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 ChannelElement *get_che(AACContext *ac, int type, int elem_id)
112 if (ac->tag_che_map[type][elem_id]) {
113 return ac->tag_che_map[type][elem_id];
115 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
118 switch (ac->m4ac.chan_config) {
120 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
122 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
125 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
126 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
127 encountered such a stream, transfer the LFE[0] element to SCE[1] */
128 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
130 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
133 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
135 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
138 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
140 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
144 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
146 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
147 } else if (ac->m4ac.chan_config == 2) {
151 if (!ac->tags_mapped && type == TYPE_SCE) {
153 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
161 * Check for the channel element in the current channel position configuration.
162 * If it exists, make sure the appropriate element is allocated and map the
163 * channel order to match the internal FFmpeg channel layout.
165 * @param che_pos current channel position configuration
166 * @param type channel element type
167 * @param id channel element id
168 * @param channels count of the number of channels in the configuration
170 * @return Returns error status. 0 - OK, !0 - error
172 static int che_configure(AACContext *ac,
173 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
177 if (che_pos[type][id]) {
178 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
179 return AVERROR(ENOMEM);
180 if (type != TYPE_CCE) {
181 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
182 if (type == TYPE_CPE) {
183 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
187 av_freep(&ac->che[type][id]);
192 * Configure output channel order based on the current program configuration element.
194 * @param che_pos current channel position configuration
195 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
197 * @return Returns error status. 0 - OK, !0 - error
199 static int output_configure(AACContext *ac,
200 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
201 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
202 int channel_config, enum OCStatus oc_type)
204 AVCodecContext *avctx = ac->avccontext;
205 int i, type, channels = 0, ret;
207 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
209 if (channel_config) {
210 for (i = 0; i < tags_per_config[channel_config]; i++) {
211 if ((ret = che_configure(ac, che_pos,
212 aac_channel_layout_map[channel_config - 1][i][0],
213 aac_channel_layout_map[channel_config - 1][i][1],
218 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
221 avctx->channel_layout = aac_channel_layout[channel_config - 1];
223 /* Allocate or free elements depending on if they are in the
224 * current program configuration.
226 * Set up default 1:1 output mapping.
228 * For a 5.1 stream the output order will be:
229 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
232 for (i = 0; i < MAX_ELEM_ID; i++) {
233 for (type = 0; type < 4; type++) {
234 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
239 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
240 ac->tags_mapped = 4 * MAX_ELEM_ID;
242 avctx->channel_layout = 0;
245 avctx->channels = channels;
247 ac->output_configured = oc_type;
253 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
255 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
256 * @param sce_map mono (Single Channel Element) map
257 * @param type speaker type/position for these channels
259 static void decode_channel_map(enum ChannelPosition *cpe_map,
260 enum ChannelPosition *sce_map,
261 enum ChannelPosition type,
262 GetBitContext *gb, int n)
265 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
266 map[get_bits(gb, 4)] = type;
271 * Decode program configuration element; reference: table 4.2.
273 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
275 * @return Returns error status. 0 - OK, !0 - error
277 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
280 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
282 skip_bits(gb, 2); // object_type
284 sampling_index = get_bits(gb, 4);
285 if (ac->m4ac.sampling_index != sampling_index)
286 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");
288 num_front = get_bits(gb, 4);
289 num_side = get_bits(gb, 4);
290 num_back = get_bits(gb, 4);
291 num_lfe = get_bits(gb, 2);
292 num_assoc_data = get_bits(gb, 3);
293 num_cc = get_bits(gb, 4);
296 skip_bits(gb, 4); // mono_mixdown_tag
298 skip_bits(gb, 4); // stereo_mixdown_tag
301 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
303 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
304 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
305 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
306 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
308 skip_bits_long(gb, 4 * num_assoc_data);
310 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
314 /* comment field, first byte is length */
315 skip_bits_long(gb, 8 * get_bits(gb, 8));
320 * Set up channel positions based on a default channel configuration
321 * as specified in table 1.17.
323 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
325 * @return Returns error status. 0 - OK, !0 - error
327 static int set_default_channel_config(AACContext *ac,
328 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
331 if (channel_config < 1 || channel_config > 7) {
332 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
337 /* default channel configurations:
339 * 1ch : front center (mono)
340 * 2ch : L + R (stereo)
341 * 3ch : front center + L + R
342 * 4ch : front center + L + R + back center
343 * 5ch : front center + L + R + back stereo
344 * 6ch : front center + L + R + back stereo + LFE
345 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
348 if (channel_config != 2)
349 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
350 if (channel_config > 1)
351 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
352 if (channel_config == 4)
353 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
354 if (channel_config > 4)
355 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
356 = AAC_CHANNEL_BACK; // back stereo
357 if (channel_config > 5)
358 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
359 if (channel_config == 7)
360 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
366 * Decode GA "General Audio" specific configuration; reference: table 4.1.
368 * @return Returns error status. 0 - OK, !0 - error
370 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
373 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
374 int extension_flag, ret;
376 if (get_bits1(gb)) { // frameLengthFlag
377 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
381 if (get_bits1(gb)) // dependsOnCoreCoder
382 skip_bits(gb, 14); // coreCoderDelay
383 extension_flag = get_bits1(gb);
385 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
386 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
387 skip_bits(gb, 3); // layerNr
389 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
390 if (channel_config == 0) {
391 skip_bits(gb, 4); // element_instance_tag
392 if ((ret = decode_pce(ac, new_che_pos, gb)))
395 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
398 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
401 if (extension_flag) {
402 switch (ac->m4ac.object_type) {
404 skip_bits(gb, 5); // numOfSubFrame
405 skip_bits(gb, 11); // layer_length
409 case AOT_ER_AAC_SCALABLE:
411 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
412 * aacScalefactorDataResilienceFlag
413 * aacSpectralDataResilienceFlag
417 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
423 * Decode audio specific configuration; reference: table 1.13.
425 * @param data pointer to AVCodecContext extradata
426 * @param data_size size of AVCCodecContext extradata
428 * @return Returns error status. 0 - OK, !0 - error
430 static int decode_audio_specific_config(AACContext *ac, void *data,
436 init_get_bits(&gb, data, data_size * 8);
438 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
440 if (ac->m4ac.sampling_index > 12) {
441 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
445 skip_bits_long(&gb, i);
447 switch (ac->m4ac.object_type) {
450 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
454 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
455 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
462 * linear congruential pseudorandom number generator
464 * @param previous_val pointer to the current state of the generator
466 * @return Returns a 32-bit pseudorandom integer
468 static av_always_inline int lcg_random(int previous_val)
470 return previous_val * 1664525 + 1013904223;
473 static void reset_predict_state(PredictorState *ps)
483 static void reset_all_predictors(PredictorState *ps)
486 for (i = 0; i < MAX_PREDICTORS; i++)
487 reset_predict_state(&ps[i]);
490 static void reset_predictor_group(PredictorState *ps, int group_num)
493 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
494 reset_predict_state(&ps[i]);
497 static av_cold int aac_decode_init(AVCodecContext *avccontext)
499 AACContext *ac = avccontext->priv_data;
502 ac->avccontext = avccontext;
504 if (avccontext->extradata_size > 0) {
505 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
507 avccontext->sample_rate = ac->m4ac.sample_rate;
508 } else if (avccontext->channels > 0) {
509 ac->m4ac.sample_rate = avccontext->sample_rate;
512 avccontext->sample_fmt = SAMPLE_FMT_S16;
513 avccontext->frame_size = 1024;
515 AAC_INIT_VLC_STATIC( 0, 304);
516 AAC_INIT_VLC_STATIC( 1, 270);
517 AAC_INIT_VLC_STATIC( 2, 550);
518 AAC_INIT_VLC_STATIC( 3, 300);
519 AAC_INIT_VLC_STATIC( 4, 328);
520 AAC_INIT_VLC_STATIC( 5, 294);
521 AAC_INIT_VLC_STATIC( 6, 306);
522 AAC_INIT_VLC_STATIC( 7, 268);
523 AAC_INIT_VLC_STATIC( 8, 510);
524 AAC_INIT_VLC_STATIC( 9, 366);
525 AAC_INIT_VLC_STATIC(10, 462);
527 dsputil_init(&ac->dsp, avccontext);
529 ac->random_state = 0x1f2e3d4c;
531 // -1024 - Compensate wrong IMDCT method.
532 // 32768 - Required to scale values to the correct range for the bias method
533 // for float to int16 conversion.
535 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
536 ac->add_bias = 385.0f;
537 ac->sf_scale = 1. / (-1024. * 32768.);
541 ac->sf_scale = 1. / -1024.;
545 #if !CONFIG_HARDCODED_TABLES
546 for (i = 0; i < 428; i++)
547 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
548 #endif /* CONFIG_HARDCODED_TABLES */
550 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
551 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
552 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
555 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
556 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
557 // window initialization
558 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
559 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
560 ff_init_ff_sine_windows(10);
561 ff_init_ff_sine_windows( 7);
563 if (!cbrt_tab[(1<<13) - 1]) {
564 for (i = 0; i < 1<<13; i++) {
575 * Skip data_stream_element; reference: table 4.10.
577 static void skip_data_stream_element(GetBitContext *gb)
579 int byte_align = get_bits1(gb);
580 int count = get_bits(gb, 8);
582 count += get_bits(gb, 8);
585 skip_bits_long(gb, 8 * count);
588 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
593 ics->predictor_reset_group = get_bits(gb, 5);
594 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
595 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
599 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
600 ics->prediction_used[sfb] = get_bits1(gb);
606 * Decode Individual Channel Stream info; reference: table 4.6.
608 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
610 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
611 GetBitContext *gb, int common_window)
614 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
615 memset(ics, 0, sizeof(IndividualChannelStream));
618 ics->window_sequence[1] = ics->window_sequence[0];
619 ics->window_sequence[0] = get_bits(gb, 2);
620 ics->use_kb_window[1] = ics->use_kb_window[0];
621 ics->use_kb_window[0] = get_bits1(gb);
622 ics->num_window_groups = 1;
623 ics->group_len[0] = 1;
624 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
626 ics->max_sfb = get_bits(gb, 4);
627 for (i = 0; i < 7; i++) {
629 ics->group_len[ics->num_window_groups - 1]++;
631 ics->num_window_groups++;
632 ics->group_len[ics->num_window_groups - 1] = 1;
635 ics->num_windows = 8;
636 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
637 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
638 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
639 ics->predictor_present = 0;
641 ics->max_sfb = get_bits(gb, 6);
642 ics->num_windows = 1;
643 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
644 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
645 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
646 ics->predictor_present = get_bits1(gb);
647 ics->predictor_reset_group = 0;
648 if (ics->predictor_present) {
649 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
650 if (decode_prediction(ac, ics, gb)) {
651 memset(ics, 0, sizeof(IndividualChannelStream));
654 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
655 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
656 memset(ics, 0, sizeof(IndividualChannelStream));
659 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
660 memset(ics, 0, sizeof(IndividualChannelStream));
666 if (ics->max_sfb > ics->num_swb) {
667 av_log(ac->avccontext, AV_LOG_ERROR,
668 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
669 ics->max_sfb, ics->num_swb);
670 memset(ics, 0, sizeof(IndividualChannelStream));
678 * Decode band types (section_data payload); reference: table 4.46.
680 * @param band_type array of the used band type
681 * @param band_type_run_end array of the last scalefactor band of a band type run
683 * @return Returns error status. 0 - OK, !0 - error
685 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
686 int band_type_run_end[120], GetBitContext *gb,
687 IndividualChannelStream *ics)
690 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
691 for (g = 0; g < ics->num_window_groups; g++) {
693 while (k < ics->max_sfb) {
694 uint8_t sect_end = k;
696 int sect_band_type = get_bits(gb, 4);
697 if (sect_band_type == 12) {
698 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
701 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
702 sect_end += sect_len_incr;
703 sect_end += sect_len_incr;
704 if (sect_end > ics->max_sfb) {
705 av_log(ac->avccontext, AV_LOG_ERROR,
706 "Number of bands (%d) exceeds limit (%d).\n",
707 sect_end, ics->max_sfb);
710 for (; k < sect_end; k++) {
711 band_type [idx] = sect_band_type;
712 band_type_run_end[idx++] = sect_end;
720 * Decode scalefactors; reference: table 4.47.
722 * @param global_gain first scalefactor value as scalefactors are differentially coded
723 * @param band_type array of the used band type
724 * @param band_type_run_end array of the last scalefactor band of a band type run
725 * @param sf array of scalefactors or intensity stereo positions
727 * @return Returns error status. 0 - OK, !0 - error
729 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
730 unsigned int global_gain,
731 IndividualChannelStream *ics,
732 enum BandType band_type[120],
733 int band_type_run_end[120])
735 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
737 int offset[3] = { global_gain, global_gain - 90, 100 };
739 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
740 for (g = 0; g < ics->num_window_groups; g++) {
741 for (i = 0; i < ics->max_sfb;) {
742 int run_end = band_type_run_end[idx];
743 if (band_type[idx] == ZERO_BT) {
744 for (; i < run_end; i++, idx++)
746 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
747 for (; i < run_end; i++, idx++) {
748 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
749 if (offset[2] > 255U) {
750 av_log(ac->avccontext, AV_LOG_ERROR,
751 "%s (%d) out of range.\n", sf_str[2], offset[2]);
754 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
756 } else if (band_type[idx] == NOISE_BT) {
757 for (; i < run_end; i++, idx++) {
758 if (noise_flag-- > 0)
759 offset[1] += get_bits(gb, 9) - 256;
761 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
762 if (offset[1] > 255U) {
763 av_log(ac->avccontext, AV_LOG_ERROR,
764 "%s (%d) out of range.\n", sf_str[1], offset[1]);
767 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
770 for (; i < run_end; i++, idx++) {
771 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
772 if (offset[0] > 255U) {
773 av_log(ac->avccontext, AV_LOG_ERROR,
774 "%s (%d) out of range.\n", sf_str[0], offset[0]);
777 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
786 * Decode pulse data; reference: table 4.7.
788 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
789 const uint16_t *swb_offset, int num_swb)
792 pulse->num_pulse = get_bits(gb, 2) + 1;
793 pulse_swb = get_bits(gb, 6);
794 if (pulse_swb >= num_swb)
796 pulse->pos[0] = swb_offset[pulse_swb];
797 pulse->pos[0] += get_bits(gb, 5);
798 if (pulse->pos[0] > 1023)
800 pulse->amp[0] = get_bits(gb, 4);
801 for (i = 1; i < pulse->num_pulse; i++) {
802 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
803 if (pulse->pos[i] > 1023)
805 pulse->amp[i] = get_bits(gb, 4);
811 * Decode Temporal Noise Shaping data; reference: table 4.48.
813 * @return Returns error status. 0 - OK, !0 - error
815 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
816 GetBitContext *gb, const IndividualChannelStream *ics)
818 int w, filt, i, coef_len, coef_res, coef_compress;
819 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
820 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
821 for (w = 0; w < ics->num_windows; w++) {
822 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
823 coef_res = get_bits1(gb);
825 for (filt = 0; filt < tns->n_filt[w]; filt++) {
827 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
829 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
830 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
831 tns->order[w][filt], tns_max_order);
832 tns->order[w][filt] = 0;
835 if (tns->order[w][filt]) {
836 tns->direction[w][filt] = get_bits1(gb);
837 coef_compress = get_bits1(gb);
838 coef_len = coef_res + 3 - coef_compress;
839 tmp2_idx = 2 * coef_compress + coef_res;
841 for (i = 0; i < tns->order[w][filt]; i++)
842 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
851 * Decode Mid/Side data; reference: table 4.54.
853 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
854 * [1] mask is decoded from bitstream; [2] mask is all 1s;
855 * [3] reserved for scalable AAC
857 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
861 if (ms_present == 1) {
862 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
863 cpe->ms_mask[idx] = get_bits1(gb);
864 } else if (ms_present == 2) {
865 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
870 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
874 *dst++ = v[idx & 15] * s;
875 *dst++ = v[idx>>4 & 15] * s;
881 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
885 *dst++ = v[idx & 3] * s;
886 *dst++ = v[idx>>2 & 3] * s;
887 *dst++ = v[idx>>4 & 3] * s;
888 *dst++ = v[idx>>6 & 3] * s;
894 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
895 unsigned sign, const float *scale)
897 union float754 s0, s1;
899 s0.f = s1.f = *scale;
900 s0.i ^= sign >> 1 << 31;
903 *dst++ = v[idx & 15] * s0.f;
904 *dst++ = v[idx>>4 & 15] * s1.f;
911 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
912 unsigned sign, const float *scale)
914 unsigned nz = idx >> 12;
915 union float754 s = { .f = *scale };
918 t.i = s.i ^ (sign & 1<<31);
919 *dst++ = v[idx & 3] * t.f;
921 sign <<= nz & 1; nz >>= 1;
922 t.i = s.i ^ (sign & 1<<31);
923 *dst++ = v[idx>>2 & 3] * t.f;
925 sign <<= nz & 1; nz >>= 1;
926 t.i = s.i ^ (sign & 1<<31);
927 *dst++ = v[idx>>4 & 3] * t.f;
929 sign <<= nz & 1; nz >>= 1;
930 t.i = s.i ^ (sign & 1<<31);
931 *dst++ = v[idx>>6 & 3] * t.f;
938 * Decode spectral data; reference: table 4.50.
939 * Dequantize and scale spectral data; reference: 4.6.3.3.
941 * @param coef array of dequantized, scaled spectral data
942 * @param sf array of scalefactors or intensity stereo positions
943 * @param pulse_present set if pulses are present
944 * @param pulse pointer to pulse data struct
945 * @param band_type array of the used band type
947 * @return Returns error status. 0 - OK, !0 - error
949 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
950 GetBitContext *gb, const float sf[120],
951 int pulse_present, const Pulse *pulse,
952 const IndividualChannelStream *ics,
953 enum BandType band_type[120])
955 int i, k, g, idx = 0;
956 const int c = 1024 / ics->num_windows;
957 const uint16_t *offsets = ics->swb_offset;
958 float *coef_base = coef;
961 for (g = 0; g < ics->num_windows; g++)
962 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
964 for (g = 0; g < ics->num_window_groups; g++) {
965 unsigned g_len = ics->group_len[g];
967 for (i = 0; i < ics->max_sfb; i++, idx++) {
968 const unsigned cbt_m1 = band_type[idx] - 1;
969 float *cfo = coef + offsets[i];
970 int off_len = offsets[i + 1] - offsets[i];
973 if (cbt_m1 >= INTENSITY_BT2 - 1) {
974 for (group = 0; group < g_len; group++, cfo+=128) {
975 memset(cfo, 0, off_len * sizeof(float));
977 } else if (cbt_m1 == NOISE_BT - 1) {
978 for (group = 0; group < g_len; group++, cfo+=128) {
982 for (k = 0; k < off_len; k++) {
983 ac->random_state = lcg_random(ac->random_state);
984 cfo[k] = ac->random_state;
987 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
988 scale = sf[idx] / sqrtf(band_energy);
989 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
992 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
993 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
994 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
995 const int cb_size = ff_aac_spectral_sizes[cbt_m1];
998 switch (cbt_m1 >> 1) {
1000 for (group = 0; group < g_len; group++, cfo+=128) {
1008 UPDATE_CACHE(re, gb);
1009 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1011 if (code >= cb_size) {
1013 goto err_cb_overflow;
1016 cb_idx = cb_vector_idx[code];
1017 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1023 for (group = 0; group < g_len; group++, cfo+=128) {
1033 UPDATE_CACHE(re, gb);
1034 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1036 if (code >= cb_size) {
1038 goto err_cb_overflow;
1041 #if MIN_CACHE_BITS < 20
1042 UPDATE_CACHE(re, gb);
1044 cb_idx = cb_vector_idx[code];
1045 nnz = cb_idx >> 8 & 15;
1046 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1047 LAST_SKIP_BITS(re, gb, nnz);
1048 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1054 for (group = 0; group < g_len; group++, cfo+=128) {
1062 UPDATE_CACHE(re, gb);
1063 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1065 if (code >= cb_size) {
1067 goto err_cb_overflow;
1070 cb_idx = cb_vector_idx[code];
1071 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1078 for (group = 0; group < g_len; group++, cfo+=128) {
1088 UPDATE_CACHE(re, gb);
1089 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1091 if (code >= cb_size) {
1093 goto err_cb_overflow;
1096 cb_idx = cb_vector_idx[code];
1097 nnz = cb_idx >> 8 & 15;
1098 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1099 LAST_SKIP_BITS(re, gb, nnz);
1100 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1106 for (group = 0; group < g_len; group++, cfo+=128) {
1108 uint32_t *icf = (uint32_t *) cf;
1118 UPDATE_CACHE(re, gb);
1119 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1127 if (code >= cb_size) {
1129 goto err_cb_overflow;
1132 cb_idx = cb_vector_idx[code];
1135 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1136 LAST_SKIP_BITS(re, gb, nnz);
1138 for (j = 0; j < 2; j++) {
1142 /* The total length of escape_sequence must be < 22 bits according
1143 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1144 UPDATE_CACHE(re, gb);
1145 b = GET_CACHE(re, gb);
1146 b = 31 - av_log2(~b);
1149 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1153 #if MIN_CACHE_BITS < 21
1154 LAST_SKIP_BITS(re, gb, b + 1);
1155 UPDATE_CACHE(re, gb);
1157 SKIP_BITS(re, gb, b + 1);
1160 n = (1 << b) + SHOW_UBITS(re, gb, b);
1161 LAST_SKIP_BITS(re, gb, b);
1162 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1165 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1166 *icf++ = (bits & 1<<31) | v;
1173 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1177 CLOSE_READER(re, gb);
1183 if (pulse_present) {
1185 for (i = 0; i < pulse->num_pulse; i++) {
1186 float co = coef_base[ pulse->pos[i] ];
1187 while (offsets[idx + 1] <= pulse->pos[i])
1189 if (band_type[idx] != NOISE_BT && sf[idx]) {
1190 float ico = -pulse->amp[i];
1193 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1195 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1202 av_log(ac->avccontext, AV_LOG_ERROR,
1203 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1204 band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1208 static av_always_inline float flt16_round(float pf)
1212 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1216 static av_always_inline float flt16_even(float pf)
1220 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1224 static av_always_inline float flt16_trunc(float pf)
1228 pun.i &= 0xFFFF0000U;
1232 static void predict(AACContext *ac, PredictorState *ps, float *coef,
1235 const float a = 0.953125; // 61.0 / 64
1236 const float alpha = 0.90625; // 29.0 / 32
1241 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1242 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1244 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1246 *coef += pv * ac->sf_scale;
1248 e0 = *coef / ac->sf_scale;
1249 e1 = e0 - k1 * ps->r0;
1251 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1252 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1253 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1254 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1256 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1257 ps->r0 = flt16_trunc(a * e0);
1261 * Apply AAC-Main style frequency domain prediction.
1263 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1267 if (!sce->ics.predictor_initialized) {
1268 reset_all_predictors(sce->predictor_state);
1269 sce->ics.predictor_initialized = 1;
1272 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1273 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1274 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1275 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1276 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1279 if (sce->ics.predictor_reset_group)
1280 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1282 reset_all_predictors(sce->predictor_state);
1286 * Decode an individual_channel_stream payload; reference: table 4.44.
1288 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1289 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1291 * @return Returns error status. 0 - OK, !0 - error
1293 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1294 GetBitContext *gb, int common_window, int scale_flag)
1297 TemporalNoiseShaping *tns = &sce->tns;
1298 IndividualChannelStream *ics = &sce->ics;
1299 float *out = sce->coeffs;
1300 int global_gain, pulse_present = 0;
1302 /* This assignment is to silence a GCC warning about the variable being used
1303 * uninitialized when in fact it always is.
1305 pulse.num_pulse = 0;
1307 global_gain = get_bits(gb, 8);
1309 if (!common_window && !scale_flag) {
1310 if (decode_ics_info(ac, ics, gb, 0) < 0)
1314 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1316 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1321 if ((pulse_present = get_bits1(gb))) {
1322 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1323 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1326 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1327 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1331 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1333 if (get_bits1(gb)) {
1334 av_log_missing_feature(ac->avccontext, "SSR", 1);
1339 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1342 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1343 apply_prediction(ac, sce);
1349 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1351 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1353 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1354 float *ch0 = cpe->ch[0].coeffs;
1355 float *ch1 = cpe->ch[1].coeffs;
1356 int g, i, group, idx = 0;
1357 const uint16_t *offsets = ics->swb_offset;
1358 for (g = 0; g < ics->num_window_groups; g++) {
1359 for (i = 0; i < ics->max_sfb; i++, idx++) {
1360 if (cpe->ms_mask[idx] &&
1361 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1362 for (group = 0; group < ics->group_len[g]; group++) {
1363 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1364 ch1 + group * 128 + offsets[i],
1365 offsets[i+1] - offsets[i]);
1369 ch0 += ics->group_len[g] * 128;
1370 ch1 += ics->group_len[g] * 128;
1375 * intensity stereo decoding; reference: 4.6.8.2.3
1377 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1378 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1379 * [3] reserved for scalable AAC
1381 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1383 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1384 SingleChannelElement *sce1 = &cpe->ch[1];
1385 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1386 const uint16_t *offsets = ics->swb_offset;
1387 int g, group, i, k, idx = 0;
1390 for (g = 0; g < ics->num_window_groups; g++) {
1391 for (i = 0; i < ics->max_sfb;) {
1392 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1393 const int bt_run_end = sce1->band_type_run_end[idx];
1394 for (; i < bt_run_end; i++, idx++) {
1395 c = -1 + 2 * (sce1->band_type[idx] - 14);
1397 c *= 1 - 2 * cpe->ms_mask[idx];
1398 scale = c * sce1->sf[idx];
1399 for (group = 0; group < ics->group_len[g]; group++)
1400 for (k = offsets[i]; k < offsets[i + 1]; k++)
1401 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1404 int bt_run_end = sce1->band_type_run_end[idx];
1405 idx += bt_run_end - i;
1409 coef0 += ics->group_len[g] * 128;
1410 coef1 += ics->group_len[g] * 128;
1415 * Decode a channel_pair_element; reference: table 4.4.
1417 * @param elem_id Identifies the instance of a syntax element.
1419 * @return Returns error status. 0 - OK, !0 - error
1421 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1423 int i, ret, common_window, ms_present = 0;
1425 common_window = get_bits1(gb);
1426 if (common_window) {
1427 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1429 i = cpe->ch[1].ics.use_kb_window[0];
1430 cpe->ch[1].ics = cpe->ch[0].ics;
1431 cpe->ch[1].ics.use_kb_window[1] = i;
1432 ms_present = get_bits(gb, 2);
1433 if (ms_present == 3) {
1434 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1436 } else if (ms_present)
1437 decode_mid_side_stereo(cpe, gb, ms_present);
1439 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1441 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1444 if (common_window) {
1446 apply_mid_side_stereo(ac, cpe);
1447 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1448 apply_prediction(ac, &cpe->ch[0]);
1449 apply_prediction(ac, &cpe->ch[1]);
1453 apply_intensity_stereo(cpe, ms_present);
1458 * Decode coupling_channel_element; reference: table 4.8.
1460 * @param elem_id Identifies the instance of a syntax element.
1462 * @return Returns error status. 0 - OK, !0 - error
1464 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1470 SingleChannelElement *sce = &che->ch[0];
1471 ChannelCoupling *coup = &che->coup;
1473 coup->coupling_point = 2 * get_bits1(gb);
1474 coup->num_coupled = get_bits(gb, 3);
1475 for (c = 0; c <= coup->num_coupled; c++) {
1477 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1478 coup->id_select[c] = get_bits(gb, 4);
1479 if (coup->type[c] == TYPE_CPE) {
1480 coup->ch_select[c] = get_bits(gb, 2);
1481 if (coup->ch_select[c] == 3)
1484 coup->ch_select[c] = 2;
1486 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1488 sign = get_bits(gb, 1);
1489 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1491 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1494 for (c = 0; c < num_gain; c++) {
1498 float gain_cache = 1.;
1500 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1501 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1502 gain_cache = pow(scale, -gain);
1504 if (coup->coupling_point == AFTER_IMDCT) {
1505 coup->gain[c][0] = gain_cache;
1507 for (g = 0; g < sce->ics.num_window_groups; g++) {
1508 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1509 if (sce->band_type[idx] != ZERO_BT) {
1511 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1519 gain_cache = pow(scale, -t) * s;
1522 coup->gain[c][idx] = gain_cache;
1532 * Decode Spectral Band Replication extension data; reference: table 4.55.
1534 * @param crc flag indicating the presence of CRC checksum
1535 * @param cnt length of TYPE_FIL syntactic element in bytes
1537 * @return Returns number of bytes consumed from the TYPE_FIL element.
1539 static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1542 // TODO : sbr_extension implementation
1543 av_log_missing_feature(ac->avccontext, "SBR", 0);
1544 skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1549 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1551 * @return Returns number of bytes consumed.
1553 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1557 int num_excl_chan = 0;
1560 for (i = 0; i < 7; i++)
1561 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1562 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1564 return num_excl_chan / 7;
1568 * Decode dynamic range information; reference: table 4.52.
1570 * @param cnt length of TYPE_FIL syntactic element in bytes
1572 * @return Returns number of bytes consumed.
1574 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1575 GetBitContext *gb, int cnt)
1578 int drc_num_bands = 1;
1581 /* pce_tag_present? */
1582 if (get_bits1(gb)) {
1583 che_drc->pce_instance_tag = get_bits(gb, 4);
1584 skip_bits(gb, 4); // tag_reserved_bits
1588 /* excluded_chns_present? */
1589 if (get_bits1(gb)) {
1590 n += decode_drc_channel_exclusions(che_drc, gb);
1593 /* drc_bands_present? */
1594 if (get_bits1(gb)) {
1595 che_drc->band_incr = get_bits(gb, 4);
1596 che_drc->interpolation_scheme = get_bits(gb, 4);
1598 drc_num_bands += che_drc->band_incr;
1599 for (i = 0; i < drc_num_bands; i++) {
1600 che_drc->band_top[i] = get_bits(gb, 8);
1605 /* prog_ref_level_present? */
1606 if (get_bits1(gb)) {
1607 che_drc->prog_ref_level = get_bits(gb, 7);
1608 skip_bits1(gb); // prog_ref_level_reserved_bits
1612 for (i = 0; i < drc_num_bands; i++) {
1613 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1614 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1622 * Decode extension data (incomplete); reference: table 4.51.
1624 * @param cnt length of TYPE_FIL syntactic element in bytes
1626 * @return Returns number of bytes consumed
1628 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1632 switch (get_bits(gb, 4)) { // extension type
1633 case EXT_SBR_DATA_CRC:
1636 res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1638 case EXT_DYNAMIC_RANGE:
1639 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1643 case EXT_DATA_ELEMENT:
1645 skip_bits_long(gb, 8 * cnt - 4);
1652 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1654 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1655 * @param coef spectral coefficients
1657 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1658 IndividualChannelStream *ics, int decode)
1660 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1662 int bottom, top, order, start, end, size, inc;
1663 float lpc[TNS_MAX_ORDER];
1665 for (w = 0; w < ics->num_windows; w++) {
1666 bottom = ics->num_swb;
1667 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1669 bottom = FFMAX(0, top - tns->length[w][filt]);
1670 order = tns->order[w][filt];
1675 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1677 start = ics->swb_offset[FFMIN(bottom, mmm)];
1678 end = ics->swb_offset[FFMIN( top, mmm)];
1679 if ((size = end - start) <= 0)
1681 if (tns->direction[w][filt]) {
1690 for (m = 0; m < size; m++, start += inc)
1691 for (i = 1; i <= FFMIN(m, order); i++)
1692 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1698 * Conduct IMDCT and windowing.
1700 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1702 IndividualChannelStream *ics = &sce->ics;
1703 float *in = sce->coeffs;
1704 float *out = sce->ret;
1705 float *saved = sce->saved;
1706 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1707 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1708 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1709 float *buf = ac->buf_mdct;
1710 float *temp = ac->temp;
1714 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1715 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1716 av_log(ac->avccontext, AV_LOG_WARNING,
1717 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1718 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1719 for (i = 0; i < 1024; i += 128)
1720 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1722 ff_imdct_half(&ac->mdct, buf, in);
1724 /* window overlapping
1725 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1726 * and long to short transitions are considered to be short to short
1727 * transitions. This leaves just two cases (long to long and short to short)
1728 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1730 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1731 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1732 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1734 for (i = 0; i < 448; i++)
1735 out[i] = saved[i] + ac->add_bias;
1737 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1738 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1739 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1740 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1741 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1742 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1743 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1745 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1746 for (i = 576; i < 1024; i++)
1747 out[i] = buf[i-512] + ac->add_bias;
1752 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1753 for (i = 0; i < 64; i++)
1754 saved[i] = temp[64 + i] - ac->add_bias;
1755 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1756 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1757 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1758 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1759 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1760 memcpy( saved, buf + 512, 448 * sizeof(float));
1761 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1762 } else { // LONG_STOP or ONLY_LONG
1763 memcpy( saved, buf + 512, 512 * sizeof(float));
1768 * Apply dependent channel coupling (applied before IMDCT).
1770 * @param index index into coupling gain array
1772 static void apply_dependent_coupling(AACContext *ac,
1773 SingleChannelElement *target,
1774 ChannelElement *cce, int index)
1776 IndividualChannelStream *ics = &cce->ch[0].ics;
1777 const uint16_t *offsets = ics->swb_offset;
1778 float *dest = target->coeffs;
1779 const float *src = cce->ch[0].coeffs;
1780 int g, i, group, k, idx = 0;
1781 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1782 av_log(ac->avccontext, AV_LOG_ERROR,
1783 "Dependent coupling is not supported together with LTP\n");
1786 for (g = 0; g < ics->num_window_groups; g++) {
1787 for (i = 0; i < ics->max_sfb; i++, idx++) {
1788 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1789 const float gain = cce->coup.gain[index][idx];
1790 for (group = 0; group < ics->group_len[g]; group++) {
1791 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1793 dest[group * 128 + k] += gain * src[group * 128 + k];
1798 dest += ics->group_len[g] * 128;
1799 src += ics->group_len[g] * 128;
1804 * Apply independent channel coupling (applied after IMDCT).
1806 * @param index index into coupling gain array
1808 static void apply_independent_coupling(AACContext *ac,
1809 SingleChannelElement *target,
1810 ChannelElement *cce, int index)
1813 const float gain = cce->coup.gain[index][0];
1814 const float bias = ac->add_bias;
1815 const float *src = cce->ch[0].ret;
1816 float *dest = target->ret;
1818 for (i = 0; i < 1024; i++)
1819 dest[i] += gain * (src[i] - bias);
1823 * channel coupling transformation interface
1825 * @param index index into coupling gain array
1826 * @param apply_coupling_method pointer to (in)dependent coupling function
1828 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1829 enum RawDataBlockType type, int elem_id,
1830 enum CouplingPoint coupling_point,
1831 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1835 for (i = 0; i < MAX_ELEM_ID; i++) {
1836 ChannelElement *cce = ac->che[TYPE_CCE][i];
1839 if (cce && cce->coup.coupling_point == coupling_point) {
1840 ChannelCoupling *coup = &cce->coup;
1842 for (c = 0; c <= coup->num_coupled; c++) {
1843 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1844 if (coup->ch_select[c] != 1) {
1845 apply_coupling_method(ac, &cc->ch[0], cce, index);
1846 if (coup->ch_select[c] != 0)
1849 if (coup->ch_select[c] != 2)
1850 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1852 index += 1 + (coup->ch_select[c] == 3);
1859 * Convert spectral data to float samples, applying all supported tools as appropriate.
1861 static void spectral_to_sample(AACContext *ac)
1864 for (type = 3; type >= 0; type--) {
1865 for (i = 0; i < MAX_ELEM_ID; i++) {
1866 ChannelElement *che = ac->che[type][i];
1868 if (type <= TYPE_CPE)
1869 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1870 if (che->ch[0].tns.present)
1871 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1872 if (che->ch[1].tns.present)
1873 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1874 if (type <= TYPE_CPE)
1875 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1876 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1877 imdct_and_windowing(ac, &che->ch[0]);
1878 if (type == TYPE_CPE)
1879 imdct_and_windowing(ac, &che->ch[1]);
1880 if (type <= TYPE_CCE)
1881 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1887 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1890 AACADTSHeaderInfo hdr_info;
1892 size = ff_aac_parse_header(gb, &hdr_info);
1894 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1895 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1896 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1897 ac->m4ac.chan_config = hdr_info.chan_config;
1898 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1900 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1902 } else if (ac->output_configured != OC_LOCKED) {
1903 ac->output_configured = OC_NONE;
1905 if (ac->output_configured != OC_LOCKED)
1907 ac->m4ac.sample_rate = hdr_info.sample_rate;
1908 ac->m4ac.sampling_index = hdr_info.sampling_index;
1909 ac->m4ac.object_type = hdr_info.object_type;
1910 if (!ac->avccontext->sample_rate)
1911 ac->avccontext->sample_rate = hdr_info.sample_rate;
1912 if (hdr_info.num_aac_frames == 1) {
1913 if (!hdr_info.crc_absent)
1916 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1923 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1924 int *data_size, AVPacket *avpkt)
1926 const uint8_t *buf = avpkt->data;
1927 int buf_size = avpkt->size;
1928 AACContext *ac = avccontext->priv_data;
1929 ChannelElement *che = NULL;
1931 enum RawDataBlockType elem_type;
1932 int err, elem_id, data_size_tmp;
1934 init_get_bits(&gb, buf, buf_size * 8);
1936 if (show_bits(&gb, 12) == 0xfff) {
1937 if (parse_adts_frame_header(ac, &gb) < 0) {
1938 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1941 if (ac->m4ac.sampling_index > 12) {
1942 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1948 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1949 elem_id = get_bits(&gb, 4);
1951 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1952 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1956 switch (elem_type) {
1959 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1963 err = decode_cpe(ac, &gb, che);
1967 err = decode_cce(ac, &gb, che);
1971 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1975 skip_data_stream_element(&gb);
1980 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1981 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1982 if ((err = decode_pce(ac, new_che_pos, &gb)))
1984 if (ac->output_configured > OC_TRIAL_PCE)
1985 av_log(avccontext, AV_LOG_ERROR,
1986 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1988 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1994 elem_id += get_bits(&gb, 8) - 1;
1996 elem_id -= decode_extension_payload(ac, &gb, elem_id);
1997 err = 0; /* FIXME */
2001 err = -1; /* should not happen, but keeps compiler happy */
2009 spectral_to_sample(ac);
2011 if (!ac->is_saved) {
2017 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
2018 if (*data_size < data_size_tmp) {
2019 av_log(avccontext, AV_LOG_ERROR,
2020 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2021 *data_size, data_size_tmp);
2024 *data_size = data_size_tmp;
2026 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
2028 if (ac->output_configured)
2029 ac->output_configured = OC_LOCKED;
2034 static av_cold int aac_decode_close(AVCodecContext *avccontext)
2036 AACContext *ac = avccontext->priv_data;
2039 for (i = 0; i < MAX_ELEM_ID; i++) {
2040 for (type = 0; type < 4; type++)
2041 av_freep(&ac->che[type][i]);
2044 ff_mdct_end(&ac->mdct);
2045 ff_mdct_end(&ac->mdct_small);
2049 AVCodec aac_decoder = {
2058 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2059 .sample_fmts = (const enum SampleFormat[]) {
2060 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2062 .channel_layouts = aac_channel_layout,