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
24 * @file libavcodec/atrac3.c
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"
45 #include "atrac3data.h"
47 #define JOINT_STEREO 0x12
51 /* These structures are needed to store the parsed gain control data. */
71 tonal_component components
[64];
72 float prevFrame
[1024];
74 gain_block gainBlock
[2];
76 DECLARE_ALIGNED_16(float, spectrum
[1024]);
77 DECLARE_ALIGNED_16(float, IMDCT_buf
[1024]);
79 float delayBuf1
[46]; ///<qmf delay buffers
92 int samples_per_channel
;
93 int samples_per_frame
;
101 /** joint-stereo related variables */
102 int matrix_coeff_index_prev
[4];
103 int matrix_coeff_index_now
[4];
104 int matrix_coeff_index_next
[4];
105 int weighting_delay
[6];
109 float outSamples
[2048];
110 uint8_t* decoded_bytes_buffer
;
117 int scrambled_stream
;
122 static DECLARE_ALIGNED_16(float,mdct_window
[512]);
123 static VLC spectral_coeff_tab
[7];
124 static float gain_tab1
[16];
125 static float gain_tab2
[31];
126 static FFTContext mdct_ctx
;
127 static DSPContext dsp
;
131 * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
132 * caused by the reverse spectra of the QMF.
134 * @param pInput float input
135 * @param pOutput float output
136 * @param odd_band 1 if the band is an odd band
139 static void IMLT(float *pInput
, float *pOutput
, int odd_band
)
145 * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
146 * or it gives better compression to do it this way.
147 * FIXME: It should be possible to handle this in ff_imdct_calc
148 * for that to happen a modification of the prerotation step of
149 * all SIMD code and C code is needed.
150 * Or fix the functions before so they generate a pre reversed spectrum.
153 for (i
=0; i
<128; i
++)
154 FFSWAP(float, pInput
[i
], pInput
[255-i
]);
157 ff_imdct_calc(&mdct_ctx
,pOutput
,pInput
);
159 /* Perform windowing on the output. */
160 dsp
.vector_fmul(pOutput
,mdct_window
,512);
166 * Atrac 3 indata descrambling, only used for data coming from the rm container
168 * @param in pointer to 8 bit array of indata
169 * @param bits amount of bits
170 * @param out pointer to 8 bit array of outdata
173 static int decode_bytes(const uint8_t* inbuffer
, uint8_t* out
, int bytes
){
177 uint32_t* obuf
= (uint32_t*) out
;
179 off
= (intptr_t)inbuffer
& 3;
180 buf
= (const uint32_t*) (inbuffer
- off
);
181 c
= be2me_32((0x537F6103 >> (off
*8)) | (0x537F6103 << (32-(off
*8))));
183 for (i
= 0; i
< bytes
/4; i
++)
184 obuf
[i
] = c
^ buf
[i
];
187 av_log(NULL
,AV_LOG_DEBUG
,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off
);
193 static av_cold
void init_atrac3_transforms(ATRAC3Context
*q
) {
194 float enc_window
[256];
197 /* Generate the mdct window, for details see
198 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
199 for (i
=0 ; i
<256; i
++)
200 enc_window
[i
] = (sin(((i
+ 0.5) / 256.0 - 0.5) * M_PI
) + 1.0) * 0.5;
203 for (i
=0 ; i
<256; i
++) {
204 mdct_window
[i
] = enc_window
[i
]/(enc_window
[i
]*enc_window
[i
] + enc_window
[255-i
]*enc_window
[255-i
]);
205 mdct_window
[511-i
] = mdct_window
[i
];
208 /* Initialize the MDCT transform. */
209 ff_mdct_init(&mdct_ctx
, 9, 1, 1.0);
213 * Atrac3 uninit, free all allocated memory
216 static av_cold
int atrac3_decode_close(AVCodecContext
*avctx
)
218 ATRAC3Context
*q
= avctx
->priv_data
;
221 av_free(q
->decoded_bytes_buffer
);
227 / * Mantissa decoding
229 * @param gb the GetBit context
230 * @param selector what table is the output values coded with
231 * @param codingFlag constant length coding or variable length coding
232 * @param mantissas mantissa output table
233 * @param numCodes amount of values to get
236 static void readQuantSpectralCoeffs (GetBitContext
*gb
, int selector
, int codingFlag
, int* mantissas
, int numCodes
)
238 int numBits
, cnt
, code
, huffSymb
;
243 if (codingFlag
!= 0) {
244 /* constant length coding (CLC) */
245 numBits
= CLCLengthTab
[selector
];
248 for (cnt
= 0; cnt
< numCodes
; cnt
++) {
250 code
= get_sbits(gb
, numBits
);
253 mantissas
[cnt
] = code
;
256 for (cnt
= 0; cnt
< numCodes
; cnt
++) {
258 code
= get_bits(gb
, numBits
); //numBits is always 4 in this case
261 mantissas
[cnt
*2] = seTab_0
[code
>> 2];
262 mantissas
[cnt
*2+1] = seTab_0
[code
& 3];
266 /* variable length coding (VLC) */
268 for (cnt
= 0; cnt
< numCodes
; cnt
++) {
269 huffSymb
= get_vlc2(gb
, spectral_coeff_tab
[selector
-1].table
, spectral_coeff_tab
[selector
-1].bits
, 3);
271 code
= huffSymb
>> 1;
274 mantissas
[cnt
] = code
;
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);
279 mantissas
[cnt
*2] = decTable1
[huffSymb
*2];
280 mantissas
[cnt
*2+1] = decTable1
[huffSymb
*2+1];
287 * Restore the quantized band spectrum coefficients
289 * @param gb the GetBit context
290 * @param pOut decoded band spectrum
291 * @return outSubbands subband counter, fix for broken specification/files
294 static int decodeSpectrum (GetBitContext
*gb
, float *pOut
)
296 int numSubbands
, codingMode
, cnt
, first
, last
, subbWidth
, *pIn
;
297 int subband_vlc_index
[32], SF_idxs
[32];
301 numSubbands
= get_bits(gb
, 5); // number of coded subbands
302 codingMode
= get_bits1(gb
); // coding Mode: 0 - VLC/ 1-CLC
304 /* Get the VLC selector table for the subbands, 0 means not coded. */
305 for (cnt
= 0; cnt
<= numSubbands
; cnt
++)
306 subband_vlc_index
[cnt
] = get_bits(gb
, 3);
308 /* Read the scale factor indexes from the stream. */
309 for (cnt
= 0; cnt
<= numSubbands
; cnt
++) {
310 if (subband_vlc_index
[cnt
] != 0)
311 SF_idxs
[cnt
] = get_bits(gb
, 6);
314 for (cnt
= 0; cnt
<= numSubbands
; cnt
++) {
315 first
= subbandTab
[cnt
];
316 last
= subbandTab
[cnt
+1];
318 subbWidth
= last
- first
;
320 if (subband_vlc_index
[cnt
] != 0) {
321 /* Decode spectral coefficients for this subband. */
322 /* TODO: This can be done faster is several blocks share the
323 * same VLC selector (subband_vlc_index) */
324 readQuantSpectralCoeffs (gb
, subband_vlc_index
[cnt
], codingMode
, mantissas
, subbWidth
);
326 /* Decode the scale factor for this subband. */
327 SF
= sf_table
[SF_idxs
[cnt
]] * iMaxQuant
[subband_vlc_index
[cnt
]];
329 /* Inverse quantize the coefficients. */
330 for (pIn
=mantissas
; first
<last
; first
++, pIn
++)
331 pOut
[first
] = *pIn
* SF
;
333 /* This subband was not coded, so zero the entire subband. */
334 memset(pOut
+first
, 0, subbWidth
*sizeof(float));
338 /* Clear the subbands that were not coded. */
339 first
= subbandTab
[cnt
];
340 memset(pOut
+first
, 0, (1024 - first
) * sizeof(float));
345 * Restore the quantized tonal components
347 * @param gb the GetBit context
348 * @param pComponent tone component
349 * @param numBands amount of coded bands
352 static int decodeTonalComponents (GetBitContext
*gb
, tonal_component
*pComponent
, int numBands
)
355 int components
, coding_mode_selector
, coding_mode
, coded_values_per_component
;
356 int sfIndx
, coded_values
, max_coded_values
, quant_step_index
, coded_components
;
357 int band_flags
[4], mantissa
[8];
360 int component_count
= 0;
362 components
= get_bits(gb
,5);
364 /* no tonal components */
368 coding_mode_selector
= get_bits(gb
,2);
369 if (coding_mode_selector
== 2)
372 coding_mode
= coding_mode_selector
& 1;
374 for (i
= 0; i
< components
; i
++) {
375 for (cnt
= 0; cnt
<= numBands
; cnt
++)
376 band_flags
[cnt
] = get_bits1(gb
);
378 coded_values_per_component
= get_bits(gb
,3);
380 quant_step_index
= get_bits(gb
,3);
381 if (quant_step_index
<= 1)
384 if (coding_mode_selector
== 3)
385 coding_mode
= get_bits1(gb
);
387 for (j
= 0; j
< (numBands
+ 1) * 4; j
++) {
388 if (band_flags
[j
>> 2] == 0)
391 coded_components
= get_bits(gb
,3);
393 for (k
=0; k
<coded_components
; k
++) {
394 sfIndx
= get_bits(gb
,6);
395 pComponent
[component_count
].pos
= j
* 64 + (get_bits(gb
,6));
396 max_coded_values
= 1024 - pComponent
[component_count
].pos
;
397 coded_values
= coded_values_per_component
+ 1;
398 coded_values
= FFMIN(max_coded_values
,coded_values
);
400 scalefactor
= sf_table
[sfIndx
] * iMaxQuant
[quant_step_index
];
402 readQuantSpectralCoeffs(gb
, quant_step_index
, coding_mode
, mantissa
, coded_values
);
404 pComponent
[component_count
].numCoefs
= coded_values
;
407 pCoef
= pComponent
[component_count
].coef
;
408 for (cnt
= 0; cnt
< coded_values
; cnt
++)
409 pCoef
[cnt
] = mantissa
[cnt
] * scalefactor
;
416 return component_count
;
420 * Decode gain parameters for the coded bands
422 * @param gb the GetBit context
423 * @param pGb the gainblock for the current band
424 * @param numBands amount of coded bands
427 static int decodeGainControl (GetBitContext
*gb
, gain_block
*pGb
, int numBands
)
432 gain_info
*pGain
= pGb
->gBlock
;
434 for (i
=0 ; i
<=numBands
; i
++)
436 numData
= get_bits(gb
,3);
437 pGain
[i
].num_gain_data
= numData
;
438 pLevel
= pGain
[i
].levcode
;
439 pLoc
= pGain
[i
].loccode
;
441 for (cf
= 0; cf
< numData
; cf
++){
442 pLevel
[cf
]= get_bits(gb
,4);
443 pLoc
[cf
]= get_bits(gb
,5);
444 if(cf
&& pLoc
[cf
] <= pLoc
[cf
-1])
449 /* Clear the unused blocks. */
451 pGain
[i
].num_gain_data
= 0;
457 * Apply gain parameters and perform the MDCT overlapping part
459 * @param pIn input float buffer
460 * @param pPrev previous float buffer to perform overlap against
461 * @param pOut output float buffer
462 * @param pGain1 current band gain info
463 * @param pGain2 next band gain info
466 static void gainCompensateAndOverlap (float *pIn
, float *pPrev
, float *pOut
, gain_info
*pGain1
, gain_info
*pGain2
)
468 /* gain compensation function */
469 float gain1
, gain2
, gain_inc
;
470 int cnt
, numdata
, nsample
, startLoc
, endLoc
;
473 if (pGain2
->num_gain_data
== 0)
476 gain1
= gain_tab1
[pGain2
->levcode
[0]];
478 if (pGain1
->num_gain_data
== 0) {
479 for (cnt
= 0; cnt
< 256; cnt
++)
480 pOut
[cnt
] = pIn
[cnt
] * gain1
+ pPrev
[cnt
];
482 numdata
= pGain1
->num_gain_data
;
483 pGain1
->loccode
[numdata
] = 32;
484 pGain1
->levcode
[numdata
] = 4;
486 nsample
= 0; // current sample = 0
488 for (cnt
= 0; cnt
< numdata
; cnt
++) {
489 startLoc
= pGain1
->loccode
[cnt
] * 8;
490 endLoc
= startLoc
+ 8;
492 gain2
= gain_tab1
[pGain1
->levcode
[cnt
]];
493 gain_inc
= gain_tab2
[(pGain1
->levcode
[cnt
+1] - pGain1
->levcode
[cnt
])+15];
496 for (; nsample
< startLoc
; nsample
++)
497 pOut
[nsample
] = (pIn
[nsample
] * gain1
+ pPrev
[nsample
]) * gain2
;
499 /* interpolation is done over eight samples */
500 for (; nsample
< endLoc
; nsample
++) {
501 pOut
[nsample
] = (pIn
[nsample
] * gain1
+ pPrev
[nsample
]) * gain2
;
506 for (; nsample
< 256; nsample
++)
507 pOut
[nsample
] = (pIn
[nsample
] * gain1
) + pPrev
[nsample
];
510 /* Delay for the overlapping part. */
511 memcpy(pPrev
, &pIn
[256], 256*sizeof(float));
515 * Combine the tonal band spectrum and regular band spectrum
516 * Return position of the last tonal coefficient
518 * @param pSpectrum output spectrum buffer
519 * @param numComponents amount of tonal components
520 * @param pComponent tonal components for this band
523 static int addTonalComponents (float *pSpectrum
, int numComponents
, tonal_component
*pComponent
)
525 int cnt
, i
, lastPos
= -1;
528 for (cnt
= 0; cnt
< numComponents
; cnt
++){
529 lastPos
= FFMAX(pComponent
[cnt
].pos
+ pComponent
[cnt
].numCoefs
, lastPos
);
530 pIn
= pComponent
[cnt
].coef
;
531 pOut
= &(pSpectrum
[pComponent
[cnt
].pos
]);
533 for (i
=0 ; i
<pComponent
[cnt
].numCoefs
; i
++)
541 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
543 static void reverseMatrixing(float *su1
, float *su2
, int *pPrevCode
, int *pCurrCode
)
545 int i
, band
, nsample
, s1
, s2
;
547 float mc1_l
, mc1_r
, mc2_l
, mc2_r
;
549 for (i
=0,band
= 0; band
< 4*256; band
+=256,i
++) {
555 /* Selector value changed, interpolation needed. */
556 mc1_l
= matrixCoeffs
[s1
*2];
557 mc1_r
= matrixCoeffs
[s1
*2+1];
558 mc2_l
= matrixCoeffs
[s2
*2];
559 mc2_r
= matrixCoeffs
[s2
*2+1];
561 /* Interpolation is done over the first eight samples. */
562 for(; nsample
< 8; nsample
++) {
563 c1
= su1
[band
+nsample
];
564 c2
= su2
[band
+nsample
];
565 c2
= c1
* INTERPOLATE(mc1_l
,mc2_l
,nsample
) + c2
* INTERPOLATE(mc1_r
,mc2_r
,nsample
);
566 su1
[band
+nsample
] = c2
;
567 su2
[band
+nsample
] = c1
* 2.0 - c2
;
571 /* Apply the matrix without interpolation. */
573 case 0: /* M/S decoding */
574 for (; nsample
< 256; nsample
++) {
575 c1
= su1
[band
+nsample
];
576 c2
= su2
[band
+nsample
];
577 su1
[band
+nsample
] = c2
* 2.0;
578 su2
[band
+nsample
] = (c1
- c2
) * 2.0;
583 for (; nsample
< 256; nsample
++) {
584 c1
= su1
[band
+nsample
];
585 c2
= su2
[band
+nsample
];
586 su1
[band
+nsample
] = (c1
+ c2
) * 2.0;
587 su2
[band
+nsample
] = c2
* -2.0;
592 for (; nsample
< 256; nsample
++) {
593 c1
= su1
[band
+nsample
];
594 c2
= su2
[band
+nsample
];
595 su1
[band
+nsample
] = c1
+ c2
;
596 su2
[band
+nsample
] = c1
- c2
;
605 static void getChannelWeights (int indx
, int flag
, float ch
[2]){
611 ch
[0] = (float)(indx
& 7) / 7.0;
612 ch
[1] = sqrt(2 - ch
[0]*ch
[0]);
614 FFSWAP(float, ch
[0], ch
[1]);
618 static void channelWeighting (float *su1
, float *su2
, int *p3
)
621 /* w[x][y] y=0 is left y=1 is right */
624 if (p3
[1] != 7 || p3
[3] != 7){
625 getChannelWeights(p3
[1], p3
[0], w
[0]);
626 getChannelWeights(p3
[3], p3
[2], w
[1]);
628 for(band
= 1; band
< 4; band
++) {
629 /* scale the channels by the weights */
630 for(nsample
= 0; nsample
< 8; nsample
++) {
631 su1
[band
*256+nsample
] *= INTERPOLATE(w
[0][0], w
[0][1], nsample
);
632 su2
[band
*256+nsample
] *= INTERPOLATE(w
[1][0], w
[1][1], nsample
);
635 for(; nsample
< 256; nsample
++) {
636 su1
[band
*256+nsample
] *= w
[1][0];
637 su2
[band
*256+nsample
] *= w
[1][1];
645 * Decode a Sound Unit
647 * @param gb the GetBit context
648 * @param pSnd the channel unit to be used
649 * @param pOut the decoded samples before IQMF in float representation
650 * @param channelNum channel number
651 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
655 static int decodeChannelSoundUnit (ATRAC3Context
*q
, GetBitContext
*gb
, channel_unit
*pSnd
, float *pOut
, int channelNum
, int codingMode
)
657 int band
, result
=0, numSubbands
, lastTonal
, numBands
;
659 if (codingMode
== JOINT_STEREO
&& channelNum
== 1) {
660 if (get_bits(gb
,2) != 3) {
661 av_log(NULL
,AV_LOG_ERROR
,"JS mono Sound Unit id != 3.\n");
665 if (get_bits(gb
,6) != 0x28) {
666 av_log(NULL
,AV_LOG_ERROR
,"Sound Unit id != 0x28.\n");
671 /* number of coded QMF bands */
672 pSnd
->bandsCoded
= get_bits(gb
,2);
674 result
= decodeGainControl (gb
, &(pSnd
->gainBlock
[pSnd
->gcBlkSwitch
]), pSnd
->bandsCoded
);
675 if (result
) return result
;
677 pSnd
->numComponents
= decodeTonalComponents (gb
, pSnd
->components
, pSnd
->bandsCoded
);
678 if (pSnd
->numComponents
== -1) return -1;
680 numSubbands
= decodeSpectrum (gb
, pSnd
->spectrum
);
682 /* Merge the decoded spectrum and tonal components. */
683 lastTonal
= addTonalComponents (pSnd
->spectrum
, pSnd
->numComponents
, pSnd
->components
);
686 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
687 numBands
= (subbandTab
[numSubbands
] - 1) >> 8;
689 numBands
= FFMAX((lastTonal
+ 256) >> 8, numBands
);
692 /* Reconstruct time domain samples. */
693 for (band
=0; band
<4; band
++) {
694 /* Perform the IMDCT step without overlapping. */
695 if (band
<= numBands
) {
696 IMLT(&(pSnd
->spectrum
[band
*256]), pSnd
->IMDCT_buf
, band
&1);
698 memset(pSnd
->IMDCT_buf
, 0, 512 * sizeof(float));
700 /* gain compensation and overlapping */
701 gainCompensateAndOverlap (pSnd
->IMDCT_buf
, &(pSnd
->prevFrame
[band
*256]), &(pOut
[band
*256]),
702 &((pSnd
->gainBlock
[1 - (pSnd
->gcBlkSwitch
)]).gBlock
[band
]),
703 &((pSnd
->gainBlock
[pSnd
->gcBlkSwitch
]).gBlock
[band
]));
706 /* Swap the gain control buffers for the next frame. */
707 pSnd
->gcBlkSwitch
^= 1;
715 * @param q Atrac3 private context
716 * @param databuf the input data
719 static int decodeFrame(ATRAC3Context
*q
, const uint8_t* databuf
)
722 float *p1
, *p2
, *p3
, *p4
;
725 if (q
->codingMode
== JOINT_STEREO
) {
727 /* channel coupling mode */
728 /* decode Sound Unit 1 */
729 init_get_bits(&q
->gb
,databuf
,q
->bits_per_frame
);
731 result
= decodeChannelSoundUnit(q
,&q
->gb
, q
->pUnits
, q
->outSamples
, 0, JOINT_STEREO
);
735 /* Framedata of the su2 in the joint-stereo mode is encoded in
736 * reverse byte order so we need to swap it first. */
737 if (databuf
== q
->decoded_bytes_buffer
) {
738 uint8_t *ptr2
= q
->decoded_bytes_buffer
+q
->bytes_per_frame
-1;
739 ptr1
= q
->decoded_bytes_buffer
;
740 for (i
= 0; i
< (q
->bytes_per_frame
/2); i
++, ptr1
++, ptr2
--) {
741 FFSWAP(uint8_t,*ptr1
,*ptr2
);
744 const uint8_t *ptr2
= databuf
+q
->bytes_per_frame
-1;
745 for (i
= 0; i
< q
->bytes_per_frame
; i
++)
746 q
->decoded_bytes_buffer
[i
] = *ptr2
--;
749 /* Skip the sync codes (0xF8). */
750 ptr1
= q
->decoded_bytes_buffer
;
751 for (i
= 4; *ptr1
== 0xF8; i
++, ptr1
++) {
752 if (i
>= q
->bytes_per_frame
)
757 /* set the bitstream reader at the start of the second Sound Unit*/
758 init_get_bits(&q
->gb
,ptr1
,q
->bits_per_frame
);
760 /* Fill the Weighting coeffs delay buffer */
761 memmove(q
->weighting_delay
,&(q
->weighting_delay
[2]),4*sizeof(int));
762 q
->weighting_delay
[4] = get_bits1(&q
->gb
);
763 q
->weighting_delay
[5] = get_bits(&q
->gb
,3);
765 for (i
= 0; i
< 4; i
++) {
766 q
->matrix_coeff_index_prev
[i
] = q
->matrix_coeff_index_now
[i
];
767 q
->matrix_coeff_index_now
[i
] = q
->matrix_coeff_index_next
[i
];
768 q
->matrix_coeff_index_next
[i
] = get_bits(&q
->gb
,2);
771 /* Decode Sound Unit 2. */
772 result
= decodeChannelSoundUnit(q
,&q
->gb
, &q
->pUnits
[1], &q
->outSamples
[1024], 1, JOINT_STEREO
);
776 /* Reconstruct the channel coefficients. */
777 reverseMatrixing(q
->outSamples
, &q
->outSamples
[1024], q
->matrix_coeff_index_prev
, q
->matrix_coeff_index_now
);
779 channelWeighting(q
->outSamples
, &q
->outSamples
[1024], q
->weighting_delay
);
782 /* normal stereo mode or mono */
783 /* Decode the channel sound units. */
784 for (i
=0 ; i
<q
->channels
; i
++) {
786 /* Set the bitstream reader at the start of a channel sound unit. */
787 init_get_bits(&q
->gb
, databuf
+((i
*q
->bytes_per_frame
)/q
->channels
), (q
->bits_per_frame
)/q
->channels
);
789 result
= decodeChannelSoundUnit(q
,&q
->gb
, &q
->pUnits
[i
], &q
->outSamples
[i
*1024], i
, q
->codingMode
);
795 /* Apply the iQMF synthesis filter. */
797 for (i
=0 ; i
<q
->channels
; i
++) {
801 atrac_iqmf (p1
, p2
, 256, p1
, q
->pUnits
[i
].delayBuf1
, q
->tempBuf
);
802 atrac_iqmf (p4
, p3
, 256, p3
, q
->pUnits
[i
].delayBuf2
, q
->tempBuf
);
803 atrac_iqmf (p1
, p3
, 512, p1
, q
->pUnits
[i
].delayBuf3
, q
->tempBuf
);
812 * Atrac frame decoding
814 * @param avctx pointer to the AVCodecContext
817 static int atrac3_decode_frame(AVCodecContext
*avctx
,
818 void *data
, int *data_size
,
820 const uint8_t *buf
= avpkt
->data
;
821 int buf_size
= avpkt
->size
;
822 ATRAC3Context
*q
= avctx
->priv_data
;
824 const uint8_t* databuf
;
825 int16_t* samples
= data
;
827 if (buf_size
< avctx
->block_align
)
830 /* Check if we need to descramble and what buffer to pass on. */
831 if (q
->scrambled_stream
) {
832 decode_bytes(buf
, q
->decoded_bytes_buffer
, avctx
->block_align
);
833 databuf
= q
->decoded_bytes_buffer
;
838 result
= decodeFrame(q
, databuf
);
841 av_log(NULL
,AV_LOG_ERROR
,"Frame decoding error!\n");
845 if (q
->channels
== 1) {
847 for (i
= 0; i
<1024; i
++)
848 samples
[i
] = av_clip_int16(round(q
->outSamples
[i
]));
849 *data_size
= 1024 * sizeof(int16_t);
852 for (i
= 0; i
< 1024; i
++) {
853 samples
[i
*2] = av_clip_int16(round(q
->outSamples
[i
]));
854 samples
[i
*2+1] = av_clip_int16(round(q
->outSamples
[1024+i
]));
856 *data_size
= 2048 * sizeof(int16_t);
859 return avctx
->block_align
;
864 * Atrac3 initialization
866 * @param avctx pointer to the AVCodecContext
869 static av_cold
int atrac3_decode_init(AVCodecContext
*avctx
)
872 const uint8_t *edata_ptr
= avctx
->extradata
;
873 ATRAC3Context
*q
= avctx
->priv_data
;
874 static VLC_TYPE atrac3_vlc_table
[4096][2];
875 static int vlcs_initialized
= 0;
877 /* Take data from the AVCodecContext (RM container). */
878 q
->sample_rate
= avctx
->sample_rate
;
879 q
->channels
= avctx
->channels
;
880 q
->bit_rate
= avctx
->bit_rate
;
881 q
->bits_per_frame
= avctx
->block_align
* 8;
882 q
->bytes_per_frame
= avctx
->block_align
;
884 /* Take care of the codec-specific extradata. */
885 if (avctx
->extradata_size
== 14) {
886 /* Parse the extradata, WAV format */
887 av_log(avctx
,AV_LOG_DEBUG
,"[0-1] %d\n",bytestream_get_le16(&edata_ptr
)); //Unknown value always 1
888 q
->samples_per_channel
= bytestream_get_le32(&edata_ptr
);
889 q
->codingMode
= bytestream_get_le16(&edata_ptr
);
890 av_log(avctx
,AV_LOG_DEBUG
,"[8-9] %d\n",bytestream_get_le16(&edata_ptr
)); //Dupe of coding mode
891 q
->frame_factor
= bytestream_get_le16(&edata_ptr
); //Unknown always 1
892 av_log(avctx
,AV_LOG_DEBUG
,"[12-13] %d\n",bytestream_get_le16(&edata_ptr
)); //Unknown always 0
895 q
->samples_per_frame
= 1024 * q
->channels
;
896 q
->atrac3version
= 4;
899 q
->codingMode
= JOINT_STEREO
;
901 q
->codingMode
= STEREO
;
903 q
->scrambled_stream
= 0;
905 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
)) {
907 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
);
911 } else if (avctx
->extradata_size
== 10) {
912 /* Parse the extradata, RM format. */
913 q
->atrac3version
= bytestream_get_be32(&edata_ptr
);
914 q
->samples_per_frame
= bytestream_get_be16(&edata_ptr
);
915 q
->delay
= bytestream_get_be16(&edata_ptr
);
916 q
->codingMode
= bytestream_get_be16(&edata_ptr
);
918 q
->samples_per_channel
= q
->samples_per_frame
/ q
->channels
;
919 q
->scrambled_stream
= 1;
922 av_log(NULL
,AV_LOG_ERROR
,"Unknown extradata size %d.\n",avctx
->extradata_size
);
924 /* Check the extradata. */
926 if (q
->atrac3version
!= 4) {
927 av_log(avctx
,AV_LOG_ERROR
,"Version %d != 4.\n",q
->atrac3version
);
931 if (q
->samples_per_frame
!= 1024 && q
->samples_per_frame
!= 2048) {
932 av_log(avctx
,AV_LOG_ERROR
,"Unknown amount of samples per frame %d.\n",q
->samples_per_frame
);
936 if (q
->delay
!= 0x88E) {
937 av_log(avctx
,AV_LOG_ERROR
,"Unknown amount of delay %x != 0x88E.\n",q
->delay
);
941 if (q
->codingMode
== STEREO
) {
942 av_log(avctx
,AV_LOG_DEBUG
,"Normal stereo detected.\n");
943 } else if (q
->codingMode
== JOINT_STEREO
) {
944 av_log(avctx
,AV_LOG_DEBUG
,"Joint stereo detected.\n");
946 av_log(avctx
,AV_LOG_ERROR
,"Unknown channel coding mode %x!\n",q
->codingMode
);
950 if (avctx
->channels
<= 0 || avctx
->channels
> 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
951 av_log(avctx
,AV_LOG_ERROR
,"Channel configuration error!\n");
956 if(avctx
->block_align
>= UINT_MAX
/2)
959 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
960 * this is for the bitstream reader. */
961 if ((q
->decoded_bytes_buffer
= av_mallocz((avctx
->block_align
+(4-avctx
->block_align
%4) + FF_INPUT_BUFFER_PADDING_SIZE
))) == NULL
)
962 return AVERROR(ENOMEM
);
965 /* Initialize the VLC tables. */
966 if (!vlcs_initialized
) {
967 for (i
=0 ; i
<7 ; i
++) {
968 spectral_coeff_tab
[i
].table
= &atrac3_vlc_table
[atrac3_vlc_offs
[i
]];
969 spectral_coeff_tab
[i
].table_allocated
= atrac3_vlc_offs
[i
+ 1] - atrac3_vlc_offs
[i
];
970 init_vlc (&spectral_coeff_tab
[i
], 9, huff_tab_sizes
[i
],
972 huff_codes
[i
], 1, 1, INIT_VLC_USE_NEW_STATIC
);
974 vlcs_initialized
= 1;
977 init_atrac3_transforms(q
);
979 atrac_generate_tables();
981 /* Generate gain tables. */
982 for (i
=0 ; i
<16 ; i
++)
983 gain_tab1
[i
] = powf (2.0, (4 - i
));
985 for (i
=-15 ; i
<16 ; i
++)
986 gain_tab2
[i
+15] = powf (2.0, i
* -0.125);
988 /* init the joint-stereo decoding data */
989 q
->weighting_delay
[0] = 0;
990 q
->weighting_delay
[1] = 7;
991 q
->weighting_delay
[2] = 0;
992 q
->weighting_delay
[3] = 7;
993 q
->weighting_delay
[4] = 0;
994 q
->weighting_delay
[5] = 7;
996 for (i
=0; i
<4; i
++) {
997 q
->matrix_coeff_index_prev
[i
] = 3;
998 q
->matrix_coeff_index_now
[i
] = 3;
999 q
->matrix_coeff_index_next
[i
] = 3;
1002 dsputil_init(&dsp
, avctx
);
1004 q
->pUnits
= av_mallocz(sizeof(channel_unit
)*q
->channels
);
1006 av_free(q
->decoded_bytes_buffer
);
1007 return AVERROR(ENOMEM
);
1010 avctx
->sample_fmt
= SAMPLE_FMT_S16
;
1015 AVCodec atrac3_decoder
=
1018 .type
= CODEC_TYPE_AUDIO
,
1019 .id
= CODEC_ID_ATRAC3
,
1020 .priv_data_size
= sizeof(ATRAC3Context
),
1021 .init
= atrac3_decode_init
,
1022 .close
= atrac3_decode_close
,
1023 .decode
= atrac3_decode_frame
,
1024 .long_name
= NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),