOT_ER_AAC_LD:
842  res_flags = get_bits(gb, 3);
843  if (res_flags) {
845  "AAC data resilience (flags %x)",
846  res_flags);
847  return AVERROR_PATCHWELCOME;
848  }
849  break;
850  }
851  skip_bits1(gb); // extensionFlag3 (TBD in version 3)
852  }
853  switch (m4ac->object_type) {
854  case AOT_ER_AAC_LC:
855  case AOT_ER_AAC_LTP:
856  case AOT_ER_AAC_SCALABLE:
857  case AOT_ER_AAC_LD:
858  ep_config = get_bits(gb, 2);
859  if (ep_config) {
861  "epConfig %d", ep_config);
862  return AVERROR_PATCHWELCOME;
863  }
864  }
865  return 0;
866 }
867 
869  GetBitContext *gb,
870  MPEG4AudioConfig *m4ac,
871  int channel_config)
872 {
873  int ret, ep_config, res_flags;
874  uint8_t layout_map[MAX_ELEM_ID*4][3];
875  int tags = 0;
876  const int ELDEXT_TERM = 0;
877 
878  m4ac->ps = 0;
879  m4ac->sbr = 0;
880 
881  if (get_bits1(gb)) { // frameLengthFlag
882  avpriv_request_sample(avctx, "960/120 MDCT window");
883  return AVERROR_PATCHWELCOME;
884  }
885 
886  res_flags = get_bits(gb, 3);
887  if (res_flags) {
889  "AAC data resilience (flags %x)",
890  res_flags);
891  return AVERROR_PATCHWELCOME;
892  }
893 
894  if (get_bits1(gb)) { // ldSbrPresentFlag
896  "Low Delay SBR");
897  return AVERROR_PATCHWELCOME;
898  }
899 
900  while (get_bits(gb, 4) != ELDEXT_TERM) {
901  int len = get_bits(gb, 4);
902  if (len == 15)
903  len += get_bits(gb, 8);
904  if (len == 15 + 255)
905  len += get_bits(gb, 16);
906  if (get_bits_left(gb) < len * 8 + 4) {
908  return AVERROR_INVALIDDATA;
909  }
910  skip_bits_long(gb, 8 * len);
911  }
912 
913  if ((ret = set_default_channel_config(avctx, layout_map,
914  &tags, channel_config)))
915  return ret;
916 
917  if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
918  return ret;
919 
920  ep_config = get_bits(gb, 2);
921  if (ep_config) {
923  "epConfig %d", ep_config);
924  return AVERROR_PATCHWELCOME;
925  }
926  return 0;
927 }
928 
929 /**
930  * Decode audio specific configuration; reference: table 1.13.
931  *
932  * @param ac pointer to AACContext, may be null
933  * @param avctx pointer to AVCCodecContext, used for logging
934  * @param m4ac pointer to MPEG4AudioConfig, used for parsing
935  * @param data pointer to buffer holding an audio specific config
936  * @param bit_size size of audio specific config or data in bits
937  * @param sync_extension look for an appended sync extension
938  *
939  * @return Returns error status or number of consumed bits. <0 - error
940  */
942  AVCodecContext *avctx,
943  MPEG4AudioConfig *m4ac,
944  const uint8_t *data, int bit_size,
945  int sync_extension)
946 {
947  GetBitContext gb;
948  int i, ret;
949 
950  av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
951  for (i = 0; i < bit_size >> 3; i++)
952  av_dlog(avctx, "%02x ", data[i]);
953  av_dlog(avctx, "\n");
954 
955  if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
956  return ret;
957 
958  if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
959  sync_extension)) < 0)
960  return AVERROR_INVALIDDATA;
961  if (m4ac->sampling_index > 12) {
962  av_log(avctx, AV_LOG_ERROR,
963  "invalid sampling rate index %d\n",
964  m4ac->sampling_index);
965  return AVERROR_INVALIDDATA;
966  }
967  if (m4ac->object_type == AOT_ER_AAC_LD &&
968  (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
969  av_log(avctx, AV_LOG_ERROR,
970  "invalid low delay sampling rate index %d\n",
971  m4ac->sampling_index);
972  return AVERROR_INVALIDDATA;
973  }
974 
975  skip_bits_long(&gb, i);
976 
977  switch (m4ac->object_type) {
978  case AOT_AAC_MAIN:
979  case AOT_AAC_LC:
980  case AOT_AAC_LTP:
981  case AOT_ER_AAC_LC:
982  case AOT_ER_AAC_LD:
983  if ((ret = decode_ga_specific_config(ac, avctx, &gb,
984  m4ac, m4ac->chan_config)) < 0)
985  return ret;
986  break;
987  case AOT_ER_AAC_ELD:
988  if ((ret = decode_eld_specific_config(ac, avctx, &gb,
989  m4ac, m4ac->chan_config)) < 0)
990  return ret;
991  break;
992  default:
994  "Audio object type %s%d",
995  m4ac->sbr == 1 ? "SBR+" : "",
996  m4ac->object_type);
997  return AVERROR(ENOSYS);
998  }
999 
1000  av_dlog(avctx,
1001  "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
1002  m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1003  m4ac->sample_rate, m4ac->sbr,
1004  m4ac->ps);
1005 
1006  return get_bits_count(&gb);
1007 }
1008 
1009 /**
1010  * linear congruential pseudorandom number generator
1011  *
1012  * @param previous_val pointer to the current state of the generator
1013  *
1014  * @return Returns a 32-bit pseudorandom integer
1015  */
1016 static av_always_inline int lcg_random(unsigned previous_val)
1017 {
1018  union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1019  return v.s;
1020 }
1021 
1023 {
1024  ps->r0 = 0.0f;
1025  ps->r1 = 0.0f;
1026  ps->cor0 = 0.0f;
1027  ps->cor1 = 0.0f;
1028  ps->var0 = 1.0f;
1029  ps->var1 = 1.0f;
1030 }
1031 
1033 {
1034  int i;
1035  for (i = 0; i < MAX_PREDICTORS; i++)
1036  reset_predict_state(&ps[i]);
1037 }
1038 
1039 static int sample_rate_idx (int rate)
1040 {
1041  if (92017 <= rate) return 0;
1042  else if (75132 <= rate) return 1;
1043  else if (55426 <= rate) return 2;
1044  else if (46009 <= rate) return 3;
1045  else if (37566 <= rate) return 4;
1046  else if (27713 <= rate) return 5;
1047  else if (23004 <= rate) return 6;
1048  else if (18783 <= rate) return 7;
1049  else if (13856 <= rate) return 8;
1050  else if (11502 <= rate) return 9;
1051  else if (9391 <= rate) return 10;
1052  else return 11;
1053 }
1054 
1055 static void reset_predictor_group(PredictorState *ps, int group_num)
1056 {
1057  int i;
1058  for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1059  reset_predict_state(&ps[i]);
1060 }
1061 
1062 #define AAC_INIT_VLC_STATIC(num, size) \
1063  INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1064  ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1065  sizeof(ff_aac_spectral_bits[num][0]), \
1066  ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1067  sizeof(ff_aac_spectral_codes[num][0]), \
1068  size);
1069 
1070 static void aacdec_init(AACContext *ac);
1071 
1073 {
1074  AACContext *ac = avctx->priv_data;
1075  int ret;
1076 
1077  ac->avctx = avctx;
1078  ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1079 
1080  aacdec_init(ac);
1081 
1082  avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1083 
1084  if (avctx->extradata_size > 0) {
1085  if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1086  avctx->extradata,
1087  avctx->extradata_size * 8,
1088  1)) < 0)
1089  return ret;
1090  } else {
1091  int sr, i;
1092  uint8_t layout_map[MAX_ELEM_ID*4][3];
1093  int layout_map_tags;
1094 
1095  sr = sample_rate_idx(avctx->sample_rate);
1096  ac->oc[1].m4ac.sampling_index = sr;
1097  ac->oc[1].m4ac.channels = avctx->channels;
1098  ac->oc[1].m4ac.sbr = -1;
1099  ac->oc[1].m4ac.ps = -1;
1100 
1101  for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1102  if (ff_mpeg4audio_channels[i] == avctx->channels)
1103  break;
1105  i = 0;
1106  }
1107  ac->oc[1].m4ac.chan_config = i;
1108 
1109  if (ac->oc[1].m4ac.chan_config) {
1110  int ret = set_default_channel_config(avctx, layout_map,
1111  &layout_map_tags, ac->oc[1].m4ac.chan_config);
1112  if (!ret)
1113  output_configure(ac, layout_map, layout_map_tags,
1114  OC_GLOBAL_HDR, 0);
1115  else if (avctx->err_recognition & AV_EF_EXPLODE)
1116  return AVERROR_INVALIDDATA;
1117  }
1118  }
1119 
1120  if (avctx->channels > MAX_CHANNELS) {
1121  av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1122  return AVERROR_INVALIDDATA;
1123  }
1124 
1125  AAC_INIT_VLC_STATIC( 0, 304);
1126  AAC_INIT_VLC_STATIC( 1, 270);
1127  AAC_INIT_VLC_STATIC( 2, 550);
1128  AAC_INIT_VLC_STATIC( 3, 300);
1129  AAC_INIT_VLC_STATIC( 4, 328);
1130  AAC_INIT_VLC_STATIC( 5, 294);
1131  AAC_INIT_VLC_STATIC( 6, 306);
1132  AAC_INIT_VLC_STATIC( 7, 268);
1133  AAC_INIT_VLC_STATIC( 8, 510);
1134  AAC_INIT_VLC_STATIC( 9, 366);
1135  AAC_INIT_VLC_STATIC(10, 462);
1136 
1137  ff_aac_sbr_init();
1138 
1139  ff_fmt_convert_init(&ac->fmt_conv, avctx);
1141 
1142  ac->random_state = 0x1f2e3d4c;
1143 
1144  ff_aac_tableinit();
1145 
1146  INIT_VLC_STATIC(&vlc_scalefactors, 7,
1149  sizeof(ff_aac_scalefactor_bits[0]),
1150  sizeof(ff_aac_scalefactor_bits[0]),
1152  sizeof(ff_aac_scalefactor_code[0]),
1153  sizeof(ff_aac_scalefactor_code[0]),
1154  352);
1155 
1156  ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1157  ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1158  ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1159  ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1160  // window initialization
1166 
1167  cbrt_tableinit();
1168 
1169  return 0;
1170 }
1171 
1172 /**
1173  * Skip data_stream_element; reference: table 4.10.
1174  */
1176 {
1177  int byte_align = get_bits1(gb);
1178  int count = get_bits(gb, 8);
1179  if (count == 255)
1180  count += get_bits(gb, 8);
1181  if (byte_align)
1182  align_get_bits(gb);
1183 
1184  if (get_bits_left(gb) < 8 * count) {
1185  av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1186  return AVERROR_INVALIDDATA;
1187  }
1188  skip_bits_long(gb, 8 * count);
1189  return 0;
1190 }
1191 
1193  GetBitContext *gb)
1194 {
1195  int sfb;
1196  if (get_bits1(gb)) {
1197  ics->predictor_reset_group = get_bits(gb, 5);
1198  if (ics->predictor_reset_group == 0 ||
1199  ics->predictor_reset_group > 30) {
1200  av_log(ac->avctx, AV_LOG_ERROR,
1201  "Invalid Predictor Reset Group.\n");
1202  return AVERROR_INVALIDDATA;
1203  }
1204  }
1205  for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1206  ics->prediction_used[sfb] = get_bits1(gb);
1207  }
1208  return 0;
1209 }
1210 
1211 /**
1212  * Decode Long Term Prediction data; reference: table 4.xx.
1213  */
1215  GetBitContext *gb, uint8_t max_sfb)
1216 {
1217  int sfb;
1218 
1219  ltp->lag = get_bits(gb, 11);
1220  ltp->coef = ltp_coef[get_bits(gb, 3)];
1221  for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1222  ltp->used[sfb] = get_bits1(gb);
1223 }
1224 
1225 /**
1226  * Decode Individual Channel Stream info; reference: table 4.6.
1227  */
1229  GetBitContext *gb)
1230 {
1231  int aot = ac->oc[1].m4ac.object_type;
1232  if (aot != AOT_ER_AAC_ELD) {
1233  if (get_bits1(gb)) {
1234  av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1235  return AVERROR_INVALIDDATA;
1236  }
1237  ics->window_sequence[1] = ics->window_sequence[0];
1238  ics->window_sequence[0] = get_bits(gb, 2);
1239  if (aot == AOT_ER_AAC_LD &&
1240  ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1241  av_log(ac->avctx, AV_LOG_ERROR,
1242  "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1243  "window sequence %d found.\n", ics->window_sequence[0]);
1245  return AVERROR_INVALIDDATA;
1246  }
1247  ics->use_kb_window[1] = ics->use_kb_window[0];
1248  ics->use_kb_window[0] = get_bits1(gb);
1249  }
1250  ics->num_window_groups = 1;
1251  ics->group_len[0] = 1;
1252  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1253  int i;
1254  ics->max_sfb = get_bits(gb, 4);
1255  for (i = 0; i < 7; i++) {
1256  if (get_bits1(gb)) {
1257  ics->group_len[ics->num_window_groups - 1]++;
1258  } else {
1259  ics->num_window_groups++;
1260  ics->group_len[ics->num_window_groups - 1] = 1;
1261  }
1262  }
1263  ics->num_windows = 8;
1267  ics->predictor_present = 0;
1268  } else {
1269  ics->max_sfb = get_bits(gb, 6);
1270  ics->num_windows = 1;
1271  if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1275  if (!ics->num_swb || !ics->swb_offset)
1276  return AVERROR_BUG;
1277  } else {
1281  }
1282  if (aot != AOT_ER_AAC_ELD) {
1283  ics->predictor_present = get_bits1(gb);
1284  ics->predictor_reset_group = 0;
1285  }
1286  if (ics->predictor_present) {
1287  if (aot == AOT_AAC_MAIN) {
1288  if (decode_prediction(ac, ics, gb)) {
1289  goto fail;
1290  }
1291  } else if (aot == AOT_AAC_LC ||
1292  aot == AOT_ER_AAC_LC) {
1293  av_log(ac->avctx, AV_LOG_ERROR,
1294  "Prediction is not allowed in AAC-LC.\n");
1295  goto fail;
1296  } else {
1297  if (aot == AOT_ER_AAC_LD) {
1298  av_log(ac->avctx, AV_LOG_ERROR,
1299  "LTP in ER AAC LD not yet implemented.\n");
1300  return AVERROR_PATCHWELCOME;
1301  }
1302  if ((ics->ltp.present = get_bits(gb, 1)))
1303  decode_ltp(&ics->ltp, gb, ics->max_sfb);
1304  }
1305  }
1306  }
1307 
1308  if (ics->max_sfb > ics->num_swb) {
1309  av_log(ac->avctx, AV_LOG_ERROR,
1310  "Number of scalefactor bands in group (%d) "
1311  "exceeds limit (%d).\n",
1312  ics->max_sfb, ics->num_swb);
1313  goto fail;
1314  }
1315 
1316  return 0;
1317 fail:
1318  ics->max_sfb = 0;
1319  return AVERROR_INVALIDDATA;
1320 }
1321 
1322 /**
1323  * Decode band types (section_data payload); reference: table 4.46.
1324  *
1325  * @param band_type array of the used band type
1326  * @param band_type_run_end array of the last scalefactor band of a band type run
1327  *
1328  * @return Returns error status. 0 - OK, !0 - error
1329  */
1330 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1331  int band_type_run_end[120], GetBitContext *gb,
1333 {
1334  int g, idx = 0;
1335  const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1336  for (g = 0; g < ics->num_window_groups; g++) {
1337  int k = 0;
1338  while (k < ics->max_sfb) {
1339  uint8_t sect_end = k;
1340  int sect_len_incr;
1341  int sect_band_type = get_bits(gb, 4);
1342  if (sect_band_type == 12) {
1343  av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1344  return AVERROR_INVALIDDATA;
1345  }
1346  do {
1347  sect_len_incr = get_bits(gb, bits);
1348  sect_end += sect_len_incr;
1349  if (get_bits_left(gb) < 0) {
1350  av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1351  return AVERROR_INVALIDDATA;
1352  }
1353  if (sect_end > ics->max_sfb) {
1354  av_log(ac->avctx, AV_LOG_ERROR,
1355  "Number of bands (%d) exceeds limit (%d).\n",
1356  sect_end, ics->max_sfb);
1357  return AVERROR_INVALIDDATA;
1358  }
1359  } while (sect_len_incr == (1 << bits) - 1);
1360  for (; k < sect_end; k++) {
1361  band_type [idx] = sect_band_type;
1362  band_type_run_end[idx++] = sect_end;
1363  }
1364  }
1365  }
1366  return 0;
1367 }
1368 
1369 /**
1370  * Decode scalefactors; reference: table 4.47.
1371  *
1372  * @param global_gain first scalefactor value as scalefactors are differentially coded
1373  * @param band_type array of the used band type
1374  * @param band_type_run_end array of the last scalefactor band of a band type run
1375  * @param sf array of scalefactors or intensity stereo positions
1376  *
1377  * @return Returns error status. 0 - OK, !0 - error
1378  */
1379 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1380  unsigned int global_gain,
1382  enum BandType band_type[120],
1383  int band_type_run_end[120])
1384 {
1385  int g, i, idx = 0;
1386  int offset[3] = { global_gain, global_gain - 90, 0 };
1387  int clipped_offset;
1388  int noise_flag = 1;
1389  for (g = 0; g < ics->num_window_groups; g++) {
1390  for (i = 0; i < ics->max_sfb;) {
1391  int run_end = band_type_run_end[idx];
1392  if (band_type[idx] == ZERO_BT) {
1393  for (; i < run_end; i++, idx++)
1394  sf[idx] = 0.0;
1395  } else if ((band_type[idx] == INTENSITY_BT) ||
1396  (band_type[idx] == INTENSITY_BT2)) {
1397  for (; i < run_end; i++, idx++) {
1398  offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1399  clipped_offset = av_clip(offset[2], -155, 100);
1400  if (offset[2] != clipped_offset) {
1402  "If you heard an audible artifact, there may be a bug in the decoder. "
1403  "Clipped intensity stereo position (%d -> %d)",
1404  offset[2], clipped_offset);
1405  }
1406  sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1407  }
1408  } else if (band_type[idx] == NOISE_BT) {
1409  for (; i < run_end; i++, idx++) {
1410  if (noise_flag-- > 0)
1411  offset[1] += get_bits(gb, 9) - 256;
1412  else
1413  offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1414  clipped_offset = av_clip(offset[1], -100, 155);
1415  if (offset[1] != clipped_offset) {
1417  "If you heard an audible artifact, there may be a bug in the decoder. "
1418  "Clipped noise gain (%d -> %d)",
1419  offset[1], clipped_offset);
1420  }
1421  sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1422  }
1423  } else {
1424  for (; i < run_end; i++, idx++) {
1425  offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1426  if (offset[0] > 255U) {
1427  av_log(ac->avctx, AV_LOG_ERROR,
1428  "Scalefactor (%d) out of range.\n", offset[0]);
1429  return AVERROR_INVALIDDATA;
1430  }
1431  sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1432  }
1433  }
1434  }
1435  }
1436  return 0;
1437 }
1438 
1439 /**
1440  * Decode pulse data; reference: table 4.7.
1441  */
1442 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1443  const uint16_t *swb_offset, int num_swb)
1444 {
1445  int i, pulse_swb;
1446  pulse->num_pulse = get_bits(gb, 2) + 1;
1447  pulse_swb = get_bits(gb, 6);
1448  if (pulse_swb >= num_swb)
1449  return -1;
1450  pulse->pos[0] = swb_offset[pulse_swb];
1451  pulse->pos[0] += get_bits(gb, 5);
1452  if (pulse->pos[0] >= swb_offset[num_swb])
1453  return -1;
1454  pulse->amp[0] = get_bits(gb, 4);
1455  for (i = 1; i < pulse->num_pulse; i++) {
1456  pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1457  if (pulse->pos[i] >= swb_offset[num_swb])
1458  return -1;
1459  pulse->amp[i] = get_bits(gb, 4);
1460  }
1461  return 0;
1462 }
1463 
1464 /**
1465  * Decode Temporal Noise Shaping data; reference: table 4.48.
1466  *
1467  * @return Returns error status. 0 - OK, !0 - error
1468  */
1470  GetBitContext *gb, const IndividualChannelStream *ics)
1471 {
1472  int w, filt, i, coef_len, coef_res, coef_compress;
1473  const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1474  const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1475  for (w = 0; w < ics->num_windows; w++) {
1476  if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1477  coef_res = get_bits1(gb);
1478 
1479  for (filt = 0; filt < tns->n_filt[w]; filt++) {
1480  int tmp2_idx;
1481  tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1482 
1483  if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1484  av_log(ac->avctx, AV_LOG_ERROR,
1485  "TNS filter order %d is greater than maximum %d.\n",
1486  tns->order[w][filt], tns_max_order);
1487  tns->order[w][filt] = 0;
1488  return AVERROR_INVALIDDATA;
1489  }
1490  if (tns->order[w][filt]) {
1491  tns->direction[w][filt] = get_bits1(gb);
1492  coef_compress = get_bits1(gb);
1493  coef_len = coef_res + 3 - coef_compress;
1494  tmp2_idx = 2 * coef_compress + coef_res;
1495 
1496  for (i = 0; i < tns->order[w][filt]; i++)
1497  tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1498  }
1499  }
1500  }
1501  }
1502  return 0;
1503 }
1504 
1505 /**
1506  * Decode Mid/Side data; reference: table 4.54.
1507  *
1508  * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1509  * [1] mask is decoded from bitstream; [2] mask is all 1s;
1510  * [3] reserved for scalable AAC
1511  */
1513  int ms_present)
1514 {
1515  int idx;
1516  if (ms_present == 1) {
1517  for (idx = 0;
1518  idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1519  idx++)
1520  cpe->ms_mask[idx] = get_bits1(gb);
1521  } else if (ms_present == 2) {
1522  memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1523  }
1524 }
1525 
1526 #ifndef VMUL2
1527 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1528  const float *scale)
1529 {
1530  float s = *scale;
1531  *dst++ = v[idx & 15] * s;
1532  *dst++ = v[idx>>4 & 15] * s;
1533  return dst;
1534 }
1535 #endif
1536 
1537 #ifndef VMUL4
1538 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1539  const float *scale)
1540 {
1541  float s = *scale;
1542  *dst++ = v[idx & 3] * s;
1543  *dst++ = v[idx>>2 & 3] * s;
1544  *dst++ = v[idx>>4 & 3] * s;
1545  *dst++ = v[idx>>6 & 3] * s;
1546  return dst;
1547 }
1548 #endif
1549 
1550 #ifndef VMUL2S
1551 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1552  unsigned sign, const float *scale)
1553 {
1554  union av_intfloat32 s0, s1;
1555 
1556  s0.f = s1.f = *scale;
1557  s0.i ^= sign >> 1 << 31;
1558  s1.i ^= sign << 31;
1559 
1560  *dst++ = v[idx & 15] * s0.f;
1561  *dst++ = v[idx>>4 & 15] * s1.f;
1562 
1563  return dst;
1564 }
1565 #endif
1566 
1567 #ifndef VMUL4S
1568 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1569  unsigned sign, const float *scale)
1570 {
1571  unsigned nz = idx >> 12;
1572  union av_intfloat32 s = { .f = *scale };
1573  union av_intfloat32 t;
1574 
1575  t.i = s.i ^ (sign & 1U<<31);
1576  *dst++ = v[idx & 3] * t.f;
1577 
1578  sign <<= nz & 1; nz >>= 1;
1579  t.i = s.i ^ (sign & 1U<<31);
1580  *dst++ = v[idx>>2 & 3] * t.f;
1581 
1582  sign <<= nz & 1; nz >>= 1;
1583  t.i = s.i ^ (sign & 1U<<31);
1584  *dst++ = v[idx>>4 & 3] * t.f;
1585 
1586  sign <<= nz & 1;
1587  t.i = s.i ^ (sign & 1U<<31);
1588  *dst++ = v[idx>>6 & 3] * t.f;
1589 
1590  return dst;
1591 }
1592 #endif
1593 
1594 /**
1595  * Decode spectral data; reference: table 4.50.
1596  * Dequantize and scale spectral data; reference: 4.6.3.3.
1597  *
1598  * @param coef array of dequantized, scaled spectral data
1599  * @param sf array of scalefactors or intensity stereo positions
1600  * @param pulse_present set if pulses are present
1601  * @param pulse pointer to pulse data struct
1602  * @param band_type array of the used band type
1603  *
1604  * @return Returns error status. 0 - OK, !0 - error
1605  */
1606 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1607  GetBitContext *gb, const float sf[120],
1608  int pulse_present, const Pulse *pulse,
1609  const IndividualChannelStream *ics,
1610  enum BandType band_type[120])
1611 {
1612  int i, k, g, idx = 0;
1613  const int c = 1024 / ics->num_windows;
1614  const uint16_t *offsets = ics->swb_offset;
1615  float *coef_base = coef;
1616 
1617  for (g = 0; g < ics->num_windows; g++)
1618  memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1619  sizeof(float) * (c - offsets[ics->max_sfb]));
1620 
1621  for (g = 0; g < ics->num_window_groups; g++) {
1622  unsigned g_len = ics->group_len[g];
1623 
1624  for (i = 0; i < ics->max_sfb; i++, idx++) {
1625  const unsigned cbt_m1 = band_type[idx] - 1;
1626  float *cfo = coef + offsets[i];
1627  int off_len = offsets[i + 1] - offsets[i];
1628  int group;
1629 
1630  if (cbt_m1 >= INTENSITY_BT2 - 1) {
1631  for (group = 0; group < g_len; group++, cfo+=128) {
1632  memset(cfo, 0, off_len * sizeof(float));
1633  }
1634  } else if (cbt_m1 == NOISE_BT - 1) {
1635  for (group = 0; group < g_len; group++, cfo+=128) {
1636  float scale;
1637  float band_energy;
1638 
1639  for (k = 0; k < off_len; k++) {
1641  cfo[k] = ac->random_state;
1642  }
1643 
1644  band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1645  scale = sf[idx] / sqrtf(band_energy);
1646  ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1647  }
1648  } else {
1649  const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1650  const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1651  VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1652  OPEN_READER(re, gb);
1653 
1654  switch (cbt_m1 >> 1) {
1655  case 0:
1656  for (group = 0; group < g_len; group++, cfo+=128) {
1657  float *cf = cfo;
1658  int len = off_len;
1659 
1660  do {
1661  int code;
1662  unsigned cb_idx;
1663 
1664  UPDATE_CACHE(re, gb);
1665  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1666  cb_idx = cb_vector_idx[code];
1667  cf = VMUL4(cf, vq, cb_idx, sf + idx);
1668  } while (len -= 4);
1669  }
1670  break;
1671 
1672  case 1:
1673  for (group = 0; group < g_len; group++, cfo+=128) {
1674  float *cf = cfo;
1675  int len = off_len;
1676 
1677  do {
1678  int code;
1679  unsigned nnz;
1680  unsigned cb_idx;
1681  uint32_t bits;
1682 
1683  UPDATE_CACHE(re, gb);
1684  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1685  cb_idx = cb_vector_idx[code];
1686  nnz = cb_idx >> 8 & 15;
1687  bits = nnz ? GET_CACHE(re, gb) : 0;
1688  LAST_SKIP_BITS(re, gb, nnz);
1689  cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1690  } while (len -= 4);
1691  }
1692  break;
1693 
1694  case 2:
1695  for (group = 0; group < g_len; group++, cfo+=128) {
1696  float *cf = cfo;
1697  int len = off_len;
1698 
1699  do {
1700  int code;
1701  unsigned cb_idx;
1702 
1703  UPDATE_CACHE(re, gb);
1704  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1705  cb_idx = cb_vector_idx[code];
1706  cf = VMUL2(cf, vq, cb_idx, sf + idx);
1707  } while (len -= 2);
1708  }
1709  break;
1710 
1711  case 3:
1712  case 4:
1713  for (group = 0; group < g_len; group++, cfo+=128) {
1714  float *cf = cfo;
1715  int len = off_len;
1716 
1717  do {
1718  int code;
1719  unsigned nnz;
1720  unsigned cb_idx;
1721  unsigned sign;
1722 
1723  UPDATE_CACHE(re, gb);
1724  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1725  cb_idx = cb_vector_idx[code];
1726  nnz = cb_idx >> 8 & 15;
1727  sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1728  LAST_SKIP_BITS(re, gb, nnz);
1729  cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1730  } while (len -= 2);
1731  }
1732  break;
1733 
1734  default:
1735  for (group = 0; group < g_len; group++, cfo+=128) {
1736  float *cf = cfo;
1737  uint32_t *icf = (uint32_t *) cf;
1738  int len = off_len;
1739 
1740  do {
1741  int code;
1742  unsigned nzt, nnz;
1743  unsigned cb_idx;
1744  uint32_t bits;
1745  int j;
1746 
1747  UPDATE_CACHE(re, gb);
1748  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1749 
1750  if (!code) {
1751  *icf++ = 0;
1752  *icf++ = 0;
1753  continue;
1754  }
1755 
1756  cb_idx = cb_vector_idx[code];
1757  nnz = cb_idx >> 12;
1758  nzt = cb_idx >> 8;
1759  bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1760  LAST_SKIP_BITS(re, gb, nnz);
1761 
1762  for (j = 0; j < 2; j++) {
1763  if (nzt & 1<<j) {
1764  uint32_t b;
1765  int n;
1766  /* The total length of escape_sequence must be < 22 bits according
1767  to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1768  UPDATE_CACHE(re, gb);
1769  b = GET_CACHE(re, gb);
1770  b = 31 - av_log2(~b);
1771 
1772  if (b > 8) {
1773  av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1774  return AVERROR_INVALIDDATA;
1775  }
1776 
1777  SKIP_BITS(re, gb, b + 1);
1778  b += 4;
1779  n = (1 << b) + SHOW_UBITS(re, gb, b);
1780  LAST_SKIP_BITS(re, gb, b);
1781  *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1782  bits <<= 1;
1783  } else {
1784  unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1785  *icf++ = (bits & 1U<<31) | v;
1786  bits <<= !!v;
1787  }
1788  cb_idx >>= 4;
1789  }
1790  } while (len -= 2);
1791 
1792  ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1793  }
1794  }
1795 
1796  CLOSE_READER(re, gb);
1797  }
1798  }
1799  coef += g_len << 7;
1800  }
1801 
1802  if (pulse_present) {
1803  idx = 0;
1804  for (i = 0; i < pulse->num_pulse; i++) {
1805  float co = coef_base[ pulse->pos[i] ];
1806  while (offsets[idx + 1] <= pulse->pos[i])
1807  idx++;
1808  if (band_type[idx] != NOISE_BT && sf[idx]) {
1809  float ico = -pulse->amp[i];
1810  if (co) {
1811  co /= sf[idx];
1812  ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1813  }
1814  coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1815  }
1816  }
1817  }
1818  return 0;
1819 }
1820 
1821 static av_always_inline float flt16_round(float pf)
1822 {
1823  union av_intfloat32 tmp;
1824  tmp.f = pf;
1825  tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1826  return tmp.f;
1827 }
1828 
1829 static av_always_inline float flt16_even(float pf)
1830 {
1831  union av_intfloat32 tmp;
1832  tmp.f = pf;
1833  tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1834  return tmp.f;
1835 }
1836 
1837 static av_always_inline float flt16_trunc(float pf)
1838 {
1839  union av_intfloat32 pun;
1840  pun.f = pf;
1841  pun.i &= 0xFFFF0000U;
1842  return pun.f;
1843 }
1844 
1845 static av_always_inline void predict(PredictorState *ps, float *coef,
1846  int output_enable)
1847 {
1848  const float a = 0.953125; // 61.0 / 64
1849  const float alpha = 0.90625; // 29.0 / 32
1850  float e0, e1;
1851  float pv;
1852  float k1, k2;
1853  float r0 = ps->r0, r1 = ps->r1;
1854  float cor0 = ps->cor0, cor1 = ps->cor1;
1855  float var0 = ps->var0, var1 = ps->var1;
1856 
1857  k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1858  k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1859 
1860  pv = flt16_round(k1 * r0 + k2 * r1);
1861  if (output_enable)
1862  *coef += pv;
1863 
1864  e0 = *coef;
1865  e1 = e0 - k1 * r0;
1866 
1867  ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1868  ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1869  ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1870  ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1871 
1872  ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1873  ps->r0 = flt16_trunc(a * e0);
1874 }
1875 
1876 /**
1877  * Apply AAC-Main style frequency domain prediction.
1878  */
1880 {
1881  int sfb, k;
1882 
1883  if (!sce->ics.predictor_initialized) {
1885  sce->ics.predictor_initialized = 1;
1886  }
1887 
1888  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1889  for (sfb = 0;
1890  sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1891  sfb++) {
1892  for (k = sce->ics.swb_offset[sfb];
1893  k < sce->ics.swb_offset[sfb + 1];
1894  k++) {
1895  predict(&sce->predictor_state[k], &sce->coeffs[k],
1896  sce->ics.predictor_present &&
1897  sce->ics.prediction_used[sfb]);
1898  }
1899  }
1900  if (sce->ics.predictor_reset_group)
1902  sce->ics.predictor_reset_group);
1903  } else
1905 }
1906 
1907 /**
1908  * Decode an individual_channel_stream payload; reference: table 4.44.
1909  *
1910  * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1911  * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1912  *
1913  * @return Returns error status. 0 - OK, !0 - error
1914  */
1916  GetBitContext *gb, int common_window, int scale_flag)
1917 {
1918  Pulse pulse;
1919  TemporalNoiseShaping *tns = &sce->tns;
1920  IndividualChannelStream *ics = &sce->ics;
1921  float *out = sce->coeffs;
1922  int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1923  int ret;
1924 
1925  eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1926  er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1927  ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1928  ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1929  ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1930 
1931  /* This assignment is to silence a GCC warning about the variable being used
1932  * uninitialized when in fact it always is.
1933  */
1934  pulse.num_pulse = 0;
1935 
1936  global_gain = get_bits(gb, 8);
1937 
1938  if (!common_window && !scale_flag) {
1939  if (decode_ics_info(ac, ics, gb) < 0)
1940  return AVERROR_INVALIDDATA;
1941  }
1942 
1943  if ((ret = decode_band_types(ac, sce->band_type,
1944  sce->band_type_run_end, gb, ics)) < 0)
1945  return ret;
1946  if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1947  sce->band_type, sce->band_type_run_end)) < 0)
1948  return ret;
1949 
1950  pulse_present = 0;
1951  if (!scale_flag) {
1952  if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1953  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1954  av_log(ac->avctx, AV_LOG_ERROR,
1955  "Pulse tool not allowed in eight short sequence.\n");
1956  return AVERROR_INVALIDDATA;
1957  }
1958  if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1959  av_log(ac->avctx, AV_LOG_ERROR,
1960  "Pulse data corrupt or invalid.\n");
1961  return AVERROR_INVALIDDATA;
1962  }
1963  }
1964  tns->present = get_bits1(gb);
1965  if (tns->present && !er_syntax)
1966  if (decode_tns(ac, tns, gb, ics) < 0)
1967  return AVERROR_INVALIDDATA;
1968  if (!eld_syntax && get_bits1(gb)) {
1969  avpriv_request_sample(ac->avctx, "SSR");
1970  return AVERROR_PATCHWELCOME;
1971  }
1972  // I see no textual basis in the spec for this occurring after SSR gain
1973  // control, but this is what both reference and real implmentations do
1974  if (tns->present && er_syntax)
1975  if (decode_tns(ac, tns, gb, ics) < 0)
1976  return AVERROR_INVALIDDATA;
1977  }
1978 
1979  if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1980  &pulse, ics, sce->band_type) < 0)
1981  return AVERROR_INVALIDDATA;
1982 
1983  if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1984  apply_prediction(ac, sce);
1985 
1986  return 0;
1987 }
1988 
1989 /**
1990  * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1991  */
1993 {
1994  const IndividualChannelStream *ics = &cpe->ch[0].ics;
1995  float *ch0 = cpe->ch[0].coeffs;
1996  float *ch1 = cpe->ch[1].coeffs;
1997  int g, i, group, idx = 0;
1998  const uint16_t *offsets = ics->swb_offset;
1999  for (g = 0; g < ics->num_window_groups; g++) {
2000  for (i = 0; i < ics->max_sfb; i++, idx++) {
2001  if (cpe->ms_mask[idx] &&
2002  cpe->ch[0].band_type[idx] < NOISE_BT &&
2003  cpe->ch[1].band_type[idx] < NOISE_BT) {
2004  for (group = 0; group < ics->group_len[g]; group++) {
2005  ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
2006  ch1 + group * 128 + offsets[i],
2007  offsets[i+1] - offsets[i]);
2008  }
2009  }
2010  }
2011  ch0 += ics->group_len[g] * 128;
2012  ch1 += ics->group_len[g] * 128;
2013  }
2014 }
2015 
2016 /**
2017  * intensity stereo decoding; reference: 4.6.8.2.3
2018  *
2019  * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2020  * [1] mask is decoded from bitstream; [2] mask is all 1s;
2021  * [3] reserved for scalable AAC
2022  */
2024  ChannelElement *cpe, int ms_present)
2025 {
2026  const IndividualChannelStream *ics = &cpe->ch[1].ics;
2027  SingleChannelElement *sce1 = &cpe->ch[1];
2028  float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2029  const uint16_t *offsets = ics->swb_offset;
2030  int g, group, i, idx = 0;
2031  int c;
2032  float scale;
2033  for (g = 0; g < ics->num_window_groups; g++) {
2034  for (i = 0; i < ics->max_sfb;) {
2035  if (sce1->band_type[idx] == INTENSITY_BT ||
2036  sce1->band_type[idx] == INTENSITY_BT2) {
2037  const int bt_run_end = sce1->band_type_run_end[idx];
2038  for (; i < bt_run_end; i++, idx++) {
2039  c = -1 + 2 * (sce1->band_type[idx] - 14);
2040  if (ms_present)
2041  c *= 1 - 2 * cpe->ms_mask[idx];
2042  scale = c * sce1->sf[idx];
2043  for (group = 0; group < ics->group_len[g]; group++)
2044  ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2045  coef0 + group * 128 + offsets[i],
2046  scale,
2047  offsets[i + 1] - offsets[i]);
2048  }
2049  } else {
2050  int bt_run_end = sce1->band_type_run_end[idx];
2051  idx += bt_run_end - i;
2052  i = bt_run_end;
2053  }
2054  }
2055  coef0 += ics->group_len[g] * 128;
2056  coef1 += ics->group_len[g] * 128;
2057  }
2058 }
2059 
2060 /**
2061  * Decode a channel_pair_element; reference: table 4.4.
2062  *
2063  * @return Returns error status. 0 - OK, !0 - error
2064  */
2066 {
2067  int i, ret, common_window, ms_present = 0;
2068  int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2069 
2070  common_window = eld_syntax || get_bits1(gb);
2071  if (common_window) {
2072  if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2073  return AVERROR_INVALIDDATA;
2074  i = cpe->ch[1].ics.use_kb_window[0];
2075  cpe->ch[1].ics = cpe->ch[0].ics;
2076  cpe->ch[1].ics.use_kb_window[1] = i;
2077  if (cpe->ch[1].ics.predictor_present &&
2078  (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2079  if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2080  decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2081  ms_present = get_bits(gb, 2);
2082  if (ms_present == 3) {
2083  av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2084  return AVERROR_INVALIDDATA;
2085  } else if (ms_present)
2086  decode_mid_side_stereo(cpe, gb, ms_present);
2087  }
2088  if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2089  return ret;
2090  if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2091  return ret;
2092 
2093  if (common_window) {
2094  if (ms_present)
2095  apply_mid_side_stereo(ac, cpe);
2096  if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2097  apply_prediction(ac, &cpe->ch[0]);
2098  apply_prediction(ac, &cpe->ch[1]);
2099  }
2100  }
2101 
2102  apply_intensity_stereo(ac, cpe, ms_present);
2103  return 0;
2104 }
2105 
2106 static const float cce_scale[] = {
2107  1.09050773266525765921, //2^(1/8)
2108  1.18920711500272106672, //2^(1/4)
2109  M_SQRT2,
2110  2,
2111 };
2112 
2113 /**
2114  * Decode coupling_channel_element; reference: table 4.8.
2115  *
2116  * @return Returns error status. 0 - OK, !0 - error
2117  */
2119 {
2120  int num_gain = 0;
2121  int c, g, sfb, ret;
2122  int sign;
2123  float scale;
2124  SingleChannelElement *sce = &che->ch[0];
2125  ChannelCoupling *coup = &che->coup;
2126 
2127  coup->coupling_point = 2 * get_bits1(gb);
2128  coup->num_coupled = get_bits(gb, 3);
2129  for (c = 0; c <= coup->num_coupled; c++) {
2130  num_gain++;
2131  coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2132  coup->id_select[c] = get_bits(gb, 4);
2133  if (coup->type[c] == TYPE_CPE) {
2134  coup->ch_select[c] = get_bits(gb, 2);
2135  if (coup->ch_select[c] == 3)
2136  num_gain++;
2137  } else
2138  coup->ch_select[c] = 2;
2139  }
2140  coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2141 
2142  sign = get_bits(gb, 1);
2143  scale = cce_scale[get_bits(gb, 2)];
2144 
2145  if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2146  return ret;
2147 
2148  for (c = 0; c < num_gain; c++) {
2149  int idx = 0;
2150  int cge = 1;
2151  int gain = 0;
2152  float gain_cache = 1.0;
2153  if (c) {
2154  cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2155  gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2156  gain_cache = powf(scale, -gain);
2157  }
2158  if (coup->coupling_point == AFTER_IMDCT) {
2159  coup->gain[c][0] = gain_cache;
2160  } else {
2161  for (g = 0; g < sce->ics.num_window_groups; g++) {
2162  for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2163  if (sce->band_type[idx] != ZERO_BT) {
2164  if (!cge) {
2165  int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2166  if (t) {
2167  int s = 1;
2168  t = gain += t;
2169  if (sign) {
2170  s -= 2 * (t & 0x1);
2171  t >>= 1;
2172  }
2173  gain_cache = powf(scale, -t) * s;
2174  }
2175  }
2176  coup->gain[c][idx] = gain_cache;
2177  }
2178  }
2179  }
2180  }
2181  }
2182  return 0;
2183 }
2184 
2185 /**
2186  * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2187  *
2188  * @return Returns number of bytes consumed.
2189  */
2191  GetBitContext *gb)
2192 {
2193  int i;
2194  int num_excl_chan = 0;
2195 
2196  do {
2197  for (i = 0; i < 7; i++)
2198  che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2199  } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2200 
2201  return num_excl_chan / 7;
2202 }
2203 
2204 /**
2205  * Decode dynamic range information; reference: table 4.52.
2206  *
2207  * @return Returns number of bytes consumed.
2208  */
2210  GetBitContext *gb)
2211 {
2212  int n = 1;
2213  int drc_num_bands = 1;
2214  int i;
2215 
2216  /* pce_tag_present? */
2217  if (get_bits1(gb)) {
2218  che_drc->pce_instance_tag = get_bits(gb, 4);
2219  skip_bits(gb, 4); // tag_reserved_bits
2220  n++;
2221  }
2222 
2223  /* excluded_chns_present? */
2224  if (get_bits1(gb)) {
2225  n += decode_drc_channel_exclusions(che_drc, gb);
2226  }
2227 
2228  /* drc_bands_present? */
2229  if (get_bits1(gb)) {
2230  che_drc->band_incr = get_bits(gb, 4);
2231  che_drc->interpolation_scheme = get_bits(gb, 4);
2232  n++;
2233  drc_num_bands += che_drc->band_incr;
2234  for (i = 0; i < drc_num_bands; i++) {
2235  che_drc->band_top[i] = get_bits(gb, 8);
2236  n++;
2237  }
2238  }
2239 
2240  /* prog_ref_level_present? */
2241  if (get_bits1(gb)) {
2242  che_drc->prog_ref_level = get_bits(gb, 7);
2243  skip_bits1(gb); // prog_ref_level_reserved_bits
2244  n++;
2245  }
2246 
2247  for (i = 0; i < drc_num_bands; i++) {
2248  che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2249  che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2250  n++;
2251  }
2252 
2253  return n;
2254 }
2255 
2256 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2257  uint8_t buf[256];
2258  int i, major, minor;
2259 
2260  if (len < 13+7*8)
2261  goto unknown;
2262 
2263  get_bits(gb, 13); len -= 13;
2264 
2265  for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2266  buf[i] = get_bits(gb, 8);
2267 
2268  buf[i] = 0;
2269  if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2270  av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2271 
2272  if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2273  ac->avctx->internal->skip_samples = 1024;
2274  }
2275 
2276 unknown:
2277  skip_bits_long(gb, len);
2278 
2279  return 0;
2280 }
2281 
2282 /**
2283  * Decode extension data (incomplete); reference: table 4.51.
2284  *
2285  * @param cnt length of TYPE_FIL syntactic element in bytes
2286  *
2287  * @return Returns number of bytes consumed
2288  */
2290  ChannelElement *che, enum RawDataBlockType elem_type)
2291 {
2292  int crc_flag = 0;
2293  int res = cnt;
2294  int type = get_bits(gb, 4);
2295 
2296  if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2297  av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2298 
2299  switch (type) { // extension type
2300  case EXT_SBR_DATA_CRC:
2301  crc_flag++;
2302  case EXT_SBR_DATA:
2303  if (!che) {
2304  av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2305  return res;
2306  } else if (!ac->oc[1].m4ac.sbr) {
2307  av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2308  skip_bits_long(gb, 8 * cnt - 4);
2309  return res;
2310  } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2311  av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2312  skip_bits_long(gb, 8 * cnt - 4);
2313  return res;
2314  } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2315  ac->oc[1].m4ac.sbr = 1;
2316  ac->oc[1].m4ac.ps = 1;
2318  output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2319  ac->oc[1].status, 1);
2320  } else {
2321  ac->oc[1].m4ac.sbr = 1;
2323  }
2324  res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2325  break;
2326  case EXT_DYNAMIC_RANGE:
2327  res = decode_dynamic_range(&ac->che_drc, gb);
2328  break;
2329  case EXT_FILL:
2330  decode_fill(ac, gb, 8 * cnt - 4);
2331  break;
2332  case EXT_FILL_DATA:
2333  case EXT_DATA_ELEMENT:
2334  default:
2335  skip_bits_long(gb, 8 * cnt - 4);
2336  break;
2337  };
2338  return res;
2339 }
2340 
2341 /**
2342  * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2343  *
2344  * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2345  * @param coef spectral coefficients
2346  */
2347 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2348  IndividualChannelStream *ics, int decode)
2349 {
2350  const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2351  int w, filt, m, i;
2352  int bottom, top, order, start, end, size, inc;
2353  float lpc[TNS_MAX_ORDER];
2354  float tmp[TNS_MAX_ORDER+1];
2355 
2356  for (w = 0; w < ics->num_windows; w++) {
2357  bottom = ics->num_swb;
2358  for (filt = 0; filt < tns->n_filt[w]; filt++) {
2359  top = bottom;
2360  bottom = FFMAX(0, top - tns->length[w][filt]);
2361  order = tns->order[w][filt];
2362  if (order == 0)
2363  continue;
2364 
2365  // tns_decode_coef
2366  compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2367 
2368  start = ics->swb_offset[FFMIN(bottom, mmm)];
2369  end = ics->swb_offset[FFMIN( top, mmm)];
2370  if ((size = end - start) <= 0)
2371  continue;
2372  if (tns->direction[w][filt]) {
2373  inc = -1;
2374  start = end - 1;
2375  } else {
2376  inc = 1;
2377  }
2378  start += w * 128;
2379 
2380  if (decode) {
2381  // ar filter
2382  for (m = 0; m < size; m++, start += inc)
2383  for (i = 1; i <= FFMIN(m, order); i++)
2384  coef[start] -= coef[start - i * inc] * lpc[i - 1];
2385  } else {
2386  // ma filter
2387  for (m = 0; m < size; m++, start += inc) {
2388  tmp[0] = coef[start];
2389  for (i = 1; i <= FFMIN(m, order); i++)
2390  coef[start] += tmp[i] * lpc[i - 1];
2391  for (i = order; i > 0; i--)
2392  tmp[i] = tmp[i - 1];
2393  }
2394  }
2395  }
2396  }
2397 }
2398 
2399 /**
2400  * Apply windowing and MDCT to obtain the spectral
2401  * coefficient from the predicted sample by LTP.
2402  */
2403 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2404  float *in, IndividualChannelStream *ics)
2405 {
2406  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2407  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2408  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2409  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2410 
2411  if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2412  ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2413  } else {
2414  memset(in, 0, 448 * sizeof(float));
2415  ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2416  }
2417  if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2418  ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2419  } else {
2420  ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2421  memset(in + 1024 + 576, 0, 448 * sizeof(float));
2422  }
2423  ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2424 }
2425 
2426 /**
2427  * Apply the long term prediction
2428  */
2430 {
2431  const LongTermPrediction *ltp = &sce->ics.ltp;
2432  const uint16_t *offsets = sce->ics.swb_offset;
2433  int i, sfb;
2434 
2435  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2436  float *predTime = sce->ret;
2437  float *predFreq = ac->buf_mdct;
2438  int16_t num_samples = 2048;
2439 
2440  if (ltp->lag < 1024)
2441  num_samples = ltp->lag + 1024;
2442  for (i = 0; i < num_samples; i++)
2443  predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2444  memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2445 
2446  ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2447 
2448  if (sce->tns.present)
2449  ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2450 
2451  for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2452  if (ltp->used[sfb])
2453  for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2454  sce->coeffs[i] += predFreq[i];
2455  }
2456 }
2457 
2458 /**
2459  * Update the LTP buffer for next frame
2460  */
2462 {
2463  IndividualChannelStream *ics = &sce->ics;
2464  float *saved = sce->saved;
2465  float *saved_ltp = sce->coeffs;
2466  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2467  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2468  int i;
2469 
2470  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2471  memcpy(saved_ltp, saved, 512 * sizeof(float));
2472  memset(saved_ltp + 576, 0, 448 * sizeof(float));
2473  ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2474  for (i = 0; i < 64; i++)
2475  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2476  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2477  memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2478  memset(saved_ltp + 576, 0, 448 * sizeof(float));
2479  ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2480  for (i = 0; i < 64; i++)
2481  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2482  } else { // LONG_STOP or ONLY_LONG
2483  ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2484  for (i = 0; i < 512; i++)
2485  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2486  }
2487 
2488  memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2489  memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2490  memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2491 }
2492 
2493 /**
2494  * Conduct IMDCT and windowing.
2495  */
2497 {
2498  IndividualChannelStream *ics = &sce->ics;
2499  float *in = sce->coeffs;
2500  float *out = sce->ret;
2501  float *saved = sce->saved;
2502  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2503  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2504  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2505  float *buf = ac->buf_mdct;
2506  float *temp = ac->temp;
2507  int i;
2508 
2509  // imdct
2510  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2511  for (i = 0; i < 1024; i += 128)
2512  ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2513  } else
2514  ac->mdct.imdct_half(&ac->mdct, buf, in);
2515 
2516  /* window overlapping
2517  * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2518  * and long to short transitions are considered to be short to short
2519  * transitions. This leaves just two cases (long to long and short to short)
2520  * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2521  */
2522  if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2524  ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2525  } else {
2526  memcpy( out, saved, 448 * sizeof(float));
2527 
2528  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2529  ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2530  ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2531  ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2532  ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2533  ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2534  memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2535  } else {
2536  ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2537  memcpy( out + 576, buf + 64, 448 * sizeof(float));
2538  }
2539  }
2540 
2541  // buffer update
2542  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2543  memcpy( saved, temp + 64, 64 * sizeof(float));
2544  ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2545  ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2546  ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2547  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2548  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2549  memcpy( saved, buf + 512, 448 * sizeof(float));
2550  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2551  } else { // LONG_STOP or ONLY_LONG
2552  memcpy( saved, buf + 512, 512 * sizeof(float));
2553  }
2554 }
2555 
2557 {
2558  IndividualChannelStream *ics = &sce->ics;
2559  float *in = sce->coeffs;
2560  float *out = sce->ret;
2561  float *saved = sce->saved;
2562  float *buf = ac->buf_mdct;
2563 
2564  // imdct
2565  ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2566 
2567  // window overlapping
2568  if (ics->use_kb_window[1]) {
2569  // AAC LD uses a low overlap sine window instead of a KBD window
2570  memcpy(out, saved, 192 * sizeof(float));
2571  ac->fdsp.vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2572  memcpy( out + 320, buf + 64, 192 * sizeof(float));
2573  } else {
2574  ac->fdsp.vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2575  }
2576 
2577  // buffer update
2578  memcpy(saved, buf + 256, 256 * sizeof(float));
2579 }
2580 
2582 {
2583  float *in = sce->coeffs;
2584  float *out = sce->ret;
2585  float *saved = sce->saved;
2586  const float *const window = ff_aac_eld_window;
2587  float *buf = ac->buf_mdct;
2588  int i;
2589  const int n = 512;
2590  const int n2 = n >> 1;
2591  const int n4 = n >> 2;
2592 
2593  // Inverse transform, mapped to the conventional IMDCT by
2594  // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2595  // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2596  // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2597  // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2598  for (i = 0; i < n2; i+=2) {
2599  float temp;
2600  temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2601  temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2602  }
2603  ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2604  for (i = 0; i < n; i+=2) {
2605  buf[i] = -buf[i];
2606  }
2607  // Like with the regular IMDCT at this point we still have the middle half
2608  // of a transform but with even symmetry on the left and odd symmetry on
2609  // the right
2610 
2611  // window overlapping
2612  // The spec says to use samples [0..511] but the reference decoder uses
2613  // samples [128..639].
2614  for (i = n4; i < n2; i ++) {
2615  out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2616  saved[ i + n2] * window[i + n - n4] +
2617  -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2618  -saved[2*n + n2 + i] * window[i + 3*n - n4];
2619  }
2620  for (i = 0; i < n2; i ++) {
2621  out[n4 + i] = buf[i] * window[i + n2 - n4] +
2622  -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2623  -saved[ n + i] * window[i + n2 + 2*n - n4] +
2624  saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2625  }
2626  for (i = 0; i < n4; i ++) {
2627  out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2628  -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2629  -saved[ n + n2 + i] * window[i + 3*n - n4];
2630  }
2631 
2632  // buffer update
2633  memmove(saved + n, saved, 2 * n * sizeof(float));
2634  memcpy( saved, buf, n * sizeof(float));
2635 }
2636 
2637 /**
2638  * Apply dependent channel coupling (applied before IMDCT).
2639  *
2640  * @param index index into coupling gain array
2641  */
2643  SingleChannelElement *target,
2644  ChannelElement *cce, int index)
2645 {
2646  IndividualChannelStream *ics = &cce->ch[0].ics;
2647  const uint16_t *offsets = ics->swb_offset;
2648  float *dest = target->coeffs;
2649  const float *src = cce->ch[0].coeffs;
2650  int g, i, group, k, idx = 0;
2651  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2652  av_log(ac->avctx, AV_LOG_ERROR,
2653  "Dependent coupling is not supported together with LTP\n");
2654  return;
2655  }
2656  for (g = 0; g < ics->num_window_groups; g++) {
2657  for (i = 0; i < ics->max_sfb; i++, idx++) {
2658  if (cce->ch[0].band_type[idx] != ZERO_BT) {
2659  const float gain = cce->coup.gain[index][idx];
2660  for (group = 0; group < ics->group_len[g]; group++) {
2661  for (k = offsets[i]; k < offsets[i + 1]; k++) {
2662  // FIXME: SIMDify
2663  dest[group * 128 + k] += gain * src[group * 128 + k];
2664  }
2665  }
2666  }
2667  }
2668  dest += ics->group_len[g] * 128;
2669  src += ics->group_len[g] * 128;
2670  }
2671 }
2672 
2673 /**
2674  * Apply independent channel coupling (applied after IMDCT).
2675  *
2676  * @param index index into coupling gain array
2677  */
2679  SingleChannelElement *target,
2680  ChannelElement *cce, int index)
2681 {
2682  int i;
2683  const float gain = cce->coup.gain[index][0];
2684  const float *src = cce->ch[0].ret;
2685  float *dest = target->ret;
2686  const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2687 
2688  for (i = 0; i < len; i++)
2689  dest[i] += gain * src[i];
2690 }
2691 
2692 /**
2693  * channel coupling transformation interface
2694  *
2695  * @param apply_coupling_method pointer to (in)dependent coupling function
2696  */
2698  enum RawDataBlockType type, int elem_id,
2699  enum CouplingPoint coupling_point,
2700  void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2701 {
2702  int i, c;
2703 
2704  for (i = 0; i < MAX_ELEM_ID; i++) {
2705  ChannelElement *cce = ac->che[TYPE_CCE][i];
2706  int index = 0;
2707 
2708  if (cce && cce->coup.coupling_point == coupling_point) {
2709  ChannelCoupling *coup = &cce->coup;
2710 
2711  for (c = 0; c <= coup->num_coupled; c++) {
2712  if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2713  if (coup->ch_select[c] != 1) {
2714  apply_coupling_method(ac, &cc->ch[0], cce, index);
2715  if (coup->ch_select[c] != 0)
2716  index++;
2717  }
2718  if (coup->ch_select[c] != 2)
2719  apply_coupling_method(ac, &cc->ch[1], cce, index++);
2720  } else
2721  index += 1 + (coup->ch_select[c] == 3);
2722  }
2723  }
2724  }
2725 }
2726 
2727 /**
2728  * Convert spectral data to float samples, applying all supported tools as appropriate.
2729  */
2731 {
2732  int i, type;
2734  switch (ac->oc[1].m4ac.object_type) {
2735  case AOT_ER_AAC_LD:
2737  break;
2738  case AOT_ER_AAC_ELD:
2740  break;
2741  default:
2743  }
2744  for (type = 3; type >= 0; type--) {
2745  for (i = 0; i < MAX_ELEM_ID; i++) {
2746  ChannelElement *che = ac->che[type][i];
2747  if (che && che->present) {
2748  if (type <= TYPE_CPE)
2750  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2751  if (che->ch[0].ics.predictor_present) {
2752  if (che->ch[0].ics.ltp.present)
2753  ac->apply_ltp(ac, &che->ch[0]);
2754  if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2755  ac->apply_ltp(ac, &che->ch[1]);
2756  }
2757  }
2758  if (che->ch[0].tns.present)
2759  ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2760  if (che->ch[1].tns.present)
2761  ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2762  if (type <= TYPE_CPE)
2764  if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2765  imdct_and_window(ac, &che->ch[0]);
2766  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2767  ac->update_ltp(ac, &che->ch[0]);
2768  if (type == TYPE_CPE) {
2769  imdct_and_window(ac, &che->ch[1]);
2770  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2771  ac->update_ltp(ac, &che->ch[1]);
2772  }
2773  if (ac->oc[1].m4ac.sbr > 0) {
2774  ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2775  }
2776  }
2777  if (type <= TYPE_CCE)
2779  che->present = 0;
2780  } else if (che) {
2781  av_log(ac->avctx, AV_LOG_WARNING, "ChannelElement %d.%d missing \n", type, i);
2782  }
2783  }
2784  }
2785 }
2786 
2788 {
2789  int size;
2790  AACADTSHeaderInfo hdr_info;
2791  uint8_t layout_map[MAX_ELEM_ID*4][3];
2792  int layout_map_tags, ret;
2793 
2794  size = avpriv_aac_parse_header(gb, &hdr_info);
2795  if (size > 0) {
2796  if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2797  // This is 2 for "VLB " audio in NSV files.
2798  // See samples/nsv/vlb_audio.
2800  "More than one AAC RDB per ADTS frame");
2801  ac->warned_num_aac_frames = 1;
2802  }
2804  if (hdr_info.chan_config) {
2805  ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2806  if ((ret = set_default_channel_config(ac->avctx,
2807  layout_map,
2808  &layout_map_tags,
2809  hdr_info.chan_config)) < 0)
2810  return ret;
2811  if ((ret = output_configure(ac, layout_map, layout_map_tags,
2812  FFMAX(ac->oc[1].status,
2813  OC_TRIAL_FRAME), 0)) < 0)
2814  return ret;
2815  } else {
2816  ac->oc[1].m4ac.chan_config = 0;
2817  /**
2818  * dual mono frames in Japanese DTV can have chan_config 0
2819  * WITHOUT specifying PCE.
2820  * thus, set dual mono as default.
2821  */
2822  if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2823  layout_map_tags = 2;
2824  layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2825  layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2826  layout_map[0][1] = 0;
2827  layout_map[1][1] = 1;
2828  if (output_configure(ac, layout_map, layout_map_tags,
2829  OC_TRIAL_FRAME, 0))
2830  return -7;
2831  }
2832  }
2833  ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2834  ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2835  ac->oc[1].m4ac.object_type = hdr_info.object_type;
2836  if (ac->oc[0].status != OC_LOCKED ||
2837  ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2838  ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2839  ac->oc[1].m4ac.sbr = -1;
2840  ac->oc[1].m4ac.ps = -1;
2841  }
2842  if (!hdr_info.crc_absent)
2843  skip_bits(gb, 16);
2844  }
2845  return size;
2846 }
2847 
2848 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2849  int *got_frame_ptr, GetBitContext *gb)
2850 {
2851  AACContext *ac = avctx->priv_data;
2852  ChannelElement *che;
2853  int err, i;
2854  int samples = 1024;
2855  int chan_config = ac->oc[1].m4ac.chan_config;
2856  int aot = ac->oc[1].m4ac.object_type;
2857 
2858  if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2859  samples >>= 1;
2860 
2861  ac->frame = data;
2862 
2863  if ((err = frame_configure_elements(avctx)) < 0)
2864  return err;
2865 
2866  // The FF_PROFILE_AAC_* defines are all object_type - 1
2867  // This may lead to an undefined profile being signaled
2868  ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2869 
2870  ac->tags_mapped = 0;
2871 
2872  if (chan_config < 0 || chan_config >= 8) {
2873  avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2874  ac->oc[1].m4ac.chan_config);
2875  return AVERROR_INVALIDDATA;
2876  }
2877  for (i = 0; i < tags_per_config[chan_config]; i++) {
2878  const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2879  const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2880  if (!(che=get_che(ac, elem_type, elem_id))) {
2881  av_log(ac->avctx, AV_LOG_ERROR,
2882  "channel element %d.%d is not allocated\n",
2883  elem_type, elem_id);
2884  return AVERROR_INVALIDDATA;
2885  }
2886  che->present = 1;
2887  if (aot != AOT_ER_AAC_ELD)
2888  skip_bits(gb, 4);
2889  switch (elem_type) {
2890  case TYPE_SCE:
2891  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2892  break;
2893  case TYPE_CPE:
2894  err = decode_cpe(ac, gb, che);
2895  break;
2896  case TYPE_LFE:
2897  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2898  break;
2899  }
2900  if (err < 0)
2901  return err;
2902  }
2903 
2904  spectral_to_sample(ac);
2905 
2906  ac->frame->nb_samples = samples;
2907  ac->frame->sample_rate = avctx->sample_rate;
2908  *got_frame_ptr = 1;
2909 
2910  skip_bits_long(gb, get_bits_left(gb));
2911  return 0;
2912 }
2913 
2914 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2915  int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2916 {
2917  AACContext *ac = avctx->priv_data;
2918  ChannelElement *che = NULL, *che_prev = NULL;
2919  enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2920  int err, elem_id;
2921  int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2922  int is_dmono, sce_count = 0;
2923 
2924  ac->frame = data;
2925 
2926  if (show_bits(gb, 12) == 0xfff) {
2927  if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2928  av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2929  goto fail;
2930  }
2931  if (ac->oc[1].m4ac.sampling_index > 12) {
2932  av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2933  err = AVERROR_INVALIDDATA;
2934  goto fail;
2935  }
2936  }
2937 
2938  if ((err = frame_configure_elements(avctx)) < 0)
2939  goto fail;
2940 
2941  // The FF_PROFILE_AAC_* defines are all object_type - 1
2942  // This may lead to an undefined profile being signaled
2943  ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2944 
2945  ac->tags_mapped = 0;
2946  // parse
2947  while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2948  elem_id = get_bits(gb, 4);
2949 
2950  if (avctx->debug & FF_DEBUG_STARTCODE)
2951  av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2952 
2953  if (elem_type < TYPE_DSE) {
2954  if (!(che=get_che(ac, elem_type, elem_id))) {
2955  av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2956  elem_type, elem_id);
2957  err = AVERROR_INVALIDDATA;
2958  goto fail;
2959  }
2960  samples = 1024;
2961  che->present = 1;
2962  }
2963 
2964  switch (elem_type) {
2965 
2966  case TYPE_SCE:
2967  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2968  audio_found = 1;
2969  sce_count++;
2970  break;
2971 
2972  case TYPE_CPE:
2973  err = decode_cpe(ac, gb, che);
2974  audio_found = 1;
2975  break;
2976 
2977  case TYPE_CCE:
2978  err = decode_cce(ac, gb, che);
2979  break;
2980 
2981  case TYPE_LFE:
2982  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2983  audio_found = 1;
2984  break;
2985 
2986  case TYPE_DSE:
2987  err = skip_data_stream_element(ac, gb);
2988  break;
2989 
2990  case TYPE_PCE: {
2991  uint8_t layout_map[MAX_ELEM_ID*4][3];
2992  int tags;
2994  tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2995  if (tags < 0) {
2996  err = tags;
2997  break;
2998  }
2999  if (pce_found) {
3000  av_log(avctx, AV_LOG_ERROR,
3001  "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3002  } else {
3003  err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3004  if (!err)
3005  ac->oc[1].m4ac.chan_config = 0;
3006  pce_found = 1;
3007  }
3008  break;
3009  }
3010 
3011  case TYPE_FIL:
3012  if (elem_id == 15)
3013  elem_id += get_bits(gb, 8) - 1;
3014  if (get_bits_left(gb) < 8 * elem_id) {
3015  av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3016  err = AVERROR_INVALIDDATA;
3017  goto fail;
3018  }
3019  while (elem_id > 0)
3020  elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3021  err = 0; /* FIXME */
3022  break;
3023 
3024  default:
3025  err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3026  break;
3027  }
3028 
3029  che_prev = che;
3030  elem_type_prev = elem_type;
3031 
3032  if (err)
3033  goto fail;
3034 
3035  if (get_bits_left(gb) < 3) {
3036  av_log(avctx, AV_LOG_ERROR, overread_err);
3037  err = AVERROR_INVALIDDATA;
3038  goto fail;
3039  }
3040  }
3041 
3042  spectral_to_sample(ac);
3043 
3044  multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3045  samples <<= multiplier;
3046 
3047  if (ac->oc[1].status && audio_found) {
3048  avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3049  avctx->frame_size = samples;
3050  ac->oc[1].status = OC_LOCKED;
3051  }
3052 
3053  if (multiplier) {
3054  int side_size;
3055  const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3056  if (side && side_size>=4)
3057  AV_WL32(side, 2*AV_RL32(side));
3058  }
3059 
3060  *got_frame_ptr = !!samples;
3061  if (samples) {
3062  ac->frame->nb_samples = samples;
3063  ac->frame->sample_rate = avctx->sample_rate;
3064  } else
3065  av_frame_unref(ac->frame);
3066  *got_frame_ptr = !!samples;
3067 
3068  /* for dual-mono audio (SCE + SCE) */
3069  is_dmono = ac->dmono_mode && sce_count == 2 &&
3071  if (is_dmono) {
3072  if (ac->dmono_mode == 1)
3073  ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3074  else if (ac->dmono_mode == 2)
3075  ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3076  }
3077 
3078  return 0;
3079 fail:
3081  return err;
3082 }
3083 
3084 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3085  int *got_frame_ptr, AVPacket *avpkt)
3086 {
3087  AACContext *ac = avctx->priv_data;
3088  const uint8_t *buf = avpkt->data;
3089  int buf_size = avpkt->size;
3090  GetBitContext gb;
3091  int buf_consumed;
3092  int buf_offset;
3093  int err;
3094  int new_extradata_size;
3095  const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3097  &new_extradata_size);
3098  int jp_dualmono_size;
3099  const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3101  &jp_dualmono_size);
3102 
3103  if (new_extradata && 0) {
3104  av_free(avctx->extradata);
3105  avctx->extradata = av_mallocz(new_extradata_size +
3107  if (!avctx->extradata)
3108  return AVERROR(ENOMEM);
3109  avctx->extradata_size = new_extradata_size;
3110  memcpy(avctx->extradata, new_extradata, new_extradata_size);
3112  if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3113  avctx->extradata,
3114  avctx->extradata_size*8, 1) < 0) {
3116  return AVERROR_INVALIDDATA;
3117  }
3118  }
3119 
3120  ac->dmono_mode = 0;
3121  if (jp_dualmono && jp_dualmono_size > 0)
3122  ac->dmono_mode = 1 + *jp_dualmono;
3123  if (ac->force_dmono_mode >= 0)
3124  ac->dmono_mode = ac->force_dmono_mode;
3125 
3126  if (INT_MAX / 8 <= buf_size)
3127  return AVERROR_INVALIDDATA;
3128 
3129  if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3130  return err;
3131 
3132  switch (ac->oc[1].m4ac.object_type) {
3133  case AOT_ER_AAC_LC:
3134  case AOT_ER_AAC_LTP:
3135  case AOT_ER_AAC_LD:
3136  case AOT_ER_AAC_ELD:
3137  err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3138  break;
3139  default:
3140  err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3141  }
3142  if (err < 0)
3143  return err;
3144 
3145  buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3146  for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3147  if (buf[buf_offset])
3148  break;
3149 
3150  return buf_size > buf_offset ? buf_consumed : buf_size;
3151 }
3152 
3154 {
3155  AACContext *ac = avctx->priv_data;
3156  int i, type;
3157 
3158  for (i = 0; i < MAX_ELEM_ID; i++) {
3159  for (type = 0; type < 4; type++) {
3160  if (ac->che[type][i])
3161  ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3162  av_freep(&ac->che[type][i]);
3163  }
3164  }
3165 
3166  ff_mdct_end(&ac->mdct);
3167  ff_mdct_end(&ac->mdct_small);
3168  ff_mdct_end(&ac->mdct_ld);
3169  ff_mdct_end(&ac->mdct_ltp);
3170  return 0;
3171 }
3172 
3173 
3174 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3175 
3176 struct LATMContext {
3177  AACContext aac_ctx; ///< containing AACContext
3178  int initialized; ///< initialized after a valid extradata was seen
3179 
3180  // parser data
3181  int audio_mux_version_A; ///< LATM syntax version
3182  int frame_length_type; ///< 0/1 variable/fixed frame length
3183  int frame_length; ///< frame length for fixed frame length
3184 };
3185 
3186 static inline uint32_t latm_get_value(GetBitContext *b)
3187 {
3188  int length = get_bits(b, 2);
3189 
3190  return get_bits_long(b, (length+1)*8);
3191 }
3192 
3194  GetBitContext *gb, int asclen)
3195 {
3196  AACContext *ac = &latmctx->aac_ctx;
3197  AVCodecContext *avctx = ac->avctx;
3198  MPEG4AudioConfig m4ac = { 0 };
3199  int config_start_bit = get_bits_count(gb);
3200  int sync_extension = 0;
3201  int bits_consumed, esize;
3202 
3203  if (asclen) {
3204  sync_extension = 1;
3205  asclen = FFMIN(asclen, get_bits_left(gb));
3206  } else
3207  asclen = get_bits_left(gb);
3208 
3209  if (config_start_bit % 8) {
3211  "Non-byte-aligned audio-specific config");
3212  return AVERROR_PATCHWELCOME;
3213  }
3214  if (asclen <= 0)
3215  return AVERROR_INVALIDDATA;
3216  bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3217  gb->buffer + (config_start_bit / 8),
3218  asclen, sync_extension);
3219 
3220  if (bits_consumed < 0)
3221  return AVERROR_INVALIDDATA;
3222 
3223  if (!latmctx->initialized ||
3224  ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3225  ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3226 
3227  if(latmctx->initialized) {
3228  av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3229  } else {
3230  av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3231  }
3232  latmctx->initialized = 0;
3233 
3234  esize = (bits_consumed+7) / 8;
3235 
3236  if (avctx->extradata_size < esize) {
3237  av_free(avctx->extradata);
3239  if (!avctx->extradata)
3240  return AVERROR(ENOMEM);
3241  }
3242 
3243  avctx->extradata_size = esize;
3244  memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3245  memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3246  }
3247  skip_bits_long(gb, bits_consumed);
3248 
3249  return bits_consumed;
3250 }
3251 
3252 static int read_stream_mux_config(struct LATMContext *latmctx,
3253  GetBitContext *gb)
3254 {
3255  int ret, audio_mux_version = get_bits(gb, 1);
3256 
3257  latmctx->audio_mux_version_A = 0;
3258  if (audio_mux_version)
3259  latmctx->audio_mux_version_A = get_bits(gb, 1);
3260 
3261  if (!latmctx->audio_mux_version_A) {
3262 
3263  if (audio_mux_version)
3264  latm_get_value(gb); // taraFullness
3265 
3266  skip_bits(gb, 1); // allStreamSameTimeFraming
3267  skip_bits(gb, 6); // numSubFrames
3268  // numPrograms
3269  if (get_bits(gb, 4)) { // numPrograms
3270  avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3271  return AVERROR_PATCHWELCOME;
3272  }
3273 
3274  // for each program (which there is only one in DVB)
3275 
3276  // for each layer (which there is only one in DVB)
3277  if (get_bits(gb, 3)) { // numLayer
3278  avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3279  return AVERROR_PATCHWELCOME;
3280  }
3281 
3282  // for all but first stream: use_same_config = get_bits(gb, 1);
3283  if (!audio_mux_version) {
3284  if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3285  return ret;
3286  } else {
3287  int ascLen = latm_get_value(gb);
3288  if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3289  return ret;
3290  ascLen -= ret;
3291  skip_bits_long(gb, ascLen);
3292  }
3293 
3294  latmctx->frame_length_type = get_bits(gb, 3);
3295  switch (latmctx->frame_length_type) {
3296  case 0:
3297  skip_bits(gb, 8); // latmBufferFullness
3298  break;
3299  case 1:
3300  latmctx->frame_length = get_bits(gb, 9);
3301  break;
3302  case 3:
3303  case 4:
3304  case 5:
3305  skip_bits(gb, 6); // CELP frame length table index
3306  break;
3307  case 6:
3308  case 7:
3309  skip_bits(gb, 1); // HVXC frame length table index
3310  break;
3311  }
3312 
3313  if (get_bits(gb, 1)) { // other data
3314  if (audio_mux_version) {
3315  latm_get_value(gb); // other_data_bits
3316  } else {
3317  int esc;
3318  do {
3319  esc = get_bits(gb, 1);
3320  skip_bits(gb, 8);
3321  } while (esc);
3322  }
3323  }
3324 
3325  if (get_bits(gb, 1)) // crc present
3326  skip_bits(gb, 8); // config_crc
3327  }
3328 
3329  return 0;
3330 }
3331 
3333 {
3334  uint8_t tmp;
3335 
3336  if (ctx->frame_length_type == 0) {
3337  int mux_slot_length = 0;
3338  do {
3339  tmp = get_bits(gb, 8);
3340  mux_slot_length += tmp;
3341  } while (tmp == 255);
3342  return mux_slot_length;
3343  } else if (ctx->frame_length_type == 1) {
3344  return ctx->frame_length;
3345  } else if (ctx->frame_length_type == 3 ||
3346  ctx->frame_length_type == 5 ||
3347  ctx->frame_length_type == 7) {
3348  skip_bits(gb, 2); // mux_slot_length_coded
3349  }
3350  return 0;
3351 }
3352 
3353 static int read_audio_mux_element(struct LATMContext *latmctx,
3354  GetBitContext *gb)
3355 {
3356  int err;
3357  uint8_t use_same_mux = get_bits(gb, 1);
3358  if (!use_same_mux) {
3359  if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3360  return err;
3361  } else if (!latmctx->aac_ctx.avctx->extradata) {
3362  av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3363  "no decoder config found\n");
3364  return AVERROR(EAGAIN);
3365  }
3366  if (latmctx->audio_mux_version_A == 0) {
3367  int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3368  if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3369  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3370  return AVERROR_INVALIDDATA;
3371  } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3372  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3373  "frame length mismatch %d << %d\n",
3374  mux_slot_length_bytes * 8, get_bits_left(gb));
3375  return AVERROR_INVALIDDATA;
3376  }
3377  }
3378  return 0;
3379 }
3380 
3381 
3382 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3383  int *got_frame_ptr, AVPacket *avpkt)
3384 {
3385  struct LATMContext *latmctx = avctx->priv_data;
3386  int muxlength, err;
3387  GetBitContext gb;
3388 
3389  if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3390  return err;
3391 
3392  // check for LOAS sync word
3393  if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3394  return AVERROR_INVALIDDATA;
3395 
3396  muxlength = get_bits(&gb, 13) + 3;
3397  // not enough data, the parser should have sorted this out
3398  if (muxlength > avpkt->size)
3399  return AVERROR_INVALIDDATA;
3400 
3401  if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3402  return err;
3403 
3404  if (!latmctx->initialized) {
3405  if (!avctx->extradata) {
3406  *got_frame_ptr = 0;
3407  return avpkt->size;
3408  } else {
3410  if ((err = decode_audio_specific_config(
3411  &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3412  avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3413  pop_output_configuration(&latmctx->aac_ctx);
3414  return err;
3415  }
3416  latmctx->initialized = 1;
3417  }
3418  }
3419 
3420  if (show_bits(&gb, 12) == 0xfff) {
3421  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3422  "ADTS header detected, probably as result of configuration "
3423  "misparsing\n");
3424  return AVERROR_INVALIDDATA;
3425  }
3426 
3427  if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
3428  return err;
3429 
3430  return muxlength;
3431 }
3432 
3434 {
3435  struct LATMContext *latmctx = avctx->priv_data;
3436  int ret = aac_decode_init(avctx);
3437 
3438  if (avctx->extradata_size > 0)
3439  latmctx->initialized = !ret;
3440 
3441  return ret;
3442 }
3443 
3444 static void aacdec_init(AACContext *c)
3445 {
3447  c->apply_ltp = apply_ltp;
3448  c->apply_tns = apply_tns;
3450  c->update_ltp = update_ltp;
3451 
3452  if(ARCH_MIPS)
3454 }
3455 /**
3456  * AVOptions for Japanese DTV specific extensions (ADTS only)
3457  */
3458 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3459 static const AVOption options[] = {
3460  {"dual_mono_mode", "Select the channel to decode for dual mono",
3461  offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3462  AACDEC_FLAGS, "dual_mono_mode"},
3463 
3464  {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3465  {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3466  {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3467  {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3468 
3469  {NULL},
3470 };
3471 
3472 static const AVClass aac_decoder_class = {
3473  .class_name = "AAC decoder",
3474  .item_name = av_default_item_name,
3475  .option = options,
3476  .version = LIBAVUTIL_VERSION_INT,
3477 };
3478 
3479 static const AVProfile profiles[] = {
3480  { FF_PROFILE_AAC_MAIN, "Main" },
3481  { FF_PROFILE_AAC_LOW, "LC" },
3482  { FF_PROFILE_AAC_SSR, "SSR" },
3483  { FF_PROFILE_AAC_LTP, "LTP" },
3484  { FF_PROFILE_AAC_HE, "HE-AAC" },
3485  { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3486  { FF_PROFILE_AAC_LD, "LD" },
3487  { FF_PROFILE_AAC_ELD, "ELD" },
3488  { FF_PROFILE_UNKNOWN },
3489 };
3490 
3492  .name = "aac",
3493  .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3494  .type = AVMEDIA_TYPE_AUDIO,
3495  .id = AV_CODEC_ID_AAC,
3496  .priv_data_size = sizeof(AACContext),
3497  .init = aac_decode_init,
3500  .sample_fmts = (const enum AVSampleFormat[]) {
3502  },
3503  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3504  .channel_layouts = aac_channel_layout,
3505  .flush = flush,
3506  .priv_class = &aac_decoder_class,
3507  .profiles = profiles,
3508 };
3509 
3510 /*
3511  Note: This decoder filter is intended to decode LATM streams transferred
3512  in MPEG transport streams which only contain one program.
3513  To do a more complex LATM demuxing a separate LATM demuxer should be used.
3514 */
3516  .name = "aac_latm",
3517  .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3518  .type = AVMEDIA_TYPE_AUDIO,
3519  .id = AV_CODEC_ID_AAC_LATM,
3520  .priv_data_size = sizeof(struct LATMContext),
3521  .init = latm_decode_init,
3522  .close = aac_decode_close,
3523  .decode = latm_decode_frame,
3524  .sample_fmts = (const enum AVSampleFormat[]) {
3526  },
3527  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3528  .channel_layouts = aac_channel_layout,
3529  .flush = flush,
3530  .profiles = profiles,
3531 };