Synchronise presets with recent subme/brdo/bidir_refine -> subme changes in
[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/**
24 * @file aac.c
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
44 * N frequency domain prediction
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"
80#include "bitstream.h"
81#include "dsputil.h"
1be0fc29 82#include "lpc.h"
71e9a1b8
RS
83
84#include "aac.h"
85#include "aactab.h"
cc0591da 86#include "aacdectab.h"
71e9a1b8
RS
87#include "mpeg4audio.h"
88
89#include <assert.h>
90#include <errno.h>
91#include <math.h>
92#include <string.h>
93
71e9a1b8
RS
94static VLC vlc_scalefactors;
95static VLC vlc_spectral[11];
96
97
9cc04edf 98/**
62a57fae
RS
99 * Configure output channel order based on the current program configuration element.
100 *
101 * @param che_pos current channel position configuration
102 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
103 *
104 * @return Returns error status. 0 - OK, !0 - error
105 */
106static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID],
107 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]) {
108 AVCodecContext *avctx = ac->avccontext;
109 int i, type, channels = 0;
110
111 if(!memcmp(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])))
112 return 0; /* no change */
113
114 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
115
116 /* Allocate or free elements depending on if they are in the
117 * current program configuration.
118 *
119 * Set up default 1:1 output mapping.
120 *
121 * For a 5.1 stream the output order will be:
f3399088 122 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
62a57fae
RS
123 */
124
125 for(i = 0; i < MAX_ELEM_ID; i++) {
126 for(type = 0; type < 4; type++) {
127 if(che_pos[type][i]) {
128 if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
129 return AVERROR(ENOMEM);
130 if(type != TYPE_CCE) {
131 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
132 if(type == TYPE_CPE) {
133 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
134 }
135 }
136 } else
137 av_freep(&ac->che[type][i]);
138 }
139 }
140
141 avctx->channels = channels;
142 return 0;
143}
144
145/**
9cc04edf
RS
146 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
147 *
148 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
149 * @param sce_map mono (Single Channel Element) map
150 * @param type speaker type/position for these channels
151 */
152static void decode_channel_map(enum ChannelPosition *cpe_map,
153 enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) {
154 while(n--) {
155 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
156 map[get_bits(gb, 4)] = type;
157 }
158}
159
160/**
161 * Decode program configuration element; reference: table 4.2.
162 *
163 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
164 *
165 * @return Returns error status. 0 - OK, !0 - error
166 */
167static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
168 GetBitContext * gb) {
169 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
170
171 skip_bits(gb, 2); // object_type
172
173 ac->m4ac.sampling_index = get_bits(gb, 4);
174 if(ac->m4ac.sampling_index > 11) {
175 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
176 return -1;
177 }
178 ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index];
71e9a1b8
RS
179 num_front = get_bits(gb, 4);
180 num_side = get_bits(gb, 4);
181 num_back = get_bits(gb, 4);
182 num_lfe = get_bits(gb, 2);
183 num_assoc_data = get_bits(gb, 3);
184 num_cc = get_bits(gb, 4);
185
cc0591da
RS
186 if (get_bits1(gb))
187 skip_bits(gb, 4); // mono_mixdown_tag
188 if (get_bits1(gb))
189 skip_bits(gb, 4); // stereo_mixdown_tag
71e9a1b8 190
cc0591da
RS
191 if (get_bits1(gb))
192 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
71e9a1b8 193
cc0591da
RS
194 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
195 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
196 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
197 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
71e9a1b8
RS
198
199 skip_bits_long(gb, 4 * num_assoc_data);
200
cc0591da 201 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
71e9a1b8
RS
202
203 align_get_bits(gb);
204
205 /* comment field, first byte is length */
206 skip_bits_long(gb, 8 * get_bits(gb, 8));
cc0591da
RS
207 return 0;
208}
71e9a1b8 209
9cc04edf
RS
210/**
211 * Set up channel positions based on a default channel configuration
212 * as specified in table 1.17.
213 *
214 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
215 *
216 * @return Returns error status. 0 - OK, !0 - error
217 */
218static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
219 int channel_config)
220{
221 if(channel_config < 1 || channel_config > 7) {
222 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
223 channel_config);
224 return -1;
225 }
226
227 /* default channel configurations:
228 *
229 * 1ch : front center (mono)
230 * 2ch : L + R (stereo)
231 * 3ch : front center + L + R
232 * 4ch : front center + L + R + back center
233 * 5ch : front center + L + R + back stereo
234 * 6ch : front center + L + R + back stereo + LFE
235 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
236 */
237
238 if(channel_config != 2)
239 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
240 if(channel_config > 1)
241 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
242 if(channel_config == 4)
243 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
244 if(channel_config > 4)
245 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
246 = AAC_CHANNEL_BACK; // back stereo
247 if(channel_config > 5)
248 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
249 if(channel_config == 7)
250 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
251
252 return 0;
253}
254
62a57fae
RS
255/**
256 * Decode GA "General Audio" specific configuration; reference: table 4.1.
257 *
258 * @return Returns error status. 0 - OK, !0 - error
259 */
260static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) {
261 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
262 int extension_flag, ret;
263
264 if(get_bits1(gb)) { // frameLengthFlag
265 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
9cc04edf
RS
266 return -1;
267 }
268
269 if (get_bits1(gb)) // dependsOnCoreCoder
270 skip_bits(gb, 14); // coreCoderDelay
271 extension_flag = get_bits1(gb);
272
273 if(ac->m4ac.object_type == AOT_AAC_SCALABLE ||
274 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
275 skip_bits(gb, 3); // layerNr
276
277 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
278 if (channel_config == 0) {
279 skip_bits(gb, 4); // element_instance_tag
280 if((ret = decode_pce(ac, new_che_pos, gb)))
281 return ret;
282 } else {
283 if((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
284 return ret;
285 }
286 if((ret = output_configure(ac, ac->che_pos, new_che_pos)))
287 return ret;
288
289 if (extension_flag) {
290 switch (ac->m4ac.object_type) {
291 case AOT_ER_BSAC:
292 skip_bits(gb, 5); // numOfSubFrame
293 skip_bits(gb, 11); // layer_length
294 break;
295 case AOT_ER_AAC_LC:
296 case AOT_ER_AAC_LTP:
297 case AOT_ER_AAC_SCALABLE:
298 case AOT_ER_AAC_LD:
299 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
300 * aacScalefactorDataResilienceFlag
301 * aacSpectralDataResilienceFlag
302 */
303 break;
304 }
305 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
306 }
307 return 0;
308}
309
310/**
311 * Decode audio specific configuration; reference: table 1.13.
312 *
313 * @param data pointer to AVCodecContext extradata
314 * @param data_size size of AVCCodecContext extradata
315 *
316 * @return Returns error status. 0 - OK, !0 - error
317 */
318static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) {
319 GetBitContext gb;
320 int i;
321
322 init_get_bits(&gb, data, data_size * 8);
323
324 if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
325 return -1;
326 if(ac->m4ac.sampling_index > 11) {
327 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
328 return -1;
329 }
330
331 skip_bits_long(&gb, i);
332
333 switch (ac->m4ac.object_type) {
334 case AOT_AAC_LC:
335 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
336 return -1;
337 break;
338 default:
339 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
340 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
341 return -1;
342 }
343 return 0;
344}
345
62a57fae
RS
346/**
347 * linear congruential pseudorandom number generator
348 *
349 * @param previous_val pointer to the current state of the generator
350 *
351 * @return Returns a 32-bit pseudorandom integer
352 */
353static av_always_inline int lcg_random(int previous_val) {
354 return previous_val * 1664525 + 1013904223;
355}
356
71e9a1b8
RS
357static av_cold int aac_decode_init(AVCodecContext * avccontext) {
358 AACContext * ac = avccontext->priv_data;
359 int i;
360
361 ac->avccontext = avccontext;
362
cc0591da
RS
363 if (avccontext->extradata_size <= 0 ||
364 decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
365 return -1;
366
9cc04edf 367 avccontext->sample_fmt = SAMPLE_FMT_S16;
71e9a1b8
RS
368 avccontext->sample_rate = ac->m4ac.sample_rate;
369 avccontext->frame_size = 1024;
370
371 AAC_INIT_VLC_STATIC( 0, 144);
372 AAC_INIT_VLC_STATIC( 1, 114);
373 AAC_INIT_VLC_STATIC( 2, 188);
374 AAC_INIT_VLC_STATIC( 3, 180);
375 AAC_INIT_VLC_STATIC( 4, 172);
376 AAC_INIT_VLC_STATIC( 5, 140);
377 AAC_INIT_VLC_STATIC( 6, 168);
378 AAC_INIT_VLC_STATIC( 7, 114);
379 AAC_INIT_VLC_STATIC( 8, 262);
380 AAC_INIT_VLC_STATIC( 9, 248);
381 AAC_INIT_VLC_STATIC(10, 384);
382
383 dsputil_init(&ac->dsp, avccontext);
384
9cc04edf
RS
385 ac->random_state = 0x1f2e3d4c;
386
71e9a1b8
RS
387 // -1024 - Compensate wrong IMDCT method.
388 // 32768 - Required to scale values to the correct range for the bias method
389 // for float to int16 conversion.
390
391 if(ac->dsp.float_to_int16 == ff_float_to_int16_c) {
392 ac->add_bias = 385.0f;
393 ac->sf_scale = 1. / (-1024. * 32768.);
394 ac->sf_offset = 0;
395 } else {
396 ac->add_bias = 0.0f;
397 ac->sf_scale = 1. / -1024.;
398 ac->sf_offset = 60;
399 }
400
401#ifndef CONFIG_HARDCODED_TABLES
cc0591da
RS
402 for (i = 0; i < 316; i++)
403 ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
71e9a1b8
RS
404#endif /* CONFIG_HARDCODED_TABLES */
405
406 INIT_VLC_STATIC(&vlc_scalefactors, 7, sizeof(ff_aac_scalefactor_code)/sizeof(ff_aac_scalefactor_code[0]),
407 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
408 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
409 352);
410
411 ff_mdct_init(&ac->mdct, 11, 1);
412 ff_mdct_init(&ac->mdct_small, 8, 1);
9ffd5c1c
RS
413 // window initialization
414 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
415 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
416 ff_sine_window_init(ff_sine_1024, 1024);
417 ff_sine_window_init(ff_sine_128, 128);
418
71e9a1b8
RS
419 return 0;
420}
421
9cc04edf
RS
422/**
423 * Skip data_stream_element; reference: table 4.10.
424 */
425static void skip_data_stream_element(GetBitContext * gb) {
71e9a1b8
RS
426 int byte_align = get_bits1(gb);
427 int count = get_bits(gb, 8);
428 if (count == 255)
429 count += get_bits(gb, 8);
430 if (byte_align)
431 align_get_bits(gb);
432 skip_bits_long(gb, 8 * count);
433}
434
435/**
9cc04edf
RS
436 * Decode Individual Channel Stream info; reference: table 4.6.
437 *
438 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
439 */
440static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) {
441 if (get_bits1(gb)) {
442 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
443 memset(ics, 0, sizeof(IndividualChannelStream));
444 return -1;
445 }
446 ics->window_sequence[1] = ics->window_sequence[0];
447 ics->window_sequence[0] = get_bits(gb, 2);
448 ics->use_kb_window[1] = ics->use_kb_window[0];
449 ics->use_kb_window[0] = get_bits1(gb);
450 ics->num_window_groups = 1;
451 ics->group_len[0] = 1;
9ffd5c1c
RS
452 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
453 int i;
454 ics->max_sfb = get_bits(gb, 4);
455 for (i = 0; i < 7; i++) {
456 if (get_bits1(gb)) {
457 ics->group_len[ics->num_window_groups-1]++;
458 } else {
459 ics->num_window_groups++;
460 ics->group_len[ics->num_window_groups-1] = 1;
461 }
462 }
463 ics->num_windows = 8;
464 ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
465 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
466 ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
467 } else {
468 ics->max_sfb = get_bits(gb, 6);
469 ics->num_windows = 1;
470 ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
471 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
472 ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
62a57fae
RS
473 if (get_bits1(gb)) {
474 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
475 memset(ics, 0, sizeof(IndividualChannelStream));
476 return -1;
477 }
478 }
479
480 if(ics->max_sfb > ics->num_swb) {
481 av_log(ac->avccontext, AV_LOG_ERROR,
482 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
483 ics->max_sfb, ics->num_swb);
484 memset(ics, 0, sizeof(IndividualChannelStream));
485 return -1;
486 }
487
9cc04edf
RS
488 return 0;
489}
490
491/**
9cc04edf
RS
492 * Decode band types (section_data payload); reference: table 4.46.
493 *
494 * @param band_type array of the used band type
495 * @param band_type_run_end array of the last scalefactor band of a band type run
496 *
497 * @return Returns error status. 0 - OK, !0 - error
498 */
499static int decode_band_types(AACContext * ac, enum BandType band_type[120],
cc0591da
RS
500 int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) {
501 int g, idx = 0;
502 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
503 for (g = 0; g < ics->num_window_groups; g++) {
504 int k = 0;
505 while (k < ics->max_sfb) {
506 uint8_t sect_len = k;
507 int sect_len_incr;
508 int sect_band_type = get_bits(gb, 4);
509 if (sect_band_type == 12) {
510 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
511 return -1;
512 }
513 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1)
514 sect_len += sect_len_incr;
515 sect_len += sect_len_incr;
516 if (sect_len > ics->max_sfb) {
517 av_log(ac->avccontext, AV_LOG_ERROR,
518 "Number of bands (%d) exceeds limit (%d).\n",
519 sect_len, ics->max_sfb);
520 return -1;
521 }
9ffd5c1c
RS
522 for (; k < sect_len; k++) {
523 band_type [idx] = sect_band_type;
524 band_type_run_end[idx++] = sect_len;
525 }
9cc04edf
RS
526 }
527 }
528 return 0;
529}
cc0591da 530
9cc04edf
RS
531/**
532 * Decode scalefactors; reference: table 4.47.
cc0591da 533 *
cc0591da
RS
534 * @param global_gain first scalefactor value as scalefactors are differentially coded
535 * @param band_type array of the used band type
536 * @param band_type_run_end array of the last scalefactor band of a band type run
537 * @param sf array of scalefactors or intensity stereo positions
538 *
539 * @return Returns error status. 0 - OK, !0 - error
540 */
541static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb,
9edae4ad 542 unsigned int global_gain, IndividualChannelStream * ics,
cc0591da
RS
543 enum BandType band_type[120], int band_type_run_end[120]) {
544 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
545 int g, i, idx = 0;
546 int offset[3] = { global_gain, global_gain - 90, 100 };
547 int noise_flag = 1;
548 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
cc0591da
RS
549 for (g = 0; g < ics->num_window_groups; g++) {
550 for (i = 0; i < ics->max_sfb;) {
551 int run_end = band_type_run_end[idx];
552 if (band_type[idx] == ZERO_BT) {
553 for(; i < run_end; i++, idx++)
554 sf[idx] = 0.;
555 }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
cc0591da
RS
556 for(; i < run_end; i++, idx++) {
557 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
558 if(offset[2] > 255U) {
559 av_log(ac->avccontext, AV_LOG_ERROR,
560 "%s (%d) out of range.\n", sf_str[2], offset[2]);
561 return -1;
562 }
563 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
cc0591da
RS
564 }
565 }else if(band_type[idx] == NOISE_BT) {
566 for(; i < run_end; i++, idx++) {
567 if(noise_flag-- > 0)
568 offset[1] += get_bits(gb, 9) - 256;
569 else
570 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
571 if(offset[1] > 255U) {
572 av_log(ac->avccontext, AV_LOG_ERROR,
573 "%s (%d) out of range.\n", sf_str[1], offset[1]);
574 return -1;
575 }
576 sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset];
cc0591da
RS
577 }
578 }else {
579 for(; i < run_end; i++, idx++) {
580 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
581 if(offset[0] > 255U) {
582 av_log(ac->avccontext, AV_LOG_ERROR,
583 "%s (%d) out of range.\n", sf_str[0], offset[0]);
584 return -1;
585 }
586 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
cc0591da
RS
587 }
588 }
589 }
590 }
591 return 0;
592}
593
594/**
595 * Decode pulse data; reference: table 4.7.
596 */
aac0eda4
AC
597static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) {
598 int i, pulse_swb;
cc0591da 599 pulse->num_pulse = get_bits(gb, 2) + 1;
aac0eda4
AC
600 pulse_swb = get_bits(gb, 6);
601 if (pulse_swb >= num_swb)
602 return -1;
603 pulse->pos[0] = swb_offset[pulse_swb];
408992ba 604 pulse->pos[0] += get_bits(gb, 5);
aac0eda4
AC
605 if (pulse->pos[0] > 1023)
606 return -1;
848a5815
RS
607 pulse->amp[0] = get_bits(gb, 4);
608 for (i = 1; i < pulse->num_pulse; i++) {
609 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1];
aac0eda4
AC
610 if (pulse->pos[i] > 1023)
611 return -1;
848a5815 612 pulse->amp[i] = get_bits(gb, 4);
cc0591da 613 }
aac0eda4 614 return 0;
cc0591da
RS
615}
616
617/**
1dece0d2
RS
618 * Decode Temporal Noise Shaping data; reference: table 4.48.
619 *
620 * @return Returns error status. 0 - OK, !0 - error
621 */
622static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns,
623 GetBitContext * gb, const IndividualChannelStream * ics) {
624 int w, filt, i, coef_len, coef_res, coef_compress;
625 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
626 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
627 for (w = 0; w < ics->num_windows; w++) {
fbd91d7c 628 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1dece0d2
RS
629 coef_res = get_bits1(gb);
630
65b20b24
RS
631 for (filt = 0; filt < tns->n_filt[w]; filt++) {
632 int tmp2_idx;
633 tns->length[w][filt] = get_bits(gb, 6 - 2*is8);
634
635 if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) {
636 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
637 tns->order[w][filt], tns_max_order);
638 tns->order[w][filt] = 0;
639 return -1;
640 }
51673647 641 if (tns->order[w][filt]) {
35445d29
RS
642 tns->direction[w][filt] = get_bits1(gb);
643 coef_compress = get_bits1(gb);
644 coef_len = coef_res + 3 - coef_compress;
645 tmp2_idx = 2*coef_compress + coef_res;
1dece0d2 646
35445d29
RS
647 for (i = 0; i < tns->order[w][filt]; i++)
648 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
51673647 649 }
65b20b24 650 }
fbd91d7c 651 }
1dece0d2
RS
652 }
653 return 0;
654}
655
656/**
9cc04edf
RS
657 * Decode Mid/Side data; reference: table 4.54.
658 *
659 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
660 * [1] mask is decoded from bitstream; [2] mask is all 1s;
661 * [3] reserved for scalable AAC
662 */
663static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb,
664 int ms_present) {
62a57fae
RS
665 int idx;
666 if (ms_present == 1) {
667 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
668 cpe->ms_mask[idx] = get_bits1(gb);
669 } else if (ms_present == 2) {
670 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
671 }
672}
9cc04edf
RS
673
674/**
9ffd5c1c
RS
675 * Decode spectral data; reference: table 4.50.
676 * Dequantize and scale spectral data; reference: 4.6.3.3.
677 *
678 * @param coef array of dequantized, scaled spectral data
679 * @param sf array of scalefactors or intensity stereo positions
680 * @param pulse_present set if pulses are present
681 * @param pulse pointer to pulse data struct
682 * @param band_type array of the used band type
683 *
684 * @return Returns error status. 0 - OK, !0 - error
685 */
686static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120],
687 int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) {
688 int i, k, g, idx = 0;
689 const int c = 1024/ics->num_windows;
690 const uint16_t * offsets = ics->swb_offset;
691 float *coef_base = coef;
692
693 for (g = 0; g < ics->num_windows; g++)
694 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
695
696 for (g = 0; g < ics->num_window_groups; g++) {
697 for (i = 0; i < ics->max_sfb; i++, idx++) {
698 const int cur_band_type = band_type[idx];
699 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
700 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
701 int group;
702 if (cur_band_type == ZERO_BT) {
703 for (group = 0; group < ics->group_len[g]; group++) {
704 memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
705 }
706 }else if (cur_band_type == NOISE_BT) {
707 const float scale = sf[idx] / ((offsets[i+1] - offsets[i]) * PNS_MEAN_ENERGY);
708 for (group = 0; group < ics->group_len[g]; group++) {
709 for (k = offsets[i]; k < offsets[i+1]; k++) {
710 ac->random_state = lcg_random(ac->random_state);
711 coef[group*128+k] = ac->random_state * scale;
712 }
713 }
714 }else if (cur_band_type != INTENSITY_BT2 && cur_band_type != INTENSITY_BT) {
715 for (group = 0; group < ics->group_len[g]; group++) {
716 for (k = offsets[i]; k < offsets[i+1]; k += dim) {
717 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
718 const int coef_tmp_idx = (group << 7) + k;
719 const float *vq_ptr;
720 int j;
721 if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
722 av_log(ac->avccontext, AV_LOG_ERROR,
723 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
724 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
725 return -1;
726 }
727 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
728 if (is_cb_unsigned) {
729 for (j = 0; j < dim; j++)
730 if (vq_ptr[j])
731 coef[coef_tmp_idx + j] = 1 - 2*(int)get_bits1(gb);
732 }else {
733 for (j = 0; j < dim; j++)
734 coef[coef_tmp_idx + j] = 1.0f;
735 }
736 if (cur_band_type == ESC_BT) {
737 for (j = 0; j < 2; j++) {
738 if (vq_ptr[j] == 64.0f) {
739 int n = 4;
740 /* The total length of escape_sequence must be < 22 bits according
741 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
742 while (get_bits1(gb) && n < 15) n++;
743 if(n == 15) {
744 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
745 return -1;
746 }
747 n = (1<<n) + get_bits(gb, n);
748 coef[coef_tmp_idx + j] *= cbrtf(fabsf(n)) * n;
749 }else
750 coef[coef_tmp_idx + j] *= vq_ptr[j];
751 }
752 }else
753 for (j = 0; j < dim; j++)
754 coef[coef_tmp_idx + j] *= vq_ptr[j];
755 for (j = 0; j < dim; j++)
756 coef[coef_tmp_idx + j] *= sf[idx];
757 }
758 }
759 }
760 }
761 coef += ics->group_len[g]<<7;
762 }
763
764 if (pulse_present) {
51436848 765 idx = 0;
9ffd5c1c
RS
766 for(i = 0; i < pulse->num_pulse; i++){
767 float co = coef_base[ pulse->pos[i] ];
51436848
AC
768 while(offsets[idx + 1] <= pulse->pos[i])
769 idx++;
770 if (band_type[idx] != NOISE_BT && sf[idx]) {
70735a3f
RS
771 float ico = -pulse->amp[i];
772 if (co) {
773 co /= sf[idx];
774 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
775 }
776 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
51436848 777 }
9ffd5c1c
RS
778 }
779 }
780 return 0;
781}
782
783/**
9cc04edf
RS
784 * Decode an individual_channel_stream payload; reference: table 4.44.
785 *
786 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
787 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
788 *
789 * @return Returns error status. 0 - OK, !0 - error
790 */
791static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) {
9cc04edf
RS
792 Pulse pulse;
793 TemporalNoiseShaping * tns = &sce->tns;
794 IndividualChannelStream * ics = &sce->ics;
795 float * out = sce->coeffs;
796 int global_gain, pulse_present = 0;
797
848a5815
RS
798 /* This assignment is to silence a GCC warning about the variable being used
799 * uninitialized when in fact it always is.
9cc04edf
RS
800 */
801 pulse.num_pulse = 0;
9cc04edf
RS
802
803 global_gain = get_bits(gb, 8);
804
805 if (!common_window && !scale_flag) {
806 if (decode_ics_info(ac, ics, gb, 0) < 0)
807 return -1;
808 }
809
810 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
811 return -1;
812 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
813 return -1;
814
815 pulse_present = 0;
816 if (!scale_flag) {
817 if ((pulse_present = get_bits1(gb))) {
818 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
819 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
820 return -1;
821 }
aac0eda4
AC
822 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
823 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
824 return -1;
825 }
9cc04edf
RS
826 }
827 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
828 return -1;
829 if (get_bits1(gb)) {
830 av_log_missing_feature(ac->avccontext, "SSR", 1);
831 return -1;
832 }
833 }
834
848a5815 835 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
9cc04edf 836 return -1;
9cc04edf
RS
837 return 0;
838}
839
840/**
9ffd5c1c
RS
841 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
842 */
843static void apply_mid_side_stereo(ChannelElement * cpe) {
844 const IndividualChannelStream * ics = &cpe->ch[0].ics;
845 float *ch0 = cpe->ch[0].coeffs;
846 float *ch1 = cpe->ch[1].coeffs;
847 int g, i, k, group, idx = 0;
848 const uint16_t * offsets = ics->swb_offset;
849 for (g = 0; g < ics->num_window_groups; g++) {
850 for (i = 0; i < ics->max_sfb; i++, idx++) {
851 if (cpe->ms_mask[idx] &&
852 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
853 for (group = 0; group < ics->group_len[g]; group++) {
854 for (k = offsets[i]; k < offsets[i+1]; k++) {
855 float tmp = ch0[group*128 + k] - ch1[group*128 + k];
856 ch0[group*128 + k] += ch1[group*128 + k];
857 ch1[group*128 + k] = tmp;
858 }
859 }
860 }
861 }
862 ch0 += ics->group_len[g]*128;
863 ch1 += ics->group_len[g]*128;
864 }
865}
866
867/**
868 * intensity stereo decoding; reference: 4.6.8.2.3
869 *
870 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
871 * [1] mask is decoded from bitstream; [2] mask is all 1s;
872 * [3] reserved for scalable AAC
873 */
874static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) {
875 const IndividualChannelStream * ics = &cpe->ch[1].ics;
876 SingleChannelElement * sce1 = &cpe->ch[1];
877 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
878 const uint16_t * offsets = ics->swb_offset;
879 int g, group, i, k, idx = 0;
880 int c;
881 float scale;
882 for (g = 0; g < ics->num_window_groups; g++) {
883 for (i = 0; i < ics->max_sfb;) {
884 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
885 const int bt_run_end = sce1->band_type_run_end[idx];
886 for (; i < bt_run_end; i++, idx++) {
887 c = -1 + 2 * (sce1->band_type[idx] - 14);
888 if (ms_present)
889 c *= 1 - 2 * cpe->ms_mask[idx];
890 scale = c * sce1->sf[idx];
891 for (group = 0; group < ics->group_len[g]; group++)
892 for (k = offsets[i]; k < offsets[i+1]; k++)
893 coef1[group*128 + k] = scale * coef0[group*128 + k];
894 }
895 } else {
896 int bt_run_end = sce1->band_type_run_end[idx];
897 idx += bt_run_end - i;
898 i = bt_run_end;
899 }
900 }
901 coef0 += ics->group_len[g]*128;
902 coef1 += ics->group_len[g]*128;
903 }
904}
905
906/**
9cc04edf
RS
907 * Decode a channel_pair_element; reference: table 4.4.
908 *
909 * @param elem_id Identifies the instance of a syntax element.
910 *
911 * @return Returns error status. 0 - OK, !0 - error
912 */
913static int decode_cpe(AACContext * ac, GetBitContext * gb, int elem_id) {
914 int i, ret, common_window, ms_present = 0;
915 ChannelElement * cpe;
916
917 cpe = ac->che[TYPE_CPE][elem_id];
918 common_window = get_bits1(gb);
919 if (common_window) {
920 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
921 return -1;
922 i = cpe->ch[1].ics.use_kb_window[0];
923 cpe->ch[1].ics = cpe->ch[0].ics;
924 cpe->ch[1].ics.use_kb_window[1] = i;
925 ms_present = get_bits(gb, 2);
926 if(ms_present == 3) {
927 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
928 return -1;
929 } else if(ms_present)
930 decode_mid_side_stereo(cpe, gb, ms_present);
931 }
932 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
933 return ret;
934 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
935 return ret;
936
937 if (common_window && ms_present)
938 apply_mid_side_stereo(cpe);
939
848a5815 940 apply_intensity_stereo(cpe, ms_present);
9cc04edf
RS
941 return 0;
942}
943
9ffd5c1c
RS
944/**
945 * Decode coupling_channel_element; reference: table 4.8.
946 *
947 * @param elem_id Identifies the instance of a syntax element.
948 *
949 * @return Returns error status. 0 - OK, !0 - error
950 */
951static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
952 int num_gain = 0;
341b28c0 953 int c, g, sfb, ret;
9ffd5c1c
RS
954 int sign;
955 float scale;
956 SingleChannelElement * sce = &che->ch[0];
957 ChannelCoupling * coup = &che->coup;
958
62a57fae
RS
959 coup->coupling_point = 2*get_bits1(gb);
960 coup->num_coupled = get_bits(gb, 3);
961 for (c = 0; c <= coup->num_coupled; c++) {
962 num_gain++;
963 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
964 coup->id_select[c] = get_bits(gb, 4);
965 if (coup->type[c] == TYPE_CPE) {
966 coup->ch_select[c] = get_bits(gb, 2);
967 if (coup->ch_select[c] == 3)
968 num_gain++;
969 } else
970 coup->ch_select[c] = 1;
971 }
972 coup->coupling_point += get_bits1(gb);
973
974 if (coup->coupling_point == 2) {
975 av_log(ac->avccontext, AV_LOG_ERROR,
976 "Independently switched CCE with 'invalid' domain signalled.\n");
977 memset(coup, 0, sizeof(ChannelCoupling));
978 return -1;
979 }
980
981 sign = get_bits(gb, 1);
c8947a56 982 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
62a57fae
RS
983
984 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
985 return ret;
986
987 for (c = 0; c < num_gain; c++) {
341b28c0 988 int idx = 0;
62a57fae
RS
989 int cge = 1;
990 int gain = 0;
991 float gain_cache = 1.;
992 if (c) {
993 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
994 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
995 gain_cache = pow(scale, gain);
996 }
f80a8ca5
RS
997 for (g = 0; g < sce->ics.num_window_groups; g++) {
998 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
62a57fae
RS
999 if (sce->band_type[idx] != ZERO_BT) {
1000 if (!cge) {
1001 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1002 if (t) {
1003 int s = 1;
1004 if (sign) {
1005 s -= 2 * (t & 0x1);
1006 t >>= 1;
1007 }
1008 gain += t;
1009 gain_cache = pow(scale, gain) * s;
1010 }
1011 }
1012 coup->gain[c][idx] = gain_cache;
1013 }
f80a8ca5
RS
1014 }
1015 }
62a57fae
RS
1016 }
1017 return 0;
1018}
1019
9cc04edf
RS
1020/**
1021 * Decode Spectral Band Replication extension data; reference: table 4.55.
cc0591da
RS
1022 *
1023 * @param crc flag indicating the presence of CRC checksum
1024 * @param cnt length of TYPE_FIL syntactic element in bytes
9cc04edf 1025 *
cc0591da
RS
1026 * @return Returns number of bytes consumed from the TYPE_FIL element.
1027 */
1028static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
1029 // TODO : sbr_extension implementation
9edae4ad 1030 av_log_missing_feature(ac->avccontext, "SBR", 0);
cc0591da
RS
1031 skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
1032 return cnt;
1033}
1034
9cc04edf 1035/**
62a57fae
RS
1036 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1037 *
1038 * @return Returns number of bytes consumed.
1039 */
1040static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
1041 int i;
1042 int num_excl_chan = 0;
1043
1044 do {
1045 for (i = 0; i < 7; i++)
1046 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1047 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1048
1049 return num_excl_chan / 7;
1050}
1051
1052/**
9cc04edf
RS
1053 * Decode dynamic range information; reference: table 4.52.
1054 *
1055 * @param cnt length of TYPE_FIL syntactic element in bytes
1056 *
1057 * @return Returns number of bytes consumed.
1058 */
1059static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
1060 int n = 1;
1061 int drc_num_bands = 1;
1062 int i;
1063
1064 /* pce_tag_present? */
1065 if(get_bits1(gb)) {
1066 che_drc->pce_instance_tag = get_bits(gb, 4);
1067 skip_bits(gb, 4); // tag_reserved_bits
1068 n++;
1069 }
1070
1071 /* excluded_chns_present? */
1072 if(get_bits1(gb)) {
1073 n += decode_drc_channel_exclusions(che_drc, gb);
1074 }
1075
1076 /* drc_bands_present? */
1077 if (get_bits1(gb)) {
1078 che_drc->band_incr = get_bits(gb, 4);
1079 che_drc->interpolation_scheme = get_bits(gb, 4);
1080 n++;
1081 drc_num_bands += che_drc->band_incr;
1082 for (i = 0; i < drc_num_bands; i++) {
1083 che_drc->band_top[i] = get_bits(gb, 8);
1084 n++;
1085 }
1086 }
1087
1088 /* prog_ref_level_present? */
1089 if (get_bits1(gb)) {
1090 che_drc->prog_ref_level = get_bits(gb, 7);
1091 skip_bits1(gb); // prog_ref_level_reserved_bits
1092 n++;
1093 }
1094
1095 for (i = 0; i < drc_num_bands; i++) {
1096 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1097 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1098 n++;
1099 }
1100
1101 return n;
1102}
1103
1104/**
1105 * Decode extension data (incomplete); reference: table 4.51.
1106 *
1107 * @param cnt length of TYPE_FIL syntactic element in bytes
1108 *
1109 * @return Returns number of bytes consumed
1110 */
1111static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
cc0591da
RS
1112 int crc_flag = 0;
1113 int res = cnt;
1114 switch (get_bits(gb, 4)) { // extension type
1115 case EXT_SBR_DATA_CRC:
1116 crc_flag++;
1117 case EXT_SBR_DATA:
1118 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1119 break;
1120 case EXT_DYNAMIC_RANGE:
1121 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1122 break;
1123 case EXT_FILL:
1124 case EXT_FILL_DATA:
1125 case EXT_DATA_ELEMENT:
1126 default:
1127 skip_bits_long(gb, 8*cnt - 4);
1128 break;
1129 };
1130 return res;
1131}
1132
7d8f3de4
RS
1133/**
1134 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1135 *
1136 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1137 * @param coef spectral coefficients
1138 */
1139static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
1140 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1098e8d2 1141 int w, filt, m, i;
7d8f3de4
RS
1142 int bottom, top, order, start, end, size, inc;
1143 float lpc[TNS_MAX_ORDER];
1144
1145 for (w = 0; w < ics->num_windows; w++) {
1146 bottom = ics->num_swb;
1147 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1148 top = bottom;
1149 bottom = FFMAX(0, top - tns->length[w][filt]);
1150 order = tns->order[w][filt];
1151 if (order == 0)
1152 continue;
1153
1be0fc29
VS
1154 // tns_decode_coef
1155 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
7d8f3de4 1156
1dece0d2
RS
1157 start = ics->swb_offset[FFMIN(bottom, mmm)];
1158 end = ics->swb_offset[FFMIN( top, mmm)];
1159 if ((size = end - start) <= 0)
1160 continue;
1161 if (tns->direction[w][filt]) {
1162 inc = -1; start = end - 1;
1163 } else {
1164 inc = 1;
1165 }
1166 start += w * 128;
1167
1168 // ar filter
1169 for (m = 0; m < size; m++, start += inc)
1170 for (i = 1; i <= FFMIN(m, order); i++)
7d8f3de4 1171 coef[start] -= coef[start - i*inc] * lpc[i-1];
1dece0d2
RS
1172 }
1173 }
1174}
1175
cc0591da 1176/**
9cc04edf
RS
1177 * Conduct IMDCT and windowing.
1178 */
1179static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
1180 IndividualChannelStream * ics = &sce->ics;
1181 float * in = sce->coeffs;
1182 float * out = sce->ret;
1183 float * saved = sce->saved;
848a5815
RS
1184 const float * swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1185 const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1186 const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
9cc04edf 1187 float * buf = ac->buf_mdct;
b0f5852a 1188 DECLARE_ALIGNED(16, float, temp[128]);
9cc04edf
RS
1189 int i;
1190
f4990558 1191 // imdct
62a57fae
RS
1192 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1193 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1194 av_log(ac->avccontext, AV_LOG_WARNING,
1195 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1196 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
b0f5852a
RS
1197 for (i = 0; i < 1024; i += 128)
1198 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
f4990558 1199 } else
b0f5852a 1200 ff_imdct_half(&ac->mdct, buf, in);
f4990558
RS
1201
1202 /* window overlapping
1203 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1204 * and long to short transitions are considered to be short to short
1205 * transitions. This leaves just two cases (long to long and short to short)
1206 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1207 */
1208 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1209 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
b0f5852a 1210 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
f4990558 1211 } else {
db38c386
RS
1212 for (i = 0; i < 448; i++)
1213 out[i] = saved[i] + ac->add_bias;
62a57fae 1214
f4990558 1215 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1216 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1217 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1218 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1219 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1220 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1221 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
f4990558 1222 } else {
b0f5852a 1223 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
db38c386 1224 for (i = 576; i < 1024; i++)
b0f5852a 1225 out[i] = buf[i-512] + ac->add_bias;
f4990558
RS
1226 }
1227 }
62a57fae 1228
f4990558
RS
1229 // buffer update
1230 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
b0f5852a
RS
1231 for (i = 0; i < 64; i++)
1232 saved[i] = temp[64 + i] - ac->add_bias;
1233 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1234 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1235 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1236 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1237 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
b0f5852a
RS
1238 memcpy( saved, buf + 512, 448 * sizeof(float));
1239 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
f4990558 1240 } else { // LONG_STOP or ONLY_LONG
b0f5852a 1241 memcpy( saved, buf + 512, 512 * sizeof(float));
62a57fae
RS
1242 }
1243}
1244
9cc04edf 1245/**
cc0591da
RS
1246 * Apply dependent channel coupling (applied before IMDCT).
1247 *
1248 * @param index index into coupling gain array
1249 */
1250static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
1251 IndividualChannelStream * ics = &cc->ch[0].ics;
1252 const uint16_t * offsets = ics->swb_offset;
1253 float * dest = sce->coeffs;
1254 const float * src = cc->ch[0].coeffs;
1255 int g, i, group, k, idx = 0;
1256 if(ac->m4ac.object_type == AOT_AAC_LTP) {
1257 av_log(ac->avccontext, AV_LOG_ERROR,
1258 "Dependent coupling is not supported together with LTP\n");
1259 return;
1260 }
1261 for (g = 0; g < ics->num_window_groups; g++) {
1262 for (i = 0; i < ics->max_sfb; i++, idx++) {
1263 if (cc->ch[0].band_type[idx] != ZERO_BT) {
cc0591da
RS
1264 for (group = 0; group < ics->group_len[g]; group++) {
1265 for (k = offsets[i]; k < offsets[i+1]; k++) {
1266 // XXX dsputil-ize
9edae4ad 1267 dest[group*128+k] += cc->coup.gain[index][idx] * src[group*128+k];
cc0591da
RS
1268 }
1269 }
1270 }
1271 }
1272 dest += ics->group_len[g]*128;
1273 src += ics->group_len[g]*128;
1274 }
1275}
1276
1277/**
1278 * Apply independent channel coupling (applied after IMDCT).
1279 *
1280 * @param index index into coupling gain array
1281 */
1282static void apply_independent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
1283 int i;
cc0591da 1284 for (i = 0; i < 1024; i++)
9edae4ad 1285 sce->ret[i] += cc->coup.gain[index][0] * (cc->ch[0].ret[i] - ac->add_bias);
cc0591da
RS
1286}
1287
9ffd5c1c
RS
1288/**
1289 * channel coupling transformation interface
1290 *
1291 * @param index index into coupling gain array
1292 * @param apply_coupling_method pointer to (in)dependent coupling function
1293 */
1294static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
1295 void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index))
1296{
1297 int c;
1298 int index = 0;
1299 ChannelCoupling * coup = &cc->coup;
1300 for (c = 0; c <= coup->num_coupled; c++) {
1301 if (ac->che[coup->type[c]][coup->id_select[c]]) {
1302 if (coup->ch_select[c] != 2) {
1303 apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[0], cc, index);
1304 if (coup->ch_select[c] != 0)
1305 index++;
1306 }
1307 if (coup->ch_select[c] != 1)
1308 apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[1], cc, index++);
1309 } else {
1310 av_log(ac->avccontext, AV_LOG_ERROR,
1311 "coupling target %sE[%d] not available\n",
1312 coup->type[c] == TYPE_CPE ? "CP" : "SC", coup->id_select[c]);
1313 break;
1314 }
1315 }
1316}
1317
1318/**
1319 * Convert spectral data to float samples, applying all supported tools as appropriate.
1320 */
1321static void spectral_to_sample(AACContext * ac) {
1322 int i, type;
1323 for (i = 0; i < MAX_ELEM_ID; i++) {
1324 for(type = 0; type < 4; type++) {
1325 ChannelElement *che = ac->che[type][i];
1326 if(che) {
1327 if(che->coup.coupling_point == BEFORE_TNS)
1328 apply_channel_coupling(ac, che, apply_dependent_coupling);
1329 if(che->ch[0].tns.present)
1330 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1331 if(che->ch[1].tns.present)
1332 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1333 if(che->coup.coupling_point == BETWEEN_TNS_AND_IMDCT)
1334 apply_channel_coupling(ac, che, apply_dependent_coupling);
1335 imdct_and_windowing(ac, &che->ch[0]);
1336 if(type == TYPE_CPE)
1337 imdct_and_windowing(ac, &che->ch[1]);
1338 if(che->coup.coupling_point == AFTER_IMDCT)
1339 apply_channel_coupling(ac, che, apply_independent_coupling);
62a57fae
RS
1340 }
1341 }
1342 }
1343}
1344
1345static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, const uint8_t * buf, int buf_size) {
1346 AACContext * ac = avccontext->priv_data;
1347 GetBitContext gb;
1348 enum RawDataBlockType elem_type;
1349 int err, elem_id, data_size_tmp;
1350
1351 init_get_bits(&gb, buf, buf_size*8);
1352
1353 // parse
1354 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1355 elem_id = get_bits(&gb, 4);
1356 err = -1;
1357
1358 if(elem_type == TYPE_SCE && elem_id == 1 &&
1359 !ac->che[TYPE_SCE][elem_id] && ac->che[TYPE_LFE][0]) {
1360 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
1361 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
1362 encountered such a stream, transfer the LFE[0] element to SCE[1] */
1363 ac->che[TYPE_SCE][elem_id] = ac->che[TYPE_LFE][0];
1364 ac->che[TYPE_LFE][0] = NULL;
1365 }
f4fae46e 1366 if(elem_type < TYPE_DSE) {
62a57fae
RS
1367 if(!ac->che[elem_type][elem_id])
1368 return -1;
1369 if(elem_type != TYPE_CCE)
1370 ac->che[elem_type][elem_id]->coup.coupling_point = 4;
1371 }
1372
1373 switch (elem_type) {
1374
1375 case TYPE_SCE:
1376 err = decode_ics(ac, &ac->che[TYPE_SCE][elem_id]->ch[0], &gb, 0, 0);
1377 break;
1378
1379 case TYPE_CPE:
1380 err = decode_cpe(ac, &gb, elem_id);
1381 break;
1382
1383 case TYPE_CCE:
93b63b1a 1384 err = decode_cce(ac, &gb, ac->che[TYPE_CCE][elem_id]);
62a57fae
RS
1385 break;
1386
1387 case TYPE_LFE:
1388 err = decode_ics(ac, &ac->che[TYPE_LFE][elem_id]->ch[0], &gb, 0, 0);
1389 break;
1390
1391 case TYPE_DSE:
1392 skip_data_stream_element(&gb);
1393 err = 0;
1394 break;
1395
1396 case TYPE_PCE:
1397 {
1398 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1399 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1400 if((err = decode_pce(ac, new_che_pos, &gb)))
1401 break;
1402 err = output_configure(ac, ac->che_pos, new_che_pos);
1403 break;
1404 }
1405
1406 case TYPE_FIL:
1407 if (elem_id == 15)
1408 elem_id += get_bits(&gb, 8) - 1;
1409 while (elem_id > 0)
1410 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1411 err = 0; /* FIXME */
1412 break;
1413
1414 default:
1415 err = -1; /* should not happen, but keeps compiler happy */
1416 break;
1417 }
1418
1419 if(err)
1420 return err;
1421 }
1422
1423 spectral_to_sample(ac);
1424
9cc04edf
RS
1425 if (!ac->is_saved) {
1426 ac->is_saved = 1;
1427 *data_size = 0;
848a5815 1428 return buf_size;
9cc04edf
RS
1429 }
1430
1431 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1432 if(*data_size < data_size_tmp) {
1433 av_log(avccontext, AV_LOG_ERROR,
1434 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1435 *data_size, data_size_tmp);
1436 return -1;
1437 }
1438 *data_size = data_size_tmp;
1439
1440 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1441
1442 return buf_size;
1443}
1444
71e9a1b8
RS
1445static av_cold int aac_decode_close(AVCodecContext * avccontext) {
1446 AACContext * ac = avccontext->priv_data;
9edae4ad 1447 int i, type;
71e9a1b8 1448
cc0591da 1449 for (i = 0; i < MAX_ELEM_ID; i++) {
9edae4ad
RS
1450 for(type = 0; type < 4; type++)
1451 av_freep(&ac->che[type][i]);
71e9a1b8
RS
1452 }
1453
1454 ff_mdct_end(&ac->mdct);
1455 ff_mdct_end(&ac->mdct_small);
71e9a1b8
RS
1456 return 0 ;
1457}
1458
1459AVCodec aac_decoder = {
1460 "aac",
1461 CODEC_TYPE_AUDIO,
1462 CODEC_ID_AAC,
1463 sizeof(AACContext),
1464 aac_decode_init,
1465 NULL,
1466 aac_decode_close,
1467 aac_decode_frame,
1468 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
cc0591da 1469 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
71e9a1b8 1470};