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"
105 static VLC vlc_scalefactors;
106 static VLC vlc_spectral[11];
108 static uint32_t cbrt_tab[1<<13];
110 static const char overread_err[] = "Input buffer exhausted before END element found\n";
112 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
114 if (ac->tag_che_map[type][elem_id]) {
115 return ac->tag_che_map[type][elem_id];
117 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
120 switch (ac->m4ac.chan_config) {
122 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
124 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
127 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
128 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
129 encountered such a stream, transfer the LFE[0] element to SCE[1] */
130 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
132 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
135 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
137 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
140 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
142 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
146 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
148 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
149 } else if (ac->m4ac.chan_config == 2) {
153 if (!ac->tags_mapped && type == TYPE_SCE) {
155 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
163 * Check for the channel element in the current channel position configuration.
164 * If it exists, make sure the appropriate element is allocated and map the
165 * channel order to match the internal FFmpeg channel layout.
167 * @param che_pos current channel position configuration
168 * @param type channel element type
169 * @param id channel element id
170 * @param channels count of the number of channels in the configuration
172 * @return Returns error status. 0 - OK, !0 - error
174 static av_cold int che_configure(AACContext *ac,
175 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
179 if (che_pos[type][id]) {
180 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
181 return AVERROR(ENOMEM);
182 if (type != TYPE_CCE) {
183 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
184 if (type == TYPE_CPE) {
185 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
189 av_freep(&ac->che[type][id]);
194 * Configure output channel order based on the current program configuration element.
196 * @param che_pos current channel position configuration
197 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
199 * @return Returns error status. 0 - OK, !0 - error
201 static av_cold int output_configure(AACContext *ac,
202 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
203 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
204 int channel_config, enum OCStatus oc_type)
206 AVCodecContext *avctx = ac->avccontext;
207 int i, type, channels = 0, ret;
209 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
211 if (channel_config) {
212 for (i = 0; i < tags_per_config[channel_config]; i++) {
213 if ((ret = che_configure(ac, che_pos,
214 aac_channel_layout_map[channel_config - 1][i][0],
215 aac_channel_layout_map[channel_config - 1][i][1],
220 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
223 avctx->channel_layout = aac_channel_layout[channel_config - 1];
225 /* Allocate or free elements depending on if they are in the
226 * current program configuration.
228 * Set up default 1:1 output mapping.
230 * For a 5.1 stream the output order will be:
231 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
234 for (i = 0; i < MAX_ELEM_ID; i++) {
235 for (type = 0; type < 4; type++) {
236 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
241 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
242 ac->tags_mapped = 4 * MAX_ELEM_ID;
244 avctx->channel_layout = 0;
247 avctx->channels = channels;
249 ac->output_configured = oc_type;
255 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
257 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
258 * @param sce_map mono (Single Channel Element) map
259 * @param type speaker type/position for these channels
261 static void decode_channel_map(enum ChannelPosition *cpe_map,
262 enum ChannelPosition *sce_map,
263 enum ChannelPosition type,
264 GetBitContext *gb, int n)
267 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
268 map[get_bits(gb, 4)] = type;
273 * Decode program configuration element; reference: table 4.2.
275 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
277 * @return Returns error status. 0 - OK, !0 - error
279 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
282 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
285 skip_bits(gb, 2); // object_type
287 sampling_index = get_bits(gb, 4);
288 if (ac->m4ac.sampling_index != sampling_index)
289 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");
291 num_front = get_bits(gb, 4);
292 num_side = get_bits(gb, 4);
293 num_back = get_bits(gb, 4);
294 num_lfe = get_bits(gb, 2);
295 num_assoc_data = get_bits(gb, 3);
296 num_cc = get_bits(gb, 4);
299 skip_bits(gb, 4); // mono_mixdown_tag
301 skip_bits(gb, 4); // stereo_mixdown_tag
304 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
306 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
307 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
308 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
309 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
311 skip_bits_long(gb, 4 * num_assoc_data);
313 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
317 /* comment field, first byte is length */
318 comment_len = get_bits(gb, 8) * 8;
319 if (get_bits_left(gb) < comment_len) {
320 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
323 skip_bits_long(gb, comment_len);
328 * Set up channel positions based on a default channel configuration
329 * as specified in table 1.17.
331 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
333 * @return Returns error status. 0 - OK, !0 - error
335 static av_cold int set_default_channel_config(AACContext *ac,
336 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
339 if (channel_config < 1 || channel_config > 7) {
340 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
345 /* default channel configurations:
347 * 1ch : front center (mono)
348 * 2ch : L + R (stereo)
349 * 3ch : front center + L + R
350 * 4ch : front center + L + R + back center
351 * 5ch : front center + L + R + back stereo
352 * 6ch : front center + L + R + back stereo + LFE
353 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
356 if (channel_config != 2)
357 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
358 if (channel_config > 1)
359 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
360 if (channel_config == 4)
361 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
362 if (channel_config > 4)
363 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
364 = AAC_CHANNEL_BACK; // back stereo
365 if (channel_config > 5)
366 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
367 if (channel_config == 7)
368 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
374 * Decode GA "General Audio" specific configuration; reference: table 4.1.
376 * @return Returns error status. 0 - OK, !0 - error
378 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
381 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
382 int extension_flag, ret;
384 if (get_bits1(gb)) { // frameLengthFlag
385 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
389 if (get_bits1(gb)) // dependsOnCoreCoder
390 skip_bits(gb, 14); // coreCoderDelay
391 extension_flag = get_bits1(gb);
393 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
394 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
395 skip_bits(gb, 3); // layerNr
397 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
398 if (channel_config == 0) {
399 skip_bits(gb, 4); // element_instance_tag
400 if ((ret = decode_pce(ac, new_che_pos, gb)))
403 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
406 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
409 if (extension_flag) {
410 switch (ac->m4ac.object_type) {
412 skip_bits(gb, 5); // numOfSubFrame
413 skip_bits(gb, 11); // layer_length
417 case AOT_ER_AAC_SCALABLE:
419 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
420 * aacScalefactorDataResilienceFlag
421 * aacSpectralDataResilienceFlag
425 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
431 * Decode audio specific configuration; reference: table 1.13.
433 * @param data pointer to AVCodecContext extradata
434 * @param data_size size of AVCCodecContext extradata
436 * @return Returns error status. 0 - OK, !0 - error
438 static int decode_audio_specific_config(AACContext *ac, void *data,
444 init_get_bits(&gb, data, data_size * 8);
446 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
448 if (ac->m4ac.sampling_index > 12) {
449 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
453 skip_bits_long(&gb, i);
455 switch (ac->m4ac.object_type) {
458 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
462 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
463 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
470 * linear congruential pseudorandom number generator
472 * @param previous_val pointer to the current state of the generator
474 * @return Returns a 32-bit pseudorandom integer
476 static av_always_inline int lcg_random(int previous_val)
478 return previous_val * 1664525 + 1013904223;
481 static av_always_inline void reset_predict_state(PredictorState *ps)
491 static void reset_all_predictors(PredictorState *ps)
494 for (i = 0; i < MAX_PREDICTORS; i++)
495 reset_predict_state(&ps[i]);
498 static void reset_predictor_group(PredictorState *ps, int group_num)
501 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
502 reset_predict_state(&ps[i]);
505 static av_cold int aac_decode_init(AVCodecContext *avccontext)
507 AACContext *ac = avccontext->priv_data;
510 ac->avccontext = avccontext;
512 if (avccontext->extradata_size > 0) {
513 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
515 avccontext->sample_rate = ac->m4ac.sample_rate;
516 } else if (avccontext->channels > 0) {
517 ac->m4ac.sample_rate = avccontext->sample_rate;
520 avccontext->sample_fmt = SAMPLE_FMT_S16;
521 avccontext->frame_size = 1024;
523 AAC_INIT_VLC_STATIC( 0, 304);
524 AAC_INIT_VLC_STATIC( 1, 270);
525 AAC_INIT_VLC_STATIC( 2, 550);
526 AAC_INIT_VLC_STATIC( 3, 300);
527 AAC_INIT_VLC_STATIC( 4, 328);
528 AAC_INIT_VLC_STATIC( 5, 294);
529 AAC_INIT_VLC_STATIC( 6, 306);
530 AAC_INIT_VLC_STATIC( 7, 268);
531 AAC_INIT_VLC_STATIC( 8, 510);
532 AAC_INIT_VLC_STATIC( 9, 366);
533 AAC_INIT_VLC_STATIC(10, 462);
535 dsputil_init(&ac->dsp, avccontext);
537 ac->random_state = 0x1f2e3d4c;
539 // -1024 - Compensate wrong IMDCT method.
540 // 32768 - Required to scale values to the correct range for the bias method
541 // for float to int16 conversion.
543 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
544 ac->add_bias = 385.0f;
545 ac->sf_scale = 1. / (-1024. * 32768.);
549 ac->sf_scale = 1. / -1024.;
553 #if !CONFIG_HARDCODED_TABLES
554 for (i = 0; i < 428; i++)
555 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
556 #endif /* CONFIG_HARDCODED_TABLES */
558 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
559 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
560 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
563 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
564 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
565 // window initialization
566 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
567 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
568 ff_init_ff_sine_windows(10);
569 ff_init_ff_sine_windows( 7);
571 if (!cbrt_tab[(1<<13) - 1]) {
572 for (i = 0; i < 1<<13; i++) {
583 * Skip data_stream_element; reference: table 4.10.
585 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
587 int byte_align = get_bits1(gb);
588 int count = get_bits(gb, 8);
590 count += get_bits(gb, 8);
594 if (get_bits_left(gb) < 8 * count) {
595 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
598 skip_bits_long(gb, 8 * count);
602 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
607 ics->predictor_reset_group = get_bits(gb, 5);
608 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
609 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
613 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
614 ics->prediction_used[sfb] = get_bits1(gb);
620 * Decode Individual Channel Stream info; reference: table 4.6.
622 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
624 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
625 GetBitContext *gb, int common_window)
628 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
629 memset(ics, 0, sizeof(IndividualChannelStream));
632 ics->window_sequence[1] = ics->window_sequence[0];
633 ics->window_sequence[0] = get_bits(gb, 2);
634 ics->use_kb_window[1] = ics->use_kb_window[0];
635 ics->use_kb_window[0] = get_bits1(gb);
636 ics->num_window_groups = 1;
637 ics->group_len[0] = 1;
638 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
640 ics->max_sfb = get_bits(gb, 4);
641 for (i = 0; i < 7; i++) {
643 ics->group_len[ics->num_window_groups - 1]++;
645 ics->num_window_groups++;
646 ics->group_len[ics->num_window_groups - 1] = 1;
649 ics->num_windows = 8;
650 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
651 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
652 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
653 ics->predictor_present = 0;
655 ics->max_sfb = get_bits(gb, 6);
656 ics->num_windows = 1;
657 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
658 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
659 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
660 ics->predictor_present = get_bits1(gb);
661 ics->predictor_reset_group = 0;
662 if (ics->predictor_present) {
663 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
664 if (decode_prediction(ac, ics, gb)) {
665 memset(ics, 0, sizeof(IndividualChannelStream));
668 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
669 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
670 memset(ics, 0, sizeof(IndividualChannelStream));
673 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
674 memset(ics, 0, sizeof(IndividualChannelStream));
680 if (ics->max_sfb > ics->num_swb) {
681 av_log(ac->avccontext, AV_LOG_ERROR,
682 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
683 ics->max_sfb, ics->num_swb);
684 memset(ics, 0, sizeof(IndividualChannelStream));
692 * Decode band types (section_data payload); reference: table 4.46.
694 * @param band_type array of the used band type
695 * @param band_type_run_end array of the last scalefactor band of a band type run
697 * @return Returns error status. 0 - OK, !0 - error
699 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
700 int band_type_run_end[120], GetBitContext *gb,
701 IndividualChannelStream *ics)
704 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
705 for (g = 0; g < ics->num_window_groups; g++) {
707 while (k < ics->max_sfb) {
708 uint8_t sect_end = k;
710 int sect_band_type = get_bits(gb, 4);
711 if (sect_band_type == 12) {
712 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
715 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
716 sect_end += sect_len_incr;
717 sect_end += sect_len_incr;
718 if (get_bits_left(gb) < 0) {
719 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
722 if (sect_end > ics->max_sfb) {
723 av_log(ac->avccontext, AV_LOG_ERROR,
724 "Number of bands (%d) exceeds limit (%d).\n",
725 sect_end, ics->max_sfb);
728 for (; k < sect_end; k++) {
729 band_type [idx] = sect_band_type;
730 band_type_run_end[idx++] = sect_end;
738 * Decode scalefactors; reference: table 4.47.
740 * @param global_gain first scalefactor value as scalefactors are differentially coded
741 * @param band_type array of the used band type
742 * @param band_type_run_end array of the last scalefactor band of a band type run
743 * @param sf array of scalefactors or intensity stereo positions
745 * @return Returns error status. 0 - OK, !0 - error
747 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
748 unsigned int global_gain,
749 IndividualChannelStream *ics,
750 enum BandType band_type[120],
751 int band_type_run_end[120])
753 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
755 int offset[3] = { global_gain, global_gain - 90, 100 };
757 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
758 for (g = 0; g < ics->num_window_groups; g++) {
759 for (i = 0; i < ics->max_sfb;) {
760 int run_end = band_type_run_end[idx];
761 if (band_type[idx] == ZERO_BT) {
762 for (; i < run_end; i++, idx++)
764 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
765 for (; i < run_end; i++, idx++) {
766 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
767 if (offset[2] > 255U) {
768 av_log(ac->avccontext, AV_LOG_ERROR,
769 "%s (%d) out of range.\n", sf_str[2], offset[2]);
772 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
774 } else if (band_type[idx] == NOISE_BT) {
775 for (; i < run_end; i++, idx++) {
776 if (noise_flag-- > 0)
777 offset[1] += get_bits(gb, 9) - 256;
779 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
780 if (offset[1] > 255U) {
781 av_log(ac->avccontext, AV_LOG_ERROR,
782 "%s (%d) out of range.\n", sf_str[1], offset[1]);
785 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
788 for (; i < run_end; i++, idx++) {
789 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
790 if (offset[0] > 255U) {
791 av_log(ac->avccontext, AV_LOG_ERROR,
792 "%s (%d) out of range.\n", sf_str[0], offset[0]);
795 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
804 * Decode pulse data; reference: table 4.7.
806 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
807 const uint16_t *swb_offset, int num_swb)
810 pulse->num_pulse = get_bits(gb, 2) + 1;
811 pulse_swb = get_bits(gb, 6);
812 if (pulse_swb >= num_swb)
814 pulse->pos[0] = swb_offset[pulse_swb];
815 pulse->pos[0] += get_bits(gb, 5);
816 if (pulse->pos[0] > 1023)
818 pulse->amp[0] = get_bits(gb, 4);
819 for (i = 1; i < pulse->num_pulse; i++) {
820 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
821 if (pulse->pos[i] > 1023)
823 pulse->amp[i] = get_bits(gb, 4);
829 * Decode Temporal Noise Shaping data; reference: table 4.48.
831 * @return Returns error status. 0 - OK, !0 - error
833 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
834 GetBitContext *gb, const IndividualChannelStream *ics)
836 int w, filt, i, coef_len, coef_res, coef_compress;
837 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
838 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
839 for (w = 0; w < ics->num_windows; w++) {
840 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
841 coef_res = get_bits1(gb);
843 for (filt = 0; filt < tns->n_filt[w]; filt++) {
845 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
847 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
848 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
849 tns->order[w][filt], tns_max_order);
850 tns->order[w][filt] = 0;
853 if (tns->order[w][filt]) {
854 tns->direction[w][filt] = get_bits1(gb);
855 coef_compress = get_bits1(gb);
856 coef_len = coef_res + 3 - coef_compress;
857 tmp2_idx = 2 * coef_compress + coef_res;
859 for (i = 0; i < tns->order[w][filt]; i++)
860 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
869 * Decode Mid/Side data; reference: table 4.54.
871 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
872 * [1] mask is decoded from bitstream; [2] mask is all 1s;
873 * [3] reserved for scalable AAC
875 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
879 if (ms_present == 1) {
880 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
881 cpe->ms_mask[idx] = get_bits1(gb);
882 } else if (ms_present == 2) {
883 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
888 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
892 *dst++ = v[idx & 15] * s;
893 *dst++ = v[idx>>4 & 15] * s;
899 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
903 *dst++ = v[idx & 3] * s;
904 *dst++ = v[idx>>2 & 3] * s;
905 *dst++ = v[idx>>4 & 3] * s;
906 *dst++ = v[idx>>6 & 3] * s;
912 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
913 unsigned sign, const float *scale)
915 union float754 s0, s1;
917 s0.f = s1.f = *scale;
918 s0.i ^= sign >> 1 << 31;
921 *dst++ = v[idx & 15] * s0.f;
922 *dst++ = v[idx>>4 & 15] * s1.f;
929 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
930 unsigned sign, const float *scale)
932 unsigned nz = idx >> 12;
933 union float754 s = { .f = *scale };
936 t.i = s.i ^ (sign & 1<<31);
937 *dst++ = v[idx & 3] * t.f;
939 sign <<= nz & 1; nz >>= 1;
940 t.i = s.i ^ (sign & 1<<31);
941 *dst++ = v[idx>>2 & 3] * t.f;
943 sign <<= nz & 1; nz >>= 1;
944 t.i = s.i ^ (sign & 1<<31);
945 *dst++ = v[idx>>4 & 3] * t.f;
947 sign <<= nz & 1; nz >>= 1;
948 t.i = s.i ^ (sign & 1<<31);
949 *dst++ = v[idx>>6 & 3] * t.f;
956 * Decode spectral data; reference: table 4.50.
957 * Dequantize and scale spectral data; reference: 4.6.3.3.
959 * @param coef array of dequantized, scaled spectral data
960 * @param sf array of scalefactors or intensity stereo positions
961 * @param pulse_present set if pulses are present
962 * @param pulse pointer to pulse data struct
963 * @param band_type array of the used band type
965 * @return Returns error status. 0 - OK, !0 - error
967 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
968 GetBitContext *gb, const float sf[120],
969 int pulse_present, const Pulse *pulse,
970 const IndividualChannelStream *ics,
971 enum BandType band_type[120])
973 int i, k, g, idx = 0;
974 const int c = 1024 / ics->num_windows;
975 const uint16_t *offsets = ics->swb_offset;
976 float *coef_base = coef;
979 for (g = 0; g < ics->num_windows; g++)
980 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
982 for (g = 0; g < ics->num_window_groups; g++) {
983 unsigned g_len = ics->group_len[g];
985 for (i = 0; i < ics->max_sfb; i++, idx++) {
986 const unsigned cbt_m1 = band_type[idx] - 1;
987 float *cfo = coef + offsets[i];
988 int off_len = offsets[i + 1] - offsets[i];
991 if (cbt_m1 >= INTENSITY_BT2 - 1) {
992 for (group = 0; group < g_len; group++, cfo+=128) {
993 memset(cfo, 0, off_len * sizeof(float));
995 } else if (cbt_m1 == NOISE_BT - 1) {
996 for (group = 0; group < g_len; group++, cfo+=128) {
1000 for (k = 0; k < off_len; k++) {
1001 ac->random_state = lcg_random(ac->random_state);
1002 cfo[k] = ac->random_state;
1005 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1006 scale = sf[idx] / sqrtf(band_energy);
1007 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1010 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1011 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1012 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1013 const int cb_size = ff_aac_spectral_sizes[cbt_m1];
1014 OPEN_READER(re, gb);
1016 switch (cbt_m1 >> 1) {
1018 for (group = 0; group < g_len; group++, cfo+=128) {
1026 UPDATE_CACHE(re, gb);
1027 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1029 if (code >= cb_size) {
1031 goto err_cb_overflow;
1034 cb_idx = cb_vector_idx[code];
1035 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1041 for (group = 0; group < g_len; group++, cfo+=128) {
1051 UPDATE_CACHE(re, gb);
1052 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1054 if (code >= cb_size) {
1056 goto err_cb_overflow;
1059 #if MIN_CACHE_BITS < 20
1060 UPDATE_CACHE(re, gb);
1062 cb_idx = cb_vector_idx[code];
1063 nnz = cb_idx >> 8 & 15;
1064 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1065 LAST_SKIP_BITS(re, gb, nnz);
1066 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1072 for (group = 0; group < g_len; group++, cfo+=128) {
1080 UPDATE_CACHE(re, gb);
1081 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1083 if (code >= cb_size) {
1085 goto err_cb_overflow;
1088 cb_idx = cb_vector_idx[code];
1089 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1096 for (group = 0; group < g_len; group++, cfo+=128) {
1106 UPDATE_CACHE(re, gb);
1107 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1109 if (code >= cb_size) {
1111 goto err_cb_overflow;
1114 cb_idx = cb_vector_idx[code];
1115 nnz = cb_idx >> 8 & 15;
1116 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1117 LAST_SKIP_BITS(re, gb, nnz);
1118 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1124 for (group = 0; group < g_len; group++, cfo+=128) {
1126 uint32_t *icf = (uint32_t *) cf;
1136 UPDATE_CACHE(re, gb);
1137 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1145 if (code >= cb_size) {
1147 goto err_cb_overflow;
1150 cb_idx = cb_vector_idx[code];
1153 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1154 LAST_SKIP_BITS(re, gb, nnz);
1156 for (j = 0; j < 2; j++) {
1160 /* The total length of escape_sequence must be < 22 bits according
1161 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1162 UPDATE_CACHE(re, gb);
1163 b = GET_CACHE(re, gb);
1164 b = 31 - av_log2(~b);
1167 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1171 #if MIN_CACHE_BITS < 21
1172 LAST_SKIP_BITS(re, gb, b + 1);
1173 UPDATE_CACHE(re, gb);
1175 SKIP_BITS(re, gb, b + 1);
1178 n = (1 << b) + SHOW_UBITS(re, gb, b);
1179 LAST_SKIP_BITS(re, gb, b);
1180 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1183 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1184 *icf++ = (bits & 1<<31) | v;
1191 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1195 CLOSE_READER(re, gb);
1201 if (pulse_present) {
1203 for (i = 0; i < pulse->num_pulse; i++) {
1204 float co = coef_base[ pulse->pos[i] ];
1205 while (offsets[idx + 1] <= pulse->pos[i])
1207 if (band_type[idx] != NOISE_BT && sf[idx]) {
1208 float ico = -pulse->amp[i];
1211 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1213 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1220 av_log(ac->avccontext, AV_LOG_ERROR,
1221 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1222 band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1226 static av_always_inline float flt16_round(float pf)
1230 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1234 static av_always_inline float flt16_even(float pf)
1238 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1242 static av_always_inline float flt16_trunc(float pf)
1246 pun.i &= 0xFFFF0000U;
1250 static av_always_inline void predict(AACContext *ac, PredictorState *ps, float *coef,
1253 const float a = 0.953125; // 61.0 / 64
1254 const float alpha = 0.90625; // 29.0 / 32
1259 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1260 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1262 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1264 *coef += pv * ac->sf_scale;
1266 e0 = *coef / ac->sf_scale;
1267 e1 = e0 - k1 * ps->r0;
1269 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1270 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1271 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1272 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1274 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1275 ps->r0 = flt16_trunc(a * e0);
1279 * Apply AAC-Main style frequency domain prediction.
1281 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1285 if (!sce->ics.predictor_initialized) {
1286 reset_all_predictors(sce->predictor_state);
1287 sce->ics.predictor_initialized = 1;
1290 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1291 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1292 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1293 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1294 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1297 if (sce->ics.predictor_reset_group)
1298 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1300 reset_all_predictors(sce->predictor_state);
1304 * Decode an individual_channel_stream payload; reference: table 4.44.
1306 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1307 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1309 * @return Returns error status. 0 - OK, !0 - error
1311 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1312 GetBitContext *gb, int common_window, int scale_flag)
1315 TemporalNoiseShaping *tns = &sce->tns;
1316 IndividualChannelStream *ics = &sce->ics;
1317 float *out = sce->coeffs;
1318 int global_gain, pulse_present = 0;
1320 /* This assignment is to silence a GCC warning about the variable being used
1321 * uninitialized when in fact it always is.
1323 pulse.num_pulse = 0;
1325 global_gain = get_bits(gb, 8);
1327 if (!common_window && !scale_flag) {
1328 if (decode_ics_info(ac, ics, gb, 0) < 0)
1332 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1334 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1339 if ((pulse_present = get_bits1(gb))) {
1340 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1341 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1344 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1345 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1349 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1351 if (get_bits1(gb)) {
1352 av_log_missing_feature(ac->avccontext, "SSR", 1);
1357 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1360 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1361 apply_prediction(ac, sce);
1367 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1369 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1371 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1372 float *ch0 = cpe->ch[0].coeffs;
1373 float *ch1 = cpe->ch[1].coeffs;
1374 int g, i, group, idx = 0;
1375 const uint16_t *offsets = ics->swb_offset;
1376 for (g = 0; g < ics->num_window_groups; g++) {
1377 for (i = 0; i < ics->max_sfb; i++, idx++) {
1378 if (cpe->ms_mask[idx] &&
1379 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1380 for (group = 0; group < ics->group_len[g]; group++) {
1381 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1382 ch1 + group * 128 + offsets[i],
1383 offsets[i+1] - offsets[i]);
1387 ch0 += ics->group_len[g] * 128;
1388 ch1 += ics->group_len[g] * 128;
1393 * intensity stereo decoding; reference: 4.6.8.2.3
1395 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1396 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1397 * [3] reserved for scalable AAC
1399 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1401 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1402 SingleChannelElement *sce1 = &cpe->ch[1];
1403 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1404 const uint16_t *offsets = ics->swb_offset;
1405 int g, group, i, k, idx = 0;
1408 for (g = 0; g < ics->num_window_groups; g++) {
1409 for (i = 0; i < ics->max_sfb;) {
1410 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1411 const int bt_run_end = sce1->band_type_run_end[idx];
1412 for (; i < bt_run_end; i++, idx++) {
1413 c = -1 + 2 * (sce1->band_type[idx] - 14);
1415 c *= 1 - 2 * cpe->ms_mask[idx];
1416 scale = c * sce1->sf[idx];
1417 for (group = 0; group < ics->group_len[g]; group++)
1418 for (k = offsets[i]; k < offsets[i + 1]; k++)
1419 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1422 int bt_run_end = sce1->band_type_run_end[idx];
1423 idx += bt_run_end - i;
1427 coef0 += ics->group_len[g] * 128;
1428 coef1 += ics->group_len[g] * 128;
1433 * Decode a channel_pair_element; reference: table 4.4.
1435 * @param elem_id Identifies the instance of a syntax element.
1437 * @return Returns error status. 0 - OK, !0 - error
1439 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1441 int i, ret, common_window, ms_present = 0;
1443 common_window = get_bits1(gb);
1444 if (common_window) {
1445 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1447 i = cpe->ch[1].ics.use_kb_window[0];
1448 cpe->ch[1].ics = cpe->ch[0].ics;
1449 cpe->ch[1].ics.use_kb_window[1] = i;
1450 ms_present = get_bits(gb, 2);
1451 if (ms_present == 3) {
1452 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1454 } else if (ms_present)
1455 decode_mid_side_stereo(cpe, gb, ms_present);
1457 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1459 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1462 if (common_window) {
1464 apply_mid_side_stereo(ac, cpe);
1465 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1466 apply_prediction(ac, &cpe->ch[0]);
1467 apply_prediction(ac, &cpe->ch[1]);
1471 apply_intensity_stereo(cpe, ms_present);
1476 * Decode coupling_channel_element; reference: table 4.8.
1478 * @param elem_id Identifies the instance of a syntax element.
1480 * @return Returns error status. 0 - OK, !0 - error
1482 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1488 SingleChannelElement *sce = &che->ch[0];
1489 ChannelCoupling *coup = &che->coup;
1491 coup->coupling_point = 2 * get_bits1(gb);
1492 coup->num_coupled = get_bits(gb, 3);
1493 for (c = 0; c <= coup->num_coupled; c++) {
1495 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1496 coup->id_select[c] = get_bits(gb, 4);
1497 if (coup->type[c] == TYPE_CPE) {
1498 coup->ch_select[c] = get_bits(gb, 2);
1499 if (coup->ch_select[c] == 3)
1502 coup->ch_select[c] = 2;
1504 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1506 sign = get_bits(gb, 1);
1507 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1509 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1512 for (c = 0; c < num_gain; c++) {
1516 float gain_cache = 1.;
1518 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1519 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1520 gain_cache = pow(scale, -gain);
1522 if (coup->coupling_point == AFTER_IMDCT) {
1523 coup->gain[c][0] = gain_cache;
1525 for (g = 0; g < sce->ics.num_window_groups; g++) {
1526 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1527 if (sce->band_type[idx] != ZERO_BT) {
1529 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1537 gain_cache = pow(scale, -t) * s;
1540 coup->gain[c][idx] = gain_cache;
1550 * Decode Spectral Band Replication extension data; reference: table 4.55.
1552 * @param crc flag indicating the presence of CRC checksum
1553 * @param cnt length of TYPE_FIL syntactic element in bytes
1555 * @return Returns number of bytes consumed from the TYPE_FIL element.
1557 static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1560 // TODO : sbr_extension implementation
1561 av_log_missing_feature(ac->avccontext, "SBR", 0);
1562 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1567 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1569 * @return Returns number of bytes consumed.
1571 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1575 int num_excl_chan = 0;
1578 for (i = 0; i < 7; i++)
1579 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1580 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1582 return num_excl_chan / 7;
1586 * Decode dynamic range information; reference: table 4.52.
1588 * @param cnt length of TYPE_FIL syntactic element in bytes
1590 * @return Returns number of bytes consumed.
1592 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1593 GetBitContext *gb, int cnt)
1596 int drc_num_bands = 1;
1599 /* pce_tag_present? */
1600 if (get_bits1(gb)) {
1601 che_drc->pce_instance_tag = get_bits(gb, 4);
1602 skip_bits(gb, 4); // tag_reserved_bits
1606 /* excluded_chns_present? */
1607 if (get_bits1(gb)) {
1608 n += decode_drc_channel_exclusions(che_drc, gb);
1611 /* drc_bands_present? */
1612 if (get_bits1(gb)) {
1613 che_drc->band_incr = get_bits(gb, 4);
1614 che_drc->interpolation_scheme = get_bits(gb, 4);
1616 drc_num_bands += che_drc->band_incr;
1617 for (i = 0; i < drc_num_bands; i++) {
1618 che_drc->band_top[i] = get_bits(gb, 8);
1623 /* prog_ref_level_present? */
1624 if (get_bits1(gb)) {
1625 che_drc->prog_ref_level = get_bits(gb, 7);
1626 skip_bits1(gb); // prog_ref_level_reserved_bits
1630 for (i = 0; i < drc_num_bands; i++) {
1631 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1632 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1640 * Decode extension data (incomplete); reference: table 4.51.
1642 * @param cnt length of TYPE_FIL syntactic element in bytes
1644 * @return Returns number of bytes consumed
1646 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1650 switch (get_bits(gb, 4)) { // extension type
1651 case EXT_SBR_DATA_CRC:
1654 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1656 case EXT_DYNAMIC_RANGE:
1657 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1661 case EXT_DATA_ELEMENT:
1663 skip_bits_long(gb, 8 * cnt - 4);
1670 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1672 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1673 * @param coef spectral coefficients
1675 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1676 IndividualChannelStream *ics, int decode)
1678 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1680 int bottom, top, order, start, end, size, inc;
1681 float lpc[TNS_MAX_ORDER];
1683 for (w = 0; w < ics->num_windows; w++) {
1684 bottom = ics->num_swb;
1685 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1687 bottom = FFMAX(0, top - tns->length[w][filt]);
1688 order = tns->order[w][filt];
1693 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1695 start = ics->swb_offset[FFMIN(bottom, mmm)];
1696 end = ics->swb_offset[FFMIN( top, mmm)];
1697 if ((size = end - start) <= 0)
1699 if (tns->direction[w][filt]) {
1708 for (m = 0; m < size; m++, start += inc)
1709 for (i = 1; i <= FFMIN(m, order); i++)
1710 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1716 * Conduct IMDCT and windowing.
1718 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1720 IndividualChannelStream *ics = &sce->ics;
1721 float *in = sce->coeffs;
1722 float *out = sce->ret;
1723 float *saved = sce->saved;
1724 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1725 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1726 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1727 float *buf = ac->buf_mdct;
1728 float *temp = ac->temp;
1732 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1733 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1734 av_log(ac->avccontext, AV_LOG_WARNING,
1735 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1736 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1737 for (i = 0; i < 1024; i += 128)
1738 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1740 ff_imdct_half(&ac->mdct, buf, in);
1742 /* window overlapping
1743 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1744 * and long to short transitions are considered to be short to short
1745 * transitions. This leaves just two cases (long to long and short to short)
1746 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1748 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1749 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1750 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1752 for (i = 0; i < 448; i++)
1753 out[i] = saved[i] + ac->add_bias;
1755 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1756 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1757 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1758 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1759 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1760 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1761 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1763 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1764 for (i = 576; i < 1024; i++)
1765 out[i] = buf[i-512] + ac->add_bias;
1770 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1771 for (i = 0; i < 64; i++)
1772 saved[i] = temp[64 + i] - ac->add_bias;
1773 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1774 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1775 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1776 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1777 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1778 memcpy( saved, buf + 512, 448 * sizeof(float));
1779 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1780 } else { // LONG_STOP or ONLY_LONG
1781 memcpy( saved, buf + 512, 512 * sizeof(float));
1786 * Apply dependent channel coupling (applied before IMDCT).
1788 * @param index index into coupling gain array
1790 static void apply_dependent_coupling(AACContext *ac,
1791 SingleChannelElement *target,
1792 ChannelElement *cce, int index)
1794 IndividualChannelStream *ics = &cce->ch[0].ics;
1795 const uint16_t *offsets = ics->swb_offset;
1796 float *dest = target->coeffs;
1797 const float *src = cce->ch[0].coeffs;
1798 int g, i, group, k, idx = 0;
1799 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1800 av_log(ac->avccontext, AV_LOG_ERROR,
1801 "Dependent coupling is not supported together with LTP\n");
1804 for (g = 0; g < ics->num_window_groups; g++) {
1805 for (i = 0; i < ics->max_sfb; i++, idx++) {
1806 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1807 const float gain = cce->coup.gain[index][idx];
1808 for (group = 0; group < ics->group_len[g]; group++) {
1809 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1811 dest[group * 128 + k] += gain * src[group * 128 + k];
1816 dest += ics->group_len[g] * 128;
1817 src += ics->group_len[g] * 128;
1822 * Apply independent channel coupling (applied after IMDCT).
1824 * @param index index into coupling gain array
1826 static void apply_independent_coupling(AACContext *ac,
1827 SingleChannelElement *target,
1828 ChannelElement *cce, int index)
1831 const float gain = cce->coup.gain[index][0];
1832 const float bias = ac->add_bias;
1833 const float *src = cce->ch[0].ret;
1834 float *dest = target->ret;
1836 for (i = 0; i < 1024; i++)
1837 dest[i] += gain * (src[i] - bias);
1841 * channel coupling transformation interface
1843 * @param index index into coupling gain array
1844 * @param apply_coupling_method pointer to (in)dependent coupling function
1846 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1847 enum RawDataBlockType type, int elem_id,
1848 enum CouplingPoint coupling_point,
1849 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1853 for (i = 0; i < MAX_ELEM_ID; i++) {
1854 ChannelElement *cce = ac->che[TYPE_CCE][i];
1857 if (cce && cce->coup.coupling_point == coupling_point) {
1858 ChannelCoupling *coup = &cce->coup;
1860 for (c = 0; c <= coup->num_coupled; c++) {
1861 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1862 if (coup->ch_select[c] != 1) {
1863 apply_coupling_method(ac, &cc->ch[0], cce, index);
1864 if (coup->ch_select[c] != 0)
1867 if (coup->ch_select[c] != 2)
1868 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1870 index += 1 + (coup->ch_select[c] == 3);
1877 * Convert spectral data to float samples, applying all supported tools as appropriate.
1879 static void spectral_to_sample(AACContext *ac)
1882 for (type = 3; type >= 0; type--) {
1883 for (i = 0; i < MAX_ELEM_ID; i++) {
1884 ChannelElement *che = ac->che[type][i];
1886 if (type <= TYPE_CPE)
1887 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1888 if (che->ch[0].tns.present)
1889 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1890 if (che->ch[1].tns.present)
1891 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1892 if (type <= TYPE_CPE)
1893 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1894 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1895 imdct_and_windowing(ac, &che->ch[0]);
1896 if (type == TYPE_CPE)
1897 imdct_and_windowing(ac, &che->ch[1]);
1898 if (type <= TYPE_CCE)
1899 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1905 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1908 AACADTSHeaderInfo hdr_info;
1910 size = ff_aac_parse_header(gb, &hdr_info);
1912 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1913 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1914 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1915 ac->m4ac.chan_config = hdr_info.chan_config;
1916 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1918 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1920 } else if (ac->output_configured != OC_LOCKED) {
1921 ac->output_configured = OC_NONE;
1923 if (ac->output_configured != OC_LOCKED)
1925 ac->m4ac.sample_rate = hdr_info.sample_rate;
1926 ac->m4ac.sampling_index = hdr_info.sampling_index;
1927 ac->m4ac.object_type = hdr_info.object_type;
1928 if (!ac->avccontext->sample_rate)
1929 ac->avccontext->sample_rate = hdr_info.sample_rate;
1930 if (hdr_info.num_aac_frames == 1) {
1931 if (!hdr_info.crc_absent)
1934 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1941 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1942 int *data_size, AVPacket *avpkt)
1944 const uint8_t *buf = avpkt->data;
1945 int buf_size = avpkt->size;
1946 AACContext *ac = avccontext->priv_data;
1947 ChannelElement *che = NULL;
1949 enum RawDataBlockType elem_type;
1950 int err, elem_id, data_size_tmp;
1953 init_get_bits(&gb, buf, buf_size * 8);
1955 if (show_bits(&gb, 12) == 0xfff) {
1956 if (parse_adts_frame_header(ac, &gb) < 0) {
1957 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1960 if (ac->m4ac.sampling_index > 12) {
1961 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1967 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1968 elem_id = get_bits(&gb, 4);
1970 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1971 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1975 switch (elem_type) {
1978 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1982 err = decode_cpe(ac, &gb, che);
1986 err = decode_cce(ac, &gb, che);
1990 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1994 err = skip_data_stream_element(ac, &gb);
1998 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1999 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2000 if ((err = decode_pce(ac, new_che_pos, &gb)))
2002 if (ac->output_configured > OC_TRIAL_PCE)
2003 av_log(avccontext, AV_LOG_ERROR,
2004 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2006 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2012 elem_id += get_bits(&gb, 8) - 1;
2013 if (get_bits_left(&gb) < 8 * elem_id) {
2014 av_log(avccontext, AV_LOG_ERROR, overread_err);
2018 elem_id -= decode_extension_payload(ac, &gb, elem_id);
2019 err = 0; /* FIXME */
2023 err = -1; /* should not happen, but keeps compiler happy */
2030 if (get_bits_left(&gb) < 3) {
2031 av_log(avccontext, AV_LOG_ERROR, overread_err);
2036 spectral_to_sample(ac);
2038 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
2039 if (*data_size < data_size_tmp) {
2040 av_log(avccontext, AV_LOG_ERROR,
2041 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2042 *data_size, data_size_tmp);
2045 *data_size = data_size_tmp;
2047 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
2049 if (ac->output_configured)
2050 ac->output_configured = OC_LOCKED;
2052 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2053 return buf_size > buf_consumed ? buf_consumed : buf_size;
2056 static av_cold int aac_decode_close(AVCodecContext *avccontext)
2058 AACContext *ac = avccontext->priv_data;
2061 for (i = 0; i < MAX_ELEM_ID; i++) {
2062 for (type = 0; type < 4; type++)
2063 av_freep(&ac->che[type][i]);
2066 ff_mdct_end(&ac->mdct);
2067 ff_mdct_end(&ac->mdct_small);
2071 AVCodec aac_decoder = {
2080 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2081 .sample_fmts = (const enum SampleFormat[]) {
2082 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2084 .channel_layouts = aac_channel_layout,