cosmetics: Align codec declarations
[libav.git] / libavcodec / atrac3.c
1 /*
2 * Atrac 3 compatible decoder
3 * Copyright (c) 2006-2008 Maxim Poliakovski
4 * Copyright (c) 2006-2008 Benjamin Larsson
5 *
6 * This file is part of Libav.
7 *
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.
12 *
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.
17 *
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
21 */
22
23 /**
24 * @file
25 * Atrac 3 compatible decoder.
26 * This decoder handles Sony's ATRAC3 data.
27 *
28 * Container formats used to store atrac 3 data:
29 * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
30 *
31 * To use this decoder, a calling application must supply the extradata
32 * bytes provided in the containers above.
33 */
34
35 #include <math.h>
36 #include <stddef.h>
37 #include <stdio.h>
38
39 #include "avcodec.h"
40 #include "get_bits.h"
41 #include "dsputil.h"
42 #include "bytestream.h"
43 #include "fft.h"
44 #include "fmtconvert.h"
45
46 #include "atrac.h"
47 #include "atrac3data.h"
48
49 #define JOINT_STEREO 0x12
50 #define STEREO 0x2
51
52 #define SAMPLES_PER_FRAME 1024
53 #define MDCT_SIZE 512
54
55 /* These structures are needed to store the parsed gain control data. */
56 typedef struct {
57 int num_gain_data;
58 int levcode[8];
59 int loccode[8];
60 } gain_info;
61
62 typedef struct {
63 gain_info gBlock[4];
64 } gain_block;
65
66 typedef struct {
67 int pos;
68 int numCoefs;
69 float coef[8];
70 } tonal_component;
71
72 typedef struct {
73 int bandsCoded;
74 int numComponents;
75 tonal_component components[64];
76 float prevFrame[SAMPLES_PER_FRAME];
77 int gcBlkSwitch;
78 gain_block gainBlock[2];
79
80 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
81 DECLARE_ALIGNED(32, float, IMDCT_buf)[SAMPLES_PER_FRAME];
82
83 float delayBuf1[46]; ///<qmf delay buffers
84 float delayBuf2[46];
85 float delayBuf3[46];
86 } channel_unit;
87
88 typedef struct {
89 AVFrame frame;
90 GetBitContext gb;
91 //@{
92 /** stream data */
93 int channels;
94 int codingMode;
95 int bit_rate;
96 int sample_rate;
97 int samples_per_channel;
98 int samples_per_frame;
99
100 int bits_per_frame;
101 int bytes_per_frame;
102 int pBs;
103 channel_unit* pUnits;
104 //@}
105 //@{
106 /** joint-stereo related variables */
107 int matrix_coeff_index_prev[4];
108 int matrix_coeff_index_now[4];
109 int matrix_coeff_index_next[4];
110 int weighting_delay[6];
111 //@}
112 //@{
113 /** data buffers */
114 float *outSamples[2];
115 uint8_t* decoded_bytes_buffer;
116 float tempBuf[1070];
117 //@}
118 //@{
119 /** extradata */
120 int atrac3version;
121 int delay;
122 int scrambled_stream;
123 int frame_factor;
124 //@}
125
126 FFTContext mdct_ctx;
127 FmtConvertContext fmt_conv;
128 } ATRAC3Context;
129
130 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
131 static VLC spectral_coeff_tab[7];
132 static float gain_tab1[16];
133 static float gain_tab2[31];
134 static DSPContext dsp;
135
136
137 /**
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.
140 *
141 * @param pInput float input
142 * @param pOutput float output
143 * @param odd_band 1 if the band is an odd band
144 */
145
146 static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
147 {
148 int i;
149
150 if (odd_band) {
151 /**
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.
158 */
159
160 for (i=0; i<128; i++)
161 FFSWAP(float, pInput[i], pInput[255-i]);
162 }
163
164 q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
165
166 /* Perform windowing on the output. */
167 dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
168
169 }
170
171
172 /**
173 * Atrac 3 indata descrambling, only used for data coming from the rm container
174 *
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
178 */
179
180 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
181 int i, off;
182 uint32_t c;
183 const uint32_t* buf;
184 uint32_t* obuf = (uint32_t*) out;
185
186 off = (intptr_t)inbuffer & 3;
187 buf = (const uint32_t*) (inbuffer - off);
188 c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
189 bytes += 3 + off;
190 for (i = 0; i < bytes/4; i++)
191 obuf[i] = c ^ buf[i];
192
193 if (off)
194 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
195
196 return off;
197 }
198
199
200 static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
201 float enc_window[256];
202 int i;
203
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;
208
209 if (!mdct_window[0])
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];
213 }
214
215 /* Initialize the MDCT transform. */
216 return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
217 }
218
219 /**
220 * Atrac3 uninit, free all allocated memory
221 */
222
223 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
224 {
225 ATRAC3Context *q = avctx->priv_data;
226
227 av_free(q->pUnits);
228 av_free(q->decoded_bytes_buffer);
229 av_freep(&q->outSamples[0]);
230
231 ff_mdct_end(&q->mdct_ctx);
232
233 return 0;
234 }
235
236 /**
237 / * Mantissa decoding
238 *
239 * @param gb the GetBit context
240 * @param selector what table is the output values coded with
241 * @param codingFlag constant length coding or variable length coding
242 * @param mantissas mantissa output table
243 * @param numCodes amount of values to get
244 */
245
246 static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int codingFlag, int* mantissas, int numCodes)
247 {
248 int numBits, cnt, code, huffSymb;
249
250 if (selector == 1)
251 numCodes /= 2;
252
253 if (codingFlag != 0) {
254 /* constant length coding (CLC) */
255 numBits = CLCLengthTab[selector];
256
257 if (selector > 1) {
258 for (cnt = 0; cnt < numCodes; cnt++) {
259 if (numBits)
260 code = get_sbits(gb, numBits);
261 else
262 code = 0;
263 mantissas[cnt] = code;
264 }
265 } else {
266 for (cnt = 0; cnt < numCodes; cnt++) {
267 if (numBits)
268 code = get_bits(gb, numBits); //numBits is always 4 in this case
269 else
270 code = 0;
271 mantissas[cnt*2] = seTab_0[code >> 2];
272 mantissas[cnt*2+1] = seTab_0[code & 3];
273 }
274 }
275 } else {
276 /* variable length coding (VLC) */
277 if (selector != 1) {
278 for (cnt = 0; cnt < numCodes; cnt++) {
279 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
280 huffSymb += 1;
281 code = huffSymb >> 1;
282 if (huffSymb & 1)
283 code = -code;
284 mantissas[cnt] = code;
285 }
286 } else {
287 for (cnt = 0; cnt < numCodes; cnt++) {
288 huffSymb = get_vlc2(gb, spectral_coeff_tab[selector-1].table, spectral_coeff_tab[selector-1].bits, 3);
289 mantissas[cnt*2] = decTable1[huffSymb*2];
290 mantissas[cnt*2+1] = decTable1[huffSymb*2+1];
291 }
292 }
293 }
294 }
295
296 /**
297 * Restore the quantized band spectrum coefficients
298 *
299 * @param gb the GetBit context
300 * @param pOut decoded band spectrum
301 * @return outSubbands subband counter, fix for broken specification/files
302 */
303
304 static int decodeSpectrum (GetBitContext *gb, float *pOut)
305 {
306 int numSubbands, codingMode, cnt, first, last, subbWidth, *pIn;
307 int subband_vlc_index[32], SF_idxs[32];
308 int mantissas[128];
309 float SF;
310
311 numSubbands = get_bits(gb, 5); // number of coded subbands
312 codingMode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
313
314 /* Get the VLC selector table for the subbands, 0 means not coded. */
315 for (cnt = 0; cnt <= numSubbands; cnt++)
316 subband_vlc_index[cnt] = get_bits(gb, 3);
317
318 /* Read the scale factor indexes from the stream. */
319 for (cnt = 0; cnt <= numSubbands; cnt++) {
320 if (subband_vlc_index[cnt] != 0)
321 SF_idxs[cnt] = get_bits(gb, 6);
322 }
323
324 for (cnt = 0; cnt <= numSubbands; cnt++) {
325 first = subbandTab[cnt];
326 last = subbandTab[cnt+1];
327
328 subbWidth = last - first;
329
330 if (subband_vlc_index[cnt] != 0) {
331 /* Decode spectral coefficients for this subband. */
332 /* TODO: This can be done faster is several blocks share the
333 * same VLC selector (subband_vlc_index) */
334 readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
335
336 /* Decode the scale factor for this subband. */
337 SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
338
339 /* Inverse quantize the coefficients. */
340 for (pIn=mantissas ; first<last; first++, pIn++)
341 pOut[first] = *pIn * SF;
342 } else {
343 /* This subband was not coded, so zero the entire subband. */
344 memset(pOut+first, 0, subbWidth*sizeof(float));
345 }
346 }
347
348 /* Clear the subbands that were not coded. */
349 first = subbandTab[cnt];
350 memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
351 return numSubbands;
352 }
353
354 /**
355 * Restore the quantized tonal components
356 *
357 * @param gb the GetBit context
358 * @param pComponent tone component
359 * @param numBands amount of coded bands
360 */
361
362 static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent, int numBands)
363 {
364 int i,j,k,cnt;
365 int components, coding_mode_selector, coding_mode, coded_values_per_component;
366 int sfIndx, coded_values, max_coded_values, quant_step_index, coded_components;
367 int band_flags[4], mantissa[8];
368 float *pCoef;
369 float scalefactor;
370 int component_count = 0;
371
372 components = get_bits(gb,5);
373
374 /* no tonal components */
375 if (components == 0)
376 return 0;
377
378 coding_mode_selector = get_bits(gb,2);
379 if (coding_mode_selector == 2)
380 return AVERROR_INVALIDDATA;
381
382 coding_mode = coding_mode_selector & 1;
383
384 for (i = 0; i < components; i++) {
385 for (cnt = 0; cnt <= numBands; cnt++)
386 band_flags[cnt] = get_bits1(gb);
387
388 coded_values_per_component = get_bits(gb,3);
389
390 quant_step_index = get_bits(gb,3);
391 if (quant_step_index <= 1)
392 return AVERROR_INVALIDDATA;
393
394 if (coding_mode_selector == 3)
395 coding_mode = get_bits1(gb);
396
397 for (j = 0; j < (numBands + 1) * 4; j++) {
398 if (band_flags[j >> 2] == 0)
399 continue;
400
401 coded_components = get_bits(gb,3);
402
403 for (k=0; k<coded_components; k++) {
404 sfIndx = get_bits(gb,6);
405 if (component_count >= 64)
406 return AVERROR_INVALIDDATA;
407 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
408 max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
409 coded_values = coded_values_per_component + 1;
410 coded_values = FFMIN(max_coded_values,coded_values);
411
412 scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
413
414 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
415
416 pComponent[component_count].numCoefs = coded_values;
417
418 /* inverse quant */
419 pCoef = pComponent[component_count].coef;
420 for (cnt = 0; cnt < coded_values; cnt++)
421 pCoef[cnt] = mantissa[cnt] * scalefactor;
422
423 component_count++;
424 }
425 }
426 }
427
428 return component_count;
429 }
430
431 /**
432 * Decode gain parameters for the coded bands
433 *
434 * @param gb the GetBit context
435 * @param pGb the gainblock for the current band
436 * @param numBands amount of coded bands
437 */
438
439 static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
440 {
441 int i, cf, numData;
442 int *pLevel, *pLoc;
443
444 gain_info *pGain = pGb->gBlock;
445
446 for (i=0 ; i<=numBands; i++)
447 {
448 numData = get_bits(gb,3);
449 pGain[i].num_gain_data = numData;
450 pLevel = pGain[i].levcode;
451 pLoc = pGain[i].loccode;
452
453 for (cf = 0; cf < numData; cf++){
454 pLevel[cf]= get_bits(gb,4);
455 pLoc [cf]= get_bits(gb,5);
456 if(cf && pLoc[cf] <= pLoc[cf-1])
457 return AVERROR_INVALIDDATA;
458 }
459 }
460
461 /* Clear the unused blocks. */
462 for (; i<4 ; i++)
463 pGain[i].num_gain_data = 0;
464
465 return 0;
466 }
467
468 /**
469 * Apply gain parameters and perform the MDCT overlapping part
470 *
471 * @param pIn input float buffer
472 * @param pPrev previous float buffer to perform overlap against
473 * @param pOut output float buffer
474 * @param pGain1 current band gain info
475 * @param pGain2 next band gain info
476 */
477
478 static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gain_info *pGain1, gain_info *pGain2)
479 {
480 /* gain compensation function */
481 float gain1, gain2, gain_inc;
482 int cnt, numdata, nsample, startLoc, endLoc;
483
484
485 if (pGain2->num_gain_data == 0)
486 gain1 = 1.0;
487 else
488 gain1 = gain_tab1[pGain2->levcode[0]];
489
490 if (pGain1->num_gain_data == 0) {
491 for (cnt = 0; cnt < 256; cnt++)
492 pOut[cnt] = pIn[cnt] * gain1 + pPrev[cnt];
493 } else {
494 numdata = pGain1->num_gain_data;
495 pGain1->loccode[numdata] = 32;
496 pGain1->levcode[numdata] = 4;
497
498 nsample = 0; // current sample = 0
499
500 for (cnt = 0; cnt < numdata; cnt++) {
501 startLoc = pGain1->loccode[cnt] * 8;
502 endLoc = startLoc + 8;
503
504 gain2 = gain_tab1[pGain1->levcode[cnt]];
505 gain_inc = gain_tab2[(pGain1->levcode[cnt+1] - pGain1->levcode[cnt])+15];
506
507 /* interpolate */
508 for (; nsample < startLoc; nsample++)
509 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
510
511 /* interpolation is done over eight samples */
512 for (; nsample < endLoc; nsample++) {
513 pOut[nsample] = (pIn[nsample] * gain1 + pPrev[nsample]) * gain2;
514 gain2 *= gain_inc;
515 }
516 }
517
518 for (; nsample < 256; nsample++)
519 pOut[nsample] = (pIn[nsample] * gain1) + pPrev[nsample];
520 }
521
522 /* Delay for the overlapping part. */
523 memcpy(pPrev, &pIn[256], 256*sizeof(float));
524 }
525
526 /**
527 * Combine the tonal band spectrum and regular band spectrum
528 * Return position of the last tonal coefficient
529 *
530 * @param pSpectrum output spectrum buffer
531 * @param numComponents amount of tonal components
532 * @param pComponent tonal components for this band
533 */
534
535 static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
536 {
537 int cnt, i, lastPos = -1;
538 float *pIn, *pOut;
539
540 for (cnt = 0; cnt < numComponents; cnt++){
541 lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
542 pIn = pComponent[cnt].coef;
543 pOut = &(pSpectrum[pComponent[cnt].pos]);
544
545 for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
546 pOut[i] += pIn[i];
547 }
548
549 return lastPos;
550 }
551
552
553 #define INTERPOLATE(old,new,nsample) ((old) + (nsample)*0.125*((new)-(old)))
554
555 static void reverseMatrixing(float *su1, float *su2, int *pPrevCode, int *pCurrCode)
556 {
557 int i, band, nsample, s1, s2;
558 float c1, c2;
559 float mc1_l, mc1_r, mc2_l, mc2_r;
560
561 for (i=0,band = 0; band < 4*256; band+=256,i++) {
562 s1 = pPrevCode[i];
563 s2 = pCurrCode[i];
564 nsample = 0;
565
566 if (s1 != s2) {
567 /* Selector value changed, interpolation needed. */
568 mc1_l = matrixCoeffs[s1*2];
569 mc1_r = matrixCoeffs[s1*2+1];
570 mc2_l = matrixCoeffs[s2*2];
571 mc2_r = matrixCoeffs[s2*2+1];
572
573 /* Interpolation is done over the first eight samples. */
574 for(; nsample < 8; nsample++) {
575 c1 = su1[band+nsample];
576 c2 = su2[band+nsample];
577 c2 = c1 * INTERPOLATE(mc1_l,mc2_l,nsample) + c2 * INTERPOLATE(mc1_r,mc2_r,nsample);
578 su1[band+nsample] = c2;
579 su2[band+nsample] = c1 * 2.0 - c2;
580 }
581 }
582
583 /* Apply the matrix without interpolation. */
584 switch (s2) {
585 case 0: /* M/S decoding */
586 for (; nsample < 256; nsample++) {
587 c1 = su1[band+nsample];
588 c2 = su2[band+nsample];
589 su1[band+nsample] = c2 * 2.0;
590 su2[band+nsample] = (c1 - c2) * 2.0;
591 }
592 break;
593
594 case 1:
595 for (; nsample < 256; nsample++) {
596 c1 = su1[band+nsample];
597 c2 = su2[band+nsample];
598 su1[band+nsample] = (c1 + c2) * 2.0;
599 su2[band+nsample] = c2 * -2.0;
600 }
601 break;
602 case 2:
603 case 3:
604 for (; nsample < 256; nsample++) {
605 c1 = su1[band+nsample];
606 c2 = su2[band+nsample];
607 su1[band+nsample] = c1 + c2;
608 su2[band+nsample] = c1 - c2;
609 }
610 break;
611 default:
612 assert(0);
613 }
614 }
615 }
616
617 static void getChannelWeights (int indx, int flag, float ch[2]){
618
619 if (indx == 7) {
620 ch[0] = 1.0;
621 ch[1] = 1.0;
622 } else {
623 ch[0] = (float)(indx & 7) / 7.0;
624 ch[1] = sqrt(2 - ch[0]*ch[0]);
625 if(flag)
626 FFSWAP(float, ch[0], ch[1]);
627 }
628 }
629
630 static void channelWeighting (float *su1, float *su2, int *p3)
631 {
632 int band, nsample;
633 /* w[x][y] y=0 is left y=1 is right */
634 float w[2][2];
635
636 if (p3[1] != 7 || p3[3] != 7){
637 getChannelWeights(p3[1], p3[0], w[0]);
638 getChannelWeights(p3[3], p3[2], w[1]);
639
640 for(band = 1; band < 4; band++) {
641 /* scale the channels by the weights */
642 for(nsample = 0; nsample < 8; nsample++) {
643 su1[band*256+nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample);
644 su2[band*256+nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample);
645 }
646
647 for(; nsample < 256; nsample++) {
648 su1[band*256+nsample] *= w[1][0];
649 su2[band*256+nsample] *= w[1][1];
650 }
651 }
652 }
653 }
654
655
656 /**
657 * Decode a Sound Unit
658 *
659 * @param gb the GetBit context
660 * @param pSnd the channel unit to be used
661 * @param pOut the decoded samples before IQMF in float representation
662 * @param channelNum channel number
663 * @param codingMode the coding mode (JOINT_STEREO or regular stereo/mono)
664 */
665
666
667 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
668 {
669 int band, result=0, numSubbands, lastTonal, numBands;
670
671 if (codingMode == JOINT_STEREO && channelNum == 1) {
672 if (get_bits(gb,2) != 3) {
673 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
674 return AVERROR_INVALIDDATA;
675 }
676 } else {
677 if (get_bits(gb,6) != 0x28) {
678 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
679 return AVERROR_INVALIDDATA;
680 }
681 }
682
683 /* number of coded QMF bands */
684 pSnd->bandsCoded = get_bits(gb,2);
685
686 result = decodeGainControl (gb, &(pSnd->gainBlock[pSnd->gcBlkSwitch]), pSnd->bandsCoded);
687 if (result) return result;
688
689 pSnd->numComponents = decodeTonalComponents (gb, pSnd->components, pSnd->bandsCoded);
690 if (pSnd->numComponents == -1) return -1;
691
692 numSubbands = decodeSpectrum (gb, pSnd->spectrum);
693
694 /* Merge the decoded spectrum and tonal components. */
695 lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
696
697
698 /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
699 numBands = (subbandTab[numSubbands] - 1) >> 8;
700 if (lastTonal >= 0)
701 numBands = FFMAX((lastTonal + 256) >> 8, numBands);
702
703
704 /* Reconstruct time domain samples. */
705 for (band=0; band<4; band++) {
706 /* Perform the IMDCT step without overlapping. */
707 if (band <= numBands) {
708 IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
709 } else
710 memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
711
712 /* gain compensation and overlapping */
713 gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
714 &pOut[band * 256],
715 &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
716 &pSnd->gainBlock[ pSnd->gcBlkSwitch].gBlock[band]);
717 }
718
719 /* Swap the gain control buffers for the next frame. */
720 pSnd->gcBlkSwitch ^= 1;
721
722 return 0;
723 }
724
725 /**
726 * Frame handling
727 *
728 * @param q Atrac3 private context
729 * @param databuf the input data
730 */
731
732 static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
733 float **out_samples)
734 {
735 int result, i;
736 float *p1, *p2, *p3, *p4;
737 uint8_t *ptr1;
738
739 if (q->codingMode == JOINT_STEREO) {
740
741 /* channel coupling mode */
742 /* decode Sound Unit 1 */
743 init_get_bits(&q->gb,databuf,q->bits_per_frame);
744
745 result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
746 if (result != 0)
747 return result;
748
749 /* Framedata of the su2 in the joint-stereo mode is encoded in
750 * reverse byte order so we need to swap it first. */
751 if (databuf == q->decoded_bytes_buffer) {
752 uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
753 ptr1 = q->decoded_bytes_buffer;
754 for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
755 FFSWAP(uint8_t,*ptr1,*ptr2);
756 }
757 } else {
758 const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
759 for (i = 0; i < q->bytes_per_frame; i++)
760 q->decoded_bytes_buffer[i] = *ptr2--;
761 }
762
763 /* Skip the sync codes (0xF8). */
764 ptr1 = q->decoded_bytes_buffer;
765 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
766 if (i >= q->bytes_per_frame)
767 return AVERROR_INVALIDDATA;
768 }
769
770
771 /* set the bitstream reader at the start of the second Sound Unit*/
772 init_get_bits(&q->gb,ptr1,q->bits_per_frame);
773
774 /* Fill the Weighting coeffs delay buffer */
775 memmove(q->weighting_delay,&(q->weighting_delay[2]),4*sizeof(int));
776 q->weighting_delay[4] = get_bits1(&q->gb);
777 q->weighting_delay[5] = get_bits(&q->gb,3);
778
779 for (i = 0; i < 4; i++) {
780 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
781 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
782 q->matrix_coeff_index_next[i] = get_bits(&q->gb,2);
783 }
784
785 /* Decode Sound Unit 2. */
786 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
787 if (result != 0)
788 return result;
789
790 /* Reconstruct the channel coefficients. */
791 reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
792
793 channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
794
795 } else {
796 /* normal stereo mode or mono */
797 /* Decode the channel sound units. */
798 for (i=0 ; i<q->channels ; i++) {
799
800 /* Set the bitstream reader at the start of a channel sound unit. */
801 init_get_bits(&q->gb,
802 databuf + i * q->bytes_per_frame / q->channels,
803 q->bits_per_frame / q->channels);
804
805 result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
806 if (result != 0)
807 return result;
808 }
809 }
810
811 /* Apply the iQMF synthesis filter. */
812 for (i=0 ; i<q->channels ; i++) {
813 p1 = out_samples[i];
814 p2= p1+256;
815 p3= p2+256;
816 p4= p3+256;
817 ff_atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
818 ff_atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
819 ff_atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
820 }
821
822 return 0;
823 }
824
825
826 /**
827 * Atrac frame decoding
828 *
829 * @param avctx pointer to the AVCodecContext
830 */
831
832 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
833 int *got_frame_ptr, AVPacket *avpkt)
834 {
835 const uint8_t *buf = avpkt->data;
836 int buf_size = avpkt->size;
837 ATRAC3Context *q = avctx->priv_data;
838 int result;
839 const uint8_t* databuf;
840 float *samples_flt;
841 int16_t *samples_s16;
842
843 if (buf_size < avctx->block_align) {
844 av_log(avctx, AV_LOG_ERROR,
845 "Frame too small (%d bytes). Truncated file?\n", buf_size);
846 return AVERROR_INVALIDDATA;
847 }
848
849 /* get output buffer */
850 q->frame.nb_samples = SAMPLES_PER_FRAME;
851 if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
852 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
853 return result;
854 }
855 samples_flt = (float *)q->frame.data[0];
856 samples_s16 = (int16_t *)q->frame.data[0];
857
858 /* Check if we need to descramble and what buffer to pass on. */
859 if (q->scrambled_stream) {
860 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
861 databuf = q->decoded_bytes_buffer;
862 } else {
863 databuf = buf;
864 }
865
866 if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
867 result = decodeFrame(q, databuf, &samples_flt);
868 else
869 result = decodeFrame(q, databuf, q->outSamples);
870
871 if (result != 0) {
872 av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
873 return result;
874 }
875
876 /* interleave */
877 if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
878 q->fmt_conv.float_interleave(samples_flt,
879 (const float **)q->outSamples,
880 SAMPLES_PER_FRAME, 2);
881 } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
882 q->fmt_conv.float_to_int16_interleave(samples_s16,
883 (const float **)q->outSamples,
884 SAMPLES_PER_FRAME, q->channels);
885 }
886
887 *got_frame_ptr = 1;
888 *(AVFrame *)data = q->frame;
889
890 return avctx->block_align;
891 }
892
893
894 /**
895 * Atrac3 initialization
896 *
897 * @param avctx pointer to the AVCodecContext
898 */
899
900 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
901 {
902 int i, ret;
903 const uint8_t *edata_ptr = avctx->extradata;
904 ATRAC3Context *q = avctx->priv_data;
905 static VLC_TYPE atrac3_vlc_table[4096][2];
906 static int vlcs_initialized = 0;
907
908 /* Take data from the AVCodecContext (RM container). */
909 q->sample_rate = avctx->sample_rate;
910 q->channels = avctx->channels;
911 q->bit_rate = avctx->bit_rate;
912 q->bits_per_frame = avctx->block_align * 8;
913 q->bytes_per_frame = avctx->block_align;
914
915 /* Take care of the codec-specific extradata. */
916 if (avctx->extradata_size == 14) {
917 /* Parse the extradata, WAV format */
918 av_log(avctx,AV_LOG_DEBUG,"[0-1] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown value always 1
919 q->samples_per_channel = bytestream_get_le32(&edata_ptr);
920 q->codingMode = bytestream_get_le16(&edata_ptr);
921 av_log(avctx,AV_LOG_DEBUG,"[8-9] %d\n",bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
922 q->frame_factor = bytestream_get_le16(&edata_ptr); //Unknown always 1
923 av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr)); //Unknown always 0
924
925 /* setup */
926 q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
927 q->atrac3version = 4;
928 q->delay = 0x88E;
929 if (q->codingMode)
930 q->codingMode = JOINT_STEREO;
931 else
932 q->codingMode = STEREO;
933
934 q->scrambled_stream = 0;
935
936 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)) {
937 } else {
938 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);
939 return AVERROR_INVALIDDATA;
940 }
941
942 } else if (avctx->extradata_size == 10) {
943 /* Parse the extradata, RM format. */
944 q->atrac3version = bytestream_get_be32(&edata_ptr);
945 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
946 q->delay = bytestream_get_be16(&edata_ptr);
947 q->codingMode = bytestream_get_be16(&edata_ptr);
948
949 q->samples_per_channel = q->samples_per_frame / q->channels;
950 q->scrambled_stream = 1;
951
952 } else {
953 av_log(NULL,AV_LOG_ERROR,"Unknown extradata size %d.\n",avctx->extradata_size);
954 }
955 /* Check the extradata. */
956
957 if (q->atrac3version != 4) {
958 av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
959 return AVERROR_INVALIDDATA;
960 }
961
962 if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
963 av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
964 return AVERROR_INVALIDDATA;
965 }
966
967 if (q->delay != 0x88E) {
968 av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
969 return AVERROR_INVALIDDATA;
970 }
971
972 if (q->codingMode == STEREO) {
973 av_log(avctx,AV_LOG_DEBUG,"Normal stereo detected.\n");
974 } else if (q->codingMode == JOINT_STEREO) {
975 av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
976 } else {
977 av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
978 return AVERROR_INVALIDDATA;
979 }
980
981 if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
982 av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
983 return AVERROR(EINVAL);
984 }
985
986
987 if(avctx->block_align >= UINT_MAX/2)
988 return AVERROR(EINVAL);
989
990 /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
991 * this is for the bitstream reader. */
992 if ((q->decoded_bytes_buffer = av_mallocz((avctx->block_align+(4-avctx->block_align%4) + FF_INPUT_BUFFER_PADDING_SIZE))) == NULL)
993 return AVERROR(ENOMEM);
994
995
996 /* Initialize the VLC tables. */
997 if (!vlcs_initialized) {
998 for (i=0 ; i<7 ; i++) {
999 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
1000 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
1001 init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
1002 huff_bits[i], 1, 1,
1003 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
1004 }
1005 vlcs_initialized = 1;
1006 }
1007
1008 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
1009 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1010 else
1011 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1012
1013 if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
1014 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
1015 av_freep(&q->decoded_bytes_buffer);
1016 return ret;
1017 }
1018
1019 ff_atrac_generate_tables();
1020
1021 /* Generate gain tables. */
1022 for (i=0 ; i<16 ; i++)
1023 gain_tab1[i] = powf (2.0, (4 - i));
1024
1025 for (i=-15 ; i<16 ; i++)
1026 gain_tab2[i+15] = powf (2.0, i * -0.125);
1027
1028 /* init the joint-stereo decoding data */
1029 q->weighting_delay[0] = 0;
1030 q->weighting_delay[1] = 7;
1031 q->weighting_delay[2] = 0;
1032 q->weighting_delay[3] = 7;
1033 q->weighting_delay[4] = 0;
1034 q->weighting_delay[5] = 7;
1035
1036 for (i=0; i<4; i++) {
1037 q->matrix_coeff_index_prev[i] = 3;
1038 q->matrix_coeff_index_now[i] = 3;
1039 q->matrix_coeff_index_next[i] = 3;
1040 }
1041
1042 ff_dsputil_init(&dsp, avctx);
1043 ff_fmt_convert_init(&q->fmt_conv, avctx);
1044
1045 q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
1046 if (!q->pUnits) {
1047 atrac3_decode_close(avctx);
1048 return AVERROR(ENOMEM);
1049 }
1050
1051 if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
1052 q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
1053 q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
1054 if (!q->outSamples[0]) {
1055 atrac3_decode_close(avctx);
1056 return AVERROR(ENOMEM);
1057 }
1058 }
1059
1060 avcodec_get_frame_defaults(&q->frame);
1061 avctx->coded_frame = &q->frame;
1062
1063 return 0;
1064 }
1065
1066
1067 AVCodec ff_atrac3_decoder =
1068 {
1069 .name = "atrac3",
1070 .type = AVMEDIA_TYPE_AUDIO,
1071 .id = CODEC_ID_ATRAC3,
1072 .priv_data_size = sizeof(ATRAC3Context),
1073 .init = atrac3_decode_init,
1074 .close = atrac3_decode_close,
1075 .decode = atrac3_decode_frame,
1076 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1077 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1078 };