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)
151 if (*channels >= MAX_CHANNELS)
152 return AVERROR_INVALIDDATA;
154 if (!ac->che[type][id]) {
155 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
156 return AVERROR(ENOMEM);
157 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
159 if (type != TYPE_CCE) {
160 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
161 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
162 return AVERROR_INVALIDDATA;
164 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
165 if (type == TYPE_CPE ||
166 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
167 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
171 if (ac->che[type][id])
172 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
173 av_freep(&ac->che[type][id]);
178 static int frame_configure_elements(AVCodecContext *avctx)
180 AACContext *ac = avctx->priv_data;
181 int type, id, ch, ret;
183 /* set channel pointers to internal buffers by default */
184 for (type = 0; type < 4; type++) {
185 for (id = 0; id < MAX_ELEM_ID; id++) {
186 ChannelElement *che = ac->che[type][id];
188 che->ch[0].ret = che->ch[0].ret_buf;
189 che->ch[1].ret = che->ch[1].ret_buf;
194 /* get output buffer */
195 av_frame_unref(ac->frame);
196 ac->frame->nb_samples = 2048;
197 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
200 /* map output channel pointers to AVFrame data */
201 for (ch = 0; ch < avctx->channels; ch++) {
202 if (ac->output_element[ch])
203 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
209 struct elem_to_channel {
210 uint64_t av_position;
213 uint8_t aac_position;
216 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
217 uint8_t (*layout_map)[3], int offset, uint64_t left,
218 uint64_t right, int pos)
220 if (layout_map[offset][0] == TYPE_CPE) {
221 e2c_vec[offset] = (struct elem_to_channel) {
222 .av_position = left | right, .syn_ele = TYPE_CPE,
223 .elem_id = layout_map[offset ][1], .aac_position = pos };
226 e2c_vec[offset] = (struct elem_to_channel) {
227 .av_position = left, .syn_ele = TYPE_SCE,
228 .elem_id = layout_map[offset ][1], .aac_position = pos };
229 e2c_vec[offset + 1] = (struct elem_to_channel) {
230 .av_position = right, .syn_ele = TYPE_SCE,
231 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
236 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
237 int num_pos_channels = 0;
241 for (i = *current; i < tags; i++) {
242 if (layout_map[i][2] != pos)
244 if (layout_map[i][0] == TYPE_CPE) {
246 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
252 num_pos_channels += 2;
260 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
263 return num_pos_channels;
266 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
268 int i, n, total_non_cc_elements;
269 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
270 int num_front_channels, num_side_channels, num_back_channels;
273 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
278 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
279 if (num_front_channels < 0)
282 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
283 if (num_side_channels < 0)
286 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
287 if (num_back_channels < 0)
291 if (num_front_channels & 1) {
292 e2c_vec[i] = (struct elem_to_channel) {
293 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
294 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
296 num_front_channels--;
298 if (num_front_channels >= 4) {
299 i += assign_pair(e2c_vec, layout_map, i,
300 AV_CH_FRONT_LEFT_OF_CENTER,
301 AV_CH_FRONT_RIGHT_OF_CENTER,
303 num_front_channels -= 2;
305 if (num_front_channels >= 2) {
306 i += assign_pair(e2c_vec, layout_map, i,
310 num_front_channels -= 2;
312 while (num_front_channels >= 2) {
313 i += assign_pair(e2c_vec, layout_map, i,
317 num_front_channels -= 2;
320 if (num_side_channels >= 2) {
321 i += assign_pair(e2c_vec, layout_map, i,
325 num_side_channels -= 2;
327 while (num_side_channels >= 2) {
328 i += assign_pair(e2c_vec, layout_map, i,
332 num_side_channels -= 2;
335 while (num_back_channels >= 4) {
336 i += assign_pair(e2c_vec, layout_map, i,
340 num_back_channels -= 2;
342 if (num_back_channels >= 2) {
343 i += assign_pair(e2c_vec, layout_map, i,
347 num_back_channels -= 2;
349 if (num_back_channels) {
350 e2c_vec[i] = (struct elem_to_channel) {
351 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
352 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
357 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
358 e2c_vec[i] = (struct elem_to_channel) {
359 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
360 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
363 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
364 e2c_vec[i] = (struct elem_to_channel) {
365 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
366 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
370 // Must choose a stable sort
371 total_non_cc_elements = n = i;
374 for (i = 1; i < n; i++) {
375 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
376 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
384 for (i = 0; i < total_non_cc_elements; i++) {
385 layout_map[i][0] = e2c_vec[i].syn_ele;
386 layout_map[i][1] = e2c_vec[i].elem_id;
387 layout_map[i][2] = e2c_vec[i].aac_position;
388 if (e2c_vec[i].av_position != UINT64_MAX) {
389 layout |= e2c_vec[i].av_position;
397 * Save current output configuration if and only if it has been locked.
399 static void push_output_configuration(AACContext *ac) {
400 if (ac->oc[1].status == OC_LOCKED) {
401 ac->oc[0] = ac->oc[1];
403 ac->oc[1].status = OC_NONE;
407 * Restore the previous output configuration if and only if the current
408 * configuration is unlocked.
410 static void pop_output_configuration(AACContext *ac) {
411 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
412 ac->oc[1] = ac->oc[0];
413 ac->avctx->channels = ac->oc[1].channels;
414 ac->avctx->channel_layout = ac->oc[1].channel_layout;
415 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
416 ac->oc[1].status, 0);
421 * Configure output channel order based on the current program configuration element.
423 * @return Returns error status. 0 - OK, !0 - error
425 static int output_configure(AACContext *ac,
426 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
427 enum OCStatus oc_type, int get_new_frame)
429 AVCodecContext *avctx = ac->avctx;
430 int i, channels = 0, ret;
433 if (ac->oc[1].layout_map != layout_map) {
434 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
435 ac->oc[1].layout_map_tags = tags;
438 // Try to sniff a reasonable channel order, otherwise output the
439 // channels in the order the PCE declared them.
440 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
441 layout = sniff_channel_order(layout_map, tags);
442 for (i = 0; i < tags; i++) {
443 int type = layout_map[i][0];
444 int id = layout_map[i][1];
445 int position = layout_map[i][2];
446 // Allocate or free elements depending on if they are in the
447 // current program configuration.
448 ret = che_configure(ac, position, type, id, &channels);
452 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
453 if (layout == AV_CH_FRONT_CENTER) {
454 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
460 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
461 if (layout) avctx->channel_layout = layout;
462 ac->oc[1].channel_layout = layout;
463 avctx->channels = ac->oc[1].channels = channels;
464 ac->oc[1].status = oc_type;
467 if ((ret = frame_configure_elements(ac->avctx)) < 0)
474 static void flush(AVCodecContext *avctx)
476 AACContext *ac= avctx->priv_data;
479 for (type = 3; type >= 0; type--) {
480 for (i = 0; i < MAX_ELEM_ID; i++) {
481 ChannelElement *che = ac->che[type][i];
483 for (j = 0; j <= 1; j++) {
484 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
492 * Set up channel positions based on a default channel configuration
493 * as specified in table 1.17.
495 * @return Returns error status. 0 - OK, !0 - error
497 static int set_default_channel_config(AVCodecContext *avctx,
498 uint8_t (*layout_map)[3],
502 if (channel_config < 1 || channel_config > 7) {
503 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
507 *tags = tags_per_config[channel_config];
508 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
512 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
514 // For PCE based channel configurations map the channels solely based on tags.
515 if (!ac->oc[1].m4ac.chan_config) {
516 return ac->tag_che_map[type][elem_id];
518 // Allow single CPE stereo files to be signalled with mono configuration.
519 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
520 uint8_t layout_map[MAX_ELEM_ID*4][3];
522 push_output_configuration(ac);
524 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
526 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
529 if (output_configure(ac, layout_map, layout_map_tags,
530 OC_TRIAL_FRAME, 1) < 0)
533 ac->oc[1].m4ac.chan_config = 2;
534 ac->oc[1].m4ac.ps = 0;
537 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
538 uint8_t layout_map[MAX_ELEM_ID*4][3];
540 push_output_configuration(ac);
542 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
544 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
547 if (output_configure(ac, layout_map, layout_map_tags,
548 OC_TRIAL_FRAME, 1) < 0)
551 ac->oc[1].m4ac.chan_config = 1;
552 if (ac->oc[1].m4ac.sbr)
553 ac->oc[1].m4ac.ps = -1;
555 // For indexed channel configurations map the channels solely based on position.
556 switch (ac->oc[1].m4ac.chan_config) {
558 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
560 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
563 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
564 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
565 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
566 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
568 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
571 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
573 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
576 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
578 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
582 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
584 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
585 } else if (ac->oc[1].m4ac.chan_config == 2) {
589 if (!ac->tags_mapped && type == TYPE_SCE) {
591 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
599 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
601 * @param type speaker type/position for these channels
603 static void decode_channel_map(uint8_t layout_map[][3],
604 enum ChannelPosition type,
605 GetBitContext *gb, int n)
608 enum RawDataBlockType syn_ele;
610 case AAC_CHANNEL_FRONT:
611 case AAC_CHANNEL_BACK:
612 case AAC_CHANNEL_SIDE:
613 syn_ele = get_bits1(gb);
619 case AAC_CHANNEL_LFE:
625 layout_map[0][0] = syn_ele;
626 layout_map[0][1] = get_bits(gb, 4);
627 layout_map[0][2] = type;
633 * Decode program configuration element; reference: table 4.2.
635 * @return Returns error status. 0 - OK, !0 - error
637 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
638 uint8_t (*layout_map)[3],
641 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
645 skip_bits(gb, 2); // object_type
647 sampling_index = get_bits(gb, 4);
648 if (m4ac->sampling_index != sampling_index)
649 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");
651 num_front = get_bits(gb, 4);
652 num_side = get_bits(gb, 4);
653 num_back = get_bits(gb, 4);
654 num_lfe = get_bits(gb, 2);
655 num_assoc_data = get_bits(gb, 3);
656 num_cc = get_bits(gb, 4);
659 skip_bits(gb, 4); // mono_mixdown_tag
661 skip_bits(gb, 4); // stereo_mixdown_tag
664 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
666 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
667 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
670 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
672 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
674 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
676 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
679 skip_bits_long(gb, 4 * num_assoc_data);
681 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
686 /* comment field, first byte is length */
687 comment_len = get_bits(gb, 8) * 8;
688 if (get_bits_left(gb) < comment_len) {
689 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
692 skip_bits_long(gb, comment_len);
697 * Decode GA "General Audio" specific configuration; reference: table 4.1.
699 * @param ac pointer to AACContext, may be null
700 * @param avctx pointer to AVCCodecContext, used for logging
702 * @return Returns error status. 0 - OK, !0 - error
704 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
706 MPEG4AudioConfig *m4ac,
709 int extension_flag, ret;
710 uint8_t layout_map[MAX_ELEM_ID*4][3];
713 if (get_bits1(gb)) { // frameLengthFlag
714 avpriv_request_sample(avctx, "960/120 MDCT window");
715 return AVERROR_PATCHWELCOME;
718 if (get_bits1(gb)) // dependsOnCoreCoder
719 skip_bits(gb, 14); // coreCoderDelay
720 extension_flag = get_bits1(gb);
722 if (m4ac->object_type == AOT_AAC_SCALABLE ||
723 m4ac->object_type == AOT_ER_AAC_SCALABLE)
724 skip_bits(gb, 3); // layerNr
726 if (channel_config == 0) {
727 skip_bits(gb, 4); // element_instance_tag
728 tags = decode_pce(avctx, m4ac, layout_map, gb);
732 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
736 if (count_channels(layout_map, tags) > 1) {
738 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
741 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
744 if (extension_flag) {
745 switch (m4ac->object_type) {
747 skip_bits(gb, 5); // numOfSubFrame
748 skip_bits(gb, 11); // layer_length
752 case AOT_ER_AAC_SCALABLE:
754 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
755 * aacScalefactorDataResilienceFlag
756 * aacSpectralDataResilienceFlag
760 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
766 * Decode audio specific configuration; reference: table 1.13.
768 * @param ac pointer to AACContext, may be null
769 * @param avctx pointer to AVCCodecContext, used for logging
770 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
771 * @param data pointer to buffer holding an audio specific config
772 * @param bit_size size of audio specific config or data in bits
773 * @param sync_extension look for an appended sync extension
775 * @return Returns error status or number of consumed bits. <0 - error
777 static int decode_audio_specific_config(AACContext *ac,
778 AVCodecContext *avctx,
779 MPEG4AudioConfig *m4ac,
780 const uint8_t *data, int bit_size,
787 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
788 for (i = 0; i < bit_size >> 3; i++)
789 av_dlog(avctx, "%02x ", data[i]);
790 av_dlog(avctx, "\n");
792 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
795 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
797 if (m4ac->sampling_index > 12) {
798 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
802 skip_bits_long(&gb, i);
804 switch (m4ac->object_type) {
808 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
812 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
813 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
817 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
818 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
819 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
821 return get_bits_count(&gb);
825 * linear congruential pseudorandom number generator
827 * @param previous_val pointer to the current state of the generator
829 * @return Returns a 32-bit pseudorandom integer
831 static av_always_inline int lcg_random(unsigned previous_val)
833 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
837 static av_always_inline void reset_predict_state(PredictorState *ps)
847 static void reset_all_predictors(PredictorState *ps)
850 for (i = 0; i < MAX_PREDICTORS; i++)
851 reset_predict_state(&ps[i]);
854 static int sample_rate_idx (int rate)
856 if (92017 <= rate) return 0;
857 else if (75132 <= rate) return 1;
858 else if (55426 <= rate) return 2;
859 else if (46009 <= rate) return 3;
860 else if (37566 <= rate) return 4;
861 else if (27713 <= rate) return 5;
862 else if (23004 <= rate) return 6;
863 else if (18783 <= rate) return 7;
864 else if (13856 <= rate) return 8;
865 else if (11502 <= rate) return 9;
866 else if (9391 <= rate) return 10;
870 static void reset_predictor_group(PredictorState *ps, int group_num)
873 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
874 reset_predict_state(&ps[i]);
877 #define AAC_INIT_VLC_STATIC(num, size) \
878 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
879 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
880 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
883 static void aacdec_init(AACContext *ac);
885 static av_cold int aac_decode_init(AVCodecContext *avctx)
887 AACContext *ac = avctx->priv_data;
890 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
894 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
896 if (avctx->extradata_size > 0) {
897 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
899 avctx->extradata_size*8, 1) < 0)
903 uint8_t layout_map[MAX_ELEM_ID*4][3];
906 sr = sample_rate_idx(avctx->sample_rate);
907 ac->oc[1].m4ac.sampling_index = sr;
908 ac->oc[1].m4ac.channels = avctx->channels;
909 ac->oc[1].m4ac.sbr = -1;
910 ac->oc[1].m4ac.ps = -1;
912 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
913 if (ff_mpeg4audio_channels[i] == avctx->channels)
915 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
918 ac->oc[1].m4ac.chan_config = i;
920 if (ac->oc[1].m4ac.chan_config) {
921 int ret = set_default_channel_config(avctx, layout_map,
922 &layout_map_tags, ac->oc[1].m4ac.chan_config);
924 output_configure(ac, layout_map, layout_map_tags,
926 else if (avctx->err_recognition & AV_EF_EXPLODE)
927 return AVERROR_INVALIDDATA;
931 if (avctx->channels > MAX_CHANNELS) {
932 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
933 return AVERROR_INVALIDDATA;
936 AAC_INIT_VLC_STATIC( 0, 304);
937 AAC_INIT_VLC_STATIC( 1, 270);
938 AAC_INIT_VLC_STATIC( 2, 550);
939 AAC_INIT_VLC_STATIC( 3, 300);
940 AAC_INIT_VLC_STATIC( 4, 328);
941 AAC_INIT_VLC_STATIC( 5, 294);
942 AAC_INIT_VLC_STATIC( 6, 306);
943 AAC_INIT_VLC_STATIC( 7, 268);
944 AAC_INIT_VLC_STATIC( 8, 510);
945 AAC_INIT_VLC_STATIC( 9, 366);
946 AAC_INIT_VLC_STATIC(10, 462);
950 ff_fmt_convert_init(&ac->fmt_conv, avctx);
951 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
953 ac->random_state = 0x1f2e3d4c;
957 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
958 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
959 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
962 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
963 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
964 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
965 // window initialization
966 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
967 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
968 ff_init_ff_sine_windows(10);
969 ff_init_ff_sine_windows( 7);
977 * Skip data_stream_element; reference: table 4.10.
979 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
981 int byte_align = get_bits1(gb);
982 int count = get_bits(gb, 8);
984 count += get_bits(gb, 8);
988 if (get_bits_left(gb) < 8 * count) {
989 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
992 skip_bits_long(gb, 8 * count);
996 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1000 if (get_bits1(gb)) {
1001 ics->predictor_reset_group = get_bits(gb, 5);
1002 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
1003 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
1007 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1008 ics->prediction_used[sfb] = get_bits1(gb);
1014 * Decode Long Term Prediction data; reference: table 4.xx.
1016 static void decode_ltp(LongTermPrediction *ltp,
1017 GetBitContext *gb, uint8_t max_sfb)
1021 ltp->lag = get_bits(gb, 11);
1022 ltp->coef = ltp_coef[get_bits(gb, 3)];
1023 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1024 ltp->used[sfb] = get_bits1(gb);
1028 * Decode Individual Channel Stream info; reference: table 4.6.
1030 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1033 if (get_bits1(gb)) {
1034 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1035 return AVERROR_INVALIDDATA;
1037 ics->window_sequence[1] = ics->window_sequence[0];
1038 ics->window_sequence[0] = get_bits(gb, 2);
1039 ics->use_kb_window[1] = ics->use_kb_window[0];
1040 ics->use_kb_window[0] = get_bits1(gb);
1041 ics->num_window_groups = 1;
1042 ics->group_len[0] = 1;
1043 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1045 ics->max_sfb = get_bits(gb, 4);
1046 for (i = 0; i < 7; i++) {
1047 if (get_bits1(gb)) {
1048 ics->group_len[ics->num_window_groups - 1]++;
1050 ics->num_window_groups++;
1051 ics->group_len[ics->num_window_groups - 1] = 1;
1054 ics->num_windows = 8;
1055 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1056 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1057 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1058 ics->predictor_present = 0;
1060 ics->max_sfb = get_bits(gb, 6);
1061 ics->num_windows = 1;
1062 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1063 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1064 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1065 ics->predictor_present = get_bits1(gb);
1066 ics->predictor_reset_group = 0;
1067 if (ics->predictor_present) {
1068 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1069 if (decode_prediction(ac, ics, gb)) {
1072 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1073 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1076 if ((ics->ltp.present = get_bits(gb, 1)))
1077 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1082 if (ics->max_sfb > ics->num_swb) {
1083 av_log(ac->avctx, AV_LOG_ERROR,
1084 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1085 ics->max_sfb, ics->num_swb);
1092 return AVERROR_INVALIDDATA;
1096 * Decode band types (section_data payload); reference: table 4.46.
1098 * @param band_type array of the used band type
1099 * @param band_type_run_end array of the last scalefactor band of a band type run
1101 * @return Returns error status. 0 - OK, !0 - error
1103 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1104 int band_type_run_end[120], GetBitContext *gb,
1105 IndividualChannelStream *ics)
1108 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1109 for (g = 0; g < ics->num_window_groups; g++) {
1111 while (k < ics->max_sfb) {
1112 uint8_t sect_end = k;
1114 int sect_band_type = get_bits(gb, 4);
1115 if (sect_band_type == 12) {
1116 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1120 sect_len_incr = get_bits(gb, bits);
1121 sect_end += sect_len_incr;
1122 if (get_bits_left(gb) < 0) {
1123 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1126 if (sect_end > ics->max_sfb) {
1127 av_log(ac->avctx, AV_LOG_ERROR,
1128 "Number of bands (%d) exceeds limit (%d).\n",
1129 sect_end, ics->max_sfb);
1132 } while (sect_len_incr == (1 << bits) - 1);
1133 for (; k < sect_end; k++) {
1134 band_type [idx] = sect_band_type;
1135 band_type_run_end[idx++] = sect_end;
1143 * Decode scalefactors; reference: table 4.47.
1145 * @param global_gain first scalefactor value as scalefactors are differentially coded
1146 * @param band_type array of the used band type
1147 * @param band_type_run_end array of the last scalefactor band of a band type run
1148 * @param sf array of scalefactors or intensity stereo positions
1150 * @return Returns error status. 0 - OK, !0 - error
1152 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1153 unsigned int global_gain,
1154 IndividualChannelStream *ics,
1155 enum BandType band_type[120],
1156 int band_type_run_end[120])
1159 int offset[3] = { global_gain, global_gain - 90, 0 };
1162 for (g = 0; g < ics->num_window_groups; g++) {
1163 for (i = 0; i < ics->max_sfb;) {
1164 int run_end = band_type_run_end[idx];
1165 if (band_type[idx] == ZERO_BT) {
1166 for (; i < run_end; i++, idx++)
1168 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1169 for (; i < run_end; i++, idx++) {
1170 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1171 clipped_offset = av_clip(offset[2], -155, 100);
1172 if (offset[2] != clipped_offset) {
1173 avpriv_request_sample(ac->avctx,
1174 "If you heard an audible artifact, there may be a bug in the decoder. "
1175 "Clipped intensity stereo position (%d -> %d)",
1176 offset[2], clipped_offset);
1178 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1180 } else if (band_type[idx] == NOISE_BT) {
1181 for (; i < run_end; i++, idx++) {
1182 if (noise_flag-- > 0)
1183 offset[1] += get_bits(gb, 9) - 256;
1185 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1186 clipped_offset = av_clip(offset[1], -100, 155);
1187 if (offset[1] != clipped_offset) {
1188 avpriv_request_sample(ac->avctx,
1189 "If you heard an audible artifact, there may be a bug in the decoder. "
1190 "Clipped noise gain (%d -> %d)",
1191 offset[1], clipped_offset);
1193 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1196 for (; i < run_end; i++, idx++) {
1197 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1198 if (offset[0] > 255U) {
1199 av_log(ac->avctx, AV_LOG_ERROR,
1200 "Scalefactor (%d) out of range.\n", offset[0]);
1203 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1212 * Decode pulse data; reference: table 4.7.
1214 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1215 const uint16_t *swb_offset, int num_swb)
1218 pulse->num_pulse = get_bits(gb, 2) + 1;
1219 pulse_swb = get_bits(gb, 6);
1220 if (pulse_swb >= num_swb)
1222 pulse->pos[0] = swb_offset[pulse_swb];
1223 pulse->pos[0] += get_bits(gb, 5);
1224 if (pulse->pos[0] > 1023)
1226 pulse->amp[0] = get_bits(gb, 4);
1227 for (i = 1; i < pulse->num_pulse; i++) {
1228 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1229 if (pulse->pos[i] > 1023)
1231 pulse->amp[i] = get_bits(gb, 4);
1237 * Decode Temporal Noise Shaping data; reference: table 4.48.
1239 * @return Returns error status. 0 - OK, !0 - error
1241 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1242 GetBitContext *gb, const IndividualChannelStream *ics)
1244 int w, filt, i, coef_len, coef_res, coef_compress;
1245 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1246 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1247 for (w = 0; w < ics->num_windows; w++) {
1248 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1249 coef_res = get_bits1(gb);
1251 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1253 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1255 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1256 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1257 tns->order[w][filt], tns_max_order);
1258 tns->order[w][filt] = 0;
1261 if (tns->order[w][filt]) {
1262 tns->direction[w][filt] = get_bits1(gb);
1263 coef_compress = get_bits1(gb);
1264 coef_len = coef_res + 3 - coef_compress;
1265 tmp2_idx = 2 * coef_compress + coef_res;
1267 for (i = 0; i < tns->order[w][filt]; i++)
1268 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1277 * Decode Mid/Side data; reference: table 4.54.
1279 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1280 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1281 * [3] reserved for scalable AAC
1283 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1287 if (ms_present == 1) {
1288 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1289 cpe->ms_mask[idx] = get_bits1(gb);
1290 } else if (ms_present == 2) {
1291 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1296 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1300 *dst++ = v[idx & 15] * s;
1301 *dst++ = v[idx>>4 & 15] * s;
1307 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1311 *dst++ = v[idx & 3] * s;
1312 *dst++ = v[idx>>2 & 3] * s;
1313 *dst++ = v[idx>>4 & 3] * s;
1314 *dst++ = v[idx>>6 & 3] * s;
1320 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1321 unsigned sign, const float *scale)
1323 union av_intfloat32 s0, s1;
1325 s0.f = s1.f = *scale;
1326 s0.i ^= sign >> 1 << 31;
1329 *dst++ = v[idx & 15] * s0.f;
1330 *dst++ = v[idx>>4 & 15] * s1.f;
1337 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1338 unsigned sign, const float *scale)
1340 unsigned nz = idx >> 12;
1341 union av_intfloat32 s = { .f = *scale };
1342 union av_intfloat32 t;
1344 t.i = s.i ^ (sign & 1U<<31);
1345 *dst++ = v[idx & 3] * t.f;
1347 sign <<= nz & 1; nz >>= 1;
1348 t.i = s.i ^ (sign & 1U<<31);
1349 *dst++ = v[idx>>2 & 3] * t.f;
1351 sign <<= nz & 1; nz >>= 1;
1352 t.i = s.i ^ (sign & 1U<<31);
1353 *dst++ = v[idx>>4 & 3] * t.f;
1356 t.i = s.i ^ (sign & 1U<<31);
1357 *dst++ = v[idx>>6 & 3] * t.f;
1364 * Decode spectral data; reference: table 4.50.
1365 * Dequantize and scale spectral data; reference: 4.6.3.3.
1367 * @param coef array of dequantized, scaled spectral data
1368 * @param sf array of scalefactors or intensity stereo positions
1369 * @param pulse_present set if pulses are present
1370 * @param pulse pointer to pulse data struct
1371 * @param band_type array of the used band type
1373 * @return Returns error status. 0 - OK, !0 - error
1375 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1376 GetBitContext *gb, const float sf[120],
1377 int pulse_present, const Pulse *pulse,
1378 const IndividualChannelStream *ics,
1379 enum BandType band_type[120])
1381 int i, k, g, idx = 0;
1382 const int c = 1024 / ics->num_windows;
1383 const uint16_t *offsets = ics->swb_offset;
1384 float *coef_base = coef;
1386 for (g = 0; g < ics->num_windows; g++)
1387 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1389 for (g = 0; g < ics->num_window_groups; g++) {
1390 unsigned g_len = ics->group_len[g];
1392 for (i = 0; i < ics->max_sfb; i++, idx++) {
1393 const unsigned cbt_m1 = band_type[idx] - 1;
1394 float *cfo = coef + offsets[i];
1395 int off_len = offsets[i + 1] - offsets[i];
1398 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1399 for (group = 0; group < g_len; group++, cfo+=128) {
1400 memset(cfo, 0, off_len * sizeof(float));
1402 } else if (cbt_m1 == NOISE_BT - 1) {
1403 for (group = 0; group < g_len; group++, cfo+=128) {
1407 for (k = 0; k < off_len; k++) {
1408 ac->random_state = lcg_random(ac->random_state);
1409 cfo[k] = ac->random_state;
1412 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1413 scale = sf[idx] / sqrtf(band_energy);
1414 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1417 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1418 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1419 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1420 OPEN_READER(re, gb);
1422 switch (cbt_m1 >> 1) {
1424 for (group = 0; group < g_len; group++, cfo+=128) {
1432 UPDATE_CACHE(re, gb);
1433 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1434 cb_idx = cb_vector_idx[code];
1435 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1441 for (group = 0; group < g_len; group++, cfo+=128) {
1451 UPDATE_CACHE(re, gb);
1452 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1453 cb_idx = cb_vector_idx[code];
1454 nnz = cb_idx >> 8 & 15;
1455 bits = nnz ? GET_CACHE(re, gb) : 0;
1456 LAST_SKIP_BITS(re, gb, nnz);
1457 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1463 for (group = 0; group < g_len; group++, cfo+=128) {
1471 UPDATE_CACHE(re, gb);
1472 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1473 cb_idx = cb_vector_idx[code];
1474 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1481 for (group = 0; group < g_len; group++, cfo+=128) {
1491 UPDATE_CACHE(re, gb);
1492 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1493 cb_idx = cb_vector_idx[code];
1494 nnz = cb_idx >> 8 & 15;
1495 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1496 LAST_SKIP_BITS(re, gb, nnz);
1497 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1503 for (group = 0; group < g_len; group++, cfo+=128) {
1505 uint32_t *icf = (uint32_t *) cf;
1515 UPDATE_CACHE(re, gb);
1516 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1524 cb_idx = cb_vector_idx[code];
1527 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1528 LAST_SKIP_BITS(re, gb, nnz);
1530 for (j = 0; j < 2; j++) {
1534 /* The total length of escape_sequence must be < 22 bits according
1535 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1536 UPDATE_CACHE(re, gb);
1537 b = GET_CACHE(re, gb);
1538 b = 31 - av_log2(~b);
1541 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1545 SKIP_BITS(re, gb, b + 1);
1547 n = (1 << b) + SHOW_UBITS(re, gb, b);
1548 LAST_SKIP_BITS(re, gb, b);
1549 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1552 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1553 *icf++ = (bits & 1U<<31) | v;
1560 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1564 CLOSE_READER(re, gb);
1570 if (pulse_present) {
1572 for (i = 0; i < pulse->num_pulse; i++) {
1573 float co = coef_base[ pulse->pos[i] ];
1574 while (offsets[idx + 1] <= pulse->pos[i])
1576 if (band_type[idx] != NOISE_BT && sf[idx]) {
1577 float ico = -pulse->amp[i];
1580 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1582 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1589 static av_always_inline float flt16_round(float pf)
1591 union av_intfloat32 tmp;
1593 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1597 static av_always_inline float flt16_even(float pf)
1599 union av_intfloat32 tmp;
1601 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1605 static av_always_inline float flt16_trunc(float pf)
1607 union av_intfloat32 pun;
1609 pun.i &= 0xFFFF0000U;
1613 static av_always_inline void predict(PredictorState *ps, float *coef,
1616 const float a = 0.953125; // 61.0 / 64
1617 const float alpha = 0.90625; // 29.0 / 32
1621 float r0 = ps->r0, r1 = ps->r1;
1622 float cor0 = ps->cor0, cor1 = ps->cor1;
1623 float var0 = ps->var0, var1 = ps->var1;
1625 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1626 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1628 pv = flt16_round(k1 * r0 + k2 * r1);
1635 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1636 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1637 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1638 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1640 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1641 ps->r0 = flt16_trunc(a * e0);
1645 * Apply AAC-Main style frequency domain prediction.
1647 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1651 if (!sce->ics.predictor_initialized) {
1652 reset_all_predictors(sce->predictor_state);
1653 sce->ics.predictor_initialized = 1;
1656 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1657 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1658 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1659 predict(&sce->predictor_state[k], &sce->coeffs[k],
1660 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1663 if (sce->ics.predictor_reset_group)
1664 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1666 reset_all_predictors(sce->predictor_state);
1670 * Decode an individual_channel_stream payload; reference: table 4.44.
1672 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1673 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1675 * @return Returns error status. 0 - OK, !0 - error
1677 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1678 GetBitContext *gb, int common_window, int scale_flag)
1681 TemporalNoiseShaping *tns = &sce->tns;
1682 IndividualChannelStream *ics = &sce->ics;
1683 float *out = sce->coeffs;
1684 int global_gain, pulse_present = 0;
1686 /* This assignment is to silence a GCC warning about the variable being used
1687 * uninitialized when in fact it always is.
1689 pulse.num_pulse = 0;
1691 global_gain = get_bits(gb, 8);
1693 if (!common_window && !scale_flag) {
1694 if (decode_ics_info(ac, ics, gb) < 0)
1695 return AVERROR_INVALIDDATA;
1698 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1700 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1705 if ((pulse_present = get_bits1(gb))) {
1706 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1707 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1710 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1711 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1715 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1717 if (get_bits1(gb)) {
1718 avpriv_request_sample(ac->avctx, "SSR");
1719 return AVERROR_PATCHWELCOME;
1723 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1726 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1727 apply_prediction(ac, sce);
1733 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1735 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1737 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1738 float *ch0 = cpe->ch[0].coeffs;
1739 float *ch1 = cpe->ch[1].coeffs;
1740 int g, i, group, idx = 0;
1741 const uint16_t *offsets = ics->swb_offset;
1742 for (g = 0; g < ics->num_window_groups; g++) {
1743 for (i = 0; i < ics->max_sfb; i++, idx++) {
1744 if (cpe->ms_mask[idx] &&
1745 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1746 for (group = 0; group < ics->group_len[g]; group++) {
1747 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1748 ch1 + group * 128 + offsets[i],
1749 offsets[i+1] - offsets[i]);
1753 ch0 += ics->group_len[g] * 128;
1754 ch1 += ics->group_len[g] * 128;
1759 * intensity stereo decoding; reference: 4.6.8.2.3
1761 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1762 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1763 * [3] reserved for scalable AAC
1765 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1767 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1768 SingleChannelElement *sce1 = &cpe->ch[1];
1769 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1770 const uint16_t *offsets = ics->swb_offset;
1771 int g, group, i, idx = 0;
1774 for (g = 0; g < ics->num_window_groups; g++) {
1775 for (i = 0; i < ics->max_sfb;) {
1776 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1777 const int bt_run_end = sce1->band_type_run_end[idx];
1778 for (; i < bt_run_end; i++, idx++) {
1779 c = -1 + 2 * (sce1->band_type[idx] - 14);
1781 c *= 1 - 2 * cpe->ms_mask[idx];
1782 scale = c * sce1->sf[idx];
1783 for (group = 0; group < ics->group_len[g]; group++)
1784 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1785 coef0 + group * 128 + offsets[i],
1787 offsets[i + 1] - offsets[i]);
1790 int bt_run_end = sce1->band_type_run_end[idx];
1791 idx += bt_run_end - i;
1795 coef0 += ics->group_len[g] * 128;
1796 coef1 += ics->group_len[g] * 128;
1801 * Decode a channel_pair_element; reference: table 4.4.
1803 * @return Returns error status. 0 - OK, !0 - error
1805 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1807 int i, ret, common_window, ms_present = 0;
1809 common_window = get_bits1(gb);
1810 if (common_window) {
1811 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1812 return AVERROR_INVALIDDATA;
1813 i = cpe->ch[1].ics.use_kb_window[0];
1814 cpe->ch[1].ics = cpe->ch[0].ics;
1815 cpe->ch[1].ics.use_kb_window[1] = i;
1816 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1817 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1818 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1819 ms_present = get_bits(gb, 2);
1820 if (ms_present == 3) {
1821 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1823 } else if (ms_present)
1824 decode_mid_side_stereo(cpe, gb, ms_present);
1826 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1828 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1831 if (common_window) {
1833 apply_mid_side_stereo(ac, cpe);
1834 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1835 apply_prediction(ac, &cpe->ch[0]);
1836 apply_prediction(ac, &cpe->ch[1]);
1840 apply_intensity_stereo(ac, cpe, ms_present);
1844 static const float cce_scale[] = {
1845 1.09050773266525765921, //2^(1/8)
1846 1.18920711500272106672, //2^(1/4)
1852 * Decode coupling_channel_element; reference: table 4.8.
1854 * @return Returns error status. 0 - OK, !0 - error
1856 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1862 SingleChannelElement *sce = &che->ch[0];
1863 ChannelCoupling *coup = &che->coup;
1865 coup->coupling_point = 2 * get_bits1(gb);
1866 coup->num_coupled = get_bits(gb, 3);
1867 for (c = 0; c <= coup->num_coupled; c++) {
1869 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1870 coup->id_select[c] = get_bits(gb, 4);
1871 if (coup->type[c] == TYPE_CPE) {
1872 coup->ch_select[c] = get_bits(gb, 2);
1873 if (coup->ch_select[c] == 3)
1876 coup->ch_select[c] = 2;
1878 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1880 sign = get_bits(gb, 1);
1881 scale = cce_scale[get_bits(gb, 2)];
1883 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1886 for (c = 0; c < num_gain; c++) {
1890 float gain_cache = 1.;
1892 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1893 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1894 gain_cache = powf(scale, -gain);
1896 if (coup->coupling_point == AFTER_IMDCT) {
1897 coup->gain[c][0] = gain_cache;
1899 for (g = 0; g < sce->ics.num_window_groups; g++) {
1900 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1901 if (sce->band_type[idx] != ZERO_BT) {
1903 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1911 gain_cache = powf(scale, -t) * s;
1914 coup->gain[c][idx] = gain_cache;
1924 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1926 * @return Returns number of bytes consumed.
1928 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1932 int num_excl_chan = 0;
1935 for (i = 0; i < 7; i++)
1936 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1937 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1939 return num_excl_chan / 7;
1943 * Decode dynamic range information; reference: table 4.52.
1945 * @return Returns number of bytes consumed.
1947 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1951 int drc_num_bands = 1;
1954 /* pce_tag_present? */
1955 if (get_bits1(gb)) {
1956 che_drc->pce_instance_tag = get_bits(gb, 4);
1957 skip_bits(gb, 4); // tag_reserved_bits
1961 /* excluded_chns_present? */
1962 if (get_bits1(gb)) {
1963 n += decode_drc_channel_exclusions(che_drc, gb);
1966 /* drc_bands_present? */
1967 if (get_bits1(gb)) {
1968 che_drc->band_incr = get_bits(gb, 4);
1969 che_drc->interpolation_scheme = get_bits(gb, 4);
1971 drc_num_bands += che_drc->band_incr;
1972 for (i = 0; i < drc_num_bands; i++) {
1973 che_drc->band_top[i] = get_bits(gb, 8);
1978 /* prog_ref_level_present? */
1979 if (get_bits1(gb)) {
1980 che_drc->prog_ref_level = get_bits(gb, 7);
1981 skip_bits1(gb); // prog_ref_level_reserved_bits
1985 for (i = 0; i < drc_num_bands; i++) {
1986 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1987 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1994 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1996 int i, major, minor;
2001 get_bits(gb, 13); len -= 13;
2003 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2004 buf[i] = get_bits(gb, 8);
2007 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2008 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2010 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2011 ac->avctx->internal->skip_samples = 1024;
2015 skip_bits_long(gb, len);
2021 * Decode extension data (incomplete); reference: table 4.51.
2023 * @param cnt length of TYPE_FIL syntactic element in bytes
2025 * @return Returns number of bytes consumed
2027 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2028 ChannelElement *che, enum RawDataBlockType elem_type)
2032 switch (get_bits(gb, 4)) { // extension type
2033 case EXT_SBR_DATA_CRC:
2037 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2039 } else if (!ac->oc[1].m4ac.sbr) {
2040 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2041 skip_bits_long(gb, 8 * cnt - 4);
2043 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2044 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2045 skip_bits_long(gb, 8 * cnt - 4);
2047 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2048 ac->oc[1].m4ac.sbr = 1;
2049 ac->oc[1].m4ac.ps = 1;
2050 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2051 ac->oc[1].status, 1);
2053 ac->oc[1].m4ac.sbr = 1;
2055 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2057 case EXT_DYNAMIC_RANGE:
2058 res = decode_dynamic_range(&ac->che_drc, gb);
2061 decode_fill(ac, gb, 8 * cnt - 4);
2064 case EXT_DATA_ELEMENT:
2066 skip_bits_long(gb, 8 * cnt - 4);
2073 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2075 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2076 * @param coef spectral coefficients
2078 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2079 IndividualChannelStream *ics, int decode)
2081 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2083 int bottom, top, order, start, end, size, inc;
2084 float lpc[TNS_MAX_ORDER];
2085 float tmp[TNS_MAX_ORDER+1];
2087 for (w = 0; w < ics->num_windows; w++) {
2088 bottom = ics->num_swb;
2089 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2091 bottom = FFMAX(0, top - tns->length[w][filt]);
2092 order = tns->order[w][filt];
2097 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2099 start = ics->swb_offset[FFMIN(bottom, mmm)];
2100 end = ics->swb_offset[FFMIN( top, mmm)];
2101 if ((size = end - start) <= 0)
2103 if (tns->direction[w][filt]) {
2113 for (m = 0; m < size; m++, start += inc)
2114 for (i = 1; i <= FFMIN(m, order); i++)
2115 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2118 for (m = 0; m < size; m++, start += inc) {
2119 tmp[0] = coef[start];
2120 for (i = 1; i <= FFMIN(m, order); i++)
2121 coef[start] += tmp[i] * lpc[i - 1];
2122 for (i = order; i > 0; i--)
2123 tmp[i] = tmp[i - 1];
2131 * Apply windowing and MDCT to obtain the spectral
2132 * coefficient from the predicted sample by LTP.
2134 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2135 float *in, IndividualChannelStream *ics)
2137 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2138 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2139 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2140 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2142 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2143 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2145 memset(in, 0, 448 * sizeof(float));
2146 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2148 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2149 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2151 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2152 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2154 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2158 * Apply the long term prediction
2160 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2162 const LongTermPrediction *ltp = &sce->ics.ltp;
2163 const uint16_t *offsets = sce->ics.swb_offset;
2166 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2167 float *predTime = sce->ret;
2168 float *predFreq = ac->buf_mdct;
2169 int16_t num_samples = 2048;
2171 if (ltp->lag < 1024)
2172 num_samples = ltp->lag + 1024;
2173 for (i = 0; i < num_samples; i++)
2174 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2175 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2177 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2179 if (sce->tns.present)
2180 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2182 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2184 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2185 sce->coeffs[i] += predFreq[i];
2190 * Update the LTP buffer for next frame
2192 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2194 IndividualChannelStream *ics = &sce->ics;
2195 float *saved = sce->saved;
2196 float *saved_ltp = sce->coeffs;
2197 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2198 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2201 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2202 memcpy(saved_ltp, saved, 512 * sizeof(float));
2203 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2204 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2205 for (i = 0; i < 64; i++)
2206 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2207 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2208 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2209 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2210 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2211 for (i = 0; i < 64; i++)
2212 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2213 } else { // LONG_STOP or ONLY_LONG
2214 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2215 for (i = 0; i < 512; i++)
2216 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2219 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2220 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2221 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2225 * Conduct IMDCT and windowing.
2227 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2229 IndividualChannelStream *ics = &sce->ics;
2230 float *in = sce->coeffs;
2231 float *out = sce->ret;
2232 float *saved = sce->saved;
2233 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2234 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2235 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2236 float *buf = ac->buf_mdct;
2237 float *temp = ac->temp;
2241 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2242 for (i = 0; i < 1024; i += 128)
2243 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2245 ac->mdct.imdct_half(&ac->mdct, buf, in);
2247 /* window overlapping
2248 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2249 * and long to short transitions are considered to be short to short
2250 * transitions. This leaves just two cases (long to long and short to short)
2251 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2253 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2254 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2255 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2257 memcpy( out, saved, 448 * sizeof(float));
2259 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2260 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2261 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2262 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2263 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2264 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2265 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2267 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2268 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2273 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2274 memcpy( saved, temp + 64, 64 * sizeof(float));
2275 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2276 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2277 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2278 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2279 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2280 memcpy( saved, buf + 512, 448 * sizeof(float));
2281 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2282 } else { // LONG_STOP or ONLY_LONG
2283 memcpy( saved, buf + 512, 512 * sizeof(float));
2288 * Apply dependent channel coupling (applied before IMDCT).
2290 * @param index index into coupling gain array
2292 static void apply_dependent_coupling(AACContext *ac,
2293 SingleChannelElement *target,
2294 ChannelElement *cce, int index)
2296 IndividualChannelStream *ics = &cce->ch[0].ics;
2297 const uint16_t *offsets = ics->swb_offset;
2298 float *dest = target->coeffs;
2299 const float *src = cce->ch[0].coeffs;
2300 int g, i, group, k, idx = 0;
2301 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2302 av_log(ac->avctx, AV_LOG_ERROR,
2303 "Dependent coupling is not supported together with LTP\n");
2306 for (g = 0; g < ics->num_window_groups; g++) {
2307 for (i = 0; i < ics->max_sfb; i++, idx++) {
2308 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2309 const float gain = cce->coup.gain[index][idx];
2310 for (group = 0; group < ics->group_len[g]; group++) {
2311 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2313 dest[group * 128 + k] += gain * src[group * 128 + k];
2318 dest += ics->group_len[g] * 128;
2319 src += ics->group_len[g] * 128;
2324 * Apply independent channel coupling (applied after IMDCT).
2326 * @param index index into coupling gain array
2328 static void apply_independent_coupling(AACContext *ac,
2329 SingleChannelElement *target,
2330 ChannelElement *cce, int index)
2333 const float gain = cce->coup.gain[index][0];
2334 const float *src = cce->ch[0].ret;
2335 float *dest = target->ret;
2336 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2338 for (i = 0; i < len; i++)
2339 dest[i] += gain * src[i];
2343 * channel coupling transformation interface
2345 * @param apply_coupling_method pointer to (in)dependent coupling function
2347 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2348 enum RawDataBlockType type, int elem_id,
2349 enum CouplingPoint coupling_point,
2350 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2354 for (i = 0; i < MAX_ELEM_ID; i++) {
2355 ChannelElement *cce = ac->che[TYPE_CCE][i];
2358 if (cce && cce->coup.coupling_point == coupling_point) {
2359 ChannelCoupling *coup = &cce->coup;
2361 for (c = 0; c <= coup->num_coupled; c++) {
2362 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2363 if (coup->ch_select[c] != 1) {
2364 apply_coupling_method(ac, &cc->ch[0], cce, index);
2365 if (coup->ch_select[c] != 0)
2368 if (coup->ch_select[c] != 2)
2369 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2371 index += 1 + (coup->ch_select[c] == 3);
2378 * Convert spectral data to float samples, applying all supported tools as appropriate.
2380 static void spectral_to_sample(AACContext *ac)
2383 for (type = 3; type >= 0; type--) {
2384 for (i = 0; i < MAX_ELEM_ID; i++) {
2385 ChannelElement *che = ac->che[type][i];
2387 if (type <= TYPE_CPE)
2388 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2389 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2390 if (che->ch[0].ics.predictor_present) {
2391 if (che->ch[0].ics.ltp.present)
2392 ac->apply_ltp(ac, &che->ch[0]);
2393 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2394 ac->apply_ltp(ac, &che->ch[1]);
2397 if (che->ch[0].tns.present)
2398 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2399 if (che->ch[1].tns.present)
2400 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2401 if (type <= TYPE_CPE)
2402 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2403 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2404 ac->imdct_and_windowing(ac, &che->ch[0]);
2405 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2406 ac->update_ltp(ac, &che->ch[0]);
2407 if (type == TYPE_CPE) {
2408 ac->imdct_and_windowing(ac, &che->ch[1]);
2409 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2410 ac->update_ltp(ac, &che->ch[1]);
2412 if (ac->oc[1].m4ac.sbr > 0) {
2413 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2416 if (type <= TYPE_CCE)
2417 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2423 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2426 AACADTSHeaderInfo hdr_info;
2427 uint8_t layout_map[MAX_ELEM_ID*4][3];
2428 int layout_map_tags;
2430 size = avpriv_aac_parse_header(gb, &hdr_info);
2432 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2433 // This is 2 for "VLB " audio in NSV files.
2434 // See samples/nsv/vlb_audio.
2435 avpriv_report_missing_feature(ac->avctx,
2436 "More than one AAC RDB per ADTS frame");
2437 ac->warned_num_aac_frames = 1;
2439 push_output_configuration(ac);
2440 if (hdr_info.chan_config) {
2441 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2442 if (set_default_channel_config(ac->avctx, layout_map,
2443 &layout_map_tags, hdr_info.chan_config))
2445 if (output_configure(ac, layout_map, layout_map_tags,
2446 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2449 ac->oc[1].m4ac.chan_config = 0;
2451 * dual mono frames in Japanese DTV can have chan_config 0
2452 * WITHOUT specifying PCE.
2453 * thus, set dual mono as default.
2455 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2456 layout_map_tags = 2;
2457 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2458 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2459 layout_map[0][1] = 0;
2460 layout_map[1][1] = 1;
2461 if (output_configure(ac, layout_map, layout_map_tags,
2466 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2467 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2468 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2469 if (ac->oc[0].status != OC_LOCKED ||
2470 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2471 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2472 ac->oc[1].m4ac.sbr = -1;
2473 ac->oc[1].m4ac.ps = -1;
2475 if (!hdr_info.crc_absent)
2481 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2482 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2484 AACContext *ac = avctx->priv_data;
2485 ChannelElement *che = NULL, *che_prev = NULL;
2486 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2488 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2489 int is_dmono, sce_count = 0;
2493 if (show_bits(gb, 12) == 0xfff) {
2494 if (parse_adts_frame_header(ac, gb) < 0) {
2495 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2499 if (ac->oc[1].m4ac.sampling_index > 12) {
2500 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2506 if (frame_configure_elements(avctx) < 0) {
2511 ac->tags_mapped = 0;
2513 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2514 elem_id = get_bits(gb, 4);
2516 if (elem_type < TYPE_DSE) {
2517 if (!(che=get_che(ac, elem_type, elem_id))) {
2518 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2519 elem_type, elem_id);
2526 switch (elem_type) {
2529 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2535 err = decode_cpe(ac, gb, che);
2540 err = decode_cce(ac, gb, che);
2544 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2549 err = skip_data_stream_element(ac, gb);
2553 uint8_t layout_map[MAX_ELEM_ID*4][3];
2555 push_output_configuration(ac);
2556 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2562 av_log(avctx, AV_LOG_ERROR,
2563 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2565 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2567 ac->oc[1].m4ac.chan_config = 0;
2575 elem_id += get_bits(gb, 8) - 1;
2576 if (get_bits_left(gb) < 8 * elem_id) {
2577 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2582 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2583 err = 0; /* FIXME */
2587 err = -1; /* should not happen, but keeps compiler happy */
2592 elem_type_prev = elem_type;
2597 if (get_bits_left(gb) < 3) {
2598 av_log(avctx, AV_LOG_ERROR, overread_err);
2604 spectral_to_sample(ac);
2606 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2607 samples <<= multiplier;
2608 /* for dual-mono audio (SCE + SCE) */
2609 is_dmono = ac->dmono_mode && sce_count == 2 &&
2610 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2613 ac->frame->nb_samples = samples;
2615 av_frame_unref(ac->frame);
2616 *got_frame_ptr = !!samples;
2619 if (ac->dmono_mode == 1)
2620 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2621 else if (ac->dmono_mode == 2)
2622 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2625 if (ac->oc[1].status && audio_found) {
2626 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2627 avctx->frame_size = samples;
2628 ac->oc[1].status = OC_LOCKED;
2633 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2634 if (side && side_size>=4)
2635 AV_WL32(side, 2*AV_RL32(side));
2639 pop_output_configuration(ac);
2643 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2644 int *got_frame_ptr, AVPacket *avpkt)
2646 AACContext *ac = avctx->priv_data;
2647 const uint8_t *buf = avpkt->data;
2648 int buf_size = avpkt->size;
2653 int new_extradata_size;
2654 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2655 AV_PKT_DATA_NEW_EXTRADATA,
2656 &new_extradata_size);
2657 int jp_dualmono_size;
2658 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2659 AV_PKT_DATA_JP_DUALMONO,
2662 if (new_extradata && 0) {
2663 av_free(avctx->extradata);
2664 avctx->extradata = av_mallocz(new_extradata_size +
2665 FF_INPUT_BUFFER_PADDING_SIZE);
2666 if (!avctx->extradata)
2667 return AVERROR(ENOMEM);
2668 avctx->extradata_size = new_extradata_size;
2669 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2670 push_output_configuration(ac);
2671 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2673 avctx->extradata_size*8, 1) < 0) {
2674 pop_output_configuration(ac);
2675 return AVERROR_INVALIDDATA;
2680 if (jp_dualmono && jp_dualmono_size > 0)
2681 ac->dmono_mode = 1 + *jp_dualmono;
2682 if (ac->force_dmono_mode >= 0)
2683 ac->dmono_mode = ac->force_dmono_mode;
2685 if (INT_MAX / 8 <= buf_size)
2686 return AVERROR_INVALIDDATA;
2688 init_get_bits(&gb, buf, buf_size * 8);
2690 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2693 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2694 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2695 if (buf[buf_offset])
2698 return buf_size > buf_offset ? buf_consumed : buf_size;
2701 static av_cold int aac_decode_close(AVCodecContext *avctx)
2703 AACContext *ac = avctx->priv_data;
2706 for (i = 0; i < MAX_ELEM_ID; i++) {
2707 for (type = 0; type < 4; type++) {
2708 if (ac->che[type][i])
2709 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2710 av_freep(&ac->che[type][i]);
2714 ff_mdct_end(&ac->mdct);
2715 ff_mdct_end(&ac->mdct_small);
2716 ff_mdct_end(&ac->mdct_ltp);
2721 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2723 struct LATMContext {
2724 AACContext aac_ctx; ///< containing AACContext
2725 int initialized; ///< initialized after a valid extradata was seen
2728 int audio_mux_version_A; ///< LATM syntax version
2729 int frame_length_type; ///< 0/1 variable/fixed frame length
2730 int frame_length; ///< frame length for fixed frame length
2733 static inline uint32_t latm_get_value(GetBitContext *b)
2735 int length = get_bits(b, 2);
2737 return get_bits_long(b, (length+1)*8);
2740 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2741 GetBitContext *gb, int asclen)
2743 AACContext *ac = &latmctx->aac_ctx;
2744 AVCodecContext *avctx = ac->avctx;
2745 MPEG4AudioConfig m4ac = { 0 };
2746 int config_start_bit = get_bits_count(gb);
2747 int sync_extension = 0;
2748 int bits_consumed, esize;
2752 asclen = FFMIN(asclen, get_bits_left(gb));
2754 asclen = get_bits_left(gb);
2756 if (config_start_bit % 8) {
2757 avpriv_request_sample(latmctx->aac_ctx.avctx,
2758 "Non-byte-aligned audio-specific config");
2759 return AVERROR_PATCHWELCOME;
2762 return AVERROR_INVALIDDATA;
2763 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2764 gb->buffer + (config_start_bit / 8),
2765 asclen, sync_extension);
2767 if (bits_consumed < 0)
2768 return AVERROR_INVALIDDATA;
2770 if (!latmctx->initialized ||
2771 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2772 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2774 if(latmctx->initialized) {
2775 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2777 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
2779 latmctx->initialized = 0;
2781 esize = (bits_consumed+7) / 8;
2783 if (avctx->extradata_size < esize) {
2784 av_free(avctx->extradata);
2785 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2786 if (!avctx->extradata)
2787 return AVERROR(ENOMEM);
2790 avctx->extradata_size = esize;
2791 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2792 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2794 skip_bits_long(gb, bits_consumed);
2796 return bits_consumed;
2799 static int read_stream_mux_config(struct LATMContext *latmctx,
2802 int ret, audio_mux_version = get_bits(gb, 1);
2804 latmctx->audio_mux_version_A = 0;
2805 if (audio_mux_version)
2806 latmctx->audio_mux_version_A = get_bits(gb, 1);
2808 if (!latmctx->audio_mux_version_A) {
2810 if (audio_mux_version)
2811 latm_get_value(gb); // taraFullness
2813 skip_bits(gb, 1); // allStreamSameTimeFraming
2814 skip_bits(gb, 6); // numSubFrames
2816 if (get_bits(gb, 4)) { // numPrograms
2817 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
2818 return AVERROR_PATCHWELCOME;
2821 // for each program (which there is only one in DVB)
2823 // for each layer (which there is only one in DVB)
2824 if (get_bits(gb, 3)) { // numLayer
2825 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
2826 return AVERROR_PATCHWELCOME;
2829 // for all but first stream: use_same_config = get_bits(gb, 1);
2830 if (!audio_mux_version) {
2831 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2834 int ascLen = latm_get_value(gb);
2835 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2838 skip_bits_long(gb, ascLen);
2841 latmctx->frame_length_type = get_bits(gb, 3);
2842 switch (latmctx->frame_length_type) {
2844 skip_bits(gb, 8); // latmBufferFullness
2847 latmctx->frame_length = get_bits(gb, 9);
2852 skip_bits(gb, 6); // CELP frame length table index
2856 skip_bits(gb, 1); // HVXC frame length table index
2860 if (get_bits(gb, 1)) { // other data
2861 if (audio_mux_version) {
2862 latm_get_value(gb); // other_data_bits
2866 esc = get_bits(gb, 1);
2872 if (get_bits(gb, 1)) // crc present
2873 skip_bits(gb, 8); // config_crc
2879 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2883 if (ctx->frame_length_type == 0) {
2884 int mux_slot_length = 0;
2886 tmp = get_bits(gb, 8);
2887 mux_slot_length += tmp;
2888 } while (tmp == 255);
2889 return mux_slot_length;
2890 } else if (ctx->frame_length_type == 1) {
2891 return ctx->frame_length;
2892 } else if (ctx->frame_length_type == 3 ||
2893 ctx->frame_length_type == 5 ||
2894 ctx->frame_length_type == 7) {
2895 skip_bits(gb, 2); // mux_slot_length_coded
2900 static int read_audio_mux_element(struct LATMContext *latmctx,
2904 uint8_t use_same_mux = get_bits(gb, 1);
2905 if (!use_same_mux) {
2906 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2908 } else if (!latmctx->aac_ctx.avctx->extradata) {
2909 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2910 "no decoder config found\n");
2911 return AVERROR(EAGAIN);
2913 if (latmctx->audio_mux_version_A == 0) {
2914 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2915 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2916 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2917 return AVERROR_INVALIDDATA;
2918 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2919 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2920 "frame length mismatch %d << %d\n",
2921 mux_slot_length_bytes * 8, get_bits_left(gb));
2922 return AVERROR_INVALIDDATA;
2929 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2930 int *got_frame_ptr, AVPacket *avpkt)
2932 struct LATMContext *latmctx = avctx->priv_data;
2936 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
2939 // check for LOAS sync word
2940 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2941 return AVERROR_INVALIDDATA;
2943 muxlength = get_bits(&gb, 13) + 3;
2944 // not enough data, the parser should have sorted this out
2945 if (muxlength > avpkt->size)
2946 return AVERROR_INVALIDDATA;
2948 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2951 if (!latmctx->initialized) {
2952 if (!avctx->extradata) {
2956 push_output_configuration(&latmctx->aac_ctx);
2957 if ((err = decode_audio_specific_config(
2958 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2959 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2960 pop_output_configuration(&latmctx->aac_ctx);
2963 latmctx->initialized = 1;
2967 if (show_bits(&gb, 12) == 0xfff) {
2968 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2969 "ADTS header detected, probably as result of configuration "
2971 return AVERROR_INVALIDDATA;
2974 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2980 static av_cold int latm_decode_init(AVCodecContext *avctx)
2982 struct LATMContext *latmctx = avctx->priv_data;
2983 int ret = aac_decode_init(avctx);
2985 if (avctx->extradata_size > 0)
2986 latmctx->initialized = !ret;
2991 static void aacdec_init(AACContext *c)
2993 c->imdct_and_windowing = imdct_and_windowing;
2994 c->apply_ltp = apply_ltp;
2995 c->apply_tns = apply_tns;
2996 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
2997 c->update_ltp = update_ltp;
3000 ff_aacdec_init_mips(c);
3003 * AVOptions for Japanese DTV specific extensions (ADTS only)
3005 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3006 static const AVOption options[] = {
3007 {"dual_mono_mode", "Select the channel to decode for dual mono",
3008 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3009 AACDEC_FLAGS, "dual_mono_mode"},
3011 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3012 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3013 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3014 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3019 static const AVClass aac_decoder_class = {
3020 .class_name = "AAC decoder",
3021 .item_name = av_default_item_name,
3023 .version = LIBAVUTIL_VERSION_INT,
3026 AVCodec ff_aac_decoder = {
3028 .type = AVMEDIA_TYPE_AUDIO,
3029 .id = AV_CODEC_ID_AAC,
3030 .priv_data_size = sizeof(AACContext),
3031 .init = aac_decode_init,
3032 .close = aac_decode_close,
3033 .decode = aac_decode_frame,
3034 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3035 .sample_fmts = (const enum AVSampleFormat[]) {
3036 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3038 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3039 .channel_layouts = aac_channel_layout,
3041 .priv_class = &aac_decoder_class,
3045 Note: This decoder filter is intended to decode LATM streams transferred
3046 in MPEG transport streams which only contain one program.
3047 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3049 AVCodec ff_aac_latm_decoder = {
3051 .type = AVMEDIA_TYPE_AUDIO,
3052 .id = AV_CODEC_ID_AAC_LATM,
3053 .priv_data_size = sizeof(struct LATMContext),
3054 .init = latm_decode_init,
3055 .close = aac_decode_close,
3056 .decode = latm_decode_frame,
3057 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3058 .sample_fmts = (const enum AVSampleFormat[]) {
3059 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3061 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3062 .channel_layouts = aac_channel_layout,