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  *
6  * AAC LATM decoder
7  * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8  * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
9  *
10  * This file is part of Libav.
11  *
12  * Libav is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU Lesser General Public
14  * License as published by the Free Software Foundation; either
15  * version 2.1 of the License, or (at your option) any later version.
16  *
17  * Libav is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20  * Lesser General Public License for more details.
21  *
22  * You should have received a copy of the GNU Lesser General Public
23  * License along with Libav; if not, write to the Free Software
24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25  */
26 
34 /*
35  * supported tools
36  *
37  * Support? Name
38  * N (code in SoC repo) gain control
39  * Y block switching
40  * Y window shapes - standard
41  * N window shapes - Low Delay
42  * Y filterbank - standard
43  * N (code in SoC repo) filterbank - Scalable Sample Rate
44  * Y Temporal Noise Shaping
45  * Y Long Term Prediction
46  * Y intensity stereo
47  * Y channel coupling
48  * Y frequency domain prediction
49  * Y Perceptual Noise Substitution
50  * Y Mid/Side stereo
51  * N Scalable Inverse AAC Quantization
52  * N Frequency Selective Switch
53  * N upsampling filter
54  * Y quantization & coding - AAC
55  * N quantization & coding - TwinVQ
56  * N quantization & coding - BSAC
57  * N AAC Error Resilience tools
58  * N Error Resilience payload syntax
59  * N Error Protection tool
60  * N CELP
61  * N Silence Compression
62  * N HVXC
63  * N HVXC 4kbits/s VR
64  * N Structured Audio tools
65  * N Structured Audio Sample Bank Format
66  * N MIDI
67  * N Harmonic and Individual Lines plus Noise
68  * N Text-To-Speech Interface
69  * Y Spectral Band Replication
70  * Y (not in this code) Layer-1
71  * Y (not in this code) Layer-2
72  * Y (not in this code) Layer-3
73  * N SinuSoidal Coding (Transient, Sinusoid, Noise)
74  * Y Parametric Stereo
75  * N Direct Stream Transfer
76  *
77  * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78  * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
79  Parametric Stereo.
80  */
81 
82 
83 #include "avcodec.h"
84 #include "internal.h"
85 #include "get_bits.h"
86 #include "dsputil.h"
87 #include "fft.h"
88 #include "fmtconvert.h"
89 #include "lpc.h"
90 #include "kbdwin.h"
91 #include "sinewin.h"
92 
93 #include "aac.h"
94 #include "aactab.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
97 #include "sbr.h"
98 #include "aacsbr.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
101 #include "libavutil/intfloat.h"
102 
103 #include <assert.h>
104 #include <errno.h>
105 #include <math.h>
106 #include <string.h>
107 
108 #if ARCH_ARM
109 # include "arm/aac.h"
110 #endif
111 
113 static VLC vlc_spectral[11];
114 
115 static const char overread_err[] = "Input buffer exhausted before END element found\n";
116 
117 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
118 {
119  // For PCE based channel configurations map the channels solely based on tags.
120  if (!ac->m4ac.chan_config) {
121  return ac->tag_che_map[type][elem_id];
122  }
123  // For indexed channel configurations map the channels solely based on position.
124  switch (ac->m4ac.chan_config) {
125  case 7:
126  if (ac->tags_mapped == 3 && type == TYPE_CPE) {
127  ac->tags_mapped++;
128  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
129  }
130  case 6:
131  /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
132  instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
133  encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
134  if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
135  ac->tags_mapped++;
136  return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
137  }
138  case 5:
139  if (ac->tags_mapped == 2 && type == TYPE_CPE) {
140  ac->tags_mapped++;
141  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
142  }
143  case 4:
144  if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
145  ac->tags_mapped++;
146  return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
147  }
148  case 3:
149  case 2:
150  if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
151  ac->tags_mapped++;
152  return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
153  } else if (ac->m4ac.chan_config == 2) {
154  return NULL;
155  }
156  case 1:
157  if (!ac->tags_mapped && type == TYPE_SCE) {
158  ac->tags_mapped++;
159  return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
160  }
161  default:
162  return NULL;
163  }
164 }
165 
166 static int count_channels(enum ChannelPosition che_pos[4][MAX_ELEM_ID])
167 {
168  int i, type, sum = 0;
169  for (i = 0; i < MAX_ELEM_ID; i++) {
170  for (type = 0; type < 4; type++) {
171  sum += (1 + (type == TYPE_CPE)) *
172  (che_pos[type][i] != AAC_CHANNEL_OFF &&
173  che_pos[type][i] != AAC_CHANNEL_CC);
174  }
175  }
176  return sum;
177 }
178 
192  enum ChannelPosition che_pos[4][MAX_ELEM_ID],
193  int type, int id, int *channels)
194 {
195  if (che_pos[type][id]) {
196  if (!ac->che[type][id]) {
197  if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
198  return AVERROR(ENOMEM);
199  ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
200  }
201  if (type != TYPE_CCE) {
202  ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
203  if (type == TYPE_CPE ||
204  (type == TYPE_SCE && ac->m4ac.ps == 1)) {
205  ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
206  }
207  }
208  } else {
209  if (ac->che[type][id])
210  ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
211  av_freep(&ac->che[type][id]);
212  }
213  return 0;
214 }
215 
225  enum ChannelPosition che_pos[4][MAX_ELEM_ID],
226  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
227  int channel_config, enum OCStatus oc_type)
228 {
229  AVCodecContext *avctx = ac->avctx;
230  int i, type, channels = 0, ret;
231 
232  if (new_che_pos != che_pos)
233  memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
234 
235  if (channel_config) {
236  for (i = 0; i < tags_per_config[channel_config]; i++) {
237  if ((ret = che_configure(ac, che_pos,
238  aac_channel_layout_map[channel_config - 1][i][0],
239  aac_channel_layout_map[channel_config - 1][i][1],
240  &channels)))
241  return ret;
242  }
243 
244  memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
245 
246  avctx->channel_layout = aac_channel_layout[channel_config - 1];
247  } else {
248  /* Allocate or free elements depending on if they are in the
249  * current program configuration.
250  *
251  * Set up default 1:1 output mapping.
252  *
253  * For a 5.1 stream the output order will be:
254  * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
255  */
256 
257  for (i = 0; i < MAX_ELEM_ID; i++) {
258  for (type = 0; type < 4; type++) {
259  if ((ret = che_configure(ac, che_pos, type, i, &channels)))
260  return ret;
261  }
262  }
263 
264  memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
265 
266  avctx->channel_layout = 0;
267  }
268 
269  avctx->channels = channels;
270 
271  ac->output_configured = oc_type;
272 
273  return 0;
274 }
275 
283 static void decode_channel_map(enum ChannelPosition *cpe_map,
284  enum ChannelPosition *sce_map,
285  enum ChannelPosition type,
286  GetBitContext *gb, int n)
287 {
288  while (n--) {
289  enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
290  map[get_bits(gb, 4)] = type;
291  }
292 }
293 
301 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
302  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
303  GetBitContext *gb)
304 {
305  int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
306  int comment_len;
307 
308  skip_bits(gb, 2); // object_type
309 
310  sampling_index = get_bits(gb, 4);
311  if (m4ac->sampling_index != sampling_index)
312  av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
313 
314  num_front = get_bits(gb, 4);
315  num_side = get_bits(gb, 4);
316  num_back = get_bits(gb, 4);
317  num_lfe = get_bits(gb, 2);
318  num_assoc_data = get_bits(gb, 3);
319  num_cc = get_bits(gb, 4);
320 
321  if (get_bits1(gb))
322  skip_bits(gb, 4); // mono_mixdown_tag
323  if (get_bits1(gb))
324  skip_bits(gb, 4); // stereo_mixdown_tag
325 
326  if (get_bits1(gb))
327  skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
328 
329  decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
330  decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
331  decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
332  decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
333 
334  skip_bits_long(gb, 4 * num_assoc_data);
335 
336  decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
337 
338  align_get_bits(gb);
339 
340  /* comment field, first byte is length */
341  comment_len = get_bits(gb, 8) * 8;
342  if (get_bits_left(gb) < comment_len) {
344  return -1;
345  }
346  skip_bits_long(gb, comment_len);
347  return 0;
348 }
349 
359  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
360  int channel_config)
361 {
362  if (channel_config < 1 || channel_config > 7) {
363  av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
364  channel_config);
365  return -1;
366  }
367 
368  /* default channel configurations:
369  *
370  * 1ch : front center (mono)
371  * 2ch : L + R (stereo)
372  * 3ch : front center + L + R
373  * 4ch : front center + L + R + back center
374  * 5ch : front center + L + R + back stereo
375  * 6ch : front center + L + R + back stereo + LFE
376  * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
377  */
378 
379  if (channel_config != 2)
380  new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
381  if (channel_config > 1)
382  new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
383  if (channel_config == 4)
384  new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
385  if (channel_config > 4)
386  new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
387  = AAC_CHANNEL_BACK; // back stereo
388  if (channel_config > 5)
389  new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
390  if (channel_config == 7)
391  new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
392 
393  return 0;
394 }
395 
405  GetBitContext *gb,
406  MPEG4AudioConfig *m4ac,
407  int channel_config)
408 {
409  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
410  int extension_flag, ret;
411 
412  if (get_bits1(gb)) { // frameLengthFlag
413  av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
414  return -1;
415  }
416 
417  if (get_bits1(gb)) // dependsOnCoreCoder
418  skip_bits(gb, 14); // coreCoderDelay
419  extension_flag = get_bits1(gb);
420 
421  if (m4ac->object_type == AOT_AAC_SCALABLE ||
423  skip_bits(gb, 3); // layerNr
424 
425  memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
426  if (channel_config == 0) {
427  skip_bits(gb, 4); // element_instance_tag
428  if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
429  return ret;
430  } else {
431  if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
432  return ret;
433  }
434 
435  if (count_channels(new_che_pos) > 1) {
436  m4ac->ps = 0;
437  } else if (m4ac->sbr == 1 && m4ac->ps == -1)
438  m4ac->ps = 1;
439 
440  if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
441  return ret;
442 
443  if (extension_flag) {
444  switch (m4ac->object_type) {
445  case AOT_ER_BSAC:
446  skip_bits(gb, 5); // numOfSubFrame
447  skip_bits(gb, 11); // layer_length
448  break;
449  case AOT_ER_AAC_LC:
450  case AOT_ER_AAC_LTP:
451  case AOT_ER_AAC_SCALABLE:
452  case AOT_ER_AAC_LD:
453  skip_bits(gb, 3); /* aacSectionDataResilienceFlag
454  * aacScalefactorDataResilienceFlag
455  * aacSpectralDataResilienceFlag
456  */
457  break;
458  }
459  skip_bits1(gb); // extensionFlag3 (TBD in version 3)
460  }
461  return 0;
462 }
463 
477  AVCodecContext *avctx,
478  MPEG4AudioConfig *m4ac,
479  const uint8_t *data, int bit_size,
480  int sync_extension)
481 {
482  GetBitContext gb;
483  int i;
484 
485  av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
486  for (i = 0; i < avctx->extradata_size; i++)
487  av_dlog(avctx, "%02x ", avctx->extradata[i]);
488  av_dlog(avctx, "\n");
489 
490  init_get_bits(&gb, data, bit_size);
491 
492  if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
493  return -1;
494  if (m4ac->sampling_index > 12) {
495  av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
496  return -1;
497  }
498 
499  skip_bits_long(&gb, i);
500 
501  switch (m4ac->object_type) {
502  case AOT_AAC_MAIN:
503  case AOT_AAC_LC:
504  case AOT_AAC_LTP:
505  if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
506  return -1;
507  break;
508  default:
509  av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
510  m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
511  return -1;
512  }
513 
514  av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
515  m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
516  m4ac->sample_rate, m4ac->sbr, m4ac->ps);
517 
518  return get_bits_count(&gb);
519 }
520 
528 static av_always_inline int lcg_random(int previous_val)
529 {
530  return previous_val * 1664525 + 1013904223;
531 }
532 
534 {
535  ps->r0 = 0.0f;
536  ps->r1 = 0.0f;
537  ps->cor0 = 0.0f;
538  ps->cor1 = 0.0f;
539  ps->var0 = 1.0f;
540  ps->var1 = 1.0f;
541 }
542 
544 {
545  int i;
546  for (i = 0; i < MAX_PREDICTORS; i++)
547  reset_predict_state(&ps[i]);
548 }
549 
550 static int sample_rate_idx (int rate)
551 {
552  if (92017 <= rate) return 0;
553  else if (75132 <= rate) return 1;
554  else if (55426 <= rate) return 2;
555  else if (46009 <= rate) return 3;
556  else if (37566 <= rate) return 4;
557  else if (27713 <= rate) return 5;
558  else if (23004 <= rate) return 6;
559  else if (18783 <= rate) return 7;
560  else if (13856 <= rate) return 8;
561  else if (11502 <= rate) return 9;
562  else if (9391 <= rate) return 10;
563  else return 11;
564 }
565 
566 static void reset_predictor_group(PredictorState *ps, int group_num)
567 {
568  int i;
569  for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
570  reset_predict_state(&ps[i]);
571 }
572 
573 #define AAC_INIT_VLC_STATIC(num, size) \
574  INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
575  ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
576  ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
577  size);
578 
580 {
581  AACContext *ac = avctx->priv_data;
582  float output_scale_factor;
583 
584  ac->avctx = avctx;
585  ac->m4ac.sample_rate = avctx->sample_rate;
586 
587  if (avctx->extradata_size > 0) {
588  if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
589  avctx->extradata,
590  avctx->extradata_size*8, 1) < 0)
591  return -1;
592  } else {
593  int sr, i;
594  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
595 
596  sr = sample_rate_idx(avctx->sample_rate);
597  ac->m4ac.sampling_index = sr;
598  ac->m4ac.channels = avctx->channels;
599  ac->m4ac.sbr = -1;
600  ac->m4ac.ps = -1;
601 
602  for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
603  if (ff_mpeg4audio_channels[i] == avctx->channels)
604  break;
606  i = 0;
607  }
608  ac->m4ac.chan_config = i;
609 
610  if (ac->m4ac.chan_config) {
611  int ret = set_default_channel_config(avctx, new_che_pos, ac->m4ac.chan_config);
612  if (!ret)
613  output_configure(ac, ac->che_pos, new_che_pos, ac->m4ac.chan_config, OC_GLOBAL_HDR);
614  else if (avctx->err_recognition & AV_EF_EXPLODE)
615  return AVERROR_INVALIDDATA;
616  }
617  }
618 
619  if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
620  avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
621  output_scale_factor = 1.0 / 32768.0;
622  } else {
623  avctx->sample_fmt = AV_SAMPLE_FMT_S16;
624  output_scale_factor = 1.0;
625  }
626 
627  AAC_INIT_VLC_STATIC( 0, 304);
628  AAC_INIT_VLC_STATIC( 1, 270);
629  AAC_INIT_VLC_STATIC( 2, 550);
630  AAC_INIT_VLC_STATIC( 3, 300);
631  AAC_INIT_VLC_STATIC( 4, 328);
632  AAC_INIT_VLC_STATIC( 5, 294);
633  AAC_INIT_VLC_STATIC( 6, 306);
634  AAC_INIT_VLC_STATIC( 7, 268);
635  AAC_INIT_VLC_STATIC( 8, 510);
636  AAC_INIT_VLC_STATIC( 9, 366);
637  AAC_INIT_VLC_STATIC(10, 462);
638 
639  ff_aac_sbr_init();
640 
641  dsputil_init(&ac->dsp, avctx);
642  ff_fmt_convert_init(&ac->fmt_conv, avctx);
643 
644  ac->random_state = 0x1f2e3d4c;
645 
647 
651  352);
652 
653  ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
654  ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
655  ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
656  // window initialization
661 
662  cbrt_tableinit();
663 
665  avctx->coded_frame = &ac->frame;
666 
667  return 0;
668 }
669 
674 {
675  int byte_align = get_bits1(gb);
676  int count = get_bits(gb, 8);
677  if (count == 255)
678  count += get_bits(gb, 8);
679  if (byte_align)
680  align_get_bits(gb);
681 
682  if (get_bits_left(gb) < 8 * count) {
684  return -1;
685  }
686  skip_bits_long(gb, 8 * count);
687  return 0;
688 }
689 
691  GetBitContext *gb)
692 {
693  int sfb;
694  if (get_bits1(gb)) {
695  ics->predictor_reset_group = get_bits(gb, 5);
696  if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
697  av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
698  return -1;
699  }
700  }
701  for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
702  ics->prediction_used[sfb] = get_bits1(gb);
703  }
704  return 0;
705 }
706 
711  GetBitContext *gb, uint8_t max_sfb)
712 {
713  int sfb;
714 
715  ltp->lag = get_bits(gb, 11);
716  ltp->coef = ltp_coef[get_bits(gb, 3)];
717  for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
718  ltp->used[sfb] = get_bits1(gb);
719 }
720 
725  GetBitContext *gb)
726 {
727  if (get_bits1(gb)) {
728  av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
729  return AVERROR_INVALIDDATA;
730  }
731  ics->window_sequence[1] = ics->window_sequence[0];
732  ics->window_sequence[0] = get_bits(gb, 2);
733  ics->use_kb_window[1] = ics->use_kb_window[0];
734  ics->use_kb_window[0] = get_bits1(gb);
735  ics->num_window_groups = 1;
736  ics->group_len[0] = 1;
737  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
738  int i;
739  ics->max_sfb = get_bits(gb, 4);
740  for (i = 0; i < 7; i++) {
741  if (get_bits1(gb)) {
742  ics->group_len[ics->num_window_groups - 1]++;
743  } else {
744  ics->num_window_groups++;
745  ics->group_len[ics->num_window_groups - 1] = 1;
746  }
747  }
748  ics->num_windows = 8;
752  ics->predictor_present = 0;
753  } else {
754  ics->max_sfb = get_bits(gb, 6);
755  ics->num_windows = 1;
759  ics->predictor_present = get_bits1(gb);
760  ics->predictor_reset_group = 0;
761  if (ics->predictor_present) {
762  if (ac->m4ac.object_type == AOT_AAC_MAIN) {
763  if (decode_prediction(ac, ics, gb)) {
764  return AVERROR_INVALIDDATA;
765  }
766  } else if (ac->m4ac.object_type == AOT_AAC_LC) {
767  av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
768  return AVERROR_INVALIDDATA;
769  } else {
770  if ((ics->ltp.present = get_bits(gb, 1)))
771  decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
772  }
773  }
774  }
775 
776  if (ics->max_sfb > ics->num_swb) {
778  "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
779  ics->max_sfb, ics->num_swb);
780  return AVERROR_INVALIDDATA;
781  }
782 
783  return 0;
784 }
785 
794 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
795  int band_type_run_end[120], GetBitContext *gb,
797 {
798  int g, idx = 0;
799  const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
800  for (g = 0; g < ics->num_window_groups; g++) {
801  int k = 0;
802  while (k < ics->max_sfb) {
803  uint8_t sect_end = k;
804  int sect_len_incr;
805  int sect_band_type = get_bits(gb, 4);
806  if (sect_band_type == 12) {
807  av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
808  return -1;
809  }
810  do {
811  sect_len_incr = get_bits(gb, bits);
812  sect_end += sect_len_incr;
813  if (get_bits_left(gb) < 0) {
815  return -1;
816  }
817  if (sect_end > ics->max_sfb) {
819  "Number of bands (%d) exceeds limit (%d).\n",
820  sect_end, ics->max_sfb);
821  return -1;
822  }
823  } while (sect_len_incr == (1 << bits) - 1);
824  for (; k < sect_end; k++) {
825  band_type [idx] = sect_band_type;
826  band_type_run_end[idx++] = sect_end;
827  }
828  }
829  }
830  return 0;
831 }
832 
843 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
844  unsigned int global_gain,
846  enum BandType band_type[120],
847  int band_type_run_end[120])
848 {
849  int g, i, idx = 0;
850  int offset[3] = { global_gain, global_gain - 90, 0 };
851  int clipped_offset;
852  int noise_flag = 1;
853  static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
854  for (g = 0; g < ics->num_window_groups; g++) {
855  for (i = 0; i < ics->max_sfb;) {
856  int run_end = band_type_run_end[idx];
857  if (band_type[idx] == ZERO_BT) {
858  for (; i < run_end; i++, idx++)
859  sf[idx] = 0.;
860  } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
861  for (; i < run_end; i++, idx++) {
862  offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
863  clipped_offset = av_clip(offset[2], -155, 100);
864  if (offset[2] != clipped_offset) {
865  av_log_ask_for_sample(ac->avctx, "Intensity stereo "
866  "position clipped (%d -> %d).\nIf you heard an "
867  "audible artifact, there may be a bug in the "
868  "decoder. ", offset[2], clipped_offset);
869  }
870  sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
871  }
872  } else if (band_type[idx] == NOISE_BT) {
873  for (; i < run_end; i++, idx++) {
874  if (noise_flag-- > 0)
875  offset[1] += get_bits(gb, 9) - 256;
876  else
877  offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
878  clipped_offset = av_clip(offset[1], -100, 155);
879  if (offset[1] != clipped_offset) {
880  av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
881  "(%d -> %d).\nIf you heard an audible "
882  "artifact, there may be a bug in the decoder. ",
883  offset[1], clipped_offset);
884  }
885  sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
886  }
887  } else {
888  for (; i < run_end; i++, idx++) {
889  offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
890  if (offset[0] > 255U) {
892  "%s (%d) out of range.\n", sf_str[0], offset[0]);
893  return -1;
894  }
895  sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
896  }
897  }
898  }
899  }
900  return 0;
901 }
902 
906 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
907  const uint16_t *swb_offset, int num_swb)
908 {
909  int i, pulse_swb;
910  pulse->num_pulse = get_bits(gb, 2) + 1;
911  pulse_swb = get_bits(gb, 6);
912  if (pulse_swb >= num_swb)
913  return -1;
914  pulse->pos[0] = swb_offset[pulse_swb];
915  pulse->pos[0] += get_bits(gb, 5);
916  if (pulse->pos[0] > 1023)
917  return -1;
918  pulse->amp[0] = get_bits(gb, 4);
919  for (i = 1; i < pulse->num_pulse; i++) {
920  pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
921  if (pulse->pos[i] > 1023)
922  return -1;
923  pulse->amp[i] = get_bits(gb, 4);
924  }
925  return 0;
926 }
927 
934  GetBitContext *gb, const IndividualChannelStream *ics)
935 {
936  int w, filt, i, coef_len, coef_res, coef_compress;
937  const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
938  const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
939  for (w = 0; w < ics->num_windows; w++) {
940  if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
941  coef_res = get_bits1(gb);
942 
943  for (filt = 0; filt < tns->n_filt[w]; filt++) {
944  int tmp2_idx;
945  tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
946 
947  if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
948  av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
949  tns->order[w][filt], tns_max_order);
950  tns->order[w][filt] = 0;
951  return -1;
952  }
953  if (tns->order[w][filt]) {
954  tns->direction[w][filt] = get_bits1(gb);
955  coef_compress = get_bits1(gb);
956  coef_len = coef_res + 3 - coef_compress;
957  tmp2_idx = 2 * coef_compress + coef_res;
958 
959  for (i = 0; i < tns->order[w][filt]; i++)
960  tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
961  }
962  }
963  }
964  }
965  return 0;
966 }
967 
976  int ms_present)
977 {
978  int idx;
979  if (ms_present == 1) {
980  for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
981  cpe->ms_mask[idx] = get_bits1(gb);
982  } else if (ms_present == 2) {
983  memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
984  }
985 }
986 
987 #ifndef VMUL2
988 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
989  const float *scale)
990 {
991  float s = *scale;
992  *dst++ = v[idx & 15] * s;
993  *dst++ = v[idx>>4 & 15] * s;
994  return dst;
995 }
996 #endif
997 
998 #ifndef VMUL4
999 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1000  const float *scale)
1001 {
1002  float s = *scale;
1003  *dst++ = v[idx & 3] * s;
1004  *dst++ = v[idx>>2 & 3] * s;
1005  *dst++ = v[idx>>4 & 3] * s;
1006  *dst++ = v[idx>>6 & 3] * s;
1007  return dst;
1008 }
1009 #endif
1010 
1011 #ifndef VMUL2S
1012 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1013  unsigned sign, const float *scale)
1014 {
1015  union av_intfloat32 s0, s1;
1016 
1017  s0.f = s1.f = *scale;
1018  s0.i ^= sign >> 1 << 31;
1019  s1.i ^= sign << 31;
1020 
1021  *dst++ = v[idx & 15] * s0.f;
1022  *dst++ = v[idx>>4 & 15] * s1.f;
1023 
1024  return dst;
1025 }
1026 #endif
1027 
1028 #ifndef VMUL4S
1029 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1030  unsigned sign, const float *scale)
1031 {
1032  unsigned nz = idx >> 12;
1033  union av_intfloat32 s = { .f = *scale };
1034  union av_intfloat32 t;
1035 
1036  t.i = s.i ^ (sign & 1U<<31);
1037  *dst++ = v[idx & 3] * t.f;
1038 
1039  sign <<= nz & 1; nz >>= 1;
1040  t.i = s.i ^ (sign & 1U<<31);
1041  *dst++ = v[idx>>2 & 3] * t.f;
1042 
1043  sign <<= nz & 1; nz >>= 1;
1044  t.i = s.i ^ (sign & 1U<<31);
1045  *dst++ = v[idx>>4 & 3] * t.f;
1046 
1047  sign <<= nz & 1; nz >>= 1;
1048  t.i = s.i ^ (sign & 1U<<31);
1049  *dst++ = v[idx>>6 & 3] * t.f;
1050 
1051  return dst;
1052 }
1053 #endif
1054 
1067 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1068  GetBitContext *gb, const float sf[120],
1069  int pulse_present, const Pulse *pulse,
1070  const IndividualChannelStream *ics,
1071  enum BandType band_type[120])
1072 {
1073  int i, k, g, idx = 0;
1074  const int c = 1024 / ics->num_windows;
1075  const uint16_t *offsets = ics->swb_offset;
1076  float *coef_base = coef;
1077 
1078  for (g = 0; g < ics->num_windows; g++)
1079  memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1080 
1081  for (g = 0; g < ics->num_window_groups; g++) {
1082  unsigned g_len = ics->group_len[g];
1083 
1084  for (i = 0; i < ics->max_sfb; i++, idx++) {
1085  const unsigned cbt_m1 = band_type[idx] - 1;
1086  float *cfo = coef + offsets[i];
1087  int off_len = offsets[i + 1] - offsets[i];
1088  int group;
1089 
1090  if (cbt_m1 >= INTENSITY_BT2 - 1) {
1091  for (group = 0; group < g_len; group++, cfo+=128) {
1092  memset(cfo, 0, off_len * sizeof(float));
1093  }
1094  } else if (cbt_m1 == NOISE_BT - 1) {
1095  for (group = 0; group < g_len; group++, cfo+=128) {
1096  float scale;
1097  float band_energy;
1098 
1099  for (k = 0; k < off_len; k++) {
1101  cfo[k] = ac->random_state;
1102  }
1103 
1104  band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1105  scale = sf[idx] / sqrtf(band_energy);
1106  ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1107  }
1108  } else {
1109  const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1110  const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1111  VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1112  OPEN_READER(re, gb);
1113 
1114  switch (cbt_m1 >> 1) {
1115  case 0:
1116  for (group = 0; group < g_len; group++, cfo+=128) {
1117  float *cf = cfo;
1118  int len = off_len;
1119 
1120  do {
1121  int code;
1122  unsigned cb_idx;
1123 
1124  UPDATE_CACHE(re, gb);
1125  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1126  cb_idx = cb_vector_idx[code];
1127  cf = VMUL4(cf, vq, cb_idx, sf + idx);
1128  } while (len -= 4);
1129  }
1130  break;
1131 
1132  case 1:
1133  for (group = 0; group < g_len; group++, cfo+=128) {
1134  float *cf = cfo;
1135  int len = off_len;
1136 
1137  do {
1138  int code;
1139  unsigned nnz;
1140  unsigned cb_idx;
1141  uint32_t bits;
1142 
1143  UPDATE_CACHE(re, gb);
1144  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1145  cb_idx = cb_vector_idx[code];
1146  nnz = cb_idx >> 8 & 15;
1147  bits = nnz ? GET_CACHE(re, gb) : 0;
1148  LAST_SKIP_BITS(re, gb, nnz);
1149  cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1150  } while (len -= 4);
1151  }
1152  break;
1153 
1154  case 2:
1155  for (group = 0; group < g_len; group++, cfo+=128) {
1156  float *cf = cfo;
1157  int len = off_len;
1158 
1159  do {
1160  int code;
1161  unsigned cb_idx;
1162 
1163  UPDATE_CACHE(re, gb);
1164  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1165  cb_idx = cb_vector_idx[code];
1166  cf = VMUL2(cf, vq, cb_idx, sf + idx);
1167  } while (len -= 2);
1168  }
1169  break;
1170 
1171  case 3:
1172  case 4:
1173  for (group = 0; group < g_len; group++, cfo+=128) {
1174  float *cf = cfo;
1175  int len = off_len;
1176 
1177  do {
1178  int code;
1179  unsigned nnz;
1180  unsigned cb_idx;
1181  unsigned sign;
1182 
1183  UPDATE_CACHE(re, gb);
1184  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1185  cb_idx = cb_vector_idx[code];
1186  nnz = cb_idx >> 8 & 15;
1187  sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1188  LAST_SKIP_BITS(re, gb, nnz);
1189  cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1190  } while (len -= 2);
1191  }
1192  break;
1193 
1194  default:
1195  for (group = 0; group < g_len; group++, cfo+=128) {
1196  float *cf = cfo;
1197  uint32_t *icf = (uint32_t *) cf;
1198  int len = off_len;
1199 
1200  do {
1201  int code;
1202  unsigned nzt, nnz;
1203  unsigned cb_idx;
1204  uint32_t bits;
1205  int j;
1206 
1207  UPDATE_CACHE(re, gb);
1208  GET_VLC(code, re, gb, vlc_tab, 8, 2);
1209 
1210  if (!code) {
1211  *icf++ = 0;
1212  *icf++ = 0;
1213  continue;
1214  }
1215 
1216  cb_idx = cb_vector_idx[code];
1217  nnz = cb_idx >> 12;
1218  nzt = cb_idx >> 8;
1219  bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1220  LAST_SKIP_BITS(re, gb, nnz);
1221 
1222  for (j = 0; j < 2; j++) {
1223  if (nzt & 1<<j) {
1224  uint32_t b;
1225  int n;
1226  /* The total length of escape_sequence must be < 22 bits according
1227  to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1228  UPDATE_CACHE(re, gb);
1229  b = GET_CACHE(re, gb);
1230  b = 31 - av_log2(~b);
1231 
1232  if (b > 8) {
1233  av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1234  return -1;
1235  }
1236 
1237  SKIP_BITS(re, gb, b + 1);
1238  b += 4;
1239  n = (1 << b) + SHOW_UBITS(re, gb, b);
1240  LAST_SKIP_BITS(re, gb, b);
1241  *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1242  bits <<= 1;
1243  } else {
1244  unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1245  *icf++ = (bits & 1U<<31) | v;
1246  bits <<= !!v;
1247  }
1248  cb_idx >>= 4;
1249  }
1250  } while (len -= 2);
1251 
1252  ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1253  }
1254  }
1255 
1256  CLOSE_READER(re, gb);
1257  }
1258  }
1259  coef += g_len << 7;
1260  }
1261 
1262  if (pulse_present) {
1263  idx = 0;
1264  for (i = 0; i < pulse->num_pulse; i++) {
1265  float co = coef_base[ pulse->pos[i] ];
1266  while (offsets[idx + 1] <= pulse->pos[i])
1267  idx++;
1268  if (band_type[idx] != NOISE_BT && sf[idx]) {
1269  float ico = -pulse->amp[i];
1270  if (co) {
1271  co /= sf[idx];
1272  ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1273  }
1274  coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1275  }
1276  }
1277  }
1278  return 0;
1279 }
1280 
1281 static av_always_inline float flt16_round(float pf)
1282 {
1283  union av_intfloat32 tmp;
1284  tmp.f = pf;
1285  tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1286  return tmp.f;
1287 }
1288 
1289 static av_always_inline float flt16_even(float pf)
1290 {
1291  union av_intfloat32 tmp;
1292  tmp.f = pf;
1293  tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1294  return tmp.f;
1295 }
1296 
1297 static av_always_inline float flt16_trunc(float pf)
1298 {
1299  union av_intfloat32 pun;
1300  pun.f = pf;
1301  pun.i &= 0xFFFF0000U;
1302  return pun.f;
1303 }
1304 
1305 static av_always_inline void predict(PredictorState *ps, float *coef,
1306  int output_enable)
1307 {
1308  const float a = 0.953125; // 61.0 / 64
1309  const float alpha = 0.90625; // 29.0 / 32
1310  float e0, e1;
1311  float pv;
1312  float k1, k2;
1313  float r0 = ps->r0, r1 = ps->r1;
1314  float cor0 = ps->cor0, cor1 = ps->cor1;
1315  float var0 = ps->var0, var1 = ps->var1;
1316 
1317  k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1318  k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1319 
1320  pv = flt16_round(k1 * r0 + k2 * r1);
1321  if (output_enable)
1322  *coef += pv;
1323 
1324  e0 = *coef;
1325  e1 = e0 - k1 * r0;
1326 
1327  ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1328  ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1329  ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1330  ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1331 
1332  ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1333  ps->r0 = flt16_trunc(a * e0);
1334 }
1335 
1340 {
1341  int sfb, k;
1342 
1343  if (!sce->ics.predictor_initialized) {
1345  sce->ics.predictor_initialized = 1;
1346  }
1347 
1348  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1349  for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1350  for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1351  predict(&sce->predictor_state[k], &sce->coeffs[k],
1352  sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1353  }
1354  }
1355  if (sce->ics.predictor_reset_group)
1357  } else
1359 }
1360 
1370  GetBitContext *gb, int common_window, int scale_flag)
1371 {
1372  Pulse pulse;
1373  TemporalNoiseShaping *tns = &sce->tns;
1374  IndividualChannelStream *ics = &sce->ics;
1375  float *out = sce->coeffs;
1376  int global_gain, pulse_present = 0;
1377 
1378  /* This assignment is to silence a GCC warning about the variable being used
1379  * uninitialized when in fact it always is.
1380  */
1381  pulse.num_pulse = 0;
1382 
1383  global_gain = get_bits(gb, 8);
1384 
1385  if (!common_window && !scale_flag) {
1386  if (decode_ics_info(ac, ics, gb) < 0)
1387  return AVERROR_INVALIDDATA;
1388  }
1389 
1390  if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1391  return -1;
1392  if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1393  return -1;
1394 
1395  pulse_present = 0;
1396  if (!scale_flag) {
1397  if ((pulse_present = get_bits1(gb))) {
1398  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1399  av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1400  return -1;
1401  }
1402  if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1403  av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1404  return -1;
1405  }
1406  }
1407  if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1408  return -1;
1409  if (get_bits1(gb)) {
1410  av_log_missing_feature(ac->avctx, "SSR", 1);
1411  return -1;
1412  }
1413  }
1414 
1415  if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1416  return -1;
1417 
1418  if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1419  apply_prediction(ac, sce);
1420 
1421  return 0;
1422 }
1423 
1428 {
1429  const IndividualChannelStream *ics = &cpe->ch[0].ics;
1430  float *ch0 = cpe->ch[0].coeffs;
1431  float *ch1 = cpe->ch[1].coeffs;
1432  int g, i, group, idx = 0;
1433  const uint16_t *offsets = ics->swb_offset;
1434  for (g = 0; g < ics->num_window_groups; g++) {
1435  for (i = 0; i < ics->max_sfb; i++, idx++) {
1436  if (cpe->ms_mask[idx] &&
1437  cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1438  for (group = 0; group < ics->group_len[g]; group++) {
1439  ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1440  ch1 + group * 128 + offsets[i],
1441  offsets[i+1] - offsets[i]);
1442  }
1443  }
1444  }
1445  ch0 += ics->group_len[g] * 128;
1446  ch1 += ics->group_len[g] * 128;
1447  }
1448 }
1449 
1457 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1458 {
1459  const IndividualChannelStream *ics = &cpe->ch[1].ics;
1460  SingleChannelElement *sce1 = &cpe->ch[1];
1461  float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1462  const uint16_t *offsets = ics->swb_offset;
1463  int g, group, i, idx = 0;
1464  int c;
1465  float scale;
1466  for (g = 0; g < ics->num_window_groups; g++) {
1467  for (i = 0; i < ics->max_sfb;) {
1468  if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1469  const int bt_run_end = sce1->band_type_run_end[idx];
1470  for (; i < bt_run_end; i++, idx++) {
1471  c = -1 + 2 * (sce1->band_type[idx] - 14);
1472  if (ms_present)
1473  c *= 1 - 2 * cpe->ms_mask[idx];
1474  scale = c * sce1->sf[idx];
1475  for (group = 0; group < ics->group_len[g]; group++)
1476  ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1477  coef0 + group * 128 + offsets[i],
1478  scale,
1479  offsets[i + 1] - offsets[i]);
1480  }
1481  } else {
1482  int bt_run_end = sce1->band_type_run_end[idx];
1483  idx += bt_run_end - i;
1484  i = bt_run_end;
1485  }
1486  }
1487  coef0 += ics->group_len[g] * 128;
1488  coef1 += ics->group_len[g] * 128;
1489  }
1490 }
1491 
1498 {
1499  int i, ret, common_window, ms_present = 0;
1500 
1501  common_window = get_bits1(gb);
1502  if (common_window) {
1503  if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1504  return AVERROR_INVALIDDATA;
1505  i = cpe->ch[1].ics.use_kb_window[0];
1506  cpe->ch[1].ics = cpe->ch[0].ics;
1507  cpe->ch[1].ics.use_kb_window[1] = i;
1508  if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1509  if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1510  decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1511  ms_present = get_bits(gb, 2);
1512  if (ms_present == 3) {
1513  av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1514  return -1;
1515  } else if (ms_present)
1516  decode_mid_side_stereo(cpe, gb, ms_present);
1517  }
1518  if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1519  return ret;
1520  if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1521  return ret;
1522 
1523  if (common_window) {
1524  if (ms_present)
1525  apply_mid_side_stereo(ac, cpe);
1526  if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1527  apply_prediction(ac, &cpe->ch[0]);
1528  apply_prediction(ac, &cpe->ch[1]);
1529  }
1530  }
1531 
1532  apply_intensity_stereo(ac, cpe, ms_present);
1533  return 0;
1534 }
1535 
1536 static const float cce_scale[] = {
1537  1.09050773266525765921, //2^(1/8)
1538  1.18920711500272106672, //2^(1/4)
1539  M_SQRT2,
1540  2,
1541 };
1542 
1549 {
1550  int num_gain = 0;
1551  int c, g, sfb, ret;
1552  int sign;
1553  float scale;
1554  SingleChannelElement *sce = &che->ch[0];
1555  ChannelCoupling *coup = &che->coup;
1556 
1557  coup->coupling_point = 2 * get_bits1(gb);
1558  coup->num_coupled = get_bits(gb, 3);
1559  for (c = 0; c <= coup->num_coupled; c++) {
1560  num_gain++;
1561  coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1562  coup->id_select[c] = get_bits(gb, 4);
1563  if (coup->type[c] == TYPE_CPE) {
1564  coup->ch_select[c] = get_bits(gb, 2);
1565  if (coup->ch_select[c] == 3)
1566  num_gain++;
1567  } else
1568  coup->ch_select[c] = 2;
1569  }
1570  coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1571 
1572  sign = get_bits(gb, 1);
1573  scale = cce_scale[get_bits(gb, 2)];
1574 
1575  if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1576  return ret;
1577 
1578  for (c = 0; c < num_gain; c++) {
1579  int idx = 0;
1580  int cge = 1;
1581  int gain = 0;
1582  float gain_cache = 1.;
1583  if (c) {
1584  cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1585  gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1586  gain_cache = powf(scale, -gain);
1587  }
1588  if (coup->coupling_point == AFTER_IMDCT) {
1589  coup->gain[c][0] = gain_cache;
1590  } else {
1591  for (g = 0; g < sce->ics.num_window_groups; g++) {
1592  for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1593  if (sce->band_type[idx] != ZERO_BT) {
1594  if (!cge) {
1595  int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1596  if (t) {
1597  int s = 1;
1598  t = gain += t;
1599  if (sign) {
1600  s -= 2 * (t & 0x1);
1601  t >>= 1;
1602  }
1603  gain_cache = powf(scale, -t) * s;
1604  }
1605  }
1606  coup->gain[c][idx] = gain_cache;
1607  }
1608  }
1609  }
1610  }
1611  }
1612  return 0;
1613 }
1614 
1621  GetBitContext *gb)
1622 {
1623  int i;
1624  int num_excl_chan = 0;
1625 
1626  do {
1627  for (i = 0; i < 7; i++)
1628  che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1629  } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1630 
1631  return num_excl_chan / 7;
1632 }
1633 
1642  GetBitContext *gb, int cnt)
1643 {
1644  int n = 1;
1645  int drc_num_bands = 1;
1646  int i;
1647 
1648  /* pce_tag_present? */
1649  if (get_bits1(gb)) {
1650  che_drc->pce_instance_tag = get_bits(gb, 4);
1651  skip_bits(gb, 4); // tag_reserved_bits
1652  n++;
1653  }
1654 
1655  /* excluded_chns_present? */
1656  if (get_bits1(gb)) {
1657  n += decode_drc_channel_exclusions(che_drc, gb);
1658  }
1659 
1660  /* drc_bands_present? */
1661  if (get_bits1(gb)) {
1662  che_drc->band_incr = get_bits(gb, 4);
1663  che_drc->interpolation_scheme = get_bits(gb, 4);
1664  n++;
1665  drc_num_bands += che_drc->band_incr;
1666  for (i = 0; i < drc_num_bands; i++) {
1667  che_drc->band_top[i] = get_bits(gb, 8);
1668  n++;
1669  }
1670  }
1671 
1672  /* prog_ref_level_present? */
1673  if (get_bits1(gb)) {
1674  che_drc->prog_ref_level = get_bits(gb, 7);
1675  skip_bits1(gb); // prog_ref_level_reserved_bits
1676  n++;
1677  }
1678 
1679  for (i = 0; i < drc_num_bands; i++) {
1680  che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1681  che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1682  n++;
1683  }
1684 
1685  return n;
1686 }
1687 
1696  ChannelElement *che, enum RawDataBlockType elem_type)
1697 {
1698  int crc_flag = 0;
1699  int res = cnt;
1700  switch (get_bits(gb, 4)) { // extension type
1701  case EXT_SBR_DATA_CRC:
1702  crc_flag++;
1703  case EXT_SBR_DATA:
1704  if (!che) {
1705  av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1706  return res;
1707  } else if (!ac->m4ac.sbr) {
1708  av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1709  skip_bits_long(gb, 8 * cnt - 4);
1710  return res;
1711  } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1712  av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1713  skip_bits_long(gb, 8 * cnt - 4);
1714  return res;
1715  } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1716  ac->m4ac.sbr = 1;
1717  ac->m4ac.ps = 1;
1719  } else {
1720  ac->m4ac.sbr = 1;
1721  }
1722  res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1723  break;
1724  case EXT_DYNAMIC_RANGE:
1725  res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1726  break;
1727  case EXT_FILL:
1728  case EXT_FILL_DATA:
1729  case EXT_DATA_ELEMENT:
1730  default:
1731  skip_bits_long(gb, 8 * cnt - 4);
1732  break;
1733  };
1734  return res;
1735 }
1736 
1743 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1744  IndividualChannelStream *ics, int decode)
1745 {
1746  const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1747  int w, filt, m, i;
1748  int bottom, top, order, start, end, size, inc;
1749  float lpc[TNS_MAX_ORDER];
1750  float tmp[TNS_MAX_ORDER + 1];
1751 
1752  for (w = 0; w < ics->num_windows; w++) {
1753  bottom = ics->num_swb;
1754  for (filt = 0; filt < tns->n_filt[w]; filt++) {
1755  top = bottom;
1756  bottom = FFMAX(0, top - tns->length[w][filt]);
1757  order = tns->order[w][filt];
1758  if (order == 0)
1759  continue;
1760 
1761  // tns_decode_coef
1762  compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1763 
1764  start = ics->swb_offset[FFMIN(bottom, mmm)];
1765  end = ics->swb_offset[FFMIN( top, mmm)];
1766  if ((size = end - start) <= 0)
1767  continue;
1768  if (tns->direction[w][filt]) {
1769  inc = -1;
1770  start = end - 1;
1771  } else {
1772  inc = 1;
1773  }
1774  start += w * 128;
1775 
1776  if (decode) {
1777  // ar filter
1778  for (m = 0; m < size; m++, start += inc)
1779  for (i = 1; i <= FFMIN(m, order); i++)
1780  coef[start] -= coef[start - i * inc] * lpc[i - 1];
1781  } else {
1782  // ma filter
1783  for (m = 0; m < size; m++, start += inc) {
1784  tmp[0] = coef[start];
1785  for (i = 1; i <= FFMIN(m, order); i++)
1786  coef[start] += tmp[i] * lpc[i - 1];
1787  for (i = order; i > 0; i--)
1788  tmp[i] = tmp[i - 1];
1789  }
1790  }
1791  }
1792  }
1793 }
1794 
1799 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1800  float *in, IndividualChannelStream *ics)
1801 {
1802  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1803  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1804  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1805  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1806 
1807  if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1808  ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1809  } else {
1810  memset(in, 0, 448 * sizeof(float));
1811  ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1812  }
1813  if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1814  ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1815  } else {
1816  ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1817  memset(in + 1024 + 576, 0, 448 * sizeof(float));
1818  }
1819  ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1820 }
1821 
1826 {
1827  const LongTermPrediction *ltp = &sce->ics.ltp;
1828  const uint16_t *offsets = sce->ics.swb_offset;
1829  int i, sfb;
1830 
1831  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1832  float *predTime = sce->ret;
1833  float *predFreq = ac->buf_mdct;
1834  int16_t num_samples = 2048;
1835 
1836  if (ltp->lag < 1024)
1837  num_samples = ltp->lag + 1024;
1838  for (i = 0; i < num_samples; i++)
1839  predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1840  memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1841 
1842  windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1843 
1844  if (sce->tns.present)
1845  apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1846 
1847  for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1848  if (ltp->used[sfb])
1849  for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1850  sce->coeffs[i] += predFreq[i];
1851  }
1852 }
1853 
1858 {
1859  IndividualChannelStream *ics = &sce->ics;
1860  float *saved = sce->saved;
1861  float *saved_ltp = sce->coeffs;
1862  const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1863  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1864  int i;
1865 
1866  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1867  memcpy(saved_ltp, saved, 512 * sizeof(float));
1868  memset(saved_ltp + 576, 0, 448 * sizeof(float));
1869  ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1870  for (i = 0; i < 64; i++)
1871  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1872  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1873  memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1874  memset(saved_ltp + 576, 0, 448 * sizeof(float));
1875  ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1876  for (i = 0; i < 64; i++)
1877  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1878  } else { // LONG_STOP or ONLY_LONG
1879  ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1880  for (i = 0; i < 512; i++)
1881  saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1882  }
1883 
1884  memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1885  memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1886  memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1887 }
1888 
1893 {
1894  IndividualChannelStream *ics = &sce->ics;
1895  float *in = sce->coeffs;
1896  float *out = sce->ret;
1897  float *saved = sce->saved;
1898  const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1899  const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1900  const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1901  float *buf = ac->buf_mdct;
1902  float *temp = ac->temp;
1903  int i;
1904 
1905  // imdct
1906  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1907  for (i = 0; i < 1024; i += 128)
1908  ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1909  } else
1910  ac->mdct.imdct_half(&ac->mdct, buf, in);
1911 
1912  /* window overlapping
1913  * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1914  * and long to short transitions are considered to be short to short
1915  * transitions. This leaves just two cases (long to long and short to short)
1916  * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1917  */
1918  if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1920  ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1921  } else {
1922  memcpy( out, saved, 448 * sizeof(float));
1923 
1924  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1925  ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1926  ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1927  ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1928  ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1929  ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1930  memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1931  } else {
1932  ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1933  memcpy( out + 576, buf + 64, 448 * sizeof(float));
1934  }
1935  }
1936 
1937  // buffer update
1938  if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1939  memcpy( saved, temp + 64, 64 * sizeof(float));
1940  ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1941  ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1942  ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1943  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1944  } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1945  memcpy( saved, buf + 512, 448 * sizeof(float));
1946  memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1947  } else { // LONG_STOP or ONLY_LONG
1948  memcpy( saved, buf + 512, 512 * sizeof(float));
1949  }
1950 }
1951 
1958  SingleChannelElement *target,
1959  ChannelElement *cce, int index)
1960 {
1961  IndividualChannelStream *ics = &cce->ch[0].ics;
1962  const uint16_t *offsets = ics->swb_offset;
1963  float *dest = target->coeffs;
1964  const float *src = cce->ch[0].coeffs;
1965  int g, i, group, k, idx = 0;
1966  if (ac->m4ac.object_type == AOT_AAC_LTP) {
1967  av_log(ac->avctx, AV_LOG_ERROR,
1968  "Dependent coupling is not supported together with LTP\n");
1969  return;
1970  }
1971  for (g = 0; g < ics->num_window_groups; g++) {
1972  for (i = 0; i < ics->max_sfb; i++, idx++) {
1973  if (cce->ch[0].band_type[idx] != ZERO_BT) {
1974  const float gain = cce->coup.gain[index][idx];
1975  for (group = 0; group < ics->group_len[g]; group++) {
1976  for (k = offsets[i]; k < offsets[i + 1]; k++) {
1977  // XXX dsputil-ize
1978  dest[group * 128 + k] += gain * src[group * 128 + k];
1979  }
1980  }
1981  }
1982  }
1983  dest += ics->group_len[g] * 128;
1984  src += ics->group_len[g] * 128;
1985  }
1986 }
1987 
1994  SingleChannelElement *target,
1995  ChannelElement *cce, int index)
1996 {
1997  int i;
1998  const float gain = cce->coup.gain[index][0];
1999  const float *src = cce->ch[0].ret;
2000  float *dest = target->ret;
2001  const int len = 1024 << (ac->m4ac.sbr == 1);
2002 
2003  for (i = 0; i < len; i++)
2004  dest[i] += gain * src[i];
2005 }
2006 
2013  enum RawDataBlockType type, int elem_id,
2014  enum CouplingPoint coupling_point,
2015  void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2016 {
2017  int i, c;
2018 
2019  for (i = 0; i < MAX_ELEM_ID; i++) {
2020  ChannelElement *cce = ac->che[TYPE_CCE][i];
2021  int index = 0;
2022 
2023  if (cce && cce->coup.coupling_point == coupling_point) {
2024  ChannelCoupling *coup = &cce->coup;
2025 
2026  for (c = 0; c <= coup->num_coupled; c++) {
2027  if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2028  if (coup->ch_select[c] != 1) {
2029  apply_coupling_method(ac, &cc->ch[0], cce, index);
2030  if (coup->ch_select[c] != 0)
2031  index++;
2032  }
2033  if (coup->ch_select[c] != 2)
2034  apply_coupling_method(ac, &cc->ch[1], cce, index++);
2035  } else
2036  index += 1 + (coup->ch_select[c] == 3);
2037  }
2038  }
2039  }
2040 }
2041 
2046 {
2047  int i, type;
2048  for (type = 3; type >= 0; type--) {
2049  for (i = 0; i < MAX_ELEM_ID; i++) {
2050  ChannelElement *che = ac->che[type][i];
2051  if (che) {
2052  if (type <= TYPE_CPE)
2054  if (ac->m4ac.object_type == AOT_AAC_LTP) {
2055  if (che->ch[0].ics.predictor_present) {
2056  if (che->ch[0].ics.ltp.present)
2057  apply_ltp(ac, &che->ch[0]);
2058  if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2059  apply_ltp(ac, &che->ch[1]);
2060  }
2061  }
2062  if (che->ch[0].tns.present)
2063  apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2064  if (che->ch[1].tns.present)
2065  apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2066  if (type <= TYPE_CPE)
2068  if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2069  imdct_and_windowing(ac, &che->ch[0]);
2070  if (ac->m4ac.object_type == AOT_AAC_LTP)
2071  update_ltp(ac, &che->ch[0]);
2072  if (type == TYPE_CPE) {
2073  imdct_and_windowing(ac, &che->ch[1]);
2074  if (ac->m4ac.object_type == AOT_AAC_LTP)
2075  update_ltp(ac, &che->ch[1]);
2076  }
2077  if (ac->m4ac.sbr > 0) {
2078  ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2079  }
2080  }
2081  if (type <= TYPE_CCE)
2083  }
2084  }
2085  }
2086 }
2087 
2089 {
2090  int size;
2091  AACADTSHeaderInfo hdr_info;
2092 
2093  size = avpriv_aac_parse_header(gb, &hdr_info);
2094  if (size > 0) {
2095  if (hdr_info.chan_config) {
2096  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2097  memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2098  ac->m4ac.chan_config = hdr_info.chan_config;
2099  if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2100  return -7;
2101  if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config,
2103  return -7;
2104  } else if (ac->output_configured != OC_LOCKED) {
2105  ac->m4ac.chan_config = 0;
2106  ac->output_configured = OC_NONE;
2107  }
2108  if (ac->output_configured != OC_LOCKED) {
2109  ac->m4ac.sbr = -1;
2110  ac->m4ac.ps = -1;
2111  ac->m4ac.sample_rate = hdr_info.sample_rate;
2112  ac->m4ac.sampling_index = hdr_info.sampling_index;
2113  ac->m4ac.object_type = hdr_info.object_type;
2114  }
2115  if (!ac->avctx->sample_rate)
2116  ac->avctx->sample_rate = hdr_info.sample_rate;
2117  if (hdr_info.num_aac_frames == 1) {
2118  if (!hdr_info.crc_absent)
2119  skip_bits(gb, 16);
2120  } else {
2121  av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2122  return -1;
2123  }
2124  }
2125  return size;
2126 }
2127 
2128 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2129  int *got_frame_ptr, GetBitContext *gb)
2130 {
2131  AACContext *ac = avctx->priv_data;
2132  ChannelElement *che = NULL, *che_prev = NULL;
2133  enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2134  int err, elem_id;
2135  int samples = 0, multiplier, audio_found = 0;
2136 
2137  if (show_bits(gb, 12) == 0xfff) {
2138  if (parse_adts_frame_header(ac, gb) < 0) {
2139  av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2140  return -1;
2141  }
2142  if (ac->m4ac.sampling_index > 12) {
2143  av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2144  return -1;
2145  }
2146  }
2147 
2148  ac->tags_mapped = 0;
2149  // parse
2150  while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2151  elem_id = get_bits(gb, 4);
2152 
2153  if (elem_type < TYPE_DSE) {
2154  if (!(che=get_che(ac, elem_type, elem_id))) {
2155  av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2156  elem_type, elem_id);
2157  return -1;
2158  }
2159  samples = 1024;
2160  }
2161 
2162  switch (elem_type) {
2163 
2164  case TYPE_SCE:
2165  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2166  audio_found = 1;
2167  break;
2168 
2169  case TYPE_CPE:
2170  err = decode_cpe(ac, gb, che);
2171  audio_found = 1;
2172  break;
2173 
2174  case TYPE_CCE:
2175  err = decode_cce(ac, gb, che);
2176  break;
2177 
2178  case TYPE_LFE:
2179  err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2180  audio_found = 1;
2181  break;
2182 
2183  case TYPE_DSE:
2184  err = skip_data_stream_element(ac, gb);
2185  break;
2186 
2187  case TYPE_PCE: {
2188  enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2189  memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2190  if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2191  break;
2192  if (ac->output_configured > OC_TRIAL_PCE)
2193  av_log(avctx, AV_LOG_ERROR,
2194  "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2195  else
2196  err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2197  break;
2198  }
2199 
2200  case TYPE_FIL:
2201  if (elem_id == 15)
2202  elem_id += get_bits(gb, 8) - 1;
2203  if (get_bits_left(gb) < 8 * elem_id) {
2204  av_log(avctx, AV_LOG_ERROR, overread_err);
2205  return -1;
2206  }
2207  while (elem_id > 0)
2208  elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2209  err = 0; /* FIXME */
2210  break;
2211 
2212  default:
2213  err = -1; /* should not happen, but keeps compiler happy */
2214  break;
2215  }
2216 
2217  che_prev = che;
2218  elem_type_prev = elem_type;
2219 
2220  if (err)
2221  return err;
2222 
2223  if (get_bits_left(gb) < 3) {
2224  av_log(avctx, AV_LOG_ERROR, overread_err);
2225  return -1;
2226  }
2227  }
2228 
2229  spectral_to_sample(ac);
2230 
2231  multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2232  samples <<= multiplier;
2233  if (ac->output_configured < OC_LOCKED) {
2234  avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2235  avctx->frame_size = samples;
2236  }
2237 
2238  if (samples) {
2239  /* get output buffer */
2240  ac->frame.nb_samples = samples;
2241  if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2242  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2243  return err;
2244  }
2245 
2246  if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2247  ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2248  (const float **)ac->output_data,
2249  samples, avctx->channels);
2250  else
2251  ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2252  (const float **)ac->output_data,
2253  samples, avctx->channels);
2254 
2255  *(AVFrame *)data = ac->frame;
2256  }
2257  *got_frame_ptr = !!samples;
2258 
2259  if (ac->output_configured && audio_found)
2261 
2262  return 0;
2263 }
2264 
2265 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2266  int *got_frame_ptr, AVPacket *avpkt)
2267 {
2268  AACContext *ac = avctx->priv_data;
2269  const uint8_t *buf = avpkt->data;
2270  int buf_size = avpkt->size;
2271  GetBitContext gb;
2272  int buf_consumed;
2273  int buf_offset;
2274  int err;
2275  int new_extradata_size;
2276  const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2278  &new_extradata_size);
2279 
2280  if (new_extradata) {
2281  av_free(avctx->extradata);
2282  avctx->extradata = av_mallocz(new_extradata_size +
2284  if (!avctx->extradata)
2285  return AVERROR(ENOMEM);
2286  avctx->extradata_size = new_extradata_size;
2287  memcpy(avctx->extradata, new_extradata, new_extradata_size);
2288  if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2289  avctx->extradata,
2290  avctx->extradata_size*8, 1) < 0)
2291  return AVERROR_INVALIDDATA;
2292  }
2293 
2294  init_get_bits(&gb, buf, buf_size * 8);
2295 
2296  if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2297  return err;
2298 
2299  buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2300  for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2301  if (buf[buf_offset])
2302  break;
2303 
2304  return buf_size > buf_offset ? buf_consumed : buf_size;
2305 }
2306 
2308 {
2309  AACContext *ac = avctx->priv_data;
2310  int i, type;
2311 
2312  for (i = 0; i < MAX_ELEM_ID; i++) {
2313  for (type = 0; type < 4; type++) {
2314  if (ac->che[type][i])
2315  ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2316  av_freep(&ac->che[type][i]);
2317  }
2318  }
2319 
2320  ff_mdct_end(&ac->mdct);
2321  ff_mdct_end(&ac->mdct_small);
2322  ff_mdct_end(&ac->mdct_ltp);
2323  return 0;
2324 }
2325 
2326 
2327 #define LOAS_SYNC_WORD 0x2b7
2328 
2329 struct LATMContext {
2332 
2333  // parser data
2337 };
2338 
2339 static inline uint32_t latm_get_value(GetBitContext *b)
2340 {
2341  int length = get_bits(b, 2);
2342 
2343  return get_bits_long(b, (length+1)*8);
2344 }
2345 
2347  GetBitContext *gb, int asclen)
2348 {
2349  AACContext *ac = &latmctx->aac_ctx;
2350  AVCodecContext *avctx = ac->avctx;
2351  MPEG4AudioConfig m4ac = {0};
2352  int config_start_bit = get_bits_count(gb);
2353  int sync_extension = 0;
2354  int bits_consumed, esize;
2355 
2356  if (asclen) {
2357  sync_extension = 1;
2358  asclen = FFMIN(asclen, get_bits_left(gb));
2359  } else
2360  asclen = get_bits_left(gb);
2361 
2362  if (config_start_bit % 8) {
2363  av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2364  "config not byte aligned.\n", 1);
2365  return AVERROR_INVALIDDATA;
2366  }
2367  if (asclen <= 0)
2368  return AVERROR_INVALIDDATA;
2369  bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2370  gb->buffer + (config_start_bit / 8),
2371  asclen, sync_extension);
2372 
2373  if (bits_consumed < 0)
2374  return AVERROR_INVALIDDATA;
2375 
2376  if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2377  ac->m4ac.chan_config != m4ac.chan_config) {
2378 
2379  av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2380  latmctx->initialized = 0;
2381 
2382  esize = (bits_consumed+7) / 8;
2383 
2384  if (avctx->extradata_size < esize) {
2385  av_free(avctx->extradata);
2387  if (!avctx->extradata)
2388  return AVERROR(ENOMEM);
2389  }
2390 
2391  avctx->extradata_size = esize;
2392  memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2393  memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2394  }
2395  skip_bits_long(gb, bits_consumed);
2396 
2397  return bits_consumed;
2398 }
2399 
2400 static int read_stream_mux_config(struct LATMContext *latmctx,
2401  GetBitContext *gb)
2402 {
2403  int ret, audio_mux_version = get_bits(gb, 1);
2404 
2405  latmctx->audio_mux_version_A = 0;
2406  if (audio_mux_version)
2407  latmctx->audio_mux_version_A = get_bits(gb, 1);
2408 
2409  if (!latmctx->audio_mux_version_A) {
2410 
2411  if (audio_mux_version)
2412  latm_get_value(gb); // taraFullness
2413 
2414  skip_bits(gb, 1); // allStreamSameTimeFraming
2415  skip_bits(gb, 6); // numSubFrames
2416  // numPrograms
2417  if (get_bits(gb, 4)) { // numPrograms
2419  "multiple programs are not supported\n", 1);
2420  return AVERROR_PATCHWELCOME;
2421  }
2422 
2423  // for each program (which there is only on in DVB)
2424 
2425  // for each layer (which there is only on in DVB)
2426  if (get_bits(gb, 3)) { // numLayer
2428  "multiple layers are not supported\n", 1);
2429  return AVERROR_PATCHWELCOME;
2430  }
2431 
2432  // for all but first stream: use_same_config = get_bits(gb, 1);
2433  if (!audio_mux_version) {
2434  if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2435  return ret;
2436  } else {
2437  int ascLen = latm_get_value(gb);
2438  if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2439  return ret;
2440  ascLen -= ret;
2441  skip_bits_long(gb, ascLen);
2442  }
2443 
2444  latmctx->frame_length_type = get_bits(gb, 3);
2445  switch (latmctx->frame_length_type) {
2446  case 0:
2447  skip_bits(gb, 8); // latmBufferFullness
2448  break;
2449  case 1:
2450  latmctx->frame_length = get_bits(gb, 9);
2451  break;
2452  case 3:
2453  case 4:
2454  case 5:
2455  skip_bits(gb, 6); // CELP frame length table index
2456  break;
2457  case 6:
2458  case 7:
2459  skip_bits(gb, 1); // HVXC frame length table index
2460  break;
2461  }
2462 
2463  if (get_bits(gb, 1)) { // other data
2464  if (audio_mux_version) {
2465  latm_get_value(gb); // other_data_bits
2466  } else {
2467  int esc;
2468  do {
2469  esc = get_bits(gb, 1);
2470  skip_bits(gb, 8);
2471  } while (esc);
2472  }
2473  }
2474 
2475  if (get_bits(gb, 1)) // crc present
2476  skip_bits(gb, 8); // config_crc
2477  }
2478 
2479  return 0;
2480 }
2481 
2483 {
2484  uint8_t tmp;
2485 
2486  if (ctx->frame_length_type == 0) {
2487  int mux_slot_length = 0;
2488  do {
2489  tmp = get_bits(gb, 8);
2490  mux_slot_length += tmp;
2491  } while (tmp == 255);
2492  return mux_slot_length;
2493  } else if (ctx->frame_length_type == 1) {
2494  return ctx->frame_length;
2495  } else if (ctx->frame_length_type == 3 ||
2496  ctx->frame_length_type == 5 ||
2497  ctx->frame_length_type == 7) {
2498  skip_bits(gb, 2); // mux_slot_length_coded
2499  }
2500  return 0;
2501 }
2502 
2503 static int read_audio_mux_element(struct LATMContext *latmctx,
2504  GetBitContext *gb)
2505 {
2506  int err;
2507  uint8_t use_same_mux = get_bits(gb, 1);
2508  if (!use_same_mux) {
2509  if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2510  return err;
2511  } else if (!latmctx->aac_ctx.avctx->extradata) {
2512  av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2513  "no decoder config found\n");
2514  return AVERROR(EAGAIN);
2515  }
2516  if (latmctx->audio_mux_version_A == 0) {
2517  int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2518  if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2519  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2520  return AVERROR_INVALIDDATA;
2521  } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2522  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2523  "frame length mismatch %d << %d\n",
2524  mux_slot_length_bytes * 8, get_bits_left(gb));
2525  return AVERROR_INVALIDDATA;
2526  }
2527  }
2528  return 0;
2529 }
2530 
2531 
2532 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2533  int *got_frame_ptr, AVPacket *avpkt)
2534 {
2535  struct LATMContext *latmctx = avctx->priv_data;
2536  int muxlength, err;
2537  GetBitContext gb;
2538 
2539  init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2540 
2541  // check for LOAS sync word
2542  if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2543  return AVERROR_INVALIDDATA;
2544 
2545  muxlength = get_bits(&gb, 13) + 3;
2546  // not enough data, the parser should have sorted this
2547  if (muxlength > avpkt->size)
2548  return AVERROR_INVALIDDATA;
2549 
2550  if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2551  return err;
2552 
2553  if (!latmctx->initialized) {
2554  if (!avctx->extradata) {
2555  *got_frame_ptr = 0;
2556  return avpkt->size;
2557  } else {
2558  if ((err = decode_audio_specific_config(
2559  &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2560  avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2561  return err;
2562  latmctx->initialized = 1;
2563  }
2564  }
2565 
2566  if (show_bits(&gb, 12) == 0xfff) {
2567  av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2568  "ADTS header detected, probably as result of configuration "
2569  "misparsing\n");
2570  return AVERROR_INVALIDDATA;
2571  }
2572 
2573  if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2574  return err;
2575 
2576  return muxlength;
2577 }
2578 
2580 {
2581  struct LATMContext *latmctx = avctx->priv_data;
2582  int ret = aac_decode_init(avctx);
2583 
2584  if (avctx->extradata_size > 0)
2585  latmctx->initialized = !ret;
2586 
2587  return ret;
2588 }
2589 
2590 
2592  .name = "aac",
2593  .type = AVMEDIA_TYPE_AUDIO,
2594  .id = CODEC_ID_AAC,
2595  .priv_data_size = sizeof(AACContext),
2596  .init = aac_decode_init,
2599  .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2600  .sample_fmts = (const enum AVSampleFormat[]) {
2602  },
2603  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2604  .channel_layouts = aac_channel_layout,
2605 };
2606 
2607 /*
2608  Note: This decoder filter is intended to decode LATM streams transferred
2609  in MPEG transport streams which only contain one program.
2610  To do a more complex LATM demuxing a separate LATM demuxer should be used.
2611 */
2613  .name = "aac_latm",
2614  .type = AVMEDIA_TYPE_AUDIO,
2615  .id = CODEC_ID_AAC_LATM,
2616  .priv_data_size = sizeof(struct LATMContext),
2617  .init = latm_decode_init,
2618  .close = aac_decode_close,
2619  .decode = latm_decode_frame,
2620  .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2621  .sample_fmts = (const enum AVSampleFormat[]) {
2623  },
2624  .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2625  .channel_layouts = aac_channel_layout,
2626 };