libavcodec/atrac3.c
Go to the documentation of this file.
00001 /*
00002  * Atrac 3 compatible decoder
00003  * Copyright (c) 2006-2008 Maxim Poliakovski
00004  * Copyright (c) 2006-2008 Benjamin Larsson
00005  *
00006  * This file is part of FFmpeg.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with FFmpeg; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 
00035 #include <math.h>
00036 #include <stddef.h>
00037 #include <stdio.h>
00038 
00039 #include "avcodec.h"
00040 #include "get_bits.h"
00041 #include "dsputil.h"
00042 #include "bytestream.h"
00043 #include "fft.h"
00044 #include "fmtconvert.h"
00045 
00046 #include "atrac.h"
00047 #include "atrac3data.h"
00048 
00049 #define JOINT_STEREO    0x12
00050 #define STEREO          0x2
00051 
00052 #define SAMPLES_PER_FRAME 1024
00053 #define MDCT_SIZE          512
00054 
00055 /* These structures are needed to store the parsed gain control data. */
00056 typedef struct {
00057     int   num_gain_data;
00058     int   levcode[8];
00059     int   loccode[8];
00060 } gain_info;
00061 
00062 typedef struct {
00063     gain_info   gBlock[4];
00064 } gain_block;
00065 
00066 typedef struct {
00067     int     pos;
00068     int     numCoefs;
00069     float   coef[8];
00070 } tonal_component;
00071 
00072 typedef struct {
00073     int               bandsCoded;
00074     int               numComponents;
00075     tonal_component   components[64];
00076     float             prevFrame[SAMPLES_PER_FRAME];
00077     int               gcBlkSwitch;
00078     gain_block        gainBlock[2];
00079 
00080     DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
00081     DECLARE_ALIGNED(32, float, IMDCT_buf)[SAMPLES_PER_FRAME];
00082 
00083     float             delayBuf1[46]; 
00084     float             delayBuf2[46];
00085     float             delayBuf3[46];
00086 } channel_unit;
00087 
00088 typedef struct {
00089     AVFrame             frame;
00090     GetBitContext       gb;
00092 
00093     int                 channels;
00094     int                 codingMode;
00095     int                 bit_rate;
00096     int                 sample_rate;
00097     int                 samples_per_channel;
00098     int                 samples_per_frame;
00099 
00100     int                 bits_per_frame;
00101     int                 bytes_per_frame;
00102     int                 pBs;
00103     channel_unit*       pUnits;
00105 
00106 
00107     int                 matrix_coeff_index_prev[4];
00108     int                 matrix_coeff_index_now[4];
00109     int                 matrix_coeff_index_next[4];
00110     int                 weighting_delay[6];
00112 
00113 
00114     float              *outSamples[2];
00115     uint8_t*            decoded_bytes_buffer;
00116     float               tempBuf[1070];
00118 
00119 
00120     int                 atrac3version;
00121     int                 delay;
00122     int                 scrambled_stream;
00123     int                 frame_factor;
00125 
00126     FFTContext          mdct_ctx;
00127     FmtConvertContext   fmt_conv;
00128 } ATRAC3Context;
00129 
00130 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
00131 static VLC              spectral_coeff_tab[7];
00132 static float            gain_tab1[16];
00133 static float            gain_tab2[31];
00134 static DSPContext       dsp;
00135 
00136 
00146 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
00147 {
00148     int     i;
00149 
00150     if (odd_band) {
00160         for (i=0; i<128; i++)
00161             FFSWAP(float, pInput[i], pInput[255-i]);
00162     }
00163 
00164     q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
00165 
00166     /* Perform windowing on the output. */
00167     dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
00168 
00169 }
00170 
00171 
00180 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
00181     int i, off;
00182     uint32_t c;
00183     const uint32_t* buf;
00184     uint32_t* obuf = (uint32_t*) out;
00185 
00186     off = (intptr_t)inbuffer & 3;
00187     buf = (const uint32_t*) (inbuffer - off);
00188     c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
00189     bytes += 3 + off;
00190     for (i = 0; i < bytes/4; i++)
00191         obuf[i] = c ^ buf[i];
00192 
00193     if (off)
00194         av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
00195 
00196     return off;
00197 }
00198 
00199 
00200 static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
00201     float enc_window[256];
00202     int i;
00203 
00204     /* Generate the mdct window, for details see
00205      * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
00206     for (i=0 ; i<256; i++)
00207         enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
00208 
00209     if (!mdct_window[0])
00210         for (i=0 ; i<256; i++) {
00211             mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
00212             mdct_window[511-i] = mdct_window[i];
00213         }
00214 
00215     /* Initialize the MDCT transform. */
00216     return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
00217 }
00218 
00223 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
00224 {
00225     ATRAC3Context *q = avctx->priv_data;
00226 
00227     av_free(q->pUnits);
00228     av_free(q->decoded_bytes_buffer);
00229     av_freep(&q->outSamples[0]);
00230 
00231     ff_mdct_end(&q->mdct_ctx);
00232 
00233     return 0;
00234 }
00235 
00246 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
00247 {
00248     int   numBits, cnt, code, huffSymb;
00249 
00250     if (selector == 1)
00251         numCodes /= 2;
00252 
00253     if (codingFlag != 0) {
00254         /* constant length coding (CLC) */
00255         numBits = CLCLengthTab[selector];
00256 
00257         if (selector > 1) {
00258             for (cnt = 0; cnt < numCodes; cnt++) {
00259                 if (numBits)
00260                     code = get_sbits(gb, numBits);
00261                 else
00262                     code = 0;
00263                 mantissas[cnt] = code;
00264             }
00265         } else {
00266             for (cnt = 0; cnt < numCodes; cnt++) {
00267                 if (numBits)
00268                     code = get_bits(gb, numBits); //numBits is always 4 in this case
00269                 else
00270                     code = 0;
00271                 mantissas[cnt*2] = seTab_0[code >> 2];
00272                 mantissas[cnt*2+1] = seTab_0[code & 3];
00273             }
00274         }
00275     } else {
00276         /* variable length coding (VLC) */
00277         if (selector != 1) {
00278             for (cnt = 0; cnt < numCodes; cnt++) {
00279                 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
00280                 huffSymb += 1;
00281                 code = huffSymb >> 1;
00282                 if (huffSymb & 1)
00283                     code = -code;
00284                 mantissas[cnt] = code;
00285             }
00286         } else {
00287             for (cnt = 0; cnt < numCodes; cnt++) {
00288                 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
00289                 mantissas[cnt*2] = decTable1[huffSymb*2];
00290                 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
00291             }
00292         }
00293     }
00294 }
00295 
00304 static int decodeSpectrum (GetBitContext *gb, float *pOut)
00305 {
00306     int   numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
00307     int   subband_vlc_index[32], SF_idxs[32];
00308     int   mantissas[128];
00309     float SF;
00310 
00311     numSubbands = get_bits(gb, 5); // number of coded subbands
00312     codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
00313 
00314     /* Get the VLC selector table for the subbands, 0 means not coded. */
00315     for (cnt = 0; cnt <= numSubbands; cnt++)
00316         subband_vlc_index[cnt] = get_bits(gb, 3);
00317 
00318     /* Read the scale factor indexes from the stream. */
00319     for (cnt = 0; cnt <= numSubbands; cnt++) {
00320         if (subband_vlc_index[cnt] != 0)
00321             SF_idxs[cnt] = get_bits(gb, 6);
00322     }
00323 
00324     for (cnt = 0; cnt <= numSubbands; cnt++) {
00325         first = subbandTab[cnt];
00326         last = subbandTab[cnt+1];
00327 
00328         subbWidth = last - first;
00329 
00330         if (subband_vlc_index[cnt] != 0) {
00331             /* Decode spectral coefficients for this subband. */
00332             /* TODO: This can be done faster is several blocks share the
00333              * same VLC selector (subband_vlc_index) */
00334             readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
00335 
00336             /* Decode the scale factor for this subband. */
00337             SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
00338 
00339             /* Inverse quantize the coefficients. */
00340             for (pIn=mantissas ; first<last; first++, pIn++)
00341                 pOut[first] = *pIn * SF;
00342         } else {
00343             /* This subband was not coded, so zero the entire subband. */
00344             memset(pOut+first, 0, subbWidth*sizeof(float));
00345         }
00346     }
00347 
00348     /* Clear the subbands that were not coded. */
00349     first = subbandTab[cnt];
00350     memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
00351     return numSubbands;
00352 }
00353 
00362 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
00363 {
00364     int i,j,k,cnt;
00365     int   components, coding_mode_selector, coding_mode, coded_values_per_component;
00366     int   sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
00367     int   band_flags[4], mantissa[8];
00368     float  *pCoef;
00369     float  scalefactor;
00370     int   component_count = 0;
00371 
00372     components = get_bits(gb,5);
00373 
00374     /* no tonal components */
00375     if (components == 0)
00376         return 0;
00377 
00378     coding_mode_selector = get_bits(gb,2);
00379     if (coding_mode_selector == 2)
00380         return AVERROR_INVALIDDATA;
00381 
00382     coding_mode = coding_mode_selector & 1;
00383 
00384     for (i = 0; i < components; i++) {
00385         for (cnt = 0; cnt <= numBands; cnt++)
00386             band_flags[cnt] = get_bits1(gb);
00387 
00388         coded_values_per_component = get_bits(gb,3);
00389 
00390         quant_step_index = get_bits(gb,3);
00391         if (quant_step_index <= 1)
00392             return AVERROR_INVALIDDATA;
00393 
00394         if (coding_mode_selector == 3)
00395             coding_mode = get_bits1(gb);
00396 
00397         for (j = 0; j < (numBands + 1) * 4; j++) {
00398             if (band_flags[j >> 2] == 0)
00399                 continue;
00400 
00401             coded_components = get_bits(gb,3);
00402 
00403             for (k=0; k<coded_components; k++) {
00404                 sfIndx = get_bits(gb,6);
00405                 if (component_count >= 64)
00406                     return AVERROR_INVALIDDATA;
00407                 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
00408                 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
00409                 coded_values = coded_values_per_component + 1;
00410                 coded_values = FFMIN(max_coded_values,coded_values);
00411 
00412                 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
00413 
00414                 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
00415 
00416                 pComponent[component_count].numCoefs = coded_values;
00417 
00418                 /* inverse quant */
00419                 pCoef = pComponent[component_count].coef;
00420                 for (cnt = 0; cnt < coded_values; cnt++)
00421                     pCoef[cnt] = mantissa[cnt] * scalefactor;
00422 
00423                 component_count++;
00424             }
00425         }
00426     }
00427 
00428     return component_count;
00429 }
00430 
00439 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
00440 {
00441     int   i, cf, numData;
00442     int   *pLevel, *pLoc;
00443 
00444     gain_info   *pGain = pGb->gBlock;
00445 
00446     for (i=0 ; i<=numBands; i++)
00447     {
00448         numData = get_bits(gb,3);
00449         pGain[i].num_gain_data = numData;
00450         pLevel = pGain[i].levcode;
00451         pLoc = pGain[i].loccode;
00452 
00453         for (cf = 0; cf < numData; cf++){
00454             pLevel[cf]= get_bits(gb,4);
00455             pLoc  [cf]= get_bits(gb,5);
00456             if(cf && pLoc[cf] <= pLoc[cf-1])
00457                 return AVERROR_INVALIDDATA;
00458         }
00459     }
00460 
00461     /* Clear the unused blocks. */
00462     for (; i<4 ; i++)
00463         pGain[i].num_gain_data = 0;
00464 
00465     return 0;
00466 }
00467 
00478 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
00479 {
00480     /* gain compensation function */
00481     float  gain1, gain2, gain_inc;
00482     int   cnt, numdata, nsample, startLoc, endLoc;
00483 
00484 
00485     if (pGain2->num_gain_data == 0)
00486         gain1 = 1.0;
00487     else
00488         gain1 = gain_tab1[pGain2->levcode[0]];
00489 
00490     if (pGain1->num_gain_data == 0) {
00491         for (cnt = 0; cnt < 256; cnt++)
00492             pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
00493     } else {
00494         numdata = pGain1->num_gain_data;
00495         pGain1->loccode[numdata] = 32;
00496         pGain1->levcode[numdata] = 4;
00497 
00498         nsample = 0; // current sample = 0
00499 
00500         for (cnt = 0; cnt < numdata; cnt++) {
00501             startLoc = pGain1->loccode[cnt] * 8;
00502             endLoc = startLoc + 8;
00503 
00504             gain2 = gain_tab1[pGain1->levcode[cnt]];
00505             gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
00506 
00507             /* interpolate */
00508             for (; nsample < startLoc; nsample++)
00509                 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
00510 
00511             /* interpolation is done over eight samples */
00512             for (; nsample < endLoc; nsample++) {
00513                 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
00514                 gain2 *= gain_inc;
00515             }
00516         }
00517 
00518         for (; nsample < 256; nsample++)
00519             pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
00520     }
00521 
00522     /* Delay for the overlapping part. */
00523     memcpy(pPrev, &pIn[256], 256*sizeof(float));
00524 }
00525 
00535 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
00536 {
00537     int   cnt, i, lastPos = -1;
00538     float   *pIn, *pOut;
00539 
00540     for (cnt = 0; cnt < numComponents; cnt++){
00541         lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
00542         pIn = pComponent[cnt].coef;
00543         pOut = &(pSpectrum[pComponent[cnt].pos]);
00544 
00545         for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
00546             pOut[i] += pIn[i];
00547     }
00548 
00549     return lastPos;
00550 }
00551 
00552 
00553 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
00554 
00555 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
00556 {
00557     int    i, band, nsample, s1, s2;
00558     float    c1, c2;
00559     float    mc1_l, mc1_r, mc2_l, mc2_r;
00560 
00561     for (i=0,band = 0; band < 4*256; band+=256,i++) {
00562         s1 = pPrevCode[i];
00563         s2 = pCurrCode[i];
00564         nsample = 0;
00565 
00566         if (s1 != s2) {
00567             /* Selector value changed, interpolation needed. */
00568             mc1_l = matrixCoeffs[s1*2];
00569             mc1_r = matrixCoeffs[s1*2+1];
00570             mc2_l = matrixCoeffs[s2*2];
00571             mc2_r = matrixCoeffs[s2*2+1];
00572 
00573             /* Interpolation is done over the first eight samples. */
00574             for(; nsample < 8; nsample++) {
00575                 c1 = su1[band+nsample];
00576                 c2 = su2[band+nsample];
00577                 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
00578                 su1[band+nsample] = c2;
00579                 su2[band+nsample] = c1 * 2.0 - c2;
00580             }
00581         }
00582 
00583         /* Apply the matrix without interpolation. */
00584         switch (s2) {
00585             case 0:     /* M/S decoding */
00586                 for (; nsample < 256; nsample++) {
00587                     c1 = su1[band+nsample];
00588                     c2 = su2[band+nsample];
00589                     su1[band+nsample] = c2 * 2.0;
00590                     su2[band+nsample] = (c1 - c2) * 2.0;
00591                 }
00592                 break;
00593 
00594             case 1:
00595                 for (; nsample < 256; nsample++) {
00596                     c1 = su1[band+nsample];
00597                     c2 = su2[band+nsample];
00598                     su1[band+nsample] = (c1 + c2) * 2.0;
00599                     su2[band+nsample] = c2 * -2.0;
00600                 }
00601                 break;
00602             case 2:
00603             case 3:
00604                 for (; nsample < 256; nsample++) {
00605                     c1 = su1[band+nsample];
00606                     c2 = su2[band+nsample];
00607                     su1[band+nsample] = c1 + c2;
00608                     su2[band+nsample] = c1 - c2;
00609                 }
00610                 break;
00611             default:
00612                 assert(0);
00613         }
00614     }
00615 }
00616 
00617 static void getChannelWeights (int indx, int flag, float ch[2]){
00618 
00619     if (indx == 7) {
00620         ch[0] = 1.0;
00621         ch[1] = 1.0;
00622     } else {
00623         ch[0] = (float)(indx & 7) / 7.0;
00624         ch[1] = sqrt(2 - ch[0]*ch[0]);
00625         if(flag)
00626             FFSWAP(float, ch[0], ch[1]);
00627     }
00628 }
00629 
00630 static void channelWeighting (float *su1, float *su2, int *p3)
00631 {
00632     int   band, nsample;
00633     /* w[x][y] y=0 is left y=1 is right */
00634     float w[2][2];
00635 
00636     if (p3[1] != 7 || p3[3] != 7){
00637         getChannelWeights(p3[1], p3[0], w[0]);
00638         getChannelWeights(p3[3], p3[2], w[1]);
00639 
00640         for(band = 1; band < 4; band++) {
00641             /* scale the channels by the weights */
00642             for(nsample = 0; nsample < 8; nsample++) {
00643                 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
00644                 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
00645             }
00646 
00647             for(; nsample < 256; nsample++) {
00648                 su1[band*256+nsample] *= w[1][0];
00649                 su2[band*256+nsample] *= w[1][1];
00650             }
00651         }
00652     }
00653 }
00654 
00655 
00667 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
00668 {
00669     int   band, result=0, numSubbands, lastTonal, numBands;
00670 
00671     if (codingMode == JOINT_STEREO && channelNum == 1) {
00672         if (get_bits(gb,2) != 3) {
00673             av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
00674             return AVERROR_INVALIDDATA;
00675         }
00676     } else {
00677         if (get_bits(gb,6) != 0x28) {
00678             av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
00679             return AVERROR_INVALIDDATA;
00680         }
00681     }
00682 
00683     /* number of coded QMF bands */
00684     pSnd->bandsCoded = get_bits(gb,2);
00685 
00686     result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
00687     if (result) return result;
00688 
00689     pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
00690     if (pSnd->numComponents == -1) return -1;
00691 
00692     numSubbands = decodeSpectrum (gb, pSnd->spectrum);
00693 
00694     /* Merge the decoded spectrum and tonal components. */
00695     lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
00696 
00697 
00698     /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
00699     numBands = (subbandTab[numSubbands] - 1) >> 8;
00700     if (lastTonal >= 0)
00701         numBands = FFMAX((lastTonal + 256) >> 8, numBands);
00702 
00703 
00704     /* Reconstruct time domain samples. */
00705     for (band=0; band<4; band++) {
00706         /* Perform the IMDCT step without overlapping. */
00707         if (band <= numBands) {
00708             IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
00709         } else
00710             memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
00711 
00712         /* gain compensation and overlapping */
00713         gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
00714                                  &pOut[band * 256],
00715                                  &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
00716                                  &pSnd->gainBlock[    pSnd->gcBlkSwitch].gBlock[band]);
00717     }
00718 
00719     /* Swap the gain control buffers for the next frame. */
00720     pSnd->gcBlkSwitch ^= 1;
00721 
00722     return 0;
00723 }
00724 
00732 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
00733                        float **out_samples)
00734 {
00735     int   result, i;
00736     float   *p1, *p2, *p3, *p4;
00737     uint8_t *ptr1;
00738 
00739     if (q->codingMode == JOINT_STEREO) {
00740 
00741         /* channel coupling mode */
00742         /* decode Sound Unit 1 */
00743         init_get_bits(&q->gb,databuf,q->bits_per_frame);
00744 
00745         result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
00746         if (result != 0)
00747             return result;
00748 
00749         /* Framedata of the su2 in the joint-stereo mode is encoded in
00750          * reverse byte order so we need to swap it first. */
00751         if (databuf == q->decoded_bytes_buffer) {
00752             uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
00753             ptr1 = q->decoded_bytes_buffer;
00754             for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
00755                 FFSWAP(uint8_t,*ptr1,*ptr2);
00756             }
00757         } else {
00758             const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
00759             for (i = 0; i < q->bytes_per_frame; i++)
00760                 q->decoded_bytes_buffer[i] = *ptr2--;
00761         }
00762 
00763         /* Skip the sync codes (0xF8). */
00764         ptr1 = q->decoded_bytes_buffer;
00765         for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
00766             if (i >= q->bytes_per_frame)
00767                 return AVERROR_INVALIDDATA;
00768         }
00769 
00770 
00771         /* set the bitstream reader at the start of the second Sound Unit*/
00772         init_get_bits(&q->gb,ptr1,q->bits_per_frame);
00773 
00774         /* Fill the Weighting coeffs delay buffer */
00775         memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
00776         q->weighting_delay[4] = get_bits1(&q->gb);
00777         q->weighting_delay[5] = get_bits(&q->gb,3);
00778 
00779         for (i = 0; i < 4; i++) {
00780             q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
00781             q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
00782             q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
00783         }
00784 
00785         /* Decode Sound Unit 2. */
00786         result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
00787         if (result != 0)
00788             return result;
00789 
00790         /* Reconstruct the channel coefficients. */
00791         reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
00792 
00793         channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
00794 
00795     } else {
00796         /* normal stereo mode or mono */
00797         /* Decode the channel sound units. */
00798         for (i=0 ; i<q->channels ; i++) {
00799 
00800             /* Set the bitstream reader at the start of a channel sound unit. */
00801             init_get_bits(&q->gb,
00802                           databuf + i * q->bytes_per_frame / q->channels,
00803                           q->bits_per_frame / q->channels);
00804 
00805             result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
00806             if (result != 0)
00807                 return result;
00808         }
00809     }
00810 
00811     /* Apply the iQMF synthesis filter. */
00812     for (i=0 ; i<q->channels ; i++) {
00813         p1 = out_samples[i];
00814         p2= p1+256;
00815         p3= p2+256;
00816         p4= p3+256;
00817         atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
00818         atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
00819         atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
00820     }
00821 
00822     return 0;
00823 }
00824 
00825 
00832 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
00833                                int *got_frame_ptr, AVPacket *avpkt)
00834 {
00835     const uint8_t *buf = avpkt->data;
00836     int buf_size = avpkt->size;
00837     ATRAC3Context *q = avctx->priv_data;
00838     int result;
00839     const uint8_t* databuf;
00840     float   *samples_flt;
00841     int16_t *samples_s16;
00842 
00843     if (buf_size < avctx->block_align) {
00844         av_log(avctx, AV_LOG_ERROR,
00845                "Frame too small (%d bytes). Truncated file?\n", buf_size);
00846         return AVERROR_INVALIDDATA;
00847     }
00848 
00849     /* get output buffer */
00850     q->frame.nb_samples = SAMPLES_PER_FRAME;
00851     if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
00852         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
00853         return result;
00854     }
00855     samples_flt = (float   *)q->frame.data[0];
00856     samples_s16 = (int16_t *)q->frame.data[0];
00857 
00858     /* Check if we need to descramble and what buffer to pass on. */
00859     if (q->scrambled_stream) {
00860         decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
00861         databuf = q->decoded_bytes_buffer;
00862     } else {
00863         databuf = buf;
00864     }
00865 
00866     if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
00867         result = decodeFrame(q, databuf, &samples_flt);
00868     else
00869         result = decodeFrame(q, databuf, q->outSamples);
00870 
00871     if (result != 0) {
00872         av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
00873         return result;
00874     }
00875 
00876     /* interleave */
00877     if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
00878         q->fmt_conv.float_interleave(samples_flt,
00879                                      (const float **)q->outSamples,
00880                                      SAMPLES_PER_FRAME, 2);
00881     } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
00882         q->fmt_conv.float_to_int16_interleave(samples_s16,
00883                                               (const float **)q->outSamples,
00884                                               SAMPLES_PER_FRAME, q->channels);
00885     }
00886 
00887     *got_frame_ptr   = 1;
00888     *(AVFrame *)data = q->frame;
00889 
00890     return avctx->block_align;
00891 }
00892 
00893 
00900 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
00901 {
00902     int i, ret;
00903     const uint8_t *edata_ptr = avctx->extradata;
00904     ATRAC3Context *q = avctx->priv_data;
00905     static VLC_TYPE atrac3_vlc_table[4096][2];
00906     static int vlcs_initialized = 0;
00907 
00908     /* Take data from the AVCodecContext (RM container). */
00909     q->sample_rate = avctx->sample_rate;
00910     q->channels = avctx->channels;
00911     q->bit_rate = avctx->bit_rate;
00912     q->bits_per_frame = avctx->block_align * 8;
00913     q->bytes_per_frame = avctx->block_align;
00914 
00915     /* Take care of the codec-specific extradata. */
00916     if (avctx->extradata_size == 14) {
00917         /* Parse the extradata, WAV format */
00918         av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr));  //Unknown value always 1
00919         q->samples_per_channel = bytestream_get_le32(&edata_ptr);
00920         q->codingMode = bytestream_get_le16(&edata_ptr);
00921         av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr));  //Dupe of coding mode
00922         q->frame_factor = bytestream_get_le16(&edata_ptr);  //Unknown always 1
00923         av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr));  //Unknown always 0
00924 
00925         /* setup */
00926         q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
00927         q->atrac3version = 4;
00928         q->delay = 0x88E;
00929         if (q->codingMode)
00930             q->codingMode = JOINT_STEREO;
00931         else
00932             q->codingMode = STEREO;
00933 
00934         q->scrambled_stream = 0;
00935 
00936         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)) {
00937         } else {
00938             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);
00939             return AVERROR_INVALIDDATA;
00940         }
00941 
00942     } else if (avctx->extradata_size == 10) {
00943         /* Parse the extradata, RM format. */
00944         q->atrac3version = bytestream_get_be32(&edata_ptr);
00945         q->samples_per_frame = bytestream_get_be16(&edata_ptr);
00946         q->delay = bytestream_get_be16(&edata_ptr);
00947         q->codingMode = bytestream_get_be16(&edata_ptr);
00948 
00949         q->samples_per_channel = q->samples_per_frame / q->channels;
00950         q->scrambled_stream = 1;
00951 
00952     } else {
00953         av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
00954     }
00955     /* Check the extradata. */
00956 
00957     if (q->atrac3version != 4) {
00958         av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
00959         return AVERROR_INVALIDDATA;
00960     }
00961 
00962     if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
00963         av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
00964         return AVERROR_INVALIDDATA;
00965     }
00966 
00967     if (q->delay != 0x88E) {
00968         av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
00969         return AVERROR_INVALIDDATA;
00970     }
00971 
00972     if (q->codingMode == STEREO) {
00973         av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
00974     } else if (q->codingMode == JOINT_STEREO) {
00975         av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
00976     } else {
00977         av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
00978         return AVERROR_INVALIDDATA;
00979     }
00980 
00981     if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
00982         av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
00983         return AVERROR(EINVAL);
00984     }
00985 
00986 
00987     if(avctx->block_align >= UINT_MAX/2)
00988         return AVERROR(EINVAL);
00989 
00990     /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
00991      * this is for the bitstream reader. */
00992     if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE)))  == NULL)
00993         return AVERROR(ENOMEM);
00994 
00995 
00996     /* Initialize the VLC tables. */
00997     if (!vlcs_initialized) {
00998         for (i=0 ; i<7 ; i++) {
00999             spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
01000             spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
01001             init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
01002                 huff_bits[i], 1, 1,
01003                 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
01004         }
01005         vlcs_initialized = 1;
01006     }
01007 
01008     if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
01009         avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
01010     else
01011         avctx->sample_fmt = AV_SAMPLE_FMT_S16;
01012 
01013     if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
01014         av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
01015         av_freep(&q->decoded_bytes_buffer);
01016         return ret;
01017     }
01018 
01019     atrac_generate_tables();
01020 
01021     /* Generate gain tables. */
01022     for (i=0 ; i<16 ; i++)
01023         gain_tab1[i] = powf (2.0, (4 - i));
01024 
01025     for (i=-15 ; i<16 ; i++)
01026         gain_tab2[i+15] = powf (2.0, i * -0.125);
01027 
01028     /* init the joint-stereo decoding data */
01029     q->weighting_delay[0] = 0;
01030     q->weighting_delay[1] = 7;
01031     q->weighting_delay[2] = 0;
01032     q->weighting_delay[3] = 7;
01033     q->weighting_delay[4] = 0;
01034     q->weighting_delay[5] = 7;
01035 
01036     for (i=0; i<4; i++) {
01037         q->matrix_coeff_index_prev[i] = 3;
01038         q->matrix_coeff_index_now[i] = 3;
01039         q->matrix_coeff_index_next[i] = 3;
01040     }
01041 
01042     dsputil_init(&dsp, avctx);
01043     ff_fmt_convert_init(&q->fmt_conv, avctx);
01044 
01045     q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
01046     if (!q->pUnits) {
01047         atrac3_decode_close(avctx);
01048         return AVERROR(ENOMEM);
01049     }
01050 
01051     if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
01052         q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
01053         q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
01054         if (!q->outSamples[0]) {
01055             atrac3_decode_close(avctx);
01056             return AVERROR(ENOMEM);
01057         }
01058     }
01059 
01060     avcodec_get_frame_defaults(&q->frame);
01061     avctx->coded_frame = &q->frame;
01062 
01063     return 0;
01064 }
01065 
01066 
01067 AVCodec ff_atrac3_decoder =
01068 {
01069     .name = "atrac3",
01070     .type = AVMEDIA_TYPE_AUDIO,
01071     .id = CODEC_ID_ATRAC3,
01072     .priv_data_size = sizeof(ATRAC3Context),
01073     .init = atrac3_decode_init,
01074     .close = atrac3_decode_close,
01075     .decode = atrac3_decode_frame,
01076     .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
01077     .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
01078 };