1/* 2 * MPEG-4 ALS decoder 3 * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ googlemail.com> 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_ googlemail.com> 26 */ 27 28 29//#define DEBUG 30 31 32#include "avcodec.h" 33#include "get_bits.h" 34#include "unary.h" 35#include "mpeg4audio.h" 36#include "bytestream.h" 37#include "bgmc.h" 38 39#include <stdint.h> 40 41/** Rice parameters and corresponding index offsets for decoding the 42 * indices of scaled PARCOR values. The table choosen is set globally 43 * by the encoder and stored in ALSSpecificConfig. 44 */ 45static const int8_t parcor_rice_table[3][20][2] = { 46 { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4}, 47 { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3}, 48 { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2}, 49 { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} }, 50 { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4}, 51 { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4}, 52 {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4}, 53 { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} }, 54 { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4}, 55 { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3}, 56 {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3}, 57 { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} } 58}; 59 60 61/** Scaled PARCOR values used for the first two PARCOR coefficients. 62 * To be indexed by the Rice coded indices. 63 * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20) 64 * Actual values are divided by 32 in order to be stored in 16 bits. 65 */ 66static const int16_t parcor_scaled_values[] = { 67 -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32, 68 -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32, 69 -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32, 70 -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32, 71 -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32, 72 -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32, 73 -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32, 74 -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32, 75 -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32, 76 -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32, 77 -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32, 78 -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32, 79 -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32, 80 -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32, 81 -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32, 82 -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32, 83 -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32, 84 -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32, 85 -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32, 86 -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32, 87 -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32, 88 -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32, 89 -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32, 90 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32, 91 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32, 92 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32, 93 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32, 94 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32, 95 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32, 96 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32, 97 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32, 98 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32 99}; 100 101 102/** Gain values of p(0) for long-term prediction. 103 * To be indexed by the Rice coded indices. 104 */ 105static const uint8_t ltp_gain_values [4][4] = { 106 { 0, 8, 16, 24}, 107 {32, 40, 48, 56}, 108 {64, 70, 76, 82}, 109 {88, 92, 96, 100} 110}; 111 112 113/** Inter-channel weighting factors for multi-channel correlation. 114 * To be indexed by the Rice coded indices. 115 */ 116static const int16_t mcc_weightings[] = { 117 204, 192, 179, 166, 153, 140, 128, 115, 118 102, 89, 76, 64, 51, 38, 25, 12, 119 0, -12, -25, -38, -51, -64, -76, -89, 120 -102, -115, -128, -140, -153, -166, -179, -192 121}; 122 123 124/** Tail codes used in arithmetic coding using block Gilbert-Moore codes. 125 */ 126static const uint8_t tail_code[16][6] = { 127 { 74, 44, 25, 13, 7, 3}, 128 { 68, 42, 24, 13, 7, 3}, 129 { 58, 39, 23, 13, 7, 3}, 130 {126, 70, 37, 19, 10, 5}, 131 {132, 70, 37, 20, 10, 5}, 132 {124, 70, 38, 20, 10, 5}, 133 {120, 69, 37, 20, 11, 5}, 134 {116, 67, 37, 20, 11, 5}, 135 {108, 66, 36, 20, 10, 5}, 136 {102, 62, 36, 20, 10, 5}, 137 { 88, 58, 34, 19, 10, 5}, 138 {162, 89, 49, 25, 13, 7}, 139 {156, 87, 49, 26, 14, 7}, 140 {150, 86, 47, 26, 14, 7}, 141 {142, 84, 47, 26, 14, 7}, 142 {131, 79, 46, 26, 14, 7} 143}; 144 145 146enum RA_Flag { 147 RA_FLAG_NONE, 148 RA_FLAG_FRAMES, 149 RA_FLAG_HEADER 150}; 151 152 153typedef struct { 154 uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown 155 int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit 156 int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer 157 int frame_length; ///< frame length for each frame (last frame may differ) 158 int ra_distance; ///< distance between RA frames (in frames, 0...255) 159 enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored 160 int adapt_order; ///< adaptive order: 1 = on, 0 = off 161 int coef_table; ///< table index of Rice code parameters 162 int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off 163 int max_order; ///< maximum prediction order (0..1023) 164 int block_switching; ///< number of block switching levels 165 int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only) 166 int sb_part; ///< sub-block partition 167 int joint_stereo; ///< joint stereo: 1 = on, 0 = off 168 int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off 169 int chan_config; ///< indicates that a chan_config_info field is present 170 int chan_sort; ///< channel rearrangement: 1 = on, 0 = off 171 int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off 172 int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented. 173 int *chan_pos; ///< original channel positions 174} ALSSpecificConfig; 175 176 177typedef struct { 178 int stop_flag; 179 int master_channel; 180 int time_diff_flag; 181 int time_diff_sign; 182 int time_diff_index; 183 int weighting[6]; 184} ALSChannelData; 185 186 187typedef struct { 188 AVCodecContext *avctx; 189 ALSSpecificConfig sconf; 190 GetBitContext gb; 191 unsigned int cur_frame_length; ///< length of the current frame to decode 192 unsigned int frame_id; ///< the frame ID / number of the current frame 193 unsigned int js_switch; ///< if true, joint-stereo decoding is enforced 194 unsigned int num_blocks; ///< number of blocks used in the current frame 195 unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding 196 uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC 197 unsigned int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC 198 int ltp_lag_length; ///< number of bits used for ltp lag value 199 int *use_ltp; ///< contains use_ltp flags for all channels 200 int *ltp_lag; ///< contains ltp lag values for all channels 201 int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel 202 int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter 203 int32_t **quant_cof; ///< quantized parcor coefficients for a channel 204 int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients 205 int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel 206 int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter 207 int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer 208 ALSChannelData **chan_data; ///< channel data for multi-channel correlation 209 ALSChannelData *chan_data_buffer; ///< contains channel data for all channels 210 int *reverted_channels; ///< stores a flag for each reverted channel 211 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block 212 int32_t **raw_samples; ///< decoded raw samples for each channel 213 int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples 214} ALSDecContext; 215 216 217typedef struct { 218 unsigned int block_length; ///< number of samples within the block 219 unsigned int ra_block; ///< if true, this is a random access block 220 int const_block; ///< if true, this is a constant value block 221 int32_t const_val; ///< the sample value of a constant block 222 int js_blocks; ///< true if this block contains a difference signal 223 unsigned int shift_lsbs; ///< shift of values for this block 224 unsigned int opt_order; ///< prediction order of this block 225 int store_prev_samples;///< if true, carryover samples have to be stored 226 int *use_ltp; ///< if true, long-term prediction is used 227 int *ltp_lag; ///< lag value for long-term prediction 228 int *ltp_gain; ///< gain values for ltp 5-tap filter 229 int32_t *quant_cof; ///< quantized parcor coefficients 230 int32_t *lpc_cof; ///< coefficients of the direct form prediction 231 int32_t *raw_samples; ///< decoded raw samples / residuals for this block 232 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block 233 int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair 234} ALSBlockData; 235 236 237static av_cold void dprint_specific_config(ALSDecContext *ctx) 238{ 239#ifdef DEBUG 240 AVCodecContext *avctx = ctx->avctx; 241 ALSSpecificConfig *sconf = &ctx->sconf; 242 243 dprintf(avctx, "resolution = %i\n", sconf->resolution); 244 dprintf(avctx, "floating = %i\n", sconf->floating); 245 dprintf(avctx, "frame_length = %i\n", sconf->frame_length); 246 dprintf(avctx, "ra_distance = %i\n", sconf->ra_distance); 247 dprintf(avctx, "ra_flag = %i\n", sconf->ra_flag); 248 dprintf(avctx, "adapt_order = %i\n", sconf->adapt_order); 249 dprintf(avctx, "coef_table = %i\n", sconf->coef_table); 250 dprintf(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction); 251 dprintf(avctx, "max_order = %i\n", sconf->max_order); 252 dprintf(avctx, "block_switching = %i\n", sconf->block_switching); 253 dprintf(avctx, "bgmc = %i\n", sconf->bgmc); 254 dprintf(avctx, "sb_part = %i\n", sconf->sb_part); 255 dprintf(avctx, "joint_stereo = %i\n", sconf->joint_stereo); 256 dprintf(avctx, "mc_coding = %i\n", sconf->mc_coding); 257 dprintf(avctx, "chan_config = %i\n", sconf->chan_config); 258 dprintf(avctx, "chan_sort = %i\n", sconf->chan_sort); 259 dprintf(avctx, "RLSLMS = %i\n", sconf->rlslms); 260 dprintf(avctx, "chan_config_info = %i\n", sconf->chan_config_info); 261#endif 262} 263 264 265/** Reads an ALSSpecificConfig from a buffer into the output struct. 266 */ 267static av_cold int read_specific_config(ALSDecContext *ctx) 268{ 269 GetBitContext gb; 270 uint64_t ht_size; 271 int i, config_offset, crc_enabled; 272 MPEG4AudioConfig m4ac; 273 ALSSpecificConfig *sconf = &ctx->sconf; 274 AVCodecContext *avctx = ctx->avctx; 275 uint32_t als_id, header_size, trailer_size; 276 277 init_get_bits(&gb, avctx->extradata, avctx->extradata_size * 8); 278 279 config_offset = ff_mpeg4audio_get_config(&m4ac, avctx->extradata, 280 avctx->extradata_size); 281 282 if (config_offset < 0) 283 return -1; 284 285 skip_bits_long(&gb, config_offset); 286 287 if (get_bits_left(&gb) < (30 << 3)) 288 return -1; 289 290 // read the fixed items 291 als_id = get_bits_long(&gb, 32); 292 avctx->sample_rate = m4ac.sample_rate; 293 skip_bits_long(&gb, 32); // sample rate already known 294 sconf->samples = get_bits_long(&gb, 32); 295 avctx->channels = m4ac.channels; 296 skip_bits(&gb, 16); // number of channels already knwon 297 skip_bits(&gb, 3); // skip file_type 298 sconf->resolution = get_bits(&gb, 3); 299 sconf->floating = get_bits1(&gb); 300 skip_bits1(&gb); // skip msb_first 301 sconf->frame_length = get_bits(&gb, 16) + 1; 302 sconf->ra_distance = get_bits(&gb, 8); 303 sconf->ra_flag = get_bits(&gb, 2); 304 sconf->adapt_order = get_bits1(&gb); 305 sconf->coef_table = get_bits(&gb, 2); 306 sconf->long_term_prediction = get_bits1(&gb); 307 sconf->max_order = get_bits(&gb, 10); 308 sconf->block_switching = get_bits(&gb, 2); 309 sconf->bgmc = get_bits1(&gb); 310 sconf->sb_part = get_bits1(&gb); 311 sconf->joint_stereo = get_bits1(&gb); 312 sconf->mc_coding = get_bits1(&gb); 313 sconf->chan_config = get_bits1(&gb); 314 sconf->chan_sort = get_bits1(&gb); 315 crc_enabled = get_bits1(&gb); 316 sconf->rlslms = get_bits1(&gb); 317 skip_bits(&gb, 5); // skip 5 reserved bits 318 skip_bits1(&gb); // skip aux_data_enabled 319 320 321 // check for ALSSpecificConfig struct 322 if (als_id != MKBETAG('A','L','S','\0')) 323 return -1; 324 325 ctx->cur_frame_length = sconf->frame_length; 326 327 // read channel config 328 if (sconf->chan_config) 329 sconf->chan_config_info = get_bits(&gb, 16); 330 // TODO: use this to set avctx->channel_layout 331 332 333 // read channel sorting 334 if (sconf->chan_sort && avctx->channels > 1) { 335 int chan_pos_bits = av_ceil_log2(avctx->channels); 336 int bits_needed = avctx->channels * chan_pos_bits + 7; 337 if (get_bits_left(&gb) < bits_needed) 338 return -1; 339 340 if (!(sconf->chan_pos = av_malloc(avctx->channels * sizeof(*sconf->chan_pos)))) 341 return AVERROR(ENOMEM); 342 343 for (i = 0; i < avctx->channels; i++) 344 sconf->chan_pos[i] = get_bits(&gb, chan_pos_bits); 345 346 align_get_bits(&gb); 347 // TODO: use this to actually do channel sorting 348 } else { 349 sconf->chan_sort = 0; 350 } 351 352 353 // read fixed header and trailer sizes, 354 // if size = 0xFFFFFFFF then there is no data field! 355 if (get_bits_left(&gb) < 64) 356 return -1; 357 358 header_size = get_bits_long(&gb, 32); 359 trailer_size = get_bits_long(&gb, 32); 360 if (header_size == 0xFFFFFFFF) 361 header_size = 0; 362 if (trailer_size == 0xFFFFFFFF) 363 trailer_size = 0; 364 365 ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3; 366 367 368 // skip the header and trailer data 369 if (get_bits_left(&gb) < ht_size) 370 return -1; 371 372 if (ht_size > INT32_MAX) 373 return -1; 374 375 skip_bits_long(&gb, ht_size); 376 377 378 // skip the crc data 379 if (crc_enabled) { 380 if (get_bits_left(&gb) < 32) 381 return -1; 382 383 skip_bits_long(&gb, 32); 384 } 385 386 387 // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data) 388 389 dprint_specific_config(ctx); 390 391 return 0; 392} 393 394 395/** Checks the ALSSpecificConfig for unsupported features. 396 */ 397static int check_specific_config(ALSDecContext *ctx) 398{ 399 ALSSpecificConfig *sconf = &ctx->sconf; 400 int error = 0; 401 402 // report unsupported feature and set error value 403 #define MISSING_ERR(cond, str, errval) \ 404 { \ 405 if (cond) { \ 406 av_log_missing_feature(ctx->avctx, str, 0); \ 407 error = errval; \ 408 } \ 409 } 410 411 MISSING_ERR(sconf->floating, "Floating point decoding", -1); 412 MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", -1); 413 MISSING_ERR(sconf->chan_sort, "Channel sorting", 0); 414 415 return error; 416} 417 418 419/** Parses the bs_info field to extract the block partitioning used in 420 * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2. 421 */ 422static void parse_bs_info(const uint32_t bs_info, unsigned int n, 423 unsigned int div, unsigned int **div_blocks, 424 unsigned int *num_blocks) 425{ 426 if (n < 31 && ((bs_info << n) & 0x40000000)) { 427 // if the level is valid and the investigated bit n is set 428 // then recursively check both children at bits (2n+1) and (2n+2) 429 n *= 2; 430 div += 1; 431 parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks); 432 parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks); 433 } else { 434 // else the bit is not set or the last level has been reached 435 // (bit implicitly not set) 436 **div_blocks = div; 437 (*div_blocks)++; 438 (*num_blocks)++; 439 } 440} 441 442 443/** Reads and decodes a Rice codeword. 444 */ 445static int32_t decode_rice(GetBitContext *gb, unsigned int k) 446{ 447 int max = get_bits_left(gb) - k; 448 int q = get_unary(gb, 0, max); 449 int r = k ? get_bits1(gb) : !(q & 1); 450 451 if (k > 1) { 452 q <<= (k - 1); 453 q += get_bits_long(gb, k - 1); 454 } else if (!k) { 455 q >>= 1; 456 } 457 return r ? q : ~q; 458} 459 460 461/** Converts PARCOR coefficient k to direct filter coefficient. 462 */ 463static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof) 464{ 465 int i, j; 466 467 for (i = 0, j = k - 1; i < j; i++, j--) { 468 int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20); 469 cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20); 470 cof[i] += tmp1; 471 } 472 if (i == j) 473 cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20); 474 475 cof[k] = par[k]; 476} 477 478 479/** Reads block switching field if necessary and sets actual block sizes. 480 * Also assures that the block sizes of the last frame correspond to the 481 * actual number of samples. 482 */ 483static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks, 484 uint32_t *bs_info) 485{ 486 ALSSpecificConfig *sconf = &ctx->sconf; 487 GetBitContext *gb = &ctx->gb; 488 unsigned int *ptr_div_blocks = div_blocks; 489 unsigned int b; 490 491 if (sconf->block_switching) { 492 unsigned int bs_info_len = 1 << (sconf->block_switching + 2); 493 *bs_info = get_bits_long(gb, bs_info_len); 494 *bs_info <<= (32 - bs_info_len); 495 } 496 497 ctx->num_blocks = 0; 498 parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks); 499 500 // The last frame may have an overdetermined block structure given in 501 // the bitstream. In that case the defined block structure would need 502 // more samples than available to be consistent. 503 // The block structure is actually used but the block sizes are adapted 504 // to fit the actual number of available samples. 505 // Example: 5 samples, 2nd level block sizes: 2 2 2 2. 506 // This results in the actual block sizes: 2 2 1 0. 507 // This is not specified in 14496-3 but actually done by the reference 508 // codec RM22 revision 2. 509 // This appears to happen in case of an odd number of samples in the last 510 // frame which is actually not allowed by the block length switching part 511 // of 14496-3. 512 // The ALS conformance files feature an odd number of samples in the last 513 // frame. 514 515 for (b = 0; b < ctx->num_blocks; b++) 516 div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b]; 517 518 if (ctx->cur_frame_length != ctx->sconf.frame_length) { 519 unsigned int remaining = ctx->cur_frame_length; 520 521 for (b = 0; b < ctx->num_blocks; b++) { 522 if (remaining <= div_blocks[b]) { 523 div_blocks[b] = remaining; 524 ctx->num_blocks = b + 1; 525 break; 526 } 527 528 remaining -= div_blocks[b]; 529 } 530 } 531} 532 533 534/** Reads the block data for a constant block 535 */ 536static void read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd) 537{ 538 ALSSpecificConfig *sconf = &ctx->sconf; 539 AVCodecContext *avctx = ctx->avctx; 540 GetBitContext *gb = &ctx->gb; 541 542 bd->const_val = 0; 543 bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence) 544 bd->js_blocks = get_bits1(gb); 545 546 // skip 5 reserved bits 547 skip_bits(gb, 5); 548 549 if (bd->const_block) { 550 unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample; 551 bd->const_val = get_sbits_long(gb, const_val_bits); 552 } 553 554 // ensure constant block decoding by reusing this field 555 bd->const_block = 1; 556} 557 558 559/** Decodes the block data for a constant block 560 */ 561static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd) 562{ 563 int smp = bd->block_length; 564 int32_t val = bd->const_val; 565 int32_t *dst = bd->raw_samples; 566 567 // write raw samples into buffer 568 for (; smp; smp--) 569 *dst++ = val; 570} 571 572 573/** Reads the block data for a non-constant block 574 */ 575static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd) 576{ 577 ALSSpecificConfig *sconf = &ctx->sconf; 578 AVCodecContext *avctx = ctx->avctx; 579 GetBitContext *gb = &ctx->gb; 580 unsigned int k; 581 unsigned int s[8]; 582 unsigned int sx[8]; 583 unsigned int sub_blocks, log2_sub_blocks, sb_length; 584 unsigned int start = 0; 585 unsigned int opt_order; 586 int sb; 587 int32_t *quant_cof = bd->quant_cof; 588 int32_t *current_res; 589 590 591 // ensure variable block decoding by reusing this field 592 bd->const_block = 0; 593 594 bd->opt_order = 1; 595 bd->js_blocks = get_bits1(gb); 596 597 opt_order = bd->opt_order; 598 599 // determine the number of subblocks for entropy decoding 600 if (!sconf->bgmc && !sconf->sb_part) { 601 log2_sub_blocks = 0; 602 } else { 603 if (sconf->bgmc && sconf->sb_part) 604 log2_sub_blocks = get_bits(gb, 2); 605 else 606 log2_sub_blocks = 2 * get_bits1(gb); 607 } 608 609 sub_blocks = 1 << log2_sub_blocks; 610 611 // do not continue in case of a damaged stream since 612 // block_length must be evenly divisible by sub_blocks 613 if (bd->block_length & (sub_blocks - 1)) { 614 av_log(avctx, AV_LOG_WARNING, 615 "Block length is not evenly divisible by the number of subblocks.\n"); 616 return -1; 617 } 618 619 sb_length = bd->block_length >> log2_sub_blocks; 620 621 if (sconf->bgmc) { 622 s[0] = get_bits(gb, 8 + (sconf->resolution > 1)); 623 for (k = 1; k < sub_blocks; k++) 624 s[k] = s[k - 1] + decode_rice(gb, 2); 625 626 for (k = 0; k < sub_blocks; k++) { 627 sx[k] = s[k] & 0x0F; 628 s [k] >>= 4; 629 } 630 } else { 631 s[0] = get_bits(gb, 4 + (sconf->resolution > 1)); 632 for (k = 1; k < sub_blocks; k++) 633 s[k] = s[k - 1] + decode_rice(gb, 0); 634 } 635 636 if (get_bits1(gb)) 637 bd->shift_lsbs = get_bits(gb, 4) + 1; 638 639 bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || bd->shift_lsbs; 640 641 642 if (!sconf->rlslms) { 643 if (sconf->adapt_order) { 644 int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1, 645 2, sconf->max_order + 1)); 646 bd->opt_order = get_bits(gb, opt_order_length); 647 } else { 648 bd->opt_order = sconf->max_order; 649 } 650 651 opt_order = bd->opt_order; 652 653 if (opt_order) { 654 int add_base; 655 656 if (sconf->coef_table == 3) { 657 add_base = 0x7F; 658 659 // read coefficient 0 660 quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)]; 661 662 // read coefficient 1 663 if (opt_order > 1) 664 quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)]; 665 666 // read coefficients 2 to opt_order 667 for (k = 2; k < opt_order; k++) 668 quant_cof[k] = get_bits(gb, 7); 669 } else { 670 int k_max; 671 add_base = 1; 672 673 // read coefficient 0 to 19 674 k_max = FFMIN(opt_order, 20); 675 for (k = 0; k < k_max; k++) { 676 int rice_param = parcor_rice_table[sconf->coef_table][k][1]; 677 int offset = parcor_rice_table[sconf->coef_table][k][0]; 678 quant_cof[k] = decode_rice(gb, rice_param) + offset; 679 } 680 681 // read coefficients 20 to 126 682 k_max = FFMIN(opt_order, 127); 683 for (; k < k_max; k++) 684 quant_cof[k] = decode_rice(gb, 2) + (k & 1); 685 686 // read coefficients 127 to opt_order 687 for (; k < opt_order; k++) 688 quant_cof[k] = decode_rice(gb, 1); 689 690 quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64]; 691 692 if (opt_order > 1) 693 quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64]; 694 } 695 696 for (k = 2; k < opt_order; k++) 697 quant_cof[k] = (quant_cof[k] << 14) + (add_base << 13); 698 } 699 } 700 701 // read LTP gain and lag values 702 if (sconf->long_term_prediction) { 703 *bd->use_ltp = get_bits1(gb); 704 705 if (*bd->use_ltp) { 706 int r, c; 707 708 bd->ltp_gain[0] = decode_rice(gb, 1) << 3; 709 bd->ltp_gain[1] = decode_rice(gb, 2) << 3; 710 711 r = get_unary(gb, 0, 4); 712 c = get_bits(gb, 2); 713 bd->ltp_gain[2] = ltp_gain_values[r][c]; 714 715 bd->ltp_gain[3] = decode_rice(gb, 2) << 3; 716 bd->ltp_gain[4] = decode_rice(gb, 1) << 3; 717 718 *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length); 719 *bd->ltp_lag += FFMAX(4, opt_order + 1); 720 } 721 } 722 723 // read first value and residuals in case of a random access block 724 if (bd->ra_block) { 725 if (opt_order) 726 bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4); 727 if (opt_order > 1) 728 bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max)); 729 if (opt_order > 2) 730 bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max)); 731 732 start = FFMIN(opt_order, 3); 733 } 734 735 // read all residuals 736 if (sconf->bgmc) { 737 unsigned int delta[sub_blocks]; 738 unsigned int k [sub_blocks]; 739 unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5); 740 unsigned int i = start; 741 742 // read most significant bits 743 unsigned int high; 744 unsigned int low; 745 unsigned int value; 746 747 ff_bgmc_decode_init(gb, &high, &low, &value); 748 749 current_res = bd->raw_samples + start; 750 751 for (sb = 0; sb < sub_blocks; sb++, i = 0) { 752 k [sb] = s[sb] > b ? s[sb] - b : 0; 753 delta[sb] = 5 - s[sb] + k[sb]; 754 755 ff_bgmc_decode(gb, sb_length, current_res, 756 delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status); 757 758 current_res += sb_length; 759 } 760 761 ff_bgmc_decode_end(gb); 762 763 764 // read least significant bits and tails 765 i = start; 766 current_res = bd->raw_samples + start; 767 768 for (sb = 0; sb < sub_blocks; sb++, i = 0) { 769 unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]]; 770 unsigned int cur_k = k[sb]; 771 unsigned int cur_s = s[sb]; 772 773 for (; i < sb_length; i++) { 774 int32_t res = *current_res; 775 776 if (res == cur_tail_code) { 777 unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10)) 778 << (5 - delta[sb]); 779 780 res = decode_rice(gb, cur_s); 781 782 if (res >= 0) { 783 res += (max_msb ) << cur_k; 784 } else { 785 res -= (max_msb - 1) << cur_k; 786 } 787 } else { 788 if (res > cur_tail_code) 789 res--; 790 791 if (res & 1) 792 res = -res; 793 794 res >>= 1; 795 796 if (cur_k) { 797 res <<= cur_k; 798 res |= get_bits_long(gb, cur_k); 799 } 800 } 801 802 *current_res++ = res; 803 } 804 } 805 } else { 806 current_res = bd->raw_samples + start; 807 808 for (sb = 0; sb < sub_blocks; sb++, start = 0) 809 for (; start < sb_length; start++) 810 *current_res++ = decode_rice(gb, s[sb]); 811 } 812 813 if (!sconf->mc_coding || ctx->js_switch) 814 align_get_bits(gb); 815 816 return 0; 817} 818 819 820/** Decodes the block data for a non-constant block 821 */ 822static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd) 823{ 824 ALSSpecificConfig *sconf = &ctx->sconf; 825 unsigned int block_length = bd->block_length; 826 unsigned int smp = 0; 827 unsigned int k; 828 int opt_order = bd->opt_order; 829 int sb; 830 int64_t y; 831 int32_t *quant_cof = bd->quant_cof; 832 int32_t *lpc_cof = bd->lpc_cof; 833 int32_t *raw_samples = bd->raw_samples; 834 int32_t *raw_samples_end = bd->raw_samples + bd->block_length; 835 int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer; 836 837 // reverse long-term prediction 838 if (*bd->use_ltp) { 839 int ltp_smp; 840 841 for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) { 842 int center = ltp_smp - *bd->ltp_lag; 843 int begin = FFMAX(0, center - 2); 844 int end = center + 3; 845 int tab = 5 - (end - begin); 846 int base; 847 848 y = 1 << 6; 849 850 for (base = begin; base < end; base++, tab++) 851 y += MUL64(bd->ltp_gain[tab], raw_samples[base]); 852 853 raw_samples[ltp_smp] += y >> 7; 854 } 855 } 856 857 // reconstruct all samples from residuals 858 if (bd->ra_block) { 859 for (smp = 0; smp < opt_order; smp++) { 860 y = 1 << 19; 861 862 for (sb = 0; sb < smp; sb++) 863 y += MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]); 864 865 *raw_samples++ -= y >> 20; 866 parcor_to_lpc(smp, quant_cof, lpc_cof); 867 } 868 } else { 869 for (k = 0; k < opt_order; k++) 870 parcor_to_lpc(k, quant_cof, lpc_cof); 871 872 // store previous samples in case that they have to be altered 873 if (bd->store_prev_samples) 874 memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order, 875 sizeof(*bd->prev_raw_samples) * sconf->max_order); 876 877 // reconstruct difference signal for prediction (joint-stereo) 878 if (bd->js_blocks && bd->raw_other) { 879 int32_t *left, *right; 880 881 if (bd->raw_other > raw_samples) { // D = R - L 882 left = raw_samples; 883 right = bd->raw_other; 884 } else { // D = R - L 885 left = bd->raw_other; 886 right = raw_samples; 887 } 888 889 for (sb = -1; sb >= -sconf->max_order; sb--) 890 raw_samples[sb] = right[sb] - left[sb]; 891 } 892 893 // reconstruct shifted signal 894 if (bd->shift_lsbs) 895 for (sb = -1; sb >= -sconf->max_order; sb--) 896 raw_samples[sb] >>= bd->shift_lsbs; 897 } 898 899 // reverse linear prediction coefficients for efficiency 900 lpc_cof = lpc_cof + opt_order; 901 902 for (sb = 0; sb < opt_order; sb++) 903 lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)]; 904 905 // reconstruct raw samples 906 raw_samples = bd->raw_samples + smp; 907 lpc_cof = lpc_cof_reversed + opt_order; 908 909 for (; raw_samples < raw_samples_end; raw_samples++) { 910 y = 1 << 19; 911 912 for (sb = -opt_order; sb < 0; sb++) 913 y += MUL64(lpc_cof[sb], raw_samples[sb]); 914 915 *raw_samples -= y >> 20; 916 } 917 918 raw_samples = bd->raw_samples; 919 920 // restore previous samples in case that they have been altered 921 if (bd->store_prev_samples) 922 memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples, 923 sizeof(*raw_samples) * sconf->max_order); 924 925 return 0; 926} 927 928 929/** Reads the block data. 930 */ 931static int read_block(ALSDecContext *ctx, ALSBlockData *bd) 932{ 933 GetBitContext *gb = &ctx->gb; 934 935 // read block type flag and read the samples accordingly 936 if (get_bits1(gb)) { 937 if (read_var_block_data(ctx, bd)) 938 return -1; 939 } else { 940 read_const_block_data(ctx, bd); 941 } 942 943 return 0; 944} 945 946 947/** Decodes the block data. 948 */ 949static int decode_block(ALSDecContext *ctx, ALSBlockData *bd) 950{ 951 unsigned int smp; 952 953 // read block type flag and read the samples accordingly 954 if (bd->const_block) 955 decode_const_block_data(ctx, bd); 956 else if (decode_var_block_data(ctx, bd)) 957 return -1; 958 959 // TODO: read RLSLMS extension data 960 961 if (bd->shift_lsbs) 962 for (smp = 0; smp < bd->block_length; smp++) 963 bd->raw_samples[smp] <<= bd->shift_lsbs; 964 965 return 0; 966} 967 968 969/** Reads and decodes block data successively. 970 */ 971static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd) 972{ 973 int ret; 974 975 ret = read_block(ctx, bd); 976 977 if (ret) 978 return ret; 979 980 ret = decode_block(ctx, bd); 981 982 return ret; 983} 984 985 986/** Computes the number of samples left to decode for the current frame and 987 * sets these samples to zero. 988 */ 989static void zero_remaining(unsigned int b, unsigned int b_max, 990 const unsigned int *div_blocks, int32_t *buf) 991{ 992 unsigned int count = 0; 993 994 while (b < b_max) 995 count += div_blocks[b]; 996 997 if (count) 998 memset(buf, 0, sizeof(*buf) * count); 999} 1000 1001 1002/** Decodes blocks independently. 1003 */ 1004static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame, 1005 unsigned int c, const unsigned int *div_blocks, 1006 unsigned int *js_blocks) 1007{ 1008 unsigned int b; 1009 ALSBlockData bd; 1010 1011 memset(&bd, 0, sizeof(ALSBlockData)); 1012 1013 bd.ra_block = ra_frame; 1014 bd.use_ltp = ctx->use_ltp; 1015 bd.ltp_lag = ctx->ltp_lag; 1016 bd.ltp_gain = ctx->ltp_gain[0]; 1017 bd.quant_cof = ctx->quant_cof[0]; 1018 bd.lpc_cof = ctx->lpc_cof[0]; 1019 bd.prev_raw_samples = ctx->prev_raw_samples; 1020 bd.raw_samples = ctx->raw_samples[c]; 1021 1022 1023 for (b = 0; b < ctx->num_blocks; b++) { 1024 bd.shift_lsbs = 0; 1025 bd.block_length = div_blocks[b]; 1026 1027 if (read_decode_block(ctx, &bd)) { 1028 // damaged block, write zero for the rest of the frame 1029 zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples); 1030 return -1; 1031 } 1032 bd.raw_samples += div_blocks[b]; 1033 bd.ra_block = 0; 1034 } 1035 1036 return 0; 1037} 1038 1039 1040/** Decodes blocks dependently. 1041 */ 1042static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame, 1043 unsigned int c, const unsigned int *div_blocks, 1044 unsigned int *js_blocks) 1045{ 1046 ALSSpecificConfig *sconf = &ctx->sconf; 1047 unsigned int offset = 0; 1048 unsigned int b; 1049 ALSBlockData bd[2]; 1050 1051 memset(bd, 0, 2 * sizeof(ALSBlockData)); 1052 1053 bd[0].ra_block = ra_frame; 1054 bd[0].use_ltp = ctx->use_ltp; 1055 bd[0].ltp_lag = ctx->ltp_lag; 1056 bd[0].ltp_gain = ctx->ltp_gain[0]; 1057 bd[0].quant_cof = ctx->quant_cof[0]; 1058 bd[0].lpc_cof = ctx->lpc_cof[0]; 1059 bd[0].prev_raw_samples = ctx->prev_raw_samples; 1060 bd[0].js_blocks = *js_blocks; 1061 1062 bd[1].ra_block = ra_frame; 1063 bd[1].use_ltp = ctx->use_ltp; 1064 bd[1].ltp_lag = ctx->ltp_lag; 1065 bd[1].ltp_gain = ctx->ltp_gain[0]; 1066 bd[1].quant_cof = ctx->quant_cof[0]; 1067 bd[1].lpc_cof = ctx->lpc_cof[0]; 1068 bd[1].prev_raw_samples = ctx->prev_raw_samples; 1069 bd[1].js_blocks = *(js_blocks + 1); 1070 1071 // decode all blocks 1072 for (b = 0; b < ctx->num_blocks; b++) { 1073 unsigned int s; 1074 1075 bd[0].shift_lsbs = 0; 1076 bd[1].shift_lsbs = 0; 1077 1078 bd[0].block_length = div_blocks[b]; 1079 bd[1].block_length = div_blocks[b]; 1080 1081 bd[0].raw_samples = ctx->raw_samples[c ] + offset; 1082 bd[1].raw_samples = ctx->raw_samples[c + 1] + offset; 1083 1084 bd[0].raw_other = bd[1].raw_samples; 1085 bd[1].raw_other = bd[0].raw_samples; 1086 1087 if(read_decode_block(ctx, &bd[0]) || read_decode_block(ctx, &bd[1])) { 1088 // damaged block, write zero for the rest of the frame 1089 zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples); 1090 zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples); 1091 return -1; 1092 } 1093 1094 // reconstruct joint-stereo blocks 1095 if (bd[0].js_blocks) { 1096 if (bd[1].js_blocks) 1097 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair!\n"); 1098 1099 for (s = 0; s < div_blocks[b]; s++) 1100 bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s]; 1101 } else if (bd[1].js_blocks) { 1102 for (s = 0; s < div_blocks[b]; s++) 1103 bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s]; 1104 } 1105 1106 offset += div_blocks[b]; 1107 bd[0].ra_block = 0; 1108 bd[1].ra_block = 0; 1109 } 1110 1111 // store carryover raw samples, 1112 // the others channel raw samples are stored by the calling function. 1113 memmove(ctx->raw_samples[c] - sconf->max_order, 1114 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1115 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1116 1117 return 0; 1118} 1119 1120 1121/** Reads the channel data. 1122 */ 1123static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c) 1124{ 1125 GetBitContext *gb = &ctx->gb; 1126 ALSChannelData *current = cd; 1127 unsigned int channels = ctx->avctx->channels; 1128 int entries = 0; 1129 1130 while (entries < channels && !(current->stop_flag = get_bits1(gb))) { 1131 current->master_channel = get_bits_long(gb, av_ceil_log2(channels)); 1132 1133 if (current->master_channel >= channels) { 1134 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel!\n"); 1135 return -1; 1136 } 1137 1138 if (current->master_channel != c) { 1139 current->time_diff_flag = get_bits1(gb); 1140 current->weighting[0] = mcc_weightings[av_clip(decode_rice(gb, 1) + 16, 0, 32)]; 1141 current->weighting[1] = mcc_weightings[av_clip(decode_rice(gb, 2) + 14, 0, 32)]; 1142 current->weighting[2] = mcc_weightings[av_clip(decode_rice(gb, 1) + 16, 0, 32)]; 1143 1144 if (current->time_diff_flag) { 1145 current->weighting[3] = mcc_weightings[av_clip(decode_rice(gb, 1) + 16, 0, 32)]; 1146 current->weighting[4] = mcc_weightings[av_clip(decode_rice(gb, 1) + 16, 0, 32)]; 1147 current->weighting[5] = mcc_weightings[av_clip(decode_rice(gb, 1) + 16, 0, 32)]; 1148 1149 current->time_diff_sign = get_bits1(gb); 1150 current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3; 1151 } 1152 } 1153 1154 current++; 1155 entries++; 1156 } 1157 1158 if (entries == channels) { 1159 av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data!\n"); 1160 return -1; 1161 } 1162 1163 align_get_bits(gb); 1164 return 0; 1165} 1166 1167 1168/** Recursively reverts the inter-channel correlation for a block. 1169 */ 1170static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd, 1171 ALSChannelData **cd, int *reverted, 1172 unsigned int offset, int c) 1173{ 1174 ALSChannelData *ch = cd[c]; 1175 unsigned int dep = 0; 1176 unsigned int channels = ctx->avctx->channels; 1177 1178 if (reverted[c]) 1179 return 0; 1180 1181 reverted[c] = 1; 1182 1183 while (dep < channels && !ch[dep].stop_flag) { 1184 revert_channel_correlation(ctx, bd, cd, reverted, offset, 1185 ch[dep].master_channel); 1186 1187 dep++; 1188 } 1189 1190 if (dep == channels) { 1191 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation!\n"); 1192 return -1; 1193 } 1194 1195 bd->use_ltp = ctx->use_ltp + c; 1196 bd->ltp_lag = ctx->ltp_lag + c; 1197 bd->ltp_gain = ctx->ltp_gain[c]; 1198 bd->lpc_cof = ctx->lpc_cof[c]; 1199 bd->quant_cof = ctx->quant_cof[c]; 1200 bd->raw_samples = ctx->raw_samples[c] + offset; 1201 1202 dep = 0; 1203 while (!ch[dep].stop_flag) { 1204 unsigned int smp; 1205 unsigned int begin = 1; 1206 unsigned int end = bd->block_length - 1; 1207 int64_t y; 1208 int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset; 1209 1210 if (ch[dep].time_diff_flag) { 1211 int t = ch[dep].time_diff_index; 1212 1213 if (ch[dep].time_diff_sign) { 1214 t = -t; 1215 begin -= t; 1216 } else { 1217 end -= t; 1218 } 1219 1220 for (smp = begin; smp < end; smp++) { 1221 y = (1 << 6) + 1222 MUL64(ch[dep].weighting[0], master[smp - 1 ]) + 1223 MUL64(ch[dep].weighting[1], master[smp ]) + 1224 MUL64(ch[dep].weighting[2], master[smp + 1 ]) + 1225 MUL64(ch[dep].weighting[3], master[smp - 1 + t]) + 1226 MUL64(ch[dep].weighting[4], master[smp + t]) + 1227 MUL64(ch[dep].weighting[5], master[smp + 1 + t]); 1228 1229 bd->raw_samples[smp] += y >> 7; 1230 } 1231 } else { 1232 for (smp = begin; smp < end; smp++) { 1233 y = (1 << 6) + 1234 MUL64(ch[dep].weighting[0], master[smp - 1]) + 1235 MUL64(ch[dep].weighting[1], master[smp ]) + 1236 MUL64(ch[dep].weighting[2], master[smp + 1]); 1237 1238 bd->raw_samples[smp] += y >> 7; 1239 } 1240 } 1241 1242 dep++; 1243 } 1244 1245 return 0; 1246} 1247 1248 1249/** Reads the frame data. 1250 */ 1251static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame) 1252{ 1253 ALSSpecificConfig *sconf = &ctx->sconf; 1254 AVCodecContext *avctx = ctx->avctx; 1255 GetBitContext *gb = &ctx->gb; 1256 unsigned int div_blocks[32]; ///< block sizes. 1257 unsigned int c; 1258 unsigned int js_blocks[2]; 1259 1260 uint32_t bs_info = 0; 1261 1262 // skip the size of the ra unit if present in the frame 1263 if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame) 1264 skip_bits_long(gb, 32); 1265 1266 if (sconf->mc_coding && sconf->joint_stereo) { 1267 ctx->js_switch = get_bits1(gb); 1268 align_get_bits(gb); 1269 } 1270 1271 if (!sconf->mc_coding || ctx->js_switch) { 1272 int independent_bs = !sconf->joint_stereo; 1273 1274 for (c = 0; c < avctx->channels; c++) { 1275 js_blocks[0] = 0; 1276 js_blocks[1] = 0; 1277 1278 get_block_sizes(ctx, div_blocks, &bs_info); 1279 1280 // if joint_stereo and block_switching is set, independent decoding 1281 // is signaled via the first bit of bs_info 1282 if (sconf->joint_stereo && sconf->block_switching) 1283 if (bs_info >> 31) 1284 independent_bs = 2; 1285 1286 // if this is the last channel, it has to be decoded independently 1287 if (c == avctx->channels - 1) 1288 independent_bs = 1; 1289 1290 if (independent_bs) { 1291 if (decode_blocks_ind(ctx, ra_frame, c, div_blocks, js_blocks)) 1292 return -1; 1293 1294 independent_bs--; 1295 } else { 1296 if (decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks)) 1297 return -1; 1298 1299 c++; 1300 } 1301 1302 // store carryover raw samples 1303 memmove(ctx->raw_samples[c] - sconf->max_order, 1304 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1305 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1306 } 1307 } else { // multi-channel coding 1308 ALSBlockData bd; 1309 int b; 1310 int *reverted_channels = ctx->reverted_channels; 1311 unsigned int offset = 0; 1312 1313 for (c = 0; c < avctx->channels; c++) 1314 if (ctx->chan_data[c] < ctx->chan_data_buffer) { 1315 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data!\n"); 1316 return -1; 1317 } 1318 1319 memset(&bd, 0, sizeof(ALSBlockData)); 1320 memset(reverted_channels, 0, sizeof(*reverted_channels) * avctx->channels); 1321 1322 bd.ra_block = ra_frame; 1323 bd.prev_raw_samples = ctx->prev_raw_samples; 1324 1325 get_block_sizes(ctx, div_blocks, &bs_info); 1326 1327 for (b = 0; b < ctx->num_blocks; b++) { 1328 bd.shift_lsbs = 0; 1329 bd.block_length = div_blocks[b]; 1330 1331 for (c = 0; c < avctx->channels; c++) { 1332 bd.use_ltp = ctx->use_ltp + c; 1333 bd.ltp_lag = ctx->ltp_lag + c; 1334 bd.ltp_gain = ctx->ltp_gain[c]; 1335 bd.lpc_cof = ctx->lpc_cof[c]; 1336 bd.quant_cof = ctx->quant_cof[c]; 1337 bd.raw_samples = ctx->raw_samples[c] + offset; 1338 bd.raw_other = NULL; 1339 1340 read_block(ctx, &bd); 1341 if (read_channel_data(ctx, ctx->chan_data[c], c)) 1342 return -1; 1343 } 1344 1345 for (c = 0; c < avctx->channels; c++) 1346 if (revert_channel_correlation(ctx, &bd, ctx->chan_data, 1347 reverted_channels, offset, c)) 1348 return -1; 1349 1350 for (c = 0; c < avctx->channels; c++) { 1351 bd.use_ltp = ctx->use_ltp + c; 1352 bd.ltp_lag = ctx->ltp_lag + c; 1353 bd.ltp_gain = ctx->ltp_gain[c]; 1354 bd.lpc_cof = ctx->lpc_cof[c]; 1355 bd.quant_cof = ctx->quant_cof[c]; 1356 bd.raw_samples = ctx->raw_samples[c] + offset; 1357 decode_block(ctx, &bd); 1358 } 1359 1360 memset(reverted_channels, 0, avctx->channels * sizeof(*reverted_channels)); 1361 offset += div_blocks[b]; 1362 bd.ra_block = 0; 1363 } 1364 1365 // store carryover raw samples 1366 for (c = 0; c < avctx->channels; c++) 1367 memmove(ctx->raw_samples[c] - sconf->max_order, 1368 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length, 1369 sizeof(*ctx->raw_samples[c]) * sconf->max_order); 1370 } 1371 1372 // TODO: read_diff_float_data 1373 1374 return 0; 1375} 1376 1377 1378/** Decodes an ALS frame. 1379 */ 1380static int decode_frame(AVCodecContext *avctx, 1381 void *data, int *data_size, 1382 AVPacket *avpkt) 1383{ 1384 ALSDecContext *ctx = avctx->priv_data; 1385 ALSSpecificConfig *sconf = &ctx->sconf; 1386 const uint8_t *buffer = avpkt->data; 1387 int buffer_size = avpkt->size; 1388 int invalid_frame, size; 1389 unsigned int c, sample, ra_frame, bytes_read, shift; 1390 1391 init_get_bits(&ctx->gb, buffer, buffer_size * 8); 1392 1393 // In the case that the distance between random access frames is set to zero 1394 // (sconf->ra_distance == 0) no frame is treated as a random access frame. 1395 // For the first frame, if prediction is used, all samples used from the 1396 // previous frame are assumed to be zero. 1397 ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance); 1398 1399 // the last frame to decode might have a different length 1400 if (sconf->samples != 0xFFFFFFFF) 1401 ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length, 1402 sconf->frame_length); 1403 else 1404 ctx->cur_frame_length = sconf->frame_length; 1405 1406 // decode the frame data 1407 if ((invalid_frame = read_frame_data(ctx, ra_frame) < 0)) 1408 av_log(ctx->avctx, AV_LOG_WARNING, 1409 "Reading frame data failed. Skipping RA unit.\n"); 1410 1411 ctx->frame_id++; 1412 1413 // check for size of decoded data 1414 size = ctx->cur_frame_length * avctx->channels * 1415 (av_get_bits_per_sample_format(avctx->sample_fmt) >> 3); 1416 1417 if (size > *data_size) { 1418 av_log(avctx, AV_LOG_ERROR, "Decoded data exceeds buffer size.\n"); 1419 return -1; 1420 } 1421 1422 *data_size = size; 1423 1424 // transform decoded frame into output format 1425 #define INTERLEAVE_OUTPUT(bps) \ 1426 { \ 1427 int##bps##_t *dest = (int##bps##_t*) data; \ 1428 shift = bps - ctx->avctx->bits_per_raw_sample; \ 1429 for (sample = 0; sample < ctx->cur_frame_length; sample++) \ 1430 for (c = 0; c < avctx->channels; c++) \ 1431 *dest++ = ctx->raw_samples[c][sample] << shift; \ 1432 } 1433 1434 if (ctx->avctx->bits_per_raw_sample <= 16) { 1435 INTERLEAVE_OUTPUT(16) 1436 } else { 1437 INTERLEAVE_OUTPUT(32) 1438 } 1439 1440 bytes_read = invalid_frame ? buffer_size : 1441 (get_bits_count(&ctx->gb) + 7) >> 3; 1442 1443 return bytes_read; 1444} 1445 1446 1447/** Uninitializes the ALS decoder. 1448 */ 1449static av_cold int decode_end(AVCodecContext *avctx) 1450{ 1451 ALSDecContext *ctx = avctx->priv_data; 1452 1453 av_freep(&ctx->sconf.chan_pos); 1454 1455 ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status); 1456 1457 av_freep(&ctx->use_ltp); 1458 av_freep(&ctx->ltp_lag); 1459 av_freep(&ctx->ltp_gain); 1460 av_freep(&ctx->ltp_gain_buffer); 1461 av_freep(&ctx->quant_cof); 1462 av_freep(&ctx->lpc_cof); 1463 av_freep(&ctx->quant_cof_buffer); 1464 av_freep(&ctx->lpc_cof_buffer); 1465 av_freep(&ctx->lpc_cof_reversed_buffer); 1466 av_freep(&ctx->prev_raw_samples); 1467 av_freep(&ctx->raw_samples); 1468 av_freep(&ctx->raw_buffer); 1469 av_freep(&ctx->chan_data); 1470 av_freep(&ctx->chan_data_buffer); 1471 av_freep(&ctx->reverted_channels); 1472 1473 return 0; 1474} 1475 1476 1477/** Initializes the ALS decoder. 1478 */ 1479static av_cold int decode_init(AVCodecContext *avctx) 1480{ 1481 unsigned int c; 1482 unsigned int channel_size; 1483 int num_buffers; 1484 ALSDecContext *ctx = avctx->priv_data; 1485 ALSSpecificConfig *sconf = &ctx->sconf; 1486 ctx->avctx = avctx; 1487 1488 if (!avctx->extradata) { 1489 av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n"); 1490 return -1; 1491 } 1492 1493 if (read_specific_config(ctx)) { 1494 av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n"); 1495 decode_end(avctx); 1496 return -1; 1497 } 1498 1499 if (check_specific_config(ctx)) { 1500 decode_end(avctx); 1501 return -1; 1502 } 1503 1504 if (sconf->bgmc) 1505 ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status); 1506 1507 if (sconf->floating) { 1508 avctx->sample_fmt = SAMPLE_FMT_FLT; 1509 avctx->bits_per_raw_sample = 32; 1510 } else { 1511 avctx->sample_fmt = sconf->resolution > 1 1512 ? SAMPLE_FMT_S32 : SAMPLE_FMT_S16; 1513 avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8; 1514 } 1515 1516 // set maximum Rice parameter for progressive decoding based on resolution 1517 // This is not specified in 14496-3 but actually done by the reference 1518 // codec RM22 revision 2. 1519 ctx->s_max = sconf->resolution > 1 ? 31 : 15; 1520 1521 // set lag value for long-term prediction 1522 ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) + 1523 (avctx->sample_rate >= 192000); 1524 1525 // allocate quantized parcor coefficient buffer 1526 num_buffers = sconf->mc_coding ? avctx->channels : 1; 1527 1528 ctx->quant_cof = av_malloc(sizeof(*ctx->quant_cof) * num_buffers); 1529 ctx->lpc_cof = av_malloc(sizeof(*ctx->lpc_cof) * num_buffers); 1530 ctx->quant_cof_buffer = av_malloc(sizeof(*ctx->quant_cof_buffer) * 1531 num_buffers * sconf->max_order); 1532 ctx->lpc_cof_buffer = av_malloc(sizeof(*ctx->lpc_cof_buffer) * 1533 num_buffers * sconf->max_order); 1534 ctx->lpc_cof_reversed_buffer = av_malloc(sizeof(*ctx->lpc_cof_buffer) * 1535 sconf->max_order); 1536 1537 if (!ctx->quant_cof || !ctx->lpc_cof || 1538 !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer || 1539 !ctx->lpc_cof_reversed_buffer) { 1540 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 1541 return AVERROR(ENOMEM); 1542 } 1543 1544 // assign quantized parcor coefficient buffers 1545 for (c = 0; c < num_buffers; c++) { 1546 ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order; 1547 ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order; 1548 } 1549 1550 // allocate and assign lag and gain data buffer for ltp mode 1551 ctx->use_ltp = av_mallocz(sizeof(*ctx->use_ltp) * num_buffers); 1552 ctx->ltp_lag = av_malloc (sizeof(*ctx->ltp_lag) * num_buffers); 1553 ctx->ltp_gain = av_malloc (sizeof(*ctx->ltp_gain) * num_buffers); 1554 ctx->ltp_gain_buffer = av_malloc (sizeof(*ctx->ltp_gain_buffer) * 1555 num_buffers * 5); 1556 1557 if (!ctx->use_ltp || !ctx->ltp_lag || 1558 !ctx->ltp_gain || !ctx->ltp_gain_buffer) { 1559 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 1560 decode_end(avctx); 1561 return AVERROR(ENOMEM); 1562 } 1563 1564 for (c = 0; c < num_buffers; c++) 1565 ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5; 1566 1567 // allocate and assign channel data buffer for mcc mode 1568 if (sconf->mc_coding) { 1569 ctx->chan_data_buffer = av_malloc(sizeof(*ctx->chan_data_buffer) * 1570 num_buffers * num_buffers); 1571 ctx->chan_data = av_malloc(sizeof(*ctx->chan_data) * 1572 num_buffers); 1573 ctx->reverted_channels = av_malloc(sizeof(*ctx->reverted_channels) * 1574 num_buffers); 1575 1576 if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) { 1577 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 1578 decode_end(avctx); 1579 return AVERROR(ENOMEM); 1580 } 1581 1582 for (c = 0; c < num_buffers; c++) 1583 ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers; 1584 } else { 1585 ctx->chan_data = NULL; 1586 ctx->chan_data_buffer = NULL; 1587 ctx->reverted_channels = NULL; 1588 } 1589 1590 avctx->frame_size = sconf->frame_length; 1591 channel_size = sconf->frame_length + sconf->max_order; 1592 1593 ctx->prev_raw_samples = av_malloc (sizeof(*ctx->prev_raw_samples) * sconf->max_order); 1594 ctx->raw_buffer = av_mallocz(sizeof(*ctx-> raw_buffer) * avctx->channels * channel_size); 1595 ctx->raw_samples = av_malloc (sizeof(*ctx-> raw_samples) * avctx->channels); 1596 1597 // allocate previous raw sample buffer 1598 if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) { 1599 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n"); 1600 decode_end(avctx); 1601 return AVERROR(ENOMEM); 1602 } 1603 1604 // assign raw samples buffers 1605 ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order; 1606 for (c = 1; c < avctx->channels; c++) 1607 ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size; 1608 1609 return 0; 1610} 1611 1612 1613/** Flushes (resets) the frame ID after seeking. 1614 */ 1615static av_cold void flush(AVCodecContext *avctx) 1616{ 1617 ALSDecContext *ctx = avctx->priv_data; 1618 1619 ctx->frame_id = 0; 1620} 1621 1622 1623AVCodec als_decoder = { 1624 "als", 1625 AVMEDIA_TYPE_AUDIO, 1626 CODEC_ID_MP4ALS, 1627 sizeof(ALSDecContext), 1628 decode_init, 1629 NULL, 1630 decode_end, 1631 decode_frame, 1632 .flush = flush, 1633 .capabilities = CODEC_CAP_SUBFRAMES, 1634 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"), 1635}; 1636 1637