1/* 2 * MPEG-4 ALS decoder 3 * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ mail.de> 4 * 5 * This file is part of FFmpeg. 6 * 7 * FFmpeg is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * FFmpeg is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with FFmpeg; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22/** 23 * @file 24 * MPEG-4 ALS decoder 25 * @author Thilo Borgmann <thilo.borgmann _at_ mail.de> 26 */ 27 28#include <inttypes.h> 29 30#include "avcodec.h" 31#include "get_bits.h" 32#include "unary.h" 33#include "mpeg4audio.h" 34#include "bgmc.h" 35#include "bswapdsp.h" 36#include "codec_internal.h" 37#include "internal.h" 38#include "mlz.h" 39#include "libavutil/samplefmt.h" 40#include "libavutil/crc.h" 41#include "libavutil/softfloat_ieee754.h" 42#include "libavutil/intfloat.h" 43#include "libavutil/intreadwrite.h" 44 45#include <stdint.h> 46 47/** Rice parameters and corresponding index offsets for decoding the 48 * indices of scaled PARCOR values. The table chosen is set globally 49 * by the encoder and stored in ALSSpecificConfig. 50 */ 51static const int8_t parcor_rice_table[3][20][2] = { 52 { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4}, 53 { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3}, 54 { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2}, 55 { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} }, 56 { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4}, 57 { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4}, 58 {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4}, 59 { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} }, 60 { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4}, 61 { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3}, 62 {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3}, 63 { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} } 64}; 65 66 67/** Scaled PARCOR values used for the first two PARCOR coefficients. 68 * To be indexed by the Rice coded indices. 69 * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20) 70 * Actual values are divided by 32 in order to be stored in 16 bits. 71 */ 72static const int16_t parcor_scaled_values[] = { 73 -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32, 74 -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32, 75 -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32, 76 -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32, 77 -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32, 78 -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32, 79 -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32, 80 -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32, 81 -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32, 82 -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32, 83 -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32, 84 -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32, 85 -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32, 86 -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32, 87 -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32, 88 -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32, 89 -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32, 90 -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32, 91 -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32, 92 -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32, 93 -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32, 94 -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32, 95 -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32, 96 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32, 97 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32, 98 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32, 99 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32, 100 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32, 101 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32, 102 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32, 103 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32, 104 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32 105}; 106 107 108/** Gain values of p(0) for long-term prediction. 109 * To be indexed by the Rice coded indices. 110 */ 111static const uint8_t ltp_gain_values [4][4] = { 112 { 0, 8, 16, 24}, 113 {32, 40, 48, 56}, 114 {64, 70, 76, 82}, 115 {88, 92, 96, 100} 116}; 117 118 119/** Inter-channel weighting factors for multi-channel correlation. 120 * To be indexed by the Rice coded indices. 121 */ 122static const int16_t mcc_weightings[] = { 123 204, 192, 179, 166, 153, 140, 128, 115, 124 102, 89, 76, 64, 51, 38, 25, 12, 125 0, -12, -25, -38, -51, -64, -76, -89, 126 -102, -115, -128, -140, -153, -166, -179, -192 127}; 128 129 130/** Tail codes used in arithmetic coding using block Gilbert-Moore codes. 131 */ 132static const uint8_t tail_code[16][6] = { 133 { 74, 44, 25, 13, 7, 3}, 134 { 68, 42, 24, 13, 7, 3}, 135 { 58, 39, 23, 13, 7, 3}, 136 {126, 70, 37, 19, 10, 5}, 137 {132, 70, 37, 20, 10, 5}, 138 {124, 70, 38, 20, 10, 5}, 139 {120, 69, 37, 20, 11, 5}, 140 {116, 67, 37, 20, 11, 5}, 141 {108, 66, 36, 20, 10, 5}, 142 {102, 62, 36, 20, 10, 5}, 143 { 88, 58, 34, 19, 10, 5}, 144 {162, 89, 49, 25, 13, 7}, 145 {156, 87, 49, 26, 14, 7}, 146 {150, 86, 47, 26, 14, 7}, 147 {142, 84, 47, 26, 14, 7}, 148 {131, 79, 46, 26, 14, 7} 149}; 150 151 152enum RA_Flag { 153 RA_FLAG_NONE, 154 RA_FLAG_FRAMES, 155 RA_FLAG_HEADER 156}; 157 158 159typedef struct ALSSpecificConfig { 160 uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown 161 int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit 162 int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer 163 int msb_first; ///< 1 = original CRC calculated on big-endian system, 0 = little-endian 164 int frame_length; ///< frame length for each frame (last frame may differ) 165 int ra_distance; ///< distance between RA frames (in frames, 0...255) 166 enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored 167 int adapt_order; ///< adaptive order: 1 = on, 0 = off 168 int coef_table; ///< table index of Rice code parameters 169 int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off 170 int max_order; ///< maximum prediction order (0..1023) 171 int block_switching; ///< number of block switching levels 172 int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only) 173 int sb_part; ///< sub-block partition 174 int joint_stereo; ///< joint stereo: 1 = on, 0 = off 175 int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off 176 int chan_config; ///< indicates that a chan_config_info field is present 177 int chan_sort; ///< channel rearrangement: 1 = on, 0 = off 178 int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off 179 int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented. 180 int *chan_pos; ///< original channel positions 181 int crc_enabled; ///< enable Cyclic Redundancy Checksum 182} ALSSpecificConfig; 183 184 185typedef struct ALSChannelData { 186 int stop_flag; 187 int master_channel; 188 int time_diff_flag; 189 int time_diff_sign; 190 int time_diff_index; 191 int weighting[6]; 192} ALSChannelData; 193 194 195typedef struct ALSDecContext { 196 AVCodecContext *avctx; 197 ALSSpecificConfig sconf; 198 GetBitContext gb; 199 BswapDSPContext bdsp; 200 const AVCRC *crc_table; 201 uint32_t crc_org; ///< CRC value of the original input data 202 uint32_t crc; ///< CRC value calculated from decoded data 203 unsigned int cur_frame_length; ///< length of the current frame to decode 204 unsigned int frame_id; ///< the frame ID / number of the current frame 205 unsigned int js_switch; ///< if true, joint-stereo decoding is enforced 206 unsigned int cs_switch; ///< if true, channel rearrangement is done 207 unsigned int num_blocks; ///< number of blocks used in the current frame 208 unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding 209 uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC 210 int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC 211 int ltp_lag_length; ///< number of bits used for ltp lag value 212 int *const_block; ///< contains const_block flags for all channels 213 unsigned int *shift_lsbs; ///< contains shift_lsbs flags for all channels 214 unsigned int *opt_order; ///< contains opt_order flags for all channels 215 int *store_prev_samples; ///< contains store_prev_samples flags for all channels 216 int *use_ltp; ///< contains use_ltp flags for all channels 217 int *ltp_lag; ///< contains ltp lag values for all channels 218 int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel 219 int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter 220 int32_t **quant_cof; ///< quantized parcor coefficients for a channel 221 int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients 222 int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel 223 int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter 224 int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer 225 ALSChannelData **chan_data; ///< channel data for multi-channel correlation 226 ALSChannelData *chan_data_buffer; ///< contains channel data for all channels 227 int *reverted_channels; ///< stores a flag for each reverted channel 228 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block 229 int32_t **raw_samples; ///< decoded raw samples for each channel 230 int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples 231 uint8_t *crc_buffer; ///< buffer of byte order corrected samples used for CRC check 232 MLZ* mlz; ///< masked lz decompression structure 233 SoftFloat_IEEE754 *acf; ///< contains common multiplier for all channels 234 int *last_acf_mantissa; ///< contains the last acf mantissa data of common multiplier for all channels 235 int *shift_value; ///< value by which the binary point is to be shifted for all channels 236 int *last_shift_value; ///< contains last shift value for all channels 237 int **raw_mantissa; ///< decoded mantissa bits of the difference signal 238 unsigned char *larray; ///< buffer to store the output of masked lz decompression 239 int *nbits; ///< contains the number of bits to read for masked lz decompression for all samples 240 int highest_decoded_channel; 241} ALSDecContext; 242 243 244typedef struct ALSBlockData { 245 unsigned int block_length; ///< number of samples within the block 246 unsigned int ra_block; ///< if true, this is a random access block 247 int *const_block; ///< if true, this is a constant value block 248 int js_blocks; ///< true if this block contains a difference signal 249 unsigned int *shift_lsbs; ///< shift of values for this block 250 unsigned int *opt_order; ///< prediction order of this block 251 int *store_prev_samples;///< if true, carryover samples have to be stored 252 int *use_ltp; ///< if true, long-term prediction is used 253 int *ltp_lag; ///< lag value for long-term prediction 254 int *ltp_gain; ///< gain values for ltp 5-tap filter 255 int32_t *quant_cof; ///< quantized parcor coefficients 256 int32_t *lpc_cof; ///< coefficients of the direct form prediction 257 int32_t *raw_samples; ///< decoded raw samples / residuals for this block 258 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block 259 int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair 260} ALSBlockData; 261 262 263static av_cold void dprint_specific_config(ALSDecContext *ctx) 264{ 265#ifdef DEBUG 266 AVCodecContext *avctx = ctx->avctx; 267 ALSSpecificConfig *sconf = &ctx->sconf; 268 269 ff_dlog(avctx, "resolution = %i\n", sconf->resolution); 270 ff_dlog(avctx, "floating = %i\n", sconf->floating); 271 ff_dlog(avctx, "frame_length = %i\n", sconf->frame_length); 272 ff_dlog(avctx, "ra_distance = %i\n", sconf->ra_distance); 273 ff_dlog(avctx, "ra_flag = %i\n", sconf->ra_flag); 274 ff_dlog(avctx, "adapt_order = %i\n", sconf->adapt_order); 275 ff_dlog(avctx, "coef_table = %i\n", sconf->coef_table); 276 ff_dlog(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction); 277 ff_dlog(avctx, "max_order = %i\n", sconf->max_order); 278 ff_dlog(avctx, "block_switching = %i\n", sconf->block_switching); 279 ff_dlog(avctx, "bgmc = %i\n", sconf->bgmc); 280 ff_dlog(avctx, "sb_part = %i\n", sconf->sb_part); 281 ff_dlog(avctx, "joint_stereo = %i\n", sconf->joint_stereo); 282 ff_dlog(avctx, "mc_coding = %i\n", sconf->mc_coding); 283 ff_dlog(avctx, "chan_config = %i\n", sconf->chan_config); 284 ff_dlog(avctx, "chan_sort = %i\n", sconf->chan_sort); 285 ff_dlog(avctx, "RLSLMS = %i\n", sconf->rlslms); 286 ff_dlog(avctx, "chan_config_info = %i\n", sconf->chan_config_info); 287#endif 288} 289 290 291/** Read an ALSSpecificConfig from a buffer into the output struct. 292 */ 293static av_cold int read_specific_config(ALSDecContext *ctx) 294{ 295 GetBitContext gb; 296 uint64_t ht_size; 297 int i, config_offset; 298 MPEG4AudioConfig m4ac = {0}; 299 ALSSpecificConfig *sconf = &ctx->sconf; 300 AVCodecContext *avctx = ctx->avctx; 301 uint32_t als_id, header_size, trailer_size; 302 int ret; 303 304 if ((ret = init_get_bits8(&gb, avctx->extradata, avctx->extradata_size)) < 0) 305 return ret; 306 307 config_offset = avpriv_mpeg4audio_get_config2(&m4ac, avctx->extradata, 308 avctx->extradata_size, 1, avctx); 309 310 if (config_offset < 0) 311 return AVERROR_INVALIDDATA; 312 313 skip_bits_long(&gb, config_offset); 314 315 if (get_bits_left(&gb) < (30 << 3)) 316 return AVERROR_INVALIDDATA; 317 318 // read the fixed items 319 als_id = get_bits_long(&gb, 32); 320 avctx->sample_rate = m4ac.sample_rate; 321 skip_bits_long(&gb, 32); // sample rate already known 322 sconf->samples = get_bits_long(&gb, 32); 323 324 if (avctx->ch_layout.nb_channels != m4ac.channels) { 325 av_channel_layout_uninit(&avctx->ch_layout); 326 avctx->ch_layout.order = AV_CHANNEL_ORDER_UNSPEC; 327 avctx->ch_layout.nb_channels = m4ac.channels; 328 } 329 330 skip_bits(&gb, 16); // number of channels already known 331 skip_bits(&gb, 3); // skip file_type 332 sconf->resolution = get_bits(&gb, 3); 333 sconf->floating = get_bits1(&gb); 334 sconf->msb_first = get_bits1(&gb); 335 sconf->frame_length = get_bits(&gb, 16) + 1; 336 sconf->ra_distance = get_bits(&gb, 8); 337 sconf->ra_flag = get_bits(&gb, 2); 338 sconf->adapt_order = get_bits1(&gb); 339 sconf->coef_table = get_bits(&gb, 2); 340 sconf->long_term_prediction = get_bits1(&gb); 341 sconf->max_order = get_bits(&gb, 10); 342 sconf->block_switching = get_bits(&gb, 2); 343 sconf->bgmc = get_bits1(&gb); 344 sconf->sb_part = get_bits1(&gb); 345 sconf->joint_stereo = get_bits1(&gb); 346 sconf->mc_coding = get_bits1(&gb); 347 sconf->chan_config = get_bits1(&gb); 348 sconf->chan_sort = get_bits1(&gb); 349 sconf->crc_enabled = get_bits1(&gb); 350 sconf->rlslms = get_bits1(&gb); 351 skip_bits(&gb, 5); // skip 5 reserved bits 352 skip_bits1(&gb); // skip aux_data_enabled 353 354 355 // check for ALSSpecificConfig struct 356 if (als_id != MKBETAG('A','L','S','\0')) 357 return AVERROR_INVALIDDATA; 358 359 if (avctx->ch_layout.nb_channels > FF_SANE_NB_CHANNELS) { 360 avpriv_request_sample(avctx, "Huge number of channels"); 361 return AVERROR_PATCHWELCOME; 362 } 363 364 if (avctx->ch_layout.nb_channels == 0) 365 return AVERROR_INVALIDDATA; 366 367 ctx->cur_frame_length = sconf->frame_length; 368 369 // read channel config 370 if (sconf->chan_config) 371 sconf->chan_config_info = get_bits(&gb, 16); 372 // TODO: use this to set avctx->channel_layout 373 374 375 // read channel sorting 376 if (sconf->chan_sort && avctx->ch_layout.nb_channels > 1) { 377 int chan_pos_bits = av_ceil_log2(avctx->ch_layout.nb_channels); 378 int bits_needed = avctx->ch_layout.nb_channels * chan_pos_bits + 7; 379 if (get_bits_left(&gb) < bits_needed) 380 return AVERROR_INVALIDDATA; 381 382 if (!(sconf->chan_pos = av_malloc_array(avctx->ch_layout.nb_channels, sizeof(*sconf->chan_pos)))) 383 return AVERROR(ENOMEM); 384 385 ctx->cs_switch = 1; 386 387 for (i = 0; i < avctx->ch_layout.nb_channels; i++) { 388 sconf->chan_pos[i] = -1; 389 } 390 391 for (i = 0; i < avctx->ch_layout.nb_channels; i++) { 392 int idx; 393 394 idx = get_bits(&gb, chan_pos_bits); 395 if (idx >= avctx->ch_layout.nb_channels || sconf->chan_pos[idx] != -1) { 396 av_log(avctx, AV_LOG_WARNING, "Invalid channel reordering.\n"); 397 ctx->cs_switch = 0; 398 break; 399 } 400 sconf->chan_pos[idx] = i; 401 } 402 403 align_get_bits(&gb); 404 } 405 406 407 // read fixed header and trailer sizes, 408 // if size = 0xFFFFFFFF then there is no data field! 409 if (get_bits_left(&gb) < 64) 410 return AVERROR_INVALIDDATA; 411 412 header_size = get_bits_long(&gb, 32); 413 trailer_size = get_bits_long(&gb, 32); 414 if (header_size == 0xFFFFFFFF) 415 header_size = 0; 416 if (trailer_size == 0xFFFFFFFF) 417 trailer_size = 0; 418 419 ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3; 420 421 422 // skip the header and trailer data 423 if (get_bits_left(&gb) < ht_size) 424 return AVERROR_INVALIDDATA; 425 426 if (ht_size > INT32_MAX) 427 return AVERROR_PATCHWELCOME; 428 429 skip_bits_long(&gb, ht_size); 430 431 432 // initialize CRC calculation 433 if (sconf->crc_enabled) { 434 if (get_bits_left(&gb) < 32) 435 return AVERROR_INVALIDDATA; 436 437 if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) { 438 ctx->crc_table = av_crc_get_table(AV_CRC_32_IEEE_LE); 439 ctx->crc = 0xFFFFFFFF; 440 ctx->crc_org = ~get_bits_long(&gb, 32); 441 } else 442 skip_bits_long(&gb, 32); 443 } 444 445 446 // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data) 447 448 dprint_specific_config(ctx); 449 450 return 0; 451} 452 453 454/** Check the ALSSpecificConfig for unsupported features. 455 */ 456static int check_specific_config(ALSDecContext *ctx) 457{ 458 ALSSpecificConfig *sconf = &ctx->sconf; 459 int error = 0; 460 461 // report unsupported feature and set error value 462 #define MISSING_ERR(cond, str, errval) \ 463 { \ 464 if (cond) { \ 465 avpriv_report_missing_feature(ctx->avctx, \ 466 str); \ 467 error = errval; \ 468 } \ 469 } 470 471 MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", AVERROR_PATCHWELCOME); 472 473 return error; 474} 475 476 477/** Parse the bs_info field to extract the block partitioning used in 478 * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2. 479 */ 480static void parse_bs_info(const uint32_t bs_info, unsigned int n, 481 unsigned int div, unsigned int **div_blocks, 482 unsigned int *num_blocks) 483{ 484 if (n < 31 && ((bs_info << n) & 0x40000000)) { 485 // if the level is valid and the investigated bit n is set 486 // then recursively check both children at bits (2n+1) and (2n+2) 487 n *= 2; 488 div += 1; 489 parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks); 490 parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks); 491 } else { 492 // else the bit is not set or the last level has been reached 493 // (bit implicitly not set) 494 **div_blocks = div; 495 (*div_blocks)++; 496 (*num_blocks)++; 497 } 498} 499 500 501/** Read and decode a Rice codeword. 502 */ 503static int32_t decode_rice(GetBitContext *gb, unsigned int k) 504{ 505 int max = get_bits_left(gb) - k; 506 unsigned q = get_unary(gb, 0, max); 507 int r = k ? get_bits1(gb) : !(q & 1); 508 509 if (k > 1) { 510 q <<= (k - 1); 511 q += get_bits_long(gb, k - 1); 512 } else if (!k) { 513 q >>= 1; 514 } 515 return r ? q : ~q; 516} 517 518 519/** Convert PARCOR coefficient k to direct filter coefficient. 520 */ 521static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof) 522{ 523 int i, j; 524 525 for (i = 0, j = k - 1; i < j; i++, j--) { 526 unsigned tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20); 527 cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20); 528 cof[i] += tmp1; 529 } 530 if (i == j) 531 cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20); 532 533 cof[k] = par[k]; 534} 535 536 537/** Read block switching field if necessary and set actual block sizes. 538 * Also assure that the block sizes of the last frame correspond to the 539 * actual number of samples. 540 */ 541static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks, 542 uint32_t *bs_info) 543{ 544 ALSSpecificConfig *sconf = &ctx->sconf; 545 GetBitContext *gb = &ctx->gb; 546 unsigned int *ptr_div_blocks = div_blocks; 547 unsigned int b; 548 549 if (sconf->block_switching) { 550 unsigned int bs_info_len = 1 << (sconf->block_switching + 2); 551 *bs_info = get_bits_long(gb, bs_info_len); 552 *bs_info <<= (32 - bs_info_len); 553 } 554 555 ctx->num_blocks = 0; 556 parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks); 557 558 // The last frame may have an overdetermined block structure given in 559 // the bitstream. In that case the defined block structure would need 560 // more samples than available to be consistent. 561 // The block structure is actually used but the block sizes are adapted 562 // to fit the actual number of available samples. 563 // Example: 5 samples, 2nd level block sizes: 2 2 2 2. 564 // This results in the actual block sizes: 2 2 1 0. 565 // This is not specified in 14496-3 but actually done by the reference 566 // codec RM22 revision 2. 567 // This appears to happen in case of an odd number of samples in the last 568 // frame which is actually not allowed by the block length switching part 569 // of 14496-3. 570 // The ALS conformance files feature an odd number of samples in the last 571 // frame. 572 573 for (b = 0; b < ctx->num_blocks; b++) 574 div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b]; 575 576 if (ctx->cur_frame_length != ctx->sconf.frame_length) { 577 unsigned int remaining = ctx->cur_frame_length; 578 579 for (b = 0; b < ctx->num_blocks; b++) { 580 if (remaining <= div_blocks[b]) { 581 div_blocks[b] = remaining; 582 ctx->num_blocks = b + 1; 583 break; 584 } 585 586 remaining -= div_blocks[b]; 587 } 588 } 589} 590 591 592/** Read the block data for a constant block 593 */ 594static int read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd) 595{ 596 ALSSpecificConfig *sconf = &ctx->sconf; 597 AVCodecContext *avctx = ctx->avctx; 598 GetBitContext *gb = &ctx->gb; 599 600 if (bd->block_length <= 0) 601 return AVERROR_INVALIDDATA; 602 603 *bd->raw_samples = 0; 604 *bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence) 605 bd->js_blocks = get_bits1(gb); 606 607 // skip 5 reserved bits 608 skip_bits(gb, 5); 609 610 if (*bd->const_block) { 611 unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample; 612 *bd->raw_samples = get_sbits_long(gb, const_val_bits); 613 } 614 615 // ensure constant block decoding by reusing this field 616 *bd->const_block = 1; 617 618 return 0; 619} 620 621 622/** Decode the block data for a constant block 623 */ 624static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd) 625{ 626 int smp = bd->block_length - 1; 627 int32_t val = *bd->raw_samples; 628 int32_t *dst = bd->raw_samples + 1; 629 630 // write raw samples into buffer 631 for (; smp; smp--) 632 *dst++ = val; 633} 634 635 636/** Read the block data for a non-constant block 637 */ 638static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd) 639{ 640 ALSSpecificConfig *sconf = &ctx->sconf; 641 AVCodecContext *avctx = ctx->avctx; 642 GetBitContext *gb = &ctx->gb; 643 unsigned int k; 644 unsigned int s[8]; 645 unsigned int sx[8]; 646 unsigned int sub_blocks, log2_sub_blocks, sb_length; 647 unsigned int start = 0; 648 unsigned int opt_order; 649 int sb; 650 int32_t *quant_cof = bd->quant_cof; 651 int32_t *current_res; 652 653 654 // ensure variable block decoding by reusing this field 655 *bd->const_block = 0; 656 657 *bd->opt_order = 1; 658 bd->js_blocks = get_bits1(gb); 659 660 opt_order = *bd->opt_order; 661 662 // determine the number of subblocks for entropy decoding 663 if (!sconf->bgmc && !sconf->sb_part) { 664 log2_sub_blocks = 0; 665 } else { 666 if (sconf->bgmc && sconf->sb_part) 667 log2_sub_blocks = get_bits(gb, 2); 668 else 669 log2_sub_blocks = 2 * get_bits1(gb); 670 } 671 672 sub_blocks = 1 << log2_sub_blocks; 673 674 // do not continue in case of a damaged stream since 675 // block_length must be evenly divisible by sub_blocks 676 if (bd->block_length & (sub_blocks - 1) || bd->block_length <= 0) { 677 av_log(avctx, AV_LOG_WARNING, 678 "Block length is not evenly divisible by the number of subblocks.\n"); 679 return AVERROR_INVALIDDATA; 680 } 681 682 sb_length = bd->block_length >> log2_sub_blocks; 683 684 if (sconf->bgmc) { 685 s[0] = get_bits(gb, 8 + (sconf->resolution > 1)); 686 for (k = 1; k < sub_blocks; k++) 687 s[k] = s[k - 1] + decode_rice(gb, 2); 688 689 for (k = 0; k < sub_blocks; k++) { 690 sx[k] = s[k] & 0x0F; 691 s [k] >>= 4; 692 } 693 } else { 694 s[0] = get_bits(gb, 4 + (sconf->resolution > 1)); 695 for (k = 1; k < sub_blocks; k++) 696 s[k] = s[k - 1] + decode_rice(gb, 0); 697 } 698 for (k = 1; k < sub_blocks; k++) 699 if (s[k] > 32) { 700 av_log(avctx, AV_LOG_ERROR, "k invalid for rice code.\n"); 701 return AVERROR_INVALIDDATA; 702 } 703 704 if (get_bits1(gb)) 705 *bd->shift_lsbs = get_bits(gb, 4) + 1; 706 707 *bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || *bd->shift_lsbs; 708 709 710 if (!sconf->rlslms) { 711 if (sconf->adapt_order && sconf->max_order) { 712 int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1, 713 2, sconf->max_order + 1)); 714 *bd->opt_order = get_bits(gb, opt_order_length); 715 if (*bd->opt_order > sconf->max_order) { 716 *bd->opt_order = sconf->max_order; 717 av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n"); 718 return AVERROR_INVALIDDATA; 719 } 720 } else { 721 *bd->opt_order = sconf->max_order; 722 } 723 opt_order = *bd->opt_order; 724 725 if (opt_order) { 726 int add_base; 727 728 if (sconf->coef_table == 3) { 729 add_base = 0x7F; 730 731 // read coefficient 0 732 quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)]; 733 734 // read coefficient 1 735 if (opt_order > 1) 736 quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)]; 737 738 // read coefficients 2 to opt_order 739 for (k = 2; k < opt_order; k++) 740 quant_cof[k] = get_bits(gb, 7); 741 } else { 742 int k_max; 743 add_base = 1; 744 745 // read coefficient 0 to 19 746 k_max = FFMIN(opt_order, 20); 747 for (k = 0; k < k_max; k++) { 748 int rice_param = parcor_rice_table[sconf->coef_table][k][1]; 749 int offset = parcor_rice_table[sconf->coef_table][k][0]; 750 quant_cof[k] = decode_rice(gb, rice_param) + offset; 751 if (quant_cof[k] < -64 || quant_cof[k] > 63) { 752 av_log(avctx, AV_LOG_ERROR, 753 "quant_cof %"PRId32" is out of range.\n", 754 quant_cof[k]); 755 return AVERROR_INVALIDDATA; 756 } 757 } 758 759 // read coefficients 20 to 126 760 k_max = FFMIN(opt_order, 127); 761 for (; k < k_max; k++) 762 quant_cof[k] = decode_rice(gb, 2) + (k & 1); 763 764 // read coefficients 127 to opt_order 765 for (; k < opt_order; k++) 766 quant_cof[k] = decode_rice(gb, 1); 767 768 quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64]; 769 770 if (opt_order > 1) 771 quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64]; 772 } 773 774 for (k = 2; k < opt_order; k++) 775 quant_cof[k] = (quant_cof[k] * (1U << 14)) + (add_base << 13); 776 } 777 } 778 779 // read LTP gain and lag values 780 if (sconf->long_term_prediction) { 781 *bd->use_ltp = get_bits1(gb); 782 783 if (*bd->use_ltp) { 784 int r, c; 785 786 bd->ltp_gain[0] = decode_rice(gb, 1) * 8; 787 bd->ltp_gain[1] = decode_rice(gb, 2) * 8; 788 789 r = get_unary(gb, 0, 4); 790 c = get_bits(gb, 2); 791 if (r >= 4) { 792 av_log(avctx, AV_LOG_ERROR, "r overflow\n"); 793 return AVERROR_INVALIDDATA; 794 } 795 796 bd->ltp_gain[2] = ltp_gain_values[r][c]; 797 798 bd->ltp_gain[3] = decode_rice(gb, 2) * 8; 799 bd->ltp_gain[4] = decode_rice(gb, 1) * 8; 800 801 *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length); 802 *bd->ltp_lag += FFMAX(4, opt_order + 1); 803 } 804 } 805 806 // read first value and residuals in case of a random access block 807 if (bd->ra_block) { 808 start = FFMIN(opt_order, 3); 809 av_assert0(sb_length <= sconf->frame_length); 810 if (sb_length <= start) { 811 // opt_order or sb_length may be corrupted, either way this is unsupported and not well defined in the specification 812 av_log(avctx, AV_LOG_ERROR, "Sub block length smaller or equal start\n"); 813 return AVERROR_PATCHWELCOME; 814 } 815 816 if (opt_order) 817 bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4); 818 if (opt_order > 1) 819 bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max)); 820 if (opt_order > 2) 821 bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max)); 822 } 823 824 // read all residuals 825 if (sconf->bgmc) { 826 int delta[8]; 827 unsigned int k [8]; 828 unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5); 829 830 // read most significant bits 831 unsigned int high; 832 unsigned int low; 833 unsigned int value; 834 835 int ret = ff_bgmc_decode_init(gb, &high, &low, &value); 836 if (ret < 0) 837 return ret; 838 839 current_res = bd->raw_samples + start; 840 841 for (sb = 0; sb < sub_blocks; sb++) { 842 unsigned int sb_len = sb_length - (sb ? 0 : start); 843 844 k [sb] = s[sb] > b ? s[sb] - b : 0; 845 delta[sb] = 5 - s[sb] + k[sb]; 846 847 if (k[sb] >= 32) 848 return AVERROR_INVALIDDATA; 849 850 ff_bgmc_decode(gb, sb_len, current_res, 851 delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status); 852 853 current_res += sb_len; 854 } 855 856 ff_bgmc_decode_end(gb); 857 858 859 // read least significant bits and tails 860 current_res = bd->raw_samples + start; 861 862 for (sb = 0; sb < sub_blocks; sb++, start = 0) { 863 unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]]; 864 unsigned int cur_k = k[sb]; 865 unsigned int cur_s = s[sb]; 866 867 for (; start < sb_length; start++) { 868 int32_t res = *current_res; 869 870 if (res == cur_tail_code) { 871 unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10)) 872 << (5 - delta[sb]); 873 874 res = decode_rice(gb, cur_s); 875 876 if (res >= 0) { 877 res += (max_msb ) << cur_k; 878 } else { 879 res -= (max_msb - 1) << cur_k; 880 } 881 } else { 882 if (res > cur_tail_code) 883 res--; 884 885 if (res & 1) 886 res = -res; 887 888 res >>= 1; 889 890 if (cur_k) { 891 res *= 1U << cur_k; 892 res |= get_bits_long(gb, cur_k); 893 } 894 } 895 896 *current_res++ = res; 897 } 898 } 899 } else { 900 current_res = bd->raw_samples + start; 901 902 for (sb = 0; sb < sub_blocks; sb++, start = 0) 903 for (; start < sb_length; start++) 904 *current_res++ = decode_rice(gb, s[sb]); 905 } 906 907 return 0; 908} 909 910 911/** Decode the block data for a non-constant block 912 */ 913static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd) 914{ 915 ALSSpecificConfig *sconf = &ctx->sconf; 916 unsigned int block_length = bd->block_length; 917 unsigned int smp = 0; 918 unsigned int k; 919 int opt_order = *bd->opt_order; 920 int sb; 921 int64_t y; 922 int32_t *quant_cof = bd->quant_cof; 923 int32_t *lpc_cof = bd->lpc_cof; 924 int32_t *raw_samples = bd->raw_samples; 925 int32_t *raw_samples_end = bd->raw_samples + bd->block_length; 926 int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer; 927 928 // reverse long-term prediction 929 if (*bd->use_ltp) { 930 int ltp_smp; 931 932 for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) { 933 int center = ltp_smp - *bd->ltp_lag; 934 int begin = FFMAX(0, center - 2); 935 int end = center + 3; 936 int tab = 5 - (end - begin); 937 int base; 938 939 y = 1 << 6; 940 941 for (base = begin; base < end; base++, tab++) 942 y += (uint64_t)MUL64(bd->ltp_gain[tab], raw_samples[base]); 943 944 raw_samples[ltp_smp] += y >> 7; 945 } 946 } 947 948 // reconstruct all samples from residuals 949 if (bd->ra_block) { 950 for (smp = 0; smp < FFMIN(opt_order, block_length); smp++) { 951 y = 1 << 19; 952 953 for (sb = 0; sb < smp; sb++) 954 y += (uint64_t)MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]); 955 956 *raw_samples++ -= y >> 20; 957 parcor_to_lpc(smp, quant_cof, lpc_cof); 958 } 959 } else { 960 for (k = 0; k < opt_order; k++) 961 parcor_to_lpc(k, quant_cof, lpc_cof); 962 963 // store previous samples in case that they have to be altered 964 if (*bd->store_prev_samples) 965 memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order, 966 sizeof(*bd->prev_raw_samples) * sconf->max_order); 967 968 // reconstruct difference signal for prediction (joint-stereo) 969 if (bd->js_blocks && bd->raw_other) { 970 uint32_t *left, *right; 971 972 if (bd->raw_other > raw_samples) { // D = R - L 973 left = raw_samples; 974 right = bd->raw_other; 975 } else { // D = R - L 976 left = bd->raw_other; 977 right = raw_samples; 978 } 979 980 for (sb = -1; sb >= -sconf->max_order; sb--) 981 raw_samples[sb] = right[sb] - left[sb]; 982 } 983 984 // reconstruct shifted signal 985 if (*bd->shift_lsbs) 986 for (sb = -1; sb >= -sconf->max_order; sb--) 987 raw_samples[sb] >>= *bd->shift_lsbs; 988 } 989 990 // reverse linear prediction coefficients for efficiency 991 lpc_cof = lpc_cof + opt_order; 992 993 for (sb = 0; sb < opt_order; sb++) 994 lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)]; 995 996 // reconstruct raw samples 997 raw_samples = bd->raw_samples + smp; 998 lpc_cof = lpc_cof_reversed + opt_order; 999 1000 for (; raw_samples < raw_samples_end; raw_samples++) { 1001 y = 1 << 19; 1002 1003 for (sb = -opt_order; sb < 0; sb++) 1004 y += (uint64_t)MUL64(lpc_cof[sb], raw_samples[sb]); 1005 1006 *raw_samples -= y >> 20; 1007 } 1008 1009 raw_samples = bd->raw_samples; 1010 1011 // restore previous samples in case that they have been altered 1012 if (*bd->store_prev_samples) 1013 memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples, 1014 sizeof(*raw_samples) * sconf->max_order); 1015 1016 return 0; 1017} 1018 1019 1020/** Read the block data. 1021 */ 1022static int read_block(ALSDecContext *ctx, ALSBlockData *bd) 1023{ 1024 int ret; 1025 GetBitContext *gb = &ctx->gb; 1026 ALSSpecificConfig *sconf = &ctx->sconf; 1027 1028 *bd->shift_lsbs = 0; 1029 1030 if (get_bits_left(gb) < 7) 1031 return AVERROR_INVALIDDATA; 1032 1033 // read block type flag and read the samples accordingly 1034 if (get_bits1(gb)) { 1035 ret = read_var_block_data(ctx, bd); 1036 } else { 1037 ret = read_const_block_data(ctx, bd); 1038 } 1039 1040 if (!sconf->mc_coding || ctx->js_switch) 1041 align_get_bits(gb); 1042 1043 return ret; 1044} 1045 1046 1047/** Decode the block data. 1048 */ 1049static int decode_block(ALSDecContext *ctx, ALSBlockData *bd) 1050{ 1051 unsigned int smp; 1052 int ret = 0; 1053 1054 // read block type flag and read the samples accordingly 1055 if (*bd->const_block) 1056 decode_const_block_data(ctx, bd); 1057 else 1058 ret = decode_var_block_data(ctx, bd); // always return 0 1059 1060 if (ret < 0) 1061 return ret; 1062 1063 // TODO: read RLSLMS extension data 1064 1065 if (*bd->shift_lsbs) 1066 for (smp = 0; smp < bd->block_length; smp++) 1067 bd->raw_samples[smp] = (unsigned)bd->raw_samples[smp] << *bd->shift_lsbs; 1068 1069 return 0; 1070} 1071 1072 1073/** Read and decode block data successively. 1074 */ 1075static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd) 1076{ 1077 int ret; 1078 1079 if ((ret = read_block(ctx, bd)) < 0) 1080 return ret; 1081 1082 return decode_block(ctx, bd); 1083} 1084 1085 1086/** Compute the number of samples left to decode for the current frame and 1087 * sets these samples to zero. 1088 */ 1089static void zero_remaining(unsigned int b, unsigned int b_max, 1090 const unsigned int *div_blocks, int32_t *buf) 1091{ 1092 unsigned int count = 0; 1093 1094 while (b < b_max) 1095 count += div_blocks[b++]; 1096 1097 if (count) 1098 memset(buf, 0, sizeof(*buf) * count); 1099} 1100 1101 1102/** Decode blocks independently. 1103 */ 1104static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame, 1105 unsigned int c, const unsigned int *div_blocks, 1106 unsigned int *js_blocks) 1107{ 1108 int ret; 1109 unsigned int b; 1110 ALSBlockData bd = { 0 }; 1111 1112 bd.ra_block = ra_frame; 1113 bd.const_block = ctx->const_block; 1114 bd.shift_lsbs = ctx->shift_lsbs; 1115 bd.opt_order = ctx->opt_order; 1116 bd.store_prev_samples = ctx->store_prev_samples; 1117 bd.use_ltp = ctx->use_ltp; 1118 bd.ltp_lag = ctx->ltp_lag; 1119 bd.ltp_gain = ctx->ltp_gain[0]; 1120 bd.quant_cof = ctx->quant_cof[0]; 1121 bd.lpc_cof = ctx->lpc_cof[0]; 1122 bd.prev_raw_samples = ctx->prev_raw_samples; 1123 bd.raw_samples = ctx->raw_samples[c]; 1124 1125 1126 for (b = 0; b < ctx->num_blocks; b++) { 1127 bd.block_length = div_blocks[b]; 1128 1129 if ((ret = read_decode_block(ctx, &bd)) < 0) { 1130 // damaged block, write zero for the rest of the frame 1131 zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples); 1132 return ret; 1133 } 1134 bd.raw_samples += div_blocks[b]; 1135 bd.ra_block = 0; 1136 } 1137 1138 return 0; 1139} 1140 1141 1142/** Decode blocks dependently. 1143 */ 1144static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame, 1145 unsigned int c, const unsigned int *div_blocks, 1146 unsigned int *js_blocks) 1147{ 1148 ALSSpecificConfig *sconf = &ctx->sconf; 1149 unsigned int offset = 0; 1150 unsigned int b; 1151 int ret; 1152 ALSBlockData bd[2] = { { 0 } }; 1153 1154 bd[0].ra_block = ra_frame; 1155 bd[0].const_block = ctx->const_block; 1156 bd[0].shift_lsbs = ctx->shift_lsbs; 1157 bd[0].opt_order = ctx->opt_order; 1158 bd[0].store_prev_samples = ctx->store_prev_samples; 1159 bd[0].use_ltp = ctx->use_ltp; 1160 bd[0].ltp_lag = ctx->ltp_lag; 1161 bd[0].ltp_gain = ctx->ltp_gain[0]; 1162 bd[0].quant_cof = ctx->quant_cof[0]; 1163 bd[0].lpc_cof = ctx->lpc_cof[0]; 1164 bd[0].prev_raw_samples = ctx->prev_raw_samples; 1165 bd[0].js_blocks = *js_blocks; 1166 1167 bd[1].ra_block = ra_frame; 1168 bd[1].const_block = ctx->const_block; 1169 bd[1].shift_lsbs = ctx->shift_lsbs; 1170 bd[1].opt_order = ctx->opt_order; 1171 bd[1].store_prev_samples = ctx->store_prev_samples; 1172 bd[1].use_ltp = ctx->use_ltp; 1173 bd[1].ltp_lag = ctx->ltp_lag; 1174 bd[1].ltp_gain = ctx->ltp_gain[0]; 1175 bd[1].quant_cof = ctx->quant_cof[0]; 1176 bd[1].lpc_cof = ctx->lpc_cof[0]; 1177 bd[1].prev_raw_samples = ctx->prev_raw_samples; 1178 bd[1].js_blocks = *(js_blocks + 1); 1179 1180 // decode all blocks 1181 for (b = 0; b < ctx->num_blocks; b++) { 1182 unsigned int s; 1183 1184 bd[0].block_length = div_blocks[b]; 1185 bd[1].block_length = div_blocks[b]; 1186 1187 bd[0].raw_samples = ctx->raw_samples[c ] + offset; 1188 bd[1].raw_samples = ctx->raw_samples[c + 1] + offset; 1189 1190 bd[0].raw_other = bd[1].raw_samples; 1191 bd[1].raw_other = bd[0].raw_samples; 1192 1193 if ((ret = read_decode_block(ctx, &bd[0])) < 0 || 1194 (ret = read_decode_block(ctx, &bd[1])) < 0) 1195 goto fail; 1196 1197 // reconstruct joint-stereo blocks 1198 if (bd[0].js_blocks) { 1199 if (bd[1].js_blocks) 1200 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair.\n"); 1201 1202 for (s = 0; s < div_blocks[b]; s++) 1203 bd[0].raw_samples[s] = bd[1].raw_samples[s] - (unsigned)bd[0].raw_samples[s]; 1204 } else if (bd[1].js_blocks) { 1205 for (s = 0; s < div_blocks[b]; s++) 1206 bd[1].raw_samples[s] = bd[1].raw_samples[s] + (unsigned)bd[0].raw_samples[s]; 1207 } 1208 1209 offset += div_blocks[b]; 1210 bd[0].ra_block = 0; 1211 bd[1].ra_block = 0; 1212 } 1213 1214 // store carryover raw samples, 1215 // the others channel raw samples are stored by the calling function. 1216 memmove(ctx->raw_samples[c] - sconf->max_order, 1217 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1218 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1219 1220 return 0; 1221fail: 1222 // damaged block, write zero for the rest of the frame 1223 zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples); 1224 zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples); 1225 return ret; 1226} 1227 1228static inline int als_weighting(GetBitContext *gb, int k, int off) 1229{ 1230 int idx = av_clip(decode_rice(gb, k) + off, 1231 0, FF_ARRAY_ELEMS(mcc_weightings) - 1); 1232 return mcc_weightings[idx]; 1233} 1234 1235/** Read the channel data. 1236 */ 1237static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c) 1238{ 1239 GetBitContext *gb = &ctx->gb; 1240 ALSChannelData *current = cd; 1241 unsigned int channels = ctx->avctx->ch_layout.nb_channels; 1242 int entries = 0; 1243 1244 while (entries < channels && !(current->stop_flag = get_bits1(gb))) { 1245 current->master_channel = get_bits_long(gb, av_ceil_log2(channels)); 1246 1247 if (current->master_channel >= channels) { 1248 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel.\n"); 1249 return AVERROR_INVALIDDATA; 1250 } 1251 1252 if (current->master_channel != c) { 1253 current->time_diff_flag = get_bits1(gb); 1254 current->weighting[0] = als_weighting(gb, 1, 16); 1255 current->weighting[1] = als_weighting(gb, 2, 14); 1256 current->weighting[2] = als_weighting(gb, 1, 16); 1257 1258 if (current->time_diff_flag) { 1259 current->weighting[3] = als_weighting(gb, 1, 16); 1260 current->weighting[4] = als_weighting(gb, 1, 16); 1261 current->weighting[5] = als_weighting(gb, 1, 16); 1262 1263 current->time_diff_sign = get_bits1(gb); 1264 current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3; 1265 } 1266 } 1267 1268 current++; 1269 entries++; 1270 } 1271 1272 if (entries == channels) { 1273 av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data.\n"); 1274 return AVERROR_INVALIDDATA; 1275 } 1276 1277 align_get_bits(gb); 1278 return 0; 1279} 1280 1281 1282/** Recursively reverts the inter-channel correlation for a block. 1283 */ 1284static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd, 1285 ALSChannelData **cd, int *reverted, 1286 unsigned int offset, int c) 1287{ 1288 ALSChannelData *ch = cd[c]; 1289 unsigned int dep = 0; 1290 unsigned int channels = ctx->avctx->ch_layout.nb_channels; 1291 unsigned int channel_size = ctx->sconf.frame_length + ctx->sconf.max_order; 1292 1293 if (reverted[c]) 1294 return 0; 1295 1296 reverted[c] = 1; 1297 1298 while (dep < channels && !ch[dep].stop_flag) { 1299 revert_channel_correlation(ctx, bd, cd, reverted, offset, 1300 ch[dep].master_channel); 1301 1302 dep++; 1303 } 1304 1305 if (dep == channels) { 1306 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation.\n"); 1307 return AVERROR_INVALIDDATA; 1308 } 1309 1310 bd->const_block = ctx->const_block + c; 1311 bd->shift_lsbs = ctx->shift_lsbs + c; 1312 bd->opt_order = ctx->opt_order + c; 1313 bd->store_prev_samples = ctx->store_prev_samples + c; 1314 bd->use_ltp = ctx->use_ltp + c; 1315 bd->ltp_lag = ctx->ltp_lag + c; 1316 bd->ltp_gain = ctx->ltp_gain[c]; 1317 bd->lpc_cof = ctx->lpc_cof[c]; 1318 bd->quant_cof = ctx->quant_cof[c]; 1319 bd->raw_samples = ctx->raw_samples[c] + offset; 1320 1321 for (dep = 0; !ch[dep].stop_flag; dep++) { 1322 ptrdiff_t smp; 1323 ptrdiff_t begin = 1; 1324 ptrdiff_t end = bd->block_length - 1; 1325 int64_t y; 1326 int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset; 1327 1328 if (ch[dep].master_channel == c) 1329 continue; 1330 1331 if (ch[dep].time_diff_flag) { 1332 int t = ch[dep].time_diff_index; 1333 1334 if (ch[dep].time_diff_sign) { 1335 t = -t; 1336 if (begin < t) { 1337 av_log(ctx->avctx, AV_LOG_ERROR, "begin %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", begin, t); 1338 return AVERROR_INVALIDDATA; 1339 } 1340 begin -= t; 1341 } else { 1342 if (end < t) { 1343 av_log(ctx->avctx, AV_LOG_ERROR, "end %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", end, t); 1344 return AVERROR_INVALIDDATA; 1345 } 1346 end -= t; 1347 } 1348 1349 if (FFMIN(begin - 1, begin - 1 + t) < ctx->raw_buffer - master || 1350 FFMAX(end + 1, end + 1 + t) > ctx->raw_buffer + channels * channel_size - master) { 1351 av_log(ctx->avctx, AV_LOG_ERROR, 1352 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n", 1353 master + FFMIN(begin - 1, begin - 1 + t), master + FFMAX(end + 1, end + 1 + t), 1354 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size); 1355 return AVERROR_INVALIDDATA; 1356 } 1357 1358 for (smp = begin; smp < end; smp++) { 1359 y = (1 << 6) + 1360 MUL64(ch[dep].weighting[0], master[smp - 1 ]) + 1361 MUL64(ch[dep].weighting[1], master[smp ]) + 1362 MUL64(ch[dep].weighting[2], master[smp + 1 ]) + 1363 MUL64(ch[dep].weighting[3], master[smp - 1 + t]) + 1364 MUL64(ch[dep].weighting[4], master[smp + t]) + 1365 MUL64(ch[dep].weighting[5], master[smp + 1 + t]); 1366 1367 bd->raw_samples[smp] += y >> 7; 1368 } 1369 } else { 1370 1371 if (begin - 1 < ctx->raw_buffer - master || 1372 end + 1 > ctx->raw_buffer + channels * channel_size - master) { 1373 av_log(ctx->avctx, AV_LOG_ERROR, 1374 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n", 1375 master + begin - 1, master + end + 1, 1376 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size); 1377 return AVERROR_INVALIDDATA; 1378 } 1379 1380 for (smp = begin; smp < end; smp++) { 1381 y = (1 << 6) + 1382 MUL64(ch[dep].weighting[0], master[smp - 1]) + 1383 MUL64(ch[dep].weighting[1], master[smp ]) + 1384 MUL64(ch[dep].weighting[2], master[smp + 1]); 1385 1386 bd->raw_samples[smp] += y >> 7; 1387 } 1388 } 1389 } 1390 1391 return 0; 1392} 1393 1394 1395/** multiply two softfloats and handle the rounding off 1396 */ 1397static SoftFloat_IEEE754 multiply(SoftFloat_IEEE754 a, SoftFloat_IEEE754 b) { 1398 uint64_t mantissa_temp; 1399 uint64_t mask_64; 1400 int cutoff_bit_count; 1401 unsigned char last_2_bits; 1402 unsigned int mantissa; 1403 int32_t sign; 1404 uint32_t return_val = 0; 1405 int bit_count = 48; 1406 1407 sign = a.sign ^ b.sign; 1408 1409 // Multiply mantissa bits in a 64-bit register 1410 mantissa_temp = (uint64_t)a.mant * (uint64_t)b.mant; 1411 mask_64 = (uint64_t)0x1 << 47; 1412 1413 if (!mantissa_temp) 1414 return FLOAT_0; 1415 1416 // Count the valid bit count 1417 while (!(mantissa_temp & mask_64) && mask_64) { 1418 bit_count--; 1419 mask_64 >>= 1; 1420 } 1421 1422 // Round off 1423 cutoff_bit_count = bit_count - 24; 1424 if (cutoff_bit_count > 0) { 1425 last_2_bits = (unsigned char)(((unsigned int)mantissa_temp >> (cutoff_bit_count - 1)) & 0x3 ); 1426 if ((last_2_bits == 0x3) || ((last_2_bits == 0x1) && ((unsigned int)mantissa_temp & ((0x1UL << (cutoff_bit_count - 1)) - 1)))) { 1427 // Need to round up 1428 mantissa_temp += (uint64_t)0x1 << cutoff_bit_count; 1429 } 1430 } 1431 1432 if (cutoff_bit_count >= 0) { 1433 mantissa = (unsigned int)(mantissa_temp >> cutoff_bit_count); 1434 } else { 1435 mantissa = (unsigned int)(mantissa_temp <<-cutoff_bit_count); 1436 } 1437 1438 // Need one more shift? 1439 if (mantissa & 0x01000000ul) { 1440 bit_count++; 1441 mantissa >>= 1; 1442 } 1443 1444 if (!sign) { 1445 return_val = 0x80000000U; 1446 } 1447 1448 return_val |= ((unsigned)av_clip(a.exp + b.exp + bit_count - 47, -126, 127) << 23) & 0x7F800000; 1449 return_val |= mantissa; 1450 return av_bits2sf_ieee754(return_val); 1451} 1452 1453 1454/** Read and decode the floating point sample data 1455 */ 1456static int read_diff_float_data(ALSDecContext *ctx, unsigned int ra_frame) { 1457 AVCodecContext *avctx = ctx->avctx; 1458 GetBitContext *gb = &ctx->gb; 1459 SoftFloat_IEEE754 *acf = ctx->acf; 1460 int *shift_value = ctx->shift_value; 1461 int *last_shift_value = ctx->last_shift_value; 1462 int *last_acf_mantissa = ctx->last_acf_mantissa; 1463 int **raw_mantissa = ctx->raw_mantissa; 1464 int *nbits = ctx->nbits; 1465 unsigned char *larray = ctx->larray; 1466 int frame_length = ctx->cur_frame_length; 1467 SoftFloat_IEEE754 scale = av_int2sf_ieee754(0x1u, 23); 1468 unsigned int partA_flag; 1469 unsigned int highest_byte; 1470 unsigned int shift_amp; 1471 uint32_t tmp_32; 1472 int use_acf; 1473 int nchars; 1474 int i; 1475 int c; 1476 long k; 1477 long nbits_aligned; 1478 unsigned long acc; 1479 unsigned long j; 1480 uint32_t sign; 1481 uint32_t e; 1482 uint32_t mantissa; 1483 1484 skip_bits_long(gb, 32); //num_bytes_diff_float 1485 use_acf = get_bits1(gb); 1486 1487 if (ra_frame) { 1488 memset(last_acf_mantissa, 0, avctx->ch_layout.nb_channels * sizeof(*last_acf_mantissa)); 1489 memset(last_shift_value, 0, avctx->ch_layout.nb_channels * sizeof(*last_shift_value) ); 1490 ff_mlz_flush_dict(ctx->mlz); 1491 } 1492 1493 if (avctx->ch_layout.nb_channels * 8 > get_bits_left(gb)) 1494 return AVERROR_INVALIDDATA; 1495 1496 for (c = 0; c < avctx->ch_layout.nb_channels; ++c) { 1497 if (use_acf) { 1498 //acf_flag 1499 if (get_bits1(gb)) { 1500 tmp_32 = get_bits(gb, 23); 1501 last_acf_mantissa[c] = tmp_32; 1502 } else { 1503 tmp_32 = last_acf_mantissa[c]; 1504 } 1505 acf[c] = av_bits2sf_ieee754(tmp_32); 1506 } else { 1507 acf[c] = FLOAT_1; 1508 } 1509 1510 highest_byte = get_bits(gb, 2); 1511 partA_flag = get_bits1(gb); 1512 shift_amp = get_bits1(gb); 1513 1514 if (shift_amp) { 1515 shift_value[c] = get_bits(gb, 8); 1516 last_shift_value[c] = shift_value[c]; 1517 } else { 1518 shift_value[c] = last_shift_value[c]; 1519 } 1520 1521 if (partA_flag) { 1522 if (!get_bits1(gb)) { //uncompressed 1523 for (i = 0; i < frame_length; ++i) { 1524 if (ctx->raw_samples[c][i] == 0) { 1525 ctx->raw_mantissa[c][i] = get_bits_long(gb, 32); 1526 } 1527 } 1528 } else { //compressed 1529 nchars = 0; 1530 for (i = 0; i < frame_length; ++i) { 1531 if (ctx->raw_samples[c][i] == 0) { 1532 nchars += 4; 1533 } 1534 } 1535 1536 tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray); 1537 if(tmp_32 != nchars) { 1538 av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%"PRId32", %d).\n", tmp_32, nchars); 1539 return AVERROR_INVALIDDATA; 1540 } 1541 1542 for (i = 0; i < frame_length; ++i) { 1543 ctx->raw_mantissa[c][i] = AV_RB32(larray); 1544 } 1545 } 1546 } 1547 1548 //decode part B 1549 if (highest_byte) { 1550 for (i = 0; i < frame_length; ++i) { 1551 if (ctx->raw_samples[c][i] != 0) { 1552 //The following logic is taken from Tabel 14.45 and 14.46 from the ISO spec 1553 if (av_cmp_sf_ieee754(acf[c], FLOAT_1)) { 1554 nbits[i] = 23 - av_log2(abs(ctx->raw_samples[c][i])); 1555 } else { 1556 nbits[i] = 23; 1557 } 1558 nbits[i] = FFMIN(nbits[i], highest_byte*8); 1559 } 1560 } 1561 1562 if (!get_bits1(gb)) { //uncompressed 1563 for (i = 0; i < frame_length; ++i) { 1564 if (ctx->raw_samples[c][i] != 0) { 1565 raw_mantissa[c][i] = get_bitsz(gb, nbits[i]); 1566 } 1567 } 1568 } else { //compressed 1569 nchars = 0; 1570 for (i = 0; i < frame_length; ++i) { 1571 if (ctx->raw_samples[c][i]) { 1572 nchars += (int) nbits[i] / 8; 1573 if (nbits[i] & 7) { 1574 ++nchars; 1575 } 1576 } 1577 } 1578 1579 tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray); 1580 if(tmp_32 != nchars) { 1581 av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%"PRId32", %d).\n", tmp_32, nchars); 1582 return AVERROR_INVALIDDATA; 1583 } 1584 1585 j = 0; 1586 for (i = 0; i < frame_length; ++i) { 1587 if (ctx->raw_samples[c][i]) { 1588 if (nbits[i] & 7) { 1589 nbits_aligned = 8 * ((unsigned int)(nbits[i] / 8) + 1); 1590 } else { 1591 nbits_aligned = nbits[i]; 1592 } 1593 acc = 0; 1594 for (k = 0; k < nbits_aligned/8; ++k) { 1595 acc = (acc << 8) + larray[j++]; 1596 } 1597 acc >>= (nbits_aligned - nbits[i]); 1598 raw_mantissa[c][i] = acc; 1599 } 1600 } 1601 } 1602 } 1603 1604 for (i = 0; i < frame_length; ++i) { 1605 SoftFloat_IEEE754 pcm_sf = av_int2sf_ieee754(ctx->raw_samples[c][i], 0); 1606 pcm_sf = av_div_sf_ieee754(pcm_sf, scale); 1607 1608 if (ctx->raw_samples[c][i] != 0) { 1609 if (!av_cmp_sf_ieee754(acf[c], FLOAT_1)) { 1610 pcm_sf = multiply(acf[c], pcm_sf); 1611 } 1612 1613 sign = pcm_sf.sign; 1614 e = pcm_sf.exp; 1615 mantissa = (pcm_sf.mant | 0x800000) + raw_mantissa[c][i]; 1616 1617 while(mantissa >= 0x1000000) { 1618 e++; 1619 mantissa >>= 1; 1620 } 1621 1622 if (mantissa) e += (shift_value[c] - 127); 1623 mantissa &= 0x007fffffUL; 1624 1625 tmp_32 = (sign << 31) | ((e + EXP_BIAS) << 23) | (mantissa); 1626 ctx->raw_samples[c][i] = tmp_32; 1627 } else { 1628 ctx->raw_samples[c][i] = raw_mantissa[c][i] & 0x007fffffUL; 1629 } 1630 } 1631 align_get_bits(gb); 1632 } 1633 return 0; 1634} 1635 1636 1637/** Read the frame data. 1638 */ 1639static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame) 1640{ 1641 ALSSpecificConfig *sconf = &ctx->sconf; 1642 AVCodecContext *avctx = ctx->avctx; 1643 GetBitContext *gb = &ctx->gb; 1644 unsigned int div_blocks[32]; ///< block sizes. 1645 int c; 1646 unsigned int js_blocks[2]; 1647 int channels = avctx->ch_layout.nb_channels; 1648 uint32_t bs_info = 0; 1649 int ret; 1650 1651 // skip the size of the ra unit if present in the frame 1652 if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame) 1653 skip_bits_long(gb, 32); 1654 1655 if (sconf->mc_coding && sconf->joint_stereo) { 1656 ctx->js_switch = get_bits1(gb); 1657 align_get_bits(gb); 1658 } 1659 1660 if (!sconf->mc_coding || ctx->js_switch) { 1661 int independent_bs = !sconf->joint_stereo; 1662 if (get_bits_left(gb) < 7*channels*ctx->num_blocks) 1663 return AVERROR_INVALIDDATA; 1664 for (c = 0; c < channels; c++) { 1665 js_blocks[0] = 0; 1666 js_blocks[1] = 0; 1667 1668 get_block_sizes(ctx, div_blocks, &bs_info); 1669 1670 // if joint_stereo and block_switching is set, independent decoding 1671 // is signaled via the first bit of bs_info 1672 if (sconf->joint_stereo && sconf->block_switching) 1673 if (bs_info >> 31) 1674 independent_bs = 2; 1675 1676 // if this is the last channel, it has to be decoded independently 1677 if (c == channels - 1 || (c & 1)) 1678 independent_bs = 1; 1679 1680 if (independent_bs) { 1681 ret = decode_blocks_ind(ctx, ra_frame, c, 1682 div_blocks, js_blocks); 1683 if (ret < 0) 1684 return ret; 1685 independent_bs--; 1686 } else { 1687 ret = decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks); 1688 if (ret < 0) 1689 return ret; 1690 1691 c++; 1692 } 1693 1694 // store carryover raw samples 1695 memmove(ctx->raw_samples[c] - sconf->max_order, 1696 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1697 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1698 ctx->highest_decoded_channel = c; 1699 } 1700 } else { // multi-channel coding 1701 ALSBlockData bd = { 0 }; 1702 int b, ret; 1703 int *reverted_channels = ctx->reverted_channels; 1704 unsigned int offset = 0; 1705 1706 for (c = 0; c < channels; c++) 1707 if (ctx->chan_data[c] < ctx->chan_data_buffer) { 1708 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data.\n"); 1709 return AVERROR_INVALIDDATA; 1710 } 1711 1712 memset(reverted_channels, 0, sizeof(*reverted_channels) * channels); 1713 1714 bd.ra_block = ra_frame; 1715 bd.prev_raw_samples = ctx->prev_raw_samples; 1716 1717 get_block_sizes(ctx, div_blocks, &bs_info); 1718 1719 for (b = 0; b < ctx->num_blocks; b++) { 1720 bd.block_length = div_blocks[b]; 1721 if (bd.block_length <= 0) { 1722 av_log(ctx->avctx, AV_LOG_WARNING, 1723 "Invalid block length %u in channel data!\n", 1724 bd.block_length); 1725 continue; 1726 } 1727 1728 for (c = 0; c < channels; c++) { 1729 bd.const_block = ctx->const_block + c; 1730 bd.shift_lsbs = ctx->shift_lsbs + c; 1731 bd.opt_order = ctx->opt_order + c; 1732 bd.store_prev_samples = ctx->store_prev_samples + c; 1733 bd.use_ltp = ctx->use_ltp + c; 1734 bd.ltp_lag = ctx->ltp_lag + c; 1735 bd.ltp_gain = ctx->ltp_gain[c]; 1736 bd.lpc_cof = ctx->lpc_cof[c]; 1737 bd.quant_cof = ctx->quant_cof[c]; 1738 bd.raw_samples = ctx->raw_samples[c] + offset; 1739 bd.raw_other = NULL; 1740 1741 if ((ret = read_block(ctx, &bd)) < 0) 1742 return ret; 1743 if ((ret = read_channel_data(ctx, ctx->chan_data[c], c)) < 0) 1744 return ret; 1745 } 1746 1747 for (c = 0; c < channels; c++) { 1748 ret = revert_channel_correlation(ctx, &bd, ctx->chan_data, 1749 reverted_channels, offset, c); 1750 if (ret < 0) 1751 return ret; 1752 } 1753 for (c = 0; c < channels; c++) { 1754 bd.const_block = ctx->const_block + c; 1755 bd.shift_lsbs = ctx->shift_lsbs + c; 1756 bd.opt_order = ctx->opt_order + c; 1757 bd.store_prev_samples = ctx->store_prev_samples + c; 1758 bd.use_ltp = ctx->use_ltp + c; 1759 bd.ltp_lag = ctx->ltp_lag + c; 1760 bd.ltp_gain = ctx->ltp_gain[c]; 1761 bd.lpc_cof = ctx->lpc_cof[c]; 1762 bd.quant_cof = ctx->quant_cof[c]; 1763 bd.raw_samples = ctx->raw_samples[c] + offset; 1764 1765 if ((ret = decode_block(ctx, &bd)) < 0) 1766 return ret; 1767 1768 ctx->highest_decoded_channel = FFMAX(ctx->highest_decoded_channel, c); 1769 } 1770 1771 memset(reverted_channels, 0, channels * sizeof(*reverted_channels)); 1772 offset += div_blocks[b]; 1773 bd.ra_block = 0; 1774 } 1775 1776 // store carryover raw samples 1777 for (c = 0; c < channels; c++) 1778 memmove(ctx->raw_samples[c] - sconf->max_order, 1779 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1780 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1781 } 1782 1783 if (sconf->floating) { 1784 read_diff_float_data(ctx, ra_frame); 1785 } 1786 1787 if (get_bits_left(gb) < 0) { 1788 av_log(ctx->avctx, AV_LOG_ERROR, "Overread %d\n", -get_bits_left(gb)); 1789 return AVERROR_INVALIDDATA; 1790 } 1791 1792 return 0; 1793} 1794 1795 1796/** Decode an ALS frame. 1797 */ 1798static int decode_frame(AVCodecContext *avctx, AVFrame *frame, 1799 int *got_frame_ptr, AVPacket *avpkt) 1800{ 1801 ALSDecContext *ctx = avctx->priv_data; 1802 ALSSpecificConfig *sconf = &ctx->sconf; 1803 const uint8_t *buffer = avpkt->data; 1804 int buffer_size = avpkt->size; 1805 int invalid_frame, ret; 1806 int channels = avctx->ch_layout.nb_channels; 1807 unsigned int c, sample, ra_frame, bytes_read, shift; 1808 1809 if ((ret = init_get_bits8(&ctx->gb, buffer, buffer_size)) < 0) 1810 return ret; 1811 1812 // In the case that the distance between random access frames is set to zero 1813 // (sconf->ra_distance == 0) no frame is treated as a random access frame. 1814 // For the first frame, if prediction is used, all samples used from the 1815 // previous frame are assumed to be zero. 1816 ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance); 1817 1818 // the last frame to decode might have a different length 1819 if (sconf->samples != 0xFFFFFFFF) 1820 ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length, 1821 sconf->frame_length); 1822 else 1823 ctx->cur_frame_length = sconf->frame_length; 1824 1825 ctx->highest_decoded_channel = -1; 1826 // decode the frame data 1827 if ((invalid_frame = read_frame_data(ctx, ra_frame)) < 0) 1828 av_log(ctx->avctx, AV_LOG_WARNING, 1829 "Reading frame data failed. Skipping RA unit.\n"); 1830 1831 if (ctx->highest_decoded_channel == -1) { 1832 av_log(ctx->avctx, AV_LOG_WARNING, 1833 "No channel data decoded.\n"); 1834 return AVERROR_INVALIDDATA; 1835 } 1836 1837 ctx->frame_id++; 1838 1839 /* get output buffer */ 1840 frame->nb_samples = ctx->cur_frame_length; 1841 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) 1842 return ret; 1843 1844 // transform decoded frame into output format 1845 #define INTERLEAVE_OUTPUT(bps) \ 1846 { \ 1847 int##bps##_t *dest = (int##bps##_t*)frame->data[0]; \ 1848 int32_t *raw_samples = ctx->raw_samples[0]; \ 1849 int raw_step = channels > 1 ? ctx->raw_samples[1] - raw_samples : 1; \ 1850 shift = bps - ctx->avctx->bits_per_raw_sample; \ 1851 if (!ctx->cs_switch) { \ 1852 for (sample = 0; sample < ctx->cur_frame_length; sample++) \ 1853 for (c = 0; c < channels; c++) \ 1854 *dest++ = raw_samples[c*raw_step + sample] * (1U << shift); \ 1855 } else { \ 1856 for (sample = 0; sample < ctx->cur_frame_length; sample++) \ 1857 for (c = 0; c < channels; c++) \ 1858 *dest++ = raw_samples[sconf->chan_pos[c]*raw_step + sample] * (1U << shift);\ 1859 } \ 1860 } 1861 1862 if (ctx->avctx->bits_per_raw_sample <= 16) { 1863 INTERLEAVE_OUTPUT(16) 1864 } else { 1865 INTERLEAVE_OUTPUT(32) 1866 } 1867 1868 // update CRC 1869 if (sconf->crc_enabled && (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) { 1870 int swap = HAVE_BIGENDIAN != sconf->msb_first; 1871 1872 if (ctx->avctx->bits_per_raw_sample == 24) { 1873 int32_t *src = (int32_t *)frame->data[0]; 1874 1875 for (sample = 0; 1876 sample < ctx->cur_frame_length * channels; 1877 sample++) { 1878 int32_t v; 1879 1880 if (swap) 1881 v = av_bswap32(src[sample]); 1882 else 1883 v = src[sample]; 1884 if (!HAVE_BIGENDIAN) 1885 v >>= 8; 1886 1887 ctx->crc = av_crc(ctx->crc_table, ctx->crc, (uint8_t*)(&v), 3); 1888 } 1889 } else { 1890 uint8_t *crc_source; 1891 1892 if (swap) { 1893 if (ctx->avctx->bits_per_raw_sample <= 16) { 1894 int16_t *src = (int16_t*) frame->data[0]; 1895 int16_t *dest = (int16_t*) ctx->crc_buffer; 1896 for (sample = 0; 1897 sample < ctx->cur_frame_length * channels; 1898 sample++) 1899 *dest++ = av_bswap16(src[sample]); 1900 } else { 1901 ctx->bdsp.bswap_buf((uint32_t *) ctx->crc_buffer, 1902 (uint32_t *) frame->data[0], 1903 ctx->cur_frame_length * channels); 1904 } 1905 crc_source = ctx->crc_buffer; 1906 } else { 1907 crc_source = frame->data[0]; 1908 } 1909 1910 ctx->crc = av_crc(ctx->crc_table, ctx->crc, crc_source, 1911 ctx->cur_frame_length * channels * 1912 av_get_bytes_per_sample(avctx->sample_fmt)); 1913 } 1914 1915 1916 // check CRC sums if this is the last frame 1917 if (ctx->cur_frame_length != sconf->frame_length && 1918 ctx->crc_org != ctx->crc) { 1919 av_log(avctx, AV_LOG_ERROR, "CRC error.\n"); 1920 if (avctx->err_recognition & AV_EF_EXPLODE) 1921 return AVERROR_INVALIDDATA; 1922 } 1923 } 1924 1925 *got_frame_ptr = 1; 1926 1927 bytes_read = invalid_frame ? buffer_size : 1928 (get_bits_count(&ctx->gb) + 7) >> 3; 1929 1930 return bytes_read; 1931} 1932 1933 1934/** Uninitialize the ALS decoder. 1935 */ 1936static av_cold int decode_end(AVCodecContext *avctx) 1937{ 1938 ALSDecContext *ctx = avctx->priv_data; 1939 int i; 1940 1941 av_freep(&ctx->sconf.chan_pos); 1942 1943 ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status); 1944 1945 av_freep(&ctx->const_block); 1946 av_freep(&ctx->shift_lsbs); 1947 av_freep(&ctx->opt_order); 1948 av_freep(&ctx->store_prev_samples); 1949 av_freep(&ctx->use_ltp); 1950 av_freep(&ctx->ltp_lag); 1951 av_freep(&ctx->ltp_gain); 1952 av_freep(&ctx->ltp_gain_buffer); 1953 av_freep(&ctx->quant_cof); 1954 av_freep(&ctx->lpc_cof); 1955 av_freep(&ctx->quant_cof_buffer); 1956 av_freep(&ctx->lpc_cof_buffer); 1957 av_freep(&ctx->lpc_cof_reversed_buffer); 1958 av_freep(&ctx->prev_raw_samples); 1959 av_freep(&ctx->raw_samples); 1960 av_freep(&ctx->raw_buffer); 1961 av_freep(&ctx->chan_data); 1962 av_freep(&ctx->chan_data_buffer); 1963 av_freep(&ctx->reverted_channels); 1964 av_freep(&ctx->crc_buffer); 1965 if (ctx->mlz) { 1966 av_freep(&ctx->mlz->dict); 1967 av_freep(&ctx->mlz); 1968 } 1969 av_freep(&ctx->acf); 1970 av_freep(&ctx->last_acf_mantissa); 1971 av_freep(&ctx->shift_value); 1972 av_freep(&ctx->last_shift_value); 1973 if (ctx->raw_mantissa) { 1974 for (i = 0; i < avctx->ch_layout.nb_channels; i++) { 1975 av_freep(&ctx->raw_mantissa[i]); 1976 } 1977 av_freep(&ctx->raw_mantissa); 1978 } 1979 av_freep(&ctx->larray); 1980 av_freep(&ctx->nbits); 1981 1982 return 0; 1983} 1984 1985 1986/** Initialize the ALS decoder. 1987 */ 1988static av_cold int decode_init(AVCodecContext *avctx) 1989{ 1990 unsigned int c; 1991 unsigned int channel_size; 1992 int num_buffers, ret; 1993 int channels; 1994 ALSDecContext *ctx = avctx->priv_data; 1995 ALSSpecificConfig *sconf = &ctx->sconf; 1996 ctx->avctx = avctx; 1997 1998 if (!avctx->extradata) { 1999 av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n"); 2000 return AVERROR_INVALIDDATA; 2001 } 2002 2003 if ((ret = read_specific_config(ctx)) < 0) { 2004 av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n"); 2005 return ret; 2006 } 2007 channels = avctx->ch_layout.nb_channels; 2008 2009 if ((ret = check_specific_config(ctx)) < 0) { 2010 return ret; 2011 } 2012 2013 if (sconf->bgmc) { 2014 ret = ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status); 2015 if (ret < 0) 2016 return ret; 2017 } 2018 if (sconf->floating) { 2019 avctx->sample_fmt = AV_SAMPLE_FMT_FLT; 2020 avctx->bits_per_raw_sample = 32; 2021 } else { 2022 avctx->sample_fmt = sconf->resolution > 1 2023 ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S16; 2024 avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8; 2025 if (avctx->bits_per_raw_sample > 32) { 2026 av_log(avctx, AV_LOG_ERROR, "Bits per raw sample %d larger than 32.\n", 2027 avctx->bits_per_raw_sample); 2028 return AVERROR_INVALIDDATA; 2029 } 2030 } 2031 2032 // set maximum Rice parameter for progressive decoding based on resolution 2033 // This is not specified in 14496-3 but actually done by the reference 2034 // codec RM22 revision 2. 2035 ctx->s_max = sconf->resolution > 1 ? 31 : 15; 2036 2037 // set lag value for long-term prediction 2038 ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) + 2039 (avctx->sample_rate >= 192000); 2040 2041 // allocate quantized parcor coefficient buffer 2042 num_buffers = sconf->mc_coding ? channels : 1; 2043 if (num_buffers * (uint64_t)num_buffers > INT_MAX) // protect chan_data_buffer allocation 2044 return AVERROR_INVALIDDATA; 2045 2046 ctx->quant_cof = av_malloc_array(num_buffers, sizeof(*ctx->quant_cof)); 2047 ctx->lpc_cof = av_malloc_array(num_buffers, sizeof(*ctx->lpc_cof)); 2048 ctx->quant_cof_buffer = av_malloc_array(num_buffers * sconf->max_order, 2049 sizeof(*ctx->quant_cof_buffer)); 2050 ctx->lpc_cof_buffer = av_malloc_array(num_buffers * sconf->max_order, 2051 sizeof(*ctx->lpc_cof_buffer)); 2052 ctx->lpc_cof_reversed_buffer = av_malloc_array(sconf->max_order, 2053 sizeof(*ctx->lpc_cof_buffer)); 2054 2055 if (!ctx->quant_cof || !ctx->lpc_cof || 2056 !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer || 2057 !ctx->lpc_cof_reversed_buffer) { 2058 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2059 return AVERROR(ENOMEM); 2060 } 2061 2062 // assign quantized parcor coefficient buffers 2063 for (c = 0; c < num_buffers; c++) { 2064 ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order; 2065 ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order; 2066 } 2067 2068 // allocate and assign lag and gain data buffer for ltp mode 2069 ctx->const_block = av_malloc_array(num_buffers, sizeof(*ctx->const_block)); 2070 ctx->shift_lsbs = av_malloc_array(num_buffers, sizeof(*ctx->shift_lsbs)); 2071 ctx->opt_order = av_malloc_array(num_buffers, sizeof(*ctx->opt_order)); 2072 ctx->store_prev_samples = av_malloc_array(num_buffers, sizeof(*ctx->store_prev_samples)); 2073 ctx->use_ltp = av_calloc(num_buffers, sizeof(*ctx->use_ltp)); 2074 ctx->ltp_lag = av_malloc_array(num_buffers, sizeof(*ctx->ltp_lag)); 2075 ctx->ltp_gain = av_malloc_array(num_buffers, sizeof(*ctx->ltp_gain)); 2076 ctx->ltp_gain_buffer = av_malloc_array(num_buffers * 5, sizeof(*ctx->ltp_gain_buffer)); 2077 2078 if (!ctx->const_block || !ctx->shift_lsbs || 2079 !ctx->opt_order || !ctx->store_prev_samples || 2080 !ctx->use_ltp || !ctx->ltp_lag || 2081 !ctx->ltp_gain || !ctx->ltp_gain_buffer) { 2082 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2083 return AVERROR(ENOMEM); 2084 } 2085 2086 for (c = 0; c < num_buffers; c++) 2087 ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5; 2088 2089 // allocate and assign channel data buffer for mcc mode 2090 if (sconf->mc_coding) { 2091 ctx->chan_data_buffer = av_calloc(num_buffers * num_buffers, 2092 sizeof(*ctx->chan_data_buffer)); 2093 ctx->chan_data = av_calloc(num_buffers, sizeof(*ctx->chan_data)); 2094 ctx->reverted_channels = av_malloc_array(num_buffers, 2095 sizeof(*ctx->reverted_channels)); 2096 2097 if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) { 2098 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2099 return AVERROR(ENOMEM); 2100 } 2101 2102 for (c = 0; c < num_buffers; c++) 2103 ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers; 2104 } else { 2105 ctx->chan_data = NULL; 2106 ctx->chan_data_buffer = NULL; 2107 ctx->reverted_channels = NULL; 2108 } 2109 2110 if (sconf->floating) { 2111 ctx->acf = av_malloc_array(channels, sizeof(*ctx->acf)); 2112 ctx->shift_value = av_malloc_array(channels, sizeof(*ctx->shift_value)); 2113 ctx->last_shift_value = av_malloc_array(channels, sizeof(*ctx->last_shift_value)); 2114 ctx->last_acf_mantissa = av_malloc_array(channels, sizeof(*ctx->last_acf_mantissa)); 2115 ctx->raw_mantissa = av_calloc(channels, sizeof(*ctx->raw_mantissa)); 2116 2117 ctx->larray = av_malloc_array(ctx->cur_frame_length * 4, sizeof(*ctx->larray)); 2118 ctx->nbits = av_malloc_array(ctx->cur_frame_length, sizeof(*ctx->nbits)); 2119 ctx->mlz = av_mallocz(sizeof(*ctx->mlz)); 2120 2121 if (!ctx->mlz || !ctx->acf || !ctx->shift_value || !ctx->last_shift_value 2122 || !ctx->last_acf_mantissa || !ctx->raw_mantissa) { 2123 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2124 return AVERROR(ENOMEM); 2125 } 2126 2127 ret = ff_mlz_init_dict(avctx, ctx->mlz); 2128 if (ret < 0) 2129 return ret; 2130 ff_mlz_flush_dict(ctx->mlz); 2131 2132 for (c = 0; c < channels; ++c) { 2133 ctx->raw_mantissa[c] = av_calloc(ctx->cur_frame_length, sizeof(**ctx->raw_mantissa)); 2134 } 2135 } 2136 2137 channel_size = sconf->frame_length + sconf->max_order; 2138 2139 // allocate previous raw sample buffer 2140 ctx->prev_raw_samples = av_malloc_array(sconf->max_order, sizeof(*ctx->prev_raw_samples)); 2141 ctx->raw_buffer = av_calloc(channels * channel_size, sizeof(*ctx->raw_buffer)); 2142 ctx->raw_samples = av_malloc_array(channels, sizeof(*ctx->raw_samples)); 2143 if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) { 2144 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2145 return AVERROR(ENOMEM); 2146 } 2147 2148 // assign raw samples buffers 2149 ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order; 2150 for (c = 1; c < channels; c++) 2151 ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size; 2152 2153 // allocate crc buffer 2154 if (HAVE_BIGENDIAN != sconf->msb_first && sconf->crc_enabled && 2155 (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) { 2156 ctx->crc_buffer = av_malloc_array(ctx->cur_frame_length * 2157 channels * 2158 av_get_bytes_per_sample(avctx->sample_fmt), 2159 sizeof(*ctx->crc_buffer)); 2160 if (!ctx->crc_buffer) { 2161 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 2162 return AVERROR(ENOMEM); 2163 } 2164 } 2165 2166 ff_bswapdsp_init(&ctx->bdsp); 2167 2168 return 0; 2169} 2170 2171 2172/** Flush (reset) the frame ID after seeking. 2173 */ 2174static av_cold void flush(AVCodecContext *avctx) 2175{ 2176 ALSDecContext *ctx = avctx->priv_data; 2177 2178 ctx->frame_id = 0; 2179} 2180 2181 2182const FFCodec ff_als_decoder = { 2183 .p.name = "als", 2184 .p.long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"), 2185 .p.type = AVMEDIA_TYPE_AUDIO, 2186 .p.id = AV_CODEC_ID_MP4ALS, 2187 .priv_data_size = sizeof(ALSDecContext), 2188 .init = decode_init, 2189 .close = decode_end, 2190 FF_CODEC_DECODE_CB(decode_frame), 2191 .flush = flush, 2192 .p.capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF, 2193 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP, 2194}; 2195