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  if (!ac->fdsp) {
1142  return AVERROR(ENOMEM);
1143  }
1144 
1145  ac->random_state = 0x1f2e3d4c;
1146 
1147  ff_aac_tableinit();
1148 
1149  INIT_VLC_STATIC(&vlc_scalefactors, 7,
1152  sizeof(ff_aac_scalefactor_bits[0]),
1153  sizeof(ff_aac_scalefactor_bits[0]),
1155  sizeof(ff_aac_scalefactor_code[0]),
1156  sizeof(ff_aac_scalefactor_code[0]),
1157  352);
1158 
1159  ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1160  ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1161  ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1162  ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1163  // window initialization
1169 
1170  cbrt_tableinit();
1171 
1172  return 0;
1173 }
1174 
1175 /**
1176  * Skip data_stream_element; reference: table 4.10.
1177  */
1179 {
1180  int byte_align = get_bits1(gb);
1181  int count = get_bits(gb, 8);
1182  if (count == 255)
1183  count += get_bits(gb, 8);
1184  if (byte_align)
1185  align_get_bits(gb);
1186 
1187  if (get_bits_left(gb) < 8 * count) {
1188  av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1189  return AVERROR_INVALIDDATA;
1190  }
1191  skip_bits_long(gb, 8 * count);
1192  return 0;
1193 }
1194 
1196  GetBitContext *gb)
1197 {
1198  int sfb;
1199  if (get_bits1(gb)) {
1200  ics->predictor_reset_group = get_bits(gb, 5);
1201  if (ics->predictor_reset_group == 0 ||
1202  ics->predictor_reset_group > 30) {
1203  av_log(ac->avctx, AV_LOG_ERROR,
1204  "Invalid Predictor Reset Group.\n");
1205  return AVERROR_INVALIDDATA;
1206  }
1207  }
1208  for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1209  ics->prediction_used[sfb] = get_bits1(gb);
1210  }
1211  return 0;
1212 }
1213 
1214 /**
1215  * Decode Long Term Prediction data; reference: table 4.xx.
1216  */
1218  GetBitContext *gb, uint8_t max_sfb)
1219 {
1220  int sfb;
1221 
1222  ltp->lag = get_bits(gb, 11);
1223  ltp->coef = ltp_coef[get_bits(gb, 3)];
1224  for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1225  ltp->used[sfb] = get_bits1(gb);
1226 }
1227 
1228 /**
1229  * Decode Individual Channel Stream info; reference: table 4.6.
1230  */
1232  GetBitContext *gb)
1233 {
1234  int aot = ac->oc[1].m4ac.object_type;
1235  if (aot != AOT_ER_AAC_ELD) {
1236  if (get_bits1(gb)) {
1237  av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1238  return AVERROR_INVALIDDATA;
1239  }
1240  ics->window_sequence[1] = ics->window_sequence[0];
1241  ics->window_sequence[0] = get_bits(gb, 2);
1242  if (aot == AOT_ER_AAC_LD &&
1243  ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1244  av_log(ac->avctx, AV_LOG_ERROR,
1245  "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1246  "window sequence %d found.\n", ics->window_sequence[0]);
1248  return AVERROR_INVALIDDATA;
1249  }
1250  ics->use_kb_window[1] = ics->use_kb_window[0];
1251  ics->use_kb_window[0] = get_bits1(gb);
1252  }
1253  ics->num_window_groups = 1;
1254  ics->group_len[0] = 1;
1255  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1256  int i;
1257  ics->max_sfb = get_bits(gb, 4);
1258  for (i = 0; i < 7; i++) {
1259  if (get_bits1(gb)) {
1260  ics->group_len[ics->num_window_groups - 1]++;
1261  } else {
1262  ics->num_window_groups++;
1263  ics->group_len[ics->num_window_groups - 1] = 1;
1264  }
1265  }
1266  ics->num_windows = 8;
1270  ics->predictor_present = 0;
1271  } else {
1272  ics->max_sfb = get_bits(gb, 6);
1273  ics->num_windows = 1;
1274  if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1278  if (!ics->num_swb || !ics->swb_offset)
1279  return AVERROR_BUG;
1280  } else {
1284  }
1285  if (aot != AOT_ER_AAC_ELD) {
1286  ics->predictor_present = get_bits1(gb);
1287  ics->predictor_reset_group = 0;
1288  }
1289  if (ics->predictor_present) {
1290  if (aot == AOT_AAC_MAIN) {
1291  if (decode_prediction(ac, ics, gb)) {
1292  goto fail;
1293  }
1294  } else if (aot == AOT_AAC_LC ||
1295  aot == AOT_ER_AAC_LC) {
1296  av_log(ac->avctx, AV_LOG_ERROR,
1297  "Prediction is not allowed in AAC-LC.\n");
1298  goto fail;
1299  } else {
1300  if (aot == AOT_ER_AAC_LD) {
1301  av_log(ac->avctx, AV_LOG_ERROR,
1302  "LTP in ER AAC LD not yet implemented.\n");
1303  return AVERROR_PATCHWELCOME;
1304  }
1305  if ((ics->ltp.present = get_bits(gb, 1)))
1306  decode_ltp(&ics->ltp, gb, ics->max_sfb);
1307  }
1308  }
1309  }
1310 
1311  if (ics->max_sfb > ics->num_swb) {
1312  av_log(ac->avctx, AV_LOG_ERROR,
1313  "Number of scalefactor bands in group (%d) "
1314  "exceeds limit (%d).\n",
1315  ics->max_sfb, ics->num_swb);
1316  goto fail;
1317  }
1318 
1319  return 0;
1320 fail:
1321  ics->max_sfb = 0;
1322  return AVERROR_INVALIDDATA;
1323 }
1324 
1325 /**
1326  * Decode band types (section_data payload); reference: table 4.46.
1327  *
1328  * @param band_type array of the used band type
1329  * @param band_type_run_end array of the last scalefactor band of a band type run
1330  *
1331  * @return Returns error status. 0 - OK, !0 - error
1332  */
1333 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1334  int band_type_run_end[120], GetBitContext *gb,
1336 {
1337  int g, idx = 0;
1338  const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1339  for (g = 0; g < ics->num_window_groups; g++) {
1340  int k = 0;
1341  while (k < ics->max_sfb) {
1342  uint8_t sect_end = k;
1343  int sect_len_incr;
1344  int sect_band_type = get_bits(gb, 4);
1345  if (sect_band_type == 12) {
1346  av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1347  return AVERROR_INVALIDDATA;
1348  }
1349  do {
1350  sect_len_incr = get_bits(gb, bits);
1351  sect_end += sect_len_incr;
1352  if (get_bits_left(gb) < 0) {
1353  av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1354  return AVERROR_INVALIDDATA;
1355  }
1356  if (sect_end > ics->max_sfb) {
1357  av_log(ac->avctx, AV_LOG_ERROR,
1358  "Number of bands (%d) exceeds limit (%d).\n",
1359  sect_end, ics->max_sfb);
1360  return AVERROR_INVALIDDATA;
1361  }
1362  } while (sect_len_incr == (1 << bits) - 1);
1363  for (; k < sect_end; k++) {
1364  band_type [idx] = sect_band_type;
1365  band_type_run_end[idx++] = sect_end;
1366  }
1367  }
1368  }
1369  return 0;
1370 }
1371 
1372 /**
1373  * Decode scalefactors; reference: table 4.47.
1374  *
1375  * @param global_gain first scalefactor value as scalefactors are differentially coded
1376  * @param band_type array of the used band type
1377  * @param band_type_run_end array of the last scalefactor band of a band type run
1378  * @param sf array of scalefactors or intensity stereo positions
1379  *
1380  * @return Returns error status. 0 - OK, !0 - error
1381  */
1382 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1383  unsigned int global_gain,
1385  enum BandType band_type[120],
1386  int band_type_run_end[120])
1387 {
1388  int g, i, idx = 0;
1389  int offset[3] = { global_gain, global_gain - 90, 0 };
1390  int clipped_offset;
1391  int noise_flag = 1;
1392  for (g = 0; g < ics->num_window_groups; g++) {
1393  for (i = 0; i < ics->max_sfb;) {
1394  int run_end = band_type_run_end[idx];
1395  if (band_type[idx] == ZERO_BT) {
1396  for (; i < run_end; i++, idx++)
1397  sf[idx] = 0.0;
1398  } else if ((band_type[idx] == INTENSITY_BT) ||
1399  (band_type[idx] == INTENSITY_BT2)) {
1400  for (; i < run_end; i++, idx++) {
1401  offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1402  clipped_offset = av_clip(offset[2], -155, 100);
1403  if (offset[2] != clipped_offset) {
1405  "If you heard an audible artifact, there may be a bug in the decoder. "
1406  "Clipped intensity stereo position (%d -> %d)",
1407  offset[2], clipped_offset);
1408  }
1409  sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1410  }
1411  } else if (band_type[idx] == NOISE_BT) {
1412  for (; i < run_end; i++, idx++) {
1413  if (noise_flag-- > 0)
1414  offset[1] += get_bits(gb, 9) - 256;
1415  else
1416  offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1417  clipped_offset = av_clip(offset[1], -100, 155);
1418  if (offset[1] != clipped_offset) {
1420  "If you heard an audible artifact, there may be a bug in the decoder. "
1421  "Clipped noise gain (%d -> %d)",
1422  offset[1], clipped_offset);
1423  }
1424  sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1425  }
1426  } else {
1427  for (; i < run_end; i++, idx++) {
1428  offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1429  if (offset[0] > 255U) {
1430  av_log(ac->avctx, AV_LOG_ERROR,
1431  "Scalefactor (%d) out of range.\n", offset[0]);
1432  return AVERROR_INVALIDDATA;
1433  }
1434  sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1435  }
1436  }
1437  }
1438  }
1439  return 0;
1440 }
1441 
1442 /**
1443  * Decode pulse data; reference: table 4.7.
1444  */
1445 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1446  const uint16_t *swb_offset, int num_swb)
1447 {
1448  int i, pulse_swb;
1449  pulse->num_pulse = get_bits(gb, 2) + 1;
1450  pulse_swb = get_bits(gb, 6);
1451  if (pulse_swb >= num_swb)
1452  return -1;
1453  pulse->pos[0] = swb_offset[pulse_swb];
1454  pulse->pos[0] += get_bits(gb, 5);
1455  if (pulse->pos[0] >= swb_offset[num_swb])
1456  return -1;
1457  pulse->amp[0] = get_bits(gb, 4);
1458  for (i = 1; i < pulse->num_pulse; i++) {
1459  pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1460  if (pulse->pos[i] >= swb_offset[num_swb])
1461  return -1;
1462  pulse->amp[i] = get_bits(gb, 4);
1463  }
1464  return 0;
1465 }
1466 
1467 /**
1468  * Decode Temporal Noise Shaping data; reference: table 4.48.
1469  *
1470  * @return Returns error status. 0 - OK, !0 - error
1471  */
1473  GetBitContext *gb, const IndividualChannelStream *ics)
1474 {
1475  int w, filt, i, coef_len, coef_res, coef_compress;
1476  const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1477  const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1478  for (w = 0; w < ics->num_windows; w++) {
1479  if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1480  coef_res = get_bits1(gb);
1481 
1482  for (filt = 0; filt < tns->n_filt[w]; filt++) {
1483  int tmp2_idx;
1484  tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1485 
1486  if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1487  av_log(ac->avctx, AV_LOG_ERROR,
1488  "TNS filter order %d is greater than maximum %d.\n",
1489  tns->order[w][filt], tns_max_order);
1490  tns->order[w][filt] = 0;
1491  return AVERROR_INVALIDDATA;
1492  }
1493  if (tns->order[w][filt]) {
1494  tns->direction[w][filt] = get_bits1(gb);
1495  coef_compress = get_bits1(gb);
1496  coef_len = coef_res + 3 - coef_compress;
1497  tmp2_idx = 2 * coef_compress + coef_res;
1498 
1499  for (i = 0; i < tns->order[w][filt]; i++)
1500  tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1501  }
1502  }
1503  }
1504  }
1505  return 0;
1506 }
1507 
1508 /**
1509  * Decode Mid/Side data; reference: table 4.54.
1510  *
1511  * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1512  * [1] mask is decoded from bitstream; [2] mask is all 1s;
1513  * [3] reserved for scalable AAC
1514  */
1516  int ms_present)
1517 {
1518  int idx;
1519  int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1520  if (ms_present == 1) {
1521  for (idx = 0; idx < max_idx; idx++)
1522  cpe->ms_mask[idx] = get_bits1(gb);
1523  } else if (ms_present == 2) {
1524  memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1525  }
1526 }
1527 
1528 #ifndef VMUL2
1529 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1530  const float *scale)
1531 {
1532  float s = *scale;
1533  *dst++ = v[idx & 15] * s;
1534  *dst++ = v[idx>>4 & 15] * s;
1535  return dst;
1536 }
1537 #endif
1538 
1539 #ifndef VMUL4
1540 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1541  const float *scale)
1542 {
1543  float s = *scale;
1544  *dst++ = v[idx & 3] * s;
1545  *dst++ = v[idx>>2 & 3] * s;
1546  *dst++ = v[idx>>4 & 3] * s;
1547  *dst++ = v[idx>>6 & 3] * s;
1548  return dst;
1549 }
1550 #endif
1551 
1552 #ifndef VMUL2S
1553 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1554  unsigned sign, const float *scale)
1555 {
1556  union av_intfloat32 s0, s1;
1557 
1558  s0.f = s1.f = *scale;
1559  s0.i ^= sign >> 1 << 31;
1560  s1.i ^= sign << 31;
1561 
1562  *dst++ = v[idx & 15] * s0.f;
1563  *dst++ = v[idx>>4 & 15] * s1.f;
1564 
1565  return dst;
1566 }
1567 #endif
1568 
1569 #ifndef VMUL4S
1570 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1571  unsigned sign, const float *scale)
1572 {
1573  unsigned nz = idx >> 12;
1574  union av_intfloat32 s = { .f = *scale };
1575  union av_intfloat32 t;
1576 
1577  t.i = s.i ^ (sign & 1U<<31);
1578  *dst++ = v[idx & 3] * t.f;
1579 
1580  sign <<= nz & 1; nz >>= 1;
1581  t.i = s.i ^ (sign & 1U<<31);
1582  *dst++ = v[idx>>2 & 3] * t.f;
1583 
1584  sign <<= nz & 1; nz >>= 1;
1585  t.i = s.i ^ (sign & 1U<<31);
1586  *dst++ = v[idx>>4 & 3] * t.f;
1587 
1588  sign <<= nz & 1;
1589  t.i = s.i ^ (sign & 1U<<31);
1590  *dst++ = v[idx>>6 & 3] * t.f;
1591 
1592  return dst;
1593 }
1594 #endif
1595 
1596 /**
1597  * Decode spectral data; reference: table 4.50.
1598  * Dequantize and scale spectral data; reference: 4.6.3.3.
1599  *
1600  * @param coef array of dequantized, scaled spectral data
1601  * @param sf array of scalefactors or intensity stereo positions
1602  * @param pulse_present set if pulses are present
1603  * @param pulse pointer to pulse data struct
1604  * @param band_type array of the used band type
1605  *
1606  * @return Returns error status. 0 - OK, !0 - error
1607  */
1608 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1609  GetBitContext *gb, const float sf[120],
1610  int pulse_present, const Pulse *pulse,
1611  const IndividualChannelStream *ics,
1612  enum BandType band_type[120])
1613 {
1614  int i, k, g, idx = 0;
1615  const int c = 1024 / ics->num_windows;
1616  const uint16_t *offsets = ics->swb_offset;
1617  float *coef_base = coef;
1618 
1619  for (g = 0; g < ics->num_windows; g++)
1620  memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1621  sizeof(float) * (c - offsets[ics->max_sfb]));
1622 
1623  for (g = 0; g < ics->num_window_groups; g++) {
1624  unsigned g_len = ics->group_len[g];
1625 
1626  for (i = 0; i < ics->max_sfb; i++, idx++) {
1627  const unsigned cbt_m1 = band_type[idx] - 1;
1628  float *cfo = coef + offsets[i];
1629  int off_len = offsets[i + 1] - offsets[i];
1630  int group;
1631 
1632  if (cbt_m1 >= INTENSITY_BT2 - 1) {
1633  for (group = 0; group < g_len; group++, cfo+=128) {
1634  memset(cfo, 0, off_len * sizeof(float));
1635  }
1636  } else if (cbt_m1 == NOISE_BT - 1) {
1637  for (group = 0; group < g_len; group++, cfo+=128) {
1638  float scale;
1639  float band_energy;
1640 
1641  for (k = 0; k < off_len; k++) {
1643  cfo[k] = ac->random_state;
1644  }
1645 
1646  band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1647  scale = sf[idx] / sqrtf(band_energy);
1648  ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1649  }
1650  } else {
1651  const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1652  const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1653  VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1654  OPEN_READER(re, gb);
1655 
1656  switch (cbt_m1 >> 1) {
1657  case 0:
1658  for (group = 0; group < g_len; group++, cfo+=128) {
1659  float *cf = cfo;
1660  int len = off_len;
1661 
1662  do {
1663  int code;
1664  unsigned cb_idx;
1665 
1666  UPDATE_CACHE(re, gb);
1667  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1668  cb_idx = cb_vector_idx[code];
1669  cf = VMUL4(cf, vq, cb_idx, sf + idx);
1670  } while (len -= 4);
1671  }
1672  break;
1673 
1674  case 1:
1675  for (group = 0; group < g_len; group++, cfo+=128) {
1676  float *cf = cfo;
1677  int len = off_len;
1678 
1679  do {
1680  int code;
1681  unsigned nnz;
1682  unsigned cb_idx;
1683  uint32_t bits;
1684 
1685  UPDATE_CACHE(re, gb);
1686  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1687  cb_idx = cb_vector_idx[code];
1688  nnz = cb_idx >> 8 & 15;
1689  bits = nnz ? GET_CACHE(re, gb) : 0;
1690  LAST_SKIP_BITS(re, gb, nnz);
1691  cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1692  } while (len -= 4);
1693  }
1694  break;
1695 
1696  case 2:
1697  for (group = 0; group < g_len; group++, cfo+=128) {
1698  float *cf = cfo;
1699  int len = off_len;
1700 
1701  do {
1702  int code;
1703  unsigned cb_idx;
1704 
1705  UPDATE_CACHE(re, gb);
1706  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1707  cb_idx = cb_vector_idx[code];
1708  cf = VMUL2(cf, vq, cb_idx, sf + idx);
1709  } while (len -= 2);
1710  }
1711  break;
1712 
1713  case 3:
1714  case 4:
1715  for (group = 0; group < g_len; group++, cfo+=128) {
1716  float *cf = cfo;
1717  int len = off_len;
1718 
1719  do {
1720  int code;
1721  unsigned nnz;
1722  unsigned cb_idx;
1723  unsigned sign;
1724 
1725  UPDATE_CACHE(re, gb);
1726  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1727  cb_idx = cb_vector_idx[code];
1728  nnz = cb_idx >> 8 & 15;
1729  sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1730  LAST_SKIP_BITS(re, gb, nnz);
1731  cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1732  } while (len -= 2);
1733  }
1734  break;
1735 
1736  default:
1737  for (group = 0; group < g_len; group++, cfo+=128) {
1738  float *cf = cfo;
1739  uint32_t *icf = (uint32_t *) cf;
1740  int len = off_len;
1741 
1742  do {
1743  int code;
1744  unsigned nzt, nnz;
1745  unsigned cb_idx;
1746  uint32_t bits;
1747  int j;
1748 
1749  UPDATE_CACHE(re, gb);
1750  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1751 
1752  if (!code) {
1753  *icf++ = 0;
1754  *icf++ = 0;
1755  continue;
1756  }
1757 
1758  cb_idx = cb_vector_idx[code];
1759  nnz = cb_idx >> 12;
1760  nzt = cb_idx >> 8;
1761  bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1762  LAST_SKIP_BITS(re, gb, nnz);
1763 
1764  for (j = 0; j < 2; j++) {
1765  if (nzt & 1<<j) {
1766  uint32_t b;
1767  int n;
1768  /* The total length of escape_sequence must be < 22 bits according
1769  to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1770  UPDATE_CACHE(re, gb);
1771  b = GET_CACHE(re, gb);
1772  b = 31 - av_log2(~b);
1773 
1774  if (b > 8) {
1775  av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1776  return AVERROR_INVALIDDATA;
1777  }
1778 
1779  SKIP_BITS(re, gb, b + 1);
1780  b += 4;
1781  n = (1 << b) + SHOW_UBITS(re, gb, b);
1782  LAST_SKIP_BITS(re, gb, b);
1783  *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1784  bits <<= 1;
1785  } else {
1786  unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1787  *icf++ = (bits & 1U<<31) | v;
1788  bits <<= !!v;
1789  }
1790  cb_idx >>= 4;
1791  }
1792  } while (len -= 2);
1793 
1794  ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1795  }
1796  }
1797 
1798  CLOSE_READER(re, gb);
1799  }
1800  }
1801  coef += g_len << 7;
1802  }
1803 
1804  if (pulse_present) {
1805  idx = 0;
1806  for (i = 0; i < pulse->num_pulse; i++) {
1807  float co = coef_base[ pulse->pos[i] ];
1808  while (offsets[idx + 1] <= pulse->pos[i])
1809  idx++;
1810  if (band_type[idx] != NOISE_BT && sf[idx]) {
1811  float ico = -pulse->amp[i];
1812  if (co) {
1813  co /= sf[idx];
1814  ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1815  }
1816  coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1817  }
1818  }
1819  }
1820  return 0;
1821 }
1822 
1823 static av_always_inline float flt16_round(float pf)
1824 {
1825  union av_intfloat32 tmp;
1826  tmp.f = pf;
1827  tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1828  return tmp.f;
1829 }
1830 
1831 static av_always_inline float flt16_even(float pf)
1832 {
1833  union av_intfloat32 tmp;
1834  tmp.f = pf;
1835  tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1836  return tmp.f;
1837 }
1838 
1839 static av_always_inline float flt16_trunc(float pf)
1840 {
1841  union av_intfloat32 pun;
1842  pun.f = pf;
1843  pun.i &= 0xFFFF0000U;
1844  return pun.f;
1845 }
1846 
1847 static av_always_inline void predict(PredictorState *ps, float *coef,
1848  int output_enable)
1849 {
1850  const float a = 0.953125; // 61.0 / 64
1851  const float alpha = 0.90625; // 29.0 / 32
1852  float e0, e1;
1853  float pv;
1854  float k1, k2;
1855  float r0 = ps->r0, r1 = ps->r1;
1856  float cor0 = ps->cor0, cor1 = ps->cor1;
1857  float var0 = ps->var0, var1 = ps->var1;
1858 
1859  k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1860  k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1861 
1862  pv = flt16_round(k1 * r0 + k2 * r1);
1863  if (output_enable)
1864  *coef += pv;
1865 
1866  e0 = *coef;
1867  e1 = e0 - k1 * r0;
1868 
1869  ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1870  ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1871  ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1872  ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1873 
1874  ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1875  ps->r0 = flt16_trunc(a * e0);
1876 }
1877 
1878 /**
1879  * Apply AAC-Main style frequency domain prediction.
1880  */
1882 {
1883  int sfb, k;
1884 
1885  if (!sce->ics.predictor_initialized) {
1887  sce->ics.predictor_initialized = 1;
1888  }
1889 
1890  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1891  for (sfb = 0;
1892  sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1893  sfb++) {
1894  for (k = sce->ics.swb_offset[sfb];
1895  k < sce->ics.swb_offset[sfb + 1];
1896  k++) {
1897  predict(&sce->predictor_state[k], &sce->coeffs[k],
1898  sce->ics.predictor_present &&
1899  sce->ics.prediction_used[sfb]);
1900  }
1901  }
1902  if (sce->ics.predictor_reset_group)
1904  sce->ics.predictor_reset_group);
1905  } else
1907 }
1908 
1909 /**
1910  * Decode an individual_channel_stream payload; reference: table 4.44.
1911  *
1912  * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1913  * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1914  *
1915  * @return Returns error status. 0 - OK, !0 - error
1916  */
1918  GetBitContext *gb, int common_window, int scale_flag)
1919 {
1920  Pulse pulse;
1921  TemporalNoiseShaping *tns = &sce->tns;
1922  IndividualChannelStream *ics = &sce->ics;
1923  float *out = sce->coeffs;
1924  int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1925  int ret;
1926 
1927  eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1928  er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1929  ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1930  ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1931  ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1932 
1933  /* This assignment is to silence a GCC warning about the variable being used
1934  * uninitialized when in fact it always is.
1935  */
1936  pulse.num_pulse = 0;
1937 
1938  global_gain = get_bits(gb, 8);
1939 
1940  if (!common_window && !scale_flag) {
1941  if (decode_ics_info(ac, ics, gb) < 0)
1942  return AVERROR_INVALIDDATA;
1943  }
1944 
1945  if ((ret = decode_band_types(ac, sce->band_type,
1946  sce->band_type_run_end, gb, ics)) < 0)
1947  return ret;
1948  if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1949  sce->band_type, sce->band_type_run_end)) < 0)
1950  return ret;
1951 
1952  pulse_present = 0;
1953  if (!scale_flag) {
1954  if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1955  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1956  av_log(ac->avctx, AV_LOG_ERROR,
1957  "Pulse tool not allowed in eight short sequence.\n");
1958  return AVERROR_INVALIDDATA;
1959  }
1960  if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1961  av_log(ac->avctx, AV_LOG_ERROR,
1962  "Pulse data corrupt or invalid.\n");
1963  return AVERROR_INVALIDDATA;
1964  }
1965  }
1966  tns->present = get_bits1(gb);
1967  if (tns->present && !er_syntax)
1968  if (decode_tns(ac, tns, gb, ics) < 0)
1969  return AVERROR_INVALIDDATA;
1970  if (!eld_syntax && get_bits1(gb)) {
1971  avpriv_request_sample(ac->avctx, "SSR");
1972  return AVERROR_PATCHWELCOME;
1973  }
1974  // I see no textual basis in the spec for this occurring after SSR gain
1975  // control, but this is what both reference and real implmentations do
1976  if (tns->present && er_syntax)
1977  if (decode_tns(ac, tns, gb, ics) < 0)
1978  return AVERROR_INVALIDDATA;
1979  }
1980 
1981  if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1982  &pulse, ics, sce->band_type) < 0)
1983  return AVERROR_INVALIDDATA;
1984 
1985  if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1986  apply_prediction(ac, sce);
1987 
1988  return 0;
1989 }
1990 
1991 /**
1992  * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1993  */
1995 {
1996  const IndividualChannelStream *ics = &cpe->ch[0].ics;
1997  float *ch0 = cpe->ch[0].coeffs;
1998  float *ch1 = cpe->ch[1].coeffs;
1999  int g, i, group, idx = 0;
2000  const uint16_t *offsets = ics->swb_offset;
2001  for (g = 0; g < ics->num_window_groups; g++) {
2002  for (i = 0; i < ics->max_sfb; i++, idx++) {
2003  if (cpe->ms_mask[idx] &&
2004  cpe->ch[0].band_type[idx] < NOISE_BT &&
2005  cpe->ch[1].band_type[idx] < NOISE_BT) {
2006  for (group = 0; group < ics->group_len[g]; group++) {
2007  ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2008  ch1 + group * 128 + offsets[i],
2009  offsets[i+1] - offsets[i]);
2010  }
2011  }
2012  }
2013  ch0 += ics->group_len[g] * 128;
2014  ch1 += ics->group_len[g] * 128;
2015  }
2016 }
2017 
2018 /**
2019  * intensity stereo decoding; reference: 4.6.8.2.3
2020  *
2021  * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2022  * [1] mask is decoded from bitstream; [2] mask is all 1s;
2023  * [3] reserved for scalable AAC
2024  */
2026  ChannelElement *cpe, int ms_present)
2027 {
2028  const IndividualChannelStream *ics = &cpe->ch[1].ics;
2029  SingleChannelElement *sce1 = &cpe->ch[1];
2030  float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2031  const uint16_t *offsets = ics->swb_offset;
2032  int g, group, i, idx = 0;
2033  int c;
2034  float scale;
2035  for (g = 0; g < ics->num_window_groups; g++) {
2036  for (i = 0; i < ics->max_sfb;) {
2037  if (sce1->band_type[idx] == INTENSITY_BT ||
2038  sce1->band_type[idx] == INTENSITY_BT2) {
2039  const int bt_run_end = sce1->band_type_run_end[idx];
2040  for (; i < bt_run_end; i++, idx++) {
2041  c = -1 + 2 * (sce1->band_type[idx] - 14);
2042  if (ms_present)
2043  c *= 1 - 2 * cpe->ms_mask[idx];
2044  scale = c * sce1->sf[idx];
2045  for (group = 0; group < ics->group_len[g]; group++)
2046  ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2047  coef0 + group * 128 + offsets[i],
2048  scale,
2049  offsets[i + 1] - offsets[i]);
2050  }
2051  } else {
2052  int bt_run_end = sce1->band_type_run_end[idx];
2053  idx += bt_run_end - i;
2054  i = bt_run_end;
2055  }
2056  }
2057  coef0 += ics->group_len[g] * 128;
2058  coef1 += ics->group_len[g] * 128;
2059  }
2060 }
2061 
2062 /**
2063  * Decode a channel_pair_element; reference: table 4.4.
2064  *
2065  * @return Returns error status. 0 - OK, !0 - error
2066  */
2068 {
2069  int i, ret, common_window, ms_present = 0;
2070  int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2071 
2072  common_window = eld_syntax || get_bits1(gb);
2073  if (common_window) {
2074  if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2075  return AVERROR_INVALIDDATA;
2076  i = cpe->ch[1].ics.use_kb_window[0];
2077  cpe->ch[1].ics = cpe->ch[0].ics;
2078  cpe->ch[1].ics.use_kb_window[1] = i;
2079  if (cpe->ch[1].ics.predictor_present &&
2080  (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2081  if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2082  decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2083  ms_present = get_bits(gb, 2);
2084  if (ms_present == 3) {
2085  av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2086  return AVERROR_INVALIDDATA;
2087  } else if (ms_present)
2088  decode_mid_side_stereo(cpe, gb, ms_present);
2089  }
2090  if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2091  return ret;
2092  if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2093  return ret;
2094 
2095  if (common_window) {
2096  if (ms_present)
2097  apply_mid_side_stereo(ac, cpe);
2098  if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2099  apply_prediction(ac, &cpe->ch[0]);
2100  apply_prediction(ac, &cpe->ch[1]);
2101  }
2102  }
2103 
2104  apply_intensity_stereo(ac, cpe, ms_present);
2105  return 0;
2106 }
2107 
2108 static const float cce_scale[] = {
2109  1.09050773266525765921, //2^(1/8)
2110  1.18920711500272106672, //2^(1/4)
2111  M_SQRT2,
2112  2,
2113 };
2114 
2115 /**
2116  * Decode coupling_channel_element; reference: table 4.8.
2117  *
2118  * @return Returns error status. 0 - OK, !0 - error
2119  */
2121 {
2122  int num_gain = 0;
2123  int c, g, sfb, ret;
2124  int sign;
2125  float scale;
2126  SingleChannelElement *sce = &che->ch[0];
2127  ChannelCoupling *coup = &che->coup;
2128 
2129  coup->coupling_point = 2 * get_bits1(gb);
2130  coup->num_coupled = get_bits(gb, 3);
2131  for (c = 0; c <= coup->num_coupled; c++) {
2132  num_gain++;
2133  coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2134  coup->id_select[c] = get_bits(gb, 4);
2135  if (coup->type[c] == TYPE_CPE) {
2136  coup->ch_select[c] = get_bits(gb, 2);
2137  if (coup->ch_select[c] == 3)
2138  num_gain++;
2139  } else
2140  coup->ch_select[c] = 2;
2141  }
2142  coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2143 
2144  sign = get_bits(gb, 1);
2145  scale = cce_scale[get_bits(gb, 2)];
2146 
2147  if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2148  return ret;
2149 
2150  for (c = 0; c < num_gain; c++) {
2151  int idx = 0;
2152  int cge = 1;
2153  int gain = 0;
2154  float gain_cache = 1.0;
2155  if (c) {
2156  cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2157  gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2158  gain_cache = powf(scale, -gain);
2159  }
2160  if (coup->coupling_point == AFTER_IMDCT) {
2161  coup->gain[c][0] = gain_cache;
2162  } else {
2163  for (g = 0; g < sce->ics.num_window_groups; g++) {
2164  for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2165  if (sce->band_type[idx] != ZERO_BT) {
2166  if (!cge) {
2167  int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2168  if (t) {
2169  int s = 1;
2170  t = gain += t;
2171  if (sign) {
2172  s -= 2 * (t & 0x1);
2173  t >>= 1;
2174  }
2175  gain_cache = powf(scale, -t) * s;
2176  }
2177  }
2178  coup->gain[c][idx] = gain_cache;
2179  }
2180  }
2181  }
2182  }
2183  }
2184  return 0;
2185 }
2186 
2187 /**
2188  * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2189  *
2190  * @return Returns number of bytes consumed.
2191  */
2193  GetBitContext *gb)
2194 {
2195  int i;
2196  int num_excl_chan = 0;
2197 
2198  do {
2199  for (i = 0; i < 7; i++)
2200  che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2201  } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2202 
2203  return num_excl_chan / 7;
2204 }
2205 
2206 /**
2207  * Decode dynamic range information; reference: table 4.52.
2208  *
2209  * @return Returns number of bytes consumed.
2210  */
2212  GetBitContext *gb)
2213 {
2214  int n = 1;
2215  int drc_num_bands = 1;
2216  int i;
2217 
2218  /* pce_tag_present? */
2219  if (get_bits1(gb)) {
2220  che_drc->pce_instance_tag = get_bits(gb, 4);
2221  skip_bits(gb, 4); // tag_reserved_bits
2222  n++;
2223  }
2224 
2225  /* excluded_chns_present? */
2226  if (get_bits1(gb)) {
2227  n += decode_drc_channel_exclusions(che_drc, gb);
2228  }
2229 
2230  /* drc_bands_present? */
2231  if (get_bits1(gb)) {
2232  che_drc->band_incr = get_bits(gb, 4);
2233  che_drc->interpolation_scheme = get_bits(gb, 4);
2234  n++;
2235  drc_num_bands += che_drc->band_incr;
2236  for (i = 0; i < drc_num_bands; i++) {
2237  che_drc->band_top[i] = get_bits(gb, 8);
2238  n++;
2239  }
2240  }
2241 
2242  /* prog_ref_level_present? */
2243  if (get_bits1(gb)) {
2244  che_drc->prog_ref_level = get_bits(gb, 7);
2245  skip_bits1(gb); // prog_ref_level_reserved_bits
2246  n++;
2247  }
2248 
2249  for (i = 0; i < drc_num_bands; i++) {
2250  che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2251  che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2252  n++;
2253  }
2254 
2255  return n;
2256 }
2257 
2258 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2259  uint8_t buf[256];
2260  int i, major, minor;
2261 
2262  if (len < 13+7*8)
2263  goto unknown;
2264 
2265  get_bits(gb, 13); len -= 13;
2266 
2267  for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2268  buf[i] = get_bits(gb, 8);
2269 
2270  buf[i] = 0;
2271  if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2272  av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2273 
2274  if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2275  ac->avctx->internal->skip_samples = 1024;
2276  }
2277 
2278 unknown:
2279  skip_bits_long(gb, len);
2280 
2281  return 0;
2282 }
2283 
2284 /**
2285  * Decode extension data (incomplete); reference: table 4.51.
2286  *
2287  * @param cnt length of TYPE_FIL syntactic element in bytes
2288  *
2289  * @return Returns number of bytes consumed
2290  */
2292  ChannelElement *che, enum RawDataBlockType elem_type)
2293 {
2294  int crc_flag = 0;
2295  int res = cnt;
2296  int type = get_bits(gb, 4);
2297 
2298  if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2299  av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2300 
2301  switch (type) { // extension type
2302  case EXT_SBR_DATA_CRC:
2303  crc_flag++;
2304  case EXT_SBR_DATA:
2305  if (!che) {
2306  av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2307  return res;
2308  } else if (!ac->oc[1].m4ac.sbr) {
2309  av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2310  skip_bits_long(gb, 8 * cnt - 4);
2311  return res;
2312  } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2313  av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2314  skip_bits_long(gb, 8 * cnt - 4);
2315  return res;
2316  } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2317  ac->oc[1].m4ac.sbr = 1;
2318  ac->oc[1].m4ac.ps = 1;
2320  output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2321  ac->oc[1].status, 1);
2322  } else {
2323  ac->oc[1].m4ac.sbr = 1;
2325  }
2326  res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2327  break;
2328  case EXT_DYNAMIC_RANGE:
2329  res = decode_dynamic_range(&ac->che_drc, gb);
2330  break;
2331  case EXT_FILL:
2332  decode_fill(ac, gb, 8 * cnt - 4);
2333  break;
2334  case EXT_FILL_DATA:
2335  case EXT_DATA_ELEMENT:
2336  default:
2337  skip_bits_long(gb, 8 * cnt - 4);
2338  break;
2339  };
2340  return res;
2341 }
2342 
2343 /**
2344  * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2345  *
2346  * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2347  * @param coef spectral coefficients
2348  */
2349 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2350  IndividualChannelStream *ics, int decode)
2351 {
2352  const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2353  int w, filt, m, i;
2354  int bottom, top, order, start, end, size, inc;
2355  float lpc[TNS_MAX_ORDER];
2356  float tmp[TNS_MAX_ORDER+1];
2357 
2358  for (w = 0; w < ics->num_windows; w++) {
2359  bottom = ics->num_swb;
2360  for (filt = 0; filt < tns->n_filt[w]; filt++) {
2361  top = bottom;
2362  bottom = FFMAX(0, top - tns->length[w][filt]);
2363  order = tns->order[w][filt];
2364  if (order == 0)
2365  continue;
2366 
2367  // tns_decode_coef
2368  compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2369 
2370  start = ics->swb_offset[FFMIN(bottom, mmm)];
2371  end = ics->swb_offset[FFMIN( top, mmm)];
2372  if ((size = end - start) <= 0)
2373  continue;
2374  if (tns->direction[w][filt]) {
2375  inc = -1;
2376  start = end - 1;
2377  } else {
2378  inc = 1;
2379  }
2380  start += w * 128;
2381 
2382  if (decode) {
2383  // ar filter
2384  for (m = 0; m < size; m++, start += inc)
2385  for (i = 1; i <= FFMIN(m, order); i++)
2386  coef[start] -= coef[start - i * inc] * lpc[i - 1];
2387  } else {
2388  // ma filter
2389  for (m = 0; m < size; m++, start += inc) {
2390  tmp[0] = coef[start];
2391  for (i = 1; i <= FFMIN(m, order); i++)
2392  coef[start] += tmp[i] * lpc[i - 1];
2393  for (i = order; i > 0; i--)
2394  tmp[i] = tmp[i - 1];
2395  }
2396  }
2397  }
2398  }
2399 }
2400 
2401 /**
2402  * Apply windowing and MDCT to obtain the spectral
2403  * coefficient from the predicted sample by LTP.
2404  */
2405 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2406  float *in, IndividualChannelStream *ics)
2407 {
2408  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2409  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2410  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2411  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2412 
2413  if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2414  ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2415  } else {
2416  memset(in, 0, 448 * sizeof(float));
2417  ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2418  }
2419  if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2420  ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2421  } else {
2422  ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2423  memset(in + 1024 + 576, 0, 448 * sizeof(float));
2424  }
2425  ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2426 }
2427 
2428 /**
2429  * Apply the long term prediction
2430  */
2432 {
2433  const LongTermPrediction *ltp = &sce->ics.ltp;
2434  const uint16_t *offsets = sce->ics.swb_offset;
2435  int i, sfb;
2436 
2437  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2438  float *predTime = sce->ret;
2439  float *predFreq = ac->buf_mdct;
2440  int16_t num_samples = 2048;
2441 
2442  if (ltp->lag < 1024)
2443  num_samples = ltp->lag + 1024;
2444  for (i = 0; i < num_samples; i++)
2445  predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2446  memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2447 
2448  ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2449 
2450  if (sce->tns.present)
2451  ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2452 
2453  for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2454  if (ltp->used[sfb])
2455  for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2456  sce->coeffs[i] += predFreq[i];
2457  }
2458 }
2459 
2460 /**
2461  * Update the LTP buffer for next frame
2462  */
2464 {
2465  IndividualChannelStream *ics = &sce->ics;
2466  float *saved = sce->saved;
2467  float *saved_ltp = sce->coeffs;
2468  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2469  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2470  int i;
2471 
2472  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2473  memcpy(saved_ltp, saved, 512 * sizeof(float));
2474  memset(saved_ltp + 576, 0, 448 * sizeof(float));
2475  ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2476  for (i = 0; i < 64; i++)
2477  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2478  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2479  memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2480  memset(saved_ltp + 576, 0, 448 * sizeof(float));
2481  ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2482  for (i = 0; i < 64; i++)
2483  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2484  } else { // LONG_STOP or ONLY_LONG
2485  ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2486  for (i = 0; i < 512; i++)
2487  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2488  }
2489 
2490  memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2491  memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2492  memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2493 }
2494 
2495 /**
2496  * Conduct IMDCT and windowing.
2497  */
2499 {
2500  IndividualChannelStream *ics = &sce->ics;
2501  float *in = sce->coeffs;
2502  float *out = sce->ret;
2503  float *saved = sce->saved;
2504  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2505  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2506  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2507  float *buf = ac->buf_mdct;
2508  float *temp = ac->temp;
2509  int i;
2510 
2511  // imdct
2512  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2513  for (i = 0; i < 1024; i += 128)
2514  ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2515  } else
2516  ac->mdct.imdct_half(&ac->mdct, buf, in);
2517 
2518  /* window overlapping
2519  * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2520  * and long to short transitions are considered to be short to short
2521  * transitions. This leaves just two cases (long to long and short to short)
2522  * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2523  */
2524  if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2526  ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2527  } else {
2528  memcpy( out, saved, 448 * sizeof(float));
2529 
2530  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2531  ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2532  ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2533  ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2534  ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2535  ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2536  memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2537  } else {
2538  ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2539  memcpy( out + 576, buf + 64, 448 * sizeof(float));
2540  }
2541  }
2542 
2543  // buffer update
2544  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2545  memcpy( saved, temp + 64, 64 * sizeof(float));
2546  ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2547  ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2548  ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2549  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2550  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2551  memcpy( saved, buf + 512, 448 * sizeof(float));
2552  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2553  } else { // LONG_STOP or ONLY_LONG
2554  memcpy( saved, buf + 512, 512 * sizeof(float));
2555  }
2556 }
2557 
2559 {
2560  IndividualChannelStream *ics = &sce->ics;
2561  float *in = sce->coeffs;
2562  float *out = sce->ret;
2563  float *saved = sce->saved;
2564  float *buf = ac->buf_mdct;
2565 
2566  // imdct
2567  ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2568 
2569  // window overlapping
2570  if (ics->use_kb_window[1]) {
2571  // AAC LD uses a low overlap sine window instead of a KBD window
2572  memcpy(out, saved, 192 * sizeof(float));
2573  ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2574  memcpy( out + 320, buf + 64, 192 * sizeof(float));
2575  } else {
2576  ac->fdsp->vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2577  }
2578 
2579  // buffer update
2580  memcpy(saved, buf + 256, 256 * sizeof(float));
2581 }
2582 
2584 {
2585  float *in = sce->coeffs;
2586  float *out = sce->ret;
2587  float *saved = sce->saved;
2588  const float *const window = ff_aac_eld_window;
2589  float *buf = ac->buf_mdct;
2590  int i;
2591  const int n = 512;
2592  const int n2 = n >> 1;
2593  const int n4 = n >> 2;
2594 
2595  // Inverse transform, mapped to the conventional IMDCT by
2596  // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2597  // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2598  // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2599  // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2600  for (i = 0; i < n2; i+=2) {
2601  float temp;
2602  temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2603  temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2604  }
2605  ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2606  for (i = 0; i < n; i+=2) {
2607  buf[i] = -buf[i];
2608  }
2609  // Like with the regular IMDCT at this point we still have the middle half
2610  // of a transform but with even symmetry on the left and odd symmetry on
2611  // the right
2612 
2613  // window overlapping
2614  // The spec says to use samples [0..511] but the reference decoder uses
2615  // samples [128..639].
2616  for (i = n4; i < n2; i ++) {
2617  out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2618  saved[ i + n2] * window[i + n - n4] +
2619  -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2620  -saved[2*n + n2 + i] * window[i + 3*n - n4];
2621  }
2622  for (i = 0; i < n2; i ++) {
2623  out[n4 + i] = buf[i] * window[i + n2 - n4] +
2624  -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2625  -saved[ n + i] * window[i + n2 + 2*n - n4] +
2626  saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2627  }
2628  for (i = 0; i < n4; i ++) {
2629  out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2630  -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2631  -saved[ n + n2 + i] * window[i + 3*n - n4];
2632  }
2633 
2634  // buffer update
2635  memmove(saved + n, saved, 2 * n * sizeof(float));
2636  memcpy( saved, buf, n * sizeof(float));
2637 }
2638 
2639 /**
2640  * Apply dependent channel coupling (applied before IMDCT).
2641  *
2642  * @param index index into coupling gain array
2643  */
2645  SingleChannelElement *target,
2646  ChannelElement *cce, int index)
2647 {
2648  IndividualChannelStream *ics = &cce->ch[0].ics;
2649  const uint16_t *offsets = ics->swb_offset;
2650  float *dest = target->coeffs;
2651  const float *src = cce->ch[0].coeffs;
2652  int g, i, group, k, idx = 0;
2653  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2654  av_log(ac->avctx, AV_LOG_ERROR,
2655  "Dependent coupling is not supported together with LTP\n");
2656  return;
2657  }
2658  for (g = 0; g < ics->num_window_groups; g++) {
2659  for (i = 0; i < ics->max_sfb; i++, idx++) {
2660  if (cce->ch[0].band_type[idx] != ZERO_BT) {
2661  const float gain = cce->coup.gain[index][idx];
2662  for (group = 0; group < ics->group_len[g]; group++) {
2663  for (k = offsets[i]; k < offsets[i + 1]; k++) {
2664  // FIXME: SIMDify
2665  dest[group * 128 + k] += gain * src[group * 128 + k];
2666  }
2667  }
2668  }
2669  }
2670  dest += ics->group_len[g] * 128;
2671  src += ics->group_len[g] * 128;
2672  }
2673 }
2674 
2675 /**
2676  * Apply independent channel coupling (applied after IMDCT).
2677  *
2678  * @param index index into coupling gain array
2679  */
2681  SingleChannelElement *target,
2682  ChannelElement *cce, int index)
2683 {
2684  int i;
2685  const float gain = cce->coup.gain[index][0];
2686  const float *src = cce->ch[0].ret;
2687  float *dest = target->ret;
2688  const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2689 
2690  for (i = 0; i < len; i++)
2691  dest[i] += gain * src[i];
2692 }
2693 
2694 /**
2695  * channel coupling transformation interface
2696  *
2697  * @param apply_coupling_method pointer to (in)dependent coupling function
2698  */
2700  enum RawDataBlockType type, int elem_id,
2701  enum CouplingPoint coupling_point,
2702  void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2703 {
2704  int i, c;
2705 
2706  for (i = 0; i < MAX_ELEM_ID; i++) {
2707  ChannelElement *cce = ac->che[TYPE_CCE][i];
2708  int index = 0;
2709 
2710  if (cce && cce->coup.coupling_point == coupling_point) {
2711  ChannelCoupling *coup = &cce->coup;
2712 
2713  for (c = 0; c <= coup->num_coupled; c++) {
2714  if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2715  if (coup->ch_select[c] != 1) {
2716  apply_coupling_method(ac, &cc->ch[0], cce, index);
2717  if (coup->ch_select[c] != 0)
2718  index++;
2719  }
2720  if (coup->ch_select[c] != 2)
2721  apply_coupling_method(ac, &cc->ch[1], cce, index++);
2722  } else
2723  index += 1 + (coup->ch_select[c] == 3);
2724  }
2725  }
2726  }
2727 }
2728 
2729 /**
2730  * Convert spectral data to float samples, applying all supported tools as appropriate.
2731  */
2733 {
2734  int i, type;
2736  switch (ac->oc[1].m4ac.object_type) {
2737  case AOT_ER_AAC_LD:
2739  break;
2740  case AOT_ER_AAC_ELD:
2742  break;
2743  default:
2745  }
2746  for (type = 3; type >= 0; type--) {
2747  for (i = 0; i < MAX_ELEM_ID; i++) {
2748  ChannelElement *che = ac->che[type][i];
2749  if (che && che->present) {
2750  if (type <= TYPE_CPE)
2752  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2753  if (che->ch[0].ics.predictor_present) {
2754  if (che->ch[0].ics.ltp.present)
2755  ac->apply_ltp(ac, &che->ch[0]);
2756  if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2757  ac->apply_ltp(ac, &che->ch[1]);
2758  }
2759  }
2760  if (che->ch[0].tns.present)
2761  ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2762  if (che->ch[1].tns.present)
2763  ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2764  if (type <= TYPE_CPE)
2766  if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2767  imdct_and_window(ac, &che->ch[0]);
2768  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2769  ac->update_ltp(ac, &che->ch[0]);
2770  if (type == TYPE_CPE) {
2771  imdct_and_window(ac, &che->ch[1]);
2772  if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2773  ac->update_ltp(ac, &che->ch[1]);
2774  }
2775  if (ac->oc[1].m4ac.sbr > 0) {
2776  ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2777  }
2778  }
2779  if (type <= TYPE_CCE)
2781  che->present = 0;
2782  } else if (che) {
2783  av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2784  }
2785  }
2786  }
2787 }
2788 
2790 {
2791  int size;
2792  AACADTSHeaderInfo hdr_info;
2793  uint8_t layout_map[MAX_ELEM_ID*4][3];
2794  int layout_map_tags, ret;
2795 
2796  size = avpriv_aac_parse_header(gb, &hdr_info);
2797  if (size > 0) {
2798  if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2799  // This is 2 for "VLB " audio in NSV files.
2800  // See samples/nsv/vlb_audio.
2802  "More than one AAC RDB per ADTS frame");
2803  ac->warned_num_aac_frames = 1;
2804  }
2806  if (hdr_info.chan_config) {
2807  ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2808  if ((ret = set_default_channel_config(ac->avctx,
2809  layout_map,
2810  &layout_map_tags,
2811  hdr_info.chan_config)) < 0)
2812  return ret;
2813  if ((ret = output_configure(ac, layout_map, layout_map_tags,
2814  FFMAX(ac->oc[1].status,
2815  OC_TRIAL_FRAME), 0)) < 0)
2816  return ret;
2817  } else {
2818  ac->oc[1].m4ac.chan_config = 0;
2819  /**
2820  * dual mono frames in Japanese DTV can have chan_config 0
2821  * WITHOUT specifying PCE.
2822  * thus, set dual mono as default.
2823  */
2824  if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2825  layout_map_tags = 2;
2826  layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2827  layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2828  layout_map[0][1] = 0;
2829  layout_map[1][1] = 1;
2830  if (output_configure(ac, layout_map, layout_map_tags,
2831  OC_TRIAL_FRAME, 0))
2832  return -7;
2833  }
2834  }
2835  ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2836  ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2837  ac->oc[1].m4ac.object_type = hdr_info.object_type;
2838  if (ac->oc[0].status != OC_LOCKED ||
2839  ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2840  ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2841  ac->oc[1].m4ac.sbr = -1;
2842  ac->oc[1].m4ac.ps = -1;
2843  }
2844  if (!hdr_info.crc_absent)
2845  skip_bits(gb, 16);
2846  }
2847  return size;
2848 }
2849 
2850 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2851  int *got_frame_ptr, GetBitContext *gb)
2852 {
2853  AACContext *ac = avctx->priv_data;
2854  ChannelElement *che;
2855  int err, i;
2856  int samples = 1024;
2857  int chan_config = ac->oc[1].m4ac.chan_config;
2858  int aot = ac->oc[1].m4ac.object_type;
2859 
2860  if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2861  samples >>= 1;
2862 
2863  ac->frame = data;
2864 
2865  if ((err = frame_configure_elements(avctx)) < 0)
2866  return err;
2867 
2868  // The FF_PROFILE_AAC_* defines are all object_type - 1
2869  // This may lead to an undefined profile being signaled
2870  ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2871 
2872  ac->tags_mapped = 0;
2873 
2874  if (chan_config < 0 || chan_config >= 8) {
2875  avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2876  ac->oc[1].m4ac.chan_config);
2877  return AVERROR_INVALIDDATA;
2878  }
2879  for (i = 0; i < tags_per_config[chan_config]; i++) {
2880  const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2881  const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2882  if (!(che=get_che(ac, elem_type, elem_id))) {
2883  av_log(ac->avctx, AV_LOG_ERROR,
2884  "channel element %d.%d is not allocated\n",
2885  elem_type, elem_id);
2886  return AVERROR_INVALIDDATA;
2887  }
2888  che->present = 1;
2889  if (aot != AOT_ER_AAC_ELD)
2890  skip_bits(gb, 4);
2891  switch (elem_type) {
2892  case TYPE_SCE:
2893  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2894  break;
2895  case TYPE_CPE:
2896  err = decode_cpe(ac, gb, che);
2897  break;
2898  case TYPE_LFE:
2899  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2900  break;
2901  }
2902  if (err < 0)
2903  return err;
2904  }
2905 
2906  spectral_to_sample(ac);
2907 
2908  ac->frame->nb_samples = samples;
2909  ac->frame->sample_rate = avctx->sample_rate;
2910  *got_frame_ptr = 1;
2911 
2912  skip_bits_long(gb, get_bits_left(gb));
2913  return 0;
2914 }
2915 
2916 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2917  int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2918 {
2919  AACContext *ac = avctx->priv_data;
2920  ChannelElement *che = NULL, *che_prev = NULL;
2921  enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2922  int err, elem_id;
2923  int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2924  int is_dmono, sce_count = 0;
2925 
2926  ac->frame = data;
2927 
2928  if (show_bits(gb, 12) == 0xfff) {
2929  if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2930  av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2931  goto fail;
2932  }
2933  if (ac->oc[1].m4ac.sampling_index > 12) {
2934  av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2935  err = AVERROR_INVALIDDATA;
2936  goto fail;
2937  }
2938  }
2939 
2940  if ((err = frame_configure_elements(avctx)) < 0)
2941  goto fail;
2942 
2943  // The FF_PROFILE_AAC_* defines are all object_type - 1
2944  // This may lead to an undefined profile being signaled
2945  ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2946 
2947  ac->tags_mapped = 0;
2948  // parse
2949  while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2950  elem_id = get_bits(gb, 4);
2951 
2952  if (avctx->debug & FF_DEBUG_STARTCODE)
2953  av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2954 
2955  if (elem_type < TYPE_DSE) {
2956  if (!(che=get_che(ac, elem_type, elem_id))) {
2957  av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2958  elem_type, elem_id);
2959  err = AVERROR_INVALIDDATA;
2960  goto fail;
2961  }
2962  samples = 1024;
2963  che->present = 1;
2964  }
2965 
2966  switch (elem_type) {
2967 
2968  case TYPE_SCE:
2969  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2970  audio_found = 1;
2971  sce_count++;
2972  break;
2973 
2974  case TYPE_CPE:
2975  err = decode_cpe(ac, gb, che);
2976  audio_found = 1;
2977  break;
2978 
2979  case TYPE_CCE:
2980  err = decode_cce(ac, gb, che);
2981  break;
2982 
2983  case TYPE_LFE:
2984  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2985  audio_found = 1;
2986  break;
2987 
2988  case TYPE_DSE:
2989  err = skip_data_stream_element(ac, gb);
2990  break;
2991 
2992  case TYPE_PCE: {
2993  uint8_t layout_map[MAX_ELEM_ID*4][3];
2994  int tags;
2996  tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2997  if (tags < 0) {
2998  err = tags;
2999  break;
3000  }
3001  if (pce_found) {
3002  av_log(avctx, AV_LOG_ERROR,
3003  "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3004  } else {
3005  err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3006  if (!err)
3007  ac->oc[1].m4ac.chan_config = 0;
3008  pce_found = 1;
3009  }
3010  break;
3011  }
3012 
3013  case TYPE_FIL:
3014  if (elem_id == 15)
3015  elem_id += get_bits(gb, 8) - 1;
3016  if (get_bits_left(gb) < 8 * elem_id) {
3017  av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3018  err = AVERROR_INVALIDDATA;
3019  goto fail;
3020  }
3021  while (elem_id > 0)
3022  elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3023  err = 0; /* FIXME */
3024  break;
3025 
3026  default:
3027  err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3028  break;
3029  }
3030 
3031  che_prev = che;
3032  elem_type_prev = elem_type;
3033 
3034  if (err)
3035  goto fail;
3036 
3037  if (get_bits_left(gb) < 3) {
3038  av_log(avctx, AV_LOG_ERROR, overread_err);
3039  err = AVERROR_INVALIDDATA;
3040  goto fail;
3041  }
3042  }
3043 
3044  spectral_to_sample(ac);
3045 
3046  multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3047  samples <<= multiplier;
3048 
3049  if (ac->oc[1].status && audio_found) {
3050  avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3051  avctx->frame_size = samples;
3052  ac->oc[1].status = OC_LOCKED;
3053  }
3054 
3055  if (multiplier) {
3056  int side_size;
3057  const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3058  if (side && side_size>=4)
3059  AV_WL32(side, 2*AV_RL32(side));
3060  }
3061 
3062  *got_frame_ptr = !!samples;
3063  if (samples) {
3064  ac->frame->nb_samples = samples;
3065  ac->frame->sample_rate = avctx->sample_rate;
3066  } else
3067  av_frame_unref(ac->frame);
3068  *got_frame_ptr = !!samples;
3069 
3070  /* for dual-mono audio (SCE + SCE) */
3071  is_dmono = ac->dmono_mode && sce_count == 2 &&
3073  if (is_dmono) {
3074  if (ac->dmono_mode == 1)
3075  ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3076  else if (ac->dmono_mode == 2)
3077  ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3078  }
3079 
3080  return 0;
3081 fail:
3083  return err;
3084 }
3085 
3086 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3087  int *got_frame_ptr, AVPacket *avpkt)
3088 {
3089  AACContext *ac = avctx->priv_data;
3090  const uint8_t *buf = avpkt->data;
3091  int buf_size = avpkt->size;
3092  GetBitContext gb;
3093  int buf_consumed;
3094  int buf_offset;
3095  int err;
3096  int new_extradata_size;
3097  const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3099  &new_extradata_size);
3100  int jp_dualmono_size;
3101  const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3103  &jp_dualmono_size);
3104 
3105  if (new_extradata && 0) {
3106  av_free(avctx->extradata);
3107  avctx->extradata = av_mallocz(new_extradata_size +
3109  if (!avctx->extradata)
3110  return AVERROR(ENOMEM);
3111  avctx->extradata_size = new_extradata_size;
3112  memcpy(avctx->extradata, new_extradata, new_extradata_size);
3114  if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3115  avctx->extradata,
3116  avctx->extradata_size*8, 1) < 0) {
3118  return AVERROR_INVALIDDATA;
3119  }
3120  }
3121 
3122  ac->dmono_mode = 0;
3123  if (jp_dualmono && jp_dualmono_size > 0)
3124  ac->dmono_mode = 1 + *jp_dualmono;
3125  if (ac->force_dmono_mode >= 0)
3126  ac->dmono_mode = ac->force_dmono_mode;
3127 
3128  if (INT_MAX / 8 <= buf_size)
3129  return AVERROR_INVALIDDATA;
3130 
3131  if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3132  return err;
3133 
3134  switch (ac->oc[1].m4ac.object_type) {
3135  case AOT_ER_AAC_LC:
3136  case AOT_ER_AAC_LTP:
3137  case AOT_ER_AAC_LD:
3138  case AOT_ER_AAC_ELD:
3139  err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3140  break;
3141  default:
3142  err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3143  }
3144  if (err < 0)
3145  return err;
3146 
3147  buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3148  for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3149  if (buf[buf_offset])
3150  break;
3151 
3152  return buf_size > buf_offset ? buf_consumed : buf_size;
3153 }
3154 
3156 {
3157  AACContext *ac = avctx->priv_data;
3158  int i, type;
3159 
3160  for (i = 0; i < MAX_ELEM_ID; i++) {
3161  for (type = 0; type < 4; type++) {
3162  if (ac->che[type][i])
3163  ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3164  av_freep(&ac->che[type][i]);
3165  }
3166  }
3167 
3168  ff_mdct_end(&ac->mdct);
3169  ff_mdct_end(&ac->mdct_small);
3170  ff_mdct_end(&ac->mdct_ld);
3171  ff_mdct_end(&ac->mdct_ltp);
3172  av_freep(&ac->fdsp);
3173  return 0;
3174 }
3175 
3176 
3177 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3178 
3179 struct LATMContext {
3180  AACContext aac_ctx; ///< containing AACContext
3181  int initialized; ///< initialized after a valid extradata was seen
3182 
3183  // parser data
3184  int audio_mux_version_A; ///< LATM syntax version
3185  int frame_length_type; ///< 0/1 variable/fixed frame length
3186  int frame_length; ///< frame length for fixed frame length
3187 };
3188 
3189 static inline uint32_t latm_get_value(GetBitContext *b)
3190 {
3191  int length = get_bits(b, 2);
3192 
3193  return get_bits_long(b, (length+1)*8);
3194 }
3195 
3197  GetBitContext *gb, int asclen)
3198 {
3199  AACContext *ac = &latmctx->aac_ctx;
3200  AVCodecContext *avctx = ac->avctx;
3201  MPEG4AudioConfig m4ac = { 0 };
3202  int config_start_bit = get_bits_count(gb);
3203  int sync_extension = 0;
3204  int bits_consumed, esize;
3205 
3206  if (asclen) {
3207  sync_extension = 1;
3208  asclen = FFMIN(asclen, get_bits_left(gb));
3209  } else
3210  asclen = get_bits_left(gb);
3211 
3212  if (config_start_bit % 8) {
3214  "Non-byte-aligned audio-specific config");
3215  return AVERROR_PATCHWELCOME;
3216  }
3217  if (asclen <= 0)
3218  return AVERROR_INVALIDDATA;
3219  bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3220  gb->buffer + (config_start_bit / 8),
3221  asclen, sync_extension);
3222 
3223  if (bits_consumed < 0)
3224  return AVERROR_INVALIDDATA;
3225 
3226  if (!latmctx->initialized ||
3227  ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3228  ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3229 
3230  if(latmctx->initialized) {
3231  av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3232  } else {
3233  av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3234  }
3235  latmctx->initialized = 0;
3236 
3237  esize = (bits_consumed+7) / 8;
3238 
3239  if (avctx->extradata_size < esize) {
3240  av_free(avctx->extradata);
3242  if (!avctx->extradata)
3243  return AVERROR(ENOMEM);
3244  }
3245 
3246  avctx->extradata_size = esize;
3247  memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3248  memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3249  }
3250  skip_bits_long(gb, bits_consumed);
3251 
3252  return bits_consumed;
3253 }
3254 
3255 static int read_stream_mux_config(struct LATMContext *latmctx,
3256  GetBitContext *gb)
3257 {
3258  int ret, audio_mux_version = get_bits(gb, 1);
3259 
3260  latmctx->audio_mux_version_A = 0;
3261  if (audio_mux_version)
3262  latmctx->audio_mux_version_A = get_bits(gb, 1);
3263 
3264  if (!latmctx->audio_mux_version_A) {
3265 
3266  if (audio_mux_version)
3267  latm_get_value(gb); // taraFullness
3268 
3269  skip_bits(gb, 1); // allStreamSameTimeFraming
3270  skip_bits(gb, 6); // numSubFrames
3271  // numPrograms
3272  if (get_bits(gb, 4)) { // numPrograms
3273  avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3274  return AVERROR_PATCHWELCOME;
3275  }
3276 
3277  // for each program (which there is only one in DVB)
3278 
3279  // for each layer (which there is only one in DVB)
3280  if (get_bits(gb, 3)) { // numLayer
3281  avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3282  return AVERROR_PATCHWELCOME;
3283  }
3284 
3285  // for all but first stream: use_same_config = get_bits(gb, 1);
3286  if (!audio_mux_version) {
3287  if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3288  return ret;
3289  } else {
3290  int ascLen = latm_get_value(gb);
3291  if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3292  return ret;
3293  ascLen -= ret;
3294  skip_bits_long(gb, ascLen);
3295  }
3296 
3297  latmctx->frame_length_type = get_bits(gb, 3);
3298  switch (latmctx->frame_length_type) {
3299  case 0:
3300  skip_bits(gb, 8); // latmBufferFullness
3301  break;
3302  case 1:
3303  latmctx->frame_length = get_bits(gb, 9);
3304  break;
3305  case 3:
3306  case 4:
3307  case 5:
3308  skip_bits(gb, 6); // CELP frame length table index
3309  break;
3310  case 6:
3311  case 7:
3312  skip_bits(gb, 1); // HVXC frame length table index
3313  break;
3314  }
3315 
3316  if (get_bits(gb, 1)) { // other data
3317  if (audio_mux_version) {
3318  latm_get_value(gb); // other_data_bits
3319  } else {
3320  int esc;
3321  do {
3322  esc = get_bits(gb, 1);
3323  skip_bits(gb, 8);
3324  } while (esc);
3325  }
3326  }
3327 
3328  if (get_bits(gb, 1)) // crc present
3329  skip_bits(gb, 8); // config_crc
3330  }
3331 
3332  return 0;
3333 }
3334 
3336 {
3337  uint8_t tmp;
3338 
3339  if (ctx->frame_length_type == 0) {
3340  int mux_slot_length = 0;
3341  do {
3342  tmp = get_bits(gb, 8);
3343  mux_slot_length += tmp;
3344  } while (tmp == 255);
3345  return mux_slot_length;
3346  } else if (ctx->frame_length_type == 1) {
3347  return ctx->frame_length;
3348  } else if (ctx->frame_length_type == 3 ||
3349  ctx->frame_length_type == 5 ||
3350  ctx->frame_length_type == 7) {
3351  skip_bits(gb, 2); // mux_slot_length_coded
3352  }
3353  return 0;
3354 }
3355 
3356 static int read_audio_mux_element(struct LATMContext *latmctx,
3357  GetBitContext *gb)
3358 {
3359  int err;
3360  uint8_t use_same_mux = get_bits(gb, 1);
3361  if (!use_same_mux) {
3362  if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3363  return err;
3364  } else if (!latmctx->aac_ctx.avctx->extradata) {
3365  av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3366  "no decoder config found\n");
3367  return AVERROR(EAGAIN);
3368  }
3369  if (latmctx->audio_mux_version_A == 0) {
3370  int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3371  if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3372  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3373  return AVERROR_INVALIDDATA;
3374  } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3375  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3376  "frame length mismatch %d << %d\n",
3377  mux_slot_length_bytes * 8, get_bits_left(gb));
3378  return AVERROR_INVALIDDATA;
3379  }
3380  }
3381  return 0;
3382 }
3383 
3384 
3385 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3386  int *got_frame_ptr, AVPacket *avpkt)
3387 {
3388  struct LATMContext *latmctx = avctx->priv_data;
3389  int muxlength, err;
3390  GetBitContext gb;
3391 
3392  if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3393  return err;
3394 
3395  // check for LOAS sync word
3396  if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3397  return AVERROR_INVALIDDATA;
3398 
3399  muxlength = get_bits(&gb, 13) + 3;
3400  // not enough data, the parser should have sorted this out
3401  if (muxlength > avpkt->size)
3402  return AVERROR_INVALIDDATA;
3403 
3404  if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3405  return err;
3406 
3407  if (!latmctx->initialized) {
3408  if (!avctx->extradata) {
3409  *got_frame_ptr = 0;
3410  return avpkt->size;
3411  } else {
3413  if ((err = decode_audio_specific_config(
3414  &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3415  avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3416  pop_output_configuration(&latmctx->aac_ctx);
3417  return err;
3418  }
3419  latmctx->initialized = 1;
3420  }
3421  }
3422 
3423  if (show_bits(&gb, 12) == 0xfff) {
3424  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3425  "ADTS header detected, probably as result of configuration "
3426  "misparsing\n");
3427  return AVERROR_INVALIDDATA;
3428  }
3429 
3430  if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
3431  return err;
3432 
3433  return muxlength;
3434 }
3435 
3437 {
3438  struct LATMContext *latmctx = avctx->priv_data;
3439  int ret = aac_decode_init(avctx);
3440 
3441  if (avctx->extradata_size > 0)
3442  latmctx->initialized = !ret;
3443 
3444  return ret;
3445 }
3446 
3447 static void aacdec_init(AACContext *c)
3448 {
3450  c->apply_ltp = apply_ltp;
3451  c->apply_tns = apply_tns;
3453  c->update_ltp = update_ltp;
3454 
3455  if(ARCH_MIPS)
3457 }
3458 /**
3459  * AVOptions for Japanese DTV specific extensions (ADTS only)
3460  */
3461 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3462 static const AVOption options[] = {
3463  {"dual_mono_mode", "Select the channel to decode for dual mono",
3464  offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3465  AACDEC_FLAGS, "dual_mono_mode"},
3466 
3467  {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3468  {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3469  {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3470  {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3471 
3472  {NULL},
3473 };
3474 
3475 static const AVClass aac_decoder_class = {
3476  .class_name = "AAC decoder",
3477  .item_name = av_default_item_name,
3478  .option = options,
3479  .version = LIBAVUTIL_VERSION_INT,
3480 };
3481 
3482 static const AVProfile profiles[] = {
3483  { FF_PROFILE_AAC_MAIN, "Main" },
3484  { FF_PROFILE_AAC_LOW, "LC" },
3485  { FF_PROFILE_AAC_SSR, "SSR" },
3486  { FF_PROFILE_AAC_LTP, "LTP" },
3487  { FF_PROFILE_AAC_HE, "HE-AAC" },
3488  { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3489  { FF_PROFILE_AAC_LD, "LD" },
3490  { FF_PROFILE_AAC_ELD, "ELD" },
3491  { FF_PROFILE_UNKNOWN },
3492 };
3493 
3495  .name = "aac",
3496  .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3497  .type = AVMEDIA_TYPE_AUDIO,
3498  .id = AV_CODEC_ID_AAC,
3499  .priv_data_size = sizeof(AACContext),
3500  .init = aac_decode_init,
3501  .close = aac_decode_close,
3503  .sample_fmts = (const enum AVSampleFormat[]) {
3505  },
3506  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3507  .channel_layouts = aac_channel_layout,
3508  .flush = flush,
3509  .priv_class = &aac_decoder_class,
3510  .profiles = profiles,
3511 };
3512 
3513 /*
3514  Note: This decoder filter is intended to decode LATM streams transferred
3515  in MPEG transport streams which only contain one program.
3516  To do a more complex LATM demuxing a separate LATM demuxer should be used.
3517 */
3519  .name = "aac_latm",
3520  .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3521  .type = AVMEDIA_TYPE_AUDIO,
3522  .id = AV_CODEC_ID_AAC_LATM,
3523  .priv_data_size = sizeof(struct LATMContext),
3524  .init = latm_decode_init,
3525  .close = aac_decode_close,
3526  .decode = latm_decode_frame,
3527  .sample_fmts = (const enum AVSampleFormat[]) {
3529  },
3530  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3531  .channel_layouts = aac_channel_layout,
3532  .flush = flush,
3533  .profiles = profiles,
3534 };