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 uint32_t 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++) {
571 * Skip data_stream_element; reference: table 4.10.
573 static void skip_data_stream_element(GetBitContext *gb)
575 int byte_align = get_bits1(gb);
576 int count = get_bits(gb, 8);
578 count += get_bits(gb, 8);
581 skip_bits_long(gb, 8 * count);
584 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
589 ics->predictor_reset_group = get_bits(gb, 5);
590 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
591 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
595 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
596 ics->prediction_used[sfb] = get_bits1(gb);
602 * Decode Individual Channel Stream info; reference: table 4.6.
604 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
606 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
607 GetBitContext *gb, int common_window)
610 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
611 memset(ics, 0, sizeof(IndividualChannelStream));
614 ics->window_sequence[1] = ics->window_sequence[0];
615 ics->window_sequence[0] = get_bits(gb, 2);
616 ics->use_kb_window[1] = ics->use_kb_window[0];
617 ics->use_kb_window[0] = get_bits1(gb);
618 ics->num_window_groups = 1;
619 ics->group_len[0] = 1;
620 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
622 ics->max_sfb = get_bits(gb, 4);
623 for (i = 0; i < 7; i++) {
625 ics->group_len[ics->num_window_groups - 1]++;
627 ics->num_window_groups++;
628 ics->group_len[ics->num_window_groups - 1] = 1;
631 ics->num_windows = 8;
632 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
633 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
634 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
635 ics->predictor_present = 0;
637 ics->max_sfb = get_bits(gb, 6);
638 ics->num_windows = 1;
639 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
640 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
641 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
642 ics->predictor_present = get_bits1(gb);
643 ics->predictor_reset_group = 0;
644 if (ics->predictor_present) {
645 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
646 if (decode_prediction(ac, ics, gb)) {
647 memset(ics, 0, sizeof(IndividualChannelStream));
650 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
651 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
652 memset(ics, 0, sizeof(IndividualChannelStream));
655 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
656 memset(ics, 0, sizeof(IndividualChannelStream));
662 if (ics->max_sfb > ics->num_swb) {
663 av_log(ac->avccontext, AV_LOG_ERROR,
664 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
665 ics->max_sfb, ics->num_swb);
666 memset(ics, 0, sizeof(IndividualChannelStream));
674 * Decode band types (section_data payload); reference: table 4.46.
676 * @param band_type array of the used band type
677 * @param band_type_run_end array of the last scalefactor band of a band type run
679 * @return Returns error status. 0 - OK, !0 - error
681 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
682 int band_type_run_end[120], GetBitContext *gb,
683 IndividualChannelStream *ics)
686 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
687 for (g = 0; g < ics->num_window_groups; g++) {
689 while (k < ics->max_sfb) {
690 uint8_t sect_end = k;
692 int sect_band_type = get_bits(gb, 4);
693 if (sect_band_type == 12) {
694 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
697 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
698 sect_end += sect_len_incr;
699 sect_end += sect_len_incr;
700 if (sect_end > ics->max_sfb) {
701 av_log(ac->avccontext, AV_LOG_ERROR,
702 "Number of bands (%d) exceeds limit (%d).\n",
703 sect_end, ics->max_sfb);
706 for (; k < sect_end; k++) {
707 band_type [idx] = sect_band_type;
708 band_type_run_end[idx++] = sect_end;
716 * Decode scalefactors; reference: table 4.47.
718 * @param global_gain first scalefactor value as scalefactors are differentially coded
719 * @param band_type array of the used band type
720 * @param band_type_run_end array of the last scalefactor band of a band type run
721 * @param sf array of scalefactors or intensity stereo positions
723 * @return Returns error status. 0 - OK, !0 - error
725 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
726 unsigned int global_gain,
727 IndividualChannelStream *ics,
728 enum BandType band_type[120],
729 int band_type_run_end[120])
731 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
733 int offset[3] = { global_gain, global_gain - 90, 100 };
735 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
736 for (g = 0; g < ics->num_window_groups; g++) {
737 for (i = 0; i < ics->max_sfb;) {
738 int run_end = band_type_run_end[idx];
739 if (band_type[idx] == ZERO_BT) {
740 for (; i < run_end; i++, idx++)
742 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
743 for (; i < run_end; i++, idx++) {
744 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
745 if (offset[2] > 255U) {
746 av_log(ac->avccontext, AV_LOG_ERROR,
747 "%s (%d) out of range.\n", sf_str[2], offset[2]);
750 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
752 } else if (band_type[idx] == NOISE_BT) {
753 for (; i < run_end; i++, idx++) {
754 if (noise_flag-- > 0)
755 offset[1] += get_bits(gb, 9) - 256;
757 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
758 if (offset[1] > 255U) {
759 av_log(ac->avccontext, AV_LOG_ERROR,
760 "%s (%d) out of range.\n", sf_str[1], offset[1]);
763 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
766 for (; i < run_end; i++, idx++) {
767 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
768 if (offset[0] > 255U) {
769 av_log(ac->avccontext, AV_LOG_ERROR,
770 "%s (%d) out of range.\n", sf_str[0], offset[0]);
773 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
782 * Decode pulse data; reference: table 4.7.
784 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
785 const uint16_t *swb_offset, int num_swb)
788 pulse->num_pulse = get_bits(gb, 2) + 1;
789 pulse_swb = get_bits(gb, 6);
790 if (pulse_swb >= num_swb)
792 pulse->pos[0] = swb_offset[pulse_swb];
793 pulse->pos[0] += get_bits(gb, 5);
794 if (pulse->pos[0] > 1023)
796 pulse->amp[0] = get_bits(gb, 4);
797 for (i = 1; i < pulse->num_pulse; i++) {
798 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
799 if (pulse->pos[i] > 1023)
801 pulse->amp[i] = get_bits(gb, 4);
807 * Decode Temporal Noise Shaping data; reference: table 4.48.
809 * @return Returns error status. 0 - OK, !0 - error
811 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
812 GetBitContext *gb, const IndividualChannelStream *ics)
814 int w, filt, i, coef_len, coef_res, coef_compress;
815 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
816 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
817 for (w = 0; w < ics->num_windows; w++) {
818 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
819 coef_res = get_bits1(gb);
821 for (filt = 0; filt < tns->n_filt[w]; filt++) {
823 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
825 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
826 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
827 tns->order[w][filt], tns_max_order);
828 tns->order[w][filt] = 0;
831 if (tns->order[w][filt]) {
832 tns->direction[w][filt] = get_bits1(gb);
833 coef_compress = get_bits1(gb);
834 coef_len = coef_res + 3 - coef_compress;
835 tmp2_idx = 2 * coef_compress + coef_res;
837 for (i = 0; i < tns->order[w][filt]; i++)
838 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
847 * Decode Mid/Side data; reference: table 4.54.
849 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
850 * [1] mask is decoded from bitstream; [2] mask is all 1s;
851 * [3] reserved for scalable AAC
853 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
857 if (ms_present == 1) {
858 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
859 cpe->ms_mask[idx] = get_bits1(gb);
860 } else if (ms_present == 2) {
861 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
865 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
869 *dst++ = v[idx & 15] * s;
870 *dst++ = v[idx>>4 & 15] * s;
874 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
878 *dst++ = v[idx & 3] * s;
879 *dst++ = v[idx>>2 & 3] * s;
880 *dst++ = v[idx>>4 & 3] * s;
881 *dst++ = v[idx>>6 & 3] * s;
885 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
886 unsigned sign, const float *scale)
888 union float754 s0, s1;
890 s0.f = s1.f = *scale;
891 s0.i ^= sign >> 1 << 31;
894 *dst++ = v[idx & 15] * s0.f;
895 *dst++ = v[idx>>4 & 15] * s1.f;
900 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
901 unsigned sign, const float *scale)
903 unsigned nz = idx >> 12;
904 union float754 s = { .f = *scale };
907 t.i = s.i ^ (sign & 1<<31);
908 *dst++ = v[idx & 3] * t.f;
910 sign <<= nz & 1; nz >>= 1;
911 t.i = s.i ^ (sign & 1<<31);
912 *dst++ = v[idx>>2 & 3] * t.f;
914 sign <<= nz & 1; nz >>= 1;
915 t.i = s.i ^ (sign & 1<<31);
916 *dst++ = v[idx>>4 & 3] * t.f;
918 sign <<= nz & 1; nz >>= 1;
919 t.i = s.i ^ (sign & 1<<31);
920 *dst++ = v[idx>>6 & 3] * t.f;
926 * Decode spectral data; reference: table 4.50.
927 * Dequantize and scale spectral data; reference: 4.6.3.3.
929 * @param coef array of dequantized, scaled spectral data
930 * @param sf array of scalefactors or intensity stereo positions
931 * @param pulse_present set if pulses are present
932 * @param pulse pointer to pulse data struct
933 * @param band_type array of the used band type
935 * @return Returns error status. 0 - OK, !0 - error
937 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
938 GetBitContext *gb, const float sf[120],
939 int pulse_present, const Pulse *pulse,
940 const IndividualChannelStream *ics,
941 enum BandType band_type[120])
943 int i, k, g, idx = 0;
944 const int c = 1024 / ics->num_windows;
945 const uint16_t *offsets = ics->swb_offset;
946 float *coef_base = coef;
949 for (g = 0; g < ics->num_windows; g++)
950 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
952 for (g = 0; g < ics->num_window_groups; g++) {
953 unsigned g_len = ics->group_len[g];
955 for (i = 0; i < ics->max_sfb; i++, idx++) {
956 const unsigned cbt_m1 = band_type[idx] - 1;
957 float *cfo = coef + offsets[i];
958 int off_len = offsets[i + 1] - offsets[i];
961 if (cbt_m1 >= INTENSITY_BT2 - 1) {
962 for (group = 0; group < g_len; group++, cfo+=128) {
963 memset(cfo, 0, off_len * sizeof(float));
965 } else if (cbt_m1 == NOISE_BT - 1) {
966 for (group = 0; group < g_len; group++, cfo+=128) {
970 for (k = 0; k < off_len; k++) {
971 ac->random_state = lcg_random(ac->random_state);
972 cfo[k] = ac->random_state;
975 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
976 scale = sf[idx] / sqrtf(band_energy);
977 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
980 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
981 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
982 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
983 const int cb_size = ff_aac_spectral_sizes[cbt_m1];
985 for (group = 0; group < g_len; group++, cfo+=128) {
987 uint32_t *icf = (uint32_t *) cf;
990 switch (cbt_m1 >> 1) {
993 const int index = get_vlc2(gb, vlc_tab, 6, 3);
996 if (index >= cb_size) {
998 goto err_cb_overflow;
1001 cb_idx = cb_vector_idx[index];
1002 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1007 const int index = get_vlc2(gb, vlc_tab, 6, 3);
1012 if (index >= cb_size) {
1014 goto err_cb_overflow;
1017 cb_idx = cb_vector_idx[index];
1018 nnz = cb_idx >> 8 & 15;
1019 bits = get_bits(gb, nnz) << (32-nnz);
1020 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1025 const int index = get_vlc2(gb, vlc_tab, 6, 3);
1028 if (index >= cb_size) {
1030 goto err_cb_overflow;
1033 cb_idx = cb_vector_idx[index];
1034 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1040 const int index = get_vlc2(gb, vlc_tab, 6, 3);
1045 if (index >= cb_size) {
1047 goto err_cb_overflow;
1050 cb_idx = cb_vector_idx[index];
1051 nnz = cb_idx >> 8 & 15;
1052 sign = get_bits(gb, nnz) << (cb_idx >> 12);
1053 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1058 const int index = get_vlc2(gb, vlc_tab, 6, 3);
1070 if (index >= cb_size) {
1072 goto err_cb_overflow;
1075 cb_idx = cb_vector_idx[index];
1078 bits = get_bits(gb, nnz) << (32-nnz);
1080 for (j = 0; j < 2; j++) {
1083 /* The total length of escape_sequence must be < 22 bits according
1084 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1085 while (get_bits1(gb) && n < 13) n++;
1087 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1090 n = (1 << n) + get_bits(gb, n);
1091 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1094 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1095 *icf++ = (bits & 1<<31) | v;
1102 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1110 if (pulse_present) {
1112 for (i = 0; i < pulse->num_pulse; i++) {
1113 float co = coef_base[ pulse->pos[i] ];
1114 while (offsets[idx + 1] <= pulse->pos[i])
1116 if (band_type[idx] != NOISE_BT && sf[idx]) {
1117 float ico = -pulse->amp[i];
1120 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1122 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1129 av_log(ac->avccontext, AV_LOG_ERROR,
1130 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1131 band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1135 static av_always_inline float flt16_round(float pf)
1139 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1143 static av_always_inline float flt16_even(float pf)
1147 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1151 static av_always_inline float flt16_trunc(float pf)
1155 pun.i &= 0xFFFF0000U;
1159 static void predict(AACContext *ac, PredictorState *ps, float *coef,
1162 const float a = 0.953125; // 61.0 / 64
1163 const float alpha = 0.90625; // 29.0 / 32
1168 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1169 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1171 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1173 *coef += pv * ac->sf_scale;
1175 e0 = *coef / ac->sf_scale;
1176 e1 = e0 - k1 * ps->r0;
1178 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1179 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1180 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1181 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1183 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1184 ps->r0 = flt16_trunc(a * e0);
1188 * Apply AAC-Main style frequency domain prediction.
1190 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1194 if (!sce->ics.predictor_initialized) {
1195 reset_all_predictors(sce->predictor_state);
1196 sce->ics.predictor_initialized = 1;
1199 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1200 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1201 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1202 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1203 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1206 if (sce->ics.predictor_reset_group)
1207 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1209 reset_all_predictors(sce->predictor_state);
1213 * Decode an individual_channel_stream payload; reference: table 4.44.
1215 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1216 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1218 * @return Returns error status. 0 - OK, !0 - error
1220 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1221 GetBitContext *gb, int common_window, int scale_flag)
1224 TemporalNoiseShaping *tns = &sce->tns;
1225 IndividualChannelStream *ics = &sce->ics;
1226 float *out = sce->coeffs;
1227 int global_gain, pulse_present = 0;
1229 /* This assignment is to silence a GCC warning about the variable being used
1230 * uninitialized when in fact it always is.
1232 pulse.num_pulse = 0;
1234 global_gain = get_bits(gb, 8);
1236 if (!common_window && !scale_flag) {
1237 if (decode_ics_info(ac, ics, gb, 0) < 0)
1241 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1243 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1248 if ((pulse_present = get_bits1(gb))) {
1249 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1250 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1253 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1254 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1258 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1260 if (get_bits1(gb)) {
1261 av_log_missing_feature(ac->avccontext, "SSR", 1);
1266 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1269 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1270 apply_prediction(ac, sce);
1276 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1278 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1280 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1281 float *ch0 = cpe->ch[0].coeffs;
1282 float *ch1 = cpe->ch[1].coeffs;
1283 int g, i, group, idx = 0;
1284 const uint16_t *offsets = ics->swb_offset;
1285 for (g = 0; g < ics->num_window_groups; g++) {
1286 for (i = 0; i < ics->max_sfb; i++, idx++) {
1287 if (cpe->ms_mask[idx] &&
1288 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1289 for (group = 0; group < ics->group_len[g]; group++) {
1290 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1291 ch1 + group * 128 + offsets[i],
1292 offsets[i+1] - offsets[i]);
1296 ch0 += ics->group_len[g] * 128;
1297 ch1 += ics->group_len[g] * 128;
1302 * intensity stereo decoding; reference: 4.6.8.2.3
1304 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1305 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1306 * [3] reserved for scalable AAC
1308 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1310 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1311 SingleChannelElement *sce1 = &cpe->ch[1];
1312 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1313 const uint16_t *offsets = ics->swb_offset;
1314 int g, group, i, k, idx = 0;
1317 for (g = 0; g < ics->num_window_groups; g++) {
1318 for (i = 0; i < ics->max_sfb;) {
1319 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1320 const int bt_run_end = sce1->band_type_run_end[idx];
1321 for (; i < bt_run_end; i++, idx++) {
1322 c = -1 + 2 * (sce1->band_type[idx] - 14);
1324 c *= 1 - 2 * cpe->ms_mask[idx];
1325 scale = c * sce1->sf[idx];
1326 for (group = 0; group < ics->group_len[g]; group++)
1327 for (k = offsets[i]; k < offsets[i + 1]; k++)
1328 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1331 int bt_run_end = sce1->band_type_run_end[idx];
1332 idx += bt_run_end - i;
1336 coef0 += ics->group_len[g] * 128;
1337 coef1 += ics->group_len[g] * 128;
1342 * Decode a channel_pair_element; reference: table 4.4.
1344 * @param elem_id Identifies the instance of a syntax element.
1346 * @return Returns error status. 0 - OK, !0 - error
1348 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1350 int i, ret, common_window, ms_present = 0;
1352 common_window = get_bits1(gb);
1353 if (common_window) {
1354 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1356 i = cpe->ch[1].ics.use_kb_window[0];
1357 cpe->ch[1].ics = cpe->ch[0].ics;
1358 cpe->ch[1].ics.use_kb_window[1] = i;
1359 ms_present = get_bits(gb, 2);
1360 if (ms_present == 3) {
1361 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1363 } else if (ms_present)
1364 decode_mid_side_stereo(cpe, gb, ms_present);
1366 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1368 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1371 if (common_window) {
1373 apply_mid_side_stereo(ac, cpe);
1374 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1375 apply_prediction(ac, &cpe->ch[0]);
1376 apply_prediction(ac, &cpe->ch[1]);
1380 apply_intensity_stereo(cpe, ms_present);
1385 * Decode coupling_channel_element; reference: table 4.8.
1387 * @param elem_id Identifies the instance of a syntax element.
1389 * @return Returns error status. 0 - OK, !0 - error
1391 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1397 SingleChannelElement *sce = &che->ch[0];
1398 ChannelCoupling *coup = &che->coup;
1400 coup->coupling_point = 2 * get_bits1(gb);
1401 coup->num_coupled = get_bits(gb, 3);
1402 for (c = 0; c <= coup->num_coupled; c++) {
1404 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1405 coup->id_select[c] = get_bits(gb, 4);
1406 if (coup->type[c] == TYPE_CPE) {
1407 coup->ch_select[c] = get_bits(gb, 2);
1408 if (coup->ch_select[c] == 3)
1411 coup->ch_select[c] = 2;
1413 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1415 sign = get_bits(gb, 1);
1416 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1418 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1421 for (c = 0; c < num_gain; c++) {
1425 float gain_cache = 1.;
1427 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1428 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1429 gain_cache = pow(scale, -gain);
1431 if (coup->coupling_point == AFTER_IMDCT) {
1432 coup->gain[c][0] = gain_cache;
1434 for (g = 0; g < sce->ics.num_window_groups; g++) {
1435 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1436 if (sce->band_type[idx] != ZERO_BT) {
1438 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1446 gain_cache = pow(scale, -t) * s;
1449 coup->gain[c][idx] = gain_cache;
1459 * Decode Spectral Band Replication extension data; reference: table 4.55.
1461 * @param crc flag indicating the presence of CRC checksum
1462 * @param cnt length of TYPE_FIL syntactic element in bytes
1464 * @return Returns number of bytes consumed from the TYPE_FIL element.
1466 static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1469 // TODO : sbr_extension implementation
1470 av_log_missing_feature(ac->avccontext, "SBR", 0);
1471 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1476 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1478 * @return Returns number of bytes consumed.
1480 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1484 int num_excl_chan = 0;
1487 for (i = 0; i < 7; i++)
1488 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1489 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1491 return num_excl_chan / 7;
1495 * Decode dynamic range information; reference: table 4.52.
1497 * @param cnt length of TYPE_FIL syntactic element in bytes
1499 * @return Returns number of bytes consumed.
1501 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1502 GetBitContext *gb, int cnt)
1505 int drc_num_bands = 1;
1508 /* pce_tag_present? */
1509 if (get_bits1(gb)) {
1510 che_drc->pce_instance_tag = get_bits(gb, 4);
1511 skip_bits(gb, 4); // tag_reserved_bits
1515 /* excluded_chns_present? */
1516 if (get_bits1(gb)) {
1517 n += decode_drc_channel_exclusions(che_drc, gb);
1520 /* drc_bands_present? */
1521 if (get_bits1(gb)) {
1522 che_drc->band_incr = get_bits(gb, 4);
1523 che_drc->interpolation_scheme = get_bits(gb, 4);
1525 drc_num_bands += che_drc->band_incr;
1526 for (i = 0; i < drc_num_bands; i++) {
1527 che_drc->band_top[i] = get_bits(gb, 8);
1532 /* prog_ref_level_present? */
1533 if (get_bits1(gb)) {
1534 che_drc->prog_ref_level = get_bits(gb, 7);
1535 skip_bits1(gb); // prog_ref_level_reserved_bits
1539 for (i = 0; i < drc_num_bands; i++) {
1540 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1541 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1549 * Decode extension data (incomplete); reference: table 4.51.
1551 * @param cnt length of TYPE_FIL syntactic element in bytes
1553 * @return Returns number of bytes consumed
1555 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1559 switch (get_bits(gb, 4)) { // extension type
1560 case EXT_SBR_DATA_CRC:
1563 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1565 case EXT_DYNAMIC_RANGE:
1566 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1570 case EXT_DATA_ELEMENT:
1572 skip_bits_long(gb, 8 * cnt - 4);
1579 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1581 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1582 * @param coef spectral coefficients
1584 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1585 IndividualChannelStream *ics, int decode)
1587 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1589 int bottom, top, order, start, end, size, inc;
1590 float lpc[TNS_MAX_ORDER];
1592 for (w = 0; w < ics->num_windows; w++) {
1593 bottom = ics->num_swb;
1594 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1596 bottom = FFMAX(0, top - tns->length[w][filt]);
1597 order = tns->order[w][filt];
1602 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1604 start = ics->swb_offset[FFMIN(bottom, mmm)];
1605 end = ics->swb_offset[FFMIN( top, mmm)];
1606 if ((size = end - start) <= 0)
1608 if (tns->direction[w][filt]) {
1617 for (m = 0; m < size; m++, start += inc)
1618 for (i = 1; i <= FFMIN(m, order); i++)
1619 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1625 * Conduct IMDCT and windowing.
1627 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1629 IndividualChannelStream *ics = &sce->ics;
1630 float *in = sce->coeffs;
1631 float *out = sce->ret;
1632 float *saved = sce->saved;
1633 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1634 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1635 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1636 float *buf = ac->buf_mdct;
1637 float *temp = ac->temp;
1641 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1642 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1643 av_log(ac->avccontext, AV_LOG_WARNING,
1644 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1645 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1646 for (i = 0; i < 1024; i += 128)
1647 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1649 ff_imdct_half(&ac->mdct, buf, in);
1651 /* window overlapping
1652 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1653 * and long to short transitions are considered to be short to short
1654 * transitions. This leaves just two cases (long to long and short to short)
1655 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1657 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1658 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1659 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1661 for (i = 0; i < 448; i++)
1662 out[i] = saved[i] + ac->add_bias;
1664 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1665 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1666 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1667 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1668 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1669 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1670 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1672 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1673 for (i = 576; i < 1024; i++)
1674 out[i] = buf[i-512] + ac->add_bias;
1679 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1680 for (i = 0; i < 64; i++)
1681 saved[i] = temp[64 + i] - ac->add_bias;
1682 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1683 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1684 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1685 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1686 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1687 memcpy( saved, buf + 512, 448 * sizeof(float));
1688 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1689 } else { // LONG_STOP or ONLY_LONG
1690 memcpy( saved, buf + 512, 512 * sizeof(float));
1695 * Apply dependent channel coupling (applied before IMDCT).
1697 * @param index index into coupling gain array
1699 static void apply_dependent_coupling(AACContext *ac,
1700 SingleChannelElement *target,
1701 ChannelElement *cce, int index)
1703 IndividualChannelStream *ics = &cce->ch[0].ics;
1704 const uint16_t *offsets = ics->swb_offset;
1705 float *dest = target->coeffs;
1706 const float *src = cce->ch[0].coeffs;
1707 int g, i, group, k, idx = 0;
1708 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1709 av_log(ac->avccontext, AV_LOG_ERROR,
1710 "Dependent coupling is not supported together with LTP\n");
1713 for (g = 0; g < ics->num_window_groups; g++) {
1714 for (i = 0; i < ics->max_sfb; i++, idx++) {
1715 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1716 const float gain = cce->coup.gain[index][idx];
1717 for (group = 0; group < ics->group_len[g]; group++) {
1718 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1720 dest[group * 128 + k] += gain * src[group * 128 + k];
1725 dest += ics->group_len[g] * 128;
1726 src += ics->group_len[g] * 128;
1731 * Apply independent channel coupling (applied after IMDCT).
1733 * @param index index into coupling gain array
1735 static void apply_independent_coupling(AACContext *ac,
1736 SingleChannelElement *target,
1737 ChannelElement *cce, int index)
1740 const float gain = cce->coup.gain[index][0];
1741 const float bias = ac->add_bias;
1742 const float *src = cce->ch[0].ret;
1743 float *dest = target->ret;
1745 for (i = 0; i < 1024; i++)
1746 dest[i] += gain * (src[i] - bias);
1750 * channel coupling transformation interface
1752 * @param index index into coupling gain array
1753 * @param apply_coupling_method pointer to (in)dependent coupling function
1755 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1756 enum RawDataBlockType type, int elem_id,
1757 enum CouplingPoint coupling_point,
1758 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1762 for (i = 0; i < MAX_ELEM_ID; i++) {
1763 ChannelElement *cce = ac->che[TYPE_CCE][i];
1766 if (cce && cce->coup.coupling_point == coupling_point) {
1767 ChannelCoupling *coup = &cce->coup;
1769 for (c = 0; c <= coup->num_coupled; c++) {
1770 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1771 if (coup->ch_select[c] != 1) {
1772 apply_coupling_method(ac, &cc->ch[0], cce, index);
1773 if (coup->ch_select[c] != 0)
1776 if (coup->ch_select[c] != 2)
1777 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1779 index += 1 + (coup->ch_select[c] == 3);
1786 * Convert spectral data to float samples, applying all supported tools as appropriate.
1788 static void spectral_to_sample(AACContext *ac)
1791 for (type = 3; type >= 0; type--) {
1792 for (i = 0; i < MAX_ELEM_ID; i++) {
1793 ChannelElement *che = ac->che[type][i];
1795 if (type <= TYPE_CPE)
1796 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1797 if (che->ch[0].tns.present)
1798 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1799 if (che->ch[1].tns.present)
1800 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1801 if (type <= TYPE_CPE)
1802 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1803 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1804 imdct_and_windowing(ac, &che->ch[0]);
1805 if (type == TYPE_CPE)
1806 imdct_and_windowing(ac, &che->ch[1]);
1807 if (type <= TYPE_CCE)
1808 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1814 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1817 AACADTSHeaderInfo hdr_info;
1819 size = ff_aac_parse_header(gb, &hdr_info);
1821 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1822 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1823 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1824 ac->m4ac.chan_config = hdr_info.chan_config;
1825 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1827 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1829 } else if (ac->output_configured != OC_LOCKED) {
1830 ac->output_configured = OC_NONE;
1832 if (ac->output_configured != OC_LOCKED)
1834 ac->m4ac.sample_rate = hdr_info.sample_rate;
1835 ac->m4ac.sampling_index = hdr_info.sampling_index;
1836 ac->m4ac.object_type = hdr_info.object_type;
1837 if (!ac->avccontext->sample_rate)
1838 ac->avccontext->sample_rate = hdr_info.sample_rate;
1839 if (hdr_info.num_aac_frames == 1) {
1840 if (!hdr_info.crc_absent)
1843 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1850 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1851 int *data_size, AVPacket *avpkt)
1853 const uint8_t *buf = avpkt->data;
1854 int buf_size = avpkt->size;
1855 AACContext *ac = avccontext->priv_data;
1856 ChannelElement *che = NULL;
1858 enum RawDataBlockType elem_type;
1859 int err, elem_id, data_size_tmp;
1861 init_get_bits(&gb, buf, buf_size * 8);
1863 if (show_bits(&gb, 12) == 0xfff) {
1864 if (parse_adts_frame_header(ac, &gb) < 0) {
1865 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1868 if (ac->m4ac.sampling_index > 12) {
1869 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1875 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1876 elem_id = get_bits(&gb, 4);
1878 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1879 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1883 switch (elem_type) {
1886 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1890 err = decode_cpe(ac, &gb, che);
1894 err = decode_cce(ac, &gb, che);
1898 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1902 skip_data_stream_element(&gb);
1907 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1908 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1909 if ((err = decode_pce(ac, new_che_pos, &gb)))
1911 if (ac->output_configured > OC_TRIAL_PCE)
1912 av_log(avccontext, AV_LOG_ERROR,
1913 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1915 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1921 elem_id += get_bits(&gb, 8) - 1;
1923 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1924 err = 0; /* FIXME */
1928 err = -1; /* should not happen, but keeps compiler happy */
1936 spectral_to_sample(ac);
1938 if (!ac->is_saved) {
1944 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1945 if (*data_size < data_size_tmp) {
1946 av_log(avccontext, AV_LOG_ERROR,
1947 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1948 *data_size, data_size_tmp);
1951 *data_size = data_size_tmp;
1953 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1955 if (ac->output_configured)
1956 ac->output_configured = OC_LOCKED;
1961 static av_cold int aac_decode_close(AVCodecContext *avccontext)
1963 AACContext *ac = avccontext->priv_data;
1966 for (i = 0; i < MAX_ELEM_ID; i++) {
1967 for (type = 0; type < 4; type++)
1968 av_freep(&ac->che[type][i]);
1971 ff_mdct_end(&ac->mdct);
1972 ff_mdct_end(&ac->mdct_small);
1976 AVCodec aac_decoder = {
1985 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1986 .sample_fmts = (const enum SampleFormat[]) {
1987 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
1989 .channel_layouts = aac_channel_layout,