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
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"
95 static VLC vlc_scalefactors;
96 static VLC vlc_spectral[11];
100 * Configure output channel order based on the current program configuration element.
102 * @param che_pos current channel position configuration
103 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
105 * @return Returns error status. 0 - OK, !0 - error
107 static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID],
108 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]) {
109 AVCodecContext *avctx = ac->avccontext;
110 int i, type, channels = 0;
112 if(!memcmp(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])))
113 return 0; /* no change */
115 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
117 /* Allocate or free elements depending on if they are in the
118 * current program configuration.
120 * Set up default 1:1 output mapping.
122 * For a 5.1 stream the output order will be:
123 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
126 for(i = 0; i < MAX_ELEM_ID; i++) {
127 for(type = 0; type < 4; type++) {
128 if(che_pos[type][i]) {
129 if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
130 return AVERROR(ENOMEM);
131 if(type != TYPE_CCE) {
132 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
133 if(type == TYPE_CPE) {
134 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
138 av_freep(&ac->che[type][i]);
142 avctx->channels = channels;
147 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
149 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
150 * @param sce_map mono (Single Channel Element) map
151 * @param type speaker type/position for these channels
153 static void decode_channel_map(enum ChannelPosition *cpe_map,
154 enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) {
156 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
157 map[get_bits(gb, 4)] = type;
162 * Decode program configuration element; reference: table 4.2.
164 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
166 * @return Returns error status. 0 - OK, !0 - error
168 static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
169 GetBitContext * gb) {
170 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
172 skip_bits(gb, 2); // object_type
174 sampling_index = get_bits(gb, 4);
175 if(sampling_index > 11) {
176 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
179 ac->m4ac.sampling_index = sampling_index;
180 ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index];
181 num_front = get_bits(gb, 4);
182 num_side = get_bits(gb, 4);
183 num_back = get_bits(gb, 4);
184 num_lfe = get_bits(gb, 2);
185 num_assoc_data = get_bits(gb, 3);
186 num_cc = get_bits(gb, 4);
189 skip_bits(gb, 4); // mono_mixdown_tag
191 skip_bits(gb, 4); // stereo_mixdown_tag
194 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
196 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
197 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
198 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
199 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
201 skip_bits_long(gb, 4 * num_assoc_data);
203 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
207 /* comment field, first byte is length */
208 skip_bits_long(gb, 8 * get_bits(gb, 8));
213 * Set up channel positions based on a default channel configuration
214 * as specified in table 1.17.
216 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
218 * @return Returns error status. 0 - OK, !0 - error
220 static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
223 if(channel_config < 1 || channel_config > 7) {
224 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
229 /* default channel configurations:
231 * 1ch : front center (mono)
232 * 2ch : L + R (stereo)
233 * 3ch : front center + L + R
234 * 4ch : front center + L + R + back center
235 * 5ch : front center + L + R + back stereo
236 * 6ch : front center + L + R + back stereo + LFE
237 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
240 if(channel_config != 2)
241 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
242 if(channel_config > 1)
243 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
244 if(channel_config == 4)
245 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
246 if(channel_config > 4)
247 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
248 = AAC_CHANNEL_BACK; // back stereo
249 if(channel_config > 5)
250 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
251 if(channel_config == 7)
252 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
258 * Decode GA "General Audio" specific configuration; reference: table 4.1.
260 * @return Returns error status. 0 - OK, !0 - error
262 static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) {
263 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
264 int extension_flag, ret;
266 if(get_bits1(gb)) { // frameLengthFlag
267 ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
271 if (get_bits1(gb)) // dependsOnCoreCoder
272 skip_bits(gb, 14); // coreCoderDelay
273 extension_flag = get_bits1(gb);
275 if(ac->m4ac.object_type == AOT_AAC_SCALABLE ||
276 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
277 skip_bits(gb, 3); // layerNr
279 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
280 if (channel_config == 0) {
281 skip_bits(gb, 4); // element_instance_tag
282 if((ret = decode_pce(ac, new_che_pos, gb)))
285 if((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
288 if((ret = output_configure(ac, ac->che_pos, new_che_pos)))
291 if (extension_flag) {
292 switch (ac->m4ac.object_type) {
294 skip_bits(gb, 5); // numOfSubFrame
295 skip_bits(gb, 11); // layer_length
299 case AOT_ER_AAC_SCALABLE:
301 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
302 * aacScalefactorDataResilienceFlag
303 * aacSpectralDataResilienceFlag
307 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
313 * Decode audio specific configuration; reference: table 1.13.
315 * @param data pointer to AVCodecContext extradata
316 * @param data_size size of AVCCodecContext extradata
318 * @return Returns error status. 0 - OK, !0 - error
320 static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) {
324 init_get_bits(&gb, data, data_size * 8);
326 if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
328 if(ac->m4ac.sampling_index > 11) {
329 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
333 skip_bits_long(&gb, i);
335 switch (ac->m4ac.object_type) {
338 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
342 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
343 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
350 * linear congruential pseudorandom number generator
352 * @param previous_val pointer to the current state of the generator
354 * @return Returns a 32-bit pseudorandom integer
356 static av_always_inline int lcg_random(int previous_val) {
357 return previous_val * 1664525 + 1013904223;
360 static void reset_predict_state(PredictorState * ps) {
369 static void reset_all_predictors(PredictorState * ps) {
371 for (i = 0; i < MAX_PREDICTORS; i++)
372 reset_predict_state(&ps[i]);
375 static void reset_predictor_group(PredictorState * ps, int group_num) {
377 for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
378 reset_predict_state(&ps[i]);
381 static av_cold int aac_decode_init(AVCodecContext * avccontext) {
382 AACContext * ac = avccontext->priv_data;
385 ac->avccontext = avccontext;
387 if (avccontext->extradata_size <= 0 ||
388 decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
391 avccontext->sample_fmt = SAMPLE_FMT_S16;
392 avccontext->sample_rate = ac->m4ac.sample_rate;
393 avccontext->frame_size = 1024;
395 AAC_INIT_VLC_STATIC( 0, 144);
396 AAC_INIT_VLC_STATIC( 1, 114);
397 AAC_INIT_VLC_STATIC( 2, 188);
398 AAC_INIT_VLC_STATIC( 3, 180);
399 AAC_INIT_VLC_STATIC( 4, 172);
400 AAC_INIT_VLC_STATIC( 5, 140);
401 AAC_INIT_VLC_STATIC( 6, 168);
402 AAC_INIT_VLC_STATIC( 7, 114);
403 AAC_INIT_VLC_STATIC( 8, 262);
404 AAC_INIT_VLC_STATIC( 9, 248);
405 AAC_INIT_VLC_STATIC(10, 384);
407 dsputil_init(&ac->dsp, avccontext);
409 ac->random_state = 0x1f2e3d4c;
411 // -1024 - Compensate wrong IMDCT method.
412 // 32768 - Required to scale values to the correct range for the bias method
413 // for float to int16 conversion.
415 if(ac->dsp.float_to_int16 == ff_float_to_int16_c) {
416 ac->add_bias = 385.0f;
417 ac->sf_scale = 1. / (-1024. * 32768.);
421 ac->sf_scale = 1. / -1024.;
425 #ifndef CONFIG_HARDCODED_TABLES
426 for (i = 0; i < 428; i++)
427 ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
428 #endif /* CONFIG_HARDCODED_TABLES */
430 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
431 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
432 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
435 ff_mdct_init(&ac->mdct, 11, 1);
436 ff_mdct_init(&ac->mdct_small, 8, 1);
437 // window initialization
438 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
439 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
440 ff_sine_window_init(ff_sine_1024, 1024);
441 ff_sine_window_init(ff_sine_128, 128);
447 * Skip data_stream_element; reference: table 4.10.
449 static void skip_data_stream_element(GetBitContext * gb) {
450 int byte_align = get_bits1(gb);
451 int count = get_bits(gb, 8);
453 count += get_bits(gb, 8);
456 skip_bits_long(gb, 8 * count);
459 static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
462 ics->predictor_reset_group = get_bits(gb, 5);
463 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
464 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
468 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
469 ics->prediction_used[sfb] = get_bits1(gb);
475 * Decode Individual Channel Stream info; reference: table 4.6.
477 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
479 static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) {
481 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
482 memset(ics, 0, sizeof(IndividualChannelStream));
485 ics->window_sequence[1] = ics->window_sequence[0];
486 ics->window_sequence[0] = get_bits(gb, 2);
487 ics->use_kb_window[1] = ics->use_kb_window[0];
488 ics->use_kb_window[0] = get_bits1(gb);
489 ics->num_window_groups = 1;
490 ics->group_len[0] = 1;
491 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
493 ics->max_sfb = get_bits(gb, 4);
494 for (i = 0; i < 7; i++) {
496 ics->group_len[ics->num_window_groups-1]++;
498 ics->num_window_groups++;
499 ics->group_len[ics->num_window_groups-1] = 1;
502 ics->num_windows = 8;
503 ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
504 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
505 ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
506 ics->predictor_present = 0;
508 ics->max_sfb = get_bits(gb, 6);
509 ics->num_windows = 1;
510 ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
511 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
512 ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
513 ics->predictor_present = get_bits1(gb);
514 ics->predictor_reset_group = 0;
515 if (ics->predictor_present) {
516 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
517 if (decode_prediction(ac, ics, gb)) {
518 memset(ics, 0, sizeof(IndividualChannelStream));
521 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
522 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
523 memset(ics, 0, sizeof(IndividualChannelStream));
526 ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
527 memset(ics, 0, sizeof(IndividualChannelStream));
533 if(ics->max_sfb > ics->num_swb) {
534 av_log(ac->avccontext, AV_LOG_ERROR,
535 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
536 ics->max_sfb, ics->num_swb);
537 memset(ics, 0, sizeof(IndividualChannelStream));
545 * Decode band types (section_data payload); reference: table 4.46.
547 * @param band_type array of the used band type
548 * @param band_type_run_end array of the last scalefactor band of a band type run
550 * @return Returns error status. 0 - OK, !0 - error
552 static int decode_band_types(AACContext * ac, enum BandType band_type[120],
553 int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) {
555 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
556 for (g = 0; g < ics->num_window_groups; g++) {
558 while (k < ics->max_sfb) {
559 uint8_t sect_len = k;
561 int sect_band_type = get_bits(gb, 4);
562 if (sect_band_type == 12) {
563 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
566 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1)
567 sect_len += sect_len_incr;
568 sect_len += sect_len_incr;
569 if (sect_len > ics->max_sfb) {
570 av_log(ac->avccontext, AV_LOG_ERROR,
571 "Number of bands (%d) exceeds limit (%d).\n",
572 sect_len, ics->max_sfb);
575 for (; k < sect_len; k++) {
576 band_type [idx] = sect_band_type;
577 band_type_run_end[idx++] = sect_len;
585 * Decode scalefactors; reference: table 4.47.
587 * @param global_gain first scalefactor value as scalefactors are differentially coded
588 * @param band_type array of the used band type
589 * @param band_type_run_end array of the last scalefactor band of a band type run
590 * @param sf array of scalefactors or intensity stereo positions
592 * @return Returns error status. 0 - OK, !0 - error
594 static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb,
595 unsigned int global_gain, IndividualChannelStream * ics,
596 enum BandType band_type[120], int band_type_run_end[120]) {
597 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
599 int offset[3] = { global_gain, global_gain - 90, 100 };
601 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
602 for (g = 0; g < ics->num_window_groups; g++) {
603 for (i = 0; i < ics->max_sfb;) {
604 int run_end = band_type_run_end[idx];
605 if (band_type[idx] == ZERO_BT) {
606 for(; i < run_end; i++, idx++)
608 }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
609 for(; i < run_end; i++, idx++) {
610 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
611 if(offset[2] > 255U) {
612 av_log(ac->avccontext, AV_LOG_ERROR,
613 "%s (%d) out of range.\n", sf_str[2], offset[2]);
616 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
618 }else if(band_type[idx] == NOISE_BT) {
619 for(; i < run_end; i++, idx++) {
621 offset[1] += get_bits(gb, 9) - 256;
623 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
624 if(offset[1] > 255U) {
625 av_log(ac->avccontext, AV_LOG_ERROR,
626 "%s (%d) out of range.\n", sf_str[1], offset[1]);
629 sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100];
632 for(; i < run_end; i++, idx++) {
633 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
634 if(offset[0] > 255U) {
635 av_log(ac->avccontext, AV_LOG_ERROR,
636 "%s (%d) out of range.\n", sf_str[0], offset[0]);
639 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
648 * Decode pulse data; reference: table 4.7.
650 static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) {
652 pulse->num_pulse = get_bits(gb, 2) + 1;
653 pulse_swb = get_bits(gb, 6);
654 if (pulse_swb >= num_swb)
656 pulse->pos[0] = swb_offset[pulse_swb];
657 pulse->pos[0] += get_bits(gb, 5);
658 if (pulse->pos[0] > 1023)
660 pulse->amp[0] = get_bits(gb, 4);
661 for (i = 1; i < pulse->num_pulse; i++) {
662 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1];
663 if (pulse->pos[i] > 1023)
665 pulse->amp[i] = get_bits(gb, 4);
671 * Decode Temporal Noise Shaping data; reference: table 4.48.
673 * @return Returns error status. 0 - OK, !0 - error
675 static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns,
676 GetBitContext * gb, const IndividualChannelStream * ics) {
677 int w, filt, i, coef_len, coef_res, coef_compress;
678 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
679 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
680 for (w = 0; w < ics->num_windows; w++) {
681 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
682 coef_res = get_bits1(gb);
684 for (filt = 0; filt < tns->n_filt[w]; filt++) {
686 tns->length[w][filt] = get_bits(gb, 6 - 2*is8);
688 if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) {
689 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
690 tns->order[w][filt], tns_max_order);
691 tns->order[w][filt] = 0;
694 if (tns->order[w][filt]) {
695 tns->direction[w][filt] = get_bits1(gb);
696 coef_compress = get_bits1(gb);
697 coef_len = coef_res + 3 - coef_compress;
698 tmp2_idx = 2*coef_compress + coef_res;
700 for (i = 0; i < tns->order[w][filt]; i++)
701 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
710 * Decode Mid/Side data; reference: table 4.54.
712 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
713 * [1] mask is decoded from bitstream; [2] mask is all 1s;
714 * [3] reserved for scalable AAC
716 static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb,
719 if (ms_present == 1) {
720 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
721 cpe->ms_mask[idx] = get_bits1(gb);
722 } else if (ms_present == 2) {
723 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
728 * Decode spectral data; reference: table 4.50.
729 * Dequantize and scale spectral data; reference: 4.6.3.3.
731 * @param coef array of dequantized, scaled spectral data
732 * @param sf array of scalefactors or intensity stereo positions
733 * @param pulse_present set if pulses are present
734 * @param pulse pointer to pulse data struct
735 * @param band_type array of the used band type
737 * @return Returns error status. 0 - OK, !0 - error
739 static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120],
740 int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) {
741 int i, k, g, idx = 0;
742 const int c = 1024/ics->num_windows;
743 const uint16_t * offsets = ics->swb_offset;
744 float *coef_base = coef;
745 static const float sign_lookup[] = { 1.0f, -1.0f };
747 for (g = 0; g < ics->num_windows; g++)
748 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
750 for (g = 0; g < ics->num_window_groups; g++) {
751 for (i = 0; i < ics->max_sfb; i++, idx++) {
752 const int cur_band_type = band_type[idx];
753 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
754 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
756 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
757 for (group = 0; group < ics->group_len[g]; group++) {
758 memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
760 }else if (cur_band_type == NOISE_BT) {
761 for (group = 0; group < ics->group_len[g]; group++) {
763 float band_energy = 0;
764 for (k = offsets[i]; k < offsets[i+1]; k++) {
765 ac->random_state = lcg_random(ac->random_state);
766 coef[group*128+k] = ac->random_state;
767 band_energy += coef[group*128+k]*coef[group*128+k];
769 scale = sf[idx] / sqrtf(band_energy);
770 for (k = offsets[i]; k < offsets[i+1]; k++) {
771 coef[group*128+k] *= scale;
775 for (group = 0; group < ics->group_len[g]; group++) {
776 for (k = offsets[i]; k < offsets[i+1]; k += dim) {
777 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
778 const int coef_tmp_idx = (group << 7) + k;
781 if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
782 av_log(ac->avccontext, AV_LOG_ERROR,
783 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
784 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
787 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
788 if (is_cb_unsigned) {
789 if (vq_ptr[0]) coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
790 if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
792 if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
793 if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
796 coef[coef_tmp_idx ] = 1.0f;
797 coef[coef_tmp_idx + 1] = 1.0f;
799 coef[coef_tmp_idx + 2] = 1.0f;
800 coef[coef_tmp_idx + 3] = 1.0f;
803 if (cur_band_type == ESC_BT) {
804 for (j = 0; j < 2; j++) {
805 if (vq_ptr[j] == 64.0f) {
807 /* The total length of escape_sequence must be < 22 bits according
808 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
809 while (get_bits1(gb) && n < 15) n++;
811 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
814 n = (1<<n) + get_bits(gb, n);
815 coef[coef_tmp_idx + j] *= cbrtf(n) * n;
817 coef[coef_tmp_idx + j] *= vq_ptr[j];
821 coef[coef_tmp_idx ] *= vq_ptr[0];
822 coef[coef_tmp_idx + 1] *= vq_ptr[1];
824 coef[coef_tmp_idx + 2] *= vq_ptr[2];
825 coef[coef_tmp_idx + 3] *= vq_ptr[3];
828 coef[coef_tmp_idx ] *= sf[idx];
829 coef[coef_tmp_idx + 1] *= sf[idx];
831 coef[coef_tmp_idx + 2] *= sf[idx];
832 coef[coef_tmp_idx + 3] *= sf[idx];
838 coef += ics->group_len[g]<<7;
843 for(i = 0; i < pulse->num_pulse; i++){
844 float co = coef_base[ pulse->pos[i] ];
845 while(offsets[idx + 1] <= pulse->pos[i])
847 if (band_type[idx] != NOISE_BT && sf[idx]) {
848 float ico = -pulse->amp[i];
851 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
853 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
860 static av_always_inline float flt16_round(float pf) {
862 pf = frexpf(pf, &exp);
863 pf = ldexpf(roundf(ldexpf(pf, 8)), exp-8);
867 static av_always_inline float flt16_even(float pf) {
869 pf = frexpf(pf, &exp);
870 pf = ldexpf(rintf(ldexpf(pf, 8)), exp-8);
874 static av_always_inline float flt16_trunc(float pf) {
876 pf = frexpf(pf, &exp);
877 pf = ldexpf(truncf(ldexpf(pf, 8)), exp-8);
881 static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
882 const float a = 0.953125; // 61.0/64
883 const float alpha = 0.90625; // 29.0/32
888 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
889 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
891 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
893 *coef += pv * ac->sf_scale;
895 e0 = *coef / ac->sf_scale;
896 e1 = e0 - k1 * ps->r0;
898 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
899 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
900 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
901 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
903 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
904 ps->r0 = flt16_trunc(a * e0);
908 * Apply AAC-Main style frequency domain prediction.
910 static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
913 if (!sce->ics.predictor_initialized) {
914 reset_all_predictors(sce->predictor_state);
915 sce->ics.predictor_initialized = 1;
918 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
919 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
920 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
921 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
922 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
925 if (sce->ics.predictor_reset_group)
926 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
928 reset_all_predictors(sce->predictor_state);
932 * Decode an individual_channel_stream payload; reference: table 4.44.
934 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
935 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
937 * @return Returns error status. 0 - OK, !0 - error
939 static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) {
941 TemporalNoiseShaping * tns = &sce->tns;
942 IndividualChannelStream * ics = &sce->ics;
943 float * out = sce->coeffs;
944 int global_gain, pulse_present = 0;
946 /* This assignment is to silence a GCC warning about the variable being used
947 * uninitialized when in fact it always is.
951 global_gain = get_bits(gb, 8);
953 if (!common_window && !scale_flag) {
954 if (decode_ics_info(ac, ics, gb, 0) < 0)
958 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
960 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
965 if ((pulse_present = get_bits1(gb))) {
966 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
967 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
970 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
971 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
975 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
978 ff_log_missing_feature(ac->avccontext, "SSR", 1);
983 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
986 if(ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
987 apply_prediction(ac, sce);
993 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
995 static void apply_mid_side_stereo(ChannelElement * cpe) {
996 const IndividualChannelStream * ics = &cpe->ch[0].ics;
997 float *ch0 = cpe->ch[0].coeffs;
998 float *ch1 = cpe->ch[1].coeffs;
999 int g, i, k, group, idx = 0;
1000 const uint16_t * offsets = ics->swb_offset;
1001 for (g = 0; g < ics->num_window_groups; g++) {
1002 for (i = 0; i < ics->max_sfb; i++, idx++) {
1003 if (cpe->ms_mask[idx] &&
1004 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1005 for (group = 0; group < ics->group_len[g]; group++) {
1006 for (k = offsets[i]; k < offsets[i+1]; k++) {
1007 float tmp = ch0[group*128 + k] - ch1[group*128 + k];
1008 ch0[group*128 + k] += ch1[group*128 + k];
1009 ch1[group*128 + k] = tmp;
1014 ch0 += ics->group_len[g]*128;
1015 ch1 += ics->group_len[g]*128;
1020 * intensity stereo decoding; reference: 4.6.8.2.3
1022 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1023 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1024 * [3] reserved for scalable AAC
1026 static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) {
1027 const IndividualChannelStream * ics = &cpe->ch[1].ics;
1028 SingleChannelElement * sce1 = &cpe->ch[1];
1029 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1030 const uint16_t * offsets = ics->swb_offset;
1031 int g, group, i, k, idx = 0;
1034 for (g = 0; g < ics->num_window_groups; g++) {
1035 for (i = 0; i < ics->max_sfb;) {
1036 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1037 const int bt_run_end = sce1->band_type_run_end[idx];
1038 for (; i < bt_run_end; i++, idx++) {
1039 c = -1 + 2 * (sce1->band_type[idx] - 14);
1041 c *= 1 - 2 * cpe->ms_mask[idx];
1042 scale = c * sce1->sf[idx];
1043 for (group = 0; group < ics->group_len[g]; group++)
1044 for (k = offsets[i]; k < offsets[i+1]; k++)
1045 coef1[group*128 + k] = scale * coef0[group*128 + k];
1048 int bt_run_end = sce1->band_type_run_end[idx];
1049 idx += bt_run_end - i;
1053 coef0 += ics->group_len[g]*128;
1054 coef1 += ics->group_len[g]*128;
1059 * Decode a channel_pair_element; reference: table 4.4.
1061 * @param elem_id Identifies the instance of a syntax element.
1063 * @return Returns error status. 0 - OK, !0 - error
1065 static int decode_cpe(AACContext * ac, GetBitContext * gb, int elem_id) {
1066 int i, ret, common_window, ms_present = 0;
1067 ChannelElement * cpe;
1069 cpe = ac->che[TYPE_CPE][elem_id];
1070 common_window = get_bits1(gb);
1071 if (common_window) {
1072 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1074 i = cpe->ch[1].ics.use_kb_window[0];
1075 cpe->ch[1].ics = cpe->ch[0].ics;
1076 cpe->ch[1].ics.use_kb_window[1] = i;
1077 ms_present = get_bits(gb, 2);
1078 if(ms_present == 3) {
1079 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1081 } else if(ms_present)
1082 decode_mid_side_stereo(cpe, gb, ms_present);
1084 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1086 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1089 if (common_window) {
1091 apply_mid_side_stereo(cpe);
1092 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1093 apply_prediction(ac, &cpe->ch[0]);
1094 apply_prediction(ac, &cpe->ch[1]);
1098 apply_intensity_stereo(cpe, ms_present);
1103 * Decode coupling_channel_element; reference: table 4.8.
1105 * @param elem_id Identifies the instance of a syntax element.
1107 * @return Returns error status. 0 - OK, !0 - error
1109 static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
1114 SingleChannelElement * sce = &che->ch[0];
1115 ChannelCoupling * coup = &che->coup;
1117 coup->coupling_point = 2*get_bits1(gb);
1118 coup->num_coupled = get_bits(gb, 3);
1119 for (c = 0; c <= coup->num_coupled; c++) {
1121 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1122 coup->id_select[c] = get_bits(gb, 4);
1123 if (coup->type[c] == TYPE_CPE) {
1124 coup->ch_select[c] = get_bits(gb, 2);
1125 if (coup->ch_select[c] == 3)
1128 coup->ch_select[c] = 2;
1130 coup->coupling_point += get_bits1(gb);
1132 if (coup->coupling_point == 2) {
1133 av_log(ac->avccontext, AV_LOG_ERROR,
1134 "Independently switched CCE with 'invalid' domain signalled.\n");
1135 memset(coup, 0, sizeof(ChannelCoupling));
1139 sign = get_bits(gb, 1);
1140 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1142 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1145 for (c = 0; c < num_gain; c++) {
1149 float gain_cache = 1.;
1151 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1152 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1153 gain_cache = pow(scale, -gain);
1155 for (g = 0; g < sce->ics.num_window_groups; g++) {
1156 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1157 if (sce->band_type[idx] != ZERO_BT) {
1159 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1167 gain_cache = pow(scale, -t) * s;
1170 coup->gain[c][idx] = gain_cache;
1179 * Decode Spectral Band Replication extension data; reference: table 4.55.
1181 * @param crc flag indicating the presence of CRC checksum
1182 * @param cnt length of TYPE_FIL syntactic element in bytes
1184 * @return Returns number of bytes consumed from the TYPE_FIL element.
1186 static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
1187 // TODO : sbr_extension implementation
1188 ff_log_missing_feature(ac->avccontext, "SBR", 0);
1189 skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
1194 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1196 * @return Returns number of bytes consumed.
1198 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
1200 int num_excl_chan = 0;
1203 for (i = 0; i < 7; i++)
1204 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1205 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1207 return num_excl_chan / 7;
1211 * Decode dynamic range information; reference: table 4.52.
1213 * @param cnt length of TYPE_FIL syntactic element in bytes
1215 * @return Returns number of bytes consumed.
1217 static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
1219 int drc_num_bands = 1;
1222 /* pce_tag_present? */
1224 che_drc->pce_instance_tag = get_bits(gb, 4);
1225 skip_bits(gb, 4); // tag_reserved_bits
1229 /* excluded_chns_present? */
1231 n += decode_drc_channel_exclusions(che_drc, gb);
1234 /* drc_bands_present? */
1235 if (get_bits1(gb)) {
1236 che_drc->band_incr = get_bits(gb, 4);
1237 che_drc->interpolation_scheme = get_bits(gb, 4);
1239 drc_num_bands += che_drc->band_incr;
1240 for (i = 0; i < drc_num_bands; i++) {
1241 che_drc->band_top[i] = get_bits(gb, 8);
1246 /* prog_ref_level_present? */
1247 if (get_bits1(gb)) {
1248 che_drc->prog_ref_level = get_bits(gb, 7);
1249 skip_bits1(gb); // prog_ref_level_reserved_bits
1253 for (i = 0; i < drc_num_bands; i++) {
1254 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1255 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1263 * Decode extension data (incomplete); reference: table 4.51.
1265 * @param cnt length of TYPE_FIL syntactic element in bytes
1267 * @return Returns number of bytes consumed
1269 static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
1272 switch (get_bits(gb, 4)) { // extension type
1273 case EXT_SBR_DATA_CRC:
1276 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1278 case EXT_DYNAMIC_RANGE:
1279 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1283 case EXT_DATA_ELEMENT:
1285 skip_bits_long(gb, 8*cnt - 4);
1292 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1294 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1295 * @param coef spectral coefficients
1297 static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
1298 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1300 int bottom, top, order, start, end, size, inc;
1301 float lpc[TNS_MAX_ORDER];
1303 for (w = 0; w < ics->num_windows; w++) {
1304 bottom = ics->num_swb;
1305 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1307 bottom = FFMAX(0, top - tns->length[w][filt]);
1308 order = tns->order[w][filt];
1313 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1315 start = ics->swb_offset[FFMIN(bottom, mmm)];
1316 end = ics->swb_offset[FFMIN( top, mmm)];
1317 if ((size = end - start) <= 0)
1319 if (tns->direction[w][filt]) {
1320 inc = -1; start = end - 1;
1327 for (m = 0; m < size; m++, start += inc)
1328 for (i = 1; i <= FFMIN(m, order); i++)
1329 coef[start] -= coef[start - i*inc] * lpc[i-1];
1335 * Conduct IMDCT and windowing.
1337 static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
1338 IndividualChannelStream * ics = &sce->ics;
1339 float * in = sce->coeffs;
1340 float * out = sce->ret;
1341 float * saved = sce->saved;
1342 const float * swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1343 const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1344 const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1345 float * buf = ac->buf_mdct;
1346 float * temp = ac->temp;
1350 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1351 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1352 av_log(ac->avccontext, AV_LOG_WARNING,
1353 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1354 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1355 for (i = 0; i < 1024; i += 128)
1356 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1358 ff_imdct_half(&ac->mdct, buf, in);
1360 /* window overlapping
1361 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1362 * and long to short transitions are considered to be short to short
1363 * transitions. This leaves just two cases (long to long and short to short)
1364 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1366 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1367 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1368 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1370 for (i = 0; i < 448; i++)
1371 out[i] = saved[i] + ac->add_bias;
1373 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1374 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1375 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1376 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1377 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1378 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1379 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1381 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1382 for (i = 576; i < 1024; i++)
1383 out[i] = buf[i-512] + ac->add_bias;
1388 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1389 for (i = 0; i < 64; i++)
1390 saved[i] = temp[64 + i] - ac->add_bias;
1391 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1392 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1393 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1394 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1395 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1396 memcpy( saved, buf + 512, 448 * sizeof(float));
1397 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1398 } else { // LONG_STOP or ONLY_LONG
1399 memcpy( saved, buf + 512, 512 * sizeof(float));
1404 * Apply dependent channel coupling (applied before IMDCT).
1406 * @param index index into coupling gain array
1408 static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1409 IndividualChannelStream * ics = &cce->ch[0].ics;
1410 const uint16_t * offsets = ics->swb_offset;
1411 float * dest = target->coeffs;
1412 const float * src = cce->ch[0].coeffs;
1413 int g, i, group, k, idx = 0;
1414 if(ac->m4ac.object_type == AOT_AAC_LTP) {
1415 av_log(ac->avccontext, AV_LOG_ERROR,
1416 "Dependent coupling is not supported together with LTP\n");
1419 for (g = 0; g < ics->num_window_groups; g++) {
1420 for (i = 0; i < ics->max_sfb; i++, idx++) {
1421 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1422 for (group = 0; group < ics->group_len[g]; group++) {
1423 for (k = offsets[i]; k < offsets[i+1]; k++) {
1425 dest[group*128+k] += cce->coup.gain[index][idx] * src[group*128+k];
1430 dest += ics->group_len[g]*128;
1431 src += ics->group_len[g]*128;
1436 * Apply independent channel coupling (applied after IMDCT).
1438 * @param index index into coupling gain array
1440 static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1442 for (i = 0; i < 1024; i++)
1443 target->ret[i] += cce->coup.gain[index][0] * (cce->ch[0].ret[i] - ac->add_bias);
1447 * channel coupling transformation interface
1449 * @param index index into coupling gain array
1450 * @param apply_coupling_method pointer to (in)dependent coupling function
1452 static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
1453 enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point,
1454 void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index))
1458 for (i = 0; i < MAX_ELEM_ID; i++) {
1459 ChannelElement *cce = ac->che[TYPE_CCE][i];
1462 if (cce && cce->coup.coupling_point == coupling_point) {
1463 ChannelCoupling * coup = &cce->coup;
1465 for (c = 0; c <= coup->num_coupled; c++) {
1466 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1467 if (coup->ch_select[c] != 1) {
1468 apply_coupling_method(ac, &cc->ch[0], cce, index);
1469 if (coup->ch_select[c] != 0)
1472 if (coup->ch_select[c] != 2)
1473 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1475 index += 1 + (coup->ch_select[c] == 3);
1482 * Convert spectral data to float samples, applying all supported tools as appropriate.
1484 static void spectral_to_sample(AACContext * ac) {
1486 enum RawDataBlockType type;
1487 for(type = 3; type >= 0; type--) {
1488 for (i = 0; i < MAX_ELEM_ID; i++) {
1489 ChannelElement *che = ac->che[type][i];
1491 if(type <= TYPE_CPE)
1492 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1493 if(che->ch[0].tns.present)
1494 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1495 if(che->ch[1].tns.present)
1496 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1497 if(type <= TYPE_CPE)
1498 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1499 if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1500 imdct_and_windowing(ac, &che->ch[0]);
1501 if(type == TYPE_CPE)
1502 imdct_and_windowing(ac, &che->ch[1]);
1503 if(type <= TYPE_CCE)
1504 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1510 static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, const uint8_t * buf, int buf_size) {
1511 AACContext * ac = avccontext->priv_data;
1513 enum RawDataBlockType elem_type;
1514 int err, elem_id, data_size_tmp;
1516 init_get_bits(&gb, buf, buf_size*8);
1519 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1520 elem_id = get_bits(&gb, 4);
1523 if(elem_type == TYPE_SCE && elem_id == 1 &&
1524 !ac->che[TYPE_SCE][elem_id] && ac->che[TYPE_LFE][0]) {
1525 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
1526 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
1527 encountered such a stream, transfer the LFE[0] element to SCE[1] */
1528 ac->che[TYPE_SCE][elem_id] = ac->che[TYPE_LFE][0];
1529 ac->che[TYPE_LFE][0] = NULL;
1531 if(elem_type < TYPE_DSE) {
1532 if(!ac->che[elem_type][elem_id])
1534 if(elem_type != TYPE_CCE)
1535 ac->che[elem_type][elem_id]->coup.coupling_point = 4;
1538 switch (elem_type) {
1541 err = decode_ics(ac, &ac->che[TYPE_SCE][elem_id]->ch[0], &gb, 0, 0);
1545 err = decode_cpe(ac, &gb, elem_id);
1549 err = decode_cce(ac, &gb, ac->che[TYPE_CCE][elem_id]);
1553 err = decode_ics(ac, &ac->che[TYPE_LFE][elem_id]->ch[0], &gb, 0, 0);
1557 skip_data_stream_element(&gb);
1563 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1564 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1565 if((err = decode_pce(ac, new_che_pos, &gb)))
1567 err = output_configure(ac, ac->che_pos, new_che_pos);
1573 elem_id += get_bits(&gb, 8) - 1;
1575 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1576 err = 0; /* FIXME */
1580 err = -1; /* should not happen, but keeps compiler happy */
1588 spectral_to_sample(ac);
1590 if (!ac->is_saved) {
1596 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1597 if(*data_size < data_size_tmp) {
1598 av_log(avccontext, AV_LOG_ERROR,
1599 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1600 *data_size, data_size_tmp);
1603 *data_size = data_size_tmp;
1605 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1610 static av_cold int aac_decode_close(AVCodecContext * avccontext) {
1611 AACContext * ac = avccontext->priv_data;
1614 for (i = 0; i < MAX_ELEM_ID; i++) {
1615 for(type = 0; type < 4; type++)
1616 av_freep(&ac->che[type][i]);
1619 ff_mdct_end(&ac->mdct);
1620 ff_mdct_end(&ac->mdct_small);
1624 AVCodec aac_decoder = {
1633 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1634 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},