Libav 0.7.1
|
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 Libav. 00007 * 00008 * Libav 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 * Libav 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 Libav; 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 00045 #include "atrac.h" 00046 #include "atrac3data.h" 00047 00048 #define JOINT_STEREO 0x12 00049 #define STEREO 0x2 00050 00051 00052 /* These structures are needed to store the parsed gain control data. */ 00053 typedef struct { 00054 int num_gain_data; 00055 int levcode[8]; 00056 int loccode[8]; 00057 } gain_info; 00058 00059 typedef struct { 00060 gain_info gBlock[4]; 00061 } gain_block; 00062 00063 typedef struct { 00064 int pos; 00065 int numCoefs; 00066 float coef[8]; 00067 } tonal_component; 00068 00069 typedef struct { 00070 int bandsCoded; 00071 int numComponents; 00072 tonal_component components[64]; 00073 float prevFrame[1024]; 00074 int gcBlkSwitch; 00075 gain_block gainBlock[2]; 00076 00077 DECLARE_ALIGNED(32, float, spectrum)[1024]; 00078 DECLARE_ALIGNED(32, float, IMDCT_buf)[1024]; 00079 00080 float delayBuf1[46]; 00081 float delayBuf2[46]; 00082 float delayBuf3[46]; 00083 } channel_unit; 00084 00085 typedef struct { 00086 GetBitContext gb; 00088 00089 int channels; 00090 int codingMode; 00091 int bit_rate; 00092 int sample_rate; 00093 int samples_per_channel; 00094 int samples_per_frame; 00095 00096 int bits_per_frame; 00097 int bytes_per_frame; 00098 int pBs; 00099 channel_unit* pUnits; 00101 00102 00103 int matrix_coeff_index_prev[4]; 00104 int matrix_coeff_index_now[4]; 00105 int matrix_coeff_index_next[4]; 00106 int weighting_delay[6]; 00108 00109 00110 float outSamples[2048]; 00111 uint8_t* decoded_bytes_buffer; 00112 float tempBuf[1070]; 00114 00115 00116 int atrac3version; 00117 int delay; 00118 int scrambled_stream; 00119 int frame_factor; 00121 00122 FFTContext mdct_ctx; 00123 } ATRAC3Context; 00124 00125 static DECLARE_ALIGNED(32, float, mdct_window)[512]; 00126 static VLC spectral_coeff_tab[7]; 00127 static float gain_tab1[16]; 00128 static float gain_tab2[31]; 00129 static DSPContext dsp; 00130 00131 00141 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band) 00142 { 00143 int i; 00144 00145 if (odd_band) { 00155 for (i=0; i<128; i++) 00156 FFSWAP(float, pInput[i], pInput[255-i]); 00157 } 00158 00159 q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput); 00160 00161 /* Perform windowing on the output. */ 00162 dsp.vector_fmul(pOutput, pOutput, mdct_window, 512); 00163 00164 } 00165 00166 00175 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){ 00176 int i, off; 00177 uint32_t c; 00178 const uint32_t* buf; 00179 uint32_t* obuf = (uint32_t*) out; 00180 00181 off = (intptr_t)inbuffer & 3; 00182 buf = (const uint32_t*) (inbuffer - off); 00183 c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8)))); 00184 bytes += 3 + off; 00185 for (i = 0; i < bytes/4; i++) 00186 obuf[i] = c ^ buf[i]; 00187 00188 if (off) 00189 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off); 00190 00191 return off; 00192 } 00193 00194 00195 static av_cold void init_atrac3_transforms(ATRAC3Context *q) { 00196 float enc_window[256]; 00197 int i; 00198 00199 /* Generate the mdct window, for details see 00200 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */ 00201 for (i=0 ; i<256; i++) 00202 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5; 00203 00204 if (!mdct_window[0]) 00205 for (i=0 ; i<256; i++) { 00206 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]); 00207 mdct_window[511-i] = mdct_window[i]; 00208 } 00209 00210 /* Initialize the MDCT transform. */ 00211 ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0); 00212 } 00213 00218 static av_cold int atrac3_decode_close(AVCodecContext *avctx) 00219 { 00220 ATRAC3Context *q = avctx->priv_data; 00221 00222 av_free(q->pUnits); 00223 av_free(q->decoded_bytes_buffer); 00224 ff_mdct_end(&q->mdct_ctx); 00225 00226 return 0; 00227 } 00228 00239 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes) 00240 { 00241 int numBits, cnt, code, huffSymb; 00242 00243 if (selector == 1) 00244 numCodes /= 2; 00245 00246 if (codingFlag != 0) { 00247 /* constant length coding (CLC) */ 00248 numBits = CLCLengthTab[selector]; 00249 00250 if (selector > 1) { 00251 for (cnt = 0; cnt < numCodes; cnt++) { 00252 if (numBits) 00253 code = get_sbits(gb, numBits); 00254 else 00255 code = 0; 00256 mantissas[cnt] = code; 00257 } 00258 } else { 00259 for (cnt = 0; cnt < numCodes; cnt++) { 00260 if (numBits) 00261 code = get_bits(gb, numBits); //numBits is always 4 in this case 00262 else 00263 code = 0; 00264 mantissas[cnt*2] = seTab_0[code >> 2]; 00265 mantissas[cnt*2+1] = seTab_0[code & 3]; 00266 } 00267 } 00268 } else { 00269 /* variable length coding (VLC) */ 00270 if (selector != 1) { 00271 for (cnt = 0; cnt < numCodes; cnt++) { 00272 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3); 00273 huffSymb += 1; 00274 code = huffSymb >> 1; 00275 if (huffSymb & 1) 00276 code = -code; 00277 mantissas[cnt] = code; 00278 } 00279 } else { 00280 for (cnt = 0; cnt < numCodes; cnt++) { 00281 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3); 00282 mantissas[cnt*2] = decTable1[huffSymb*2]; 00283 mantissas[cnt*2+1] = decTable1[huffSymb*2+1]; 00284 } 00285 } 00286 } 00287 } 00288 00297 static int decodeSpectrum (GetBitContext *gb, float *pOut) 00298 { 00299 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn; 00300 int subband_vlc_index[32], SF_idxs[32]; 00301 int mantissas[128]; 00302 float SF; 00303 00304 numSubbands = get_bits(gb, 5); // number of coded subbands 00305 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC 00306 00307 /* Get the VLC selector table for the subbands, 0 means not coded. */ 00308 for (cnt = 0; cnt <= numSubbands; cnt++) 00309 subband_vlc_index[cnt] = get_bits(gb, 3); 00310 00311 /* Read the scale factor indexes from the stream. */ 00312 for (cnt = 0; cnt <= numSubbands; cnt++) { 00313 if (subband_vlc_index[cnt] != 0) 00314 SF_idxs[cnt] = get_bits(gb, 6); 00315 } 00316 00317 for (cnt = 0; cnt <= numSubbands; cnt++) { 00318 first = subbandTab[cnt]; 00319 last = subbandTab[cnt+1]; 00320 00321 subbWidth = last - first; 00322 00323 if (subband_vlc_index[cnt] != 0) { 00324 /* Decode spectral coefficients for this subband. */ 00325 /* TODO: This can be done faster is several blocks share the 00326 * same VLC selector (subband_vlc_index) */ 00327 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth); 00328 00329 /* Decode the scale factor for this subband. */ 00330 SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]]; 00331 00332 /* Inverse quantize the coefficients. */ 00333 for (pIn=mantissas ; first<last; first++, pIn++) 00334 pOut[first] = *pIn * SF; 00335 } else { 00336 /* This subband was not coded, so zero the entire subband. */ 00337 memset(pOut+first, 0, subbWidth*sizeof(float)); 00338 } 00339 } 00340 00341 /* Clear the subbands that were not coded. */ 00342 first = subbandTab[cnt]; 00343 memset(pOut+first, 0, (1024 - first) * sizeof(float)); 00344 return numSubbands; 00345 } 00346 00355 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands) 00356 { 00357 int i,j,k,cnt; 00358 int components, coding_mode_selector, coding_mode, coded_values_per_component; 00359 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components; 00360 int band_flags[4], mantissa[8]; 00361 float *pCoef; 00362 float scalefactor; 00363 int component_count = 0; 00364 00365 components = get_bits(gb,5); 00366 00367 /* no tonal components */ 00368 if (components == 0) 00369 return 0; 00370 00371 coding_mode_selector = get_bits(gb,2); 00372 if (coding_mode_selector == 2) 00373 return -1; 00374 00375 coding_mode = coding_mode_selector & 1; 00376 00377 for (i = 0; i < components; i++) { 00378 for (cnt = 0; cnt <= numBands; cnt++) 00379 band_flags[cnt] = get_bits1(gb); 00380 00381 coded_values_per_component = get_bits(gb,3); 00382 00383 quant_step_index = get_bits(gb,3); 00384 if (quant_step_index <= 1) 00385 return -1; 00386 00387 if (coding_mode_selector == 3) 00388 coding_mode = get_bits1(gb); 00389 00390 for (j = 0; j < (numBands + 1) * 4; j++) { 00391 if (band_flags[j >> 2] == 0) 00392 continue; 00393 00394 coded_components = get_bits(gb,3); 00395 00396 for (k=0; k<coded_components; k++) { 00397 sfIndx = get_bits(gb,6); 00398 pComponent[component_count].pos = j * 64 + (get_bits(gb,6)); 00399 max_coded_values = 1024 - pComponent[component_count].pos; 00400 coded_values = coded_values_per_component + 1; 00401 coded_values = FFMIN(max_coded_values,coded_values); 00402 00403 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index]; 00404 00405 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values); 00406 00407 pComponent[component_count].numCoefs = coded_values; 00408 00409 /* inverse quant */ 00410 pCoef = pComponent[component_count].coef; 00411 for (cnt = 0; cnt < coded_values; cnt++) 00412 pCoef[cnt] = mantissa[cnt] * scalefactor; 00413 00414 component_count++; 00415 } 00416 } 00417 } 00418 00419 return component_count; 00420 } 00421 00430 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands) 00431 { 00432 int i, cf, numData; 00433 int *pLevel, *pLoc; 00434 00435 gain_info *pGain = pGb->gBlock; 00436 00437 for (i=0 ; i<=numBands; i++) 00438 { 00439 numData = get_bits(gb,3); 00440 pGain[i].num_gain_data = numData; 00441 pLevel = pGain[i].levcode; 00442 pLoc = pGain[i].loccode; 00443 00444 for (cf = 0; cf < numData; cf++){ 00445 pLevel[cf]= get_bits(gb,4); 00446 pLoc [cf]= get_bits(gb,5); 00447 if(cf && pLoc[cf] <= pLoc[cf-1]) 00448 return -1; 00449 } 00450 } 00451 00452 /* Clear the unused blocks. */ 00453 for (; i<4 ; i++) 00454 pGain[i].num_gain_data = 0; 00455 00456 return 0; 00457 } 00458 00469 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2) 00470 { 00471 /* gain compensation function */ 00472 float gain1, gain2, gain_inc; 00473 int cnt, numdata, nsample, startLoc, endLoc; 00474 00475 00476 if (pGain2->num_gain_data == 0) 00477 gain1 = 1.0; 00478 else 00479 gain1 = gain_tab1[pGain2->levcode[0]]; 00480 00481 if (pGain1->num_gain_data == 0) { 00482 for (cnt = 0; cnt < 256; cnt++) 00483 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt]; 00484 } else { 00485 numdata = pGain1->num_gain_data; 00486 pGain1->loccode[numdata] = 32; 00487 pGain1->levcode[numdata] = 4; 00488 00489 nsample = 0; // current sample = 0 00490 00491 for (cnt = 0; cnt < numdata; cnt++) { 00492 startLoc = pGain1->loccode[cnt] * 8; 00493 endLoc = startLoc + 8; 00494 00495 gain2 = gain_tab1[pGain1->levcode[cnt]]; 00496 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15]; 00497 00498 /* interpolate */ 00499 for (; nsample < startLoc; nsample++) 00500 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2; 00501 00502 /* interpolation is done over eight samples */ 00503 for (; nsample < endLoc; nsample++) { 00504 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2; 00505 gain2 *= gain_inc; 00506 } 00507 } 00508 00509 for (; nsample < 256; nsample++) 00510 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample]; 00511 } 00512 00513 /* Delay for the overlapping part. */ 00514 memcpy(pPrev, &pIn[256], 256*sizeof(float)); 00515 } 00516 00526 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent) 00527 { 00528 int cnt, i, lastPos = -1; 00529 float *pIn, *pOut; 00530 00531 for (cnt = 0; cnt < numComponents; cnt++){ 00532 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos); 00533 pIn = pComponent[cnt].coef; 00534 pOut = &(pSpectrum[pComponent[cnt].pos]); 00535 00536 for (i=0 ; i<pComponent[cnt].numCoefs ; i++) 00537 pOut[i] += pIn[i]; 00538 } 00539 00540 return lastPos; 00541 } 00542 00543 00544 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old))) 00545 00546 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode) 00547 { 00548 int i, band, nsample, s1, s2; 00549 float c1, c2; 00550 float mc1_l, mc1_r, mc2_l, mc2_r; 00551 00552 for (i=0,band = 0; band < 4*256; band+=256,i++) { 00553 s1 = pPrevCode[i]; 00554 s2 = pCurrCode[i]; 00555 nsample = 0; 00556 00557 if (s1 != s2) { 00558 /* Selector value changed, interpolation needed. */ 00559 mc1_l = matrixCoeffs[s1*2]; 00560 mc1_r = matrixCoeffs[s1*2+1]; 00561 mc2_l = matrixCoeffs[s2*2]; 00562 mc2_r = matrixCoeffs[s2*2+1]; 00563 00564 /* Interpolation is done over the first eight samples. */ 00565 for(; nsample < 8; nsample++) { 00566 c1 = su1[band+nsample]; 00567 c2 = su2[band+nsample]; 00568 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample); 00569 su1[band+nsample] = c2; 00570 su2[band+nsample] = c1 * 2.0 - c2; 00571 } 00572 } 00573 00574 /* Apply the matrix without interpolation. */ 00575 switch (s2) { 00576 case 0: /* M/S decoding */ 00577 for (; nsample < 256; nsample++) { 00578 c1 = su1[band+nsample]; 00579 c2 = su2[band+nsample]; 00580 su1[band+nsample] = c2 * 2.0; 00581 su2[band+nsample] = (c1 - c2) * 2.0; 00582 } 00583 break; 00584 00585 case 1: 00586 for (; nsample < 256; nsample++) { 00587 c1 = su1[band+nsample]; 00588 c2 = su2[band+nsample]; 00589 su1[band+nsample] = (c1 + c2) * 2.0; 00590 su2[band+nsample] = c2 * -2.0; 00591 } 00592 break; 00593 case 2: 00594 case 3: 00595 for (; nsample < 256; nsample++) { 00596 c1 = su1[band+nsample]; 00597 c2 = su2[band+nsample]; 00598 su1[band+nsample] = c1 + c2; 00599 su2[band+nsample] = c1 - c2; 00600 } 00601 break; 00602 default: 00603 assert(0); 00604 } 00605 } 00606 } 00607 00608 static void getChannelWeights (int indx, int flag, float ch[2]){ 00609 00610 if (indx == 7) { 00611 ch[0] = 1.0; 00612 ch[1] = 1.0; 00613 } else { 00614 ch[0] = (float)(indx & 7) / 7.0; 00615 ch[1] = sqrt(2 - ch[0]*ch[0]); 00616 if(flag) 00617 FFSWAP(float, ch[0], ch[1]); 00618 } 00619 } 00620 00621 static void channelWeighting (float *su1, float *su2, int *p3) 00622 { 00623 int band, nsample; 00624 /* w[x][y] y=0 is left y=1 is right */ 00625 float w[2][2]; 00626 00627 if (p3[1] != 7 || p3[3] != 7){ 00628 getChannelWeights(p3[1], p3[0], w[0]); 00629 getChannelWeights(p3[3], p3[2], w[1]); 00630 00631 for(band = 1; band < 4; band++) { 00632 /* scale the channels by the weights */ 00633 for(nsample = 0; nsample < 8; nsample++) { 00634 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample); 00635 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample); 00636 } 00637 00638 for(; nsample < 256; nsample++) { 00639 su1[band*256+nsample] *= w[1][0]; 00640 su2[band*256+nsample] *= w[1][1]; 00641 } 00642 } 00643 } 00644 } 00645 00646 00658 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode) 00659 { 00660 int band, result=0, numSubbands, lastTonal, numBands; 00661 00662 if (codingMode == JOINT_STEREO && channelNum == 1) { 00663 if (get_bits(gb,2) != 3) { 00664 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n"); 00665 return -1; 00666 } 00667 } else { 00668 if (get_bits(gb,6) != 0x28) { 00669 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n"); 00670 return -1; 00671 } 00672 } 00673 00674 /* number of coded QMF bands */ 00675 pSnd->bandsCoded = get_bits(gb,2); 00676 00677 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded); 00678 if (result) return result; 00679 00680 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded); 00681 if (pSnd->numComponents == -1) return -1; 00682 00683 numSubbands = decodeSpectrum (gb, pSnd->spectrum); 00684 00685 /* Merge the decoded spectrum and tonal components. */ 00686 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components); 00687 00688 00689 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */ 00690 numBands = (subbandTab[numSubbands] - 1) >> 8; 00691 if (lastTonal >= 0) 00692 numBands = FFMAX((lastTonal + 256) >> 8, numBands); 00693 00694 00695 /* Reconstruct time domain samples. */ 00696 for (band=0; band<4; band++) { 00697 /* Perform the IMDCT step without overlapping. */ 00698 if (band <= numBands) { 00699 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1); 00700 } else 00701 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float)); 00702 00703 /* gain compensation and overlapping */ 00704 gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]), 00705 &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]), 00706 &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band])); 00707 } 00708 00709 /* Swap the gain control buffers for the next frame. */ 00710 pSnd->gcBlkSwitch ^= 1; 00711 00712 return 0; 00713 } 00714 00722 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf) 00723 { 00724 int result, i; 00725 float *p1, *p2, *p3, *p4; 00726 uint8_t *ptr1; 00727 00728 if (q->codingMode == JOINT_STEREO) { 00729 00730 /* channel coupling mode */ 00731 /* decode Sound Unit 1 */ 00732 init_get_bits(&q->gb,databuf,q->bits_per_frame); 00733 00734 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO); 00735 if (result != 0) 00736 return (result); 00737 00738 /* Framedata of the su2 in the joint-stereo mode is encoded in 00739 * reverse byte order so we need to swap it first. */ 00740 if (databuf == q->decoded_bytes_buffer) { 00741 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1; 00742 ptr1 = q->decoded_bytes_buffer; 00743 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) { 00744 FFSWAP(uint8_t,*ptr1,*ptr2); 00745 } 00746 } else { 00747 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1; 00748 for (i = 0; i < q->bytes_per_frame; i++) 00749 q->decoded_bytes_buffer[i] = *ptr2--; 00750 } 00751 00752 /* Skip the sync codes (0xF8). */ 00753 ptr1 = q->decoded_bytes_buffer; 00754 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) { 00755 if (i >= q->bytes_per_frame) 00756 return -1; 00757 } 00758 00759 00760 /* set the bitstream reader at the start of the second Sound Unit*/ 00761 init_get_bits(&q->gb,ptr1,q->bits_per_frame); 00762 00763 /* Fill the Weighting coeffs delay buffer */ 00764 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int)); 00765 q->weighting_delay[4] = get_bits1(&q->gb); 00766 q->weighting_delay[5] = get_bits(&q->gb,3); 00767 00768 for (i = 0; i < 4; i++) { 00769 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i]; 00770 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i]; 00771 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2); 00772 } 00773 00774 /* Decode Sound Unit 2. */ 00775 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO); 00776 if (result != 0) 00777 return (result); 00778 00779 /* Reconstruct the channel coefficients. */ 00780 reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now); 00781 00782 channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay); 00783 00784 } else { 00785 /* normal stereo mode or mono */ 00786 /* Decode the channel sound units. */ 00787 for (i=0 ; i<q->channels ; i++) { 00788 00789 /* Set the bitstream reader at the start of a channel sound unit. */ 00790 init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels); 00791 00792 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode); 00793 if (result != 0) 00794 return (result); 00795 } 00796 } 00797 00798 /* Apply the iQMF synthesis filter. */ 00799 p1= q->outSamples; 00800 for (i=0 ; i<q->channels ; i++) { 00801 p2= p1+256; 00802 p3= p2+256; 00803 p4= p3+256; 00804 atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf); 00805 atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf); 00806 atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf); 00807 p1 +=1024; 00808 } 00809 00810 return 0; 00811 } 00812 00813 00820 static int atrac3_decode_frame(AVCodecContext *avctx, 00821 void *data, int *data_size, 00822 AVPacket *avpkt) { 00823 const uint8_t *buf = avpkt->data; 00824 int buf_size = avpkt->size; 00825 ATRAC3Context *q = avctx->priv_data; 00826 int result = 0, i; 00827 const uint8_t* databuf; 00828 int16_t* samples = data; 00829 00830 if (buf_size < avctx->block_align) { 00831 av_log(avctx, AV_LOG_ERROR, 00832 "Frame too small (%d bytes). Truncated file?\n", buf_size); 00833 *data_size = 0; 00834 return buf_size; 00835 } 00836 00837 /* Check if we need to descramble and what buffer to pass on. */ 00838 if (q->scrambled_stream) { 00839 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align); 00840 databuf = q->decoded_bytes_buffer; 00841 } else { 00842 databuf = buf; 00843 } 00844 00845 result = decodeFrame(q, databuf); 00846 00847 if (result != 0) { 00848 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n"); 00849 return -1; 00850 } 00851 00852 if (q->channels == 1) { 00853 /* mono */ 00854 for (i = 0; i<1024; i++) 00855 samples[i] = av_clip_int16(round(q->outSamples[i])); 00856 *data_size = 1024 * sizeof(int16_t); 00857 } else { 00858 /* stereo */ 00859 for (i = 0; i < 1024; i++) { 00860 samples[i*2] = av_clip_int16(round(q->outSamples[i])); 00861 samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i])); 00862 } 00863 *data_size = 2048 * sizeof(int16_t); 00864 } 00865 00866 return avctx->block_align; 00867 } 00868 00869 00876 static av_cold int atrac3_decode_init(AVCodecContext *avctx) 00877 { 00878 int i; 00879 const uint8_t *edata_ptr = avctx->extradata; 00880 ATRAC3Context *q = avctx->priv_data; 00881 static VLC_TYPE atrac3_vlc_table[4096][2]; 00882 static int vlcs_initialized = 0; 00883 00884 /* Take data from the AVCodecContext (RM container). */ 00885 q->sample_rate = avctx->sample_rate; 00886 q->channels = avctx->channels; 00887 q->bit_rate = avctx->bit_rate; 00888 q->bits_per_frame = avctx->block_align * 8; 00889 q->bytes_per_frame = avctx->block_align; 00890 00891 /* Take care of the codec-specific extradata. */ 00892 if (avctx->extradata_size == 14) { 00893 /* Parse the extradata, WAV format */ 00894 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1 00895 q->samples_per_channel = bytestream_get_le32(&edata_ptr); 00896 q->codingMode = bytestream_get_le16(&edata_ptr); 00897 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode 00898 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1 00899 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0 00900 00901 /* setup */ 00902 q->samples_per_frame = 1024 * q->channels; 00903 q->atrac3version = 4; 00904 q->delay = 0x88E; 00905 if (q->codingMode) 00906 q->codingMode = JOINT_STEREO; 00907 else 00908 q->codingMode = STEREO; 00909 00910 q->scrambled_stream = 0; 00911 00912 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)) { 00913 } else { 00914 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); 00915 return -1; 00916 } 00917 00918 } else if (avctx->extradata_size == 10) { 00919 /* Parse the extradata, RM format. */ 00920 q->atrac3version = bytestream_get_be32(&edata_ptr); 00921 q->samples_per_frame = bytestream_get_be16(&edata_ptr); 00922 q->delay = bytestream_get_be16(&edata_ptr); 00923 q->codingMode = bytestream_get_be16(&edata_ptr); 00924 00925 q->samples_per_channel = q->samples_per_frame / q->channels; 00926 q->scrambled_stream = 1; 00927 00928 } else { 00929 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size); 00930 } 00931 /* Check the extradata. */ 00932 00933 if (q->atrac3version != 4) { 00934 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version); 00935 return -1; 00936 } 00937 00938 if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) { 00939 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame); 00940 return -1; 00941 } 00942 00943 if (q->delay != 0x88E) { 00944 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay); 00945 return -1; 00946 } 00947 00948 if (q->codingMode == STEREO) { 00949 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n"); 00950 } else if (q->codingMode == JOINT_STEREO) { 00951 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n"); 00952 } else { 00953 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode); 00954 return -1; 00955 } 00956 00957 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) { 00958 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n"); 00959 return -1; 00960 } 00961 00962 00963 if(avctx->block_align >= UINT_MAX/2) 00964 return -1; 00965 00966 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE, 00967 * this is for the bitstream reader. */ 00968 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL) 00969 return AVERROR(ENOMEM); 00970 00971 00972 /* Initialize the VLC tables. */ 00973 if (!vlcs_initialized) { 00974 for (i=0 ; i<7 ; i++) { 00975 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]]; 00976 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i]; 00977 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i], 00978 huff_bits[i], 1, 1, 00979 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC); 00980 } 00981 vlcs_initialized = 1; 00982 } 00983 00984 init_atrac3_transforms(q); 00985 00986 atrac_generate_tables(); 00987 00988 /* Generate gain tables. */ 00989 for (i=0 ; i<16 ; i++) 00990 gain_tab1[i] = powf (2.0, (4 - i)); 00991 00992 for (i=-15 ; i<16 ; i++) 00993 gain_tab2[i+15] = powf (2.0, i * -0.125); 00994 00995 /* init the joint-stereo decoding data */ 00996 q->weighting_delay[0] = 0; 00997 q->weighting_delay[1] = 7; 00998 q->weighting_delay[2] = 0; 00999 q->weighting_delay[3] = 7; 01000 q->weighting_delay[4] = 0; 01001 q->weighting_delay[5] = 7; 01002 01003 for (i=0; i<4; i++) { 01004 q->matrix_coeff_index_prev[i] = 3; 01005 q->matrix_coeff_index_now[i] = 3; 01006 q->matrix_coeff_index_next[i] = 3; 01007 } 01008 01009 dsputil_init(&dsp, avctx); 01010 01011 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels); 01012 if (!q->pUnits) { 01013 av_free(q->decoded_bytes_buffer); 01014 return AVERROR(ENOMEM); 01015 } 01016 01017 avctx->sample_fmt = AV_SAMPLE_FMT_S16; 01018 return 0; 01019 } 01020 01021 01022 AVCodec ff_atrac3_decoder = 01023 { 01024 .name = "atrac3", 01025 .type = AVMEDIA_TYPE_AUDIO, 01026 .id = CODEC_ID_ATRAC3, 01027 .priv_data_size = sizeof(ATRAC3Context), 01028 .init = atrac3_decode_init, 01029 .close = atrac3_decode_close, 01030 .decode = atrac3_decode_frame, 01031 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"), 01032 };