2 * Atrac 3 compatible decoder
3 * Copyright (c) 2006-2008 Maxim Poliakovski
4 * Copyright (c) 2006-2008 Benjamin Larsson
6 * This file is part of FFmpeg.
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.
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.
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
24 * @file libavcodec/atrac3.c
25 * Atrac 3 compatible decoder.
26 * This decoder handles Sony's ATRAC3 data.
28 * Container formats used to store atrac 3 data:
29 * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
31 * To use this decoder, a calling application must supply the extradata
32 * bytes provided in the containers above.
42 #include "bytestream.h"
45 #include "atrac3data.h"
47 #define JOINT_STEREO 0x12
51 /* These structures are needed to store the parsed gain control data. */
71 tonal_component components[64];
72 float prevFrame[1024];
74 gain_block gainBlock[2];
76 DECLARE_ALIGNED_16(float, spectrum[1024]);
77 DECLARE_ALIGNED_16(float, IMDCT_buf[1024]);
79 float delayBuf1[46]; ///<qmf delay buffers
92 int samples_per_channel;
93 int samples_per_frame;
101 /** joint-stereo related variables */
102 int matrix_coeff_index_prev[4];
103 int matrix_coeff_index_now[4];
104 int matrix_coeff_index_next[4];
105 int weighting_delay[6];
109 float outSamples[2048];
110 uint8_t* decoded_bytes_buffer;
117 int scrambled_stream;
122 static DECLARE_ALIGNED_16(float,mdct_window[512]);
123 static VLC spectral_coeff_tab[7];
124 static float gain_tab1[16];
125 static float gain_tab2[31];
126 static MDCTContext mdct_ctx;
127 static DSPContext dsp;
131 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
132 * caused by the reverse spectra of the QMF.
134 * @param pInput float input
135 * @param pOutput float output
136 * @param odd_band 1 if the band is an odd band
139 static void IMLT(float *pInput, float *pOutput, int odd_band)
145 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
146 * or it gives better compression to do it this way.
147 * FIXME: It should be possible to handle this in ff_imdct_calc
148 * for that to happen a modification of the prerotation step of
149 * all SIMD code and C code is needed.
150 * Or fix the functions before so they generate a pre reversed spectrum.
153 for (i=0; i<128; i++)
154 FFSWAP(float, pInput[i], pInput[255-i]);
157 ff_imdct_calc(&mdct_ctx,pOutput,pInput);
159 /* Perform windowing on the output. */
160 dsp.vector_fmul(pOutput,mdct_window,512);
166 * Atrac 3 indata descrambling, only used for data coming from the rm container
168 * @param in pointer to 8 bit array of indata
169 * @param bits amount of bits
170 * @param out pointer to 8 bit array of outdata
173 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
177 uint32_t* obuf = (uint32_t*) out;
179 off = (intptr_t)inbuffer & 3;
180 buf = (const uint32_t*) (inbuffer - off);
181 c = be2me_32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
183 for (i = 0; i < bytes/4; i++)
184 obuf[i] = c ^ buf[i];
187 av_log(NULL,AV_LOG_DEBUG,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off);
193 static av_cold void init_atrac3_transforms(ATRAC3Context *q) {
194 float enc_window[256];
198 /* Generate the mdct window, for details see
199 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
200 for (i=0 ; i<256; i++)
201 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
204 for (i=0 ; i<256; i++) {
205 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
206 mdct_window[511-i] = mdct_window[i];
209 /* Generate the QMF window. */
210 for (i=0 ; i<24; i++) {
211 s = qmf_48tap_half[i] * 2.0;
213 qmf_window[47 - i] = s;
216 /* Initialize the MDCT transform. */
217 ff_mdct_init(&mdct_ctx, 9, 1, 1.0);
221 * Atrac3 uninit, free all allocated memory
224 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
226 ATRAC3Context *q = avctx->priv_data;
229 av_free(q->decoded_bytes_buffer);
235 / * Mantissa decoding
237 * @param gb the GetBit context
238 * @param selector what table is the output values coded with
239 * @param codingFlag constant length coding or variable length coding
240 * @param mantissas mantissa output table
241 * @param numCodes amount of values to get
244 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
246 int numBits, cnt, code, huffSymb;
251 if (codingFlag != 0) {
252 /* constant length coding (CLC) */
253 numBits = CLCLengthTab[selector];
256 for (cnt = 0; cnt < numCodes; cnt++) {
258 code = get_sbits(gb, numBits);
261 mantissas[cnt] = code;
264 for (cnt = 0; cnt < numCodes; cnt++) {
266 code = get_bits(gb, numBits); //numBits is always 4 in this case
269 mantissas[cnt*2] = seTab_0[code >> 2];
270 mantissas[cnt*2+1] = seTab_0[code & 3];
274 /* variable length coding (VLC) */
276 for (cnt = 0; cnt < numCodes; cnt++) {
277 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
279 code = huffSymb >> 1;
282 mantissas[cnt] = code;
285 for (cnt = 0; cnt < numCodes; cnt++) {
286 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
287 mantissas[cnt*2] = decTable1[huffSymb*2];
288 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
295 * Restore the quantized band spectrum coefficients
297 * @param gb the GetBit context
298 * @param pOut decoded band spectrum
299 * @return outSubbands subband counter, fix for broken specification/files
302 static int decodeSpectrum (GetBitContext *gb, float *pOut)
304 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
305 int subband_vlc_index[32], SF_idxs[32];
309 numSubbands = get_bits(gb, 5); // number of coded subbands
310 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
312 /* Get the VLC selector table for the subbands, 0 means not coded. */
313 for (cnt = 0; cnt <= numSubbands; cnt++)
314 subband_vlc_index[cnt] = get_bits(gb, 3);
316 /* Read the scale factor indexes from the stream. */
317 for (cnt = 0; cnt <= numSubbands; cnt++) {
318 if (subband_vlc_index[cnt] != 0)
319 SF_idxs[cnt] = get_bits(gb, 6);
322 for (cnt = 0; cnt <= numSubbands; cnt++) {
323 first = subbandTab[cnt];
324 last = subbandTab[cnt+1];
326 subbWidth = last - first;
328 if (subband_vlc_index[cnt] != 0) {
329 /* Decode spectral coefficients for this subband. */
330 /* TODO: This can be done faster is several blocks share the
331 * same VLC selector (subband_vlc_index) */
332 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
334 /* Decode the scale factor for this subband. */
335 SF = sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
337 /* Inverse quantize the coefficients. */
338 for (pIn=mantissas ; first<last; first++, pIn++)
339 pOut[first] = *pIn * SF;
341 /* This subband was not coded, so zero the entire subband. */
342 memset(pOut+first, 0, subbWidth*sizeof(float));
346 /* Clear the subbands that were not coded. */
347 first = subbandTab[cnt];
348 memset(pOut+first, 0, (1024 - first) * sizeof(float));
353 * Restore the quantized tonal components
355 * @param gb the GetBit context
356 * @param pComponent tone component
357 * @param numBands amount of coded bands
360 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
363 int components, coding_mode_selector, coding_mode, coded_values_per_component;
364 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
365 int band_flags[4], mantissa[8];
368 int component_count = 0;
370 components = get_bits(gb,5);
372 /* no tonal components */
376 coding_mode_selector = get_bits(gb,2);
377 if (coding_mode_selector == 2)
380 coding_mode = coding_mode_selector & 1;
382 for (i = 0; i < components; i++) {
383 for (cnt = 0; cnt <= numBands; cnt++)
384 band_flags[cnt] = get_bits1(gb);
386 coded_values_per_component = get_bits(gb,3);
388 quant_step_index = get_bits(gb,3);
389 if (quant_step_index <= 1)
392 if (coding_mode_selector == 3)
393 coding_mode = get_bits1(gb);
395 for (j = 0; j < (numBands + 1) * 4; j++) {
396 if (band_flags[j >> 2] == 0)
399 coded_components = get_bits(gb,3);
401 for (k=0; k<coded_components; k++) {
402 sfIndx = get_bits(gb,6);
403 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
404 max_coded_values = 1024 - pComponent[component_count].pos;
405 coded_values = coded_values_per_component + 1;
406 coded_values = FFMIN(max_coded_values,coded_values);
408 scalefactor = sf_table[sfIndx] * iMaxQuant[quant_step_index];
410 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
412 pComponent[component_count].numCoefs = coded_values;
415 pCoef = pComponent[component_count].coef;
416 for (cnt = 0; cnt < coded_values; cnt++)
417 pCoef[cnt] = mantissa[cnt] * scalefactor;
424 return component_count;
428 * Decode gain parameters for the coded bands
430 * @param gb the GetBit context
431 * @param pGb the gainblock for the current band
432 * @param numBands amount of coded bands
435 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
440 gain_info *pGain = pGb->gBlock;
442 for (i=0 ; i<=numBands; i++)
444 numData = get_bits(gb,3);
445 pGain[i].num_gain_data = numData;
446 pLevel = pGain[i].levcode;
447 pLoc = pGain[i].loccode;
449 for (cf = 0; cf < numData; cf++){
450 pLevel[cf]= get_bits(gb,4);
451 pLoc [cf]= get_bits(gb,5);
452 if(cf && pLoc[cf] <= pLoc[cf-1])
457 /* Clear the unused blocks. */
459 pGain[i].num_gain_data = 0;
465 * Apply gain parameters and perform the MDCT overlapping part
467 * @param pIn input float buffer
468 * @param pPrev previous float buffer to perform overlap against
469 * @param pOut output float buffer
470 * @param pGain1 current band gain info
471 * @param pGain2 next band gain info
474 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
476 /* gain compensation function */
477 float gain1, gain2, gain_inc;
478 int cnt, numdata, nsample, startLoc, endLoc;
481 if (pGain2->num_gain_data == 0)
484 gain1 = gain_tab1[pGain2->levcode[0]];
486 if (pGain1->num_gain_data == 0) {
487 for (cnt = 0; cnt < 256; cnt++)
488 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
490 numdata = pGain1->num_gain_data;
491 pGain1->loccode[numdata] = 32;
492 pGain1->levcode[numdata] = 4;
494 nsample = 0; // current sample = 0
496 for (cnt = 0; cnt < numdata; cnt++) {
497 startLoc = pGain1->loccode[cnt] * 8;
498 endLoc = startLoc + 8;
500 gain2 = gain_tab1[pGain1->levcode[cnt]];
501 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
504 for (; nsample < startLoc; nsample++)
505 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
507 /* interpolation is done over eight samples */
508 for (; nsample < endLoc; nsample++) {
509 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
514 for (; nsample < 256; nsample++)
515 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
518 /* Delay for the overlapping part. */
519 memcpy(pPrev, &pIn[256], 256*sizeof(float));
523 * Combine the tonal band spectrum and regular band spectrum
524 * Return position of the last tonal coefficient
526 * @param pSpectrum output spectrum buffer
527 * @param numComponents amount of tonal components
528 * @param pComponent tonal components for this band
531 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
533 int cnt, i, lastPos = -1;
536 for (cnt = 0; cnt < numComponents; cnt++){
537 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
538 pIn = pComponent[cnt].coef;
539 pOut = &(pSpectrum[pComponent[cnt].pos]);
541 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
549 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
551 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
553 int i, band, nsample, s1, s2;
555 float mc1_l, mc1_r, mc2_l, mc2_r;
557 for (i=0,band = 0; band < 4*256; band+=256,i++) {
563 /* Selector value changed, interpolation needed. */
564 mc1_l = matrixCoeffs[s1*2];
565 mc1_r = matrixCoeffs[s1*2+1];
566 mc2_l = matrixCoeffs[s2*2];
567 mc2_r = matrixCoeffs[s2*2+1];
569 /* Interpolation is done over the first eight samples. */
570 for(; nsample < 8; nsample++) {
571 c1 = su1[band+nsample];
572 c2 = su2[band+nsample];
573 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
574 su1[band+nsample] = c2;
575 su2[band+nsample] = c1 * 2.0 - c2;
579 /* Apply the matrix without interpolation. */
581 case 0: /* M/S decoding */
582 for (; nsample < 256; nsample++) {
583 c1 = su1[band+nsample];
584 c2 = su2[band+nsample];
585 su1[band+nsample] = c2 * 2.0;
586 su2[band+nsample] = (c1 - c2) * 2.0;
591 for (; nsample < 256; nsample++) {
592 c1 = su1[band+nsample];
593 c2 = su2[band+nsample];
594 su1[band+nsample] = (c1 + c2) * 2.0;
595 su2[band+nsample] = c2 * -2.0;
600 for (; nsample < 256; nsample++) {
601 c1 = su1[band+nsample];
602 c2 = su2[band+nsample];
603 su1[band+nsample] = c1 + c2;
604 su2[band+nsample] = c1 - c2;
613 static void getChannelWeights (int indx, int flag, float ch[2]){
619 ch[0] = (float)(indx & 7) / 7.0;
620 ch[1] = sqrt(2 - ch[0]*ch[0]);
622 FFSWAP(float, ch[0], ch[1]);
626 static void channelWeighting (float *su1, float *su2, int *p3)
629 /* w[x][y] y=0 is left y=1 is right */
632 if (p3[1] != 7 || p3[3] != 7){
633 getChannelWeights(p3[1], p3[0], w[0]);
634 getChannelWeights(p3[3], p3[2], w[1]);
636 for(band = 1; band < 4; band++) {
637 /* scale the channels by the weights */
638 for(nsample = 0; nsample < 8; nsample++) {
639 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
640 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
643 for(; nsample < 256; nsample++) {
644 su1[band*256+nsample] *= w[1][0];
645 su2[band*256+nsample] *= w[1][1];
653 * Decode a Sound Unit
655 * @param gb the GetBit context
656 * @param pSnd the channel unit to be used
657 * @param pOut the decoded samples before IQMF in float representation
658 * @param channelNum channel number
659 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
663 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
665 int band, result=0, numSubbands, lastTonal, numBands;
667 if (codingMode == JOINT_STEREO && channelNum == 1) {
668 if (get_bits(gb,2) != 3) {
669 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
673 if (get_bits(gb,6) != 0x28) {
674 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
679 /* number of coded QMF bands */
680 pSnd->bandsCoded = get_bits(gb,2);
682 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
683 if (result) return result;
685 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
686 if (pSnd->numComponents == -1) return -1;
688 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
690 /* Merge the decoded spectrum and tonal components. */
691 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
694 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
695 numBands = (subbandTab[numSubbands] - 1) >> 8;
697 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
700 /* Reconstruct time domain samples. */
701 for (band=0; band<4; band++) {
702 /* Perform the IMDCT step without overlapping. */
703 if (band <= numBands) {
704 IMLT(&(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
706 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
708 /* gain compensation and overlapping */
709 gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]),
710 &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]),
711 &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band]));
714 /* Swap the gain control buffers for the next frame. */
715 pSnd->gcBlkSwitch ^= 1;
723 * @param q Atrac3 private context
724 * @param databuf the input data
727 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf)
730 float *p1, *p2, *p3, *p4;
733 if (q->codingMode == JOINT_STEREO) {
735 /* channel coupling mode */
736 /* decode Sound Unit 1 */
737 init_get_bits(&q->gb,databuf,q->bits_per_frame);
739 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO);
743 /* Framedata of the su2 in the joint-stereo mode is encoded in
744 * reverse byte order so we need to swap it first. */
745 if (databuf == q->decoded_bytes_buffer) {
746 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
747 ptr1 = q->decoded_bytes_buffer;
748 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
749 FFSWAP(uint8_t,*ptr1,*ptr2);
752 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
753 for (i = 0; i < q->bytes_per_frame; i++)
754 q->decoded_bytes_buffer[i] = *ptr2--;
757 /* Skip the sync codes (0xF8). */
758 ptr1 = q->decoded_bytes_buffer;
759 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
760 if (i >= q->bytes_per_frame)
765 /* set the bitstream reader at the start of the second Sound Unit*/
766 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
768 /* Fill the Weighting coeffs delay buffer */
769 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
770 q->weighting_delay[4] = get_bits1(&q->gb);
771 q->weighting_delay[5] = get_bits(&q->gb,3);
773 for (i = 0; i < 4; i++) {
774 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
775 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
776 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
779 /* Decode Sound Unit 2. */
780 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO);
784 /* Reconstruct the channel coefficients. */
785 reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
787 channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay);
790 /* normal stereo mode or mono */
791 /* Decode the channel sound units. */
792 for (i=0 ; i<q->channels ; i++) {
794 /* Set the bitstream reader at the start of a channel sound unit. */
795 init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels);
797 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode);
803 /* Apply the iQMF synthesis filter. */
805 for (i=0 ; i<q->channels ; i++) {
809 atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
810 atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
811 atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
820 * Atrac frame decoding
822 * @param avctx pointer to the AVCodecContext
825 static int atrac3_decode_frame(AVCodecContext *avctx,
826 void *data, int *data_size,
828 const uint8_t *buf = avpkt->data;
829 int buf_size = avpkt->size;
830 ATRAC3Context *q = avctx->priv_data;
832 const uint8_t* databuf;
833 int16_t* samples = data;
835 if (buf_size < avctx->block_align)
838 /* Check if we need to descramble and what buffer to pass on. */
839 if (q->scrambled_stream) {
840 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
841 databuf = q->decoded_bytes_buffer;
846 result = decodeFrame(q, databuf);
849 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
853 if (q->channels == 1) {
855 for (i = 0; i<1024; i++)
856 samples[i] = av_clip_int16(round(q->outSamples[i]));
857 *data_size = 1024 * sizeof(int16_t);
860 for (i = 0; i < 1024; i++) {
861 samples[i*2] = av_clip_int16(round(q->outSamples[i]));
862 samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i]));
864 *data_size = 2048 * sizeof(int16_t);
867 return avctx->block_align;
872 * Atrac3 initialization
874 * @param avctx pointer to the AVCodecContext
877 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
880 const uint8_t *edata_ptr = avctx->extradata;
881 ATRAC3Context *q = avctx->priv_data;
882 static VLC_TYPE atrac3_vlc_table[4096][2];
883 static int vlcs_initialized = 0;
885 /* Take data from the AVCodecContext (RM container). */
886 q->sample_rate = avctx->sample_rate;
887 q->channels = avctx->channels;
888 q->bit_rate = avctx->bit_rate;
889 q->bits_per_frame = avctx->block_align * 8;
890 q->bytes_per_frame = avctx->block_align;
892 /* Take care of the codec-specific extradata. */
893 if (avctx->extradata_size == 14) {
894 /* Parse the extradata, WAV format */
895 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
896 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
897 q->codingMode = bytestream_get_le16(&edata_ptr);
898 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
899 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
900 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
903 q->samples_per_frame = 1024 * q->channels;
904 q->atrac3version = 4;
907 q->codingMode = JOINT_STEREO;
909 q->codingMode = STEREO;
911 q->scrambled_stream = 0;
913 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)) {
915 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);
919 } else if (avctx->extradata_size == 10) {
920 /* Parse the extradata, RM format. */
921 q->atrac3version = bytestream_get_be32(&edata_ptr);
922 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
923 q->delay = bytestream_get_be16(&edata_ptr);
924 q->codingMode = bytestream_get_be16(&edata_ptr);
926 q->samples_per_channel = q->samples_per_frame / q->channels;
927 q->scrambled_stream = 1;
930 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
932 /* Check the extradata. */
934 if (q->atrac3version != 4) {
935 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
939 if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) {
940 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
944 if (q->delay != 0x88E) {
945 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
949 if (q->codingMode == STEREO) {
950 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
951 } else if (q->codingMode == JOINT_STEREO) {
952 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
954 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
958 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
959 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
964 if(avctx->block_align >= UINT_MAX/2)
967 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
968 * this is for the bitstream reader. */
969 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
970 return AVERROR(ENOMEM);
973 /* Initialize the VLC tables. */
974 if (!vlcs_initialized) {
975 for (i=0 ; i<7 ; i++) {
976 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
977 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
978 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
980 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
982 vlcs_initialized = 1;
985 init_atrac3_transforms(q);
987 atrac_generate_tables();
989 /* Generate gain tables. */
990 for (i=0 ; i<16 ; i++)
991 gain_tab1[i] = powf (2.0, (4 - i));
993 for (i=-15 ; i<16 ; i++)
994 gain_tab2[i+15] = powf (2.0, i * -0.125);
996 /* init the joint-stereo decoding data */
997 q->weighting_delay[0] = 0;
998 q->weighting_delay[1] = 7;
999 q->weighting_delay[2] = 0;
1000 q->weighting_delay[3] = 7;
1001 q->weighting_delay[4] = 0;
1002 q->weighting_delay[5] = 7;
1004 for (i=0; i<4; i++) {
1005 q->matrix_coeff_index_prev[i] = 3;
1006 q->matrix_coeff_index_now[i] = 3;
1007 q->matrix_coeff_index_next[i] = 3;
1010 dsputil_init(&dsp, avctx);
1012 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1014 av_free(q->decoded_bytes_buffer);
1015 return AVERROR(ENOMEM);
1018 avctx->sample_fmt = SAMPLE_FMT_S16;
1023 AVCodec atrac3_decoder =
1026 .type = CODEC_TYPE_AUDIO,
1027 .id = CODEC_ID_ATRAC3,
1028 .priv_data_size = sizeof(ATRAC3Context),
1029 .init = atrac3_decode_init,
1030 .close = atrac3_decode_close,
1031 .decode = atrac3_decode_frame,
1032 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),