| 1 | /* |
| 2 | * ATRAC3 compatible decoder |
| 3 | * Copyright (c) 2006-2008 Maxim Poliakovski |
| 4 | * Copyright (c) 2006-2008 Benjamin Larsson |
| 5 | * |
| 6 | * This file is part of FFmpeg. |
| 7 | * |
| 8 | * FFmpeg is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; either |
| 11 | * version 2.1 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * FFmpeg is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with FFmpeg; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | /** |
| 24 | * @file |
| 25 | * ATRAC3 compatible decoder. |
| 26 | * This decoder handles Sony's ATRAC3 data. |
| 27 | * |
| 28 | * Container formats used to store ATRAC3 data: |
| 29 | * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3). |
| 30 | * |
| 31 | * To use this decoder, a calling application must supply the extradata |
| 32 | * bytes provided in the containers above. |
| 33 | */ |
| 34 | |
| 35 | #include <math.h> |
| 36 | #include <stddef.h> |
| 37 | #include <stdio.h> |
| 38 | |
| 39 | #include "libavutil/attributes.h" |
| 40 | #include "libavutil/float_dsp.h" |
| 41 | #include "libavutil/libm.h" |
| 42 | #include "avcodec.h" |
| 43 | #include "bytestream.h" |
| 44 | #include "fft.h" |
| 45 | #include "fmtconvert.h" |
| 46 | #include "get_bits.h" |
| 47 | #include "internal.h" |
| 48 | |
| 49 | #include "atrac.h" |
| 50 | #include "atrac3data.h" |
| 51 | |
| 52 | #define JOINT_STEREO 0x12 |
| 53 | #define STEREO 0x2 |
| 54 | |
| 55 | #define SAMPLES_PER_FRAME 1024 |
| 56 | #define MDCT_SIZE 512 |
| 57 | |
| 58 | typedef struct GainBlock { |
| 59 | AtracGainInfo g_block[4]; |
| 60 | } GainBlock; |
| 61 | |
| 62 | typedef struct TonalComponent { |
| 63 | int pos; |
| 64 | int num_coefs; |
| 65 | float coef[8]; |
| 66 | } TonalComponent; |
| 67 | |
| 68 | typedef struct ChannelUnit { |
| 69 | int bands_coded; |
| 70 | int num_components; |
| 71 | float prev_frame[SAMPLES_PER_FRAME]; |
| 72 | int gc_blk_switch; |
| 73 | TonalComponent components[64]; |
| 74 | GainBlock gain_block[2]; |
| 75 | |
| 76 | DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME]; |
| 77 | DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME]; |
| 78 | |
| 79 | float delay_buf1[46]; ///<qmf delay buffers |
| 80 | float delay_buf2[46]; |
| 81 | float delay_buf3[46]; |
| 82 | } ChannelUnit; |
| 83 | |
| 84 | typedef struct ATRAC3Context { |
| 85 | GetBitContext gb; |
| 86 | //@{ |
| 87 | /** stream data */ |
| 88 | int coding_mode; |
| 89 | |
| 90 | ChannelUnit *units; |
| 91 | //@} |
| 92 | //@{ |
| 93 | /** joint-stereo related variables */ |
| 94 | int matrix_coeff_index_prev[4]; |
| 95 | int matrix_coeff_index_now[4]; |
| 96 | int matrix_coeff_index_next[4]; |
| 97 | int weighting_delay[6]; |
| 98 | //@} |
| 99 | //@{ |
| 100 | /** data buffers */ |
| 101 | uint8_t *decoded_bytes_buffer; |
| 102 | float temp_buf[1070]; |
| 103 | //@} |
| 104 | //@{ |
| 105 | /** extradata */ |
| 106 | int scrambled_stream; |
| 107 | //@} |
| 108 | |
| 109 | AtracGCContext gainc_ctx; |
| 110 | FFTContext mdct_ctx; |
| 111 | FmtConvertContext fmt_conv; |
| 112 | AVFloatDSPContext fdsp; |
| 113 | } ATRAC3Context; |
| 114 | |
| 115 | static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE]; |
| 116 | static VLC_TYPE atrac3_vlc_table[4096][2]; |
| 117 | static VLC spectral_coeff_tab[7]; |
| 118 | |
| 119 | /** |
| 120 | * Regular 512 points IMDCT without overlapping, with the exception of the |
| 121 | * swapping of odd bands caused by the reverse spectra of the QMF. |
| 122 | * |
| 123 | * @param odd_band 1 if the band is an odd band |
| 124 | */ |
| 125 | static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band) |
| 126 | { |
| 127 | int i; |
| 128 | |
| 129 | if (odd_band) { |
| 130 | /** |
| 131 | * Reverse the odd bands before IMDCT, this is an effect of the QMF |
| 132 | * transform or it gives better compression to do it this way. |
| 133 | * FIXME: It should be possible to handle this in imdct_calc |
| 134 | * for that to happen a modification of the prerotation step of |
| 135 | * all SIMD code and C code is needed. |
| 136 | * Or fix the functions before so they generate a pre reversed spectrum. |
| 137 | */ |
| 138 | for (i = 0; i < 128; i++) |
| 139 | FFSWAP(float, input[i], input[255 - i]); |
| 140 | } |
| 141 | |
| 142 | q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input); |
| 143 | |
| 144 | /* Perform windowing on the output. */ |
| 145 | q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE); |
| 146 | } |
| 147 | |
| 148 | /* |
| 149 | * indata descrambling, only used for data coming from the rm container |
| 150 | */ |
| 151 | static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes) |
| 152 | { |
| 153 | int i, off; |
| 154 | uint32_t c; |
| 155 | const uint32_t *buf; |
| 156 | uint32_t *output = (uint32_t *)out; |
| 157 | |
| 158 | off = (intptr_t)input & 3; |
| 159 | buf = (const uint32_t *)(input - off); |
| 160 | if (off) |
| 161 | c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8)))); |
| 162 | else |
| 163 | c = av_be2ne32(0x537F6103U); |
| 164 | bytes += 3 + off; |
| 165 | for (i = 0; i < bytes / 4; i++) |
| 166 | output[i] = c ^ buf[i]; |
| 167 | |
| 168 | if (off) |
| 169 | avpriv_request_sample(NULL, "Offset of %d", off); |
| 170 | |
| 171 | return off; |
| 172 | } |
| 173 | |
| 174 | static av_cold void init_imdct_window(void) |
| 175 | { |
| 176 | int i, j; |
| 177 | |
| 178 | /* generate the mdct window, for details see |
| 179 | * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */ |
| 180 | for (i = 0, j = 255; i < 128; i++, j--) { |
| 181 | float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0; |
| 182 | float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0; |
| 183 | float w = 0.5 * (wi * wi + wj * wj); |
| 184 | mdct_window[i] = mdct_window[511 - i] = wi / w; |
| 185 | mdct_window[j] = mdct_window[511 - j] = wj / w; |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | static av_cold int atrac3_decode_close(AVCodecContext *avctx) |
| 190 | { |
| 191 | ATRAC3Context *q = avctx->priv_data; |
| 192 | |
| 193 | av_free(q->units); |
| 194 | av_free(q->decoded_bytes_buffer); |
| 195 | |
| 196 | ff_mdct_end(&q->mdct_ctx); |
| 197 | |
| 198 | return 0; |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * Mantissa decoding |
| 203 | * |
| 204 | * @param selector which table the output values are coded with |
| 205 | * @param coding_flag constant length coding or variable length coding |
| 206 | * @param mantissas mantissa output table |
| 207 | * @param num_codes number of values to get |
| 208 | */ |
| 209 | static void read_quant_spectral_coeffs(GetBitContext *gb, int selector, |
| 210 | int coding_flag, int *mantissas, |
| 211 | int num_codes) |
| 212 | { |
| 213 | int i, code, huff_symb; |
| 214 | |
| 215 | if (selector == 1) |
| 216 | num_codes /= 2; |
| 217 | |
| 218 | if (coding_flag != 0) { |
| 219 | /* constant length coding (CLC) */ |
| 220 | int num_bits = clc_length_tab[selector]; |
| 221 | |
| 222 | if (selector > 1) { |
| 223 | for (i = 0; i < num_codes; i++) { |
| 224 | if (num_bits) |
| 225 | code = get_sbits(gb, num_bits); |
| 226 | else |
| 227 | code = 0; |
| 228 | mantissas[i] = code; |
| 229 | } |
| 230 | } else { |
| 231 | for (i = 0; i < num_codes; i++) { |
| 232 | if (num_bits) |
| 233 | code = get_bits(gb, num_bits); // num_bits is always 4 in this case |
| 234 | else |
| 235 | code = 0; |
| 236 | mantissas[i * 2 ] = mantissa_clc_tab[code >> 2]; |
| 237 | mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3]; |
| 238 | } |
| 239 | } |
| 240 | } else { |
| 241 | /* variable length coding (VLC) */ |
| 242 | if (selector != 1) { |
| 243 | for (i = 0; i < num_codes; i++) { |
| 244 | huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, |
| 245 | spectral_coeff_tab[selector-1].bits, 3); |
| 246 | huff_symb += 1; |
| 247 | code = huff_symb >> 1; |
| 248 | if (huff_symb & 1) |
| 249 | code = -code; |
| 250 | mantissas[i] = code; |
| 251 | } |
| 252 | } else { |
| 253 | for (i = 0; i < num_codes; i++) { |
| 254 | huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table, |
| 255 | spectral_coeff_tab[selector - 1].bits, 3); |
| 256 | mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ]; |
| 257 | mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1]; |
| 258 | } |
| 259 | } |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | /** |
| 264 | * Restore the quantized band spectrum coefficients |
| 265 | * |
| 266 | * @return subband count, fix for broken specification/files |
| 267 | */ |
| 268 | static int decode_spectrum(GetBitContext *gb, float *output) |
| 269 | { |
| 270 | int num_subbands, coding_mode, i, j, first, last, subband_size; |
| 271 | int subband_vlc_index[32], sf_index[32]; |
| 272 | int mantissas[128]; |
| 273 | float scale_factor; |
| 274 | |
| 275 | num_subbands = get_bits(gb, 5); // number of coded subbands |
| 276 | coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC |
| 277 | |
| 278 | /* get the VLC selector table for the subbands, 0 means not coded */ |
| 279 | for (i = 0; i <= num_subbands; i++) |
| 280 | subband_vlc_index[i] = get_bits(gb, 3); |
| 281 | |
| 282 | /* read the scale factor indexes from the stream */ |
| 283 | for (i = 0; i <= num_subbands; i++) { |
| 284 | if (subband_vlc_index[i] != 0) |
| 285 | sf_index[i] = get_bits(gb, 6); |
| 286 | } |
| 287 | |
| 288 | for (i = 0; i <= num_subbands; i++) { |
| 289 | first = subband_tab[i ]; |
| 290 | last = subband_tab[i + 1]; |
| 291 | |
| 292 | subband_size = last - first; |
| 293 | |
| 294 | if (subband_vlc_index[i] != 0) { |
| 295 | /* decode spectral coefficients for this subband */ |
| 296 | /* TODO: This can be done faster is several blocks share the |
| 297 | * same VLC selector (subband_vlc_index) */ |
| 298 | read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode, |
| 299 | mantissas, subband_size); |
| 300 | |
| 301 | /* decode the scale factor for this subband */ |
| 302 | scale_factor = ff_atrac_sf_table[sf_index[i]] * |
| 303 | inv_max_quant[subband_vlc_index[i]]; |
| 304 | |
| 305 | /* inverse quantize the coefficients */ |
| 306 | for (j = 0; first < last; first++, j++) |
| 307 | output[first] = mantissas[j] * scale_factor; |
| 308 | } else { |
| 309 | /* this subband was not coded, so zero the entire subband */ |
| 310 | memset(output + first, 0, subband_size * sizeof(*output)); |
| 311 | } |
| 312 | } |
| 313 | |
| 314 | /* clear the subbands that were not coded */ |
| 315 | first = subband_tab[i]; |
| 316 | memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output)); |
| 317 | return num_subbands; |
| 318 | } |
| 319 | |
| 320 | /** |
| 321 | * Restore the quantized tonal components |
| 322 | * |
| 323 | * @param components tonal components |
| 324 | * @param num_bands number of coded bands |
| 325 | */ |
| 326 | static int decode_tonal_components(GetBitContext *gb, |
| 327 | TonalComponent *components, int num_bands) |
| 328 | { |
| 329 | int i, b, c, m; |
| 330 | int nb_components, coding_mode_selector, coding_mode; |
| 331 | int band_flags[4], mantissa[8]; |
| 332 | int component_count = 0; |
| 333 | |
| 334 | nb_components = get_bits(gb, 5); |
| 335 | |
| 336 | /* no tonal components */ |
| 337 | if (nb_components == 0) |
| 338 | return 0; |
| 339 | |
| 340 | coding_mode_selector = get_bits(gb, 2); |
| 341 | if (coding_mode_selector == 2) |
| 342 | return AVERROR_INVALIDDATA; |
| 343 | |
| 344 | coding_mode = coding_mode_selector & 1; |
| 345 | |
| 346 | for (i = 0; i < nb_components; i++) { |
| 347 | int coded_values_per_component, quant_step_index; |
| 348 | |
| 349 | for (b = 0; b <= num_bands; b++) |
| 350 | band_flags[b] = get_bits1(gb); |
| 351 | |
| 352 | coded_values_per_component = get_bits(gb, 3); |
| 353 | |
| 354 | quant_step_index = get_bits(gb, 3); |
| 355 | if (quant_step_index <= 1) |
| 356 | return AVERROR_INVALIDDATA; |
| 357 | |
| 358 | if (coding_mode_selector == 3) |
| 359 | coding_mode = get_bits1(gb); |
| 360 | |
| 361 | for (b = 0; b < (num_bands + 1) * 4; b++) { |
| 362 | int coded_components; |
| 363 | |
| 364 | if (band_flags[b >> 2] == 0) |
| 365 | continue; |
| 366 | |
| 367 | coded_components = get_bits(gb, 3); |
| 368 | |
| 369 | for (c = 0; c < coded_components; c++) { |
| 370 | TonalComponent *cmp = &components[component_count]; |
| 371 | int sf_index, coded_values, max_coded_values; |
| 372 | float scale_factor; |
| 373 | |
| 374 | sf_index = get_bits(gb, 6); |
| 375 | if (component_count >= 64) |
| 376 | return AVERROR_INVALIDDATA; |
| 377 | |
| 378 | cmp->pos = b * 64 + get_bits(gb, 6); |
| 379 | |
| 380 | max_coded_values = SAMPLES_PER_FRAME - cmp->pos; |
| 381 | coded_values = coded_values_per_component + 1; |
| 382 | coded_values = FFMIN(max_coded_values, coded_values); |
| 383 | |
| 384 | scale_factor = ff_atrac_sf_table[sf_index] * |
| 385 | inv_max_quant[quant_step_index]; |
| 386 | |
| 387 | read_quant_spectral_coeffs(gb, quant_step_index, coding_mode, |
| 388 | mantissa, coded_values); |
| 389 | |
| 390 | cmp->num_coefs = coded_values; |
| 391 | |
| 392 | /* inverse quant */ |
| 393 | for (m = 0; m < coded_values; m++) |
| 394 | cmp->coef[m] = mantissa[m] * scale_factor; |
| 395 | |
| 396 | component_count++; |
| 397 | } |
| 398 | } |
| 399 | } |
| 400 | |
| 401 | return component_count; |
| 402 | } |
| 403 | |
| 404 | /** |
| 405 | * Decode gain parameters for the coded bands |
| 406 | * |
| 407 | * @param block the gainblock for the current band |
| 408 | * @param num_bands amount of coded bands |
| 409 | */ |
| 410 | static int decode_gain_control(GetBitContext *gb, GainBlock *block, |
| 411 | int num_bands) |
| 412 | { |
| 413 | int b, j; |
| 414 | int *level, *loc; |
| 415 | |
| 416 | AtracGainInfo *gain = block->g_block; |
| 417 | |
| 418 | for (b = 0; b <= num_bands; b++) { |
| 419 | gain[b].num_points = get_bits(gb, 3); |
| 420 | level = gain[b].lev_code; |
| 421 | loc = gain[b].loc_code; |
| 422 | |
| 423 | for (j = 0; j < gain[b].num_points; j++) { |
| 424 | level[j] = get_bits(gb, 4); |
| 425 | loc[j] = get_bits(gb, 5); |
| 426 | if (j && loc[j] <= loc[j - 1]) |
| 427 | return AVERROR_INVALIDDATA; |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | /* Clear the unused blocks. */ |
| 432 | for (; b < 4 ; b++) |
| 433 | gain[b].num_points = 0; |
| 434 | |
| 435 | return 0; |
| 436 | } |
| 437 | |
| 438 | /** |
| 439 | * Combine the tonal band spectrum and regular band spectrum |
| 440 | * |
| 441 | * @param spectrum output spectrum buffer |
| 442 | * @param num_components number of tonal components |
| 443 | * @param components tonal components for this band |
| 444 | * @return position of the last tonal coefficient |
| 445 | */ |
| 446 | static int add_tonal_components(float *spectrum, int num_components, |
| 447 | TonalComponent *components) |
| 448 | { |
| 449 | int i, j, last_pos = -1; |
| 450 | float *input, *output; |
| 451 | |
| 452 | for (i = 0; i < num_components; i++) { |
| 453 | last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos); |
| 454 | input = components[i].coef; |
| 455 | output = &spectrum[components[i].pos]; |
| 456 | |
| 457 | for (j = 0; j < components[i].num_coefs; j++) |
| 458 | output[j] += input[j]; |
| 459 | } |
| 460 | |
| 461 | return last_pos; |
| 462 | } |
| 463 | |
| 464 | #define INTERPOLATE(old, new, nsample) \ |
| 465 | ((old) + (nsample) * 0.125 * ((new) - (old))) |
| 466 | |
| 467 | static void reverse_matrixing(float *su1, float *su2, int *prev_code, |
| 468 | int *curr_code) |
| 469 | { |
| 470 | int i, nsample, band; |
| 471 | float mc1_l, mc1_r, mc2_l, mc2_r; |
| 472 | |
| 473 | for (i = 0, band = 0; band < 4 * 256; band += 256, i++) { |
| 474 | int s1 = prev_code[i]; |
| 475 | int s2 = curr_code[i]; |
| 476 | nsample = band; |
| 477 | |
| 478 | if (s1 != s2) { |
| 479 | /* Selector value changed, interpolation needed. */ |
| 480 | mc1_l = matrix_coeffs[s1 * 2 ]; |
| 481 | mc1_r = matrix_coeffs[s1 * 2 + 1]; |
| 482 | mc2_l = matrix_coeffs[s2 * 2 ]; |
| 483 | mc2_r = matrix_coeffs[s2 * 2 + 1]; |
| 484 | |
| 485 | /* Interpolation is done over the first eight samples. */ |
| 486 | for (; nsample < band + 8; nsample++) { |
| 487 | float c1 = su1[nsample]; |
| 488 | float c2 = su2[nsample]; |
| 489 | c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) + |
| 490 | c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band); |
| 491 | su1[nsample] = c2; |
| 492 | su2[nsample] = c1 * 2.0 - c2; |
| 493 | } |
| 494 | } |
| 495 | |
| 496 | /* Apply the matrix without interpolation. */ |
| 497 | switch (s2) { |
| 498 | case 0: /* M/S decoding */ |
| 499 | for (; nsample < band + 256; nsample++) { |
| 500 | float c1 = su1[nsample]; |
| 501 | float c2 = su2[nsample]; |
| 502 | su1[nsample] = c2 * 2.0; |
| 503 | su2[nsample] = (c1 - c2) * 2.0; |
| 504 | } |
| 505 | break; |
| 506 | case 1: |
| 507 | for (; nsample < band + 256; nsample++) { |
| 508 | float c1 = su1[nsample]; |
| 509 | float c2 = su2[nsample]; |
| 510 | su1[nsample] = (c1 + c2) * 2.0; |
| 511 | su2[nsample] = c2 * -2.0; |
| 512 | } |
| 513 | break; |
| 514 | case 2: |
| 515 | case 3: |
| 516 | for (; nsample < band + 256; nsample++) { |
| 517 | float c1 = su1[nsample]; |
| 518 | float c2 = su2[nsample]; |
| 519 | su1[nsample] = c1 + c2; |
| 520 | su2[nsample] = c1 - c2; |
| 521 | } |
| 522 | break; |
| 523 | default: |
| 524 | av_assert1(0); |
| 525 | } |
| 526 | } |
| 527 | } |
| 528 | |
| 529 | static void get_channel_weights(int index, int flag, float ch[2]) |
| 530 | { |
| 531 | if (index == 7) { |
| 532 | ch[0] = 1.0; |
| 533 | ch[1] = 1.0; |
| 534 | } else { |
| 535 | ch[0] = (index & 7) / 7.0; |
| 536 | ch[1] = sqrt(2 - ch[0] * ch[0]); |
| 537 | if (flag) |
| 538 | FFSWAP(float, ch[0], ch[1]); |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | static void channel_weighting(float *su1, float *su2, int *p3) |
| 543 | { |
| 544 | int band, nsample; |
| 545 | /* w[x][y] y=0 is left y=1 is right */ |
| 546 | float w[2][2]; |
| 547 | |
| 548 | if (p3[1] != 7 || p3[3] != 7) { |
| 549 | get_channel_weights(p3[1], p3[0], w[0]); |
| 550 | get_channel_weights(p3[3], p3[2], w[1]); |
| 551 | |
| 552 | for (band = 256; band < 4 * 256; band += 256) { |
| 553 | for (nsample = band; nsample < band + 8; nsample++) { |
| 554 | su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band); |
| 555 | su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band); |
| 556 | } |
| 557 | for(; nsample < band + 256; nsample++) { |
| 558 | su1[nsample] *= w[1][0]; |
| 559 | su2[nsample] *= w[1][1]; |
| 560 | } |
| 561 | } |
| 562 | } |
| 563 | } |
| 564 | |
| 565 | /** |
| 566 | * Decode a Sound Unit |
| 567 | * |
| 568 | * @param snd the channel unit to be used |
| 569 | * @param output the decoded samples before IQMF in float representation |
| 570 | * @param channel_num channel number |
| 571 | * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono) |
| 572 | */ |
| 573 | static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb, |
| 574 | ChannelUnit *snd, float *output, |
| 575 | int channel_num, int coding_mode) |
| 576 | { |
| 577 | int band, ret, num_subbands, last_tonal, num_bands; |
| 578 | GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch]; |
| 579 | GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch]; |
| 580 | |
| 581 | if (coding_mode == JOINT_STEREO && channel_num == 1) { |
| 582 | if (get_bits(gb, 2) != 3) { |
| 583 | av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n"); |
| 584 | return AVERROR_INVALIDDATA; |
| 585 | } |
| 586 | } else { |
| 587 | if (get_bits(gb, 6) != 0x28) { |
| 588 | av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n"); |
| 589 | return AVERROR_INVALIDDATA; |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | /* number of coded QMF bands */ |
| 594 | snd->bands_coded = get_bits(gb, 2); |
| 595 | |
| 596 | ret = decode_gain_control(gb, gain2, snd->bands_coded); |
| 597 | if (ret) |
| 598 | return ret; |
| 599 | |
| 600 | snd->num_components = decode_tonal_components(gb, snd->components, |
| 601 | snd->bands_coded); |
| 602 | if (snd->num_components < 0) |
| 603 | return snd->num_components; |
| 604 | |
| 605 | num_subbands = decode_spectrum(gb, snd->spectrum); |
| 606 | |
| 607 | /* Merge the decoded spectrum and tonal components. */ |
| 608 | last_tonal = add_tonal_components(snd->spectrum, snd->num_components, |
| 609 | snd->components); |
| 610 | |
| 611 | |
| 612 | /* calculate number of used MLT/QMF bands according to the amount of coded |
| 613 | spectral lines */ |
| 614 | num_bands = (subband_tab[num_subbands] - 1) >> 8; |
| 615 | if (last_tonal >= 0) |
| 616 | num_bands = FFMAX((last_tonal + 256) >> 8, num_bands); |
| 617 | |
| 618 | |
| 619 | /* Reconstruct time domain samples. */ |
| 620 | for (band = 0; band < 4; band++) { |
| 621 | /* Perform the IMDCT step without overlapping. */ |
| 622 | if (band <= num_bands) |
| 623 | imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1); |
| 624 | else |
| 625 | memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf)); |
| 626 | |
| 627 | /* gain compensation and overlapping */ |
| 628 | ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf, |
| 629 | &snd->prev_frame[band * 256], |
| 630 | &gain1->g_block[band], &gain2->g_block[band], |
| 631 | 256, &output[band * 256]); |
| 632 | } |
| 633 | |
| 634 | /* Swap the gain control buffers for the next frame. */ |
| 635 | snd->gc_blk_switch ^= 1; |
| 636 | |
| 637 | return 0; |
| 638 | } |
| 639 | |
| 640 | static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf, |
| 641 | float **out_samples) |
| 642 | { |
| 643 | ATRAC3Context *q = avctx->priv_data; |
| 644 | int ret, i; |
| 645 | uint8_t *ptr1; |
| 646 | |
| 647 | if (q->coding_mode == JOINT_STEREO) { |
| 648 | /* channel coupling mode */ |
| 649 | /* decode Sound Unit 1 */ |
| 650 | init_get_bits(&q->gb, databuf, avctx->block_align * 8); |
| 651 | |
| 652 | ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0, |
| 653 | JOINT_STEREO); |
| 654 | if (ret != 0) |
| 655 | return ret; |
| 656 | |
| 657 | /* Framedata of the su2 in the joint-stereo mode is encoded in |
| 658 | * reverse byte order so we need to swap it first. */ |
| 659 | if (databuf == q->decoded_bytes_buffer) { |
| 660 | uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1; |
| 661 | ptr1 = q->decoded_bytes_buffer; |
| 662 | for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--) |
| 663 | FFSWAP(uint8_t, *ptr1, *ptr2); |
| 664 | } else { |
| 665 | const uint8_t *ptr2 = databuf + avctx->block_align - 1; |
| 666 | for (i = 0; i < avctx->block_align; i++) |
| 667 | q->decoded_bytes_buffer[i] = *ptr2--; |
| 668 | } |
| 669 | |
| 670 | /* Skip the sync codes (0xF8). */ |
| 671 | ptr1 = q->decoded_bytes_buffer; |
| 672 | for (i = 4; *ptr1 == 0xF8; i++, ptr1++) { |
| 673 | if (i >= avctx->block_align) |
| 674 | return AVERROR_INVALIDDATA; |
| 675 | } |
| 676 | |
| 677 | |
| 678 | /* set the bitstream reader at the start of the second Sound Unit*/ |
| 679 | init_get_bits8(&q->gb, ptr1, q->decoded_bytes_buffer + avctx->block_align - ptr1); |
| 680 | |
| 681 | /* Fill the Weighting coeffs delay buffer */ |
| 682 | memmove(q->weighting_delay, &q->weighting_delay[2], |
| 683 | 4 * sizeof(*q->weighting_delay)); |
| 684 | q->weighting_delay[4] = get_bits1(&q->gb); |
| 685 | q->weighting_delay[5] = get_bits(&q->gb, 3); |
| 686 | |
| 687 | for (i = 0; i < 4; i++) { |
| 688 | q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i]; |
| 689 | q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i]; |
| 690 | q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2); |
| 691 | } |
| 692 | |
| 693 | /* Decode Sound Unit 2. */ |
| 694 | ret = decode_channel_sound_unit(q, &q->gb, &q->units[1], |
| 695 | out_samples[1], 1, JOINT_STEREO); |
| 696 | if (ret != 0) |
| 697 | return ret; |
| 698 | |
| 699 | /* Reconstruct the channel coefficients. */ |
| 700 | reverse_matrixing(out_samples[0], out_samples[1], |
| 701 | q->matrix_coeff_index_prev, |
| 702 | q->matrix_coeff_index_now); |
| 703 | |
| 704 | channel_weighting(out_samples[0], out_samples[1], q->weighting_delay); |
| 705 | } else { |
| 706 | /* normal stereo mode or mono */ |
| 707 | /* Decode the channel sound units. */ |
| 708 | for (i = 0; i < avctx->channels; i++) { |
| 709 | /* Set the bitstream reader at the start of a channel sound unit. */ |
| 710 | init_get_bits(&q->gb, |
| 711 | databuf + i * avctx->block_align / avctx->channels, |
| 712 | avctx->block_align * 8 / avctx->channels); |
| 713 | |
| 714 | ret = decode_channel_sound_unit(q, &q->gb, &q->units[i], |
| 715 | out_samples[i], i, q->coding_mode); |
| 716 | if (ret != 0) |
| 717 | return ret; |
| 718 | } |
| 719 | } |
| 720 | |
| 721 | /* Apply the iQMF synthesis filter. */ |
| 722 | for (i = 0; i < avctx->channels; i++) { |
| 723 | float *p1 = out_samples[i]; |
| 724 | float *p2 = p1 + 256; |
| 725 | float *p3 = p2 + 256; |
| 726 | float *p4 = p3 + 256; |
| 727 | ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf); |
| 728 | ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf); |
| 729 | ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf); |
| 730 | } |
| 731 | |
| 732 | return 0; |
| 733 | } |
| 734 | |
| 735 | static int atrac3_decode_frame(AVCodecContext *avctx, void *data, |
| 736 | int *got_frame_ptr, AVPacket *avpkt) |
| 737 | { |
| 738 | AVFrame *frame = data; |
| 739 | const uint8_t *buf = avpkt->data; |
| 740 | int buf_size = avpkt->size; |
| 741 | ATRAC3Context *q = avctx->priv_data; |
| 742 | int ret; |
| 743 | const uint8_t *databuf; |
| 744 | |
| 745 | if (buf_size < avctx->block_align) { |
| 746 | av_log(avctx, AV_LOG_ERROR, |
| 747 | "Frame too small (%d bytes). Truncated file?\n", buf_size); |
| 748 | return AVERROR_INVALIDDATA; |
| 749 | } |
| 750 | |
| 751 | /* get output buffer */ |
| 752 | frame->nb_samples = SAMPLES_PER_FRAME; |
| 753 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
| 754 | return ret; |
| 755 | |
| 756 | /* Check if we need to descramble and what buffer to pass on. */ |
| 757 | if (q->scrambled_stream) { |
| 758 | decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align); |
| 759 | databuf = q->decoded_bytes_buffer; |
| 760 | } else { |
| 761 | databuf = buf; |
| 762 | } |
| 763 | |
| 764 | ret = decode_frame(avctx, databuf, (float **)frame->extended_data); |
| 765 | if (ret) { |
| 766 | av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n"); |
| 767 | return ret; |
| 768 | } |
| 769 | |
| 770 | *got_frame_ptr = 1; |
| 771 | |
| 772 | return avctx->block_align; |
| 773 | } |
| 774 | |
| 775 | static av_cold void atrac3_init_static_data(void) |
| 776 | { |
| 777 | int i; |
| 778 | |
| 779 | init_imdct_window(); |
| 780 | ff_atrac_generate_tables(); |
| 781 | |
| 782 | /* Initialize the VLC tables. */ |
| 783 | for (i = 0; i < 7; i++) { |
| 784 | spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]]; |
| 785 | spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - |
| 786 | atrac3_vlc_offs[i ]; |
| 787 | init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i], |
| 788 | huff_bits[i], 1, 1, |
| 789 | huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC); |
| 790 | } |
| 791 | } |
| 792 | |
| 793 | static av_cold int atrac3_decode_init(AVCodecContext *avctx) |
| 794 | { |
| 795 | static int static_init_done; |
| 796 | int i, ret; |
| 797 | int version, delay, samples_per_frame, frame_factor; |
| 798 | const uint8_t *edata_ptr = avctx->extradata; |
| 799 | ATRAC3Context *q = avctx->priv_data; |
| 800 | |
| 801 | if (avctx->channels <= 0 || avctx->channels > 2) { |
| 802 | av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n"); |
| 803 | return AVERROR(EINVAL); |
| 804 | } |
| 805 | |
| 806 | if (!static_init_done) |
| 807 | atrac3_init_static_data(); |
| 808 | static_init_done = 1; |
| 809 | |
| 810 | /* Take care of the codec-specific extradata. */ |
| 811 | if (avctx->extradata_size == 14) { |
| 812 | /* Parse the extradata, WAV format */ |
| 813 | av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n", |
| 814 | bytestream_get_le16(&edata_ptr)); // Unknown value always 1 |
| 815 | edata_ptr += 4; // samples per channel |
| 816 | q->coding_mode = bytestream_get_le16(&edata_ptr); |
| 817 | av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n", |
| 818 | bytestream_get_le16(&edata_ptr)); //Dupe of coding mode |
| 819 | frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1 |
| 820 | av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n", |
| 821 | bytestream_get_le16(&edata_ptr)); // Unknown always 0 |
| 822 | |
| 823 | /* setup */ |
| 824 | samples_per_frame = SAMPLES_PER_FRAME * avctx->channels; |
| 825 | version = 4; |
| 826 | delay = 0x88E; |
| 827 | q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO; |
| 828 | q->scrambled_stream = 0; |
| 829 | |
| 830 | if (avctx->block_align != 96 * avctx->channels * frame_factor && |
| 831 | avctx->block_align != 152 * avctx->channels * frame_factor && |
| 832 | avctx->block_align != 192 * avctx->channels * frame_factor) { |
| 833 | av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor " |
| 834 | "configuration %d/%d/%d\n", avctx->block_align, |
| 835 | avctx->channels, frame_factor); |
| 836 | return AVERROR_INVALIDDATA; |
| 837 | } |
| 838 | } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) { |
| 839 | /* Parse the extradata, RM format. */ |
| 840 | version = bytestream_get_be32(&edata_ptr); |
| 841 | samples_per_frame = bytestream_get_be16(&edata_ptr); |
| 842 | delay = bytestream_get_be16(&edata_ptr); |
| 843 | q->coding_mode = bytestream_get_be16(&edata_ptr); |
| 844 | q->scrambled_stream = 1; |
| 845 | |
| 846 | } else { |
| 847 | av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n", |
| 848 | avctx->extradata_size); |
| 849 | return AVERROR(EINVAL); |
| 850 | } |
| 851 | |
| 852 | /* Check the extradata */ |
| 853 | |
| 854 | if (version != 4) { |
| 855 | av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version); |
| 856 | return AVERROR_INVALIDDATA; |
| 857 | } |
| 858 | |
| 859 | if (samples_per_frame != SAMPLES_PER_FRAME && |
| 860 | samples_per_frame != SAMPLES_PER_FRAME * 2) { |
| 861 | av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n", |
| 862 | samples_per_frame); |
| 863 | return AVERROR_INVALIDDATA; |
| 864 | } |
| 865 | |
| 866 | if (delay != 0x88E) { |
| 867 | av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n", |
| 868 | delay); |
| 869 | return AVERROR_INVALIDDATA; |
| 870 | } |
| 871 | |
| 872 | if (q->coding_mode == STEREO) |
| 873 | av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n"); |
| 874 | else if (q->coding_mode == JOINT_STEREO) { |
| 875 | if (avctx->channels != 2) { |
| 876 | av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n"); |
| 877 | return AVERROR_INVALIDDATA; |
| 878 | } |
| 879 | av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n"); |
| 880 | } else { |
| 881 | av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n", |
| 882 | q->coding_mode); |
| 883 | return AVERROR_INVALIDDATA; |
| 884 | } |
| 885 | |
| 886 | if (avctx->block_align >= UINT_MAX / 2) |
| 887 | return AVERROR(EINVAL); |
| 888 | |
| 889 | q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) + |
| 890 | FF_INPUT_BUFFER_PADDING_SIZE); |
| 891 | if (!q->decoded_bytes_buffer) |
| 892 | return AVERROR(ENOMEM); |
| 893 | |
| 894 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; |
| 895 | |
| 896 | /* initialize the MDCT transform */ |
| 897 | if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) { |
| 898 | av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n"); |
| 899 | av_freep(&q->decoded_bytes_buffer); |
| 900 | return ret; |
| 901 | } |
| 902 | |
| 903 | /* init the joint-stereo decoding data */ |
| 904 | q->weighting_delay[0] = 0; |
| 905 | q->weighting_delay[1] = 7; |
| 906 | q->weighting_delay[2] = 0; |
| 907 | q->weighting_delay[3] = 7; |
| 908 | q->weighting_delay[4] = 0; |
| 909 | q->weighting_delay[5] = 7; |
| 910 | |
| 911 | for (i = 0; i < 4; i++) { |
| 912 | q->matrix_coeff_index_prev[i] = 3; |
| 913 | q->matrix_coeff_index_now[i] = 3; |
| 914 | q->matrix_coeff_index_next[i] = 3; |
| 915 | } |
| 916 | |
| 917 | ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3); |
| 918 | avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); |
| 919 | ff_fmt_convert_init(&q->fmt_conv, avctx); |
| 920 | |
| 921 | q->units = av_mallocz_array(avctx->channels, sizeof(*q->units)); |
| 922 | if (!q->units) { |
| 923 | atrac3_decode_close(avctx); |
| 924 | return AVERROR(ENOMEM); |
| 925 | } |
| 926 | |
| 927 | return 0; |
| 928 | } |
| 929 | |
| 930 | AVCodec ff_atrac3_decoder = { |
| 931 | .name = "atrac3", |
| 932 | .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"), |
| 933 | .type = AVMEDIA_TYPE_AUDIO, |
| 934 | .id = AV_CODEC_ID_ATRAC3, |
| 935 | .priv_data_size = sizeof(ATRAC3Context), |
| 936 | .init = atrac3_decode_init, |
| 937 | .close = atrac3_decode_close, |
| 938 | .decode = atrac3_decode_frame, |
| 939 | .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1, |
| 940 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
| 941 | AV_SAMPLE_FMT_NONE }, |
| 942 | }; |