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 ac->frame->nb_samples = 2048;
194 if ((ret = ff_get_buffer(avctx, ac->frame)) < 0) {
195 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
199 /* map output channel pointers to AVFrame data */
200 for (ch = 0; ch < avctx->channels; ch++) {
201 if (ac->output_element[ch])
202 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
208 struct elem_to_channel {
209 uint64_t av_position;
212 uint8_t aac_position;
215 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
216 uint8_t (*layout_map)[3], int offset, uint64_t left,
217 uint64_t right, int pos)
219 if (layout_map[offset][0] == TYPE_CPE) {
220 e2c_vec[offset] = (struct elem_to_channel) {
221 .av_position = left | right, .syn_ele = TYPE_CPE,
222 .elem_id = layout_map[offset ][1], .aac_position = pos };
225 e2c_vec[offset] = (struct elem_to_channel) {
226 .av_position = left, .syn_ele = TYPE_SCE,
227 .elem_id = layout_map[offset ][1], .aac_position = pos };
228 e2c_vec[offset + 1] = (struct elem_to_channel) {
229 .av_position = right, .syn_ele = TYPE_SCE,
230 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
235 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
236 int num_pos_channels = 0;
240 for (i = *current; i < tags; i++) {
241 if (layout_map[i][2] != pos)
243 if (layout_map[i][0] == TYPE_CPE) {
245 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
251 num_pos_channels += 2;
259 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
262 return num_pos_channels;
265 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
267 int i, n, total_non_cc_elements;
268 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
269 int num_front_channels, num_side_channels, num_back_channels;
272 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
277 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
278 if (num_front_channels < 0)
281 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
282 if (num_side_channels < 0)
285 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
286 if (num_back_channels < 0)
290 if (num_front_channels & 1) {
291 e2c_vec[i] = (struct elem_to_channel) {
292 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
293 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
295 num_front_channels--;
297 if (num_front_channels >= 4) {
298 i += assign_pair(e2c_vec, layout_map, i,
299 AV_CH_FRONT_LEFT_OF_CENTER,
300 AV_CH_FRONT_RIGHT_OF_CENTER,
302 num_front_channels -= 2;
304 if (num_front_channels >= 2) {
305 i += assign_pair(e2c_vec, layout_map, i,
309 num_front_channels -= 2;
311 while (num_front_channels >= 2) {
312 i += assign_pair(e2c_vec, layout_map, i,
316 num_front_channels -= 2;
319 if (num_side_channels >= 2) {
320 i += assign_pair(e2c_vec, layout_map, i,
324 num_side_channels -= 2;
326 while (num_side_channels >= 2) {
327 i += assign_pair(e2c_vec, layout_map, i,
331 num_side_channels -= 2;
334 while (num_back_channels >= 4) {
335 i += assign_pair(e2c_vec, layout_map, i,
339 num_back_channels -= 2;
341 if (num_back_channels >= 2) {
342 i += assign_pair(e2c_vec, layout_map, i,
346 num_back_channels -= 2;
348 if (num_back_channels) {
349 e2c_vec[i] = (struct elem_to_channel) {
350 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
351 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
356 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
357 e2c_vec[i] = (struct elem_to_channel) {
358 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
359 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
362 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
363 e2c_vec[i] = (struct elem_to_channel) {
364 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
365 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
369 // Must choose a stable sort
370 total_non_cc_elements = n = i;
373 for (i = 1; i < n; i++) {
374 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
375 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
383 for (i = 0; i < total_non_cc_elements; i++) {
384 layout_map[i][0] = e2c_vec[i].syn_ele;
385 layout_map[i][1] = e2c_vec[i].elem_id;
386 layout_map[i][2] = e2c_vec[i].aac_position;
387 if (e2c_vec[i].av_position != UINT64_MAX) {
388 layout |= e2c_vec[i].av_position;
396 * Save current output configuration if and only if it has been locked.
398 static void push_output_configuration(AACContext *ac) {
399 if (ac->oc[1].status == OC_LOCKED) {
400 ac->oc[0] = ac->oc[1];
402 ac->oc[1].status = OC_NONE;
406 * Restore the previous output configuration if and only if the current
407 * configuration is unlocked.
409 static void pop_output_configuration(AACContext *ac) {
410 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
411 ac->oc[1] = ac->oc[0];
412 ac->avctx->channels = ac->oc[1].channels;
413 ac->avctx->channel_layout = ac->oc[1].channel_layout;
414 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
415 ac->oc[1].status, 0);
420 * Configure output channel order based on the current program configuration element.
422 * @return Returns error status. 0 - OK, !0 - error
424 static int output_configure(AACContext *ac,
425 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
426 enum OCStatus oc_type, int get_new_frame)
428 AVCodecContext *avctx = ac->avctx;
429 int i, channels = 0, ret;
432 if (ac->oc[1].layout_map != layout_map) {
433 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
434 ac->oc[1].layout_map_tags = tags;
437 // Try to sniff a reasonable channel order, otherwise output the
438 // channels in the order the PCE declared them.
439 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
440 layout = sniff_channel_order(layout_map, tags);
441 for (i = 0; i < tags; i++) {
442 int type = layout_map[i][0];
443 int id = layout_map[i][1];
444 int position = layout_map[i][2];
445 // Allocate or free elements depending on if they are in the
446 // current program configuration.
447 ret = che_configure(ac, position, type, id, &channels);
451 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
452 if (layout == AV_CH_FRONT_CENTER) {
453 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
459 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
460 if (layout) avctx->channel_layout = layout;
461 ac->oc[1].channel_layout = layout;
462 avctx->channels = ac->oc[1].channels = channels;
463 ac->oc[1].status = oc_type;
466 if ((ret = frame_configure_elements(ac->avctx)) < 0)
473 static void flush(AVCodecContext *avctx)
475 AACContext *ac= avctx->priv_data;
478 for (type = 3; type >= 0; type--) {
479 for (i = 0; i < MAX_ELEM_ID; i++) {
480 ChannelElement *che = ac->che[type][i];
482 for (j = 0; j <= 1; j++) {
483 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
491 * Set up channel positions based on a default channel configuration
492 * as specified in table 1.17.
494 * @return Returns error status. 0 - OK, !0 - error
496 static int set_default_channel_config(AVCodecContext *avctx,
497 uint8_t (*layout_map)[3],
501 if (channel_config < 1 || channel_config > 7) {
502 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
506 *tags = tags_per_config[channel_config];
507 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
510 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
511 * However, at least Nero AAC encoder encodes 7.1 streams using the default
512 * channel config 7, mapping the side channels of the original audio stream
513 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
514 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
515 * the incorrect streams as if they were correct (and as the encoder intended).
517 * As actual intended 7.1(wide) streams are very rare, default to assuming a
518 * 7.1 layout was intended.
520 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
521 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
522 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
523 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
524 layout_map[2][2] = AAC_CHANNEL_SIDE;
530 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
532 // For PCE based channel configurations map the channels solely based on tags.
533 if (!ac->oc[1].m4ac.chan_config) {
534 return ac->tag_che_map[type][elem_id];
536 // Allow single CPE stereo files to be signalled with mono configuration.
537 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
538 uint8_t layout_map[MAX_ELEM_ID*4][3];
540 push_output_configuration(ac);
542 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\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 = 2;
552 ac->oc[1].m4ac.ps = 0;
555 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
556 uint8_t layout_map[MAX_ELEM_ID*4][3];
558 push_output_configuration(ac);
560 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
562 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
565 if (output_configure(ac, layout_map, layout_map_tags,
566 OC_TRIAL_FRAME, 1) < 0)
569 ac->oc[1].m4ac.chan_config = 1;
570 if (ac->oc[1].m4ac.sbr)
571 ac->oc[1].m4ac.ps = -1;
573 // For indexed channel configurations map the channels solely based on position.
574 switch (ac->oc[1].m4ac.chan_config) {
576 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
578 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
581 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
582 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
583 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
584 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
586 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
589 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
591 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
594 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
596 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
600 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
602 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
603 } else if (ac->oc[1].m4ac.chan_config == 2) {
607 if (!ac->tags_mapped && type == TYPE_SCE) {
609 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
617 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
619 * @param type speaker type/position for these channels
621 static void decode_channel_map(uint8_t layout_map[][3],
622 enum ChannelPosition type,
623 GetBitContext *gb, int n)
626 enum RawDataBlockType syn_ele;
628 case AAC_CHANNEL_FRONT:
629 case AAC_CHANNEL_BACK:
630 case AAC_CHANNEL_SIDE:
631 syn_ele = get_bits1(gb);
637 case AAC_CHANNEL_LFE:
643 layout_map[0][0] = syn_ele;
644 layout_map[0][1] = get_bits(gb, 4);
645 layout_map[0][2] = type;
651 * Decode program configuration element; reference: table 4.2.
653 * @return Returns error status. 0 - OK, !0 - error
655 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
656 uint8_t (*layout_map)[3],
659 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
663 skip_bits(gb, 2); // object_type
665 sampling_index = get_bits(gb, 4);
666 if (m4ac->sampling_index != sampling_index)
667 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");
669 num_front = get_bits(gb, 4);
670 num_side = get_bits(gb, 4);
671 num_back = get_bits(gb, 4);
672 num_lfe = get_bits(gb, 2);
673 num_assoc_data = get_bits(gb, 3);
674 num_cc = get_bits(gb, 4);
677 skip_bits(gb, 4); // mono_mixdown_tag
679 skip_bits(gb, 4); // stereo_mixdown_tag
682 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
684 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
685 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
688 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
690 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
692 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
694 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
697 skip_bits_long(gb, 4 * num_assoc_data);
699 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
704 /* comment field, first byte is length */
705 comment_len = get_bits(gb, 8) * 8;
706 if (get_bits_left(gb) < comment_len) {
707 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
710 skip_bits_long(gb, comment_len);
715 * Decode GA "General Audio" specific configuration; reference: table 4.1.
717 * @param ac pointer to AACContext, may be null
718 * @param avctx pointer to AVCCodecContext, used for logging
720 * @return Returns error status. 0 - OK, !0 - error
722 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
724 MPEG4AudioConfig *m4ac,
727 int extension_flag, ret;
728 uint8_t layout_map[MAX_ELEM_ID*4][3];
731 if (get_bits1(gb)) { // frameLengthFlag
732 av_log_missing_feature(avctx, "960/120 MDCT window", 1);
733 return AVERROR_PATCHWELCOME;
736 if (get_bits1(gb)) // dependsOnCoreCoder
737 skip_bits(gb, 14); // coreCoderDelay
738 extension_flag = get_bits1(gb);
740 if (m4ac->object_type == AOT_AAC_SCALABLE ||
741 m4ac->object_type == AOT_ER_AAC_SCALABLE)
742 skip_bits(gb, 3); // layerNr
744 if (channel_config == 0) {
745 skip_bits(gb, 4); // element_instance_tag
746 tags = decode_pce(avctx, m4ac, layout_map, gb);
750 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
754 if (count_channels(layout_map, tags) > 1) {
756 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
759 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
762 if (extension_flag) {
763 switch (m4ac->object_type) {
765 skip_bits(gb, 5); // numOfSubFrame
766 skip_bits(gb, 11); // layer_length
770 case AOT_ER_AAC_SCALABLE:
772 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
773 * aacScalefactorDataResilienceFlag
774 * aacSpectralDataResilienceFlag
778 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
784 * Decode audio specific configuration; reference: table 1.13.
786 * @param ac pointer to AACContext, may be null
787 * @param avctx pointer to AVCCodecContext, used for logging
788 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
789 * @param data pointer to buffer holding an audio specific config
790 * @param bit_size size of audio specific config or data in bits
791 * @param sync_extension look for an appended sync extension
793 * @return Returns error status or number of consumed bits. <0 - error
795 static int decode_audio_specific_config(AACContext *ac,
796 AVCodecContext *avctx,
797 MPEG4AudioConfig *m4ac,
798 const uint8_t *data, int bit_size,
805 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
806 for (i = 0; i < bit_size >> 3; i++)
807 av_dlog(avctx, "%02x ", data[i]);
808 av_dlog(avctx, "\n");
810 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
813 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
815 if (m4ac->sampling_index > 12) {
816 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
820 skip_bits_long(&gb, i);
822 switch (m4ac->object_type) {
826 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
830 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
831 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
835 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
836 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
837 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
839 return get_bits_count(&gb);
843 * linear congruential pseudorandom number generator
845 * @param previous_val pointer to the current state of the generator
847 * @return Returns a 32-bit pseudorandom integer
849 static av_always_inline int lcg_random(unsigned previous_val)
851 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
855 static av_always_inline void reset_predict_state(PredictorState *ps)
865 static void reset_all_predictors(PredictorState *ps)
868 for (i = 0; i < MAX_PREDICTORS; i++)
869 reset_predict_state(&ps[i]);
872 static int sample_rate_idx (int rate)
874 if (92017 <= rate) return 0;
875 else if (75132 <= rate) return 1;
876 else if (55426 <= rate) return 2;
877 else if (46009 <= rate) return 3;
878 else if (37566 <= rate) return 4;
879 else if (27713 <= rate) return 5;
880 else if (23004 <= rate) return 6;
881 else if (18783 <= rate) return 7;
882 else if (13856 <= rate) return 8;
883 else if (11502 <= rate) return 9;
884 else if (9391 <= rate) return 10;
888 static void reset_predictor_group(PredictorState *ps, int group_num)
891 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
892 reset_predict_state(&ps[i]);
895 #define AAC_INIT_VLC_STATIC(num, size) \
896 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
897 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
898 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
901 static void aacdec_init(AACContext *ac);
903 static av_cold int aac_decode_init(AVCodecContext *avctx)
905 AACContext *ac = avctx->priv_data;
908 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
912 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
914 if (avctx->extradata_size > 0) {
915 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
917 avctx->extradata_size*8, 1) < 0)
921 uint8_t layout_map[MAX_ELEM_ID*4][3];
924 sr = sample_rate_idx(avctx->sample_rate);
925 ac->oc[1].m4ac.sampling_index = sr;
926 ac->oc[1].m4ac.channels = avctx->channels;
927 ac->oc[1].m4ac.sbr = -1;
928 ac->oc[1].m4ac.ps = -1;
930 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
931 if (ff_mpeg4audio_channels[i] == avctx->channels)
933 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
936 ac->oc[1].m4ac.chan_config = i;
938 if (ac->oc[1].m4ac.chan_config) {
939 int ret = set_default_channel_config(avctx, layout_map,
940 &layout_map_tags, ac->oc[1].m4ac.chan_config);
942 output_configure(ac, layout_map, layout_map_tags,
944 else if (avctx->err_recognition & AV_EF_EXPLODE)
945 return AVERROR_INVALIDDATA;
949 if (avctx->channels > MAX_CHANNELS) {
950 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
951 return AVERROR_INVALIDDATA;
954 AAC_INIT_VLC_STATIC( 0, 304);
955 AAC_INIT_VLC_STATIC( 1, 270);
956 AAC_INIT_VLC_STATIC( 2, 550);
957 AAC_INIT_VLC_STATIC( 3, 300);
958 AAC_INIT_VLC_STATIC( 4, 328);
959 AAC_INIT_VLC_STATIC( 5, 294);
960 AAC_INIT_VLC_STATIC( 6, 306);
961 AAC_INIT_VLC_STATIC( 7, 268);
962 AAC_INIT_VLC_STATIC( 8, 510);
963 AAC_INIT_VLC_STATIC( 9, 366);
964 AAC_INIT_VLC_STATIC(10, 462);
968 ff_fmt_convert_init(&ac->fmt_conv, avctx);
969 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
971 ac->random_state = 0x1f2e3d4c;
975 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
976 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
977 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
980 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
981 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
982 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
983 // window initialization
984 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
985 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
986 ff_init_ff_sine_windows(10);
987 ff_init_ff_sine_windows( 7);
995 * Skip data_stream_element; reference: table 4.10.
997 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
999 int byte_align = get_bits1(gb);
1000 int count = get_bits(gb, 8);
1002 count += get_bits(gb, 8);
1006 if (get_bits_left(gb) < 8 * count) {
1007 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1010 skip_bits_long(gb, 8 * count);
1014 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1018 if (get_bits1(gb)) {
1019 ics->predictor_reset_group = get_bits(gb, 5);
1020 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
1021 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
1025 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1026 ics->prediction_used[sfb] = get_bits1(gb);
1032 * Decode Long Term Prediction data; reference: table 4.xx.
1034 static void decode_ltp(LongTermPrediction *ltp,
1035 GetBitContext *gb, uint8_t max_sfb)
1039 ltp->lag = get_bits(gb, 11);
1040 ltp->coef = ltp_coef[get_bits(gb, 3)];
1041 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1042 ltp->used[sfb] = get_bits1(gb);
1046 * Decode Individual Channel Stream info; reference: table 4.6.
1048 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1051 if (get_bits1(gb)) {
1052 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1053 return AVERROR_INVALIDDATA;
1055 ics->window_sequence[1] = ics->window_sequence[0];
1056 ics->window_sequence[0] = get_bits(gb, 2);
1057 ics->use_kb_window[1] = ics->use_kb_window[0];
1058 ics->use_kb_window[0] = get_bits1(gb);
1059 ics->num_window_groups = 1;
1060 ics->group_len[0] = 1;
1061 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1063 ics->max_sfb = get_bits(gb, 4);
1064 for (i = 0; i < 7; i++) {
1065 if (get_bits1(gb)) {
1066 ics->group_len[ics->num_window_groups - 1]++;
1068 ics->num_window_groups++;
1069 ics->group_len[ics->num_window_groups - 1] = 1;
1072 ics->num_windows = 8;
1073 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1074 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1075 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1076 ics->predictor_present = 0;
1078 ics->max_sfb = get_bits(gb, 6);
1079 ics->num_windows = 1;
1080 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1081 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1082 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1083 ics->predictor_present = get_bits1(gb);
1084 ics->predictor_reset_group = 0;
1085 if (ics->predictor_present) {
1086 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1087 if (decode_prediction(ac, ics, gb)) {
1090 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1091 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1094 if ((ics->ltp.present = get_bits(gb, 1)))
1095 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1100 if (ics->max_sfb > ics->num_swb) {
1101 av_log(ac->avctx, AV_LOG_ERROR,
1102 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1103 ics->max_sfb, ics->num_swb);
1110 return AVERROR_INVALIDDATA;
1114 * Decode band types (section_data payload); reference: table 4.46.
1116 * @param band_type array of the used band type
1117 * @param band_type_run_end array of the last scalefactor band of a band type run
1119 * @return Returns error status. 0 - OK, !0 - error
1121 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1122 int band_type_run_end[120], GetBitContext *gb,
1123 IndividualChannelStream *ics)
1126 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1127 for (g = 0; g < ics->num_window_groups; g++) {
1129 while (k < ics->max_sfb) {
1130 uint8_t sect_end = k;
1132 int sect_band_type = get_bits(gb, 4);
1133 if (sect_band_type == 12) {
1134 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1138 sect_len_incr = get_bits(gb, bits);
1139 sect_end += sect_len_incr;
1140 if (get_bits_left(gb) < 0) {
1141 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1144 if (sect_end > ics->max_sfb) {
1145 av_log(ac->avctx, AV_LOG_ERROR,
1146 "Number of bands (%d) exceeds limit (%d).\n",
1147 sect_end, ics->max_sfb);
1150 } while (sect_len_incr == (1 << bits) - 1);
1151 for (; k < sect_end; k++) {
1152 band_type [idx] = sect_band_type;
1153 band_type_run_end[idx++] = sect_end;
1161 * Decode scalefactors; reference: table 4.47.
1163 * @param global_gain first scalefactor value as scalefactors are differentially coded
1164 * @param band_type array of the used band type
1165 * @param band_type_run_end array of the last scalefactor band of a band type run
1166 * @param sf array of scalefactors or intensity stereo positions
1168 * @return Returns error status. 0 - OK, !0 - error
1170 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1171 unsigned int global_gain,
1172 IndividualChannelStream *ics,
1173 enum BandType band_type[120],
1174 int band_type_run_end[120])
1177 int offset[3] = { global_gain, global_gain - 90, 0 };
1180 for (g = 0; g < ics->num_window_groups; g++) {
1181 for (i = 0; i < ics->max_sfb;) {
1182 int run_end = band_type_run_end[idx];
1183 if (band_type[idx] == ZERO_BT) {
1184 for (; i < run_end; i++, idx++)
1186 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1187 for (; i < run_end; i++, idx++) {
1188 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1189 clipped_offset = av_clip(offset[2], -155, 100);
1190 if (offset[2] != clipped_offset) {
1191 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1192 "position clipped (%d -> %d).\nIf you heard an "
1193 "audible artifact, there may be a bug in the "
1194 "decoder. ", offset[2], clipped_offset);
1196 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1198 } else if (band_type[idx] == NOISE_BT) {
1199 for (; i < run_end; i++, idx++) {
1200 if (noise_flag-- > 0)
1201 offset[1] += get_bits(gb, 9) - 256;
1203 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1204 clipped_offset = av_clip(offset[1], -100, 155);
1205 if (offset[1] != clipped_offset) {
1206 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1207 "(%d -> %d).\nIf you heard an audible "
1208 "artifact, there may be a bug in the decoder. ",
1209 offset[1], clipped_offset);
1211 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1214 for (; i < run_end; i++, idx++) {
1215 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1216 if (offset[0] > 255U) {
1217 av_log(ac->avctx, AV_LOG_ERROR,
1218 "Scalefactor (%d) out of range.\n", offset[0]);
1221 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1230 * Decode pulse data; reference: table 4.7.
1232 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1233 const uint16_t *swb_offset, int num_swb)
1236 pulse->num_pulse = get_bits(gb, 2) + 1;
1237 pulse_swb = get_bits(gb, 6);
1238 if (pulse_swb >= num_swb)
1240 pulse->pos[0] = swb_offset[pulse_swb];
1241 pulse->pos[0] += get_bits(gb, 5);
1242 if (pulse->pos[0] > 1023)
1244 pulse->amp[0] = get_bits(gb, 4);
1245 for (i = 1; i < pulse->num_pulse; i++) {
1246 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1247 if (pulse->pos[i] > 1023)
1249 pulse->amp[i] = get_bits(gb, 4);
1255 * Decode Temporal Noise Shaping data; reference: table 4.48.
1257 * @return Returns error status. 0 - OK, !0 - error
1259 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1260 GetBitContext *gb, const IndividualChannelStream *ics)
1262 int w, filt, i, coef_len, coef_res, coef_compress;
1263 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1264 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1265 for (w = 0; w < ics->num_windows; w++) {
1266 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1267 coef_res = get_bits1(gb);
1269 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1271 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1273 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1274 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1275 tns->order[w][filt], tns_max_order);
1276 tns->order[w][filt] = 0;
1279 if (tns->order[w][filt]) {
1280 tns->direction[w][filt] = get_bits1(gb);
1281 coef_compress = get_bits1(gb);
1282 coef_len = coef_res + 3 - coef_compress;
1283 tmp2_idx = 2 * coef_compress + coef_res;
1285 for (i = 0; i < tns->order[w][filt]; i++)
1286 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1295 * Decode Mid/Side data; reference: table 4.54.
1297 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1298 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1299 * [3] reserved for scalable AAC
1301 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1305 if (ms_present == 1) {
1306 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1307 cpe->ms_mask[idx] = get_bits1(gb);
1308 } else if (ms_present == 2) {
1309 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1314 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1318 *dst++ = v[idx & 15] * s;
1319 *dst++ = v[idx>>4 & 15] * s;
1325 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1329 *dst++ = v[idx & 3] * s;
1330 *dst++ = v[idx>>2 & 3] * s;
1331 *dst++ = v[idx>>4 & 3] * s;
1332 *dst++ = v[idx>>6 & 3] * s;
1338 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1339 unsigned sign, const float *scale)
1341 union av_intfloat32 s0, s1;
1343 s0.f = s1.f = *scale;
1344 s0.i ^= sign >> 1 << 31;
1347 *dst++ = v[idx & 15] * s0.f;
1348 *dst++ = v[idx>>4 & 15] * s1.f;
1355 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1356 unsigned sign, const float *scale)
1358 unsigned nz = idx >> 12;
1359 union av_intfloat32 s = { .f = *scale };
1360 union av_intfloat32 t;
1362 t.i = s.i ^ (sign & 1U<<31);
1363 *dst++ = v[idx & 3] * t.f;
1365 sign <<= nz & 1; nz >>= 1;
1366 t.i = s.i ^ (sign & 1U<<31);
1367 *dst++ = v[idx>>2 & 3] * t.f;
1369 sign <<= nz & 1; nz >>= 1;
1370 t.i = s.i ^ (sign & 1U<<31);
1371 *dst++ = v[idx>>4 & 3] * t.f;
1374 t.i = s.i ^ (sign & 1U<<31);
1375 *dst++ = v[idx>>6 & 3] * t.f;
1382 * Decode spectral data; reference: table 4.50.
1383 * Dequantize and scale spectral data; reference: 4.6.3.3.
1385 * @param coef array of dequantized, scaled spectral data
1386 * @param sf array of scalefactors or intensity stereo positions
1387 * @param pulse_present set if pulses are present
1388 * @param pulse pointer to pulse data struct
1389 * @param band_type array of the used band type
1391 * @return Returns error status. 0 - OK, !0 - error
1393 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1394 GetBitContext *gb, const float sf[120],
1395 int pulse_present, const Pulse *pulse,
1396 const IndividualChannelStream *ics,
1397 enum BandType band_type[120])
1399 int i, k, g, idx = 0;
1400 const int c = 1024 / ics->num_windows;
1401 const uint16_t *offsets = ics->swb_offset;
1402 float *coef_base = coef;
1404 for (g = 0; g < ics->num_windows; g++)
1405 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1407 for (g = 0; g < ics->num_window_groups; g++) {
1408 unsigned g_len = ics->group_len[g];
1410 for (i = 0; i < ics->max_sfb; i++, idx++) {
1411 const unsigned cbt_m1 = band_type[idx] - 1;
1412 float *cfo = coef + offsets[i];
1413 int off_len = offsets[i + 1] - offsets[i];
1416 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1417 for (group = 0; group < g_len; group++, cfo+=128) {
1418 memset(cfo, 0, off_len * sizeof(float));
1420 } else if (cbt_m1 == NOISE_BT - 1) {
1421 for (group = 0; group < g_len; group++, cfo+=128) {
1425 for (k = 0; k < off_len; k++) {
1426 ac->random_state = lcg_random(ac->random_state);
1427 cfo[k] = ac->random_state;
1430 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1431 scale = sf[idx] / sqrtf(band_energy);
1432 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1435 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1436 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1437 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1438 OPEN_READER(re, gb);
1440 switch (cbt_m1 >> 1) {
1442 for (group = 0; group < g_len; group++, cfo+=128) {
1450 UPDATE_CACHE(re, gb);
1451 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1452 cb_idx = cb_vector_idx[code];
1453 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1459 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 nnz = cb_idx >> 8 & 15;
1473 bits = nnz ? GET_CACHE(re, gb) : 0;
1474 LAST_SKIP_BITS(re, gb, nnz);
1475 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1481 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 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1499 for (group = 0; group < g_len; group++, cfo+=128) {
1509 UPDATE_CACHE(re, gb);
1510 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1511 cb_idx = cb_vector_idx[code];
1512 nnz = cb_idx >> 8 & 15;
1513 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1514 LAST_SKIP_BITS(re, gb, nnz);
1515 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1521 for (group = 0; group < g_len; group++, cfo+=128) {
1523 uint32_t *icf = (uint32_t *) cf;
1533 UPDATE_CACHE(re, gb);
1534 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1542 cb_idx = cb_vector_idx[code];
1545 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1546 LAST_SKIP_BITS(re, gb, nnz);
1548 for (j = 0; j < 2; j++) {
1552 /* The total length of escape_sequence must be < 22 bits according
1553 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1554 UPDATE_CACHE(re, gb);
1555 b = GET_CACHE(re, gb);
1556 b = 31 - av_log2(~b);
1559 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1563 SKIP_BITS(re, gb, b + 1);
1565 n = (1 << b) + SHOW_UBITS(re, gb, b);
1566 LAST_SKIP_BITS(re, gb, b);
1567 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1570 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1571 *icf++ = (bits & 1U<<31) | v;
1578 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1582 CLOSE_READER(re, gb);
1588 if (pulse_present) {
1590 for (i = 0; i < pulse->num_pulse; i++) {
1591 float co = coef_base[ pulse->pos[i] ];
1592 while (offsets[idx + 1] <= pulse->pos[i])
1594 if (band_type[idx] != NOISE_BT && sf[idx]) {
1595 float ico = -pulse->amp[i];
1598 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1600 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1607 static av_always_inline float flt16_round(float pf)
1609 union av_intfloat32 tmp;
1611 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1615 static av_always_inline float flt16_even(float pf)
1617 union av_intfloat32 tmp;
1619 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1623 static av_always_inline float flt16_trunc(float pf)
1625 union av_intfloat32 pun;
1627 pun.i &= 0xFFFF0000U;
1631 static av_always_inline void predict(PredictorState *ps, float *coef,
1634 const float a = 0.953125; // 61.0 / 64
1635 const float alpha = 0.90625; // 29.0 / 32
1639 float r0 = ps->r0, r1 = ps->r1;
1640 float cor0 = ps->cor0, cor1 = ps->cor1;
1641 float var0 = ps->var0, var1 = ps->var1;
1643 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1644 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1646 pv = flt16_round(k1 * r0 + k2 * r1);
1653 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1654 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1655 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1656 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1658 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1659 ps->r0 = flt16_trunc(a * e0);
1663 * Apply AAC-Main style frequency domain prediction.
1665 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1669 if (!sce->ics.predictor_initialized) {
1670 reset_all_predictors(sce->predictor_state);
1671 sce->ics.predictor_initialized = 1;
1674 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1675 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1676 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1677 predict(&sce->predictor_state[k], &sce->coeffs[k],
1678 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1681 if (sce->ics.predictor_reset_group)
1682 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1684 reset_all_predictors(sce->predictor_state);
1688 * Decode an individual_channel_stream payload; reference: table 4.44.
1690 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1691 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1693 * @return Returns error status. 0 - OK, !0 - error
1695 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1696 GetBitContext *gb, int common_window, int scale_flag)
1699 TemporalNoiseShaping *tns = &sce->tns;
1700 IndividualChannelStream *ics = &sce->ics;
1701 float *out = sce->coeffs;
1702 int global_gain, pulse_present = 0;
1704 /* This assignment is to silence a GCC warning about the variable being used
1705 * uninitialized when in fact it always is.
1707 pulse.num_pulse = 0;
1709 global_gain = get_bits(gb, 8);
1711 if (!common_window && !scale_flag) {
1712 if (decode_ics_info(ac, ics, gb) < 0)
1713 return AVERROR_INVALIDDATA;
1716 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1718 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1723 if ((pulse_present = get_bits1(gb))) {
1724 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1725 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1728 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1729 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1733 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1735 if (get_bits1(gb)) {
1736 av_log_missing_feature(ac->avctx, "SSR", 1);
1737 return AVERROR_PATCHWELCOME;
1741 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1744 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1745 apply_prediction(ac, sce);
1751 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1753 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1755 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1756 float *ch0 = cpe->ch[0].coeffs;
1757 float *ch1 = cpe->ch[1].coeffs;
1758 int g, i, group, idx = 0;
1759 const uint16_t *offsets = ics->swb_offset;
1760 for (g = 0; g < ics->num_window_groups; g++) {
1761 for (i = 0; i < ics->max_sfb; i++, idx++) {
1762 if (cpe->ms_mask[idx] &&
1763 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1764 for (group = 0; group < ics->group_len[g]; group++) {
1765 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1766 ch1 + group * 128 + offsets[i],
1767 offsets[i+1] - offsets[i]);
1771 ch0 += ics->group_len[g] * 128;
1772 ch1 += ics->group_len[g] * 128;
1777 * intensity stereo decoding; reference: 4.6.8.2.3
1779 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1780 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1781 * [3] reserved for scalable AAC
1783 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1785 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1786 SingleChannelElement *sce1 = &cpe->ch[1];
1787 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1788 const uint16_t *offsets = ics->swb_offset;
1789 int g, group, i, idx = 0;
1792 for (g = 0; g < ics->num_window_groups; g++) {
1793 for (i = 0; i < ics->max_sfb;) {
1794 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1795 const int bt_run_end = sce1->band_type_run_end[idx];
1796 for (; i < bt_run_end; i++, idx++) {
1797 c = -1 + 2 * (sce1->band_type[idx] - 14);
1799 c *= 1 - 2 * cpe->ms_mask[idx];
1800 scale = c * sce1->sf[idx];
1801 for (group = 0; group < ics->group_len[g]; group++)
1802 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1803 coef0 + group * 128 + offsets[i],
1805 offsets[i + 1] - offsets[i]);
1808 int bt_run_end = sce1->band_type_run_end[idx];
1809 idx += bt_run_end - i;
1813 coef0 += ics->group_len[g] * 128;
1814 coef1 += ics->group_len[g] * 128;
1819 * Decode a channel_pair_element; reference: table 4.4.
1821 * @return Returns error status. 0 - OK, !0 - error
1823 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1825 int i, ret, common_window, ms_present = 0;
1827 common_window = get_bits1(gb);
1828 if (common_window) {
1829 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1830 return AVERROR_INVALIDDATA;
1831 i = cpe->ch[1].ics.use_kb_window[0];
1832 cpe->ch[1].ics = cpe->ch[0].ics;
1833 cpe->ch[1].ics.use_kb_window[1] = i;
1834 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1835 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1836 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1837 ms_present = get_bits(gb, 2);
1838 if (ms_present == 3) {
1839 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1841 } else if (ms_present)
1842 decode_mid_side_stereo(cpe, gb, ms_present);
1844 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1846 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1849 if (common_window) {
1851 apply_mid_side_stereo(ac, cpe);
1852 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1853 apply_prediction(ac, &cpe->ch[0]);
1854 apply_prediction(ac, &cpe->ch[1]);
1858 apply_intensity_stereo(ac, cpe, ms_present);
1862 static const float cce_scale[] = {
1863 1.09050773266525765921, //2^(1/8)
1864 1.18920711500272106672, //2^(1/4)
1870 * Decode coupling_channel_element; reference: table 4.8.
1872 * @return Returns error status. 0 - OK, !0 - error
1874 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1880 SingleChannelElement *sce = &che->ch[0];
1881 ChannelCoupling *coup = &che->coup;
1883 coup->coupling_point = 2 * get_bits1(gb);
1884 coup->num_coupled = get_bits(gb, 3);
1885 for (c = 0; c <= coup->num_coupled; c++) {
1887 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1888 coup->id_select[c] = get_bits(gb, 4);
1889 if (coup->type[c] == TYPE_CPE) {
1890 coup->ch_select[c] = get_bits(gb, 2);
1891 if (coup->ch_select[c] == 3)
1894 coup->ch_select[c] = 2;
1896 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1898 sign = get_bits(gb, 1);
1899 scale = cce_scale[get_bits(gb, 2)];
1901 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1904 for (c = 0; c < num_gain; c++) {
1908 float gain_cache = 1.;
1910 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1911 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1912 gain_cache = powf(scale, -gain);
1914 if (coup->coupling_point == AFTER_IMDCT) {
1915 coup->gain[c][0] = gain_cache;
1917 for (g = 0; g < sce->ics.num_window_groups; g++) {
1918 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1919 if (sce->band_type[idx] != ZERO_BT) {
1921 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1929 gain_cache = powf(scale, -t) * s;
1932 coup->gain[c][idx] = gain_cache;
1942 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1944 * @return Returns number of bytes consumed.
1946 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1950 int num_excl_chan = 0;
1953 for (i = 0; i < 7; i++)
1954 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1955 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1957 return num_excl_chan / 7;
1961 * Decode dynamic range information; reference: table 4.52.
1963 * @return Returns number of bytes consumed.
1965 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1969 int drc_num_bands = 1;
1972 /* pce_tag_present? */
1973 if (get_bits1(gb)) {
1974 che_drc->pce_instance_tag = get_bits(gb, 4);
1975 skip_bits(gb, 4); // tag_reserved_bits
1979 /* excluded_chns_present? */
1980 if (get_bits1(gb)) {
1981 n += decode_drc_channel_exclusions(che_drc, gb);
1984 /* drc_bands_present? */
1985 if (get_bits1(gb)) {
1986 che_drc->band_incr = get_bits(gb, 4);
1987 che_drc->interpolation_scheme = get_bits(gb, 4);
1989 drc_num_bands += che_drc->band_incr;
1990 for (i = 0; i < drc_num_bands; i++) {
1991 che_drc->band_top[i] = get_bits(gb, 8);
1996 /* prog_ref_level_present? */
1997 if (get_bits1(gb)) {
1998 che_drc->prog_ref_level = get_bits(gb, 7);
1999 skip_bits1(gb); // prog_ref_level_reserved_bits
2003 for (i = 0; i < drc_num_bands; i++) {
2004 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2005 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2012 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2014 int i, major, minor;
2019 get_bits(gb, 13); len -= 13;
2021 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2022 buf[i] = get_bits(gb, 8);
2025 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2026 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2028 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2029 ac->avctx->internal->skip_samples = 1024;
2033 skip_bits_long(gb, len);
2039 * Decode extension data (incomplete); reference: table 4.51.
2041 * @param cnt length of TYPE_FIL syntactic element in bytes
2043 * @return Returns number of bytes consumed
2045 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2046 ChannelElement *che, enum RawDataBlockType elem_type)
2050 switch (get_bits(gb, 4)) { // extension type
2051 case EXT_SBR_DATA_CRC:
2055 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2057 } else if (!ac->oc[1].m4ac.sbr) {
2058 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2059 skip_bits_long(gb, 8 * cnt - 4);
2061 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2062 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2063 skip_bits_long(gb, 8 * cnt - 4);
2065 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2066 ac->oc[1].m4ac.sbr = 1;
2067 ac->oc[1].m4ac.ps = 1;
2068 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2069 ac->oc[1].status, 1);
2071 ac->oc[1].m4ac.sbr = 1;
2073 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2075 case EXT_DYNAMIC_RANGE:
2076 res = decode_dynamic_range(&ac->che_drc, gb);
2079 decode_fill(ac, gb, 8 * cnt - 4);
2082 case EXT_DATA_ELEMENT:
2084 skip_bits_long(gb, 8 * cnt - 4);
2091 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2093 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2094 * @param coef spectral coefficients
2096 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2097 IndividualChannelStream *ics, int decode)
2099 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2101 int bottom, top, order, start, end, size, inc;
2102 float lpc[TNS_MAX_ORDER];
2103 float tmp[TNS_MAX_ORDER+1];
2105 for (w = 0; w < ics->num_windows; w++) {
2106 bottom = ics->num_swb;
2107 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2109 bottom = FFMAX(0, top - tns->length[w][filt]);
2110 order = tns->order[w][filt];
2115 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2117 start = ics->swb_offset[FFMIN(bottom, mmm)];
2118 end = ics->swb_offset[FFMIN( top, mmm)];
2119 if ((size = end - start) <= 0)
2121 if (tns->direction[w][filt]) {
2131 for (m = 0; m < size; m++, start += inc)
2132 for (i = 1; i <= FFMIN(m, order); i++)
2133 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2136 for (m = 0; m < size; m++, start += inc) {
2137 tmp[0] = coef[start];
2138 for (i = 1; i <= FFMIN(m, order); i++)
2139 coef[start] += tmp[i] * lpc[i - 1];
2140 for (i = order; i > 0; i--)
2141 tmp[i] = tmp[i - 1];
2149 * Apply windowing and MDCT to obtain the spectral
2150 * coefficient from the predicted sample by LTP.
2152 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2153 float *in, IndividualChannelStream *ics)
2155 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2156 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2157 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2158 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2160 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2161 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2163 memset(in, 0, 448 * sizeof(float));
2164 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2166 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2167 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2169 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2170 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2172 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2176 * Apply the long term prediction
2178 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2180 const LongTermPrediction *ltp = &sce->ics.ltp;
2181 const uint16_t *offsets = sce->ics.swb_offset;
2184 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2185 float *predTime = sce->ret;
2186 float *predFreq = ac->buf_mdct;
2187 int16_t num_samples = 2048;
2189 if (ltp->lag < 1024)
2190 num_samples = ltp->lag + 1024;
2191 for (i = 0; i < num_samples; i++)
2192 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2193 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2195 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2197 if (sce->tns.present)
2198 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2200 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2202 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2203 sce->coeffs[i] += predFreq[i];
2208 * Update the LTP buffer for next frame
2210 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2212 IndividualChannelStream *ics = &sce->ics;
2213 float *saved = sce->saved;
2214 float *saved_ltp = sce->coeffs;
2215 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2216 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2219 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2220 memcpy(saved_ltp, saved, 512 * sizeof(float));
2221 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2222 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2223 for (i = 0; i < 64; i++)
2224 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2225 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2226 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2227 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2228 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2229 for (i = 0; i < 64; i++)
2230 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2231 } else { // LONG_STOP or ONLY_LONG
2232 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2233 for (i = 0; i < 512; i++)
2234 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2237 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2238 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2239 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2243 * Conduct IMDCT and windowing.
2245 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2247 IndividualChannelStream *ics = &sce->ics;
2248 float *in = sce->coeffs;
2249 float *out = sce->ret;
2250 float *saved = sce->saved;
2251 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2252 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2253 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2254 float *buf = ac->buf_mdct;
2255 float *temp = ac->temp;
2259 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2260 for (i = 0; i < 1024; i += 128)
2261 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2263 ac->mdct.imdct_half(&ac->mdct, buf, in);
2265 /* window overlapping
2266 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2267 * and long to short transitions are considered to be short to short
2268 * transitions. This leaves just two cases (long to long and short to short)
2269 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2271 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2272 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2273 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2275 memcpy( out, saved, 448 * sizeof(float));
2277 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2278 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2279 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2280 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2281 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2282 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2283 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2285 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2286 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2291 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2292 memcpy( saved, temp + 64, 64 * sizeof(float));
2293 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2294 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2295 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2296 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2297 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2298 memcpy( saved, buf + 512, 448 * sizeof(float));
2299 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2300 } else { // LONG_STOP or ONLY_LONG
2301 memcpy( saved, buf + 512, 512 * sizeof(float));
2306 * Apply dependent channel coupling (applied before IMDCT).
2308 * @param index index into coupling gain array
2310 static void apply_dependent_coupling(AACContext *ac,
2311 SingleChannelElement *target,
2312 ChannelElement *cce, int index)
2314 IndividualChannelStream *ics = &cce->ch[0].ics;
2315 const uint16_t *offsets = ics->swb_offset;
2316 float *dest = target->coeffs;
2317 const float *src = cce->ch[0].coeffs;
2318 int g, i, group, k, idx = 0;
2319 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2320 av_log(ac->avctx, AV_LOG_ERROR,
2321 "Dependent coupling is not supported together with LTP\n");
2324 for (g = 0; g < ics->num_window_groups; g++) {
2325 for (i = 0; i < ics->max_sfb; i++, idx++) {
2326 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2327 const float gain = cce->coup.gain[index][idx];
2328 for (group = 0; group < ics->group_len[g]; group++) {
2329 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2331 dest[group * 128 + k] += gain * src[group * 128 + k];
2336 dest += ics->group_len[g] * 128;
2337 src += ics->group_len[g] * 128;
2342 * Apply independent channel coupling (applied after IMDCT).
2344 * @param index index into coupling gain array
2346 static void apply_independent_coupling(AACContext *ac,
2347 SingleChannelElement *target,
2348 ChannelElement *cce, int index)
2351 const float gain = cce->coup.gain[index][0];
2352 const float *src = cce->ch[0].ret;
2353 float *dest = target->ret;
2354 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2356 for (i = 0; i < len; i++)
2357 dest[i] += gain * src[i];
2361 * channel coupling transformation interface
2363 * @param apply_coupling_method pointer to (in)dependent coupling function
2365 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2366 enum RawDataBlockType type, int elem_id,
2367 enum CouplingPoint coupling_point,
2368 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2372 for (i = 0; i < MAX_ELEM_ID; i++) {
2373 ChannelElement *cce = ac->che[TYPE_CCE][i];
2376 if (cce && cce->coup.coupling_point == coupling_point) {
2377 ChannelCoupling *coup = &cce->coup;
2379 for (c = 0; c <= coup->num_coupled; c++) {
2380 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2381 if (coup->ch_select[c] != 1) {
2382 apply_coupling_method(ac, &cc->ch[0], cce, index);
2383 if (coup->ch_select[c] != 0)
2386 if (coup->ch_select[c] != 2)
2387 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2389 index += 1 + (coup->ch_select[c] == 3);
2396 * Convert spectral data to float samples, applying all supported tools as appropriate.
2398 static void spectral_to_sample(AACContext *ac)
2401 for (type = 3; type >= 0; type--) {
2402 for (i = 0; i < MAX_ELEM_ID; i++) {
2403 ChannelElement *che = ac->che[type][i];
2405 if (type <= TYPE_CPE)
2406 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2407 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2408 if (che->ch[0].ics.predictor_present) {
2409 if (che->ch[0].ics.ltp.present)
2410 ac->apply_ltp(ac, &che->ch[0]);
2411 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2412 ac->apply_ltp(ac, &che->ch[1]);
2415 if (che->ch[0].tns.present)
2416 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2417 if (che->ch[1].tns.present)
2418 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2419 if (type <= TYPE_CPE)
2420 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2421 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2422 ac->imdct_and_windowing(ac, &che->ch[0]);
2423 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2424 ac->update_ltp(ac, &che->ch[0]);
2425 if (type == TYPE_CPE) {
2426 ac->imdct_and_windowing(ac, &che->ch[1]);
2427 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2428 ac->update_ltp(ac, &che->ch[1]);
2430 if (ac->oc[1].m4ac.sbr > 0) {
2431 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2434 if (type <= TYPE_CCE)
2435 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2441 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2444 AACADTSHeaderInfo hdr_info;
2445 uint8_t layout_map[MAX_ELEM_ID*4][3];
2446 int layout_map_tags;
2448 size = avpriv_aac_parse_header(gb, &hdr_info);
2450 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2451 // This is 2 for "VLB " audio in NSV files.
2452 // See samples/nsv/vlb_audio.
2453 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2454 ac->warned_num_aac_frames = 1;
2456 push_output_configuration(ac);
2457 if (hdr_info.chan_config) {
2458 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2459 if (set_default_channel_config(ac->avctx, layout_map,
2460 &layout_map_tags, hdr_info.chan_config))
2462 if (output_configure(ac, layout_map, layout_map_tags,
2463 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2466 ac->oc[1].m4ac.chan_config = 0;
2468 * dual mono frames in Japanese DTV can have chan_config 0
2469 * WITHOUT specifying PCE.
2470 * thus, set dual mono as default.
2472 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2473 layout_map_tags = 2;
2474 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2475 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2476 layout_map[0][1] = 0;
2477 layout_map[1][1] = 1;
2478 if (output_configure(ac, layout_map, layout_map_tags,
2483 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2484 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2485 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2486 if (ac->oc[0].status != OC_LOCKED ||
2487 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2488 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2489 ac->oc[1].m4ac.sbr = -1;
2490 ac->oc[1].m4ac.ps = -1;
2492 if (!hdr_info.crc_absent)
2498 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2499 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2501 AACContext *ac = avctx->priv_data;
2502 ChannelElement *che = NULL, *che_prev = NULL;
2503 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2505 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2506 int is_dmono, sce_count = 0;
2510 if (show_bits(gb, 12) == 0xfff) {
2511 if (parse_adts_frame_header(ac, gb) < 0) {
2512 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2516 if (ac->oc[1].m4ac.sampling_index > 12) {
2517 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2523 if (frame_configure_elements(avctx) < 0) {
2528 ac->tags_mapped = 0;
2530 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2531 elem_id = get_bits(gb, 4);
2533 if (elem_type < TYPE_DSE) {
2534 if (!(che=get_che(ac, elem_type, elem_id))) {
2535 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2536 elem_type, elem_id);
2543 switch (elem_type) {
2546 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2552 err = decode_cpe(ac, gb, che);
2557 err = decode_cce(ac, gb, che);
2561 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2566 err = skip_data_stream_element(ac, gb);
2570 uint8_t layout_map[MAX_ELEM_ID*4][3];
2572 push_output_configuration(ac);
2573 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2579 av_log(avctx, AV_LOG_ERROR,
2580 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2582 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2584 ac->oc[1].m4ac.chan_config = 0;
2592 elem_id += get_bits(gb, 8) - 1;
2593 if (get_bits_left(gb) < 8 * elem_id) {
2594 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2599 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2600 err = 0; /* FIXME */
2604 err = -1; /* should not happen, but keeps compiler happy */
2609 elem_type_prev = elem_type;
2614 if (get_bits_left(gb) < 3) {
2615 av_log(avctx, AV_LOG_ERROR, overread_err);
2621 spectral_to_sample(ac);
2623 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2624 samples <<= multiplier;
2625 /* for dual-mono audio (SCE + SCE) */
2626 is_dmono = ac->dmono_mode && sce_count == 2 &&
2627 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2630 ac->frame->nb_samples = samples;
2631 *got_frame_ptr = !!samples;
2634 if (ac->dmono_mode == 1)
2635 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2636 else if (ac->dmono_mode == 2)
2637 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2640 if (ac->oc[1].status && audio_found) {
2641 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2642 avctx->frame_size = samples;
2643 ac->oc[1].status = OC_LOCKED;
2648 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2649 if (side && side_size>=4)
2650 AV_WL32(side, 2*AV_RL32(side));
2654 pop_output_configuration(ac);
2658 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2659 int *got_frame_ptr, AVPacket *avpkt)
2661 AACContext *ac = avctx->priv_data;
2662 const uint8_t *buf = avpkt->data;
2663 int buf_size = avpkt->size;
2668 int new_extradata_size;
2669 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2670 AV_PKT_DATA_NEW_EXTRADATA,
2671 &new_extradata_size);
2672 int jp_dualmono_size;
2673 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2674 AV_PKT_DATA_JP_DUALMONO,
2677 if (new_extradata && 0) {
2678 av_free(avctx->extradata);
2679 avctx->extradata = av_mallocz(new_extradata_size +
2680 FF_INPUT_BUFFER_PADDING_SIZE);
2681 if (!avctx->extradata)
2682 return AVERROR(ENOMEM);
2683 avctx->extradata_size = new_extradata_size;
2684 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2685 push_output_configuration(ac);
2686 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2688 avctx->extradata_size*8, 1) < 0) {
2689 pop_output_configuration(ac);
2690 return AVERROR_INVALIDDATA;
2695 if (jp_dualmono && jp_dualmono_size > 0)
2696 ac->dmono_mode = 1 + *jp_dualmono;
2697 if (ac->force_dmono_mode >= 0)
2698 ac->dmono_mode = ac->force_dmono_mode;
2700 if (INT_MAX / 8 <= buf_size)
2701 return AVERROR_INVALIDDATA;
2703 init_get_bits(&gb, buf, buf_size * 8);
2705 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2708 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2709 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2710 if (buf[buf_offset])
2713 return buf_size > buf_offset ? buf_consumed : buf_size;
2716 static av_cold int aac_decode_close(AVCodecContext *avctx)
2718 AACContext *ac = avctx->priv_data;
2721 for (i = 0; i < MAX_ELEM_ID; i++) {
2722 for (type = 0; type < 4; type++) {
2723 if (ac->che[type][i])
2724 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2725 av_freep(&ac->che[type][i]);
2729 ff_mdct_end(&ac->mdct);
2730 ff_mdct_end(&ac->mdct_small);
2731 ff_mdct_end(&ac->mdct_ltp);
2736 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2738 struct LATMContext {
2739 AACContext aac_ctx; ///< containing AACContext
2740 int initialized; ///< initialized after a valid extradata was seen
2743 int audio_mux_version_A; ///< LATM syntax version
2744 int frame_length_type; ///< 0/1 variable/fixed frame length
2745 int frame_length; ///< frame length for fixed frame length
2748 static inline uint32_t latm_get_value(GetBitContext *b)
2750 int length = get_bits(b, 2);
2752 return get_bits_long(b, (length+1)*8);
2755 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2756 GetBitContext *gb, int asclen)
2758 AACContext *ac = &latmctx->aac_ctx;
2759 AVCodecContext *avctx = ac->avctx;
2760 MPEG4AudioConfig m4ac = { 0 };
2761 int config_start_bit = get_bits_count(gb);
2762 int sync_extension = 0;
2763 int bits_consumed, esize;
2767 asclen = FFMIN(asclen, get_bits_left(gb));
2769 asclen = get_bits_left(gb);
2771 if (config_start_bit % 8) {
2772 av_log_missing_feature(latmctx->aac_ctx.avctx,
2773 "Non-byte-aligned audio-specific config", 1);
2774 return AVERROR_PATCHWELCOME;
2777 return AVERROR_INVALIDDATA;
2778 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2779 gb->buffer + (config_start_bit / 8),
2780 asclen, sync_extension);
2782 if (bits_consumed < 0)
2783 return AVERROR_INVALIDDATA;
2785 if (!latmctx->initialized ||
2786 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2787 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2789 if(latmctx->initialized) {
2790 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2792 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
2794 latmctx->initialized = 0;
2796 esize = (bits_consumed+7) / 8;
2798 if (avctx->extradata_size < esize) {
2799 av_free(avctx->extradata);
2800 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2801 if (!avctx->extradata)
2802 return AVERROR(ENOMEM);
2805 avctx->extradata_size = esize;
2806 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2807 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2809 skip_bits_long(gb, bits_consumed);
2811 return bits_consumed;
2814 static int read_stream_mux_config(struct LATMContext *latmctx,
2817 int ret, audio_mux_version = get_bits(gb, 1);
2819 latmctx->audio_mux_version_A = 0;
2820 if (audio_mux_version)
2821 latmctx->audio_mux_version_A = get_bits(gb, 1);
2823 if (!latmctx->audio_mux_version_A) {
2825 if (audio_mux_version)
2826 latm_get_value(gb); // taraFullness
2828 skip_bits(gb, 1); // allStreamSameTimeFraming
2829 skip_bits(gb, 6); // numSubFrames
2831 if (get_bits(gb, 4)) { // numPrograms
2832 av_log_missing_feature(latmctx->aac_ctx.avctx,
2833 "Multiple programs", 1);
2834 return AVERROR_PATCHWELCOME;
2837 // for each program (which there is only one in DVB)
2839 // for each layer (which there is only one in DVB)
2840 if (get_bits(gb, 3)) { // numLayer
2841 av_log_missing_feature(latmctx->aac_ctx.avctx,
2842 "Multiple layers", 1);
2843 return AVERROR_PATCHWELCOME;
2846 // for all but first stream: use_same_config = get_bits(gb, 1);
2847 if (!audio_mux_version) {
2848 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2851 int ascLen = latm_get_value(gb);
2852 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2855 skip_bits_long(gb, ascLen);
2858 latmctx->frame_length_type = get_bits(gb, 3);
2859 switch (latmctx->frame_length_type) {
2861 skip_bits(gb, 8); // latmBufferFullness
2864 latmctx->frame_length = get_bits(gb, 9);
2869 skip_bits(gb, 6); // CELP frame length table index
2873 skip_bits(gb, 1); // HVXC frame length table index
2877 if (get_bits(gb, 1)) { // other data
2878 if (audio_mux_version) {
2879 latm_get_value(gb); // other_data_bits
2883 esc = get_bits(gb, 1);
2889 if (get_bits(gb, 1)) // crc present
2890 skip_bits(gb, 8); // config_crc
2896 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2900 if (ctx->frame_length_type == 0) {
2901 int mux_slot_length = 0;
2903 tmp = get_bits(gb, 8);
2904 mux_slot_length += tmp;
2905 } while (tmp == 255);
2906 return mux_slot_length;
2907 } else if (ctx->frame_length_type == 1) {
2908 return ctx->frame_length;
2909 } else if (ctx->frame_length_type == 3 ||
2910 ctx->frame_length_type == 5 ||
2911 ctx->frame_length_type == 7) {
2912 skip_bits(gb, 2); // mux_slot_length_coded
2917 static int read_audio_mux_element(struct LATMContext *latmctx,
2921 uint8_t use_same_mux = get_bits(gb, 1);
2922 if (!use_same_mux) {
2923 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2925 } else if (!latmctx->aac_ctx.avctx->extradata) {
2926 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2927 "no decoder config found\n");
2928 return AVERROR(EAGAIN);
2930 if (latmctx->audio_mux_version_A == 0) {
2931 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2932 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2933 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2934 return AVERROR_INVALIDDATA;
2935 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2936 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2937 "frame length mismatch %d << %d\n",
2938 mux_slot_length_bytes * 8, get_bits_left(gb));
2939 return AVERROR_INVALIDDATA;
2946 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2947 int *got_frame_ptr, AVPacket *avpkt)
2949 struct LATMContext *latmctx = avctx->priv_data;
2953 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
2956 // check for LOAS sync word
2957 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2958 return AVERROR_INVALIDDATA;
2960 muxlength = get_bits(&gb, 13) + 3;
2961 // not enough data, the parser should have sorted this out
2962 if (muxlength > avpkt->size)
2963 return AVERROR_INVALIDDATA;
2965 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2968 if (!latmctx->initialized) {
2969 if (!avctx->extradata) {
2973 push_output_configuration(&latmctx->aac_ctx);
2974 if ((err = decode_audio_specific_config(
2975 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2976 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2977 pop_output_configuration(&latmctx->aac_ctx);
2980 latmctx->initialized = 1;
2984 if (show_bits(&gb, 12) == 0xfff) {
2985 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2986 "ADTS header detected, probably as result of configuration "
2988 return AVERROR_INVALIDDATA;
2991 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2997 static av_cold int latm_decode_init(AVCodecContext *avctx)
2999 struct LATMContext *latmctx = avctx->priv_data;
3000 int ret = aac_decode_init(avctx);
3002 if (avctx->extradata_size > 0)
3003 latmctx->initialized = !ret;
3008 static void aacdec_init(AACContext *c)
3010 c->imdct_and_windowing = imdct_and_windowing;
3011 c->apply_ltp = apply_ltp;
3012 c->apply_tns = apply_tns;
3013 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3014 c->update_ltp = update_ltp;
3017 ff_aacdec_init_mips(c);
3020 * AVOptions for Japanese DTV specific extensions (ADTS only)
3022 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3023 static const AVOption options[] = {
3024 {"dual_mono_mode", "Select the channel to decode for dual mono",
3025 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3026 AACDEC_FLAGS, "dual_mono_mode"},
3028 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3029 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3030 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3031 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3036 static const AVClass aac_decoder_class = {
3037 .class_name = "AAC decoder",
3038 .item_name = av_default_item_name,
3040 .version = LIBAVUTIL_VERSION_INT,
3043 AVCodec ff_aac_decoder = {
3045 .type = AVMEDIA_TYPE_AUDIO,
3046 .id = AV_CODEC_ID_AAC,
3047 .priv_data_size = sizeof(AACContext),
3048 .init = aac_decode_init,
3049 .close = aac_decode_close,
3050 .decode = aac_decode_frame,
3051 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3052 .sample_fmts = (const enum AVSampleFormat[]) {
3053 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3055 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3056 .channel_layouts = aac_channel_layout,
3058 .priv_class = &aac_decoder_class,
3062 Note: This decoder filter is intended to decode LATM streams transferred
3063 in MPEG transport streams which only contain one program.
3064 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3066 AVCodec ff_aac_latm_decoder = {
3068 .type = AVMEDIA_TYPE_AUDIO,
3069 .id = AV_CODEC_ID_AAC_LATM,
3070 .priv_data_size = sizeof(struct LATMContext),
3071 .init = latm_decode_init,
3072 .close = aac_decode_close,
3073 .decode = latm_decode_frame,
3074 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3075 .sample_fmts = (const enum AVSampleFormat[]) {
3076 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3078 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3079 .channel_layouts = aac_channel_layout,