3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5 * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
11 * AAC decoder fixed-point implementation
13 * MIPS Technologies, Inc., California.
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU Lesser General Public
19 * License as published by the Free Software Foundation; either
20 * version 2.1 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * Lesser General Public License for more details.
27 * You should have received a copy of the GNU Lesser General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 * @author Oded Shimon ( ods15 ods15 dyndns org )
36 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * AAC decoder fixed-point implementation
39 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
40 * @author Nedeljko Babic ( nedeljko.babic imgtec com )
47 * N (code in SoC repo) gain control
49 * Y window shapes - standard
50 * N window shapes - Low Delay
51 * Y filterbank - standard
52 * N (code in SoC repo) filterbank - Scalable Sample Rate
53 * Y Temporal Noise Shaping
54 * Y Long Term Prediction
57 * Y frequency domain prediction
58 * Y Perceptual Noise Substitution
60 * N Scalable Inverse AAC Quantization
61 * N Frequency Selective Switch
63 * Y quantization & coding - AAC
64 * N quantization & coding - TwinVQ
65 * N quantization & coding - BSAC
66 * N AAC Error Resilience tools
67 * N Error Resilience payload syntax
68 * N Error Protection tool
70 * N Silence Compression
73 * N Structured Audio tools
74 * N Structured Audio Sample Bank Format
76 * N Harmonic and Individual Lines plus Noise
77 * N Text-To-Speech Interface
78 * Y Spectral Band Replication
79 * Y (not in this code) Layer-1
80 * Y (not in this code) Layer-2
81 * Y (not in this code) Layer-3
82 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
84 * N Direct Stream Transfer
85 * Y (not in fixed point code) Enhanced AAC Low Delay (ER AAC ELD)
87 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
88 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
92 #include "libavutil/thread.h"
94 static VLC vlc_scalefactors;
95 static VLC vlc_spectral[11];
97 static int output_configure(AACContext *ac,
98 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
99 enum OCStatus oc_type, int get_new_frame);
101 #define overread_err "Input buffer exhausted before END element found\n"
103 static int count_channels(uint8_t (*layout)[3], int tags)
106 for (i = 0; i < tags; i++) {
107 int syn_ele = layout[i][0];
108 int pos = layout[i][2];
109 sum += (1 + (syn_ele == TYPE_CPE)) *
110 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
116 * Check for the channel element in the current channel position configuration.
117 * If it exists, make sure the appropriate element is allocated and map the
118 * channel order to match the internal FFmpeg channel layout.
120 * @param che_pos current channel position configuration
121 * @param type channel element type
122 * @param id channel element id
123 * @param channels count of the number of channels in the configuration
125 * @return Returns error status. 0 - OK, !0 - error
127 static av_cold int che_configure(AACContext *ac,
128 enum ChannelPosition che_pos,
129 int type, int id, int *channels)
131 if (*channels >= MAX_CHANNELS)
132 return AVERROR_INVALIDDATA;
134 if (!ac->che[type][id]) {
135 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
136 return AVERROR(ENOMEM);
137 AAC_RENAME(ff_aac_sbr_ctx_init)(ac, &ac->che[type][id]->sbr, type);
139 if (type != TYPE_CCE) {
140 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
141 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
142 return AVERROR_INVALIDDATA;
144 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
145 if (type == TYPE_CPE ||
146 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
147 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
151 if (ac->che[type][id])
152 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][id]->sbr);
153 av_freep(&ac->che[type][id]);
158 static int frame_configure_elements(AVCodecContext *avctx)
160 AACContext *ac = avctx->priv_data;
161 int type, id, ch, ret;
163 /* set channel pointers to internal buffers by default */
164 for (type = 0; type < 4; type++) {
165 for (id = 0; id < MAX_ELEM_ID; id++) {
166 ChannelElement *che = ac->che[type][id];
168 che->ch[0].ret = che->ch[0].ret_buf;
169 che->ch[1].ret = che->ch[1].ret_buf;
174 /* get output buffer */
175 av_frame_unref(ac->frame);
176 if (!avctx->channels)
179 ac->frame->nb_samples = 2048;
180 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
183 /* map output channel pointers to AVFrame data */
184 for (ch = 0; ch < avctx->channels; ch++) {
185 if (ac->output_element[ch])
186 ac->output_element[ch]->ret = (INTFLOAT *)ac->frame->extended_data[ch];
192 struct elem_to_channel {
193 uint64_t av_position;
196 uint8_t aac_position;
199 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
200 uint8_t (*layout_map)[3], int offset, uint64_t left,
201 uint64_t right, int pos)
203 if (layout_map[offset][0] == TYPE_CPE) {
204 e2c_vec[offset] = (struct elem_to_channel) {
205 .av_position = left | right,
207 .elem_id = layout_map[offset][1],
212 e2c_vec[offset] = (struct elem_to_channel) {
215 .elem_id = layout_map[offset][1],
218 e2c_vec[offset + 1] = (struct elem_to_channel) {
219 .av_position = right,
221 .elem_id = layout_map[offset + 1][1],
228 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
231 int num_pos_channels = 0;
235 for (i = *current; i < tags; i++) {
236 if (layout_map[i][2] != pos)
238 if (layout_map[i][0] == TYPE_CPE) {
240 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
246 num_pos_channels += 2;
254 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
257 return num_pos_channels;
260 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
262 int i, n, total_non_cc_elements;
263 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
264 int num_front_channels, num_side_channels, num_back_channels;
267 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
272 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
273 if (num_front_channels < 0)
276 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
277 if (num_side_channels < 0)
280 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
281 if (num_back_channels < 0)
284 if (num_side_channels == 0 && num_back_channels >= 4) {
285 num_side_channels = 2;
286 num_back_channels -= 2;
290 if (num_front_channels & 1) {
291 e2c_vec[i] = (struct elem_to_channel) {
292 .av_position = AV_CH_FRONT_CENTER,
294 .elem_id = layout_map[i][1],
295 .aac_position = AAC_CHANNEL_FRONT
298 num_front_channels--;
300 if (num_front_channels >= 4) {
301 i += assign_pair(e2c_vec, layout_map, i,
302 AV_CH_FRONT_LEFT_OF_CENTER,
303 AV_CH_FRONT_RIGHT_OF_CENTER,
305 num_front_channels -= 2;
307 if (num_front_channels >= 2) {
308 i += assign_pair(e2c_vec, layout_map, i,
312 num_front_channels -= 2;
314 while (num_front_channels >= 2) {
315 i += assign_pair(e2c_vec, layout_map, i,
319 num_front_channels -= 2;
322 if (num_side_channels >= 2) {
323 i += assign_pair(e2c_vec, layout_map, i,
327 num_side_channels -= 2;
329 while (num_side_channels >= 2) {
330 i += assign_pair(e2c_vec, layout_map, i,
334 num_side_channels -= 2;
337 while (num_back_channels >= 4) {
338 i += assign_pair(e2c_vec, layout_map, i,
342 num_back_channels -= 2;
344 if (num_back_channels >= 2) {
345 i += assign_pair(e2c_vec, layout_map, i,
349 num_back_channels -= 2;
351 if (num_back_channels) {
352 e2c_vec[i] = (struct elem_to_channel) {
353 .av_position = AV_CH_BACK_CENTER,
355 .elem_id = layout_map[i][1],
356 .aac_position = AAC_CHANNEL_BACK
362 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
363 e2c_vec[i] = (struct elem_to_channel) {
364 .av_position = AV_CH_LOW_FREQUENCY,
366 .elem_id = layout_map[i][1],
367 .aac_position = AAC_CHANNEL_LFE
371 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
372 e2c_vec[i] = (struct elem_to_channel) {
373 .av_position = UINT64_MAX,
375 .elem_id = layout_map[i][1],
376 .aac_position = AAC_CHANNEL_LFE
381 // Must choose a stable sort
382 total_non_cc_elements = n = i;
385 for (i = 1; i < n; i++)
386 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
387 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
394 for (i = 0; i < total_non_cc_elements; i++) {
395 layout_map[i][0] = e2c_vec[i].syn_ele;
396 layout_map[i][1] = e2c_vec[i].elem_id;
397 layout_map[i][2] = e2c_vec[i].aac_position;
398 if (e2c_vec[i].av_position != UINT64_MAX) {
399 layout |= e2c_vec[i].av_position;
407 * Save current output configuration if and only if it has been locked.
409 static int push_output_configuration(AACContext *ac) {
412 if (ac->oc[1].status == OC_LOCKED || ac->oc[0].status == OC_NONE) {
413 ac->oc[0] = ac->oc[1];
416 ac->oc[1].status = OC_NONE;
421 * Restore the previous output configuration if and only if the current
422 * configuration is unlocked.
424 static void pop_output_configuration(AACContext *ac) {
425 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
426 ac->oc[1] = ac->oc[0];
427 ac->avctx->channels = ac->oc[1].channels;
428 ac->avctx->channel_layout = ac->oc[1].channel_layout;
429 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
430 ac->oc[1].status, 0);
435 * Configure output channel order based on the current program
436 * configuration element.
438 * @return Returns error status. 0 - OK, !0 - error
440 static int output_configure(AACContext *ac,
441 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
442 enum OCStatus oc_type, int get_new_frame)
444 AVCodecContext *avctx = ac->avctx;
445 int i, channels = 0, ret;
447 uint8_t id_map[TYPE_END][MAX_ELEM_ID] = {{ 0 }};
448 uint8_t type_counts[TYPE_END] = { 0 };
450 if (ac->oc[1].layout_map != layout_map) {
451 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
452 ac->oc[1].layout_map_tags = tags;
454 for (i = 0; i < tags; i++) {
455 int type = layout_map[i][0];
456 int id = layout_map[i][1];
457 id_map[type][id] = type_counts[type]++;
458 if (id_map[type][id] >= MAX_ELEM_ID) {
459 avpriv_request_sample(ac->avctx, "Too large remapped id");
460 return AVERROR_PATCHWELCOME;
463 // Try to sniff a reasonable channel order, otherwise output the
464 // channels in the order the PCE declared them.
465 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
466 layout = sniff_channel_order(layout_map, tags);
467 for (i = 0; i < tags; i++) {
468 int type = layout_map[i][0];
469 int id = layout_map[i][1];
470 int iid = id_map[type][id];
471 int position = layout_map[i][2];
472 // Allocate or free elements depending on if they are in the
473 // current program configuration.
474 ret = che_configure(ac, position, type, iid, &channels);
477 ac->tag_che_map[type][id] = ac->che[type][iid];
479 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
480 if (layout == AV_CH_FRONT_CENTER) {
481 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
487 if (layout) avctx->channel_layout = layout;
488 ac->oc[1].channel_layout = layout;
489 avctx->channels = ac->oc[1].channels = channels;
490 ac->oc[1].status = oc_type;
493 if ((ret = frame_configure_elements(ac->avctx)) < 0)
500 static void flush(AVCodecContext *avctx)
502 AACContext *ac= avctx->priv_data;
505 for (type = 3; type >= 0; type--) {
506 for (i = 0; i < MAX_ELEM_ID; i++) {
507 ChannelElement *che = ac->che[type][i];
509 for (j = 0; j <= 1; j++) {
510 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
518 * Set up channel positions based on a default channel configuration
519 * as specified in table 1.17.
521 * @return Returns error status. 0 - OK, !0 - error
523 static int set_default_channel_config(AVCodecContext *avctx,
524 uint8_t (*layout_map)[3],
528 if (channel_config < 1 || (channel_config > 7 && channel_config < 11) ||
529 channel_config > 12) {
530 av_log(avctx, AV_LOG_ERROR,
531 "invalid default channel configuration (%d)\n",
533 return AVERROR_INVALIDDATA;
535 *tags = tags_per_config[channel_config];
536 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
537 *tags * sizeof(*layout_map));
540 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
541 * However, at least Nero AAC encoder encodes 7.1 streams using the default
542 * channel config 7, mapping the side channels of the original audio stream
543 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
544 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
545 * the incorrect streams as if they were correct (and as the encoder intended).
547 * As actual intended 7.1(wide) streams are very rare, default to assuming a
548 * 7.1 layout was intended.
550 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
551 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
552 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
553 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
554 layout_map[2][2] = AAC_CHANNEL_SIDE;
560 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
562 /* For PCE based channel configurations map the channels solely based
564 if (!ac->oc[1].m4ac.chan_config) {
565 return ac->tag_che_map[type][elem_id];
567 // Allow single CPE stereo files to be signalled with mono configuration.
568 if (!ac->tags_mapped && type == TYPE_CPE &&
569 ac->oc[1].m4ac.chan_config == 1) {
570 uint8_t layout_map[MAX_ELEM_ID*4][3];
572 push_output_configuration(ac);
574 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
576 if (set_default_channel_config(ac->avctx, layout_map,
577 &layout_map_tags, 2) < 0)
579 if (output_configure(ac, layout_map, layout_map_tags,
580 OC_TRIAL_FRAME, 1) < 0)
583 ac->oc[1].m4ac.chan_config = 2;
584 ac->oc[1].m4ac.ps = 0;
587 if (!ac->tags_mapped && type == TYPE_SCE &&
588 ac->oc[1].m4ac.chan_config == 2) {
589 uint8_t layout_map[MAX_ELEM_ID * 4][3];
591 push_output_configuration(ac);
593 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
595 if (set_default_channel_config(ac->avctx, layout_map,
596 &layout_map_tags, 1) < 0)
598 if (output_configure(ac, layout_map, layout_map_tags,
599 OC_TRIAL_FRAME, 1) < 0)
602 ac->oc[1].m4ac.chan_config = 1;
603 if (ac->oc[1].m4ac.sbr)
604 ac->oc[1].m4ac.ps = -1;
606 /* For indexed channel configurations map the channels solely based
608 switch (ac->oc[1].m4ac.chan_config) {
611 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
613 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
616 if (ac->tags_mapped == 2 &&
617 ac->oc[1].m4ac.chan_config == 11 &&
620 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
623 /* Some streams incorrectly code 5.1 audio as
624 * SCE[0] CPE[0] CPE[1] SCE[1]
626 * SCE[0] CPE[0] CPE[1] LFE[0].
627 * If we seem to have encountered such a stream, transfer
628 * the LFE[0] element to the SCE[1]'s mapping */
629 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
630 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
631 av_log(ac->avctx, AV_LOG_WARNING,
632 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
633 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
634 ac->warned_remapping_once++;
637 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
640 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
642 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
645 /* Some streams incorrectly code 4.0 audio as
646 * SCE[0] CPE[0] LFE[0]
648 * SCE[0] CPE[0] SCE[1].
649 * If we seem to have encountered such a stream, transfer
650 * the SCE[1] element to the LFE[0]'s mapping */
651 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
652 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
653 av_log(ac->avctx, AV_LOG_WARNING,
654 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
655 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
656 ac->warned_remapping_once++;
659 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
661 if (ac->tags_mapped == 2 &&
662 ac->oc[1].m4ac.chan_config == 4 &&
665 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
669 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
672 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
673 } else if (ac->oc[1].m4ac.chan_config == 2) {
677 if (!ac->tags_mapped && type == TYPE_SCE) {
679 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
687 * Decode an array of 4 bit element IDs, optionally interleaved with a
688 * stereo/mono switching bit.
690 * @param type speaker type/position for these channels
692 static void decode_channel_map(uint8_t layout_map[][3],
693 enum ChannelPosition type,
694 GetBitContext *gb, int n)
697 enum RawDataBlockType syn_ele;
699 case AAC_CHANNEL_FRONT:
700 case AAC_CHANNEL_BACK:
701 case AAC_CHANNEL_SIDE:
702 syn_ele = get_bits1(gb);
708 case AAC_CHANNEL_LFE:
712 // AAC_CHANNEL_OFF has no channel map
715 layout_map[0][0] = syn_ele;
716 layout_map[0][1] = get_bits(gb, 4);
717 layout_map[0][2] = type;
722 static inline void relative_align_get_bits(GetBitContext *gb,
723 int reference_position) {
724 int n = (reference_position - get_bits_count(gb) & 7);
730 * Decode program configuration element; reference: table 4.2.
732 * @return Returns error status. 0 - OK, !0 - error
734 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
735 uint8_t (*layout_map)[3],
736 GetBitContext *gb, int byte_align_ref)
738 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
743 skip_bits(gb, 2); // object_type
745 sampling_index = get_bits(gb, 4);
746 if (m4ac->sampling_index != sampling_index)
747 av_log(avctx, AV_LOG_WARNING,
748 "Sample rate index in program config element does not "
749 "match the sample rate index configured by the container.\n");
751 num_front = get_bits(gb, 4);
752 num_side = get_bits(gb, 4);
753 num_back = get_bits(gb, 4);
754 num_lfe = get_bits(gb, 2);
755 num_assoc_data = get_bits(gb, 3);
756 num_cc = get_bits(gb, 4);
759 skip_bits(gb, 4); // mono_mixdown_tag
761 skip_bits(gb, 4); // stereo_mixdown_tag
764 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
766 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
767 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
770 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
772 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
774 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
776 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
779 skip_bits_long(gb, 4 * num_assoc_data);
781 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
784 relative_align_get_bits(gb, byte_align_ref);
786 /* comment field, first byte is length */
787 comment_len = get_bits(gb, 8) * 8;
788 if (get_bits_left(gb) < comment_len) {
789 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
790 return AVERROR_INVALIDDATA;
792 skip_bits_long(gb, comment_len);
797 * Decode GA "General Audio" specific configuration; reference: table 4.1.
799 * @param ac pointer to AACContext, may be null
800 * @param avctx pointer to AVCCodecContext, used for logging
802 * @return Returns error status. 0 - OK, !0 - error
804 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
806 int get_bit_alignment,
807 MPEG4AudioConfig *m4ac,
810 int extension_flag, ret, ep_config, res_flags;
811 uint8_t layout_map[MAX_ELEM_ID*4][3];
814 if (get_bits1(gb)) { // frameLengthFlag
815 avpriv_request_sample(avctx, "960/120 MDCT window");
816 return AVERROR_PATCHWELCOME;
818 m4ac->frame_length_short = 0;
820 if (get_bits1(gb)) // dependsOnCoreCoder
821 skip_bits(gb, 14); // coreCoderDelay
822 extension_flag = get_bits1(gb);
824 if (m4ac->object_type == AOT_AAC_SCALABLE ||
825 m4ac->object_type == AOT_ER_AAC_SCALABLE)
826 skip_bits(gb, 3); // layerNr
828 if (channel_config == 0) {
829 skip_bits(gb, 4); // element_instance_tag
830 tags = decode_pce(avctx, m4ac, layout_map, gb, get_bit_alignment);
834 if ((ret = set_default_channel_config(avctx, layout_map,
835 &tags, channel_config)))
839 if (count_channels(layout_map, tags) > 1) {
841 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
844 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
847 if (extension_flag) {
848 switch (m4ac->object_type) {
850 skip_bits(gb, 5); // numOfSubFrame
851 skip_bits(gb, 11); // layer_length
855 case AOT_ER_AAC_SCALABLE:
857 res_flags = get_bits(gb, 3);
859 avpriv_report_missing_feature(avctx,
860 "AAC data resilience (flags %x)",
862 return AVERROR_PATCHWELCOME;
866 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
868 switch (m4ac->object_type) {
871 case AOT_ER_AAC_SCALABLE:
873 ep_config = get_bits(gb, 2);
875 avpriv_report_missing_feature(avctx,
876 "epConfig %d", ep_config);
877 return AVERROR_PATCHWELCOME;
883 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
885 MPEG4AudioConfig *m4ac,
888 int ret, ep_config, res_flags;
889 uint8_t layout_map[MAX_ELEM_ID*4][3];
891 const int ELDEXT_TERM = 0;
896 if (get_bits1(gb)) { // frameLengthFlag
897 avpriv_request_sample(avctx, "960/120 MDCT window");
898 return AVERROR_PATCHWELCOME;
901 m4ac->frame_length_short = get_bits1(gb);
903 res_flags = get_bits(gb, 3);
905 avpriv_report_missing_feature(avctx,
906 "AAC data resilience (flags %x)",
908 return AVERROR_PATCHWELCOME;
911 if (get_bits1(gb)) { // ldSbrPresentFlag
912 avpriv_report_missing_feature(avctx,
914 return AVERROR_PATCHWELCOME;
917 while (get_bits(gb, 4) != ELDEXT_TERM) {
918 int len = get_bits(gb, 4);
920 len += get_bits(gb, 8);
922 len += get_bits(gb, 16);
923 if (get_bits_left(gb) < len * 8 + 4) {
924 av_log(avctx, AV_LOG_ERROR, overread_err);
925 return AVERROR_INVALIDDATA;
927 skip_bits_long(gb, 8 * len);
930 if ((ret = set_default_channel_config(avctx, layout_map,
931 &tags, channel_config)))
934 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
937 ep_config = get_bits(gb, 2);
939 avpriv_report_missing_feature(avctx,
940 "epConfig %d", ep_config);
941 return AVERROR_PATCHWELCOME;
947 * Decode audio specific configuration; reference: table 1.13.
949 * @param ac pointer to AACContext, may be null
950 * @param avctx pointer to AVCCodecContext, used for logging
951 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
952 * @param gb buffer holding an audio specific config
953 * @param get_bit_alignment relative alignment for byte align operations
954 * @param sync_extension look for an appended sync extension
956 * @return Returns error status or number of consumed bits. <0 - error
958 static int decode_audio_specific_config_gb(AACContext *ac,
959 AVCodecContext *avctx,
960 MPEG4AudioConfig *m4ac,
962 int get_bit_alignment,
966 GetBitContext gbc = *gb;
968 if ((i = ff_mpeg4audio_get_config_gb(m4ac, &gbc, sync_extension)) < 0)
969 return AVERROR_INVALIDDATA;
971 if (m4ac->sampling_index > 12) {
972 av_log(avctx, AV_LOG_ERROR,
973 "invalid sampling rate index %d\n",
974 m4ac->sampling_index);
975 return AVERROR_INVALIDDATA;
977 if (m4ac->object_type == AOT_ER_AAC_LD &&
978 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
979 av_log(avctx, AV_LOG_ERROR,
980 "invalid low delay sampling rate index %d\n",
981 m4ac->sampling_index);
982 return AVERROR_INVALIDDATA;
985 skip_bits_long(gb, i);
987 switch (m4ac->object_type) {
993 if ((ret = decode_ga_specific_config(ac, avctx, gb, get_bit_alignment,
994 m4ac, m4ac->chan_config)) < 0)
998 if ((ret = decode_eld_specific_config(ac, avctx, gb,
999 m4ac, m4ac->chan_config)) < 0)
1003 avpriv_report_missing_feature(avctx,
1004 "Audio object type %s%d",
1005 m4ac->sbr == 1 ? "SBR+" : "",
1007 return AVERROR(ENOSYS);
1011 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
1012 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1013 m4ac->sample_rate, m4ac->sbr,
1016 return get_bits_count(gb);
1019 static int decode_audio_specific_config(AACContext *ac,
1020 AVCodecContext *avctx,
1021 MPEG4AudioConfig *m4ac,
1022 const uint8_t *data, int64_t bit_size,
1028 if (bit_size < 0 || bit_size > INT_MAX) {
1029 av_log(avctx, AV_LOG_ERROR, "Audio specific config size is invalid\n");
1030 return AVERROR_INVALIDDATA;
1033 ff_dlog(avctx, "audio specific config size %d\n", (int)bit_size >> 3);
1034 for (i = 0; i < bit_size >> 3; i++)
1035 ff_dlog(avctx, "%02x ", data[i]);
1036 ff_dlog(avctx, "\n");
1038 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
1041 return decode_audio_specific_config_gb(ac, avctx, m4ac, &gb, 0,
1046 * linear congruential pseudorandom number generator
1048 * @param previous_val pointer to the current state of the generator
1050 * @return Returns a 32-bit pseudorandom integer
1052 static av_always_inline int lcg_random(unsigned previous_val)
1054 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1058 static void reset_all_predictors(PredictorState *ps)
1061 for (i = 0; i < MAX_PREDICTORS; i++)
1062 reset_predict_state(&ps[i]);
1065 static int sample_rate_idx (int rate)
1067 if (92017 <= rate) return 0;
1068 else if (75132 <= rate) return 1;
1069 else if (55426 <= rate) return 2;
1070 else if (46009 <= rate) return 3;
1071 else if (37566 <= rate) return 4;
1072 else if (27713 <= rate) return 5;
1073 else if (23004 <= rate) return 6;
1074 else if (18783 <= rate) return 7;
1075 else if (13856 <= rate) return 8;
1076 else if (11502 <= rate) return 9;
1077 else if (9391 <= rate) return 10;
1081 static void reset_predictor_group(PredictorState *ps, int group_num)
1084 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1085 reset_predict_state(&ps[i]);
1088 #define AAC_INIT_VLC_STATIC(num, size) \
1089 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1090 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1091 sizeof(ff_aac_spectral_bits[num][0]), \
1092 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1093 sizeof(ff_aac_spectral_codes[num][0]), \
1096 static void aacdec_init(AACContext *ac);
1098 static av_cold void aac_static_table_init(void)
1100 AAC_INIT_VLC_STATIC( 0, 304);
1101 AAC_INIT_VLC_STATIC( 1, 270);
1102 AAC_INIT_VLC_STATIC( 2, 550);
1103 AAC_INIT_VLC_STATIC( 3, 300);
1104 AAC_INIT_VLC_STATIC( 4, 328);
1105 AAC_INIT_VLC_STATIC( 5, 294);
1106 AAC_INIT_VLC_STATIC( 6, 306);
1107 AAC_INIT_VLC_STATIC( 7, 268);
1108 AAC_INIT_VLC_STATIC( 8, 510);
1109 AAC_INIT_VLC_STATIC( 9, 366);
1110 AAC_INIT_VLC_STATIC(10, 462);
1112 AAC_RENAME(ff_aac_sbr_init)();
1116 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1117 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1118 ff_aac_scalefactor_bits,
1119 sizeof(ff_aac_scalefactor_bits[0]),
1120 sizeof(ff_aac_scalefactor_bits[0]),
1121 ff_aac_scalefactor_code,
1122 sizeof(ff_aac_scalefactor_code[0]),
1123 sizeof(ff_aac_scalefactor_code[0]),
1126 // window initialization
1127 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_long_1024), 4.0, 1024);
1128 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_short_128), 6.0, 128);
1129 AAC_RENAME(ff_init_ff_sine_windows)(10);
1130 AAC_RENAME(ff_init_ff_sine_windows)( 9);
1131 AAC_RENAME(ff_init_ff_sine_windows)( 7);
1133 AAC_RENAME(ff_cbrt_tableinit)();
1136 static AVOnce aac_table_init = AV_ONCE_INIT;
1138 static av_cold int aac_decode_init(AVCodecContext *avctx)
1140 AACContext *ac = avctx->priv_data;
1143 ret = ff_thread_once(&aac_table_init, &aac_static_table_init);
1145 return AVERROR_UNKNOWN;
1148 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1152 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1154 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1155 #endif /* USE_FIXED */
1157 if (avctx->extradata_size > 0) {
1158 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1160 avctx->extradata_size * 8LL,
1165 uint8_t layout_map[MAX_ELEM_ID*4][3];
1166 int layout_map_tags;
1168 sr = sample_rate_idx(avctx->sample_rate);
1169 ac->oc[1].m4ac.sampling_index = sr;
1170 ac->oc[1].m4ac.channels = avctx->channels;
1171 ac->oc[1].m4ac.sbr = -1;
1172 ac->oc[1].m4ac.ps = -1;
1174 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1175 if (ff_mpeg4audio_channels[i] == avctx->channels)
1177 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1180 ac->oc[1].m4ac.chan_config = i;
1182 if (ac->oc[1].m4ac.chan_config) {
1183 int ret = set_default_channel_config(avctx, layout_map,
1184 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1186 output_configure(ac, layout_map, layout_map_tags,
1188 else if (avctx->err_recognition & AV_EF_EXPLODE)
1189 return AVERROR_INVALIDDATA;
1193 if (avctx->channels > MAX_CHANNELS) {
1194 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1195 return AVERROR_INVALIDDATA;
1199 ac->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1201 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1202 #endif /* USE_FIXED */
1204 return AVERROR(ENOMEM);
1207 ac->random_state = 0x1f2e3d4c;
1209 AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
1210 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
1211 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
1212 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
1214 ret = ff_mdct15_init(&ac->mdct480, 1, 5, -1.0f);
1223 * Skip data_stream_element; reference: table 4.10.
1225 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1227 int byte_align = get_bits1(gb);
1228 int count = get_bits(gb, 8);
1230 count += get_bits(gb, 8);
1234 if (get_bits_left(gb) < 8 * count) {
1235 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1236 return AVERROR_INVALIDDATA;
1238 skip_bits_long(gb, 8 * count);
1242 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1246 if (get_bits1(gb)) {
1247 ics->predictor_reset_group = get_bits(gb, 5);
1248 if (ics->predictor_reset_group == 0 ||
1249 ics->predictor_reset_group > 30) {
1250 av_log(ac->avctx, AV_LOG_ERROR,
1251 "Invalid Predictor Reset Group.\n");
1252 return AVERROR_INVALIDDATA;
1255 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1256 ics->prediction_used[sfb] = get_bits1(gb);
1262 * Decode Long Term Prediction data; reference: table 4.xx.
1264 static void decode_ltp(LongTermPrediction *ltp,
1265 GetBitContext *gb, uint8_t max_sfb)
1269 ltp->lag = get_bits(gb, 11);
1270 ltp->coef = ltp_coef[get_bits(gb, 3)];
1271 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1272 ltp->used[sfb] = get_bits1(gb);
1276 * Decode Individual Channel Stream info; reference: table 4.6.
1278 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1281 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1282 const int aot = m4ac->object_type;
1283 const int sampling_index = m4ac->sampling_index;
1284 if (aot != AOT_ER_AAC_ELD) {
1285 if (get_bits1(gb)) {
1286 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1287 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1288 return AVERROR_INVALIDDATA;
1290 ics->window_sequence[1] = ics->window_sequence[0];
1291 ics->window_sequence[0] = get_bits(gb, 2);
1292 if (aot == AOT_ER_AAC_LD &&
1293 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1294 av_log(ac->avctx, AV_LOG_ERROR,
1295 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1296 "window sequence %d found.\n", ics->window_sequence[0]);
1297 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1298 return AVERROR_INVALIDDATA;
1300 ics->use_kb_window[1] = ics->use_kb_window[0];
1301 ics->use_kb_window[0] = get_bits1(gb);
1303 ics->num_window_groups = 1;
1304 ics->group_len[0] = 1;
1305 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1307 ics->max_sfb = get_bits(gb, 4);
1308 for (i = 0; i < 7; i++) {
1309 if (get_bits1(gb)) {
1310 ics->group_len[ics->num_window_groups - 1]++;
1312 ics->num_window_groups++;
1313 ics->group_len[ics->num_window_groups - 1] = 1;
1316 ics->num_windows = 8;
1317 ics->swb_offset = ff_swb_offset_128[sampling_index];
1318 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1319 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1320 ics->predictor_present = 0;
1322 ics->max_sfb = get_bits(gb, 6);
1323 ics->num_windows = 1;
1324 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1325 if (m4ac->frame_length_short) {
1326 ics->swb_offset = ff_swb_offset_480[sampling_index];
1327 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1328 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1330 ics->swb_offset = ff_swb_offset_512[sampling_index];
1331 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1332 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1334 if (!ics->num_swb || !ics->swb_offset)
1337 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1338 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1339 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1341 if (aot != AOT_ER_AAC_ELD) {
1342 ics->predictor_present = get_bits1(gb);
1343 ics->predictor_reset_group = 0;
1345 if (ics->predictor_present) {
1346 if (aot == AOT_AAC_MAIN) {
1347 if (decode_prediction(ac, ics, gb)) {
1350 } else if (aot == AOT_AAC_LC ||
1351 aot == AOT_ER_AAC_LC) {
1352 av_log(ac->avctx, AV_LOG_ERROR,
1353 "Prediction is not allowed in AAC-LC.\n");
1356 if (aot == AOT_ER_AAC_LD) {
1357 av_log(ac->avctx, AV_LOG_ERROR,
1358 "LTP in ER AAC LD not yet implemented.\n");
1359 return AVERROR_PATCHWELCOME;
1361 if ((ics->ltp.present = get_bits(gb, 1)))
1362 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1367 if (ics->max_sfb > ics->num_swb) {
1368 av_log(ac->avctx, AV_LOG_ERROR,
1369 "Number of scalefactor bands in group (%d) "
1370 "exceeds limit (%d).\n",
1371 ics->max_sfb, ics->num_swb);
1378 return AVERROR_INVALIDDATA;
1382 * Decode band types (section_data payload); reference: table 4.46.
1384 * @param band_type array of the used band type
1385 * @param band_type_run_end array of the last scalefactor band of a band type run
1387 * @return Returns error status. 0 - OK, !0 - error
1389 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1390 int band_type_run_end[120], GetBitContext *gb,
1391 IndividualChannelStream *ics)
1394 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1395 for (g = 0; g < ics->num_window_groups; g++) {
1397 while (k < ics->max_sfb) {
1398 uint8_t sect_end = k;
1400 int sect_band_type = get_bits(gb, 4);
1401 if (sect_band_type == 12) {
1402 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1403 return AVERROR_INVALIDDATA;
1406 sect_len_incr = get_bits(gb, bits);
1407 sect_end += sect_len_incr;
1408 if (get_bits_left(gb) < 0) {
1409 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1410 return AVERROR_INVALIDDATA;
1412 if (sect_end > ics->max_sfb) {
1413 av_log(ac->avctx, AV_LOG_ERROR,
1414 "Number of bands (%d) exceeds limit (%d).\n",
1415 sect_end, ics->max_sfb);
1416 return AVERROR_INVALIDDATA;
1418 } while (sect_len_incr == (1 << bits) - 1);
1419 for (; k < sect_end; k++) {
1420 band_type [idx] = sect_band_type;
1421 band_type_run_end[idx++] = sect_end;
1429 * Decode scalefactors; reference: table 4.47.
1431 * @param global_gain first scalefactor value as scalefactors are differentially coded
1432 * @param band_type array of the used band type
1433 * @param band_type_run_end array of the last scalefactor band of a band type run
1434 * @param sf array of scalefactors or intensity stereo positions
1436 * @return Returns error status. 0 - OK, !0 - error
1438 static int decode_scalefactors(AACContext *ac, INTFLOAT sf[120], GetBitContext *gb,
1439 unsigned int global_gain,
1440 IndividualChannelStream *ics,
1441 enum BandType band_type[120],
1442 int band_type_run_end[120])
1445 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1448 for (g = 0; g < ics->num_window_groups; g++) {
1449 for (i = 0; i < ics->max_sfb;) {
1450 int run_end = band_type_run_end[idx];
1451 if (band_type[idx] == ZERO_BT) {
1452 for (; i < run_end; i++, idx++)
1454 } else if ((band_type[idx] == INTENSITY_BT) ||
1455 (band_type[idx] == INTENSITY_BT2)) {
1456 for (; i < run_end; i++, idx++) {
1457 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1458 clipped_offset = av_clip(offset[2], -155, 100);
1459 if (offset[2] != clipped_offset) {
1460 avpriv_request_sample(ac->avctx,
1461 "If you heard an audible artifact, there may be a bug in the decoder. "
1462 "Clipped intensity stereo position (%d -> %d)",
1463 offset[2], clipped_offset);
1466 sf[idx] = 100 - clipped_offset;
1468 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1469 #endif /* USE_FIXED */
1471 } else if (band_type[idx] == NOISE_BT) {
1472 for (; i < run_end; i++, idx++) {
1473 if (noise_flag-- > 0)
1474 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1476 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1477 clipped_offset = av_clip(offset[1], -100, 155);
1478 if (offset[1] != clipped_offset) {
1479 avpriv_request_sample(ac->avctx,
1480 "If you heard an audible artifact, there may be a bug in the decoder. "
1481 "Clipped noise gain (%d -> %d)",
1482 offset[1], clipped_offset);
1485 sf[idx] = -(100 + clipped_offset);
1487 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1488 #endif /* USE_FIXED */
1491 for (; i < run_end; i++, idx++) {
1492 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1493 if (offset[0] > 255U) {
1494 av_log(ac->avctx, AV_LOG_ERROR,
1495 "Scalefactor (%d) out of range.\n", offset[0]);
1496 return AVERROR_INVALIDDATA;
1499 sf[idx] = -offset[0];
1501 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1502 #endif /* USE_FIXED */
1511 * Decode pulse data; reference: table 4.7.
1513 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1514 const uint16_t *swb_offset, int num_swb)
1517 pulse->num_pulse = get_bits(gb, 2) + 1;
1518 pulse_swb = get_bits(gb, 6);
1519 if (pulse_swb >= num_swb)
1521 pulse->pos[0] = swb_offset[pulse_swb];
1522 pulse->pos[0] += get_bits(gb, 5);
1523 if (pulse->pos[0] >= swb_offset[num_swb])
1525 pulse->amp[0] = get_bits(gb, 4);
1526 for (i = 1; i < pulse->num_pulse; i++) {
1527 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1528 if (pulse->pos[i] >= swb_offset[num_swb])
1530 pulse->amp[i] = get_bits(gb, 4);
1536 * Decode Temporal Noise Shaping data; reference: table 4.48.
1538 * @return Returns error status. 0 - OK, !0 - error
1540 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1541 GetBitContext *gb, const IndividualChannelStream *ics)
1543 int w, filt, i, coef_len, coef_res, coef_compress;
1544 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1545 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1546 for (w = 0; w < ics->num_windows; w++) {
1547 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1548 coef_res = get_bits1(gb);
1550 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1552 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1554 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1555 av_log(ac->avctx, AV_LOG_ERROR,
1556 "TNS filter order %d is greater than maximum %d.\n",
1557 tns->order[w][filt], tns_max_order);
1558 tns->order[w][filt] = 0;
1559 return AVERROR_INVALIDDATA;
1561 if (tns->order[w][filt]) {
1562 tns->direction[w][filt] = get_bits1(gb);
1563 coef_compress = get_bits1(gb);
1564 coef_len = coef_res + 3 - coef_compress;
1565 tmp2_idx = 2 * coef_compress + coef_res;
1567 for (i = 0; i < tns->order[w][filt]; i++)
1568 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1577 * Decode Mid/Side data; reference: table 4.54.
1579 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1580 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1581 * [3] reserved for scalable AAC
1583 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1587 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1588 if (ms_present == 1) {
1589 for (idx = 0; idx < max_idx; idx++)
1590 cpe->ms_mask[idx] = get_bits1(gb);
1591 } else if (ms_present == 2) {
1592 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1597 * Decode spectral data; reference: table 4.50.
1598 * Dequantize and scale spectral data; reference: 4.6.3.3.
1600 * @param coef array of dequantized, scaled spectral data
1601 * @param sf array of scalefactors or intensity stereo positions
1602 * @param pulse_present set if pulses are present
1603 * @param pulse pointer to pulse data struct
1604 * @param band_type array of the used band type
1606 * @return Returns error status. 0 - OK, !0 - error
1608 static int decode_spectrum_and_dequant(AACContext *ac, INTFLOAT coef[1024],
1609 GetBitContext *gb, const INTFLOAT sf[120],
1610 int pulse_present, const Pulse *pulse,
1611 const IndividualChannelStream *ics,
1612 enum BandType band_type[120])
1614 int i, k, g, idx = 0;
1615 const int c = 1024 / ics->num_windows;
1616 const uint16_t *offsets = ics->swb_offset;
1617 INTFLOAT *coef_base = coef;
1619 for (g = 0; g < ics->num_windows; g++)
1620 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1621 sizeof(INTFLOAT) * (c - offsets[ics->max_sfb]));
1623 for (g = 0; g < ics->num_window_groups; g++) {
1624 unsigned g_len = ics->group_len[g];
1626 for (i = 0; i < ics->max_sfb; i++, idx++) {
1627 const unsigned cbt_m1 = band_type[idx] - 1;
1628 INTFLOAT *cfo = coef + offsets[i];
1629 int off_len = offsets[i + 1] - offsets[i];
1632 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1633 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1634 memset(cfo, 0, off_len * sizeof(*cfo));
1636 } else if (cbt_m1 == NOISE_BT - 1) {
1637 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1640 #endif /* !USE_FIXED */
1641 INTFLOAT band_energy;
1643 for (k = 0; k < off_len; k++) {
1644 ac->random_state = lcg_random(ac->random_state);
1646 cfo[k] = ac->random_state >> 3;
1648 cfo[k] = ac->random_state;
1649 #endif /* USE_FIXED */
1653 band_energy = ac->fdsp->scalarproduct_fixed(cfo, cfo, off_len);
1654 band_energy = fixed_sqrt(band_energy, 31);
1655 noise_scale(cfo, sf[idx], band_energy, off_len);
1657 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1658 scale = sf[idx] / sqrtf(band_energy);
1659 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1660 #endif /* USE_FIXED */
1664 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1665 #endif /* !USE_FIXED */
1666 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1667 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1668 OPEN_READER(re, gb);
1670 switch (cbt_m1 >> 1) {
1672 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1680 UPDATE_CACHE(re, gb);
1681 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1682 cb_idx = cb_vector_idx[code];
1684 cf = DEC_SQUAD(cf, cb_idx);
1686 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1687 #endif /* USE_FIXED */
1693 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1703 UPDATE_CACHE(re, gb);
1704 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1705 cb_idx = cb_vector_idx[code];
1706 nnz = cb_idx >> 8 & 15;
1707 bits = nnz ? GET_CACHE(re, gb) : 0;
1708 LAST_SKIP_BITS(re, gb, nnz);
1710 cf = DEC_UQUAD(cf, cb_idx, bits);
1712 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1713 #endif /* USE_FIXED */
1719 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1727 UPDATE_CACHE(re, gb);
1728 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1729 cb_idx = cb_vector_idx[code];
1731 cf = DEC_SPAIR(cf, cb_idx);
1733 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1734 #endif /* USE_FIXED */
1741 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1751 UPDATE_CACHE(re, gb);
1752 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1753 cb_idx = cb_vector_idx[code];
1754 nnz = cb_idx >> 8 & 15;
1755 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1756 LAST_SKIP_BITS(re, gb, nnz);
1758 cf = DEC_UPAIR(cf, cb_idx, sign);
1760 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1761 #endif /* USE_FIXED */
1767 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1773 uint32_t *icf = (uint32_t *) cf;
1774 #endif /* USE_FIXED */
1784 UPDATE_CACHE(re, gb);
1785 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1793 cb_idx = cb_vector_idx[code];
1796 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1797 LAST_SKIP_BITS(re, gb, nnz);
1799 for (j = 0; j < 2; j++) {
1803 /* The total length of escape_sequence must be < 22 bits according
1804 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1805 UPDATE_CACHE(re, gb);
1806 b = GET_CACHE(re, gb);
1807 b = 31 - av_log2(~b);
1810 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1811 return AVERROR_INVALIDDATA;
1814 SKIP_BITS(re, gb, b + 1);
1816 n = (1 << b) + SHOW_UBITS(re, gb, b);
1817 LAST_SKIP_BITS(re, gb, b);
1824 *icf++ = ff_cbrt_tab[n] | (bits & 1U<<31);
1825 #endif /* USE_FIXED */
1834 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1835 *icf++ = (bits & 1U<<31) | v;
1836 #endif /* USE_FIXED */
1843 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1844 #endif /* !USE_FIXED */
1848 CLOSE_READER(re, gb);
1854 if (pulse_present) {
1856 for (i = 0; i < pulse->num_pulse; i++) {
1857 INTFLOAT co = coef_base[ pulse->pos[i] ];
1858 while (offsets[idx + 1] <= pulse->pos[i])
1860 if (band_type[idx] != NOISE_BT && sf[idx]) {
1861 INTFLOAT ico = -pulse->amp[i];
1864 ico = co + (co > 0 ? -ico : ico);
1866 coef_base[ pulse->pos[i] ] = ico;
1870 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1872 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1873 #endif /* USE_FIXED */
1880 for (g = 0; g < ics->num_window_groups; g++) {
1881 unsigned g_len = ics->group_len[g];
1883 for (i = 0; i < ics->max_sfb; i++, idx++) {
1884 const unsigned cbt_m1 = band_type[idx] - 1;
1885 int *cfo = coef + offsets[i];
1886 int off_len = offsets[i + 1] - offsets[i];
1889 if (cbt_m1 < NOISE_BT - 1) {
1890 for (group = 0; group < (int)g_len; group++, cfo+=128) {
1891 ac->vector_pow43(cfo, off_len);
1892 ac->subband_scale(cfo, cfo, sf[idx], 34, off_len);
1898 #endif /* USE_FIXED */
1903 * Apply AAC-Main style frequency domain prediction.
1905 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1909 if (!sce->ics.predictor_initialized) {
1910 reset_all_predictors(sce->predictor_state);
1911 sce->ics.predictor_initialized = 1;
1914 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1916 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1918 for (k = sce->ics.swb_offset[sfb];
1919 k < sce->ics.swb_offset[sfb + 1];
1921 predict(&sce->predictor_state[k], &sce->coeffs[k],
1922 sce->ics.predictor_present &&
1923 sce->ics.prediction_used[sfb]);
1926 if (sce->ics.predictor_reset_group)
1927 reset_predictor_group(sce->predictor_state,
1928 sce->ics.predictor_reset_group);
1930 reset_all_predictors(sce->predictor_state);
1934 * Decode an individual_channel_stream payload; reference: table 4.44.
1936 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1937 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1939 * @return Returns error status. 0 - OK, !0 - error
1941 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1942 GetBitContext *gb, int common_window, int scale_flag)
1945 TemporalNoiseShaping *tns = &sce->tns;
1946 IndividualChannelStream *ics = &sce->ics;
1947 INTFLOAT *out = sce->coeffs;
1948 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1951 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1952 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1953 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1954 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1955 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1957 /* This assignment is to silence a GCC warning about the variable being used
1958 * uninitialized when in fact it always is.
1960 pulse.num_pulse = 0;
1962 global_gain = get_bits(gb, 8);
1964 if (!common_window && !scale_flag) {
1965 if (decode_ics_info(ac, ics, gb) < 0)
1966 return AVERROR_INVALIDDATA;
1969 if ((ret = decode_band_types(ac, sce->band_type,
1970 sce->band_type_run_end, gb, ics)) < 0)
1972 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1973 sce->band_type, sce->band_type_run_end)) < 0)
1978 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1979 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1980 av_log(ac->avctx, AV_LOG_ERROR,
1981 "Pulse tool not allowed in eight short sequence.\n");
1982 return AVERROR_INVALIDDATA;
1984 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1985 av_log(ac->avctx, AV_LOG_ERROR,
1986 "Pulse data corrupt or invalid.\n");
1987 return AVERROR_INVALIDDATA;
1990 tns->present = get_bits1(gb);
1991 if (tns->present && !er_syntax)
1992 if (decode_tns(ac, tns, gb, ics) < 0)
1993 return AVERROR_INVALIDDATA;
1994 if (!eld_syntax && get_bits1(gb)) {
1995 avpriv_request_sample(ac->avctx, "SSR");
1996 return AVERROR_PATCHWELCOME;
1998 // I see no textual basis in the spec for this occurring after SSR gain
1999 // control, but this is what both reference and real implmentations do
2000 if (tns->present && er_syntax)
2001 if (decode_tns(ac, tns, gb, ics) < 0)
2002 return AVERROR_INVALIDDATA;
2005 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
2006 &pulse, ics, sce->band_type) < 0)
2007 return AVERROR_INVALIDDATA;
2009 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
2010 apply_prediction(ac, sce);
2016 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
2018 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
2020 const IndividualChannelStream *ics = &cpe->ch[0].ics;
2021 INTFLOAT *ch0 = cpe->ch[0].coeffs;
2022 INTFLOAT *ch1 = cpe->ch[1].coeffs;
2023 int g, i, group, idx = 0;
2024 const uint16_t *offsets = ics->swb_offset;
2025 for (g = 0; g < ics->num_window_groups; g++) {
2026 for (i = 0; i < ics->max_sfb; i++, idx++) {
2027 if (cpe->ms_mask[idx] &&
2028 cpe->ch[0].band_type[idx] < NOISE_BT &&
2029 cpe->ch[1].band_type[idx] < NOISE_BT) {
2031 for (group = 0; group < ics->group_len[g]; group++) {
2032 ac->fdsp->butterflies_fixed(ch0 + group * 128 + offsets[i],
2033 ch1 + group * 128 + offsets[i],
2034 offsets[i+1] - offsets[i]);
2036 for (group = 0; group < ics->group_len[g]; group++) {
2037 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2038 ch1 + group * 128 + offsets[i],
2039 offsets[i+1] - offsets[i]);
2040 #endif /* USE_FIXED */
2044 ch0 += ics->group_len[g] * 128;
2045 ch1 += ics->group_len[g] * 128;
2050 * intensity stereo decoding; reference: 4.6.8.2.3
2052 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2053 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2054 * [3] reserved for scalable AAC
2056 static void apply_intensity_stereo(AACContext *ac,
2057 ChannelElement *cpe, int ms_present)
2059 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2060 SingleChannelElement *sce1 = &cpe->ch[1];
2061 INTFLOAT *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2062 const uint16_t *offsets = ics->swb_offset;
2063 int g, group, i, idx = 0;
2066 for (g = 0; g < ics->num_window_groups; g++) {
2067 for (i = 0; i < ics->max_sfb;) {
2068 if (sce1->band_type[idx] == INTENSITY_BT ||
2069 sce1->band_type[idx] == INTENSITY_BT2) {
2070 const int bt_run_end = sce1->band_type_run_end[idx];
2071 for (; i < bt_run_end; i++, idx++) {
2072 c = -1 + 2 * (sce1->band_type[idx] - 14);
2074 c *= 1 - 2 * cpe->ms_mask[idx];
2075 scale = c * sce1->sf[idx];
2076 for (group = 0; group < ics->group_len[g]; group++)
2078 ac->subband_scale(coef1 + group * 128 + offsets[i],
2079 coef0 + group * 128 + offsets[i],
2082 offsets[i + 1] - offsets[i]);
2084 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2085 coef0 + group * 128 + offsets[i],
2087 offsets[i + 1] - offsets[i]);
2088 #endif /* USE_FIXED */
2091 int bt_run_end = sce1->band_type_run_end[idx];
2092 idx += bt_run_end - i;
2096 coef0 += ics->group_len[g] * 128;
2097 coef1 += ics->group_len[g] * 128;
2102 * Decode a channel_pair_element; reference: table 4.4.
2104 * @return Returns error status. 0 - OK, !0 - error
2106 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2108 int i, ret, common_window, ms_present = 0;
2109 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2111 common_window = eld_syntax || get_bits1(gb);
2112 if (common_window) {
2113 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2114 return AVERROR_INVALIDDATA;
2115 i = cpe->ch[1].ics.use_kb_window[0];
2116 cpe->ch[1].ics = cpe->ch[0].ics;
2117 cpe->ch[1].ics.use_kb_window[1] = i;
2118 if (cpe->ch[1].ics.predictor_present &&
2119 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2120 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2121 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2122 ms_present = get_bits(gb, 2);
2123 if (ms_present == 3) {
2124 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2125 return AVERROR_INVALIDDATA;
2126 } else if (ms_present)
2127 decode_mid_side_stereo(cpe, gb, ms_present);
2129 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2131 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2134 if (common_window) {
2136 apply_mid_side_stereo(ac, cpe);
2137 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2138 apply_prediction(ac, &cpe->ch[0]);
2139 apply_prediction(ac, &cpe->ch[1]);
2143 apply_intensity_stereo(ac, cpe, ms_present);
2147 static const float cce_scale[] = {
2148 1.09050773266525765921, //2^(1/8)
2149 1.18920711500272106672, //2^(1/4)
2155 * Decode coupling_channel_element; reference: table 4.8.
2157 * @return Returns error status. 0 - OK, !0 - error
2159 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2165 SingleChannelElement *sce = &che->ch[0];
2166 ChannelCoupling *coup = &che->coup;
2168 coup->coupling_point = 2 * get_bits1(gb);
2169 coup->num_coupled = get_bits(gb, 3);
2170 for (c = 0; c <= coup->num_coupled; c++) {
2172 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2173 coup->id_select[c] = get_bits(gb, 4);
2174 if (coup->type[c] == TYPE_CPE) {
2175 coup->ch_select[c] = get_bits(gb, 2);
2176 if (coup->ch_select[c] == 3)
2179 coup->ch_select[c] = 2;
2181 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2183 sign = get_bits(gb, 1);
2185 scale = get_bits(gb, 2);
2187 scale = cce_scale[get_bits(gb, 2)];
2190 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2193 for (c = 0; c < num_gain; c++) {
2197 INTFLOAT gain_cache = FIXR10(1.);
2199 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2200 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2201 gain_cache = GET_GAIN(scale, gain);
2203 if (coup->coupling_point == AFTER_IMDCT) {
2204 coup->gain[c][0] = gain_cache;
2206 for (g = 0; g < sce->ics.num_window_groups; g++) {
2207 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2208 if (sce->band_type[idx] != ZERO_BT) {
2210 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2218 gain_cache = GET_GAIN(scale, t) * s;
2221 coup->gain[c][idx] = gain_cache;
2231 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2233 * @return Returns number of bytes consumed.
2235 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2239 int num_excl_chan = 0;
2242 for (i = 0; i < 7; i++)
2243 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2244 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2246 return num_excl_chan / 7;
2250 * Decode dynamic range information; reference: table 4.52.
2252 * @return Returns number of bytes consumed.
2254 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2258 int drc_num_bands = 1;
2261 /* pce_tag_present? */
2262 if (get_bits1(gb)) {
2263 che_drc->pce_instance_tag = get_bits(gb, 4);
2264 skip_bits(gb, 4); // tag_reserved_bits
2268 /* excluded_chns_present? */
2269 if (get_bits1(gb)) {
2270 n += decode_drc_channel_exclusions(che_drc, gb);
2273 /* drc_bands_present? */
2274 if (get_bits1(gb)) {
2275 che_drc->band_incr = get_bits(gb, 4);
2276 che_drc->interpolation_scheme = get_bits(gb, 4);
2278 drc_num_bands += che_drc->band_incr;
2279 for (i = 0; i < drc_num_bands; i++) {
2280 che_drc->band_top[i] = get_bits(gb, 8);
2285 /* prog_ref_level_present? */
2286 if (get_bits1(gb)) {
2287 che_drc->prog_ref_level = get_bits(gb, 7);
2288 skip_bits1(gb); // prog_ref_level_reserved_bits
2292 for (i = 0; i < drc_num_bands; i++) {
2293 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2294 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2301 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2303 int i, major, minor;
2308 get_bits(gb, 13); len -= 13;
2310 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2311 buf[i] = get_bits(gb, 8);
2314 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2315 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2317 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2318 ac->avctx->internal->skip_samples = 1024;
2322 skip_bits_long(gb, len);
2328 * Decode extension data (incomplete); reference: table 4.51.
2330 * @param cnt length of TYPE_FIL syntactic element in bytes
2332 * @return Returns number of bytes consumed
2334 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2335 ChannelElement *che, enum RawDataBlockType elem_type)
2339 int type = get_bits(gb, 4);
2341 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2342 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2344 switch (type) { // extension type
2345 case EXT_SBR_DATA_CRC:
2349 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2351 } else if (!ac->oc[1].m4ac.sbr) {
2352 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2353 skip_bits_long(gb, 8 * cnt - 4);
2355 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2356 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2357 skip_bits_long(gb, 8 * cnt - 4);
2359 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2360 ac->oc[1].m4ac.sbr = 1;
2361 ac->oc[1].m4ac.ps = 1;
2362 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2363 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2364 ac->oc[1].status, 1);
2366 ac->oc[1].m4ac.sbr = 1;
2367 ac->avctx->profile = FF_PROFILE_AAC_HE;
2369 res = AAC_RENAME(ff_decode_sbr_extension)(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2371 case EXT_DYNAMIC_RANGE:
2372 res = decode_dynamic_range(&ac->che_drc, gb);
2375 decode_fill(ac, gb, 8 * cnt - 4);
2378 case EXT_DATA_ELEMENT:
2380 skip_bits_long(gb, 8 * cnt - 4);
2387 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2389 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2390 * @param coef spectral coefficients
2392 static void apply_tns(INTFLOAT coef[1024], TemporalNoiseShaping *tns,
2393 IndividualChannelStream *ics, int decode)
2395 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2397 int bottom, top, order, start, end, size, inc;
2398 INTFLOAT lpc[TNS_MAX_ORDER];
2399 INTFLOAT tmp[TNS_MAX_ORDER+1];
2401 for (w = 0; w < ics->num_windows; w++) {
2402 bottom = ics->num_swb;
2403 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2405 bottom = FFMAX(0, top - tns->length[w][filt]);
2406 order = tns->order[w][filt];
2411 AAC_RENAME(compute_lpc_coefs)(tns->coef[w][filt], order, lpc, 0, 0, 0);
2413 start = ics->swb_offset[FFMIN(bottom, mmm)];
2414 end = ics->swb_offset[FFMIN( top, mmm)];
2415 if ((size = end - start) <= 0)
2417 if (tns->direction[w][filt]) {
2427 for (m = 0; m < size; m++, start += inc)
2428 for (i = 1; i <= FFMIN(m, order); i++)
2429 coef[start] -= AAC_MUL26(coef[start - i * inc], lpc[i - 1]);
2432 for (m = 0; m < size; m++, start += inc) {
2433 tmp[0] = coef[start];
2434 for (i = 1; i <= FFMIN(m, order); i++)
2435 coef[start] += AAC_MUL26(tmp[i], lpc[i - 1]);
2436 for (i = order; i > 0; i--)
2437 tmp[i] = tmp[i - 1];
2445 * Apply windowing and MDCT to obtain the spectral
2446 * coefficient from the predicted sample by LTP.
2448 static void windowing_and_mdct_ltp(AACContext *ac, INTFLOAT *out,
2449 INTFLOAT *in, IndividualChannelStream *ics)
2451 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2452 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2453 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2454 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2456 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2457 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2459 memset(in, 0, 448 * sizeof(*in));
2460 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2462 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2463 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2465 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2466 memset(in + 1024 + 576, 0, 448 * sizeof(*in));
2468 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2472 * Apply the long term prediction
2474 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2476 const LongTermPrediction *ltp = &sce->ics.ltp;
2477 const uint16_t *offsets = sce->ics.swb_offset;
2480 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2481 INTFLOAT *predTime = sce->ret;
2482 INTFLOAT *predFreq = ac->buf_mdct;
2483 int16_t num_samples = 2048;
2485 if (ltp->lag < 1024)
2486 num_samples = ltp->lag + 1024;
2487 for (i = 0; i < num_samples; i++)
2488 predTime[i] = AAC_MUL30(sce->ltp_state[i + 2048 - ltp->lag], ltp->coef);
2489 memset(&predTime[i], 0, (2048 - i) * sizeof(*predTime));
2491 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2493 if (sce->tns.present)
2494 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2496 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2498 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2499 sce->coeffs[i] += predFreq[i];
2504 * Update the LTP buffer for next frame
2506 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2508 IndividualChannelStream *ics = &sce->ics;
2509 INTFLOAT *saved = sce->saved;
2510 INTFLOAT *saved_ltp = sce->coeffs;
2511 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2512 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2515 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2516 memcpy(saved_ltp, saved, 512 * sizeof(*saved_ltp));
2517 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2518 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2520 for (i = 0; i < 64; i++)
2521 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2522 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2523 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(*saved_ltp));
2524 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2525 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2527 for (i = 0; i < 64; i++)
2528 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2529 } else { // LONG_STOP or ONLY_LONG
2530 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2532 for (i = 0; i < 512; i++)
2533 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], lwindow[511 - i]);
2536 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2537 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2538 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2542 * Conduct IMDCT and windowing.
2544 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2546 IndividualChannelStream *ics = &sce->ics;
2547 INTFLOAT *in = sce->coeffs;
2548 INTFLOAT *out = sce->ret;
2549 INTFLOAT *saved = sce->saved;
2550 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2551 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2552 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2553 INTFLOAT *buf = ac->buf_mdct;
2554 INTFLOAT *temp = ac->temp;
2558 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2559 for (i = 0; i < 1024; i += 128)
2560 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2562 ac->mdct.imdct_half(&ac->mdct, buf, in);
2564 for (i=0; i<1024; i++)
2565 buf[i] = (buf[i] + 4) >> 3;
2566 #endif /* USE_FIXED */
2569 /* window overlapping
2570 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2571 * and long to short transitions are considered to be short to short
2572 * transitions. This leaves just two cases (long to long and short to short)
2573 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2575 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2576 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2577 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2579 memcpy( out, saved, 448 * sizeof(*out));
2581 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2582 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2583 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2584 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2585 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2586 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2587 memcpy( out + 448 + 4*128, temp, 64 * sizeof(*out));
2589 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2590 memcpy( out + 576, buf + 64, 448 * sizeof(*out));
2595 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2596 memcpy( saved, temp + 64, 64 * sizeof(*saved));
2597 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2598 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2599 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2600 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2601 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2602 memcpy( saved, buf + 512, 448 * sizeof(*saved));
2603 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2604 } else { // LONG_STOP or ONLY_LONG
2605 memcpy( saved, buf + 512, 512 * sizeof(*saved));
2609 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2611 IndividualChannelStream *ics = &sce->ics;
2612 INTFLOAT *in = sce->coeffs;
2613 INTFLOAT *out = sce->ret;
2614 INTFLOAT *saved = sce->saved;
2615 INTFLOAT *buf = ac->buf_mdct;
2618 #endif /* USE_FIXED */
2621 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2624 for (i = 0; i < 1024; i++)
2625 buf[i] = (buf[i] + 2) >> 2;
2626 #endif /* USE_FIXED */
2628 // window overlapping
2629 if (ics->use_kb_window[1]) {
2630 // AAC LD uses a low overlap sine window instead of a KBD window
2631 memcpy(out, saved, 192 * sizeof(*out));
2632 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, AAC_RENAME(ff_sine_128), 64);
2633 memcpy( out + 320, buf + 64, 192 * sizeof(*out));
2635 ac->fdsp->vector_fmul_window(out, saved, buf, AAC_RENAME(ff_sine_512), 256);
2639 memcpy(saved, buf + 256, 256 * sizeof(*saved));
2642 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2644 INTFLOAT *in = sce->coeffs;
2645 INTFLOAT *out = sce->ret;
2646 INTFLOAT *saved = sce->saved;
2647 INTFLOAT *buf = ac->buf_mdct;
2649 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2650 const int n2 = n >> 1;
2651 const int n4 = n >> 2;
2652 const INTFLOAT *const window = n == 480 ? AAC_RENAME(ff_aac_eld_window_480) :
2653 AAC_RENAME(ff_aac_eld_window_512);
2655 // Inverse transform, mapped to the conventional IMDCT by
2656 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2657 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2658 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2659 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2660 for (i = 0; i < n2; i+=2) {
2662 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2663 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2667 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2670 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2673 for (i = 0; i < 1024; i++)
2674 buf[i] = (buf[i] + 1) >> 1;
2675 #endif /* USE_FIXED */
2677 for (i = 0; i < n; i+=2) {
2680 // Like with the regular IMDCT at this point we still have the middle half
2681 // of a transform but with even symmetry on the left and odd symmetry on
2684 // window overlapping
2685 // The spec says to use samples [0..511] but the reference decoder uses
2686 // samples [128..639].
2687 for (i = n4; i < n2; i ++) {
2688 out[i - n4] = AAC_MUL31( buf[ n2 - 1 - i] , window[i - n4]) +
2689 AAC_MUL31( saved[ i + n2] , window[i + n - n4]) +
2690 AAC_MUL31(-saved[n + n2 - 1 - i] , window[i + 2*n - n4]) +
2691 AAC_MUL31(-saved[ 2*n + n2 + i] , window[i + 3*n - n4]);
2693 for (i = 0; i < n2; i ++) {
2694 out[n4 + i] = AAC_MUL31( buf[ i] , window[i + n2 - n4]) +
2695 AAC_MUL31(-saved[ n - 1 - i] , window[i + n2 + n - n4]) +
2696 AAC_MUL31(-saved[ n + i] , window[i + n2 + 2*n - n4]) +
2697 AAC_MUL31( saved[2*n + n - 1 - i] , window[i + n2 + 3*n - n4]);
2699 for (i = 0; i < n4; i ++) {
2700 out[n2 + n4 + i] = AAC_MUL31( buf[ i + n2] , window[i + n - n4]) +
2701 AAC_MUL31(-saved[n2 - 1 - i] , window[i + 2*n - n4]) +
2702 AAC_MUL31(-saved[n + n2 + i] , window[i + 3*n - n4]);
2706 memmove(saved + n, saved, 2 * n * sizeof(*saved));
2707 memcpy( saved, buf, n * sizeof(*saved));
2711 * channel coupling transformation interface
2713 * @param apply_coupling_method pointer to (in)dependent coupling function
2715 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2716 enum RawDataBlockType type, int elem_id,
2717 enum CouplingPoint coupling_point,
2718 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2722 for (i = 0; i < MAX_ELEM_ID; i++) {
2723 ChannelElement *cce = ac->che[TYPE_CCE][i];
2726 if (cce && cce->coup.coupling_point == coupling_point) {
2727 ChannelCoupling *coup = &cce->coup;
2729 for (c = 0; c <= coup->num_coupled; c++) {
2730 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2731 if (coup->ch_select[c] != 1) {
2732 apply_coupling_method(ac, &cc->ch[0], cce, index);
2733 if (coup->ch_select[c] != 0)
2736 if (coup->ch_select[c] != 2)
2737 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2739 index += 1 + (coup->ch_select[c] == 3);
2746 * Convert spectral data to samples, applying all supported tools as appropriate.
2748 static void spectral_to_sample(AACContext *ac, int samples)
2751 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2752 switch (ac->oc[1].m4ac.object_type) {
2754 imdct_and_window = imdct_and_windowing_ld;
2756 case AOT_ER_AAC_ELD:
2757 imdct_and_window = imdct_and_windowing_eld;
2760 imdct_and_window = ac->imdct_and_windowing;
2762 for (type = 3; type >= 0; type--) {
2763 for (i = 0; i < MAX_ELEM_ID; i++) {
2764 ChannelElement *che = ac->che[type][i];
2765 if (che && che->present) {
2766 if (type <= TYPE_CPE)
2767 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, AAC_RENAME(apply_dependent_coupling));
2768 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2769 if (che->ch[0].ics.predictor_present) {
2770 if (che->ch[0].ics.ltp.present)
2771 ac->apply_ltp(ac, &che->ch[0]);
2772 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2773 ac->apply_ltp(ac, &che->ch[1]);
2776 if (che->ch[0].tns.present)
2777 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2778 if (che->ch[1].tns.present)
2779 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2780 if (type <= TYPE_CPE)
2781 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, AAC_RENAME(apply_dependent_coupling));
2782 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2783 imdct_and_window(ac, &che->ch[0]);
2784 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2785 ac->update_ltp(ac, &che->ch[0]);
2786 if (type == TYPE_CPE) {
2787 imdct_and_window(ac, &che->ch[1]);
2788 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2789 ac->update_ltp(ac, &che->ch[1]);
2791 if (ac->oc[1].m4ac.sbr > 0) {
2792 AAC_RENAME(ff_sbr_apply)(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2795 if (type <= TYPE_CCE)
2796 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, AAC_RENAME(apply_independent_coupling));
2801 /* preparation for resampler */
2802 for(j = 0; j<samples; j++){
2803 che->ch[0].ret[j] = (int32_t)av_clip64((int64_t)che->ch[0].ret[j]*128, INT32_MIN, INT32_MAX-0x8000)+0x8000;
2804 if(type == TYPE_CPE)
2805 che->ch[1].ret[j] = (int32_t)av_clip64((int64_t)che->ch[1].ret[j]*128, INT32_MIN, INT32_MAX-0x8000)+0x8000;
2808 #endif /* USE_FIXED */
2811 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2817 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2820 AACADTSHeaderInfo hdr_info;
2821 uint8_t layout_map[MAX_ELEM_ID*4][3];
2822 int layout_map_tags, ret;
2824 size = avpriv_aac_parse_header(gb, &hdr_info);
2826 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2827 // This is 2 for "VLB " audio in NSV files.
2828 // See samples/nsv/vlb_audio.
2829 avpriv_report_missing_feature(ac->avctx,
2830 "More than one AAC RDB per ADTS frame");
2831 ac->warned_num_aac_frames = 1;
2833 push_output_configuration(ac);
2834 if (hdr_info.chan_config) {
2835 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2836 if ((ret = set_default_channel_config(ac->avctx,
2839 hdr_info.chan_config)) < 0)
2841 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2842 FFMAX(ac->oc[1].status,
2843 OC_TRIAL_FRAME), 0)) < 0)
2846 ac->oc[1].m4ac.chan_config = 0;
2848 * dual mono frames in Japanese DTV can have chan_config 0
2849 * WITHOUT specifying PCE.
2850 * thus, set dual mono as default.
2852 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2853 layout_map_tags = 2;
2854 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2855 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2856 layout_map[0][1] = 0;
2857 layout_map[1][1] = 1;
2858 if (output_configure(ac, layout_map, layout_map_tags,
2863 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2864 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2865 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2866 ac->oc[1].m4ac.frame_length_short = 0;
2867 if (ac->oc[0].status != OC_LOCKED ||
2868 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2869 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2870 ac->oc[1].m4ac.sbr = -1;
2871 ac->oc[1].m4ac.ps = -1;
2873 if (!hdr_info.crc_absent)
2879 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2880 int *got_frame_ptr, GetBitContext *gb)
2882 AACContext *ac = avctx->priv_data;
2883 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2884 ChannelElement *che;
2886 int samples = m4ac->frame_length_short ? 960 : 1024;
2887 int chan_config = m4ac->chan_config;
2888 int aot = m4ac->object_type;
2890 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2895 if ((err = frame_configure_elements(avctx)) < 0)
2898 // The FF_PROFILE_AAC_* defines are all object_type - 1
2899 // This may lead to an undefined profile being signaled
2900 ac->avctx->profile = aot - 1;
2902 ac->tags_mapped = 0;
2904 if (chan_config < 0 || (chan_config >= 8 && chan_config < 11) || chan_config >= 13) {
2905 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2907 return AVERROR_INVALIDDATA;
2909 for (i = 0; i < tags_per_config[chan_config]; i++) {
2910 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2911 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2912 if (!(che=get_che(ac, elem_type, elem_id))) {
2913 av_log(ac->avctx, AV_LOG_ERROR,
2914 "channel element %d.%d is not allocated\n",
2915 elem_type, elem_id);
2916 return AVERROR_INVALIDDATA;
2919 if (aot != AOT_ER_AAC_ELD)
2921 switch (elem_type) {
2923 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2926 err = decode_cpe(ac, gb, che);
2929 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2936 spectral_to_sample(ac, samples);
2938 if (!ac->frame->data[0] && samples) {
2939 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
2940 return AVERROR_INVALIDDATA;
2943 ac->frame->nb_samples = samples;
2944 ac->frame->sample_rate = avctx->sample_rate;
2947 skip_bits_long(gb, get_bits_left(gb));
2951 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2952 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2954 AACContext *ac = avctx->priv_data;
2955 ChannelElement *che = NULL, *che_prev = NULL;
2956 enum RawDataBlockType elem_type, che_prev_type = TYPE_END;
2958 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2959 int is_dmono, sce_count = 0;
2960 int payload_alignment;
2964 if (show_bits(gb, 12) == 0xfff) {
2965 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2966 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2969 if (ac->oc[1].m4ac.sampling_index > 12) {
2970 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2971 err = AVERROR_INVALIDDATA;
2976 if ((err = frame_configure_elements(avctx)) < 0)
2979 // The FF_PROFILE_AAC_* defines are all object_type - 1
2980 // This may lead to an undefined profile being signaled
2981 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2983 payload_alignment = get_bits_count(gb);
2984 ac->tags_mapped = 0;
2986 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2987 elem_id = get_bits(gb, 4);
2989 if (avctx->debug & FF_DEBUG_STARTCODE)
2990 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2992 if (!avctx->channels && elem_type != TYPE_PCE) {
2993 err = AVERROR_INVALIDDATA;
2997 if (elem_type < TYPE_DSE) {
2998 if (!(che=get_che(ac, elem_type, elem_id))) {
2999 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
3000 elem_type, elem_id);
3001 err = AVERROR_INVALIDDATA;
3008 switch (elem_type) {
3011 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3017 err = decode_cpe(ac, gb, che);
3022 err = decode_cce(ac, gb, che);
3026 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3031 err = skip_data_stream_element(ac, gb);
3035 uint8_t layout_map[MAX_ELEM_ID*4][3];
3038 int pushed = push_output_configuration(ac);
3039 if (pce_found && !pushed) {
3040 err = AVERROR_INVALIDDATA;
3044 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb,
3051 av_log(avctx, AV_LOG_ERROR,
3052 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3053 pop_output_configuration(ac);
3055 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3057 ac->oc[1].m4ac.chan_config = 0;
3065 elem_id += get_bits(gb, 8) - 1;
3066 if (get_bits_left(gb) < 8 * elem_id) {
3067 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3068 err = AVERROR_INVALIDDATA;
3072 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, che_prev_type);
3073 err = 0; /* FIXME */
3077 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3081 if (elem_type < TYPE_DSE) {
3083 che_prev_type = elem_type;
3089 if (get_bits_left(gb) < 3) {
3090 av_log(avctx, AV_LOG_ERROR, overread_err);
3091 err = AVERROR_INVALIDDATA;
3096 if (!avctx->channels) {
3101 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3102 samples <<= multiplier;
3104 spectral_to_sample(ac, samples);
3106 if (ac->oc[1].status && audio_found) {
3107 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3108 avctx->frame_size = samples;
3109 ac->oc[1].status = OC_LOCKED;
3113 avctx->internal->skip_samples_multiplier = 2;
3115 if (!ac->frame->data[0] && samples) {
3116 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3117 err = AVERROR_INVALIDDATA;
3122 ac->frame->nb_samples = samples;
3123 ac->frame->sample_rate = avctx->sample_rate;
3125 av_frame_unref(ac->frame);
3126 *got_frame_ptr = !!samples;
3128 /* for dual-mono audio (SCE + SCE) */
3129 is_dmono = ac->dmono_mode && sce_count == 2 &&
3130 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3132 if (ac->dmono_mode == 1)
3133 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3134 else if (ac->dmono_mode == 2)
3135 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3140 pop_output_configuration(ac);
3144 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3145 int *got_frame_ptr, AVPacket *avpkt)
3147 AACContext *ac = avctx->priv_data;
3148 const uint8_t *buf = avpkt->data;
3149 int buf_size = avpkt->size;
3154 int new_extradata_size;
3155 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3156 AV_PKT_DATA_NEW_EXTRADATA,
3157 &new_extradata_size);
3158 int jp_dualmono_size;
3159 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3160 AV_PKT_DATA_JP_DUALMONO,
3163 if (new_extradata && 0) {
3164 av_free(avctx->extradata);
3165 avctx->extradata = av_mallocz(new_extradata_size +
3166 AV_INPUT_BUFFER_PADDING_SIZE);
3167 if (!avctx->extradata)
3168 return AVERROR(ENOMEM);
3169 avctx->extradata_size = new_extradata_size;
3170 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3171 push_output_configuration(ac);
3172 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3174 avctx->extradata_size*8LL, 1) < 0) {
3175 pop_output_configuration(ac);
3176 return AVERROR_INVALIDDATA;
3181 if (jp_dualmono && jp_dualmono_size > 0)
3182 ac->dmono_mode = 1 + *jp_dualmono;
3183 if (ac->force_dmono_mode >= 0)
3184 ac->dmono_mode = ac->force_dmono_mode;
3186 if (INT_MAX / 8 <= buf_size)
3187 return AVERROR_INVALIDDATA;
3189 if ((err = init_get_bits8(&gb, buf, buf_size)) < 0)
3192 switch (ac->oc[1].m4ac.object_type) {
3194 case AOT_ER_AAC_LTP:
3196 case AOT_ER_AAC_ELD:
3197 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3200 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3205 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3206 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3207 if (buf[buf_offset])
3210 return buf_size > buf_offset ? buf_consumed : buf_size;
3213 static av_cold int aac_decode_close(AVCodecContext *avctx)
3215 AACContext *ac = avctx->priv_data;
3218 for (i = 0; i < MAX_ELEM_ID; i++) {
3219 for (type = 0; type < 4; type++) {
3220 if (ac->che[type][i])
3221 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][i]->sbr);
3222 av_freep(&ac->che[type][i]);
3226 ff_mdct_end(&ac->mdct);
3227 ff_mdct_end(&ac->mdct_small);
3228 ff_mdct_end(&ac->mdct_ld);
3229 ff_mdct_end(&ac->mdct_ltp);
3231 ff_mdct15_uninit(&ac->mdct480);
3233 av_freep(&ac->fdsp);
3237 static void aacdec_init(AACContext *c)
3239 c->imdct_and_windowing = imdct_and_windowing;
3240 c->apply_ltp = apply_ltp;
3241 c->apply_tns = apply_tns;
3242 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3243 c->update_ltp = update_ltp;
3245 c->vector_pow43 = vector_pow43;
3246 c->subband_scale = subband_scale;
3251 ff_aacdec_init_mips(c);
3252 #endif /* !USE_FIXED */
3255 * AVOptions for Japanese DTV specific extensions (ADTS only)
3257 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3258 static const AVOption options[] = {
3259 {"dual_mono_mode", "Select the channel to decode for dual mono",
3260 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3261 AACDEC_FLAGS, "dual_mono_mode"},
3263 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3264 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3265 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3266 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3271 static const AVClass aac_decoder_class = {
3272 .class_name = "AAC decoder",
3273 .item_name = av_default_item_name,
3275 .version = LIBAVUTIL_VERSION_INT,