cbaecf9e2f9cf8a7d131b85faaa8f47089b05fb7
[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 // -1024 - Compensate wrong IMDCT method.
582 // 60 - Required to scale values to the correct range [-32768,32767]
583 // for float to int16 conversion. (1 << (60 / 4)) == 32768
584 ac->sf_scale = 1. / -1024.;
585 ac->sf_offset = 60;
586
587 ff_aac_tableinit();
588
589 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
590 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
591 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
592 352);
593
594 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
595 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
596 ff_mdct_init(&ac->mdct_ltp, 11, 0, 2.0);
597 // window initialization
598 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
599 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
600 ff_init_ff_sine_windows(10);
601 ff_init_ff_sine_windows( 7);
602
603 cbrt_tableinit();
604
605 return 0;
606 }
607
608 /**
609 * Skip data_stream_element; reference: table 4.10.
610 */
611 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
612 {
613 int byte_align = get_bits1(gb);
614 int count = get_bits(gb, 8);
615 if (count == 255)
616 count += get_bits(gb, 8);
617 if (byte_align)
618 align_get_bits(gb);
619
620 if (get_bits_left(gb) < 8 * count) {
621 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
622 return -1;
623 }
624 skip_bits_long(gb, 8 * count);
625 return 0;
626 }
627
628 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
629 GetBitContext *gb)
630 {
631 int sfb;
632 if (get_bits1(gb)) {
633 ics->predictor_reset_group = get_bits(gb, 5);
634 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
635 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
636 return -1;
637 }
638 }
639 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
640 ics->prediction_used[sfb] = get_bits1(gb);
641 }
642 return 0;
643 }
644
645 /**
646 * Decode Long Term Prediction data; reference: table 4.xx.
647 */
648 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
649 GetBitContext *gb, uint8_t max_sfb)
650 {
651 int sfb;
652
653 ltp->lag = get_bits(gb, 11);
654 ltp->coef = ltp_coef[get_bits(gb, 3)] * ac->sf_scale;
655 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
656 ltp->used[sfb] = get_bits1(gb);
657 }
658
659 /**
660 * Decode Individual Channel Stream info; reference: table 4.6.
661 *
662 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
663 */
664 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
665 GetBitContext *gb, int common_window)
666 {
667 if (get_bits1(gb)) {
668 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
669 memset(ics, 0, sizeof(IndividualChannelStream));
670 return -1;
671 }
672 ics->window_sequence[1] = ics->window_sequence[0];
673 ics->window_sequence[0] = get_bits(gb, 2);
674 ics->use_kb_window[1] = ics->use_kb_window[0];
675 ics->use_kb_window[0] = get_bits1(gb);
676 ics->num_window_groups = 1;
677 ics->group_len[0] = 1;
678 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
679 int i;
680 ics->max_sfb = get_bits(gb, 4);
681 for (i = 0; i < 7; i++) {
682 if (get_bits1(gb)) {
683 ics->group_len[ics->num_window_groups - 1]++;
684 } else {
685 ics->num_window_groups++;
686 ics->group_len[ics->num_window_groups - 1] = 1;
687 }
688 }
689 ics->num_windows = 8;
690 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
691 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
692 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
693 ics->predictor_present = 0;
694 } else {
695 ics->max_sfb = get_bits(gb, 6);
696 ics->num_windows = 1;
697 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
698 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
699 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
700 ics->predictor_present = get_bits1(gb);
701 ics->predictor_reset_group = 0;
702 if (ics->predictor_present) {
703 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
704 if (decode_prediction(ac, ics, gb)) {
705 memset(ics, 0, sizeof(IndividualChannelStream));
706 return -1;
707 }
708 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
709 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
710 memset(ics, 0, sizeof(IndividualChannelStream));
711 return -1;
712 } else {
713 if ((ics->ltp.present = get_bits(gb, 1)))
714 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
715 }
716 }
717 }
718
719 if (ics->max_sfb > ics->num_swb) {
720 av_log(ac->avctx, AV_LOG_ERROR,
721 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
722 ics->max_sfb, ics->num_swb);
723 memset(ics, 0, sizeof(IndividualChannelStream));
724 return -1;
725 }
726
727 return 0;
728 }
729
730 /**
731 * Decode band types (section_data payload); reference: table 4.46.
732 *
733 * @param band_type array of the used band type
734 * @param band_type_run_end array of the last scalefactor band of a band type run
735 *
736 * @return Returns error status. 0 - OK, !0 - error
737 */
738 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
739 int band_type_run_end[120], GetBitContext *gb,
740 IndividualChannelStream *ics)
741 {
742 int g, idx = 0;
743 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
744 for (g = 0; g < ics->num_window_groups; g++) {
745 int k = 0;
746 while (k < ics->max_sfb) {
747 uint8_t sect_end = k;
748 int sect_len_incr;
749 int sect_band_type = get_bits(gb, 4);
750 if (sect_band_type == 12) {
751 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
752 return -1;
753 }
754 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
755 sect_end += sect_len_incr;
756 sect_end += sect_len_incr;
757 if (get_bits_left(gb) < 0) {
758 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
759 return -1;
760 }
761 if (sect_end > ics->max_sfb) {
762 av_log(ac->avctx, AV_LOG_ERROR,
763 "Number of bands (%d) exceeds limit (%d).\n",
764 sect_end, ics->max_sfb);
765 return -1;
766 }
767 for (; k < sect_end; k++) {
768 band_type [idx] = sect_band_type;
769 band_type_run_end[idx++] = sect_end;
770 }
771 }
772 }
773 return 0;
774 }
775
776 /**
777 * Decode scalefactors; reference: table 4.47.
778 *
779 * @param global_gain first scalefactor value as scalefactors are differentially coded
780 * @param band_type array of the used band type
781 * @param band_type_run_end array of the last scalefactor band of a band type run
782 * @param sf array of scalefactors or intensity stereo positions
783 *
784 * @return Returns error status. 0 - OK, !0 - error
785 */
786 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
787 unsigned int global_gain,
788 IndividualChannelStream *ics,
789 enum BandType band_type[120],
790 int band_type_run_end[120])
791 {
792 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
793 int g, i, idx = 0;
794 int offset[3] = { global_gain, global_gain - 90, 0 };
795 int clipped_offset;
796 int noise_flag = 1;
797 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
798 for (g = 0; g < ics->num_window_groups; g++) {
799 for (i = 0; i < ics->max_sfb;) {
800 int run_end = band_type_run_end[idx];
801 if (band_type[idx] == ZERO_BT) {
802 for (; i < run_end; i++, idx++)
803 sf[idx] = 0.;
804 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
805 for (; i < run_end; i++, idx++) {
806 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
807 clipped_offset = av_clip(offset[2], -155, 100);
808 if (offset[2] != clipped_offset) {
809 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
810 "position clipped (%d -> %d).\nIf you heard an "
811 "audible artifact, there may be a bug in the "
812 "decoder. ", offset[2], clipped_offset);
813 }
814 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
815 }
816 } else if (band_type[idx] == NOISE_BT) {
817 for (; i < run_end; i++, idx++) {
818 if (noise_flag-- > 0)
819 offset[1] += get_bits(gb, 9) - 256;
820 else
821 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
822 clipped_offset = av_clip(offset[1], -100, 155);
823 if (offset[2] != clipped_offset) {
824 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
825 "(%d -> %d).\nIf you heard an audible "
826 "artifact, there may be a bug in the decoder. ",
827 offset[1], clipped_offset);
828 }
829 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + sf_offset - 100 + POW_SF2_ZERO];
830 }
831 } else {
832 for (; i < run_end; i++, idx++) {
833 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
834 if (offset[0] > 255U) {
835 av_log(ac->avctx, AV_LOG_ERROR,
836 "%s (%d) out of range.\n", sf_str[0], offset[0]);
837 return -1;
838 }
839 sf[idx] = -ff_aac_pow2sf_tab[offset[0] + sf_offset - 200 + POW_SF2_ZERO];
840 }
841 }
842 }
843 }
844 return 0;
845 }
846
847 /**
848 * Decode pulse data; reference: table 4.7.
849 */
850 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
851 const uint16_t *swb_offset, int num_swb)
852 {
853 int i, pulse_swb;
854 pulse->num_pulse = get_bits(gb, 2) + 1;
855 pulse_swb = get_bits(gb, 6);
856 if (pulse_swb >= num_swb)
857 return -1;
858 pulse->pos[0] = swb_offset[pulse_swb];
859 pulse->pos[0] += get_bits(gb, 5);
860 if (pulse->pos[0] > 1023)
861 return -1;
862 pulse->amp[0] = get_bits(gb, 4);
863 for (i = 1; i < pulse->num_pulse; i++) {
864 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
865 if (pulse->pos[i] > 1023)
866 return -1;
867 pulse->amp[i] = get_bits(gb, 4);
868 }
869 return 0;
870 }
871
872 /**
873 * Decode Temporal Noise Shaping data; reference: table 4.48.
874 *
875 * @return Returns error status. 0 - OK, !0 - error
876 */
877 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
878 GetBitContext *gb, const IndividualChannelStream *ics)
879 {
880 int w, filt, i, coef_len, coef_res, coef_compress;
881 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
882 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
883 for (w = 0; w < ics->num_windows; w++) {
884 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
885 coef_res = get_bits1(gb);
886
887 for (filt = 0; filt < tns->n_filt[w]; filt++) {
888 int tmp2_idx;
889 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
890
891 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
892 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
893 tns->order[w][filt], tns_max_order);
894 tns->order[w][filt] = 0;
895 return -1;
896 }
897 if (tns->order[w][filt]) {
898 tns->direction[w][filt] = get_bits1(gb);
899 coef_compress = get_bits1(gb);
900 coef_len = coef_res + 3 - coef_compress;
901 tmp2_idx = 2 * coef_compress + coef_res;
902
903 for (i = 0; i < tns->order[w][filt]; i++)
904 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
905 }
906 }
907 }
908 }
909 return 0;
910 }
911
912 /**
913 * Decode Mid/Side data; reference: table 4.54.
914 *
915 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
916 * [1] mask is decoded from bitstream; [2] mask is all 1s;
917 * [3] reserved for scalable AAC
918 */
919 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
920 int ms_present)
921 {
922 int idx;
923 if (ms_present == 1) {
924 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
925 cpe->ms_mask[idx] = get_bits1(gb);
926 } else if (ms_present == 2) {
927 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
928 }
929 }
930
931 #ifndef VMUL2
932 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
933 const float *scale)
934 {
935 float s = *scale;
936 *dst++ = v[idx & 15] * s;
937 *dst++ = v[idx>>4 & 15] * s;
938 return dst;
939 }
940 #endif
941
942 #ifndef VMUL4
943 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
944 const float *scale)
945 {
946 float s = *scale;
947 *dst++ = v[idx & 3] * s;
948 *dst++ = v[idx>>2 & 3] * s;
949 *dst++ = v[idx>>4 & 3] * s;
950 *dst++ = v[idx>>6 & 3] * s;
951 return dst;
952 }
953 #endif
954
955 #ifndef VMUL2S
956 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
957 unsigned sign, const float *scale)
958 {
959 union float754 s0, s1;
960
961 s0.f = s1.f = *scale;
962 s0.i ^= sign >> 1 << 31;
963 s1.i ^= sign << 31;
964
965 *dst++ = v[idx & 15] * s0.f;
966 *dst++ = v[idx>>4 & 15] * s1.f;
967
968 return dst;
969 }
970 #endif
971
972 #ifndef VMUL4S
973 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
974 unsigned sign, const float *scale)
975 {
976 unsigned nz = idx >> 12;
977 union float754 s = { .f = *scale };
978 union float754 t;
979
980 t.i = s.i ^ (sign & 1U<<31);
981 *dst++ = v[idx & 3] * t.f;
982
983 sign <<= nz & 1; nz >>= 1;
984 t.i = s.i ^ (sign & 1U<<31);
985 *dst++ = v[idx>>2 & 3] * t.f;
986
987 sign <<= nz & 1; nz >>= 1;
988 t.i = s.i ^ (sign & 1U<<31);
989 *dst++ = v[idx>>4 & 3] * t.f;
990
991 sign <<= nz & 1; nz >>= 1;
992 t.i = s.i ^ (sign & 1U<<31);
993 *dst++ = v[idx>>6 & 3] * t.f;
994
995 return dst;
996 }
997 #endif
998
999 /**
1000 * Decode spectral data; reference: table 4.50.
1001 * Dequantize and scale spectral data; reference: 4.6.3.3.
1002 *
1003 * @param coef array of dequantized, scaled spectral data
1004 * @param sf array of scalefactors or intensity stereo positions
1005 * @param pulse_present set if pulses are present
1006 * @param pulse pointer to pulse data struct
1007 * @param band_type array of the used band type
1008 *
1009 * @return Returns error status. 0 - OK, !0 - error
1010 */
1011 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1012 GetBitContext *gb, const float sf[120],
1013 int pulse_present, const Pulse *pulse,
1014 const IndividualChannelStream *ics,
1015 enum BandType band_type[120])
1016 {
1017 int i, k, g, idx = 0;
1018 const int c = 1024 / ics->num_windows;
1019 const uint16_t *offsets = ics->swb_offset;
1020 float *coef_base = coef;
1021
1022 for (g = 0; g < ics->num_windows; g++)
1023 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1024
1025 for (g = 0; g < ics->num_window_groups; g++) {
1026 unsigned g_len = ics->group_len[g];
1027
1028 for (i = 0; i < ics->max_sfb; i++, idx++) {
1029 const unsigned cbt_m1 = band_type[idx] - 1;
1030 float *cfo = coef + offsets[i];
1031 int off_len = offsets[i + 1] - offsets[i];
1032 int group;
1033
1034 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1035 for (group = 0; group < g_len; group++, cfo+=128) {
1036 memset(cfo, 0, off_len * sizeof(float));
1037 }
1038 } else if (cbt_m1 == NOISE_BT - 1) {
1039 for (group = 0; group < g_len; group++, cfo+=128) {
1040 float scale;
1041 float band_energy;
1042
1043 for (k = 0; k < off_len; k++) {
1044 ac->random_state = lcg_random(ac->random_state);
1045 cfo[k] = ac->random_state;
1046 }
1047
1048 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1049 scale = sf[idx] / sqrtf(band_energy);
1050 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1051 }
1052 } else {
1053 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1054 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1055 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1056 OPEN_READER(re, gb);
1057
1058 switch (cbt_m1 >> 1) {
1059 case 0:
1060 for (group = 0; group < g_len; group++, cfo+=128) {
1061 float *cf = cfo;
1062 int len = off_len;
1063
1064 do {
1065 int code;
1066 unsigned cb_idx;
1067
1068 UPDATE_CACHE(re, gb);
1069 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1070 cb_idx = cb_vector_idx[code];
1071 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1072 } while (len -= 4);
1073 }
1074 break;
1075
1076 case 1:
1077 for (group = 0; group < g_len; group++, cfo+=128) {
1078 float *cf = cfo;
1079 int len = off_len;
1080
1081 do {
1082 int code;
1083 unsigned nnz;
1084 unsigned cb_idx;
1085 uint32_t bits;
1086
1087 UPDATE_CACHE(re, gb);
1088 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1089 cb_idx = cb_vector_idx[code];
1090 nnz = cb_idx >> 8 & 15;
1091 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1092 LAST_SKIP_BITS(re, gb, nnz);
1093 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1094 } while (len -= 4);
1095 }
1096 break;
1097
1098 case 2:
1099 for (group = 0; group < g_len; group++, cfo+=128) {
1100 float *cf = cfo;
1101 int len = off_len;
1102
1103 do {
1104 int code;
1105 unsigned cb_idx;
1106
1107 UPDATE_CACHE(re, gb);
1108 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1109 cb_idx = cb_vector_idx[code];
1110 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1111 } while (len -= 2);
1112 }
1113 break;
1114
1115 case 3:
1116 case 4:
1117 for (group = 0; group < g_len; group++, cfo+=128) {
1118 float *cf = cfo;
1119 int len = off_len;
1120
1121 do {
1122 int code;
1123 unsigned nnz;
1124 unsigned cb_idx;
1125 unsigned sign;
1126
1127 UPDATE_CACHE(re, gb);
1128 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1129 cb_idx = cb_vector_idx[code];
1130 nnz = cb_idx >> 8 & 15;
1131 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1132 LAST_SKIP_BITS(re, gb, nnz);
1133 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1134 } while (len -= 2);
1135 }
1136 break;
1137
1138 default:
1139 for (group = 0; group < g_len; group++, cfo+=128) {
1140 float *cf = cfo;
1141 uint32_t *icf = (uint32_t *) cf;
1142 int len = off_len;
1143
1144 do {
1145 int code;
1146 unsigned nzt, nnz;
1147 unsigned cb_idx;
1148 uint32_t bits;
1149 int j;
1150
1151 UPDATE_CACHE(re, gb);
1152 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1153
1154 if (!code) {
1155 *icf++ = 0;
1156 *icf++ = 0;
1157 continue;
1158 }
1159
1160 cb_idx = cb_vector_idx[code];
1161 nnz = cb_idx >> 12;
1162 nzt = cb_idx >> 8;
1163 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1164 LAST_SKIP_BITS(re, gb, nnz);
1165
1166 for (j = 0; j < 2; j++) {
1167 if (nzt & 1<<j) {
1168 uint32_t b;
1169 int n;
1170 /* The total length of escape_sequence must be < 22 bits according
1171 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1172 UPDATE_CACHE(re, gb);
1173 b = GET_CACHE(re, gb);
1174 b = 31 - av_log2(~b);
1175
1176 if (b > 8) {
1177 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1178 return -1;
1179 }
1180
1181 SKIP_BITS(re, gb, b + 1);
1182 b += 4;
1183 n = (1 << b) + SHOW_UBITS(re, gb, b);
1184 LAST_SKIP_BITS(re, gb, b);
1185 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1186 bits <<= 1;
1187 } else {
1188 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1189 *icf++ = (bits & 1U<<31) | v;
1190 bits <<= !!v;
1191 }
1192 cb_idx >>= 4;
1193 }
1194 } while (len -= 2);
1195
1196 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1197 }
1198 }
1199
1200 CLOSE_READER(re, gb);
1201 }
1202 }
1203 coef += g_len << 7;
1204 }
1205
1206 if (pulse_present) {
1207 idx = 0;
1208 for (i = 0; i < pulse->num_pulse; i++) {
1209 float co = coef_base[ pulse->pos[i] ];
1210 while (offsets[idx + 1] <= pulse->pos[i])
1211 idx++;
1212 if (band_type[idx] != NOISE_BT && sf[idx]) {
1213 float ico = -pulse->amp[i];
1214 if (co) {
1215 co /= sf[idx];
1216 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1217 }
1218 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1219 }
1220 }
1221 }
1222 return 0;
1223 }
1224
1225 static av_always_inline float flt16_round(float pf)
1226 {
1227 union float754 tmp;
1228 tmp.f = pf;
1229 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1230 return tmp.f;
1231 }
1232
1233 static av_always_inline float flt16_even(float pf)
1234 {
1235 union float754 tmp;
1236 tmp.f = pf;
1237 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1238 return tmp.f;
1239 }
1240
1241 static av_always_inline float flt16_trunc(float pf)
1242 {
1243 union float754 pun;
1244 pun.f = pf;
1245 pun.i &= 0xFFFF0000U;
1246 return pun.f;
1247 }
1248
1249 static av_always_inline void predict(PredictorState *ps, float *coef,
1250 float sf_scale, float inv_sf_scale,
1251 int output_enable)
1252 {
1253 const float a = 0.953125; // 61.0 / 64
1254 const float alpha = 0.90625; // 29.0 / 32
1255 float e0, e1;
1256 float pv;
1257 float k1, k2;
1258 float r0 = ps->r0, r1 = ps->r1;
1259 float cor0 = ps->cor0, cor1 = ps->cor1;
1260 float var0 = ps->var0, var1 = ps->var1;
1261
1262 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1263 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1264
1265 pv = flt16_round(k1 * r0 + k2 * r1);
1266 if (output_enable)
1267 *coef += pv * sf_scale;
1268
1269 e0 = *coef * inv_sf_scale;
1270 e1 = e0 - k1 * r0;
1271
1272 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1273 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1274 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1275 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1276
1277 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1278 ps->r0 = flt16_trunc(a * e0);
1279 }
1280
1281 /**
1282 * Apply AAC-Main style frequency domain prediction.
1283 */
1284 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1285 {
1286 int sfb, k;
1287 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1288
1289 if (!sce->ics.predictor_initialized) {
1290 reset_all_predictors(sce->predictor_state);
1291 sce->ics.predictor_initialized = 1;
1292 }
1293
1294 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1295 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1296 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1297 predict(&sce->predictor_state[k], &sce->coeffs[k],
1298 sf_scale, inv_sf_scale,
1299 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1300 }
1301 }
1302 if (sce->ics.predictor_reset_group)
1303 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1304 } else
1305 reset_all_predictors(sce->predictor_state);
1306 }
1307
1308 /**
1309 * Decode an individual_channel_stream payload; reference: table 4.44.
1310 *
1311 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1312 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1313 *
1314 * @return Returns error status. 0 - OK, !0 - error
1315 */
1316 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1317 GetBitContext *gb, int common_window, int scale_flag)
1318 {
1319 Pulse pulse;
1320 TemporalNoiseShaping *tns = &sce->tns;
1321 IndividualChannelStream *ics = &sce->ics;
1322 float *out = sce->coeffs;
1323 int global_gain, pulse_present = 0;
1324
1325 /* This assignment is to silence a GCC warning about the variable being used
1326 * uninitialized when in fact it always is.
1327 */
1328 pulse.num_pulse = 0;
1329
1330 global_gain = get_bits(gb, 8);
1331
1332 if (!common_window && !scale_flag) {
1333 if (decode_ics_info(ac, ics, gb, 0) < 0)
1334 return -1;
1335 }
1336
1337 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1338 return -1;
1339 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1340 return -1;
1341
1342 pulse_present = 0;
1343 if (!scale_flag) {
1344 if ((pulse_present = get_bits1(gb))) {
1345 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1346 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1347 return -1;
1348 }
1349 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1350 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1351 return -1;
1352 }
1353 }
1354 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1355 return -1;
1356 if (get_bits1(gb)) {
1357 av_log_missing_feature(ac->avctx, "SSR", 1);
1358 return -1;
1359 }
1360 }
1361
1362 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1363 return -1;
1364
1365 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1366 apply_prediction(ac, sce);
1367
1368 return 0;
1369 }
1370
1371 /**
1372 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1373 */
1374 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1375 {
1376 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1377 float *ch0 = cpe->ch[0].coeffs;
1378 float *ch1 = cpe->ch[1].coeffs;
1379 int g, i, group, idx = 0;
1380 const uint16_t *offsets = ics->swb_offset;
1381 for (g = 0; g < ics->num_window_groups; g++) {
1382 for (i = 0; i < ics->max_sfb; i++, idx++) {
1383 if (cpe->ms_mask[idx] &&
1384 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1385 for (group = 0; group < ics->group_len[g]; group++) {
1386 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1387 ch1 + group * 128 + offsets[i],
1388 offsets[i+1] - offsets[i]);
1389 }
1390 }
1391 }
1392 ch0 += ics->group_len[g] * 128;
1393 ch1 += ics->group_len[g] * 128;
1394 }
1395 }
1396
1397 /**
1398 * intensity stereo decoding; reference: 4.6.8.2.3
1399 *
1400 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1401 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1402 * [3] reserved for scalable AAC
1403 */
1404 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1405 {
1406 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1407 SingleChannelElement *sce1 = &cpe->ch[1];
1408 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1409 const uint16_t *offsets = ics->swb_offset;
1410 int g, group, i, idx = 0;
1411 int c;
1412 float scale;
1413 for (g = 0; g < ics->num_window_groups; g++) {
1414 for (i = 0; i < ics->max_sfb;) {
1415 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1416 const int bt_run_end = sce1->band_type_run_end[idx];
1417 for (; i < bt_run_end; i++, idx++) {
1418 c = -1 + 2 * (sce1->band_type[idx] - 14);
1419 if (ms_present)
1420 c *= 1 - 2 * cpe->ms_mask[idx];
1421 scale = c * sce1->sf[idx];
1422 for (group = 0; group < ics->group_len[g]; group++)
1423 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1424 coef0 + group * 128 + offsets[i],
1425 scale,
1426 offsets[i + 1] - offsets[i]);
1427 }
1428 } else {
1429 int bt_run_end = sce1->band_type_run_end[idx];
1430 idx += bt_run_end - i;
1431 i = bt_run_end;
1432 }
1433 }
1434 coef0 += ics->group_len[g] * 128;
1435 coef1 += ics->group_len[g] * 128;
1436 }
1437 }
1438
1439 /**
1440 * Decode a channel_pair_element; reference: table 4.4.
1441 *
1442 * @return Returns error status. 0 - OK, !0 - error
1443 */
1444 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1445 {
1446 int i, ret, common_window, ms_present = 0;
1447
1448 common_window = get_bits1(gb);
1449 if (common_window) {
1450 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1451 return -1;
1452 i = cpe->ch[1].ics.use_kb_window[0];
1453 cpe->ch[1].ics = cpe->ch[0].ics;
1454 cpe->ch[1].ics.use_kb_window[1] = i;
1455 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1456 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1457 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1458 ms_present = get_bits(gb, 2);
1459 if (ms_present == 3) {
1460 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1461 return -1;
1462 } else if (ms_present)
1463 decode_mid_side_stereo(cpe, gb, ms_present);
1464 }
1465 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1466 return ret;
1467 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1468 return ret;
1469
1470 if (common_window) {
1471 if (ms_present)
1472 apply_mid_side_stereo(ac, cpe);
1473 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1474 apply_prediction(ac, &cpe->ch[0]);
1475 apply_prediction(ac, &cpe->ch[1]);
1476 }
1477 }
1478
1479 apply_intensity_stereo(ac, cpe, ms_present);
1480 return 0;
1481 }
1482
1483 static const float cce_scale[] = {
1484 1.09050773266525765921, //2^(1/8)
1485 1.18920711500272106672, //2^(1/4)
1486 M_SQRT2,
1487 2,
1488 };
1489
1490 /**
1491 * Decode coupling_channel_element; reference: table 4.8.
1492 *
1493 * @return Returns error status. 0 - OK, !0 - error
1494 */
1495 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1496 {
1497 int num_gain = 0;
1498 int c, g, sfb, ret;
1499 int sign;
1500 float scale;
1501 SingleChannelElement *sce = &che->ch[0];
1502 ChannelCoupling *coup = &che->coup;
1503
1504 coup->coupling_point = 2 * get_bits1(gb);
1505 coup->num_coupled = get_bits(gb, 3);
1506 for (c = 0; c <= coup->num_coupled; c++) {
1507 num_gain++;
1508 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1509 coup->id_select[c] = get_bits(gb, 4);
1510 if (coup->type[c] == TYPE_CPE) {
1511 coup->ch_select[c] = get_bits(gb, 2);
1512 if (coup->ch_select[c] == 3)
1513 num_gain++;
1514 } else
1515 coup->ch_select[c] = 2;
1516 }
1517 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1518
1519 sign = get_bits(gb, 1);
1520 scale = cce_scale[get_bits(gb, 2)];
1521
1522 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1523 return ret;
1524
1525 for (c = 0; c < num_gain; c++) {
1526 int idx = 0;
1527 int cge = 1;
1528 int gain = 0;
1529 float gain_cache = 1.;
1530 if (c) {
1531 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1532 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1533 gain_cache = powf(scale, -gain);
1534 }
1535 if (coup->coupling_point == AFTER_IMDCT) {
1536 coup->gain[c][0] = gain_cache;
1537 } else {
1538 for (g = 0; g < sce->ics.num_window_groups; g++) {
1539 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1540 if (sce->band_type[idx] != ZERO_BT) {
1541 if (!cge) {
1542 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1543 if (t) {
1544 int s = 1;
1545 t = gain += t;
1546 if (sign) {
1547 s -= 2 * (t & 0x1);
1548 t >>= 1;
1549 }
1550 gain_cache = powf(scale, -t) * s;
1551 }
1552 }
1553 coup->gain[c][idx] = gain_cache;
1554 }
1555 }
1556 }
1557 }
1558 }
1559 return 0;
1560 }
1561
1562 /**
1563 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1564 *
1565 * @return Returns number of bytes consumed.
1566 */
1567 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1568 GetBitContext *gb)
1569 {
1570 int i;
1571 int num_excl_chan = 0;
1572
1573 do {
1574 for (i = 0; i < 7; i++)
1575 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1576 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1577
1578 return num_excl_chan / 7;
1579 }
1580
1581 /**
1582 * Decode dynamic range information; reference: table 4.52.
1583 *
1584 * @param cnt length of TYPE_FIL syntactic element in bytes
1585 *
1586 * @return Returns number of bytes consumed.
1587 */
1588 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1589 GetBitContext *gb, int cnt)
1590 {
1591 int n = 1;
1592 int drc_num_bands = 1;
1593 int i;
1594
1595 /* pce_tag_present? */
1596 if (get_bits1(gb)) {
1597 che_drc->pce_instance_tag = get_bits(gb, 4);
1598 skip_bits(gb, 4); // tag_reserved_bits
1599 n++;
1600 }
1601
1602 /* excluded_chns_present? */
1603 if (get_bits1(gb)) {
1604 n += decode_drc_channel_exclusions(che_drc, gb);
1605 }
1606
1607 /* drc_bands_present? */
1608 if (get_bits1(gb)) {
1609 che_drc->band_incr = get_bits(gb, 4);
1610 che_drc->interpolation_scheme = get_bits(gb, 4);
1611 n++;
1612 drc_num_bands += che_drc->band_incr;
1613 for (i = 0; i < drc_num_bands; i++) {
1614 che_drc->band_top[i] = get_bits(gb, 8);
1615 n++;
1616 }
1617 }
1618
1619 /* prog_ref_level_present? */
1620 if (get_bits1(gb)) {
1621 che_drc->prog_ref_level = get_bits(gb, 7);
1622 skip_bits1(gb); // prog_ref_level_reserved_bits
1623 n++;
1624 }
1625
1626 for (i = 0; i < drc_num_bands; i++) {
1627 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1628 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1629 n++;
1630 }
1631
1632 return n;
1633 }
1634
1635 /**
1636 * Decode extension data (incomplete); reference: table 4.51.
1637 *
1638 * @param cnt length of TYPE_FIL syntactic element in bytes
1639 *
1640 * @return Returns number of bytes consumed
1641 */
1642 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1643 ChannelElement *che, enum RawDataBlockType elem_type)
1644 {
1645 int crc_flag = 0;
1646 int res = cnt;
1647 switch (get_bits(gb, 4)) { // extension type
1648 case EXT_SBR_DATA_CRC:
1649 crc_flag++;
1650 case EXT_SBR_DATA:
1651 if (!che) {
1652 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1653 return res;
1654 } else if (!ac->m4ac.sbr) {
1655 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1656 skip_bits_long(gb, 8 * cnt - 4);
1657 return res;
1658 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1659 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1660 skip_bits_long(gb, 8 * cnt - 4);
1661 return res;
1662 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1663 ac->m4ac.sbr = 1;
1664 ac->m4ac.ps = 1;
1665 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1666 } else {
1667 ac->m4ac.sbr = 1;
1668 }
1669 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1670 break;
1671 case EXT_DYNAMIC_RANGE:
1672 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1673 break;
1674 case EXT_FILL:
1675 case EXT_FILL_DATA:
1676 case EXT_DATA_ELEMENT:
1677 default:
1678 skip_bits_long(gb, 8 * cnt - 4);
1679 break;
1680 };
1681 return res;
1682 }
1683
1684 /**
1685 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1686 *
1687 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1688 * @param coef spectral coefficients
1689 */
1690 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1691 IndividualChannelStream *ics, int decode)
1692 {
1693 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1694 int w, filt, m, i;
1695 int bottom, top, order, start, end, size, inc;
1696 float lpc[TNS_MAX_ORDER];
1697 float tmp[TNS_MAX_ORDER];
1698
1699 for (w = 0; w < ics->num_windows; w++) {
1700 bottom = ics->num_swb;
1701 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1702 top = bottom;
1703 bottom = FFMAX(0, top - tns->length[w][filt]);
1704 order = tns->order[w][filt];
1705 if (order == 0)
1706 continue;
1707
1708 // tns_decode_coef
1709 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1710
1711 start = ics->swb_offset[FFMIN(bottom, mmm)];
1712 end = ics->swb_offset[FFMIN( top, mmm)];
1713 if ((size = end - start) <= 0)
1714 continue;
1715 if (tns->direction[w][filt]) {
1716 inc = -1;
1717 start = end - 1;
1718 } else {
1719 inc = 1;
1720 }
1721 start += w * 128;
1722
1723 if (decode) {
1724 // ar filter
1725 for (m = 0; m < size; m++, start += inc)
1726 for (i = 1; i <= FFMIN(m, order); i++)
1727 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1728 } else {
1729 // ma filter
1730 for (m = 0; m < size; m++, start += inc) {
1731 tmp[0] = coef[start];
1732 for (i = 1; i <= FFMIN(m, order); i++)
1733 coef[start] += tmp[i] * lpc[i - 1];
1734 for (i = order; i > 0; i--)
1735 tmp[i] = tmp[i - 1];
1736 }
1737 }
1738 }
1739 }
1740 }
1741
1742 /**
1743 * Apply windowing and MDCT to obtain the spectral
1744 * coefficient from the predicted sample by LTP.
1745 */
1746 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1747 float *in, IndividualChannelStream *ics)
1748 {
1749 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1750 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1751 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1752 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1753
1754 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1755 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1756 } else {
1757 memset(in, 0, 448 * sizeof(float));
1758 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1759 memcpy(in + 576, in + 576, 448 * sizeof(float));
1760 }
1761 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1762 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1763 } else {
1764 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1765 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1766 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1767 }
1768 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1769 }
1770
1771 /**
1772 * Apply the long term prediction
1773 */
1774 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1775 {
1776 const LongTermPrediction *ltp = &sce->ics.ltp;
1777 const uint16_t *offsets = sce->ics.swb_offset;
1778 int i, sfb;
1779
1780 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1781 float *predTime = sce->ret;
1782 float *predFreq = ac->buf_mdct;
1783 int16_t num_samples = 2048;
1784
1785 if (ltp->lag < 1024)
1786 num_samples = ltp->lag + 1024;
1787 for (i = 0; i < num_samples; i++)
1788 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1789 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1790
1791 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1792
1793 if (sce->tns.present)
1794 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1795
1796 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1797 if (ltp->used[sfb])
1798 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1799 sce->coeffs[i] += predFreq[i];
1800 }
1801 }
1802
1803 /**
1804 * Update the LTP buffer for next frame
1805 */
1806 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1807 {
1808 IndividualChannelStream *ics = &sce->ics;
1809 float *saved = sce->saved;
1810 float *saved_ltp = sce->coeffs;
1811 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1812 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1813 int i;
1814
1815 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1816 memcpy(saved_ltp, saved, 512 * sizeof(float));
1817 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1818 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1819 for (i = 0; i < 64; i++)
1820 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1821 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1822 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1823 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1824 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1825 for (i = 0; i < 64; i++)
1826 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1827 } else { // LONG_STOP or ONLY_LONG
1828 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1829 for (i = 0; i < 512; i++)
1830 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1831 }
1832
1833 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1834 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1835 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1836 }
1837
1838 /**
1839 * Conduct IMDCT and windowing.
1840 */
1841 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1842 {
1843 IndividualChannelStream *ics = &sce->ics;
1844 float *in = sce->coeffs;
1845 float *out = sce->ret;
1846 float *saved = sce->saved;
1847 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1848 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1849 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1850 float *buf = ac->buf_mdct;
1851 float *temp = ac->temp;
1852 int i;
1853
1854 // imdct
1855 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1856 for (i = 0; i < 1024; i += 128)
1857 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1858 } else
1859 ac->mdct.imdct_half(&ac->mdct, buf, in);
1860
1861 /* window overlapping
1862 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1863 * and long to short transitions are considered to be short to short
1864 * transitions. This leaves just two cases (long to long and short to short)
1865 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1866 */
1867 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1868 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1869 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1870 } else {
1871 memcpy( out, saved, 448 * sizeof(float));
1872
1873 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1874 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1875 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1876 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1877 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1878 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1879 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1880 } else {
1881 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1882 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1883 }
1884 }
1885
1886 // buffer update
1887 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1888 memcpy( saved, temp + 64, 64 * sizeof(float));
1889 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1890 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1891 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1892 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1893 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1894 memcpy( saved, buf + 512, 448 * sizeof(float));
1895 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1896 } else { // LONG_STOP or ONLY_LONG
1897 memcpy( saved, buf + 512, 512 * sizeof(float));
1898 }
1899 }
1900
1901 /**
1902 * Apply dependent channel coupling (applied before IMDCT).
1903 *
1904 * @param index index into coupling gain array
1905 */
1906 static void apply_dependent_coupling(AACContext *ac,
1907 SingleChannelElement *target,
1908 ChannelElement *cce, int index)
1909 {
1910 IndividualChannelStream *ics = &cce->ch[0].ics;
1911 const uint16_t *offsets = ics->swb_offset;
1912 float *dest = target->coeffs;
1913 const float *src = cce->ch[0].coeffs;
1914 int g, i, group, k, idx = 0;
1915 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1916 av_log(ac->avctx, AV_LOG_ERROR,
1917 "Dependent coupling is not supported together with LTP\n");
1918 return;
1919 }
1920 for (g = 0; g < ics->num_window_groups; g++) {
1921 for (i = 0; i < ics->max_sfb; i++, idx++) {
1922 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1923 const float gain = cce->coup.gain[index][idx];
1924 for (group = 0; group < ics->group_len[g]; group++) {
1925 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1926 // XXX dsputil-ize
1927 dest[group * 128 + k] += gain * src[group * 128 + k];
1928 }
1929 }
1930 }
1931 }
1932 dest += ics->group_len[g] * 128;
1933 src += ics->group_len[g] * 128;
1934 }
1935 }
1936
1937 /**
1938 * Apply independent channel coupling (applied after IMDCT).
1939 *
1940 * @param index index into coupling gain array
1941 */
1942 static void apply_independent_coupling(AACContext *ac,
1943 SingleChannelElement *target,
1944 ChannelElement *cce, int index)
1945 {
1946 int i;
1947 const float gain = cce->coup.gain[index][0];
1948 const float *src = cce->ch[0].ret;
1949 float *dest = target->ret;
1950 const int len = 1024 << (ac->m4ac.sbr == 1);
1951
1952 for (i = 0; i < len; i++)
1953 dest[i] += gain * src[i];
1954 }
1955
1956 /**
1957 * channel coupling transformation interface
1958 *
1959 * @param apply_coupling_method pointer to (in)dependent coupling function
1960 */
1961 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1962 enum RawDataBlockType type, int elem_id,
1963 enum CouplingPoint coupling_point,
1964 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1965 {
1966 int i, c;
1967
1968 for (i = 0; i < MAX_ELEM_ID; i++) {
1969 ChannelElement *cce = ac->che[TYPE_CCE][i];
1970 int index = 0;
1971
1972 if (cce && cce->coup.coupling_point == coupling_point) {
1973 ChannelCoupling *coup = &cce->coup;
1974
1975 for (c = 0; c <= coup->num_coupled; c++) {
1976 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1977 if (coup->ch_select[c] != 1) {
1978 apply_coupling_method(ac, &cc->ch[0], cce, index);
1979 if (coup->ch_select[c] != 0)
1980 index++;
1981 }
1982 if (coup->ch_select[c] != 2)
1983 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1984 } else
1985 index += 1 + (coup->ch_select[c] == 3);
1986 }
1987 }
1988 }
1989 }
1990
1991 /**
1992 * Convert spectral data to float samples, applying all supported tools as appropriate.
1993 */
1994 static void spectral_to_sample(AACContext *ac)
1995 {
1996 int i, type;
1997 for (type = 3; type >= 0; type--) {
1998 for (i = 0; i < MAX_ELEM_ID; i++) {
1999 ChannelElement *che = ac->che[type][i];
2000 if (che) {
2001 if (type <= TYPE_CPE)
2002 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2003 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2004 if (che->ch[0].ics.predictor_present) {
2005 if (che->ch[0].ics.ltp.present)
2006 apply_ltp(ac, &che->ch[0]);
2007 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2008 apply_ltp(ac, &che->ch[1]);
2009 }
2010 }
2011 if (che->ch[0].tns.present)
2012 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2013 if (che->ch[1].tns.present)
2014 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2015 if (type <= TYPE_CPE)
2016 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2017 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2018 imdct_and_windowing(ac, &che->ch[0]);
2019 if (ac->m4ac.object_type == AOT_AAC_LTP)
2020 update_ltp(ac, &che->ch[0]);
2021 if (type == TYPE_CPE) {
2022 imdct_and_windowing(ac, &che->ch[1]);
2023 if (ac->m4ac.object_type == AOT_AAC_LTP)
2024 update_ltp(ac, &che->ch[1]);
2025 }
2026 if (ac->m4ac.sbr > 0) {
2027 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2028 }
2029 }
2030 if (type <= TYPE_CCE)
2031 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2032 }
2033 }
2034 }
2035 }
2036
2037 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2038 {
2039 int size;
2040 AACADTSHeaderInfo hdr_info;
2041
2042 size = ff_aac_parse_header(gb, &hdr_info);
2043 if (size > 0) {
2044 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2045 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2046 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2047 ac->m4ac.chan_config = hdr_info.chan_config;
2048 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2049 return -7;
2050 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2051 return -7;
2052 } else if (ac->output_configured != OC_LOCKED) {
2053 ac->output_configured = OC_NONE;
2054 }
2055 if (ac->output_configured != OC_LOCKED) {
2056 ac->m4ac.sbr = -1;
2057 ac->m4ac.ps = -1;
2058 }
2059 ac->m4ac.sample_rate = hdr_info.sample_rate;
2060 ac->m4ac.sampling_index = hdr_info.sampling_index;
2061 ac->m4ac.object_type = hdr_info.object_type;
2062 if (!ac->avctx->sample_rate)
2063 ac->avctx->sample_rate = hdr_info.sample_rate;
2064 if (hdr_info.num_aac_frames == 1) {
2065 if (!hdr_info.crc_absent)
2066 skip_bits(gb, 16);
2067 } else {
2068 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2069 return -1;
2070 }
2071 }
2072 return size;
2073 }
2074
2075 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2076 int *data_size, GetBitContext *gb)
2077 {
2078 AACContext *ac = avctx->priv_data;
2079 ChannelElement *che = NULL, *che_prev = NULL;
2080 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2081 int err, elem_id, data_size_tmp;
2082 int samples = 0, multiplier;
2083
2084 if (show_bits(gb, 12) == 0xfff) {
2085 if (parse_adts_frame_header(ac, gb) < 0) {
2086 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2087 return -1;
2088 }
2089 if (ac->m4ac.sampling_index > 12) {
2090 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2091 return -1;
2092 }
2093 }
2094
2095 ac->tags_mapped = 0;
2096 // parse
2097 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2098 elem_id = get_bits(gb, 4);
2099
2100 if (elem_type < TYPE_DSE) {
2101 if (!(che=get_che(ac, elem_type, elem_id))) {
2102 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2103 elem_type, elem_id);
2104 return -1;
2105 }
2106 samples = 1024;
2107 }
2108
2109 switch (elem_type) {
2110
2111 case TYPE_SCE:
2112 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2113 break;
2114
2115 case TYPE_CPE:
2116 err = decode_cpe(ac, gb, che);
2117 break;
2118
2119 case TYPE_CCE:
2120 err = decode_cce(ac, gb, che);
2121 break;
2122
2123 case TYPE_LFE:
2124 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2125 break;
2126
2127 case TYPE_DSE:
2128 err = skip_data_stream_element(ac, gb);
2129 break;
2130
2131 case TYPE_PCE: {
2132 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2133 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2134 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2135 break;
2136 if (ac->output_configured > OC_TRIAL_PCE)
2137 av_log(avctx, AV_LOG_ERROR,
2138 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2139 else
2140 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2141 break;
2142 }
2143
2144 case TYPE_FIL:
2145 if (elem_id == 15)
2146 elem_id += get_bits(gb, 8) - 1;
2147 if (get_bits_left(gb) < 8 * elem_id) {
2148 av_log(avctx, AV_LOG_ERROR, overread_err);
2149 return -1;
2150 }
2151 while (elem_id > 0)
2152 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2153 err = 0; /* FIXME */
2154 break;
2155
2156 default:
2157 err = -1; /* should not happen, but keeps compiler happy */
2158 break;
2159 }
2160
2161 che_prev = che;
2162 elem_type_prev = elem_type;
2163
2164 if (err)
2165 return err;
2166
2167 if (get_bits_left(gb) < 3) {
2168 av_log(avctx, AV_LOG_ERROR, overread_err);
2169 return -1;
2170 }
2171 }
2172
2173 spectral_to_sample(ac);
2174
2175 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2176 samples <<= multiplier;
2177 if (ac->output_configured < OC_LOCKED) {
2178 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2179 avctx->frame_size = samples;
2180 }
2181
2182 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2183 if (*data_size < data_size_tmp) {
2184 av_log(avctx, AV_LOG_ERROR,
2185 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2186 *data_size, data_size_tmp);
2187 return -1;
2188 }
2189 *data_size = data_size_tmp;
2190
2191 if (samples)
2192 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2193
2194 if (ac->output_configured)
2195 ac->output_configured = OC_LOCKED;
2196
2197 return 0;
2198 }
2199
2200 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2201 int *data_size, AVPacket *avpkt)
2202 {
2203 const uint8_t *buf = avpkt->data;
2204 int buf_size = avpkt->size;
2205 GetBitContext gb;
2206 int buf_consumed;
2207 int buf_offset;
2208 int err;
2209
2210 init_get_bits(&gb, buf, buf_size * 8);
2211
2212 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2213 return err;
2214
2215 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2216 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2217 if (buf[buf_offset])
2218 break;
2219
2220 return buf_size > buf_offset ? buf_consumed : buf_size;
2221 }
2222
2223 static av_cold int aac_decode_close(AVCodecContext *avctx)
2224 {
2225 AACContext *ac = avctx->priv_data;
2226 int i, type;
2227
2228 for (i = 0; i < MAX_ELEM_ID; i++) {
2229 for (type = 0; type < 4; type++) {
2230 if (ac->che[type][i])
2231 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2232 av_freep(&ac->che[type][i]);
2233 }
2234 }
2235
2236 ff_mdct_end(&ac->mdct);
2237 ff_mdct_end(&ac->mdct_small);
2238 ff_mdct_end(&ac->mdct_ltp);
2239 return 0;
2240 }
2241
2242
2243 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2244
2245 struct LATMContext {
2246 AACContext aac_ctx; ///< containing AACContext
2247 int initialized; ///< initilized after a valid extradata was seen
2248
2249 // parser data
2250 int audio_mux_version_A; ///< LATM syntax version
2251 int frame_length_type; ///< 0/1 variable/fixed frame length
2252 int frame_length; ///< frame length for fixed frame length
2253 };
2254
2255 static inline uint32_t latm_get_value(GetBitContext *b)
2256 {
2257 int length = get_bits(b, 2);
2258
2259 return get_bits_long(b, (length+1)*8);
2260 }
2261
2262 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2263 GetBitContext *gb)
2264 {
2265 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2266 MPEG4AudioConfig m4ac;
2267 int config_start_bit = get_bits_count(gb);
2268 int bits_consumed, esize;
2269
2270 if (config_start_bit % 8) {
2271 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2272 "config not byte aligned.\n", 1);
2273 return AVERROR_INVALIDDATA;
2274 } else {
2275 bits_consumed =
2276 decode_audio_specific_config(NULL, avctx, &m4ac,
2277 gb->buffer + (config_start_bit / 8),
2278 get_bits_left(gb) / 8);
2279
2280 if (bits_consumed < 0)
2281 return AVERROR_INVALIDDATA;
2282
2283 esize = (bits_consumed+7) / 8;
2284
2285 if (avctx->extradata_size <= esize) {
2286 av_free(avctx->extradata);
2287 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2288 if (!avctx->extradata)
2289 return AVERROR(ENOMEM);
2290 }
2291
2292 avctx->extradata_size = esize;
2293 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2294 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2295
2296 skip_bits_long(gb, bits_consumed);
2297 }
2298
2299 return bits_consumed;
2300 }
2301
2302 static int read_stream_mux_config(struct LATMContext *latmctx,
2303 GetBitContext *gb)
2304 {
2305 int ret, audio_mux_version = get_bits(gb, 1);
2306
2307 latmctx->audio_mux_version_A = 0;
2308 if (audio_mux_version)
2309 latmctx->audio_mux_version_A = get_bits(gb, 1);
2310
2311 if (!latmctx->audio_mux_version_A) {
2312
2313 if (audio_mux_version)
2314 latm_get_value(gb); // taraFullness
2315
2316 skip_bits(gb, 1); // allStreamSameTimeFraming
2317 skip_bits(gb, 6); // numSubFrames
2318 // numPrograms
2319 if (get_bits(gb, 4)) { // numPrograms
2320 av_log_missing_feature(latmctx->aac_ctx.avctx,
2321 "multiple programs are not supported\n", 1);
2322 return AVERROR_PATCHWELCOME;
2323 }
2324
2325 // for each program (which there is only on in DVB)
2326
2327 // for each layer (which there is only on in DVB)
2328 if (get_bits(gb, 3)) { // numLayer
2329 av_log_missing_feature(latmctx->aac_ctx.avctx,
2330 "multiple layers are not supported\n", 1);
2331 return AVERROR_PATCHWELCOME;
2332 }
2333
2334 // for all but first stream: use_same_config = get_bits(gb, 1);
2335 if (!audio_mux_version) {
2336 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2337 return ret;
2338 } else {
2339 int ascLen = latm_get_value(gb);
2340 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2341 return ret;
2342 ascLen -= ret;
2343 skip_bits_long(gb, ascLen);
2344 }
2345
2346 latmctx->frame_length_type = get_bits(gb, 3);
2347 switch (latmctx->frame_length_type) {
2348 case 0:
2349 skip_bits(gb, 8); // latmBufferFullness
2350 break;
2351 case 1:
2352 latmctx->frame_length = get_bits(gb, 9);
2353 break;
2354 case 3:
2355 case 4:
2356 case 5:
2357 skip_bits(gb, 6); // CELP frame length table index
2358 break;
2359 case 6:
2360 case 7:
2361 skip_bits(gb, 1); // HVXC frame length table index
2362 break;
2363 }
2364
2365 if (get_bits(gb, 1)) { // other data
2366 if (audio_mux_version) {
2367 latm_get_value(gb); // other_data_bits
2368 } else {
2369 int esc;
2370 do {
2371 esc = get_bits(gb, 1);
2372 skip_bits(gb, 8);
2373 } while (esc);
2374 }
2375 }
2376
2377 if (get_bits(gb, 1)) // crc present
2378 skip_bits(gb, 8); // config_crc
2379 }
2380
2381 return 0;
2382 }
2383
2384 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2385 {
2386 uint8_t tmp;
2387
2388 if (ctx->frame_length_type == 0) {
2389 int mux_slot_length = 0;
2390 do {
2391 tmp = get_bits(gb, 8);
2392 mux_slot_length += tmp;
2393 } while (tmp == 255);
2394 return mux_slot_length;
2395 } else if (ctx->frame_length_type == 1) {
2396 return ctx->frame_length;
2397 } else if (ctx->frame_length_type == 3 ||
2398 ctx->frame_length_type == 5 ||
2399 ctx->frame_length_type == 7) {
2400 skip_bits(gb, 2); // mux_slot_length_coded
2401 }
2402 return 0;
2403 }
2404
2405 static int read_audio_mux_element(struct LATMContext *latmctx,
2406 GetBitContext *gb)
2407 {
2408 int err;
2409 uint8_t use_same_mux = get_bits(gb, 1);
2410 if (!use_same_mux) {
2411 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2412 return err;
2413 } else if (!latmctx->aac_ctx.avctx->extradata) {
2414 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2415 "no decoder config found\n");
2416 return AVERROR(EAGAIN);
2417 }
2418 if (latmctx->audio_mux_version_A == 0) {
2419 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2420 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2421 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2422 return AVERROR_INVALIDDATA;
2423 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2424 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2425 "frame length mismatch %d << %d\n",
2426 mux_slot_length_bytes * 8, get_bits_left(gb));
2427 return AVERROR_INVALIDDATA;
2428 }
2429 }
2430 return 0;
2431 }
2432
2433
2434 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2435 AVPacket *avpkt)
2436 {
2437 struct LATMContext *latmctx = avctx->priv_data;
2438 int muxlength, err;
2439 GetBitContext gb;
2440
2441 if (avpkt->size == 0)
2442 return 0;
2443
2444 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2445
2446 // check for LOAS sync word
2447 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2448 return AVERROR_INVALIDDATA;
2449
2450 muxlength = get_bits(&gb, 13) + 3;
2451 // not enough data, the parser should have sorted this
2452 if (muxlength > avpkt->size)
2453 return AVERROR_INVALIDDATA;
2454
2455 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2456 return err;
2457
2458 if (!latmctx->initialized) {
2459 if (!avctx->extradata) {
2460 *out_size = 0;
2461 return avpkt->size;
2462 } else {
2463 if ((err = aac_decode_init(avctx)) < 0)
2464 return err;
2465 latmctx->initialized = 1;
2466 }
2467 }
2468
2469 if (show_bits(&gb, 12) == 0xfff) {
2470 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2471 "ADTS header detected, probably as result of configuration "
2472 "misparsing\n");
2473 return AVERROR_INVALIDDATA;
2474 }
2475
2476 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2477 return err;
2478
2479 return muxlength;
2480 }
2481
2482 av_cold static int latm_decode_init(AVCodecContext *avctx)
2483 {
2484 struct LATMContext *latmctx = avctx->priv_data;
2485 int ret;
2486
2487 ret = aac_decode_init(avctx);
2488
2489 if (avctx->extradata_size > 0) {
2490 latmctx->initialized = !ret;
2491 } else {
2492 latmctx->initialized = 0;
2493 }
2494
2495 return ret;
2496 }
2497
2498
2499 AVCodec ff_aac_decoder = {
2500 "aac",
2501 AVMEDIA_TYPE_AUDIO,
2502 CODEC_ID_AAC,
2503 sizeof(AACContext),
2504 aac_decode_init,
2505 NULL,
2506 aac_decode_close,
2507 aac_decode_frame,
2508 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2509 .sample_fmts = (const enum AVSampleFormat[]) {
2510 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2511 },
2512 .channel_layouts = aac_channel_layout,
2513 };
2514
2515 /*
2516 Note: This decoder filter is intended to decode LATM streams transferred
2517 in MPEG transport streams which only contain one program.
2518 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2519 */
2520 AVCodec ff_aac_latm_decoder = {
2521 .name = "aac_latm",
2522 .type = AVMEDIA_TYPE_AUDIO,
2523 .id = CODEC_ID_AAC_LATM,
2524 .priv_data_size = sizeof(struct LATMContext),
2525 .init = latm_decode_init,
2526 .close = aac_decode_close,
2527 .decode = latm_decode_frame,
2528 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2529 .sample_fmts = (const enum AVSampleFormat[]) {
2530 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2531 },
2532 .channel_layouts = aac_channel_layout,
2533 };