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 * Y 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
88 #include "aacdectab.h"
91 #include "mpeg4audio.h"
92 #include "aac_parser.h"
100 # include "arm/aac.h"
108 static VLC vlc_scalefactors;
109 static VLC vlc_spectral[11];
111 static uint32_t cbrt_tab[1<<13];
113 static const char overread_err[] = "Input buffer exhausted before END element found\n";
115 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
117 if (ac->tag_che_map[type][elem_id]) {
118 return ac->tag_che_map[type][elem_id];
120 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
123 switch (ac->m4ac.chan_config) {
125 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
127 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
130 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
131 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
132 encountered such a stream, transfer the LFE[0] element to SCE[1] */
133 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
135 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
138 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
140 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
143 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
145 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
149 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
151 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
152 } else if (ac->m4ac.chan_config == 2) {
156 if (!ac->tags_mapped && type == TYPE_SCE) {
158 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
166 * Check for the channel element in the current channel position configuration.
167 * If it exists, make sure the appropriate element is allocated and map the
168 * channel order to match the internal FFmpeg channel layout.
170 * @param che_pos current channel position configuration
171 * @param type channel element type
172 * @param id channel element id
173 * @param channels count of the number of channels in the configuration
175 * @return Returns error status. 0 - OK, !0 - error
177 static av_cold int che_configure(AACContext *ac,
178 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
182 if (che_pos[type][id]) {
183 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
184 return AVERROR(ENOMEM);
185 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
186 if (type != TYPE_CCE) {
187 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
188 if (type == TYPE_CPE) {
189 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
193 if (ac->che[type][id])
194 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
195 av_freep(&ac->che[type][id]);
201 * Configure output channel order based on the current program configuration element.
203 * @param che_pos current channel position configuration
204 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
206 * @return Returns error status. 0 - OK, !0 - error
208 static av_cold int output_configure(AACContext *ac,
209 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
210 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
211 int channel_config, enum OCStatus oc_type)
213 AVCodecContext *avctx = ac->avccontext;
214 int i, type, channels = 0, ret;
216 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
218 if (channel_config) {
219 for (i = 0; i < tags_per_config[channel_config]; i++) {
220 if ((ret = che_configure(ac, che_pos,
221 aac_channel_layout_map[channel_config - 1][i][0],
222 aac_channel_layout_map[channel_config - 1][i][1],
227 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
230 avctx->channel_layout = aac_channel_layout[channel_config - 1];
232 /* Allocate or free elements depending on if they are in the
233 * current program configuration.
235 * Set up default 1:1 output mapping.
237 * For a 5.1 stream the output order will be:
238 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
241 for (i = 0; i < MAX_ELEM_ID; i++) {
242 for (type = 0; type < 4; type++) {
243 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
248 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
249 ac->tags_mapped = 4 * MAX_ELEM_ID;
251 avctx->channel_layout = 0;
254 avctx->channels = channels;
256 ac->output_configured = oc_type;
262 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
264 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
265 * @param sce_map mono (Single Channel Element) map
266 * @param type speaker type/position for these channels
268 static void decode_channel_map(enum ChannelPosition *cpe_map,
269 enum ChannelPosition *sce_map,
270 enum ChannelPosition type,
271 GetBitContext *gb, int n)
274 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
275 map[get_bits(gb, 4)] = type;
280 * Decode program configuration element; reference: table 4.2.
282 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
284 * @return Returns error status. 0 - OK, !0 - error
286 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
289 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
292 skip_bits(gb, 2); // object_type
294 sampling_index = get_bits(gb, 4);
295 if (ac->m4ac.sampling_index != sampling_index)
296 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");
298 num_front = get_bits(gb, 4);
299 num_side = get_bits(gb, 4);
300 num_back = get_bits(gb, 4);
301 num_lfe = get_bits(gb, 2);
302 num_assoc_data = get_bits(gb, 3);
303 num_cc = get_bits(gb, 4);
306 skip_bits(gb, 4); // mono_mixdown_tag
308 skip_bits(gb, 4); // stereo_mixdown_tag
311 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
313 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
314 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
315 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
316 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
318 skip_bits_long(gb, 4 * num_assoc_data);
320 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
324 /* comment field, first byte is length */
325 comment_len = get_bits(gb, 8) * 8;
326 if (get_bits_left(gb) < comment_len) {
327 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
330 skip_bits_long(gb, comment_len);
335 * Set up channel positions based on a default channel configuration
336 * as specified in table 1.17.
338 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
340 * @return Returns error status. 0 - OK, !0 - error
342 static av_cold int set_default_channel_config(AACContext *ac,
343 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
346 if (channel_config < 1 || channel_config > 7) {
347 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
352 /* default channel configurations:
354 * 1ch : front center (mono)
355 * 2ch : L + R (stereo)
356 * 3ch : front center + L + R
357 * 4ch : front center + L + R + back center
358 * 5ch : front center + L + R + back stereo
359 * 6ch : front center + L + R + back stereo + LFE
360 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
363 if (channel_config != 2)
364 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
365 if (channel_config > 1)
366 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
367 if (channel_config == 4)
368 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
369 if (channel_config > 4)
370 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
371 = AAC_CHANNEL_BACK; // back stereo
372 if (channel_config > 5)
373 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
374 if (channel_config == 7)
375 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
381 * Decode GA "General Audio" specific configuration; reference: table 4.1.
383 * @return Returns error status. 0 - OK, !0 - error
385 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
388 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
389 int extension_flag, ret;
391 if (get_bits1(gb)) { // frameLengthFlag
392 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
396 if (get_bits1(gb)) // dependsOnCoreCoder
397 skip_bits(gb, 14); // coreCoderDelay
398 extension_flag = get_bits1(gb);
400 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
401 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
402 skip_bits(gb, 3); // layerNr
404 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
405 if (channel_config == 0) {
406 skip_bits(gb, 4); // element_instance_tag
407 if ((ret = decode_pce(ac, new_che_pos, gb)))
410 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
413 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
416 if (extension_flag) {
417 switch (ac->m4ac.object_type) {
419 skip_bits(gb, 5); // numOfSubFrame
420 skip_bits(gb, 11); // layer_length
424 case AOT_ER_AAC_SCALABLE:
426 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
427 * aacScalefactorDataResilienceFlag
428 * aacSpectralDataResilienceFlag
432 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
438 * Decode audio specific configuration; reference: table 1.13.
440 * @param data pointer to AVCodecContext extradata
441 * @param data_size size of AVCCodecContext extradata
443 * @return Returns error status. 0 - OK, !0 - error
445 static int decode_audio_specific_config(AACContext *ac, void *data,
451 init_get_bits(&gb, data, data_size * 8);
453 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
455 if (ac->m4ac.sampling_index > 12) {
456 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
460 skip_bits_long(&gb, i);
462 switch (ac->m4ac.object_type) {
465 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
469 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
470 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
477 * linear congruential pseudorandom number generator
479 * @param previous_val pointer to the current state of the generator
481 * @return Returns a 32-bit pseudorandom integer
483 static av_always_inline int lcg_random(int previous_val)
485 return previous_val * 1664525 + 1013904223;
488 static av_always_inline void reset_predict_state(PredictorState *ps)
498 static void reset_all_predictors(PredictorState *ps)
501 for (i = 0; i < MAX_PREDICTORS; i++)
502 reset_predict_state(&ps[i]);
505 static void reset_predictor_group(PredictorState *ps, int group_num)
508 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
509 reset_predict_state(&ps[i]);
512 static av_cold int aac_decode_init(AVCodecContext *avccontext)
514 AACContext *ac = avccontext->priv_data;
517 ac->avccontext = avccontext;
518 ac->m4ac.sample_rate = avccontext->sample_rate;
520 if (avccontext->extradata_size > 0) {
521 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
525 avccontext->sample_fmt = SAMPLE_FMT_S16;
527 AAC_INIT_VLC_STATIC( 0, 304);
528 AAC_INIT_VLC_STATIC( 1, 270);
529 AAC_INIT_VLC_STATIC( 2, 550);
530 AAC_INIT_VLC_STATIC( 3, 300);
531 AAC_INIT_VLC_STATIC( 4, 328);
532 AAC_INIT_VLC_STATIC( 5, 294);
533 AAC_INIT_VLC_STATIC( 6, 306);
534 AAC_INIT_VLC_STATIC( 7, 268);
535 AAC_INIT_VLC_STATIC( 8, 510);
536 AAC_INIT_VLC_STATIC( 9, 366);
537 AAC_INIT_VLC_STATIC(10, 462);
541 dsputil_init(&ac->dsp, avccontext);
543 ac->random_state = 0x1f2e3d4c;
545 // -1024 - Compensate wrong IMDCT method.
546 // 32768 - Required to scale values to the correct range for the bias method
547 // for float to int16 conversion.
549 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
550 ac->add_bias = 385.0f;
551 ac->sf_scale = 1. / (-1024. * 32768.);
555 ac->sf_scale = 1. / -1024.;
559 #if !CONFIG_HARDCODED_TABLES
560 for (i = 0; i < 428; i++)
561 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
562 #endif /* CONFIG_HARDCODED_TABLES */
564 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
565 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
566 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
569 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
570 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
571 // window initialization
572 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
573 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
574 ff_init_ff_sine_windows(10);
575 ff_init_ff_sine_windows( 7);
577 if (!cbrt_tab[(1<<13) - 1]) {
578 for (i = 0; i < 1<<13; i++) {
589 * Skip data_stream_element; reference: table 4.10.
591 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
593 int byte_align = get_bits1(gb);
594 int count = get_bits(gb, 8);
596 count += get_bits(gb, 8);
600 if (get_bits_left(gb) < 8 * count) {
601 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
604 skip_bits_long(gb, 8 * count);
608 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
613 ics->predictor_reset_group = get_bits(gb, 5);
614 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
615 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
619 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
620 ics->prediction_used[sfb] = get_bits1(gb);
626 * Decode Individual Channel Stream info; reference: table 4.6.
628 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
630 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
631 GetBitContext *gb, int common_window)
634 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
635 memset(ics, 0, sizeof(IndividualChannelStream));
638 ics->window_sequence[1] = ics->window_sequence[0];
639 ics->window_sequence[0] = get_bits(gb, 2);
640 ics->use_kb_window[1] = ics->use_kb_window[0];
641 ics->use_kb_window[0] = get_bits1(gb);
642 ics->num_window_groups = 1;
643 ics->group_len[0] = 1;
644 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
646 ics->max_sfb = get_bits(gb, 4);
647 for (i = 0; i < 7; i++) {
649 ics->group_len[ics->num_window_groups - 1]++;
651 ics->num_window_groups++;
652 ics->group_len[ics->num_window_groups - 1] = 1;
655 ics->num_windows = 8;
656 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
657 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
658 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
659 ics->predictor_present = 0;
661 ics->max_sfb = get_bits(gb, 6);
662 ics->num_windows = 1;
663 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
664 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
665 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
666 ics->predictor_present = get_bits1(gb);
667 ics->predictor_reset_group = 0;
668 if (ics->predictor_present) {
669 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
670 if (decode_prediction(ac, ics, gb)) {
671 memset(ics, 0, sizeof(IndividualChannelStream));
674 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
675 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
676 memset(ics, 0, sizeof(IndividualChannelStream));
679 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
680 memset(ics, 0, sizeof(IndividualChannelStream));
686 if (ics->max_sfb > ics->num_swb) {
687 av_log(ac->avccontext, AV_LOG_ERROR,
688 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
689 ics->max_sfb, ics->num_swb);
690 memset(ics, 0, sizeof(IndividualChannelStream));
698 * Decode band types (section_data payload); reference: table 4.46.
700 * @param band_type array of the used band type
701 * @param band_type_run_end array of the last scalefactor band of a band type run
703 * @return Returns error status. 0 - OK, !0 - error
705 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
706 int band_type_run_end[120], GetBitContext *gb,
707 IndividualChannelStream *ics)
710 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
711 for (g = 0; g < ics->num_window_groups; g++) {
713 while (k < ics->max_sfb) {
714 uint8_t sect_end = k;
716 int sect_band_type = get_bits(gb, 4);
717 if (sect_band_type == 12) {
718 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
721 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
722 sect_end += sect_len_incr;
723 sect_end += sect_len_incr;
724 if (get_bits_left(gb) < 0) {
725 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
728 if (sect_end > ics->max_sfb) {
729 av_log(ac->avccontext, AV_LOG_ERROR,
730 "Number of bands (%d) exceeds limit (%d).\n",
731 sect_end, ics->max_sfb);
734 for (; k < sect_end; k++) {
735 band_type [idx] = sect_band_type;
736 band_type_run_end[idx++] = sect_end;
744 * Decode scalefactors; reference: table 4.47.
746 * @param global_gain first scalefactor value as scalefactors are differentially coded
747 * @param band_type array of the used band type
748 * @param band_type_run_end array of the last scalefactor band of a band type run
749 * @param sf array of scalefactors or intensity stereo positions
751 * @return Returns error status. 0 - OK, !0 - error
753 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
754 unsigned int global_gain,
755 IndividualChannelStream *ics,
756 enum BandType band_type[120],
757 int band_type_run_end[120])
759 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
761 int offset[3] = { global_gain, global_gain - 90, 100 };
763 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
764 for (g = 0; g < ics->num_window_groups; g++) {
765 for (i = 0; i < ics->max_sfb;) {
766 int run_end = band_type_run_end[idx];
767 if (band_type[idx] == ZERO_BT) {
768 for (; i < run_end; i++, idx++)
770 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
771 for (; i < run_end; i++, idx++) {
772 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
773 if (offset[2] > 255U) {
774 av_log(ac->avccontext, AV_LOG_ERROR,
775 "%s (%d) out of range.\n", sf_str[2], offset[2]);
778 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
780 } else if (band_type[idx] == NOISE_BT) {
781 for (; i < run_end; i++, idx++) {
782 if (noise_flag-- > 0)
783 offset[1] += get_bits(gb, 9) - 256;
785 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
786 if (offset[1] > 255U) {
787 av_log(ac->avccontext, AV_LOG_ERROR,
788 "%s (%d) out of range.\n", sf_str[1], offset[1]);
791 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
794 for (; i < run_end; i++, idx++) {
795 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
796 if (offset[0] > 255U) {
797 av_log(ac->avccontext, AV_LOG_ERROR,
798 "%s (%d) out of range.\n", sf_str[0], offset[0]);
801 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
810 * Decode pulse data; reference: table 4.7.
812 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
813 const uint16_t *swb_offset, int num_swb)
816 pulse->num_pulse = get_bits(gb, 2) + 1;
817 pulse_swb = get_bits(gb, 6);
818 if (pulse_swb >= num_swb)
820 pulse->pos[0] = swb_offset[pulse_swb];
821 pulse->pos[0] += get_bits(gb, 5);
822 if (pulse->pos[0] > 1023)
824 pulse->amp[0] = get_bits(gb, 4);
825 for (i = 1; i < pulse->num_pulse; i++) {
826 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
827 if (pulse->pos[i] > 1023)
829 pulse->amp[i] = get_bits(gb, 4);
835 * Decode Temporal Noise Shaping data; reference: table 4.48.
837 * @return Returns error status. 0 - OK, !0 - error
839 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
840 GetBitContext *gb, const IndividualChannelStream *ics)
842 int w, filt, i, coef_len, coef_res, coef_compress;
843 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
844 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
845 for (w = 0; w < ics->num_windows; w++) {
846 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
847 coef_res = get_bits1(gb);
849 for (filt = 0; filt < tns->n_filt[w]; filt++) {
851 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
853 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
854 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
855 tns->order[w][filt], tns_max_order);
856 tns->order[w][filt] = 0;
859 if (tns->order[w][filt]) {
860 tns->direction[w][filt] = get_bits1(gb);
861 coef_compress = get_bits1(gb);
862 coef_len = coef_res + 3 - coef_compress;
863 tmp2_idx = 2 * coef_compress + coef_res;
865 for (i = 0; i < tns->order[w][filt]; i++)
866 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
875 * Decode Mid/Side data; reference: table 4.54.
877 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
878 * [1] mask is decoded from bitstream; [2] mask is all 1s;
879 * [3] reserved for scalable AAC
881 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
885 if (ms_present == 1) {
886 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
887 cpe->ms_mask[idx] = get_bits1(gb);
888 } else if (ms_present == 2) {
889 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
894 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
898 *dst++ = v[idx & 15] * s;
899 *dst++ = v[idx>>4 & 15] * s;
905 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
909 *dst++ = v[idx & 3] * s;
910 *dst++ = v[idx>>2 & 3] * s;
911 *dst++ = v[idx>>4 & 3] * s;
912 *dst++ = v[idx>>6 & 3] * s;
918 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
919 unsigned sign, const float *scale)
921 union float754 s0, s1;
923 s0.f = s1.f = *scale;
924 s0.i ^= sign >> 1 << 31;
927 *dst++ = v[idx & 15] * s0.f;
928 *dst++ = v[idx>>4 & 15] * s1.f;
935 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
936 unsigned sign, const float *scale)
938 unsigned nz = idx >> 12;
939 union float754 s = { .f = *scale };
942 t.i = s.i ^ (sign & 1<<31);
943 *dst++ = v[idx & 3] * t.f;
945 sign <<= nz & 1; nz >>= 1;
946 t.i = s.i ^ (sign & 1<<31);
947 *dst++ = v[idx>>2 & 3] * t.f;
949 sign <<= nz & 1; nz >>= 1;
950 t.i = s.i ^ (sign & 1<<31);
951 *dst++ = v[idx>>4 & 3] * t.f;
953 sign <<= nz & 1; nz >>= 1;
954 t.i = s.i ^ (sign & 1<<31);
955 *dst++ = v[idx>>6 & 3] * t.f;
962 * Decode spectral data; reference: table 4.50.
963 * Dequantize and scale spectral data; reference: 4.6.3.3.
965 * @param coef array of dequantized, scaled spectral data
966 * @param sf array of scalefactors or intensity stereo positions
967 * @param pulse_present set if pulses are present
968 * @param pulse pointer to pulse data struct
969 * @param band_type array of the used band type
971 * @return Returns error status. 0 - OK, !0 - error
973 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
974 GetBitContext *gb, const float sf[120],
975 int pulse_present, const Pulse *pulse,
976 const IndividualChannelStream *ics,
977 enum BandType band_type[120])
979 int i, k, g, idx = 0;
980 const int c = 1024 / ics->num_windows;
981 const uint16_t *offsets = ics->swb_offset;
982 float *coef_base = coef;
985 for (g = 0; g < ics->num_windows; g++)
986 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
988 for (g = 0; g < ics->num_window_groups; g++) {
989 unsigned g_len = ics->group_len[g];
991 for (i = 0; i < ics->max_sfb; i++, idx++) {
992 const unsigned cbt_m1 = band_type[idx] - 1;
993 float *cfo = coef + offsets[i];
994 int off_len = offsets[i + 1] - offsets[i];
997 if (cbt_m1 >= INTENSITY_BT2 - 1) {
998 for (group = 0; group < g_len; group++, cfo+=128) {
999 memset(cfo, 0, off_len * sizeof(float));
1001 } else if (cbt_m1 == NOISE_BT - 1) {
1002 for (group = 0; group < g_len; group++, cfo+=128) {
1006 for (k = 0; k < off_len; k++) {
1007 ac->random_state = lcg_random(ac->random_state);
1008 cfo[k] = ac->random_state;
1011 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1012 scale = sf[idx] / sqrtf(band_energy);
1013 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1016 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1017 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1018 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1019 const int cb_size = ff_aac_spectral_sizes[cbt_m1];
1020 OPEN_READER(re, gb);
1022 switch (cbt_m1 >> 1) {
1024 for (group = 0; group < g_len; group++, cfo+=128) {
1032 UPDATE_CACHE(re, gb);
1033 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1035 if (code >= cb_size) {
1037 goto err_cb_overflow;
1040 cb_idx = cb_vector_idx[code];
1041 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1047 for (group = 0; group < g_len; group++, cfo+=128) {
1057 UPDATE_CACHE(re, gb);
1058 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1060 if (code >= cb_size) {
1062 goto err_cb_overflow;
1065 #if MIN_CACHE_BITS < 20
1066 UPDATE_CACHE(re, gb);
1068 cb_idx = cb_vector_idx[code];
1069 nnz = cb_idx >> 8 & 15;
1070 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1071 LAST_SKIP_BITS(re, gb, nnz);
1072 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1078 for (group = 0; group < g_len; group++, cfo+=128) {
1086 UPDATE_CACHE(re, gb);
1087 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1089 if (code >= cb_size) {
1091 goto err_cb_overflow;
1094 cb_idx = cb_vector_idx[code];
1095 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1102 for (group = 0; group < g_len; group++, cfo+=128) {
1112 UPDATE_CACHE(re, gb);
1113 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1115 if (code >= cb_size) {
1117 goto err_cb_overflow;
1120 cb_idx = cb_vector_idx[code];
1121 nnz = cb_idx >> 8 & 15;
1122 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1123 LAST_SKIP_BITS(re, gb, nnz);
1124 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1130 for (group = 0; group < g_len; group++, cfo+=128) {
1132 uint32_t *icf = (uint32_t *) cf;
1142 UPDATE_CACHE(re, gb);
1143 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1151 if (code >= cb_size) {
1153 goto err_cb_overflow;
1156 cb_idx = cb_vector_idx[code];
1159 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1160 LAST_SKIP_BITS(re, gb, nnz);
1162 for (j = 0; j < 2; j++) {
1166 /* The total length of escape_sequence must be < 22 bits according
1167 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1168 UPDATE_CACHE(re, gb);
1169 b = GET_CACHE(re, gb);
1170 b = 31 - av_log2(~b);
1173 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1177 #if MIN_CACHE_BITS < 21
1178 LAST_SKIP_BITS(re, gb, b + 1);
1179 UPDATE_CACHE(re, gb);
1181 SKIP_BITS(re, gb, b + 1);
1184 n = (1 << b) + SHOW_UBITS(re, gb, b);
1185 LAST_SKIP_BITS(re, gb, b);
1186 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1189 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1190 *icf++ = (bits & 1<<31) | v;
1197 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1201 CLOSE_READER(re, gb);
1207 if (pulse_present) {
1209 for (i = 0; i < pulse->num_pulse; i++) {
1210 float co = coef_base[ pulse->pos[i] ];
1211 while (offsets[idx + 1] <= pulse->pos[i])
1213 if (band_type[idx] != NOISE_BT && sf[idx]) {
1214 float ico = -pulse->amp[i];
1217 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1219 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1226 av_log(ac->avccontext, AV_LOG_ERROR,
1227 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1228 band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1232 static av_always_inline float flt16_round(float pf)
1236 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1240 static av_always_inline float flt16_even(float pf)
1244 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1248 static av_always_inline float flt16_trunc(float pf)
1252 pun.i &= 0xFFFF0000U;
1256 static av_always_inline void predict(AACContext *ac, PredictorState *ps, float *coef,
1259 const float a = 0.953125; // 61.0 / 64
1260 const float alpha = 0.90625; // 29.0 / 32
1265 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1266 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1268 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1270 *coef += pv * ac->sf_scale;
1272 e0 = *coef / ac->sf_scale;
1273 e1 = e0 - k1 * ps->r0;
1275 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1276 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1277 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1278 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1280 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1281 ps->r0 = flt16_trunc(a * e0);
1285 * Apply AAC-Main style frequency domain prediction.
1287 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1291 if (!sce->ics.predictor_initialized) {
1292 reset_all_predictors(sce->predictor_state);
1293 sce->ics.predictor_initialized = 1;
1296 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1297 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1298 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1299 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1300 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1303 if (sce->ics.predictor_reset_group)
1304 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1306 reset_all_predictors(sce->predictor_state);
1310 * Decode an individual_channel_stream payload; reference: table 4.44.
1312 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1313 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1315 * @return Returns error status. 0 - OK, !0 - error
1317 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1318 GetBitContext *gb, int common_window, int scale_flag)
1321 TemporalNoiseShaping *tns = &sce->tns;
1322 IndividualChannelStream *ics = &sce->ics;
1323 float *out = sce->coeffs;
1324 int global_gain, pulse_present = 0;
1326 /* This assignment is to silence a GCC warning about the variable being used
1327 * uninitialized when in fact it always is.
1329 pulse.num_pulse = 0;
1331 global_gain = get_bits(gb, 8);
1333 if (!common_window && !scale_flag) {
1334 if (decode_ics_info(ac, ics, gb, 0) < 0)
1338 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1340 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1345 if ((pulse_present = get_bits1(gb))) {
1346 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1347 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1350 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1351 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1355 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1357 if (get_bits1(gb)) {
1358 av_log_missing_feature(ac->avccontext, "SSR", 1);
1363 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1366 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1367 apply_prediction(ac, sce);
1373 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1375 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1377 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1378 float *ch0 = cpe->ch[0].coeffs;
1379 float *ch1 = cpe->ch[1].coeffs;
1380 int g, i, group, idx = 0;
1381 const uint16_t *offsets = ics->swb_offset;
1382 for (g = 0; g < ics->num_window_groups; g++) {
1383 for (i = 0; i < ics->max_sfb; i++, idx++) {
1384 if (cpe->ms_mask[idx] &&
1385 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1386 for (group = 0; group < ics->group_len[g]; group++) {
1387 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1388 ch1 + group * 128 + offsets[i],
1389 offsets[i+1] - offsets[i]);
1393 ch0 += ics->group_len[g] * 128;
1394 ch1 += ics->group_len[g] * 128;
1399 * intensity stereo decoding; reference: 4.6.8.2.3
1401 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1402 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1403 * [3] reserved for scalable AAC
1405 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1407 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1408 SingleChannelElement *sce1 = &cpe->ch[1];
1409 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1410 const uint16_t *offsets = ics->swb_offset;
1411 int g, group, i, k, idx = 0;
1414 for (g = 0; g < ics->num_window_groups; g++) {
1415 for (i = 0; i < ics->max_sfb;) {
1416 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1417 const int bt_run_end = sce1->band_type_run_end[idx];
1418 for (; i < bt_run_end; i++, idx++) {
1419 c = -1 + 2 * (sce1->band_type[idx] - 14);
1421 c *= 1 - 2 * cpe->ms_mask[idx];
1422 scale = c * sce1->sf[idx];
1423 for (group = 0; group < ics->group_len[g]; group++)
1424 for (k = offsets[i]; k < offsets[i + 1]; k++)
1425 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1428 int bt_run_end = sce1->band_type_run_end[idx];
1429 idx += bt_run_end - i;
1433 coef0 += ics->group_len[g] * 128;
1434 coef1 += ics->group_len[g] * 128;
1439 * Decode a channel_pair_element; reference: table 4.4.
1441 * @param elem_id Identifies the instance of a syntax element.
1443 * @return Returns error status. 0 - OK, !0 - error
1445 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1447 int i, ret, common_window, ms_present = 0;
1449 common_window = get_bits1(gb);
1450 if (common_window) {
1451 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1453 i = cpe->ch[1].ics.use_kb_window[0];
1454 cpe->ch[1].ics = cpe->ch[0].ics;
1455 cpe->ch[1].ics.use_kb_window[1] = i;
1456 ms_present = get_bits(gb, 2);
1457 if (ms_present == 3) {
1458 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1460 } else if (ms_present)
1461 decode_mid_side_stereo(cpe, gb, ms_present);
1463 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1465 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1468 if (common_window) {
1470 apply_mid_side_stereo(ac, cpe);
1471 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1472 apply_prediction(ac, &cpe->ch[0]);
1473 apply_prediction(ac, &cpe->ch[1]);
1477 apply_intensity_stereo(cpe, ms_present);
1482 * Decode coupling_channel_element; reference: table 4.8.
1484 * @param elem_id Identifies the instance of a syntax element.
1486 * @return Returns error status. 0 - OK, !0 - error
1488 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1494 SingleChannelElement *sce = &che->ch[0];
1495 ChannelCoupling *coup = &che->coup;
1497 coup->coupling_point = 2 * get_bits1(gb);
1498 coup->num_coupled = get_bits(gb, 3);
1499 for (c = 0; c <= coup->num_coupled; c++) {
1501 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1502 coup->id_select[c] = get_bits(gb, 4);
1503 if (coup->type[c] == TYPE_CPE) {
1504 coup->ch_select[c] = get_bits(gb, 2);
1505 if (coup->ch_select[c] == 3)
1508 coup->ch_select[c] = 2;
1510 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1512 sign = get_bits(gb, 1);
1513 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1515 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1518 for (c = 0; c < num_gain; c++) {
1522 float gain_cache = 1.;
1524 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1525 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1526 gain_cache = pow(scale, -gain);
1528 if (coup->coupling_point == AFTER_IMDCT) {
1529 coup->gain[c][0] = gain_cache;
1531 for (g = 0; g < sce->ics.num_window_groups; g++) {
1532 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1533 if (sce->band_type[idx] != ZERO_BT) {
1535 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1543 gain_cache = pow(scale, -t) * s;
1546 coup->gain[c][idx] = gain_cache;
1556 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1558 * @return Returns number of bytes consumed.
1560 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1564 int num_excl_chan = 0;
1567 for (i = 0; i < 7; i++)
1568 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1569 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1571 return num_excl_chan / 7;
1575 * Decode dynamic range information; reference: table 4.52.
1577 * @param cnt length of TYPE_FIL syntactic element in bytes
1579 * @return Returns number of bytes consumed.
1581 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1582 GetBitContext *gb, int cnt)
1585 int drc_num_bands = 1;
1588 /* pce_tag_present? */
1589 if (get_bits1(gb)) {
1590 che_drc->pce_instance_tag = get_bits(gb, 4);
1591 skip_bits(gb, 4); // tag_reserved_bits
1595 /* excluded_chns_present? */
1596 if (get_bits1(gb)) {
1597 n += decode_drc_channel_exclusions(che_drc, gb);
1600 /* drc_bands_present? */
1601 if (get_bits1(gb)) {
1602 che_drc->band_incr = get_bits(gb, 4);
1603 che_drc->interpolation_scheme = get_bits(gb, 4);
1605 drc_num_bands += che_drc->band_incr;
1606 for (i = 0; i < drc_num_bands; i++) {
1607 che_drc->band_top[i] = get_bits(gb, 8);
1612 /* prog_ref_level_present? */
1613 if (get_bits1(gb)) {
1614 che_drc->prog_ref_level = get_bits(gb, 7);
1615 skip_bits1(gb); // prog_ref_level_reserved_bits
1619 for (i = 0; i < drc_num_bands; i++) {
1620 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1621 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1629 * Decode extension data (incomplete); reference: table 4.51.
1631 * @param cnt length of TYPE_FIL syntactic element in bytes
1633 * @return Returns number of bytes consumed
1635 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1636 ChannelElement *che, enum RawDataBlockType elem_type)
1640 switch (get_bits(gb, 4)) { // extension type
1641 case EXT_SBR_DATA_CRC:
1645 av_log(ac->avccontext, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1647 } else if (!ac->m4ac.sbr) {
1648 av_log(ac->avccontext, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1649 skip_bits_long(gb, 8 * cnt - 4);
1651 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1652 av_log(ac->avccontext, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1653 skip_bits_long(gb, 8 * cnt - 4);
1658 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1660 case EXT_DYNAMIC_RANGE:
1661 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1665 case EXT_DATA_ELEMENT:
1667 skip_bits_long(gb, 8 * cnt - 4);
1674 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1676 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1677 * @param coef spectral coefficients
1679 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1680 IndividualChannelStream *ics, int decode)
1682 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1684 int bottom, top, order, start, end, size, inc;
1685 float lpc[TNS_MAX_ORDER];
1687 for (w = 0; w < ics->num_windows; w++) {
1688 bottom = ics->num_swb;
1689 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1691 bottom = FFMAX(0, top - tns->length[w][filt]);
1692 order = tns->order[w][filt];
1697 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1699 start = ics->swb_offset[FFMIN(bottom, mmm)];
1700 end = ics->swb_offset[FFMIN( top, mmm)];
1701 if ((size = end - start) <= 0)
1703 if (tns->direction[w][filt]) {
1712 for (m = 0; m < size; m++, start += inc)
1713 for (i = 1; i <= FFMIN(m, order); i++)
1714 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1720 * Conduct IMDCT and windowing.
1722 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1724 IndividualChannelStream *ics = &sce->ics;
1725 float *in = sce->coeffs;
1726 float *out = sce->ret;
1727 float *saved = sce->saved;
1728 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1729 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1730 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1731 float *buf = ac->buf_mdct;
1732 float *temp = ac->temp;
1736 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1737 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1738 av_log(ac->avccontext, AV_LOG_WARNING,
1739 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1740 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1741 for (i = 0; i < 1024; i += 128)
1742 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1744 ff_imdct_half(&ac->mdct, buf, in);
1746 /* window overlapping
1747 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1748 * and long to short transitions are considered to be short to short
1749 * transitions. This leaves just two cases (long to long and short to short)
1750 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1752 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1753 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1754 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512);
1756 for (i = 0; i < 448; i++)
1757 out[i] = saved[i] + ac->add_bias;
1759 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1760 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64);
1761 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64);
1762 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64);
1763 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64);
1764 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64);
1765 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1767 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64);
1768 for (i = 576; i < 1024; i++)
1769 out[i] = buf[i-512] + ac->add_bias;
1774 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1775 for (i = 0; i < 64; i++)
1776 saved[i] = temp[64 + i] - ac->add_bias;
1777 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1778 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1779 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1780 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1781 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1782 memcpy( saved, buf + 512, 448 * sizeof(float));
1783 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1784 } else { // LONG_STOP or ONLY_LONG
1785 memcpy( saved, buf + 512, 512 * sizeof(float));
1790 * Apply dependent channel coupling (applied before IMDCT).
1792 * @param index index into coupling gain array
1794 static void apply_dependent_coupling(AACContext *ac,
1795 SingleChannelElement *target,
1796 ChannelElement *cce, int index)
1798 IndividualChannelStream *ics = &cce->ch[0].ics;
1799 const uint16_t *offsets = ics->swb_offset;
1800 float *dest = target->coeffs;
1801 const float *src = cce->ch[0].coeffs;
1802 int g, i, group, k, idx = 0;
1803 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1804 av_log(ac->avccontext, AV_LOG_ERROR,
1805 "Dependent coupling is not supported together with LTP\n");
1808 for (g = 0; g < ics->num_window_groups; g++) {
1809 for (i = 0; i < ics->max_sfb; i++, idx++) {
1810 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1811 const float gain = cce->coup.gain[index][idx];
1812 for (group = 0; group < ics->group_len[g]; group++) {
1813 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1815 dest[group * 128 + k] += gain * src[group * 128 + k];
1820 dest += ics->group_len[g] * 128;
1821 src += ics->group_len[g] * 128;
1826 * Apply independent channel coupling (applied after IMDCT).
1828 * @param index index into coupling gain array
1830 static void apply_independent_coupling(AACContext *ac,
1831 SingleChannelElement *target,
1832 ChannelElement *cce, int index)
1835 const float gain = cce->coup.gain[index][0];
1836 const float bias = ac->add_bias;
1837 const float *src = cce->ch[0].ret;
1838 float *dest = target->ret;
1839 const int len = 1024 << (ac->m4ac.sbr == 1);
1841 for (i = 0; i < len; i++)
1842 dest[i] += gain * (src[i] - bias);
1846 * channel coupling transformation interface
1848 * @param index index into coupling gain array
1849 * @param apply_coupling_method pointer to (in)dependent coupling function
1851 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1852 enum RawDataBlockType type, int elem_id,
1853 enum CouplingPoint coupling_point,
1854 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1858 for (i = 0; i < MAX_ELEM_ID; i++) {
1859 ChannelElement *cce = ac->che[TYPE_CCE][i];
1862 if (cce && cce->coup.coupling_point == coupling_point) {
1863 ChannelCoupling *coup = &cce->coup;
1865 for (c = 0; c <= coup->num_coupled; c++) {
1866 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1867 if (coup->ch_select[c] != 1) {
1868 apply_coupling_method(ac, &cc->ch[0], cce, index);
1869 if (coup->ch_select[c] != 0)
1872 if (coup->ch_select[c] != 2)
1873 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1875 index += 1 + (coup->ch_select[c] == 3);
1882 * Convert spectral data to float samples, applying all supported tools as appropriate.
1884 static void spectral_to_sample(AACContext *ac)
1887 for (type = 3; type >= 0; type--) {
1888 for (i = 0; i < MAX_ELEM_ID; i++) {
1889 ChannelElement *che = ac->che[type][i];
1891 if (type <= TYPE_CPE)
1892 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1893 if (che->ch[0].tns.present)
1894 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1895 if (che->ch[1].tns.present)
1896 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1897 if (type <= TYPE_CPE)
1898 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1899 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1900 imdct_and_windowing(ac, &che->ch[0]);
1901 if (ac->m4ac.sbr > 0) {
1902 ff_sbr_dequant(ac, &che->sbr, type == TYPE_CPE ? TYPE_CPE : TYPE_SCE);
1903 ff_sbr_apply(ac, &che->sbr, 0, che->ch[0].ret, che->ch[0].ret);
1906 if (type == TYPE_CPE) {
1907 imdct_and_windowing(ac, &che->ch[1]);
1908 if (ac->m4ac.sbr > 0)
1909 ff_sbr_apply(ac, &che->sbr, 1, che->ch[1].ret, che->ch[1].ret);
1911 if (type <= TYPE_CCE)
1912 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1918 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1921 AACADTSHeaderInfo hdr_info;
1923 size = ff_aac_parse_header(gb, &hdr_info);
1925 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1926 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1927 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1928 ac->m4ac.chan_config = hdr_info.chan_config;
1929 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1931 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1933 } else if (ac->output_configured != OC_LOCKED) {
1934 ac->output_configured = OC_NONE;
1936 if (ac->output_configured != OC_LOCKED)
1938 ac->m4ac.sample_rate = hdr_info.sample_rate;
1939 ac->m4ac.sampling_index = hdr_info.sampling_index;
1940 ac->m4ac.object_type = hdr_info.object_type;
1941 if (!ac->avccontext->sample_rate)
1942 ac->avccontext->sample_rate = hdr_info.sample_rate;
1943 if (hdr_info.num_aac_frames == 1) {
1944 if (!hdr_info.crc_absent)
1947 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1954 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1955 int *data_size, AVPacket *avpkt)
1957 const uint8_t *buf = avpkt->data;
1958 int buf_size = avpkt->size;
1959 AACContext *ac = avccontext->priv_data;
1960 ChannelElement *che = NULL, *che_prev = NULL;
1962 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1963 int err, elem_id, data_size_tmp;
1965 int samples = 1024, multiplier;
1967 init_get_bits(&gb, buf, buf_size * 8);
1969 if (show_bits(&gb, 12) == 0xfff) {
1970 if (parse_adts_frame_header(ac, &gb) < 0) {
1971 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1974 if (ac->m4ac.sampling_index > 12) {
1975 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1981 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1982 elem_id = get_bits(&gb, 4);
1984 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1985 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1989 switch (elem_type) {
1992 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1996 err = decode_cpe(ac, &gb, che);
2000 err = decode_cce(ac, &gb, che);
2004 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
2008 err = skip_data_stream_element(ac, &gb);
2012 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2013 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2014 if ((err = decode_pce(ac, new_che_pos, &gb)))
2016 if (ac->output_configured > OC_TRIAL_PCE)
2017 av_log(avccontext, AV_LOG_ERROR,
2018 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2020 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2026 elem_id += get_bits(&gb, 8) - 1;
2027 if (get_bits_left(&gb) < 8 * elem_id) {
2028 av_log(avccontext, AV_LOG_ERROR, overread_err);
2032 elem_id -= decode_extension_payload(ac, &gb, elem_id, che_prev, elem_type_prev);
2033 err = 0; /* FIXME */
2037 err = -1; /* should not happen, but keeps compiler happy */
2042 elem_type_prev = elem_type;
2047 if (get_bits_left(&gb) < 3) {
2048 av_log(avccontext, AV_LOG_ERROR, overread_err);
2053 spectral_to_sample(ac);
2055 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2056 samples <<= multiplier;
2057 if (ac->output_configured < OC_LOCKED) {
2058 avccontext->sample_rate = ac->m4ac.sample_rate << multiplier;
2059 avccontext->frame_size = samples;
2062 data_size_tmp = samples * avccontext->channels * sizeof(int16_t);
2063 if (*data_size < data_size_tmp) {
2064 av_log(avccontext, AV_LOG_ERROR,
2065 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2066 *data_size, data_size_tmp);
2069 *data_size = data_size_tmp;
2071 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avccontext->channels);
2073 if (ac->output_configured)
2074 ac->output_configured = OC_LOCKED;
2076 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2077 return buf_size > buf_consumed ? buf_consumed : buf_size;
2080 static av_cold int aac_decode_close(AVCodecContext *avccontext)
2082 AACContext *ac = avccontext->priv_data;
2085 for (i = 0; i < MAX_ELEM_ID; i++) {
2086 for (type = 0; type < 4; type++) {
2087 if (ac->che[type][i])
2088 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2089 av_freep(&ac->che[type][i]);
2093 ff_mdct_end(&ac->mdct);
2094 ff_mdct_end(&ac->mdct_small);
2098 AVCodec aac_decoder = {
2107 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2108 .sample_fmts = (const enum SampleFormat[]) {
2109 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2111 .channel_layouts = aac_channel_layout,