aacdec: remove sf_scale and sf_offset.
[libav.git] / libavcodec / aacdec.c
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
27 /**
28 * @file
29 * AAC decoder
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
32 */
33
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
102 #include <assert.h>
103 #include <errno.h>
104 #include <math.h>
105 #include <string.h>
106
107 #if ARCH_ARM
108 # include "arm/aac.h"
109 #endif
110
111 union float754 {
112 float f;
113 uint32_t i;
114 };
115
116 static VLC vlc_scalefactors;
117 static VLC vlc_spectral[11];
118
119 static const char overread_err[] = "Input buffer exhausted before END element found\n";
120
121 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
122 {
123 // For PCE based channel configurations map the channels solely based on tags.
124 if (!ac->m4ac.chan_config) {
125 return ac->tag_che_map[type][elem_id];
126 }
127 // For indexed channel configurations map the channels solely based on position.
128 switch (ac->m4ac.chan_config) {
129 case 7:
130 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
131 ac->tags_mapped++;
132 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
133 }
134 case 6:
135 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
136 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
137 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
138 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
139 ac->tags_mapped++;
140 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
141 }
142 case 5:
143 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
144 ac->tags_mapped++;
145 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
146 }
147 case 4:
148 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
149 ac->tags_mapped++;
150 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
151 }
152 case 3:
153 case 2:
154 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
155 ac->tags_mapped++;
156 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
157 } else if (ac->m4ac.chan_config == 2) {
158 return NULL;
159 }
160 case 1:
161 if (!ac->tags_mapped && type == TYPE_SCE) {
162 ac->tags_mapped++;
163 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
164 }
165 default:
166 return NULL;
167 }
168 }
169
170 /**
171 * Check for the channel element in the current channel position configuration.
172 * If it exists, make sure the appropriate element is allocated and map the
173 * channel order to match the internal Libav channel layout.
174 *
175 * @param che_pos current channel position configuration
176 * @param type channel element type
177 * @param id channel element id
178 * @param channels count of the number of channels in the configuration
179 *
180 * @return Returns error status. 0 - OK, !0 - error
181 */
182 static av_cold int che_configure(AACContext *ac,
183 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
184 int type, int id, int *channels)
185 {
186 if (che_pos[type][id]) {
187 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
188 return AVERROR(ENOMEM);
189 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
190 if (type != TYPE_CCE) {
191 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
192 if (type == TYPE_CPE ||
193 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
194 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
195 }
196 }
197 } else {
198 if (ac->che[type][id])
199 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
200 av_freep(&ac->che[type][id]);
201 }
202 return 0;
203 }
204
205 /**
206 * Configure output channel order based on the current program configuration element.
207 *
208 * @param che_pos current channel position configuration
209 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
210 *
211 * @return Returns error status. 0 - OK, !0 - error
212 */
213 static av_cold int output_configure(AACContext *ac,
214 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
215 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
216 int channel_config, enum OCStatus oc_type)
217 {
218 AVCodecContext *avctx = ac->avctx;
219 int i, type, channels = 0, ret;
220
221 if (new_che_pos != che_pos)
222 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
223
224 if (channel_config) {
225 for (i = 0; i < tags_per_config[channel_config]; i++) {
226 if ((ret = che_configure(ac, che_pos,
227 aac_channel_layout_map[channel_config - 1][i][0],
228 aac_channel_layout_map[channel_config - 1][i][1],
229 &channels)))
230 return ret;
231 }
232
233 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
234
235 avctx->channel_layout = aac_channel_layout[channel_config - 1];
236 } else {
237 /* Allocate or free elements depending on if they are in the
238 * current program configuration.
239 *
240 * Set up default 1:1 output mapping.
241 *
242 * For a 5.1 stream the output order will be:
243 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
244 */
245
246 for (i = 0; i < MAX_ELEM_ID; i++) {
247 for (type = 0; type < 4; type++) {
248 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
249 return ret;
250 }
251 }
252
253 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
254
255 avctx->channel_layout = 0;
256 }
257
258 avctx->channels = channels;
259
260 ac->output_configured = oc_type;
261
262 return 0;
263 }
264
265 /**
266 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
267 *
268 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
269 * @param sce_map mono (Single Channel Element) map
270 * @param type speaker type/position for these channels
271 */
272 static void decode_channel_map(enum ChannelPosition *cpe_map,
273 enum ChannelPosition *sce_map,
274 enum ChannelPosition type,
275 GetBitContext *gb, int n)
276 {
277 while (n--) {
278 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
279 map[get_bits(gb, 4)] = type;
280 }
281 }
282
283 /**
284 * Decode program configuration element; reference: table 4.2.
285 *
286 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
287 *
288 * @return Returns error status. 0 - OK, !0 - error
289 */
290 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
291 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
292 GetBitContext *gb)
293 {
294 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
295 int comment_len;
296
297 skip_bits(gb, 2); // object_type
298
299 sampling_index = get_bits(gb, 4);
300 if (m4ac->sampling_index != sampling_index)
301 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");
302
303 num_front = get_bits(gb, 4);
304 num_side = get_bits(gb, 4);
305 num_back = get_bits(gb, 4);
306 num_lfe = get_bits(gb, 2);
307 num_assoc_data = get_bits(gb, 3);
308 num_cc = get_bits(gb, 4);
309
310 if (get_bits1(gb))
311 skip_bits(gb, 4); // mono_mixdown_tag
312 if (get_bits1(gb))
313 skip_bits(gb, 4); // stereo_mixdown_tag
314
315 if (get_bits1(gb))
316 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
317
318 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
321 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
322
323 skip_bits_long(gb, 4 * num_assoc_data);
324
325 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
326
327 align_get_bits(gb);
328
329 /* comment field, first byte is length */
330 comment_len = get_bits(gb, 8) * 8;
331 if (get_bits_left(gb) < comment_len) {
332 av_log(avctx, AV_LOG_ERROR, overread_err);
333 return -1;
334 }
335 skip_bits_long(gb, comment_len);
336 return 0;
337 }
338
339 /**
340 * Set up channel positions based on a default channel configuration
341 * as specified in table 1.17.
342 *
343 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
344 *
345 * @return Returns error status. 0 - OK, !0 - error
346 */
347 static av_cold int set_default_channel_config(AVCodecContext *avctx,
348 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
349 int channel_config)
350 {
351 if (channel_config < 1 || channel_config > 7) {
352 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
353 channel_config);
354 return -1;
355 }
356
357 /* default channel configurations:
358 *
359 * 1ch : front center (mono)
360 * 2ch : L + R (stereo)
361 * 3ch : front center + L + R
362 * 4ch : front center + L + R + back center
363 * 5ch : front center + L + R + back stereo
364 * 6ch : front center + L + R + back stereo + LFE
365 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
366 */
367
368 if (channel_config != 2)
369 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
370 if (channel_config > 1)
371 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
372 if (channel_config == 4)
373 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
374 if (channel_config > 4)
375 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
376 = AAC_CHANNEL_BACK; // back stereo
377 if (channel_config > 5)
378 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
379 if (channel_config == 7)
380 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
381
382 return 0;
383 }
384
385 /**
386 * Decode GA "General Audio" specific configuration; reference: table 4.1.
387 *
388 * @param ac pointer to AACContext, may be null
389 * @param avctx pointer to AVCCodecContext, used for logging
390 *
391 * @return Returns error status. 0 - OK, !0 - error
392 */
393 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
394 GetBitContext *gb,
395 MPEG4AudioConfig *m4ac,
396 int channel_config)
397 {
398 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
399 int extension_flag, ret;
400
401 if (get_bits1(gb)) { // frameLengthFlag
402 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
403 return -1;
404 }
405
406 if (get_bits1(gb)) // dependsOnCoreCoder
407 skip_bits(gb, 14); // coreCoderDelay
408 extension_flag = get_bits1(gb);
409
410 if (m4ac->object_type == AOT_AAC_SCALABLE ||
411 m4ac->object_type == AOT_ER_AAC_SCALABLE)
412 skip_bits(gb, 3); // layerNr
413
414 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
415 if (channel_config == 0) {
416 skip_bits(gb, 4); // element_instance_tag
417 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
418 return ret;
419 } else {
420 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
421 return ret;
422 }
423 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
424 return ret;
425
426 if (extension_flag) {
427 switch (m4ac->object_type) {
428 case AOT_ER_BSAC:
429 skip_bits(gb, 5); // numOfSubFrame
430 skip_bits(gb, 11); // layer_length
431 break;
432 case AOT_ER_AAC_LC:
433 case AOT_ER_AAC_LTP:
434 case AOT_ER_AAC_SCALABLE:
435 case AOT_ER_AAC_LD:
436 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
437 * aacScalefactorDataResilienceFlag
438 * aacSpectralDataResilienceFlag
439 */
440 break;
441 }
442 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
443 }
444 return 0;
445 }
446
447 /**
448 * Decode audio specific configuration; reference: table 1.13.
449 *
450 * @param ac pointer to AACContext, may be null
451 * @param avctx pointer to AVCCodecContext, used for logging
452 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
453 * @param data pointer to AVCodecContext extradata
454 * @param data_size size of AVCCodecContext extradata
455 *
456 * @return Returns error status or number of consumed bits. <0 - error
457 */
458 static int decode_audio_specific_config(AACContext *ac,
459 AVCodecContext *avctx,
460 MPEG4AudioConfig *m4ac,
461 const uint8_t *data, int data_size)
462 {
463 GetBitContext gb;
464 int i;
465
466 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
467 for (i = 0; i < avctx->extradata_size; i++)
468 av_dlog(avctx, "%02x ", avctx->extradata[i]);
469 av_dlog(avctx, "\n");
470
471 init_get_bits(&gb, data, data_size * 8);
472
473 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
474 return -1;
475 if (m4ac->sampling_index > 12) {
476 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
477 return -1;
478 }
479 if (m4ac->sbr == 1 && m4ac->ps == -1)
480 m4ac->ps = 1;
481
482 skip_bits_long(&gb, i);
483
484 switch (m4ac->object_type) {
485 case AOT_AAC_MAIN:
486 case AOT_AAC_LC:
487 case AOT_AAC_LTP:
488 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
489 return -1;
490 break;
491 default:
492 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
493 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
494 return -1;
495 }
496
497 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
498 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
499 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
500
501 return get_bits_count(&gb);
502 }
503
504 /**
505 * linear congruential pseudorandom number generator
506 *
507 * @param previous_val pointer to the current state of the generator
508 *
509 * @return Returns a 32-bit pseudorandom integer
510 */
511 static av_always_inline int lcg_random(int previous_val)
512 {
513 return previous_val * 1664525 + 1013904223;
514 }
515
516 static av_always_inline void reset_predict_state(PredictorState *ps)
517 {
518 ps->r0 = 0.0f;
519 ps->r1 = 0.0f;
520 ps->cor0 = 0.0f;
521 ps->cor1 = 0.0f;
522 ps->var0 = 1.0f;
523 ps->var1 = 1.0f;
524 }
525
526 static void reset_all_predictors(PredictorState *ps)
527 {
528 int i;
529 for (i = 0; i < MAX_PREDICTORS; i++)
530 reset_predict_state(&ps[i]);
531 }
532
533 static void reset_predictor_group(PredictorState *ps, int group_num)
534 {
535 int i;
536 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
537 reset_predict_state(&ps[i]);
538 }
539
540 #define AAC_INIT_VLC_STATIC(num, size) \
541 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
542 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
543 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
544 size);
545
546 static av_cold int aac_decode_init(AVCodecContext *avctx)
547 {
548 AACContext *ac = avctx->priv_data;
549
550 ac->avctx = avctx;
551 ac->m4ac.sample_rate = avctx->sample_rate;
552
553 if (avctx->extradata_size > 0) {
554 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
555 avctx->extradata,
556 avctx->extradata_size) < 0)
557 return -1;
558 }
559
560 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
561
562 AAC_INIT_VLC_STATIC( 0, 304);
563 AAC_INIT_VLC_STATIC( 1, 270);
564 AAC_INIT_VLC_STATIC( 2, 550);
565 AAC_INIT_VLC_STATIC( 3, 300);
566 AAC_INIT_VLC_STATIC( 4, 328);
567 AAC_INIT_VLC_STATIC( 5, 294);
568 AAC_INIT_VLC_STATIC( 6, 306);
569 AAC_INIT_VLC_STATIC( 7, 268);
570 AAC_INIT_VLC_STATIC( 8, 510);
571 AAC_INIT_VLC_STATIC( 9, 366);
572 AAC_INIT_VLC_STATIC(10, 462);
573
574 ff_aac_sbr_init();
575
576 dsputil_init(&ac->dsp, avctx);
577 ff_fmt_convert_init(&ac->fmt_conv, avctx);
578
579 ac->random_state = 0x1f2e3d4c;
580
581 ff_aac_tableinit();
582
583 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
584 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
585 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
586 352);
587
588 ff_mdct_init(&ac->mdct, 11, 1, 1.0/1024.0);
589 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0/128.0);
590 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0);
591 // window initialization
592 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
593 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
594 ff_init_ff_sine_windows(10);
595 ff_init_ff_sine_windows( 7);
596
597 cbrt_tableinit();
598
599 return 0;
600 }
601
602 /**
603 * Skip data_stream_element; reference: table 4.10.
604 */
605 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
606 {
607 int byte_align = get_bits1(gb);
608 int count = get_bits(gb, 8);
609 if (count == 255)
610 count += get_bits(gb, 8);
611 if (byte_align)
612 align_get_bits(gb);
613
614 if (get_bits_left(gb) < 8 * count) {
615 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
616 return -1;
617 }
618 skip_bits_long(gb, 8 * count);
619 return 0;
620 }
621
622 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
623 GetBitContext *gb)
624 {
625 int sfb;
626 if (get_bits1(gb)) {
627 ics->predictor_reset_group = get_bits(gb, 5);
628 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
629 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
630 return -1;
631 }
632 }
633 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
634 ics->prediction_used[sfb] = get_bits1(gb);
635 }
636 return 0;
637 }
638
639 /**
640 * Decode Long Term Prediction data; reference: table 4.xx.
641 */
642 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
643 GetBitContext *gb, uint8_t max_sfb)
644 {
645 int sfb;
646
647 ltp->lag = get_bits(gb, 11);
648 ltp->coef = ltp_coef[get_bits(gb, 3)];
649 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
650 ltp->used[sfb] = get_bits1(gb);
651 }
652
653 /**
654 * Decode Individual Channel Stream info; reference: table 4.6.
655 *
656 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
657 */
658 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
659 GetBitContext *gb, int common_window)
660 {
661 if (get_bits1(gb)) {
662 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
663 memset(ics, 0, sizeof(IndividualChannelStream));
664 return -1;
665 }
666 ics->window_sequence[1] = ics->window_sequence[0];
667 ics->window_sequence[0] = get_bits(gb, 2);
668 ics->use_kb_window[1] = ics->use_kb_window[0];
669 ics->use_kb_window[0] = get_bits1(gb);
670 ics->num_window_groups = 1;
671 ics->group_len[0] = 1;
672 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
673 int i;
674 ics->max_sfb = get_bits(gb, 4);
675 for (i = 0; i < 7; i++) {
676 if (get_bits1(gb)) {
677 ics->group_len[ics->num_window_groups - 1]++;
678 } else {
679 ics->num_window_groups++;
680 ics->group_len[ics->num_window_groups - 1] = 1;
681 }
682 }
683 ics->num_windows = 8;
684 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
685 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
686 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
687 ics->predictor_present = 0;
688 } else {
689 ics->max_sfb = get_bits(gb, 6);
690 ics->num_windows = 1;
691 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
692 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
693 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
694 ics->predictor_present = get_bits1(gb);
695 ics->predictor_reset_group = 0;
696 if (ics->predictor_present) {
697 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
698 if (decode_prediction(ac, ics, gb)) {
699 memset(ics, 0, sizeof(IndividualChannelStream));
700 return -1;
701 }
702 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
703 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
704 memset(ics, 0, sizeof(IndividualChannelStream));
705 return -1;
706 } else {
707 if ((ics->ltp.present = get_bits(gb, 1)))
708 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
709 }
710 }
711 }
712
713 if (ics->max_sfb > ics->num_swb) {
714 av_log(ac->avctx, AV_LOG_ERROR,
715 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
716 ics->max_sfb, ics->num_swb);
717 memset(ics, 0, sizeof(IndividualChannelStream));
718 return -1;
719 }
720
721 return 0;
722 }
723
724 /**
725 * Decode band types (section_data payload); reference: table 4.46.
726 *
727 * @param band_type array of the used band type
728 * @param band_type_run_end array of the last scalefactor band of a band type run
729 *
730 * @return Returns error status. 0 - OK, !0 - error
731 */
732 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
733 int band_type_run_end[120], GetBitContext *gb,
734 IndividualChannelStream *ics)
735 {
736 int g, idx = 0;
737 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
738 for (g = 0; g < ics->num_window_groups; g++) {
739 int k = 0;
740 while (k < ics->max_sfb) {
741 uint8_t sect_end = k;
742 int sect_len_incr;
743 int sect_band_type = get_bits(gb, 4);
744 if (sect_band_type == 12) {
745 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
746 return -1;
747 }
748 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
749 sect_end += sect_len_incr;
750 sect_end += sect_len_incr;
751 if (get_bits_left(gb) < 0) {
752 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
753 return -1;
754 }
755 if (sect_end > ics->max_sfb) {
756 av_log(ac->avctx, AV_LOG_ERROR,
757 "Number of bands (%d) exceeds limit (%d).\n",
758 sect_end, ics->max_sfb);
759 return -1;
760 }
761 for (; k < sect_end; k++) {
762 band_type [idx] = sect_band_type;
763 band_type_run_end[idx++] = sect_end;
764 }
765 }
766 }
767 return 0;
768 }
769
770 /**
771 * Decode scalefactors; reference: table 4.47.
772 *
773 * @param global_gain first scalefactor value as scalefactors are differentially coded
774 * @param band_type array of the used band type
775 * @param band_type_run_end array of the last scalefactor band of a band type run
776 * @param sf array of scalefactors or intensity stereo positions
777 *
778 * @return Returns error status. 0 - OK, !0 - error
779 */
780 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
781 unsigned int global_gain,
782 IndividualChannelStream *ics,
783 enum BandType band_type[120],
784 int band_type_run_end[120])
785 {
786 int g, i, idx = 0;
787 int offset[3] = { global_gain, global_gain - 90, 0 };
788 int clipped_offset;
789 int noise_flag = 1;
790 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
791 for (g = 0; g < ics->num_window_groups; g++) {
792 for (i = 0; i < ics->max_sfb;) {
793 int run_end = band_type_run_end[idx];
794 if (band_type[idx] == ZERO_BT) {
795 for (; i < run_end; i++, idx++)
796 sf[idx] = 0.;
797 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
798 for (; i < run_end; i++, idx++) {
799 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
800 clipped_offset = av_clip(offset[2], -155, 100);
801 if (offset[2] != clipped_offset) {
802 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
803 "position clipped (%d -> %d).\nIf you heard an "
804 "audible artifact, there may be a bug in the "
805 "decoder. ", offset[2], clipped_offset);
806 }
807 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
808 }
809 } else if (band_type[idx] == NOISE_BT) {
810 for (; i < run_end; i++, idx++) {
811 if (noise_flag-- > 0)
812 offset[1] += get_bits(gb, 9) - 256;
813 else
814 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
815 clipped_offset = av_clip(offset[1], -100, 155);
816 if (offset[2] != clipped_offset) {
817 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
818 "(%d -> %d).\nIf you heard an audible "
819 "artifact, there may be a bug in the decoder. ",
820 offset[1], clipped_offset);
821 }
822 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
823 }
824 } else {
825 for (; i < run_end; i++, idx++) {
826 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
827 if (offset[0] > 255U) {
828 av_log(ac->avctx, AV_LOG_ERROR,
829 "%s (%d) out of range.\n", sf_str[0], offset[0]);
830 return -1;
831 }
832 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
833 }
834 }
835 }
836 }
837 return 0;
838 }
839
840 /**
841 * Decode pulse data; reference: table 4.7.
842 */
843 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
844 const uint16_t *swb_offset, int num_swb)
845 {
846 int i, pulse_swb;
847 pulse->num_pulse = get_bits(gb, 2) + 1;
848 pulse_swb = get_bits(gb, 6);
849 if (pulse_swb >= num_swb)
850 return -1;
851 pulse->pos[0] = swb_offset[pulse_swb];
852 pulse->pos[0] += get_bits(gb, 5);
853 if (pulse->pos[0] > 1023)
854 return -1;
855 pulse->amp[0] = get_bits(gb, 4);
856 for (i = 1; i < pulse->num_pulse; i++) {
857 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
858 if (pulse->pos[i] > 1023)
859 return -1;
860 pulse->amp[i] = get_bits(gb, 4);
861 }
862 return 0;
863 }
864
865 /**
866 * Decode Temporal Noise Shaping data; reference: table 4.48.
867 *
868 * @return Returns error status. 0 - OK, !0 - error
869 */
870 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
871 GetBitContext *gb, const IndividualChannelStream *ics)
872 {
873 int w, filt, i, coef_len, coef_res, coef_compress;
874 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
875 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
876 for (w = 0; w < ics->num_windows; w++) {
877 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
878 coef_res = get_bits1(gb);
879
880 for (filt = 0; filt < tns->n_filt[w]; filt++) {
881 int tmp2_idx;
882 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
883
884 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
885 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
886 tns->order[w][filt], tns_max_order);
887 tns->order[w][filt] = 0;
888 return -1;
889 }
890 if (tns->order[w][filt]) {
891 tns->direction[w][filt] = get_bits1(gb);
892 coef_compress = get_bits1(gb);
893 coef_len = coef_res + 3 - coef_compress;
894 tmp2_idx = 2 * coef_compress + coef_res;
895
896 for (i = 0; i < tns->order[w][filt]; i++)
897 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
898 }
899 }
900 }
901 }
902 return 0;
903 }
904
905 /**
906 * Decode Mid/Side data; reference: table 4.54.
907 *
908 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
909 * [1] mask is decoded from bitstream; [2] mask is all 1s;
910 * [3] reserved for scalable AAC
911 */
912 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
913 int ms_present)
914 {
915 int idx;
916 if (ms_present == 1) {
917 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
918 cpe->ms_mask[idx] = get_bits1(gb);
919 } else if (ms_present == 2) {
920 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
921 }
922 }
923
924 #ifndef VMUL2
925 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
926 const float *scale)
927 {
928 float s = *scale;
929 *dst++ = v[idx & 15] * s;
930 *dst++ = v[idx>>4 & 15] * s;
931 return dst;
932 }
933 #endif
934
935 #ifndef VMUL4
936 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
937 const float *scale)
938 {
939 float s = *scale;
940 *dst++ = v[idx & 3] * s;
941 *dst++ = v[idx>>2 & 3] * s;
942 *dst++ = v[idx>>4 & 3] * s;
943 *dst++ = v[idx>>6 & 3] * s;
944 return dst;
945 }
946 #endif
947
948 #ifndef VMUL2S
949 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
950 unsigned sign, const float *scale)
951 {
952 union float754 s0, s1;
953
954 s0.f = s1.f = *scale;
955 s0.i ^= sign >> 1 << 31;
956 s1.i ^= sign << 31;
957
958 *dst++ = v[idx & 15] * s0.f;
959 *dst++ = v[idx>>4 & 15] * s1.f;
960
961 return dst;
962 }
963 #endif
964
965 #ifndef VMUL4S
966 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
967 unsigned sign, const float *scale)
968 {
969 unsigned nz = idx >> 12;
970 union float754 s = { .f = *scale };
971 union float754 t;
972
973 t.i = s.i ^ (sign & 1U<<31);
974 *dst++ = v[idx & 3] * t.f;
975
976 sign <<= nz & 1; nz >>= 1;
977 t.i = s.i ^ (sign & 1U<<31);
978 *dst++ = v[idx>>2 & 3] * t.f;
979
980 sign <<= nz & 1; nz >>= 1;
981 t.i = s.i ^ (sign & 1U<<31);
982 *dst++ = v[idx>>4 & 3] * t.f;
983
984 sign <<= nz & 1; nz >>= 1;
985 t.i = s.i ^ (sign & 1U<<31);
986 *dst++ = v[idx>>6 & 3] * t.f;
987
988 return dst;
989 }
990 #endif
991
992 /**
993 * Decode spectral data; reference: table 4.50.
994 * Dequantize and scale spectral data; reference: 4.6.3.3.
995 *
996 * @param coef array of dequantized, scaled spectral data
997 * @param sf array of scalefactors or intensity stereo positions
998 * @param pulse_present set if pulses are present
999 * @param pulse pointer to pulse data struct
1000 * @param band_type array of the used band type
1001 *
1002 * @return Returns error status. 0 - OK, !0 - error
1003 */
1004 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1005 GetBitContext *gb, const float sf[120],
1006 int pulse_present, const Pulse *pulse,
1007 const IndividualChannelStream *ics,
1008 enum BandType band_type[120])
1009 {
1010 int i, k, g, idx = 0;
1011 const int c = 1024 / ics->num_windows;
1012 const uint16_t *offsets = ics->swb_offset;
1013 float *coef_base = coef;
1014
1015 for (g = 0; g < ics->num_windows; g++)
1016 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1017
1018 for (g = 0; g < ics->num_window_groups; g++) {
1019 unsigned g_len = ics->group_len[g];
1020
1021 for (i = 0; i < ics->max_sfb; i++, idx++) {
1022 const unsigned cbt_m1 = band_type[idx] - 1;
1023 float *cfo = coef + offsets[i];
1024 int off_len = offsets[i + 1] - offsets[i];
1025 int group;
1026
1027 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1028 for (group = 0; group < g_len; group++, cfo+=128) {
1029 memset(cfo, 0, off_len * sizeof(float));
1030 }
1031 } else if (cbt_m1 == NOISE_BT - 1) {
1032 for (group = 0; group < g_len; group++, cfo+=128) {
1033 float scale;
1034 float band_energy;
1035
1036 for (k = 0; k < off_len; k++) {
1037 ac->random_state = lcg_random(ac->random_state);
1038 cfo[k] = ac->random_state;
1039 }
1040
1041 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1042 scale = sf[idx] / sqrtf(band_energy);
1043 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1044 }
1045 } else {
1046 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1047 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1048 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1049 OPEN_READER(re, gb);
1050
1051 switch (cbt_m1 >> 1) {
1052 case 0:
1053 for (group = 0; group < g_len; group++, cfo+=128) {
1054 float *cf = cfo;
1055 int len = off_len;
1056
1057 do {
1058 int code;
1059 unsigned cb_idx;
1060
1061 UPDATE_CACHE(re, gb);
1062 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1063 cb_idx = cb_vector_idx[code];
1064 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1065 } while (len -= 4);
1066 }
1067 break;
1068
1069 case 1:
1070 for (group = 0; group < g_len; group++, cfo+=128) {
1071 float *cf = cfo;
1072 int len = off_len;
1073
1074 do {
1075 int code;
1076 unsigned nnz;
1077 unsigned cb_idx;
1078 uint32_t bits;
1079
1080 UPDATE_CACHE(re, gb);
1081 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1082 cb_idx = cb_vector_idx[code];
1083 nnz = cb_idx >> 8 & 15;
1084 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1085 LAST_SKIP_BITS(re, gb, nnz);
1086 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1087 } while (len -= 4);
1088 }
1089 break;
1090
1091 case 2:
1092 for (group = 0; group < g_len; group++, cfo+=128) {
1093 float *cf = cfo;
1094 int len = off_len;
1095
1096 do {
1097 int code;
1098 unsigned cb_idx;
1099
1100 UPDATE_CACHE(re, gb);
1101 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1102 cb_idx = cb_vector_idx[code];
1103 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1104 } while (len -= 2);
1105 }
1106 break;
1107
1108 case 3:
1109 case 4:
1110 for (group = 0; group < g_len; group++, cfo+=128) {
1111 float *cf = cfo;
1112 int len = off_len;
1113
1114 do {
1115 int code;
1116 unsigned nnz;
1117 unsigned cb_idx;
1118 unsigned sign;
1119
1120 UPDATE_CACHE(re, gb);
1121 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1122 cb_idx = cb_vector_idx[code];
1123 nnz = cb_idx >> 8 & 15;
1124 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1125 LAST_SKIP_BITS(re, gb, nnz);
1126 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1127 } while (len -= 2);
1128 }
1129 break;
1130
1131 default:
1132 for (group = 0; group < g_len; group++, cfo+=128) {
1133 float *cf = cfo;
1134 uint32_t *icf = (uint32_t *) cf;
1135 int len = off_len;
1136
1137 do {
1138 int code;
1139 unsigned nzt, nnz;
1140 unsigned cb_idx;
1141 uint32_t bits;
1142 int j;
1143
1144 UPDATE_CACHE(re, gb);
1145 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1146
1147 if (!code) {
1148 *icf++ = 0;
1149 *icf++ = 0;
1150 continue;
1151 }
1152
1153 cb_idx = cb_vector_idx[code];
1154 nnz = cb_idx >> 12;
1155 nzt = cb_idx >> 8;
1156 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1157 LAST_SKIP_BITS(re, gb, nnz);
1158
1159 for (j = 0; j < 2; j++) {
1160 if (nzt & 1<<j) {
1161 uint32_t b;
1162 int n;
1163 /* The total length of escape_sequence must be < 22 bits according
1164 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1165 UPDATE_CACHE(re, gb);
1166 b = GET_CACHE(re, gb);
1167 b = 31 - av_log2(~b);
1168
1169 if (b > 8) {
1170 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1171 return -1;
1172 }
1173
1174 SKIP_BITS(re, gb, b + 1);
1175 b += 4;
1176 n = (1 << b) + SHOW_UBITS(re, gb, b);
1177 LAST_SKIP_BITS(re, gb, b);
1178 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1179 bits <<= 1;
1180 } else {
1181 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1182 *icf++ = (bits & 1U<<31) | v;
1183 bits <<= !!v;
1184 }
1185 cb_idx >>= 4;
1186 }
1187 } while (len -= 2);
1188
1189 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1190 }
1191 }
1192
1193 CLOSE_READER(re, gb);
1194 }
1195 }
1196 coef += g_len << 7;
1197 }
1198
1199 if (pulse_present) {
1200 idx = 0;
1201 for (i = 0; i < pulse->num_pulse; i++) {
1202 float co = coef_base[ pulse->pos[i] ];
1203 while (offsets[idx + 1] <= pulse->pos[i])
1204 idx++;
1205 if (band_type[idx] != NOISE_BT && sf[idx]) {
1206 float ico = -pulse->amp[i];
1207 if (co) {
1208 co /= sf[idx];
1209 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1210 }
1211 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1212 }
1213 }
1214 }
1215 return 0;
1216 }
1217
1218 static av_always_inline float flt16_round(float pf)
1219 {
1220 union float754 tmp;
1221 tmp.f = pf;
1222 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1223 return tmp.f;
1224 }
1225
1226 static av_always_inline float flt16_even(float pf)
1227 {
1228 union float754 tmp;
1229 tmp.f = pf;
1230 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1231 return tmp.f;
1232 }
1233
1234 static av_always_inline float flt16_trunc(float pf)
1235 {
1236 union float754 pun;
1237 pun.f = pf;
1238 pun.i &= 0xFFFF0000U;
1239 return pun.f;
1240 }
1241
1242 static av_always_inline void predict(PredictorState *ps, float *coef,
1243 int output_enable)
1244 {
1245 const float a = 0.953125; // 61.0 / 64
1246 const float alpha = 0.90625; // 29.0 / 32
1247 float e0, e1;
1248 float pv;
1249 float k1, k2;
1250 float r0 = ps->r0, r1 = ps->r1;
1251 float cor0 = ps->cor0, cor1 = ps->cor1;
1252 float var0 = ps->var0, var1 = ps->var1;
1253
1254 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1255 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1256
1257 pv = flt16_round(k1 * r0 + k2 * r1);
1258 if (output_enable)
1259 *coef += pv;
1260
1261 e0 = *coef;
1262 e1 = e0 - k1 * r0;
1263
1264 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1265 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1266 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1267 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1268
1269 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1270 ps->r0 = flt16_trunc(a * e0);
1271 }
1272
1273 /**
1274 * Apply AAC-Main style frequency domain prediction.
1275 */
1276 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1277 {
1278 int sfb, k;
1279
1280 if (!sce->ics.predictor_initialized) {
1281 reset_all_predictors(sce->predictor_state);
1282 sce->ics.predictor_initialized = 1;
1283 }
1284
1285 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1286 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1287 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1288 predict(&sce->predictor_state[k], &sce->coeffs[k],
1289 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1290 }
1291 }
1292 if (sce->ics.predictor_reset_group)
1293 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1294 } else
1295 reset_all_predictors(sce->predictor_state);
1296 }
1297
1298 /**
1299 * Decode an individual_channel_stream payload; reference: table 4.44.
1300 *
1301 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1302 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1303 *
1304 * @return Returns error status. 0 - OK, !0 - error
1305 */
1306 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1307 GetBitContext *gb, int common_window, int scale_flag)
1308 {
1309 Pulse pulse;
1310 TemporalNoiseShaping *tns = &sce->tns;
1311 IndividualChannelStream *ics = &sce->ics;
1312 float *out = sce->coeffs;
1313 int global_gain, pulse_present = 0;
1314
1315 /* This assignment is to silence a GCC warning about the variable being used
1316 * uninitialized when in fact it always is.
1317 */
1318 pulse.num_pulse = 0;
1319
1320 global_gain = get_bits(gb, 8);
1321
1322 if (!common_window && !scale_flag) {
1323 if (decode_ics_info(ac, ics, gb, 0) < 0)
1324 return -1;
1325 }
1326
1327 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1328 return -1;
1329 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1330 return -1;
1331
1332 pulse_present = 0;
1333 if (!scale_flag) {
1334 if ((pulse_present = get_bits1(gb))) {
1335 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1336 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1337 return -1;
1338 }
1339 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1340 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1341 return -1;
1342 }
1343 }
1344 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1345 return -1;
1346 if (get_bits1(gb)) {
1347 av_log_missing_feature(ac->avctx, "SSR", 1);
1348 return -1;
1349 }
1350 }
1351
1352 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1353 return -1;
1354
1355 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1356 apply_prediction(ac, sce);
1357
1358 return 0;
1359 }
1360
1361 /**
1362 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1363 */
1364 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1365 {
1366 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1367 float *ch0 = cpe->ch[0].coeffs;
1368 float *ch1 = cpe->ch[1].coeffs;
1369 int g, i, group, idx = 0;
1370 const uint16_t *offsets = ics->swb_offset;
1371 for (g = 0; g < ics->num_window_groups; g++) {
1372 for (i = 0; i < ics->max_sfb; i++, idx++) {
1373 if (cpe->ms_mask[idx] &&
1374 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1375 for (group = 0; group < ics->group_len[g]; group++) {
1376 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1377 ch1 + group * 128 + offsets[i],
1378 offsets[i+1] - offsets[i]);
1379 }
1380 }
1381 }
1382 ch0 += ics->group_len[g] * 128;
1383 ch1 += ics->group_len[g] * 128;
1384 }
1385 }
1386
1387 /**
1388 * intensity stereo decoding; reference: 4.6.8.2.3
1389 *
1390 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1391 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1392 * [3] reserved for scalable AAC
1393 */
1394 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1395 {
1396 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1397 SingleChannelElement *sce1 = &cpe->ch[1];
1398 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1399 const uint16_t *offsets = ics->swb_offset;
1400 int g, group, i, idx = 0;
1401 int c;
1402 float scale;
1403 for (g = 0; g < ics->num_window_groups; g++) {
1404 for (i = 0; i < ics->max_sfb;) {
1405 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1406 const int bt_run_end = sce1->band_type_run_end[idx];
1407 for (; i < bt_run_end; i++, idx++) {
1408 c = -1 + 2 * (sce1->band_type[idx] - 14);
1409 if (ms_present)
1410 c *= 1 - 2 * cpe->ms_mask[idx];
1411 scale = c * sce1->sf[idx];
1412 for (group = 0; group < ics->group_len[g]; group++)
1413 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1414 coef0 + group * 128 + offsets[i],
1415 scale,
1416 offsets[i + 1] - offsets[i]);
1417 }
1418 } else {
1419 int bt_run_end = sce1->band_type_run_end[idx];
1420 idx += bt_run_end - i;
1421 i = bt_run_end;
1422 }
1423 }
1424 coef0 += ics->group_len[g] * 128;
1425 coef1 += ics->group_len[g] * 128;
1426 }
1427 }
1428
1429 /**
1430 * Decode a channel_pair_element; reference: table 4.4.
1431 *
1432 * @return Returns error status. 0 - OK, !0 - error
1433 */
1434 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1435 {
1436 int i, ret, common_window, ms_present = 0;
1437
1438 common_window = get_bits1(gb);
1439 if (common_window) {
1440 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1441 return -1;
1442 i = cpe->ch[1].ics.use_kb_window[0];
1443 cpe->ch[1].ics = cpe->ch[0].ics;
1444 cpe->ch[1].ics.use_kb_window[1] = i;
1445 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1446 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1447 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1448 ms_present = get_bits(gb, 2);
1449 if (ms_present == 3) {
1450 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1451 return -1;
1452 } else if (ms_present)
1453 decode_mid_side_stereo(cpe, gb, ms_present);
1454 }
1455 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1456 return ret;
1457 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1458 return ret;
1459
1460 if (common_window) {
1461 if (ms_present)
1462 apply_mid_side_stereo(ac, cpe);
1463 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1464 apply_prediction(ac, &cpe->ch[0]);
1465 apply_prediction(ac, &cpe->ch[1]);
1466 }
1467 }
1468
1469 apply_intensity_stereo(ac, cpe, ms_present);
1470 return 0;
1471 }
1472
1473 static const float cce_scale[] = {
1474 1.09050773266525765921, //2^(1/8)
1475 1.18920711500272106672, //2^(1/4)
1476 M_SQRT2,
1477 2,
1478 };
1479
1480 /**
1481 * Decode coupling_channel_element; reference: table 4.8.
1482 *
1483 * @return Returns error status. 0 - OK, !0 - error
1484 */
1485 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1486 {
1487 int num_gain = 0;
1488 int c, g, sfb, ret;
1489 int sign;
1490 float scale;
1491 SingleChannelElement *sce = &che->ch[0];
1492 ChannelCoupling *coup = &che->coup;
1493
1494 coup->coupling_point = 2 * get_bits1(gb);
1495 coup->num_coupled = get_bits(gb, 3);
1496 for (c = 0; c <= coup->num_coupled; c++) {
1497 num_gain++;
1498 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1499 coup->id_select[c] = get_bits(gb, 4);
1500 if (coup->type[c] == TYPE_CPE) {
1501 coup->ch_select[c] = get_bits(gb, 2);
1502 if (coup->ch_select[c] == 3)
1503 num_gain++;
1504 } else
1505 coup->ch_select[c] = 2;
1506 }
1507 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1508
1509 sign = get_bits(gb, 1);
1510 scale = cce_scale[get_bits(gb, 2)];
1511
1512 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1513 return ret;
1514
1515 for (c = 0; c < num_gain; c++) {
1516 int idx = 0;
1517 int cge = 1;
1518 int gain = 0;
1519 float gain_cache = 1.;
1520 if (c) {
1521 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1522 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1523 gain_cache = powf(scale, -gain);
1524 }
1525 if (coup->coupling_point == AFTER_IMDCT) {
1526 coup->gain[c][0] = gain_cache;
1527 } else {
1528 for (g = 0; g < sce->ics.num_window_groups; g++) {
1529 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1530 if (sce->band_type[idx] != ZERO_BT) {
1531 if (!cge) {
1532 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1533 if (t) {
1534 int s = 1;
1535 t = gain += t;
1536 if (sign) {
1537 s -= 2 * (t & 0x1);
1538 t >>= 1;
1539 }
1540 gain_cache = powf(scale, -t) * s;
1541 }
1542 }
1543 coup->gain[c][idx] = gain_cache;
1544 }
1545 }
1546 }
1547 }
1548 }
1549 return 0;
1550 }
1551
1552 /**
1553 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1554 *
1555 * @return Returns number of bytes consumed.
1556 */
1557 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1558 GetBitContext *gb)
1559 {
1560 int i;
1561 int num_excl_chan = 0;
1562
1563 do {
1564 for (i = 0; i < 7; i++)
1565 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1566 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1567
1568 return num_excl_chan / 7;
1569 }
1570
1571 /**
1572 * Decode dynamic range information; reference: table 4.52.
1573 *
1574 * @param cnt length of TYPE_FIL syntactic element in bytes
1575 *
1576 * @return Returns number of bytes consumed.
1577 */
1578 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1579 GetBitContext *gb, int cnt)
1580 {
1581 int n = 1;
1582 int drc_num_bands = 1;
1583 int i;
1584
1585 /* pce_tag_present? */
1586 if (get_bits1(gb)) {
1587 che_drc->pce_instance_tag = get_bits(gb, 4);
1588 skip_bits(gb, 4); // tag_reserved_bits
1589 n++;
1590 }
1591
1592 /* excluded_chns_present? */
1593 if (get_bits1(gb)) {
1594 n += decode_drc_channel_exclusions(che_drc, gb);
1595 }
1596
1597 /* drc_bands_present? */
1598 if (get_bits1(gb)) {
1599 che_drc->band_incr = get_bits(gb, 4);
1600 che_drc->interpolation_scheme = get_bits(gb, 4);
1601 n++;
1602 drc_num_bands += che_drc->band_incr;
1603 for (i = 0; i < drc_num_bands; i++) {
1604 che_drc->band_top[i] = get_bits(gb, 8);
1605 n++;
1606 }
1607 }
1608
1609 /* prog_ref_level_present? */
1610 if (get_bits1(gb)) {
1611 che_drc->prog_ref_level = get_bits(gb, 7);
1612 skip_bits1(gb); // prog_ref_level_reserved_bits
1613 n++;
1614 }
1615
1616 for (i = 0; i < drc_num_bands; i++) {
1617 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1618 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1619 n++;
1620 }
1621
1622 return n;
1623 }
1624
1625 /**
1626 * Decode extension data (incomplete); reference: table 4.51.
1627 *
1628 * @param cnt length of TYPE_FIL syntactic element in bytes
1629 *
1630 * @return Returns number of bytes consumed
1631 */
1632 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1633 ChannelElement *che, enum RawDataBlockType elem_type)
1634 {
1635 int crc_flag = 0;
1636 int res = cnt;
1637 switch (get_bits(gb, 4)) { // extension type
1638 case EXT_SBR_DATA_CRC:
1639 crc_flag++;
1640 case EXT_SBR_DATA:
1641 if (!che) {
1642 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1643 return res;
1644 } else if (!ac->m4ac.sbr) {
1645 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1646 skip_bits_long(gb, 8 * cnt - 4);
1647 return res;
1648 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1649 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1650 skip_bits_long(gb, 8 * cnt - 4);
1651 return res;
1652 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1653 ac->m4ac.sbr = 1;
1654 ac->m4ac.ps = 1;
1655 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1656 } else {
1657 ac->m4ac.sbr = 1;
1658 }
1659 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1660 break;
1661 case EXT_DYNAMIC_RANGE:
1662 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1663 break;
1664 case EXT_FILL:
1665 case EXT_FILL_DATA:
1666 case EXT_DATA_ELEMENT:
1667 default:
1668 skip_bits_long(gb, 8 * cnt - 4);
1669 break;
1670 };
1671 return res;
1672 }
1673
1674 /**
1675 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1676 *
1677 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1678 * @param coef spectral coefficients
1679 */
1680 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1681 IndividualChannelStream *ics, int decode)
1682 {
1683 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1684 int w, filt, m, i;
1685 int bottom, top, order, start, end, size, inc;
1686 float lpc[TNS_MAX_ORDER];
1687 float tmp[TNS_MAX_ORDER];
1688
1689 for (w = 0; w < ics->num_windows; w++) {
1690 bottom = ics->num_swb;
1691 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1692 top = bottom;
1693 bottom = FFMAX(0, top - tns->length[w][filt]);
1694 order = tns->order[w][filt];
1695 if (order == 0)
1696 continue;
1697
1698 // tns_decode_coef
1699 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1700
1701 start = ics->swb_offset[FFMIN(bottom, mmm)];
1702 end = ics->swb_offset[FFMIN( top, mmm)];
1703 if ((size = end - start) <= 0)
1704 continue;
1705 if (tns->direction[w][filt]) {
1706 inc = -1;
1707 start = end - 1;
1708 } else {
1709 inc = 1;
1710 }
1711 start += w * 128;
1712
1713 if (decode) {
1714 // ar filter
1715 for (m = 0; m < size; m++, start += inc)
1716 for (i = 1; i <= FFMIN(m, order); i++)
1717 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1718 } else {
1719 // ma filter
1720 for (m = 0; m < size; m++, start += inc) {
1721 tmp[0] = coef[start];
1722 for (i = 1; i <= FFMIN(m, order); i++)
1723 coef[start] += tmp[i] * lpc[i - 1];
1724 for (i = order; i > 0; i--)
1725 tmp[i] = tmp[i - 1];
1726 }
1727 }
1728 }
1729 }
1730 }
1731
1732 /**
1733 * Apply windowing and MDCT to obtain the spectral
1734 * coefficient from the predicted sample by LTP.
1735 */
1736 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1737 float *in, IndividualChannelStream *ics)
1738 {
1739 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1740 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1741 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1742 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1743
1744 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1745 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1746 } else {
1747 memset(in, 0, 448 * sizeof(float));
1748 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1749 memcpy(in + 576, in + 576, 448 * sizeof(float));
1750 }
1751 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1752 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1753 } else {
1754 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1755 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1756 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1757 }
1758 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1759 }
1760
1761 /**
1762 * Apply the long term prediction
1763 */
1764 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1765 {
1766 const LongTermPrediction *ltp = &sce->ics.ltp;
1767 const uint16_t *offsets = sce->ics.swb_offset;
1768 int i, sfb;
1769
1770 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1771 float *predTime = sce->ret;
1772 float *predFreq = ac->buf_mdct;
1773 int16_t num_samples = 2048;
1774
1775 if (ltp->lag < 1024)
1776 num_samples = ltp->lag + 1024;
1777 for (i = 0; i < num_samples; i++)
1778 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1779 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1780
1781 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1782
1783 if (sce->tns.present)
1784 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1785
1786 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1787 if (ltp->used[sfb])
1788 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1789 sce->coeffs[i] += predFreq[i];
1790 }
1791 }
1792
1793 /**
1794 * Update the LTP buffer for next frame
1795 */
1796 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1797 {
1798 IndividualChannelStream *ics = &sce->ics;
1799 float *saved = sce->saved;
1800 float *saved_ltp = sce->coeffs;
1801 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1802 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1803 int i;
1804
1805 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1806 memcpy(saved_ltp, saved, 512 * sizeof(float));
1807 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1808 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1809 for (i = 0; i < 64; i++)
1810 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1811 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1812 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1813 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1814 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1815 for (i = 0; i < 64; i++)
1816 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1817 } else { // LONG_STOP or ONLY_LONG
1818 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1819 for (i = 0; i < 512; i++)
1820 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1821 }
1822
1823 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1824 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1825 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1826 }
1827
1828 /**
1829 * Conduct IMDCT and windowing.
1830 */
1831 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1832 {
1833 IndividualChannelStream *ics = &sce->ics;
1834 float *in = sce->coeffs;
1835 float *out = sce->ret;
1836 float *saved = sce->saved;
1837 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1838 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1839 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1840 float *buf = ac->buf_mdct;
1841 float *temp = ac->temp;
1842 int i;
1843
1844 // imdct
1845 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1846 for (i = 0; i < 1024; i += 128)
1847 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1848 } else
1849 ac->mdct.imdct_half(&ac->mdct, buf, in);
1850
1851 /* window overlapping
1852 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1853 * and long to short transitions are considered to be short to short
1854 * transitions. This leaves just two cases (long to long and short to short)
1855 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1856 */
1857 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1858 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1859 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1860 } else {
1861 memcpy( out, saved, 448 * sizeof(float));
1862
1863 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1864 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1865 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1866 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1867 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1868 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1869 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1870 } else {
1871 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1872 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1873 }
1874 }
1875
1876 // buffer update
1877 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1878 memcpy( saved, temp + 64, 64 * sizeof(float));
1879 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1880 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1881 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1882 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1883 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1884 memcpy( saved, buf + 512, 448 * sizeof(float));
1885 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1886 } else { // LONG_STOP or ONLY_LONG
1887 memcpy( saved, buf + 512, 512 * sizeof(float));
1888 }
1889 }
1890
1891 /**
1892 * Apply dependent channel coupling (applied before IMDCT).
1893 *
1894 * @param index index into coupling gain array
1895 */
1896 static void apply_dependent_coupling(AACContext *ac,
1897 SingleChannelElement *target,
1898 ChannelElement *cce, int index)
1899 {
1900 IndividualChannelStream *ics = &cce->ch[0].ics;
1901 const uint16_t *offsets = ics->swb_offset;
1902 float *dest = target->coeffs;
1903 const float *src = cce->ch[0].coeffs;
1904 int g, i, group, k, idx = 0;
1905 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1906 av_log(ac->avctx, AV_LOG_ERROR,
1907 "Dependent coupling is not supported together with LTP\n");
1908 return;
1909 }
1910 for (g = 0; g < ics->num_window_groups; g++) {
1911 for (i = 0; i < ics->max_sfb; i++, idx++) {
1912 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1913 const float gain = cce->coup.gain[index][idx];
1914 for (group = 0; group < ics->group_len[g]; group++) {
1915 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1916 // XXX dsputil-ize
1917 dest[group * 128 + k] += gain * src[group * 128 + k];
1918 }
1919 }
1920 }
1921 }
1922 dest += ics->group_len[g] * 128;
1923 src += ics->group_len[g] * 128;
1924 }
1925 }
1926
1927 /**
1928 * Apply independent channel coupling (applied after IMDCT).
1929 *
1930 * @param index index into coupling gain array
1931 */
1932 static void apply_independent_coupling(AACContext *ac,
1933 SingleChannelElement *target,
1934 ChannelElement *cce, int index)
1935 {
1936 int i;
1937 const float gain = cce->coup.gain[index][0];
1938 const float *src = cce->ch[0].ret;
1939 float *dest = target->ret;
1940 const int len = 1024 << (ac->m4ac.sbr == 1);
1941
1942 for (i = 0; i < len; i++)
1943 dest[i] += gain * src[i];
1944 }
1945
1946 /**
1947 * channel coupling transformation interface
1948 *
1949 * @param apply_coupling_method pointer to (in)dependent coupling function
1950 */
1951 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1952 enum RawDataBlockType type, int elem_id,
1953 enum CouplingPoint coupling_point,
1954 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1955 {
1956 int i, c;
1957
1958 for (i = 0; i < MAX_ELEM_ID; i++) {
1959 ChannelElement *cce = ac->che[TYPE_CCE][i];
1960 int index = 0;
1961
1962 if (cce && cce->coup.coupling_point == coupling_point) {
1963 ChannelCoupling *coup = &cce->coup;
1964
1965 for (c = 0; c <= coup->num_coupled; c++) {
1966 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1967 if (coup->ch_select[c] != 1) {
1968 apply_coupling_method(ac, &cc->ch[0], cce, index);
1969 if (coup->ch_select[c] != 0)
1970 index++;
1971 }
1972 if (coup->ch_select[c] != 2)
1973 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1974 } else
1975 index += 1 + (coup->ch_select[c] == 3);
1976 }
1977 }
1978 }
1979 }
1980
1981 /**
1982 * Convert spectral data to float samples, applying all supported tools as appropriate.
1983 */
1984 static void spectral_to_sample(AACContext *ac)
1985 {
1986 int i, type;
1987 for (type = 3; type >= 0; type--) {
1988 for (i = 0; i < MAX_ELEM_ID; i++) {
1989 ChannelElement *che = ac->che[type][i];
1990 if (che) {
1991 if (type <= TYPE_CPE)
1992 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1993 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1994 if (che->ch[0].ics.predictor_present) {
1995 if (che->ch[0].ics.ltp.present)
1996 apply_ltp(ac, &che->ch[0]);
1997 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
1998 apply_ltp(ac, &che->ch[1]);
1999 }
2000 }
2001 if (che->ch[0].tns.present)
2002 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2003 if (che->ch[1].tns.present)
2004 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2005 if (type <= TYPE_CPE)
2006 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2007 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2008 imdct_and_windowing(ac, &che->ch[0]);
2009 if (ac->m4ac.object_type == AOT_AAC_LTP)
2010 update_ltp(ac, &che->ch[0]);
2011 if (type == TYPE_CPE) {
2012 imdct_and_windowing(ac, &che->ch[1]);
2013 if (ac->m4ac.object_type == AOT_AAC_LTP)
2014 update_ltp(ac, &che->ch[1]);
2015 }
2016 if (ac->m4ac.sbr > 0) {
2017 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2018 }
2019 }
2020 if (type <= TYPE_CCE)
2021 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2022 }
2023 }
2024 }
2025 }
2026
2027 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2028 {
2029 int size;
2030 AACADTSHeaderInfo hdr_info;
2031
2032 size = ff_aac_parse_header(gb, &hdr_info);
2033 if (size > 0) {
2034 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2035 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2036 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2037 ac->m4ac.chan_config = hdr_info.chan_config;
2038 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2039 return -7;
2040 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2041 return -7;
2042 } else if (ac->output_configured != OC_LOCKED) {
2043 ac->output_configured = OC_NONE;
2044 }
2045 if (ac->output_configured != OC_LOCKED) {
2046 ac->m4ac.sbr = -1;
2047 ac->m4ac.ps = -1;
2048 }
2049 ac->m4ac.sample_rate = hdr_info.sample_rate;
2050 ac->m4ac.sampling_index = hdr_info.sampling_index;
2051 ac->m4ac.object_type = hdr_info.object_type;
2052 if (!ac->avctx->sample_rate)
2053 ac->avctx->sample_rate = hdr_info.sample_rate;
2054 if (hdr_info.num_aac_frames == 1) {
2055 if (!hdr_info.crc_absent)
2056 skip_bits(gb, 16);
2057 } else {
2058 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2059 return -1;
2060 }
2061 }
2062 return size;
2063 }
2064
2065 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2066 int *data_size, GetBitContext *gb)
2067 {
2068 AACContext *ac = avctx->priv_data;
2069 ChannelElement *che = NULL, *che_prev = NULL;
2070 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2071 int err, elem_id, data_size_tmp;
2072 int samples = 0, multiplier;
2073
2074 if (show_bits(gb, 12) == 0xfff) {
2075 if (parse_adts_frame_header(ac, gb) < 0) {
2076 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2077 return -1;
2078 }
2079 if (ac->m4ac.sampling_index > 12) {
2080 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2081 return -1;
2082 }
2083 }
2084
2085 ac->tags_mapped = 0;
2086 // parse
2087 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2088 elem_id = get_bits(gb, 4);
2089
2090 if (elem_type < TYPE_DSE) {
2091 if (!(che=get_che(ac, elem_type, elem_id))) {
2092 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2093 elem_type, elem_id);
2094 return -1;
2095 }
2096 samples = 1024;
2097 }
2098
2099 switch (elem_type) {
2100
2101 case TYPE_SCE:
2102 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2103 break;
2104
2105 case TYPE_CPE:
2106 err = decode_cpe(ac, gb, che);
2107 break;
2108
2109 case TYPE_CCE:
2110 err = decode_cce(ac, gb, che);
2111 break;
2112
2113 case TYPE_LFE:
2114 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2115 break;
2116
2117 case TYPE_DSE:
2118 err = skip_data_stream_element(ac, gb);
2119 break;
2120
2121 case TYPE_PCE: {
2122 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2123 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2124 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2125 break;
2126 if (ac->output_configured > OC_TRIAL_PCE)
2127 av_log(avctx, AV_LOG_ERROR,
2128 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2129 else
2130 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2131 break;
2132 }
2133
2134 case TYPE_FIL:
2135 if (elem_id == 15)
2136 elem_id += get_bits(gb, 8) - 1;
2137 if (get_bits_left(gb) < 8 * elem_id) {
2138 av_log(avctx, AV_LOG_ERROR, overread_err);
2139 return -1;
2140 }
2141 while (elem_id > 0)
2142 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2143 err = 0; /* FIXME */
2144 break;
2145
2146 default:
2147 err = -1; /* should not happen, but keeps compiler happy */
2148 break;
2149 }
2150
2151 che_prev = che;
2152 elem_type_prev = elem_type;
2153
2154 if (err)
2155 return err;
2156
2157 if (get_bits_left(gb) < 3) {
2158 av_log(avctx, AV_LOG_ERROR, overread_err);
2159 return -1;
2160 }
2161 }
2162
2163 spectral_to_sample(ac);
2164
2165 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2166 samples <<= multiplier;
2167 if (ac->output_configured < OC_LOCKED) {
2168 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2169 avctx->frame_size = samples;
2170 }
2171
2172 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2173 if (*data_size < data_size_tmp) {
2174 av_log(avctx, AV_LOG_ERROR,
2175 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2176 *data_size, data_size_tmp);
2177 return -1;
2178 }
2179 *data_size = data_size_tmp;
2180
2181 if (samples)
2182 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2183
2184 if (ac->output_configured)
2185 ac->output_configured = OC_LOCKED;
2186
2187 return 0;
2188 }
2189
2190 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2191 int *data_size, AVPacket *avpkt)
2192 {
2193 const uint8_t *buf = avpkt->data;
2194 int buf_size = avpkt->size;
2195 GetBitContext gb;
2196 int buf_consumed;
2197 int buf_offset;
2198 int err;
2199
2200 init_get_bits(&gb, buf, buf_size * 8);
2201
2202 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2203 return err;
2204
2205 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2206 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2207 if (buf[buf_offset])
2208 break;
2209
2210 return buf_size > buf_offset ? buf_consumed : buf_size;
2211 }
2212
2213 static av_cold int aac_decode_close(AVCodecContext *avctx)
2214 {
2215 AACContext *ac = avctx->priv_data;
2216 int i, type;
2217
2218 for (i = 0; i < MAX_ELEM_ID; i++) {
2219 for (type = 0; type < 4; type++) {
2220 if (ac->che[type][i])
2221 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2222 av_freep(&ac->che[type][i]);
2223 }
2224 }
2225
2226 ff_mdct_end(&ac->mdct);
2227 ff_mdct_end(&ac->mdct_small);
2228 ff_mdct_end(&ac->mdct_ltp);
2229 return 0;
2230 }
2231
2232
2233 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2234
2235 struct LATMContext {
2236 AACContext aac_ctx; ///< containing AACContext
2237 int initialized; ///< initilized after a valid extradata was seen
2238
2239 // parser data
2240 int audio_mux_version_A; ///< LATM syntax version
2241 int frame_length_type; ///< 0/1 variable/fixed frame length
2242 int frame_length; ///< frame length for fixed frame length
2243 };
2244
2245 static inline uint32_t latm_get_value(GetBitContext *b)
2246 {
2247 int length = get_bits(b, 2);
2248
2249 return get_bits_long(b, (length+1)*8);
2250 }
2251
2252 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2253 GetBitContext *gb)
2254 {
2255 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2256 MPEG4AudioConfig m4ac;
2257 int config_start_bit = get_bits_count(gb);
2258 int bits_consumed, esize;
2259
2260 if (config_start_bit % 8) {
2261 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2262 "config not byte aligned.\n", 1);
2263 return AVERROR_INVALIDDATA;
2264 } else {
2265 bits_consumed =
2266 decode_audio_specific_config(NULL, avctx, &m4ac,
2267 gb->buffer + (config_start_bit / 8),
2268 get_bits_left(gb) / 8);
2269
2270 if (bits_consumed < 0)
2271 return AVERROR_INVALIDDATA;
2272
2273 esize = (bits_consumed+7) / 8;
2274
2275 if (avctx->extradata_size <= esize) {
2276 av_free(avctx->extradata);
2277 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2278 if (!avctx->extradata)
2279 return AVERROR(ENOMEM);
2280 }
2281
2282 avctx->extradata_size = esize;
2283 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2284 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2285
2286 skip_bits_long(gb, bits_consumed);
2287 }
2288
2289 return bits_consumed;
2290 }
2291
2292 static int read_stream_mux_config(struct LATMContext *latmctx,
2293 GetBitContext *gb)
2294 {
2295 int ret, audio_mux_version = get_bits(gb, 1);
2296
2297 latmctx->audio_mux_version_A = 0;
2298 if (audio_mux_version)
2299 latmctx->audio_mux_version_A = get_bits(gb, 1);
2300
2301 if (!latmctx->audio_mux_version_A) {
2302
2303 if (audio_mux_version)
2304 latm_get_value(gb); // taraFullness
2305
2306 skip_bits(gb, 1); // allStreamSameTimeFraming
2307 skip_bits(gb, 6); // numSubFrames
2308 // numPrograms
2309 if (get_bits(gb, 4)) { // numPrograms
2310 av_log_missing_feature(latmctx->aac_ctx.avctx,
2311 "multiple programs are not supported\n", 1);
2312 return AVERROR_PATCHWELCOME;
2313 }
2314
2315 // for each program (which there is only on in DVB)
2316
2317 // for each layer (which there is only on in DVB)
2318 if (get_bits(gb, 3)) { // numLayer
2319 av_log_missing_feature(latmctx->aac_ctx.avctx,
2320 "multiple layers are not supported\n", 1);
2321 return AVERROR_PATCHWELCOME;
2322 }
2323
2324 // for all but first stream: use_same_config = get_bits(gb, 1);
2325 if (!audio_mux_version) {
2326 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2327 return ret;
2328 } else {
2329 int ascLen = latm_get_value(gb);
2330 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2331 return ret;
2332 ascLen -= ret;
2333 skip_bits_long(gb, ascLen);
2334 }
2335
2336 latmctx->frame_length_type = get_bits(gb, 3);
2337 switch (latmctx->frame_length_type) {
2338 case 0:
2339 skip_bits(gb, 8); // latmBufferFullness
2340 break;
2341 case 1:
2342 latmctx->frame_length = get_bits(gb, 9);
2343 break;
2344 case 3:
2345 case 4:
2346 case 5:
2347 skip_bits(gb, 6); // CELP frame length table index
2348 break;
2349 case 6:
2350 case 7:
2351 skip_bits(gb, 1); // HVXC frame length table index
2352 break;
2353 }
2354
2355 if (get_bits(gb, 1)) { // other data
2356 if (audio_mux_version) {
2357 latm_get_value(gb); // other_data_bits
2358 } else {
2359 int esc;
2360 do {
2361 esc = get_bits(gb, 1);
2362 skip_bits(gb, 8);
2363 } while (esc);
2364 }
2365 }
2366
2367 if (get_bits(gb, 1)) // crc present
2368 skip_bits(gb, 8); // config_crc
2369 }
2370
2371 return 0;
2372 }
2373
2374 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2375 {
2376 uint8_t tmp;
2377
2378 if (ctx->frame_length_type == 0) {
2379 int mux_slot_length = 0;
2380 do {
2381 tmp = get_bits(gb, 8);
2382 mux_slot_length += tmp;
2383 } while (tmp == 255);
2384 return mux_slot_length;
2385 } else if (ctx->frame_length_type == 1) {
2386 return ctx->frame_length;
2387 } else if (ctx->frame_length_type == 3 ||
2388 ctx->frame_length_type == 5 ||
2389 ctx->frame_length_type == 7) {
2390 skip_bits(gb, 2); // mux_slot_length_coded
2391 }
2392 return 0;
2393 }
2394
2395 static int read_audio_mux_element(struct LATMContext *latmctx,
2396 GetBitContext *gb)
2397 {
2398 int err;
2399 uint8_t use_same_mux = get_bits(gb, 1);
2400 if (!use_same_mux) {
2401 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2402 return err;
2403 } else if (!latmctx->aac_ctx.avctx->extradata) {
2404 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2405 "no decoder config found\n");
2406 return AVERROR(EAGAIN);
2407 }
2408 if (latmctx->audio_mux_version_A == 0) {
2409 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2410 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2411 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2412 return AVERROR_INVALIDDATA;
2413 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2414 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2415 "frame length mismatch %d << %d\n",
2416 mux_slot_length_bytes * 8, get_bits_left(gb));
2417 return AVERROR_INVALIDDATA;
2418 }
2419 }
2420 return 0;
2421 }
2422
2423
2424 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2425 AVPacket *avpkt)
2426 {
2427 struct LATMContext *latmctx = avctx->priv_data;
2428 int muxlength, err;
2429 GetBitContext gb;
2430
2431 if (avpkt->size == 0)
2432 return 0;
2433
2434 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2435
2436 // check for LOAS sync word
2437 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2438 return AVERROR_INVALIDDATA;
2439
2440 muxlength = get_bits(&gb, 13) + 3;
2441 // not enough data, the parser should have sorted this
2442 if (muxlength > avpkt->size)
2443 return AVERROR_INVALIDDATA;
2444
2445 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2446 return err;
2447
2448 if (!latmctx->initialized) {
2449 if (!avctx->extradata) {
2450 *out_size = 0;
2451 return avpkt->size;
2452 } else {
2453 if ((err = aac_decode_init(avctx)) < 0)
2454 return err;
2455 latmctx->initialized = 1;
2456 }
2457 }
2458
2459 if (show_bits(&gb, 12) == 0xfff) {
2460 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2461 "ADTS header detected, probably as result of configuration "
2462 "misparsing\n");
2463 return AVERROR_INVALIDDATA;
2464 }
2465
2466 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2467 return err;
2468
2469 return muxlength;
2470 }
2471
2472 av_cold static int latm_decode_init(AVCodecContext *avctx)
2473 {
2474 struct LATMContext *latmctx = avctx->priv_data;
2475 int ret;
2476
2477 ret = aac_decode_init(avctx);
2478
2479 if (avctx->extradata_size > 0) {
2480 latmctx->initialized = !ret;
2481 } else {
2482 latmctx->initialized = 0;
2483 }
2484
2485 return ret;
2486 }
2487
2488
2489 AVCodec ff_aac_decoder = {
2490 "aac",
2491 AVMEDIA_TYPE_AUDIO,
2492 CODEC_ID_AAC,
2493 sizeof(AACContext),
2494 aac_decode_init,
2495 NULL,
2496 aac_decode_close,
2497 aac_decode_frame,
2498 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2499 .sample_fmts = (const enum AVSampleFormat[]) {
2500 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2501 },
2502 .channel_layouts = aac_channel_layout,
2503 };
2504
2505 /*
2506 Note: This decoder filter is intended to decode LATM streams transferred
2507 in MPEG transport streams which only contain one program.
2508 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2509 */
2510 AVCodec ff_aac_latm_decoder = {
2511 .name = "aac_latm",
2512 .type = AVMEDIA_TYPE_AUDIO,
2513 .id = CODEC_ID_AAC_LATM,
2514 .priv_data_size = sizeof(struct LATMContext),
2515 .init = latm_decode_init,
2516 .close = aac_decode_close,
2517 .decode = latm_decode_frame,
2518 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2519 .sample_fmts = (const enum AVSampleFormat[]) {
2520 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2521 },
2522 .channel_layouts = aac_channel_layout,
2523 };