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"
101 static VLC vlc_scalefactors;
102 static VLC vlc_spectral[11];
104 static float cbrt_tab[1<<13];
106 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
108 if (ac->tag_che_map[type][elem_id]) {
109 return ac->tag_che_map[type][elem_id];
111 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
114 switch (ac->m4ac.chan_config) {
116 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
118 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
121 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
122 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
123 encountered such a stream, transfer the LFE[0] element to SCE[1] */
124 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
126 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
129 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
131 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
134 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
136 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
140 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
142 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
143 } else if (ac->m4ac.chan_config == 2) {
147 if (!ac->tags_mapped && type == TYPE_SCE) {
149 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
157 * Check for the channel element in the current channel position configuration.
158 * If it exists, make sure the appropriate element is allocated and map the
159 * channel order to match the internal FFmpeg channel layout.
161 * @param che_pos current channel position configuration
162 * @param type channel element type
163 * @param id channel element id
164 * @param channels count of the number of channels in the configuration
166 * @return Returns error status. 0 - OK, !0 - error
168 static int che_configure(AACContext *ac,
169 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
173 if (che_pos[type][id]) {
174 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
175 return AVERROR(ENOMEM);
176 if (type != TYPE_CCE) {
177 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
178 if (type == TYPE_CPE) {
179 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
183 av_freep(&ac->che[type][id]);
188 * Configure output channel order based on the current program configuration element.
190 * @param che_pos current channel position configuration
191 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
193 * @return Returns error status. 0 - OK, !0 - error
195 static int output_configure(AACContext *ac,
196 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
197 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
198 int channel_config, enum OCStatus oc_type)
200 AVCodecContext *avctx = ac->avccontext;
201 int i, type, channels = 0, ret;
203 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
205 if (channel_config) {
206 for (i = 0; i < tags_per_config[channel_config]; i++) {
207 if ((ret = che_configure(ac, che_pos,
208 aac_channel_layout_map[channel_config - 1][i][0],
209 aac_channel_layout_map[channel_config - 1][i][1],
214 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
217 avctx->channel_layout = aac_channel_layout[channel_config - 1];
219 /* Allocate or free elements depending on if they are in the
220 * current program configuration.
222 * Set up default 1:1 output mapping.
224 * For a 5.1 stream the output order will be:
225 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
228 for (i = 0; i < MAX_ELEM_ID; i++) {
229 for (type = 0; type < 4; type++) {
230 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
235 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
236 ac->tags_mapped = 4 * MAX_ELEM_ID;
238 avctx->channel_layout = 0;
241 avctx->channels = channels;
243 ac->output_configured = oc_type;
249 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
251 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
252 * @param sce_map mono (Single Channel Element) map
253 * @param type speaker type/position for these channels
255 static void decode_channel_map(enum ChannelPosition *cpe_map,
256 enum ChannelPosition *sce_map,
257 enum ChannelPosition type,
258 GetBitContext *gb, int n)
261 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
262 map[get_bits(gb, 4)] = type;
267 * Decode program configuration element; reference: table 4.2.
269 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
271 * @return Returns error status. 0 - OK, !0 - error
273 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
276 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
278 skip_bits(gb, 2); // object_type
280 sampling_index = get_bits(gb, 4);
281 if (ac->m4ac.sampling_index != sampling_index)
282 av_log(ac->avccontext, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
284 num_front = get_bits(gb, 4);
285 num_side = get_bits(gb, 4);
286 num_back = get_bits(gb, 4);
287 num_lfe = get_bits(gb, 2);
288 num_assoc_data = get_bits(gb, 3);
289 num_cc = get_bits(gb, 4);
292 skip_bits(gb, 4); // mono_mixdown_tag
294 skip_bits(gb, 4); // stereo_mixdown_tag
297 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
299 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
300 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
301 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
302 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
304 skip_bits_long(gb, 4 * num_assoc_data);
306 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
310 /* comment field, first byte is length */
311 skip_bits_long(gb, 8 * get_bits(gb, 8));
316 * Set up channel positions based on a default channel configuration
317 * as specified in table 1.17.
319 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
321 * @return Returns error status. 0 - OK, !0 - error
323 static int set_default_channel_config(AACContext *ac,
324 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
327 if (channel_config < 1 || channel_config > 7) {
328 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
333 /* default channel configurations:
335 * 1ch : front center (mono)
336 * 2ch : L + R (stereo)
337 * 3ch : front center + L + R
338 * 4ch : front center + L + R + back center
339 * 5ch : front center + L + R + back stereo
340 * 6ch : front center + L + R + back stereo + LFE
341 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
344 if (channel_config != 2)
345 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
346 if (channel_config > 1)
347 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
348 if (channel_config == 4)
349 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
350 if (channel_config > 4)
351 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
352 = AAC_CHANNEL_BACK; // back stereo
353 if (channel_config > 5)
354 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
355 if (channel_config == 7)
356 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
362 * Decode GA "General Audio" specific configuration; reference: table 4.1.
364 * @return Returns error status. 0 - OK, !0 - error
366 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
369 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
370 int extension_flag, ret;
372 if (get_bits1(gb)) { // frameLengthFlag
373 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
377 if (get_bits1(gb)) // dependsOnCoreCoder
378 skip_bits(gb, 14); // coreCoderDelay
379 extension_flag = get_bits1(gb);
381 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
382 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
383 skip_bits(gb, 3); // layerNr
385 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
386 if (channel_config == 0) {
387 skip_bits(gb, 4); // element_instance_tag
388 if ((ret = decode_pce(ac, new_che_pos, gb)))
391 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
394 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
397 if (extension_flag) {
398 switch (ac->m4ac.object_type) {
400 skip_bits(gb, 5); // numOfSubFrame
401 skip_bits(gb, 11); // layer_length
405 case AOT_ER_AAC_SCALABLE:
407 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
408 * aacScalefactorDataResilienceFlag
409 * aacSpectralDataResilienceFlag
413 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
419 * Decode audio specific configuration; reference: table 1.13.
421 * @param data pointer to AVCodecContext extradata
422 * @param data_size size of AVCCodecContext extradata
424 * @return Returns error status. 0 - OK, !0 - error
426 static int decode_audio_specific_config(AACContext *ac, void *data,
432 init_get_bits(&gb, data, data_size * 8);
434 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
436 if (ac->m4ac.sampling_index > 12) {
437 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
441 skip_bits_long(&gb, i);
443 switch (ac->m4ac.object_type) {
446 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
450 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
451 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
458 * linear congruential pseudorandom number generator
460 * @param previous_val pointer to the current state of the generator
462 * @return Returns a 32-bit pseudorandom integer
464 static av_always_inline int lcg_random(int previous_val)
466 return previous_val * 1664525 + 1013904223;
469 static void reset_predict_state(PredictorState *ps)
479 static void reset_all_predictors(PredictorState *ps)
482 for (i = 0; i < MAX_PREDICTORS; i++)
483 reset_predict_state(&ps[i]);
486 static void reset_predictor_group(PredictorState *ps, int group_num)
489 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
490 reset_predict_state(&ps[i]);
493 static av_cold int aac_decode_init(AVCodecContext *avccontext)
495 AACContext *ac = avccontext->priv_data;
498 ac->avccontext = avccontext;
500 if (avccontext->extradata_size > 0) {
501 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
503 avccontext->sample_rate = ac->m4ac.sample_rate;
504 } else if (avccontext->channels > 0) {
505 ac->m4ac.sample_rate = avccontext->sample_rate;
508 avccontext->sample_fmt = SAMPLE_FMT_S16;
509 avccontext->frame_size = 1024;
511 AAC_INIT_VLC_STATIC( 0, 144);
512 AAC_INIT_VLC_STATIC( 1, 114);
513 AAC_INIT_VLC_STATIC( 2, 188);
514 AAC_INIT_VLC_STATIC( 3, 180);
515 AAC_INIT_VLC_STATIC( 4, 172);
516 AAC_INIT_VLC_STATIC( 5, 140);
517 AAC_INIT_VLC_STATIC( 6, 168);
518 AAC_INIT_VLC_STATIC( 7, 114);
519 AAC_INIT_VLC_STATIC( 8, 262);
520 AAC_INIT_VLC_STATIC( 9, 248);
521 AAC_INIT_VLC_STATIC(10, 384);
523 dsputil_init(&ac->dsp, avccontext);
525 ac->random_state = 0x1f2e3d4c;
527 // -1024 - Compensate wrong IMDCT method.
528 // 32768 - Required to scale values to the correct range for the bias method
529 // for float to int16 conversion.
531 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
532 ac->add_bias = 385.0f;
533 ac->sf_scale = 1. / (-1024. * 32768.);
537 ac->sf_scale = 1. / -1024.;
541 #if !CONFIG_HARDCODED_TABLES
542 for (i = 0; i < 428; i++)
543 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
544 #endif /* CONFIG_HARDCODED_TABLES */
546 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
547 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
548 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
551 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
552 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
553 // window initialization
554 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
555 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
556 ff_init_ff_sine_windows(10);
557 ff_init_ff_sine_windows( 7);
559 if (!cbrt_tab[(1<<13) - 1])
560 for (i = 0; i < 1<<13; i++)
561 cbrt_tab[i] = cbrtf(i) * i;
567 * Skip data_stream_element; reference: table 4.10.
569 static void skip_data_stream_element(GetBitContext *gb)
571 int byte_align = get_bits1(gb);
572 int count = get_bits(gb, 8);
574 count += get_bits(gb, 8);
577 skip_bits_long(gb, 8 * count);
580 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
585 ics->predictor_reset_group = get_bits(gb, 5);
586 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
587 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
591 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
592 ics->prediction_used[sfb] = get_bits1(gb);
598 * Decode Individual Channel Stream info; reference: table 4.6.
600 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
602 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
603 GetBitContext *gb, int common_window)
606 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
607 memset(ics, 0, sizeof(IndividualChannelStream));
610 ics->window_sequence[1] = ics->window_sequence[0];
611 ics->window_sequence[0] = get_bits(gb, 2);
612 ics->use_kb_window[1] = ics->use_kb_window[0];
613 ics->use_kb_window[0] = get_bits1(gb);
614 ics->num_window_groups = 1;
615 ics->group_len[0] = 1;
616 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
618 ics->max_sfb = get_bits(gb, 4);
619 for (i = 0; i < 7; i++) {
621 ics->group_len[ics->num_window_groups - 1]++;
623 ics->num_window_groups++;
624 ics->group_len[ics->num_window_groups - 1] = 1;
627 ics->num_windows = 8;
628 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
629 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
630 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
631 ics->predictor_present = 0;
633 ics->max_sfb = get_bits(gb, 6);
634 ics->num_windows = 1;
635 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
636 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
637 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
638 ics->predictor_present = get_bits1(gb);
639 ics->predictor_reset_group = 0;
640 if (ics->predictor_present) {
641 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
642 if (decode_prediction(ac, ics, gb)) {
643 memset(ics, 0, sizeof(IndividualChannelStream));
646 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
647 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
648 memset(ics, 0, sizeof(IndividualChannelStream));
651 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
652 memset(ics, 0, sizeof(IndividualChannelStream));
658 if (ics->max_sfb > ics->num_swb) {
659 av_log(ac->avccontext, AV_LOG_ERROR,
660 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
661 ics->max_sfb, ics->num_swb);
662 memset(ics, 0, sizeof(IndividualChannelStream));
670 * Decode band types (section_data payload); reference: table 4.46.
672 * @param band_type array of the used band type
673 * @param band_type_run_end array of the last scalefactor band of a band type run
675 * @return Returns error status. 0 - OK, !0 - error
677 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
678 int band_type_run_end[120], GetBitContext *gb,
679 IndividualChannelStream *ics)
682 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
683 for (g = 0; g < ics->num_window_groups; g++) {
685 while (k < ics->max_sfb) {
686 uint8_t sect_end = k;
688 int sect_band_type = get_bits(gb, 4);
689 if (sect_band_type == 12) {
690 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
693 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
694 sect_end += sect_len_incr;
695 sect_end += sect_len_incr;
696 if (sect_end > ics->max_sfb) {
697 av_log(ac->avccontext, AV_LOG_ERROR,
698 "Number of bands (%d) exceeds limit (%d).\n",
699 sect_end, ics->max_sfb);
702 for (; k < sect_end; k++) {
703 band_type [idx] = sect_band_type;
704 band_type_run_end[idx++] = sect_end;
712 * Decode scalefactors; reference: table 4.47.
714 * @param global_gain first scalefactor value as scalefactors are differentially coded
715 * @param band_type array of the used band type
716 * @param band_type_run_end array of the last scalefactor band of a band type run
717 * @param sf array of scalefactors or intensity stereo positions
719 * @return Returns error status. 0 - OK, !0 - error
721 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
722 unsigned int global_gain,
723 IndividualChannelStream *ics,
724 enum BandType band_type[120],
725 int band_type_run_end[120])
727 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
729 int offset[3] = { global_gain, global_gain - 90, 100 };
731 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
732 for (g = 0; g < ics->num_window_groups; g++) {
733 for (i = 0; i < ics->max_sfb;) {
734 int run_end = band_type_run_end[idx];
735 if (band_type[idx] == ZERO_BT) {
736 for (; i < run_end; i++, idx++)
738 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
739 for (; i < run_end; i++, idx++) {
740 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
741 if (offset[2] > 255U) {
742 av_log(ac->avccontext, AV_LOG_ERROR,
743 "%s (%d) out of range.\n", sf_str[2], offset[2]);
746 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
748 } else if (band_type[idx] == NOISE_BT) {
749 for (; i < run_end; i++, idx++) {
750 if (noise_flag-- > 0)
751 offset[1] += get_bits(gb, 9) - 256;
753 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
754 if (offset[1] > 255U) {
755 av_log(ac->avccontext, AV_LOG_ERROR,
756 "%s (%d) out of range.\n", sf_str[1], offset[1]);
759 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
762 for (; i < run_end; i++, idx++) {
763 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
764 if (offset[0] > 255U) {
765 av_log(ac->avccontext, AV_LOG_ERROR,
766 "%s (%d) out of range.\n", sf_str[0], offset[0]);
769 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
778 * Decode pulse data; reference: table 4.7.
780 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
781 const uint16_t *swb_offset, int num_swb)
784 pulse->num_pulse = get_bits(gb, 2) + 1;
785 pulse_swb = get_bits(gb, 6);
786 if (pulse_swb >= num_swb)
788 pulse->pos[0] = swb_offset[pulse_swb];
789 pulse->pos[0] += get_bits(gb, 5);
790 if (pulse->pos[0] > 1023)
792 pulse->amp[0] = get_bits(gb, 4);
793 for (i = 1; i < pulse->num_pulse; i++) {
794 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
795 if (pulse->pos[i] > 1023)
797 pulse->amp[i] = get_bits(gb, 4);
803 * Decode Temporal Noise Shaping data; reference: table 4.48.
805 * @return Returns error status. 0 - OK, !0 - error
807 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
808 GetBitContext *gb, const IndividualChannelStream *ics)
810 int w, filt, i, coef_len, coef_res, coef_compress;
811 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
812 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
813 for (w = 0; w < ics->num_windows; w++) {
814 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
815 coef_res = get_bits1(gb);
817 for (filt = 0; filt < tns->n_filt[w]; filt++) {
819 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
821 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
822 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
823 tns->order[w][filt], tns_max_order);
824 tns->order[w][filt] = 0;
827 if (tns->order[w][filt]) {
828 tns->direction[w][filt] = get_bits1(gb);
829 coef_compress = get_bits1(gb);
830 coef_len = coef_res + 3 - coef_compress;
831 tmp2_idx = 2 * coef_compress + coef_res;
833 for (i = 0; i < tns->order[w][filt]; i++)
834 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
843 * Decode Mid/Side data; reference: table 4.54.
845 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
846 * [1] mask is decoded from bitstream; [2] mask is all 1s;
847 * [3] reserved for scalable AAC
849 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
853 if (ms_present == 1) {
854 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
855 cpe->ms_mask[idx] = get_bits1(gb);
856 } else if (ms_present == 2) {
857 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
862 * Decode spectral data; reference: table 4.50.
863 * Dequantize and scale spectral data; reference: 4.6.3.3.
865 * @param coef array of dequantized, scaled spectral data
866 * @param sf array of scalefactors or intensity stereo positions
867 * @param pulse_present set if pulses are present
868 * @param pulse pointer to pulse data struct
869 * @param band_type array of the used band type
871 * @return Returns error status. 0 - OK, !0 - error
873 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
874 GetBitContext *gb, const float sf[120],
875 int pulse_present, const Pulse *pulse,
876 const IndividualChannelStream *ics,
877 enum BandType band_type[120])
879 int i, k, g, idx = 0;
880 const int c = 1024 / ics->num_windows;
881 const uint16_t *offsets = ics->swb_offset;
882 float *coef_base = coef;
883 static const float sign_lookup[] = { 1.0f, -1.0f };
885 for (g = 0; g < ics->num_windows; g++)
886 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
888 for (g = 0; g < ics->num_window_groups; g++) {
889 for (i = 0; i < ics->max_sfb; i++, idx++) {
890 const int cur_band_type = band_type[idx];
891 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
892 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
894 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
895 for (group = 0; group < ics->group_len[g]; group++) {
896 memset(coef + group * 128 + offsets[i], 0, (offsets[i + 1] - offsets[i]) * sizeof(float));
898 } else if (cur_band_type == NOISE_BT) {
899 for (group = 0; group < ics->group_len[g]; group++) {
902 float *cf = coef + group * 128 + offsets[i];
903 int len = offsets[i+1] - offsets[i];
905 for (k = 0; k < len; k++) {
906 ac->random_state = lcg_random(ac->random_state);
907 cf[k] = ac->random_state;
910 band_energy = ac->dsp.scalarproduct_float(cf, cf, len);
911 scale = sf[idx] / sqrtf(band_energy);
912 ac->dsp.vector_fmul_scalar(cf, cf, scale, len);
915 for (group = 0; group < ics->group_len[g]; group++) {
917 const float **vqp = vq;
918 float *cf = coef + (group << 7) + offsets[i];
919 int len = offsets[i + 1] - offsets[i];
921 for (k = offsets[i]; k < offsets[i + 1]; k += dim) {
922 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
923 const int coef_tmp_idx = (group << 7) + k;
926 if (index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
927 av_log(ac->avccontext, AV_LOG_ERROR,
928 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
929 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
932 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
934 if (is_cb_unsigned) {
936 coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
938 coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
941 coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
943 coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
945 if (cur_band_type == ESC_BT) {
946 for (j = 0; j < 2; j++) {
947 if (vq_ptr[j] == 64.0f) {
949 /* The total length of escape_sequence must be < 22 bits according
950 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
951 while (get_bits1(gb) && n < 13) n++;
953 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
956 n = (1 << n) + get_bits(gb, n);
957 coef[coef_tmp_idx + j] *= cbrt_tab[n];
959 coef[coef_tmp_idx + j] *= vq_ptr[j];
965 if (is_cb_unsigned && cur_band_type != ESC_BT) {
966 ac->dsp.vector_fmul_sv_scalar[dim>>2](
967 cf, cf, vq, sf[idx], len);
968 } else if (cur_band_type == ESC_BT) {
969 ac->dsp.vector_fmul_scalar(cf, cf, sf[idx], len);
970 } else { /* !is_cb_unsigned */
971 ac->dsp.sv_fmul_scalar[dim>>2](cf, vq, sf[idx], len);
976 coef += ics->group_len[g] << 7;
981 for (i = 0; i < pulse->num_pulse; i++) {
982 float co = coef_base[ pulse->pos[i] ];
983 while (offsets[idx + 1] <= pulse->pos[i])
985 if (band_type[idx] != NOISE_BT && sf[idx]) {
986 float ico = -pulse->amp[i];
989 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
991 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
998 static av_always_inline float flt16_round(float pf)
1002 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1006 static av_always_inline float flt16_even(float pf)
1010 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1014 static av_always_inline float flt16_trunc(float pf)
1018 pun.i &= 0xFFFF0000U;
1022 static void predict(AACContext *ac, PredictorState *ps, float *coef,
1025 const float a = 0.953125; // 61.0 / 64
1026 const float alpha = 0.90625; // 29.0 / 32
1031 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1032 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1034 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1036 *coef += pv * ac->sf_scale;
1038 e0 = *coef / ac->sf_scale;
1039 e1 = e0 - k1 * ps->r0;
1041 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1042 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1043 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1044 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1046 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1047 ps->r0 = flt16_trunc(a * e0);
1051 * Apply AAC-Main style frequency domain prediction.
1053 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1057 if (!sce->ics.predictor_initialized) {
1058 reset_all_predictors(sce->predictor_state);
1059 sce->ics.predictor_initialized = 1;
1062 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1063 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1064 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1065 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1066 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1069 if (sce->ics.predictor_reset_group)
1070 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1072 reset_all_predictors(sce->predictor_state);
1076 * Decode an individual_channel_stream payload; reference: table 4.44.
1078 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1079 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1081 * @return Returns error status. 0 - OK, !0 - error
1083 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1084 GetBitContext *gb, int common_window, int scale_flag)
1087 TemporalNoiseShaping *tns = &sce->tns;
1088 IndividualChannelStream *ics = &sce->ics;
1089 float *out = sce->coeffs;
1090 int global_gain, pulse_present = 0;
1092 /* This assignment is to silence a GCC warning about the variable being used
1093 * uninitialized when in fact it always is.
1095 pulse.num_pulse = 0;
1097 global_gain = get_bits(gb, 8);
1099 if (!common_window && !scale_flag) {
1100 if (decode_ics_info(ac, ics, gb, 0) < 0)
1104 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1106 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1111 if ((pulse_present = get_bits1(gb))) {
1112 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1113 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1116 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1117 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1121 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1123 if (get_bits1(gb)) {
1124 av_log_missing_feature(ac->avccontext, "SSR", 1);
1129 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1132 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1133 apply_prediction(ac, sce);
1139 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1141 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1143 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1144 float *ch0 = cpe->ch[0].coeffs;
1145 float *ch1 = cpe->ch[1].coeffs;
1146 int g, i, group, idx = 0;
1147 const uint16_t *offsets = ics->swb_offset;
1148 for (g = 0; g < ics->num_window_groups; g++) {
1149 for (i = 0; i < ics->max_sfb; i++, idx++) {
1150 if (cpe->ms_mask[idx] &&
1151 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1152 for (group = 0; group < ics->group_len[g]; group++) {
1153 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1154 ch1 + group * 128 + offsets[i],
1155 offsets[i+1] - offsets[i]);
1159 ch0 += ics->group_len[g] * 128;
1160 ch1 += ics->group_len[g] * 128;
1165 * intensity stereo decoding; reference: 4.6.8.2.3
1167 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1168 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1169 * [3] reserved for scalable AAC
1171 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1173 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1174 SingleChannelElement *sce1 = &cpe->ch[1];
1175 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1176 const uint16_t *offsets = ics->swb_offset;
1177 int g, group, i, k, idx = 0;
1180 for (g = 0; g < ics->num_window_groups; g++) {
1181 for (i = 0; i < ics->max_sfb;) {
1182 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1183 const int bt_run_end = sce1->band_type_run_end[idx];
1184 for (; i < bt_run_end; i++, idx++) {
1185 c = -1 + 2 * (sce1->band_type[idx] - 14);
1187 c *= 1 - 2 * cpe->ms_mask[idx];
1188 scale = c * sce1->sf[idx];
1189 for (group = 0; group < ics->group_len[g]; group++)
1190 for (k = offsets[i]; k < offsets[i + 1]; k++)
1191 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1194 int bt_run_end = sce1->band_type_run_end[idx];
1195 idx += bt_run_end - i;
1199 coef0 += ics->group_len[g] * 128;
1200 coef1 += ics->group_len[g] * 128;
1205 * Decode a channel_pair_element; reference: table 4.4.
1207 * @param elem_id Identifies the instance of a syntax element.
1209 * @return Returns error status. 0 - OK, !0 - error
1211 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1213 int i, ret, common_window, ms_present = 0;
1215 common_window = get_bits1(gb);
1216 if (common_window) {
1217 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1219 i = cpe->ch[1].ics.use_kb_window[0];
1220 cpe->ch[1].ics = cpe->ch[0].ics;
1221 cpe->ch[1].ics.use_kb_window[1] = i;
1222 ms_present = get_bits(gb, 2);
1223 if (ms_present == 3) {
1224 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1226 } else if (ms_present)
1227 decode_mid_side_stereo(cpe, gb, ms_present);
1229 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1231 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1234 if (common_window) {
1236 apply_mid_side_stereo(ac, cpe);
1237 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1238 apply_prediction(ac, &cpe->ch[0]);
1239 apply_prediction(ac, &cpe->ch[1]);
1243 apply_intensity_stereo(cpe, ms_present);
1248 * Decode coupling_channel_element; reference: table 4.8.
1250 * @param elem_id Identifies the instance of a syntax element.
1252 * @return Returns error status. 0 - OK, !0 - error
1254 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1260 SingleChannelElement *sce = &che->ch[0];
1261 ChannelCoupling *coup = &che->coup;
1263 coup->coupling_point = 2 * get_bits1(gb);
1264 coup->num_coupled = get_bits(gb, 3);
1265 for (c = 0; c <= coup->num_coupled; c++) {
1267 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1268 coup->id_select[c] = get_bits(gb, 4);
1269 if (coup->type[c] == TYPE_CPE) {
1270 coup->ch_select[c] = get_bits(gb, 2);
1271 if (coup->ch_select[c] == 3)
1274 coup->ch_select[c] = 2;
1276 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1278 sign = get_bits(gb, 1);
1279 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1281 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1284 for (c = 0; c < num_gain; c++) {
1288 float gain_cache = 1.;
1290 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1291 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1292 gain_cache = pow(scale, -gain);
1294 if (coup->coupling_point == AFTER_IMDCT) {
1295 coup->gain[c][0] = gain_cache;
1297 for (g = 0; g < sce->ics.num_window_groups; g++) {
1298 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1299 if (sce->band_type[idx] != ZERO_BT) {
1301 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1309 gain_cache = pow(scale, -t) * s;
1312 coup->gain[c][idx] = gain_cache;
1322 * Decode Spectral Band Replication extension data; reference: table 4.55.
1324 * @param crc flag indicating the presence of CRC checksum
1325 * @param cnt length of TYPE_FIL syntactic element in bytes
1327 * @return Returns number of bytes consumed from the TYPE_FIL element.
1329 static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1332 // TODO : sbr_extension implementation
1333 av_log_missing_feature(ac->avccontext, "SBR", 0);
1334 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1339 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1341 * @return Returns number of bytes consumed.
1343 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1347 int num_excl_chan = 0;
1350 for (i = 0; i < 7; i++)
1351 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1352 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1354 return num_excl_chan / 7;
1358 * Decode dynamic range information; reference: table 4.52.
1360 * @param cnt length of TYPE_FIL syntactic element in bytes
1362 * @return Returns number of bytes consumed.
1364 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1365 GetBitContext *gb, int cnt)
1368 int drc_num_bands = 1;
1371 /* pce_tag_present? */
1372 if (get_bits1(gb)) {
1373 che_drc->pce_instance_tag = get_bits(gb, 4);
1374 skip_bits(gb, 4); // tag_reserved_bits
1378 /* excluded_chns_present? */
1379 if (get_bits1(gb)) {
1380 n += decode_drc_channel_exclusions(che_drc, gb);
1383 /* drc_bands_present? */
1384 if (get_bits1(gb)) {
1385 che_drc->band_incr = get_bits(gb, 4);
1386 che_drc->interpolation_scheme = get_bits(gb, 4);
1388 drc_num_bands += che_drc->band_incr;
1389 for (i = 0; i < drc_num_bands; i++) {
1390 che_drc->band_top[i] = get_bits(gb, 8);
1395 /* prog_ref_level_present? */
1396 if (get_bits1(gb)) {
1397 che_drc->prog_ref_level = get_bits(gb, 7);
1398 skip_bits1(gb); // prog_ref_level_reserved_bits
1402 for (i = 0; i < drc_num_bands; i++) {
1403 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1404 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1412 * Decode extension data (incomplete); reference: table 4.51.
1414 * @param cnt length of TYPE_FIL syntactic element in bytes
1416 * @return Returns number of bytes consumed
1418 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1422 switch (get_bits(gb, 4)) { // extension type
1423 case EXT_SBR_DATA_CRC:
1426 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1428 case EXT_DYNAMIC_RANGE:
1429 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1433 case EXT_DATA_ELEMENT:
1435 skip_bits_long(gb, 8 * cnt - 4);
1442 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1444 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1445 * @param coef spectral coefficients
1447 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1448 IndividualChannelStream *ics, int decode)
1450 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1452 int bottom, top, order, start, end, size, inc;
1453 float lpc[TNS_MAX_ORDER];
1455 for (w = 0; w < ics->num_windows; w++) {
1456 bottom = ics->num_swb;
1457 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1459 bottom = FFMAX(0, top - tns->length[w][filt]);
1460 order = tns->order[w][filt];
1465 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1467 start = ics->swb_offset[FFMIN(bottom, mmm)];
1468 end = ics->swb_offset[FFMIN( top, mmm)];
1469 if ((size = end - start) <= 0)
1471 if (tns->direction[w][filt]) {
1480 for (m = 0; m < size; m++, start += inc)
1481 for (i = 1; i <= FFMIN(m, order); i++)
1482 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1488 * Conduct IMDCT and windowing.
1490 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1492 IndividualChannelStream *ics = &sce->ics;
1493 float *in = sce->coeffs;
1494 float *out = sce->ret;
1495 float *saved = sce->saved;
1496 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1497 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1498 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1499 float *buf = ac->buf_mdct;
1500 float *temp = ac->temp;
1504 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1505 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1506 av_log(ac->avccontext, AV_LOG_WARNING,
1507 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1508 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1509 for (i = 0; i < 1024; i += 128)
1510 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1512 ff_imdct_half(&ac->mdct, buf, in);
1514 /* window overlapping
1515 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1516 * and long to short transitions are considered to be short to short
1517 * transitions. This leaves just two cases (long to long and short to short)
1518 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1520 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1521 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1522 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1524 for (i = 0; i < 448; i++)
1525 out[i] = saved[i] + ac->add_bias;
1527 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1528 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1529 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1530 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1531 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1532 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1533 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1535 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1536 for (i = 576; i < 1024; i++)
1537 out[i] = buf[i-512] + ac->add_bias;
1542 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1543 for (i = 0; i < 64; i++)
1544 saved[i] = temp[64 + i] - ac->add_bias;
1545 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1546 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1547 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1548 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1549 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1550 memcpy( saved, buf + 512, 448 * sizeof(float));
1551 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1552 } else { // LONG_STOP or ONLY_LONG
1553 memcpy( saved, buf + 512, 512 * sizeof(float));
1558 * Apply dependent channel coupling (applied before IMDCT).
1560 * @param index index into coupling gain array
1562 static void apply_dependent_coupling(AACContext *ac,
1563 SingleChannelElement *target,
1564 ChannelElement *cce, int index)
1566 IndividualChannelStream *ics = &cce->ch[0].ics;
1567 const uint16_t *offsets = ics->swb_offset;
1568 float *dest = target->coeffs;
1569 const float *src = cce->ch[0].coeffs;
1570 int g, i, group, k, idx = 0;
1571 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1572 av_log(ac->avccontext, AV_LOG_ERROR,
1573 "Dependent coupling is not supported together with LTP\n");
1576 for (g = 0; g < ics->num_window_groups; g++) {
1577 for (i = 0; i < ics->max_sfb; i++, idx++) {
1578 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1579 const float gain = cce->coup.gain[index][idx];
1580 for (group = 0; group < ics->group_len[g]; group++) {
1581 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1583 dest[group * 128 + k] += gain * src[group * 128 + k];
1588 dest += ics->group_len[g] * 128;
1589 src += ics->group_len[g] * 128;
1594 * Apply independent channel coupling (applied after IMDCT).
1596 * @param index index into coupling gain array
1598 static void apply_independent_coupling(AACContext *ac,
1599 SingleChannelElement *target,
1600 ChannelElement *cce, int index)
1603 const float gain = cce->coup.gain[index][0];
1604 const float bias = ac->add_bias;
1605 const float *src = cce->ch[0].ret;
1606 float *dest = target->ret;
1608 for (i = 0; i < 1024; i++)
1609 dest[i] += gain * (src[i] - bias);
1613 * channel coupling transformation interface
1615 * @param index index into coupling gain array
1616 * @param apply_coupling_method pointer to (in)dependent coupling function
1618 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1619 enum RawDataBlockType type, int elem_id,
1620 enum CouplingPoint coupling_point,
1621 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1625 for (i = 0; i < MAX_ELEM_ID; i++) {
1626 ChannelElement *cce = ac->che[TYPE_CCE][i];
1629 if (cce && cce->coup.coupling_point == coupling_point) {
1630 ChannelCoupling *coup = &cce->coup;
1632 for (c = 0; c <= coup->num_coupled; c++) {
1633 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1634 if (coup->ch_select[c] != 1) {
1635 apply_coupling_method(ac, &cc->ch[0], cce, index);
1636 if (coup->ch_select[c] != 0)
1639 if (coup->ch_select[c] != 2)
1640 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1642 index += 1 + (coup->ch_select[c] == 3);
1649 * Convert spectral data to float samples, applying all supported tools as appropriate.
1651 static void spectral_to_sample(AACContext *ac)
1654 for (type = 3; type >= 0; type--) {
1655 for (i = 0; i < MAX_ELEM_ID; i++) {
1656 ChannelElement *che = ac->che[type][i];
1658 if (type <= TYPE_CPE)
1659 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1660 if (che->ch[0].tns.present)
1661 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1662 if (che->ch[1].tns.present)
1663 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1664 if (type <= TYPE_CPE)
1665 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1666 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1667 imdct_and_windowing(ac, &che->ch[0]);
1668 if (type == TYPE_CPE)
1669 imdct_and_windowing(ac, &che->ch[1]);
1670 if (type <= TYPE_CCE)
1671 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1677 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1680 AACADTSHeaderInfo hdr_info;
1682 size = ff_aac_parse_header(gb, &hdr_info);
1684 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1685 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1686 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1687 ac->m4ac.chan_config = hdr_info.chan_config;
1688 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1690 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1692 } else if (ac->output_configured != OC_LOCKED) {
1693 ac->output_configured = OC_NONE;
1695 if (ac->output_configured != OC_LOCKED)
1697 ac->m4ac.sample_rate = hdr_info.sample_rate;
1698 ac->m4ac.sampling_index = hdr_info.sampling_index;
1699 ac->m4ac.object_type = hdr_info.object_type;
1700 if (!ac->avccontext->sample_rate)
1701 ac->avccontext->sample_rate = hdr_info.sample_rate;
1702 if (hdr_info.num_aac_frames == 1) {
1703 if (!hdr_info.crc_absent)
1706 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1713 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1714 int *data_size, AVPacket *avpkt)
1716 const uint8_t *buf = avpkt->data;
1717 int buf_size = avpkt->size;
1718 AACContext *ac = avccontext->priv_data;
1719 ChannelElement *che = NULL;
1721 enum RawDataBlockType elem_type;
1722 int err, elem_id, data_size_tmp;
1724 init_get_bits(&gb, buf, buf_size * 8);
1726 if (show_bits(&gb, 12) == 0xfff) {
1727 if (parse_adts_frame_header(ac, &gb) < 0) {
1728 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1731 if (ac->m4ac.sampling_index > 12) {
1732 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1738 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1739 elem_id = get_bits(&gb, 4);
1741 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1742 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1746 switch (elem_type) {
1749 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1753 err = decode_cpe(ac, &gb, che);
1757 err = decode_cce(ac, &gb, che);
1761 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1765 skip_data_stream_element(&gb);
1770 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1771 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1772 if ((err = decode_pce(ac, new_che_pos, &gb)))
1774 if (ac->output_configured > OC_TRIAL_PCE)
1775 av_log(avccontext, AV_LOG_ERROR,
1776 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1778 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1784 elem_id += get_bits(&gb, 8) - 1;
1786 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1787 err = 0; /* FIXME */
1791 err = -1; /* should not happen, but keeps compiler happy */
1799 spectral_to_sample(ac);
1801 if (!ac->is_saved) {
1807 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1808 if (*data_size < data_size_tmp) {
1809 av_log(avccontext, AV_LOG_ERROR,
1810 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1811 *data_size, data_size_tmp);
1814 *data_size = data_size_tmp;
1816 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1818 if (ac->output_configured)
1819 ac->output_configured = OC_LOCKED;
1824 static av_cold int aac_decode_close(AVCodecContext *avccontext)
1826 AACContext *ac = avccontext->priv_data;
1829 for (i = 0; i < MAX_ELEM_ID; i++) {
1830 for (type = 0; type < 4; type++)
1831 av_freep(&ac->che[type][i]);
1834 ff_mdct_end(&ac->mdct);
1835 ff_mdct_end(&ac->mdct_small);
1839 AVCodec aac_decoder = {
1848 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1849 .sample_fmts = (const enum SampleFormat[]) {
1850 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
1852 .channel_layouts = aac_channel_layout,