AAC: Add channel layout support for channel configuration as defined in the
[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
104
577d383b
DB
105static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
106{
bb5c0988
AC
107 if (ac->tag_che_map[type][elem_id]) {
108 return ac->tag_che_map[type][elem_id];
109 }
110 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
111 return NULL;
112 }
113 switch (ac->m4ac.chan_config) {
577d383b
DB
114 case 7:
115 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
116 ac->tags_mapped++;
117 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
118 }
119 case 6:
120 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
121 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
122 encountered such a stream, transfer the LFE[0] element to SCE[1] */
123 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
124 ac->tags_mapped++;
125 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
126 }
127 case 5:
128 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
129 ac->tags_mapped++;
130 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
131 }
132 case 4:
133 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
134 ac->tags_mapped++;
135 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
136 }
137 case 3:
138 case 2:
139 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
140 ac->tags_mapped++;
141 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
142 } else if (ac->m4ac.chan_config == 2) {
bb5c0988 143 return NULL;
577d383b
DB
144 }
145 case 1:
146 if (!ac->tags_mapped && type == TYPE_SCE) {
147 ac->tags_mapped++;
148 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
149 }
150 default:
151 return NULL;
bb5c0988
AC
152 }
153}
154
9cc04edf 155/**
62a57fae
RS
156 * Configure output channel order based on the current program configuration element.
157 *
158 * @param che_pos current channel position configuration
159 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
160 *
161 * @return Returns error status. 0 - OK, !0 - error
162 */
577d383b
DB
163static int output_configure(AACContext *ac,
164 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
165 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
166 int channel_config)
167{
62a57fae
RS
168 AVCodecContext *avctx = ac->avccontext;
169 int i, type, channels = 0;
170
62a57fae
RS
171 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
172
e22da6b6
RS
173 if (channel_config) {
174 for (i = 0; i < tags_per_config[channel_config]; i++) {
175 const int id = aac_channel_layout_map[channel_config - 1][i][1];
176 type = aac_channel_layout_map[channel_config - 1][i][0];
177
178 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
179 return AVERROR(ENOMEM);
180
181 if (type != TYPE_CCE) {
182 ac->output_data[channels++] = ac->che[type][id]->ch[0].ret;
183 if (type == TYPE_CPE)
184 ac->output_data[channels++] = ac->che[type][id]->ch[1].ret;
185 }
186 }
187
188 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
189 ac->tags_mapped = 0;
190
191 avctx->channel_layout = aac_channel_layout[channel_config - 1];
192 } else {
62a57fae
RS
193 /* Allocate or free elements depending on if they are in the
194 * current program configuration.
195 *
196 * Set up default 1:1 output mapping.
197 *
198 * For a 5.1 stream the output order will be:
f3399088 199 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
62a57fae
RS
200 */
201
577d383b
DB
202 for (i = 0; i < MAX_ELEM_ID; i++) {
203 for (type = 0; type < 4; type++) {
204 if (che_pos[type][i]) {
205 if (!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
62a57fae 206 return AVERROR(ENOMEM);
577d383b 207 if (type != TYPE_CCE) {
62a57fae 208 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
577d383b 209 if (type == TYPE_CPE) {
62a57fae
RS
210 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
211 }
212 }
213 } else
214 av_freep(&ac->che[type][i]);
215 }
216 }
217
bb5c0988 218 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
577d383b 219 ac->tags_mapped = 4 * MAX_ELEM_ID;
e22da6b6
RS
220
221 avctx->channel_layout = 0;
bb5c0988
AC
222 }
223
62a57fae 224 avctx->channels = channels;
bb5c0988 225
6308765c
AC
226 ac->output_configured = 1;
227
62a57fae
RS
228 return 0;
229}
230
231/**
9cc04edf
RS
232 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
233 *
234 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
235 * @param sce_map mono (Single Channel Element) map
236 * @param type speaker type/position for these channels
237 */
238static void decode_channel_map(enum ChannelPosition *cpe_map,
577d383b
DB
239 enum ChannelPosition *sce_map,
240 enum ChannelPosition type,
241 GetBitContext *gb, int n)
242{
243 while (n--) {
9cc04edf
RS
244 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
245 map[get_bits(gb, 4)] = type;
246 }
247}
248
249/**
250 * Decode program configuration element; reference: table 4.2.
251 *
252 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
253 *
254 * @return Returns error status. 0 - OK, !0 - error
255 */
577d383b
DB
256static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
257 GetBitContext *gb)
258{
99665a21 259 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
9cc04edf
RS
260
261 skip_bits(gb, 2); // object_type
262
99665a21 263 sampling_index = get_bits(gb, 4);
401a9950
AC
264 if (ac->m4ac.sampling_index != sampling_index)
265 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");
266
71e9a1b8
RS
267 num_front = get_bits(gb, 4);
268 num_side = get_bits(gb, 4);
269 num_back = get_bits(gb, 4);
270 num_lfe = get_bits(gb, 2);
271 num_assoc_data = get_bits(gb, 3);
272 num_cc = get_bits(gb, 4);
273
cc0591da
RS
274 if (get_bits1(gb))
275 skip_bits(gb, 4); // mono_mixdown_tag
276 if (get_bits1(gb))
277 skip_bits(gb, 4); // stereo_mixdown_tag
71e9a1b8 278
cc0591da
RS
279 if (get_bits1(gb))
280 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
71e9a1b8 281
cc0591da
RS
282 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
283 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
284 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
285 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
71e9a1b8
RS
286
287 skip_bits_long(gb, 4 * num_assoc_data);
288
cc0591da 289 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
71e9a1b8
RS
290
291 align_get_bits(gb);
292
293 /* comment field, first byte is length */
294 skip_bits_long(gb, 8 * get_bits(gb, 8));
cc0591da
RS
295 return 0;
296}
71e9a1b8 297
9cc04edf
RS
298/**
299 * Set up channel positions based on a default channel configuration
300 * as specified in table 1.17.
301 *
302 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
303 *
304 * @return Returns error status. 0 - OK, !0 - error
305 */
577d383b
DB
306static int set_default_channel_config(AACContext *ac,
307 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
308 int channel_config)
9cc04edf 309{
577d383b 310 if (channel_config < 1 || channel_config > 7) {
9cc04edf
RS
311 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
312 channel_config);
313 return -1;
314 }
315
316 /* default channel configurations:
317 *
318 * 1ch : front center (mono)
319 * 2ch : L + R (stereo)
320 * 3ch : front center + L + R
321 * 4ch : front center + L + R + back center
322 * 5ch : front center + L + R + back stereo
323 * 6ch : front center + L + R + back stereo + LFE
324 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
325 */
326
577d383b 327 if (channel_config != 2)
9cc04edf 328 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
577d383b 329 if (channel_config > 1)
9cc04edf 330 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
577d383b 331 if (channel_config == 4)
9cc04edf 332 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
577d383b 333 if (channel_config > 4)
9cc04edf 334 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
577d383b
DB
335 = AAC_CHANNEL_BACK; // back stereo
336 if (channel_config > 5)
9cc04edf 337 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
577d383b 338 if (channel_config == 7)
9cc04edf
RS
339 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
340
341 return 0;
342}
343
62a57fae
RS
344/**
345 * Decode GA "General Audio" specific configuration; reference: table 4.1.
346 *
347 * @return Returns error status. 0 - OK, !0 - error
348 */
577d383b
DB
349static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
350 int channel_config)
351{
62a57fae
RS
352 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
353 int extension_flag, ret;
354
577d383b 355 if (get_bits1(gb)) { // frameLengthFlag
ce863d7f 356 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
9cc04edf
RS
357 return -1;
358 }
359
360 if (get_bits1(gb)) // dependsOnCoreCoder
361 skip_bits(gb, 14); // coreCoderDelay
362 extension_flag = get_bits1(gb);
363
577d383b
DB
364 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
365 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
9cc04edf
RS
366 skip_bits(gb, 3); // layerNr
367
368 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
369 if (channel_config == 0) {
370 skip_bits(gb, 4); // element_instance_tag
577d383b 371 if ((ret = decode_pce(ac, new_che_pos, gb)))
9cc04edf
RS
372 return ret;
373 } else {
577d383b 374 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
9cc04edf
RS
375 return ret;
376 }
577d383b 377 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config)))
9cc04edf
RS
378 return ret;
379
380 if (extension_flag) {
381 switch (ac->m4ac.object_type) {
577d383b
DB
382 case AOT_ER_BSAC:
383 skip_bits(gb, 5); // numOfSubFrame
384 skip_bits(gb, 11); // layer_length
385 break;
386 case AOT_ER_AAC_LC:
387 case AOT_ER_AAC_LTP:
388 case AOT_ER_AAC_SCALABLE:
389 case AOT_ER_AAC_LD:
390 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
9cc04edf
RS
391 * aacScalefactorDataResilienceFlag
392 * aacSpectralDataResilienceFlag
393 */
577d383b 394 break;
9cc04edf
RS
395 }
396 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
397 }
398 return 0;
399}
400
401/**
402 * Decode audio specific configuration; reference: table 1.13.
403 *
404 * @param data pointer to AVCodecContext extradata
405 * @param data_size size of AVCCodecContext extradata
406 *
407 * @return Returns error status. 0 - OK, !0 - error
408 */
577d383b
DB
409static int decode_audio_specific_config(AACContext *ac, void *data,
410 int data_size)
411{
9cc04edf
RS
412 GetBitContext gb;
413 int i;
414
415 init_get_bits(&gb, data, data_size * 8);
416
577d383b 417 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
9cc04edf 418 return -1;
577d383b 419 if (ac->m4ac.sampling_index > 12) {
9cc04edf
RS
420 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
421 return -1;
422 }
423
424 skip_bits_long(&gb, i);
425
426 switch (ac->m4ac.object_type) {
7633a041 427 case AOT_AAC_MAIN:
9cc04edf
RS
428 case AOT_AAC_LC:
429 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
430 return -1;
431 break;
432 default:
433 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
434 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
435 return -1;
436 }
437 return 0;
438}
439
62a57fae
RS
440/**
441 * linear congruential pseudorandom number generator
442 *
443 * @param previous_val pointer to the current state of the generator
444 *
445 * @return Returns a 32-bit pseudorandom integer
446 */
577d383b
DB
447static av_always_inline int lcg_random(int previous_val)
448{
62a57fae
RS
449 return previous_val * 1664525 + 1013904223;
450}
451
577d383b
DB
452static void reset_predict_state(PredictorState *ps)
453{
454 ps->r0 = 0.0f;
455 ps->r1 = 0.0f;
7633a041
AC
456 ps->cor0 = 0.0f;
457 ps->cor1 = 0.0f;
458 ps->var0 = 1.0f;
459 ps->var1 = 1.0f;
460}
461
577d383b
DB
462static void reset_all_predictors(PredictorState *ps)
463{
7633a041
AC
464 int i;
465 for (i = 0; i < MAX_PREDICTORS; i++)
466 reset_predict_state(&ps[i]);
467}
468
577d383b
DB
469static void reset_predictor_group(PredictorState *ps, int group_num)
470{
7633a041 471 int i;
577d383b 472 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
7633a041
AC
473 reset_predict_state(&ps[i]);
474}
475
577d383b
DB
476static av_cold int aac_decode_init(AVCodecContext *avccontext)
477{
478 AACContext *ac = avccontext->priv_data;
71e9a1b8
RS
479 int i;
480
481 ac->avccontext = avccontext;
482
158b3912 483 if (avccontext->extradata_size > 0) {
577d383b 484 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
158b3912
RS
485 return -1;
486 avccontext->sample_rate = ac->m4ac.sample_rate;
487 } else if (avccontext->channels > 0) {
158b3912 488 ac->m4ac.sample_rate = avccontext->sample_rate;
158b3912 489 }
cc0591da 490
577d383b
DB
491 avccontext->sample_fmt = SAMPLE_FMT_S16;
492 avccontext->frame_size = 1024;
71e9a1b8
RS
493
494 AAC_INIT_VLC_STATIC( 0, 144);
495 AAC_INIT_VLC_STATIC( 1, 114);
496 AAC_INIT_VLC_STATIC( 2, 188);
497 AAC_INIT_VLC_STATIC( 3, 180);
498 AAC_INIT_VLC_STATIC( 4, 172);
499 AAC_INIT_VLC_STATIC( 5, 140);
500 AAC_INIT_VLC_STATIC( 6, 168);
501 AAC_INIT_VLC_STATIC( 7, 114);
502 AAC_INIT_VLC_STATIC( 8, 262);
503 AAC_INIT_VLC_STATIC( 9, 248);
504 AAC_INIT_VLC_STATIC(10, 384);
505
506 dsputil_init(&ac->dsp, avccontext);
507
9cc04edf
RS
508 ac->random_state = 0x1f2e3d4c;
509
71e9a1b8
RS
510 // -1024 - Compensate wrong IMDCT method.
511 // 32768 - Required to scale values to the correct range for the bias method
512 // for float to int16 conversion.
513
577d383b
DB
514 if (ac->dsp.float_to_int16 == ff_float_to_int16_c) {
515 ac->add_bias = 385.0f;
516 ac->sf_scale = 1. / (-1024. * 32768.);
71e9a1b8
RS
517 ac->sf_offset = 0;
518 } else {
577d383b
DB
519 ac->add_bias = 0.0f;
520 ac->sf_scale = 1. / -1024.;
71e9a1b8
RS
521 ac->sf_offset = 60;
522 }
523
b250f9c6 524#if !CONFIG_HARDCODED_TABLES
d0ee5021 525 for (i = 0; i < 428; i++)
577d383b 526 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
71e9a1b8
RS
527#endif /* CONFIG_HARDCODED_TABLES */
528
37d3e066 529 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
577d383b
DB
530 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
531 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
532 352);
71e9a1b8 533
7d485f16
SS
534 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
535 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
9ffd5c1c
RS
536 // window initialization
537 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
538 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
539 ff_sine_window_init(ff_sine_1024, 1024);
540 ff_sine_window_init(ff_sine_128, 128);
541
71e9a1b8
RS
542 return 0;
543}
544
9cc04edf
RS
545/**
546 * Skip data_stream_element; reference: table 4.10.
547 */
577d383b
DB
548static void skip_data_stream_element(GetBitContext *gb)
549{
71e9a1b8
RS
550 int byte_align = get_bits1(gb);
551 int count = get_bits(gb, 8);
552 if (count == 255)
553 count += get_bits(gb, 8);
554 if (byte_align)
555 align_get_bits(gb);
556 skip_bits_long(gb, 8 * count);
557}
558
577d383b
DB
559static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
560 GetBitContext *gb)
561{
7633a041
AC
562 int sfb;
563 if (get_bits1(gb)) {
564 ics->predictor_reset_group = get_bits(gb, 5);
565 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
566 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
567 return -1;
568 }
569 }
570 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
571 ics->prediction_used[sfb] = get_bits1(gb);
572 }
573 return 0;
574}
575
71e9a1b8 576/**
9cc04edf
RS
577 * Decode Individual Channel Stream info; reference: table 4.6.
578 *
579 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
580 */
577d383b
DB
581static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
582 GetBitContext *gb, int common_window)
583{
9cc04edf
RS
584 if (get_bits1(gb)) {
585 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
586 memset(ics, 0, sizeof(IndividualChannelStream));
587 return -1;
588 }
589 ics->window_sequence[1] = ics->window_sequence[0];
590 ics->window_sequence[0] = get_bits(gb, 2);
577d383b
DB
591 ics->use_kb_window[1] = ics->use_kb_window[0];
592 ics->use_kb_window[0] = get_bits1(gb);
593 ics->num_window_groups = 1;
594 ics->group_len[0] = 1;
9ffd5c1c
RS
595 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
596 int i;
597 ics->max_sfb = get_bits(gb, 4);
598 for (i = 0; i < 7; i++) {
599 if (get_bits1(gb)) {
577d383b 600 ics->group_len[ics->num_window_groups - 1]++;
9ffd5c1c
RS
601 } else {
602 ics->num_window_groups++;
577d383b 603 ics->group_len[ics->num_window_groups - 1] = 1;
9ffd5c1c
RS
604 }
605 }
577d383b
DB
606 ics->num_windows = 8;
607 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
608 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
609 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
7633a041 610 ics->predictor_present = 0;
9ffd5c1c 611 } else {
577d383b
DB
612 ics->max_sfb = get_bits(gb, 6);
613 ics->num_windows = 1;
614 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
615 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
616 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
617 ics->predictor_present = get_bits1(gb);
7633a041
AC
618 ics->predictor_reset_group = 0;
619 if (ics->predictor_present) {
620 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
621 if (decode_prediction(ac, ics, gb)) {
622 memset(ics, 0, sizeof(IndividualChannelStream));
623 return -1;
624 }
625 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
626 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
627 memset(ics, 0, sizeof(IndividualChannelStream));
628 return -1;
629 } else {
ce863d7f 630 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
8f5aaa6d
RS
631 memset(ics, 0, sizeof(IndividualChannelStream));
632 return -1;
7633a041 633 }
62a57fae
RS
634 }
635 }
636
577d383b 637 if (ics->max_sfb > ics->num_swb) {
62a57fae 638 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b
DB
639 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
640 ics->max_sfb, ics->num_swb);
62a57fae
RS
641 memset(ics, 0, sizeof(IndividualChannelStream));
642 return -1;
643 }
644
9cc04edf
RS
645 return 0;
646}
647
648/**
9cc04edf
RS
649 * Decode band types (section_data payload); reference: table 4.46.
650 *
651 * @param band_type array of the used band type
652 * @param band_type_run_end array of the last scalefactor band of a band type run
653 *
654 * @return Returns error status. 0 - OK, !0 - error
655 */
577d383b
DB
656static int decode_band_types(AACContext *ac, enum BandType band_type[120],
657 int band_type_run_end[120], GetBitContext *gb,
658 IndividualChannelStream *ics)
659{
cc0591da
RS
660 int g, idx = 0;
661 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
662 for (g = 0; g < ics->num_window_groups; g++) {
663 int k = 0;
664 while (k < ics->max_sfb) {
665 uint8_t sect_len = k;
666 int sect_len_incr;
667 int sect_band_type = get_bits(gb, 4);
668 if (sect_band_type == 12) {
669 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
670 return -1;
671 }
577d383b 672 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
cc0591da
RS
673 sect_len += sect_len_incr;
674 sect_len += sect_len_incr;
675 if (sect_len > ics->max_sfb) {
676 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b
DB
677 "Number of bands (%d) exceeds limit (%d).\n",
678 sect_len, ics->max_sfb);
cc0591da
RS
679 return -1;
680 }
9ffd5c1c
RS
681 for (; k < sect_len; k++) {
682 band_type [idx] = sect_band_type;
683 band_type_run_end[idx++] = sect_len;
684 }
9cc04edf
RS
685 }
686 }
687 return 0;
688}
cc0591da 689
9cc04edf
RS
690/**
691 * Decode scalefactors; reference: table 4.47.
cc0591da 692 *
cc0591da
RS
693 * @param global_gain first scalefactor value as scalefactors are differentially coded
694 * @param band_type array of the used band type
695 * @param band_type_run_end array of the last scalefactor band of a band type run
696 * @param sf array of scalefactors or intensity stereo positions
697 *
698 * @return Returns error status. 0 - OK, !0 - error
699 */
577d383b
DB
700static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
701 unsigned int global_gain,
702 IndividualChannelStream *ics,
703 enum BandType band_type[120],
704 int band_type_run_end[120])
705{
cc0591da
RS
706 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
707 int g, i, idx = 0;
708 int offset[3] = { global_gain, global_gain - 90, 100 };
709 int noise_flag = 1;
710 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
cc0591da
RS
711 for (g = 0; g < ics->num_window_groups; g++) {
712 for (i = 0; i < ics->max_sfb;) {
713 int run_end = band_type_run_end[idx];
714 if (band_type[idx] == ZERO_BT) {
577d383b 715 for (; i < run_end; i++, idx++)
cc0591da 716 sf[idx] = 0.;
577d383b
DB
717 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
718 for (; i < run_end; i++, idx++) {
cc0591da 719 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 720 if (offset[2] > 255U) {
cc0591da 721 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 722 "%s (%d) out of range.\n", sf_str[2], offset[2]);
cc0591da
RS
723 return -1;
724 }
577d383b 725 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
cc0591da 726 }
577d383b
DB
727 } else if (band_type[idx] == NOISE_BT) {
728 for (; i < run_end; i++, idx++) {
729 if (noise_flag-- > 0)
cc0591da
RS
730 offset[1] += get_bits(gb, 9) - 256;
731 else
732 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 733 if (offset[1] > 255U) {
cc0591da 734 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 735 "%s (%d) out of range.\n", sf_str[1], offset[1]);
cc0591da
RS
736 return -1;
737 }
577d383b 738 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
cc0591da 739 }
577d383b
DB
740 } else {
741 for (; i < run_end; i++, idx++) {
cc0591da 742 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 743 if (offset[0] > 255U) {
cc0591da 744 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b 745 "%s (%d) out of range.\n", sf_str[0], offset[0]);
cc0591da
RS
746 return -1;
747 }
748 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
cc0591da
RS
749 }
750 }
751 }
752 }
753 return 0;
754}
755
756/**
757 * Decode pulse data; reference: table 4.7.
758 */
577d383b
DB
759static int decode_pulses(Pulse *pulse, GetBitContext *gb,
760 const uint16_t *swb_offset, int num_swb)
761{
aac0eda4 762 int i, pulse_swb;
cc0591da 763 pulse->num_pulse = get_bits(gb, 2) + 1;
aac0eda4
AC
764 pulse_swb = get_bits(gb, 6);
765 if (pulse_swb >= num_swb)
766 return -1;
767 pulse->pos[0] = swb_offset[pulse_swb];
408992ba 768 pulse->pos[0] += get_bits(gb, 5);
aac0eda4
AC
769 if (pulse->pos[0] > 1023)
770 return -1;
848a5815
RS
771 pulse->amp[0] = get_bits(gb, 4);
772 for (i = 1; i < pulse->num_pulse; i++) {
577d383b 773 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
aac0eda4
AC
774 if (pulse->pos[i] > 1023)
775 return -1;
848a5815 776 pulse->amp[i] = get_bits(gb, 4);
cc0591da 777 }
aac0eda4 778 return 0;
cc0591da
RS
779}
780
781/**
1dece0d2
RS
782 * Decode Temporal Noise Shaping data; reference: table 4.48.
783 *
784 * @return Returns error status. 0 - OK, !0 - error
785 */
577d383b
DB
786static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
787 GetBitContext *gb, const IndividualChannelStream *ics)
788{
1dece0d2
RS
789 int w, filt, i, coef_len, coef_res, coef_compress;
790 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
791 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
792 for (w = 0; w < ics->num_windows; w++) {
fbd91d7c 793 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1dece0d2
RS
794 coef_res = get_bits1(gb);
795
65b20b24
RS
796 for (filt = 0; filt < tns->n_filt[w]; filt++) {
797 int tmp2_idx;
577d383b 798 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
65b20b24 799
577d383b 800 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
65b20b24
RS
801 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
802 tns->order[w][filt], tns_max_order);
803 tns->order[w][filt] = 0;
804 return -1;
805 }
51673647 806 if (tns->order[w][filt]) {
35445d29
RS
807 tns->direction[w][filt] = get_bits1(gb);
808 coef_compress = get_bits1(gb);
809 coef_len = coef_res + 3 - coef_compress;
577d383b 810 tmp2_idx = 2 * coef_compress + coef_res;
1dece0d2 811
35445d29
RS
812 for (i = 0; i < tns->order[w][filt]; i++)
813 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
51673647 814 }
65b20b24 815 }
fbd91d7c 816 }
1dece0d2
RS
817 }
818 return 0;
819}
820
821/**
9cc04edf
RS
822 * Decode Mid/Side data; reference: table 4.54.
823 *
824 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
825 * [1] mask is decoded from bitstream; [2] mask is all 1s;
826 * [3] reserved for scalable AAC
827 */
577d383b
DB
828static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
829 int ms_present)
830{
62a57fae
RS
831 int idx;
832 if (ms_present == 1) {
833 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
834 cpe->ms_mask[idx] = get_bits1(gb);
835 } else if (ms_present == 2) {
836 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
837 }
838}
9cc04edf
RS
839
840/**
9ffd5c1c
RS
841 * Decode spectral data; reference: table 4.50.
842 * Dequantize and scale spectral data; reference: 4.6.3.3.
843 *
844 * @param coef array of dequantized, scaled spectral data
845 * @param sf array of scalefactors or intensity stereo positions
846 * @param pulse_present set if pulses are present
847 * @param pulse pointer to pulse data struct
848 * @param band_type array of the used band type
849 *
850 * @return Returns error status. 0 - OK, !0 - error
851 */
577d383b
DB
852static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
853 GetBitContext *gb, float sf[120],
854 int pulse_present, const Pulse *pulse,
855 const IndividualChannelStream *ics,
856 enum BandType band_type[120])
857{
9ffd5c1c 858 int i, k, g, idx = 0;
577d383b
DB
859 const int c = 1024 / ics->num_windows;
860 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c 861 float *coef_base = coef;
4a904903 862 static const float sign_lookup[] = { 1.0f, -1.0f };
9ffd5c1c
RS
863
864 for (g = 0; g < ics->num_windows; g++)
577d383b 865 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
9ffd5c1c
RS
866
867 for (g = 0; g < ics->num_window_groups; g++) {
868 for (i = 0; i < ics->max_sfb; i++, idx++) {
869 const int cur_band_type = band_type[idx];
870 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
871 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
872 int group;
aab54133 873 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
9ffd5c1c 874 for (group = 0; group < ics->group_len[g]; group++) {
577d383b 875 memset(coef + group * 128 + offsets[i], 0, (offsets[i + 1] - offsets[i]) * sizeof(float));
9ffd5c1c 876 }
577d383b 877 } else if (cur_band_type == NOISE_BT) {
9ffd5c1c 878 for (group = 0; group < ics->group_len[g]; group++) {
d0ee5021 879 float scale;
b418a6ca 880 float band_energy;
0cb71412 881 float *cf = coef + group * 128 + offsets[i];
42d3fbb3
MR
882 int len = offsets[i+1] - offsets[i];
883
b418a6ca 884 for (k = 0; k < len; k++) {
9ffd5c1c 885 ac->random_state = lcg_random(ac->random_state);
b418a6ca 886 cf[k] = ac->random_state;
d0ee5021 887 }
42d3fbb3 888
b418a6ca 889 band_energy = ac->dsp.scalarproduct_float(cf, cf, len);
d0ee5021 890 scale = sf[idx] / sqrtf(band_energy);
42d3fbb3 891 ac->dsp.vector_fmul_scalar(cf, cf, scale, len);
9ffd5c1c 892 }
577d383b 893 } else {
9ffd5c1c 894 for (group = 0; group < ics->group_len[g]; group++) {
42d3fbb3
MR
895 const float *vq[96];
896 const float **vqp = vq;
897 float *cf = coef + (group << 7) + offsets[i];
898 int len = offsets[i + 1] - offsets[i];
899
577d383b 900 for (k = offsets[i]; k < offsets[i + 1]; k += dim) {
9ffd5c1c
RS
901 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
902 const int coef_tmp_idx = (group << 7) + k;
903 const float *vq_ptr;
904 int j;
577d383b 905 if (index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
9ffd5c1c 906 av_log(ac->avccontext, AV_LOG_ERROR,
577d383b
DB
907 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
908 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
9ffd5c1c
RS
909 return -1;
910 }
911 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
42d3fbb3 912 *vqp++ = vq_ptr;
9ffd5c1c 913 if (is_cb_unsigned) {
577d383b
DB
914 if (vq_ptr[0])
915 coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
916 if (vq_ptr[1])
917 coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
c0893c3a 918 if (dim == 4) {
577d383b
DB
919 if (vq_ptr[2])
920 coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
921 if (vq_ptr[3])
922 coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
c0893c3a 923 }
e8d5c07b
AC
924 if (cur_band_type == ESC_BT) {
925 for (j = 0; j < 2; j++) {
926 if (vq_ptr[j] == 64.0f) {
927 int n = 4;
928 /* The total length of escape_sequence must be < 22 bits according
929 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
930 while (get_bits1(gb) && n < 15) n++;
577d383b 931 if (n == 15) {
e8d5c07b
AC
932 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
933 return -1;
934 }
577d383b 935 n = (1 << n) + get_bits(gb, n);
e8d5c07b 936 coef[coef_tmp_idx + j] *= cbrtf(n) * n;
577d383b 937 } else
e8d5c07b
AC
938 coef[coef_tmp_idx + j] *= vq_ptr[j];
939 }
9ffd5c1c 940 }
c0893c3a 941 }
9ffd5c1c 942 }
42d3fbb3
MR
943
944 if (is_cb_unsigned && cur_band_type != ESC_BT) {
945 ac->dsp.vector_fmul_sv_scalar[dim>>2](
946 cf, cf, vq, sf[idx], len);
5419da8d 947 } else if (cur_band_type == ESC_BT) {
42d3fbb3
MR
948 ac->dsp.vector_fmul_scalar(cf, cf, sf[idx], len);
949 } else { /* !is_cb_unsigned */
950 ac->dsp.sv_fmul_scalar[dim>>2](cf, vq, sf[idx], len);
951 }
9ffd5c1c
RS
952 }
953 }
954 }
577d383b 955 coef += ics->group_len[g] << 7;
9ffd5c1c
RS
956 }
957
958 if (pulse_present) {
51436848 959 idx = 0;
577d383b
DB
960 for (i = 0; i < pulse->num_pulse; i++) {
961 float co = coef_base[ pulse->pos[i] ];
962 while (offsets[idx + 1] <= pulse->pos[i])
51436848
AC
963 idx++;
964 if (band_type[idx] != NOISE_BT && sf[idx]) {
70735a3f
RS
965 float ico = -pulse->amp[i];
966 if (co) {
967 co /= sf[idx];
968 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
969 }
970 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
51436848 971 }
9ffd5c1c
RS
972 }
973 }
974 return 0;
975}
976
577d383b
DB
977static av_always_inline float flt16_round(float pf)
978{
4a39ccb4
AC
979 union float754 tmp;
980 tmp.f = pf;
981 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
982 return tmp.f;
7633a041
AC
983}
984
577d383b
DB
985static av_always_inline float flt16_even(float pf)
986{
4a39ccb4
AC
987 union float754 tmp;
988 tmp.f = pf;
577d383b 989 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
4a39ccb4 990 return tmp.f;
7633a041
AC
991}
992
577d383b
DB
993static av_always_inline float flt16_trunc(float pf)
994{
4a39ccb4
AC
995 union float754 pun;
996 pun.f = pf;
997 pun.i &= 0xFFFF0000U;
998 return pun.f;
7633a041
AC
999}
1000
577d383b
DB
1001static void predict(AACContext *ac, PredictorState *ps, float *coef,
1002 int output_enable)
1003{
1004 const float a = 0.953125; // 61.0 / 64
1005 const float alpha = 0.90625; // 29.0 / 32
7633a041
AC
1006 float e0, e1;
1007 float pv;
1008 float k1, k2;
1009
1010 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1011 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1012
1013 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1014 if (output_enable)
1015 *coef += pv * ac->sf_scale;
1016
1017 e0 = *coef / ac->sf_scale;
1018 e1 = e0 - k1 * ps->r0;
1019
1020 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1021 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1022 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1023 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1024
1025 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1026 ps->r0 = flt16_trunc(a * e0);
1027}
1028
1029/**
1030 * Apply AAC-Main style frequency domain prediction.
1031 */
577d383b
DB
1032static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1033{
7633a041
AC
1034 int sfb, k;
1035
1036 if (!sce->ics.predictor_initialized) {
aab54133 1037 reset_all_predictors(sce->predictor_state);
7633a041
AC
1038 sce->ics.predictor_initialized = 1;
1039 }
1040
1041 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1042 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1043 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
aab54133 1044 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
577d383b 1045 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
7633a041
AC
1046 }
1047 }
1048 if (sce->ics.predictor_reset_group)
aab54133 1049 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
7633a041 1050 } else
aab54133 1051 reset_all_predictors(sce->predictor_state);
7633a041
AC
1052}
1053
9ffd5c1c 1054/**
9cc04edf
RS
1055 * Decode an individual_channel_stream payload; reference: table 4.44.
1056 *
1057 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1058 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1059 *
1060 * @return Returns error status. 0 - OK, !0 - error
1061 */
577d383b
DB
1062static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1063 GetBitContext *gb, int common_window, int scale_flag)
1064{
9cc04edf 1065 Pulse pulse;
577d383b
DB
1066 TemporalNoiseShaping *tns = &sce->tns;
1067 IndividualChannelStream *ics = &sce->ics;
1068 float *out = sce->coeffs;
9cc04edf
RS
1069 int global_gain, pulse_present = 0;
1070
848a5815
RS
1071 /* This assignment is to silence a GCC warning about the variable being used
1072 * uninitialized when in fact it always is.
9cc04edf
RS
1073 */
1074 pulse.num_pulse = 0;
9cc04edf
RS
1075
1076 global_gain = get_bits(gb, 8);
1077
1078 if (!common_window && !scale_flag) {
1079 if (decode_ics_info(ac, ics, gb, 0) < 0)
1080 return -1;
1081 }
1082
1083 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1084 return -1;
1085 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1086 return -1;
1087
1088 pulse_present = 0;
1089 if (!scale_flag) {
1090 if ((pulse_present = get_bits1(gb))) {
1091 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1092 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1093 return -1;
1094 }
aac0eda4
AC
1095 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1096 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1097 return -1;
1098 }
9cc04edf
RS
1099 }
1100 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1101 return -1;
1102 if (get_bits1(gb)) {
ce863d7f 1103 av_log_missing_feature(ac->avccontext, "SSR", 1);
9cc04edf
RS
1104 return -1;
1105 }
1106 }
1107
848a5815 1108 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
9cc04edf 1109 return -1;
7633a041 1110
577d383b 1111 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
7633a041
AC
1112 apply_prediction(ac, sce);
1113
9cc04edf
RS
1114 return 0;
1115}
1116
1117/**
9ffd5c1c
RS
1118 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1119 */
42d3fbb3 1120static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
577d383b
DB
1121{
1122 const IndividualChannelStream *ics = &cpe->ch[0].ics;
9ffd5c1c
RS
1123 float *ch0 = cpe->ch[0].coeffs;
1124 float *ch1 = cpe->ch[1].coeffs;
42d3fbb3 1125 int g, i, group, idx = 0;
577d383b 1126 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c
RS
1127 for (g = 0; g < ics->num_window_groups; g++) {
1128 for (i = 0; i < ics->max_sfb; i++, idx++) {
1129 if (cpe->ms_mask[idx] &&
577d383b 1130 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
9ffd5c1c 1131 for (group = 0; group < ics->group_len[g]; group++) {
42d3fbb3
MR
1132 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1133 ch1 + group * 128 + offsets[i],
1134 offsets[i+1] - offsets[i]);
9ffd5c1c
RS
1135 }
1136 }
1137 }
577d383b
DB
1138 ch0 += ics->group_len[g] * 128;
1139 ch1 += ics->group_len[g] * 128;
9ffd5c1c
RS
1140 }
1141}
1142
1143/**
1144 * intensity stereo decoding; reference: 4.6.8.2.3
1145 *
1146 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1147 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1148 * [3] reserved for scalable AAC
1149 */
577d383b
DB
1150static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1151{
1152 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1153 SingleChannelElement *sce1 = &cpe->ch[1];
9ffd5c1c 1154 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
577d383b 1155 const uint16_t *offsets = ics->swb_offset;
9ffd5c1c
RS
1156 int g, group, i, k, idx = 0;
1157 int c;
1158 float scale;
1159 for (g = 0; g < ics->num_window_groups; g++) {
1160 for (i = 0; i < ics->max_sfb;) {
1161 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1162 const int bt_run_end = sce1->band_type_run_end[idx];
1163 for (; i < bt_run_end; i++, idx++) {
1164 c = -1 + 2 * (sce1->band_type[idx] - 14);
1165 if (ms_present)
1166 c *= 1 - 2 * cpe->ms_mask[idx];
1167 scale = c * sce1->sf[idx];
1168 for (group = 0; group < ics->group_len[g]; group++)
577d383b
DB
1169 for (k = offsets[i]; k < offsets[i + 1]; k++)
1170 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
9ffd5c1c
RS
1171 }
1172 } else {
1173 int bt_run_end = sce1->band_type_run_end[idx];
1174 idx += bt_run_end - i;
1175 i = bt_run_end;
1176 }
1177 }
577d383b
DB
1178 coef0 += ics->group_len[g] * 128;
1179 coef1 += ics->group_len[g] * 128;
9ffd5c1c
RS
1180 }
1181}
1182
1183/**
9cc04edf
RS
1184 * Decode a channel_pair_element; reference: table 4.4.
1185 *
1186 * @param elem_id Identifies the instance of a syntax element.
1187 *
1188 * @return Returns error status. 0 - OK, !0 - error
1189 */
577d383b
DB
1190static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1191{
9cc04edf 1192 int i, ret, common_window, ms_present = 0;
9cc04edf 1193
9cc04edf
RS
1194 common_window = get_bits1(gb);
1195 if (common_window) {
1196 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1197 return -1;
1198 i = cpe->ch[1].ics.use_kb_window[0];
1199 cpe->ch[1].ics = cpe->ch[0].ics;
1200 cpe->ch[1].ics.use_kb_window[1] = i;
1201 ms_present = get_bits(gb, 2);
577d383b 1202 if (ms_present == 3) {
9cc04edf
RS
1203 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1204 return -1;
577d383b 1205 } else if (ms_present)
9cc04edf
RS
1206 decode_mid_side_stereo(cpe, gb, ms_present);
1207 }
1208 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1209 return ret;
1210 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1211 return ret;
1212
aab54133
AC
1213 if (common_window) {
1214 if (ms_present)
42d3fbb3 1215 apply_mid_side_stereo(ac, cpe);
aab54133
AC
1216 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1217 apply_prediction(ac, &cpe->ch[0]);
1218 apply_prediction(ac, &cpe->ch[1]);
1219 }
1220 }
9cc04edf 1221
848a5815 1222 apply_intensity_stereo(cpe, ms_present);
9cc04edf
RS
1223 return 0;
1224}
1225
9ffd5c1c
RS
1226/**
1227 * Decode coupling_channel_element; reference: table 4.8.
1228 *
1229 * @param elem_id Identifies the instance of a syntax element.
1230 *
1231 * @return Returns error status. 0 - OK, !0 - error
1232 */
577d383b
DB
1233static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1234{
9ffd5c1c 1235 int num_gain = 0;
341b28c0 1236 int c, g, sfb, ret;
9ffd5c1c
RS
1237 int sign;
1238 float scale;
577d383b
DB
1239 SingleChannelElement *sce = &che->ch[0];
1240 ChannelCoupling *coup = &che->coup;
9ffd5c1c 1241
577d383b 1242 coup->coupling_point = 2 * get_bits1(gb);
62a57fae
RS
1243 coup->num_coupled = get_bits(gb, 3);
1244 for (c = 0; c <= coup->num_coupled; c++) {
1245 num_gain++;
1246 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1247 coup->id_select[c] = get_bits(gb, 4);
1248 if (coup->type[c] == TYPE_CPE) {
1249 coup->ch_select[c] = get_bits(gb, 2);
1250 if (coup->ch_select[c] == 3)
1251 num_gain++;
1252 } else
88de95c2 1253 coup->ch_select[c] = 2;
62a57fae 1254 }
577d383b 1255 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
62a57fae 1256
577d383b 1257 sign = get_bits(gb, 1);
c8947a56 1258 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
62a57fae
RS
1259
1260 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1261 return ret;
1262
1263 for (c = 0; c < num_gain; c++) {
577d383b
DB
1264 int idx = 0;
1265 int cge = 1;
62a57fae
RS
1266 int gain = 0;
1267 float gain_cache = 1.;
1268 if (c) {
1269 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1270 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
88de95c2 1271 gain_cache = pow(scale, -gain);
62a57fae 1272 }
f1ade11e
AC
1273 if (coup->coupling_point == AFTER_IMDCT) {
1274 coup->gain[c][0] = gain_cache;
1275 } else {
03b12747
AC
1276 for (g = 0; g < sce->ics.num_window_groups; g++) {
1277 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1278 if (sce->band_type[idx] != ZERO_BT) {
1279 if (!cge) {
1280 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
577d383b 1281 if (t) {
03b12747
AC
1282 int s = 1;
1283 t = gain += t;
1284 if (sign) {
1285 s -= 2 * (t & 0x1);
1286 t >>= 1;
1287 }
1288 gain_cache = pow(scale, -t) * s;
62a57fae 1289 }
62a57fae 1290 }
03b12747 1291 coup->gain[c][idx] = gain_cache;
62a57fae 1292 }
62a57fae 1293 }
f80a8ca5
RS
1294 }
1295 }
62a57fae
RS
1296 }
1297 return 0;
1298}
1299
9cc04edf
RS
1300/**
1301 * Decode Spectral Band Replication extension data; reference: table 4.55.
cc0591da
RS
1302 *
1303 * @param crc flag indicating the presence of CRC checksum
1304 * @param cnt length of TYPE_FIL syntactic element in bytes
9cc04edf 1305 *
cc0591da
RS
1306 * @return Returns number of bytes consumed from the TYPE_FIL element.
1307 */
577d383b
DB
1308static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1309 int crc, int cnt)
1310{
cc0591da 1311 // TODO : sbr_extension implementation
ce863d7f 1312 av_log_missing_feature(ac->avccontext, "SBR", 0);
577d383b 1313 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
cc0591da
RS
1314 return cnt;
1315}
1316
9cc04edf 1317/**
62a57fae
RS
1318 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1319 *
1320 * @return Returns number of bytes consumed.
1321 */
577d383b
DB
1322static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1323 GetBitContext *gb)
1324{
62a57fae
RS
1325 int i;
1326 int num_excl_chan = 0;
1327
1328 do {
1329 for (i = 0; i < 7; i++)
1330 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1331 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1332
1333 return num_excl_chan / 7;
1334}
1335
1336/**
9cc04edf
RS
1337 * Decode dynamic range information; reference: table 4.52.
1338 *
1339 * @param cnt length of TYPE_FIL syntactic element in bytes
1340 *
1341 * @return Returns number of bytes consumed.
1342 */
577d383b
DB
1343static int decode_dynamic_range(DynamicRangeControl *che_drc,
1344 GetBitContext *gb, int cnt)
1345{
1346 int n = 1;
9cc04edf
RS
1347 int drc_num_bands = 1;
1348 int i;
1349
1350 /* pce_tag_present? */
577d383b 1351 if (get_bits1(gb)) {
9cc04edf
RS
1352 che_drc->pce_instance_tag = get_bits(gb, 4);
1353 skip_bits(gb, 4); // tag_reserved_bits
1354 n++;
1355 }
1356
1357 /* excluded_chns_present? */
577d383b 1358 if (get_bits1(gb)) {
9cc04edf
RS
1359 n += decode_drc_channel_exclusions(che_drc, gb);
1360 }
1361
1362 /* drc_bands_present? */
1363 if (get_bits1(gb)) {
1364 che_drc->band_incr = get_bits(gb, 4);
1365 che_drc->interpolation_scheme = get_bits(gb, 4);
1366 n++;
1367 drc_num_bands += che_drc->band_incr;
1368 for (i = 0; i < drc_num_bands; i++) {
1369 che_drc->band_top[i] = get_bits(gb, 8);
1370 n++;
1371 }
1372 }
1373
1374 /* prog_ref_level_present? */
1375 if (get_bits1(gb)) {
1376 che_drc->prog_ref_level = get_bits(gb, 7);
1377 skip_bits1(gb); // prog_ref_level_reserved_bits
1378 n++;
1379 }
1380
1381 for (i = 0; i < drc_num_bands; i++) {
1382 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1383 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1384 n++;
1385 }
1386
1387 return n;
1388}
1389
1390/**
1391 * Decode extension data (incomplete); reference: table 4.51.
1392 *
1393 * @param cnt length of TYPE_FIL syntactic element in bytes
1394 *
1395 * @return Returns number of bytes consumed
1396 */
577d383b
DB
1397static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1398{
cc0591da
RS
1399 int crc_flag = 0;
1400 int res = cnt;
1401 switch (get_bits(gb, 4)) { // extension type
577d383b
DB
1402 case EXT_SBR_DATA_CRC:
1403 crc_flag++;
1404 case EXT_SBR_DATA:
1405 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1406 break;
1407 case EXT_DYNAMIC_RANGE:
1408 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1409 break;
1410 case EXT_FILL:
1411 case EXT_FILL_DATA:
1412 case EXT_DATA_ELEMENT:
1413 default:
1414 skip_bits_long(gb, 8 * cnt - 4);
1415 break;
cc0591da
RS
1416 };
1417 return res;
1418}
1419
7d8f3de4
RS
1420/**
1421 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1422 *
1423 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1424 * @param coef spectral coefficients
1425 */
577d383b
DB
1426static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1427 IndividualChannelStream *ics, int decode)
1428{
1429 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1098e8d2 1430 int w, filt, m, i;
7d8f3de4
RS
1431 int bottom, top, order, start, end, size, inc;
1432 float lpc[TNS_MAX_ORDER];
1433
1434 for (w = 0; w < ics->num_windows; w++) {
1435 bottom = ics->num_swb;
1436 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1437 top = bottom;
1438 bottom = FFMAX(0, top - tns->length[w][filt]);
1439 order = tns->order[w][filt];
1440 if (order == 0)
1441 continue;
1442
1be0fc29
VS
1443 // tns_decode_coef
1444 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
7d8f3de4 1445
1dece0d2
RS
1446 start = ics->swb_offset[FFMIN(bottom, mmm)];
1447 end = ics->swb_offset[FFMIN( top, mmm)];
1448 if ((size = end - start) <= 0)
1449 continue;
1450 if (tns->direction[w][filt]) {
577d383b
DB
1451 inc = -1;
1452 start = end - 1;
1dece0d2
RS
1453 } else {
1454 inc = 1;
1455 }
1456 start += w * 128;
1457
1458 // ar filter
1459 for (m = 0; m < size; m++, start += inc)
1460 for (i = 1; i <= FFMIN(m, order); i++)
577d383b 1461 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1dece0d2
RS
1462 }
1463 }
1464}
1465
cc0591da 1466/**
9cc04edf
RS
1467 * Conduct IMDCT and windowing.
1468 */
577d383b
DB
1469static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1470{
1471 IndividualChannelStream *ics = &sce->ics;
1472 float *in = sce->coeffs;
1473 float *out = sce->ret;
1474 float *saved = sce->saved;
1475 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1476 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1477 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1478 float *buf = ac->buf_mdct;
1479 float *temp = ac->temp;
9cc04edf
RS
1480 int i;
1481
f4990558 1482 // imdct
62a57fae
RS
1483 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1484 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1485 av_log(ac->avccontext, AV_LOG_WARNING,
1486 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1487 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
b0f5852a
RS
1488 for (i = 0; i < 1024; i += 128)
1489 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
f4990558 1490 } else
b0f5852a 1491 ff_imdct_half(&ac->mdct, buf, in);
f4990558
RS
1492
1493 /* window overlapping
1494 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1495 * and long to short transitions are considered to be short to short
1496 * transitions. This leaves just two cases (long to long and short to short)
1497 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1498 */
1499 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
577d383b 1500 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
b0f5852a 1501 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
f4990558 1502 } else {
db38c386
RS
1503 for (i = 0; i < 448; i++)
1504 out[i] = saved[i] + ac->add_bias;
62a57fae 1505
f4990558 1506 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1507 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1508 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1509 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1510 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1511 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1512 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
f4990558 1513 } else {
b0f5852a 1514 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
db38c386 1515 for (i = 576; i < 1024; i++)
b0f5852a 1516 out[i] = buf[i-512] + ac->add_bias;
f4990558
RS
1517 }
1518 }
62a57fae 1519
f4990558
RS
1520 // buffer update
1521 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1522 for (i = 0; i < 64; i++)
1523 saved[i] = temp[64 + i] - ac->add_bias;
1524 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1525 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1526 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1527 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1528 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
b0f5852a
RS
1529 memcpy( saved, buf + 512, 448 * sizeof(float));
1530 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1531 } else { // LONG_STOP or ONLY_LONG
b0f5852a 1532 memcpy( saved, buf + 512, 512 * sizeof(float));
62a57fae
RS
1533 }
1534}
1535
9cc04edf 1536/**
cc0591da
RS
1537 * Apply dependent channel coupling (applied before IMDCT).
1538 *
1539 * @param index index into coupling gain array
1540 */
577d383b
DB
1541static void apply_dependent_coupling(AACContext *ac,
1542 SingleChannelElement *target,
1543 ChannelElement *cce, int index)
1544{
1545 IndividualChannelStream *ics = &cce->ch[0].ics;
1546 const uint16_t *offsets = ics->swb_offset;
1547 float *dest = target->coeffs;
1548 const float *src = cce->ch[0].coeffs;
cc0591da 1549 int g, i, group, k, idx = 0;
577d383b 1550 if (ac->m4ac.object_type == AOT_AAC_LTP) {
cc0591da
RS
1551 av_log(ac->avccontext, AV_LOG_ERROR,
1552 "Dependent coupling is not supported together with LTP\n");
1553 return;
1554 }
1555 for (g = 0; g < ics->num_window_groups; g++) {
1556 for (i = 0; i < ics->max_sfb; i++, idx++) {
fbdae895 1557 if (cce->ch[0].band_type[idx] != ZERO_BT) {
cfd937b0 1558 const float gain = cce->coup.gain[index][idx];
cc0591da 1559 for (group = 0; group < ics->group_len[g]; group++) {
577d383b 1560 for (k = offsets[i]; k < offsets[i + 1]; k++) {
cc0591da 1561 // XXX dsputil-ize
577d383b 1562 dest[group * 128 + k] += gain * src[group * 128 + k];
cc0591da
RS
1563 }
1564 }
1565 }
1566 }
577d383b
DB
1567 dest += ics->group_len[g] * 128;
1568 src += ics->group_len[g] * 128;
cc0591da
RS
1569 }
1570}
1571
1572/**
1573 * Apply independent channel coupling (applied after IMDCT).
1574 *
1575 * @param index index into coupling gain array
1576 */
577d383b
DB
1577static void apply_independent_coupling(AACContext *ac,
1578 SingleChannelElement *target,
1579 ChannelElement *cce, int index)
1580{
cc0591da 1581 int i;
039821a8
AC
1582 const float gain = cce->coup.gain[index][0];
1583 const float bias = ac->add_bias;
577d383b
DB
1584 const float *src = cce->ch[0].ret;
1585 float *dest = target->ret;
039821a8 1586
cc0591da 1587 for (i = 0; i < 1024; i++)
039821a8 1588 dest[i] += gain * (src[i] - bias);
cc0591da
RS
1589}
1590
9ffd5c1c
RS
1591/**
1592 * channel coupling transformation interface
1593 *
1594 * @param index index into coupling gain array
1595 * @param apply_coupling_method pointer to (in)dependent coupling function
1596 */
577d383b
DB
1597static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1598 enum RawDataBlockType type, int elem_id,
1599 enum CouplingPoint coupling_point,
1600 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
9ffd5c1c 1601{
88de95c2
AC
1602 int i, c;
1603
1604 for (i = 0; i < MAX_ELEM_ID; i++) {
1605 ChannelElement *cce = ac->che[TYPE_CCE][i];
1606 int index = 0;
1607
1608 if (cce && cce->coup.coupling_point == coupling_point) {
577d383b 1609 ChannelCoupling *coup = &cce->coup;
88de95c2
AC
1610
1611 for (c = 0; c <= coup->num_coupled; c++) {
1612 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1613 if (coup->ch_select[c] != 1) {
1614 apply_coupling_method(ac, &cc->ch[0], cce, index);
1615 if (coup->ch_select[c] != 0)
1616 index++;
1617 }
1618 if (coup->ch_select[c] != 2)
1619 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1620 } else
1621 index += 1 + (coup->ch_select[c] == 3);
9ffd5c1c 1622 }
9ffd5c1c
RS
1623 }
1624 }
1625}
1626
1627/**
1628 * Convert spectral data to float samples, applying all supported tools as appropriate.
1629 */
577d383b
DB
1630static void spectral_to_sample(AACContext *ac)
1631{
6735e2c8 1632 int i, type;
577d383b 1633 for (type = 3; type >= 0; type--) {
88de95c2 1634 for (i = 0; i < MAX_ELEM_ID; i++) {
9ffd5c1c 1635 ChannelElement *che = ac->che[type][i];
577d383b
DB
1636 if (che) {
1637 if (type <= TYPE_CPE)
88de95c2 1638 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
577d383b 1639 if (che->ch[0].tns.present)
9ffd5c1c 1640 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
577d383b 1641 if (che->ch[1].tns.present)
9ffd5c1c 1642 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
577d383b 1643 if (type <= TYPE_CPE)
88de95c2 1644 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
577d383b 1645 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
88de95c2 1646 imdct_and_windowing(ac, &che->ch[0]);
577d383b 1647 if (type == TYPE_CPE)
9ffd5c1c 1648 imdct_and_windowing(ac, &che->ch[1]);
577d383b 1649 if (type <= TYPE_CCE)
88de95c2 1650 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
62a57fae
RS
1651 }
1652 }
1653 }
1654}
1655
577d383b
DB
1656static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1657{
158b3912
RS
1658 int size;
1659 AACADTSHeaderInfo hdr_info;
1660
1661 size = ff_aac_parse_header(gb, &hdr_info);
1662 if (size > 0) {
6308765c
AC
1663 if (!ac->output_configured && hdr_info.chan_config) {
1664 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1665 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
158b3912 1666 ac->m4ac.chan_config = hdr_info.chan_config;
6308765c
AC
1667 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1668 return -7;
1669 if (output_configure(ac, ac->che_pos, new_che_pos, 1))
1670 return -7;
1671 }
158b3912
RS
1672 ac->m4ac.sample_rate = hdr_info.sample_rate;
1673 ac->m4ac.sampling_index = hdr_info.sampling_index;
1674 ac->m4ac.object_type = hdr_info.object_type;
7d87e2ce
AC
1675 if (hdr_info.num_aac_frames == 1) {
1676 if (!hdr_info.crc_absent)
1677 skip_bits(gb, 16);
1678 } else {
ce863d7f 1679 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
7d87e2ce
AC
1680 return -1;
1681 }
51741a82 1682 }
158b3912
RS
1683 return size;
1684}
1685
577d383b
DB
1686static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1687 int *data_size, AVPacket *avpkt)
1688{
7a00bbad
TB
1689 const uint8_t *buf = avpkt->data;
1690 int buf_size = avpkt->size;
577d383b
DB
1691 AACContext *ac = avccontext->priv_data;
1692 ChannelElement *che = NULL;
62a57fae
RS
1693 GetBitContext gb;
1694 enum RawDataBlockType elem_type;
1695 int err, elem_id, data_size_tmp;
1696
577d383b 1697 init_get_bits(&gb, buf, buf_size * 8);
62a57fae 1698
158b3912 1699 if (show_bits(&gb, 12) == 0xfff) {
5967e141 1700 if (parse_adts_frame_header(ac, &gb) < 0) {
158b3912
RS
1701 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1702 return -1;
1703 }
30272450 1704 if (ac->m4ac.sampling_index > 12) {
f418b861
JM
1705 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1706 return -1;
1707 }
158b3912
RS
1708 }
1709
62a57fae
RS
1710 // parse
1711 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1712 elem_id = get_bits(&gb, 4);
62a57fae 1713
577d383b 1714 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
5f401ee0
RS
1715 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1716 return -1;
62a57fae
RS
1717 }
1718
1719 switch (elem_type) {
1720
1721 case TYPE_SCE:
bb5c0988 1722 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
62a57fae
RS
1723 break;
1724
1725 case TYPE_CPE:
bb5c0988 1726 err = decode_cpe(ac, &gb, che);
62a57fae
RS
1727 break;
1728
1729 case TYPE_CCE:
bb5c0988 1730 err = decode_cce(ac, &gb, che);
62a57fae
RS
1731 break;
1732
1733 case TYPE_LFE:
bb5c0988 1734 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
62a57fae
RS
1735 break;
1736
1737 case TYPE_DSE:
1738 skip_data_stream_element(&gb);
1739 err = 0;
1740 break;
1741
577d383b 1742 case TYPE_PCE: {
62a57fae
RS
1743 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1744 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
577d383b 1745 if ((err = decode_pce(ac, new_che_pos, &gb)))
62a57fae 1746 break;
6308765c
AC
1747 if (ac->output_configured)
1748 av_log(avccontext, AV_LOG_ERROR,
1749 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1750 else
36f51773 1751 err = output_configure(ac, ac->che_pos, new_che_pos, 0);
62a57fae
RS
1752 break;
1753 }
1754
1755 case TYPE_FIL:
1756 if (elem_id == 15)
1757 elem_id += get_bits(&gb, 8) - 1;
1758 while (elem_id > 0)
1759 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1760 err = 0; /* FIXME */
1761 break;
1762
1763 default:
1764 err = -1; /* should not happen, but keeps compiler happy */
1765 break;
1766 }
1767
577d383b 1768 if (err)
62a57fae
RS
1769 return err;
1770 }
1771
1772 spectral_to_sample(ac);
1773
9cc04edf
RS
1774 if (!ac->is_saved) {
1775 ac->is_saved = 1;
1776 *data_size = 0;
848a5815 1777 return buf_size;
9cc04edf
RS
1778 }
1779
1780 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
577d383b 1781 if (*data_size < data_size_tmp) {
9cc04edf
RS
1782 av_log(avccontext, AV_LOG_ERROR,
1783 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1784 *data_size, data_size_tmp);
1785 return -1;
1786 }
1787 *data_size = data_size_tmp;
1788
1789 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1790
1791 return buf_size;
1792}
1793
577d383b
DB
1794static av_cold int aac_decode_close(AVCodecContext *avccontext)
1795{
1796 AACContext *ac = avccontext->priv_data;
9edae4ad 1797 int i, type;
71e9a1b8 1798
cc0591da 1799 for (i = 0; i < MAX_ELEM_ID; i++) {
577d383b 1800 for (type = 0; type < 4; type++)
9edae4ad 1801 av_freep(&ac->che[type][i]);
71e9a1b8
RS
1802 }
1803
1804 ff_mdct_end(&ac->mdct);
1805 ff_mdct_end(&ac->mdct_small);
577d383b 1806 return 0;
71e9a1b8
RS
1807}
1808
1809AVCodec aac_decoder = {
1810 "aac",
1811 CODEC_TYPE_AUDIO,
1812 CODEC_ID_AAC,
1813 sizeof(AACContext),
1814 aac_decode_init,
1815 NULL,
1816 aac_decode_close,
1817 aac_decode_frame,
1818 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
b5f09d31 1819 .sample_fmts = (const enum SampleFormat[]) {
577d383b
DB
1820 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
1821 },
e22da6b6 1822 .channel_layouts = aac_channel_layout,
71e9a1b8 1823};