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
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)
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"
89 #include "cbrt_tablegen.h"
92 #include "mpeg4audio.h"
93 #include "aacadtsdec.h"
101 # include "arm/aac.h"
109 static VLC vlc_scalefactors;
110 static VLC vlc_spectral[11];
112 static const char overread_err[] = "Input buffer exhausted before END element found\n";
114 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
116 // For PCE based channel configurations map the channels solely based on tags.
117 if (!ac->m4ac.chan_config) {
118 return ac->tag_che_map[type][elem_id];
120 // For indexed channel configurations map the channels solely based on position.
121 switch (ac->m4ac.chan_config) {
123 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
125 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
128 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
129 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
130 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
131 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
133 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
136 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
138 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
141 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
143 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
147 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
149 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
150 } else if (ac->m4ac.chan_config == 2) {
154 if (!ac->tags_mapped && type == TYPE_SCE) {
156 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
164 * Check for the channel element in the current channel position configuration.
165 * If it exists, make sure the appropriate element is allocated and map the
166 * channel order to match the internal FFmpeg channel layout.
168 * @param che_pos current channel position configuration
169 * @param type channel element type
170 * @param id channel element id
171 * @param channels count of the number of channels in the configuration
173 * @return Returns error status. 0 - OK, !0 - error
175 static av_cold int che_configure(AACContext *ac,
176 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
180 if (che_pos[type][id]) {
181 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
182 return AVERROR(ENOMEM);
183 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
184 if (type != TYPE_CCE) {
185 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
186 if (type == TYPE_CPE ||
187 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
188 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
192 if (ac->che[type][id])
193 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
194 av_freep(&ac->che[type][id]);
200 * Configure output channel order based on the current program configuration element.
202 * @param che_pos current channel position configuration
203 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
205 * @return Returns error status. 0 - OK, !0 - error
207 static av_cold int output_configure(AACContext *ac,
208 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
209 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
210 int channel_config, enum OCStatus oc_type)
212 AVCodecContext *avctx = ac->avctx;
213 int i, type, channels = 0, ret;
215 if (new_che_pos != che_pos)
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]));
229 avctx->channel_layout = aac_channel_layout[channel_config - 1];
231 /* Allocate or free elements depending on if they are in the
232 * current program configuration.
234 * Set up default 1:1 output mapping.
236 * For a 5.1 stream the output order will be:
237 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
240 for (i = 0; i < MAX_ELEM_ID; i++) {
241 for (type = 0; type < 4; type++) {
242 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
247 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
249 avctx->channel_layout = 0;
252 avctx->channels = channels;
254 ac->output_configured = oc_type;
260 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
262 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
263 * @param sce_map mono (Single Channel Element) map
264 * @param type speaker type/position for these channels
266 static void decode_channel_map(enum ChannelPosition *cpe_map,
267 enum ChannelPosition *sce_map,
268 enum ChannelPosition type,
269 GetBitContext *gb, int n)
272 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
273 map[get_bits(gb, 4)] = type;
278 * Decode program configuration element; reference: table 4.2.
280 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
282 * @return Returns error status. 0 - OK, !0 - error
284 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
287 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
290 skip_bits(gb, 2); // object_type
292 sampling_index = get_bits(gb, 4);
293 if (ac->m4ac.sampling_index != sampling_index)
294 av_log(ac->avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
296 num_front = get_bits(gb, 4);
297 num_side = get_bits(gb, 4);
298 num_back = get_bits(gb, 4);
299 num_lfe = get_bits(gb, 2);
300 num_assoc_data = get_bits(gb, 3);
301 num_cc = get_bits(gb, 4);
304 skip_bits(gb, 4); // mono_mixdown_tag
306 skip_bits(gb, 4); // stereo_mixdown_tag
309 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
311 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
312 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
313 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
314 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
316 skip_bits_long(gb, 4 * num_assoc_data);
318 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
322 /* comment field, first byte is length */
323 comment_len = get_bits(gb, 8) * 8;
324 if (get_bits_left(gb) < comment_len) {
325 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
328 skip_bits_long(gb, comment_len);
333 * Set up channel positions based on a default channel configuration
334 * as specified in table 1.17.
336 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
338 * @return Returns error status. 0 - OK, !0 - error
340 static av_cold int set_default_channel_config(AACContext *ac,
341 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
344 if (channel_config < 1 || channel_config > 7) {
345 av_log(ac->avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
350 /* default channel configurations:
352 * 1ch : front center (mono)
353 * 2ch : L + R (stereo)
354 * 3ch : front center + L + R
355 * 4ch : front center + L + R + back center
356 * 5ch : front center + L + R + back stereo
357 * 6ch : front center + L + R + back stereo + LFE
358 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
361 if (channel_config != 2)
362 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
363 if (channel_config > 1)
364 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
365 if (channel_config == 4)
366 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
367 if (channel_config > 4)
368 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
369 = AAC_CHANNEL_BACK; // back stereo
370 if (channel_config > 5)
371 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
372 if (channel_config == 7)
373 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
379 * Decode GA "General Audio" specific configuration; reference: table 4.1.
381 * @return Returns error status. 0 - OK, !0 - error
383 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
384 MPEG4AudioConfig *m4ac,
387 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
388 int extension_flag, ret;
390 if (get_bits1(gb)) { // frameLengthFlag
391 av_log_missing_feature(ac->avctx, "960/120 MDCT window is", 1);
395 if (get_bits1(gb)) // dependsOnCoreCoder
396 skip_bits(gb, 14); // coreCoderDelay
397 extension_flag = get_bits1(gb);
399 if (m4ac->object_type == AOT_AAC_SCALABLE ||
400 m4ac->object_type == AOT_ER_AAC_SCALABLE)
401 skip_bits(gb, 3); // layerNr
403 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
404 if (channel_config == 0) {
405 skip_bits(gb, 4); // element_instance_tag
406 if ((ret = decode_pce(ac, new_che_pos, gb)))
409 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
412 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
415 if (extension_flag) {
416 switch (m4ac->object_type) {
418 skip_bits(gb, 5); // numOfSubFrame
419 skip_bits(gb, 11); // layer_length
423 case AOT_ER_AAC_SCALABLE:
425 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
426 * aacScalefactorDataResilienceFlag
427 * aacSpectralDataResilienceFlag
431 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
437 * Decode audio specific configuration; reference: table 1.13.
439 * @param data pointer to AVCodecContext extradata
440 * @param data_size size of AVCCodecContext extradata
442 * @return Returns error status or number of consumed bits. <0 - error
444 static int decode_audio_specific_config(AACContext *ac,
445 MPEG4AudioConfig *m4ac, void *data,
451 init_get_bits(&gb, data, data_size * 8);
453 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
455 if (m4ac->sampling_index > 12) {
456 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
459 if (m4ac->sbr == 1 && m4ac->ps == -1)
462 skip_bits_long(&gb, i);
464 switch (m4ac->object_type) {
467 if (decode_ga_specific_config(ac, &gb, m4ac, m4ac->chan_config))
471 av_log(ac->avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
472 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
476 return get_bits_count(&gb);
480 * linear congruential pseudorandom number generator
482 * @param previous_val pointer to the current state of the generator
484 * @return Returns a 32-bit pseudorandom integer
486 static av_always_inline int lcg_random(int previous_val)
488 return previous_val * 1664525 + 1013904223;
491 static av_always_inline void reset_predict_state(PredictorState *ps)
501 static void reset_all_predictors(PredictorState *ps)
504 for (i = 0; i < MAX_PREDICTORS; i++)
505 reset_predict_state(&ps[i]);
508 static void reset_predictor_group(PredictorState *ps, int group_num)
511 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
512 reset_predict_state(&ps[i]);
515 #define AAC_INIT_VLC_STATIC(num, size) \
516 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
517 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
518 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
521 static av_cold int aac_decode_init(AVCodecContext *avctx)
523 AACContext *ac = avctx->priv_data;
526 ac->m4ac.sample_rate = avctx->sample_rate;
528 if (avctx->extradata_size > 0) {
529 if (decode_audio_specific_config(ac, &ac->m4ac, avctx->extradata, avctx->extradata_size) < 0)
533 avctx->sample_fmt = SAMPLE_FMT_S16;
535 AAC_INIT_VLC_STATIC( 0, 304);
536 AAC_INIT_VLC_STATIC( 1, 270);
537 AAC_INIT_VLC_STATIC( 2, 550);
538 AAC_INIT_VLC_STATIC( 3, 300);
539 AAC_INIT_VLC_STATIC( 4, 328);
540 AAC_INIT_VLC_STATIC( 5, 294);
541 AAC_INIT_VLC_STATIC( 6, 306);
542 AAC_INIT_VLC_STATIC( 7, 268);
543 AAC_INIT_VLC_STATIC( 8, 510);
544 AAC_INIT_VLC_STATIC( 9, 366);
545 AAC_INIT_VLC_STATIC(10, 462);
549 dsputil_init(&ac->dsp, avctx);
551 ac->random_state = 0x1f2e3d4c;
553 // -1024 - Compensate wrong IMDCT method.
554 // 32768 - Required to scale values to the correct range for the bias method
555 // for float to int16 conversion.
557 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
558 ac->add_bias = 385.0f;
559 ac->sf_scale = 1. / (-1024. * 32768.);
563 ac->sf_scale = 1. / -1024.;
569 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
570 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
571 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
574 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
575 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
576 // window initialization
577 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
578 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
579 ff_init_ff_sine_windows(10);
580 ff_init_ff_sine_windows( 7);
588 * Skip data_stream_element; reference: table 4.10.
590 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
592 int byte_align = get_bits1(gb);
593 int count = get_bits(gb, 8);
595 count += get_bits(gb, 8);
599 if (get_bits_left(gb) < 8 * count) {
600 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
603 skip_bits_long(gb, 8 * count);
607 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
612 ics->predictor_reset_group = get_bits(gb, 5);
613 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
614 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
618 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
619 ics->prediction_used[sfb] = get_bits1(gb);
625 * Decode Individual Channel Stream info; reference: table 4.6.
627 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
629 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
630 GetBitContext *gb, int common_window)
633 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
634 memset(ics, 0, sizeof(IndividualChannelStream));
637 ics->window_sequence[1] = ics->window_sequence[0];
638 ics->window_sequence[0] = get_bits(gb, 2);
639 ics->use_kb_window[1] = ics->use_kb_window[0];
640 ics->use_kb_window[0] = get_bits1(gb);
641 ics->num_window_groups = 1;
642 ics->group_len[0] = 1;
643 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
645 ics->max_sfb = get_bits(gb, 4);
646 for (i = 0; i < 7; i++) {
648 ics->group_len[ics->num_window_groups - 1]++;
650 ics->num_window_groups++;
651 ics->group_len[ics->num_window_groups - 1] = 1;
654 ics->num_windows = 8;
655 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
656 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
657 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
658 ics->predictor_present = 0;
660 ics->max_sfb = get_bits(gb, 6);
661 ics->num_windows = 1;
662 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
663 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
664 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
665 ics->predictor_present = get_bits1(gb);
666 ics->predictor_reset_group = 0;
667 if (ics->predictor_present) {
668 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
669 if (decode_prediction(ac, ics, gb)) {
670 memset(ics, 0, sizeof(IndividualChannelStream));
673 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
674 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
675 memset(ics, 0, sizeof(IndividualChannelStream));
678 av_log_missing_feature(ac->avctx, "Predictor bit set but LTP is", 1);
679 memset(ics, 0, sizeof(IndividualChannelStream));
685 if (ics->max_sfb > ics->num_swb) {
686 av_log(ac->avctx, AV_LOG_ERROR,
687 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
688 ics->max_sfb, ics->num_swb);
689 memset(ics, 0, sizeof(IndividualChannelStream));
697 * Decode band types (section_data payload); reference: table 4.46.
699 * @param band_type array of the used band type
700 * @param band_type_run_end array of the last scalefactor band of a band type run
702 * @return Returns error status. 0 - OK, !0 - error
704 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
705 int band_type_run_end[120], GetBitContext *gb,
706 IndividualChannelStream *ics)
709 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
710 for (g = 0; g < ics->num_window_groups; g++) {
712 while (k < ics->max_sfb) {
713 uint8_t sect_end = k;
715 int sect_band_type = get_bits(gb, 4);
716 if (sect_band_type == 12) {
717 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
720 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
721 sect_end += sect_len_incr;
722 sect_end += sect_len_incr;
723 if (get_bits_left(gb) < 0) {
724 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
727 if (sect_end > ics->max_sfb) {
728 av_log(ac->avctx, AV_LOG_ERROR,
729 "Number of bands (%d) exceeds limit (%d).\n",
730 sect_end, ics->max_sfb);
733 for (; k < sect_end; k++) {
734 band_type [idx] = sect_band_type;
735 band_type_run_end[idx++] = sect_end;
743 * Decode scalefactors; reference: table 4.47.
745 * @param global_gain first scalefactor value as scalefactors are differentially coded
746 * @param band_type array of the used band type
747 * @param band_type_run_end array of the last scalefactor band of a band type run
748 * @param sf array of scalefactors or intensity stereo positions
750 * @return Returns error status. 0 - OK, !0 - error
752 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
753 unsigned int global_gain,
754 IndividualChannelStream *ics,
755 enum BandType band_type[120],
756 int band_type_run_end[120])
758 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
760 int offset[3] = { global_gain, global_gain - 90, 100 };
762 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
763 for (g = 0; g < ics->num_window_groups; g++) {
764 for (i = 0; i < ics->max_sfb;) {
765 int run_end = band_type_run_end[idx];
766 if (band_type[idx] == ZERO_BT) {
767 for (; i < run_end; i++, idx++)
769 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
770 for (; i < run_end; i++, idx++) {
771 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
772 if (offset[2] > 255U) {
773 av_log(ac->avctx, AV_LOG_ERROR,
774 "%s (%d) out of range.\n", sf_str[2], offset[2]);
777 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
779 } else if (band_type[idx] == NOISE_BT) {
780 for (; i < run_end; i++, idx++) {
781 if (noise_flag-- > 0)
782 offset[1] += get_bits(gb, 9) - 256;
784 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
785 if (offset[1] > 255U) {
786 av_log(ac->avctx, AV_LOG_ERROR,
787 "%s (%d) out of range.\n", sf_str[1], offset[1]);
790 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
793 for (; i < run_end; i++, idx++) {
794 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
795 if (offset[0] > 255U) {
796 av_log(ac->avctx, AV_LOG_ERROR,
797 "%s (%d) out of range.\n", sf_str[0], offset[0]);
800 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
809 * Decode pulse data; reference: table 4.7.
811 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
812 const uint16_t *swb_offset, int num_swb)
815 pulse->num_pulse = get_bits(gb, 2) + 1;
816 pulse_swb = get_bits(gb, 6);
817 if (pulse_swb >= num_swb)
819 pulse->pos[0] = swb_offset[pulse_swb];
820 pulse->pos[0] += get_bits(gb, 5);
821 if (pulse->pos[0] > 1023)
823 pulse->amp[0] = get_bits(gb, 4);
824 for (i = 1; i < pulse->num_pulse; i++) {
825 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
826 if (pulse->pos[i] > 1023)
828 pulse->amp[i] = get_bits(gb, 4);
834 * Decode Temporal Noise Shaping data; reference: table 4.48.
836 * @return Returns error status. 0 - OK, !0 - error
838 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
839 GetBitContext *gb, const IndividualChannelStream *ics)
841 int w, filt, i, coef_len, coef_res, coef_compress;
842 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
843 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
844 for (w = 0; w < ics->num_windows; w++) {
845 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
846 coef_res = get_bits1(gb);
848 for (filt = 0; filt < tns->n_filt[w]; filt++) {
850 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
852 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
853 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
854 tns->order[w][filt], tns_max_order);
855 tns->order[w][filt] = 0;
858 if (tns->order[w][filt]) {
859 tns->direction[w][filt] = get_bits1(gb);
860 coef_compress = get_bits1(gb);
861 coef_len = coef_res + 3 - coef_compress;
862 tmp2_idx = 2 * coef_compress + coef_res;
864 for (i = 0; i < tns->order[w][filt]; i++)
865 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
874 * Decode Mid/Side data; reference: table 4.54.
876 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
877 * [1] mask is decoded from bitstream; [2] mask is all 1s;
878 * [3] reserved for scalable AAC
880 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
884 if (ms_present == 1) {
885 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
886 cpe->ms_mask[idx] = get_bits1(gb);
887 } else if (ms_present == 2) {
888 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
893 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
897 *dst++ = v[idx & 15] * s;
898 *dst++ = v[idx>>4 & 15] * s;
904 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
908 *dst++ = v[idx & 3] * s;
909 *dst++ = v[idx>>2 & 3] * s;
910 *dst++ = v[idx>>4 & 3] * s;
911 *dst++ = v[idx>>6 & 3] * s;
917 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
918 unsigned sign, const float *scale)
920 union float754 s0, s1;
922 s0.f = s1.f = *scale;
923 s0.i ^= sign >> 1 << 31;
926 *dst++ = v[idx & 15] * s0.f;
927 *dst++ = v[idx>>4 & 15] * s1.f;
934 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
935 unsigned sign, const float *scale)
937 unsigned nz = idx >> 12;
938 union float754 s = { .f = *scale };
941 t.i = s.i ^ (sign & 1<<31);
942 *dst++ = v[idx & 3] * t.f;
944 sign <<= nz & 1; nz >>= 1;
945 t.i = s.i ^ (sign & 1<<31);
946 *dst++ = v[idx>>2 & 3] * t.f;
948 sign <<= nz & 1; nz >>= 1;
949 t.i = s.i ^ (sign & 1<<31);
950 *dst++ = v[idx>>4 & 3] * t.f;
952 sign <<= nz & 1; nz >>= 1;
953 t.i = s.i ^ (sign & 1<<31);
954 *dst++ = v[idx>>6 & 3] * t.f;
961 * Decode spectral data; reference: table 4.50.
962 * Dequantize and scale spectral data; reference: 4.6.3.3.
964 * @param coef array of dequantized, scaled spectral data
965 * @param sf array of scalefactors or intensity stereo positions
966 * @param pulse_present set if pulses are present
967 * @param pulse pointer to pulse data struct
968 * @param band_type array of the used band type
970 * @return Returns error status. 0 - OK, !0 - error
972 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
973 GetBitContext *gb, const float sf[120],
974 int pulse_present, const Pulse *pulse,
975 const IndividualChannelStream *ics,
976 enum BandType band_type[120])
978 int i, k, g, idx = 0;
979 const int c = 1024 / ics->num_windows;
980 const uint16_t *offsets = ics->swb_offset;
981 float *coef_base = coef;
983 for (g = 0; g < ics->num_windows; g++)
984 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
986 for (g = 0; g < ics->num_window_groups; g++) {
987 unsigned g_len = ics->group_len[g];
989 for (i = 0; i < ics->max_sfb; i++, idx++) {
990 const unsigned cbt_m1 = band_type[idx] - 1;
991 float *cfo = coef + offsets[i];
992 int off_len = offsets[i + 1] - offsets[i];
995 if (cbt_m1 >= INTENSITY_BT2 - 1) {
996 for (group = 0; group < g_len; group++, cfo+=128) {
997 memset(cfo, 0, off_len * sizeof(float));
999 } else if (cbt_m1 == NOISE_BT - 1) {
1000 for (group = 0; group < g_len; group++, cfo+=128) {
1004 for (k = 0; k < off_len; k++) {
1005 ac->random_state = lcg_random(ac->random_state);
1006 cfo[k] = ac->random_state;
1009 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1010 scale = sf[idx] / sqrtf(band_energy);
1011 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1014 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1015 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1016 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1017 OPEN_READER(re, gb);
1019 switch (cbt_m1 >> 1) {
1021 for (group = 0; group < g_len; group++, cfo+=128) {
1029 UPDATE_CACHE(re, gb);
1030 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1031 cb_idx = cb_vector_idx[code];
1032 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1038 for (group = 0; group < g_len; group++, cfo+=128) {
1048 UPDATE_CACHE(re, gb);
1049 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1050 #if MIN_CACHE_BITS < 20
1051 UPDATE_CACHE(re, gb);
1053 cb_idx = cb_vector_idx[code];
1054 nnz = cb_idx >> 8 & 15;
1055 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1056 LAST_SKIP_BITS(re, gb, nnz);
1057 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1063 for (group = 0; group < g_len; group++, cfo+=128) {
1071 UPDATE_CACHE(re, gb);
1072 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1073 cb_idx = cb_vector_idx[code];
1074 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1081 for (group = 0; group < g_len; group++, cfo+=128) {
1091 UPDATE_CACHE(re, gb);
1092 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1093 cb_idx = cb_vector_idx[code];
1094 nnz = cb_idx >> 8 & 15;
1095 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1096 LAST_SKIP_BITS(re, gb, nnz);
1097 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1103 for (group = 0; group < g_len; group++, cfo+=128) {
1105 uint32_t *icf = (uint32_t *) cf;
1115 UPDATE_CACHE(re, gb);
1116 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1124 cb_idx = cb_vector_idx[code];
1127 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1128 LAST_SKIP_BITS(re, gb, nnz);
1130 for (j = 0; j < 2; j++) {
1134 /* The total length of escape_sequence must be < 22 bits according
1135 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1136 UPDATE_CACHE(re, gb);
1137 b = GET_CACHE(re, gb);
1138 b = 31 - av_log2(~b);
1141 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1145 #if MIN_CACHE_BITS < 21
1146 LAST_SKIP_BITS(re, gb, b + 1);
1147 UPDATE_CACHE(re, gb);
1149 SKIP_BITS(re, gb, b + 1);
1152 n = (1 << b) + SHOW_UBITS(re, gb, b);
1153 LAST_SKIP_BITS(re, gb, b);
1154 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1157 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1158 *icf++ = (bits & 1<<31) | v;
1165 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1169 CLOSE_READER(re, gb);
1175 if (pulse_present) {
1177 for (i = 0; i < pulse->num_pulse; i++) {
1178 float co = coef_base[ pulse->pos[i] ];
1179 while (offsets[idx + 1] <= pulse->pos[i])
1181 if (band_type[idx] != NOISE_BT && sf[idx]) {
1182 float ico = -pulse->amp[i];
1185 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1187 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1194 static av_always_inline float flt16_round(float pf)
1198 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1202 static av_always_inline float flt16_even(float pf)
1206 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1210 static av_always_inline float flt16_trunc(float pf)
1214 pun.i &= 0xFFFF0000U;
1218 static av_always_inline void predict(PredictorState *ps, float *coef,
1219 float sf_scale, float inv_sf_scale,
1222 const float a = 0.953125; // 61.0 / 64
1223 const float alpha = 0.90625; // 29.0 / 32
1227 float r0 = ps->r0, r1 = ps->r1;
1228 float cor0 = ps->cor0, cor1 = ps->cor1;
1229 float var0 = ps->var0, var1 = ps->var1;
1231 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1232 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1234 pv = flt16_round(k1 * r0 + k2 * r1);
1236 *coef += pv * sf_scale;
1238 e0 = *coef * inv_sf_scale;
1241 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1242 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1243 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1244 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1246 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1247 ps->r0 = flt16_trunc(a * e0);
1251 * Apply AAC-Main style frequency domain prediction.
1253 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1256 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1258 if (!sce->ics.predictor_initialized) {
1259 reset_all_predictors(sce->predictor_state);
1260 sce->ics.predictor_initialized = 1;
1263 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1264 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1265 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1266 predict(&sce->predictor_state[k], &sce->coeffs[k],
1267 sf_scale, inv_sf_scale,
1268 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1271 if (sce->ics.predictor_reset_group)
1272 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1274 reset_all_predictors(sce->predictor_state);
1278 * Decode an individual_channel_stream payload; reference: table 4.44.
1280 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1281 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1283 * @return Returns error status. 0 - OK, !0 - error
1285 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1286 GetBitContext *gb, int common_window, int scale_flag)
1289 TemporalNoiseShaping *tns = &sce->tns;
1290 IndividualChannelStream *ics = &sce->ics;
1291 float *out = sce->coeffs;
1292 int global_gain, pulse_present = 0;
1294 /* This assignment is to silence a GCC warning about the variable being used
1295 * uninitialized when in fact it always is.
1297 pulse.num_pulse = 0;
1299 global_gain = get_bits(gb, 8);
1301 if (!common_window && !scale_flag) {
1302 if (decode_ics_info(ac, ics, gb, 0) < 0)
1306 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1308 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1313 if ((pulse_present = get_bits1(gb))) {
1314 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1315 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1318 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1319 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1323 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1325 if (get_bits1(gb)) {
1326 av_log_missing_feature(ac->avctx, "SSR", 1);
1331 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1334 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1335 apply_prediction(ac, sce);
1341 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1343 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1345 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1346 float *ch0 = cpe->ch[0].coeffs;
1347 float *ch1 = cpe->ch[1].coeffs;
1348 int g, i, group, idx = 0;
1349 const uint16_t *offsets = ics->swb_offset;
1350 for (g = 0; g < ics->num_window_groups; g++) {
1351 for (i = 0; i < ics->max_sfb; i++, idx++) {
1352 if (cpe->ms_mask[idx] &&
1353 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1354 for (group = 0; group < ics->group_len[g]; group++) {
1355 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1356 ch1 + group * 128 + offsets[i],
1357 offsets[i+1] - offsets[i]);
1361 ch0 += ics->group_len[g] * 128;
1362 ch1 += ics->group_len[g] * 128;
1367 * intensity stereo decoding; reference: 4.6.8.2.3
1369 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1370 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1371 * [3] reserved for scalable AAC
1373 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1375 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1376 SingleChannelElement *sce1 = &cpe->ch[1];
1377 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1378 const uint16_t *offsets = ics->swb_offset;
1379 int g, group, i, k, idx = 0;
1382 for (g = 0; g < ics->num_window_groups; g++) {
1383 for (i = 0; i < ics->max_sfb;) {
1384 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1385 const int bt_run_end = sce1->band_type_run_end[idx];
1386 for (; i < bt_run_end; i++, idx++) {
1387 c = -1 + 2 * (sce1->band_type[idx] - 14);
1389 c *= 1 - 2 * cpe->ms_mask[idx];
1390 scale = c * sce1->sf[idx];
1391 for (group = 0; group < ics->group_len[g]; group++)
1392 for (k = offsets[i]; k < offsets[i + 1]; k++)
1393 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1396 int bt_run_end = sce1->band_type_run_end[idx];
1397 idx += bt_run_end - i;
1401 coef0 += ics->group_len[g] * 128;
1402 coef1 += ics->group_len[g] * 128;
1407 * Decode a channel_pair_element; reference: table 4.4.
1409 * @return Returns error status. 0 - OK, !0 - error
1411 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1413 int i, ret, common_window, ms_present = 0;
1415 common_window = get_bits1(gb);
1416 if (common_window) {
1417 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1419 i = cpe->ch[1].ics.use_kb_window[0];
1420 cpe->ch[1].ics = cpe->ch[0].ics;
1421 cpe->ch[1].ics.use_kb_window[1] = i;
1422 ms_present = get_bits(gb, 2);
1423 if (ms_present == 3) {
1424 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1426 } else if (ms_present)
1427 decode_mid_side_stereo(cpe, gb, ms_present);
1429 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1431 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1434 if (common_window) {
1436 apply_mid_side_stereo(ac, cpe);
1437 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1438 apply_prediction(ac, &cpe->ch[0]);
1439 apply_prediction(ac, &cpe->ch[1]);
1443 apply_intensity_stereo(cpe, ms_present);
1447 static const float cce_scale[] = {
1448 1.09050773266525765921, //2^(1/8)
1449 1.18920711500272106672, //2^(1/4)
1455 * Decode coupling_channel_element; reference: table 4.8.
1457 * @return Returns error status. 0 - OK, !0 - error
1459 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1465 SingleChannelElement *sce = &che->ch[0];
1466 ChannelCoupling *coup = &che->coup;
1468 coup->coupling_point = 2 * get_bits1(gb);
1469 coup->num_coupled = get_bits(gb, 3);
1470 for (c = 0; c <= coup->num_coupled; c++) {
1472 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1473 coup->id_select[c] = get_bits(gb, 4);
1474 if (coup->type[c] == TYPE_CPE) {
1475 coup->ch_select[c] = get_bits(gb, 2);
1476 if (coup->ch_select[c] == 3)
1479 coup->ch_select[c] = 2;
1481 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1483 sign = get_bits(gb, 1);
1484 scale = cce_scale[get_bits(gb, 2)];
1486 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1489 for (c = 0; c < num_gain; c++) {
1493 float gain_cache = 1.;
1495 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1496 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1497 gain_cache = powf(scale, -gain);
1499 if (coup->coupling_point == AFTER_IMDCT) {
1500 coup->gain[c][0] = gain_cache;
1502 for (g = 0; g < sce->ics.num_window_groups; g++) {
1503 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1504 if (sce->band_type[idx] != ZERO_BT) {
1506 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1514 gain_cache = powf(scale, -t) * s;
1517 coup->gain[c][idx] = gain_cache;
1527 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1529 * @return Returns number of bytes consumed.
1531 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1535 int num_excl_chan = 0;
1538 for (i = 0; i < 7; i++)
1539 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1540 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1542 return num_excl_chan / 7;
1546 * Decode dynamic range information; reference: table 4.52.
1548 * @param cnt length of TYPE_FIL syntactic element in bytes
1550 * @return Returns number of bytes consumed.
1552 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1553 GetBitContext *gb, int cnt)
1556 int drc_num_bands = 1;
1559 /* pce_tag_present? */
1560 if (get_bits1(gb)) {
1561 che_drc->pce_instance_tag = get_bits(gb, 4);
1562 skip_bits(gb, 4); // tag_reserved_bits
1566 /* excluded_chns_present? */
1567 if (get_bits1(gb)) {
1568 n += decode_drc_channel_exclusions(che_drc, gb);
1571 /* drc_bands_present? */
1572 if (get_bits1(gb)) {
1573 che_drc->band_incr = get_bits(gb, 4);
1574 che_drc->interpolation_scheme = get_bits(gb, 4);
1576 drc_num_bands += che_drc->band_incr;
1577 for (i = 0; i < drc_num_bands; i++) {
1578 che_drc->band_top[i] = get_bits(gb, 8);
1583 /* prog_ref_level_present? */
1584 if (get_bits1(gb)) {
1585 che_drc->prog_ref_level = get_bits(gb, 7);
1586 skip_bits1(gb); // prog_ref_level_reserved_bits
1590 for (i = 0; i < drc_num_bands; i++) {
1591 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1592 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1600 * Decode extension data (incomplete); reference: table 4.51.
1602 * @param cnt length of TYPE_FIL syntactic element in bytes
1604 * @return Returns number of bytes consumed
1606 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1607 ChannelElement *che, enum RawDataBlockType elem_type)
1611 switch (get_bits(gb, 4)) { // extension type
1612 case EXT_SBR_DATA_CRC:
1616 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1618 } else if (!ac->m4ac.sbr) {
1619 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1620 skip_bits_long(gb, 8 * cnt - 4);
1622 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1623 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1624 skip_bits_long(gb, 8 * cnt - 4);
1626 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1629 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1633 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1635 case EXT_DYNAMIC_RANGE:
1636 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1640 case EXT_DATA_ELEMENT:
1642 skip_bits_long(gb, 8 * cnt - 4);
1649 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1651 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1652 * @param coef spectral coefficients
1654 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1655 IndividualChannelStream *ics, int decode)
1657 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1659 int bottom, top, order, start, end, size, inc;
1660 float lpc[TNS_MAX_ORDER];
1662 for (w = 0; w < ics->num_windows; w++) {
1663 bottom = ics->num_swb;
1664 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1666 bottom = FFMAX(0, top - tns->length[w][filt]);
1667 order = tns->order[w][filt];
1672 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1674 start = ics->swb_offset[FFMIN(bottom, mmm)];
1675 end = ics->swb_offset[FFMIN( top, mmm)];
1676 if ((size = end - start) <= 0)
1678 if (tns->direction[w][filt]) {
1687 for (m = 0; m < size; m++, start += inc)
1688 for (i = 1; i <= FFMIN(m, order); i++)
1689 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1695 * Conduct IMDCT and windowing.
1697 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
1699 IndividualChannelStream *ics = &sce->ics;
1700 float *in = sce->coeffs;
1701 float *out = sce->ret;
1702 float *saved = sce->saved;
1703 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1704 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1705 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1706 float *buf = ac->buf_mdct;
1707 float *temp = ac->temp;
1711 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1712 for (i = 0; i < 1024; i += 128)
1713 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1715 ff_imdct_half(&ac->mdct, buf, in);
1717 /* window overlapping
1718 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1719 * and long to short transitions are considered to be short to short
1720 * transitions. This leaves just two cases (long to long and short to short)
1721 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1723 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1724 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1725 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, bias, 512);
1727 for (i = 0; i < 448; i++)
1728 out[i] = saved[i] + bias;
1730 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1731 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, bias, 64);
1732 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, bias, 64);
1733 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, bias, 64);
1734 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, bias, 64);
1735 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, bias, 64);
1736 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1738 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, bias, 64);
1739 for (i = 576; i < 1024; i++)
1740 out[i] = buf[i-512] + bias;
1745 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1746 for (i = 0; i < 64; i++)
1747 saved[i] = temp[64 + i] - bias;
1748 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1749 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1750 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1751 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1752 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1753 memcpy( saved, buf + 512, 448 * sizeof(float));
1754 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1755 } else { // LONG_STOP or ONLY_LONG
1756 memcpy( saved, buf + 512, 512 * sizeof(float));
1761 * Apply dependent channel coupling (applied before IMDCT).
1763 * @param index index into coupling gain array
1765 static void apply_dependent_coupling(AACContext *ac,
1766 SingleChannelElement *target,
1767 ChannelElement *cce, int index)
1769 IndividualChannelStream *ics = &cce->ch[0].ics;
1770 const uint16_t *offsets = ics->swb_offset;
1771 float *dest = target->coeffs;
1772 const float *src = cce->ch[0].coeffs;
1773 int g, i, group, k, idx = 0;
1774 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1775 av_log(ac->avctx, AV_LOG_ERROR,
1776 "Dependent coupling is not supported together with LTP\n");
1779 for (g = 0; g < ics->num_window_groups; g++) {
1780 for (i = 0; i < ics->max_sfb; i++, idx++) {
1781 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1782 const float gain = cce->coup.gain[index][idx];
1783 for (group = 0; group < ics->group_len[g]; group++) {
1784 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1786 dest[group * 128 + k] += gain * src[group * 128 + k];
1791 dest += ics->group_len[g] * 128;
1792 src += ics->group_len[g] * 128;
1797 * Apply independent channel coupling (applied after IMDCT).
1799 * @param index index into coupling gain array
1801 static void apply_independent_coupling(AACContext *ac,
1802 SingleChannelElement *target,
1803 ChannelElement *cce, int index)
1806 const float gain = cce->coup.gain[index][0];
1807 const float bias = ac->add_bias;
1808 const float *src = cce->ch[0].ret;
1809 float *dest = target->ret;
1810 const int len = 1024 << (ac->m4ac.sbr == 1);
1812 for (i = 0; i < len; i++)
1813 dest[i] += gain * (src[i] - bias);
1817 * channel coupling transformation interface
1819 * @param apply_coupling_method pointer to (in)dependent coupling function
1821 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1822 enum RawDataBlockType type, int elem_id,
1823 enum CouplingPoint coupling_point,
1824 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1828 for (i = 0; i < MAX_ELEM_ID; i++) {
1829 ChannelElement *cce = ac->che[TYPE_CCE][i];
1832 if (cce && cce->coup.coupling_point == coupling_point) {
1833 ChannelCoupling *coup = &cce->coup;
1835 for (c = 0; c <= coup->num_coupled; c++) {
1836 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1837 if (coup->ch_select[c] != 1) {
1838 apply_coupling_method(ac, &cc->ch[0], cce, index);
1839 if (coup->ch_select[c] != 0)
1842 if (coup->ch_select[c] != 2)
1843 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1845 index += 1 + (coup->ch_select[c] == 3);
1852 * Convert spectral data to float samples, applying all supported tools as appropriate.
1854 static void spectral_to_sample(AACContext *ac)
1857 float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
1858 for (type = 3; type >= 0; type--) {
1859 for (i = 0; i < MAX_ELEM_ID; i++) {
1860 ChannelElement *che = ac->che[type][i];
1862 if (type <= TYPE_CPE)
1863 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1864 if (che->ch[0].tns.present)
1865 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1866 if (che->ch[1].tns.present)
1867 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1868 if (type <= TYPE_CPE)
1869 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1870 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1871 imdct_and_windowing(ac, &che->ch[0], imdct_bias);
1872 if (type == TYPE_CPE) {
1873 imdct_and_windowing(ac, &che->ch[1], imdct_bias);
1875 if (ac->m4ac.sbr > 0) {
1876 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
1879 if (type <= TYPE_CCE)
1880 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1886 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1889 AACADTSHeaderInfo hdr_info;
1891 size = ff_aac_parse_header(gb, &hdr_info);
1893 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1894 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1895 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1896 ac->m4ac.chan_config = hdr_info.chan_config;
1897 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1899 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1901 } else if (ac->output_configured != OC_LOCKED) {
1902 ac->output_configured = OC_NONE;
1904 if (ac->output_configured != OC_LOCKED) {
1908 ac->m4ac.sample_rate = hdr_info.sample_rate;
1909 ac->m4ac.sampling_index = hdr_info.sampling_index;
1910 ac->m4ac.object_type = hdr_info.object_type;
1911 if (!ac->avctx->sample_rate)
1912 ac->avctx->sample_rate = hdr_info.sample_rate;
1913 if (hdr_info.num_aac_frames == 1) {
1914 if (!hdr_info.crc_absent)
1917 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
1924 static int aac_decode_frame(AVCodecContext *avctx, void *data,
1925 int *data_size, AVPacket *avpkt)
1927 const uint8_t *buf = avpkt->data;
1928 int buf_size = avpkt->size;
1929 AACContext *ac = avctx->priv_data;
1930 ChannelElement *che = NULL, *che_prev = NULL;
1932 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1933 int err, elem_id, data_size_tmp;
1935 int samples = 0, multiplier;
1938 init_get_bits(&gb, buf, buf_size * 8);
1940 if (show_bits(&gb, 12) == 0xfff) {
1941 if (parse_adts_frame_header(ac, &gb) < 0) {
1942 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1945 if (ac->m4ac.sampling_index > 12) {
1946 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1951 ac->tags_mapped = 0;
1953 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1954 elem_id = get_bits(&gb, 4);
1956 if (elem_type < TYPE_DSE) {
1957 if (!(che=get_che(ac, elem_type, elem_id))) {
1958 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
1959 elem_type, elem_id);
1965 switch (elem_type) {
1968 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1972 err = decode_cpe(ac, &gb, che);
1976 err = decode_cce(ac, &gb, che);
1980 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1984 err = skip_data_stream_element(ac, &gb);
1988 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1989 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1990 if ((err = decode_pce(ac, new_che_pos, &gb)))
1992 if (ac->output_configured > OC_TRIAL_PCE)
1993 av_log(avctx, AV_LOG_ERROR,
1994 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1996 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2002 elem_id += get_bits(&gb, 8) - 1;
2003 if (get_bits_left(&gb) < 8 * elem_id) {
2004 av_log(avctx, AV_LOG_ERROR, overread_err);
2008 elem_id -= decode_extension_payload(ac, &gb, elem_id, che_prev, elem_type_prev);
2009 err = 0; /* FIXME */
2013 err = -1; /* should not happen, but keeps compiler happy */
2018 elem_type_prev = elem_type;
2023 if (get_bits_left(&gb) < 3) {
2024 av_log(avctx, AV_LOG_ERROR, overread_err);
2029 spectral_to_sample(ac);
2031 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2032 samples <<= multiplier;
2033 if (ac->output_configured < OC_LOCKED) {
2034 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2035 avctx->frame_size = samples;
2038 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2039 if (*data_size < data_size_tmp) {
2040 av_log(avctx, AV_LOG_ERROR,
2041 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2042 *data_size, data_size_tmp);
2045 *data_size = data_size_tmp;
2048 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2050 if (ac->output_configured)
2051 ac->output_configured = OC_LOCKED;
2053 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2054 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2055 if (buf[buf_offset])
2058 return buf_size > buf_offset ? buf_consumed : buf_size;
2061 static av_cold int aac_decode_close(AVCodecContext *avctx)
2063 AACContext *ac = avctx->priv_data;
2066 for (i = 0; i < MAX_ELEM_ID; i++) {
2067 for (type = 0; type < 4; type++) {
2068 if (ac->che[type][i])
2069 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2070 av_freep(&ac->che[type][i]);
2074 ff_mdct_end(&ac->mdct);
2075 ff_mdct_end(&ac->mdct_small);
2079 AVCodec aac_decoder = {
2088 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2089 .sample_fmts = (const enum SampleFormat[]) {
2090 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2092 .channel_layouts = aac_channel_layout,