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 uint8_t* decoded_bytes_buffer;
122 int scrambled_stream;
127 FmtConvertContext fmt_conv;
128 AVFloatDSPContext fdsp;
131 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
132 static VLC spectral_coeff_tab[7];
133 static float gain_tab1[16];
134 static float gain_tab2[31];
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 q->fdsp.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) {
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, 1.0 / 32768);
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);
230 ff_mdct_end(&q->mdct_ctx);
236 / * Mantissa decoding
238 * @param gb the GetBit context
239 * @param selector what table is the output values coded with
240 * @param codingFlag constant length coding or variable length coding
241 * @param mantissas mantissa output table
242 * @param numCodes amount of values to get
245 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
247 int numBits, cnt, code, huffSymb;
252 if (codingFlag != 0) {
253 /* constant length coding (CLC) */
254 numBits = CLCLengthTab[selector];
257 for (cnt = 0; cnt < numCodes; cnt++) {
259 code = get_sbits(gb, numBits);
262 mantissas[cnt] = code;
265 for (cnt = 0; cnt < numCodes; cnt++) {
267 code = get_bits(gb, numBits); //numBits is always 4 in this case
270 mantissas[cnt*2] = seTab_0[code >> 2];
271 mantissas[cnt*2+1] = seTab_0[code & 3];
275 /* variable length coding (VLC) */
277 for (cnt = 0; cnt < numCodes; cnt++) {
278 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
280 code = huffSymb >> 1;
283 mantissas[cnt] = code;
286 for (cnt = 0; cnt < numCodes; cnt++) {
287 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
288 mantissas[cnt*2] = decTable1[huffSymb*2];
289 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
296 * Restore the quantized band spectrum coefficients
298 * @param gb the GetBit context
299 * @param pOut decoded band spectrum
300 * @return outSubbands subband counter, fix for broken specification/files
303 static int decodeSpectrum (GetBitContext *gb, float *pOut)
305 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
306 int subband_vlc_index[32], SF_idxs[32];
310 numSubbands = get_bits(gb, 5); // number of coded subbands
311 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
313 /* Get the VLC selector table for the subbands, 0 means not coded. */
314 for (cnt = 0; cnt <= numSubbands; cnt++)
315 subband_vlc_index[cnt] = get_bits(gb, 3);
317 /* Read the scale factor indexes from the stream. */
318 for (cnt = 0; cnt <= numSubbands; cnt++) {
319 if (subband_vlc_index[cnt] != 0)
320 SF_idxs[cnt] = get_bits(gb, 6);
323 for (cnt = 0; cnt <= numSubbands; cnt++) {
324 first = subbandTab[cnt];
325 last = subbandTab[cnt+1];
327 subbWidth = last - first;
329 if (subband_vlc_index[cnt] != 0) {
330 /* Decode spectral coefficients for this subband. */
331 /* TODO: This can be done faster is several blocks share the
332 * same VLC selector (subband_vlc_index) */
333 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
335 /* Decode the scale factor for this subband. */
336 SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
338 /* Inverse quantize the coefficients. */
339 for (pIn=mantissas ; first<last; first++, pIn++)
340 pOut[first] = *pIn * SF;
342 /* This subband was not coded, so zero the entire subband. */
343 memset(pOut+first, 0, subbWidth*sizeof(float));
347 /* Clear the subbands that were not coded. */
348 first = subbandTab[cnt];
349 memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
354 * Restore the quantized tonal components
356 * @param gb the GetBit context
357 * @param pComponent tone component
358 * @param numBands amount of coded bands
361 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
364 int components, coding_mode_selector, coding_mode, coded_values_per_component;
365 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
366 int band_flags[4], mantissa[8];
369 int component_count = 0;
371 components = get_bits(gb,5);
373 /* no tonal components */
377 coding_mode_selector = get_bits(gb,2);
378 if (coding_mode_selector == 2)
379 return AVERROR_INVALIDDATA;
381 coding_mode = coding_mode_selector & 1;
383 for (i = 0; i < components; i++) {
384 for (cnt = 0; cnt <= numBands; cnt++)
385 band_flags[cnt] = get_bits1(gb);
387 coded_values_per_component = get_bits(gb,3);
389 quant_step_index = get_bits(gb,3);
390 if (quant_step_index <= 1)
391 return AVERROR_INVALIDDATA;
393 if (coding_mode_selector == 3)
394 coding_mode = get_bits1(gb);
396 for (j = 0; j < (numBands + 1) * 4; j++) {
397 if (band_flags[j >> 2] == 0)
400 coded_components = get_bits(gb,3);
402 for (k=0; k<coded_components; k++) {
403 sfIndx = get_bits(gb,6);
404 if (component_count >= 64)
405 return AVERROR_INVALIDDATA;
406 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
407 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
408 coded_values = coded_values_per_component + 1;
409 coded_values = FFMIN(max_coded_values,coded_values);
411 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
413 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
415 pComponent[component_count].numCoefs = coded_values;
418 pCoef = pComponent[component_count].coef;
419 for (cnt = 0; cnt < coded_values; cnt++)
420 pCoef[cnt] = mantissa[cnt] * scalefactor;
427 return component_count;
431 * Decode gain parameters for the coded bands
433 * @param gb the GetBit context
434 * @param pGb the gainblock for the current band
435 * @param numBands amount of coded bands
438 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
443 gain_info *pGain = pGb->gBlock;
445 for (i=0 ; i<=numBands; i++)
447 numData = get_bits(gb,3);
448 pGain[i].num_gain_data = numData;
449 pLevel = pGain[i].levcode;
450 pLoc = pGain[i].loccode;
452 for (cf = 0; cf < numData; cf++){
453 pLevel[cf]= get_bits(gb,4);
454 pLoc [cf]= get_bits(gb,5);
455 if(cf && pLoc[cf] <= pLoc[cf-1])
456 return AVERROR_INVALIDDATA;
460 /* Clear the unused blocks. */
462 pGain[i].num_gain_data = 0;
468 * Apply gain parameters and perform the MDCT overlapping part
470 * @param pIn input float buffer
471 * @param pPrev previous float buffer to perform overlap against
472 * @param pOut output float buffer
473 * @param pGain1 current band gain info
474 * @param pGain2 next band gain info
477 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
479 /* gain compensation function */
480 float gain1, gain2, gain_inc;
481 int cnt, numdata, nsample, startLoc, endLoc;
484 if (pGain2->num_gain_data == 0)
487 gain1 = gain_tab1[pGain2->levcode[0]];
489 if (pGain1->num_gain_data == 0) {
490 for (cnt = 0; cnt < 256; cnt++)
491 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
493 numdata = pGain1->num_gain_data;
494 pGain1->loccode[numdata] = 32;
495 pGain1->levcode[numdata] = 4;
497 nsample = 0; // current sample = 0
499 for (cnt = 0; cnt < numdata; cnt++) {
500 startLoc = pGain1->loccode[cnt] * 8;
501 endLoc = startLoc + 8;
503 gain2 = gain_tab1[pGain1->levcode[cnt]];
504 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
507 for (; nsample < startLoc; nsample++)
508 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
510 /* interpolation is done over eight samples */
511 for (; nsample < endLoc; nsample++) {
512 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
517 for (; nsample < 256; nsample++)
518 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
521 /* Delay for the overlapping part. */
522 memcpy(pPrev, &pIn[256], 256*sizeof(float));
526 * Combine the tonal band spectrum and regular band spectrum
527 * Return position of the last tonal coefficient
529 * @param pSpectrum output spectrum buffer
530 * @param numComponents amount of tonal components
531 * @param pComponent tonal components for this band
534 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
536 int cnt, i, lastPos = -1;
539 for (cnt = 0; cnt < numComponents; cnt++){
540 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
541 pIn = pComponent[cnt].coef;
542 pOut = &(pSpectrum[pComponent[cnt].pos]);
544 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
552 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
554 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
556 int i, band, nsample, s1, s2;
558 float mc1_l, mc1_r, mc2_l, mc2_r;
560 for (i=0,band = 0; band < 4*256; band+=256,i++) {
566 /* Selector value changed, interpolation needed. */
567 mc1_l = matrixCoeffs[s1*2];
568 mc1_r = matrixCoeffs[s1*2+1];
569 mc2_l = matrixCoeffs[s2*2];
570 mc2_r = matrixCoeffs[s2*2+1];
572 /* Interpolation is done over the first eight samples. */
573 for(; nsample < 8; nsample++) {
574 c1 = su1[band+nsample];
575 c2 = su2[band+nsample];
576 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
577 su1[band+nsample] = c2;
578 su2[band+nsample] = c1 * 2.0 - c2;
582 /* Apply the matrix without interpolation. */
584 case 0: /* M/S decoding */
585 for (; nsample < 256; nsample++) {
586 c1 = su1[band+nsample];
587 c2 = su2[band+nsample];
588 su1[band+nsample] = c2 * 2.0;
589 su2[band+nsample] = (c1 - c2) * 2.0;
594 for (; nsample < 256; nsample++) {
595 c1 = su1[band+nsample];
596 c2 = su2[band+nsample];
597 su1[band+nsample] = (c1 + c2) * 2.0;
598 su2[band+nsample] = c2 * -2.0;
603 for (; nsample < 256; nsample++) {
604 c1 = su1[band+nsample];
605 c2 = su2[band+nsample];
606 su1[band+nsample] = c1 + c2;
607 su2[band+nsample] = c1 - c2;
616 static void getChannelWeights (int indx, int flag, float ch[2]){
622 ch[0] = (float)(indx & 7) / 7.0;
623 ch[1] = sqrt(2 - ch[0]*ch[0]);
625 FFSWAP(float, ch[0], ch[1]);
629 static void channelWeighting (float *su1, float *su2, int *p3)
632 /* w[x][y] y=0 is left y=1 is right */
635 if (p3[1] != 7 || p3[3] != 7){
636 getChannelWeights(p3[1], p3[0], w[0]);
637 getChannelWeights(p3[3], p3[2], w[1]);
639 for(band = 1; band < 4; band++) {
640 /* scale the channels by the weights */
641 for(nsample = 0; nsample < 8; nsample++) {
642 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
643 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
646 for(; nsample < 256; nsample++) {
647 su1[band*256+nsample] *= w[1][0];
648 su2[band*256+nsample] *= w[1][1];
656 * Decode a Sound Unit
658 * @param gb the GetBit context
659 * @param pSnd the channel unit to be used
660 * @param pOut the decoded samples before IQMF in float representation
661 * @param channelNum channel number
662 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
666 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
668 int band, result=0, numSubbands, lastTonal, numBands;
670 if (codingMode == JOINT_STEREO && channelNum == 1) {
671 if (get_bits(gb,2) != 3) {
672 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
673 return AVERROR_INVALIDDATA;
676 if (get_bits(gb,6) != 0x28) {
677 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
678 return AVERROR_INVALIDDATA;
682 /* number of coded QMF bands */
683 pSnd->bandsCoded = get_bits(gb,2);
685 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
686 if (result) return result;
688 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
689 if (pSnd->numComponents == -1) return -1;
691 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
693 /* Merge the decoded spectrum and tonal components. */
694 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
697 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
698 numBands = (subbandTab[numSubbands] - 1) >> 8;
700 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
703 /* Reconstruct time domain samples. */
704 for (band=0; band<4; band++) {
705 /* Perform the IMDCT step without overlapping. */
706 if (band <= numBands) {
707 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
709 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
711 /* gain compensation and overlapping */
712 gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
714 &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
715 &pSnd->gainBlock[ pSnd->gcBlkSwitch].gBlock[band]);
718 /* Swap the gain control buffers for the next frame. */
719 pSnd->gcBlkSwitch ^= 1;
727 * @param q Atrac3 private context
728 * @param databuf the input data
731 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
735 float *p1, *p2, *p3, *p4;
738 if (q->codingMode == JOINT_STEREO) {
740 /* channel coupling mode */
741 /* decode Sound Unit 1 */
742 init_get_bits(&q->gb,databuf,q->bits_per_frame);
744 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
748 /* Framedata of the su2 in the joint-stereo mode is encoded in
749 * reverse byte order so we need to swap it first. */
750 if (databuf == q->decoded_bytes_buffer) {
751 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
752 ptr1 = q->decoded_bytes_buffer;
753 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
754 FFSWAP(uint8_t,*ptr1,*ptr2);
757 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
758 for (i = 0; i < q->bytes_per_frame; i++)
759 q->decoded_bytes_buffer[i] = *ptr2--;
762 /* Skip the sync codes (0xF8). */
763 ptr1 = q->decoded_bytes_buffer;
764 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
765 if (i >= q->bytes_per_frame)
766 return AVERROR_INVALIDDATA;
770 /* set the bitstream reader at the start of the second Sound Unit*/
771 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
773 /* Fill the Weighting coeffs delay buffer */
774 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
775 q->weighting_delay[4] = get_bits1(&q->gb);
776 q->weighting_delay[5] = get_bits(&q->gb,3);
778 for (i = 0; i < 4; i++) {
779 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
780 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
781 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
784 /* Decode Sound Unit 2. */
785 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
789 /* Reconstruct the channel coefficients. */
790 reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
792 channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
795 /* normal stereo mode or mono */
796 /* Decode the channel sound units. */
797 for (i=0 ; i<q->channels ; i++) {
799 /* Set the bitstream reader at the start of a channel sound unit. */
800 init_get_bits(&q->gb,
801 databuf + i * q->bytes_per_frame / q->channels,
802 q->bits_per_frame / q->channels);
804 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
810 /* Apply the iQMF synthesis filter. */
811 for (i=0 ; i<q->channels ; i++) {
816 ff_atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
817 ff_atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
818 ff_atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
826 * Atrac frame decoding
828 * @param avctx pointer to the AVCodecContext
831 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
832 int *got_frame_ptr, AVPacket *avpkt)
834 const uint8_t *buf = avpkt->data;
835 int buf_size = avpkt->size;
836 ATRAC3Context *q = avctx->priv_data;
838 const uint8_t* databuf;
840 if (buf_size < avctx->block_align) {
841 av_log(avctx, AV_LOG_ERROR,
842 "Frame too small (%d bytes). Truncated file?\n", buf_size);
843 return AVERROR_INVALIDDATA;
846 /* get output buffer */
847 q->frame.nb_samples = SAMPLES_PER_FRAME;
848 if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
849 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
853 /* Check if we need to descramble and what buffer to pass on. */
854 if (q->scrambled_stream) {
855 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
856 databuf = q->decoded_bytes_buffer;
861 result = decodeFrame(q, databuf, (float **)q->frame.extended_data);
864 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
869 *(AVFrame *)data = q->frame;
871 return avctx->block_align;
876 * Atrac3 initialization
878 * @param avctx pointer to the AVCodecContext
881 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
884 const uint8_t *edata_ptr = avctx->extradata;
885 ATRAC3Context *q = avctx->priv_data;
886 static VLC_TYPE atrac3_vlc_table[4096][2];
887 static int vlcs_initialized = 0;
889 /* Take data from the AVCodecContext (RM container). */
890 q->sample_rate = avctx->sample_rate;
891 q->channels = avctx->channels;
892 q->bit_rate = avctx->bit_rate;
893 q->bits_per_frame = avctx->block_align * 8;
894 q->bytes_per_frame = avctx->block_align;
896 /* Take care of the codec-specific extradata. */
897 if (avctx->extradata_size == 14) {
898 /* Parse the extradata, WAV format */
899 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
900 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
901 q->codingMode = bytestream_get_le16(&edata_ptr);
902 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
903 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
904 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
907 q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
908 q->atrac3version = 4;
911 q->codingMode = JOINT_STEREO;
913 q->codingMode = STEREO;
915 q->scrambled_stream = 0;
917 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)) {
919 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);
920 return AVERROR_INVALIDDATA;
923 } else if (avctx->extradata_size == 10) {
924 /* Parse the extradata, RM format. */
925 q->atrac3version = bytestream_get_be32(&edata_ptr);
926 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
927 q->delay = bytestream_get_be16(&edata_ptr);
928 q->codingMode = bytestream_get_be16(&edata_ptr);
930 q->samples_per_channel = q->samples_per_frame / q->channels;
931 q->scrambled_stream = 1;
934 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
936 /* Check the extradata. */
938 if (q->atrac3version != 4) {
939 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
940 return AVERROR_INVALIDDATA;
943 if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
944 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
945 return AVERROR_INVALIDDATA;
948 if (q->delay != 0x88E) {
949 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
950 return AVERROR_INVALIDDATA;
953 if (q->codingMode == STEREO) {
954 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
955 } else if (q->codingMode == JOINT_STEREO) {
956 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
958 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
959 return AVERROR_INVALIDDATA;
962 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
963 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
964 return AVERROR(EINVAL);
968 if(avctx->block_align >= UINT_MAX/2)
969 return AVERROR(EINVAL);
971 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
972 * this is for the bitstream reader. */
973 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
974 return AVERROR(ENOMEM);
977 /* Initialize the VLC tables. */
978 if (!vlcs_initialized) {
979 for (i=0 ; i<7 ; i++) {
980 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
981 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
982 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
984 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
986 vlcs_initialized = 1;
989 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
991 if ((ret = init_atrac3_transforms(q))) {
992 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
993 av_freep(&q->decoded_bytes_buffer);
997 ff_atrac_generate_tables();
999 /* Generate gain tables. */
1000 for (i=0 ; i<16 ; i++)
1001 gain_tab1[i] = exp2f (4 - i);
1003 for (i=-15 ; i<16 ; i++)
1004 gain_tab2[i+15] = exp2f (i * -0.125);
1006 /* init the joint-stereo decoding data */
1007 q->weighting_delay[0] = 0;
1008 q->weighting_delay[1] = 7;
1009 q->weighting_delay[2] = 0;
1010 q->weighting_delay[3] = 7;
1011 q->weighting_delay[4] = 0;
1012 q->weighting_delay[5] = 7;
1014 for (i=0; i<4; i++) {
1015 q->matrix_coeff_index_prev[i] = 3;
1016 q->matrix_coeff_index_now[i] = 3;
1017 q->matrix_coeff_index_next[i] = 3;
1020 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1021 ff_fmt_convert_init(&q->fmt_conv, avctx);
1023 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1025 atrac3_decode_close(avctx);
1026 return AVERROR(ENOMEM);
1029 avcodec_get_frame_defaults(&q->frame);
1030 avctx->coded_frame = &q->frame;
1036 AVCodec ff_atrac3_decoder =
1039 .type = AVMEDIA_TYPE_AUDIO,
1040 .id = AV_CODEC_ID_ATRAC3,
1041 .priv_data_size = sizeof(ATRAC3Context),
1042 .init = atrac3_decode_init,
1043 .close = atrac3_decode_close,
1044 .decode = atrac3_decode_frame,
1045 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1046 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1047 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1048 AV_SAMPLE_FMT_NONE },