libavcodec/atrac3.c

   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 atrac3.c
  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 "bitstream.h"
  41 #include "dsputil.h"
  42 #include "bytestream.h"
  43
  44 #include "atrac3data.h"
  45
  46 #define JOINT_STEREO    0x12
  47 #define STEREO          0x2
  48
  49
  50 /* These structures are needed to store the parsed gain control data. */
  51 typedef struct {
  52     int   num_gain_data;
  53     int   levcode[8];
  54     int   loccode[8];
  55 } gain_info;
  56
  57 typedef struct {
  58     gain_info   gBlock[4];
  59 } gain_block;
  60
  61 typedef struct {
  62     int     pos;
  63     int     numCoefs;
  64     float   coef[8];
  65 } tonal_component;
  66
  67 typedef struct {
  68     int               bandsCoded;
  69     int               numComponents;
  70     tonal_component   components[64];
  71     float             prevFrame[1024];
  72     int               gcBlkSwitch;
  73     gain_block        gainBlock[2];
  74
  75     DECLARE_ALIGNED_16(float, spectrum[1024]);
  76     DECLARE_ALIGNED_16(float, IMDCT_buf[1024]);
  77
  78     float             delayBuf1[46]; ///<qmf delay buffers
  79     float             delayBuf2[46];
  80     float             delayBuf3[46];
  81 } channel_unit;
  82
  83 typedef struct {
  84     GetBitContext       gb;
  85     //@{
  86     /** stream data */
  87     int                 channels;
  88     int                 codingMode;
  89     int                 bit_rate;
  90     int                 sample_rate;
  91     int                 samples_per_channel;
  92     int                 samples_per_frame;
  93
  94     int                 bits_per_frame;
  95     int                 bytes_per_frame;
  96     int                 pBs;
  97     channel_unit*       pUnits;
  98     //@}
  99     //@{
 100     /** joint-stereo related variables */
 101     int                 matrix_coeff_index_prev[4];
 102     int                 matrix_coeff_index_now[4];
 103     int                 matrix_coeff_index_next[4];
 104     int                 weighting_delay[6];
 105     //@}
 106     //@{
 107     /** data buffers */
 108     float               outSamples[2048];
 109     uint8_t*            decoded_bytes_buffer;
 110     float               tempBuf[1070];
 111     DECLARE_ALIGNED_16(float,mdct_tmp[512]);
 112     //@}
 113     //@{
 114     /** extradata */
 115     int                 atrac3version;
 116     int                 delay;
 117     int                 scrambled_stream;
 118     int                 frame_factor;
 119     //@}
 120 } ATRAC3Context;
 121
 122 static DECLARE_ALIGNED_16(float,mdct_window[512]);
 123 static float            qmf_window[48];
 124 static VLC              spectral_coeff_tab[7];
 125 static float            SFTable[64];
 126 static float            gain_tab1[16];
 127 static float            gain_tab2[31];
 128 static MDCTContext      mdct_ctx;
 129 static DSPContext       dsp;
 130
 131
 132 /* quadrature mirror synthesis filter */
 133
 134 /**
 135  * Quadrature mirror synthesis filter.
 136  *
 137  * @param inlo      lower part of spectrum
 138  * @param inhi      higher part of spectrum
 139  * @param nIn       size of spectrum buffer
 140  * @param pOut      out buffer
 141  * @param delayBuf  delayBuf buffer
 142  * @param temp      temp buffer
 143  */
 144
 145
 146 static void iqmf (float *inlo, float *inhi, unsigned int nIn, float *pOut, float *delayBuf, float *temp)
 147 {
 148     int   i, j;
 149     float   *p1, *p3;
 150
 151     memcpy(temp, delayBuf, 46*sizeof(float));
 152
 153     p3 = temp + 46;
 154
 155     /* loop1 */
 156     for(i=0; i<nIn; i+=2){
 157         p3[2*i+0] = inlo[i  ] + inhi[i  ];
 158         p3[2*i+1] = inlo[i  ] - inhi[i  ];
 159         p3[2*i+2] = inlo[i+1] + inhi[i+1];
 160         p3[2*i+3] = inlo[i+1] - inhi[i+1];
 161     }
 162
 163     /* loop2 */
 164     p1 = temp;
 165     for (j = nIn; j != 0; j--) {
 166         float s1 = 0.0;
 167         float s2 = 0.0;
 168
 169         for (i = 0; i < 48; i += 2) {
 170             s1 += p1[i] * qmf_window[i];
 171             s2 += p1[i+1] * qmf_window[i+1];
 172         }
 173
 174         pOut[0] = s2;
 175         pOut[1] = s1;
 176
 177         p1 += 2;
 178         pOut += 2;
 179     }
 180
 181     /* Update the delay buffer. */
 182     memcpy(delayBuf, temp + nIn*2, 46*sizeof(float));
 183 }
 184
 185 /**
 186  * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
 187  * caused by the reverse spectra of the QMF.
 188  *
 189  * @param pInput    float input
 190  * @param pOutput   float output
 191  * @param odd_band  1 if the band is an odd band
 192  * @param mdct_tmp  aligned temporary buffer for the mdct
 193  */
 194
 195 static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
 196 {
 197     int     i;
 198
 199     if (odd_band) {
 200         /**
 201         * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
 202         * or it gives better compression to do it this way.
 203         * FIXME: It should be possible to handle this in ff_imdct_calc
 204         * for that to happen a modification of the prerotation step of
 205         * all SIMD code and C code is needed.
 206         * Or fix the functions before so they generate a pre reversed spectrum.
 207         */
 208
 209         for (i=0; i<128; i++)
 210             FFSWAP(float, pInput[i], pInput[255-i]);
 211     }
 212
 213     mdct_ctx.fft.imdct_calc(&mdct_ctx,pOutput,pInput,mdct_tmp);
 214
 215     /* Perform windowing on the output. */
 216     dsp.vector_fmul(pOutput,mdct_window,512);
 217
 218 }
 219
 220
 221 /**
 222  * Atrac 3 indata descrambling, only used for data coming from the rm container
 223  *
 224  * @param in        pointer to 8 bit array of indata
 225  * @param bits      amount of bits
 226  * @param out       pointer to 8 bit array of outdata
 227  */
 228
 229 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
 230     int i, off;
 231     uint32_t c;
 232     const uint32_t* buf;
 233     uint32_t* obuf = (uint32_t*) out;
 234
 235     off = (int)((long)inbuffer & 3);
 236     buf = (const uint32_t*) (inbuffer - off);
 237     c = be2me_32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
 238     bytes += 3 + off;
 239     for (i = 0; i < bytes/4; i++)
 240         obuf[i] = c ^ buf[i];
 241
 242     if (off)
 243         av_log(NULL,AV_LOG_DEBUG,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off);
 244
 245     return off;
 246 }
 247
 248
 249 static void init_atrac3_transforms(ATRAC3Context *q) {
 250     float enc_window[256];
 251     float s;
 252     int i;
 253
 254     /* Generate the mdct window, for details see
 255      * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
 256     for (i=0 ; i<256; i++)
 257         enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
 258
 259     if (!mdct_window[0])
 260         for (i=0 ; i<256; i++) {
 261             mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
 262             mdct_window[511-i] = mdct_window[i];
 263         }
 264
 265     /* Generate the QMF window. */
 266     for (i=0 ; i<24; i++) {
 267         s = qmf_48tap_half[i] * 2.0;
 268         qmf_window[i] = s;
 269         qmf_window[47 - i] = s;
 270     }
 271
 272     /* Initialize the MDCT transform. */
 273     ff_mdct_init(&mdct_ctx, 9, 1);
 274 }
 275
 276 /**
 277  * Atrac3 uninit, free all allocated memory
 278  */
 279
 280 static int atrac3_decode_close(AVCodecContext *avctx)
 281 {
 282     ATRAC3Context *q = avctx->priv_data;
 283
 284     av_free(q->pUnits);
 285     av_free(q->decoded_bytes_buffer);
 286
 287     return 0;
 288 }
 289
 290 /**
 291 / * Mantissa decoding
 292  *
 293  * @param gb            the GetBit context
 294  * @param selector      what table is the output values coded with
 295  * @param codingFlag    constant length coding or variable length coding
 296  * @param mantissas     mantissa output table
 297  * @param numCodes      amount of values to get
 298  */
 299
 300 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
 301 {
 302     int   numBits, cnt, code, huffSymb;
 303
 304     if (selector == 1)
 305         numCodes /= 2;
 306
 307     if (codingFlag != 0) {
 308         /* constant length coding (CLC) */
 309         //FIXME we don't have any samples coded in CLC mode
 310         numBits = CLCLengthTab[selector];
 311
 312         if (selector > 1) {
 313             for (cnt = 0; cnt < numCodes; cnt++) {
 314                 if (numBits)
 315                     code = get_sbits(gb, numBits);
 316                 else
 317                     code = 0;
 318                 mantissas[cnt] = code;
 319             }
 320         } else {
 321             for (cnt = 0; cnt < numCodes; cnt++) {
 322                 if (numBits)
 323                     code = get_bits(gb, numBits); //numBits is always 4 in this case
 324                 else
 325                     code = 0;
 326                 mantissas[cnt*2] = seTab_0[code >> 2];
 327                 mantissas[cnt*2+1] = seTab_0[code & 3];
 328             }
 329         }
 330     } else {
 331         /* variable length coding (VLC) */
 332         if (selector != 1) {
 333             for (cnt = 0; cnt < numCodes; cnt++) {
 334                 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
 335                 huffSymb += 1;
 336                 code = huffSymb >> 1;
 337                 if (huffSymb & 1)
 338                     code = -code;
 339                 mantissas[cnt] = code;
 340             }
 341         } else {
 342             for (cnt = 0; cnt < numCodes; cnt++) {
 343                 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
 344                 mantissas[cnt*2] = decTable1[huffSymb*2];
 345                 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
 346             }
 347         }
 348     }
 349 }
 350
 351 /**
 352  * Restore the quantized band spectrum coefficients
 353  *
 354  * @param gb            the GetBit context
 355  * @param pOut          decoded band spectrum
 356  * @return outSubbands   subband counter, fix for broken specification/files
 357  */
 358
 359 static int decodeSpectrum (GetBitContext *gb, float *pOut)
 360 {
 361     int   numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
 362     int   subband_vlc_index[32], SF_idxs[32];
 363     int   mantissas[128];
 364     float SF;
 365
 366     numSubbands = get_bits(gb, 5); // number of coded subbands
 367     codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
 368
 369     /* Get the VLC selector table for the subbands, 0 means not coded. */
 370     for (cnt = 0; cnt <= numSubbands; cnt++)
 371         subband_vlc_index[cnt] = get_bits(gb, 3);
 372
 373     /* Read the scale factor indexes from the stream. */
 374     for (cnt = 0; cnt <= numSubbands; cnt++) {
 375         if (subband_vlc_index[cnt] != 0)
 376             SF_idxs[cnt] = get_bits(gb, 6);
 377     }
 378
 379     for (cnt = 0; cnt <= numSubbands; cnt++) {
 380         first = subbandTab[cnt];
 381         last = subbandTab[cnt+1];
 382
 383         subbWidth = last - first;
 384
 385         if (subband_vlc_index[cnt] != 0) {
 386             /* Decode spectral coefficients for this subband. */
 387             /* TODO: This can be done faster is several blocks share the
 388              * same VLC selector (subband_vlc_index) */
 389             readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
 390
 391             /* Decode the scale factor for this subband. */
 392             SF = SFTable[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
 393
 394             /* Inverse quantize the coefficients. */
 395             for (pIn=mantissas ; first<last; first++, pIn++)
 396                 pOut[first] = *pIn * SF;
 397         } else {
 398             /* This subband was not coded, so zero the entire subband. */
 399             memset(pOut+first, 0, subbWidth*sizeof(float));
 400         }
 401     }
 402
 403     /* Clear the subbands that were not coded. */
 404     first = subbandTab[cnt];
 405     memset(pOut+first, 0, (1024 - first) * sizeof(float));
 406     return numSubbands;
 407 }
 408
 409 /**
 410  * Restore the quantized tonal components
 411  *
 412  * @param gb            the GetBit context
 413  * @param pComponent    tone component
 414  * @param numBands      amount of coded bands
 415  */
 416
 417 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
 418 {
 419     int i,j,k,cnt;
 420     int   components, coding_mode_selector, coding_mode, coded_values_per_component;
 421     int   sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
 422     int   band_flags[4], mantissa[8];
 423     float  *pCoef;
 424     float  scalefactor;
 425     int   component_count = 0;
 426
 427     components = get_bits(gb,5);
 428
 429     /* no tonal components */
 430     if (components == 0)
 431         return 0;
 432
 433     coding_mode_selector = get_bits(gb,2);
 434     if (coding_mode_selector == 2)
 435         return -1;
 436
 437     coding_mode = coding_mode_selector & 1;
 438
 439     for (i = 0; i < components; i++) {
 440         for (cnt = 0; cnt <= numBands; cnt++)
 441             band_flags[cnt] = get_bits1(gb);
 442
 443         coded_values_per_component = get_bits(gb,3);
 444
 445         quant_step_index = get_bits(gb,3);
 446         if (quant_step_index <= 1)
 447             return -1;
 448
 449         if (coding_mode_selector == 3)
 450             coding_mode = get_bits1(gb);
 451
 452         for (j = 0; j < (numBands + 1) * 4; j++) {
 453             if (band_flags[j >> 2] == 0)
 454                 continue;
 455
 456             coded_components = get_bits(gb,3);
 457
 458             for (k=0; k<coded_components; k++) {
 459                 sfIndx = get_bits(gb,6);
 460                 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
 461                 max_coded_values = 1024 - pComponent[component_count].pos;
 462                 coded_values = coded_values_per_component + 1;
 463                 coded_values = FFMIN(max_coded_values,coded_values);
 464
 465                 scalefactor = SFTable[sfIndx] * iMaxQuant[quant_step_index];
 466
 467                 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
 468
 469                 pComponent[component_count].numCoefs = coded_values;
 470
 471                 /* inverse quant */
 472                 pCoef = pComponent[component_count].coef;
 473                 for (cnt = 0; cnt < coded_values; cnt++)
 474                     pCoef[cnt] = mantissa[cnt] * scalefactor;
 475
 476                 component_count++;
 477             }
 478         }
 479     }
 480
 481     return component_count;
 482 }
 483
 484 /**
 485  * Decode gain parameters for the coded bands
 486  *
 487  * @param gb            the GetBit context
 488  * @param pGb           the gainblock for the current band
 489  * @param numBands      amount of coded bands
 490  */
 491
 492 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
 493 {
 494     int   i, cf, numData;
 495     int   *pLevel, *pLoc;
 496
 497     gain_info   *pGain = pGb->gBlock;
 498
 499     for (i=0 ; i<=numBands; i++)
 500     {
 501         numData = get_bits(gb,3);
 502         pGain[i].num_gain_data = numData;
 503         pLevel = pGain[i].levcode;
 504         pLoc = pGain[i].loccode;
 505
 506         for (cf = 0; cf < numData; cf++){
 507             pLevel[cf]= get_bits(gb,4);
 508             pLoc  [cf]= get_bits(gb,5);
 509             if(cf && pLoc[cf] <= pLoc[cf-1])
 510                 return -1;
 511         }
 512     }
 513
 514     /* Clear the unused blocks. */
 515     for (; i<4 ; i++)
 516         pGain[i].num_gain_data = 0;
 517
 518     return 0;
 519 }
 520
 521 /**
 522  * Apply gain parameters and perform the MDCT overlapping part
 523  *
 524  * @param pIn           input float buffer
 525  * @param pPrev         previous float buffer to perform overlap against
 526  * @param pOut          output float buffer
 527  * @param pGain1        current band gain info
 528  * @param pGain2        next band gain info
 529  */
 530
 531 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
 532 {
 533     /* gain compensation function */
 534     float  gain1, gain2, gain_inc;
 535     int   cnt, numdata, nsample, startLoc, endLoc;
 536
 537
 538     if (pGain2->num_gain_data == 0)
 539         gain1 = 1.0;
 540     else
 541         gain1 = gain_tab1[pGain2->levcode[0]];
 542
 543     if (pGain1->num_gain_data == 0) {
 544         for (cnt = 0; cnt < 256; cnt++)
 545             pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
 546     } else {
 547         numdata = pGain1->num_gain_data;
 548         pGain1->loccode[numdata] = 32;
 549         pGain1->levcode[numdata] = 4;
 550
 551         nsample = 0; // current sample = 0
 552
 553         for (cnt = 0; cnt < numdata; cnt++) {
 554             startLoc = pGain1->loccode[cnt] * 8;
 555             endLoc = startLoc + 8;
 556
 557             gain2 = gain_tab1[pGain1->levcode[cnt]];
 558             gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
 559
 560             /* interpolate */
 561             for (; nsample < startLoc; nsample++)
 562                 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
 563
 564             /* interpolation is done over eight samples */
 565             for (; nsample < endLoc; nsample++) {
 566                 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
 567                 gain2 *= gain_inc;
 568             }
 569         }
 570
 571         for (; nsample < 256; nsample++)
 572             pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
 573     }
 574
 575     /* Delay for the overlapping part. */
 576     memcpy(pPrev, &pIn[256], 256*sizeof(float));
 577 }
 578
 579 /**
 580  * Combine the tonal band spectrum and regular band spectrum
 581  * Return position of the last tonal coefficient
 582  *
 583  * @param pSpectrum     output spectrum buffer
 584  * @param numComponents amount of tonal components
 585  * @param pComponent    tonal components for this band
 586  */
 587
 588 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
 589 {
 590     int   cnt, i, lastPos = -1;
 591     float   *pIn, *pOut;
 592
 593     for (cnt = 0; cnt < numComponents; cnt++){
 594         lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
 595         pIn = pComponent[cnt].coef;
 596         pOut = &(pSpectrum[pComponent[cnt].pos]);
 597
 598         for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
 599             pOut[i] += pIn[i];
 600     }
 601
 602     return lastPos;
 603 }
 604
 605
 606 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
 607
 608 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
 609 {
 610     int    i, band, nsample, s1, s2;
 611     float    c1, c2;
 612     float    mc1_l, mc1_r, mc2_l, mc2_r;
 613
 614     for (i=0,band = 0; band < 4*256; band+=256,i++) {
 615         s1 = pPrevCode[i];
 616         s2 = pCurrCode[i];
 617         nsample = 0;
 618
 619         if (s1 != s2) {
 620             /* Selector value changed, interpolation needed. */
 621             mc1_l = matrixCoeffs[s1*2];
 622             mc1_r = matrixCoeffs[s1*2+1];
 623             mc2_l = matrixCoeffs[s2*2];
 624             mc2_r = matrixCoeffs[s2*2+1];
 625
 626             /* Interpolation is done over the first eight samples. */
 627             for(; nsample < 8; nsample++) {
 628                 c1 = su1[band+nsample];
 629                 c2 = su2[band+nsample];
 630                 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
 631                 su1[band+nsample] = c2;
 632                 su2[band+nsample] = c1 * 2.0 - c2;
 633             }
 634         }
 635
 636         /* Apply the matrix without interpolation. */
 637         switch (s2) {
 638             case 0:     /* M/S decoding */
 639                 for (; nsample < 256; nsample++) {
 640                     c1 = su1[band+nsample];
 641                     c2 = su2[band+nsample];
 642                     su1[band+nsample] = c2 * 2.0;
 643                     su2[band+nsample] = (c1 - c2) * 2.0;
 644                 }
 645                 break;
 646
 647             case 1:
 648                 for (; nsample < 256; nsample++) {
 649                     c1 = su1[band+nsample];
 650                     c2 = su2[band+nsample];
 651                     su1[band+nsample] = (c1 + c2) * 2.0;
 652                     su2[band+nsample] = c2 * -2.0;
 653                 }
 654                 break;
 655             case 2:
 656             case 3:
 657                 for (; nsample < 256; nsample++) {
 658                     c1 = su1[band+nsample];
 659                     c2 = su2[band+nsample];
 660                     su1[band+nsample] = c1 + c2;
 661                     su2[band+nsample] = c1 - c2;
 662                 }
 663                 break;
 664             default:
 665                 assert(0);
 666         }
 667     }
 668 }
 669
 670 static void getChannelWeights (int indx, int flag, float ch[2]){
 671
 672     if (indx == 7) {
 673         ch[0] = 1.0;
 674         ch[1] = 1.0;
 675     } else {
 676         ch[0] = (float)(indx & 7) / 7.0;
 677         ch[1] = sqrt(2 - ch[0]*ch[0]);
 678         if(flag)
 679             FFSWAP(float, ch[0], ch[1]);
 680     }
 681 }
 682
 683 static void channelWeighting (float *su1, float *su2, int *p3)
 684 {
 685     int   band, nsample;
 686     /* w[x][y] y=0 is left y=1 is right */
 687     float w[2][2];
 688
 689     if (p3[1] != 7 || p3[3] != 7){
 690         getChannelWeights(p3[1], p3[0], w[0]);
 691         getChannelWeights(p3[3], p3[2], w[1]);
 692
 693         for(band = 1; band < 4; band++) {
 694             /* scale the channels by the weights */
 695             for(nsample = 0; nsample < 8; nsample++) {
 696                 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
 697                 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
 698             }
 699
 700             for(; nsample < 256; nsample++) {
 701                 su1[band*256+nsample] *= w[1][0];
 702                 su2[band*256+nsample] *= w[1][1];
 703             }
 704         }
 705     }
 706 }
 707
 708
 709 /**
 710  * Decode a Sound Unit
 711  *
 712  * @param gb            the GetBit context
 713  * @param pSnd          the channel unit to be used
 714  * @param pOut          the decoded samples before IQMF in float representation
 715  * @param channelNum    channel number
 716  * @param codingMode    the coding mode (JOINT_STEREO or regular stereo/mono)
 717  */
 718
 719
 720 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
 721 {
 722     int   band, result=0, numSubbands, lastTonal, numBands;
 723
 724     if (codingMode == JOINT_STEREO && channelNum == 1) {
 725         if (get_bits(gb,2) != 3) {
 726             av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
 727             return -1;
 728         }
 729     } else {
 730         if (get_bits(gb,6) != 0x28) {
 731             av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
 732             return -1;
 733         }
 734     }
 735
 736     /* number of coded QMF bands */
 737     pSnd->bandsCoded = get_bits(gb,2);
 738
 739     result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
 740     if (result) return result;
 741
 742     pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
 743     if (pSnd->numComponents == -1) return -1;
 744
 745     numSubbands = decodeSpectrum (gb, pSnd->spectrum);
 746
 747     /* Merge the decoded spectrum and tonal components. */
 748     lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
 749
 750
 751     /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
 752     numBands = (subbandTab[numSubbands] - 1) >> 8;
 753     if (lastTonal >= 0)
 754         numBands = FFMAX((lastTonal + 256) >> 8, numBands);
 755
 756
 757     /* Reconstruct time domain samples. */
 758     for (band=0; band<4; band++) {
 759         /* Perform the IMDCT step without overlapping. */
 760         if (band <= numBands) {
 761             IMLT(&(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1,q->mdct_tmp);
 762         } else
 763             memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
 764
 765         /* gain compensation and overlapping */
 766         gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]),
 767                                     &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]),
 768                                     &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band]));
 769     }
 770
 771     /* Swap the gain control buffers for the next frame. */
 772     pSnd->gcBlkSwitch ^= 1;
 773
 774     return 0;
 775 }
 776
 777 /**
 778  * Frame handling
 779  *
 780  * @param q             Atrac3 private context
 781  * @param databuf       the input data
 782  */
 783
 784 static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
 785 {
 786     int   result, i;
 787     float   *p1, *p2, *p3, *p4;
 788     uint8_t    *ptr1, *ptr2;
 789
 790     if (q->codingMode == JOINT_STEREO) {
 791
 792         /* channel coupling mode */
 793         /* decode Sound Unit 1 */
 794         init_get_bits(&q->gb,databuf,q->bits_per_frame);
 795
 796         result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO);
 797         if (result != 0)
 798             return (result);
 799
 800         /* Framedata of the su2 in the joint-stereo mode is encoded in
 801          * reverse byte order so we need to swap it first. */
 802         ptr1 = databuf;
 803         ptr2 = databuf+q->bytes_per_frame-1;
 804         for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
 805             FFSWAP(uint8_t,*ptr1,*ptr2);
 806         }
 807
 808         /* Skip the sync codes (0xF8). */
 809         ptr1 = databuf;
 810         for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
 811             if (i >= q->bytes_per_frame)
 812                 return -1;
 813         }
 814
 815
 816         /* set the bitstream reader at the start of the second Sound Unit*/
 817         init_get_bits(&q->gb,ptr1,q->bits_per_frame);
 818
 819         /* Fill the Weighting coeffs delay buffer */
 820         memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
 821         q->weighting_delay[4] = get_bits1(&q->gb);
 822         q->weighting_delay[5] = get_bits(&q->gb,3);
 823
 824         for (i = 0; i < 4; i++) {
 825             q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
 826             q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
 827             q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
 828         }
 829
 830         /* Decode Sound Unit 2. */
 831         result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO);
 832         if (result != 0)
 833             return (result);
 834
 835         /* Reconstruct the channel coefficients. */
 836         reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
 837
 838         channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay);
 839
 840     } else {
 841         /* normal stereo mode or mono */
 842         /* Decode the channel sound units. */
 843         for (i=0 ; i<q->channels ; i++) {
 844
 845             /* Set the bitstream reader at the start of a channel sound unit. */
 846             init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels);
 847
 848             result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode);
 849             if (result != 0)
 850                 return (result);
 851         }
 852     }
 853
 854     /* Apply the iQMF synthesis filter. */
 855     p1= q->outSamples;
 856     for (i=0 ; i<q->channels ; i++) {
 857         p2= p1+256;
 858         p3= p2+256;
 859         p4= p3+256;
 860         iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
 861         iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
 862         iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
 863         p1 +=1024;
 864     }
 865
 866     return 0;
 867 }
 868
 869
 870 /**
 871  * Atrac frame decoding
 872  *
 873  * @param avctx     pointer to the AVCodecContext
 874  */
 875
 876 static int atrac3_decode_frame(AVCodecContext *avctx,
 877             void *data, int *data_size,
 878             const uint8_t *buf, int buf_size) {
 879     ATRAC3Context *q = avctx->priv_data;
 880     int result = 0, i;
 881     uint8_t* databuf;
 882     int16_t* samples = data;
 883
 884     if (buf_size < avctx->block_align)
 885         return buf_size;
 886
 887     /* Check if we need to descramble and what buffer to pass on. */
 888     if (q->scrambled_stream) {
 889         decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
 890         databuf = q->decoded_bytes_buffer;
 891     } else {
 892         databuf = buf;
 893     }
 894
 895     result = decodeFrame(q, databuf);
 896
 897     if (result != 0) {
 898         av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
 899         return -1;
 900     }
 901
 902     if (q->channels == 1) {
 903         /* mono */
 904         for (i = 0; i<1024; i++)
 905             samples[i] = av_clip_int16(round(q->outSamples[i]));
 906         *data_size = 1024 * sizeof(int16_t);
 907     } else {
 908         /* stereo */
 909         for (i = 0; i < 1024; i++) {
 910             samples[i*2] = av_clip_int16(round(q->outSamples[i]));
 911             samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i]));
 912         }
 913         *data_size = 2048 * sizeof(int16_t);
 914     }
 915
 916     return avctx->block_align;
 917 }
 918
 919
 920 /**
 921  * Atrac3 initialization
 922  *
 923  * @param avctx     pointer to the AVCodecContext
 924  */
 925
 926 static int atrac3_decode_init(AVCodecContext *avctx)
 927 {
 928     int i;
 929     const uint8_t *edata_ptr = avctx->extradata;
 930     ATRAC3Context *q = avctx->priv_data;
 931
 932     /* Take data from the AVCodecContext (RM container). */
 933     q->sample_rate = avctx->sample_rate;
 934     q->channels = avctx->channels;
 935     q->bit_rate = avctx->bit_rate;
 936     q->bits_per_frame = avctx->block_align * 8;
 937     q->bytes_per_frame = avctx->block_align;
 938
 939     /* Take care of the codec-specific extradata. */
 940     if (avctx->extradata_size == 14) {
 941         /* Parse the extradata, WAV format */
 942         av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr));  //Unknown value always 1
 943         q->samples_per_channel = bytestream_get_le32(&edata_ptr);
 944         q->codingMode = bytestream_get_le16(&edata_ptr);
 945         av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr));  //Dupe of coding mode
 946         q->frame_factor = bytestream_get_le16(&edata_ptr);  //Unknown always 1
 947         av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr));  //Unknown always 0
 948
 949         /* setup */
 950         q->samples_per_frame = 1024 * q->channels;
 951         q->atrac3version = 4;
 952         q->delay = 0x88E;
 953         if (q->codingMode)
 954             q->codingMode = JOINT_STEREO;
 955         else
 956             q->codingMode = STEREO;
 957
 958         q->scrambled_stream = 0;
 959
 960         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)) {
 961         } else {
 962             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);
 963             return -1;
 964         }
 965
 966     } else if (avctx->extradata_size == 10) {
 967         /* Parse the extradata, RM format. */
 968         q->atrac3version = bytestream_get_be32(&edata_ptr);
 969         q->samples_per_frame = bytestream_get_be16(&edata_ptr);
 970         q->delay = bytestream_get_be16(&edata_ptr);
 971         q->codingMode = bytestream_get_be16(&edata_ptr);
 972
 973         q->samples_per_channel = q->samples_per_frame / q->channels;
 974         q->scrambled_stream = 1;
 975
 976     } else {
 977         av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
 978     }
 979     /* Check the extradata. */
 980
 981     if (q->atrac3version != 4) {
 982         av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
 983         return -1;
 984     }
 985
 986     if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) {
 987         av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
 988         return -1;
 989     }
 990
 991     if (q->delay != 0x88E) {
 992         av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
 993         return -1;
 994     }
 995
 996     if (q->codingMode == STEREO) {
 997         av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
 998     } else if (q->codingMode == JOINT_STEREO) {
 999         av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
1000     } else {
1001         av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
1002         return -1;
1003     }
1004
1005     if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
1006         av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
1007         return -1;
1008     }
1009
1010
1011     if(avctx->block_align >= UINT_MAX/2)
1012         return -1;
1013
1014     /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
1015      * this is for the bitstream reader. */
1016     if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE)))  == NULL)
1017         return AVERROR(ENOMEM);
1018
1019
1020     /* Initialize the VLC tables. */
1021     for (i=0 ; i<7 ; i++) {
1022         init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
1023             huff_bits[i], 1, 1,
1024             huff_codes[i], 1, 1, INIT_VLC_USE_STATIC);
1025     }
1026
1027     init_atrac3_transforms(q);
1028
1029     /* Generate the scale factors. */
1030     for (i=0 ; i<64 ; i++)
1031         SFTable[i] = pow(2.0, (i - 15) / 3.0);
1032
1033     /* Generate gain tables. */
1034     for (i=0 ; i<16 ; i++)
1035         gain_tab1[i] = powf (2.0, (4 - i));
1036
1037     for (i=-15 ; i<16 ; i++)
1038         gain_tab2[i+15] = powf (2.0, i * -0.125);
1039
1040     /* init the joint-stereo decoding data */
1041     q->weighting_delay[0] = 0;
1042     q->weighting_delay[1] = 7;
1043     q->weighting_delay[2] = 0;
1044     q->weighting_delay[3] = 7;
1045     q->weighting_delay[4] = 0;
1046     q->weighting_delay[5] = 7;
1047
1048     for (i=0; i<4; i++) {
1049         q->matrix_coeff_index_prev[i] = 3;
1050         q->matrix_coeff_index_now[i] = 3;
1051         q->matrix_coeff_index_next[i] = 3;
1052     }
1053
1054     dsputil_init(&dsp, avctx);
1055
1056     q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1057     if (!q->pUnits) {
1058         av_free(q->decoded_bytes_buffer);
1059         return AVERROR(ENOMEM);
1060     }
1061
1062     return 0;
1063 }
1064
1065
1066 AVCodec atrac3_decoder =
1067 {
1068     .name = "atrac3",
1069     .type = CODEC_TYPE_AUDIO,
1070     .id = CODEC_ID_ATRAC3,
1071     .priv_data_size = sizeof(ATRAC3Context),
1072     .init = atrac3_decode_init,
1073     .close = atrac3_decode_close,
1074     .decode = atrac3_decode_frame,
1075     .long_name = "Atrac 3 (Adaptive TRansform Acoustic Coding 3)",
1076 };