1 /*
2 * AAC decoder
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>
6 *
7 * AAC LATM decoder
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
10 *
11 * AAC decoder fixed-point implementation
12 * Copyright (c) 2013
13 * MIPS Technologies, Inc., California.
14 *
15 * This file is part of FFmpeg.
16 *
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.
21 *
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.
26 *
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
30 */
31
32 /**
33 * @file
34 * AAC decoder
35 * @author Oded Shimon ( ods15 ods15 dyndns org )
36 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
37 *
38 * AAC decoder fixed-point implementation
39 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
40 * @author Nedeljko Babic ( nedeljko.babic imgtec com )
41 */
42
43 /*
44 * supported tools
45 *
46 * Support? Name
47 * N (code in SoC repo) gain control
48 * Y block switching
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
55 * Y intensity stereo
56 * Y channel coupling
57 * Y frequency domain prediction
58 * Y Perceptual Noise Substitution
59 * Y Mid/Side stereo
60 * N Scalable Inverse AAC Quantization
61 * N Frequency Selective Switch
62 * N upsampling filter
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
69 * N CELP
70 * N Silence Compression
71 * N HVXC
72 * N HVXC 4kbits/s VR
73 * N Structured Audio tools
74 * N Structured Audio Sample Bank Format
75 * N MIDI
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)
83 * Y Parametric Stereo
84 * N Direct Stream Transfer
85 * Y (not in fixed point code) Enhanced AAC Low Delay (ER AAC ELD)
86 *
87 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
88 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
89 Parametric Stereo.
90 */
91
92 #include "libavutil/thread.h"
93
94 static VLC vlc_scalefactors;
95 static VLC vlc_spectral[11];
96
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);
100
101 #define overread_err "Input buffer exhausted before END element found\n"
102
count_channels(uint8_t (* layout)[3],int tags)103 static int count_channels(uint8_t (*layout)[3], int tags)
104 {
105 int i, sum = 0;
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);
111 }
112 return sum;
113 }
114
115 /**
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.
119 *
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
124 *
125 * @return Returns error status. 0 - OK, !0 - error
126 */
che_configure(AACContext * ac,enum ChannelPosition che_pos,int type,int id,int * channels)127 static av_cold int che_configure(AACContext *ac,
128 enum ChannelPosition che_pos,
129 int type, int id, int *channels)
130 {
131 if (*channels >= MAX_CHANNELS)
132 return AVERROR_INVALIDDATA;
133 if (che_pos) {
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);
138 }
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;
143 }
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];
148 }
149 }
150 } else {
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]);
154 }
155 return 0;
156 }
157
frame_configure_elements(AVCodecContext * avctx)158 static int frame_configure_elements(AVCodecContext *avctx)
159 {
160 AACContext *ac = avctx->priv_data;
161 int type, id, ch, ret;
162
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];
167 if (che) {
168 che->ch[0].ret = che->ch[0].ret_buf;
169 che->ch[1].ret = che->ch[1].ret_buf;
170 }
171 }
172 }
173
174 /* get output buffer */
175 av_frame_unref(ac->frame);
176 if (!avctx->channels)
177 return 1;
178
179 ac->frame->nb_samples = 2048;
180 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
181 return ret;
182
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];
187 }
188
189 return 0;
190 }
191
192 struct elem_to_channel {
193 uint64_t av_position;
194 uint8_t syn_ele;
195 uint8_t elem_id;
196 uint8_t aac_position;
197 };
198
assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],uint8_t (* layout_map)[3],int offset,uint64_t left,uint64_t right,int pos,uint64_t * layout)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, uint64_t *layout)
202 {
203 if (layout_map[offset][0] == TYPE_CPE) {
204 e2c_vec[offset] = (struct elem_to_channel) {
205 .av_position = left | right,
206 .syn_ele = TYPE_CPE,
207 .elem_id = layout_map[offset][1],
208 .aac_position = pos
209 };
210 if (e2c_vec[offset].av_position != UINT64_MAX)
211 *layout |= e2c_vec[offset].av_position;
212
213 return 1;
214 } else {
215 e2c_vec[offset] = (struct elem_to_channel) {
216 .av_position = left,
217 .syn_ele = TYPE_SCE,
218 .elem_id = layout_map[offset][1],
219 .aac_position = pos
220 };
221 e2c_vec[offset + 1] = (struct elem_to_channel) {
222 .av_position = right,
223 .syn_ele = TYPE_SCE,
224 .elem_id = layout_map[offset + 1][1],
225 .aac_position = pos
226 };
227 if (left != UINT64_MAX)
228 *layout |= left;
229
230 if (right != UINT64_MAX)
231 *layout |= right;
232
233 return 2;
234 }
235 }
236
count_paired_channels(uint8_t (* layout_map)[3],int tags,int pos,int * current)237 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
238 int *current)
239 {
240 int num_pos_channels = 0;
241 int first_cpe = 0;
242 int sce_parity = 0;
243 int i;
244 for (i = *current; i < tags; i++) {
245 if (layout_map[i][2] != pos)
246 break;
247 if (layout_map[i][0] == TYPE_CPE) {
248 if (sce_parity) {
249 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
250 sce_parity = 0;
251 } else {
252 return -1;
253 }
254 }
255 num_pos_channels += 2;
256 first_cpe = 1;
257 } else {
258 num_pos_channels++;
259 sce_parity ^= 1;
260 }
261 }
262 if (sce_parity &&
263 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
264 return -1;
265 *current = i;
266 return num_pos_channels;
267 }
268
269 #define PREFIX_FOR_22POINT2 (AV_CH_LAYOUT_7POINT1_WIDE_BACK|AV_CH_BACK_CENTER|AV_CH_SIDE_LEFT|AV_CH_SIDE_RIGHT|AV_CH_LOW_FREQUENCY_2)
sniff_channel_order(uint8_t (* layout_map)[3],int tags)270 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
271 {
272 int i, n, total_non_cc_elements;
273 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
274 int num_front_channels, num_side_channels, num_back_channels;
275 uint64_t layout = 0;
276
277 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
278 return 0;
279
280 i = 0;
281 num_front_channels =
282 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
283 if (num_front_channels < 0)
284 return 0;
285 num_side_channels =
286 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
287 if (num_side_channels < 0)
288 return 0;
289 num_back_channels =
290 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
291 if (num_back_channels < 0)
292 return 0;
293
294 if (num_side_channels == 0 && num_back_channels >= 4) {
295 num_side_channels = 2;
296 num_back_channels -= 2;
297 }
298
299 i = 0;
300 if (num_front_channels & 1) {
301 e2c_vec[i] = (struct elem_to_channel) {
302 .av_position = AV_CH_FRONT_CENTER,
303 .syn_ele = TYPE_SCE,
304 .elem_id = layout_map[i][1],
305 .aac_position = AAC_CHANNEL_FRONT
306 };
307 layout |= e2c_vec[i].av_position;
308 i++;
309 num_front_channels--;
310 }
311 if (num_front_channels >= 4) {
312 i += assign_pair(e2c_vec, layout_map, i,
313 AV_CH_FRONT_LEFT_OF_CENTER,
314 AV_CH_FRONT_RIGHT_OF_CENTER,
315 AAC_CHANNEL_FRONT, &layout);
316 num_front_channels -= 2;
317 }
318 if (num_front_channels >= 2) {
319 i += assign_pair(e2c_vec, layout_map, i,
320 AV_CH_FRONT_LEFT,
321 AV_CH_FRONT_RIGHT,
322 AAC_CHANNEL_FRONT, &layout);
323 num_front_channels -= 2;
324 }
325 while (num_front_channels >= 2) {
326 i += assign_pair(e2c_vec, layout_map, i,
327 UINT64_MAX,
328 UINT64_MAX,
329 AAC_CHANNEL_FRONT, &layout);
330 num_front_channels -= 2;
331 }
332
333 if (num_side_channels >= 2) {
334 i += assign_pair(e2c_vec, layout_map, i,
335 AV_CH_SIDE_LEFT,
336 AV_CH_SIDE_RIGHT,
337 AAC_CHANNEL_FRONT, &layout);
338 num_side_channels -= 2;
339 }
340 while (num_side_channels >= 2) {
341 i += assign_pair(e2c_vec, layout_map, i,
342 UINT64_MAX,
343 UINT64_MAX,
344 AAC_CHANNEL_SIDE, &layout);
345 num_side_channels -= 2;
346 }
347
348 while (num_back_channels >= 4) {
349 i += assign_pair(e2c_vec, layout_map, i,
350 UINT64_MAX,
351 UINT64_MAX,
352 AAC_CHANNEL_BACK, &layout);
353 num_back_channels -= 2;
354 }
355 if (num_back_channels >= 2) {
356 i += assign_pair(e2c_vec, layout_map, i,
357 AV_CH_BACK_LEFT,
358 AV_CH_BACK_RIGHT,
359 AAC_CHANNEL_BACK, &layout);
360 num_back_channels -= 2;
361 }
362 if (num_back_channels) {
363 e2c_vec[i] = (struct elem_to_channel) {
364 .av_position = AV_CH_BACK_CENTER,
365 .syn_ele = TYPE_SCE,
366 .elem_id = layout_map[i][1],
367 .aac_position = AAC_CHANNEL_BACK
368 };
369 layout |= e2c_vec[i].av_position;
370 i++;
371 num_back_channels--;
372 }
373
374 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
375 e2c_vec[i] = (struct elem_to_channel) {
376 .av_position = AV_CH_LOW_FREQUENCY,
377 .syn_ele = TYPE_LFE,
378 .elem_id = layout_map[i][1],
379 .aac_position = AAC_CHANNEL_LFE
380 };
381 layout |= e2c_vec[i].av_position;
382 i++;
383 }
384 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
385 e2c_vec[i] = (struct elem_to_channel) {
386 .av_position = AV_CH_LOW_FREQUENCY_2,
387 .syn_ele = TYPE_LFE,
388 .elem_id = layout_map[i][1],
389 .aac_position = AAC_CHANNEL_LFE
390 };
391 layout |= e2c_vec[i].av_position;
392 i++;
393 }
394 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
395 e2c_vec[i] = (struct elem_to_channel) {
396 .av_position = UINT64_MAX,
397 .syn_ele = TYPE_LFE,
398 .elem_id = layout_map[i][1],
399 .aac_position = AAC_CHANNEL_LFE
400 };
401 i++;
402 }
403
404 // The previous checks would end up at 8 at this point for 22.2
405 if (layout == PREFIX_FOR_22POINT2 && tags == 16 && i == 8) {
406 const uint8_t (*reference_layout_map)[3] = aac_channel_layout_map[12];
407 for (int j = 0; j < tags; j++) {
408 if (layout_map[j][0] != reference_layout_map[j][0] ||
409 layout_map[j][2] != reference_layout_map[j][2])
410 goto end_of_layout_definition;
411 }
412
413 e2c_vec[i] = (struct elem_to_channel) {
414 .av_position = AV_CH_TOP_FRONT_CENTER,
415 .syn_ele = layout_map[i][0],
416 .elem_id = layout_map[i][1],
417 .aac_position = layout_map[i][2]
418 }; layout |= e2c_vec[i].av_position; i++;
419 i += assign_pair(e2c_vec, layout_map, i,
420 AV_CH_TOP_FRONT_LEFT,
421 AV_CH_TOP_FRONT_RIGHT,
422 AAC_CHANNEL_FRONT,
423 &layout);
424 i += assign_pair(e2c_vec, layout_map, i,
425 AV_CH_TOP_SIDE_LEFT,
426 AV_CH_TOP_SIDE_RIGHT,
427 AAC_CHANNEL_SIDE,
428 &layout);
429 e2c_vec[i] = (struct elem_to_channel) {
430 .av_position = AV_CH_TOP_CENTER,
431 .syn_ele = layout_map[i][0],
432 .elem_id = layout_map[i][1],
433 .aac_position = layout_map[i][2]
434 }; layout |= e2c_vec[i].av_position; i++;
435 i += assign_pair(e2c_vec, layout_map, i,
436 AV_CH_TOP_BACK_LEFT,
437 AV_CH_TOP_BACK_RIGHT,
438 AAC_CHANNEL_BACK,
439 &layout);
440 e2c_vec[i] = (struct elem_to_channel) {
441 .av_position = AV_CH_TOP_BACK_CENTER,
442 .syn_ele = layout_map[i][0],
443 .elem_id = layout_map[i][1],
444 .aac_position = layout_map[i][2]
445 }; layout |= e2c_vec[i].av_position; i++;
446 e2c_vec[i] = (struct elem_to_channel) {
447 .av_position = AV_CH_BOTTOM_FRONT_CENTER,
448 .syn_ele = layout_map[i][0],
449 .elem_id = layout_map[i][1],
450 .aac_position = layout_map[i][2]
451 }; layout |= e2c_vec[i].av_position; i++;
452 i += assign_pair(e2c_vec, layout_map, i,
453 AV_CH_BOTTOM_FRONT_LEFT,
454 AV_CH_BOTTOM_FRONT_RIGHT,
455 AAC_CHANNEL_FRONT,
456 &layout);
457 }
458
459 end_of_layout_definition:
460
461 total_non_cc_elements = n = i;
462
463 if (layout == AV_CH_LAYOUT_22POINT2) {
464 // For 22.2 reorder the result as needed
465 FFSWAP(struct elem_to_channel, e2c_vec[2], e2c_vec[0]); // FL & FR first (final), FC third
466 FFSWAP(struct elem_to_channel, e2c_vec[2], e2c_vec[1]); // FC second (final), FLc & FRc third
467 FFSWAP(struct elem_to_channel, e2c_vec[6], e2c_vec[2]); // LFE1 third (final), FLc & FRc seventh
468 FFSWAP(struct elem_to_channel, e2c_vec[4], e2c_vec[3]); // BL & BR fourth (final), SiL & SiR fifth
469 FFSWAP(struct elem_to_channel, e2c_vec[6], e2c_vec[4]); // FLc & FRc fifth (final), SiL & SiR seventh
470 FFSWAP(struct elem_to_channel, e2c_vec[7], e2c_vec[6]); // LFE2 seventh (final), SiL & SiR eight (final)
471 FFSWAP(struct elem_to_channel, e2c_vec[9], e2c_vec[8]); // TpFL & TpFR ninth (final), TFC tenth (final)
472 FFSWAP(struct elem_to_channel, e2c_vec[11], e2c_vec[10]); // TC eleventh (final), TpSiL & TpSiR twelth
473 FFSWAP(struct elem_to_channel, e2c_vec[12], e2c_vec[11]); // TpBL & TpBR twelth (final), TpSiL & TpSiR thirteenth (final)
474 } else {
475 // For everything else, utilize the AV channel position define as a
476 // stable sort.
477 do {
478 int next_n = 0;
479 for (i = 1; i < n; i++)
480 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
481 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
482 next_n = i;
483 }
484 n = next_n;
485 } while (n > 0);
486
487 }
488
489 for (i = 0; i < total_non_cc_elements; i++) {
490 layout_map[i][0] = e2c_vec[i].syn_ele;
491 layout_map[i][1] = e2c_vec[i].elem_id;
492 layout_map[i][2] = e2c_vec[i].aac_position;
493 }
494
495 return layout;
496 }
497
498 /**
499 * Save current output configuration if and only if it has been locked.
500 */
push_output_configuration(AACContext * ac)501 static int push_output_configuration(AACContext *ac) {
502 int pushed = 0;
503
504 if (ac->oc[1].status == OC_LOCKED || ac->oc[0].status == OC_NONE) {
505 ac->oc[0] = ac->oc[1];
506 pushed = 1;
507 }
508 ac->oc[1].status = OC_NONE;
509 return pushed;
510 }
511
512 /**
513 * Restore the previous output configuration if and only if the current
514 * configuration is unlocked.
515 */
pop_output_configuration(AACContext * ac)516 static void pop_output_configuration(AACContext *ac) {
517 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
518 ac->oc[1] = ac->oc[0];
519 ac->avctx->channels = ac->oc[1].channels;
520 ac->avctx->channel_layout = ac->oc[1].channel_layout;
521 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
522 ac->oc[1].status, 0);
523 }
524 }
525
526 /**
527 * Configure output channel order based on the current program
528 * configuration element.
529 *
530 * @return Returns error status. 0 - OK, !0 - error
531 */
output_configure(AACContext * ac,uint8_t layout_map[MAX_ELEM_ID * 4][3],int tags,enum OCStatus oc_type,int get_new_frame)532 static int output_configure(AACContext *ac,
533 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
534 enum OCStatus oc_type, int get_new_frame)
535 {
536 AVCodecContext *avctx = ac->avctx;
537 int i, channels = 0, ret;
538 uint64_t layout = 0;
539 uint8_t id_map[TYPE_END][MAX_ELEM_ID] = {{ 0 }};
540 uint8_t type_counts[TYPE_END] = { 0 };
541
542 if (ac->oc[1].layout_map != layout_map) {
543 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
544 ac->oc[1].layout_map_tags = tags;
545 }
546 for (i = 0; i < tags; i++) {
547 int type = layout_map[i][0];
548 int id = layout_map[i][1];
549 id_map[type][id] = type_counts[type]++;
550 if (id_map[type][id] >= MAX_ELEM_ID) {
551 avpriv_request_sample(ac->avctx, "Too large remapped id");
552 return AVERROR_PATCHWELCOME;
553 }
554 }
555 // Try to sniff a reasonable channel order, otherwise output the
556 // channels in the order the PCE declared them.
557 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
558 layout = sniff_channel_order(layout_map, tags);
559 for (i = 0; i < tags; i++) {
560 int type = layout_map[i][0];
561 int id = layout_map[i][1];
562 int iid = id_map[type][id];
563 int position = layout_map[i][2];
564 // Allocate or free elements depending on if they are in the
565 // current program configuration.
566 ret = che_configure(ac, position, type, iid, &channels);
567 if (ret < 0)
568 return ret;
569 ac->tag_che_map[type][id] = ac->che[type][iid];
570 }
571 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
572 if (layout == AV_CH_FRONT_CENTER) {
573 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
574 } else {
575 layout = 0;
576 }
577 }
578
579 if (layout) avctx->channel_layout = layout;
580 ac->oc[1].channel_layout = layout;
581 avctx->channels = ac->oc[1].channels = channels;
582 ac->oc[1].status = oc_type;
583
584 if (get_new_frame) {
585 if ((ret = frame_configure_elements(ac->avctx)) < 0)
586 return ret;
587 }
588
589 return 0;
590 }
591
flush(AVCodecContext * avctx)592 static void flush(AVCodecContext *avctx)
593 {
594 AACContext *ac= avctx->priv_data;
595 int type, i, j;
596
597 for (type = 3; type >= 0; type--) {
598 for (i = 0; i < MAX_ELEM_ID; i++) {
599 ChannelElement *che = ac->che[type][i];
600 if (che) {
601 for (j = 0; j <= 1; j++) {
602 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
603 }
604 }
605 }
606 }
607 }
608
609 /**
610 * Set up channel positions based on a default channel configuration
611 * as specified in table 1.17.
612 *
613 * @return Returns error status. 0 - OK, !0 - error
614 */
set_default_channel_config(AACContext * ac,AVCodecContext * avctx,uint8_t (* layout_map)[3],int * tags,int channel_config)615 static int set_default_channel_config(AACContext *ac, AVCodecContext *avctx,
616 uint8_t (*layout_map)[3],
617 int *tags,
618 int channel_config)
619 {
620 if (channel_config < 1 || (channel_config > 7 && channel_config < 11) ||
621 channel_config > 13) {
622 av_log(avctx, AV_LOG_ERROR,
623 "invalid default channel configuration (%d)\n",
624 channel_config);
625 return AVERROR_INVALIDDATA;
626 }
627 *tags = tags_per_config[channel_config];
628 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
629 *tags * sizeof(*layout_map));
630
631 /*
632 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
633 * However, at least Nero AAC encoder encodes 7.1 streams using the default
634 * channel config 7, mapping the side channels of the original audio stream
635 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
636 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
637 * the incorrect streams as if they were correct (and as the encoder intended).
638 *
639 * As actual intended 7.1(wide) streams are very rare, default to assuming a
640 * 7.1 layout was intended.
641 */
642 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
643 layout_map[2][2] = AAC_CHANNEL_SIDE;
644
645 if (!ac || !ac->warned_71_wide++) {
646 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
647 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
648 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
649 }
650 }
651
652 return 0;
653 }
654
get_che(AACContext * ac,int type,int elem_id)655 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
656 {
657 /* For PCE based channel configurations map the channels solely based
658 * on tags. */
659 if (!ac->oc[1].m4ac.chan_config) {
660 return ac->tag_che_map[type][elem_id];
661 }
662 // Allow single CPE stereo files to be signalled with mono configuration.
663 if (!ac->tags_mapped && type == TYPE_CPE &&
664 ac->oc[1].m4ac.chan_config == 1) {
665 uint8_t layout_map[MAX_ELEM_ID*4][3];
666 int layout_map_tags;
667 push_output_configuration(ac);
668
669 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
670
671 if (set_default_channel_config(ac, ac->avctx, layout_map,
672 &layout_map_tags, 2) < 0)
673 return NULL;
674 if (output_configure(ac, layout_map, layout_map_tags,
675 OC_TRIAL_FRAME, 1) < 0)
676 return NULL;
677
678 ac->oc[1].m4ac.chan_config = 2;
679 ac->oc[1].m4ac.ps = 0;
680 }
681 // And vice-versa
682 if (!ac->tags_mapped && type == TYPE_SCE &&
683 ac->oc[1].m4ac.chan_config == 2) {
684 uint8_t layout_map[MAX_ELEM_ID * 4][3];
685 int layout_map_tags;
686 push_output_configuration(ac);
687
688 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
689
690 if (set_default_channel_config(ac, ac->avctx, layout_map,
691 &layout_map_tags, 1) < 0)
692 return NULL;
693 if (output_configure(ac, layout_map, layout_map_tags,
694 OC_TRIAL_FRAME, 1) < 0)
695 return NULL;
696
697 ac->oc[1].m4ac.chan_config = 1;
698 if (ac->oc[1].m4ac.sbr)
699 ac->oc[1].m4ac.ps = -1;
700 }
701 /* For indexed channel configurations map the channels solely based
702 * on position. */
703 switch (ac->oc[1].m4ac.chan_config) {
704 case 13:
705 if (ac->tags_mapped > 3 && ((type == TYPE_CPE && elem_id < 8) ||
706 (type == TYPE_SCE && elem_id < 6) ||
707 (type == TYPE_LFE && elem_id < 2))) {
708 ac->tags_mapped++;
709 return ac->tag_che_map[type][elem_id] = ac->che[type][elem_id];
710 }
711 case 12:
712 case 7:
713 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
714 ac->tags_mapped++;
715 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
716 }
717 case 11:
718 if (ac->tags_mapped == 2 &&
719 ac->oc[1].m4ac.chan_config == 11 &&
720 type == TYPE_SCE) {
721 ac->tags_mapped++;
722 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
723 }
724 case 6:
725 /* Some streams incorrectly code 5.1 audio as
726 * SCE[0] CPE[0] CPE[1] SCE[1]
727 * instead of
728 * SCE[0] CPE[0] CPE[1] LFE[0].
729 * If we seem to have encountered such a stream, transfer
730 * the LFE[0] element to the SCE[1]'s mapping */
731 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
732 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
733 av_log(ac->avctx, AV_LOG_WARNING,
734 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
735 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
736 ac->warned_remapping_once++;
737 }
738 ac->tags_mapped++;
739 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
740 }
741 case 5:
742 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
743 ac->tags_mapped++;
744 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
745 }
746 case 4:
747 /* Some streams incorrectly code 4.0 audio as
748 * SCE[0] CPE[0] LFE[0]
749 * instead of
750 * SCE[0] CPE[0] SCE[1].
751 * If we seem to have encountered such a stream, transfer
752 * the SCE[1] element to the LFE[0]'s mapping */
753 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
754 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
755 av_log(ac->avctx, AV_LOG_WARNING,
756 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
757 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
758 ac->warned_remapping_once++;
759 }
760 ac->tags_mapped++;
761 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
762 }
763 if (ac->tags_mapped == 2 &&
764 ac->oc[1].m4ac.chan_config == 4 &&
765 type == TYPE_SCE) {
766 ac->tags_mapped++;
767 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
768 }
769 case 3:
770 case 2:
771 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
772 type == TYPE_CPE) {
773 ac->tags_mapped++;
774 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
775 } else if (ac->oc[1].m4ac.chan_config == 2) {
776 return NULL;
777 }
778 case 1:
779 if (!ac->tags_mapped && type == TYPE_SCE) {
780 ac->tags_mapped++;
781 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
782 }
783 default:
784 return NULL;
785 }
786 }
787
788 /**
789 * Decode an array of 4 bit element IDs, optionally interleaved with a
790 * stereo/mono switching bit.
791 *
792 * @param type speaker type/position for these channels
793 */
decode_channel_map(uint8_t layout_map[][3],enum ChannelPosition type,GetBitContext * gb,int n)794 static void decode_channel_map(uint8_t layout_map[][3],
795 enum ChannelPosition type,
796 GetBitContext *gb, int n)
797 {
798 while (n--) {
799 enum RawDataBlockType syn_ele;
800 switch (type) {
801 case AAC_CHANNEL_FRONT:
802 case AAC_CHANNEL_BACK:
803 case AAC_CHANNEL_SIDE:
804 syn_ele = get_bits1(gb);
805 break;
806 case AAC_CHANNEL_CC:
807 skip_bits1(gb);
808 syn_ele = TYPE_CCE;
809 break;
810 case AAC_CHANNEL_LFE:
811 syn_ele = TYPE_LFE;
812 break;
813 default:
814 // AAC_CHANNEL_OFF has no channel map
815 av_assert0(0);
816 }
817 layout_map[0][0] = syn_ele;
818 layout_map[0][1] = get_bits(gb, 4);
819 layout_map[0][2] = type;
820 layout_map++;
821 }
822 }
823
relative_align_get_bits(GetBitContext * gb,int reference_position)824 static inline void relative_align_get_bits(GetBitContext *gb,
825 int reference_position) {
826 int n = (reference_position - get_bits_count(gb) & 7);
827 if (n)
828 skip_bits(gb, n);
829 }
830
831 /**
832 * Decode program configuration element; reference: table 4.2.
833 *
834 * @return Returns error status. 0 - OK, !0 - error
835 */
decode_pce(AVCodecContext * avctx,MPEG4AudioConfig * m4ac,uint8_t (* layout_map)[3],GetBitContext * gb,int byte_align_ref)836 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
837 uint8_t (*layout_map)[3],
838 GetBitContext *gb, int byte_align_ref)
839 {
840 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
841 int sampling_index;
842 int comment_len;
843 int tags;
844
845 skip_bits(gb, 2); // object_type
846
847 sampling_index = get_bits(gb, 4);
848 if (m4ac->sampling_index != sampling_index)
849 av_log(avctx, AV_LOG_WARNING,
850 "Sample rate index in program config element does not "
851 "match the sample rate index configured by the container.\n");
852
853 num_front = get_bits(gb, 4);
854 num_side = get_bits(gb, 4);
855 num_back = get_bits(gb, 4);
856 num_lfe = get_bits(gb, 2);
857 num_assoc_data = get_bits(gb, 3);
858 num_cc = get_bits(gb, 4);
859
860 if (get_bits1(gb))
861 skip_bits(gb, 4); // mono_mixdown_tag
862 if (get_bits1(gb))
863 skip_bits(gb, 4); // stereo_mixdown_tag
864
865 if (get_bits1(gb))
866 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
867
868 if (get_bits_left(gb) < 5 * (num_front + num_side + num_back + num_cc) + 4 *(num_lfe + num_assoc_data + num_cc)) {
869 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
870 return -1;
871 }
872 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
873 tags = num_front;
874 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
875 tags += num_side;
876 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
877 tags += num_back;
878 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
879 tags += num_lfe;
880
881 skip_bits_long(gb, 4 * num_assoc_data);
882
883 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
884 tags += num_cc;
885
886 relative_align_get_bits(gb, byte_align_ref);
887
888 /* comment field, first byte is length */
889 comment_len = get_bits(gb, 8) * 8;
890 if (get_bits_left(gb) < comment_len) {
891 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
892 return AVERROR_INVALIDDATA;
893 }
894 skip_bits_long(gb, comment_len);
895 return tags;
896 }
897
898 /**
899 * Decode GA "General Audio" specific configuration; reference: table 4.1.
900 *
901 * @param ac pointer to AACContext, may be null
902 * @param avctx pointer to AVCCodecContext, used for logging
903 *
904 * @return Returns error status. 0 - OK, !0 - error
905 */
decode_ga_specific_config(AACContext * ac,AVCodecContext * avctx,GetBitContext * gb,int get_bit_alignment,MPEG4AudioConfig * m4ac,int channel_config)906 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
907 GetBitContext *gb,
908 int get_bit_alignment,
909 MPEG4AudioConfig *m4ac,
910 int channel_config)
911 {
912 int extension_flag, ret, ep_config, res_flags;
913 uint8_t layout_map[MAX_ELEM_ID*4][3];
914 int tags = 0;
915
916 #if USE_FIXED
917 if (get_bits1(gb)) { // frameLengthFlag
918 avpriv_report_missing_feature(avctx, "Fixed point 960/120 MDCT window");
919 return AVERROR_PATCHWELCOME;
920 }
921 m4ac->frame_length_short = 0;
922 #else
923 m4ac->frame_length_short = get_bits1(gb);
924 if (m4ac->frame_length_short && m4ac->sbr == 1) {
925 avpriv_report_missing_feature(avctx, "SBR with 960 frame length");
926 if (ac) ac->warned_960_sbr = 1;
927 m4ac->sbr = 0;
928 m4ac->ps = 0;
929 }
930 #endif
931
932 if (get_bits1(gb)) // dependsOnCoreCoder
933 skip_bits(gb, 14); // coreCoderDelay
934 extension_flag = get_bits1(gb);
935
936 if (m4ac->object_type == AOT_AAC_SCALABLE ||
937 m4ac->object_type == AOT_ER_AAC_SCALABLE)
938 skip_bits(gb, 3); // layerNr
939
940 if (channel_config == 0) {
941 skip_bits(gb, 4); // element_instance_tag
942 tags = decode_pce(avctx, m4ac, layout_map, gb, get_bit_alignment);
943 if (tags < 0)
944 return tags;
945 } else {
946 if ((ret = set_default_channel_config(ac, avctx, layout_map,
947 &tags, channel_config)))
948 return ret;
949 }
950
951 if (count_channels(layout_map, tags) > 1) {
952 m4ac->ps = 0;
953 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
954 m4ac->ps = 1;
955
956 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
957 return ret;
958
959 if (extension_flag) {
960 switch (m4ac->object_type) {
961 case AOT_ER_BSAC:
962 skip_bits(gb, 5); // numOfSubFrame
963 skip_bits(gb, 11); // layer_length
964 break;
965 case AOT_ER_AAC_LC:
966 case AOT_ER_AAC_LTP:
967 case AOT_ER_AAC_SCALABLE:
968 case AOT_ER_AAC_LD:
969 res_flags = get_bits(gb, 3);
970 if (res_flags) {
971 avpriv_report_missing_feature(avctx,
972 "AAC data resilience (flags %x)",
973 res_flags);
974 return AVERROR_PATCHWELCOME;
975 }
976 break;
977 }
978 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
979 }
980 switch (m4ac->object_type) {
981 case AOT_ER_AAC_LC:
982 case AOT_ER_AAC_LTP:
983 case AOT_ER_AAC_SCALABLE:
984 case AOT_ER_AAC_LD:
985 ep_config = get_bits(gb, 2);
986 if (ep_config) {
987 avpriv_report_missing_feature(avctx,
988 "epConfig %d", ep_config);
989 return AVERROR_PATCHWELCOME;
990 }
991 }
992 return 0;
993 }
994
decode_eld_specific_config(AACContext * ac,AVCodecContext * avctx,GetBitContext * gb,MPEG4AudioConfig * m4ac,int channel_config)995 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
996 GetBitContext *gb,
997 MPEG4AudioConfig *m4ac,
998 int channel_config)
999 {
1000 int ret, ep_config, res_flags;
1001 uint8_t layout_map[MAX_ELEM_ID*4][3];
1002 int tags = 0;
1003 const int ELDEXT_TERM = 0;
1004
1005 m4ac->ps = 0;
1006 m4ac->sbr = 0;
1007 #if USE_FIXED
1008 if (get_bits1(gb)) { // frameLengthFlag
1009 avpriv_request_sample(avctx, "960/120 MDCT window");
1010 return AVERROR_PATCHWELCOME;
1011 }
1012 #else
1013 m4ac->frame_length_short = get_bits1(gb);
1014 #endif
1015 res_flags = get_bits(gb, 3);
1016 if (res_flags) {
1017 avpriv_report_missing_feature(avctx,
1018 "AAC data resilience (flags %x)",
1019 res_flags);
1020 return AVERROR_PATCHWELCOME;
1021 }
1022
1023 if (get_bits1(gb)) { // ldSbrPresentFlag
1024 avpriv_report_missing_feature(avctx,
1025 "Low Delay SBR");
1026 return AVERROR_PATCHWELCOME;
1027 }
1028
1029 while (get_bits(gb, 4) != ELDEXT_TERM) {
1030 int len = get_bits(gb, 4);
1031 if (len == 15)
1032 len += get_bits(gb, 8);
1033 if (len == 15 + 255)
1034 len += get_bits(gb, 16);
1035 if (get_bits_left(gb) < len * 8 + 4) {
1036 av_log(avctx, AV_LOG_ERROR, overread_err);
1037 return AVERROR_INVALIDDATA;
1038 }
1039 skip_bits_long(gb, 8 * len);
1040 }
1041
1042 if ((ret = set_default_channel_config(ac, avctx, layout_map,
1043 &tags, channel_config)))
1044 return ret;
1045
1046 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
1047 return ret;
1048
1049 ep_config = get_bits(gb, 2);
1050 if (ep_config) {
1051 avpriv_report_missing_feature(avctx,
1052 "epConfig %d", ep_config);
1053 return AVERROR_PATCHWELCOME;
1054 }
1055 return 0;
1056 }
1057
1058 /**
1059 * Decode audio specific configuration; reference: table 1.13.
1060 *
1061 * @param ac pointer to AACContext, may be null
1062 * @param avctx pointer to AVCCodecContext, used for logging
1063 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
1064 * @param gb buffer holding an audio specific config
1065 * @param get_bit_alignment relative alignment for byte align operations
1066 * @param sync_extension look for an appended sync extension
1067 *
1068 * @return Returns error status or number of consumed bits. <0 - error
1069 */
decode_audio_specific_config_gb(AACContext * ac,AVCodecContext * avctx,MPEG4AudioConfig * m4ac,GetBitContext * gb,int get_bit_alignment,int sync_extension)1070 static int decode_audio_specific_config_gb(AACContext *ac,
1071 AVCodecContext *avctx,
1072 MPEG4AudioConfig *m4ac,
1073 GetBitContext *gb,
1074 int get_bit_alignment,
1075 int sync_extension)
1076 {
1077 int i, ret;
1078 GetBitContext gbc = *gb;
1079 MPEG4AudioConfig m4ac_bak = *m4ac;
1080
1081 if ((i = ff_mpeg4audio_get_config_gb(m4ac, &gbc, sync_extension, avctx)) < 0) {
1082 *m4ac = m4ac_bak;
1083 return AVERROR_INVALIDDATA;
1084 }
1085
1086 if (m4ac->sampling_index > 12) {
1087 av_log(avctx, AV_LOG_ERROR,
1088 "invalid sampling rate index %d\n",
1089 m4ac->sampling_index);
1090 *m4ac = m4ac_bak;
1091 return AVERROR_INVALIDDATA;
1092 }
1093 if (m4ac->object_type == AOT_ER_AAC_LD &&
1094 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
1095 av_log(avctx, AV_LOG_ERROR,
1096 "invalid low delay sampling rate index %d\n",
1097 m4ac->sampling_index);
1098 *m4ac = m4ac_bak;
1099 return AVERROR_INVALIDDATA;
1100 }
1101
1102 skip_bits_long(gb, i);
1103
1104 switch (m4ac->object_type) {
1105 case AOT_AAC_MAIN:
1106 case AOT_AAC_LC:
1107 case AOT_AAC_SSR:
1108 case AOT_AAC_LTP:
1109 case AOT_ER_AAC_LC:
1110 case AOT_ER_AAC_LD:
1111 if ((ret = decode_ga_specific_config(ac, avctx, gb, get_bit_alignment,
1112 m4ac, m4ac->chan_config)) < 0)
1113 return ret;
1114 break;
1115 case AOT_ER_AAC_ELD:
1116 if ((ret = decode_eld_specific_config(ac, avctx, gb,
1117 m4ac, m4ac->chan_config)) < 0)
1118 return ret;
1119 break;
1120 default:
1121 avpriv_report_missing_feature(avctx,
1122 "Audio object type %s%d",
1123 m4ac->sbr == 1 ? "SBR+" : "",
1124 m4ac->object_type);
1125 return AVERROR(ENOSYS);
1126 }
1127
1128 ff_dlog(avctx,
1129 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
1130 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1131 m4ac->sample_rate, m4ac->sbr,
1132 m4ac->ps);
1133
1134 return get_bits_count(gb);
1135 }
1136
decode_audio_specific_config(AACContext * ac,AVCodecContext * avctx,MPEG4AudioConfig * m4ac,const uint8_t * data,int64_t bit_size,int sync_extension)1137 static int decode_audio_specific_config(AACContext *ac,
1138 AVCodecContext *avctx,
1139 MPEG4AudioConfig *m4ac,
1140 const uint8_t *data, int64_t bit_size,
1141 int sync_extension)
1142 {
1143 int i, ret;
1144 GetBitContext gb;
1145
1146 if (bit_size < 0 || bit_size > INT_MAX) {
1147 av_log(avctx, AV_LOG_ERROR, "Audio specific config size is invalid\n");
1148 return AVERROR_INVALIDDATA;
1149 }
1150
1151 ff_dlog(avctx, "audio specific config size %d\n", (int)bit_size >> 3);
1152 for (i = 0; i < bit_size >> 3; i++)
1153 ff_dlog(avctx, "%02x ", data[i]);
1154 ff_dlog(avctx, "\n");
1155
1156 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
1157 return ret;
1158
1159 return decode_audio_specific_config_gb(ac, avctx, m4ac, &gb, 0,
1160 sync_extension);
1161 }
1162
1163 /**
1164 * linear congruential pseudorandom number generator
1165 *
1166 * @param previous_val pointer to the current state of the generator
1167 *
1168 * @return Returns a 32-bit pseudorandom integer
1169 */
lcg_random(unsigned previous_val)1170 static av_always_inline int lcg_random(unsigned previous_val)
1171 {
1172 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1173 return v.s;
1174 }
1175
reset_all_predictors(PredictorState * ps)1176 static void reset_all_predictors(PredictorState *ps)
1177 {
1178 int i;
1179 for (i = 0; i < MAX_PREDICTORS; i++)
1180 reset_predict_state(&ps[i]);
1181 }
1182
sample_rate_idx(int rate)1183 static int sample_rate_idx (int rate)
1184 {
1185 if (92017 <= rate) return 0;
1186 else if (75132 <= rate) return 1;
1187 else if (55426 <= rate) return 2;
1188 else if (46009 <= rate) return 3;
1189 else if (37566 <= rate) return 4;
1190 else if (27713 <= rate) return 5;
1191 else if (23004 <= rate) return 6;
1192 else if (18783 <= rate) return 7;
1193 else if (13856 <= rate) return 8;
1194 else if (11502 <= rate) return 9;
1195 else if (9391 <= rate) return 10;
1196 else return 11;
1197 }
1198
reset_predictor_group(PredictorState * ps,int group_num)1199 static void reset_predictor_group(PredictorState *ps, int group_num)
1200 {
1201 int i;
1202 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1203 reset_predict_state(&ps[i]);
1204 }
1205
1206 static void aacdec_init(AACContext *ac);
1207
aac_static_table_init(void)1208 static av_cold void aac_static_table_init(void)
1209 {
1210 static VLC_TYPE vlc_buf[304 + 270 + 550 + 300 + 328 +
1211 294 + 306 + 268 + 510 + 366 + 462][2];
1212 for (unsigned i = 0, offset = 0; i < 11; i++) {
1213 vlc_spectral[i].table = &vlc_buf[offset];
1214 vlc_spectral[i].table_allocated = FF_ARRAY_ELEMS(vlc_buf) - offset;
1215 ff_init_vlc_sparse(&vlc_spectral[i], 8, ff_aac_spectral_sizes[i],
1216 ff_aac_spectral_bits[i], sizeof(ff_aac_spectral_bits[i][0]),
1217 sizeof(ff_aac_spectral_bits[i][0]),
1218 ff_aac_spectral_codes[i], sizeof(ff_aac_spectral_codes[i][0]),
1219 sizeof(ff_aac_spectral_codes[i][0]),
1220 ff_aac_codebook_vector_idx[i], sizeof(ff_aac_codebook_vector_idx[i][0]),
1221 sizeof(ff_aac_codebook_vector_idx[i][0]),
1222 INIT_VLC_STATIC_OVERLONG);
1223 offset += vlc_spectral[i].table_size;
1224 }
1225
1226 AAC_RENAME(ff_aac_sbr_init)();
1227
1228 ff_aac_tableinit();
1229
1230 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1231 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1232 ff_aac_scalefactor_bits,
1233 sizeof(ff_aac_scalefactor_bits[0]),
1234 sizeof(ff_aac_scalefactor_bits[0]),
1235 ff_aac_scalefactor_code,
1236 sizeof(ff_aac_scalefactor_code[0]),
1237 sizeof(ff_aac_scalefactor_code[0]),
1238 352);
1239
1240 // window initialization
1241 #if !USE_FIXED
1242 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(aac_kbd_long_960), 4.0, 960);
1243 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(aac_kbd_short_120), 6.0, 120);
1244 AAC_RENAME(ff_sine_window_init)(AAC_RENAME(sine_960), 960);
1245 AAC_RENAME(ff_sine_window_init)(AAC_RENAME(sine_120), 120);
1246 AAC_RENAME(ff_init_ff_sine_windows)(9);
1247 ff_aac_float_common_init();
1248 #else
1249 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME2(aac_kbd_long_1024), 4.0, 1024);
1250 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME2(aac_kbd_short_128), 6.0, 128);
1251 init_sine_windows_fixed();
1252 #endif
1253
1254 AAC_RENAME(ff_cbrt_tableinit)();
1255 }
1256
1257 static AVOnce aac_table_init = AV_ONCE_INIT;
1258
aac_decode_init(AVCodecContext * avctx)1259 static av_cold int aac_decode_init(AVCodecContext *avctx)
1260 {
1261 AACContext *ac = avctx->priv_data;
1262 int ret;
1263
1264 if (avctx->sample_rate > 96000)
1265 return AVERROR_INVALIDDATA;
1266
1267 ret = ff_thread_once(&aac_table_init, &aac_static_table_init);
1268 if (ret != 0)
1269 return AVERROR_UNKNOWN;
1270
1271 ac->avctx = avctx;
1272 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1273
1274 aacdec_init(ac);
1275 #if USE_FIXED
1276 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1277 #else
1278 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1279 #endif /* USE_FIXED */
1280
1281 if (avctx->extradata_size > 0) {
1282 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1283 avctx->extradata,
1284 avctx->extradata_size * 8LL,
1285 1)) < 0)
1286 return ret;
1287 } else {
1288 int sr, i;
1289 uint8_t layout_map[MAX_ELEM_ID*4][3];
1290 int layout_map_tags;
1291
1292 sr = sample_rate_idx(avctx->sample_rate);
1293 ac->oc[1].m4ac.sampling_index = sr;
1294 ac->oc[1].m4ac.channels = avctx->channels;
1295 ac->oc[1].m4ac.sbr = -1;
1296 ac->oc[1].m4ac.ps = -1;
1297
1298 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1299 if (ff_mpeg4audio_channels[i] == avctx->channels)
1300 break;
1301 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1302 i = 0;
1303 }
1304 ac->oc[1].m4ac.chan_config = i;
1305
1306 if (ac->oc[1].m4ac.chan_config) {
1307 int ret = set_default_channel_config(ac, avctx, layout_map,
1308 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1309 if (!ret)
1310 output_configure(ac, layout_map, layout_map_tags,
1311 OC_GLOBAL_HDR, 0);
1312 else if (avctx->err_recognition & AV_EF_EXPLODE)
1313 return AVERROR_INVALIDDATA;
1314 }
1315 }
1316
1317 if (avctx->channels > MAX_CHANNELS) {
1318 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1319 return AVERROR_INVALIDDATA;
1320 }
1321
1322 #if USE_FIXED
1323 ac->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1324 #else
1325 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1326 #endif /* USE_FIXED */
1327 if (!ac->fdsp) {
1328 return AVERROR(ENOMEM);
1329 }
1330
1331 ac->random_state = 0x1f2e3d4c;
1332
1333 AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
1334 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
1335 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
1336 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
1337 #if !USE_FIXED
1338 ret = ff_mdct15_init(&ac->mdct120, 1, 3, 1.0f/(16*1024*120*2));
1339 if (ret < 0)
1340 return ret;
1341 ret = ff_mdct15_init(&ac->mdct480, 1, 5, 1.0f/(16*1024*960));
1342 if (ret < 0)
1343 return ret;
1344 ret = ff_mdct15_init(&ac->mdct960, 1, 6, 1.0f/(16*1024*960*2));
1345 if (ret < 0)
1346 return ret;
1347 #endif
1348
1349 return 0;
1350 }
1351
1352 /**
1353 * Skip data_stream_element; reference: table 4.10.
1354 */
skip_data_stream_element(AACContext * ac,GetBitContext * gb)1355 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1356 {
1357 int byte_align = get_bits1(gb);
1358 int count = get_bits(gb, 8);
1359 if (count == 255)
1360 count += get_bits(gb, 8);
1361 if (byte_align)
1362 align_get_bits(gb);
1363
1364 if (get_bits_left(gb) < 8 * count) {
1365 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1366 return AVERROR_INVALIDDATA;
1367 }
1368 skip_bits_long(gb, 8 * count);
1369 return 0;
1370 }
1371
decode_prediction(AACContext * ac,IndividualChannelStream * ics,GetBitContext * gb)1372 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1373 GetBitContext *gb)
1374 {
1375 int sfb;
1376 if (get_bits1(gb)) {
1377 ics->predictor_reset_group = get_bits(gb, 5);
1378 if (ics->predictor_reset_group == 0 ||
1379 ics->predictor_reset_group > 30) {
1380 av_log(ac->avctx, AV_LOG_ERROR,
1381 "Invalid Predictor Reset Group.\n");
1382 return AVERROR_INVALIDDATA;
1383 }
1384 }
1385 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1386 ics->prediction_used[sfb] = get_bits1(gb);
1387 }
1388 return 0;
1389 }
1390
1391 /**
1392 * Decode Long Term Prediction data; reference: table 4.xx.
1393 */
decode_ltp(LongTermPrediction * ltp,GetBitContext * gb,uint8_t max_sfb)1394 static void decode_ltp(LongTermPrediction *ltp,
1395 GetBitContext *gb, uint8_t max_sfb)
1396 {
1397 int sfb;
1398
1399 ltp->lag = get_bits(gb, 11);
1400 ltp->coef = ltp_coef[get_bits(gb, 3)];
1401 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1402 ltp->used[sfb] = get_bits1(gb);
1403 }
1404
1405 /**
1406 * Decode Individual Channel Stream info; reference: table 4.6.
1407 */
decode_ics_info(AACContext * ac,IndividualChannelStream * ics,GetBitContext * gb)1408 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1409 GetBitContext *gb)
1410 {
1411 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1412 const int aot = m4ac->object_type;
1413 const int sampling_index = m4ac->sampling_index;
1414 int ret_fail = AVERROR_INVALIDDATA;
1415
1416 if (aot != AOT_ER_AAC_ELD) {
1417 if (get_bits1(gb)) {
1418 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1419 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1420 return AVERROR_INVALIDDATA;
1421 }
1422 ics->window_sequence[1] = ics->window_sequence[0];
1423 ics->window_sequence[0] = get_bits(gb, 2);
1424 if (aot == AOT_ER_AAC_LD &&
1425 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1426 av_log(ac->avctx, AV_LOG_ERROR,
1427 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1428 "window sequence %d found.\n", ics->window_sequence[0]);
1429 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1430 return AVERROR_INVALIDDATA;
1431 }
1432 ics->use_kb_window[1] = ics->use_kb_window[0];
1433 ics->use_kb_window[0] = get_bits1(gb);
1434 }
1435 ics->num_window_groups = 1;
1436 ics->group_len[0] = 1;
1437 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1438 int i;
1439 ics->max_sfb = get_bits(gb, 4);
1440 for (i = 0; i < 7; i++) {
1441 if (get_bits1(gb)) {
1442 ics->group_len[ics->num_window_groups - 1]++;
1443 } else {
1444 ics->num_window_groups++;
1445 ics->group_len[ics->num_window_groups - 1] = 1;
1446 }
1447 }
1448 ics->num_windows = 8;
1449 if (m4ac->frame_length_short) {
1450 ics->swb_offset = ff_swb_offset_120[sampling_index];
1451 ics->num_swb = ff_aac_num_swb_120[sampling_index];
1452 } else {
1453 ics->swb_offset = ff_swb_offset_128[sampling_index];
1454 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1455 }
1456 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1457 ics->predictor_present = 0;
1458 } else {
1459 ics->max_sfb = get_bits(gb, 6);
1460 ics->num_windows = 1;
1461 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1462 if (m4ac->frame_length_short) {
1463 ics->swb_offset = ff_swb_offset_480[sampling_index];
1464 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1465 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1466 } else {
1467 ics->swb_offset = ff_swb_offset_512[sampling_index];
1468 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1469 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1470 }
1471 if (!ics->num_swb || !ics->swb_offset) {
1472 ret_fail = AVERROR_BUG;
1473 goto fail;
1474 }
1475 } else {
1476 if (m4ac->frame_length_short) {
1477 ics->num_swb = ff_aac_num_swb_960[sampling_index];
1478 ics->swb_offset = ff_swb_offset_960[sampling_index];
1479 } else {
1480 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1481 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1482 }
1483 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1484 }
1485 if (aot != AOT_ER_AAC_ELD) {
1486 ics->predictor_present = get_bits1(gb);
1487 ics->predictor_reset_group = 0;
1488 }
1489 if (ics->predictor_present) {
1490 if (aot == AOT_AAC_MAIN) {
1491 if (decode_prediction(ac, ics, gb)) {
1492 goto fail;
1493 }
1494 } else if (aot == AOT_AAC_LC ||
1495 aot == AOT_ER_AAC_LC) {
1496 av_log(ac->avctx, AV_LOG_ERROR,
1497 "Prediction is not allowed in AAC-LC.\n");
1498 goto fail;
1499 } else {
1500 if (aot == AOT_ER_AAC_LD) {
1501 av_log(ac->avctx, AV_LOG_ERROR,
1502 "LTP in ER AAC LD not yet implemented.\n");
1503 ret_fail = AVERROR_PATCHWELCOME;
1504 goto fail;
1505 }
1506 if ((ics->ltp.present = get_bits(gb, 1)))
1507 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1508 }
1509 }
1510 }
1511
1512 if (ics->max_sfb > ics->num_swb) {
1513 av_log(ac->avctx, AV_LOG_ERROR,
1514 "Number of scalefactor bands in group (%d) "
1515 "exceeds limit (%d).\n",
1516 ics->max_sfb, ics->num_swb);
1517 goto fail;
1518 }
1519
1520 return 0;
1521 fail:
1522 ics->max_sfb = 0;
1523 return ret_fail;
1524 }
1525
1526 /**
1527 * Decode band types (section_data payload); reference: table 4.46.
1528 *
1529 * @param band_type array of the used band type
1530 * @param band_type_run_end array of the last scalefactor band of a band type run
1531 *
1532 * @return Returns error status. 0 - OK, !0 - error
1533 */
decode_band_types(AACContext * ac,enum BandType band_type[120],int band_type_run_end[120],GetBitContext * gb,IndividualChannelStream * ics)1534 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1535 int band_type_run_end[120], GetBitContext *gb,
1536 IndividualChannelStream *ics)
1537 {
1538 int g, idx = 0;
1539 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1540 for (g = 0; g < ics->num_window_groups; g++) {
1541 int k = 0;
1542 while (k < ics->max_sfb) {
1543 uint8_t sect_end = k;
1544 int sect_len_incr;
1545 int sect_band_type = get_bits(gb, 4);
1546 if (sect_band_type == 12) {
1547 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1548 return AVERROR_INVALIDDATA;
1549 }
1550 do {
1551 sect_len_incr = get_bits(gb, bits);
1552 sect_end += sect_len_incr;
1553 if (get_bits_left(gb) < 0) {
1554 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1555 return AVERROR_INVALIDDATA;
1556 }
1557 if (sect_end > ics->max_sfb) {
1558 av_log(ac->avctx, AV_LOG_ERROR,
1559 "Number of bands (%d) exceeds limit (%d).\n",
1560 sect_end, ics->max_sfb);
1561 return AVERROR_INVALIDDATA;
1562 }
1563 } while (sect_len_incr == (1 << bits) - 1);
1564 for (; k < sect_end; k++) {
1565 band_type [idx] = sect_band_type;
1566 band_type_run_end[idx++] = sect_end;
1567 }
1568 }
1569 }
1570 return 0;
1571 }
1572
1573 /**
1574 * Decode scalefactors; reference: table 4.47.
1575 *
1576 * @param global_gain first scalefactor value as scalefactors are differentially coded
1577 * @param band_type array of the used band type
1578 * @param band_type_run_end array of the last scalefactor band of a band type run
1579 * @param sf array of scalefactors or intensity stereo positions
1580 *
1581 * @return Returns error status. 0 - OK, !0 - error
1582 */
decode_scalefactors(AACContext * ac,INTFLOAT sf[120],GetBitContext * gb,unsigned int global_gain,IndividualChannelStream * ics,enum BandType band_type[120],int band_type_run_end[120])1583 static int decode_scalefactors(AACContext *ac, INTFLOAT sf[120], GetBitContext *gb,
1584 unsigned int global_gain,
1585 IndividualChannelStream *ics,
1586 enum BandType band_type[120],
1587 int band_type_run_end[120])
1588 {
1589 int g, i, idx = 0;
1590 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1591 int clipped_offset;
1592 int noise_flag = 1;
1593 for (g = 0; g < ics->num_window_groups; g++) {
1594 for (i = 0; i < ics->max_sfb;) {
1595 int run_end = band_type_run_end[idx];
1596 if (band_type[idx] == ZERO_BT) {
1597 for (; i < run_end; i++, idx++)
1598 sf[idx] = FIXR(0.);
1599 } else if ((band_type[idx] == INTENSITY_BT) ||
1600 (band_type[idx] == INTENSITY_BT2)) {
1601 for (; i < run_end; i++, idx++) {
1602 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1603 clipped_offset = av_clip(offset[2], -155, 100);
1604 if (offset[2] != clipped_offset) {
1605 avpriv_request_sample(ac->avctx,
1606 "If you heard an audible artifact, there may be a bug in the decoder. "
1607 "Clipped intensity stereo position (%d -> %d)",
1608 offset[2], clipped_offset);
1609 }
1610 #if USE_FIXED
1611 sf[idx] = 100 - clipped_offset;
1612 #else
1613 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1614 #endif /* USE_FIXED */
1615 }
1616 } else if (band_type[idx] == NOISE_BT) {
1617 for (; i < run_end; i++, idx++) {
1618 if (noise_flag-- > 0)
1619 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1620 else
1621 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1622 clipped_offset = av_clip(offset[1], -100, 155);
1623 if (offset[1] != clipped_offset) {
1624 avpriv_request_sample(ac->avctx,
1625 "If you heard an audible artifact, there may be a bug in the decoder. "
1626 "Clipped noise gain (%d -> %d)",
1627 offset[1], clipped_offset);
1628 }
1629 #if USE_FIXED
1630 sf[idx] = -(100 + clipped_offset);
1631 #else
1632 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1633 #endif /* USE_FIXED */
1634 }
1635 } else {
1636 for (; i < run_end; i++, idx++) {
1637 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1638 if (offset[0] > 255U) {
1639 av_log(ac->avctx, AV_LOG_ERROR,
1640 "Scalefactor (%d) out of range.\n", offset[0]);
1641 return AVERROR_INVALIDDATA;
1642 }
1643 #if USE_FIXED
1644 sf[idx] = -offset[0];
1645 #else
1646 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1647 #endif /* USE_FIXED */
1648 }
1649 }
1650 }
1651 }
1652 return 0;
1653 }
1654
1655 /**
1656 * Decode pulse data; reference: table 4.7.
1657 */
decode_pulses(Pulse * pulse,GetBitContext * gb,const uint16_t * swb_offset,int num_swb)1658 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1659 const uint16_t *swb_offset, int num_swb)
1660 {
1661 int i, pulse_swb;
1662 pulse->num_pulse = get_bits(gb, 2) + 1;
1663 pulse_swb = get_bits(gb, 6);
1664 if (pulse_swb >= num_swb)
1665 return -1;
1666 pulse->pos[0] = swb_offset[pulse_swb];
1667 pulse->pos[0] += get_bits(gb, 5);
1668 if (pulse->pos[0] >= swb_offset[num_swb])
1669 return -1;
1670 pulse->amp[0] = get_bits(gb, 4);
1671 for (i = 1; i < pulse->num_pulse; i++) {
1672 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1673 if (pulse->pos[i] >= swb_offset[num_swb])
1674 return -1;
1675 pulse->amp[i] = get_bits(gb, 4);
1676 }
1677 return 0;
1678 }
1679
1680 /**
1681 * Decode Temporal Noise Shaping data; reference: table 4.48.
1682 *
1683 * @return Returns error status. 0 - OK, !0 - error
1684 */
decode_tns(AACContext * ac,TemporalNoiseShaping * tns,GetBitContext * gb,const IndividualChannelStream * ics)1685 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1686 GetBitContext *gb, const IndividualChannelStream *ics)
1687 {
1688 int w, filt, i, coef_len, coef_res, coef_compress;
1689 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1690 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1691 for (w = 0; w < ics->num_windows; w++) {
1692 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1693 coef_res = get_bits1(gb);
1694
1695 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1696 int tmp2_idx;
1697 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1698
1699 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1700 av_log(ac->avctx, AV_LOG_ERROR,
1701 "TNS filter order %d is greater than maximum %d.\n",
1702 tns->order[w][filt], tns_max_order);
1703 tns->order[w][filt] = 0;
1704 return AVERROR_INVALIDDATA;
1705 }
1706 if (tns->order[w][filt]) {
1707 tns->direction[w][filt] = get_bits1(gb);
1708 coef_compress = get_bits1(gb);
1709 coef_len = coef_res + 3 - coef_compress;
1710 tmp2_idx = 2 * coef_compress + coef_res;
1711
1712 for (i = 0; i < tns->order[w][filt]; i++)
1713 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1714 }
1715 }
1716 }
1717 }
1718 return 0;
1719 }
1720
1721 /**
1722 * Decode Mid/Side data; reference: table 4.54.
1723 *
1724 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1725 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1726 * [3] reserved for scalable AAC
1727 */
decode_mid_side_stereo(ChannelElement * cpe,GetBitContext * gb,int ms_present)1728 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1729 int ms_present)
1730 {
1731 int idx;
1732 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1733 if (ms_present == 1) {
1734 for (idx = 0; idx < max_idx; idx++)
1735 cpe->ms_mask[idx] = get_bits1(gb);
1736 } else if (ms_present == 2) {
1737 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1738 }
1739 }
1740
1741 /**
1742 * Decode spectral data; reference: table 4.50.
1743 * Dequantize and scale spectral data; reference: 4.6.3.3.
1744 *
1745 * @param coef array of dequantized, scaled spectral data
1746 * @param sf array of scalefactors or intensity stereo positions
1747 * @param pulse_present set if pulses are present
1748 * @param pulse pointer to pulse data struct
1749 * @param band_type array of the used band type
1750 *
1751 * @return Returns error status. 0 - OK, !0 - error
1752 */
decode_spectrum_and_dequant(AACContext * ac,INTFLOAT coef[1024],GetBitContext * gb,const INTFLOAT sf[120],int pulse_present,const Pulse * pulse,const IndividualChannelStream * ics,enum BandType band_type[120])1753 static int decode_spectrum_and_dequant(AACContext *ac, INTFLOAT coef[1024],
1754 GetBitContext *gb, const INTFLOAT sf[120],
1755 int pulse_present, const Pulse *pulse,
1756 const IndividualChannelStream *ics,
1757 enum BandType band_type[120])
1758 {
1759 int i, k, g, idx = 0;
1760 const int c = 1024 / ics->num_windows;
1761 const uint16_t *offsets = ics->swb_offset;
1762 INTFLOAT *coef_base = coef;
1763
1764 for (g = 0; g < ics->num_windows; g++)
1765 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1766 sizeof(INTFLOAT) * (c - offsets[ics->max_sfb]));
1767
1768 for (g = 0; g < ics->num_window_groups; g++) {
1769 unsigned g_len = ics->group_len[g];
1770
1771 for (i = 0; i < ics->max_sfb; i++, idx++) {
1772 const unsigned cbt_m1 = band_type[idx] - 1;
1773 INTFLOAT *cfo = coef + offsets[i];
1774 int off_len = offsets[i + 1] - offsets[i];
1775 int group;
1776
1777 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1778 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1779 memset(cfo, 0, off_len * sizeof(*cfo));
1780 }
1781 } else if (cbt_m1 == NOISE_BT - 1) {
1782 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1783 INTFLOAT band_energy;
1784 #if USE_FIXED
1785 for (k = 0; k < off_len; k++) {
1786 ac->random_state = lcg_random(ac->random_state);
1787 cfo[k] = ac->random_state >> 3;
1788 }
1789
1790 band_energy = ac->fdsp->scalarproduct_fixed(cfo, cfo, off_len);
1791 band_energy = fixed_sqrt(band_energy, 31);
1792 noise_scale(cfo, sf[idx], band_energy, off_len);
1793 #else
1794 float scale;
1795
1796 for (k = 0; k < off_len; k++) {
1797 ac->random_state = lcg_random(ac->random_state);
1798 cfo[k] = ac->random_state;
1799 }
1800
1801 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1802 scale = sf[idx] / sqrtf(band_energy);
1803 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1804 #endif /* USE_FIXED */
1805 }
1806 } else {
1807 #if !USE_FIXED
1808 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1809 #endif /* !USE_FIXED */
1810 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1811 OPEN_READER(re, gb);
1812
1813 switch (cbt_m1 >> 1) {
1814 case 0:
1815 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1816 INTFLOAT *cf = cfo;
1817 int len = off_len;
1818
1819 do {
1820 int code;
1821 unsigned cb_idx;
1822
1823 UPDATE_CACHE(re, gb);
1824 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1825 cb_idx = code;
1826 #if USE_FIXED
1827 cf = DEC_SQUAD(cf, cb_idx);
1828 #else
1829 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1830 #endif /* USE_FIXED */
1831 } while (len -= 4);
1832 }
1833 break;
1834
1835 case 1:
1836 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1837 INTFLOAT *cf = cfo;
1838 int len = off_len;
1839
1840 do {
1841 int code;
1842 unsigned nnz;
1843 unsigned cb_idx;
1844 uint32_t bits;
1845
1846 UPDATE_CACHE(re, gb);
1847 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1848 cb_idx = code;
1849 nnz = cb_idx >> 8 & 15;
1850 bits = nnz ? GET_CACHE(re, gb) : 0;
1851 LAST_SKIP_BITS(re, gb, nnz);
1852 #if USE_FIXED
1853 cf = DEC_UQUAD(cf, cb_idx, bits);
1854 #else
1855 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1856 #endif /* USE_FIXED */
1857 } while (len -= 4);
1858 }
1859 break;
1860
1861 case 2:
1862 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1863 INTFLOAT *cf = cfo;
1864 int len = off_len;
1865
1866 do {
1867 int code;
1868 unsigned cb_idx;
1869
1870 UPDATE_CACHE(re, gb);
1871 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1872 cb_idx = code;
1873 #if USE_FIXED
1874 cf = DEC_SPAIR(cf, cb_idx);
1875 #else
1876 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1877 #endif /* USE_FIXED */
1878 } while (len -= 2);
1879 }
1880 break;
1881
1882 case 3:
1883 case 4:
1884 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1885 INTFLOAT *cf = cfo;
1886 int len = off_len;
1887
1888 do {
1889 int code;
1890 unsigned nnz;
1891 unsigned cb_idx;
1892 unsigned sign;
1893
1894 UPDATE_CACHE(re, gb);
1895 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1896 cb_idx = code;
1897 nnz = cb_idx >> 8 & 15;
1898 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1899 LAST_SKIP_BITS(re, gb, nnz);
1900 #if USE_FIXED
1901 cf = DEC_UPAIR(cf, cb_idx, sign);
1902 #else
1903 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1904 #endif /* USE_FIXED */
1905 } while (len -= 2);
1906 }
1907 break;
1908
1909 default:
1910 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1911 #if USE_FIXED
1912 int *icf = cfo;
1913 int v;
1914 #else
1915 float *cf = cfo;
1916 uint32_t *icf = (uint32_t *) cf;
1917 #endif /* USE_FIXED */
1918 int len = off_len;
1919
1920 do {
1921 int code;
1922 unsigned nzt, nnz;
1923 unsigned cb_idx;
1924 uint32_t bits;
1925 int j;
1926
1927 UPDATE_CACHE(re, gb);
1928 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1929 cb_idx = code;
1930
1931 if (cb_idx == 0x0000) {
1932 *icf++ = 0;
1933 *icf++ = 0;
1934 continue;
1935 }
1936
1937 nnz = cb_idx >> 12;
1938 nzt = cb_idx >> 8;
1939 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1940 LAST_SKIP_BITS(re, gb, nnz);
1941
1942 for (j = 0; j < 2; j++) {
1943 if (nzt & 1<<j) {
1944 uint32_t b;
1945 int n;
1946 /* The total length of escape_sequence must be < 22 bits according
1947 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1948 UPDATE_CACHE(re, gb);
1949 b = GET_CACHE(re, gb);
1950 b = 31 - av_log2(~b);
1951
1952 if (b > 8) {
1953 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1954 return AVERROR_INVALIDDATA;
1955 }
1956
1957 SKIP_BITS(re, gb, b + 1);
1958 b += 4;
1959 n = (1 << b) + SHOW_UBITS(re, gb, b);
1960 LAST_SKIP_BITS(re, gb, b);
1961 #if USE_FIXED
1962 v = n;
1963 if (bits & 1U<<31)
1964 v = -v;
1965 *icf++ = v;
1966 #else
1967 *icf++ = ff_cbrt_tab[n] | (bits & 1U<<31);
1968 #endif /* USE_FIXED */
1969 bits <<= 1;
1970 } else {
1971 #if USE_FIXED
1972 v = cb_idx & 15;
1973 if (bits & 1U<<31)
1974 v = -v;
1975 *icf++ = v;
1976 #else
1977 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1978 *icf++ = (bits & 1U<<31) | v;
1979 #endif /* USE_FIXED */
1980 bits <<= !!v;
1981 }
1982 cb_idx >>= 4;
1983 }
1984 } while (len -= 2);
1985 #if !USE_FIXED
1986 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1987 #endif /* !USE_FIXED */
1988 }
1989 }
1990
1991 CLOSE_READER(re, gb);
1992 }
1993 }
1994 coef += g_len << 7;
1995 }
1996
1997 if (pulse_present) {
1998 idx = 0;
1999 for (i = 0; i < pulse->num_pulse; i++) {
2000 INTFLOAT co = coef_base[ pulse->pos[i] ];
2001 while (offsets[idx + 1] <= pulse->pos[i])
2002 idx++;
2003 if (band_type[idx] != NOISE_BT && sf[idx]) {
2004 INTFLOAT ico = -pulse->amp[i];
2005 #if USE_FIXED
2006 if (co) {
2007 ico = co + (co > 0 ? -ico : ico);
2008 }
2009 coef_base[ pulse->pos[i] ] = ico;
2010 #else
2011 if (co) {
2012 co /= sf[idx];
2013 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
2014 }
2015 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
2016 #endif /* USE_FIXED */
2017 }
2018 }
2019 }
2020 #if USE_FIXED
2021 coef = coef_base;
2022 idx = 0;
2023 for (g = 0; g < ics->num_window_groups; g++) {
2024 unsigned g_len = ics->group_len[g];
2025
2026 for (i = 0; i < ics->max_sfb; i++, idx++) {
2027 const unsigned cbt_m1 = band_type[idx] - 1;
2028 int *cfo = coef + offsets[i];
2029 int off_len = offsets[i + 1] - offsets[i];
2030 int group;
2031
2032 if (cbt_m1 < NOISE_BT - 1) {
2033 for (group = 0; group < (int)g_len; group++, cfo+=128) {
2034 ac->vector_pow43(cfo, off_len);
2035 ac->subband_scale(cfo, cfo, sf[idx], 34, off_len, ac->avctx);
2036 }
2037 }
2038 }
2039 coef += g_len << 7;
2040 }
2041 #endif /* USE_FIXED */
2042 return 0;
2043 }
2044
2045 /**
2046 * Apply AAC-Main style frequency domain prediction.
2047 */
apply_prediction(AACContext * ac,SingleChannelElement * sce)2048 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
2049 {
2050 int sfb, k;
2051
2052 if (!sce->ics.predictor_initialized) {
2053 reset_all_predictors(sce->predictor_state);
2054 sce->ics.predictor_initialized = 1;
2055 }
2056
2057 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2058 for (sfb = 0;
2059 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
2060 sfb++) {
2061 for (k = sce->ics.swb_offset[sfb];
2062 k < sce->ics.swb_offset[sfb + 1];
2063 k++) {
2064 predict(&sce->predictor_state[k], &sce->coeffs[k],
2065 sce->ics.predictor_present &&
2066 sce->ics.prediction_used[sfb]);
2067 }
2068 }
2069 if (sce->ics.predictor_reset_group)
2070 reset_predictor_group(sce->predictor_state,
2071 sce->ics.predictor_reset_group);
2072 } else
2073 reset_all_predictors(sce->predictor_state);
2074 }
2075
decode_gain_control(SingleChannelElement * sce,GetBitContext * gb)2076 static void decode_gain_control(SingleChannelElement * sce, GetBitContext * gb)
2077 {
2078 // wd_num, wd_test, aloc_size
2079 static const uint8_t gain_mode[4][3] = {
2080 {1, 0, 5}, // ONLY_LONG_SEQUENCE = 0,
2081 {2, 1, 2}, // LONG_START_SEQUENCE,
2082 {8, 0, 2}, // EIGHT_SHORT_SEQUENCE,
2083 {2, 1, 5}, // LONG_STOP_SEQUENCE
2084 };
2085
2086 const int mode = sce->ics.window_sequence[0];
2087 uint8_t bd, wd, ad;
2088
2089 // FIXME: Store the gain control data on |sce| and do something with it.
2090 uint8_t max_band = get_bits(gb, 2);
2091 for (bd = 0; bd < max_band; bd++) {
2092 for (wd = 0; wd < gain_mode[mode][0]; wd++) {
2093 uint8_t adjust_num = get_bits(gb, 3);
2094 for (ad = 0; ad < adjust_num; ad++) {
2095 skip_bits(gb, 4 + ((wd == 0 && gain_mode[mode][1])
2096 ? 4
2097 : gain_mode[mode][2]));
2098 }
2099 }
2100 }
2101 }
2102
2103 /**
2104 * Decode an individual_channel_stream payload; reference: table 4.44.
2105 *
2106 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
2107 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
2108 *
2109 * @return Returns error status. 0 - OK, !0 - error
2110 */
decode_ics(AACContext * ac,SingleChannelElement * sce,GetBitContext * gb,int common_window,int scale_flag)2111 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
2112 GetBitContext *gb, int common_window, int scale_flag)
2113 {
2114 Pulse pulse;
2115 TemporalNoiseShaping *tns = &sce->tns;
2116 IndividualChannelStream *ics = &sce->ics;
2117 INTFLOAT *out = sce->coeffs;
2118 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
2119 int ret;
2120
2121 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2122 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
2123 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
2124 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
2125 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2126
2127 /* This assignment is to silence a GCC warning about the variable being used
2128 * uninitialized when in fact it always is.
2129 */
2130 pulse.num_pulse = 0;
2131
2132 global_gain = get_bits(gb, 8);
2133
2134 if (!common_window && !scale_flag) {
2135 ret = decode_ics_info(ac, ics, gb);
2136 if (ret < 0)
2137 goto fail;
2138 }
2139
2140 if ((ret = decode_band_types(ac, sce->band_type,
2141 sce->band_type_run_end, gb, ics)) < 0)
2142 goto fail;
2143 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
2144 sce->band_type, sce->band_type_run_end)) < 0)
2145 goto fail;
2146
2147 pulse_present = 0;
2148 if (!scale_flag) {
2149 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
2150 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2151 av_log(ac->avctx, AV_LOG_ERROR,
2152 "Pulse tool not allowed in eight short sequence.\n");
2153 ret = AVERROR_INVALIDDATA;
2154 goto fail;
2155 }
2156 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
2157 av_log(ac->avctx, AV_LOG_ERROR,
2158 "Pulse data corrupt or invalid.\n");
2159 ret = AVERROR_INVALIDDATA;
2160 goto fail;
2161 }
2162 }
2163 tns->present = get_bits1(gb);
2164 if (tns->present && !er_syntax) {
2165 ret = decode_tns(ac, tns, gb, ics);
2166 if (ret < 0)
2167 goto fail;
2168 }
2169 if (!eld_syntax && get_bits1(gb)) {
2170 decode_gain_control(sce, gb);
2171 if (!ac->warned_gain_control) {
2172 avpriv_report_missing_feature(ac->avctx, "Gain control");
2173 ac->warned_gain_control = 1;
2174 }
2175 }
2176 // I see no textual basis in the spec for this occurring after SSR gain
2177 // control, but this is what both reference and real implmentations do
2178 if (tns->present && er_syntax) {
2179 ret = decode_tns(ac, tns, gb, ics);
2180 if (ret < 0)
2181 goto fail;
2182 }
2183 }
2184
2185 ret = decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
2186 &pulse, ics, sce->band_type);
2187 if (ret < 0)
2188 goto fail;
2189
2190 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
2191 apply_prediction(ac, sce);
2192
2193 return 0;
2194 fail:
2195 tns->present = 0;
2196 return ret;
2197 }
2198
2199 /**
2200 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
2201 */
apply_mid_side_stereo(AACContext * ac,ChannelElement * cpe)2202 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
2203 {
2204 const IndividualChannelStream *ics = &cpe->ch[0].ics;
2205 INTFLOAT *ch0 = cpe->ch[0].coeffs;
2206 INTFLOAT *ch1 = cpe->ch[1].coeffs;
2207 int g, i, group, idx = 0;
2208 const uint16_t *offsets = ics->swb_offset;
2209 for (g = 0; g < ics->num_window_groups; g++) {
2210 for (i = 0; i < ics->max_sfb; i++, idx++) {
2211 if (cpe->ms_mask[idx] &&
2212 cpe->ch[0].band_type[idx] < NOISE_BT &&
2213 cpe->ch[1].band_type[idx] < NOISE_BT) {
2214 #if USE_FIXED
2215 for (group = 0; group < ics->group_len[g]; group++) {
2216 ac->fdsp->butterflies_fixed(ch0 + group * 128 + offsets[i],
2217 ch1 + group * 128 + offsets[i],
2218 offsets[i+1] - offsets[i]);
2219 #else
2220 for (group = 0; group < ics->group_len[g]; group++) {
2221 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2222 ch1 + group * 128 + offsets[i],
2223 offsets[i+1] - offsets[i]);
2224 #endif /* USE_FIXED */
2225 }
2226 }
2227 }
2228 ch0 += ics->group_len[g] * 128;
2229 ch1 += ics->group_len[g] * 128;
2230 }
2231 }
2232
2233 /**
2234 * intensity stereo decoding; reference: 4.6.8.2.3
2235 *
2236 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2237 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2238 * [3] reserved for scalable AAC
2239 */
2240 static void apply_intensity_stereo(AACContext *ac,
2241 ChannelElement *cpe, int ms_present)
2242 {
2243 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2244 SingleChannelElement *sce1 = &cpe->ch[1];
2245 INTFLOAT *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2246 const uint16_t *offsets = ics->swb_offset;
2247 int g, group, i, idx = 0;
2248 int c;
2249 INTFLOAT scale;
2250 for (g = 0; g < ics->num_window_groups; g++) {
2251 for (i = 0; i < ics->max_sfb;) {
2252 if (sce1->band_type[idx] == INTENSITY_BT ||
2253 sce1->band_type[idx] == INTENSITY_BT2) {
2254 const int bt_run_end = sce1->band_type_run_end[idx];
2255 for (; i < bt_run_end; i++, idx++) {
2256 c = -1 + 2 * (sce1->band_type[idx] - 14);
2257 if (ms_present)
2258 c *= 1 - 2 * cpe->ms_mask[idx];
2259 scale = c * sce1->sf[idx];
2260 for (group = 0; group < ics->group_len[g]; group++)
2261 #if USE_FIXED
2262 ac->subband_scale(coef1 + group * 128 + offsets[i],
2263 coef0 + group * 128 + offsets[i],
2264 scale,
2265 23,
2266 offsets[i + 1] - offsets[i] ,ac->avctx);
2267 #else
2268 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2269 coef0 + group * 128 + offsets[i],
2270 scale,
2271 offsets[i + 1] - offsets[i]);
2272 #endif /* USE_FIXED */
2273 }
2274 } else {
2275 int bt_run_end = sce1->band_type_run_end[idx];
2276 idx += bt_run_end - i;
2277 i = bt_run_end;
2278 }
2279 }
2280 coef0 += ics->group_len[g] * 128;
2281 coef1 += ics->group_len[g] * 128;
2282 }
2283 }
2284
2285 /**
2286 * Decode a channel_pair_element; reference: table 4.4.
2287 *
2288 * @return Returns error status. 0 - OK, !0 - error
2289 */
2290 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2291 {
2292 int i, ret, common_window, ms_present = 0;
2293 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2294
2295 common_window = eld_syntax || get_bits1(gb);
2296 if (common_window) {
2297 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2298 return AVERROR_INVALIDDATA;
2299 i = cpe->ch[1].ics.use_kb_window[0];
2300 cpe->ch[1].ics = cpe->ch[0].ics;
2301 cpe->ch[1].ics.use_kb_window[1] = i;
2302 if (cpe->ch[1].ics.predictor_present &&
2303 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2304 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2305 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2306 ms_present = get_bits(gb, 2);
2307 if (ms_present == 3) {
2308 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2309 return AVERROR_INVALIDDATA;
2310 } else if (ms_present)
2311 decode_mid_side_stereo(cpe, gb, ms_present);
2312 }
2313 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2314 return ret;
2315 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2316 return ret;
2317
2318 if (common_window) {
2319 if (ms_present)
2320 apply_mid_side_stereo(ac, cpe);
2321 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2322 apply_prediction(ac, &cpe->ch[0]);
2323 apply_prediction(ac, &cpe->ch[1]);
2324 }
2325 }
2326
2327 apply_intensity_stereo(ac, cpe, ms_present);
2328 return 0;
2329 }
2330
2331 static const float cce_scale[] = {
2332 1.09050773266525765921, //2^(1/8)
2333 1.18920711500272106672, //2^(1/4)
2334 M_SQRT2,
2335 2,
2336 };
2337
2338 /**
2339 * Decode coupling_channel_element; reference: table 4.8.
2340 *
2341 * @return Returns error status. 0 - OK, !0 - error
2342 */
2343 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2344 {
2345 int num_gain = 0;
2346 int c, g, sfb, ret;
2347 int sign;
2348 INTFLOAT scale;
2349 SingleChannelElement *sce = &che->ch[0];
2350 ChannelCoupling *coup = &che->coup;
2351
2352 coup->coupling_point = 2 * get_bits1(gb);
2353 coup->num_coupled = get_bits(gb, 3);
2354 for (c = 0; c <= coup->num_coupled; c++) {
2355 num_gain++;
2356 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2357 coup->id_select[c] = get_bits(gb, 4);
2358 if (coup->type[c] == TYPE_CPE) {
2359 coup->ch_select[c] = get_bits(gb, 2);
2360 if (coup->ch_select[c] == 3)
2361 num_gain++;
2362 } else
2363 coup->ch_select[c] = 2;
2364 }
2365 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2366
2367 sign = get_bits(gb, 1);
2368 #if USE_FIXED
2369 scale = get_bits(gb, 2);
2370 #else
2371 scale = cce_scale[get_bits(gb, 2)];
2372 #endif
2373
2374 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2375 return ret;
2376
2377 for (c = 0; c < num_gain; c++) {
2378 int idx = 0;
2379 int cge = 1;
2380 int gain = 0;
2381 INTFLOAT gain_cache = FIXR10(1.);
2382 if (c) {
2383 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2384 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2385 gain_cache = GET_GAIN(scale, gain);
2386 #if USE_FIXED
2387 if ((abs(gain_cache)-1024) >> 3 > 30)
2388 return AVERROR(ERANGE);
2389 #endif
2390 }
2391 if (coup->coupling_point == AFTER_IMDCT) {
2392 coup->gain[c][0] = gain_cache;
2393 } else {
2394 for (g = 0; g < sce->ics.num_window_groups; g++) {
2395 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2396 if (sce->band_type[idx] != ZERO_BT) {
2397 if (!cge) {
2398 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2399 if (t) {
2400 int s = 1;
2401 t = gain += t;
2402 if (sign) {
2403 s -= 2 * (t & 0x1);
2404 t >>= 1;
2405 }
2406 gain_cache = GET_GAIN(scale, t) * s;
2407 #if USE_FIXED
2408 if ((abs(gain_cache)-1024) >> 3 > 30)
2409 return AVERROR(ERANGE);
2410 #endif
2411 }
2412 }
2413 coup->gain[c][idx] = gain_cache;
2414 }
2415 }
2416 }
2417 }
2418 }
2419 return 0;
2420 }
2421
2422 /**
2423 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2424 *
2425 * @return Returns number of bytes consumed.
2426 */
2427 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2428 GetBitContext *gb)
2429 {
2430 int i;
2431 int num_excl_chan = 0;
2432
2433 do {
2434 for (i = 0; i < 7; i++)
2435 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2436 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2437
2438 return num_excl_chan / 7;
2439 }
2440
2441 /**
2442 * Decode dynamic range information; reference: table 4.52.
2443 *
2444 * @return Returns number of bytes consumed.
2445 */
2446 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2447 GetBitContext *gb)
2448 {
2449 int n = 1;
2450 int drc_num_bands = 1;
2451 int i;
2452
2453 /* pce_tag_present? */
2454 if (get_bits1(gb)) {
2455 che_drc->pce_instance_tag = get_bits(gb, 4);
2456 skip_bits(gb, 4); // tag_reserved_bits
2457 n++;
2458 }
2459
2460 /* excluded_chns_present? */
2461 if (get_bits1(gb)) {
2462 n += decode_drc_channel_exclusions(che_drc, gb);
2463 }
2464
2465 /* drc_bands_present? */
2466 if (get_bits1(gb)) {
2467 che_drc->band_incr = get_bits(gb, 4);
2468 che_drc->interpolation_scheme = get_bits(gb, 4);
2469 n++;
2470 drc_num_bands += che_drc->band_incr;
2471 for (i = 0; i < drc_num_bands; i++) {
2472 che_drc->band_top[i] = get_bits(gb, 8);
2473 n++;
2474 }
2475 }
2476
2477 /* prog_ref_level_present? */
2478 if (get_bits1(gb)) {
2479 che_drc->prog_ref_level = get_bits(gb, 7);
2480 skip_bits1(gb); // prog_ref_level_reserved_bits
2481 n++;
2482 }
2483
2484 for (i = 0; i < drc_num_bands; i++) {
2485 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2486 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2487 n++;
2488 }
2489
2490 return n;
2491 }
2492
2493 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2494 uint8_t buf[256];
2495 int i, major, minor;
2496
2497 if (len < 13+7*8)
2498 goto unknown;
2499
2500 get_bits(gb, 13); len -= 13;
2501
2502 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2503 buf[i] = get_bits(gb, 8);
2504
2505 buf[i] = 0;
2506 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2507 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2508
2509 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2510 ac->avctx->internal->skip_samples = 1024;
2511 }
2512
2513 unknown:
2514 skip_bits_long(gb, len);
2515
2516 return 0;
2517 }
2518
2519 /**
2520 * Decode extension data (incomplete); reference: table 4.51.
2521 *
2522 * @param cnt length of TYPE_FIL syntactic element in bytes
2523 *
2524 * @return Returns number of bytes consumed
2525 */
2526 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2527 ChannelElement *che, enum RawDataBlockType elem_type)
2528 {
2529 int crc_flag = 0;
2530 int res = cnt;
2531 int type = get_bits(gb, 4);
2532
2533 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2534 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2535
2536 switch (type) { // extension type
2537 case EXT_SBR_DATA_CRC:
2538 crc_flag++;
2539 case EXT_SBR_DATA:
2540 if (!che) {
2541 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2542 return res;
2543 } else if (ac->oc[1].m4ac.frame_length_short) {
2544 if (!ac->warned_960_sbr)
2545 avpriv_report_missing_feature(ac->avctx,
2546 "SBR with 960 frame length");
2547 ac->warned_960_sbr = 1;
2548 skip_bits_long(gb, 8 * cnt - 4);
2549 return res;
2550 } else if (!ac->oc[1].m4ac.sbr) {
2551 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2552 skip_bits_long(gb, 8 * cnt - 4);
2553 return res;
2554 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2555 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2556 skip_bits_long(gb, 8 * cnt - 4);
2557 return res;
2558 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2559 ac->oc[1].m4ac.sbr = 1;
2560 ac->oc[1].m4ac.ps = 1;
2561 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2562 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2563 ac->oc[1].status, 1);
2564 } else {
2565 ac->oc[1].m4ac.sbr = 1;
2566 ac->avctx->profile = FF_PROFILE_AAC_HE;
2567 }
2568 res = AAC_RENAME(ff_decode_sbr_extension)(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2569 break;
2570 case EXT_DYNAMIC_RANGE:
2571 res = decode_dynamic_range(&ac->che_drc, gb);
2572 break;
2573 case EXT_FILL:
2574 decode_fill(ac, gb, 8 * cnt - 4);
2575 break;
2576 case EXT_FILL_DATA:
2577 case EXT_DATA_ELEMENT:
2578 default:
2579 skip_bits_long(gb, 8 * cnt - 4);
2580 break;
2581 };
2582 return res;
2583 }
2584
2585 /**
2586 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2587 *
2588 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2589 * @param coef spectral coefficients
2590 */
2591 static void apply_tns(INTFLOAT coef_param[1024], TemporalNoiseShaping *tns,
2592 IndividualChannelStream *ics, int decode)
2593 {
2594 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2595 int w, filt, m, i;
2596 int bottom, top, order, start, end, size, inc;
2597 INTFLOAT lpc[TNS_MAX_ORDER];
2598 INTFLOAT tmp[TNS_MAX_ORDER+1];
2599 UINTFLOAT *coef = coef_param;
2600
2601 if(!mmm)
2602 return;
2603
2604 for (w = 0; w < ics->num_windows; w++) {
2605 bottom = ics->num_swb;
2606 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2607 top = bottom;
2608 bottom = FFMAX(0, top - tns->length[w][filt]);
2609 order = tns->order[w][filt];
2610 if (order == 0)
2611 continue;
2612
2613 // tns_decode_coef
2614 AAC_RENAME(compute_lpc_coefs)(tns->coef[w][filt], order, lpc, 0, 0, 0);
2615
2616 start = ics->swb_offset[FFMIN(bottom, mmm)];
2617 end = ics->swb_offset[FFMIN( top, mmm)];
2618 if ((size = end - start) <= 0)
2619 continue;
2620 if (tns->direction[w][filt]) {
2621 inc = -1;
2622 start = end - 1;
2623 } else {
2624 inc = 1;
2625 }
2626 start += w * 128;
2627
2628 if (decode) {
2629 // ar filter
2630 for (m = 0; m < size; m++, start += inc)
2631 for (i = 1; i <= FFMIN(m, order); i++)
2632 coef[start] -= AAC_MUL26((INTFLOAT)coef[start - i * inc], lpc[i - 1]);
2633 } else {
2634 // ma filter
2635 for (m = 0; m < size; m++, start += inc) {
2636 tmp[0] = coef[start];
2637 for (i = 1; i <= FFMIN(m, order); i++)
2638 coef[start] += AAC_MUL26(tmp[i], lpc[i - 1]);
2639 for (i = order; i > 0; i--)
2640 tmp[i] = tmp[i - 1];
2641 }
2642 }
2643 }
2644 }
2645 }
2646
2647 /**
2648 * Apply windowing and MDCT to obtain the spectral
2649 * coefficient from the predicted sample by LTP.
2650 */
2651 static void windowing_and_mdct_ltp(AACContext *ac, INTFLOAT *out,
2652 INTFLOAT *in, IndividualChannelStream *ics)
2653 {
2654 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME2(aac_kbd_long_1024) : AAC_RENAME2(sine_1024);
2655 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME2(aac_kbd_short_128) : AAC_RENAME2(sine_128);
2656 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME2(aac_kbd_long_1024) : AAC_RENAME2(sine_1024);
2657 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME2(aac_kbd_short_128) : AAC_RENAME2(sine_128);
2658
2659 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2660 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2661 } else {
2662 memset(in, 0, 448 * sizeof(*in));
2663 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2664 }
2665 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2666 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2667 } else {
2668 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2669 memset(in + 1024 + 576, 0, 448 * sizeof(*in));
2670 }
2671 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2672 }
2673
2674 /**
2675 * Apply the long term prediction
2676 */
2677 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2678 {
2679 const LongTermPrediction *ltp = &sce->ics.ltp;
2680 const uint16_t *offsets = sce->ics.swb_offset;
2681 int i, sfb;
2682
2683 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2684 INTFLOAT *predTime = sce->ret;
2685 INTFLOAT *predFreq = ac->buf_mdct;
2686 int16_t num_samples = 2048;
2687
2688 if (ltp->lag < 1024)
2689 num_samples = ltp->lag + 1024;
2690 for (i = 0; i < num_samples; i++)
2691 predTime[i] = AAC_MUL30(sce->ltp_state[i + 2048 - ltp->lag], ltp->coef);
2692 memset(&predTime[i], 0, (2048 - i) * sizeof(*predTime));
2693
2694 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2695
2696 if (sce->tns.present)
2697 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2698
2699 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2700 if (ltp->used[sfb])
2701 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2702 sce->coeffs[i] += (UINTFLOAT)predFreq[i];
2703 }
2704 }
2705
2706 /**
2707 * Update the LTP buffer for next frame
2708 */
2709 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2710 {
2711 IndividualChannelStream *ics = &sce->ics;
2712 INTFLOAT *saved = sce->saved;
2713 INTFLOAT *saved_ltp = sce->coeffs;
2714 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME2(aac_kbd_long_1024) : AAC_RENAME2(sine_1024);
2715 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME2(aac_kbd_short_128) : AAC_RENAME2(sine_128);
2716 int i;
2717
2718 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2719 memcpy(saved_ltp, saved, 512 * sizeof(*saved_ltp));
2720 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2721 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2722
2723 for (i = 0; i < 64; i++)
2724 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2725 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2726 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(*saved_ltp));
2727 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2728 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2729
2730 for (i = 0; i < 64; i++)
2731 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2732 } else { // LONG_STOP or ONLY_LONG
2733 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2734
2735 for (i = 0; i < 512; i++)
2736 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], lwindow[511 - i]);
2737 }
2738
2739 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2740 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2741 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2742 }
2743
2744 /**
2745 * Conduct IMDCT and windowing.
2746 */
2747 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2748 {
2749 IndividualChannelStream *ics = &sce->ics;
2750 INTFLOAT *in = sce->coeffs;
2751 INTFLOAT *out = sce->ret;
2752 INTFLOAT *saved = sce->saved;
2753 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME2(aac_kbd_short_128) : AAC_RENAME2(sine_128);
2754 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME2(aac_kbd_long_1024) : AAC_RENAME2(sine_1024);
2755 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME2(aac_kbd_short_128) : AAC_RENAME2(sine_128);
2756 INTFLOAT *buf = ac->buf_mdct;
2757 INTFLOAT *temp = ac->temp;
2758 int i;
2759
2760 // imdct
2761 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2762 for (i = 0; i < 1024; i += 128)
2763 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2764 } else {
2765 ac->mdct.imdct_half(&ac->mdct, buf, in);
2766 #if USE_FIXED
2767 for (i=0; i<1024; i++)
2768 buf[i] = (buf[i] + 4LL) >> 3;
2769 #endif /* USE_FIXED */
2770 }
2771
2772 /* window overlapping
2773 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2774 * and long to short transitions are considered to be short to short
2775 * transitions. This leaves just two cases (long to long and short to short)
2776 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2777 */
2778 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2779 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2780 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2781 } else {
2782 memcpy( out, saved, 448 * sizeof(*out));
2783
2784 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2785 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2786 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2787 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2788 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2789 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2790 memcpy( out + 448 + 4*128, temp, 64 * sizeof(*out));
2791 } else {
2792 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2793 memcpy( out + 576, buf + 64, 448 * sizeof(*out));
2794 }
2795 }
2796
2797 // buffer update
2798 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2799 memcpy( saved, temp + 64, 64 * sizeof(*saved));
2800 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2801 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2802 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2803 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2804 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2805 memcpy( saved, buf + 512, 448 * sizeof(*saved));
2806 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2807 } else { // LONG_STOP or ONLY_LONG
2808 memcpy( saved, buf + 512, 512 * sizeof(*saved));
2809 }
2810 }
2811
2812 /**
2813 * Conduct IMDCT and windowing.
2814 */
2815 static void imdct_and_windowing_960(AACContext *ac, SingleChannelElement *sce)
2816 {
2817 #if !USE_FIXED
2818 IndividualChannelStream *ics = &sce->ics;
2819 INTFLOAT *in = sce->coeffs;
2820 INTFLOAT *out = sce->ret;
2821 INTFLOAT *saved = sce->saved;
2822 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(aac_kbd_short_120) : AAC_RENAME(sine_120);
2823 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(aac_kbd_long_960) : AAC_RENAME(sine_960);
2824 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(aac_kbd_short_120) : AAC_RENAME(sine_120);
2825 INTFLOAT *buf = ac->buf_mdct;
2826 INTFLOAT *temp = ac->temp;
2827 int i;
2828
2829 // imdct
2830 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2831 for (i = 0; i < 8; i++)
2832 ac->mdct120->imdct_half(ac->mdct120, buf + i * 120, in + i * 128, 1);
2833 } else {
2834 ac->mdct960->imdct_half(ac->mdct960, buf, in, 1);
2835 }
2836
2837 /* window overlapping
2838 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2839 * and long to short transitions are considered to be short to short
2840 * transitions. This leaves just two cases (long to long and short to short)
2841 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2842 */
2843
2844 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2845 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2846 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 480);
2847 } else {
2848 memcpy( out, saved, 420 * sizeof(*out));
2849
2850 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2851 ac->fdsp->vector_fmul_window(out + 420 + 0*120, saved + 420, buf + 0*120, swindow_prev, 60);
2852 ac->fdsp->vector_fmul_window(out + 420 + 1*120, buf + 0*120 + 60, buf + 1*120, swindow, 60);
2853 ac->fdsp->vector_fmul_window(out + 420 + 2*120, buf + 1*120 + 60, buf + 2*120, swindow, 60);
2854 ac->fdsp->vector_fmul_window(out + 420 + 3*120, buf + 2*120 + 60, buf + 3*120, swindow, 60);
2855 ac->fdsp->vector_fmul_window(temp, buf + 3*120 + 60, buf + 4*120, swindow, 60);
2856 memcpy( out + 420 + 4*120, temp, 60 * sizeof(*out));
2857 } else {
2858 ac->fdsp->vector_fmul_window(out + 420, saved + 420, buf, swindow_prev, 60);
2859 memcpy( out + 540, buf + 60, 420 * sizeof(*out));
2860 }
2861 }
2862
2863 // buffer update
2864 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2865 memcpy( saved, temp + 60, 60 * sizeof(*saved));
2866 ac->fdsp->vector_fmul_window(saved + 60, buf + 4*120 + 60, buf + 5*120, swindow, 60);
2867 ac->fdsp->vector_fmul_window(saved + 180, buf + 5*120 + 60, buf + 6*120, swindow, 60);
2868 ac->fdsp->vector_fmul_window(saved + 300, buf + 6*120 + 60, buf + 7*120, swindow, 60);
2869 memcpy( saved + 420, buf + 7*120 + 60, 60 * sizeof(*saved));
2870 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2871 memcpy( saved, buf + 480, 420 * sizeof(*saved));
2872 memcpy( saved + 420, buf + 7*120 + 60, 60 * sizeof(*saved));
2873 } else { // LONG_STOP or ONLY_LONG
2874 memcpy( saved, buf + 480, 480 * sizeof(*saved));
2875 }
2876 #endif
2877 }
2878 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2879 {
2880 IndividualChannelStream *ics = &sce->ics;
2881 INTFLOAT *in = sce->coeffs;
2882 INTFLOAT *out = sce->ret;
2883 INTFLOAT *saved = sce->saved;
2884 INTFLOAT *buf = ac->buf_mdct;
2885 #if USE_FIXED
2886 int i;
2887 #endif /* USE_FIXED */
2888
2889 // imdct
2890 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2891
2892 #if USE_FIXED
2893 for (i = 0; i < 1024; i++)
2894 buf[i] = (buf[i] + 2) >> 2;
2895 #endif /* USE_FIXED */
2896
2897 // window overlapping
2898 if (ics->use_kb_window[1]) {
2899 // AAC LD uses a low overlap sine window instead of a KBD window
2900 memcpy(out, saved, 192 * sizeof(*out));
2901 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, AAC_RENAME2(sine_128), 64);
2902 memcpy( out + 320, buf + 64, 192 * sizeof(*out));
2903 } else {
2904 ac->fdsp->vector_fmul_window(out, saved, buf, AAC_RENAME2(sine_512), 256);
2905 }
2906
2907 // buffer update
2908 memcpy(saved, buf + 256, 256 * sizeof(*saved));
2909 }
2910
2911 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2912 {
2913 UINTFLOAT *in = sce->coeffs;
2914 INTFLOAT *out = sce->ret;
2915 INTFLOAT *saved = sce->saved;
2916 INTFLOAT *buf = ac->buf_mdct;
2917 int i;
2918 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2919 const int n2 = n >> 1;
2920 const int n4 = n >> 2;
2921 const INTFLOAT *const window = n == 480 ? AAC_RENAME(ff_aac_eld_window_480) :
2922 AAC_RENAME(ff_aac_eld_window_512);
2923
2924 // Inverse transform, mapped to the conventional IMDCT by
2925 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2926 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2927 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2928 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2929 for (i = 0; i < n2; i+=2) {
2930 INTFLOAT temp;
2931 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2932 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2933 }
2934 #if !USE_FIXED
2935 if (n == 480)
2936 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1);
2937 else
2938 #endif
2939 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2940
2941 #if USE_FIXED
2942 for (i = 0; i < 1024; i++)
2943 buf[i] = (buf[i] + 1) >> 1;
2944 #endif /* USE_FIXED */
2945
2946 for (i = 0; i < n; i+=2) {
2947 buf[i] = -buf[i];
2948 }
2949 // Like with the regular IMDCT at this point we still have the middle half
2950 // of a transform but with even symmetry on the left and odd symmetry on
2951 // the right
2952
2953 // window overlapping
2954 // The spec says to use samples [0..511] but the reference decoder uses
2955 // samples [128..639].
2956 for (i = n4; i < n2; i ++) {
2957 out[i - n4] = AAC_MUL31( buf[ n2 - 1 - i] , window[i - n4]) +
2958 AAC_MUL31( saved[ i + n2] , window[i + n - n4]) +
2959 AAC_MUL31(-saved[n + n2 - 1 - i] , window[i + 2*n - n4]) +
2960 AAC_MUL31(-saved[ 2*n + n2 + i] , window[i + 3*n - n4]);
2961 }
2962 for (i = 0; i < n2; i ++) {
2963 out[n4 + i] = AAC_MUL31( buf[ i] , window[i + n2 - n4]) +
2964 AAC_MUL31(-saved[ n - 1 - i] , window[i + n2 + n - n4]) +
2965 AAC_MUL31(-saved[ n + i] , window[i + n2 + 2*n - n4]) +
2966 AAC_MUL31( saved[2*n + n - 1 - i] , window[i + n2 + 3*n - n4]);
2967 }
2968 for (i = 0; i < n4; i ++) {
2969 out[n2 + n4 + i] = AAC_MUL31( buf[ i + n2] , window[i + n - n4]) +
2970 AAC_MUL31(-saved[n2 - 1 - i] , window[i + 2*n - n4]) +
2971 AAC_MUL31(-saved[n + n2 + i] , window[i + 3*n - n4]);
2972 }
2973
2974 // buffer update
2975 memmove(saved + n, saved, 2 * n * sizeof(*saved));
2976 memcpy( saved, buf, n * sizeof(*saved));
2977 }
2978
2979 /**
2980 * channel coupling transformation interface
2981 *
2982 * @param apply_coupling_method pointer to (in)dependent coupling function
2983 */
2984 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2985 enum RawDataBlockType type, int elem_id,
2986 enum CouplingPoint coupling_point,
2987 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2988 {
2989 int i, c;
2990
2991 for (i = 0; i < MAX_ELEM_ID; i++) {
2992 ChannelElement *cce = ac->che[TYPE_CCE][i];
2993 int index = 0;
2994
2995 if (cce && cce->coup.coupling_point == coupling_point) {
2996 ChannelCoupling *coup = &cce->coup;
2997
2998 for (c = 0; c <= coup->num_coupled; c++) {
2999 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
3000 if (coup->ch_select[c] != 1) {
3001 apply_coupling_method(ac, &cc->ch[0], cce, index);
3002 if (coup->ch_select[c] != 0)
3003 index++;
3004 }
3005 if (coup->ch_select[c] != 2)
3006 apply_coupling_method(ac, &cc->ch[1], cce, index++);
3007 } else
3008 index += 1 + (coup->ch_select[c] == 3);
3009 }
3010 }
3011 }
3012 }
3013
3014 /**
3015 * Convert spectral data to samples, applying all supported tools as appropriate.
3016 */
3017 static void spectral_to_sample(AACContext *ac, int samples)
3018 {
3019 int i, type;
3020 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
3021 switch (ac->oc[1].m4ac.object_type) {
3022 case AOT_ER_AAC_LD:
3023 imdct_and_window = imdct_and_windowing_ld;
3024 break;
3025 case AOT_ER_AAC_ELD:
3026 imdct_and_window = imdct_and_windowing_eld;
3027 break;
3028 default:
3029 if (ac->oc[1].m4ac.frame_length_short)
3030 imdct_and_window = imdct_and_windowing_960;
3031 else
3032 imdct_and_window = ac->imdct_and_windowing;
3033 }
3034 for (type = 3; type >= 0; type--) {
3035 for (i = 0; i < MAX_ELEM_ID; i++) {
3036 ChannelElement *che = ac->che[type][i];
3037 if (che && che->present) {
3038 if (type <= TYPE_CPE)
3039 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, AAC_RENAME(apply_dependent_coupling));
3040 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
3041 if (che->ch[0].ics.predictor_present) {
3042 if (che->ch[0].ics.ltp.present)
3043 ac->apply_ltp(ac, &che->ch[0]);
3044 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
3045 ac->apply_ltp(ac, &che->ch[1]);
3046 }
3047 }
3048 if (che->ch[0].tns.present)
3049 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
3050 if (che->ch[1].tns.present)
3051 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
3052 if (type <= TYPE_CPE)
3053 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, AAC_RENAME(apply_dependent_coupling));
3054 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
3055 imdct_and_window(ac, &che->ch[0]);
3056 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
3057 ac->update_ltp(ac, &che->ch[0]);
3058 if (type == TYPE_CPE) {
3059 imdct_and_window(ac, &che->ch[1]);
3060 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
3061 ac->update_ltp(ac, &che->ch[1]);
3062 }
3063 if (ac->oc[1].m4ac.sbr > 0) {
3064 AAC_RENAME(ff_sbr_apply)(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
3065 }
3066 }
3067 if (type <= TYPE_CCE)
3068 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, AAC_RENAME(apply_independent_coupling));
3069
3070 #if USE_FIXED
3071 {
3072 int j;
3073 /* preparation for resampler */
3074 for(j = 0; j<samples; j++){
3075 che->ch[0].ret[j] = (int32_t)av_clip64((int64_t)che->ch[0].ret[j]*128, INT32_MIN, INT32_MAX-0x8000)+0x8000;
3076 if(type == TYPE_CPE)
3077 che->ch[1].ret[j] = (int32_t)av_clip64((int64_t)che->ch[1].ret[j]*128, INT32_MIN, INT32_MAX-0x8000)+0x8000;
3078 }
3079 }
3080 #endif /* USE_FIXED */
3081 che->present = 0;
3082 } else if (che) {
3083 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
3084 }
3085 }
3086 }
3087 }
3088
3089 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
3090 {
3091 int size;
3092 AACADTSHeaderInfo hdr_info;
3093 uint8_t layout_map[MAX_ELEM_ID*4][3];
3094 int layout_map_tags, ret;
3095
3096 size = ff_adts_header_parse(gb, &hdr_info);
3097 if (size > 0) {
3098 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
3099 // This is 2 for "VLB " audio in NSV files.
3100 // See samples/nsv/vlb_audio.
3101 avpriv_report_missing_feature(ac->avctx,
3102 "More than one AAC RDB per ADTS frame");
3103 ac->warned_num_aac_frames = 1;
3104 }
3105 push_output_configuration(ac);
3106 if (hdr_info.chan_config) {
3107 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
3108 if ((ret = set_default_channel_config(ac, ac->avctx,
3109 layout_map,
3110 &layout_map_tags,
3111 hdr_info.chan_config)) < 0)
3112 return ret;
3113 if ((ret = output_configure(ac, layout_map, layout_map_tags,
3114 FFMAX(ac->oc[1].status,
3115 OC_TRIAL_FRAME), 0)) < 0)
3116 return ret;
3117 } else {
3118 ac->oc[1].m4ac.chan_config = 0;
3119 /**
3120 * dual mono frames in Japanese DTV can have chan_config 0
3121 * WITHOUT specifying PCE.
3122 * thus, set dual mono as default.
3123 */
3124 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
3125 layout_map_tags = 2;
3126 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
3127 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
3128 layout_map[0][1] = 0;
3129 layout_map[1][1] = 1;
3130 if (output_configure(ac, layout_map, layout_map_tags,
3131 OC_TRIAL_FRAME, 0))
3132 return -7;
3133 }
3134 }
3135 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
3136 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
3137 ac->oc[1].m4ac.object_type = hdr_info.object_type;
3138 ac->oc[1].m4ac.frame_length_short = 0;
3139 if (ac->oc[0].status != OC_LOCKED ||
3140 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
3141 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
3142 ac->oc[1].m4ac.sbr = -1;
3143 ac->oc[1].m4ac.ps = -1;
3144 }
3145 if (!hdr_info.crc_absent)
3146 skip_bits(gb, 16);
3147 }
3148 return size;
3149 }
3150
3151 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
3152 int *got_frame_ptr, GetBitContext *gb)
3153 {
3154 AACContext *ac = avctx->priv_data;
3155 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
3156 ChannelElement *che;
3157 int err, i;
3158 int samples = m4ac->frame_length_short ? 960 : 1024;
3159 int chan_config = m4ac->chan_config;
3160 int aot = m4ac->object_type;
3161
3162 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
3163 samples >>= 1;
3164
3165 ac->frame = data;
3166
3167 if ((err = frame_configure_elements(avctx)) < 0)
3168 return err;
3169
3170 // The FF_PROFILE_AAC_* defines are all object_type - 1
3171 // This may lead to an undefined profile being signaled
3172 ac->avctx->profile = aot - 1;
3173
3174 ac->tags_mapped = 0;
3175
3176 if (chan_config < 0 || (chan_config >= 8 && chan_config < 11) || chan_config >= 13) {
3177 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
3178 chan_config);
3179 return AVERROR_INVALIDDATA;
3180 }
3181 for (i = 0; i < tags_per_config[chan_config]; i++) {
3182 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
3183 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
3184 if (!(che=get_che(ac, elem_type, elem_id))) {
3185 av_log(ac->avctx, AV_LOG_ERROR,
3186 "channel element %d.%d is not allocated\n",
3187 elem_type, elem_id);
3188 return AVERROR_INVALIDDATA;
3189 }
3190 che->present = 1;
3191 if (aot != AOT_ER_AAC_ELD)
3192 skip_bits(gb, 4);
3193 switch (elem_type) {
3194 case TYPE_SCE:
3195 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3196 break;
3197 case TYPE_CPE:
3198 err = decode_cpe(ac, gb, che);
3199 break;
3200 case TYPE_LFE:
3201 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3202 break;
3203 }
3204 if (err < 0)
3205 return err;
3206 }
3207
3208 spectral_to_sample(ac, samples);
3209
3210 if (!ac->frame->data[0] && samples) {
3211 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3212 return AVERROR_INVALIDDATA;
3213 }
3214
3215 ac->frame->nb_samples = samples;
3216 ac->frame->sample_rate = avctx->sample_rate;
3217 *got_frame_ptr = 1;
3218
3219 skip_bits_long(gb, get_bits_left(gb));
3220 return 0;
3221 }
3222
3223 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
3224 int *got_frame_ptr, GetBitContext *gb,
3225 const AVPacket *avpkt)
3226 {
3227 AACContext *ac = avctx->priv_data;
3228 ChannelElement *che = NULL, *che_prev = NULL;
3229 enum RawDataBlockType elem_type, che_prev_type = TYPE_END;
3230 int err, elem_id;
3231 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
3232 int is_dmono, sce_count = 0;
3233 int payload_alignment;
3234 uint8_t che_presence[4][MAX_ELEM_ID] = {{0}};
3235
3236 ac->frame = data;
3237
3238 if (show_bits(gb, 12) == 0xfff) {
3239 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
3240 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
3241 goto fail;
3242 }
3243 if (ac->oc[1].m4ac.sampling_index > 12) {
3244 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
3245 err = AVERROR_INVALIDDATA;
3246 goto fail;
3247 }
3248 }
3249
3250 if ((err = frame_configure_elements(avctx)) < 0)
3251 goto fail;
3252
3253 // The FF_PROFILE_AAC_* defines are all object_type - 1
3254 // This may lead to an undefined profile being signaled
3255 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
3256
3257 payload_alignment = get_bits_count(gb);
3258 ac->tags_mapped = 0;
3259 // parse
3260 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
3261 elem_id = get_bits(gb, 4);
3262
3263 if (avctx->debug & FF_DEBUG_STARTCODE)
3264 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
3265
3266 if (!avctx->channels && elem_type != TYPE_PCE) {
3267 err = AVERROR_INVALIDDATA;
3268 goto fail;
3269 }
3270
3271 if (elem_type < TYPE_DSE) {
3272 if (che_presence[elem_type][elem_id]) {
3273 int error = che_presence[elem_type][elem_id] > 1;
3274 av_log(ac->avctx, error ? AV_LOG_ERROR : AV_LOG_DEBUG, "channel element %d.%d duplicate\n",
3275 elem_type, elem_id);
3276 if (error) {
3277 err = AVERROR_INVALIDDATA;
3278 goto fail;
3279 }
3280 }
3281 che_presence[elem_type][elem_id]++;
3282
3283 if (!(che=get_che(ac, elem_type, elem_id))) {
3284 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
3285 elem_type, elem_id);
3286 err = AVERROR_INVALIDDATA;
3287 goto fail;
3288 }
3289 samples = ac->oc[1].m4ac.frame_length_short ? 960 : 1024;
3290 che->present = 1;
3291 }
3292
3293 switch (elem_type) {
3294
3295 case TYPE_SCE:
3296 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3297 audio_found = 1;
3298 sce_count++;
3299 break;
3300
3301 case TYPE_CPE:
3302 err = decode_cpe(ac, gb, che);
3303 audio_found = 1;
3304 break;
3305
3306 case TYPE_CCE:
3307 err = decode_cce(ac, gb, che);
3308 break;
3309
3310 case TYPE_LFE:
3311 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3312 audio_found = 1;
3313 break;
3314
3315 case TYPE_DSE:
3316 err = skip_data_stream_element(ac, gb);
3317 break;
3318
3319 case TYPE_PCE: {
3320 uint8_t layout_map[MAX_ELEM_ID*4][3] = {{0}};
3321 int tags;
3322
3323 int pushed = push_output_configuration(ac);
3324 if (pce_found && !pushed) {
3325 err = AVERROR_INVALIDDATA;
3326 goto fail;
3327 }
3328
3329 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb,
3330 payload_alignment);
3331 if (tags < 0) {
3332 err = tags;
3333 break;
3334 }
3335 if (pce_found) {
3336 av_log(avctx, AV_LOG_ERROR,
3337 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3338 pop_output_configuration(ac);
3339 } else {
3340 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3341 if (!err)
3342 ac->oc[1].m4ac.chan_config = 0;
3343 pce_found = 1;
3344 }
3345 break;
3346 }
3347
3348 case TYPE_FIL:
3349 if (elem_id == 15)
3350 elem_id += get_bits(gb, 8) - 1;
3351 if (get_bits_left(gb) < 8 * elem_id) {
3352 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3353 err = AVERROR_INVALIDDATA;
3354 goto fail;
3355 }
3356 err = 0;
3357 while (elem_id > 0) {
3358 int ret = decode_extension_payload(ac, gb, elem_id, che_prev, che_prev_type);
3359 if (ret < 0) {
3360 err = ret;
3361 break;
3362 }
3363 elem_id -= ret;
3364 }
3365 break;
3366
3367 default:
3368 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3369 break;
3370 }
3371
3372 if (elem_type < TYPE_DSE) {
3373 che_prev = che;
3374 che_prev_type = elem_type;
3375 }
3376
3377 if (err)
3378 goto fail;
3379
3380 if (get_bits_left(gb) < 3) {
3381 av_log(avctx, AV_LOG_ERROR, overread_err);
3382 err = AVERROR_INVALIDDATA;
3383 goto fail;
3384 }
3385 }
3386
3387 if (!avctx->channels) {
3388 *got_frame_ptr = 0;
3389 return 0;
3390 }
3391
3392 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3393 samples <<= multiplier;
3394
3395 spectral_to_sample(ac, samples);
3396
3397 if (ac->oc[1].status && audio_found) {
3398 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3399 avctx->frame_size = samples;
3400 ac->oc[1].status = OC_LOCKED;
3401 }
3402
3403 if (multiplier)
3404 avctx->internal->skip_samples_multiplier = 2;
3405
3406 if (!ac->frame->data[0] && samples) {
3407 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3408 err = AVERROR_INVALIDDATA;
3409 goto fail;
3410 }
3411
3412 if (samples) {
3413 ac->frame->nb_samples = samples;
3414 ac->frame->sample_rate = avctx->sample_rate;
3415 } else
3416 av_frame_unref(ac->frame);
3417 *got_frame_ptr = !!samples;
3418
3419 /* for dual-mono audio (SCE + SCE) */
3420 is_dmono = ac->dmono_mode && sce_count == 2 &&
3421 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3422 if (is_dmono) {
3423 if (ac->dmono_mode == 1)
3424 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3425 else if (ac->dmono_mode == 2)
3426 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3427 }
3428
3429 return 0;
3430 fail:
3431 pop_output_configuration(ac);
3432 return err;
3433 }
3434
3435 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3436 int *got_frame_ptr, AVPacket *avpkt)
3437 {
3438 AACContext *ac = avctx->priv_data;
3439 const uint8_t *buf = avpkt->data;
3440 int buf_size = avpkt->size;
3441 GetBitContext gb;
3442 int buf_consumed;
3443 int buf_offset;
3444 int err;
3445 buffer_size_t new_extradata_size;
3446 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3447 AV_PKT_DATA_NEW_EXTRADATA,
3448 &new_extradata_size);
3449 buffer_size_t jp_dualmono_size;
3450 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3451 AV_PKT_DATA_JP_DUALMONO,
3452 &jp_dualmono_size);
3453
3454 if (new_extradata) {
3455 /* discard previous configuration */
3456 ac->oc[1].status = OC_NONE;
3457 err = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3458 new_extradata,
3459 new_extradata_size * 8LL, 1);
3460 if (err < 0) {
3461 return err;
3462 }
3463 }
3464
3465 ac->dmono_mode = 0;
3466 if (jp_dualmono && jp_dualmono_size > 0)
3467 ac->dmono_mode = 1 + *jp_dualmono;
3468 if (ac->force_dmono_mode >= 0)
3469 ac->dmono_mode = ac->force_dmono_mode;
3470
3471 if (INT_MAX / 8 <= buf_size)
3472 return AVERROR_INVALIDDATA;
3473
3474 if ((err = init_get_bits8(&gb, buf, buf_size)) < 0)
3475 return err;
3476
3477 switch (ac->oc[1].m4ac.object_type) {
3478 case AOT_ER_AAC_LC:
3479 case AOT_ER_AAC_LTP:
3480 case AOT_ER_AAC_LD:
3481 case AOT_ER_AAC_ELD:
3482 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3483 break;
3484 default:
3485 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3486 }
3487 if (err < 0)
3488 return err;
3489
3490 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3491 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3492 if (buf[buf_offset])
3493 break;
3494
3495 return buf_size > buf_offset ? buf_consumed : buf_size;
3496 }
3497
3498 static av_cold int aac_decode_close(AVCodecContext *avctx)
3499 {
3500 AACContext *ac = avctx->priv_data;
3501 int i, type;
3502
3503 for (i = 0; i < MAX_ELEM_ID; i++) {
3504 for (type = 0; type < 4; type++) {
3505 if (ac->che[type][i])
3506 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][i]->sbr);
3507 av_freep(&ac->che[type][i]);
3508 }
3509 }
3510
3511 ff_mdct_end(&ac->mdct);
3512 ff_mdct_end(&ac->mdct_small);
3513 ff_mdct_end(&ac->mdct_ld);
3514 ff_mdct_end(&ac->mdct_ltp);
3515 #if !USE_FIXED
3516 ff_mdct15_uninit(&ac->mdct120);
3517 ff_mdct15_uninit(&ac->mdct480);
3518 ff_mdct15_uninit(&ac->mdct960);
3519 #endif
3520 av_freep(&ac->fdsp);
3521 return 0;
3522 }
3523
3524 static void aacdec_init(AACContext *c)
3525 {
3526 c->imdct_and_windowing = imdct_and_windowing;
3527 c->apply_ltp = apply_ltp;
3528 c->apply_tns = apply_tns;
3529 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3530 c->update_ltp = update_ltp;
3531 #if USE_FIXED
3532 c->vector_pow43 = vector_pow43;
3533 c->subband_scale = subband_scale;
3534 #endif
3535
3536 #if !USE_FIXED
3537 if(ARCH_MIPS)
3538 ff_aacdec_init_mips(c);
3539 #endif /* !USE_FIXED */
3540 }
3541 /**
3542 * AVOptions for Japanese DTV specific extensions (ADTS only)
3543 */
3544 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3545 static const AVOption options[] = {
3546 {"dual_mono_mode", "Select the channel to decode for dual mono",
3547 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3548 AACDEC_FLAGS, "dual_mono_mode"},
3549
3550 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3551 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3552 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3553 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3554
3555 {NULL},
3556 };
3557
3558 static const AVClass aac_decoder_class = {
3559 .class_name = "AAC decoder",
3560 .item_name = av_default_item_name,
3561 .option = options,
3562 .version = LIBAVUTIL_VERSION_INT,
3563 };
3564