1/* 2 * Atrac 3 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 * Atrac 3 compatible decoder. 26 * This decoder handles Sony's ATRAC3 data. 27 * 28 * Container formats used to store atrac 3 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 "avcodec.h" 40#include "get_bits.h" 41#include "dsputil.h" 42#include "bytestream.h" 43#include "fft.h" 44 45#include "atrac.h" 46#include "atrac3data.h" 47 48#define JOINT_STEREO 0x12 49#define STEREO 0x2 50 51 52/* These structures are needed to store the parsed gain control data. */ 53typedef struct { 54 int num_gain_data; 55 int levcode[8]; 56 int loccode[8]; 57} gain_info; 58 59typedef struct { 60 gain_info gBlock[4]; 61} gain_block; 62 63typedef struct { 64 int pos; 65 int numCoefs; 66 float coef[8]; 67} tonal_component; 68 69typedef struct { 70 int bandsCoded; 71 int numComponents; 72 tonal_component components[64]; 73 float prevFrame[1024]; 74 int gcBlkSwitch; 75 gain_block gainBlock[2]; 76 77 DECLARE_ALIGNED(16, float, spectrum)[1024]; 78 DECLARE_ALIGNED(16, float, IMDCT_buf)[1024]; 79 80 float delayBuf1[46]; ///<qmf delay buffers 81 float delayBuf2[46]; 82 float delayBuf3[46]; 83} channel_unit; 84 85typedef struct { 86 GetBitContext gb; 87 //@{ 88 /** stream data */ 89 int channels; 90 int codingMode; 91 int bit_rate; 92 int sample_rate; 93 int samples_per_channel; 94 int samples_per_frame; 95 96 int bits_per_frame; 97 int bytes_per_frame; 98 int pBs; 99 channel_unit* pUnits; 100 //@} 101 //@{ 102 /** joint-stereo related variables */ 103 int matrix_coeff_index_prev[4]; 104 int matrix_coeff_index_now[4]; 105 int matrix_coeff_index_next[4]; 106 int weighting_delay[6]; 107 //@} 108 //@{ 109 /** data buffers */ 110 float outSamples[2048]; 111 uint8_t* decoded_bytes_buffer; 112 float tempBuf[1070]; 113 //@} 114 //@{ 115 /** extradata */ 116 int atrac3version; 117 int delay; 118 int scrambled_stream; 119 int frame_factor; 120 //@} 121} ATRAC3Context; 122 123static DECLARE_ALIGNED(16, float,mdct_window)[512]; 124static VLC spectral_coeff_tab[7]; 125static float gain_tab1[16]; 126static float gain_tab2[31]; 127static FFTContext mdct_ctx; 128static DSPContext dsp; 129 130 131/** 132 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands 133 * caused by the reverse spectra of the QMF. 134 * 135 * @param pInput float input 136 * @param pOutput float output 137 * @param odd_band 1 if the band is an odd band 138 */ 139 140static void IMLT(float *pInput, float *pOutput, int odd_band) 141{ 142 int i; 143 144 if (odd_band) { 145 /** 146 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform 147 * or it gives better compression to do it this way. 148 * FIXME: It should be possible to handle this in ff_imdct_calc 149 * for that to happen a modification of the prerotation step of 150 * all SIMD code and C code is needed. 151 * Or fix the functions before so they generate a pre reversed spectrum. 152 */ 153 154 for (i=0; i<128; i++) 155 FFSWAP(float, pInput[i], pInput[255-i]); 156 } 157 158 ff_imdct_calc(&mdct_ctx,pOutput,pInput); 159 160 /* Perform windowing on the output. */ 161 dsp.vector_fmul(pOutput,mdct_window,512); 162 163} 164 165 166/** 167 * Atrac 3 indata descrambling, only used for data coming from the rm container 168 * 169 * @param in pointer to 8 bit array of indata 170 * @param bits amount of bits 171 * @param out pointer to 8 bit array of outdata 172 */ 173 174static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){ 175 int i, off; 176 uint32_t c; 177 const uint32_t* buf; 178 uint32_t* obuf = (uint32_t*) out; 179 180 off = (intptr_t)inbuffer & 3; 181 buf = (const uint32_t*) (inbuffer - off); 182 c = be2me_32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8)))); 183 bytes += 3 + off; 184 for (i = 0; i < bytes/4; i++) 185 obuf[i] = c ^ buf[i]; 186 187 if (off) 188 av_log(NULL,AV_LOG_DEBUG,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off); 189 190 return off; 191} 192 193 194static av_cold void init_atrac3_transforms(ATRAC3Context *q) { 195 float enc_window[256]; 196 int i; 197 198 /* Generate the mdct window, for details see 199 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */ 200 for (i=0 ; i<256; i++) 201 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5; 202 203 if (!mdct_window[0]) 204 for (i=0 ; i<256; i++) { 205 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]); 206 mdct_window[511-i] = mdct_window[i]; 207 } 208 209 /* Initialize the MDCT transform. */ 210 ff_mdct_init(&mdct_ctx, 9, 1, 1.0); 211} 212 213/** 214 * Atrac3 uninit, free all allocated memory 215 */ 216 217static av_cold int atrac3_decode_close(AVCodecContext *avctx) 218{ 219 ATRAC3Context *q = avctx->priv_data; 220 221 av_free(q->pUnits); 222 av_free(q->decoded_bytes_buffer); 223 224 return 0; 225} 226 227/** 228/ * Mantissa decoding 229 * 230 * @param gb the GetBit context 231 * @param selector what table is the output values coded with 232 * @param codingFlag constant length coding or variable length coding 233 * @param mantissas mantissa output table 234 * @param numCodes amount of values to get 235 */ 236 237static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes) 238{ 239 int numBits, cnt, code, huffSymb; 240 241 if (selector == 1) 242 numCodes /= 2; 243 244 if (codingFlag != 0) { 245 /* constant length coding (CLC) */ 246 numBits = CLCLengthTab[selector]; 247 248 if (selector > 1) { 249 for (cnt = 0; cnt < numCodes; cnt++) { 250 if (numBits) 251 code = get_sbits(gb, numBits); 252 else 253 code = 0; 254 mantissas[cnt] = code; 255 } 256 } else { 257 for (cnt = 0; cnt < numCodes; cnt++) { 258 if (numBits) 259 code = get_bits(gb, numBits); //numBits is always 4 in this case 260 else 261 code = 0; 262 mantissas[cnt*2] = seTab_0[code >> 2]; 263 mantissas[cnt*2+1] = seTab_0[code & 3]; 264 } 265 } 266 } else { 267 /* variable length coding (VLC) */ 268 if (selector != 1) { 269 for (cnt = 0; cnt < numCodes; cnt++) { 270 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3); 271 huffSymb += 1; 272 code = huffSymb >> 1; 273 if (huffSymb & 1) 274 code = -code; 275 mantissas[cnt] = code; 276 } 277 } else { 278 for (cnt = 0; cnt < numCodes; cnt++) { 279 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3); 280 mantissas[cnt*2] = decTable1[huffSymb*2]; 281 mantissas[cnt*2+1] = decTable1[huffSymb*2+1]; 282 } 283 } 284 } 285} 286 287/** 288 * Restore the quantized band spectrum coefficients 289 * 290 * @param gb the GetBit context 291 * @param pOut decoded band spectrum 292 * @return outSubbands subband counter, fix for broken specification/files 293 */ 294 295static int decodeSpectrum (GetBitContext *gb, float *pOut) 296{ 297 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn; 298 int subband_vlc_index[32], SF_idxs[32]; 299 int mantissas[128]; 300 float SF; 301 302 numSubbands = get_bits(gb, 5); // number of coded subbands 303 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC 304 305 /* Get the VLC selector table for the subbands, 0 means not coded. */ 306 for (cnt = 0; cnt <= numSubbands; cnt++) 307 subband_vlc_index[cnt] = get_bits(gb, 3); 308 309 /* Read the scale factor indexes from the stream. */ 310 for (cnt = 0; cnt <= numSubbands; cnt++) { 311 if (subband_vlc_index[cnt] != 0) 312 SF_idxs[cnt] = get_bits(gb, 6); 313 } 314 315 for (cnt = 0; cnt <= numSubbands; cnt++) { 316 first = subbandTab[cnt]; 317 last = subbandTab[cnt+1]; 318 319 subbWidth = last - first; 320 321 if (subband_vlc_index[cnt] != 0) { 322 /* Decode spectral coefficients for this subband. */ 323 /* TODO: This can be done faster is several blocks share the 324 * same VLC selector (subband_vlc_index) */ 325 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth); 326 327 /* Decode the scale factor for this subband. */ 328 SF = sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]]; 329 330 /* Inverse quantize the coefficients. */ 331 for (pIn=mantissas ; first<last; first++, pIn++) 332 pOut[first] = *pIn * SF; 333 } else { 334 /* This subband was not coded, so zero the entire subband. */ 335 memset(pOut+first, 0, subbWidth*sizeof(float)); 336 } 337 } 338 339 /* Clear the subbands that were not coded. */ 340 first = subbandTab[cnt]; 341 memset(pOut+first, 0, (1024 - first) * sizeof(float)); 342 return numSubbands; 343} 344 345/** 346 * Restore the quantized tonal components 347 * 348 * @param gb the GetBit context 349 * @param pComponent tone component 350 * @param numBands amount of coded bands 351 */ 352 353static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands) 354{ 355 int i,j,k,cnt; 356 int components, coding_mode_selector, coding_mode, coded_values_per_component; 357 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components; 358 int band_flags[4], mantissa[8]; 359 float *pCoef; 360 float scalefactor; 361 int component_count = 0; 362 363 components = get_bits(gb,5); 364 365 /* no tonal components */ 366 if (components == 0) 367 return 0; 368 369 coding_mode_selector = get_bits(gb,2); 370 if (coding_mode_selector == 2) 371 return -1; 372 373 coding_mode = coding_mode_selector & 1; 374 375 for (i = 0; i < components; i++) { 376 for (cnt = 0; cnt <= numBands; cnt++) 377 band_flags[cnt] = get_bits1(gb); 378 379 coded_values_per_component = get_bits(gb,3); 380 381 quant_step_index = get_bits(gb,3); 382 if (quant_step_index <= 1) 383 return -1; 384 385 if (coding_mode_selector == 3) 386 coding_mode = get_bits1(gb); 387 388 for (j = 0; j < (numBands + 1) * 4; j++) { 389 if (band_flags[j >> 2] == 0) 390 continue; 391 392 coded_components = get_bits(gb,3); 393 394 for (k=0; k<coded_components; k++) { 395 sfIndx = get_bits(gb,6); 396 pComponent[component_count].pos = j * 64 + (get_bits(gb,6)); 397 max_coded_values = 1024 - pComponent[component_count].pos; 398 coded_values = coded_values_per_component + 1; 399 coded_values = FFMIN(max_coded_values,coded_values); 400 401 scalefactor = sf_table[sfIndx] * iMaxQuant[quant_step_index]; 402 403 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values); 404 405 pComponent[component_count].numCoefs = coded_values; 406 407 /* inverse quant */ 408 pCoef = pComponent[component_count].coef; 409 for (cnt = 0; cnt < coded_values; cnt++) 410 pCoef[cnt] = mantissa[cnt] * scalefactor; 411 412 component_count++; 413 } 414 } 415 } 416 417 return component_count; 418} 419 420/** 421 * Decode gain parameters for the coded bands 422 * 423 * @param gb the GetBit context 424 * @param pGb the gainblock for the current band 425 * @param numBands amount of coded bands 426 */ 427 428static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands) 429{ 430 int i, cf, numData; 431 int *pLevel, *pLoc; 432 433 gain_info *pGain = pGb->gBlock; 434 435 for (i=0 ; i<=numBands; i++) 436 { 437 numData = get_bits(gb,3); 438 pGain[i].num_gain_data = numData; 439 pLevel = pGain[i].levcode; 440 pLoc = pGain[i].loccode; 441 442 for (cf = 0; cf < numData; cf++){ 443 pLevel[cf]= get_bits(gb,4); 444 pLoc [cf]= get_bits(gb,5); 445 if(cf && pLoc[cf] <= pLoc[cf-1]) 446 return -1; 447 } 448 } 449 450 /* Clear the unused blocks. */ 451 for (; i<4 ; i++) 452 pGain[i].num_gain_data = 0; 453 454 return 0; 455} 456 457/** 458 * Apply gain parameters and perform the MDCT overlapping part 459 * 460 * @param pIn input float buffer 461 * @param pPrev previous float buffer to perform overlap against 462 * @param pOut output float buffer 463 * @param pGain1 current band gain info 464 * @param pGain2 next band gain info 465 */ 466 467static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2) 468{ 469 /* gain compensation function */ 470 float gain1, gain2, gain_inc; 471 int cnt, numdata, nsample, startLoc, endLoc; 472 473 474 if (pGain2->num_gain_data == 0) 475 gain1 = 1.0; 476 else 477 gain1 = gain_tab1[pGain2->levcode[0]]; 478 479 if (pGain1->num_gain_data == 0) { 480 for (cnt = 0; cnt < 256; cnt++) 481 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt]; 482 } else { 483 numdata = pGain1->num_gain_data; 484 pGain1->loccode[numdata] = 32; 485 pGain1->levcode[numdata] = 4; 486 487 nsample = 0; // current sample = 0 488 489 for (cnt = 0; cnt < numdata; cnt++) { 490 startLoc = pGain1->loccode[cnt] * 8; 491 endLoc = startLoc + 8; 492 493 gain2 = gain_tab1[pGain1->levcode[cnt]]; 494 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15]; 495 496 /* interpolate */ 497 for (; nsample < startLoc; nsample++) 498 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2; 499 500 /* interpolation is done over eight samples */ 501 for (; nsample < endLoc; nsample++) { 502 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2; 503 gain2 *= gain_inc; 504 } 505 } 506 507 for (; nsample < 256; nsample++) 508 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample]; 509 } 510 511 /* Delay for the overlapping part. */ 512 memcpy(pPrev, &pIn[256], 256*sizeof(float)); 513} 514 515/** 516 * Combine the tonal band spectrum and regular band spectrum 517 * Return position of the last tonal coefficient 518 * 519 * @param pSpectrum output spectrum buffer 520 * @param numComponents amount of tonal components 521 * @param pComponent tonal components for this band 522 */ 523 524static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent) 525{ 526 int cnt, i, lastPos = -1; 527 float *pIn, *pOut; 528 529 for (cnt = 0; cnt < numComponents; cnt++){ 530 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos); 531 pIn = pComponent[cnt].coef; 532 pOut = &(pSpectrum[pComponent[cnt].pos]); 533 534 for (i=0 ; i<pComponent[cnt].numCoefs ; i++) 535 pOut[i] += pIn[i]; 536 } 537 538 return lastPos; 539} 540 541 542#define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old))) 543 544static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode) 545{ 546 int i, band, nsample, s1, s2; 547 float c1, c2; 548 float mc1_l, mc1_r, mc2_l, mc2_r; 549 550 for (i=0,band = 0; band < 4*256; band+=256,i++) { 551 s1 = pPrevCode[i]; 552 s2 = pCurrCode[i]; 553 nsample = 0; 554 555 if (s1 != s2) { 556 /* Selector value changed, interpolation needed. */ 557 mc1_l = matrixCoeffs[s1*2]; 558 mc1_r = matrixCoeffs[s1*2+1]; 559 mc2_l = matrixCoeffs[s2*2]; 560 mc2_r = matrixCoeffs[s2*2+1]; 561 562 /* Interpolation is done over the first eight samples. */ 563 for(; nsample < 8; nsample++) { 564 c1 = su1[band+nsample]; 565 c2 = su2[band+nsample]; 566 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample); 567 su1[band+nsample] = c2; 568 su2[band+nsample] = c1 * 2.0 - c2; 569 } 570 } 571 572 /* Apply the matrix without interpolation. */ 573 switch (s2) { 574 case 0: /* M/S decoding */ 575 for (; nsample < 256; nsample++) { 576 c1 = su1[band+nsample]; 577 c2 = su2[band+nsample]; 578 su1[band+nsample] = c2 * 2.0; 579 su2[band+nsample] = (c1 - c2) * 2.0; 580 } 581 break; 582 583 case 1: 584 for (; nsample < 256; nsample++) { 585 c1 = su1[band+nsample]; 586 c2 = su2[band+nsample]; 587 su1[band+nsample] = (c1 + c2) * 2.0; 588 su2[band+nsample] = c2 * -2.0; 589 } 590 break; 591 case 2: 592 case 3: 593 for (; nsample < 256; nsample++) { 594 c1 = su1[band+nsample]; 595 c2 = su2[band+nsample]; 596 su1[band+nsample] = c1 + c2; 597 su2[band+nsample] = c1 - c2; 598 } 599 break; 600 default: 601 assert(0); 602 } 603 } 604} 605 606static void getChannelWeights (int indx, int flag, float ch[2]){ 607 608 if (indx == 7) { 609 ch[0] = 1.0; 610 ch[1] = 1.0; 611 } else { 612 ch[0] = (float)(indx & 7) / 7.0; 613 ch[1] = sqrt(2 - ch[0]*ch[0]); 614 if(flag) 615 FFSWAP(float, ch[0], ch[1]); 616 } 617} 618 619static void channelWeighting (float *su1, float *su2, int *p3) 620{ 621 int band, nsample; 622 /* w[x][y] y=0 is left y=1 is right */ 623 float w[2][2]; 624 625 if (p3[1] != 7 || p3[3] != 7){ 626 getChannelWeights(p3[1], p3[0], w[0]); 627 getChannelWeights(p3[3], p3[2], w[1]); 628 629 for(band = 1; band < 4; band++) { 630 /* scale the channels by the weights */ 631 for(nsample = 0; nsample < 8; nsample++) { 632 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample); 633 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample); 634 } 635 636 for(; nsample < 256; nsample++) { 637 su1[band*256+nsample] *= w[1][0]; 638 su2[band*256+nsample] *= w[1][1]; 639 } 640 } 641 } 642} 643 644 645/** 646 * Decode a Sound Unit 647 * 648 * @param gb the GetBit context 649 * @param pSnd the channel unit to be used 650 * @param pOut the decoded samples before IQMF in float representation 651 * @param channelNum channel number 652 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono) 653 */ 654 655 656static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode) 657{ 658 int band, result=0, numSubbands, lastTonal, numBands; 659 660 if (codingMode == JOINT_STEREO && channelNum == 1) { 661 if (get_bits(gb,2) != 3) { 662 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n"); 663 return -1; 664 } 665 } else { 666 if (get_bits(gb,6) != 0x28) { 667 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n"); 668 return -1; 669 } 670 } 671 672 /* number of coded QMF bands */ 673 pSnd->bandsCoded = get_bits(gb,2); 674 675 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded); 676 if (result) return result; 677 678 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded); 679 if (pSnd->numComponents == -1) return -1; 680 681 numSubbands = decodeSpectrum (gb, pSnd->spectrum); 682 683 /* Merge the decoded spectrum and tonal components. */ 684 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components); 685 686 687 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */ 688 numBands = (subbandTab[numSubbands] - 1) >> 8; 689 if (lastTonal >= 0) 690 numBands = FFMAX((lastTonal + 256) >> 8, numBands); 691 692 693 /* Reconstruct time domain samples. */ 694 for (band=0; band<4; band++) { 695 /* Perform the IMDCT step without overlapping. */ 696 if (band <= numBands) { 697 IMLT(&(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1); 698 } else 699 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float)); 700 701 /* gain compensation and overlapping */ 702 gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]), 703 &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]), 704 &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band])); 705 } 706 707 /* Swap the gain control buffers for the next frame. */ 708 pSnd->gcBlkSwitch ^= 1; 709 710 return 0; 711} 712 713/** 714 * Frame handling 715 * 716 * @param q Atrac3 private context 717 * @param databuf the input data 718 */ 719 720static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf) 721{ 722 int result, i; 723 float *p1, *p2, *p3, *p4; 724 uint8_t *ptr1; 725 726 if (q->codingMode == JOINT_STEREO) { 727 728 /* channel coupling mode */ 729 /* decode Sound Unit 1 */ 730 init_get_bits(&q->gb,databuf,q->bits_per_frame); 731 732 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO); 733 if (result != 0) 734 return (result); 735 736 /* Framedata of the su2 in the joint-stereo mode is encoded in 737 * reverse byte order so we need to swap it first. */ 738 if (databuf == q->decoded_bytes_buffer) { 739 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1; 740 ptr1 = q->decoded_bytes_buffer; 741 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) { 742 FFSWAP(uint8_t,*ptr1,*ptr2); 743 } 744 } else { 745 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1; 746 for (i = 0; i < q->bytes_per_frame; i++) 747 q->decoded_bytes_buffer[i] = *ptr2--; 748 } 749 750 /* Skip the sync codes (0xF8). */ 751 ptr1 = q->decoded_bytes_buffer; 752 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) { 753 if (i >= q->bytes_per_frame) 754 return -1; 755 } 756 757 758 /* set the bitstream reader at the start of the second Sound Unit*/ 759 init_get_bits(&q->gb,ptr1,q->bits_per_frame); 760 761 /* Fill the Weighting coeffs delay buffer */ 762 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int)); 763 q->weighting_delay[4] = get_bits1(&q->gb); 764 q->weighting_delay[5] = get_bits(&q->gb,3); 765 766 for (i = 0; i < 4; i++) { 767 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i]; 768 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i]; 769 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2); 770 } 771 772 /* Decode Sound Unit 2. */ 773 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO); 774 if (result != 0) 775 return (result); 776 777 /* Reconstruct the channel coefficients. */ 778 reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now); 779 780 channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay); 781 782 } else { 783 /* normal stereo mode or mono */ 784 /* Decode the channel sound units. */ 785 for (i=0 ; i<q->channels ; i++) { 786 787 /* Set the bitstream reader at the start of a channel sound unit. */ 788 init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels); 789 790 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode); 791 if (result != 0) 792 return (result); 793 } 794 } 795 796 /* Apply the iQMF synthesis filter. */ 797 p1= q->outSamples; 798 for (i=0 ; i<q->channels ; i++) { 799 p2= p1+256; 800 p3= p2+256; 801 p4= p3+256; 802 atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf); 803 atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf); 804 atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf); 805 p1 +=1024; 806 } 807 808 return 0; 809} 810 811 812/** 813 * Atrac frame decoding 814 * 815 * @param avctx pointer to the AVCodecContext 816 */ 817 818static int atrac3_decode_frame(AVCodecContext *avctx, 819 void *data, int *data_size, 820 AVPacket *avpkt) { 821 const uint8_t *buf = avpkt->data; 822 int buf_size = avpkt->size; 823 ATRAC3Context *q = avctx->priv_data; 824 int result = 0, i; 825 const uint8_t* databuf; 826 int16_t* samples = data; 827 828 if (buf_size < avctx->block_align) 829 return buf_size; 830 831 /* Check if we need to descramble and what buffer to pass on. */ 832 if (q->scrambled_stream) { 833 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align); 834 databuf = q->decoded_bytes_buffer; 835 } else { 836 databuf = buf; 837 } 838 839 result = decodeFrame(q, databuf); 840 841 if (result != 0) { 842 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n"); 843 return -1; 844 } 845 846 if (q->channels == 1) { 847 /* mono */ 848 for (i = 0; i<1024; i++) 849 samples[i] = av_clip_int16(round(q->outSamples[i])); 850 *data_size = 1024 * sizeof(int16_t); 851 } else { 852 /* stereo */ 853 for (i = 0; i < 1024; i++) { 854 samples[i*2] = av_clip_int16(round(q->outSamples[i])); 855 samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i])); 856 } 857 *data_size = 2048 * sizeof(int16_t); 858 } 859 860 return avctx->block_align; 861} 862 863 864/** 865 * Atrac3 initialization 866 * 867 * @param avctx pointer to the AVCodecContext 868 */ 869 870static av_cold int atrac3_decode_init(AVCodecContext *avctx) 871{ 872 int i; 873 const uint8_t *edata_ptr = avctx->extradata; 874 ATRAC3Context *q = avctx->priv_data; 875 static VLC_TYPE atrac3_vlc_table[4096][2]; 876 static int vlcs_initialized = 0; 877 878 /* Take data from the AVCodecContext (RM container). */ 879 q->sample_rate = avctx->sample_rate; 880 q->channels = avctx->channels; 881 q->bit_rate = avctx->bit_rate; 882 q->bits_per_frame = avctx->block_align * 8; 883 q->bytes_per_frame = avctx->block_align; 884 885 /* Take care of the codec-specific extradata. */ 886 if (avctx->extradata_size == 14) { 887 /* Parse the extradata, WAV format */ 888 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1 889 q->samples_per_channel = bytestream_get_le32(&edata_ptr); 890 q->codingMode = bytestream_get_le16(&edata_ptr); 891 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode 892 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1 893 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0 894 895 /* setup */ 896 q->samples_per_frame = 1024 * q->channels; 897 q->atrac3version = 4; 898 q->delay = 0x88E; 899 if (q->codingMode) 900 q->codingMode = JOINT_STEREO; 901 else 902 q->codingMode = STEREO; 903 904 q->scrambled_stream = 0; 905 906 if ((q->bytes_per_frame == 96*q->channels*q->frame_factor) || (q->bytes_per_frame == 152*q->channels*q->frame_factor) || (q->bytes_per_frame == 192*q->channels*q->frame_factor)) { 907 } else { 908 av_log(avctx,AV_LOG_ERROR,"Unknown frame/channel/frame_factor configuration %d/%d/%d\n", q->bytes_per_frame, q->channels, q->frame_factor); 909 return -1; 910 } 911 912 } else if (avctx->extradata_size == 10) { 913 /* Parse the extradata, RM format. */ 914 q->atrac3version = bytestream_get_be32(&edata_ptr); 915 q->samples_per_frame = bytestream_get_be16(&edata_ptr); 916 q->delay = bytestream_get_be16(&edata_ptr); 917 q->codingMode = bytestream_get_be16(&edata_ptr); 918 919 q->samples_per_channel = q->samples_per_frame / q->channels; 920 q->scrambled_stream = 1; 921 922 } else { 923 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size); 924 } 925 /* Check the extradata. */ 926 927 if (q->atrac3version != 4) { 928 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version); 929 return -1; 930 } 931 932 if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) { 933 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame); 934 return -1; 935 } 936 937 if (q->delay != 0x88E) { 938 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay); 939 return -1; 940 } 941 942 if (q->codingMode == STEREO) { 943 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n"); 944 } else if (q->codingMode == JOINT_STEREO) { 945 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n"); 946 } else { 947 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode); 948 return -1; 949 } 950 951 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) { 952 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n"); 953 return -1; 954 } 955 956 957 if(avctx->block_align >= UINT_MAX/2) 958 return -1; 959 960 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE, 961 * this is for the bitstream reader. */ 962 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL) 963 return AVERROR(ENOMEM); 964 965 966 /* Initialize the VLC tables. */ 967 if (!vlcs_initialized) { 968 for (i=0 ; i<7 ; i++) { 969 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]]; 970 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i]; 971 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i], 972 huff_bits[i], 1, 1, 973 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC); 974 } 975 vlcs_initialized = 1; 976 } 977 978 init_atrac3_transforms(q); 979 980 atrac_generate_tables(); 981 982 /* Generate gain tables. */ 983 for (i=0 ; i<16 ; i++) 984 gain_tab1[i] = powf (2.0, (4 - i)); 985 986 for (i=-15 ; i<16 ; i++) 987 gain_tab2[i+15] = powf (2.0, i * -0.125); 988 989 /* init the joint-stereo decoding data */ 990 q->weighting_delay[0] = 0; 991 q->weighting_delay[1] = 7; 992 q->weighting_delay[2] = 0; 993 q->weighting_delay[3] = 7; 994 q->weighting_delay[4] = 0; 995 q->weighting_delay[5] = 7; 996 997 for (i=0; i<4; i++) { 998 q->matrix_coeff_index_prev[i] = 3; 999 q->matrix_coeff_index_now[i] = 3; 1000 q->matrix_coeff_index_next[i] = 3; 1001 } 1002 1003 dsputil_init(&dsp, avctx); 1004 1005 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels); 1006 if (!q->pUnits) { 1007 av_free(q->decoded_bytes_buffer); 1008 return AVERROR(ENOMEM); 1009 } 1010 1011 avctx->sample_fmt = SAMPLE_FMT_S16; 1012 return 0; 1013} 1014 1015 1016AVCodec atrac3_decoder = 1017{ 1018 .name = "atrac3", 1019 .type = AVMEDIA_TYPE_AUDIO, 1020 .id = CODEC_ID_ATRAC3, 1021 .priv_data_size = sizeof(ATRAC3Context), 1022 .init = atrac3_decode_init, 1023 .close = atrac3_decode_close, 1024 .decode = atrac3_decode_frame, 1025 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"), 1026}; 1027