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
87 #include "aacdectab.h"
88 #include "mpeg4audio.h"
89 #include "aac_parser.h"
96 union float754 { float f; uint32_t i; };
98 static VLC vlc_scalefactors;
99 static VLC vlc_spectral[11];
102 static ChannelElement* get_che(AACContext *ac, int type, int elem_id) {
103 static const int8_t tags_per_config[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 };
104 if (ac->tag_che_map[type][elem_id]) {
105 return ac->tag_che_map[type][elem_id];
107 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
110 switch (ac->m4ac.chan_config) {
112 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
114 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
117 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
118 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
119 encountered such a stream, transfer the LFE[0] element to SCE[1] */
120 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
122 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
125 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
127 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
130 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
132 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
136 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
138 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
139 } else if (ac->m4ac.chan_config == 2) {
143 if (!ac->tags_mapped && type == TYPE_SCE) {
145 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
153 * Configure output channel order based on the current program configuration element.
155 * @param che_pos current channel position configuration
156 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
158 * @return Returns error status. 0 - OK, !0 - error
160 static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID],
161 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID], int channel_config) {
162 AVCodecContext *avctx = ac->avccontext;
163 int i, type, channels = 0;
165 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
167 /* Allocate or free elements depending on if they are in the
168 * current program configuration.
170 * Set up default 1:1 output mapping.
172 * For a 5.1 stream the output order will be:
173 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
176 for(i = 0; i < MAX_ELEM_ID; i++) {
177 for(type = 0; type < 4; type++) {
178 if(che_pos[type][i]) {
179 if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
180 return AVERROR(ENOMEM);
181 if(type != TYPE_CCE) {
182 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
183 if(type == TYPE_CPE) {
184 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
188 av_freep(&ac->che[type][i]);
192 if (channel_config) {
193 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
196 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
197 ac->tags_mapped = 4*MAX_ELEM_ID;
200 avctx->channels = channels;
206 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
208 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
209 * @param sce_map mono (Single Channel Element) map
210 * @param type speaker type/position for these channels
212 static void decode_channel_map(enum ChannelPosition *cpe_map,
213 enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) {
215 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
216 map[get_bits(gb, 4)] = type;
221 * Decode program configuration element; reference: table 4.2.
223 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
225 * @return Returns error status. 0 - OK, !0 - error
227 static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
228 GetBitContext * gb) {
229 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
231 skip_bits(gb, 2); // object_type
233 sampling_index = get_bits(gb, 4);
234 if(sampling_index > 12) {
235 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
238 ac->m4ac.sampling_index = sampling_index;
239 ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index];
240 num_front = get_bits(gb, 4);
241 num_side = get_bits(gb, 4);
242 num_back = get_bits(gb, 4);
243 num_lfe = get_bits(gb, 2);
244 num_assoc_data = get_bits(gb, 3);
245 num_cc = get_bits(gb, 4);
248 skip_bits(gb, 4); // mono_mixdown_tag
250 skip_bits(gb, 4); // stereo_mixdown_tag
253 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
255 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
256 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
257 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
258 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
260 skip_bits_long(gb, 4 * num_assoc_data);
262 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
266 /* comment field, first byte is length */
267 skip_bits_long(gb, 8 * get_bits(gb, 8));
272 * Set up channel positions based on a default channel configuration
273 * as specified in table 1.17.
275 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
277 * @return Returns error status. 0 - OK, !0 - error
279 static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
282 if(channel_config < 1 || channel_config > 7) {
283 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
288 /* default channel configurations:
290 * 1ch : front center (mono)
291 * 2ch : L + R (stereo)
292 * 3ch : front center + L + R
293 * 4ch : front center + L + R + back center
294 * 5ch : front center + L + R + back stereo
295 * 6ch : front center + L + R + back stereo + LFE
296 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
299 if(channel_config != 2)
300 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
301 if(channel_config > 1)
302 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
303 if(channel_config == 4)
304 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
305 if(channel_config > 4)
306 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
307 = AAC_CHANNEL_BACK; // back stereo
308 if(channel_config > 5)
309 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
310 if(channel_config == 7)
311 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
317 * Decode GA "General Audio" specific configuration; reference: table 4.1.
319 * @return Returns error status. 0 - OK, !0 - error
321 static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) {
322 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
323 int extension_flag, ret;
325 if(get_bits1(gb)) { // frameLengthFlag
326 ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
330 if (get_bits1(gb)) // dependsOnCoreCoder
331 skip_bits(gb, 14); // coreCoderDelay
332 extension_flag = get_bits1(gb);
334 if(ac->m4ac.object_type == AOT_AAC_SCALABLE ||
335 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
336 skip_bits(gb, 3); // layerNr
338 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
339 if (channel_config == 0) {
340 skip_bits(gb, 4); // element_instance_tag
341 if((ret = decode_pce(ac, new_che_pos, gb)))
344 if((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
347 if((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config)))
350 if (extension_flag) {
351 switch (ac->m4ac.object_type) {
353 skip_bits(gb, 5); // numOfSubFrame
354 skip_bits(gb, 11); // layer_length
358 case AOT_ER_AAC_SCALABLE:
360 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
361 * aacScalefactorDataResilienceFlag
362 * aacSpectralDataResilienceFlag
366 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
372 * Decode audio specific configuration; reference: table 1.13.
374 * @param data pointer to AVCodecContext extradata
375 * @param data_size size of AVCCodecContext extradata
377 * @return Returns error status. 0 - OK, !0 - error
379 static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) {
383 init_get_bits(&gb, data, data_size * 8);
385 if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
387 if(ac->m4ac.sampling_index > 12) {
388 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
392 skip_bits_long(&gb, i);
394 switch (ac->m4ac.object_type) {
397 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
401 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
402 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
409 * linear congruential pseudorandom number generator
411 * @param previous_val pointer to the current state of the generator
413 * @return Returns a 32-bit pseudorandom integer
415 static av_always_inline int lcg_random(int previous_val) {
416 return previous_val * 1664525 + 1013904223;
419 static void reset_predict_state(PredictorState * ps) {
428 static void reset_all_predictors(PredictorState * ps) {
430 for (i = 0; i < MAX_PREDICTORS; i++)
431 reset_predict_state(&ps[i]);
434 static void reset_predictor_group(PredictorState * ps, int group_num) {
436 for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
437 reset_predict_state(&ps[i]);
440 static av_cold int aac_decode_init(AVCodecContext * avccontext) {
441 AACContext * ac = avccontext->priv_data;
444 ac->avccontext = avccontext;
446 if (avccontext->extradata_size > 0) {
447 if(decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
449 avccontext->sample_rate = ac->m4ac.sample_rate;
450 } else if (avccontext->channels > 0) {
451 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
452 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
453 if(set_default_channel_config(ac, new_che_pos, avccontext->channels - (avccontext->channels == 8)))
455 if(output_configure(ac, ac->che_pos, new_che_pos, 1))
457 ac->m4ac.sample_rate = avccontext->sample_rate;
460 avccontext->sample_fmt = SAMPLE_FMT_S16;
461 avccontext->frame_size = 1024;
463 AAC_INIT_VLC_STATIC( 0, 144);
464 AAC_INIT_VLC_STATIC( 1, 114);
465 AAC_INIT_VLC_STATIC( 2, 188);
466 AAC_INIT_VLC_STATIC( 3, 180);
467 AAC_INIT_VLC_STATIC( 4, 172);
468 AAC_INIT_VLC_STATIC( 5, 140);
469 AAC_INIT_VLC_STATIC( 6, 168);
470 AAC_INIT_VLC_STATIC( 7, 114);
471 AAC_INIT_VLC_STATIC( 8, 262);
472 AAC_INIT_VLC_STATIC( 9, 248);
473 AAC_INIT_VLC_STATIC(10, 384);
475 dsputil_init(&ac->dsp, avccontext);
477 ac->random_state = 0x1f2e3d4c;
479 // -1024 - Compensate wrong IMDCT method.
480 // 32768 - Required to scale values to the correct range for the bias method
481 // for float to int16 conversion.
483 if(ac->dsp.float_to_int16 == ff_float_to_int16_c) {
484 ac->add_bias = 385.0f;
485 ac->sf_scale = 1. / (-1024. * 32768.);
489 ac->sf_scale = 1. / -1024.;
493 #if !CONFIG_HARDCODED_TABLES
494 for (i = 0; i < 428; i++)
495 ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
496 #endif /* CONFIG_HARDCODED_TABLES */
498 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
499 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
500 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
503 ff_mdct_init(&ac->mdct, 11, 1);
504 ff_mdct_init(&ac->mdct_small, 8, 1);
505 // window initialization
506 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
507 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
508 ff_sine_window_init(ff_sine_1024, 1024);
509 ff_sine_window_init(ff_sine_128, 128);
515 * Skip data_stream_element; reference: table 4.10.
517 static void skip_data_stream_element(GetBitContext * gb) {
518 int byte_align = get_bits1(gb);
519 int count = get_bits(gb, 8);
521 count += get_bits(gb, 8);
524 skip_bits_long(gb, 8 * count);
527 static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
530 ics->predictor_reset_group = get_bits(gb, 5);
531 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
532 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
536 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
537 ics->prediction_used[sfb] = get_bits1(gb);
543 * Decode Individual Channel Stream info; reference: table 4.6.
545 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
547 static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) {
549 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
550 memset(ics, 0, sizeof(IndividualChannelStream));
553 ics->window_sequence[1] = ics->window_sequence[0];
554 ics->window_sequence[0] = get_bits(gb, 2);
555 ics->use_kb_window[1] = ics->use_kb_window[0];
556 ics->use_kb_window[0] = get_bits1(gb);
557 ics->num_window_groups = 1;
558 ics->group_len[0] = 1;
559 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
561 ics->max_sfb = get_bits(gb, 4);
562 for (i = 0; i < 7; i++) {
564 ics->group_len[ics->num_window_groups-1]++;
566 ics->num_window_groups++;
567 ics->group_len[ics->num_window_groups-1] = 1;
570 ics->num_windows = 8;
571 ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
572 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
573 ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
574 ics->predictor_present = 0;
576 ics->max_sfb = get_bits(gb, 6);
577 ics->num_windows = 1;
578 ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
579 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
580 ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
581 ics->predictor_present = get_bits1(gb);
582 ics->predictor_reset_group = 0;
583 if (ics->predictor_present) {
584 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
585 if (decode_prediction(ac, ics, gb)) {
586 memset(ics, 0, sizeof(IndividualChannelStream));
589 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
590 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
591 memset(ics, 0, sizeof(IndividualChannelStream));
594 ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
595 memset(ics, 0, sizeof(IndividualChannelStream));
601 if(ics->max_sfb > ics->num_swb) {
602 av_log(ac->avccontext, AV_LOG_ERROR,
603 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
604 ics->max_sfb, ics->num_swb);
605 memset(ics, 0, sizeof(IndividualChannelStream));
613 * Decode band types (section_data payload); reference: table 4.46.
615 * @param band_type array of the used band type
616 * @param band_type_run_end array of the last scalefactor band of a band type run
618 * @return Returns error status. 0 - OK, !0 - error
620 static int decode_band_types(AACContext * ac, enum BandType band_type[120],
621 int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) {
623 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
624 for (g = 0; g < ics->num_window_groups; g++) {
626 while (k < ics->max_sfb) {
627 uint8_t sect_len = k;
629 int sect_band_type = get_bits(gb, 4);
630 if (sect_band_type == 12) {
631 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
634 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1)
635 sect_len += sect_len_incr;
636 sect_len += sect_len_incr;
637 if (sect_len > ics->max_sfb) {
638 av_log(ac->avccontext, AV_LOG_ERROR,
639 "Number of bands (%d) exceeds limit (%d).\n",
640 sect_len, ics->max_sfb);
643 for (; k < sect_len; k++) {
644 band_type [idx] = sect_band_type;
645 band_type_run_end[idx++] = sect_len;
653 * Decode scalefactors; reference: table 4.47.
655 * @param global_gain first scalefactor value as scalefactors are differentially coded
656 * @param band_type array of the used band type
657 * @param band_type_run_end array of the last scalefactor band of a band type run
658 * @param sf array of scalefactors or intensity stereo positions
660 * @return Returns error status. 0 - OK, !0 - error
662 static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb,
663 unsigned int global_gain, IndividualChannelStream * ics,
664 enum BandType band_type[120], int band_type_run_end[120]) {
665 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
667 int offset[3] = { global_gain, global_gain - 90, 100 };
669 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
670 for (g = 0; g < ics->num_window_groups; g++) {
671 for (i = 0; i < ics->max_sfb;) {
672 int run_end = band_type_run_end[idx];
673 if (band_type[idx] == ZERO_BT) {
674 for(; i < run_end; i++, idx++)
676 }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
677 for(; i < run_end; i++, idx++) {
678 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
679 if(offset[2] > 255U) {
680 av_log(ac->avccontext, AV_LOG_ERROR,
681 "%s (%d) out of range.\n", sf_str[2], offset[2]);
684 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
686 }else if(band_type[idx] == NOISE_BT) {
687 for(; i < run_end; i++, idx++) {
689 offset[1] += get_bits(gb, 9) - 256;
691 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
692 if(offset[1] > 255U) {
693 av_log(ac->avccontext, AV_LOG_ERROR,
694 "%s (%d) out of range.\n", sf_str[1], offset[1]);
697 sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100];
700 for(; i < run_end; i++, idx++) {
701 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
702 if(offset[0] > 255U) {
703 av_log(ac->avccontext, AV_LOG_ERROR,
704 "%s (%d) out of range.\n", sf_str[0], offset[0]);
707 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
716 * Decode pulse data; reference: table 4.7.
718 static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) {
720 pulse->num_pulse = get_bits(gb, 2) + 1;
721 pulse_swb = get_bits(gb, 6);
722 if (pulse_swb >= num_swb)
724 pulse->pos[0] = swb_offset[pulse_swb];
725 pulse->pos[0] += get_bits(gb, 5);
726 if (pulse->pos[0] > 1023)
728 pulse->amp[0] = get_bits(gb, 4);
729 for (i = 1; i < pulse->num_pulse; i++) {
730 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1];
731 if (pulse->pos[i] > 1023)
733 pulse->amp[i] = get_bits(gb, 4);
739 * Decode Temporal Noise Shaping data; reference: table 4.48.
741 * @return Returns error status. 0 - OK, !0 - error
743 static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns,
744 GetBitContext * gb, const IndividualChannelStream * ics) {
745 int w, filt, i, coef_len, coef_res, coef_compress;
746 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
747 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
748 for (w = 0; w < ics->num_windows; w++) {
749 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
750 coef_res = get_bits1(gb);
752 for (filt = 0; filt < tns->n_filt[w]; filt++) {
754 tns->length[w][filt] = get_bits(gb, 6 - 2*is8);
756 if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) {
757 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
758 tns->order[w][filt], tns_max_order);
759 tns->order[w][filt] = 0;
762 if (tns->order[w][filt]) {
763 tns->direction[w][filt] = get_bits1(gb);
764 coef_compress = get_bits1(gb);
765 coef_len = coef_res + 3 - coef_compress;
766 tmp2_idx = 2*coef_compress + coef_res;
768 for (i = 0; i < tns->order[w][filt]; i++)
769 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
778 * Decode Mid/Side data; reference: table 4.54.
780 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
781 * [1] mask is decoded from bitstream; [2] mask is all 1s;
782 * [3] reserved for scalable AAC
784 static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb,
787 if (ms_present == 1) {
788 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
789 cpe->ms_mask[idx] = get_bits1(gb);
790 } else if (ms_present == 2) {
791 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
796 * Decode spectral data; reference: table 4.50.
797 * Dequantize and scale spectral data; reference: 4.6.3.3.
799 * @param coef array of dequantized, scaled spectral data
800 * @param sf array of scalefactors or intensity stereo positions
801 * @param pulse_present set if pulses are present
802 * @param pulse pointer to pulse data struct
803 * @param band_type array of the used band type
805 * @return Returns error status. 0 - OK, !0 - error
807 static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120],
808 int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) {
809 int i, k, g, idx = 0;
810 const int c = 1024/ics->num_windows;
811 const uint16_t * offsets = ics->swb_offset;
812 float *coef_base = coef;
813 static const float sign_lookup[] = { 1.0f, -1.0f };
815 for (g = 0; g < ics->num_windows; g++)
816 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
818 for (g = 0; g < ics->num_window_groups; g++) {
819 for (i = 0; i < ics->max_sfb; i++, idx++) {
820 const int cur_band_type = band_type[idx];
821 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
822 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
824 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
825 for (group = 0; group < ics->group_len[g]; group++) {
826 memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
828 }else if (cur_band_type == NOISE_BT) {
829 for (group = 0; group < ics->group_len[g]; group++) {
831 float band_energy = 0;
832 for (k = offsets[i]; k < offsets[i+1]; k++) {
833 ac->random_state = lcg_random(ac->random_state);
834 coef[group*128+k] = ac->random_state;
835 band_energy += coef[group*128+k]*coef[group*128+k];
837 scale = sf[idx] / sqrtf(band_energy);
838 for (k = offsets[i]; k < offsets[i+1]; k++) {
839 coef[group*128+k] *= scale;
843 for (group = 0; group < ics->group_len[g]; group++) {
844 for (k = offsets[i]; k < offsets[i+1]; k += dim) {
845 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
846 const int coef_tmp_idx = (group << 7) + k;
849 if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
850 av_log(ac->avccontext, AV_LOG_ERROR,
851 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
852 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
855 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
856 if (is_cb_unsigned) {
857 if (vq_ptr[0]) coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
858 if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
860 if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
861 if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
863 if (cur_band_type == ESC_BT) {
864 for (j = 0; j < 2; j++) {
865 if (vq_ptr[j] == 64.0f) {
867 /* The total length of escape_sequence must be < 22 bits according
868 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
869 while (get_bits1(gb) && n < 15) n++;
871 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
874 n = (1<<n) + get_bits(gb, n);
875 coef[coef_tmp_idx + j] *= cbrtf(n) * n;
877 coef[coef_tmp_idx + j] *= vq_ptr[j];
881 coef[coef_tmp_idx ] *= vq_ptr[0];
882 coef[coef_tmp_idx + 1] *= vq_ptr[1];
884 coef[coef_tmp_idx + 2] *= vq_ptr[2];
885 coef[coef_tmp_idx + 3] *= vq_ptr[3];
889 coef[coef_tmp_idx ] = vq_ptr[0];
890 coef[coef_tmp_idx + 1] = vq_ptr[1];
892 coef[coef_tmp_idx + 2] = vq_ptr[2];
893 coef[coef_tmp_idx + 3] = vq_ptr[3];
896 coef[coef_tmp_idx ] *= sf[idx];
897 coef[coef_tmp_idx + 1] *= sf[idx];
899 coef[coef_tmp_idx + 2] *= sf[idx];
900 coef[coef_tmp_idx + 3] *= sf[idx];
906 coef += ics->group_len[g]<<7;
911 for(i = 0; i < pulse->num_pulse; i++){
912 float co = coef_base[ pulse->pos[i] ];
913 while(offsets[idx + 1] <= pulse->pos[i])
915 if (band_type[idx] != NOISE_BT && sf[idx]) {
916 float ico = -pulse->amp[i];
919 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
921 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
928 static av_always_inline float flt16_round(float pf) {
931 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
935 static av_always_inline float flt16_even(float pf) {
938 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U>>16)) & 0xFFFF0000U;
942 static av_always_inline float flt16_trunc(float pf) {
945 pun.i &= 0xFFFF0000U;
949 static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
950 const float a = 0.953125; // 61.0/64
951 const float alpha = 0.90625; // 29.0/32
956 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
957 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
959 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
961 *coef += pv * ac->sf_scale;
963 e0 = *coef / ac->sf_scale;
964 e1 = e0 - k1 * ps->r0;
966 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
967 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
968 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
969 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
971 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
972 ps->r0 = flt16_trunc(a * e0);
976 * Apply AAC-Main style frequency domain prediction.
978 static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
981 if (!sce->ics.predictor_initialized) {
982 reset_all_predictors(sce->predictor_state);
983 sce->ics.predictor_initialized = 1;
986 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
987 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
988 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
989 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
990 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
993 if (sce->ics.predictor_reset_group)
994 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
996 reset_all_predictors(sce->predictor_state);
1000 * Decode an individual_channel_stream payload; reference: table 4.44.
1002 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1003 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1005 * @return Returns error status. 0 - OK, !0 - error
1007 static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) {
1009 TemporalNoiseShaping * tns = &sce->tns;
1010 IndividualChannelStream * ics = &sce->ics;
1011 float * out = sce->coeffs;
1012 int global_gain, pulse_present = 0;
1014 /* This assignment is to silence a GCC warning about the variable being used
1015 * uninitialized when in fact it always is.
1017 pulse.num_pulse = 0;
1019 global_gain = get_bits(gb, 8);
1021 if (!common_window && !scale_flag) {
1022 if (decode_ics_info(ac, ics, gb, 0) < 0)
1026 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1028 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1033 if ((pulse_present = get_bits1(gb))) {
1034 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1035 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1038 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1039 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1043 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1045 if (get_bits1(gb)) {
1046 ff_log_missing_feature(ac->avccontext, "SSR", 1);
1051 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1054 if(ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1055 apply_prediction(ac, sce);
1061 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1063 static void apply_mid_side_stereo(ChannelElement * cpe) {
1064 const IndividualChannelStream * ics = &cpe->ch[0].ics;
1065 float *ch0 = cpe->ch[0].coeffs;
1066 float *ch1 = cpe->ch[1].coeffs;
1067 int g, i, k, group, idx = 0;
1068 const uint16_t * offsets = ics->swb_offset;
1069 for (g = 0; g < ics->num_window_groups; g++) {
1070 for (i = 0; i < ics->max_sfb; i++, idx++) {
1071 if (cpe->ms_mask[idx] &&
1072 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1073 for (group = 0; group < ics->group_len[g]; group++) {
1074 for (k = offsets[i]; k < offsets[i+1]; k++) {
1075 float tmp = ch0[group*128 + k] - ch1[group*128 + k];
1076 ch0[group*128 + k] += ch1[group*128 + k];
1077 ch1[group*128 + k] = tmp;
1082 ch0 += ics->group_len[g]*128;
1083 ch1 += ics->group_len[g]*128;
1088 * intensity stereo decoding; reference: 4.6.8.2.3
1090 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1091 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1092 * [3] reserved for scalable AAC
1094 static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) {
1095 const IndividualChannelStream * ics = &cpe->ch[1].ics;
1096 SingleChannelElement * sce1 = &cpe->ch[1];
1097 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1098 const uint16_t * offsets = ics->swb_offset;
1099 int g, group, i, k, idx = 0;
1102 for (g = 0; g < ics->num_window_groups; g++) {
1103 for (i = 0; i < ics->max_sfb;) {
1104 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1105 const int bt_run_end = sce1->band_type_run_end[idx];
1106 for (; i < bt_run_end; i++, idx++) {
1107 c = -1 + 2 * (sce1->band_type[idx] - 14);
1109 c *= 1 - 2 * cpe->ms_mask[idx];
1110 scale = c * sce1->sf[idx];
1111 for (group = 0; group < ics->group_len[g]; group++)
1112 for (k = offsets[i]; k < offsets[i+1]; k++)
1113 coef1[group*128 + k] = scale * coef0[group*128 + k];
1116 int bt_run_end = sce1->band_type_run_end[idx];
1117 idx += bt_run_end - i;
1121 coef0 += ics->group_len[g]*128;
1122 coef1 += ics->group_len[g]*128;
1127 * Decode a channel_pair_element; reference: table 4.4.
1129 * @param elem_id Identifies the instance of a syntax element.
1131 * @return Returns error status. 0 - OK, !0 - error
1133 static int decode_cpe(AACContext * ac, GetBitContext * gb, ChannelElement * cpe) {
1134 int i, ret, common_window, ms_present = 0;
1136 common_window = get_bits1(gb);
1137 if (common_window) {
1138 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1140 i = cpe->ch[1].ics.use_kb_window[0];
1141 cpe->ch[1].ics = cpe->ch[0].ics;
1142 cpe->ch[1].ics.use_kb_window[1] = i;
1143 ms_present = get_bits(gb, 2);
1144 if(ms_present == 3) {
1145 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1147 } else if(ms_present)
1148 decode_mid_side_stereo(cpe, gb, ms_present);
1150 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1152 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1155 if (common_window) {
1157 apply_mid_side_stereo(cpe);
1158 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1159 apply_prediction(ac, &cpe->ch[0]);
1160 apply_prediction(ac, &cpe->ch[1]);
1164 apply_intensity_stereo(cpe, ms_present);
1169 * Decode coupling_channel_element; reference: table 4.8.
1171 * @param elem_id Identifies the instance of a syntax element.
1173 * @return Returns error status. 0 - OK, !0 - error
1175 static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
1180 SingleChannelElement * sce = &che->ch[0];
1181 ChannelCoupling * coup = &che->coup;
1183 coup->coupling_point = 2*get_bits1(gb);
1184 coup->num_coupled = get_bits(gb, 3);
1185 for (c = 0; c <= coup->num_coupled; c++) {
1187 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1188 coup->id_select[c] = get_bits(gb, 4);
1189 if (coup->type[c] == TYPE_CPE) {
1190 coup->ch_select[c] = get_bits(gb, 2);
1191 if (coup->ch_select[c] == 3)
1194 coup->ch_select[c] = 2;
1196 coup->coupling_point += get_bits1(gb) || (coup->coupling_point>>1);
1198 sign = get_bits(gb, 1);
1199 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1201 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1204 for (c = 0; c < num_gain; c++) {
1208 float gain_cache = 1.;
1210 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1211 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1212 gain_cache = pow(scale, -gain);
1214 if (coup->coupling_point == AFTER_IMDCT) {
1215 coup->gain[c][0] = gain_cache;
1217 for (g = 0; g < sce->ics.num_window_groups; g++) {
1218 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1219 if (sce->band_type[idx] != ZERO_BT) {
1221 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1229 gain_cache = pow(scale, -t) * s;
1232 coup->gain[c][idx] = gain_cache;
1242 * Decode Spectral Band Replication extension data; reference: table 4.55.
1244 * @param crc flag indicating the presence of CRC checksum
1245 * @param cnt length of TYPE_FIL syntactic element in bytes
1247 * @return Returns number of bytes consumed from the TYPE_FIL element.
1249 static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
1250 // TODO : sbr_extension implementation
1251 ff_log_missing_feature(ac->avccontext, "SBR", 0);
1252 skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
1257 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1259 * @return Returns number of bytes consumed.
1261 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
1263 int num_excl_chan = 0;
1266 for (i = 0; i < 7; i++)
1267 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1268 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1270 return num_excl_chan / 7;
1274 * Decode dynamic range information; reference: table 4.52.
1276 * @param cnt length of TYPE_FIL syntactic element in bytes
1278 * @return Returns number of bytes consumed.
1280 static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
1282 int drc_num_bands = 1;
1285 /* pce_tag_present? */
1287 che_drc->pce_instance_tag = get_bits(gb, 4);
1288 skip_bits(gb, 4); // tag_reserved_bits
1292 /* excluded_chns_present? */
1294 n += decode_drc_channel_exclusions(che_drc, gb);
1297 /* drc_bands_present? */
1298 if (get_bits1(gb)) {
1299 che_drc->band_incr = get_bits(gb, 4);
1300 che_drc->interpolation_scheme = get_bits(gb, 4);
1302 drc_num_bands += che_drc->band_incr;
1303 for (i = 0; i < drc_num_bands; i++) {
1304 che_drc->band_top[i] = get_bits(gb, 8);
1309 /* prog_ref_level_present? */
1310 if (get_bits1(gb)) {
1311 che_drc->prog_ref_level = get_bits(gb, 7);
1312 skip_bits1(gb); // prog_ref_level_reserved_bits
1316 for (i = 0; i < drc_num_bands; i++) {
1317 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1318 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1326 * Decode extension data (incomplete); reference: table 4.51.
1328 * @param cnt length of TYPE_FIL syntactic element in bytes
1330 * @return Returns number of bytes consumed
1332 static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
1335 switch (get_bits(gb, 4)) { // extension type
1336 case EXT_SBR_DATA_CRC:
1339 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1341 case EXT_DYNAMIC_RANGE:
1342 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1346 case EXT_DATA_ELEMENT:
1348 skip_bits_long(gb, 8*cnt - 4);
1355 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1357 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1358 * @param coef spectral coefficients
1360 static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
1361 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1363 int bottom, top, order, start, end, size, inc;
1364 float lpc[TNS_MAX_ORDER];
1366 for (w = 0; w < ics->num_windows; w++) {
1367 bottom = ics->num_swb;
1368 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1370 bottom = FFMAX(0, top - tns->length[w][filt]);
1371 order = tns->order[w][filt];
1376 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1378 start = ics->swb_offset[FFMIN(bottom, mmm)];
1379 end = ics->swb_offset[FFMIN( top, mmm)];
1380 if ((size = end - start) <= 0)
1382 if (tns->direction[w][filt]) {
1383 inc = -1; start = end - 1;
1390 for (m = 0; m < size; m++, start += inc)
1391 for (i = 1; i <= FFMIN(m, order); i++)
1392 coef[start] -= coef[start - i*inc] * lpc[i-1];
1398 * Conduct IMDCT and windowing.
1400 static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
1401 IndividualChannelStream * ics = &sce->ics;
1402 float * in = sce->coeffs;
1403 float * out = sce->ret;
1404 float * saved = sce->saved;
1405 const float * swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1406 const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1407 const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1408 float * buf = ac->buf_mdct;
1409 float * temp = ac->temp;
1413 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1414 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1415 av_log(ac->avccontext, AV_LOG_WARNING,
1416 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1417 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1418 for (i = 0; i < 1024; i += 128)
1419 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1421 ff_imdct_half(&ac->mdct, buf, in);
1423 /* window overlapping
1424 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1425 * and long to short transitions are considered to be short to short
1426 * transitions. This leaves just two cases (long to long and short to short)
1427 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1429 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1430 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1431 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1433 for (i = 0; i < 448; i++)
1434 out[i] = saved[i] + ac->add_bias;
1436 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1437 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1438 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1439 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1440 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1441 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1442 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1444 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1445 for (i = 576; i < 1024; i++)
1446 out[i] = buf[i-512] + ac->add_bias;
1451 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1452 for (i = 0; i < 64; i++)
1453 saved[i] = temp[64 + i] - ac->add_bias;
1454 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1455 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1456 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1457 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1458 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1459 memcpy( saved, buf + 512, 448 * sizeof(float));
1460 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1461 } else { // LONG_STOP or ONLY_LONG
1462 memcpy( saved, buf + 512, 512 * sizeof(float));
1467 * Apply dependent channel coupling (applied before IMDCT).
1469 * @param index index into coupling gain array
1471 static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1472 IndividualChannelStream * ics = &cce->ch[0].ics;
1473 const uint16_t * offsets = ics->swb_offset;
1474 float * dest = target->coeffs;
1475 const float * src = cce->ch[0].coeffs;
1476 int g, i, group, k, idx = 0;
1477 if(ac->m4ac.object_type == AOT_AAC_LTP) {
1478 av_log(ac->avccontext, AV_LOG_ERROR,
1479 "Dependent coupling is not supported together with LTP\n");
1482 for (g = 0; g < ics->num_window_groups; g++) {
1483 for (i = 0; i < ics->max_sfb; i++, idx++) {
1484 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1485 const float gain = cce->coup.gain[index][idx];
1486 for (group = 0; group < ics->group_len[g]; group++) {
1487 for (k = offsets[i]; k < offsets[i+1]; k++) {
1489 dest[group*128+k] += gain * src[group*128+k];
1494 dest += ics->group_len[g]*128;
1495 src += ics->group_len[g]*128;
1500 * Apply independent channel coupling (applied after IMDCT).
1502 * @param index index into coupling gain array
1504 static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1506 const float gain = cce->coup.gain[index][0];
1507 const float bias = ac->add_bias;
1508 const float* src = cce->ch[0].ret;
1509 float* dest = target->ret;
1511 for (i = 0; i < 1024; i++)
1512 dest[i] += gain * (src[i] - bias);
1516 * channel coupling transformation interface
1518 * @param index index into coupling gain array
1519 * @param apply_coupling_method pointer to (in)dependent coupling function
1521 static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
1522 enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point,
1523 void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index))
1527 for (i = 0; i < MAX_ELEM_ID; i++) {
1528 ChannelElement *cce = ac->che[TYPE_CCE][i];
1531 if (cce && cce->coup.coupling_point == coupling_point) {
1532 ChannelCoupling * coup = &cce->coup;
1534 for (c = 0; c <= coup->num_coupled; c++) {
1535 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1536 if (coup->ch_select[c] != 1) {
1537 apply_coupling_method(ac, &cc->ch[0], cce, index);
1538 if (coup->ch_select[c] != 0)
1541 if (coup->ch_select[c] != 2)
1542 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1544 index += 1 + (coup->ch_select[c] == 3);
1551 * Convert spectral data to float samples, applying all supported tools as appropriate.
1553 static void spectral_to_sample(AACContext * ac) {
1555 for(type = 3; type >= 0; type--) {
1556 for (i = 0; i < MAX_ELEM_ID; i++) {
1557 ChannelElement *che = ac->che[type][i];
1559 if(type <= TYPE_CPE)
1560 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1561 if(che->ch[0].tns.present)
1562 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1563 if(che->ch[1].tns.present)
1564 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1565 if(type <= TYPE_CPE)
1566 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1567 if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1568 imdct_and_windowing(ac, &che->ch[0]);
1569 if(type == TYPE_CPE)
1570 imdct_and_windowing(ac, &che->ch[1]);
1571 if(type <= TYPE_CCE)
1572 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1578 static int parse_adts_frame_header(AACContext * ac, GetBitContext * gb) {
1581 AACADTSHeaderInfo hdr_info;
1583 size = ff_aac_parse_header(gb, &hdr_info);
1585 if (hdr_info.chan_config)
1586 ac->m4ac.chan_config = hdr_info.chan_config;
1587 ac->m4ac.sample_rate = hdr_info.sample_rate;
1588 ac->m4ac.sampling_index = hdr_info.sampling_index;
1589 ac->m4ac.object_type = hdr_info.object_type;
1590 if (hdr_info.num_aac_frames == 1) {
1591 if (!hdr_info.crc_absent)
1594 ff_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1601 static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, AVPacket *avpkt) {
1602 const uint8_t *buf = avpkt->data;
1603 int buf_size = avpkt->size;
1604 AACContext * ac = avccontext->priv_data;
1605 ChannelElement * che = NULL;
1607 enum RawDataBlockType elem_type;
1608 int err, elem_id, data_size_tmp;
1610 init_get_bits(&gb, buf, buf_size*8);
1612 if (show_bits(&gb, 12) == 0xfff) {
1613 if (parse_adts_frame_header(ac, &gb) < 0) {
1614 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1617 if (ac->m4ac.sampling_index > 12) {
1618 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1624 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1625 elem_id = get_bits(&gb, 4);
1627 if(elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1628 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1632 switch (elem_type) {
1635 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1639 err = decode_cpe(ac, &gb, che);
1643 err = decode_cce(ac, &gb, che);
1647 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1651 skip_data_stream_element(&gb);
1657 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1658 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1659 if((err = decode_pce(ac, new_che_pos, &gb)))
1661 err = output_configure(ac, ac->che_pos, new_che_pos, 0);
1667 elem_id += get_bits(&gb, 8) - 1;
1669 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1670 err = 0; /* FIXME */
1674 err = -1; /* should not happen, but keeps compiler happy */
1682 spectral_to_sample(ac);
1684 if (!ac->is_saved) {
1690 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1691 if(*data_size < data_size_tmp) {
1692 av_log(avccontext, AV_LOG_ERROR,
1693 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1694 *data_size, data_size_tmp);
1697 *data_size = data_size_tmp;
1699 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1704 static av_cold int aac_decode_close(AVCodecContext * avccontext) {
1705 AACContext * ac = avccontext->priv_data;
1708 for (i = 0; i < MAX_ELEM_ID; i++) {
1709 for(type = 0; type < 4; type++)
1710 av_freep(&ac->che[type][i]);
1713 ff_mdct_end(&ac->mdct);
1714 ff_mdct_end(&ac->mdct_small);
1718 AVCodec aac_decoder = {
1727 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1728 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},