3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
6 * This file is part of FFmpeg.
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.
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.
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
24 * @file libavcodec/aac.c
26 * @author Oded Shimon ( ods15 ods15 dyndns org )
27 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
34 * N (code in SoC repo) gain control
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
44 * Y frequency domain prediction
45 * Y Perceptual Noise Substitution
47 * N Scalable Inverse AAC Quantization
48 * N Frequency Selective Switch
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
57 * N Silence Compression
60 * N Structured Audio tools
61 * N Structured Audio Sample Bank Format
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
73 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
74 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
81 #include "bitstream.h"
87 #include "aacdectab.h"
88 #include "mpeg4audio.h"
89 #include "aac_parser.h"
96 static VLC vlc_scalefactors;
97 static VLC vlc_spectral[11];
101 * Configure output channel order based on the current program configuration element.
103 * @param che_pos current channel position configuration
104 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
106 * @return Returns error status. 0 - OK, !0 - error
108 static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID],
109 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]) {
110 AVCodecContext *avctx = ac->avccontext;
111 int i, type, channels = 0;
113 if(!memcmp(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])))
114 return 0; /* no change */
116 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
118 /* Allocate or free elements depending on if they are in the
119 * current program configuration.
121 * Set up default 1:1 output mapping.
123 * For a 5.1 stream the output order will be:
124 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
127 for(i = 0; i < MAX_ELEM_ID; i++) {
128 for(type = 0; type < 4; type++) {
129 if(che_pos[type][i]) {
130 if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
131 return AVERROR(ENOMEM);
132 if(type != TYPE_CCE) {
133 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
134 if(type == TYPE_CPE) {
135 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
139 av_freep(&ac->che[type][i]);
143 avctx->channels = channels;
148 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
150 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
151 * @param sce_map mono (Single Channel Element) map
152 * @param type speaker type/position for these channels
154 static void decode_channel_map(enum ChannelPosition *cpe_map,
155 enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) {
157 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
158 map[get_bits(gb, 4)] = type;
163 * Decode program configuration element; reference: table 4.2.
165 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
167 * @return Returns error status. 0 - OK, !0 - error
169 static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
170 GetBitContext * gb) {
171 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
173 skip_bits(gb, 2); // object_type
175 sampling_index = get_bits(gb, 4);
176 if(sampling_index > 12) {
177 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
180 ac->m4ac.sampling_index = sampling_index;
181 ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index];
182 num_front = get_bits(gb, 4);
183 num_side = get_bits(gb, 4);
184 num_back = get_bits(gb, 4);
185 num_lfe = get_bits(gb, 2);
186 num_assoc_data = get_bits(gb, 3);
187 num_cc = get_bits(gb, 4);
190 skip_bits(gb, 4); // mono_mixdown_tag
192 skip_bits(gb, 4); // stereo_mixdown_tag
195 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
197 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
198 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
199 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
200 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
202 skip_bits_long(gb, 4 * num_assoc_data);
204 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
208 /* comment field, first byte is length */
209 skip_bits_long(gb, 8 * get_bits(gb, 8));
214 * Set up channel positions based on a default channel configuration
215 * as specified in table 1.17.
217 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
219 * @return Returns error status. 0 - OK, !0 - error
221 static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
224 if(channel_config < 1 || channel_config > 7) {
225 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
230 /* default channel configurations:
232 * 1ch : front center (mono)
233 * 2ch : L + R (stereo)
234 * 3ch : front center + L + R
235 * 4ch : front center + L + R + back center
236 * 5ch : front center + L + R + back stereo
237 * 6ch : front center + L + R + back stereo + LFE
238 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
241 if(channel_config != 2)
242 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
243 if(channel_config > 1)
244 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
245 if(channel_config == 4)
246 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
247 if(channel_config > 4)
248 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
249 = AAC_CHANNEL_BACK; // back stereo
250 if(channel_config > 5)
251 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
252 if(channel_config == 7)
253 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
259 * Decode GA "General Audio" specific configuration; reference: table 4.1.
261 * @return Returns error status. 0 - OK, !0 - error
263 static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) {
264 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
265 int extension_flag, ret;
267 if(get_bits1(gb)) { // frameLengthFlag
268 ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
272 if (get_bits1(gb)) // dependsOnCoreCoder
273 skip_bits(gb, 14); // coreCoderDelay
274 extension_flag = get_bits1(gb);
276 if(ac->m4ac.object_type == AOT_AAC_SCALABLE ||
277 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
278 skip_bits(gb, 3); // layerNr
280 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
281 if (channel_config == 0) {
282 skip_bits(gb, 4); // element_instance_tag
283 if((ret = decode_pce(ac, new_che_pos, gb)))
286 if((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
289 if((ret = output_configure(ac, ac->che_pos, new_che_pos)))
292 if (extension_flag) {
293 switch (ac->m4ac.object_type) {
295 skip_bits(gb, 5); // numOfSubFrame
296 skip_bits(gb, 11); // layer_length
300 case AOT_ER_AAC_SCALABLE:
302 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
303 * aacScalefactorDataResilienceFlag
304 * aacSpectralDataResilienceFlag
308 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
314 * Decode audio specific configuration; reference: table 1.13.
316 * @param data pointer to AVCodecContext extradata
317 * @param data_size size of AVCCodecContext extradata
319 * @return Returns error status. 0 - OK, !0 - error
321 static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) {
325 init_get_bits(&gb, data, data_size * 8);
327 if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
329 if(ac->m4ac.sampling_index > 12) {
330 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
334 skip_bits_long(&gb, i);
336 switch (ac->m4ac.object_type) {
339 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
343 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
344 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
351 * linear congruential pseudorandom number generator
353 * @param previous_val pointer to the current state of the generator
355 * @return Returns a 32-bit pseudorandom integer
357 static av_always_inline int lcg_random(int previous_val) {
358 return previous_val * 1664525 + 1013904223;
361 static void reset_predict_state(PredictorState * ps) {
370 static void reset_all_predictors(PredictorState * ps) {
372 for (i = 0; i < MAX_PREDICTORS; i++)
373 reset_predict_state(&ps[i]);
376 static void reset_predictor_group(PredictorState * ps, int group_num) {
378 for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
379 reset_predict_state(&ps[i]);
382 static av_cold int aac_decode_init(AVCodecContext * avccontext) {
383 AACContext * ac = avccontext->priv_data;
386 ac->avccontext = avccontext;
388 if (avccontext->extradata_size > 0) {
389 if(decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
391 avccontext->sample_rate = ac->m4ac.sample_rate;
392 } else if (avccontext->channels > 0) {
393 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
394 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
395 if(set_default_channel_config(ac, new_che_pos, avccontext->channels - (avccontext->channels == 8)))
397 if(output_configure(ac, ac->che_pos, new_che_pos))
399 ac->m4ac.sample_rate = avccontext->sample_rate;
401 ff_log_missing_feature(ac->avccontext, "Implicit channel configuration is", 0);
405 avccontext->sample_fmt = SAMPLE_FMT_S16;
406 avccontext->frame_size = 1024;
408 AAC_INIT_VLC_STATIC( 0, 144);
409 AAC_INIT_VLC_STATIC( 1, 114);
410 AAC_INIT_VLC_STATIC( 2, 188);
411 AAC_INIT_VLC_STATIC( 3, 180);
412 AAC_INIT_VLC_STATIC( 4, 172);
413 AAC_INIT_VLC_STATIC( 5, 140);
414 AAC_INIT_VLC_STATIC( 6, 168);
415 AAC_INIT_VLC_STATIC( 7, 114);
416 AAC_INIT_VLC_STATIC( 8, 262);
417 AAC_INIT_VLC_STATIC( 9, 248);
418 AAC_INIT_VLC_STATIC(10, 384);
420 dsputil_init(&ac->dsp, avccontext);
422 ac->random_state = 0x1f2e3d4c;
424 // -1024 - Compensate wrong IMDCT method.
425 // 32768 - Required to scale values to the correct range for the bias method
426 // for float to int16 conversion.
428 if(ac->dsp.float_to_int16 == ff_float_to_int16_c) {
429 ac->add_bias = 385.0f;
430 ac->sf_scale = 1. / (-1024. * 32768.);
434 ac->sf_scale = 1. / -1024.;
438 #if !CONFIG_HARDCODED_TABLES
439 for (i = 0; i < 428; i++)
440 ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
441 #endif /* CONFIG_HARDCODED_TABLES */
443 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
444 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
445 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
448 ff_mdct_init(&ac->mdct, 11, 1);
449 ff_mdct_init(&ac->mdct_small, 8, 1);
450 // window initialization
451 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
452 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
453 ff_sine_window_init(ff_sine_1024, 1024);
454 ff_sine_window_init(ff_sine_128, 128);
460 * Skip data_stream_element; reference: table 4.10.
462 static void skip_data_stream_element(GetBitContext * gb) {
463 int byte_align = get_bits1(gb);
464 int count = get_bits(gb, 8);
466 count += get_bits(gb, 8);
469 skip_bits_long(gb, 8 * count);
472 static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
475 ics->predictor_reset_group = get_bits(gb, 5);
476 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
477 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
481 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
482 ics->prediction_used[sfb] = get_bits1(gb);
488 * Decode Individual Channel Stream info; reference: table 4.6.
490 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
492 static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) {
494 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
495 memset(ics, 0, sizeof(IndividualChannelStream));
498 ics->window_sequence[1] = ics->window_sequence[0];
499 ics->window_sequence[0] = get_bits(gb, 2);
500 ics->use_kb_window[1] = ics->use_kb_window[0];
501 ics->use_kb_window[0] = get_bits1(gb);
502 ics->num_window_groups = 1;
503 ics->group_len[0] = 1;
504 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
506 ics->max_sfb = get_bits(gb, 4);
507 for (i = 0; i < 7; i++) {
509 ics->group_len[ics->num_window_groups-1]++;
511 ics->num_window_groups++;
512 ics->group_len[ics->num_window_groups-1] = 1;
515 ics->num_windows = 8;
516 ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
517 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
518 ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
519 ics->predictor_present = 0;
521 ics->max_sfb = get_bits(gb, 6);
522 ics->num_windows = 1;
523 ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
524 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
525 ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
526 ics->predictor_present = get_bits1(gb);
527 ics->predictor_reset_group = 0;
528 if (ics->predictor_present) {
529 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
530 if (decode_prediction(ac, ics, gb)) {
531 memset(ics, 0, sizeof(IndividualChannelStream));
534 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
535 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
536 memset(ics, 0, sizeof(IndividualChannelStream));
539 ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
540 memset(ics, 0, sizeof(IndividualChannelStream));
546 if(ics->max_sfb > ics->num_swb) {
547 av_log(ac->avccontext, AV_LOG_ERROR,
548 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
549 ics->max_sfb, ics->num_swb);
550 memset(ics, 0, sizeof(IndividualChannelStream));
558 * Decode band types (section_data payload); reference: table 4.46.
560 * @param band_type array of the used band type
561 * @param band_type_run_end array of the last scalefactor band of a band type run
563 * @return Returns error status. 0 - OK, !0 - error
565 static int decode_band_types(AACContext * ac, enum BandType band_type[120],
566 int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) {
568 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
569 for (g = 0; g < ics->num_window_groups; g++) {
571 while (k < ics->max_sfb) {
572 uint8_t sect_len = k;
574 int sect_band_type = get_bits(gb, 4);
575 if (sect_band_type == 12) {
576 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
579 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1)
580 sect_len += sect_len_incr;
581 sect_len += sect_len_incr;
582 if (sect_len > ics->max_sfb) {
583 av_log(ac->avccontext, AV_LOG_ERROR,
584 "Number of bands (%d) exceeds limit (%d).\n",
585 sect_len, ics->max_sfb);
588 for (; k < sect_len; k++) {
589 band_type [idx] = sect_band_type;
590 band_type_run_end[idx++] = sect_len;
598 * Decode scalefactors; reference: table 4.47.
600 * @param global_gain first scalefactor value as scalefactors are differentially coded
601 * @param band_type array of the used band type
602 * @param band_type_run_end array of the last scalefactor band of a band type run
603 * @param sf array of scalefactors or intensity stereo positions
605 * @return Returns error status. 0 - OK, !0 - error
607 static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb,
608 unsigned int global_gain, IndividualChannelStream * ics,
609 enum BandType band_type[120], int band_type_run_end[120]) {
610 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
612 int offset[3] = { global_gain, global_gain - 90, 100 };
614 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
615 for (g = 0; g < ics->num_window_groups; g++) {
616 for (i = 0; i < ics->max_sfb;) {
617 int run_end = band_type_run_end[idx];
618 if (band_type[idx] == ZERO_BT) {
619 for(; i < run_end; i++, idx++)
621 }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
622 for(; i < run_end; i++, idx++) {
623 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
624 if(offset[2] > 255U) {
625 av_log(ac->avccontext, AV_LOG_ERROR,
626 "%s (%d) out of range.\n", sf_str[2], offset[2]);
629 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
631 }else if(band_type[idx] == NOISE_BT) {
632 for(; i < run_end; i++, idx++) {
634 offset[1] += get_bits(gb, 9) - 256;
636 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
637 if(offset[1] > 255U) {
638 av_log(ac->avccontext, AV_LOG_ERROR,
639 "%s (%d) out of range.\n", sf_str[1], offset[1]);
642 sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100];
645 for(; i < run_end; i++, idx++) {
646 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
647 if(offset[0] > 255U) {
648 av_log(ac->avccontext, AV_LOG_ERROR,
649 "%s (%d) out of range.\n", sf_str[0], offset[0]);
652 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
661 * Decode pulse data; reference: table 4.7.
663 static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) {
665 pulse->num_pulse = get_bits(gb, 2) + 1;
666 pulse_swb = get_bits(gb, 6);
667 if (pulse_swb >= num_swb)
669 pulse->pos[0] = swb_offset[pulse_swb];
670 pulse->pos[0] += get_bits(gb, 5);
671 if (pulse->pos[0] > 1023)
673 pulse->amp[0] = get_bits(gb, 4);
674 for (i = 1; i < pulse->num_pulse; i++) {
675 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1];
676 if (pulse->pos[i] > 1023)
678 pulse->amp[i] = get_bits(gb, 4);
684 * Decode Temporal Noise Shaping data; reference: table 4.48.
686 * @return Returns error status. 0 - OK, !0 - error
688 static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns,
689 GetBitContext * gb, const IndividualChannelStream * ics) {
690 int w, filt, i, coef_len, coef_res, coef_compress;
691 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
692 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
693 for (w = 0; w < ics->num_windows; w++) {
694 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
695 coef_res = get_bits1(gb);
697 for (filt = 0; filt < tns->n_filt[w]; filt++) {
699 tns->length[w][filt] = get_bits(gb, 6 - 2*is8);
701 if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) {
702 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
703 tns->order[w][filt], tns_max_order);
704 tns->order[w][filt] = 0;
707 if (tns->order[w][filt]) {
708 tns->direction[w][filt] = get_bits1(gb);
709 coef_compress = get_bits1(gb);
710 coef_len = coef_res + 3 - coef_compress;
711 tmp2_idx = 2*coef_compress + coef_res;
713 for (i = 0; i < tns->order[w][filt]; i++)
714 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
723 * Decode Mid/Side data; reference: table 4.54.
725 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
726 * [1] mask is decoded from bitstream; [2] mask is all 1s;
727 * [3] reserved for scalable AAC
729 static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb,
732 if (ms_present == 1) {
733 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
734 cpe->ms_mask[idx] = get_bits1(gb);
735 } else if (ms_present == 2) {
736 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
741 * Decode spectral data; reference: table 4.50.
742 * Dequantize and scale spectral data; reference: 4.6.3.3.
744 * @param coef array of dequantized, scaled spectral data
745 * @param sf array of scalefactors or intensity stereo positions
746 * @param pulse_present set if pulses are present
747 * @param pulse pointer to pulse data struct
748 * @param band_type array of the used band type
750 * @return Returns error status. 0 - OK, !0 - error
752 static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120],
753 int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) {
754 int i, k, g, idx = 0;
755 const int c = 1024/ics->num_windows;
756 const uint16_t * offsets = ics->swb_offset;
757 float *coef_base = coef;
758 static const float sign_lookup[] = { 1.0f, -1.0f };
760 for (g = 0; g < ics->num_windows; g++)
761 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
763 for (g = 0; g < ics->num_window_groups; g++) {
764 for (i = 0; i < ics->max_sfb; i++, idx++) {
765 const int cur_band_type = band_type[idx];
766 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
767 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
769 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
770 for (group = 0; group < ics->group_len[g]; group++) {
771 memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
773 }else if (cur_band_type == NOISE_BT) {
774 for (group = 0; group < ics->group_len[g]; group++) {
776 float band_energy = 0;
777 for (k = offsets[i]; k < offsets[i+1]; k++) {
778 ac->random_state = lcg_random(ac->random_state);
779 coef[group*128+k] = ac->random_state;
780 band_energy += coef[group*128+k]*coef[group*128+k];
782 scale = sf[idx] / sqrtf(band_energy);
783 for (k = offsets[i]; k < offsets[i+1]; k++) {
784 coef[group*128+k] *= scale;
788 for (group = 0; group < ics->group_len[g]; group++) {
789 for (k = offsets[i]; k < offsets[i+1]; k += dim) {
790 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
791 const int coef_tmp_idx = (group << 7) + k;
794 if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
795 av_log(ac->avccontext, AV_LOG_ERROR,
796 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
797 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
800 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
801 if (is_cb_unsigned) {
802 if (vq_ptr[0]) coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
803 if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
805 if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
806 if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
809 coef[coef_tmp_idx ] = 1.0f;
810 coef[coef_tmp_idx + 1] = 1.0f;
812 coef[coef_tmp_idx + 2] = 1.0f;
813 coef[coef_tmp_idx + 3] = 1.0f;
816 if (cur_band_type == ESC_BT) {
817 for (j = 0; j < 2; j++) {
818 if (vq_ptr[j] == 64.0f) {
820 /* The total length of escape_sequence must be < 22 bits according
821 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
822 while (get_bits1(gb) && n < 15) n++;
824 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
827 n = (1<<n) + get_bits(gb, n);
828 coef[coef_tmp_idx + j] *= cbrtf(n) * n;
830 coef[coef_tmp_idx + j] *= vq_ptr[j];
834 coef[coef_tmp_idx ] *= vq_ptr[0];
835 coef[coef_tmp_idx + 1] *= vq_ptr[1];
837 coef[coef_tmp_idx + 2] *= vq_ptr[2];
838 coef[coef_tmp_idx + 3] *= vq_ptr[3];
841 coef[coef_tmp_idx ] *= sf[idx];
842 coef[coef_tmp_idx + 1] *= sf[idx];
844 coef[coef_tmp_idx + 2] *= sf[idx];
845 coef[coef_tmp_idx + 3] *= sf[idx];
851 coef += ics->group_len[g]<<7;
856 for(i = 0; i < pulse->num_pulse; i++){
857 float co = coef_base[ pulse->pos[i] ];
858 while(offsets[idx + 1] <= pulse->pos[i])
860 if (band_type[idx] != NOISE_BT && sf[idx]) {
861 float ico = -pulse->amp[i];
864 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
866 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
873 static av_always_inline float flt16_round(float pf) {
875 pf = frexpf(pf, &exp);
876 pf = ldexpf(roundf(ldexpf(pf, 8)), exp-8);
880 static av_always_inline float flt16_even(float pf) {
882 pf = frexpf(pf, &exp);
883 pf = ldexpf(rintf(ldexpf(pf, 8)), exp-8);
887 static av_always_inline float flt16_trunc(float pf) {
889 pf = frexpf(pf, &exp);
890 pf = ldexpf(truncf(ldexpf(pf, 8)), exp-8);
894 static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
895 const float a = 0.953125; // 61.0/64
896 const float alpha = 0.90625; // 29.0/32
901 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
902 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
904 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
906 *coef += pv * ac->sf_scale;
908 e0 = *coef / ac->sf_scale;
909 e1 = e0 - k1 * ps->r0;
911 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
912 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
913 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
914 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
916 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
917 ps->r0 = flt16_trunc(a * e0);
921 * Apply AAC-Main style frequency domain prediction.
923 static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
926 if (!sce->ics.predictor_initialized) {
927 reset_all_predictors(sce->predictor_state);
928 sce->ics.predictor_initialized = 1;
931 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
932 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
933 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
934 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
935 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
938 if (sce->ics.predictor_reset_group)
939 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
941 reset_all_predictors(sce->predictor_state);
945 * Decode an individual_channel_stream payload; reference: table 4.44.
947 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
948 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
950 * @return Returns error status. 0 - OK, !0 - error
952 static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) {
954 TemporalNoiseShaping * tns = &sce->tns;
955 IndividualChannelStream * ics = &sce->ics;
956 float * out = sce->coeffs;
957 int global_gain, pulse_present = 0;
959 /* This assignment is to silence a GCC warning about the variable being used
960 * uninitialized when in fact it always is.
964 global_gain = get_bits(gb, 8);
966 if (!common_window && !scale_flag) {
967 if (decode_ics_info(ac, ics, gb, 0) < 0)
971 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
973 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
978 if ((pulse_present = get_bits1(gb))) {
979 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
980 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
983 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
984 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
988 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
991 ff_log_missing_feature(ac->avccontext, "SSR", 1);
996 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
999 if(ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1000 apply_prediction(ac, sce);
1006 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1008 static void apply_mid_side_stereo(ChannelElement * cpe) {
1009 const IndividualChannelStream * ics = &cpe->ch[0].ics;
1010 float *ch0 = cpe->ch[0].coeffs;
1011 float *ch1 = cpe->ch[1].coeffs;
1012 int g, i, k, group, idx = 0;
1013 const uint16_t * offsets = ics->swb_offset;
1014 for (g = 0; g < ics->num_window_groups; g++) {
1015 for (i = 0; i < ics->max_sfb; i++, idx++) {
1016 if (cpe->ms_mask[idx] &&
1017 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1018 for (group = 0; group < ics->group_len[g]; group++) {
1019 for (k = offsets[i]; k < offsets[i+1]; k++) {
1020 float tmp = ch0[group*128 + k] - ch1[group*128 + k];
1021 ch0[group*128 + k] += ch1[group*128 + k];
1022 ch1[group*128 + k] = tmp;
1027 ch0 += ics->group_len[g]*128;
1028 ch1 += ics->group_len[g]*128;
1033 * intensity stereo decoding; reference: 4.6.8.2.3
1035 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1036 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1037 * [3] reserved for scalable AAC
1039 static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) {
1040 const IndividualChannelStream * ics = &cpe->ch[1].ics;
1041 SingleChannelElement * sce1 = &cpe->ch[1];
1042 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1043 const uint16_t * offsets = ics->swb_offset;
1044 int g, group, i, k, idx = 0;
1047 for (g = 0; g < ics->num_window_groups; g++) {
1048 for (i = 0; i < ics->max_sfb;) {
1049 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1050 const int bt_run_end = sce1->band_type_run_end[idx];
1051 for (; i < bt_run_end; i++, idx++) {
1052 c = -1 + 2 * (sce1->band_type[idx] - 14);
1054 c *= 1 - 2 * cpe->ms_mask[idx];
1055 scale = c * sce1->sf[idx];
1056 for (group = 0; group < ics->group_len[g]; group++)
1057 for (k = offsets[i]; k < offsets[i+1]; k++)
1058 coef1[group*128 + k] = scale * coef0[group*128 + k];
1061 int bt_run_end = sce1->band_type_run_end[idx];
1062 idx += bt_run_end - i;
1066 coef0 += ics->group_len[g]*128;
1067 coef1 += ics->group_len[g]*128;
1072 * Decode a channel_pair_element; reference: table 4.4.
1074 * @param elem_id Identifies the instance of a syntax element.
1076 * @return Returns error status. 0 - OK, !0 - error
1078 static int decode_cpe(AACContext * ac, GetBitContext * gb, int elem_id) {
1079 int i, ret, common_window, ms_present = 0;
1080 ChannelElement * cpe;
1082 cpe = ac->che[TYPE_CPE][elem_id];
1083 common_window = get_bits1(gb);
1084 if (common_window) {
1085 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1087 i = cpe->ch[1].ics.use_kb_window[0];
1088 cpe->ch[1].ics = cpe->ch[0].ics;
1089 cpe->ch[1].ics.use_kb_window[1] = i;
1090 ms_present = get_bits(gb, 2);
1091 if(ms_present == 3) {
1092 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1094 } else if(ms_present)
1095 decode_mid_side_stereo(cpe, gb, ms_present);
1097 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1099 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1102 if (common_window) {
1104 apply_mid_side_stereo(cpe);
1105 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1106 apply_prediction(ac, &cpe->ch[0]);
1107 apply_prediction(ac, &cpe->ch[1]);
1111 apply_intensity_stereo(cpe, ms_present);
1116 * Decode coupling_channel_element; reference: table 4.8.
1118 * @param elem_id Identifies the instance of a syntax element.
1120 * @return Returns error status. 0 - OK, !0 - error
1122 static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
1127 SingleChannelElement * sce = &che->ch[0];
1128 ChannelCoupling * coup = &che->coup;
1130 coup->coupling_point = 2*get_bits1(gb);
1131 coup->num_coupled = get_bits(gb, 3);
1132 for (c = 0; c <= coup->num_coupled; c++) {
1134 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1135 coup->id_select[c] = get_bits(gb, 4);
1136 if (coup->type[c] == TYPE_CPE) {
1137 coup->ch_select[c] = get_bits(gb, 2);
1138 if (coup->ch_select[c] == 3)
1141 coup->ch_select[c] = 2;
1143 coup->coupling_point += get_bits1(gb);
1145 if (coup->coupling_point == 2) {
1146 av_log(ac->avccontext, AV_LOG_ERROR,
1147 "Independently switched CCE with 'invalid' domain signalled.\n");
1148 memset(coup, 0, sizeof(ChannelCoupling));
1152 sign = get_bits(gb, 1);
1153 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1155 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1158 for (c = 0; c < num_gain; c++) {
1162 float gain_cache = 1.;
1164 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1165 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1166 gain_cache = pow(scale, -gain);
1168 if (coup->coupling_point == AFTER_IMDCT) {
1169 coup->gain[c][0] = gain_cache;
1171 for (g = 0; g < sce->ics.num_window_groups; g++) {
1172 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1173 if (sce->band_type[idx] != ZERO_BT) {
1175 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1183 gain_cache = pow(scale, -t) * s;
1186 coup->gain[c][idx] = gain_cache;
1196 * Decode Spectral Band Replication extension data; reference: table 4.55.
1198 * @param crc flag indicating the presence of CRC checksum
1199 * @param cnt length of TYPE_FIL syntactic element in bytes
1201 * @return Returns number of bytes consumed from the TYPE_FIL element.
1203 static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
1204 // TODO : sbr_extension implementation
1205 ff_log_missing_feature(ac->avccontext, "SBR", 0);
1206 skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
1211 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1213 * @return Returns number of bytes consumed.
1215 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
1217 int num_excl_chan = 0;
1220 for (i = 0; i < 7; i++)
1221 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1222 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1224 return num_excl_chan / 7;
1228 * Decode dynamic range information; reference: table 4.52.
1230 * @param cnt length of TYPE_FIL syntactic element in bytes
1232 * @return Returns number of bytes consumed.
1234 static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
1236 int drc_num_bands = 1;
1239 /* pce_tag_present? */
1241 che_drc->pce_instance_tag = get_bits(gb, 4);
1242 skip_bits(gb, 4); // tag_reserved_bits
1246 /* excluded_chns_present? */
1248 n += decode_drc_channel_exclusions(che_drc, gb);
1251 /* drc_bands_present? */
1252 if (get_bits1(gb)) {
1253 che_drc->band_incr = get_bits(gb, 4);
1254 che_drc->interpolation_scheme = get_bits(gb, 4);
1256 drc_num_bands += che_drc->band_incr;
1257 for (i = 0; i < drc_num_bands; i++) {
1258 che_drc->band_top[i] = get_bits(gb, 8);
1263 /* prog_ref_level_present? */
1264 if (get_bits1(gb)) {
1265 che_drc->prog_ref_level = get_bits(gb, 7);
1266 skip_bits1(gb); // prog_ref_level_reserved_bits
1270 for (i = 0; i < drc_num_bands; i++) {
1271 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1272 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1280 * Decode extension data (incomplete); reference: table 4.51.
1282 * @param cnt length of TYPE_FIL syntactic element in bytes
1284 * @return Returns number of bytes consumed
1286 static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
1289 switch (get_bits(gb, 4)) { // extension type
1290 case EXT_SBR_DATA_CRC:
1293 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1295 case EXT_DYNAMIC_RANGE:
1296 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1300 case EXT_DATA_ELEMENT:
1302 skip_bits_long(gb, 8*cnt - 4);
1309 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1311 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1312 * @param coef spectral coefficients
1314 static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
1315 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1317 int bottom, top, order, start, end, size, inc;
1318 float lpc[TNS_MAX_ORDER];
1320 for (w = 0; w < ics->num_windows; w++) {
1321 bottom = ics->num_swb;
1322 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1324 bottom = FFMAX(0, top - tns->length[w][filt]);
1325 order = tns->order[w][filt];
1330 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1332 start = ics->swb_offset[FFMIN(bottom, mmm)];
1333 end = ics->swb_offset[FFMIN( top, mmm)];
1334 if ((size = end - start) <= 0)
1336 if (tns->direction[w][filt]) {
1337 inc = -1; start = end - 1;
1344 for (m = 0; m < size; m++, start += inc)
1345 for (i = 1; i <= FFMIN(m, order); i++)
1346 coef[start] -= coef[start - i*inc] * lpc[i-1];
1352 * Conduct IMDCT and windowing.
1354 static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
1355 IndividualChannelStream * ics = &sce->ics;
1356 float * in = sce->coeffs;
1357 float * out = sce->ret;
1358 float * saved = sce->saved;
1359 const float * swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1360 const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1361 const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1362 float * buf = ac->buf_mdct;
1363 float * temp = ac->temp;
1367 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1368 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1369 av_log(ac->avccontext, AV_LOG_WARNING,
1370 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1371 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1372 for (i = 0; i < 1024; i += 128)
1373 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1375 ff_imdct_half(&ac->mdct, buf, in);
1377 /* window overlapping
1378 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1379 * and long to short transitions are considered to be short to short
1380 * transitions. This leaves just two cases (long to long and short to short)
1381 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1383 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1384 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1385 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1387 for (i = 0; i < 448; i++)
1388 out[i] = saved[i] + ac->add_bias;
1390 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1391 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1392 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1393 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1394 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1395 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1396 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1398 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1399 for (i = 576; i < 1024; i++)
1400 out[i] = buf[i-512] + ac->add_bias;
1405 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1406 for (i = 0; i < 64; i++)
1407 saved[i] = temp[64 + i] - ac->add_bias;
1408 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1409 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1410 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1411 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1412 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1413 memcpy( saved, buf + 512, 448 * sizeof(float));
1414 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1415 } else { // LONG_STOP or ONLY_LONG
1416 memcpy( saved, buf + 512, 512 * sizeof(float));
1421 * Apply dependent channel coupling (applied before IMDCT).
1423 * @param index index into coupling gain array
1425 static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1426 IndividualChannelStream * ics = &cce->ch[0].ics;
1427 const uint16_t * offsets = ics->swb_offset;
1428 float * dest = target->coeffs;
1429 const float * src = cce->ch[0].coeffs;
1430 int g, i, group, k, idx = 0;
1431 if(ac->m4ac.object_type == AOT_AAC_LTP) {
1432 av_log(ac->avccontext, AV_LOG_ERROR,
1433 "Dependent coupling is not supported together with LTP\n");
1436 for (g = 0; g < ics->num_window_groups; g++) {
1437 for (i = 0; i < ics->max_sfb; i++, idx++) {
1438 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1439 for (group = 0; group < ics->group_len[g]; group++) {
1440 for (k = offsets[i]; k < offsets[i+1]; k++) {
1442 dest[group*128+k] += cce->coup.gain[index][idx] * src[group*128+k];
1447 dest += ics->group_len[g]*128;
1448 src += ics->group_len[g]*128;
1453 * Apply independent channel coupling (applied after IMDCT).
1455 * @param index index into coupling gain array
1457 static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1459 for (i = 0; i < 1024; i++)
1460 target->ret[i] += cce->coup.gain[index][0] * (cce->ch[0].ret[i] - ac->add_bias);
1464 * channel coupling transformation interface
1466 * @param index index into coupling gain array
1467 * @param apply_coupling_method pointer to (in)dependent coupling function
1469 static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
1470 enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point,
1471 void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index))
1475 for (i = 0; i < MAX_ELEM_ID; i++) {
1476 ChannelElement *cce = ac->che[TYPE_CCE][i];
1479 if (cce && cce->coup.coupling_point == coupling_point) {
1480 ChannelCoupling * coup = &cce->coup;
1482 for (c = 0; c <= coup->num_coupled; c++) {
1483 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1484 if (coup->ch_select[c] != 1) {
1485 apply_coupling_method(ac, &cc->ch[0], cce, index);
1486 if (coup->ch_select[c] != 0)
1489 if (coup->ch_select[c] != 2)
1490 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1492 index += 1 + (coup->ch_select[c] == 3);
1499 * Convert spectral data to float samples, applying all supported tools as appropriate.
1501 static void spectral_to_sample(AACContext * ac) {
1503 for(type = 3; type >= 0; type--) {
1504 for (i = 0; i < MAX_ELEM_ID; i++) {
1505 ChannelElement *che = ac->che[type][i];
1507 if(type <= TYPE_CPE)
1508 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1509 if(che->ch[0].tns.present)
1510 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1511 if(che->ch[1].tns.present)
1512 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1513 if(type <= TYPE_CPE)
1514 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1515 if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1516 imdct_and_windowing(ac, &che->ch[0]);
1517 if(type == TYPE_CPE)
1518 imdct_and_windowing(ac, &che->ch[1]);
1519 if(type <= TYPE_CCE)
1520 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1526 static int parse_adts_frame_header(AACContext * ac, GetBitContext * gb) {
1529 AACADTSHeaderInfo hdr_info;
1531 size = ff_aac_parse_header(gb, &hdr_info);
1533 if (hdr_info.chan_config)
1534 ac->m4ac.chan_config = hdr_info.chan_config;
1535 ac->m4ac.sample_rate = hdr_info.sample_rate;
1536 ac->m4ac.sampling_index = hdr_info.sampling_index;
1537 ac->m4ac.object_type = hdr_info.object_type;
1539 if (hdr_info.num_aac_frames == 1) {
1540 if (!hdr_info.crc_absent)
1543 ff_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1549 static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, const uint8_t * buf, int buf_size) {
1550 AACContext * ac = avccontext->priv_data;
1552 enum RawDataBlockType elem_type;
1553 int err, elem_id, data_size_tmp;
1555 init_get_bits(&gb, buf, buf_size*8);
1557 if (show_bits(&gb, 12) == 0xfff) {
1558 if ((err = parse_adts_frame_header(ac, &gb)) < 0) {
1559 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1562 if (ac->m4ac.sampling_index > 12) {
1563 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1569 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1570 elem_id = get_bits(&gb, 4);
1573 if(elem_type == TYPE_SCE && elem_id == 1 &&
1574 !ac->che[TYPE_SCE][elem_id] && ac->che[TYPE_LFE][0]) {
1575 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
1576 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
1577 encountered such a stream, transfer the LFE[0] element to SCE[1] */
1578 ac->che[TYPE_SCE][elem_id] = ac->che[TYPE_LFE][0];
1579 ac->che[TYPE_LFE][0] = NULL;
1581 if(elem_type < TYPE_DSE && !ac->che[elem_type][elem_id]) {
1582 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1586 switch (elem_type) {
1589 err = decode_ics(ac, &ac->che[TYPE_SCE][elem_id]->ch[0], &gb, 0, 0);
1593 err = decode_cpe(ac, &gb, elem_id);
1597 err = decode_cce(ac, &gb, ac->che[TYPE_CCE][elem_id]);
1601 err = decode_ics(ac, &ac->che[TYPE_LFE][elem_id]->ch[0], &gb, 0, 0);
1605 skip_data_stream_element(&gb);
1611 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1612 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1613 if((err = decode_pce(ac, new_che_pos, &gb)))
1615 err = output_configure(ac, ac->che_pos, new_che_pos);
1621 elem_id += get_bits(&gb, 8) - 1;
1623 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1624 err = 0; /* FIXME */
1628 err = -1; /* should not happen, but keeps compiler happy */
1636 spectral_to_sample(ac);
1638 if (!ac->is_saved) {
1644 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1645 if(*data_size < data_size_tmp) {
1646 av_log(avccontext, AV_LOG_ERROR,
1647 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1648 *data_size, data_size_tmp);
1651 *data_size = data_size_tmp;
1653 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1658 static av_cold int aac_decode_close(AVCodecContext * avccontext) {
1659 AACContext * ac = avccontext->priv_data;
1662 for (i = 0; i < MAX_ELEM_ID; i++) {
1663 for(type = 0; type < 4; type++)
1664 av_freep(&ac->che[type][i]);
1667 ff_mdct_end(&ac->mdct);
1668 ff_mdct_end(&ac->mdct_small);
1672 AVCodec aac_decoder = {
1681 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1682 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},