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
96 int samples_per_channel;
97 int samples_per_frame;
102 channel_unit* pUnits;
105 /** joint-stereo related variables */
106 int matrix_coeff_index_prev[4];
107 int matrix_coeff_index_now[4];
108 int matrix_coeff_index_next[4];
109 int weighting_delay[6];
113 float *outSamples[2];
114 uint8_t* decoded_bytes_buffer;
121 int scrambled_stream;
126 FmtConvertContext fmt_conv;
129 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
130 static VLC spectral_coeff_tab[7];
131 static float gain_tab1[16];
132 static float gain_tab2[31];
133 static DSPContext dsp;
137 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
138 * caused by the reverse spectra of the QMF.
140 * @param pInput float input
141 * @param pOutput float output
142 * @param odd_band 1 if the band is an odd band
145 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
151 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
152 * or it gives better compression to do it this way.
153 * FIXME: It should be possible to handle this in imdct_calc
154 * for that to happen a modification of the prerotation step of
155 * all SIMD code and C code is needed.
156 * Or fix the functions before so they generate a pre reversed spectrum.
159 for (i=0; i<128; i++)
160 FFSWAP(float, pInput[i], pInput[255-i]);
163 q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
165 /* Perform windowing on the output. */
166 dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
172 * Atrac 3 indata descrambling, only used for data coming from the rm container
174 * @param inbuffer pointer to 8 bit array of indata
175 * @param out pointer to 8 bit array of outdata
176 * @param bytes amount of bytes
179 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
183 uint32_t* obuf = (uint32_t*) out;
185 off = (intptr_t)inbuffer & 3;
186 buf = (const uint32_t*) (inbuffer - off);
187 c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
189 for (i = 0; i < bytes/4; i++)
190 obuf[i] = c ^ buf[i];
193 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
199 static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
200 float enc_window[256];
203 /* Generate the mdct window, for details see
204 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
205 for (i=0 ; i<256; i++)
206 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
209 for (i=0 ; i<256; i++) {
210 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
211 mdct_window[511-i] = mdct_window[i];
214 /* Initialize the MDCT transform. */
215 return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
219 * Atrac3 uninit, free all allocated memory
222 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
224 ATRAC3Context *q = avctx->priv_data;
227 av_free(q->decoded_bytes_buffer);
228 av_freep(&q->outSamples[0]);
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 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
405 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
406 coded_values = coded_values_per_component + 1;
407 coded_values = FFMIN(max_coded_values,coded_values);
409 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
411 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
413 pComponent[component_count].numCoefs = coded_values;
416 pCoef = pComponent[component_count].coef;
417 for (cnt = 0; cnt < coded_values; cnt++)
418 pCoef[cnt] = mantissa[cnt] * scalefactor;
425 return component_count;
429 * Decode gain parameters for the coded bands
431 * @param gb the GetBit context
432 * @param pGb the gainblock for the current band
433 * @param numBands amount of coded bands
436 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
441 gain_info *pGain = pGb->gBlock;
443 for (i=0 ; i<=numBands; i++)
445 numData = get_bits(gb,3);
446 pGain[i].num_gain_data = numData;
447 pLevel = pGain[i].levcode;
448 pLoc = pGain[i].loccode;
450 for (cf = 0; cf < numData; cf++){
451 pLevel[cf]= get_bits(gb,4);
452 pLoc [cf]= get_bits(gb,5);
453 if(cf && pLoc[cf] <= pLoc[cf-1])
454 return AVERROR_INVALIDDATA;
458 /* Clear the unused blocks. */
460 pGain[i].num_gain_data = 0;
466 * Apply gain parameters and perform the MDCT overlapping part
468 * @param pIn input float buffer
469 * @param pPrev previous float buffer to perform overlap against
470 * @param pOut output float buffer
471 * @param pGain1 current band gain info
472 * @param pGain2 next band gain info
475 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
477 /* gain compensation function */
478 float gain1, gain2, gain_inc;
479 int cnt, numdata, nsample, startLoc, endLoc;
482 if (pGain2->num_gain_data == 0)
485 gain1 = gain_tab1[pGain2->levcode[0]];
487 if (pGain1->num_gain_data == 0) {
488 for (cnt = 0; cnt < 256; cnt++)
489 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
491 numdata = pGain1->num_gain_data;
492 pGain1->loccode[numdata] = 32;
493 pGain1->levcode[numdata] = 4;
495 nsample = 0; // current sample = 0
497 for (cnt = 0; cnt < numdata; cnt++) {
498 startLoc = pGain1->loccode[cnt] * 8;
499 endLoc = startLoc + 8;
501 gain2 = gain_tab1[pGain1->levcode[cnt]];
502 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
505 for (; nsample < startLoc; nsample++)
506 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
508 /* interpolation is done over eight samples */
509 for (; nsample < endLoc; nsample++) {
510 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
515 for (; nsample < 256; nsample++)
516 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
519 /* Delay for the overlapping part. */
520 memcpy(pPrev, &pIn[256], 256*sizeof(float));
524 * Combine the tonal band spectrum and regular band spectrum
525 * Return position of the last tonal coefficient
527 * @param pSpectrum output spectrum buffer
528 * @param numComponents amount of tonal components
529 * @param pComponent tonal components for this band
532 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
534 int cnt, i, lastPos = -1;
537 for (cnt = 0; cnt < numComponents; cnt++){
538 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
539 pIn = pComponent[cnt].coef;
540 pOut = &(pSpectrum[pComponent[cnt].pos]);
542 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
550 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
552 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
554 int i, band, nsample, s1, s2;
556 float mc1_l, mc1_r, mc2_l, mc2_r;
558 for (i=0,band = 0; band < 4*256; band+=256,i++) {
564 /* Selector value changed, interpolation needed. */
565 mc1_l = matrixCoeffs[s1*2];
566 mc1_r = matrixCoeffs[s1*2+1];
567 mc2_l = matrixCoeffs[s2*2];
568 mc2_r = matrixCoeffs[s2*2+1];
570 /* Interpolation is done over the first eight samples. */
571 for(; nsample < 8; nsample++) {
572 c1 = su1[band+nsample];
573 c2 = su2[band+nsample];
574 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
575 su1[band+nsample] = c2;
576 su2[band+nsample] = c1 * 2.0 - c2;
580 /* Apply the matrix without interpolation. */
582 case 0: /* M/S decoding */
583 for (; nsample < 256; nsample++) {
584 c1 = su1[band+nsample];
585 c2 = su2[band+nsample];
586 su1[band+nsample] = c2 * 2.0;
587 su2[band+nsample] = (c1 - c2) * 2.0;
592 for (; nsample < 256; nsample++) {
593 c1 = su1[band+nsample];
594 c2 = su2[band+nsample];
595 su1[band+nsample] = (c1 + c2) * 2.0;
596 su2[band+nsample] = c2 * -2.0;
601 for (; nsample < 256; nsample++) {
602 c1 = su1[band+nsample];
603 c2 = su2[band+nsample];
604 su1[band+nsample] = c1 + c2;
605 su2[band+nsample] = c1 - c2;
614 static void getChannelWeights (int indx, int flag, float ch[2]){
620 ch[0] = (float)(indx & 7) / 7.0;
621 ch[1] = sqrt(2 - ch[0]*ch[0]);
623 FFSWAP(float, ch[0], ch[1]);
627 static void channelWeighting (float *su1, float *su2, int *p3)
630 /* w[x][y] y=0 is left y=1 is right */
633 if (p3[1] != 7 || p3[3] != 7){
634 getChannelWeights(p3[1], p3[0], w[0]);
635 getChannelWeights(p3[3], p3[2], w[1]);
637 for(band = 1; band < 4; band++) {
638 /* scale the channels by the weights */
639 for(nsample = 0; nsample < 8; nsample++) {
640 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
641 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
644 for(; nsample < 256; nsample++) {
645 su1[band*256+nsample] *= w[1][0];
646 su2[band*256+nsample] *= w[1][1];
654 * Decode a Sound Unit
656 * @param gb the GetBit context
657 * @param pSnd the channel unit to be used
658 * @param pOut the decoded samples before IQMF in float representation
659 * @param channelNum channel number
660 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
664 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
666 int band, result=0, numSubbands, lastTonal, numBands;
668 if (codingMode == JOINT_STEREO && channelNum == 1) {
669 if (get_bits(gb,2) != 3) {
670 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
671 return AVERROR_INVALIDDATA;
674 if (get_bits(gb,6) != 0x28) {
675 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
676 return AVERROR_INVALIDDATA;
680 /* number of coded QMF bands */
681 pSnd->bandsCoded = get_bits(gb,2);
683 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
684 if (result) return result;
686 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
687 if (pSnd->numComponents == -1) return -1;
689 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
691 /* Merge the decoded spectrum and tonal components. */
692 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
695 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
696 numBands = (subbandTab[numSubbands] - 1) >> 8;
698 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
701 /* Reconstruct time domain samples. */
702 for (band=0; band<4; band++) {
703 /* Perform the IMDCT step without overlapping. */
704 if (band <= numBands) {
705 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
707 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
709 /* gain compensation and overlapping */
710 gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]),
711 &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]),
712 &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band]));
715 /* Swap the gain control buffers for the next frame. */
716 pSnd->gcBlkSwitch ^= 1;
724 * @param q Atrac3 private context
725 * @param databuf the input data
728 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
732 float *p1, *p2, *p3, *p4;
735 if (q->codingMode == JOINT_STEREO) {
737 /* channel coupling mode */
738 /* decode Sound Unit 1 */
739 init_get_bits(&q->gb,databuf,q->bits_per_frame);
741 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
745 /* Framedata of the su2 in the joint-stereo mode is encoded in
746 * reverse byte order so we need to swap it first. */
747 if (databuf == q->decoded_bytes_buffer) {
748 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
749 ptr1 = q->decoded_bytes_buffer;
750 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
751 FFSWAP(uint8_t,*ptr1,*ptr2);
754 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
755 for (i = 0; i < q->bytes_per_frame; i++)
756 q->decoded_bytes_buffer[i] = *ptr2--;
759 /* Skip the sync codes (0xF8). */
760 ptr1 = q->decoded_bytes_buffer;
761 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
762 if (i >= q->bytes_per_frame)
763 return AVERROR_INVALIDDATA;
767 /* set the bitstream reader at the start of the second Sound Unit*/
768 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
770 /* Fill the Weighting coeffs delay buffer */
771 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
772 q->weighting_delay[4] = get_bits1(&q->gb);
773 q->weighting_delay[5] = get_bits(&q->gb,3);
775 for (i = 0; i < 4; i++) {
776 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
777 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
778 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
781 /* Decode Sound Unit 2. */
782 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
786 /* Reconstruct the channel coefficients. */
787 reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
789 channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
792 /* normal stereo mode or mono */
793 /* Decode the channel sound units. */
794 for (i=0 ; i<q->channels ; i++) {
796 /* Set the bitstream reader at the start of a channel sound unit. */
797 init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels);
799 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
805 /* Apply the iQMF synthesis filter. */
806 for (i=0 ; i<q->channels ; i++) {
811 atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
812 atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
813 atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
821 * Atrac frame decoding
823 * @param avctx pointer to the AVCodecContext
826 static int atrac3_decode_frame(AVCodecContext *avctx,
827 void *data, int *data_size,
829 const uint8_t *buf = avpkt->data;
830 int buf_size = avpkt->size;
831 ATRAC3Context *q = avctx->priv_data;
832 int result = 0, out_size;
833 const uint8_t* databuf;
834 float *samples_flt = data;
835 int16_t *samples_s16 = data;
837 if (buf_size < avctx->block_align) {
838 av_log(avctx, AV_LOG_ERROR,
839 "Frame too small (%d bytes). Truncated file?\n", buf_size);
840 return AVERROR_INVALIDDATA;
843 out_size = SAMPLES_PER_FRAME * q->channels *
844 av_get_bytes_per_sample(avctx->sample_fmt);
845 if (*data_size < out_size) {
846 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
847 return AVERROR(EINVAL);
850 /* Check if we need to descramble and what buffer to pass on. */
851 if (q->scrambled_stream) {
852 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
853 databuf = q->decoded_bytes_buffer;
858 if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
859 result = decodeFrame(q, databuf, &samples_flt);
861 result = decodeFrame(q, databuf, q->outSamples);
864 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
869 if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
870 q->fmt_conv.float_interleave(samples_flt,
871 (const float **)q->outSamples,
872 SAMPLES_PER_FRAME, 2);
873 } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
874 q->fmt_conv.float_to_int16_interleave(samples_s16,
875 (const float **)q->outSamples,
876 SAMPLES_PER_FRAME, q->channels);
878 *data_size = out_size;
880 return avctx->block_align;
885 * Atrac3 initialization
887 * @param avctx pointer to the AVCodecContext
890 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
893 const uint8_t *edata_ptr = avctx->extradata;
894 ATRAC3Context *q = avctx->priv_data;
895 static VLC_TYPE atrac3_vlc_table[4096][2];
896 static int vlcs_initialized = 0;
898 /* Take data from the AVCodecContext (RM container). */
899 q->sample_rate = avctx->sample_rate;
900 q->channels = avctx->channels;
901 q->bit_rate = avctx->bit_rate;
902 q->bits_per_frame = avctx->block_align * 8;
903 q->bytes_per_frame = avctx->block_align;
905 /* Take care of the codec-specific extradata. */
906 if (avctx->extradata_size == 14) {
907 /* Parse the extradata, WAV format */
908 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
909 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
910 q->codingMode = bytestream_get_le16(&edata_ptr);
911 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
912 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
913 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
916 q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
917 q->atrac3version = 4;
920 q->codingMode = JOINT_STEREO;
922 q->codingMode = STEREO;
924 q->scrambled_stream = 0;
926 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)) {
928 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);
929 return AVERROR_INVALIDDATA;
932 } else if (avctx->extradata_size == 10) {
933 /* Parse the extradata, RM format. */
934 q->atrac3version = bytestream_get_be32(&edata_ptr);
935 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
936 q->delay = bytestream_get_be16(&edata_ptr);
937 q->codingMode = bytestream_get_be16(&edata_ptr);
939 q->samples_per_channel = q->samples_per_frame / q->channels;
940 q->scrambled_stream = 1;
943 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
945 /* Check the extradata. */
947 if (q->atrac3version != 4) {
948 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
949 return AVERROR_INVALIDDATA;
952 if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
953 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
954 return AVERROR_INVALIDDATA;
957 if (q->delay != 0x88E) {
958 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
959 return AVERROR_INVALIDDATA;
962 if (q->codingMode == STEREO) {
963 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
964 } else if (q->codingMode == JOINT_STEREO) {
965 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
967 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
968 return AVERROR_INVALIDDATA;
971 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
972 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
973 return AVERROR(EINVAL);
977 if(avctx->block_align >= UINT_MAX/2)
978 return AVERROR(EINVAL);
980 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
981 * this is for the bitstream reader. */
982 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
983 return AVERROR(ENOMEM);
986 /* Initialize the VLC tables. */
987 if (!vlcs_initialized) {
988 for (i=0 ; i<7 ; i++) {
989 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
990 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
991 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
993 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
995 vlcs_initialized = 1;
998 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
999 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1001 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1003 if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
1004 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
1005 av_freep(&q->decoded_bytes_buffer);
1009 atrac_generate_tables();
1011 /* Generate gain tables. */
1012 for (i=0 ; i<16 ; i++)
1013 gain_tab1[i] = powf (2.0, (4 - i));
1015 for (i=-15 ; i<16 ; i++)
1016 gain_tab2[i+15] = powf (2.0, i * -0.125);
1018 /* init the joint-stereo decoding data */
1019 q->weighting_delay[0] = 0;
1020 q->weighting_delay[1] = 7;
1021 q->weighting_delay[2] = 0;
1022 q->weighting_delay[3] = 7;
1023 q->weighting_delay[4] = 0;
1024 q->weighting_delay[5] = 7;
1026 for (i=0; i<4; i++) {
1027 q->matrix_coeff_index_prev[i] = 3;
1028 q->matrix_coeff_index_now[i] = 3;
1029 q->matrix_coeff_index_next[i] = 3;
1032 dsputil_init(&dsp, avctx);
1033 ff_fmt_convert_init(&q->fmt_conv, avctx);
1035 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1037 atrac3_decode_close(avctx);
1038 return AVERROR(ENOMEM);
1041 if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
1042 q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
1043 q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
1044 if (!q->outSamples[0]) {
1045 atrac3_decode_close(avctx);
1046 return AVERROR(ENOMEM);
1054 AVCodec ff_atrac3_decoder =
1057 .type = AVMEDIA_TYPE_AUDIO,
1058 .id = CODEC_ID_ATRAC3,
1059 .priv_data_size = sizeof(ATRAC3Context),
1060 .init = atrac3_decode_init,
1061 .close = atrac3_decode_close,
1062 .decode = atrac3_decode_frame,
1063 .capabilities = CODEC_CAP_SUBFRAMES,
1064 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),