3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
82 #include "libavutil/float_dsp.h"
83 #include "libavutil/opt.h"
88 #include "fmtconvert.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
101 #include "libavutil/intfloat.h"
109 # include "arm/aac.h"
111 # include "mips/aacdec_mips.h"
114 static VLC vlc_scalefactors;
115 static VLC vlc_spectral[11];
117 static int output_configure(AACContext *ac,
118 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
119 enum OCStatus oc_type, int get_new_frame);
121 #define overread_err "Input buffer exhausted before END element found\n"
123 static int count_channels(uint8_t (*layout)[3], int tags)
126 for (i = 0; i < tags; i++) {
127 int syn_ele = layout[i][0];
128 int pos = layout[i][2];
129 sum += (1 + (syn_ele == TYPE_CPE)) *
130 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
136 * Check for the channel element in the current channel position configuration.
137 * If it exists, make sure the appropriate element is allocated and map the
138 * channel order to match the internal FFmpeg channel layout.
140 * @param che_pos current channel position configuration
141 * @param type channel element type
142 * @param id channel element id
143 * @param channels count of the number of channels in the configuration
145 * @return Returns error status. 0 - OK, !0 - error
147 static av_cold int che_configure(AACContext *ac,
148 enum ChannelPosition che_pos,
149 int type, int id, int *channels)
152 if (!ac->che[type][id]) {
153 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
154 return AVERROR(ENOMEM);
155 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
157 if (type != TYPE_CCE) {
158 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
159 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
160 return AVERROR_INVALIDDATA;
162 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
163 if (type == TYPE_CPE ||
164 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
165 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
169 if (ac->che[type][id])
170 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
171 av_freep(&ac->che[type][id]);
176 static int frame_configure_elements(AVCodecContext *avctx)
178 AACContext *ac = avctx->priv_data;
179 int type, id, ch, ret;
181 /* set channel pointers to internal buffers by default */
182 for (type = 0; type < 4; type++) {
183 for (id = 0; id < MAX_ELEM_ID; id++) {
184 ChannelElement *che = ac->che[type][id];
186 che->ch[0].ret = che->ch[0].ret_buf;
187 che->ch[1].ret = che->ch[1].ret_buf;
192 /* get output buffer */
193 av_frame_unref(ac->frame);
194 ac->frame->nb_samples = 2048;
195 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
198 /* map output channel pointers to AVFrame data */
199 for (ch = 0; ch < avctx->channels; ch++) {
200 if (ac->output_element[ch])
201 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
207 struct elem_to_channel {
208 uint64_t av_position;
211 uint8_t aac_position;
214 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
215 uint8_t (*layout_map)[3], int offset, uint64_t left,
216 uint64_t right, int pos)
218 if (layout_map[offset][0] == TYPE_CPE) {
219 e2c_vec[offset] = (struct elem_to_channel) {
220 .av_position = left | right, .syn_ele = TYPE_CPE,
221 .elem_id = layout_map[offset ][1], .aac_position = pos };
224 e2c_vec[offset] = (struct elem_to_channel) {
225 .av_position = left, .syn_ele = TYPE_SCE,
226 .elem_id = layout_map[offset ][1], .aac_position = pos };
227 e2c_vec[offset + 1] = (struct elem_to_channel) {
228 .av_position = right, .syn_ele = TYPE_SCE,
229 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
234 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
235 int num_pos_channels = 0;
239 for (i = *current; i < tags; i++) {
240 if (layout_map[i][2] != pos)
242 if (layout_map[i][0] == TYPE_CPE) {
244 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
250 num_pos_channels += 2;
258 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
261 return num_pos_channels;
264 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
266 int i, n, total_non_cc_elements;
267 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
268 int num_front_channels, num_side_channels, num_back_channels;
271 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
276 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
277 if (num_front_channels < 0)
280 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
281 if (num_side_channels < 0)
284 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
285 if (num_back_channels < 0)
289 if (num_front_channels & 1) {
290 e2c_vec[i] = (struct elem_to_channel) {
291 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
292 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
294 num_front_channels--;
296 if (num_front_channels >= 4) {
297 i += assign_pair(e2c_vec, layout_map, i,
298 AV_CH_FRONT_LEFT_OF_CENTER,
299 AV_CH_FRONT_RIGHT_OF_CENTER,
301 num_front_channels -= 2;
303 if (num_front_channels >= 2) {
304 i += assign_pair(e2c_vec, layout_map, i,
308 num_front_channels -= 2;
310 while (num_front_channels >= 2) {
311 i += assign_pair(e2c_vec, layout_map, i,
315 num_front_channels -= 2;
318 if (num_side_channels >= 2) {
319 i += assign_pair(e2c_vec, layout_map, i,
323 num_side_channels -= 2;
325 while (num_side_channels >= 2) {
326 i += assign_pair(e2c_vec, layout_map, i,
330 num_side_channels -= 2;
333 while (num_back_channels >= 4) {
334 i += assign_pair(e2c_vec, layout_map, i,
338 num_back_channels -= 2;
340 if (num_back_channels >= 2) {
341 i += assign_pair(e2c_vec, layout_map, i,
345 num_back_channels -= 2;
347 if (num_back_channels) {
348 e2c_vec[i] = (struct elem_to_channel) {
349 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
350 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
355 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
356 e2c_vec[i] = (struct elem_to_channel) {
357 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
358 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
361 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
362 e2c_vec[i] = (struct elem_to_channel) {
363 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
364 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
368 // Must choose a stable sort
369 total_non_cc_elements = n = i;
372 for (i = 1; i < n; i++) {
373 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
374 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
382 for (i = 0; i < total_non_cc_elements; i++) {
383 layout_map[i][0] = e2c_vec[i].syn_ele;
384 layout_map[i][1] = e2c_vec[i].elem_id;
385 layout_map[i][2] = e2c_vec[i].aac_position;
386 if (e2c_vec[i].av_position != UINT64_MAX) {
387 layout |= e2c_vec[i].av_position;
395 * Save current output configuration if and only if it has been locked.
397 static void push_output_configuration(AACContext *ac) {
398 if (ac->oc[1].status == OC_LOCKED) {
399 ac->oc[0] = ac->oc[1];
401 ac->oc[1].status = OC_NONE;
405 * Restore the previous output configuration if and only if the current
406 * configuration is unlocked.
408 static void pop_output_configuration(AACContext *ac) {
409 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
410 ac->oc[1] = ac->oc[0];
411 ac->avctx->channels = ac->oc[1].channels;
412 ac->avctx->channel_layout = ac->oc[1].channel_layout;
413 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
414 ac->oc[1].status, 0);
419 * Configure output channel order based on the current program configuration element.
421 * @return Returns error status. 0 - OK, !0 - error
423 static int output_configure(AACContext *ac,
424 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
425 enum OCStatus oc_type, int get_new_frame)
427 AVCodecContext *avctx = ac->avctx;
428 int i, channels = 0, ret;
431 if (ac->oc[1].layout_map != layout_map) {
432 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
433 ac->oc[1].layout_map_tags = tags;
436 // Try to sniff a reasonable channel order, otherwise output the
437 // channels in the order the PCE declared them.
438 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
439 layout = sniff_channel_order(layout_map, tags);
440 for (i = 0; i < tags; i++) {
441 int type = layout_map[i][0];
442 int id = layout_map[i][1];
443 int position = layout_map[i][2];
444 // Allocate or free elements depending on if they are in the
445 // current program configuration.
446 ret = che_configure(ac, position, type, id, &channels);
450 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
451 if (layout == AV_CH_FRONT_CENTER) {
452 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
458 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
459 if (layout) avctx->channel_layout = layout;
460 ac->oc[1].channel_layout = layout;
461 avctx->channels = ac->oc[1].channels = channels;
462 ac->oc[1].status = oc_type;
465 if ((ret = frame_configure_elements(ac->avctx)) < 0)
472 static void flush(AVCodecContext *avctx)
474 AACContext *ac= avctx->priv_data;
477 for (type = 3; type >= 0; type--) {
478 for (i = 0; i < MAX_ELEM_ID; i++) {
479 ChannelElement *che = ac->che[type][i];
481 for (j = 0; j <= 1; j++) {
482 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
490 * Set up channel positions based on a default channel configuration
491 * as specified in table 1.17.
493 * @return Returns error status. 0 - OK, !0 - error
495 static int set_default_channel_config(AVCodecContext *avctx,
496 uint8_t (*layout_map)[3],
500 if (channel_config < 1 || channel_config > 7) {
501 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
505 *tags = tags_per_config[channel_config];
506 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
510 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
512 // For PCE based channel configurations map the channels solely based on tags.
513 if (!ac->oc[1].m4ac.chan_config) {
514 return ac->tag_che_map[type][elem_id];
516 // Allow single CPE stereo files to be signalled with mono configuration.
517 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
518 uint8_t layout_map[MAX_ELEM_ID*4][3];
520 push_output_configuration(ac);
522 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
524 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
527 if (output_configure(ac, layout_map, layout_map_tags,
528 OC_TRIAL_FRAME, 1) < 0)
531 ac->oc[1].m4ac.chan_config = 2;
532 ac->oc[1].m4ac.ps = 0;
535 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
536 uint8_t layout_map[MAX_ELEM_ID*4][3];
538 push_output_configuration(ac);
540 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
542 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
545 if (output_configure(ac, layout_map, layout_map_tags,
546 OC_TRIAL_FRAME, 1) < 0)
549 ac->oc[1].m4ac.chan_config = 1;
550 if (ac->oc[1].m4ac.sbr)
551 ac->oc[1].m4ac.ps = -1;
553 // For indexed channel configurations map the channels solely based on position.
554 switch (ac->oc[1].m4ac.chan_config) {
556 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
558 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
561 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
562 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
563 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
564 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
566 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
569 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
571 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
574 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
576 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
580 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
582 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
583 } else if (ac->oc[1].m4ac.chan_config == 2) {
587 if (!ac->tags_mapped && type == TYPE_SCE) {
589 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
597 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
599 * @param type speaker type/position for these channels
601 static void decode_channel_map(uint8_t layout_map[][3],
602 enum ChannelPosition type,
603 GetBitContext *gb, int n)
606 enum RawDataBlockType syn_ele;
608 case AAC_CHANNEL_FRONT:
609 case AAC_CHANNEL_BACK:
610 case AAC_CHANNEL_SIDE:
611 syn_ele = get_bits1(gb);
617 case AAC_CHANNEL_LFE:
623 layout_map[0][0] = syn_ele;
624 layout_map[0][1] = get_bits(gb, 4);
625 layout_map[0][2] = type;
631 * Decode program configuration element; reference: table 4.2.
633 * @return Returns error status. 0 - OK, !0 - error
635 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
636 uint8_t (*layout_map)[3],
639 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
643 skip_bits(gb, 2); // object_type
645 sampling_index = get_bits(gb, 4);
646 if (m4ac->sampling_index != sampling_index)
647 av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
649 num_front = get_bits(gb, 4);
650 num_side = get_bits(gb, 4);
651 num_back = get_bits(gb, 4);
652 num_lfe = get_bits(gb, 2);
653 num_assoc_data = get_bits(gb, 3);
654 num_cc = get_bits(gb, 4);
657 skip_bits(gb, 4); // mono_mixdown_tag
659 skip_bits(gb, 4); // stereo_mixdown_tag
662 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
664 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
665 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
668 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
670 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
672 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
674 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
677 skip_bits_long(gb, 4 * num_assoc_data);
679 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
684 /* comment field, first byte is length */
685 comment_len = get_bits(gb, 8) * 8;
686 if (get_bits_left(gb) < comment_len) {
687 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
690 skip_bits_long(gb, comment_len);
695 * Decode GA "General Audio" specific configuration; reference: table 4.1.
697 * @param ac pointer to AACContext, may be null
698 * @param avctx pointer to AVCCodecContext, used for logging
700 * @return Returns error status. 0 - OK, !0 - error
702 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
704 MPEG4AudioConfig *m4ac,
707 int extension_flag, ret;
708 uint8_t layout_map[MAX_ELEM_ID*4][3];
711 if (get_bits1(gb)) { // frameLengthFlag
712 avpriv_request_sample(avctx, "960/120 MDCT window");
713 return AVERROR_PATCHWELCOME;
716 if (get_bits1(gb)) // dependsOnCoreCoder
717 skip_bits(gb, 14); // coreCoderDelay
718 extension_flag = get_bits1(gb);
720 if (m4ac->object_type == AOT_AAC_SCALABLE ||
721 m4ac->object_type == AOT_ER_AAC_SCALABLE)
722 skip_bits(gb, 3); // layerNr
724 if (channel_config == 0) {
725 skip_bits(gb, 4); // element_instance_tag
726 tags = decode_pce(avctx, m4ac, layout_map, gb);
730 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
734 if (count_channels(layout_map, tags) > 1) {
736 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
739 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
742 if (extension_flag) {
743 switch (m4ac->object_type) {
745 skip_bits(gb, 5); // numOfSubFrame
746 skip_bits(gb, 11); // layer_length
750 case AOT_ER_AAC_SCALABLE:
752 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
753 * aacScalefactorDataResilienceFlag
754 * aacSpectralDataResilienceFlag
758 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
764 * Decode audio specific configuration; reference: table 1.13.
766 * @param ac pointer to AACContext, may be null
767 * @param avctx pointer to AVCCodecContext, used for logging
768 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
769 * @param data pointer to buffer holding an audio specific config
770 * @param bit_size size of audio specific config or data in bits
771 * @param sync_extension look for an appended sync extension
773 * @return Returns error status or number of consumed bits. <0 - error
775 static int decode_audio_specific_config(AACContext *ac,
776 AVCodecContext *avctx,
777 MPEG4AudioConfig *m4ac,
778 const uint8_t *data, int bit_size,
785 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
786 for (i = 0; i < bit_size >> 3; i++)
787 av_dlog(avctx, "%02x ", data[i]);
788 av_dlog(avctx, "\n");
790 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
793 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
795 if (m4ac->sampling_index > 12) {
796 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
800 skip_bits_long(&gb, i);
802 switch (m4ac->object_type) {
806 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
810 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
811 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
815 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
816 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
817 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
819 return get_bits_count(&gb);
823 * linear congruential pseudorandom number generator
825 * @param previous_val pointer to the current state of the generator
827 * @return Returns a 32-bit pseudorandom integer
829 static av_always_inline int lcg_random(unsigned previous_val)
831 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
835 static av_always_inline void reset_predict_state(PredictorState *ps)
845 static void reset_all_predictors(PredictorState *ps)
848 for (i = 0; i < MAX_PREDICTORS; i++)
849 reset_predict_state(&ps[i]);
852 static int sample_rate_idx (int rate)
854 if (92017 <= rate) return 0;
855 else if (75132 <= rate) return 1;
856 else if (55426 <= rate) return 2;
857 else if (46009 <= rate) return 3;
858 else if (37566 <= rate) return 4;
859 else if (27713 <= rate) return 5;
860 else if (23004 <= rate) return 6;
861 else if (18783 <= rate) return 7;
862 else if (13856 <= rate) return 8;
863 else if (11502 <= rate) return 9;
864 else if (9391 <= rate) return 10;
868 static void reset_predictor_group(PredictorState *ps, int group_num)
871 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
872 reset_predict_state(&ps[i]);
875 #define AAC_INIT_VLC_STATIC(num, size) \
876 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
877 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
878 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
881 static void aacdec_init(AACContext *ac);
883 static av_cold int aac_decode_init(AVCodecContext *avctx)
885 AACContext *ac = avctx->priv_data;
888 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
892 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
894 if (avctx->extradata_size > 0) {
895 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
897 avctx->extradata_size*8, 1) < 0)
901 uint8_t layout_map[MAX_ELEM_ID*4][3];
904 sr = sample_rate_idx(avctx->sample_rate);
905 ac->oc[1].m4ac.sampling_index = sr;
906 ac->oc[1].m4ac.channels = avctx->channels;
907 ac->oc[1].m4ac.sbr = -1;
908 ac->oc[1].m4ac.ps = -1;
910 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
911 if (ff_mpeg4audio_channels[i] == avctx->channels)
913 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
916 ac->oc[1].m4ac.chan_config = i;
918 if (ac->oc[1].m4ac.chan_config) {
919 int ret = set_default_channel_config(avctx, layout_map,
920 &layout_map_tags, ac->oc[1].m4ac.chan_config);
922 output_configure(ac, layout_map, layout_map_tags,
924 else if (avctx->err_recognition & AV_EF_EXPLODE)
925 return AVERROR_INVALIDDATA;
929 if (avctx->channels > MAX_CHANNELS) {
930 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
931 return AVERROR_INVALIDDATA;
934 AAC_INIT_VLC_STATIC( 0, 304);
935 AAC_INIT_VLC_STATIC( 1, 270);
936 AAC_INIT_VLC_STATIC( 2, 550);
937 AAC_INIT_VLC_STATIC( 3, 300);
938 AAC_INIT_VLC_STATIC( 4, 328);
939 AAC_INIT_VLC_STATIC( 5, 294);
940 AAC_INIT_VLC_STATIC( 6, 306);
941 AAC_INIT_VLC_STATIC( 7, 268);
942 AAC_INIT_VLC_STATIC( 8, 510);
943 AAC_INIT_VLC_STATIC( 9, 366);
944 AAC_INIT_VLC_STATIC(10, 462);
948 ff_fmt_convert_init(&ac->fmt_conv, avctx);
949 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
951 ac->random_state = 0x1f2e3d4c;
955 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
956 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
957 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
960 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
961 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
962 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
963 // window initialization
964 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
965 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
966 ff_init_ff_sine_windows(10);
967 ff_init_ff_sine_windows( 7);
975 * Skip data_stream_element; reference: table 4.10.
977 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
979 int byte_align = get_bits1(gb);
980 int count = get_bits(gb, 8);
982 count += get_bits(gb, 8);
986 if (get_bits_left(gb) < 8 * count) {
987 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
990 skip_bits_long(gb, 8 * count);
994 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
999 ics->predictor_reset_group = get_bits(gb, 5);
1000 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
1001 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
1005 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1006 ics->prediction_used[sfb] = get_bits1(gb);
1012 * Decode Long Term Prediction data; reference: table 4.xx.
1014 static void decode_ltp(LongTermPrediction *ltp,
1015 GetBitContext *gb, uint8_t max_sfb)
1019 ltp->lag = get_bits(gb, 11);
1020 ltp->coef = ltp_coef[get_bits(gb, 3)];
1021 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1022 ltp->used[sfb] = get_bits1(gb);
1026 * Decode Individual Channel Stream info; reference: table 4.6.
1028 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1031 if (get_bits1(gb)) {
1032 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1033 return AVERROR_INVALIDDATA;
1035 ics->window_sequence[1] = ics->window_sequence[0];
1036 ics->window_sequence[0] = get_bits(gb, 2);
1037 ics->use_kb_window[1] = ics->use_kb_window[0];
1038 ics->use_kb_window[0] = get_bits1(gb);
1039 ics->num_window_groups = 1;
1040 ics->group_len[0] = 1;
1041 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1043 ics->max_sfb = get_bits(gb, 4);
1044 for (i = 0; i < 7; i++) {
1045 if (get_bits1(gb)) {
1046 ics->group_len[ics->num_window_groups - 1]++;
1048 ics->num_window_groups++;
1049 ics->group_len[ics->num_window_groups - 1] = 1;
1052 ics->num_windows = 8;
1053 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1054 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1055 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1056 ics->predictor_present = 0;
1058 ics->max_sfb = get_bits(gb, 6);
1059 ics->num_windows = 1;
1060 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1061 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1062 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1063 ics->predictor_present = get_bits1(gb);
1064 ics->predictor_reset_group = 0;
1065 if (ics->predictor_present) {
1066 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1067 if (decode_prediction(ac, ics, gb)) {
1070 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1071 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1074 if ((ics->ltp.present = get_bits(gb, 1)))
1075 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1080 if (ics->max_sfb > ics->num_swb) {
1081 av_log(ac->avctx, AV_LOG_ERROR,
1082 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1083 ics->max_sfb, ics->num_swb);
1090 return AVERROR_INVALIDDATA;
1094 * Decode band types (section_data payload); reference: table 4.46.
1096 * @param band_type array of the used band type
1097 * @param band_type_run_end array of the last scalefactor band of a band type run
1099 * @return Returns error status. 0 - OK, !0 - error
1101 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1102 int band_type_run_end[120], GetBitContext *gb,
1103 IndividualChannelStream *ics)
1106 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1107 for (g = 0; g < ics->num_window_groups; g++) {
1109 while (k < ics->max_sfb) {
1110 uint8_t sect_end = k;
1112 int sect_band_type = get_bits(gb, 4);
1113 if (sect_band_type == 12) {
1114 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1118 sect_len_incr = get_bits(gb, bits);
1119 sect_end += sect_len_incr;
1120 if (get_bits_left(gb) < 0) {
1121 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1124 if (sect_end > ics->max_sfb) {
1125 av_log(ac->avctx, AV_LOG_ERROR,
1126 "Number of bands (%d) exceeds limit (%d).\n",
1127 sect_end, ics->max_sfb);
1130 } while (sect_len_incr == (1 << bits) - 1);
1131 for (; k < sect_end; k++) {
1132 band_type [idx] = sect_band_type;
1133 band_type_run_end[idx++] = sect_end;
1141 * Decode scalefactors; reference: table 4.47.
1143 * @param global_gain first scalefactor value as scalefactors are differentially coded
1144 * @param band_type array of the used band type
1145 * @param band_type_run_end array of the last scalefactor band of a band type run
1146 * @param sf array of scalefactors or intensity stereo positions
1148 * @return Returns error status. 0 - OK, !0 - error
1150 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1151 unsigned int global_gain,
1152 IndividualChannelStream *ics,
1153 enum BandType band_type[120],
1154 int band_type_run_end[120])
1157 int offset[3] = { global_gain, global_gain - 90, 0 };
1160 for (g = 0; g < ics->num_window_groups; g++) {
1161 for (i = 0; i < ics->max_sfb;) {
1162 int run_end = band_type_run_end[idx];
1163 if (band_type[idx] == ZERO_BT) {
1164 for (; i < run_end; i++, idx++)
1166 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1167 for (; i < run_end; i++, idx++) {
1168 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1169 clipped_offset = av_clip(offset[2], -155, 100);
1170 if (offset[2] != clipped_offset) {
1171 avpriv_request_sample(ac->avctx,
1172 "If you heard an audible artifact, there may be a bug in the decoder. "
1173 "Clipped intensity stereo position (%d -> %d)",
1174 offset[2], clipped_offset);
1176 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1178 } else if (band_type[idx] == NOISE_BT) {
1179 for (; i < run_end; i++, idx++) {
1180 if (noise_flag-- > 0)
1181 offset[1] += get_bits(gb, 9) - 256;
1183 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1184 clipped_offset = av_clip(offset[1], -100, 155);
1185 if (offset[1] != clipped_offset) {
1186 avpriv_request_sample(ac->avctx,
1187 "If you heard an audible artifact, there may be a bug in the decoder. "
1188 "Clipped noise gain (%d -> %d)",
1189 offset[1], clipped_offset);
1191 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1194 for (; i < run_end; i++, idx++) {
1195 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1196 if (offset[0] > 255U) {
1197 av_log(ac->avctx, AV_LOG_ERROR,
1198 "Scalefactor (%d) out of range.\n", offset[0]);
1201 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1210 * Decode pulse data; reference: table 4.7.
1212 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1213 const uint16_t *swb_offset, int num_swb)
1216 pulse->num_pulse = get_bits(gb, 2) + 1;
1217 pulse_swb = get_bits(gb, 6);
1218 if (pulse_swb >= num_swb)
1220 pulse->pos[0] = swb_offset[pulse_swb];
1221 pulse->pos[0] += get_bits(gb, 5);
1222 if (pulse->pos[0] > 1023)
1224 pulse->amp[0] = get_bits(gb, 4);
1225 for (i = 1; i < pulse->num_pulse; i++) {
1226 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1227 if (pulse->pos[i] > 1023)
1229 pulse->amp[i] = get_bits(gb, 4);
1235 * Decode Temporal Noise Shaping data; reference: table 4.48.
1237 * @return Returns error status. 0 - OK, !0 - error
1239 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1240 GetBitContext *gb, const IndividualChannelStream *ics)
1242 int w, filt, i, coef_len, coef_res, coef_compress;
1243 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1244 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1245 for (w = 0; w < ics->num_windows; w++) {
1246 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1247 coef_res = get_bits1(gb);
1249 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1251 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1253 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1254 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1255 tns->order[w][filt], tns_max_order);
1256 tns->order[w][filt] = 0;
1259 if (tns->order[w][filt]) {
1260 tns->direction[w][filt] = get_bits1(gb);
1261 coef_compress = get_bits1(gb);
1262 coef_len = coef_res + 3 - coef_compress;
1263 tmp2_idx = 2 * coef_compress + coef_res;
1265 for (i = 0; i < tns->order[w][filt]; i++)
1266 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1275 * Decode Mid/Side data; reference: table 4.54.
1277 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1278 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1279 * [3] reserved for scalable AAC
1281 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1285 if (ms_present == 1) {
1286 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1287 cpe->ms_mask[idx] = get_bits1(gb);
1288 } else if (ms_present == 2) {
1289 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1294 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1298 *dst++ = v[idx & 15] * s;
1299 *dst++ = v[idx>>4 & 15] * s;
1305 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1309 *dst++ = v[idx & 3] * s;
1310 *dst++ = v[idx>>2 & 3] * s;
1311 *dst++ = v[idx>>4 & 3] * s;
1312 *dst++ = v[idx>>6 & 3] * s;
1318 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1319 unsigned sign, const float *scale)
1321 union av_intfloat32 s0, s1;
1323 s0.f = s1.f = *scale;
1324 s0.i ^= sign >> 1 << 31;
1327 *dst++ = v[idx & 15] * s0.f;
1328 *dst++ = v[idx>>4 & 15] * s1.f;
1335 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1336 unsigned sign, const float *scale)
1338 unsigned nz = idx >> 12;
1339 union av_intfloat32 s = { .f = *scale };
1340 union av_intfloat32 t;
1342 t.i = s.i ^ (sign & 1U<<31);
1343 *dst++ = v[idx & 3] * t.f;
1345 sign <<= nz & 1; nz >>= 1;
1346 t.i = s.i ^ (sign & 1U<<31);
1347 *dst++ = v[idx>>2 & 3] * t.f;
1349 sign <<= nz & 1; nz >>= 1;
1350 t.i = s.i ^ (sign & 1U<<31);
1351 *dst++ = v[idx>>4 & 3] * t.f;
1354 t.i = s.i ^ (sign & 1U<<31);
1355 *dst++ = v[idx>>6 & 3] * t.f;
1362 * Decode spectral data; reference: table 4.50.
1363 * Dequantize and scale spectral data; reference: 4.6.3.3.
1365 * @param coef array of dequantized, scaled spectral data
1366 * @param sf array of scalefactors or intensity stereo positions
1367 * @param pulse_present set if pulses are present
1368 * @param pulse pointer to pulse data struct
1369 * @param band_type array of the used band type
1371 * @return Returns error status. 0 - OK, !0 - error
1373 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1374 GetBitContext *gb, const float sf[120],
1375 int pulse_present, const Pulse *pulse,
1376 const IndividualChannelStream *ics,
1377 enum BandType band_type[120])
1379 int i, k, g, idx = 0;
1380 const int c = 1024 / ics->num_windows;
1381 const uint16_t *offsets = ics->swb_offset;
1382 float *coef_base = coef;
1384 for (g = 0; g < ics->num_windows; g++)
1385 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1387 for (g = 0; g < ics->num_window_groups; g++) {
1388 unsigned g_len = ics->group_len[g];
1390 for (i = 0; i < ics->max_sfb; i++, idx++) {
1391 const unsigned cbt_m1 = band_type[idx] - 1;
1392 float *cfo = coef + offsets[i];
1393 int off_len = offsets[i + 1] - offsets[i];
1396 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1397 for (group = 0; group < g_len; group++, cfo+=128) {
1398 memset(cfo, 0, off_len * sizeof(float));
1400 } else if (cbt_m1 == NOISE_BT - 1) {
1401 for (group = 0; group < g_len; group++, cfo+=128) {
1405 for (k = 0; k < off_len; k++) {
1406 ac->random_state = lcg_random(ac->random_state);
1407 cfo[k] = ac->random_state;
1410 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1411 scale = sf[idx] / sqrtf(band_energy);
1412 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1415 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1416 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1417 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1418 OPEN_READER(re, gb);
1420 switch (cbt_m1 >> 1) {
1422 for (group = 0; group < g_len; group++, cfo+=128) {
1430 UPDATE_CACHE(re, gb);
1431 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1432 cb_idx = cb_vector_idx[code];
1433 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1439 for (group = 0; group < g_len; group++, cfo+=128) {
1449 UPDATE_CACHE(re, gb);
1450 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1451 cb_idx = cb_vector_idx[code];
1452 nnz = cb_idx >> 8 & 15;
1453 bits = nnz ? GET_CACHE(re, gb) : 0;
1454 LAST_SKIP_BITS(re, gb, nnz);
1455 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1461 for (group = 0; group < g_len; group++, cfo+=128) {
1469 UPDATE_CACHE(re, gb);
1470 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1471 cb_idx = cb_vector_idx[code];
1472 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1479 for (group = 0; group < g_len; group++, cfo+=128) {
1489 UPDATE_CACHE(re, gb);
1490 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1491 cb_idx = cb_vector_idx[code];
1492 nnz = cb_idx >> 8 & 15;
1493 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1494 LAST_SKIP_BITS(re, gb, nnz);
1495 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1501 for (group = 0; group < g_len; group++, cfo+=128) {
1503 uint32_t *icf = (uint32_t *) cf;
1513 UPDATE_CACHE(re, gb);
1514 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1522 cb_idx = cb_vector_idx[code];
1525 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1526 LAST_SKIP_BITS(re, gb, nnz);
1528 for (j = 0; j < 2; j++) {
1532 /* The total length of escape_sequence must be < 22 bits according
1533 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1534 UPDATE_CACHE(re, gb);
1535 b = GET_CACHE(re, gb);
1536 b = 31 - av_log2(~b);
1539 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1543 SKIP_BITS(re, gb, b + 1);
1545 n = (1 << b) + SHOW_UBITS(re, gb, b);
1546 LAST_SKIP_BITS(re, gb, b);
1547 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1550 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1551 *icf++ = (bits & 1U<<31) | v;
1558 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1562 CLOSE_READER(re, gb);
1568 if (pulse_present) {
1570 for (i = 0; i < pulse->num_pulse; i++) {
1571 float co = coef_base[ pulse->pos[i] ];
1572 while (offsets[idx + 1] <= pulse->pos[i])
1574 if (band_type[idx] != NOISE_BT && sf[idx]) {
1575 float ico = -pulse->amp[i];
1578 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1580 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1587 static av_always_inline float flt16_round(float pf)
1589 union av_intfloat32 tmp;
1591 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1595 static av_always_inline float flt16_even(float pf)
1597 union av_intfloat32 tmp;
1599 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1603 static av_always_inline float flt16_trunc(float pf)
1605 union av_intfloat32 pun;
1607 pun.i &= 0xFFFF0000U;
1611 static av_always_inline void predict(PredictorState *ps, float *coef,
1614 const float a = 0.953125; // 61.0 / 64
1615 const float alpha = 0.90625; // 29.0 / 32
1619 float r0 = ps->r0, r1 = ps->r1;
1620 float cor0 = ps->cor0, cor1 = ps->cor1;
1621 float var0 = ps->var0, var1 = ps->var1;
1623 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1624 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1626 pv = flt16_round(k1 * r0 + k2 * r1);
1633 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1634 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1635 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1636 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1638 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1639 ps->r0 = flt16_trunc(a * e0);
1643 * Apply AAC-Main style frequency domain prediction.
1645 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1649 if (!sce->ics.predictor_initialized) {
1650 reset_all_predictors(sce->predictor_state);
1651 sce->ics.predictor_initialized = 1;
1654 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1655 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1656 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1657 predict(&sce->predictor_state[k], &sce->coeffs[k],
1658 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1661 if (sce->ics.predictor_reset_group)
1662 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1664 reset_all_predictors(sce->predictor_state);
1668 * Decode an individual_channel_stream payload; reference: table 4.44.
1670 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1671 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1673 * @return Returns error status. 0 - OK, !0 - error
1675 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1676 GetBitContext *gb, int common_window, int scale_flag)
1679 TemporalNoiseShaping *tns = &sce->tns;
1680 IndividualChannelStream *ics = &sce->ics;
1681 float *out = sce->coeffs;
1682 int global_gain, pulse_present = 0;
1684 /* This assignment is to silence a GCC warning about the variable being used
1685 * uninitialized when in fact it always is.
1687 pulse.num_pulse = 0;
1689 global_gain = get_bits(gb, 8);
1691 if (!common_window && !scale_flag) {
1692 if (decode_ics_info(ac, ics, gb) < 0)
1693 return AVERROR_INVALIDDATA;
1696 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1698 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1703 if ((pulse_present = get_bits1(gb))) {
1704 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1705 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1708 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1709 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1713 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1715 if (get_bits1(gb)) {
1716 avpriv_request_sample(ac->avctx, "SSR");
1717 return AVERROR_PATCHWELCOME;
1721 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1724 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1725 apply_prediction(ac, sce);
1731 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1733 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1735 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1736 float *ch0 = cpe->ch[0].coeffs;
1737 float *ch1 = cpe->ch[1].coeffs;
1738 int g, i, group, idx = 0;
1739 const uint16_t *offsets = ics->swb_offset;
1740 for (g = 0; g < ics->num_window_groups; g++) {
1741 for (i = 0; i < ics->max_sfb; i++, idx++) {
1742 if (cpe->ms_mask[idx] &&
1743 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1744 for (group = 0; group < ics->group_len[g]; group++) {
1745 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1746 ch1 + group * 128 + offsets[i],
1747 offsets[i+1] - offsets[i]);
1751 ch0 += ics->group_len[g] * 128;
1752 ch1 += ics->group_len[g] * 128;
1757 * intensity stereo decoding; reference: 4.6.8.2.3
1759 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1760 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1761 * [3] reserved for scalable AAC
1763 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1765 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1766 SingleChannelElement *sce1 = &cpe->ch[1];
1767 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1768 const uint16_t *offsets = ics->swb_offset;
1769 int g, group, i, idx = 0;
1772 for (g = 0; g < ics->num_window_groups; g++) {
1773 for (i = 0; i < ics->max_sfb;) {
1774 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1775 const int bt_run_end = sce1->band_type_run_end[idx];
1776 for (; i < bt_run_end; i++, idx++) {
1777 c = -1 + 2 * (sce1->band_type[idx] - 14);
1779 c *= 1 - 2 * cpe->ms_mask[idx];
1780 scale = c * sce1->sf[idx];
1781 for (group = 0; group < ics->group_len[g]; group++)
1782 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1783 coef0 + group * 128 + offsets[i],
1785 offsets[i + 1] - offsets[i]);
1788 int bt_run_end = sce1->band_type_run_end[idx];
1789 idx += bt_run_end - i;
1793 coef0 += ics->group_len[g] * 128;
1794 coef1 += ics->group_len[g] * 128;
1799 * Decode a channel_pair_element; reference: table 4.4.
1801 * @return Returns error status. 0 - OK, !0 - error
1803 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1805 int i, ret, common_window, ms_present = 0;
1807 common_window = get_bits1(gb);
1808 if (common_window) {
1809 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1810 return AVERROR_INVALIDDATA;
1811 i = cpe->ch[1].ics.use_kb_window[0];
1812 cpe->ch[1].ics = cpe->ch[0].ics;
1813 cpe->ch[1].ics.use_kb_window[1] = i;
1814 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1815 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1816 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1817 ms_present = get_bits(gb, 2);
1818 if (ms_present == 3) {
1819 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1821 } else if (ms_present)
1822 decode_mid_side_stereo(cpe, gb, ms_present);
1824 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1826 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1829 if (common_window) {
1831 apply_mid_side_stereo(ac, cpe);
1832 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1833 apply_prediction(ac, &cpe->ch[0]);
1834 apply_prediction(ac, &cpe->ch[1]);
1838 apply_intensity_stereo(ac, cpe, ms_present);
1842 static const float cce_scale[] = {
1843 1.09050773266525765921, //2^(1/8)
1844 1.18920711500272106672, //2^(1/4)
1850 * Decode coupling_channel_element; reference: table 4.8.
1852 * @return Returns error status. 0 - OK, !0 - error
1854 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1860 SingleChannelElement *sce = &che->ch[0];
1861 ChannelCoupling *coup = &che->coup;
1863 coup->coupling_point = 2 * get_bits1(gb);
1864 coup->num_coupled = get_bits(gb, 3);
1865 for (c = 0; c <= coup->num_coupled; c++) {
1867 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1868 coup->id_select[c] = get_bits(gb, 4);
1869 if (coup->type[c] == TYPE_CPE) {
1870 coup->ch_select[c] = get_bits(gb, 2);
1871 if (coup->ch_select[c] == 3)
1874 coup->ch_select[c] = 2;
1876 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1878 sign = get_bits(gb, 1);
1879 scale = cce_scale[get_bits(gb, 2)];
1881 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1884 for (c = 0; c < num_gain; c++) {
1888 float gain_cache = 1.;
1890 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1891 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1892 gain_cache = powf(scale, -gain);
1894 if (coup->coupling_point == AFTER_IMDCT) {
1895 coup->gain[c][0] = gain_cache;
1897 for (g = 0; g < sce->ics.num_window_groups; g++) {
1898 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1899 if (sce->band_type[idx] != ZERO_BT) {
1901 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1909 gain_cache = powf(scale, -t) * s;
1912 coup->gain[c][idx] = gain_cache;
1922 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1924 * @return Returns number of bytes consumed.
1926 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1930 int num_excl_chan = 0;
1933 for (i = 0; i < 7; i++)
1934 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1935 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1937 return num_excl_chan / 7;
1941 * Decode dynamic range information; reference: table 4.52.
1943 * @return Returns number of bytes consumed.
1945 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1949 int drc_num_bands = 1;
1952 /* pce_tag_present? */
1953 if (get_bits1(gb)) {
1954 che_drc->pce_instance_tag = get_bits(gb, 4);
1955 skip_bits(gb, 4); // tag_reserved_bits
1959 /* excluded_chns_present? */
1960 if (get_bits1(gb)) {
1961 n += decode_drc_channel_exclusions(che_drc, gb);
1964 /* drc_bands_present? */
1965 if (get_bits1(gb)) {
1966 che_drc->band_incr = get_bits(gb, 4);
1967 che_drc->interpolation_scheme = get_bits(gb, 4);
1969 drc_num_bands += che_drc->band_incr;
1970 for (i = 0; i < drc_num_bands; i++) {
1971 che_drc->band_top[i] = get_bits(gb, 8);
1976 /* prog_ref_level_present? */
1977 if (get_bits1(gb)) {
1978 che_drc->prog_ref_level = get_bits(gb, 7);
1979 skip_bits1(gb); // prog_ref_level_reserved_bits
1983 for (i = 0; i < drc_num_bands; i++) {
1984 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1985 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1992 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1994 int i, major, minor;
1999 get_bits(gb, 13); len -= 13;
2001 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2002 buf[i] = get_bits(gb, 8);
2005 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2006 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2008 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2009 ac->avctx->internal->skip_samples = 1024;
2013 skip_bits_long(gb, len);
2019 * Decode extension data (incomplete); reference: table 4.51.
2021 * @param cnt length of TYPE_FIL syntactic element in bytes
2023 * @return Returns number of bytes consumed
2025 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2026 ChannelElement *che, enum RawDataBlockType elem_type)
2030 switch (get_bits(gb, 4)) { // extension type
2031 case EXT_SBR_DATA_CRC:
2035 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2037 } else if (!ac->oc[1].m4ac.sbr) {
2038 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2039 skip_bits_long(gb, 8 * cnt - 4);
2041 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2042 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2043 skip_bits_long(gb, 8 * cnt - 4);
2045 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2046 ac->oc[1].m4ac.sbr = 1;
2047 ac->oc[1].m4ac.ps = 1;
2048 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2049 ac->oc[1].status, 1);
2051 ac->oc[1].m4ac.sbr = 1;
2053 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2055 case EXT_DYNAMIC_RANGE:
2056 res = decode_dynamic_range(&ac->che_drc, gb);
2059 decode_fill(ac, gb, 8 * cnt - 4);
2062 case EXT_DATA_ELEMENT:
2064 skip_bits_long(gb, 8 * cnt - 4);
2071 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2073 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2074 * @param coef spectral coefficients
2076 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2077 IndividualChannelStream *ics, int decode)
2079 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2081 int bottom, top, order, start, end, size, inc;
2082 float lpc[TNS_MAX_ORDER];
2083 float tmp[TNS_MAX_ORDER+1];
2085 for (w = 0; w < ics->num_windows; w++) {
2086 bottom = ics->num_swb;
2087 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2089 bottom = FFMAX(0, top - tns->length[w][filt]);
2090 order = tns->order[w][filt];
2095 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2097 start = ics->swb_offset[FFMIN(bottom, mmm)];
2098 end = ics->swb_offset[FFMIN( top, mmm)];
2099 if ((size = end - start) <= 0)
2101 if (tns->direction[w][filt]) {
2111 for (m = 0; m < size; m++, start += inc)
2112 for (i = 1; i <= FFMIN(m, order); i++)
2113 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2116 for (m = 0; m < size; m++, start += inc) {
2117 tmp[0] = coef[start];
2118 for (i = 1; i <= FFMIN(m, order); i++)
2119 coef[start] += tmp[i] * lpc[i - 1];
2120 for (i = order; i > 0; i--)
2121 tmp[i] = tmp[i - 1];
2129 * Apply windowing and MDCT to obtain the spectral
2130 * coefficient from the predicted sample by LTP.
2132 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2133 float *in, IndividualChannelStream *ics)
2135 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2136 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2137 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2138 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2140 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2141 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2143 memset(in, 0, 448 * sizeof(float));
2144 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2146 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2147 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2149 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2150 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2152 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2156 * Apply the long term prediction
2158 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2160 const LongTermPrediction *ltp = &sce->ics.ltp;
2161 const uint16_t *offsets = sce->ics.swb_offset;
2164 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2165 float *predTime = sce->ret;
2166 float *predFreq = ac->buf_mdct;
2167 int16_t num_samples = 2048;
2169 if (ltp->lag < 1024)
2170 num_samples = ltp->lag + 1024;
2171 for (i = 0; i < num_samples; i++)
2172 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2173 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2175 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2177 if (sce->tns.present)
2178 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2180 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2182 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2183 sce->coeffs[i] += predFreq[i];
2188 * Update the LTP buffer for next frame
2190 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2192 IndividualChannelStream *ics = &sce->ics;
2193 float *saved = sce->saved;
2194 float *saved_ltp = sce->coeffs;
2195 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2196 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2199 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2200 memcpy(saved_ltp, saved, 512 * sizeof(float));
2201 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2202 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2203 for (i = 0; i < 64; i++)
2204 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2205 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2206 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2207 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2208 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2209 for (i = 0; i < 64; i++)
2210 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2211 } else { // LONG_STOP or ONLY_LONG
2212 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2213 for (i = 0; i < 512; i++)
2214 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2217 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2218 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2219 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2223 * Conduct IMDCT and windowing.
2225 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2227 IndividualChannelStream *ics = &sce->ics;
2228 float *in = sce->coeffs;
2229 float *out = sce->ret;
2230 float *saved = sce->saved;
2231 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2232 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2233 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2234 float *buf = ac->buf_mdct;
2235 float *temp = ac->temp;
2239 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2240 for (i = 0; i < 1024; i += 128)
2241 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2243 ac->mdct.imdct_half(&ac->mdct, buf, in);
2245 /* window overlapping
2246 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2247 * and long to short transitions are considered to be short to short
2248 * transitions. This leaves just two cases (long to long and short to short)
2249 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2251 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2252 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2253 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2255 memcpy( out, saved, 448 * sizeof(float));
2257 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2258 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2259 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2260 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2261 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2262 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2263 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2265 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2266 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2271 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2272 memcpy( saved, temp + 64, 64 * sizeof(float));
2273 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2274 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2275 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2276 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2277 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2278 memcpy( saved, buf + 512, 448 * sizeof(float));
2279 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2280 } else { // LONG_STOP or ONLY_LONG
2281 memcpy( saved, buf + 512, 512 * sizeof(float));
2286 * Apply dependent channel coupling (applied before IMDCT).
2288 * @param index index into coupling gain array
2290 static void apply_dependent_coupling(AACContext *ac,
2291 SingleChannelElement *target,
2292 ChannelElement *cce, int index)
2294 IndividualChannelStream *ics = &cce->ch[0].ics;
2295 const uint16_t *offsets = ics->swb_offset;
2296 float *dest = target->coeffs;
2297 const float *src = cce->ch[0].coeffs;
2298 int g, i, group, k, idx = 0;
2299 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2300 av_log(ac->avctx, AV_LOG_ERROR,
2301 "Dependent coupling is not supported together with LTP\n");
2304 for (g = 0; g < ics->num_window_groups; g++) {
2305 for (i = 0; i < ics->max_sfb; i++, idx++) {
2306 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2307 const float gain = cce->coup.gain[index][idx];
2308 for (group = 0; group < ics->group_len[g]; group++) {
2309 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2311 dest[group * 128 + k] += gain * src[group * 128 + k];
2316 dest += ics->group_len[g] * 128;
2317 src += ics->group_len[g] * 128;
2322 * Apply independent channel coupling (applied after IMDCT).
2324 * @param index index into coupling gain array
2326 static void apply_independent_coupling(AACContext *ac,
2327 SingleChannelElement *target,
2328 ChannelElement *cce, int index)
2331 const float gain = cce->coup.gain[index][0];
2332 const float *src = cce->ch[0].ret;
2333 float *dest = target->ret;
2334 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2336 for (i = 0; i < len; i++)
2337 dest[i] += gain * src[i];
2341 * channel coupling transformation interface
2343 * @param apply_coupling_method pointer to (in)dependent coupling function
2345 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2346 enum RawDataBlockType type, int elem_id,
2347 enum CouplingPoint coupling_point,
2348 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2352 for (i = 0; i < MAX_ELEM_ID; i++) {
2353 ChannelElement *cce = ac->che[TYPE_CCE][i];
2356 if (cce && cce->coup.coupling_point == coupling_point) {
2357 ChannelCoupling *coup = &cce->coup;
2359 for (c = 0; c <= coup->num_coupled; c++) {
2360 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2361 if (coup->ch_select[c] != 1) {
2362 apply_coupling_method(ac, &cc->ch[0], cce, index);
2363 if (coup->ch_select[c] != 0)
2366 if (coup->ch_select[c] != 2)
2367 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2369 index += 1 + (coup->ch_select[c] == 3);
2376 * Convert spectral data to float samples, applying all supported tools as appropriate.
2378 static void spectral_to_sample(AACContext *ac)
2381 for (type = 3; type >= 0; type--) {
2382 for (i = 0; i < MAX_ELEM_ID; i++) {
2383 ChannelElement *che = ac->che[type][i];
2385 if (type <= TYPE_CPE)
2386 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2387 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2388 if (che->ch[0].ics.predictor_present) {
2389 if (che->ch[0].ics.ltp.present)
2390 ac->apply_ltp(ac, &che->ch[0]);
2391 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2392 ac->apply_ltp(ac, &che->ch[1]);
2395 if (che->ch[0].tns.present)
2396 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2397 if (che->ch[1].tns.present)
2398 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2399 if (type <= TYPE_CPE)
2400 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2401 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2402 ac->imdct_and_windowing(ac, &che->ch[0]);
2403 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2404 ac->update_ltp(ac, &che->ch[0]);
2405 if (type == TYPE_CPE) {
2406 ac->imdct_and_windowing(ac, &che->ch[1]);
2407 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2408 ac->update_ltp(ac, &che->ch[1]);
2410 if (ac->oc[1].m4ac.sbr > 0) {
2411 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2414 if (type <= TYPE_CCE)
2415 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2421 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2424 AACADTSHeaderInfo hdr_info;
2425 uint8_t layout_map[MAX_ELEM_ID*4][3];
2426 int layout_map_tags;
2428 size = avpriv_aac_parse_header(gb, &hdr_info);
2430 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2431 // This is 2 for "VLB " audio in NSV files.
2432 // See samples/nsv/vlb_audio.
2433 avpriv_report_missing_feature(ac->avctx,
2434 "More than one AAC RDB per ADTS frame");
2435 ac->warned_num_aac_frames = 1;
2437 push_output_configuration(ac);
2438 if (hdr_info.chan_config) {
2439 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2440 if (set_default_channel_config(ac->avctx, layout_map,
2441 &layout_map_tags, hdr_info.chan_config))
2443 if (output_configure(ac, layout_map, layout_map_tags,
2444 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2447 ac->oc[1].m4ac.chan_config = 0;
2449 * dual mono frames in Japanese DTV can have chan_config 0
2450 * WITHOUT specifying PCE.
2451 * thus, set dual mono as default.
2453 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2454 layout_map_tags = 2;
2455 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2456 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2457 layout_map[0][1] = 0;
2458 layout_map[1][1] = 1;
2459 if (output_configure(ac, layout_map, layout_map_tags,
2464 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2465 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2466 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2467 if (ac->oc[0].status != OC_LOCKED ||
2468 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2469 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2470 ac->oc[1].m4ac.sbr = -1;
2471 ac->oc[1].m4ac.ps = -1;
2473 if (!hdr_info.crc_absent)
2479 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2480 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2482 AACContext *ac = avctx->priv_data;
2483 ChannelElement *che = NULL, *che_prev = NULL;
2484 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2486 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2487 int is_dmono, sce_count = 0;
2491 if (show_bits(gb, 12) == 0xfff) {
2492 if (parse_adts_frame_header(ac, gb) < 0) {
2493 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2497 if (ac->oc[1].m4ac.sampling_index > 12) {
2498 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2504 if (frame_configure_elements(avctx) < 0) {
2509 ac->tags_mapped = 0;
2511 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2512 elem_id = get_bits(gb, 4);
2514 if (elem_type < TYPE_DSE) {
2515 if (!(che=get_che(ac, elem_type, elem_id))) {
2516 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2517 elem_type, elem_id);
2524 switch (elem_type) {
2527 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2533 err = decode_cpe(ac, gb, che);
2538 err = decode_cce(ac, gb, che);
2542 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2547 err = skip_data_stream_element(ac, gb);
2551 uint8_t layout_map[MAX_ELEM_ID*4][3];
2553 push_output_configuration(ac);
2554 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2560 av_log(avctx, AV_LOG_ERROR,
2561 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2563 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2565 ac->oc[1].m4ac.chan_config = 0;
2573 elem_id += get_bits(gb, 8) - 1;
2574 if (get_bits_left(gb) < 8 * elem_id) {
2575 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2580 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2581 err = 0; /* FIXME */
2585 err = -1; /* should not happen, but keeps compiler happy */
2590 elem_type_prev = elem_type;
2595 if (get_bits_left(gb) < 3) {
2596 av_log(avctx, AV_LOG_ERROR, overread_err);
2602 spectral_to_sample(ac);
2604 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2605 samples <<= multiplier;
2606 /* for dual-mono audio (SCE + SCE) */
2607 is_dmono = ac->dmono_mode && sce_count == 2 &&
2608 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2611 ac->frame->nb_samples = samples;
2612 *got_frame_ptr = !!samples;
2615 if (ac->dmono_mode == 1)
2616 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2617 else if (ac->dmono_mode == 2)
2618 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2621 if (ac->oc[1].status && audio_found) {
2622 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2623 avctx->frame_size = samples;
2624 ac->oc[1].status = OC_LOCKED;
2629 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2630 if (side && side_size>=4)
2631 AV_WL32(side, 2*AV_RL32(side));
2635 pop_output_configuration(ac);
2639 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2640 int *got_frame_ptr, AVPacket *avpkt)
2642 AACContext *ac = avctx->priv_data;
2643 const uint8_t *buf = avpkt->data;
2644 int buf_size = avpkt->size;
2649 int new_extradata_size;
2650 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2651 AV_PKT_DATA_NEW_EXTRADATA,
2652 &new_extradata_size);
2653 int jp_dualmono_size;
2654 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2655 AV_PKT_DATA_JP_DUALMONO,
2658 if (new_extradata && 0) {
2659 av_free(avctx->extradata);
2660 avctx->extradata = av_mallocz(new_extradata_size +
2661 FF_INPUT_BUFFER_PADDING_SIZE);
2662 if (!avctx->extradata)
2663 return AVERROR(ENOMEM);
2664 avctx->extradata_size = new_extradata_size;
2665 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2666 push_output_configuration(ac);
2667 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2669 avctx->extradata_size*8, 1) < 0) {
2670 pop_output_configuration(ac);
2671 return AVERROR_INVALIDDATA;
2676 if (jp_dualmono && jp_dualmono_size > 0)
2677 ac->dmono_mode = 1 + *jp_dualmono;
2678 if (ac->force_dmono_mode >= 0)
2679 ac->dmono_mode = ac->force_dmono_mode;
2681 if (INT_MAX / 8 <= buf_size)
2682 return AVERROR_INVALIDDATA;
2684 init_get_bits(&gb, buf, buf_size * 8);
2686 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2689 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2690 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2691 if (buf[buf_offset])
2694 return buf_size > buf_offset ? buf_consumed : buf_size;
2697 static av_cold int aac_decode_close(AVCodecContext *avctx)
2699 AACContext *ac = avctx->priv_data;
2702 for (i = 0; i < MAX_ELEM_ID; i++) {
2703 for (type = 0; type < 4; type++) {
2704 if (ac->che[type][i])
2705 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2706 av_freep(&ac->che[type][i]);
2710 ff_mdct_end(&ac->mdct);
2711 ff_mdct_end(&ac->mdct_small);
2712 ff_mdct_end(&ac->mdct_ltp);
2717 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2719 struct LATMContext {
2720 AACContext aac_ctx; ///< containing AACContext
2721 int initialized; ///< initialized after a valid extradata was seen
2724 int audio_mux_version_A; ///< LATM syntax version
2725 int frame_length_type; ///< 0/1 variable/fixed frame length
2726 int frame_length; ///< frame length for fixed frame length
2729 static inline uint32_t latm_get_value(GetBitContext *b)
2731 int length = get_bits(b, 2);
2733 return get_bits_long(b, (length+1)*8);
2736 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2737 GetBitContext *gb, int asclen)
2739 AACContext *ac = &latmctx->aac_ctx;
2740 AVCodecContext *avctx = ac->avctx;
2741 MPEG4AudioConfig m4ac = { 0 };
2742 int config_start_bit = get_bits_count(gb);
2743 int sync_extension = 0;
2744 int bits_consumed, esize;
2748 asclen = FFMIN(asclen, get_bits_left(gb));
2750 asclen = get_bits_left(gb);
2752 if (config_start_bit % 8) {
2753 avpriv_request_sample(latmctx->aac_ctx.avctx,
2754 "Non-byte-aligned audio-specific config");
2755 return AVERROR_PATCHWELCOME;
2758 return AVERROR_INVALIDDATA;
2759 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2760 gb->buffer + (config_start_bit / 8),
2761 asclen, sync_extension);
2763 if (bits_consumed < 0)
2764 return AVERROR_INVALIDDATA;
2766 if (!latmctx->initialized ||
2767 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2768 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2770 if(latmctx->initialized) {
2771 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2773 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
2775 latmctx->initialized = 0;
2777 esize = (bits_consumed+7) / 8;
2779 if (avctx->extradata_size < esize) {
2780 av_free(avctx->extradata);
2781 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2782 if (!avctx->extradata)
2783 return AVERROR(ENOMEM);
2786 avctx->extradata_size = esize;
2787 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2788 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2790 skip_bits_long(gb, bits_consumed);
2792 return bits_consumed;
2795 static int read_stream_mux_config(struct LATMContext *latmctx,
2798 int ret, audio_mux_version = get_bits(gb, 1);
2800 latmctx->audio_mux_version_A = 0;
2801 if (audio_mux_version)
2802 latmctx->audio_mux_version_A = get_bits(gb, 1);
2804 if (!latmctx->audio_mux_version_A) {
2806 if (audio_mux_version)
2807 latm_get_value(gb); // taraFullness
2809 skip_bits(gb, 1); // allStreamSameTimeFraming
2810 skip_bits(gb, 6); // numSubFrames
2812 if (get_bits(gb, 4)) { // numPrograms
2813 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
2814 return AVERROR_PATCHWELCOME;
2817 // for each program (which there is only one in DVB)
2819 // for each layer (which there is only one in DVB)
2820 if (get_bits(gb, 3)) { // numLayer
2821 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
2822 return AVERROR_PATCHWELCOME;
2825 // for all but first stream: use_same_config = get_bits(gb, 1);
2826 if (!audio_mux_version) {
2827 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2830 int ascLen = latm_get_value(gb);
2831 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2834 skip_bits_long(gb, ascLen);
2837 latmctx->frame_length_type = get_bits(gb, 3);
2838 switch (latmctx->frame_length_type) {
2840 skip_bits(gb, 8); // latmBufferFullness
2843 latmctx->frame_length = get_bits(gb, 9);
2848 skip_bits(gb, 6); // CELP frame length table index
2852 skip_bits(gb, 1); // HVXC frame length table index
2856 if (get_bits(gb, 1)) { // other data
2857 if (audio_mux_version) {
2858 latm_get_value(gb); // other_data_bits
2862 esc = get_bits(gb, 1);
2868 if (get_bits(gb, 1)) // crc present
2869 skip_bits(gb, 8); // config_crc
2875 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2879 if (ctx->frame_length_type == 0) {
2880 int mux_slot_length = 0;
2882 tmp = get_bits(gb, 8);
2883 mux_slot_length += tmp;
2884 } while (tmp == 255);
2885 return mux_slot_length;
2886 } else if (ctx->frame_length_type == 1) {
2887 return ctx->frame_length;
2888 } else if (ctx->frame_length_type == 3 ||
2889 ctx->frame_length_type == 5 ||
2890 ctx->frame_length_type == 7) {
2891 skip_bits(gb, 2); // mux_slot_length_coded
2896 static int read_audio_mux_element(struct LATMContext *latmctx,
2900 uint8_t use_same_mux = get_bits(gb, 1);
2901 if (!use_same_mux) {
2902 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2904 } else if (!latmctx->aac_ctx.avctx->extradata) {
2905 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2906 "no decoder config found\n");
2907 return AVERROR(EAGAIN);
2909 if (latmctx->audio_mux_version_A == 0) {
2910 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2911 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2912 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2913 return AVERROR_INVALIDDATA;
2914 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2915 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2916 "frame length mismatch %d << %d\n",
2917 mux_slot_length_bytes * 8, get_bits_left(gb));
2918 return AVERROR_INVALIDDATA;
2925 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2926 int *got_frame_ptr, AVPacket *avpkt)
2928 struct LATMContext *latmctx = avctx->priv_data;
2932 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
2935 // check for LOAS sync word
2936 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2937 return AVERROR_INVALIDDATA;
2939 muxlength = get_bits(&gb, 13) + 3;
2940 // not enough data, the parser should have sorted this out
2941 if (muxlength > avpkt->size)
2942 return AVERROR_INVALIDDATA;
2944 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2947 if (!latmctx->initialized) {
2948 if (!avctx->extradata) {
2952 push_output_configuration(&latmctx->aac_ctx);
2953 if ((err = decode_audio_specific_config(
2954 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2955 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2956 pop_output_configuration(&latmctx->aac_ctx);
2959 latmctx->initialized = 1;
2963 if (show_bits(&gb, 12) == 0xfff) {
2964 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2965 "ADTS header detected, probably as result of configuration "
2967 return AVERROR_INVALIDDATA;
2970 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2976 static av_cold int latm_decode_init(AVCodecContext *avctx)
2978 struct LATMContext *latmctx = avctx->priv_data;
2979 int ret = aac_decode_init(avctx);
2981 if (avctx->extradata_size > 0)
2982 latmctx->initialized = !ret;
2987 static void aacdec_init(AACContext *c)
2989 c->imdct_and_windowing = imdct_and_windowing;
2990 c->apply_ltp = apply_ltp;
2991 c->apply_tns = apply_tns;
2992 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
2993 c->update_ltp = update_ltp;
2996 ff_aacdec_init_mips(c);
2999 * AVOptions for Japanese DTV specific extensions (ADTS only)
3001 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3002 static const AVOption options[] = {
3003 {"dual_mono_mode", "Select the channel to decode for dual mono",
3004 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3005 AACDEC_FLAGS, "dual_mono_mode"},
3007 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3008 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3009 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3010 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3015 static const AVClass aac_decoder_class = {
3016 .class_name = "AAC decoder",
3017 .item_name = av_default_item_name,
3019 .version = LIBAVUTIL_VERSION_INT,
3022 AVCodec ff_aac_decoder = {
3024 .type = AVMEDIA_TYPE_AUDIO,
3025 .id = AV_CODEC_ID_AAC,
3026 .priv_data_size = sizeof(AACContext),
3027 .init = aac_decode_init,
3028 .close = aac_decode_close,
3029 .decode = aac_decode_frame,
3030 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3031 .sample_fmts = (const enum AVSampleFormat[]) {
3032 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3034 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3035 .channel_layouts = aac_channel_layout,
3037 .priv_class = &aac_decoder_class,
3041 Note: This decoder filter is intended to decode LATM streams transferred
3042 in MPEG transport streams which only contain one program.
3043 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3045 AVCodec ff_aac_latm_decoder = {
3047 .type = AVMEDIA_TYPE_AUDIO,
3048 .id = AV_CODEC_ID_AAC_LATM,
3049 .priv_data_size = sizeof(struct LATMContext),
3050 .init = latm_decode_init,
3051 .close = aac_decode_close,
3052 .decode = latm_decode_frame,
3053 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3054 .sample_fmts = (const enum AVSampleFormat[]) {
3055 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3057 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3058 .channel_layouts = aac_channel_layout,