Fix issue1658 (trailing space in rtpmap descriptor).
[libav.git] / libavcodec / aac.c
CommitLineData
71e9a1b8
RS
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 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23/**
bad5537e 24 * @file libavcodec/aac.c
71e9a1b8
RS
25 * AAC decoder
26 * @author Oded Shimon ( ods15 ods15 dyndns org )
27 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
28 */
29
30/*
31 * supported tools
32 *
33 * Support? Name
34 * N (code in SoC repo) gain control
35 * Y block switching
36 * Y window shapes - standard
37 * N window shapes - Low Delay
38 * Y filterbank - standard
39 * N (code in SoC repo) filterbank - Scalable Sample Rate
40 * Y Temporal Noise Shaping
41 * N (code in SoC repo) Long Term Prediction
42 * Y intensity stereo
43 * Y channel coupling
7633a041 44 * Y frequency domain prediction
71e9a1b8
RS
45 * Y Perceptual Noise Substitution
46 * Y Mid/Side stereo
47 * N Scalable Inverse AAC Quantization
48 * N Frequency Selective Switch
49 * N upsampling filter
50 * Y quantization & coding - AAC
51 * N quantization & coding - TwinVQ
52 * N quantization & coding - BSAC
53 * N AAC Error Resilience tools
54 * N Error Resilience payload syntax
55 * N Error Protection tool
56 * N CELP
57 * N Silence Compression
58 * N HVXC
59 * N HVXC 4kbits/s VR
60 * N Structured Audio tools
61 * N Structured Audio Sample Bank Format
62 * N MIDI
63 * N Harmonic and Individual Lines plus Noise
64 * N Text-To-Speech Interface
65 * N (in progress) Spectral Band Replication
66 * Y (not in this code) Layer-1
67 * Y (not in this code) Layer-2
68 * Y (not in this code) Layer-3
69 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
70 * N (planned) Parametric Stereo
71 * N Direct Stream Transfer
72 *
73 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
74 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
75 Parametric Stereo.
76 */
77
78
79#include "avcodec.h"
dbbec0c2 80#include "internal.h"
9106a698 81#include "get_bits.h"
71e9a1b8 82#include "dsputil.h"
1be0fc29 83#include "lpc.h"
71e9a1b8
RS
84
85#include "aac.h"
86#include "aactab.h"
cc0591da 87#include "aacdectab.h"
71e9a1b8 88#include "mpeg4audio.h"
158b3912 89#include "aac_parser.h"
71e9a1b8
RS
90
91#include <assert.h>
92#include <errno.h>
93#include <math.h>
94#include <string.h>
95
577d383b
DB
96union float754 {
97 float f;
98 uint32_t i;
99};
4a39ccb4 100
71e9a1b8
RS
101static VLC vlc_scalefactors;
102static VLC vlc_spectral[11];
103
dc0d86fa 104static float cbrt_tab[1<<13];
71e9a1b8 105
577d383b
DB
106static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
107{
bb5c0988
AC
108 if (ac->tag_che_map[type][elem_id]) {
109 return ac->tag_che_map[type][elem_id];
110 }
111 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
112 return NULL;
113 }
114 switch (ac->m4ac.chan_config) {
577d383b
DB
115 case 7:
116 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
117 ac->tags_mapped++;
118 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
119 }
120 case 6:
121 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
122 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
123 encountered such a stream, transfer the LFE[0] element to SCE[1] */
124 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
125 ac->tags_mapped++;
126 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
127 }
128 case 5:
129 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
130 ac->tags_mapped++;
131 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
132 }
133 case 4:
134 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
135 ac->tags_mapped++;
136 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
137 }
138 case 3:
139 case 2:
140 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
141 ac->tags_mapped++;
142 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
143 } else if (ac->m4ac.chan_config == 2) {
bb5c0988 144 return NULL;
577d383b
DB
145 }
146 case 1:
147 if (!ac->tags_mapped && type == TYPE_SCE) {
148 ac->tags_mapped++;
149 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
150 }
151 default:
152 return NULL;
bb5c0988
AC
153 }
154}
155
9cc04edf 156/**
754ff9a7
RS
157 * Check for the channel element in the current channel position configuration.
158 * If it exists, make sure the appropriate element is allocated and map the
159 * channel order to match the internal FFmpeg channel layout.
160 *
161 * @param che_pos current channel position configuration
162 * @param type channel element type
163 * @param id channel element id
164 * @param channels count of the number of channels in the configuration
165 *
166 * @return Returns error status. 0 - OK, !0 - error
167 */
168static int che_configure(AACContext *ac,
169 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
170 int type, int id,
171 int *channels)
172{
173 if (che_pos[type][id]) {
174 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
175 return AVERROR(ENOMEM);
176 if (type != TYPE_CCE) {
177 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
178 if (type == TYPE_CPE) {
179 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
180 }
181 }
182 } else
183 av_freep(&ac->che[type][id]);
184 return 0;
185}
186
187/**
62a57fae
RS
188 * Configure output channel order based on the current program configuration element.
189 *
190 * @param che_pos current channel position configuration
191 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
192 *
193 * @return Returns error status. 0 - OK, !0 - error
194 */
577d383b
DB
195static int output_configure(AACContext *ac,
196 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
197 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
981b8fd7 198 int channel_config, enum OCStatus oc_type)
577d383b 199{
62a57fae 200 AVCodecContext *avctx = ac->avccontext;
754ff9a7 201 int i, type, channels = 0, ret;
62a57fae 202
62a57fae
RS
203 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
204
e22da6b6
RS
205 if (channel_config) {
206 for (i = 0; i < tags_per_config[channel_config]; i++) {
754ff9a7
RS
207 if ((ret = che_configure(ac, che_pos,
208 aac_channel_layout_map[channel_config - 1][i][0],
209 aac_channel_layout_map[channel_config - 1][i][1],
210 &channels)))
211 return ret;
e22da6b6
RS
212 }
213
214 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
215 ac->tags_mapped = 0;
216
217 avctx->channel_layout = aac_channel_layout[channel_config - 1];
218 } else {
2309923c
RS
219 /* Allocate or free elements depending on if they are in the
220 * current program configuration.
221 *
222 * Set up default 1:1 output mapping.
223 *
224 * For a 5.1 stream the output order will be:
225 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
226 */
227
228 for (i = 0; i < MAX_ELEM_ID; i++) {
229 for (type = 0; type < 4; type++) {
754ff9a7
RS
230 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
231 return ret;
2309923c 232 }
62a57fae 233 }
62a57fae 234
bb5c0988 235 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
577d383b 236 ac->tags_mapped = 4 * MAX_ELEM_ID;
e22da6b6
RS
237
238 avctx->channel_layout = 0;
bb5c0988
AC
239 }
240
62a57fae 241 avctx->channels = channels;
bb5c0988 242
981b8fd7 243 ac->output_configured = oc_type;
6308765c 244
62a57fae
RS
245 return 0;
246}
247
248/**
9cc04edf
RS
249 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
250 *
251 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
252 * @param sce_map mono (Single Channel Element) map
253 * @param type speaker type/position for these channels
254 */
255static void decode_channel_map(enum ChannelPosition *cpe_map,
577d383b
DB
256 enum ChannelPosition *sce_map,
257 enum ChannelPosition type,
258 GetBitContext *gb, int n)
259{
260 while (n--) {
9cc04edf
RS
261 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
262 map[get_bits(gb, 4)] = type;
263 }
264}
265
266/**
267 * Decode program configuration element; reference: table 4.2.
268 *
269 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
270 *
271 * @return Returns error status. 0 - OK, !0 - error
272 */
577d383b
DB
273static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
274 GetBitContext *gb)
275{
99665a21 276 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
9cc04edf
RS
277
278 skip_bits(gb, 2); // object_type
279
99665a21 280 sampling_index = get_bits(gb, 4);
401a9950
AC
281 if (ac->m4ac.sampling_index != sampling_index)
282 av_log(ac->avccontext, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
283
71e9a1b8
RS
284 num_front = get_bits(gb, 4);
285 num_side = get_bits(gb, 4);
286 num_back = get_bits(gb, 4);
287 num_lfe = get_bits(gb, 2);
288 num_assoc_data = get_bits(gb, 3);
289 num_cc = get_bits(gb, 4);
290
cc0591da
RS
291 if (get_bits1(gb))
292 skip_bits(gb, 4); // mono_mixdown_tag
293 if (get_bits1(gb))
294 skip_bits(gb, 4); // stereo_mixdown_tag
71e9a1b8 295
cc0591da
RS
296 if (get_bits1(gb))
297 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
71e9a1b8 298
cc0591da
RS
299 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
300 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
301 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
302 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
71e9a1b8
RS
303
304 skip_bits_long(gb, 4 * num_assoc_data);
305
cc0591da 306 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
71e9a1b8
RS
307
308 align_get_bits(gb);
309
310 /* comment field, first byte is length */
311 skip_bits_long(gb, 8 * get_bits(gb, 8));
cc0591da
RS
312 return 0;
313}
71e9a1b8 314
9cc04edf
RS
315/**
316 * Set up channel positions based on a default channel configuration
317 * as specified in table 1.17.
318 *
319 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
320 *
321 * @return Returns error status. 0 - OK, !0 - error
322 */
577d383b
DB
323static int set_default_channel_config(AACContext *ac,
324 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
325 int channel_config)
9cc04edf 326{
577d383b 327 if (channel_config < 1 || channel_config > 7) {
9cc04edf
RS
328 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
329 channel_config);
330 return -1;
331 }
332
333 /* default channel configurations:
334 *
335 * 1ch : front center (mono)
336 * 2ch : L + R (stereo)
337 * 3ch : front center + L + R
338 * 4ch : front center + L + R + back center
339 * 5ch : front center + L + R + back stereo
340 * 6ch : front center + L + R + back stereo + LFE
341 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
342 */
343
577d383b 344 if (channel_config != 2)
9cc04edf 345 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
577d383b 346 if (channel_config > 1)
9cc04edf 347 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
577d383b 348 if (channel_config == 4)
9cc04edf 349 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
577d383b 350 if (channel_config > 4)
9cc04edf 351 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
577d383b
DB
352 = AAC_CHANNEL_BACK; // back stereo
353 if (channel_config > 5)
9cc04edf 354 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
577d383b 355 if (channel_config == 7)
9cc04edf
RS
356 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
357
358 return 0;
359}
360
62a57fae
RS
361/**
362 * Decode GA "General Audio" specific configuration; reference: table 4.1.
363 *
364 * @return Returns error status. 0 - OK, !0 - error
365 */
577d383b
DB
366static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
367 int channel_config)
368{
62a57fae
RS
369 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
370 int extension_flag, ret;
371
577d383b 372 if (get_bits1(gb)) { // frameLengthFlag
ce863d7f 373 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
9cc04edf
RS
374 return -1;
375 }
376
377 if (get_bits1(gb)) // dependsOnCoreCoder
378 skip_bits(gb, 14); // coreCoderDelay
379 extension_flag = get_bits1(gb);
380
577d383b
DB
381 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
382 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
9cc04edf
RS
383 skip_bits(gb, 3); // layerNr
384
385 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
386 if (channel_config == 0) {
387 skip_bits(gb, 4); // element_instance_tag
577d383b 388 if ((ret = decode_pce(ac, new_che_pos, gb)))
9cc04edf
RS
389 return ret;
390 } else {
577d383b 391 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
9cc04edf
RS
392 return ret;
393 }
4fab6627 394 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
9cc04edf
RS
395 return ret;
396
397 if (extension_flag) {
398 switch (ac->m4ac.object_type) {
577d383b
DB
399 case AOT_ER_BSAC:
400 skip_bits(gb, 5); // numOfSubFrame
401 skip_bits(gb, 11); // layer_length
402 break;
403 case AOT_ER_AAC_LC:
404 case AOT_ER_AAC_LTP:
405 case AOT_ER_AAC_SCALABLE:
406 case AOT_ER_AAC_LD:
407 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
9cc04edf
RS
408 * aacScalefactorDataResilienceFlag
409 * aacSpectralDataResilienceFlag
410 */
577d383b 411 break;
9cc04edf
RS
412 }
413 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
414 }
415 return 0;
416}
417
418/**
419 * Decode audio specific configuration; reference: table 1.13.
420 *
421 * @param data pointer to AVCodecContext extradata
422 * @param data_size size of AVCCodecContext extradata
423 *
424 * @return Returns error status. 0 - OK, !0 - error
425 */
577d383b
DB
426static int decode_audio_specific_config(AACContext *ac, void *data,
427 int data_size)
428{
9cc04edf
RS
429 GetBitContext gb;
430 int i;
431
432 init_get_bits(&gb, data, data_size * 8);
433
577d383b 434 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
9cc04edf 435 return -1;
577d383b 436 if (ac->m4ac.sampling_index > 12) {
9cc04edf
RS
437 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
438 return -1;
439 }
440
441 skip_bits_long(&gb, i);
442
443 switch (ac->m4ac.object_type) {
7633a041 444 case AOT_AAC_MAIN:
9cc04edf
RS
445 case AOT_AAC_LC:
446 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
447 return -1;
448 break;
449 default:
450 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
451 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
452 return -1;
453 }
454 return 0;
455}
456
62a57fae
RS
457/**
458 * linear congruential pseudorandom number generator
459 *
460 * @param previous_val pointer to the current state of the generator
461 *
462 * @return Returns a 32-bit pseudorandom integer
463 */
577d383b
DB
464static av_always_inline int lcg_random(int previous_val)
465{
62a57fae
RS
466 return previous_val * 1664525 + 1013904223;
467}
468
577d383b
DB
469static void reset_predict_state(PredictorState *ps)
470{
471 ps->r0 = 0.0f;
472 ps->r1 = 0.0f;
7633a041
AC
473 ps->cor0 = 0.0f;
474 ps->cor1 = 0.0f;
475 ps->var0 = 1.0f;
476 ps->var1 = 1.0f;
477}
478
577d383b
DB
479static void reset_all_predictors(PredictorState *ps)
480{
7633a041
AC
481 int i;
482 for (i = 0; i < MAX_PREDICTORS; i++)
483 reset_predict_state(&ps[i]);
484}
485
577d383b
DB
486static void reset_predictor_group(PredictorState *ps, int group_num)
487{
7633a041 488 int i;
577d383b 489 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
7633a041
AC
490 reset_predict_state(&ps[i]);
491}
492
577d383b
DB
493static av_cold int aac_decode_init(AVCodecContext *avccontext)
494{
495 AACContext *ac = avccontext->priv_data;
71e9a1b8
RS
496 int i;
497
498 ac->avccontext = avccontext;
499
158b3912 500 if (avccontext->extradata_size > 0) {
577d383b 501 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
158b3912
RS
502 return -1;
503 avccontext->sample_rate = ac->m4ac.sample_rate;
504 } else if (avccontext->channels > 0) {
158b3912 505 ac->m4ac.sample_rate = avccontext->sample_rate;
158b3912 506 }
cc0591da 507
577d383b
DB
508 avccontext->sample_fmt = SAMPLE_FMT_S16;
509 avccontext->frame_size = 1024;
71e9a1b8
RS
510
511 AAC_INIT_VLC_STATIC( 0, 144);
512 AAC_INIT_VLC_STATIC( 1, 114);
513 AAC_INIT_VLC_STATIC( 2, 188);
514 AAC_INIT_VLC_STATIC( 3, 180);
515 AAC_INIT_VLC_STATIC( 4, 172);
516 AAC_INIT_VLC_STATIC( 5, 140);
517 AAC_INIT_VLC_STATIC( 6, 168);
518 AAC_INIT_VLC_STATIC( 7, 114);
519 AAC_INIT_VLC_STATIC( 8, 262);
520 AAC_INIT_VLC_STATIC( 9, 248);
521 AAC_INIT_VLC_STATIC(10, 384);
522
523 dsputil_init(&ac->dsp, avccontext);
524
9cc04edf
RS
525 ac->random_state = 0x1f2e3d4c;
526
71e9a1b8
RS
527 // -1024 - Compensate wrong IMDCT method.
528 // 32768 - Required to scale values to the correct range for the bias method
529 // for float to int16 conversion.
530
144fec83 531 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
577d383b
DB
532 ac->add_bias = 385.0f;
533 ac->sf_scale = 1. / (-1024. * 32768.);
71e9a1b8
RS
534 ac->sf_offset = 0;
535 } else {
577d383b
DB
536 ac->add_bias = 0.0f;
537 ac->sf_scale = 1. / -1024.;
71e9a1b8
RS
538 ac->sf_offset = 60;
539 }
540
b250f9c6 541#if !CONFIG_HARDCODED_TABLES
d0ee5021 542 for (i = 0; i < 428; i++)
577d383b 543 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
71e9a1b8
RS
544#endif /* CONFIG_HARDCODED_TABLES */
545
37d3e066 546 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
577d383b
DB
547 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
548 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
549 352);
71e9a1b8 550
7d485f16
SS
551 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
552 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
9ffd5c1c
RS
553 // window initialization
554 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
555 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
14b86070
RD
556 ff_init_ff_sine_windows(10);
557 ff_init_ff_sine_windows( 7);
9ffd5c1c 558
dc0d86fa
MR
559 if (!cbrt_tab[(1<<13) - 1])
560 for (i = 0; i < 1<<13; i++)
561 cbrt_tab[i] = cbrtf(i) * i;
562
71e9a1b8
RS
563 return 0;
564}
565
9cc04edf
RS
566/**
567 * Skip data_stream_element; reference: table 4.10.
568 */
577d383b
DB
569static void skip_data_stream_element(GetBitContext *gb)
570{
71e9a1b8
RS
571 int byte_align = get_bits1(gb);
572 int count = get_bits(gb, 8);
573 if (count == 255)
574 count += get_bits(gb, 8);
575 if (byte_align)
576 align_get_bits(gb);
577 skip_bits_long(gb, 8 * count);
578}
579
577d383b
DB
580static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
581 GetBitContext *gb)
582{
7633a041
AC
583 int sfb;
584 if (get_bits1(gb)) {
585 ics->predictor_reset_group = get_bits(gb, 5);
586 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
587 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
588 return -1;
589 }
590 }
591 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
592 ics->prediction_used[sfb] = get_bits1(gb);
593 }
594 return 0;
595}
596
71e9a1b8 597/**
9cc04edf
RS
598 * Decode Individual Channel Stream info; reference: table 4.6.
599 *
600 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
601 */
577d383b
DB
602static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
603 GetBitContext *gb, int common_window)
604{
9cc04edf
RS
605 if (get_bits1(gb)) {
606 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
607 memset(ics, 0, sizeof(IndividualChannelStream));
608 return -1;
609 }
610 ics->window_sequence[1] = ics->window_sequence[0];
611 ics->window_sequence[0] = get_bits(gb, 2);
577d383b
DB
612 ics->use_kb_window[1] = ics->use_kb_window[0];
613 ics->use_kb_window[0] = get_bits1(gb);
614 ics->num_window_groups = 1;
615 ics->group_len[0] = 1;
9ffd5c1c
RS
616 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
617 int i;
618 ics->max_sfb = get_bits(gb, 4);
619 for (i = 0; i < 7; i++) {
620 if (get_bits1(gb)) {
577d383b 621 ics->group_len[ics->num_window_groups - 1]++;
9ffd5c1c
RS
622 } else {
623 ics->num_window_groups++;
577d383b 624 ics->group_len[ics->num_window_groups - 1] = 1;
9ffd5c1c
RS
625 }
626 }
577d383b
DB
627 ics->num_windows = 8;
628 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
629 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
630 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
7633a041 631 ics->predictor_present = 0;
9ffd5c1c 632 } else {
577d383b
DB
633 ics->max_sfb = get_bits(gb, 6);
634 ics->num_windows = 1;
635 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
636 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
637 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
638 ics->predictor_present = get_bits1(gb);
7633a041
AC
639 ics->predictor_reset_group = 0;
640 if (ics->predictor_present) {
641 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
642 if (decode_prediction(ac, ics, gb)) {
643 memset(ics, 0, sizeof(IndividualChannelStream));
644 return -1;
645 }
646 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
647 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
648 memset(ics, 0, sizeof(IndividualChannelStream));
649 return -1;
650 } else {
ce863d7f 651 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
8f5aaa6d
RS
652 memset(ics, 0, sizeof(IndividualChannelStream));
653 return -1;
7633a041 654 }
62a57fae
RS
655 }
656 }
657
577d383b 658 if (ics->max_sfb > ics->num_swb) {
62a57fae 659 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b
DB
660 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
661 ics->max_sfb, ics->num_swb);
62a57fae
RS
662 memset(ics, 0, sizeof(IndividualChannelStream));
663 return -1;
664 }
665
9cc04edf
RS
666 return 0;
667}
668
669/**
9cc04edf
RS
670 * Decode band types (section_data payload); reference: table 4.46.
671 *
672 * @param band_type array of the used band type
673 * @param band_type_run_end array of the last scalefactor band of a band type run
674 *
675 * @return Returns error status. 0 - OK, !0 - error
676 */
577d383b
DB
677static int decode_band_types(AACContext *ac, enum BandType band_type[120],
678 int band_type_run_end[120], GetBitContext *gb,
679 IndividualChannelStream *ics)
680{
cc0591da
RS
681 int g, idx = 0;
682 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
683 for (g = 0; g < ics->num_window_groups; g++) {
684 int k = 0;
685 while (k < ics->max_sfb) {
01d19fbc 686 uint8_t sect_end = k;
cc0591da
RS
687 int sect_len_incr;
688 int sect_band_type = get_bits(gb, 4);
689 if (sect_band_type == 12) {
690 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
691 return -1;
692 }
577d383b 693 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
01d19fbc
AC
694 sect_end += sect_len_incr;
695 sect_end += sect_len_incr;
696 if (sect_end > ics->max_sfb) {
cc0591da 697 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 698 "Number of bands (%d) exceeds limit (%d).\n",
01d19fbc 699 sect_end, ics->max_sfb);
cc0591da
RS
700 return -1;
701 }
01d19fbc 702 for (; k < sect_end; k++) {
9ffd5c1c 703 band_type [idx] = sect_band_type;
01d19fbc 704 band_type_run_end[idx++] = sect_end;
9ffd5c1c 705 }
9cc04edf
RS
706 }
707 }
708 return 0;
709}
cc0591da 710
9cc04edf
RS
711/**
712 * Decode scalefactors; reference: table 4.47.
cc0591da 713 *
cc0591da
RS
714 * @param global_gain first scalefactor value as scalefactors are differentially coded
715 * @param band_type array of the used band type
716 * @param band_type_run_end array of the last scalefactor band of a band type run
717 * @param sf array of scalefactors or intensity stereo positions
718 *
719 * @return Returns error status. 0 - OK, !0 - error
720 */
577d383b
DB
721static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
722 unsigned int global_gain,
723 IndividualChannelStream *ics,
724 enum BandType band_type[120],
725 int band_type_run_end[120])
726{
cc0591da
RS
727 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
728 int g, i, idx = 0;
729 int offset[3] = { global_gain, global_gain - 90, 100 };
730 int noise_flag = 1;
731 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
cc0591da
RS
732 for (g = 0; g < ics->num_window_groups; g++) {
733 for (i = 0; i < ics->max_sfb;) {
734 int run_end = band_type_run_end[idx];
735 if (band_type[idx] == ZERO_BT) {
577d383b 736 for (; i < run_end; i++, idx++)
cc0591da 737 sf[idx] = 0.;
577d383b
DB
738 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
739 for (; i < run_end; i++, idx++) {
cc0591da 740 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 741 if (offset[2] > 255U) {
cc0591da 742 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 743 "%s (%d) out of range.\n", sf_str[2], offset[2]);
cc0591da
RS
744 return -1;
745 }
577d383b 746 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
cc0591da 747 }
577d383b
DB
748 } else if (band_type[idx] == NOISE_BT) {
749 for (; i < run_end; i++, idx++) {
750 if (noise_flag-- > 0)
cc0591da
RS
751 offset[1] += get_bits(gb, 9) - 256;
752 else
753 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 754 if (offset[1] > 255U) {
cc0591da 755 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 756 "%s (%d) out of range.\n", sf_str[1], offset[1]);
cc0591da
RS
757 return -1;
758 }
577d383b 759 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
cc0591da 760 }
577d383b
DB
761 } else {
762 for (; i < run_end; i++, idx++) {
cc0591da 763 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 764 if (offset[0] > 255U) {
cc0591da 765 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 766 "%s (%d) out of range.\n", sf_str[0], offset[0]);
cc0591da
RS
767 return -1;
768 }
769 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
cc0591da
RS
770 }
771 }
772 }
773 }
774 return 0;
775}
776
777/**
778 * Decode pulse data; reference: table 4.7.
779 */
577d383b
DB
780static int decode_pulses(Pulse *pulse, GetBitContext *gb,
781 const uint16_t *swb_offset, int num_swb)
782{
aac0eda4 783 int i, pulse_swb;
cc0591da 784 pulse->num_pulse = get_bits(gb, 2) + 1;
aac0eda4
AC
785 pulse_swb = get_bits(gb, 6);
786 if (pulse_swb >= num_swb)
787 return -1;
788 pulse->pos[0] = swb_offset[pulse_swb];
408992ba 789 pulse->pos[0] += get_bits(gb, 5);
aac0eda4
AC
790 if (pulse->pos[0] > 1023)
791 return -1;
848a5815
RS
792 pulse->amp[0] = get_bits(gb, 4);
793 for (i = 1; i < pulse->num_pulse; i++) {
577d383b 794 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
aac0eda4
AC
795 if (pulse->pos[i] > 1023)
796 return -1;
848a5815 797 pulse->amp[i] = get_bits(gb, 4);
cc0591da 798 }
aac0eda4 799 return 0;
cc0591da
RS
800}
801
802/**
1dece0d2
RS
803 * Decode Temporal Noise Shaping data; reference: table 4.48.
804 *
805 * @return Returns error status. 0 - OK, !0 - error
806 */
577d383b
DB
807static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
808 GetBitContext *gb, const IndividualChannelStream *ics)
809{
1dece0d2
RS
810 int w, filt, i, coef_len, coef_res, coef_compress;
811 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
812 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
813 for (w = 0; w < ics->num_windows; w++) {
fbd91d7c 814 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1dece0d2
RS
815 coef_res = get_bits1(gb);
816
65b20b24
RS
817 for (filt = 0; filt < tns->n_filt[w]; filt++) {
818 int tmp2_idx;
577d383b 819 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
65b20b24 820
577d383b 821 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
65b20b24
RS
822 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
823 tns->order[w][filt], tns_max_order);
824 tns->order[w][filt] = 0;
825 return -1;
826 }
51673647 827 if (tns->order[w][filt]) {
35445d29
RS
828 tns->direction[w][filt] = get_bits1(gb);
829 coef_compress = get_bits1(gb);
830 coef_len = coef_res + 3 - coef_compress;
577d383b 831 tmp2_idx = 2 * coef_compress + coef_res;
1dece0d2 832
35445d29
RS
833 for (i = 0; i < tns->order[w][filt]; i++)
834 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
51673647 835 }
65b20b24 836 }
fbd91d7c 837 }
1dece0d2
RS
838 }
839 return 0;
840}
841
842/**
9cc04edf
RS
843 * Decode Mid/Side data; reference: table 4.54.
844 *
845 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
846 * [1] mask is decoded from bitstream; [2] mask is all 1s;
847 * [3] reserved for scalable AAC
848 */
577d383b
DB
849static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
850 int ms_present)
851{
62a57fae
RS
852 int idx;
853 if (ms_present == 1) {
854 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
855 cpe->ms_mask[idx] = get_bits1(gb);
856 } else if (ms_present == 2) {
857 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
858 }
859}
9cc04edf
RS
860
861/**
9ffd5c1c
RS
862 * Decode spectral data; reference: table 4.50.
863 * Dequantize and scale spectral data; reference: 4.6.3.3.
864 *
865 * @param coef array of dequantized, scaled spectral data
866 * @param sf array of scalefactors or intensity stereo positions
867 * @param pulse_present set if pulses are present
868 * @param pulse pointer to pulse data struct
869 * @param band_type array of the used band type
870 *
871 * @return Returns error status. 0 - OK, !0 - error
872 */
577d383b 873static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
3963a17d 874 GetBitContext *gb, const float sf[120],
577d383b
DB
875 int pulse_present, const Pulse *pulse,
876 const IndividualChannelStream *ics,
877 enum BandType band_type[120])
878{
9ffd5c1c 879 int i, k, g, idx = 0;
577d383b
DB
880 const int c = 1024 / ics->num_windows;
881 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c 882 float *coef_base = coef;
4a904903 883 static const float sign_lookup[] = { 1.0f, -1.0f };
9ffd5c1c
RS
884
885 for (g = 0; g < ics->num_windows; g++)
577d383b 886 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
9ffd5c1c
RS
887
888 for (g = 0; g < ics->num_window_groups; g++) {
889 for (i = 0; i < ics->max_sfb; i++, idx++) {
890 const int cur_band_type = band_type[idx];
891 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
892 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
893 int group;
aab54133 894 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
9ffd5c1c 895 for (group = 0; group < ics->group_len[g]; group++) {
577d383b 896 memset(coef + group * 128 + offsets[i], 0, (offsets[i + 1] - offsets[i]) * sizeof(float));
9ffd5c1c 897 }
577d383b 898 } else if (cur_band_type == NOISE_BT) {
9ffd5c1c 899 for (group = 0; group < ics->group_len[g]; group++) {
d0ee5021 900 float scale;
b418a6ca 901 float band_energy;
0cb71412 902 float *cf = coef + group * 128 + offsets[i];
42d3fbb3
MR
903 int len = offsets[i+1] - offsets[i];
904
b418a6ca 905 for (k = 0; k < len; k++) {
9ffd5c1c 906 ac->random_state = lcg_random(ac->random_state);
b418a6ca 907 cf[k] = ac->random_state;
d0ee5021 908 }
42d3fbb3 909
b418a6ca 910 band_energy = ac->dsp.scalarproduct_float(cf, cf, len);
d0ee5021 911 scale = sf[idx] / sqrtf(band_energy);
42d3fbb3 912 ac->dsp.vector_fmul_scalar(cf, cf, scale, len);
9ffd5c1c 913 }
577d383b 914 } else {
9ffd5c1c 915 for (group = 0; group < ics->group_len[g]; group++) {
42d3fbb3
MR
916 const float *vq[96];
917 const float **vqp = vq;
918 float *cf = coef + (group << 7) + offsets[i];
919 int len = offsets[i + 1] - offsets[i];
920
577d383b 921 for (k = offsets[i]; k < offsets[i + 1]; k += dim) {
9ffd5c1c
RS
922 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
923 const int coef_tmp_idx = (group << 7) + k;
924 const float *vq_ptr;
925 int j;
577d383b 926 if (index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
9ffd5c1c 927 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b
DB
928 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
929 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
9ffd5c1c
RS
930 return -1;
931 }
932 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
42d3fbb3 933 *vqp++ = vq_ptr;
9ffd5c1c 934 if (is_cb_unsigned) {
577d383b
DB
935 if (vq_ptr[0])
936 coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
937 if (vq_ptr[1])
938 coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
c0893c3a 939 if (dim == 4) {
577d383b
DB
940 if (vq_ptr[2])
941 coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
942 if (vq_ptr[3])
943 coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
c0893c3a 944 }
e8d5c07b
AC
945 if (cur_band_type == ESC_BT) {
946 for (j = 0; j < 2; j++) {
947 if (vq_ptr[j] == 64.0f) {
948 int n = 4;
949 /* The total length of escape_sequence must be < 22 bits according
7c8af53f
MR
950 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
951 while (get_bits1(gb) && n < 13) n++;
952 if (n == 13) {
e8d5c07b
AC
953 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
954 return -1;
955 }
577d383b 956 n = (1 << n) + get_bits(gb, n);
dc0d86fa 957 coef[coef_tmp_idx + j] *= cbrt_tab[n];
577d383b 958 } else
e8d5c07b
AC
959 coef[coef_tmp_idx + j] *= vq_ptr[j];
960 }
9ffd5c1c 961 }
c0893c3a 962 }
9ffd5c1c 963 }
42d3fbb3
MR
964
965 if (is_cb_unsigned && cur_band_type != ESC_BT) {
966 ac->dsp.vector_fmul_sv_scalar[dim>>2](
967 cf, cf, vq, sf[idx], len);
5419da8d 968 } else if (cur_band_type == ESC_BT) {
42d3fbb3
MR
969 ac->dsp.vector_fmul_scalar(cf, cf, sf[idx], len);
970 } else { /* !is_cb_unsigned */
971 ac->dsp.sv_fmul_scalar[dim>>2](cf, vq, sf[idx], len);
972 }
9ffd5c1c
RS
973 }
974 }
975 }
577d383b 976 coef += ics->group_len[g] << 7;
9ffd5c1c
RS
977 }
978
979 if (pulse_present) {
51436848 980 idx = 0;
577d383b
DB
981 for (i = 0; i < pulse->num_pulse; i++) {
982 float co = coef_base[ pulse->pos[i] ];
983 while (offsets[idx + 1] <= pulse->pos[i])
51436848
AC
984 idx++;
985 if (band_type[idx] != NOISE_BT && sf[idx]) {
70735a3f
RS
986 float ico = -pulse->amp[i];
987 if (co) {
988 co /= sf[idx];
989 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
990 }
991 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
51436848 992 }
9ffd5c1c
RS
993 }
994 }
995 return 0;
996}
997
577d383b
DB
998static av_always_inline float flt16_round(float pf)
999{
4a39ccb4
AC
1000 union float754 tmp;
1001 tmp.f = pf;
1002 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1003 return tmp.f;
7633a041
AC
1004}
1005
577d383b
DB
1006static av_always_inline float flt16_even(float pf)
1007{
4a39ccb4
AC
1008 union float754 tmp;
1009 tmp.f = pf;
577d383b 1010 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
4a39ccb4 1011 return tmp.f;
7633a041
AC
1012}
1013
577d383b
DB
1014static av_always_inline float flt16_trunc(float pf)
1015{
4a39ccb4
AC
1016 union float754 pun;
1017 pun.f = pf;
1018 pun.i &= 0xFFFF0000U;
1019 return pun.f;
7633a041
AC
1020}
1021
577d383b
DB
1022static void predict(AACContext *ac, PredictorState *ps, float *coef,
1023 int output_enable)
1024{
1025 const float a = 0.953125; // 61.0 / 64
1026 const float alpha = 0.90625; // 29.0 / 32
7633a041
AC
1027 float e0, e1;
1028 float pv;
1029 float k1, k2;
1030
1031 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1032 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1033
1034 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1035 if (output_enable)
1036 *coef += pv * ac->sf_scale;
1037
1038 e0 = *coef / ac->sf_scale;
1039 e1 = e0 - k1 * ps->r0;
1040
1041 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1042 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1043 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1044 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1045
1046 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1047 ps->r0 = flt16_trunc(a * e0);
1048}
1049
1050/**
1051 * Apply AAC-Main style frequency domain prediction.
1052 */
577d383b
DB
1053static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1054{
7633a041
AC
1055 int sfb, k;
1056
1057 if (!sce->ics.predictor_initialized) {
aab54133 1058 reset_all_predictors(sce->predictor_state);
7633a041
AC
1059 sce->ics.predictor_initialized = 1;
1060 }
1061
1062 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1063 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1064 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
aab54133 1065 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
577d383b 1066 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
7633a041
AC
1067 }
1068 }
1069 if (sce->ics.predictor_reset_group)
aab54133 1070 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
7633a041 1071 } else
aab54133 1072 reset_all_predictors(sce->predictor_state);
7633a041
AC
1073}
1074
9ffd5c1c 1075/**
9cc04edf
RS
1076 * Decode an individual_channel_stream payload; reference: table 4.44.
1077 *
1078 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1079 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1080 *
1081 * @return Returns error status. 0 - OK, !0 - error
1082 */
577d383b
DB
1083static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1084 GetBitContext *gb, int common_window, int scale_flag)
1085{
9cc04edf 1086 Pulse pulse;
577d383b
DB
1087 TemporalNoiseShaping *tns = &sce->tns;
1088 IndividualChannelStream *ics = &sce->ics;
1089 float *out = sce->coeffs;
9cc04edf
RS
1090 int global_gain, pulse_present = 0;
1091
848a5815
RS
1092 /* This assignment is to silence a GCC warning about the variable being used
1093 * uninitialized when in fact it always is.
9cc04edf
RS
1094 */
1095 pulse.num_pulse = 0;
9cc04edf
RS
1096
1097 global_gain = get_bits(gb, 8);
1098
1099 if (!common_window && !scale_flag) {
1100 if (decode_ics_info(ac, ics, gb, 0) < 0)
1101 return -1;
1102 }
1103
1104 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1105 return -1;
1106 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1107 return -1;
1108
1109 pulse_present = 0;
1110 if (!scale_flag) {
1111 if ((pulse_present = get_bits1(gb))) {
1112 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1113 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1114 return -1;
1115 }
aac0eda4
AC
1116 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1117 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1118 return -1;
1119 }
9cc04edf
RS
1120 }
1121 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1122 return -1;
1123 if (get_bits1(gb)) {
ce863d7f 1124 av_log_missing_feature(ac->avccontext, "SSR", 1);
9cc04edf
RS
1125 return -1;
1126 }
1127 }
1128
848a5815 1129 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
9cc04edf 1130 return -1;
7633a041 1131
577d383b 1132 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
7633a041
AC
1133 apply_prediction(ac, sce);
1134
9cc04edf
RS
1135 return 0;
1136}
1137
1138/**
9ffd5c1c
RS
1139 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1140 */
42d3fbb3 1141static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
577d383b
DB
1142{
1143 const IndividualChannelStream *ics = &cpe->ch[0].ics;
9ffd5c1c
RS
1144 float *ch0 = cpe->ch[0].coeffs;
1145 float *ch1 = cpe->ch[1].coeffs;
42d3fbb3 1146 int g, i, group, idx = 0;
577d383b 1147 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c
RS
1148 for (g = 0; g < ics->num_window_groups; g++) {
1149 for (i = 0; i < ics->max_sfb; i++, idx++) {
1150 if (cpe->ms_mask[idx] &&
577d383b 1151 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
9ffd5c1c 1152 for (group = 0; group < ics->group_len[g]; group++) {
42d3fbb3
MR
1153 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1154 ch1 + group * 128 + offsets[i],
1155 offsets[i+1] - offsets[i]);
9ffd5c1c
RS
1156 }
1157 }
1158 }
577d383b
DB
1159 ch0 += ics->group_len[g] * 128;
1160 ch1 += ics->group_len[g] * 128;
9ffd5c1c
RS
1161 }
1162}
1163
1164/**
1165 * intensity stereo decoding; reference: 4.6.8.2.3
1166 *
1167 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1168 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1169 * [3] reserved for scalable AAC
1170 */
577d383b
DB
1171static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1172{
1173 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1174 SingleChannelElement *sce1 = &cpe->ch[1];
9ffd5c1c 1175 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
577d383b 1176 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c
RS
1177 int g, group, i, k, idx = 0;
1178 int c;
1179 float scale;
1180 for (g = 0; g < ics->num_window_groups; g++) {
1181 for (i = 0; i < ics->max_sfb;) {
1182 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1183 const int bt_run_end = sce1->band_type_run_end[idx];
1184 for (; i < bt_run_end; i++, idx++) {
1185 c = -1 + 2 * (sce1->band_type[idx] - 14);
1186 if (ms_present)
1187 c *= 1 - 2 * cpe->ms_mask[idx];
1188 scale = c * sce1->sf[idx];
1189 for (group = 0; group < ics->group_len[g]; group++)
577d383b
DB
1190 for (k = offsets[i]; k < offsets[i + 1]; k++)
1191 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
9ffd5c1c
RS
1192 }
1193 } else {
1194 int bt_run_end = sce1->band_type_run_end[idx];
1195 idx += bt_run_end - i;
1196 i = bt_run_end;
1197 }
1198 }
577d383b
DB
1199 coef0 += ics->group_len[g] * 128;
1200 coef1 += ics->group_len[g] * 128;
9ffd5c1c
RS
1201 }
1202}
1203
1204/**
9cc04edf
RS
1205 * Decode a channel_pair_element; reference: table 4.4.
1206 *
1207 * @param elem_id Identifies the instance of a syntax element.
1208 *
1209 * @return Returns error status. 0 - OK, !0 - error
1210 */
577d383b
DB
1211static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1212{
9cc04edf 1213 int i, ret, common_window, ms_present = 0;
9cc04edf 1214
9cc04edf
RS
1215 common_window = get_bits1(gb);
1216 if (common_window) {
1217 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1218 return -1;
1219 i = cpe->ch[1].ics.use_kb_window[0];
1220 cpe->ch[1].ics = cpe->ch[0].ics;
1221 cpe->ch[1].ics.use_kb_window[1] = i;
1222 ms_present = get_bits(gb, 2);
577d383b 1223 if (ms_present == 3) {
9cc04edf
RS
1224 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1225 return -1;
577d383b 1226 } else if (ms_present)
9cc04edf
RS
1227 decode_mid_side_stereo(cpe, gb, ms_present);
1228 }
1229 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1230 return ret;
1231 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1232 return ret;
1233
aab54133
AC
1234 if (common_window) {
1235 if (ms_present)
42d3fbb3 1236 apply_mid_side_stereo(ac, cpe);
aab54133
AC
1237 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1238 apply_prediction(ac, &cpe->ch[0]);
1239 apply_prediction(ac, &cpe->ch[1]);
1240 }
1241 }
9cc04edf 1242
848a5815 1243 apply_intensity_stereo(cpe, ms_present);
9cc04edf
RS
1244 return 0;
1245}
1246
9ffd5c1c
RS
1247/**
1248 * Decode coupling_channel_element; reference: table 4.8.
1249 *
1250 * @param elem_id Identifies the instance of a syntax element.
1251 *
1252 * @return Returns error status. 0 - OK, !0 - error
1253 */
577d383b
DB
1254static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1255{
9ffd5c1c 1256 int num_gain = 0;
341b28c0 1257 int c, g, sfb, ret;
9ffd5c1c
RS
1258 int sign;
1259 float scale;
577d383b
DB
1260 SingleChannelElement *sce = &che->ch[0];
1261 ChannelCoupling *coup = &che->coup;
9ffd5c1c 1262
577d383b 1263 coup->coupling_point = 2 * get_bits1(gb);
62a57fae
RS
1264 coup->num_coupled = get_bits(gb, 3);
1265 for (c = 0; c <= coup->num_coupled; c++) {
1266 num_gain++;
1267 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1268 coup->id_select[c] = get_bits(gb, 4);
1269 if (coup->type[c] == TYPE_CPE) {
1270 coup->ch_select[c] = get_bits(gb, 2);
1271 if (coup->ch_select[c] == 3)
1272 num_gain++;
1273 } else
88de95c2 1274 coup->ch_select[c] = 2;
62a57fae 1275 }
577d383b 1276 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
62a57fae 1277
577d383b 1278 sign = get_bits(gb, 1);
c8947a56 1279 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
62a57fae
RS
1280
1281 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1282 return ret;
1283
1284 for (c = 0; c < num_gain; c++) {
577d383b
DB
1285 int idx = 0;
1286 int cge = 1;
62a57fae
RS
1287 int gain = 0;
1288 float gain_cache = 1.;
1289 if (c) {
1290 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1291 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
88de95c2 1292 gain_cache = pow(scale, -gain);
62a57fae 1293 }
f1ade11e
AC
1294 if (coup->coupling_point == AFTER_IMDCT) {
1295 coup->gain[c][0] = gain_cache;
1296 } else {
03b12747
AC
1297 for (g = 0; g < sce->ics.num_window_groups; g++) {
1298 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1299 if (sce->band_type[idx] != ZERO_BT) {
1300 if (!cge) {
1301 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 1302 if (t) {
03b12747
AC
1303 int s = 1;
1304 t = gain += t;
1305 if (sign) {
1306 s -= 2 * (t & 0x1);
1307 t >>= 1;
1308 }
1309 gain_cache = pow(scale, -t) * s;
62a57fae 1310 }
62a57fae 1311 }
03b12747 1312 coup->gain[c][idx] = gain_cache;
62a57fae 1313 }
62a57fae 1314 }
f80a8ca5
RS
1315 }
1316 }
62a57fae
RS
1317 }
1318 return 0;
1319}
1320
9cc04edf
RS
1321/**
1322 * Decode Spectral Band Replication extension data; reference: table 4.55.
cc0591da
RS
1323 *
1324 * @param crc flag indicating the presence of CRC checksum
1325 * @param cnt length of TYPE_FIL syntactic element in bytes
9cc04edf 1326 *
cc0591da
RS
1327 * @return Returns number of bytes consumed from the TYPE_FIL element.
1328 */
577d383b
DB
1329static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1330 int crc, int cnt)
1331{
cc0591da 1332 // TODO : sbr_extension implementation
ce863d7f 1333 av_log_missing_feature(ac->avccontext, "SBR", 0);
577d383b 1334 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
cc0591da
RS
1335 return cnt;
1336}
1337
9cc04edf 1338/**
62a57fae
RS
1339 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1340 *
1341 * @return Returns number of bytes consumed.
1342 */
577d383b
DB
1343static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1344 GetBitContext *gb)
1345{
62a57fae
RS
1346 int i;
1347 int num_excl_chan = 0;
1348
1349 do {
1350 for (i = 0; i < 7; i++)
1351 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1352 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1353
1354 return num_excl_chan / 7;
1355}
1356
1357/**
9cc04edf
RS
1358 * Decode dynamic range information; reference: table 4.52.
1359 *
1360 * @param cnt length of TYPE_FIL syntactic element in bytes
1361 *
1362 * @return Returns number of bytes consumed.
1363 */
577d383b
DB
1364static int decode_dynamic_range(DynamicRangeControl *che_drc,
1365 GetBitContext *gb, int cnt)
1366{
1367 int n = 1;
9cc04edf
RS
1368 int drc_num_bands = 1;
1369 int i;
1370
1371 /* pce_tag_present? */
577d383b 1372 if (get_bits1(gb)) {
9cc04edf
RS
1373 che_drc->pce_instance_tag = get_bits(gb, 4);
1374 skip_bits(gb, 4); // tag_reserved_bits
1375 n++;
1376 }
1377
1378 /* excluded_chns_present? */
577d383b 1379 if (get_bits1(gb)) {
9cc04edf
RS
1380 n += decode_drc_channel_exclusions(che_drc, gb);
1381 }
1382
1383 /* drc_bands_present? */
1384 if (get_bits1(gb)) {
1385 che_drc->band_incr = get_bits(gb, 4);
1386 che_drc->interpolation_scheme = get_bits(gb, 4);
1387 n++;
1388 drc_num_bands += che_drc->band_incr;
1389 for (i = 0; i < drc_num_bands; i++) {
1390 che_drc->band_top[i] = get_bits(gb, 8);
1391 n++;
1392 }
1393 }
1394
1395 /* prog_ref_level_present? */
1396 if (get_bits1(gb)) {
1397 che_drc->prog_ref_level = get_bits(gb, 7);
1398 skip_bits1(gb); // prog_ref_level_reserved_bits
1399 n++;
1400 }
1401
1402 for (i = 0; i < drc_num_bands; i++) {
1403 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1404 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1405 n++;
1406 }
1407
1408 return n;
1409}
1410
1411/**
1412 * Decode extension data (incomplete); reference: table 4.51.
1413 *
1414 * @param cnt length of TYPE_FIL syntactic element in bytes
1415 *
1416 * @return Returns number of bytes consumed
1417 */
577d383b
DB
1418static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1419{
cc0591da
RS
1420 int crc_flag = 0;
1421 int res = cnt;
1422 switch (get_bits(gb, 4)) { // extension type
577d383b
DB
1423 case EXT_SBR_DATA_CRC:
1424 crc_flag++;
1425 case EXT_SBR_DATA:
1426 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1427 break;
1428 case EXT_DYNAMIC_RANGE:
1429 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1430 break;
1431 case EXT_FILL:
1432 case EXT_FILL_DATA:
1433 case EXT_DATA_ELEMENT:
1434 default:
1435 skip_bits_long(gb, 8 * cnt - 4);
1436 break;
cc0591da
RS
1437 };
1438 return res;
1439}
1440
7d8f3de4
RS
1441/**
1442 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1443 *
1444 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1445 * @param coef spectral coefficients
1446 */
577d383b
DB
1447static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1448 IndividualChannelStream *ics, int decode)
1449{
1450 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1098e8d2 1451 int w, filt, m, i;
7d8f3de4
RS
1452 int bottom, top, order, start, end, size, inc;
1453 float lpc[TNS_MAX_ORDER];
1454
1455 for (w = 0; w < ics->num_windows; w++) {
1456 bottom = ics->num_swb;
1457 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1458 top = bottom;
1459 bottom = FFMAX(0, top - tns->length[w][filt]);
1460 order = tns->order[w][filt];
1461 if (order == 0)
1462 continue;
1463
1be0fc29
VS
1464 // tns_decode_coef
1465 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
7d8f3de4 1466
1dece0d2
RS
1467 start = ics->swb_offset[FFMIN(bottom, mmm)];
1468 end = ics->swb_offset[FFMIN( top, mmm)];
1469 if ((size = end - start) <= 0)
1470 continue;
1471 if (tns->direction[w][filt]) {
577d383b
DB
1472 inc = -1;
1473 start = end - 1;
1dece0d2
RS
1474 } else {
1475 inc = 1;
1476 }
1477 start += w * 128;
1478
1479 // ar filter
1480 for (m = 0; m < size; m++, start += inc)
1481 for (i = 1; i <= FFMIN(m, order); i++)
577d383b 1482 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1dece0d2
RS
1483 }
1484 }
1485}
1486
cc0591da 1487/**
9cc04edf
RS
1488 * Conduct IMDCT and windowing.
1489 */
577d383b
DB
1490static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1491{
1492 IndividualChannelStream *ics = &sce->ics;
1493 float *in = sce->coeffs;
1494 float *out = sce->ret;
1495 float *saved = sce->saved;
1496 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1497 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1498 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1499 float *buf = ac->buf_mdct;
1500 float *temp = ac->temp;
9cc04edf
RS
1501 int i;
1502
f4990558 1503 // imdct
62a57fae
RS
1504 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1505 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1506 av_log(ac->avccontext, AV_LOG_WARNING,
1507 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1508 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
b0f5852a
RS
1509 for (i = 0; i < 1024; i += 128)
1510 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
f4990558 1511 } else
b0f5852a 1512 ff_imdct_half(&ac->mdct, buf, in);
f4990558
RS
1513
1514 /* window overlapping
1515 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1516 * and long to short transitions are considered to be short to short
1517 * transitions. This leaves just two cases (long to long and short to short)
1518 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1519 */
1520 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
577d383b 1521 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
b0f5852a 1522 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
f4990558 1523 } else {
db38c386
RS
1524 for (i = 0; i < 448; i++)
1525 out[i] = saved[i] + ac->add_bias;
62a57fae 1526
f4990558 1527 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1528 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1529 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1530 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1531 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1532 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1533 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
f4990558 1534 } else {
b0f5852a 1535 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
db38c386 1536 for (i = 576; i < 1024; i++)
b0f5852a 1537 out[i] = buf[i-512] + ac->add_bias;
f4990558
RS
1538 }
1539 }
62a57fae 1540
f4990558
RS
1541 // buffer update
1542 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1543 for (i = 0; i < 64; i++)
1544 saved[i] = temp[64 + i] - ac->add_bias;
1545 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1546 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1547 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1548 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1549 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
b0f5852a
RS
1550 memcpy( saved, buf + 512, 448 * sizeof(float));
1551 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1552 } else { // LONG_STOP or ONLY_LONG
b0f5852a 1553 memcpy( saved, buf + 512, 512 * sizeof(float));
62a57fae
RS
1554 }
1555}
1556
9cc04edf 1557/**
cc0591da
RS
1558 * Apply dependent channel coupling (applied before IMDCT).
1559 *
1560 * @param index index into coupling gain array
1561 */
577d383b
DB
1562static void apply_dependent_coupling(AACContext *ac,
1563 SingleChannelElement *target,
1564 ChannelElement *cce, int index)
1565{
1566 IndividualChannelStream *ics = &cce->ch[0].ics;
1567 const uint16_t *offsets = ics->swb_offset;
1568 float *dest = target->coeffs;
1569 const float *src = cce->ch[0].coeffs;
cc0591da 1570 int g, i, group, k, idx = 0;
577d383b 1571 if (ac->m4ac.object_type == AOT_AAC_LTP) {
cc0591da
RS
1572 av_log(ac->avccontext, AV_LOG_ERROR,
1573 "Dependent coupling is not supported together with LTP\n");
1574 return;
1575 }
1576 for (g = 0; g < ics->num_window_groups; g++) {
1577 for (i = 0; i < ics->max_sfb; i++, idx++) {
fbdae895 1578 if (cce->ch[0].band_type[idx] != ZERO_BT) {
cfd937b0 1579 const float gain = cce->coup.gain[index][idx];
cc0591da 1580 for (group = 0; group < ics->group_len[g]; group++) {
577d383b 1581 for (k = offsets[i]; k < offsets[i + 1]; k++) {
cc0591da 1582 // XXX dsputil-ize
577d383b 1583 dest[group * 128 + k] += gain * src[group * 128 + k];
cc0591da
RS
1584 }
1585 }
1586 }
1587 }
577d383b
DB
1588 dest += ics->group_len[g] * 128;
1589 src += ics->group_len[g] * 128;
cc0591da
RS
1590 }
1591}
1592
1593/**
1594 * Apply independent channel coupling (applied after IMDCT).
1595 *
1596 * @param index index into coupling gain array
1597 */
577d383b
DB
1598static void apply_independent_coupling(AACContext *ac,
1599 SingleChannelElement *target,
1600 ChannelElement *cce, int index)
1601{
cc0591da 1602 int i;
039821a8
AC
1603 const float gain = cce->coup.gain[index][0];
1604 const float bias = ac->add_bias;
577d383b
DB
1605 const float *src = cce->ch[0].ret;
1606 float *dest = target->ret;
039821a8 1607
cc0591da 1608 for (i = 0; i < 1024; i++)
039821a8 1609 dest[i] += gain * (src[i] - bias);
cc0591da
RS
1610}
1611
9ffd5c1c
RS
1612/**
1613 * channel coupling transformation interface
1614 *
1615 * @param index index into coupling gain array
1616 * @param apply_coupling_method pointer to (in)dependent coupling function
1617 */
577d383b
DB
1618static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1619 enum RawDataBlockType type, int elem_id,
1620 enum CouplingPoint coupling_point,
1621 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
9ffd5c1c 1622{
88de95c2
AC
1623 int i, c;
1624
1625 for (i = 0; i < MAX_ELEM_ID; i++) {
1626 ChannelElement *cce = ac->che[TYPE_CCE][i];
1627 int index = 0;
1628
1629 if (cce && cce->coup.coupling_point == coupling_point) {
577d383b 1630 ChannelCoupling *coup = &cce->coup;
88de95c2
AC
1631
1632 for (c = 0; c <= coup->num_coupled; c++) {
1633 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1634 if (coup->ch_select[c] != 1) {
1635 apply_coupling_method(ac, &cc->ch[0], cce, index);
1636 if (coup->ch_select[c] != 0)
1637 index++;
1638 }
1639 if (coup->ch_select[c] != 2)
1640 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1641 } else
1642 index += 1 + (coup->ch_select[c] == 3);
9ffd5c1c 1643 }
9ffd5c1c
RS
1644 }
1645 }
1646}
1647
1648/**
1649 * Convert spectral data to float samples, applying all supported tools as appropriate.
1650 */
577d383b
DB
1651static void spectral_to_sample(AACContext *ac)
1652{
b0bc928b
CEH
1653 int i, type;
1654 for (type = 3; type >= 0; type--) {
88de95c2 1655 for (i = 0; i < MAX_ELEM_ID; i++) {
9ffd5c1c 1656 ChannelElement *che = ac->che[type][i];
577d383b
DB
1657 if (che) {
1658 if (type <= TYPE_CPE)
88de95c2 1659 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
577d383b 1660 if (che->ch[0].tns.present)
9ffd5c1c 1661 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
577d383b 1662 if (che->ch[1].tns.present)
9ffd5c1c 1663 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
577d383b 1664 if (type <= TYPE_CPE)
88de95c2 1665 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
577d383b 1666 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
88de95c2 1667 imdct_and_windowing(ac, &che->ch[0]);
577d383b 1668 if (type == TYPE_CPE)
9ffd5c1c 1669 imdct_and_windowing(ac, &che->ch[1]);
577d383b 1670 if (type <= TYPE_CCE)
88de95c2 1671 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
62a57fae
RS
1672 }
1673 }
1674 }
1675}
1676
577d383b
DB
1677static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1678{
158b3912
RS
1679 int size;
1680 AACADTSHeaderInfo hdr_info;
1681
1682 size = ff_aac_parse_header(gb, &hdr_info);
1683 if (size > 0) {
981b8fd7 1684 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
6308765c
AC
1685 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1686 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
158b3912 1687 ac->m4ac.chan_config = hdr_info.chan_config;
6308765c
AC
1688 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1689 return -7;
981b8fd7 1690 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
6308765c 1691 return -7;
981b8fd7
AC
1692 } else if (ac->output_configured != OC_LOCKED) {
1693 ac->output_configured = OC_NONE;
6308765c 1694 }
38610d92
AC
1695 if (ac->output_configured != OC_LOCKED)
1696 ac->m4ac.sbr = -1;
158b3912
RS
1697 ac->m4ac.sample_rate = hdr_info.sample_rate;
1698 ac->m4ac.sampling_index = hdr_info.sampling_index;
1699 ac->m4ac.object_type = hdr_info.object_type;
f6586314
1700 if (!ac->avccontext->sample_rate)
1701 ac->avccontext->sample_rate = hdr_info.sample_rate;
7d87e2ce
AC
1702 if (hdr_info.num_aac_frames == 1) {
1703 if (!hdr_info.crc_absent)
1704 skip_bits(gb, 16);
1705 } else {
ce863d7f 1706 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
7d87e2ce
AC
1707 return -1;
1708 }
51741a82 1709 }
158b3912
RS
1710 return size;
1711}
1712
577d383b
DB
1713static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1714 int *data_size, AVPacket *avpkt)
1715{
7a00bbad
TB
1716 const uint8_t *buf = avpkt->data;
1717 int buf_size = avpkt->size;
577d383b
DB
1718 AACContext *ac = avccontext->priv_data;
1719 ChannelElement *che = NULL;
62a57fae
RS
1720 GetBitContext gb;
1721 enum RawDataBlockType elem_type;
1722 int err, elem_id, data_size_tmp;
1723
577d383b 1724 init_get_bits(&gb, buf, buf_size * 8);
62a57fae 1725
158b3912 1726 if (show_bits(&gb, 12) == 0xfff) {
5967e141 1727 if (parse_adts_frame_header(ac, &gb) < 0) {
158b3912
RS
1728 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1729 return -1;
1730 }
30272450 1731 if (ac->m4ac.sampling_index > 12) {
f418b861
JM
1732 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1733 return -1;
1734 }
158b3912
RS
1735 }
1736
62a57fae
RS
1737 // parse
1738 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1739 elem_id = get_bits(&gb, 4);
62a57fae 1740
577d383b 1741 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
5f401ee0
RS
1742 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1743 return -1;
62a57fae
RS
1744 }
1745
1746 switch (elem_type) {
1747
1748 case TYPE_SCE:
bb5c0988 1749 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
62a57fae
RS
1750 break;
1751
1752 case TYPE_CPE:
bb5c0988 1753 err = decode_cpe(ac, &gb, che);
62a57fae
RS
1754 break;
1755
1756 case TYPE_CCE:
bb5c0988 1757 err = decode_cce(ac, &gb, che);
62a57fae
RS
1758 break;
1759
1760 case TYPE_LFE:
bb5c0988 1761 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
62a57fae
RS
1762 break;
1763
1764 case TYPE_DSE:
1765 skip_data_stream_element(&gb);
1766 err = 0;
1767 break;
1768
577d383b 1769 case TYPE_PCE: {
62a57fae
RS
1770 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1771 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
577d383b 1772 if ((err = decode_pce(ac, new_che_pos, &gb)))
62a57fae 1773 break;
4e878a18 1774 if (ac->output_configured > OC_TRIAL_PCE)
6308765c
AC
1775 av_log(avccontext, AV_LOG_ERROR,
1776 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1777 else
981b8fd7 1778 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
62a57fae
RS
1779 break;
1780 }
1781
1782 case TYPE_FIL:
1783 if (elem_id == 15)
1784 elem_id += get_bits(&gb, 8) - 1;
1785 while (elem_id > 0)
1786 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1787 err = 0; /* FIXME */
1788 break;
1789
1790 default:
1791 err = -1; /* should not happen, but keeps compiler happy */
1792 break;
1793 }
1794
577d383b 1795 if (err)
62a57fae
RS
1796 return err;
1797 }
1798
1799 spectral_to_sample(ac);
1800
9cc04edf
RS
1801 if (!ac->is_saved) {
1802 ac->is_saved = 1;
1803 *data_size = 0;
848a5815 1804 return buf_size;
9cc04edf
RS
1805 }
1806
1807 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
577d383b 1808 if (*data_size < data_size_tmp) {
9cc04edf
RS
1809 av_log(avccontext, AV_LOG_ERROR,
1810 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1811 *data_size, data_size_tmp);
1812 return -1;
1813 }
1814 *data_size = data_size_tmp;
1815
1816 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1817
981b8fd7
AC
1818 if (ac->output_configured)
1819 ac->output_configured = OC_LOCKED;
1820
9cc04edf
RS
1821 return buf_size;
1822}
1823
577d383b
DB
1824static av_cold int aac_decode_close(AVCodecContext *avccontext)
1825{
1826 AACContext *ac = avccontext->priv_data;
9edae4ad 1827 int i, type;
71e9a1b8 1828
cc0591da 1829 for (i = 0; i < MAX_ELEM_ID; i++) {
577d383b 1830 for (type = 0; type < 4; type++)
9edae4ad 1831 av_freep(&ac->che[type][i]);
71e9a1b8
RS
1832 }
1833
1834 ff_mdct_end(&ac->mdct);
1835 ff_mdct_end(&ac->mdct_small);
577d383b 1836 return 0;
71e9a1b8
RS
1837}
1838
1839AVCodec aac_decoder = {
1840 "aac",
1841 CODEC_TYPE_AUDIO,
1842 CODEC_ID_AAC,
1843 sizeof(AACContext),
1844 aac_decode_init,
1845 NULL,
1846 aac_decode_close,
1847 aac_decode_frame,
1848 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
b5f09d31 1849 .sample_fmts = (const enum SampleFormat[]) {
577d383b
DB
1850 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
1851 },
e22da6b6 1852 .channel_layouts = aac_channel_layout,
71e9a1b8 1853};