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 Libav.
8 * Libav 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 * Libav 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 Libav; 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.
42 #include "bytestream.h"
44 #include "fmtconvert.h"
47 #include "atrac3data.h"
49 #define JOINT_STEREO 0x12
52 #define SAMPLES_PER_FRAME 1024
55 /* These structures are needed to store the parsed gain control data. */
75 tonal_component components[64];
76 float prevFrame[SAMPLES_PER_FRAME];
78 gain_block gainBlock[2];
80 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
81 DECLARE_ALIGNED(32, float, IMDCT_buf)[SAMPLES_PER_FRAME];
83 float delayBuf1[46]; ///<qmf delay buffers
97 int samples_per_channel;
98 int samples_per_frame;
103 channel_unit* pUnits;
106 /** joint-stereo related variables */
107 int matrix_coeff_index_prev[4];
108 int matrix_coeff_index_now[4];
109 int matrix_coeff_index_next[4];
110 int weighting_delay[6];
114 float *outSamples[2];
115 uint8_t* decoded_bytes_buffer;
122 int scrambled_stream;
127 FmtConvertContext fmt_conv;
130 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
131 static VLC spectral_coeff_tab[7];
132 static float gain_tab1[16];
133 static float gain_tab2[31];
134 static DSPContext dsp;
138 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
139 * caused by the reverse spectra of the QMF.
141 * @param pInput float input
142 * @param pOutput float output
143 * @param odd_band 1 if the band is an odd band
146 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
152 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
153 * or it gives better compression to do it this way.
154 * FIXME: It should be possible to handle this in imdct_calc
155 * for that to happen a modification of the prerotation step of
156 * all SIMD code and C code is needed.
157 * Or fix the functions before so they generate a pre reversed spectrum.
160 for (i=0; i<128; i++)
161 FFSWAP(float, pInput[i], pInput[255-i]);
164 q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
166 /* Perform windowing on the output. */
167 dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
173 * Atrac 3 indata descrambling, only used for data coming from the rm container
175 * @param inbuffer pointer to 8 bit array of indata
176 * @param out pointer to 8 bit array of outdata
177 * @param bytes amount of bytes
180 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
184 uint32_t* obuf = (uint32_t*) out;
186 off = (intptr_t)inbuffer & 3;
187 buf = (const uint32_t*) (inbuffer - off);
188 c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
190 for (i = 0; i < bytes/4; i++)
191 obuf[i] = c ^ buf[i];
194 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
200 static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
201 float enc_window[256];
204 /* Generate the mdct window, for details see
205 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
206 for (i=0 ; i<256; i++)
207 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
210 for (i=0 ; i<256; i++) {
211 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
212 mdct_window[511-i] = mdct_window[i];
215 /* Initialize the MDCT transform. */
216 return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
220 * Atrac3 uninit, free all allocated memory
223 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
225 ATRAC3Context *q = avctx->priv_data;
228 av_free(q->decoded_bytes_buffer);
229 av_freep(&q->outSamples[0]);
231 ff_mdct_end(&q->mdct_ctx);
237 / * Mantissa decoding
239 * @param gb the GetBit context
240 * @param selector what table is the output values coded with
241 * @param codingFlag constant length coding or variable length coding
242 * @param mantissas mantissa output table
243 * @param numCodes amount of values to get
246 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
248 int numBits, cnt, code, huffSymb;
253 if (codingFlag != 0) {
254 /* constant length coding (CLC) */
255 numBits = CLCLengthTab[selector];
258 for (cnt = 0; cnt < numCodes; cnt++) {
260 code = get_sbits(gb, numBits);
263 mantissas[cnt] = code;
266 for (cnt = 0; cnt < numCodes; cnt++) {
268 code = get_bits(gb, numBits); //numBits is always 4 in this case
271 mantissas[cnt*2] = seTab_0[code >> 2];
272 mantissas[cnt*2+1] = seTab_0[code & 3];
276 /* variable length coding (VLC) */
278 for (cnt = 0; cnt < numCodes; cnt++) {
279 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
281 code = huffSymb >> 1;
284 mantissas[cnt] = code;
287 for (cnt = 0; cnt < numCodes; cnt++) {
288 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
289 mantissas[cnt*2] = decTable1[huffSymb*2];
290 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
297 * Restore the quantized band spectrum coefficients
299 * @param gb the GetBit context
300 * @param pOut decoded band spectrum
301 * @return outSubbands subband counter, fix for broken specification/files
304 static int decodeSpectrum (GetBitContext *gb, float *pOut)
306 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
307 int subband_vlc_index[32], SF_idxs[32];
311 numSubbands = get_bits(gb, 5); // number of coded subbands
312 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
314 /* Get the VLC selector table for the subbands, 0 means not coded. */
315 for (cnt = 0; cnt <= numSubbands; cnt++)
316 subband_vlc_index[cnt] = get_bits(gb, 3);
318 /* Read the scale factor indexes from the stream. */
319 for (cnt = 0; cnt <= numSubbands; cnt++) {
320 if (subband_vlc_index[cnt] != 0)
321 SF_idxs[cnt] = get_bits(gb, 6);
324 for (cnt = 0; cnt <= numSubbands; cnt++) {
325 first = subbandTab[cnt];
326 last = subbandTab[cnt+1];
328 subbWidth = last - first;
330 if (subband_vlc_index[cnt] != 0) {
331 /* Decode spectral coefficients for this subband. */
332 /* TODO: This can be done faster is several blocks share the
333 * same VLC selector (subband_vlc_index) */
334 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
336 /* Decode the scale factor for this subband. */
337 SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
339 /* Inverse quantize the coefficients. */
340 for (pIn=mantissas ; first<last; first++, pIn++)
341 pOut[first] = *pIn * SF;
343 /* This subband was not coded, so zero the entire subband. */
344 memset(pOut+first, 0, subbWidth*sizeof(float));
348 /* Clear the subbands that were not coded. */
349 first = subbandTab[cnt];
350 memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
355 * Restore the quantized tonal components
357 * @param gb the GetBit context
358 * @param pComponent tone component
359 * @param numBands amount of coded bands
362 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
365 int components, coding_mode_selector, coding_mode, coded_values_per_component;
366 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
367 int band_flags[4], mantissa[8];
370 int component_count = 0;
372 components = get_bits(gb,5);
374 /* no tonal components */
378 coding_mode_selector = get_bits(gb,2);
379 if (coding_mode_selector == 2)
380 return AVERROR_INVALIDDATA;
382 coding_mode = coding_mode_selector & 1;
384 for (i = 0; i < components; i++) {
385 for (cnt = 0; cnt <= numBands; cnt++)
386 band_flags[cnt] = get_bits1(gb);
388 coded_values_per_component = get_bits(gb,3);
390 quant_step_index = get_bits(gb,3);
391 if (quant_step_index <= 1)
392 return AVERROR_INVALIDDATA;
394 if (coding_mode_selector == 3)
395 coding_mode = get_bits1(gb);
397 for (j = 0; j < (numBands + 1) * 4; j++) {
398 if (band_flags[j >> 2] == 0)
401 coded_components = get_bits(gb,3);
403 for (k=0; k<coded_components; k++) {
404 sfIndx = get_bits(gb,6);
405 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
406 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
407 coded_values = coded_values_per_component + 1;
408 coded_values = FFMIN(max_coded_values,coded_values);
410 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
412 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
414 pComponent[component_count].numCoefs = coded_values;
417 pCoef = pComponent[component_count].coef;
418 for (cnt = 0; cnt < coded_values; cnt++)
419 pCoef[cnt] = mantissa[cnt] * scalefactor;
426 return component_count;
430 * Decode gain parameters for the coded bands
432 * @param gb the GetBit context
433 * @param pGb the gainblock for the current band
434 * @param numBands amount of coded bands
437 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
442 gain_info *pGain = pGb->gBlock;
444 for (i=0 ; i<=numBands; i++)
446 numData = get_bits(gb,3);
447 pGain[i].num_gain_data = numData;
448 pLevel = pGain[i].levcode;
449 pLoc = pGain[i].loccode;
451 for (cf = 0; cf < numData; cf++){
452 pLevel[cf]= get_bits(gb,4);
453 pLoc [cf]= get_bits(gb,5);
454 if(cf && pLoc[cf] <= pLoc[cf-1])
455 return AVERROR_INVALIDDATA;
459 /* Clear the unused blocks. */
461 pGain[i].num_gain_data = 0;
467 * Apply gain parameters and perform the MDCT overlapping part
469 * @param pIn input float buffer
470 * @param pPrev previous float buffer to perform overlap against
471 * @param pOut output float buffer
472 * @param pGain1 current band gain info
473 * @param pGain2 next band gain info
476 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
478 /* gain compensation function */
479 float gain1, gain2, gain_inc;
480 int cnt, numdata, nsample, startLoc, endLoc;
483 if (pGain2->num_gain_data == 0)
486 gain1 = gain_tab1[pGain2->levcode[0]];
488 if (pGain1->num_gain_data == 0) {
489 for (cnt = 0; cnt < 256; cnt++)
490 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
492 numdata = pGain1->num_gain_data;
493 pGain1->loccode[numdata] = 32;
494 pGain1->levcode[numdata] = 4;
496 nsample = 0; // current sample = 0
498 for (cnt = 0; cnt < numdata; cnt++) {
499 startLoc = pGain1->loccode[cnt] * 8;
500 endLoc = startLoc + 8;
502 gain2 = gain_tab1[pGain1->levcode[cnt]];
503 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
506 for (; nsample < startLoc; nsample++)
507 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
509 /* interpolation is done over eight samples */
510 for (; nsample < endLoc; nsample++) {
511 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
516 for (; nsample < 256; nsample++)
517 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
520 /* Delay for the overlapping part. */
521 memcpy(pPrev, &pIn[256], 256*sizeof(float));
525 * Combine the tonal band spectrum and regular band spectrum
526 * Return position of the last tonal coefficient
528 * @param pSpectrum output spectrum buffer
529 * @param numComponents amount of tonal components
530 * @param pComponent tonal components for this band
533 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
535 int cnt, i, lastPos = -1;
538 for (cnt = 0; cnt < numComponents; cnt++){
539 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
540 pIn = pComponent[cnt].coef;
541 pOut = &(pSpectrum[pComponent[cnt].pos]);
543 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
551 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
553 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
555 int i, band, nsample, s1, s2;
557 float mc1_l, mc1_r, mc2_l, mc2_r;
559 for (i=0,band = 0; band < 4*256; band+=256,i++) {
565 /* Selector value changed, interpolation needed. */
566 mc1_l = matrixCoeffs[s1*2];
567 mc1_r = matrixCoeffs[s1*2+1];
568 mc2_l = matrixCoeffs[s2*2];
569 mc2_r = matrixCoeffs[s2*2+1];
571 /* Interpolation is done over the first eight samples. */
572 for(; nsample < 8; nsample++) {
573 c1 = su1[band+nsample];
574 c2 = su2[band+nsample];
575 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
576 su1[band+nsample] = c2;
577 su2[band+nsample] = c1 * 2.0 - c2;
581 /* Apply the matrix without interpolation. */
583 case 0: /* M/S decoding */
584 for (; nsample < 256; nsample++) {
585 c1 = su1[band+nsample];
586 c2 = su2[band+nsample];
587 su1[band+nsample] = c2 * 2.0;
588 su2[band+nsample] = (c1 - c2) * 2.0;
593 for (; nsample < 256; nsample++) {
594 c1 = su1[band+nsample];
595 c2 = su2[band+nsample];
596 su1[band+nsample] = (c1 + c2) * 2.0;
597 su2[band+nsample] = c2 * -2.0;
602 for (; nsample < 256; nsample++) {
603 c1 = su1[band+nsample];
604 c2 = su2[band+nsample];
605 su1[band+nsample] = c1 + c2;
606 su2[band+nsample] = c1 - c2;
615 static void getChannelWeights (int indx, int flag, float ch[2]){
621 ch[0] = (float)(indx & 7) / 7.0;
622 ch[1] = sqrt(2 - ch[0]*ch[0]);
624 FFSWAP(float, ch[0], ch[1]);
628 static void channelWeighting (float *su1, float *su2, int *p3)
631 /* w[x][y] y=0 is left y=1 is right */
634 if (p3[1] != 7 || p3[3] != 7){
635 getChannelWeights(p3[1], p3[0], w[0]);
636 getChannelWeights(p3[3], p3[2], w[1]);
638 for(band = 1; band < 4; band++) {
639 /* scale the channels by the weights */
640 for(nsample = 0; nsample < 8; nsample++) {
641 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
642 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
645 for(; nsample < 256; nsample++) {
646 su1[band*256+nsample] *= w[1][0];
647 su2[band*256+nsample] *= w[1][1];
655 * Decode a Sound Unit
657 * @param gb the GetBit context
658 * @param pSnd the channel unit to be used
659 * @param pOut the decoded samples before IQMF in float representation
660 * @param channelNum channel number
661 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
665 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
667 int band, result=0, numSubbands, lastTonal, numBands;
669 if (codingMode == JOINT_STEREO && channelNum == 1) {
670 if (get_bits(gb,2) != 3) {
671 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
672 return AVERROR_INVALIDDATA;
675 if (get_bits(gb,6) != 0x28) {
676 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
677 return AVERROR_INVALIDDATA;
681 /* number of coded QMF bands */
682 pSnd->bandsCoded = get_bits(gb,2);
684 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
685 if (result) return result;
687 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
688 if (pSnd->numComponents == -1) return -1;
690 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
692 /* Merge the decoded spectrum and tonal components. */
693 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
696 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
697 numBands = (subbandTab[numSubbands] - 1) >> 8;
699 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
702 /* Reconstruct time domain samples. */
703 for (band=0; band<4; band++) {
704 /* Perform the IMDCT step without overlapping. */
705 if (band <= numBands) {
706 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
708 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
710 /* gain compensation and overlapping */
711 gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
713 &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
714 &pSnd->gainBlock[ pSnd->gcBlkSwitch].gBlock[band]);
717 /* Swap the gain control buffers for the next frame. */
718 pSnd->gcBlkSwitch ^= 1;
726 * @param q Atrac3 private context
727 * @param databuf the input data
730 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
734 float *p1, *p2, *p3, *p4;
737 if (q->codingMode == JOINT_STEREO) {
739 /* channel coupling mode */
740 /* decode Sound Unit 1 */
741 init_get_bits(&q->gb,databuf,q->bits_per_frame);
743 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
747 /* Framedata of the su2 in the joint-stereo mode is encoded in
748 * reverse byte order so we need to swap it first. */
749 if (databuf == q->decoded_bytes_buffer) {
750 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
751 ptr1 = q->decoded_bytes_buffer;
752 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
753 FFSWAP(uint8_t,*ptr1,*ptr2);
756 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
757 for (i = 0; i < q->bytes_per_frame; i++)
758 q->decoded_bytes_buffer[i] = *ptr2--;
761 /* Skip the sync codes (0xF8). */
762 ptr1 = q->decoded_bytes_buffer;
763 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
764 if (i >= q->bytes_per_frame)
765 return AVERROR_INVALIDDATA;
769 /* set the bitstream reader at the start of the second Sound Unit*/
770 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
772 /* Fill the Weighting coeffs delay buffer */
773 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
774 q->weighting_delay[4] = get_bits1(&q->gb);
775 q->weighting_delay[5] = get_bits(&q->gb,3);
777 for (i = 0; i < 4; i++) {
778 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
779 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
780 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
783 /* Decode Sound Unit 2. */
784 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
788 /* Reconstruct the channel coefficients. */
789 reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
791 channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
794 /* normal stereo mode or mono */
795 /* Decode the channel sound units. */
796 for (i=0 ; i<q->channels ; i++) {
798 /* Set the bitstream reader at the start of a channel sound unit. */
799 init_get_bits(&q->gb,
800 databuf + i * q->bytes_per_frame / q->channels,
801 q->bits_per_frame / q->channels);
803 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
809 /* Apply the iQMF synthesis filter. */
810 for (i=0 ; i<q->channels ; i++) {
815 atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
816 atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
817 atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
825 * Atrac frame decoding
827 * @param avctx pointer to the AVCodecContext
830 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
831 int *got_frame_ptr, AVPacket *avpkt)
833 const uint8_t *buf = avpkt->data;
834 int buf_size = avpkt->size;
835 ATRAC3Context *q = avctx->priv_data;
837 const uint8_t* databuf;
839 int16_t *samples_s16;
841 if (buf_size < avctx->block_align) {
842 av_log(avctx, AV_LOG_ERROR,
843 "Frame too small (%d bytes). Truncated file?\n", buf_size);
844 return AVERROR_INVALIDDATA;
847 /* get output buffer */
848 q->frame.nb_samples = SAMPLES_PER_FRAME;
849 if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
850 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
853 samples_flt = (float *)q->frame.data[0];
854 samples_s16 = (int16_t *)q->frame.data[0];
856 /* Check if we need to descramble and what buffer to pass on. */
857 if (q->scrambled_stream) {
858 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
859 databuf = q->decoded_bytes_buffer;
864 if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
865 result = decodeFrame(q, databuf, &samples_flt);
867 result = decodeFrame(q, databuf, q->outSamples);
870 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
875 if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
876 q->fmt_conv.float_interleave(samples_flt,
877 (const float **)q->outSamples,
878 SAMPLES_PER_FRAME, 2);
879 } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
880 q->fmt_conv.float_to_int16_interleave(samples_s16,
881 (const float **)q->outSamples,
882 SAMPLES_PER_FRAME, q->channels);
886 *(AVFrame *)data = q->frame;
888 return avctx->block_align;
893 * Atrac3 initialization
895 * @param avctx pointer to the AVCodecContext
898 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
901 const uint8_t *edata_ptr = avctx->extradata;
902 ATRAC3Context *q = avctx->priv_data;
903 static VLC_TYPE atrac3_vlc_table[4096][2];
904 static int vlcs_initialized = 0;
906 /* Take data from the AVCodecContext (RM container). */
907 q->sample_rate = avctx->sample_rate;
908 q->channels = avctx->channels;
909 q->bit_rate = avctx->bit_rate;
910 q->bits_per_frame = avctx->block_align * 8;
911 q->bytes_per_frame = avctx->block_align;
913 /* Take care of the codec-specific extradata. */
914 if (avctx->extradata_size == 14) {
915 /* Parse the extradata, WAV format */
916 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
917 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
918 q->codingMode = bytestream_get_le16(&edata_ptr);
919 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
920 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
921 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
924 q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
925 q->atrac3version = 4;
928 q->codingMode = JOINT_STEREO;
930 q->codingMode = STEREO;
932 q->scrambled_stream = 0;
934 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)) {
936 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);
937 return AVERROR_INVALIDDATA;
940 } else if (avctx->extradata_size == 10) {
941 /* Parse the extradata, RM format. */
942 q->atrac3version = bytestream_get_be32(&edata_ptr);
943 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
944 q->delay = bytestream_get_be16(&edata_ptr);
945 q->codingMode = bytestream_get_be16(&edata_ptr);
947 q->samples_per_channel = q->samples_per_frame / q->channels;
948 q->scrambled_stream = 1;
951 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
953 /* Check the extradata. */
955 if (q->atrac3version != 4) {
956 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
957 return AVERROR_INVALIDDATA;
960 if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
961 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
962 return AVERROR_INVALIDDATA;
965 if (q->delay != 0x88E) {
966 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
967 return AVERROR_INVALIDDATA;
970 if (q->codingMode == STEREO) {
971 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
972 } else if (q->codingMode == JOINT_STEREO) {
973 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
975 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
976 return AVERROR_INVALIDDATA;
979 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
980 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
981 return AVERROR(EINVAL);
985 if(avctx->block_align >= UINT_MAX/2)
986 return AVERROR(EINVAL);
988 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
989 * this is for the bitstream reader. */
990 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
991 return AVERROR(ENOMEM);
994 /* Initialize the VLC tables. */
995 if (!vlcs_initialized) {
996 for (i=0 ; i<7 ; i++) {
997 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
998 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
999 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
1001 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
1003 vlcs_initialized = 1;
1006 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
1007 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1009 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1011 if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
1012 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
1013 av_freep(&q->decoded_bytes_buffer);
1017 atrac_generate_tables();
1019 /* Generate gain tables. */
1020 for (i=0 ; i<16 ; i++)
1021 gain_tab1[i] = powf (2.0, (4 - i));
1023 for (i=-15 ; i<16 ; i++)
1024 gain_tab2[i+15] = powf (2.0, i * -0.125);
1026 /* init the joint-stereo decoding data */
1027 q->weighting_delay[0] = 0;
1028 q->weighting_delay[1] = 7;
1029 q->weighting_delay[2] = 0;
1030 q->weighting_delay[3] = 7;
1031 q->weighting_delay[4] = 0;
1032 q->weighting_delay[5] = 7;
1034 for (i=0; i<4; i++) {
1035 q->matrix_coeff_index_prev[i] = 3;
1036 q->matrix_coeff_index_now[i] = 3;
1037 q->matrix_coeff_index_next[i] = 3;
1040 dsputil_init(&dsp, avctx);
1041 ff_fmt_convert_init(&q->fmt_conv, avctx);
1043 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1045 atrac3_decode_close(avctx);
1046 return AVERROR(ENOMEM);
1049 if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
1050 q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
1051 q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
1052 if (!q->outSamples[0]) {
1053 atrac3_decode_close(avctx);
1054 return AVERROR(ENOMEM);
1058 avcodec_get_frame_defaults(&q->frame);
1059 avctx->coded_frame = &q->frame;
1065 AVCodec ff_atrac3_decoder =
1068 .type = AVMEDIA_TYPE_AUDIO,
1069 .id = CODEC_ID_ATRAC3,
1070 .priv_data_size = sizeof(ATRAC3Context),
1071 .init = atrac3_decode_init,
1072 .close = atrac3_decode_close,
1073 .decode = atrac3_decode_frame,
1074 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1075 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),