check for request_channels at codec init
[ffmpeg-lucabe.git] / libavcodec / atrac3.c
blobe7239a3bab465d736948cbae5269ca58bb037f2f
1 /*
2 * Atrac 3 compatible decoder
3 * Copyright (c) 2006-2007 Maxim Poliakovski
4 * Copyright (c) 2006-2007 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
23 /**
24 * @file atrac3.c
25 * Atrac 3 compatible decoder.
26 * This decoder handles RealNetworks, RealAudio atrc data.
27 * Atrac 3 is identified by the codec name atrc in RealMedia files.
29 * To use this decoder, a calling application must supply the extradata
30 * bytes provided from the RealMedia container: 10 bytes or 14 bytes
31 * from the WAV container.
34 #include <math.h>
35 #include <stddef.h>
36 #include <stdio.h>
38 #include "avcodec.h"
39 #include "bitstream.h"
40 #include "dsputil.h"
41 #include "bytestream.h"
43 #include "atrac3data.h"
45 #define JOINT_STEREO 0x12
46 #define STEREO 0x2
49 /* These structures are needed to store the parsed gain control data. */
50 typedef struct {
51 int num_gain_data;
52 int levcode[8];
53 int loccode[8];
54 } gain_info;
56 typedef struct {
57 gain_info gBlock[4];
58 } gain_block;
60 typedef struct {
61 int pos;
62 int numCoefs;
63 float coef[8];
64 } tonal_component;
66 typedef struct {
67 int bandsCoded;
68 int numComponents;
69 tonal_component components[64];
70 float prevFrame[1024];
71 int gcBlkSwitch;
72 gain_block gainBlock[2];
74 DECLARE_ALIGNED_16(float, spectrum[1024]);
75 DECLARE_ALIGNED_16(float, IMDCT_buf[1024]);
77 float delayBuf1[46]; ///<qmf delay buffers
78 float delayBuf2[46];
79 float delayBuf3[46];
80 } channel_unit;
82 typedef struct {
83 GetBitContext gb;
84 //@{
85 /** stream data */
86 int channels;
87 int codingMode;
88 int bit_rate;
89 int sample_rate;
90 int samples_per_channel;
91 int samples_per_frame;
93 int bits_per_frame;
94 int bytes_per_frame;
95 int pBs;
96 channel_unit* pUnits;
97 //@}
98 //@{
99 /** joint-stereo related variables */
100 int matrix_coeff_index_prev[4];
101 int matrix_coeff_index_now[4];
102 int matrix_coeff_index_next[4];
103 int weighting_delay[6];
104 //@}
105 //@{
106 /** data buffers */
107 float outSamples[2048];
108 uint8_t* decoded_bytes_buffer;
109 float tempBuf[1070];
110 DECLARE_ALIGNED_16(float,mdct_tmp[512]);
111 //@}
112 //@{
113 /** extradata */
114 int atrac3version;
115 int delay;
116 int scrambled_stream;
117 int frame_factor;
118 //@}
119 } ATRAC3Context;
121 static DECLARE_ALIGNED_16(float,mdct_window[512]);
122 static float qmf_window[48];
123 static VLC spectral_coeff_tab[7];
124 static float SFTable[64];
125 static float gain_tab1[16];
126 static float gain_tab2[31];
127 static MDCTContext mdct_ctx;
128 static DSPContext dsp;
131 /* quadrature mirror synthesis filter */
134 * Quadrature mirror synthesis filter.
136 * @param inlo lower part of spectrum
137 * @param inhi higher part of spectrum
138 * @param nIn size of spectrum buffer
139 * @param pOut out buffer
140 * @param delayBuf delayBuf buffer
141 * @param temp temp buffer
145 static void iqmf (float *inlo, float *inhi, unsigned int nIn, float *pOut, float *delayBuf, float *temp)
147 int i, j;
148 float *p1, *p3;
150 memcpy(temp, delayBuf, 46*sizeof(float));
152 p3 = temp + 46;
154 /* loop1 */
155 for(i=0; i<nIn; i+=2){
156 p3[2*i+0] = inlo[i ] + inhi[i ];
157 p3[2*i+1] = inlo[i ] - inhi[i ];
158 p3[2*i+2] = inlo[i+1] + inhi[i+1];
159 p3[2*i+3] = inlo[i+1] - inhi[i+1];
162 /* loop2 */
163 p1 = temp;
164 for (j = nIn; j != 0; j--) {
165 float s1 = 0.0;
166 float s2 = 0.0;
168 for (i = 0; i < 48; i += 2) {
169 s1 += p1[i] * qmf_window[i];
170 s2 += p1[i+1] * qmf_window[i+1];
173 pOut[0] = s2;
174 pOut[1] = s1;
176 p1 += 2;
177 pOut += 2;
180 /* Update the delay buffer. */
181 memcpy(delayBuf, temp + nIn*2, 46*sizeof(float));
185 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
186 * caused by the reverse spectra of the QMF.
188 * @param pInput float input
189 * @param pOutput float output
190 * @param odd_band 1 if the band is an odd band
191 * @param mdct_tmp aligned temporary buffer for the mdct
194 static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
196 int i;
198 if (odd_band) {
200 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
201 * or it gives better compression to do it this way.
202 * FIXME: It should be possible to handle this in ff_imdct_calc
203 * for that to happen a modification of the prerotation step of
204 * all SIMD code and C code is needed.
205 * Or fix the functions before so they generate a pre reversed spectrum.
208 for (i=0; i<128; i++)
209 FFSWAP(float, pInput[i], pInput[255-i]);
212 mdct_ctx.fft.imdct_calc(&mdct_ctx,pOutput,pInput,mdct_tmp);
214 /* Perform windowing on the output. */
215 dsp.vector_fmul(pOutput,mdct_window,512);
221 * Atrac 3 indata descrambling, only used for data coming from the rm container
223 * @param in pointer to 8 bit array of indata
224 * @param bits amount of bits
225 * @param out pointer to 8 bit array of outdata
228 static int decode_bytes(uint8_t* inbuffer, uint8_t* out, int bytes){
229 int i, off;
230 uint32_t c;
231 uint32_t* buf;
232 uint32_t* obuf = (uint32_t*) out;
234 off = (int)((long)inbuffer & 3);
235 buf = (uint32_t*) (inbuffer - off);
236 c = be2me_32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
237 bytes += 3 + off;
238 for (i = 0; i < bytes/4; i++)
239 obuf[i] = c ^ buf[i];
241 if (off)
242 av_log(NULL,AV_LOG_DEBUG,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off);
244 return off;
248 static void init_atrac3_transforms(ATRAC3Context *q) {
249 float enc_window[256];
250 float s;
251 int i;
253 /* Generate the mdct window, for details see
254 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
255 for (i=0 ; i<256; i++)
256 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
258 if (!mdct_window[0])
259 for (i=0 ; i<256; i++) {
260 mdct_window[i] = enc_window[i]/(enc_window[i]*enc_window[i] + enc_window[255-i]*enc_window[255-i]);
261 mdct_window[511-i] = mdct_window[i];
264 /* Generate the QMF window. */
265 for (i=0 ; i<24; i++) {
266 s = qmf_48tap_half[i] * 2.0;
267 qmf_window[i] = s;
268 qmf_window[47 - i] = s;
271 /* Initialize the MDCT transform. */
272 ff_mdct_init(&mdct_ctx, 9, 1);
276 * Atrac3 uninit, free all allocated memory
279 static int atrac3_decode_close(AVCodecContext *avctx)
281 ATRAC3Context *q = avctx->priv_data;
283 av_free(q->pUnits);
284 av_free(q->decoded_bytes_buffer);
286 return 0;
290 / * Mantissa decoding
292 * @param gb the GetBit context
293 * @param selector what table is the output values coded with
294 * @param codingFlag constant length coding or variable length coding
295 * @param mantissas mantissa output table
296 * @param numCodes amount of values to get
299 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
301 int numBits, cnt, code, huffSymb;
303 if (selector == 1)
304 numCodes /= 2;
306 if (codingFlag != 0) {
307 /* constant length coding (CLC) */
308 //FIXME we don't have any samples coded in CLC mode
309 numBits = CLCLengthTab[selector];
311 if (selector > 1) {
312 for (cnt = 0; cnt < numCodes; cnt++) {
313 if (numBits)
314 code = get_sbits(gb, numBits);
315 else
316 code = 0;
317 mantissas[cnt] = code;
319 } else {
320 for (cnt = 0; cnt < numCodes; cnt++) {
321 if (numBits)
322 code = get_bits(gb, numBits); //numBits is always 4 in this case
323 else
324 code = 0;
325 mantissas[cnt*2] = seTab_0[code >> 2];
326 mantissas[cnt*2+1] = seTab_0[code & 3];
329 } else {
330 /* variable length coding (VLC) */
331 if (selector != 1) {
332 for (cnt = 0; cnt < numCodes; cnt++) {
333 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
334 huffSymb += 1;
335 code = huffSymb >> 1;
336 if (huffSymb & 1)
337 code = -code;
338 mantissas[cnt] = code;
340 } else {
341 for (cnt = 0; cnt < numCodes; cnt++) {
342 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
343 mantissas[cnt*2] = decTable1[huffSymb*2];
344 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
351 * Restore the quantized band spectrum coefficients
353 * @param gb the GetBit context
354 * @param pOut decoded band spectrum
355 * @return outSubbands subband counter, fix for broken specification/files
358 static int decodeSpectrum (GetBitContext *gb, float *pOut)
360 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
361 int subband_vlc_index[32], SF_idxs[32];
362 int mantissas[128];
363 float SF;
365 numSubbands = get_bits(gb, 5); // number of coded subbands
366 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
368 /* Get the VLC selector table for the subbands, 0 means not coded. */
369 for (cnt = 0; cnt <= numSubbands; cnt++)
370 subband_vlc_index[cnt] = get_bits(gb, 3);
372 /* Read the scale factor indexes from the stream. */
373 for (cnt = 0; cnt <= numSubbands; cnt++) {
374 if (subband_vlc_index[cnt] != 0)
375 SF_idxs[cnt] = get_bits(gb, 6);
378 for (cnt = 0; cnt <= numSubbands; cnt++) {
379 first = subbandTab[cnt];
380 last = subbandTab[cnt+1];
382 subbWidth = last - first;
384 if (subband_vlc_index[cnt] != 0) {
385 /* Decode spectral coefficients for this subband. */
386 /* TODO: This can be done faster is several blocks share the
387 * same VLC selector (subband_vlc_index) */
388 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
390 /* Decode the scale factor for this subband. */
391 SF = SFTable[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
393 /* Inverse quantize the coefficients. */
394 for (pIn=mantissas ; first<last; first++, pIn++)
395 pOut[first] = *pIn * SF;
396 } else {
397 /* This subband was not coded, so zero the entire subband. */
398 memset(pOut+first, 0, subbWidth*sizeof(float));
402 /* Clear the subbands that were not coded. */
403 first = subbandTab[cnt];
404 memset(pOut+first, 0, (1024 - first) * sizeof(float));
405 return numSubbands;
409 * Restore the quantized tonal components
411 * @param gb the GetBit context
412 * @param pComponent tone component
413 * @param numBands amount of coded bands
416 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
418 int i,j,k,cnt;
419 int components, coding_mode_selector, coding_mode, coded_values_per_component;
420 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
421 int band_flags[4], mantissa[8];
422 float *pCoef;
423 float scalefactor;
424 int component_count = 0;
426 components = get_bits(gb,5);
428 /* no tonal components */
429 if (components == 0)
430 return 0;
432 coding_mode_selector = get_bits(gb,2);
433 if (coding_mode_selector == 2)
434 return -1;
436 coding_mode = coding_mode_selector & 1;
438 for (i = 0; i < components; i++) {
439 for (cnt = 0; cnt <= numBands; cnt++)
440 band_flags[cnt] = get_bits1(gb);
442 coded_values_per_component = get_bits(gb,3);
444 quant_step_index = get_bits(gb,3);
445 if (quant_step_index <= 1)
446 return -1;
448 if (coding_mode_selector == 3)
449 coding_mode = get_bits1(gb);
451 for (j = 0; j < (numBands + 1) * 4; j++) {
452 if (band_flags[j >> 2] == 0)
453 continue;
455 coded_components = get_bits(gb,3);
457 for (k=0; k<coded_components; k++) {
458 sfIndx = get_bits(gb,6);
459 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
460 max_coded_values = 1024 - pComponent[component_count].pos;
461 coded_values = coded_values_per_component + 1;
462 coded_values = FFMIN(max_coded_values,coded_values);
464 scalefactor = SFTable[sfIndx] * iMaxQuant[quant_step_index];
466 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
468 pComponent[component_count].numCoefs = coded_values;
470 /* inverse quant */
471 pCoef = pComponent[k].coef;
472 for (cnt = 0; cnt < coded_values; cnt++)
473 pCoef[cnt] = mantissa[cnt] * scalefactor;
475 component_count++;
480 return component_count;
484 * Decode gain parameters for the coded bands
486 * @param gb the GetBit context
487 * @param pGb the gainblock for the current band
488 * @param numBands amount of coded bands
491 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
493 int i, cf, numData;
494 int *pLevel, *pLoc;
496 gain_info *pGain = pGb->gBlock;
498 for (i=0 ; i<=numBands; i++)
500 numData = get_bits(gb,3);
501 pGain[i].num_gain_data = numData;
502 pLevel = pGain[i].levcode;
503 pLoc = pGain[i].loccode;
505 for (cf = 0; cf < numData; cf++){
506 pLevel[cf]= get_bits(gb,4);
507 pLoc [cf]= get_bits(gb,5);
508 if(cf && pLoc[cf] <= pLoc[cf-1])
509 return -1;
513 /* Clear the unused blocks. */
514 for (; i<4 ; i++)
515 pGain[i].num_gain_data = 0;
517 return 0;
521 * Apply gain parameters and perform the MDCT overlapping part
523 * @param pIn input float buffer
524 * @param pPrev previous float buffer to perform overlap against
525 * @param pOut output float buffer
526 * @param pGain1 current band gain info
527 * @param pGain2 next band gain info
530 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
532 /* gain compensation function */
533 float gain1, gain2, gain_inc;
534 int cnt, numdata, nsample, startLoc, endLoc;
537 if (pGain2->num_gain_data == 0)
538 gain1 = 1.0;
539 else
540 gain1 = gain_tab1[pGain2->levcode[0]];
542 if (pGain1->num_gain_data == 0) {
543 for (cnt = 0; cnt < 256; cnt++)
544 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
545 } else {
546 numdata = pGain1->num_gain_data;
547 pGain1->loccode[numdata] = 32;
548 pGain1->levcode[numdata] = 4;
550 nsample = 0; // current sample = 0
552 for (cnt = 0; cnt < numdata; cnt++) {
553 startLoc = pGain1->loccode[cnt] * 8;
554 endLoc = startLoc + 8;
556 gain2 = gain_tab1[pGain1->levcode[cnt]];
557 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
559 /* interpolate */
560 for (; nsample < startLoc; nsample++)
561 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
563 /* interpolation is done over eight samples */
564 for (; nsample < endLoc; nsample++) {
565 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
566 gain2 *= gain_inc;
570 for (; nsample < 256; nsample++)
571 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
574 /* Delay for the overlapping part. */
575 memcpy(pPrev, &pIn[256], 256*sizeof(float));
579 * Combine the tonal band spectrum and regular band spectrum
581 * @param pSpectrum output spectrum buffer
582 * @param numComponents amount of tonal components
583 * @param pComponent tonal components for this band
586 static void addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
588 int cnt, i;
589 float *pIn, *pOut;
591 for (cnt = 0; cnt < numComponents; cnt++){
592 pIn = pComponent[cnt].coef;
593 pOut = &(pSpectrum[pComponent[cnt].pos]);
595 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
596 pOut[i] += pIn[i];
601 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
603 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
605 int i, band, nsample, s1, s2;
606 float c1, c2;
607 float mc1_l, mc1_r, mc2_l, mc2_r;
609 for (i=0,band = 0; band < 4*256; band+=256,i++) {
610 s1 = pPrevCode[i];
611 s2 = pCurrCode[i];
612 nsample = 0;
614 if (s1 != s2) {
615 /* Selector value changed, interpolation needed. */
616 mc1_l = matrixCoeffs[s1*2];
617 mc1_r = matrixCoeffs[s1*2+1];
618 mc2_l = matrixCoeffs[s2*2];
619 mc2_r = matrixCoeffs[s2*2+1];
621 /* Interpolation is done over the first eight samples. */
622 for(; nsample < 8; nsample++) {
623 c1 = su1[band+nsample];
624 c2 = su2[band+nsample];
625 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
626 su1[band+nsample] = c2;
627 su2[band+nsample] = c1 * 2.0 - c2;
631 /* Apply the matrix without interpolation. */
632 switch (s2) {
633 case 0: /* M/S decoding */
634 for (; nsample < 256; nsample++) {
635 c1 = su1[band+nsample];
636 c2 = su2[band+nsample];
637 su1[band+nsample] = c2 * 2.0;
638 su2[band+nsample] = (c1 - c2) * 2.0;
640 break;
642 case 1:
643 for (; nsample < 256; nsample++) {
644 c1 = su1[band+nsample];
645 c2 = su2[band+nsample];
646 su1[band+nsample] = (c1 + c2) * 2.0;
647 su2[band+nsample] = c2 * -2.0;
649 break;
650 case 2:
651 case 3:
652 for (; nsample < 256; nsample++) {
653 c1 = su1[band+nsample];
654 c2 = su2[band+nsample];
655 su1[band+nsample] = c1 + c2;
656 su2[band+nsample] = c1 - c2;
658 break;
659 default:
660 assert(0);
665 static void getChannelWeights (int indx, int flag, float ch[2]){
667 if (indx == 7) {
668 ch[0] = 1.0;
669 ch[1] = 1.0;
670 } else {
671 ch[0] = (float)(indx & 7) / 7.0;
672 ch[1] = sqrt(2 - ch[0]*ch[0]);
673 if(flag)
674 FFSWAP(float, ch[0], ch[1]);
678 static void channelWeighting (float *su1, float *su2, int *p3)
680 int band, nsample;
681 /* w[x][y] y=0 is left y=1 is right */
682 float w[2][2];
684 if (p3[1] != 7 || p3[3] != 7){
685 getChannelWeights(p3[1], p3[0], w[0]);
686 getChannelWeights(p3[3], p3[2], w[1]);
688 for(band = 1; band < 4; band++) {
689 /* scale the channels by the weights */
690 for(nsample = 0; nsample < 8; nsample++) {
691 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
692 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
695 for(; nsample < 256; nsample++) {
696 su1[band*256+nsample] *= w[1][0];
697 su2[band*256+nsample] *= w[1][1];
705 * Decode a Sound Unit
707 * @param gb the GetBit context
708 * @param pSnd the channel unit to be used
709 * @param pOut the decoded samples before IQMF in float representation
710 * @param channelNum channel number
711 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
715 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
717 int band, result=0, numSubbands, numBands;
719 if (codingMode == JOINT_STEREO && channelNum == 1) {
720 if (get_bits(gb,2) != 3) {
721 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
722 return -1;
724 } else {
725 if (get_bits(gb,6) != 0x28) {
726 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
727 return -1;
731 /* number of coded QMF bands */
732 pSnd->bandsCoded = get_bits(gb,2);
734 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
735 if (result) return result;
737 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
738 if (pSnd->numComponents == -1) return -1;
740 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
742 /* Merge the decoded spectrum and tonal components. */
743 addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
746 /* Convert number of subbands into number of MLT/QMF bands */
747 numBands = (subbandTab[numSubbands] - 1) >> 8;
750 /* Reconstruct time domain samples. */
751 for (band=0; band<4; band++) {
752 /* Perform the IMDCT step without overlapping. */
753 if (band <= numBands) {
754 IMLT(&(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1,q->mdct_tmp);
755 } else
756 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
758 /* gain compensation and overlapping */
759 gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]),
760 &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]),
761 &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band]));
764 /* Swap the gain control buffers for the next frame. */
765 pSnd->gcBlkSwitch ^= 1;
767 return 0;
771 * Frame handling
773 * @param q Atrac3 private context
774 * @param databuf the input data
777 static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
779 int result, i;
780 float *p1, *p2, *p3, *p4;
781 uint8_t *ptr1, *ptr2;
783 if (q->codingMode == JOINT_STEREO) {
785 /* channel coupling mode */
786 /* decode Sound Unit 1 */
787 init_get_bits(&q->gb,databuf,q->bits_per_frame);
789 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO);
790 if (result != 0)
791 return (result);
793 /* Framedata of the su2 in the joint-stereo mode is encoded in
794 * reverse byte order so we need to swap it first. */
795 ptr1 = databuf;
796 ptr2 = databuf+q->bytes_per_frame-1;
797 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
798 FFSWAP(uint8_t,*ptr1,*ptr2);
801 /* Skip the sync codes (0xF8). */
802 ptr1 = databuf;
803 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
804 if (i >= q->bytes_per_frame)
805 return -1;
809 /* set the bitstream reader at the start of the second Sound Unit*/
810 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
812 /* Fill the Weighting coeffs delay buffer */
813 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
814 q->weighting_delay[4] = get_bits1(&q->gb);
815 q->weighting_delay[5] = get_bits(&q->gb,3);
817 for (i = 0; i < 4; i++) {
818 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
819 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
820 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
823 /* Decode Sound Unit 2. */
824 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO);
825 if (result != 0)
826 return (result);
828 /* Reconstruct the channel coefficients. */
829 reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
831 channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay);
833 } else {
834 /* normal stereo mode or mono */
835 /* Decode the channel sound units. */
836 for (i=0 ; i<q->channels ; i++) {
838 /* Set the bitstream reader at the start of a channel sound unit. */
839 init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels);
841 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode);
842 if (result != 0)
843 return (result);
847 /* Apply the iQMF synthesis filter. */
848 p1= q->outSamples;
849 for (i=0 ; i<q->channels ; i++) {
850 p2= p1+256;
851 p3= p2+256;
852 p4= p3+256;
853 iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
854 iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
855 iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
856 p1 +=1024;
859 return 0;
864 * Atrac frame decoding
866 * @param avctx pointer to the AVCodecContext
869 static int atrac3_decode_frame(AVCodecContext *avctx,
870 void *data, int *data_size,
871 uint8_t *buf, int buf_size) {
872 ATRAC3Context *q = avctx->priv_data;
873 int result = 0, i;
874 uint8_t* databuf;
875 int16_t* samples = data;
877 if (buf_size < avctx->block_align)
878 return buf_size;
880 /* Check if we need to descramble and what buffer to pass on. */
881 if (q->scrambled_stream) {
882 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
883 databuf = q->decoded_bytes_buffer;
884 } else {
885 databuf = buf;
888 result = decodeFrame(q, databuf);
890 if (result != 0) {
891 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
892 return -1;
895 if (q->channels == 1) {
896 /* mono */
897 for (i = 0; i<1024; i++)
898 samples[i] = av_clip_int16(round(q->outSamples[i]));
899 *data_size = 1024 * sizeof(int16_t);
900 } else {
901 /* stereo */
902 for (i = 0; i < 1024; i++) {
903 samples[i*2] = av_clip_int16(round(q->outSamples[i]));
904 samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i]));
906 *data_size = 2048 * sizeof(int16_t);
909 return avctx->block_align;
914 * Atrac3 initialization
916 * @param avctx pointer to the AVCodecContext
919 static int atrac3_decode_init(AVCodecContext *avctx)
921 int i;
922 uint8_t *edata_ptr = avctx->extradata;
923 ATRAC3Context *q = avctx->priv_data;
925 /* Take data from the AVCodecContext (RM container). */
926 q->sample_rate = avctx->sample_rate;
927 q->channels = avctx->channels;
928 q->bit_rate = avctx->bit_rate;
929 q->bits_per_frame = avctx->block_align * 8;
930 q->bytes_per_frame = avctx->block_align;
932 /* Take care of the codec-specific extradata. */
933 if (avctx->extradata_size == 14) {
934 /* Parse the extradata, WAV format */
935 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
936 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
937 q->codingMode = bytestream_get_le16(&edata_ptr);
938 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
939 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
940 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
942 /* setup */
943 q->samples_per_frame = 1024 * q->channels;
944 q->atrac3version = 4;
945 q->delay = 0x88E;
946 if (q->codingMode)
947 q->codingMode = JOINT_STEREO;
948 else
949 q->codingMode = STEREO;
951 q->scrambled_stream = 0;
953 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)) {
954 } else {
955 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);
956 return -1;
959 } else if (avctx->extradata_size == 10) {
960 /* Parse the extradata, RM format. */
961 q->atrac3version = bytestream_get_be32(&edata_ptr);
962 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
963 q->delay = bytestream_get_be16(&edata_ptr);
964 q->codingMode = bytestream_get_be16(&edata_ptr);
966 q->samples_per_channel = q->samples_per_frame / q->channels;
967 q->scrambled_stream = 1;
969 } else {
970 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
972 /* Check the extradata. */
974 if (q->atrac3version != 4) {
975 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
976 return -1;
979 if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) {
980 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
981 return -1;
984 if (q->delay != 0x88E) {
985 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
986 return -1;
989 if (q->codingMode == STEREO) {
990 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
991 } else if (q->codingMode == JOINT_STEREO) {
992 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
993 } else {
994 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
995 return -1;
998 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
999 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
1000 return -1;
1004 if(avctx->block_align >= UINT_MAX/2)
1005 return -1;
1007 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
1008 * this is for the bitstream reader. */
1009 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
1010 return AVERROR(ENOMEM);
1013 /* Initialize the VLC tables. */
1014 for (i=0 ; i<7 ; i++) {
1015 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
1016 huff_bits[i], 1, 1,
1017 huff_codes[i], 1, 1, INIT_VLC_USE_STATIC);
1020 init_atrac3_transforms(q);
1022 /* Generate the scale factors. */
1023 for (i=0 ; i<64 ; i++)
1024 SFTable[i] = pow(2.0, (i - 15) / 3.0);
1026 /* Generate gain tables. */
1027 for (i=0 ; i<16 ; i++)
1028 gain_tab1[i] = powf (2.0, (4 - i));
1030 for (i=-15 ; i<16 ; i++)
1031 gain_tab2[i+15] = powf (2.0, i * -0.125);
1033 /* init the joint-stereo decoding data */
1034 q->weighting_delay[0] = 0;
1035 q->weighting_delay[1] = 7;
1036 q->weighting_delay[2] = 0;
1037 q->weighting_delay[3] = 7;
1038 q->weighting_delay[4] = 0;
1039 q->weighting_delay[5] = 7;
1041 for (i=0; i<4; i++) {
1042 q->matrix_coeff_index_prev[i] = 3;
1043 q->matrix_coeff_index_now[i] = 3;
1044 q->matrix_coeff_index_next[i] = 3;
1047 dsputil_init(&dsp, avctx);
1049 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1050 if (!q->pUnits) {
1051 av_free(q->decoded_bytes_buffer);
1052 return AVERROR(ENOMEM);
1055 return 0;
1059 AVCodec atrac3_decoder =
1061 .name = "atrac 3",
1062 .type = CODEC_TYPE_AUDIO,
1063 .id = CODEC_ID_ATRAC3,
1064 .priv_data_size = sizeof(ATRAC3Context),
1065 .init = atrac3_decode_init,
1066 .close = atrac3_decode_close,
1067 .decode = atrac3_decode_frame,