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
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.
39 #include "libavutil/float_dsp.h"
40 #include "libavutil/libm.h"
43 #include "bytestream.h"
45 #include "fmtconvert.h"
48 #include "atrac3data.h"
50 #define JOINT_STEREO 0x12
53 #define SAMPLES_PER_FRAME 1024
56 /* These structures are needed to store the parsed gain control data. */
76 tonal_component components[64];
77 float prevFrame[SAMPLES_PER_FRAME];
79 gain_block gainBlock[2];
81 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
82 DECLARE_ALIGNED(32, float, IMDCT_buf)[SAMPLES_PER_FRAME];
84 float delayBuf1[46]; ///<qmf delay buffers
98 int samples_per_channel;
99 int samples_per_frame;
104 channel_unit* pUnits;
107 /** joint-stereo related variables */
108 int matrix_coeff_index_prev[4];
109 int matrix_coeff_index_now[4];
110 int matrix_coeff_index_next[4];
111 int weighting_delay[6];
115 float *outSamples[2];
116 uint8_t* decoded_bytes_buffer;
123 int scrambled_stream;
128 FmtConvertContext fmt_conv;
129 AVFloatDSPContext fdsp;
132 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
133 static VLC spectral_coeff_tab[7];
134 static float gain_tab1[16];
135 static float gain_tab2[31];
139 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
140 * caused by the reverse spectra of the QMF.
142 * @param pInput float input
143 * @param pOutput float output
144 * @param odd_band 1 if the band is an odd band
147 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
153 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
154 * or it gives better compression to do it this way.
155 * FIXME: It should be possible to handle this in imdct_calc
156 * for that to happen a modification of the prerotation step of
157 * all SIMD code and C code is needed.
158 * Or fix the functions before so they generate a pre reversed spectrum.
161 for (i=0; i<128; i++)
162 FFSWAP(float, pInput[i], pInput[255-i]);
165 q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
167 /* Perform windowing on the output. */
168 q->fdsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
174 * Atrac 3 indata descrambling, only used for data coming from the rm container
176 * @param inbuffer pointer to 8 bit array of indata
177 * @param out pointer to 8 bit array of outdata
178 * @param bytes amount of bytes
181 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
185 uint32_t* obuf = (uint32_t*) out;
187 off = (intptr_t)inbuffer & 3;
188 buf = (const uint32_t*) (inbuffer - off);
189 c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
191 for (i = 0; i < bytes/4; i++)
192 obuf[i] = c ^ buf[i];
195 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
201 static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
202 float enc_window[256];
205 /* Generate the mdct window, for details see
206 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
207 for (i=0 ; i<256; i++)
208 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
211 for (i=0 ; i<256; i++) {
212 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
213 mdct_window[511-i] = mdct_window[i];
216 /* Initialize the MDCT transform. */
217 return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 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);
230 av_freep(&q->outSamples[0]);
232 ff_mdct_end(&q->mdct_ctx);
238 / * Mantissa decoding
240 * @param gb the GetBit context
241 * @param selector what table is the output values coded with
242 * @param codingFlag constant length coding or variable length coding
243 * @param mantissas mantissa output table
244 * @param numCodes amount of values to get
247 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
249 int numBits, cnt, code, huffSymb;
254 if (codingFlag != 0) {
255 /* constant length coding (CLC) */
256 numBits = CLCLengthTab[selector];
259 for (cnt = 0; cnt < numCodes; cnt++) {
261 code = get_sbits(gb, numBits);
264 mantissas[cnt] = code;
267 for (cnt = 0; cnt < numCodes; cnt++) {
269 code = get_bits(gb, numBits); //numBits is always 4 in this case
272 mantissas[cnt*2] = seTab_0[code >> 2];
273 mantissas[cnt*2+1] = seTab_0[code & 3];
277 /* variable length coding (VLC) */
279 for (cnt = 0; cnt < numCodes; cnt++) {
280 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
282 code = huffSymb >> 1;
285 mantissas[cnt] = code;
288 for (cnt = 0; cnt < numCodes; cnt++) {
289 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
290 mantissas[cnt*2] = decTable1[huffSymb*2];
291 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
298 * Restore the quantized band spectrum coefficients
300 * @param gb the GetBit context
301 * @param pOut decoded band spectrum
302 * @return outSubbands subband counter, fix for broken specification/files
305 static int decodeSpectrum (GetBitContext *gb, float *pOut)
307 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
308 int subband_vlc_index[32], SF_idxs[32];
312 numSubbands = get_bits(gb, 5); // number of coded subbands
313 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
315 /* Get the VLC selector table for the subbands, 0 means not coded. */
316 for (cnt = 0; cnt <= numSubbands; cnt++)
317 subband_vlc_index[cnt] = get_bits(gb, 3);
319 /* Read the scale factor indexes from the stream. */
320 for (cnt = 0; cnt <= numSubbands; cnt++) {
321 if (subband_vlc_index[cnt] != 0)
322 SF_idxs[cnt] = get_bits(gb, 6);
325 for (cnt = 0; cnt <= numSubbands; cnt++) {
326 first = subbandTab[cnt];
327 last = subbandTab[cnt+1];
329 subbWidth = last - first;
331 if (subband_vlc_index[cnt] != 0) {
332 /* Decode spectral coefficients for this subband. */
333 /* TODO: This can be done faster is several blocks share the
334 * same VLC selector (subband_vlc_index) */
335 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
337 /* Decode the scale factor for this subband. */
338 SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
340 /* Inverse quantize the coefficients. */
341 for (pIn=mantissas ; first<last; first++, pIn++)
342 pOut[first] = *pIn * SF;
344 /* This subband was not coded, so zero the entire subband. */
345 memset(pOut+first, 0, subbWidth*sizeof(float));
349 /* Clear the subbands that were not coded. */
350 first = subbandTab[cnt];
351 memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
356 * Restore the quantized tonal components
358 * @param gb the GetBit context
359 * @param pComponent tone component
360 * @param numBands amount of coded bands
363 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
366 int components, coding_mode_selector, coding_mode, coded_values_per_component;
367 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
368 int band_flags[4], mantissa[8];
371 int component_count = 0;
373 components = get_bits(gb,5);
375 /* no tonal components */
379 coding_mode_selector = get_bits(gb,2);
380 if (coding_mode_selector == 2)
381 return AVERROR_INVALIDDATA;
383 coding_mode = coding_mode_selector & 1;
385 for (i = 0; i < components; i++) {
386 for (cnt = 0; cnt <= numBands; cnt++)
387 band_flags[cnt] = get_bits1(gb);
389 coded_values_per_component = get_bits(gb,3);
391 quant_step_index = get_bits(gb,3);
392 if (quant_step_index <= 1)
393 return AVERROR_INVALIDDATA;
395 if (coding_mode_selector == 3)
396 coding_mode = get_bits1(gb);
398 for (j = 0; j < (numBands + 1) * 4; j++) {
399 if (band_flags[j >> 2] == 0)
402 coded_components = get_bits(gb,3);
404 for (k=0; k<coded_components; k++) {
405 sfIndx = get_bits(gb,6);
406 if (component_count >= 64)
407 return AVERROR_INVALIDDATA;
408 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
409 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
410 coded_values = coded_values_per_component + 1;
411 coded_values = FFMIN(max_coded_values,coded_values);
413 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
415 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
417 pComponent[component_count].numCoefs = coded_values;
420 pCoef = pComponent[component_count].coef;
421 for (cnt = 0; cnt < coded_values; cnt++)
422 pCoef[cnt] = mantissa[cnt] * scalefactor;
429 return component_count;
433 * Decode gain parameters for the coded bands
435 * @param gb the GetBit context
436 * @param pGb the gainblock for the current band
437 * @param numBands amount of coded bands
440 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
445 gain_info *pGain = pGb->gBlock;
447 for (i=0 ; i<=numBands; i++)
449 numData = get_bits(gb,3);
450 pGain[i].num_gain_data = numData;
451 pLevel = pGain[i].levcode;
452 pLoc = pGain[i].loccode;
454 for (cf = 0; cf < numData; cf++){
455 pLevel[cf]= get_bits(gb,4);
456 pLoc [cf]= get_bits(gb,5);
457 if(cf && pLoc[cf] <= pLoc[cf-1])
458 return AVERROR_INVALIDDATA;
462 /* Clear the unused blocks. */
464 pGain[i].num_gain_data = 0;
470 * Apply gain parameters and perform the MDCT overlapping part
472 * @param pIn input float buffer
473 * @param pPrev previous float buffer to perform overlap against
474 * @param pOut output float buffer
475 * @param pGain1 current band gain info
476 * @param pGain2 next band gain info
479 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
481 /* gain compensation function */
482 float gain1, gain2, gain_inc;
483 int cnt, numdata, nsample, startLoc, endLoc;
486 if (pGain2->num_gain_data == 0)
489 gain1 = gain_tab1[pGain2->levcode[0]];
491 if (pGain1->num_gain_data == 0) {
492 for (cnt = 0; cnt < 256; cnt++)
493 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
495 numdata = pGain1->num_gain_data;
496 pGain1->loccode[numdata] = 32;
497 pGain1->levcode[numdata] = 4;
499 nsample = 0; // current sample = 0
501 for (cnt = 0; cnt < numdata; cnt++) {
502 startLoc = pGain1->loccode[cnt] * 8;
503 endLoc = startLoc + 8;
505 gain2 = gain_tab1[pGain1->levcode[cnt]];
506 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
509 for (; nsample < startLoc; nsample++)
510 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
512 /* interpolation is done over eight samples */
513 for (; nsample < endLoc; nsample++) {
514 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
519 for (; nsample < 256; nsample++)
520 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
523 /* Delay for the overlapping part. */
524 memcpy(pPrev, &pIn[256], 256*sizeof(float));
528 * Combine the tonal band spectrum and regular band spectrum
529 * Return position of the last tonal coefficient
531 * @param pSpectrum output spectrum buffer
532 * @param numComponents amount of tonal components
533 * @param pComponent tonal components for this band
536 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
538 int cnt, i, lastPos = -1;
541 for (cnt = 0; cnt < numComponents; cnt++){
542 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
543 pIn = pComponent[cnt].coef;
544 pOut = &(pSpectrum[pComponent[cnt].pos]);
546 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
554 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
556 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
558 int i, band, nsample, s1, s2;
560 float mc1_l, mc1_r, mc2_l, mc2_r;
562 for (i=0,band = 0; band < 4*256; band+=256,i++) {
568 /* Selector value changed, interpolation needed. */
569 mc1_l = matrixCoeffs[s1*2];
570 mc1_r = matrixCoeffs[s1*2+1];
571 mc2_l = matrixCoeffs[s2*2];
572 mc2_r = matrixCoeffs[s2*2+1];
574 /* Interpolation is done over the first eight samples. */
575 for(; nsample < 8; nsample++) {
576 c1 = su1[band+nsample];
577 c2 = su2[band+nsample];
578 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
579 su1[band+nsample] = c2;
580 su2[band+nsample] = c1 * 2.0 - c2;
584 /* Apply the matrix without interpolation. */
586 case 0: /* M/S decoding */
587 for (; nsample < 256; nsample++) {
588 c1 = su1[band+nsample];
589 c2 = su2[band+nsample];
590 su1[band+nsample] = c2 * 2.0;
591 su2[band+nsample] = (c1 - c2) * 2.0;
596 for (; nsample < 256; nsample++) {
597 c1 = su1[band+nsample];
598 c2 = su2[band+nsample];
599 su1[band+nsample] = (c1 + c2) * 2.0;
600 su2[band+nsample] = c2 * -2.0;
605 for (; nsample < 256; nsample++) {
606 c1 = su1[band+nsample];
607 c2 = su2[band+nsample];
608 su1[band+nsample] = c1 + c2;
609 su2[band+nsample] = c1 - c2;
618 static void getChannelWeights (int indx, int flag, float ch[2]){
624 ch[0] = (float)(indx & 7) / 7.0;
625 ch[1] = sqrt(2 - ch[0]*ch[0]);
627 FFSWAP(float, ch[0], ch[1]);
631 static void channelWeighting (float *su1, float *su2, int *p3)
634 /* w[x][y] y=0 is left y=1 is right */
637 if (p3[1] != 7 || p3[3] != 7){
638 getChannelWeights(p3[1], p3[0], w[0]);
639 getChannelWeights(p3[3], p3[2], w[1]);
641 for(band = 1; band < 4; band++) {
642 /* scale the channels by the weights */
643 for(nsample = 0; nsample < 8; nsample++) {
644 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
645 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
648 for(; nsample < 256; nsample++) {
649 su1[band*256+nsample] *= w[1][0];
650 su2[band*256+nsample] *= w[1][1];
658 * Decode a Sound Unit
660 * @param gb the GetBit context
661 * @param pSnd the channel unit to be used
662 * @param pOut the decoded samples before IQMF in float representation
663 * @param channelNum channel number
664 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
668 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
670 int band, result=0, numSubbands, lastTonal, numBands;
672 if (codingMode == JOINT_STEREO && channelNum == 1) {
673 if (get_bits(gb,2) != 3) {
674 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
675 return AVERROR_INVALIDDATA;
678 if (get_bits(gb,6) != 0x28) {
679 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
680 return AVERROR_INVALIDDATA;
684 /* number of coded QMF bands */
685 pSnd->bandsCoded = get_bits(gb,2);
687 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
688 if (result) return result;
690 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
691 if (pSnd->numComponents == -1) return -1;
693 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
695 /* Merge the decoded spectrum and tonal components. */
696 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
699 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
700 numBands = (subbandTab[numSubbands] - 1) >> 8;
702 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
705 /* Reconstruct time domain samples. */
706 for (band=0; band<4; band++) {
707 /* Perform the IMDCT step without overlapping. */
708 if (band <= numBands) {
709 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
711 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
713 /* gain compensation and overlapping */
714 gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
716 &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
717 &pSnd->gainBlock[ pSnd->gcBlkSwitch].gBlock[band]);
720 /* Swap the gain control buffers for the next frame. */
721 pSnd->gcBlkSwitch ^= 1;
729 * @param q Atrac3 private context
730 * @param databuf the input data
733 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
737 float *p1, *p2, *p3, *p4;
740 if (q->codingMode == JOINT_STEREO) {
742 /* channel coupling mode */
743 /* decode Sound Unit 1 */
744 init_get_bits(&q->gb,databuf,q->bits_per_frame);
746 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
750 /* Framedata of the su2 in the joint-stereo mode is encoded in
751 * reverse byte order so we need to swap it first. */
752 if (databuf == q->decoded_bytes_buffer) {
753 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
754 ptr1 = q->decoded_bytes_buffer;
755 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
756 FFSWAP(uint8_t,*ptr1,*ptr2);
759 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
760 for (i = 0; i < q->bytes_per_frame; i++)
761 q->decoded_bytes_buffer[i] = *ptr2--;
764 /* Skip the sync codes (0xF8). */
765 ptr1 = q->decoded_bytes_buffer;
766 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
767 if (i >= q->bytes_per_frame)
768 return AVERROR_INVALIDDATA;
772 /* set the bitstream reader at the start of the second Sound Unit*/
773 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
775 /* Fill the Weighting coeffs delay buffer */
776 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
777 q->weighting_delay[4] = get_bits1(&q->gb);
778 q->weighting_delay[5] = get_bits(&q->gb,3);
780 for (i = 0; i < 4; i++) {
781 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
782 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
783 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
786 /* Decode Sound Unit 2. */
787 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
791 /* Reconstruct the channel coefficients. */
792 reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
794 channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
797 /* normal stereo mode or mono */
798 /* Decode the channel sound units. */
799 for (i=0 ; i<q->channels ; i++) {
801 /* Set the bitstream reader at the start of a channel sound unit. */
802 init_get_bits(&q->gb,
803 databuf + i * q->bytes_per_frame / q->channels,
804 q->bits_per_frame / q->channels);
806 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
812 /* Apply the iQMF synthesis filter. */
813 for (i=0 ; i<q->channels ; i++) {
818 ff_atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
819 ff_atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
820 ff_atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
828 * Atrac frame decoding
830 * @param avctx pointer to the AVCodecContext
833 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
834 int *got_frame_ptr, AVPacket *avpkt)
836 const uint8_t *buf = avpkt->data;
837 int buf_size = avpkt->size;
838 ATRAC3Context *q = avctx->priv_data;
840 const uint8_t* databuf;
842 int16_t *samples_s16;
844 if (buf_size < avctx->block_align) {
845 av_log(avctx, AV_LOG_ERROR,
846 "Frame too small (%d bytes). Truncated file?\n", buf_size);
847 return AVERROR_INVALIDDATA;
850 /* get output buffer */
851 q->frame.nb_samples = SAMPLES_PER_FRAME;
852 if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
853 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
856 samples_flt = (float *)q->frame.data[0];
857 samples_s16 = (int16_t *)q->frame.data[0];
859 /* Check if we need to descramble and what buffer to pass on. */
860 if (q->scrambled_stream) {
861 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
862 databuf = q->decoded_bytes_buffer;
867 if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
868 result = decodeFrame(q, databuf, &samples_flt);
870 result = decodeFrame(q, databuf, q->outSamples);
873 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
878 if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
879 q->fmt_conv.float_interleave(samples_flt,
880 (const float **)q->outSamples,
881 SAMPLES_PER_FRAME, 2);
882 } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
883 q->fmt_conv.float_to_int16_interleave(samples_s16,
884 (const float **)q->outSamples,
885 SAMPLES_PER_FRAME, q->channels);
889 *(AVFrame *)data = q->frame;
891 return avctx->block_align;
896 * Atrac3 initialization
898 * @param avctx pointer to the AVCodecContext
901 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
904 const uint8_t *edata_ptr = avctx->extradata;
905 ATRAC3Context *q = avctx->priv_data;
906 static VLC_TYPE atrac3_vlc_table[4096][2];
907 static int vlcs_initialized = 0;
909 /* Take data from the AVCodecContext (RM container). */
910 q->sample_rate = avctx->sample_rate;
911 q->channels = avctx->channels;
912 q->bit_rate = avctx->bit_rate;
913 q->bits_per_frame = avctx->block_align * 8;
914 q->bytes_per_frame = avctx->block_align;
916 /* Take care of the codec-specific extradata. */
917 if (avctx->extradata_size == 14) {
918 /* Parse the extradata, WAV format */
919 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
920 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
921 q->codingMode = bytestream_get_le16(&edata_ptr);
922 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
923 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
924 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
927 q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
928 q->atrac3version = 4;
931 q->codingMode = JOINT_STEREO;
933 q->codingMode = STEREO;
935 q->scrambled_stream = 0;
937 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)) {
939 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);
940 return AVERROR_INVALIDDATA;
943 } else if (avctx->extradata_size == 10) {
944 /* Parse the extradata, RM format. */
945 q->atrac3version = bytestream_get_be32(&edata_ptr);
946 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
947 q->delay = bytestream_get_be16(&edata_ptr);
948 q->codingMode = bytestream_get_be16(&edata_ptr);
950 q->samples_per_channel = q->samples_per_frame / q->channels;
951 q->scrambled_stream = 1;
954 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
956 /* Check the extradata. */
958 if (q->atrac3version != 4) {
959 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
960 return AVERROR_INVALIDDATA;
963 if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
964 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
965 return AVERROR_INVALIDDATA;
968 if (q->delay != 0x88E) {
969 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
970 return AVERROR_INVALIDDATA;
973 if (q->codingMode == STEREO) {
974 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
975 } else if (q->codingMode == JOINT_STEREO) {
976 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
978 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
979 return AVERROR_INVALIDDATA;
982 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
983 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
984 return AVERROR(EINVAL);
988 if(avctx->block_align >= UINT_MAX/2)
989 return AVERROR(EINVAL);
991 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
992 * this is for the bitstream reader. */
993 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
994 return AVERROR(ENOMEM);
997 /* Initialize the VLC tables. */
998 if (!vlcs_initialized) {
999 for (i=0 ; i<7 ; i++) {
1000 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
1001 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
1002 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
1004 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
1006 vlcs_initialized = 1;
1009 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
1010 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1012 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1014 if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
1015 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
1016 av_freep(&q->decoded_bytes_buffer);
1020 ff_atrac_generate_tables();
1022 /* Generate gain tables. */
1023 for (i=0 ; i<16 ; i++)
1024 gain_tab1[i] = exp2f (4 - i);
1026 for (i=-15 ; i<16 ; i++)
1027 gain_tab2[i+15] = exp2f (i * -0.125);
1029 /* init the joint-stereo decoding data */
1030 q->weighting_delay[0] = 0;
1031 q->weighting_delay[1] = 7;
1032 q->weighting_delay[2] = 0;
1033 q->weighting_delay[3] = 7;
1034 q->weighting_delay[4] = 0;
1035 q->weighting_delay[5] = 7;
1037 for (i=0; i<4; i++) {
1038 q->matrix_coeff_index_prev[i] = 3;
1039 q->matrix_coeff_index_now[i] = 3;
1040 q->matrix_coeff_index_next[i] = 3;
1043 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1044 ff_fmt_convert_init(&q->fmt_conv, avctx);
1046 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1048 atrac3_decode_close(avctx);
1049 return AVERROR(ENOMEM);
1052 if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
1053 q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
1054 q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
1055 if (!q->outSamples[0]) {
1056 atrac3_decode_close(avctx);
1057 return AVERROR(ENOMEM);
1061 avcodec_get_frame_defaults(&q->frame);
1062 avctx->coded_frame = &q->frame;
1068 AVCodec ff_atrac3_decoder =
1071 .type = AVMEDIA_TYPE_AUDIO,
1072 .id = AV_CODEC_ID_ATRAC3,
1073 .priv_data_size = sizeof(ATRAC3Context),
1074 .init = atrac3_decode_init,
1075 .close = atrac3_decode_close,
1076 .decode = atrac3_decode_frame,
1077 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1078 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),