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 * 6 * This file is part of FFmpeg. 7 * 8 * FFmpeg is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU Lesser General Public 10 * License as published by the Free Software Foundation; either 11 * version 2.1 of the License, or (at your option) any later version. 12 * 13 * FFmpeg is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * Lesser General Public License for more details. 17 * 18 * You should have received a copy of the GNU Lesser General Public 19 * License along with FFmpeg; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 21 */ 22 23/** 24 * @file libavcodec/aac.c 25 * AAC decoder 26 * @author Oded Shimon ( ods15 ods15 dyndns org ) 27 * @author Maxim Gavrilov ( maxim.gavrilov gmail com ) 28 */ 29 30/* 31 * supported tools 32 * 33 * Support? Name 34 * N (code in SoC repo) gain control 35 * Y block switching 36 * Y window shapes - standard 37 * N window shapes - Low Delay 38 * Y filterbank - standard 39 * N (code in SoC repo) filterbank - Scalable Sample Rate 40 * Y Temporal Noise Shaping 41 * N (code in SoC repo) Long Term Prediction 42 * Y intensity stereo 43 * Y channel coupling 44 * Y frequency domain prediction 45 * Y Perceptual Noise Substitution 46 * Y Mid/Side stereo 47 * N Scalable Inverse AAC Quantization 48 * N Frequency Selective Switch 49 * N upsampling filter 50 * Y quantization & coding - AAC 51 * N quantization & coding - TwinVQ 52 * N quantization & coding - BSAC 53 * N AAC Error Resilience tools 54 * N Error Resilience payload syntax 55 * N Error Protection tool 56 * N CELP 57 * N Silence Compression 58 * N HVXC 59 * N HVXC 4kbits/s VR 60 * N Structured Audio tools 61 * N Structured Audio Sample Bank Format 62 * N MIDI 63 * N Harmonic and Individual Lines plus Noise 64 * N Text-To-Speech Interface 65 * N (in progress) Spectral Band Replication 66 * Y (not in this code) Layer-1 67 * Y (not in this code) Layer-2 68 * Y (not in this code) Layer-3 69 * N SinuSoidal Coding (Transient, Sinusoid, Noise) 70 * N (planned) Parametric Stereo 71 * N Direct Stream Transfer 72 * 73 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication. 74 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and 75 Parametric Stereo. 76 */ 77 78 79#include "avcodec.h" 80#include "internal.h" 81#include "bitstream.h" 82#include "dsputil.h" 83#include "lpc.h" 84 85#include "aac.h" 86#include "aactab.h" 87#include "aacdectab.h" 88#include "mpeg4audio.h" 89#include "aac_parser.h" 90 91#include <assert.h> 92#include <errno.h> 93#include <math.h> 94#include <string.h> 95 96static VLC vlc_scalefactors; 97static VLC vlc_spectral[11]; 98 99 100static ChannelElement* get_che(AACContext *ac, int type, int elem_id) { 101 static const int8_t tags_per_config[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 }; 102 if (ac->tag_che_map[type][elem_id]) { 103 return ac->tag_che_map[type][elem_id]; 104 } 105 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) { 106 return NULL; 107 } 108 switch (ac->m4ac.chan_config) { 109 case 7: 110 if (ac->tags_mapped == 3 && type == TYPE_CPE) { 111 ac->tags_mapped++; 112 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2]; 113 } 114 case 6: 115 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1] 116 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have 117 encountered such a stream, transfer the LFE[0] element to SCE[1] */ 118 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) { 119 ac->tags_mapped++; 120 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0]; 121 } 122 case 5: 123 if (ac->tags_mapped == 2 && type == TYPE_CPE) { 124 ac->tags_mapped++; 125 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1]; 126 } 127 case 4: 128 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) { 129 ac->tags_mapped++; 130 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1]; 131 } 132 case 3: 133 case 2: 134 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) { 135 ac->tags_mapped++; 136 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0]; 137 } else if (ac->m4ac.chan_config == 2) { 138 return NULL; 139 } 140 case 1: 141 if (!ac->tags_mapped && type == TYPE_SCE) { 142 ac->tags_mapped++; 143 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0]; 144 } 145 default: 146 return NULL; 147 } 148} 149 150/** 151 * Configure output channel order based on the current program configuration element. 152 * 153 * @param che_pos current channel position configuration 154 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one. 155 * 156 * @return Returns error status. 0 - OK, !0 - error 157 */ 158static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID], 159 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID], int channel_config) { 160 AVCodecContext *avctx = ac->avccontext; 161 int i, type, channels = 0; 162 163 if(!memcmp(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]))) 164 return 0; /* no change */ 165 166 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])); 167 168 /* Allocate or free elements depending on if they are in the 169 * current program configuration. 170 * 171 * Set up default 1:1 output mapping. 172 * 173 * For a 5.1 stream the output order will be: 174 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ] 175 */ 176 177 for(i = 0; i < MAX_ELEM_ID; i++) { 178 for(type = 0; type < 4; type++) { 179 if(che_pos[type][i]) { 180 if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement)))) 181 return AVERROR(ENOMEM); 182 if(type != TYPE_CCE) { 183 ac->output_data[channels++] = ac->che[type][i]->ch[0].ret; 184 if(type == TYPE_CPE) { 185 ac->output_data[channels++] = ac->che[type][i]->ch[1].ret; 186 } 187 } 188 } else 189 av_freep(&ac->che[type][i]); 190 } 191 } 192 193 if (channel_config) { 194 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0])); 195 ac->tags_mapped = 0; 196 } else { 197 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0])); 198 ac->tags_mapped = 4*MAX_ELEM_ID; 199 } 200 201 avctx->channels = channels; 202 203 return 0; 204} 205 206/** 207 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit. 208 * 209 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present. 210 * @param sce_map mono (Single Channel Element) map 211 * @param type speaker type/position for these channels 212 */ 213static void decode_channel_map(enum ChannelPosition *cpe_map, 214 enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) { 215 while(n--) { 216 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map 217 map[get_bits(gb, 4)] = type; 218 } 219} 220 221/** 222 * Decode program configuration element; reference: table 4.2. 223 * 224 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one. 225 * 226 * @return Returns error status. 0 - OK, !0 - error 227 */ 228static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID], 229 GetBitContext * gb) { 230 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index; 231 232 skip_bits(gb, 2); // object_type 233 234 sampling_index = get_bits(gb, 4); 235 if(sampling_index > 12) { 236 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index); 237 return -1; 238 } 239 ac->m4ac.sampling_index = sampling_index; 240 ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index]; 241 num_front = get_bits(gb, 4); 242 num_side = get_bits(gb, 4); 243 num_back = get_bits(gb, 4); 244 num_lfe = get_bits(gb, 2); 245 num_assoc_data = get_bits(gb, 3); 246 num_cc = get_bits(gb, 4); 247 248 if (get_bits1(gb)) 249 skip_bits(gb, 4); // mono_mixdown_tag 250 if (get_bits1(gb)) 251 skip_bits(gb, 4); // stereo_mixdown_tag 252 253 if (get_bits1(gb)) 254 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround 255 256 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front); 257 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side ); 258 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back ); 259 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe ); 260 261 skip_bits_long(gb, 4 * num_assoc_data); 262 263 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc ); 264 265 align_get_bits(gb); 266 267 /* comment field, first byte is length */ 268 skip_bits_long(gb, 8 * get_bits(gb, 8)); 269 return 0; 270} 271 272/** 273 * Set up channel positions based on a default channel configuration 274 * as specified in table 1.17. 275 * 276 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one. 277 * 278 * @return Returns error status. 0 - OK, !0 - error 279 */ 280static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID], 281 int channel_config) 282{ 283 if(channel_config < 1 || channel_config > 7) { 284 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n", 285 channel_config); 286 return -1; 287 } 288 289 /* default channel configurations: 290 * 291 * 1ch : front center (mono) 292 * 2ch : L + R (stereo) 293 * 3ch : front center + L + R 294 * 4ch : front center + L + R + back center 295 * 5ch : front center + L + R + back stereo 296 * 6ch : front center + L + R + back stereo + LFE 297 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE 298 */ 299 300 if(channel_config != 2) 301 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono) 302 if(channel_config > 1) 303 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo) 304 if(channel_config == 4) 305 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center 306 if(channel_config > 4) 307 new_che_pos[TYPE_CPE][(channel_config == 7) + 1] 308 = AAC_CHANNEL_BACK; // back stereo 309 if(channel_config > 5) 310 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE 311 if(channel_config == 7) 312 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right 313 314 return 0; 315} 316 317/** 318 * Decode GA "General Audio" specific configuration; reference: table 4.1. 319 * 320 * @return Returns error status. 0 - OK, !0 - error 321 */ 322static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) { 323 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]; 324 int extension_flag, ret; 325 326 if(get_bits1(gb)) { // frameLengthFlag 327 ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1); 328 return -1; 329 } 330 331 if (get_bits1(gb)) // dependsOnCoreCoder 332 skip_bits(gb, 14); // coreCoderDelay 333 extension_flag = get_bits1(gb); 334 335 if(ac->m4ac.object_type == AOT_AAC_SCALABLE || 336 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE) 337 skip_bits(gb, 3); // layerNr 338 339 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])); 340 if (channel_config == 0) { 341 skip_bits(gb, 4); // element_instance_tag 342 if((ret = decode_pce(ac, new_che_pos, gb))) 343 return ret; 344 } else { 345 if((ret = set_default_channel_config(ac, new_che_pos, channel_config))) 346 return ret; 347 } 348 if((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config))) 349 return ret; 350 351 if (extension_flag) { 352 switch (ac->m4ac.object_type) { 353 case AOT_ER_BSAC: 354 skip_bits(gb, 5); // numOfSubFrame 355 skip_bits(gb, 11); // layer_length 356 break; 357 case AOT_ER_AAC_LC: 358 case AOT_ER_AAC_LTP: 359 case AOT_ER_AAC_SCALABLE: 360 case AOT_ER_AAC_LD: 361 skip_bits(gb, 3); /* aacSectionDataResilienceFlag 362 * aacScalefactorDataResilienceFlag 363 * aacSpectralDataResilienceFlag 364 */ 365 break; 366 } 367 skip_bits1(gb); // extensionFlag3 (TBD in version 3) 368 } 369 return 0; 370} 371 372/** 373 * Decode audio specific configuration; reference: table 1.13. 374 * 375 * @param data pointer to AVCodecContext extradata 376 * @param data_size size of AVCCodecContext extradata 377 * 378 * @return Returns error status. 0 - OK, !0 - error 379 */ 380static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) { 381 GetBitContext gb; 382 int i; 383 384 init_get_bits(&gb, data, data_size * 8); 385 386 if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0) 387 return -1; 388 if(ac->m4ac.sampling_index > 12) { 389 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index); 390 return -1; 391 } 392 393 skip_bits_long(&gb, i); 394 395 switch (ac->m4ac.object_type) { 396 case AOT_AAC_MAIN: 397 case AOT_AAC_LC: 398 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config)) 399 return -1; 400 break; 401 default: 402 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n", 403 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type); 404 return -1; 405 } 406 return 0; 407} 408 409/** 410 * linear congruential pseudorandom number generator 411 * 412 * @param previous_val pointer to the current state of the generator 413 * 414 * @return Returns a 32-bit pseudorandom integer 415 */ 416static av_always_inline int lcg_random(int previous_val) { 417 return previous_val * 1664525 + 1013904223; 418} 419 420static void reset_predict_state(PredictorState * ps) { 421 ps->r0 = 0.0f; 422 ps->r1 = 0.0f; 423 ps->cor0 = 0.0f; 424 ps->cor1 = 0.0f; 425 ps->var0 = 1.0f; 426 ps->var1 = 1.0f; 427} 428 429static void reset_all_predictors(PredictorState * ps) { 430 int i; 431 for (i = 0; i < MAX_PREDICTORS; i++) 432 reset_predict_state(&ps[i]); 433} 434 435static void reset_predictor_group(PredictorState * ps, int group_num) { 436 int i; 437 for (i = group_num-1; i < MAX_PREDICTORS; i+=30) 438 reset_predict_state(&ps[i]); 439} 440 441static av_cold int aac_decode_init(AVCodecContext * avccontext) { 442 AACContext * ac = avccontext->priv_data; 443 int i; 444 445 ac->avccontext = avccontext; 446 447 if (avccontext->extradata_size > 0) { 448 if(decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size)) 449 return -1; 450 avccontext->sample_rate = ac->m4ac.sample_rate; 451 } else if (avccontext->channels > 0) { 452 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]; 453 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])); 454 if(set_default_channel_config(ac, new_che_pos, avccontext->channels - (avccontext->channels == 8))) 455 return -1; 456 if(output_configure(ac, ac->che_pos, new_che_pos, 1)) 457 return -1; 458 ac->m4ac.sample_rate = avccontext->sample_rate; 459 } else { 460 ff_log_missing_feature(ac->avccontext, "Implicit channel configuration is", 0); 461 return -1; 462 } 463 464 avccontext->sample_fmt = SAMPLE_FMT_S16; 465 avccontext->frame_size = 1024; 466 467 AAC_INIT_VLC_STATIC( 0, 144); 468 AAC_INIT_VLC_STATIC( 1, 114); 469 AAC_INIT_VLC_STATIC( 2, 188); 470 AAC_INIT_VLC_STATIC( 3, 180); 471 AAC_INIT_VLC_STATIC( 4, 172); 472 AAC_INIT_VLC_STATIC( 5, 140); 473 AAC_INIT_VLC_STATIC( 6, 168); 474 AAC_INIT_VLC_STATIC( 7, 114); 475 AAC_INIT_VLC_STATIC( 8, 262); 476 AAC_INIT_VLC_STATIC( 9, 248); 477 AAC_INIT_VLC_STATIC(10, 384); 478 479 dsputil_init(&ac->dsp, avccontext); 480 481 ac->random_state = 0x1f2e3d4c; 482 483 // -1024 - Compensate wrong IMDCT method. 484 // 32768 - Required to scale values to the correct range for the bias method 485 // for float to int16 conversion. 486 487 if(ac->dsp.float_to_int16 == ff_float_to_int16_c) { 488 ac->add_bias = 385.0f; 489 ac->sf_scale = 1. / (-1024. * 32768.); 490 ac->sf_offset = 0; 491 } else { 492 ac->add_bias = 0.0f; 493 ac->sf_scale = 1. / -1024.; 494 ac->sf_offset = 60; 495 } 496 497#if !CONFIG_HARDCODED_TABLES 498 for (i = 0; i < 428; i++) 499 ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.); 500#endif /* CONFIG_HARDCODED_TABLES */ 501 502 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code), 503 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]), 504 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]), 505 352); 506 507 ff_mdct_init(&ac->mdct, 11, 1); 508 ff_mdct_init(&ac->mdct_small, 8, 1); 509 // window initialization 510 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024); 511 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128); 512 ff_sine_window_init(ff_sine_1024, 1024); 513 ff_sine_window_init(ff_sine_128, 128); 514 515 return 0; 516} 517 518/** 519 * Skip data_stream_element; reference: table 4.10. 520 */ 521static void skip_data_stream_element(GetBitContext * gb) { 522 int byte_align = get_bits1(gb); 523 int count = get_bits(gb, 8); 524 if (count == 255) 525 count += get_bits(gb, 8); 526 if (byte_align) 527 align_get_bits(gb); 528 skip_bits_long(gb, 8 * count); 529} 530 531static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) { 532 int sfb; 533 if (get_bits1(gb)) { 534 ics->predictor_reset_group = get_bits(gb, 5); 535 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) { 536 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n"); 537 return -1; 538 } 539 } 540 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) { 541 ics->prediction_used[sfb] = get_bits1(gb); 542 } 543 return 0; 544} 545 546/** 547 * Decode Individual Channel Stream info; reference: table 4.6. 548 * 549 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information. 550 */ 551static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) { 552 if (get_bits1(gb)) { 553 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n"); 554 memset(ics, 0, sizeof(IndividualChannelStream)); 555 return -1; 556 } 557 ics->window_sequence[1] = ics->window_sequence[0]; 558 ics->window_sequence[0] = get_bits(gb, 2); 559 ics->use_kb_window[1] = ics->use_kb_window[0]; 560 ics->use_kb_window[0] = get_bits1(gb); 561 ics->num_window_groups = 1; 562 ics->group_len[0] = 1; 563 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) { 564 int i; 565 ics->max_sfb = get_bits(gb, 4); 566 for (i = 0; i < 7; i++) { 567 if (get_bits1(gb)) { 568 ics->group_len[ics->num_window_groups-1]++; 569 } else { 570 ics->num_window_groups++; 571 ics->group_len[ics->num_window_groups-1] = 1; 572 } 573 } 574 ics->num_windows = 8; 575 ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index]; 576 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index]; 577 ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index]; 578 ics->predictor_present = 0; 579 } else { 580 ics->max_sfb = get_bits(gb, 6); 581 ics->num_windows = 1; 582 ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index]; 583 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index]; 584 ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index]; 585 ics->predictor_present = get_bits1(gb); 586 ics->predictor_reset_group = 0; 587 if (ics->predictor_present) { 588 if (ac->m4ac.object_type == AOT_AAC_MAIN) { 589 if (decode_prediction(ac, ics, gb)) { 590 memset(ics, 0, sizeof(IndividualChannelStream)); 591 return -1; 592 } 593 } else if (ac->m4ac.object_type == AOT_AAC_LC) { 594 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n"); 595 memset(ics, 0, sizeof(IndividualChannelStream)); 596 return -1; 597 } else { 598 ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1); 599 memset(ics, 0, sizeof(IndividualChannelStream)); 600 return -1; 601 } 602 } 603 } 604 605 if(ics->max_sfb > ics->num_swb) { 606 av_log(ac->avccontext, AV_LOG_ERROR, 607 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n", 608 ics->max_sfb, ics->num_swb); 609 memset(ics, 0, sizeof(IndividualChannelStream)); 610 return -1; 611 } 612 613 return 0; 614} 615 616/** 617 * Decode band types (section_data payload); reference: table 4.46. 618 * 619 * @param band_type array of the used band type 620 * @param band_type_run_end array of the last scalefactor band of a band type run 621 * 622 * @return Returns error status. 0 - OK, !0 - error 623 */ 624static int decode_band_types(AACContext * ac, enum BandType band_type[120], 625 int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) { 626 int g, idx = 0; 627 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5; 628 for (g = 0; g < ics->num_window_groups; g++) { 629 int k = 0; 630 while (k < ics->max_sfb) { 631 uint8_t sect_len = k; 632 int sect_len_incr; 633 int sect_band_type = get_bits(gb, 4); 634 if (sect_band_type == 12) { 635 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n"); 636 return -1; 637 } 638 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1) 639 sect_len += sect_len_incr; 640 sect_len += sect_len_incr; 641 if (sect_len > ics->max_sfb) { 642 av_log(ac->avccontext, AV_LOG_ERROR, 643 "Number of bands (%d) exceeds limit (%d).\n", 644 sect_len, ics->max_sfb); 645 return -1; 646 } 647 for (; k < sect_len; k++) { 648 band_type [idx] = sect_band_type; 649 band_type_run_end[idx++] = sect_len; 650 } 651 } 652 } 653 return 0; 654} 655 656/** 657 * Decode scalefactors; reference: table 4.47. 658 * 659 * @param global_gain first scalefactor value as scalefactors are differentially coded 660 * @param band_type array of the used band type 661 * @param band_type_run_end array of the last scalefactor band of a band type run 662 * @param sf array of scalefactors or intensity stereo positions 663 * 664 * @return Returns error status. 0 - OK, !0 - error 665 */ 666static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb, 667 unsigned int global_gain, IndividualChannelStream * ics, 668 enum BandType band_type[120], int band_type_run_end[120]) { 669 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0); 670 int g, i, idx = 0; 671 int offset[3] = { global_gain, global_gain - 90, 100 }; 672 int noise_flag = 1; 673 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" }; 674 for (g = 0; g < ics->num_window_groups; g++) { 675 for (i = 0; i < ics->max_sfb;) { 676 int run_end = band_type_run_end[idx]; 677 if (band_type[idx] == ZERO_BT) { 678 for(; i < run_end; i++, idx++) 679 sf[idx] = 0.; 680 }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) { 681 for(; i < run_end; i++, idx++) { 682 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60; 683 if(offset[2] > 255U) { 684 av_log(ac->avccontext, AV_LOG_ERROR, 685 "%s (%d) out of range.\n", sf_str[2], offset[2]); 686 return -1; 687 } 688 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300]; 689 } 690 }else if(band_type[idx] == NOISE_BT) { 691 for(; i < run_end; i++, idx++) { 692 if(noise_flag-- > 0) 693 offset[1] += get_bits(gb, 9) - 256; 694 else 695 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60; 696 if(offset[1] > 255U) { 697 av_log(ac->avccontext, AV_LOG_ERROR, 698 "%s (%d) out of range.\n", sf_str[1], offset[1]); 699 return -1; 700 } 701 sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100]; 702 } 703 }else { 704 for(; i < run_end; i++, idx++) { 705 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60; 706 if(offset[0] > 255U) { 707 av_log(ac->avccontext, AV_LOG_ERROR, 708 "%s (%d) out of range.\n", sf_str[0], offset[0]); 709 return -1; 710 } 711 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset]; 712 } 713 } 714 } 715 } 716 return 0; 717} 718 719/** 720 * Decode pulse data; reference: table 4.7. 721 */ 722static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) { 723 int i, pulse_swb; 724 pulse->num_pulse = get_bits(gb, 2) + 1; 725 pulse_swb = get_bits(gb, 6); 726 if (pulse_swb >= num_swb) 727 return -1; 728 pulse->pos[0] = swb_offset[pulse_swb]; 729 pulse->pos[0] += get_bits(gb, 5); 730 if (pulse->pos[0] > 1023) 731 return -1; 732 pulse->amp[0] = get_bits(gb, 4); 733 for (i = 1; i < pulse->num_pulse; i++) { 734 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1]; 735 if (pulse->pos[i] > 1023) 736 return -1; 737 pulse->amp[i] = get_bits(gb, 4); 738 } 739 return 0; 740} 741 742/** 743 * Decode Temporal Noise Shaping data; reference: table 4.48. 744 * 745 * @return Returns error status. 0 - OK, !0 - error 746 */ 747static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns, 748 GetBitContext * gb, const IndividualChannelStream * ics) { 749 int w, filt, i, coef_len, coef_res, coef_compress; 750 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE; 751 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12; 752 for (w = 0; w < ics->num_windows; w++) { 753 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) { 754 coef_res = get_bits1(gb); 755 756 for (filt = 0; filt < tns->n_filt[w]; filt++) { 757 int tmp2_idx; 758 tns->length[w][filt] = get_bits(gb, 6 - 2*is8); 759 760 if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) { 761 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.", 762 tns->order[w][filt], tns_max_order); 763 tns->order[w][filt] = 0; 764 return -1; 765 } 766 if (tns->order[w][filt]) { 767 tns->direction[w][filt] = get_bits1(gb); 768 coef_compress = get_bits1(gb); 769 coef_len = coef_res + 3 - coef_compress; 770 tmp2_idx = 2*coef_compress + coef_res; 771 772 for (i = 0; i < tns->order[w][filt]; i++) 773 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)]; 774 } 775 } 776 } 777 } 778 return 0; 779} 780 781/** 782 * Decode Mid/Side data; reference: table 4.54. 783 * 784 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s; 785 * [1] mask is decoded from bitstream; [2] mask is all 1s; 786 * [3] reserved for scalable AAC 787 */ 788static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb, 789 int ms_present) { 790 int idx; 791 if (ms_present == 1) { 792 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++) 793 cpe->ms_mask[idx] = get_bits1(gb); 794 } else if (ms_present == 2) { 795 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0])); 796 } 797} 798 799/** 800 * Decode spectral data; reference: table 4.50. 801 * Dequantize and scale spectral data; reference: 4.6.3.3. 802 * 803 * @param coef array of dequantized, scaled spectral data 804 * @param sf array of scalefactors or intensity stereo positions 805 * @param pulse_present set if pulses are present 806 * @param pulse pointer to pulse data struct 807 * @param band_type array of the used band type 808 * 809 * @return Returns error status. 0 - OK, !0 - error 810 */ 811static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120], 812 int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) { 813 int i, k, g, idx = 0; 814 const int c = 1024/ics->num_windows; 815 const uint16_t * offsets = ics->swb_offset; 816 float *coef_base = coef; 817 static const float sign_lookup[] = { 1.0f, -1.0f }; 818 819 for (g = 0; g < ics->num_windows; g++) 820 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb])); 821 822 for (g = 0; g < ics->num_window_groups; g++) { 823 for (i = 0; i < ics->max_sfb; i++, idx++) { 824 const int cur_band_type = band_type[idx]; 825 const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4; 826 const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type); 827 int group; 828 if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) { 829 for (group = 0; group < ics->group_len[g]; group++) { 830 memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float)); 831 } 832 }else if (cur_band_type == NOISE_BT) { 833 for (group = 0; group < ics->group_len[g]; group++) { 834 float scale; 835 float band_energy = 0; 836 for (k = offsets[i]; k < offsets[i+1]; k++) { 837 ac->random_state = lcg_random(ac->random_state); 838 coef[group*128+k] = ac->random_state; 839 band_energy += coef[group*128+k]*coef[group*128+k]; 840 } 841 scale = sf[idx] / sqrtf(band_energy); 842 for (k = offsets[i]; k < offsets[i+1]; k++) { 843 coef[group*128+k] *= scale; 844 } 845 } 846 }else { 847 for (group = 0; group < ics->group_len[g]; group++) { 848 for (k = offsets[i]; k < offsets[i+1]; k += dim) { 849 const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3); 850 const int coef_tmp_idx = (group << 7) + k; 851 const float *vq_ptr; 852 int j; 853 if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) { 854 av_log(ac->avccontext, AV_LOG_ERROR, 855 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n", 856 cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]); 857 return -1; 858 } 859 vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim]; 860 if (is_cb_unsigned) { 861 if (vq_ptr[0]) coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)]; 862 if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)]; 863 if (dim == 4) { 864 if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)]; 865 if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)]; 866 } 867 if (cur_band_type == ESC_BT) { 868 for (j = 0; j < 2; j++) { 869 if (vq_ptr[j] == 64.0f) { 870 int n = 4; 871 /* The total length of escape_sequence must be < 22 bits according 872 to the specification (i.e. max is 11111111110xxxxxxxxxx). */ 873 while (get_bits1(gb) && n < 15) n++; 874 if(n == 15) { 875 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n"); 876 return -1; 877 } 878 n = (1<<n) + get_bits(gb, n); 879 coef[coef_tmp_idx + j] *= cbrtf(n) * n; 880 }else 881 coef[coef_tmp_idx + j] *= vq_ptr[j]; 882 } 883 }else 884 { 885 coef[coef_tmp_idx ] *= vq_ptr[0]; 886 coef[coef_tmp_idx + 1] *= vq_ptr[1]; 887 if (dim == 4) { 888 coef[coef_tmp_idx + 2] *= vq_ptr[2]; 889 coef[coef_tmp_idx + 3] *= vq_ptr[3]; 890 } 891 } 892 }else { 893 coef[coef_tmp_idx ] = vq_ptr[0]; 894 coef[coef_tmp_idx + 1] = vq_ptr[1]; 895 if (dim == 4) { 896 coef[coef_tmp_idx + 2] = vq_ptr[2]; 897 coef[coef_tmp_idx + 3] = vq_ptr[3]; 898 } 899 } 900 coef[coef_tmp_idx ] *= sf[idx]; 901 coef[coef_tmp_idx + 1] *= sf[idx]; 902 if (dim == 4) { 903 coef[coef_tmp_idx + 2] *= sf[idx]; 904 coef[coef_tmp_idx + 3] *= sf[idx]; 905 } 906 } 907 } 908 } 909 } 910 coef += ics->group_len[g]<<7; 911 } 912 913 if (pulse_present) { 914 idx = 0; 915 for(i = 0; i < pulse->num_pulse; i++){ 916 float co = coef_base[ pulse->pos[i] ]; 917 while(offsets[idx + 1] <= pulse->pos[i]) 918 idx++; 919 if (band_type[idx] != NOISE_BT && sf[idx]) { 920 float ico = -pulse->amp[i]; 921 if (co) { 922 co /= sf[idx]; 923 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico); 924 } 925 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx]; 926 } 927 } 928 } 929 return 0; 930} 931 932static av_always_inline float flt16_round(float pf) { 933 int exp; 934 pf = frexpf(pf, &exp); 935 pf = ldexpf(roundf(ldexpf(pf, 8)), exp-8); 936 return pf; 937} 938 939static av_always_inline float flt16_even(float pf) { 940 int exp; 941 pf = frexpf(pf, &exp); 942 pf = ldexpf(rintf(ldexpf(pf, 8)), exp-8); 943 return pf; 944} 945 946 947// Foxconn, added by Michael J. 948av_always_inline av_const float truncf(float x) 949{ 950 return (x > 0) ? floor(x) : ceil(x); 951} 952 953 954static av_always_inline float flt16_trunc(float pf) { 955 int exp; 956 pf = frexpf(pf, &exp); 957 pf = ldexpf(truncf(ldexpf(pf, 8)), exp-8); 958 return pf; 959} 960 961static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) { 962 const float a = 0.953125; // 61.0/64 963 const float alpha = 0.90625; // 29.0/32 964 float e0, e1; 965 float pv; 966 float k1, k2; 967 968 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0; 969 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0; 970 971 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1); 972 if (output_enable) 973 *coef += pv * ac->sf_scale; 974 975 e0 = *coef / ac->sf_scale; 976 e1 = e0 - k1 * ps->r0; 977 978 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1); 979 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1)); 980 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0); 981 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0)); 982 983 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0)); 984 ps->r0 = flt16_trunc(a * e0); 985} 986 987/** 988 * Apply AAC-Main style frequency domain prediction. 989 */ 990static void apply_prediction(AACContext * ac, SingleChannelElement * sce) { 991 int sfb, k; 992 993 if (!sce->ics.predictor_initialized) { 994 reset_all_predictors(sce->predictor_state); 995 sce->ics.predictor_initialized = 1; 996 } 997 998 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) { 999 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) { 1000 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) { 1001 predict(ac, &sce->predictor_state[k], &sce->coeffs[k], 1002 sce->ics.predictor_present && sce->ics.prediction_used[sfb]); 1003 } 1004 } 1005 if (sce->ics.predictor_reset_group) 1006 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group); 1007 } else 1008 reset_all_predictors(sce->predictor_state); 1009} 1010 1011/** 1012 * Decode an individual_channel_stream payload; reference: table 4.44. 1013 * 1014 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information. 1015 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.) 1016 * 1017 * @return Returns error status. 0 - OK, !0 - error 1018 */ 1019static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) { 1020 Pulse pulse; 1021 TemporalNoiseShaping * tns = &sce->tns; 1022 IndividualChannelStream * ics = &sce->ics; 1023 float * out = sce->coeffs; 1024 int global_gain, pulse_present = 0; 1025 1026 /* This assignment is to silence a GCC warning about the variable being used 1027 * uninitialized when in fact it always is. 1028 */ 1029 pulse.num_pulse = 0; 1030 1031 global_gain = get_bits(gb, 8); 1032 1033 if (!common_window && !scale_flag) { 1034 if (decode_ics_info(ac, ics, gb, 0) < 0) 1035 return -1; 1036 } 1037 1038 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0) 1039 return -1; 1040 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0) 1041 return -1; 1042 1043 pulse_present = 0; 1044 if (!scale_flag) { 1045 if ((pulse_present = get_bits1(gb))) { 1046 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) { 1047 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n"); 1048 return -1; 1049 } 1050 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) { 1051 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n"); 1052 return -1; 1053 } 1054 } 1055 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics)) 1056 return -1; 1057 if (get_bits1(gb)) { 1058 ff_log_missing_feature(ac->avccontext, "SSR", 1); 1059 return -1; 1060 } 1061 } 1062 1063 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0) 1064 return -1; 1065 1066 if(ac->m4ac.object_type == AOT_AAC_MAIN && !common_window) 1067 apply_prediction(ac, sce); 1068 1069 return 0; 1070} 1071 1072/** 1073 * Mid/Side stereo decoding; reference: 4.6.8.1.3. 1074 */ 1075static void apply_mid_side_stereo(ChannelElement * cpe) { 1076 const IndividualChannelStream * ics = &cpe->ch[0].ics; 1077 float *ch0 = cpe->ch[0].coeffs; 1078 float *ch1 = cpe->ch[1].coeffs; 1079 int g, i, k, group, idx = 0; 1080 const uint16_t * offsets = ics->swb_offset; 1081 for (g = 0; g < ics->num_window_groups; g++) { 1082 for (i = 0; i < ics->max_sfb; i++, idx++) { 1083 if (cpe->ms_mask[idx] && 1084 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) { 1085 for (group = 0; group < ics->group_len[g]; group++) { 1086 for (k = offsets[i]; k < offsets[i+1]; k++) { 1087 float tmp = ch0[group*128 + k] - ch1[group*128 + k]; 1088 ch0[group*128 + k] += ch1[group*128 + k]; 1089 ch1[group*128 + k] = tmp; 1090 } 1091 } 1092 } 1093 } 1094 ch0 += ics->group_len[g]*128; 1095 ch1 += ics->group_len[g]*128; 1096 } 1097} 1098 1099/** 1100 * intensity stereo decoding; reference: 4.6.8.2.3 1101 * 1102 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s; 1103 * [1] mask is decoded from bitstream; [2] mask is all 1s; 1104 * [3] reserved for scalable AAC 1105 */ 1106static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) { 1107 const IndividualChannelStream * ics = &cpe->ch[1].ics; 1108 SingleChannelElement * sce1 = &cpe->ch[1]; 1109 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs; 1110 const uint16_t * offsets = ics->swb_offset; 1111 int g, group, i, k, idx = 0; 1112 int c; 1113 float scale; 1114 for (g = 0; g < ics->num_window_groups; g++) { 1115 for (i = 0; i < ics->max_sfb;) { 1116 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) { 1117 const int bt_run_end = sce1->band_type_run_end[idx]; 1118 for (; i < bt_run_end; i++, idx++) { 1119 c = -1 + 2 * (sce1->band_type[idx] - 14); 1120 if (ms_present) 1121 c *= 1 - 2 * cpe->ms_mask[idx]; 1122 scale = c * sce1->sf[idx]; 1123 for (group = 0; group < ics->group_len[g]; group++) 1124 for (k = offsets[i]; k < offsets[i+1]; k++) 1125 coef1[group*128 + k] = scale * coef0[group*128 + k]; 1126 } 1127 } else { 1128 int bt_run_end = sce1->band_type_run_end[idx]; 1129 idx += bt_run_end - i; 1130 i = bt_run_end; 1131 } 1132 } 1133 coef0 += ics->group_len[g]*128; 1134 coef1 += ics->group_len[g]*128; 1135 } 1136} 1137 1138/** 1139 * Decode a channel_pair_element; reference: table 4.4. 1140 * 1141 * @param elem_id Identifies the instance of a syntax element. 1142 * 1143 * @return Returns error status. 0 - OK, !0 - error 1144 */ 1145static int decode_cpe(AACContext * ac, GetBitContext * gb, ChannelElement * cpe) { 1146 int i, ret, common_window, ms_present = 0; 1147 1148 common_window = get_bits1(gb); 1149 if (common_window) { 1150 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1)) 1151 return -1; 1152 i = cpe->ch[1].ics.use_kb_window[0]; 1153 cpe->ch[1].ics = cpe->ch[0].ics; 1154 cpe->ch[1].ics.use_kb_window[1] = i; 1155 ms_present = get_bits(gb, 2); 1156 if(ms_present == 3) { 1157 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n"); 1158 return -1; 1159 } else if(ms_present) 1160 decode_mid_side_stereo(cpe, gb, ms_present); 1161 } 1162 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0))) 1163 return ret; 1164 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0))) 1165 return ret; 1166 1167 if (common_window) { 1168 if (ms_present) 1169 apply_mid_side_stereo(cpe); 1170 if (ac->m4ac.object_type == AOT_AAC_MAIN) { 1171 apply_prediction(ac, &cpe->ch[0]); 1172 apply_prediction(ac, &cpe->ch[1]); 1173 } 1174 } 1175 1176 apply_intensity_stereo(cpe, ms_present); 1177 return 0; 1178} 1179 1180/** 1181 * Decode coupling_channel_element; reference: table 4.8. 1182 * 1183 * @param elem_id Identifies the instance of a syntax element. 1184 * 1185 * @return Returns error status. 0 - OK, !0 - error 1186 */ 1187static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) { 1188 int num_gain = 0; 1189 int c, g, sfb, ret; 1190 int sign; 1191 float scale; 1192 SingleChannelElement * sce = &che->ch[0]; 1193 ChannelCoupling * coup = &che->coup; 1194 1195 coup->coupling_point = 2*get_bits1(gb); 1196 coup->num_coupled = get_bits(gb, 3); 1197 for (c = 0; c <= coup->num_coupled; c++) { 1198 num_gain++; 1199 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE; 1200 coup->id_select[c] = get_bits(gb, 4); 1201 if (coup->type[c] == TYPE_CPE) { 1202 coup->ch_select[c] = get_bits(gb, 2); 1203 if (coup->ch_select[c] == 3) 1204 num_gain++; 1205 } else 1206 coup->ch_select[c] = 2; 1207 } 1208 coup->coupling_point += get_bits1(gb); 1209 1210 if (coup->coupling_point == 2) { 1211 av_log(ac->avccontext, AV_LOG_ERROR, 1212 "Independently switched CCE with 'invalid' domain signalled.\n"); 1213 memset(coup, 0, sizeof(ChannelCoupling)); 1214 return -1; 1215 } 1216 1217 sign = get_bits(gb, 1); 1218 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3)); 1219 1220 if ((ret = decode_ics(ac, sce, gb, 0, 0))) 1221 return ret; 1222 1223 for (c = 0; c < num_gain; c++) { 1224 int idx = 0; 1225 int cge = 1; 1226 int gain = 0; 1227 float gain_cache = 1.; 1228 if (c) { 1229 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb); 1230 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0; 1231 gain_cache = pow(scale, -gain); 1232 } 1233 if (coup->coupling_point == AFTER_IMDCT) { 1234 coup->gain[c][0] = gain_cache; 1235 } else { 1236 for (g = 0; g < sce->ics.num_window_groups; g++) { 1237 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) { 1238 if (sce->band_type[idx] != ZERO_BT) { 1239 if (!cge) { 1240 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60; 1241 if (t) { 1242 int s = 1; 1243 t = gain += t; 1244 if (sign) { 1245 s -= 2 * (t & 0x1); 1246 t >>= 1; 1247 } 1248 gain_cache = pow(scale, -t) * s; 1249 } 1250 } 1251 coup->gain[c][idx] = gain_cache; 1252 } 1253 } 1254 } 1255 } 1256 } 1257 return 0; 1258} 1259 1260/** 1261 * Decode Spectral Band Replication extension data; reference: table 4.55. 1262 * 1263 * @param crc flag indicating the presence of CRC checksum 1264 * @param cnt length of TYPE_FIL syntactic element in bytes 1265 * 1266 * @return Returns number of bytes consumed from the TYPE_FIL element. 1267 */ 1268static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) { 1269 // TODO : sbr_extension implementation 1270 ff_log_missing_feature(ac->avccontext, "SBR", 0); 1271 skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type 1272 return cnt; 1273} 1274 1275/** 1276 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53. 1277 * 1278 * @return Returns number of bytes consumed. 1279 */ 1280static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) { 1281 int i; 1282 int num_excl_chan = 0; 1283 1284 do { 1285 for (i = 0; i < 7; i++) 1286 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb); 1287 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb)); 1288 1289 return num_excl_chan / 7; 1290} 1291 1292/** 1293 * Decode dynamic range information; reference: table 4.52. 1294 * 1295 * @param cnt length of TYPE_FIL syntactic element in bytes 1296 * 1297 * @return Returns number of bytes consumed. 1298 */ 1299static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) { 1300 int n = 1; 1301 int drc_num_bands = 1; 1302 int i; 1303 1304 /* pce_tag_present? */ 1305 if(get_bits1(gb)) { 1306 che_drc->pce_instance_tag = get_bits(gb, 4); 1307 skip_bits(gb, 4); // tag_reserved_bits 1308 n++; 1309 } 1310 1311 /* excluded_chns_present? */ 1312 if(get_bits1(gb)) { 1313 n += decode_drc_channel_exclusions(che_drc, gb); 1314 } 1315 1316 /* drc_bands_present? */ 1317 if (get_bits1(gb)) { 1318 che_drc->band_incr = get_bits(gb, 4); 1319 che_drc->interpolation_scheme = get_bits(gb, 4); 1320 n++; 1321 drc_num_bands += che_drc->band_incr; 1322 for (i = 0; i < drc_num_bands; i++) { 1323 che_drc->band_top[i] = get_bits(gb, 8); 1324 n++; 1325 } 1326 } 1327 1328 /* prog_ref_level_present? */ 1329 if (get_bits1(gb)) { 1330 che_drc->prog_ref_level = get_bits(gb, 7); 1331 skip_bits1(gb); // prog_ref_level_reserved_bits 1332 n++; 1333 } 1334 1335 for (i = 0; i < drc_num_bands; i++) { 1336 che_drc->dyn_rng_sgn[i] = get_bits1(gb); 1337 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7); 1338 n++; 1339 } 1340 1341 return n; 1342} 1343 1344/** 1345 * Decode extension data (incomplete); reference: table 4.51. 1346 * 1347 * @param cnt length of TYPE_FIL syntactic element in bytes 1348 * 1349 * @return Returns number of bytes consumed 1350 */ 1351static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) { 1352 int crc_flag = 0; 1353 int res = cnt; 1354 switch (get_bits(gb, 4)) { // extension type 1355 case EXT_SBR_DATA_CRC: 1356 crc_flag++; 1357 case EXT_SBR_DATA: 1358 res = decode_sbr_extension(ac, gb, crc_flag, cnt); 1359 break; 1360 case EXT_DYNAMIC_RANGE: 1361 res = decode_dynamic_range(&ac->che_drc, gb, cnt); 1362 break; 1363 case EXT_FILL: 1364 case EXT_FILL_DATA: 1365 case EXT_DATA_ELEMENT: 1366 default: 1367 skip_bits_long(gb, 8*cnt - 4); 1368 break; 1369 }; 1370 return res; 1371} 1372 1373/** 1374 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3. 1375 * 1376 * @param decode 1 if tool is used normally, 0 if tool is used in LTP. 1377 * @param coef spectral coefficients 1378 */ 1379static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) { 1380 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb); 1381 int w, filt, m, i; 1382 int bottom, top, order, start, end, size, inc; 1383 float lpc[TNS_MAX_ORDER]; 1384 1385 for (w = 0; w < ics->num_windows; w++) { 1386 bottom = ics->num_swb; 1387 for (filt = 0; filt < tns->n_filt[w]; filt++) { 1388 top = bottom; 1389 bottom = FFMAX(0, top - tns->length[w][filt]); 1390 order = tns->order[w][filt]; 1391 if (order == 0) 1392 continue; 1393 1394 // tns_decode_coef 1395 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0); 1396 1397 start = ics->swb_offset[FFMIN(bottom, mmm)]; 1398 end = ics->swb_offset[FFMIN( top, mmm)]; 1399 if ((size = end - start) <= 0) 1400 continue; 1401 if (tns->direction[w][filt]) { 1402 inc = -1; start = end - 1; 1403 } else { 1404 inc = 1; 1405 } 1406 start += w * 128; 1407 1408 // ar filter 1409 for (m = 0; m < size; m++, start += inc) 1410 for (i = 1; i <= FFMIN(m, order); i++) 1411 coef[start] -= coef[start - i*inc] * lpc[i-1]; 1412 } 1413 } 1414} 1415 1416/** 1417 * Conduct IMDCT and windowing. 1418 */ 1419static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) { 1420 IndividualChannelStream * ics = &sce->ics; 1421 float * in = sce->coeffs; 1422 float * out = sce->ret; 1423 float * saved = sce->saved; 1424 const float * swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128; 1425 const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024; 1426 const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128; 1427 float * buf = ac->buf_mdct; 1428 float * temp = ac->temp; 1429 int i; 1430 1431 // imdct 1432 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) { 1433 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) 1434 av_log(ac->avccontext, AV_LOG_WARNING, 1435 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. " 1436 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n"); 1437 for (i = 0; i < 1024; i += 128) 1438 ff_imdct_half(&ac->mdct_small, buf + i, in + i); 1439 } else 1440 ff_imdct_half(&ac->mdct, buf, in); 1441 1442 /* window overlapping 1443 * NOTE: To simplify the overlapping code, all 'meaningless' short to long 1444 * and long to short transitions are considered to be short to short 1445 * transitions. This leaves just two cases (long to long and short to short) 1446 * with a little special sauce for EIGHT_SHORT_SEQUENCE. 1447 */ 1448 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) && 1449 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) { 1450 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, ac->add_bias, 512); 1451 } else { 1452 for (i = 0; i < 448; i++) 1453 out[i] = saved[i] + ac->add_bias; 1454 1455 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) { 1456 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, ac->add_bias, 64); 1457 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, ac->add_bias, 64); 1458 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, ac->add_bias, 64); 1459 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, ac->add_bias, 64); 1460 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, ac->add_bias, 64); 1461 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float)); 1462 } else { 1463 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, ac->add_bias, 64); 1464 for (i = 576; i < 1024; i++) 1465 out[i] = buf[i-512] + ac->add_bias; 1466 } 1467 } 1468 1469 // buffer update 1470 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) { 1471 for (i = 0; i < 64; i++) 1472 saved[i] = temp[64 + i] - ac->add_bias; 1473 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64); 1474 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64); 1475 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64); 1476 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float)); 1477 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) { 1478 memcpy( saved, buf + 512, 448 * sizeof(float)); 1479 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float)); 1480 } else { // LONG_STOP or ONLY_LONG 1481 memcpy( saved, buf + 512, 512 * sizeof(float)); 1482 } 1483} 1484 1485/** 1486 * Apply dependent channel coupling (applied before IMDCT). 1487 * 1488 * @param index index into coupling gain array 1489 */ 1490static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) { 1491 IndividualChannelStream * ics = &cce->ch[0].ics; 1492 const uint16_t * offsets = ics->swb_offset; 1493 float * dest = target->coeffs; 1494 const float * src = cce->ch[0].coeffs; 1495 int g, i, group, k, idx = 0; 1496 if(ac->m4ac.object_type == AOT_AAC_LTP) { 1497 av_log(ac->avccontext, AV_LOG_ERROR, 1498 "Dependent coupling is not supported together with LTP\n"); 1499 return; 1500 } 1501 for (g = 0; g < ics->num_window_groups; g++) { 1502 for (i = 0; i < ics->max_sfb; i++, idx++) { 1503 if (cce->ch[0].band_type[idx] != ZERO_BT) { 1504 for (group = 0; group < ics->group_len[g]; group++) { 1505 for (k = offsets[i]; k < offsets[i+1]; k++) { 1506 // XXX dsputil-ize 1507 dest[group*128+k] += cce->coup.gain[index][idx] * src[group*128+k]; 1508 } 1509 } 1510 } 1511 } 1512 dest += ics->group_len[g]*128; 1513 src += ics->group_len[g]*128; 1514 } 1515} 1516 1517/** 1518 * Apply independent channel coupling (applied after IMDCT). 1519 * 1520 * @param index index into coupling gain array 1521 */ 1522static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) { 1523 int i; 1524 const float gain = cce->coup.gain[index][0]; 1525 const float bias = ac->add_bias; 1526 const float* src = cce->ch[0].ret; 1527 float* dest = target->ret; 1528 1529 for (i = 0; i < 1024; i++) 1530 dest[i] += gain * (src[i] - bias); 1531} 1532 1533/** 1534 * channel coupling transformation interface 1535 * 1536 * @param index index into coupling gain array 1537 * @param apply_coupling_method pointer to (in)dependent coupling function 1538 */ 1539static void apply_channel_coupling(AACContext * ac, ChannelElement * cc, 1540 enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point, 1541 void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index)) 1542{ 1543 int i, c; 1544 1545 for (i = 0; i < MAX_ELEM_ID; i++) { 1546 ChannelElement *cce = ac->che[TYPE_CCE][i]; 1547 int index = 0; 1548 1549 if (cce && cce->coup.coupling_point == coupling_point) { 1550 ChannelCoupling * coup = &cce->coup; 1551 1552 for (c = 0; c <= coup->num_coupled; c++) { 1553 if (coup->type[c] == type && coup->id_select[c] == elem_id) { 1554 if (coup->ch_select[c] != 1) { 1555 apply_coupling_method(ac, &cc->ch[0], cce, index); 1556 if (coup->ch_select[c] != 0) 1557 index++; 1558 } 1559 if (coup->ch_select[c] != 2) 1560 apply_coupling_method(ac, &cc->ch[1], cce, index++); 1561 } else 1562 index += 1 + (coup->ch_select[c] == 3); 1563 } 1564 } 1565 } 1566} 1567 1568/** 1569 * Convert spectral data to float samples, applying all supported tools as appropriate. 1570 */ 1571static void spectral_to_sample(AACContext * ac) { 1572 int i, type; 1573 for(type = 3; type >= 0; type--) { 1574 for (i = 0; i < MAX_ELEM_ID; i++) { 1575 ChannelElement *che = ac->che[type][i]; 1576 if(che) { 1577 if(type <= TYPE_CPE) 1578 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling); 1579 if(che->ch[0].tns.present) 1580 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1); 1581 if(che->ch[1].tns.present) 1582 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1); 1583 if(type <= TYPE_CPE) 1584 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling); 1585 if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) 1586 imdct_and_windowing(ac, &che->ch[0]); 1587 if(type == TYPE_CPE) 1588 imdct_and_windowing(ac, &che->ch[1]); 1589 if(type <= TYPE_CCE) 1590 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling); 1591 } 1592 } 1593 } 1594} 1595 1596static int parse_adts_frame_header(AACContext * ac, GetBitContext * gb) { 1597 1598 int size; 1599 AACADTSHeaderInfo hdr_info; 1600 1601 size = ff_aac_parse_header(gb, &hdr_info); 1602 if (size > 0) { 1603 if (hdr_info.chan_config) 1604 ac->m4ac.chan_config = hdr_info.chan_config; 1605 ac->m4ac.sample_rate = hdr_info.sample_rate; 1606 ac->m4ac.sampling_index = hdr_info.sampling_index; 1607 ac->m4ac.object_type = hdr_info.object_type; 1608 if (hdr_info.num_aac_frames == 1) { 1609 if (!hdr_info.crc_absent) 1610 skip_bits(gb, 16); 1611 } else { 1612 ff_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0); 1613 return -1; 1614 } 1615 } 1616 return size; 1617} 1618 1619static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, const uint8_t * buf, int buf_size) { 1620 AACContext * ac = avccontext->priv_data; 1621 ChannelElement * che = NULL; 1622 GetBitContext gb; 1623 enum RawDataBlockType elem_type; 1624 int err, elem_id, data_size_tmp; 1625 1626 init_get_bits(&gb, buf, buf_size*8); 1627 1628 if (show_bits(&gb, 12) == 0xfff) { 1629 if ((err = parse_adts_frame_header(ac, &gb)) < 0) { 1630 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n"); 1631 return -1; 1632 } 1633 if (ac->m4ac.sampling_index > 12) { 1634 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index); 1635 return -1; 1636 } 1637 } 1638 1639 // parse 1640 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) { 1641 elem_id = get_bits(&gb, 4); 1642 err = -1; 1643 1644 if(elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) { 1645 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id); 1646 return -1; 1647 } 1648 1649 switch (elem_type) { 1650 1651 case TYPE_SCE: 1652 err = decode_ics(ac, &che->ch[0], &gb, 0, 0); 1653 break; 1654 1655 case TYPE_CPE: 1656 err = decode_cpe(ac, &gb, che); 1657 break; 1658 1659 case TYPE_CCE: 1660 err = decode_cce(ac, &gb, che); 1661 break; 1662 1663 case TYPE_LFE: 1664 err = decode_ics(ac, &che->ch[0], &gb, 0, 0); 1665 break; 1666 1667 case TYPE_DSE: 1668 skip_data_stream_element(&gb); 1669 err = 0; 1670 break; 1671 1672 case TYPE_PCE: 1673 { 1674 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]; 1675 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0])); 1676 if((err = decode_pce(ac, new_che_pos, &gb))) 1677 break; 1678 err = output_configure(ac, ac->che_pos, new_che_pos, 0); 1679 break; 1680 } 1681 1682 case TYPE_FIL: 1683 if (elem_id == 15) 1684 elem_id += get_bits(&gb, 8) - 1; 1685 while (elem_id > 0) 1686 elem_id -= decode_extension_payload(ac, &gb, elem_id); 1687 err = 0; /* FIXME */ 1688 break; 1689 1690 default: 1691 err = -1; /* should not happen, but keeps compiler happy */ 1692 break; 1693 } 1694 1695 if(err) 1696 return err; 1697 } 1698 1699 spectral_to_sample(ac); 1700 1701 if (!ac->is_saved) { 1702 ac->is_saved = 1; 1703 *data_size = 0; 1704 return buf_size; 1705 } 1706 1707 data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t); 1708 if(*data_size < data_size_tmp) { 1709 av_log(avccontext, AV_LOG_ERROR, 1710 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n", 1711 *data_size, data_size_tmp); 1712 return -1; 1713 } 1714 *data_size = data_size_tmp; 1715 1716 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels); 1717 1718 return buf_size; 1719} 1720 1721static av_cold int aac_decode_close(AVCodecContext * avccontext) { 1722 AACContext * ac = avccontext->priv_data; 1723 int i, type; 1724 1725 for (i = 0; i < MAX_ELEM_ID; i++) { 1726 for(type = 0; type < 4; type++) 1727 av_freep(&ac->che[type][i]); 1728 } 1729 1730 ff_mdct_end(&ac->mdct); 1731 ff_mdct_end(&ac->mdct_small); 1732 return 0 ; 1733} 1734 1735AVCodec aac_decoder = { 1736 "aac", 1737 CODEC_TYPE_AUDIO, 1738 CODEC_ID_AAC, 1739 sizeof(AACContext), 1740 aac_decode_init, 1741 NULL, 1742 aac_decode_close, 1743 aac_decode_frame, 1744 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"), 1745 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE}, 1746}; 1747