FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
atrac3.c
Go to the documentation of this file.
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 FFmpeg.
7  *
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.
12  *
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.
17  *
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
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 "libavutil/float_dsp.h"
40 #include "libavutil/libm.h"
41 #include "avcodec.h"
42 #include "bytestream.h"
43 #include "fft.h"
44 #include "fmtconvert.h"
45 #include "get_bits.h"
46 #include "internal.h"
47 
48 #include "atrac.h"
49 #include "atrac3data.h"
50 
51 #define JOINT_STEREO 0x12
52 #define STEREO 0x2
53 
54 #define SAMPLES_PER_FRAME 1024
55 #define MDCT_SIZE 512
56 
57 typedef struct GainInfo {
59  int lev_code[8];
60  int loc_code[8];
61 } GainInfo;
62 
63 typedef struct GainBlock {
65 } GainBlock;
66 
67 typedef struct TonalComponent {
68  int pos;
69  int num_coefs;
70  float coef[8];
72 
73 typedef struct ChannelUnit {
80 
83 
84  float delay_buf1[46]; ///<qmf delay buffers
85  float delay_buf2[46];
86  float delay_buf3[46];
87 } ChannelUnit;
88 
89 typedef struct ATRAC3Context {
92  //@{
93  /** stream data */
95 
97  //@}
98  //@{
99  /** joint-stereo related variables */
104  //@}
105  //@{
106  /** data buffers */
108  float temp_buf[1070];
109  //@}
110  //@{
111  /** extradata */
113  //@}
114 
118 } ATRAC3Context;
119 
121 static VLC_TYPE atrac3_vlc_table[4096][2];
123 static float gain_tab1[16];
124 static float gain_tab2[31];
125 
126 
127 /**
128  * Regular 512 points IMDCT without overlapping, with the exception of the
129  * swapping of odd bands caused by the reverse spectra of the QMF.
130  *
131  * @param odd_band 1 if the band is an odd band
132  */
133 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
134 {
135  int i;
136 
137  if (odd_band) {
138  /**
139  * Reverse the odd bands before IMDCT, this is an effect of the QMF
140  * transform or it gives better compression to do it this way.
141  * FIXME: It should be possible to handle this in imdct_calc
142  * for that to happen a modification of the prerotation step of
143  * all SIMD code and C code is needed.
144  * Or fix the functions before so they generate a pre reversed spectrum.
145  */
146  for (i = 0; i < 128; i++)
147  FFSWAP(float, input[i], input[255 - i]);
148  }
149 
150  q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
151 
152  /* Perform windowing on the output. */
153  q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
154 }
155 
156 /*
157  * indata descrambling, only used for data coming from the rm container
158  */
159 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
160 {
161  int i, off;
162  uint32_t c;
163  const uint32_t *buf;
164  uint32_t *output = (uint32_t *)out;
165 
166  off = (intptr_t)input & 3;
167  buf = (const uint32_t *)(input - off);
168  if (off)
169  c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
170  else
171  c = av_be2ne32(0x537F6103U);
172  bytes += 3 + off;
173  for (i = 0; i < bytes / 4; i++)
174  output[i] = c ^ buf[i];
175 
176  if (off)
177  av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
178 
179  return off;
180 }
181 
182 static av_cold void init_atrac3_window(void)
183 {
184  int i, j;
185 
186  /* generate the mdct window, for details see
187  * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
188  for (i = 0, j = 255; i < 128; i++, j--) {
189  float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
190  float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
191  float w = 0.5 * (wi * wi + wj * wj);
192  mdct_window[i] = mdct_window[511 - i] = wi / w;
193  mdct_window[j] = mdct_window[511 - j] = wj / w;
194  }
195 }
196 
198 {
199  ATRAC3Context *q = avctx->priv_data;
200 
201  av_free(q->units);
203 
204  ff_mdct_end(&q->mdct_ctx);
205 
206  return 0;
207 }
208 
209 /**
210  * Mantissa decoding
211  *
212  * @param selector which table the output values are coded with
213  * @param coding_flag constant length coding or variable length coding
214  * @param mantissas mantissa output table
215  * @param num_codes number of values to get
216  */
217 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
218  int coding_flag, int *mantissas,
219  int num_codes)
220 {
221  int i, code, huff_symb;
222 
223  if (selector == 1)
224  num_codes /= 2;
225 
226  if (coding_flag != 0) {
227  /* constant length coding (CLC) */
228  int num_bits = clc_length_tab[selector];
229 
230  if (selector > 1) {
231  for (i = 0; i < num_codes; i++) {
232  if (num_bits)
233  code = get_sbits(gb, num_bits);
234  else
235  code = 0;
236  mantissas[i] = code;
237  }
238  } else {
239  for (i = 0; i < num_codes; i++) {
240  if (num_bits)
241  code = get_bits(gb, num_bits); // num_bits is always 4 in this case
242  else
243  code = 0;
244  mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
245  mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
246  }
247  }
248  } else {
249  /* variable length coding (VLC) */
250  if (selector != 1) {
251  for (i = 0; i < num_codes; i++) {
252  huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
253  spectral_coeff_tab[selector-1].bits, 3);
254  huff_symb += 1;
255  code = huff_symb >> 1;
256  if (huff_symb & 1)
257  code = -code;
258  mantissas[i] = code;
259  }
260  } else {
261  for (i = 0; i < num_codes; i++) {
262  huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
263  spectral_coeff_tab[selector - 1].bits, 3);
264  mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
265  mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
266  }
267  }
268  }
269 }
270 
271 /**
272  * Restore the quantized band spectrum coefficients
273  *
274  * @return subband count, fix for broken specification/files
275  */
276 static int decode_spectrum(GetBitContext *gb, float *output)
277 {
278  int num_subbands, coding_mode, i, j, first, last, subband_size;
279  int subband_vlc_index[32], sf_index[32];
280  int mantissas[128];
281  float scale_factor;
282 
283  num_subbands = get_bits(gb, 5); // number of coded subbands
284  coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
285 
286  /* get the VLC selector table for the subbands, 0 means not coded */
287  for (i = 0; i <= num_subbands; i++)
288  subband_vlc_index[i] = get_bits(gb, 3);
289 
290  /* read the scale factor indexes from the stream */
291  for (i = 0; i <= num_subbands; i++) {
292  if (subband_vlc_index[i] != 0)
293  sf_index[i] = get_bits(gb, 6);
294  }
295 
296  for (i = 0; i <= num_subbands; i++) {
297  first = subband_tab[i ];
298  last = subband_tab[i + 1];
299 
300  subband_size = last - first;
301 
302  if (subband_vlc_index[i] != 0) {
303  /* decode spectral coefficients for this subband */
304  /* TODO: This can be done faster is several blocks share the
305  * same VLC selector (subband_vlc_index) */
306  read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
307  mantissas, subband_size);
308 
309  /* decode the scale factor for this subband */
310  scale_factor = ff_atrac_sf_table[sf_index[i]] *
311  inv_max_quant[subband_vlc_index[i]];
312 
313  /* inverse quantize the coefficients */
314  for (j = 0; first < last; first++, j++)
315  output[first] = mantissas[j] * scale_factor;
316  } else {
317  /* this subband was not coded, so zero the entire subband */
318  memset(output + first, 0, subband_size * sizeof(*output));
319  }
320  }
321 
322  /* clear the subbands that were not coded */
323  first = subband_tab[i];
324  memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
325  return num_subbands;
326 }
327 
328 /**
329  * Restore the quantized tonal components
330  *
331  * @param components tonal components
332  * @param num_bands number of coded bands
333  */
335  TonalComponent *components, int num_bands)
336 {
337  int i, b, c, m;
338  int nb_components, coding_mode_selector, coding_mode;
339  int band_flags[4], mantissa[8];
340  int component_count = 0;
341 
342  nb_components = get_bits(gb, 5);
343 
344  /* no tonal components */
345  if (nb_components == 0)
346  return 0;
347 
348  coding_mode_selector = get_bits(gb, 2);
349  if (coding_mode_selector == 2)
350  return AVERROR_INVALIDDATA;
351 
352  coding_mode = coding_mode_selector & 1;
353 
354  for (i = 0; i < nb_components; i++) {
355  int coded_values_per_component, quant_step_index;
356 
357  for (b = 0; b <= num_bands; b++)
358  band_flags[b] = get_bits1(gb);
359 
360  coded_values_per_component = get_bits(gb, 3);
361 
362  quant_step_index = get_bits(gb, 3);
363  if (quant_step_index <= 1)
364  return AVERROR_INVALIDDATA;
365 
366  if (coding_mode_selector == 3)
367  coding_mode = get_bits1(gb);
368 
369  for (b = 0; b < (num_bands + 1) * 4; b++) {
370  int coded_components;
371 
372  if (band_flags[b >> 2] == 0)
373  continue;
374 
375  coded_components = get_bits(gb, 3);
376 
377  for (c = 0; c < coded_components; c++) {
378  TonalComponent *cmp = &components[component_count];
379  int sf_index, coded_values, max_coded_values;
380  float scale_factor;
381 
382  sf_index = get_bits(gb, 6);
383  if (component_count >= 64)
384  return AVERROR_INVALIDDATA;
385 
386  cmp->pos = b * 64 + get_bits(gb, 6);
387 
388  max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
389  coded_values = coded_values_per_component + 1;
390  coded_values = FFMIN(max_coded_values, coded_values);
391 
392  scale_factor = ff_atrac_sf_table[sf_index] *
393  inv_max_quant[quant_step_index];
394 
395  read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
396  mantissa, coded_values);
397 
398  cmp->num_coefs = coded_values;
399 
400  /* inverse quant */
401  for (m = 0; m < coded_values; m++)
402  cmp->coef[m] = mantissa[m] * scale_factor;
403 
404  component_count++;
405  }
406  }
407  }
408 
409  return component_count;
410 }
411 
412 /**
413  * Decode gain parameters for the coded bands
414  *
415  * @param block the gainblock for the current band
416  * @param num_bands amount of coded bands
417  */
419  int num_bands)
420 {
421  int i, cf, num_data;
422  int *level, *loc;
423 
424  GainInfo *gain = block->g_block;
425 
426  for (i = 0; i <= num_bands; i++) {
427  num_data = get_bits(gb, 3);
428  gain[i].num_gain_data = num_data;
429  level = gain[i].lev_code;
430  loc = gain[i].loc_code;
431 
432  for (cf = 0; cf < gain[i].num_gain_data; cf++) {
433  level[cf] = get_bits(gb, 4);
434  loc [cf] = get_bits(gb, 5);
435  if (cf && loc[cf] <= loc[cf - 1])
436  return AVERROR_INVALIDDATA;
437  }
438  }
439 
440  /* Clear the unused blocks. */
441  for (; i < 4 ; i++)
442  gain[i].num_gain_data = 0;
443 
444  return 0;
445 }
446 
447 /**
448  * Apply gain parameters and perform the MDCT overlapping part
449  *
450  * @param input input buffer
451  * @param prev previous buffer to perform overlap against
452  * @param output output buffer
453  * @param gain1 current band gain info
454  * @param gain2 next band gain info
455  */
456 static void gain_compensate_and_overlap(float *input, float *prev,
457  float *output, GainInfo *gain1,
458  GainInfo *gain2)
459 {
460  float g1, g2, gain_inc;
461  int i, j, num_data, start_loc, end_loc;
462 
463 
464  if (gain2->num_gain_data == 0)
465  g1 = 1.0;
466  else
467  g1 = gain_tab1[gain2->lev_code[0]];
468 
469  if (gain1->num_gain_data == 0) {
470  for (i = 0; i < 256; i++)
471  output[i] = input[i] * g1 + prev[i];
472  } else {
473  num_data = gain1->num_gain_data;
474  gain1->loc_code[num_data] = 32;
475  gain1->lev_code[num_data] = 4;
476 
477  for (i = 0, j = 0; i < num_data; i++) {
478  start_loc = gain1->loc_code[i] * 8;
479  end_loc = start_loc + 8;
480 
481  g2 = gain_tab1[gain1->lev_code[i]];
482  gain_inc = gain_tab2[gain1->lev_code[i + 1] -
483  gain1->lev_code[i ] + 15];
484 
485  /* interpolate */
486  for (; j < start_loc; j++)
487  output[j] = (input[j] * g1 + prev[j]) * g2;
488 
489  /* interpolation is done over eight samples */
490  for (; j < end_loc; j++) {
491  output[j] = (input[j] * g1 + prev[j]) * g2;
492  g2 *= gain_inc;
493  }
494  }
495 
496  for (; j < 256; j++)
497  output[j] = input[j] * g1 + prev[j];
498  }
499 
500  /* Delay for the overlapping part. */
501  memcpy(prev, &input[256], 256 * sizeof(*prev));
502 }
503 
504 /**
505  * Combine the tonal band spectrum and regular band spectrum
506  *
507  * @param spectrum output spectrum buffer
508  * @param num_components number of tonal components
509  * @param components tonal components for this band
510  * @return position of the last tonal coefficient
511  */
512 static int add_tonal_components(float *spectrum, int num_components,
513  TonalComponent *components)
514 {
515  int i, j, last_pos = -1;
516  float *input, *output;
517 
518  for (i = 0; i < num_components; i++) {
519  last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
520  input = components[i].coef;
521  output = &spectrum[components[i].pos];
522 
523  for (j = 0; j < components[i].num_coefs; j++)
524  output[j] += input[j];
525  }
526 
527  return last_pos;
528 }
529 
530 #define INTERPOLATE(old, new, nsample) \
531  ((old) + (nsample) * 0.125 * ((new) - (old)))
532 
533 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
534  int *curr_code)
535 {
536  int i, nsample, band;
537  float mc1_l, mc1_r, mc2_l, mc2_r;
538 
539  for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
540  int s1 = prev_code[i];
541  int s2 = curr_code[i];
542  nsample = band;
543 
544  if (s1 != s2) {
545  /* Selector value changed, interpolation needed. */
546  mc1_l = matrix_coeffs[s1 * 2 ];
547  mc1_r = matrix_coeffs[s1 * 2 + 1];
548  mc2_l = matrix_coeffs[s2 * 2 ];
549  mc2_r = matrix_coeffs[s2 * 2 + 1];
550 
551  /* Interpolation is done over the first eight samples. */
552  for (; nsample < band + 8; nsample++) {
553  float c1 = su1[nsample];
554  float c2 = su2[nsample];
555  c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
556  c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
557  su1[nsample] = c2;
558  su2[nsample] = c1 * 2.0 - c2;
559  }
560  }
561 
562  /* Apply the matrix without interpolation. */
563  switch (s2) {
564  case 0: /* M/S decoding */
565  for (; nsample < band + 256; nsample++) {
566  float c1 = su1[nsample];
567  float c2 = su2[nsample];
568  su1[nsample] = c2 * 2.0;
569  su2[nsample] = (c1 - c2) * 2.0;
570  }
571  break;
572  case 1:
573  for (; nsample < band + 256; nsample++) {
574  float c1 = su1[nsample];
575  float c2 = su2[nsample];
576  su1[nsample] = (c1 + c2) * 2.0;
577  su2[nsample] = c2 * -2.0;
578  }
579  break;
580  case 2:
581  case 3:
582  for (; nsample < band + 256; nsample++) {
583  float c1 = su1[nsample];
584  float c2 = su2[nsample];
585  su1[nsample] = c1 + c2;
586  su2[nsample] = c1 - c2;
587  }
588  break;
589  default:
590  av_assert1(0);
591  }
592  }
593 }
594 
595 static void get_channel_weights(int index, int flag, float ch[2])
596 {
597  if (index == 7) {
598  ch[0] = 1.0;
599  ch[1] = 1.0;
600  } else {
601  ch[0] = (index & 7) / 7.0;
602  ch[1] = sqrt(2 - ch[0] * ch[0]);
603  if (flag)
604  FFSWAP(float, ch[0], ch[1]);
605  }
606 }
607 
608 static void channel_weighting(float *su1, float *su2, int *p3)
609 {
610  int band, nsample;
611  /* w[x][y] y=0 is left y=1 is right */
612  float w[2][2];
613 
614  if (p3[1] != 7 || p3[3] != 7) {
615  get_channel_weights(p3[1], p3[0], w[0]);
616  get_channel_weights(p3[3], p3[2], w[1]);
617 
618  for (band = 256; band < 4 * 256; band += 256) {
619  for (nsample = band; nsample < band + 8; nsample++) {
620  su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
621  su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
622  }
623  for(; nsample < band + 256; nsample++) {
624  su1[nsample] *= w[1][0];
625  su2[nsample] *= w[1][1];
626  }
627  }
628  }
629 }
630 
631 /**
632  * Decode a Sound Unit
633  *
634  * @param snd the channel unit to be used
635  * @param output the decoded samples before IQMF in float representation
636  * @param channel_num channel number
637  * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
638  */
640  ChannelUnit *snd, float *output,
641  int channel_num, int coding_mode)
642 {
643  int band, ret, num_subbands, last_tonal, num_bands;
644  GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
645  GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
646 
647  if (coding_mode == JOINT_STEREO && channel_num == 1) {
648  if (get_bits(gb, 2) != 3) {
649  av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
650  return AVERROR_INVALIDDATA;
651  }
652  } else {
653  if (get_bits(gb, 6) != 0x28) {
654  av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
655  return AVERROR_INVALIDDATA;
656  }
657  }
658 
659  /* number of coded QMF bands */
660  snd->bands_coded = get_bits(gb, 2);
661 
662  ret = decode_gain_control(gb, gain2, snd->bands_coded);
663  if (ret)
664  return ret;
665 
667  snd->bands_coded);
668  if (snd->num_components == -1)
669  return -1;
670 
671  num_subbands = decode_spectrum(gb, snd->spectrum);
672 
673  /* Merge the decoded spectrum and tonal components. */
674  last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
675  snd->components);
676 
677 
678  /* calculate number of used MLT/QMF bands according to the amount of coded
679  spectral lines */
680  num_bands = (subband_tab[num_subbands] - 1) >> 8;
681  if (last_tonal >= 0)
682  num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
683 
684 
685  /* Reconstruct time domain samples. */
686  for (band = 0; band < 4; band++) {
687  /* Perform the IMDCT step without overlapping. */
688  if (band <= num_bands)
689  imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
690  else
691  memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
692 
693  /* gain compensation and overlapping */
695  &snd->prev_frame[band * 256],
696  &output[band * 256],
697  &gain1->g_block[band],
698  &gain2->g_block[band]);
699  }
700 
701  /* Swap the gain control buffers for the next frame. */
702  snd->gc_blk_switch ^= 1;
703 
704  return 0;
705 }
706 
707 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
708  float **out_samples)
709 {
710  ATRAC3Context *q = avctx->priv_data;
711  int ret, i;
712  uint8_t *ptr1;
713 
714  if (q->coding_mode == JOINT_STEREO) {
715  /* channel coupling mode */
716  /* decode Sound Unit 1 */
717  init_get_bits(&q->gb, databuf, avctx->block_align * 8);
718 
719  ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
720  JOINT_STEREO);
721  if (ret != 0)
722  return ret;
723 
724  /* Framedata of the su2 in the joint-stereo mode is encoded in
725  * reverse byte order so we need to swap it first. */
726  if (databuf == q->decoded_bytes_buffer) {
727  uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
728  ptr1 = q->decoded_bytes_buffer;
729  for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
730  FFSWAP(uint8_t, *ptr1, *ptr2);
731  } else {
732  const uint8_t *ptr2 = databuf + avctx->block_align - 1;
733  for (i = 0; i < avctx->block_align; i++)
734  q->decoded_bytes_buffer[i] = *ptr2--;
735  }
736 
737  /* Skip the sync codes (0xF8). */
738  ptr1 = q->decoded_bytes_buffer;
739  for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
740  if (i >= avctx->block_align)
741  return AVERROR_INVALIDDATA;
742  }
743 
744 
745  /* set the bitstream reader at the start of the second Sound Unit*/
746  init_get_bits(&q->gb, ptr1, avctx->block_align * 8);
747 
748  /* Fill the Weighting coeffs delay buffer */
749  memmove(q->weighting_delay, &q->weighting_delay[2],
750  4 * sizeof(*q->weighting_delay));
751  q->weighting_delay[4] = get_bits1(&q->gb);
752  q->weighting_delay[5] = get_bits(&q->gb, 3);
753 
754  for (i = 0; i < 4; i++) {
757  q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
758  }
759 
760  /* Decode Sound Unit 2. */
761  ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
762  out_samples[1], 1, JOINT_STEREO);
763  if (ret != 0)
764  return ret;
765 
766  /* Reconstruct the channel coefficients. */
767  reverse_matrixing(out_samples[0], out_samples[1],
770 
771  channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
772  } else {
773  /* normal stereo mode or mono */
774  /* Decode the channel sound units. */
775  for (i = 0; i < avctx->channels; i++) {
776  /* Set the bitstream reader at the start of a channel sound unit. */
777  init_get_bits(&q->gb,
778  databuf + i * avctx->block_align / avctx->channels,
779  avctx->block_align * 8 / avctx->channels);
780 
781  ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
782  out_samples[i], i, q->coding_mode);
783  if (ret != 0)
784  return ret;
785  }
786  }
787 
788  /* Apply the iQMF synthesis filter. */
789  for (i = 0; i < avctx->channels; i++) {
790  float *p1 = out_samples[i];
791  float *p2 = p1 + 256;
792  float *p3 = p2 + 256;
793  float *p4 = p3 + 256;
794  ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
795  ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
796  ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
797  }
798 
799  return 0;
800 }
801 
802 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
803  int *got_frame_ptr, AVPacket *avpkt)
804 {
805  const uint8_t *buf = avpkt->data;
806  int buf_size = avpkt->size;
807  ATRAC3Context *q = avctx->priv_data;
808  int ret;
809  const uint8_t *databuf;
810 
811  if (buf_size < avctx->block_align) {
812  av_log(avctx, AV_LOG_ERROR,
813  "Frame too small (%d bytes). Truncated file?\n", buf_size);
814  return AVERROR_INVALIDDATA;
815  }
816 
817  /* get output buffer */
819  if ((ret = ff_get_buffer(avctx, &q->frame)) < 0) {
820  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
821  return ret;
822  }
823 
824  /* Check if we need to descramble and what buffer to pass on. */
825  if (q->scrambled_stream) {
827  databuf = q->decoded_bytes_buffer;
828  } else {
829  databuf = buf;
830  }
831 
832  ret = decode_frame(avctx, databuf, (float **)q->frame.extended_data);
833  if (ret) {
834  av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
835  return ret;
836  }
837 
838  *got_frame_ptr = 1;
839  *(AVFrame *)data = q->frame;
840 
841  return avctx->block_align;
842 }
843 
844 static void atrac3_init_static_data(void)
845 {
846  int i;
847 
850 
851  /* Initialize the VLC tables. */
852  for (i = 0; i < 7; i++) {
853  spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
854  spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
855  atrac3_vlc_offs[i ];
856  init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
857  huff_bits[i], 1, 1,
859  }
860 
861  /* Generate gain tables */
862  for (i = 0; i < 16; i++)
863  gain_tab1[i] = exp2f (4 - i);
864 
865  for (i = -15; i < 16; i++)
866  gain_tab2[i + 15] = exp2f (i * -0.125);
867 }
868 
870 {
871  static int static_init_done;
872  int i, ret;
873  int version, delay, samples_per_frame, frame_factor;
874  const uint8_t *edata_ptr = avctx->extradata;
875  ATRAC3Context *q = avctx->priv_data;
876 
877  if (avctx->channels <= 0 || avctx->channels > 2) {
878  av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
879  return AVERROR(EINVAL);
880  }
881 
882  if (!static_init_done)
884  static_init_done = 1;
885 
886  /* Take care of the codec-specific extradata. */
887  if (avctx->extradata_size == 14) {
888  /* Parse the extradata, WAV format */
889  av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
890  bytestream_get_le16(&edata_ptr)); // Unknown value always 1
891  edata_ptr += 4; // samples per channel
892  q->coding_mode = bytestream_get_le16(&edata_ptr);
893  av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
894  bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
895  frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
896  av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
897  bytestream_get_le16(&edata_ptr)); // Unknown always 0
898 
899  /* setup */
900  samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
901  version = 4;
902  delay = 0x88E;
904  q->scrambled_stream = 0;
905 
906  if (avctx->block_align != 96 * avctx->channels * frame_factor &&
907  avctx->block_align != 152 * avctx->channels * frame_factor &&
908  avctx->block_align != 192 * avctx->channels * frame_factor) {
909  av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
910  "configuration %d/%d/%d\n", avctx->block_align,
911  avctx->channels, frame_factor);
912  return AVERROR_INVALIDDATA;
913  }
914  } else if (avctx->extradata_size == 10) {
915  /* Parse the extradata, RM format. */
916  version = bytestream_get_be32(&edata_ptr);
917  samples_per_frame = bytestream_get_be16(&edata_ptr);
918  delay = bytestream_get_be16(&edata_ptr);
919  q->coding_mode = bytestream_get_be16(&edata_ptr);
920  q->scrambled_stream = 1;
921 
922  } else {
923  av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
924  avctx->extradata_size);
925  return AVERROR(EINVAL);
926  }
927 
928  if (q->coding_mode == JOINT_STEREO && avctx->channels < 2) {
929  av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n");
930  return AVERROR_INVALIDDATA;
931  }
932 
933  /* Check the extradata */
934 
935  if (version != 4) {
936  av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
937  return AVERROR_INVALIDDATA;
938  }
939 
940  if (samples_per_frame != SAMPLES_PER_FRAME &&
941  samples_per_frame != SAMPLES_PER_FRAME * 2) {
942  av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
943  samples_per_frame);
944  return AVERROR_INVALIDDATA;
945  }
946 
947  if (delay != 0x88E) {
948  av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
949  delay);
950  return AVERROR_INVALIDDATA;
951  }
952 
953  if (q->coding_mode == STEREO)
954  av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
955  else if (q->coding_mode == JOINT_STEREO)
956  av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
957  else {
958  av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
959  q->coding_mode);
960  return AVERROR_INVALIDDATA;
961  }
962 
963  if (avctx->block_align >= UINT_MAX / 2)
964  return AVERROR(EINVAL);
965 
968  if (q->decoded_bytes_buffer == NULL)
969  return AVERROR(ENOMEM);
970 
972 
973  /* initialize the MDCT transform */
974  if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
975  av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
977  return ret;
978  }
979 
980  /* init the joint-stereo decoding data */
981  q->weighting_delay[0] = 0;
982  q->weighting_delay[1] = 7;
983  q->weighting_delay[2] = 0;
984  q->weighting_delay[3] = 7;
985  q->weighting_delay[4] = 0;
986  q->weighting_delay[5] = 7;
987 
988  for (i = 0; i < 4; i++) {
989  q->matrix_coeff_index_prev[i] = 3;
990  q->matrix_coeff_index_now[i] = 3;
991  q->matrix_coeff_index_next[i] = 3;
992  }
993 
995  ff_fmt_convert_init(&q->fmt_conv, avctx);
996 
997  q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
998  if (!q->units) {
999  atrac3_decode_close(avctx);
1000  return AVERROR(ENOMEM);
1001  }
1002 
1004  avctx->coded_frame = &q->frame;
1005 
1006  return 0;
1007 }
1008 
1010  .name = "atrac3",
1011  .type = AVMEDIA_TYPE_AUDIO,
1012  .id = AV_CODEC_ID_ATRAC3,
1013  .priv_data_size = sizeof(ATRAC3Context),
1017  .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1018  .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1019  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1021 };