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 {
91  //@{
92  /** stream data */
94 
96  //@}
97  //@{
98  /** joint-stereo related variables */
103  //@}
104  //@{
105  /** data buffers */
107  float temp_buf[1070];
108  //@}
109  //@{
110  /** extradata */
112  //@}
113 
117 } ATRAC3Context;
118 
120 static VLC_TYPE atrac3_vlc_table[4096][2];
122 static float gain_tab1[16];
123 static float gain_tab2[31];
124 
125 
126 /**
127  * Regular 512 points IMDCT without overlapping, with the exception of the
128  * swapping of odd bands caused by the reverse spectra of the QMF.
129  *
130  * @param odd_band 1 if the band is an odd band
131  */
132 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
133 {
134  int i;
135 
136  if (odd_band) {
137  /**
138  * Reverse the odd bands before IMDCT, this is an effect of the QMF
139  * transform or it gives better compression to do it this way.
140  * FIXME: It should be possible to handle this in imdct_calc
141  * for that to happen a modification of the prerotation step of
142  * all SIMD code and C code is needed.
143  * Or fix the functions before so they generate a pre reversed spectrum.
144  */
145  for (i = 0; i < 128; i++)
146  FFSWAP(float, input[i], input[255 - i]);
147  }
148 
149  q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
150 
151  /* Perform windowing on the output. */
152  q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
153 }
154 
155 /*
156  * indata descrambling, only used for data coming from the rm container
157  */
158 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
159 {
160  int i, off;
161  uint32_t c;
162  const uint32_t *buf;
163  uint32_t *output = (uint32_t *)out;
164 
165  off = (intptr_t)input & 3;
166  buf = (const uint32_t *)(input - off);
167  c = av_be2ne32((0x537F6103 >> (off * 8)) | (0x537F6103 << (32 - (off * 8))));
168  bytes += 3 + off;
169  for (i = 0; i < bytes / 4; i++)
170  output[i] = c ^ buf[i];
171 
172  if (off)
173  av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
174 
175  return off;
176 }
177 
178 static av_cold void init_atrac3_window(void)
179 {
180  int i, j;
181 
182  /* generate the mdct window, for details see
183  * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
184  for (i = 0, j = 255; i < 128; i++, j--) {
185  float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
186  float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
187  float w = 0.5 * (wi * wi + wj * wj);
188  mdct_window[i] = mdct_window[511 - i] = wi / w;
189  mdct_window[j] = mdct_window[511 - j] = wj / w;
190  }
191 }
192 
194 {
195  ATRAC3Context *q = avctx->priv_data;
196 
197  av_free(q->units);
199 
200  ff_mdct_end(&q->mdct_ctx);
201 
202  return 0;
203 }
204 
205 /**
206  * Mantissa decoding
207  *
208  * @param selector which table the output values are coded with
209  * @param coding_flag constant length coding or variable length coding
210  * @param mantissas mantissa output table
211  * @param num_codes number of values to get
212  */
213 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
214  int coding_flag, int *mantissas,
215  int num_codes)
216 {
217  int i, code, huff_symb;
218 
219  if (selector == 1)
220  num_codes /= 2;
221 
222  if (coding_flag != 0) {
223  /* constant length coding (CLC) */
224  int num_bits = clc_length_tab[selector];
225 
226  if (selector > 1) {
227  for (i = 0; i < num_codes; i++) {
228  if (num_bits)
229  code = get_sbits(gb, num_bits);
230  else
231  code = 0;
232  mantissas[i] = code;
233  }
234  } else {
235  for (i = 0; i < num_codes; i++) {
236  if (num_bits)
237  code = get_bits(gb, num_bits); // num_bits is always 4 in this case
238  else
239  code = 0;
240  mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
241  mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
242  }
243  }
244  } else {
245  /* variable length coding (VLC) */
246  if (selector != 1) {
247  for (i = 0; i < num_codes; i++) {
248  huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
249  spectral_coeff_tab[selector-1].bits, 3);
250  huff_symb += 1;
251  code = huff_symb >> 1;
252  if (huff_symb & 1)
253  code = -code;
254  mantissas[i] = code;
255  }
256  } else {
257  for (i = 0; i < num_codes; i++) {
258  huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
259  spectral_coeff_tab[selector - 1].bits, 3);
260  mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
261  mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
262  }
263  }
264  }
265 }
266 
267 /**
268  * Restore the quantized band spectrum coefficients
269  *
270  * @return subband count, fix for broken specification/files
271  */
272 static int decode_spectrum(GetBitContext *gb, float *output)
273 {
274  int num_subbands, coding_mode, i, j, first, last, subband_size;
275  int subband_vlc_index[32], sf_index[32];
276  int mantissas[128];
277  float scale_factor;
278 
279  num_subbands = get_bits(gb, 5); // number of coded subbands
280  coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
281 
282  /* get the VLC selector table for the subbands, 0 means not coded */
283  for (i = 0; i <= num_subbands; i++)
284  subband_vlc_index[i] = get_bits(gb, 3);
285 
286  /* read the scale factor indexes from the stream */
287  for (i = 0; i <= num_subbands; i++) {
288  if (subband_vlc_index[i] != 0)
289  sf_index[i] = get_bits(gb, 6);
290  }
291 
292  for (i = 0; i <= num_subbands; i++) {
293  first = subband_tab[i ];
294  last = subband_tab[i + 1];
295 
296  subband_size = last - first;
297 
298  if (subband_vlc_index[i] != 0) {
299  /* decode spectral coefficients for this subband */
300  /* TODO: This can be done faster is several blocks share the
301  * same VLC selector (subband_vlc_index) */
302  read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
303  mantissas, subband_size);
304 
305  /* decode the scale factor for this subband */
306  scale_factor = ff_atrac_sf_table[sf_index[i]] *
307  inv_max_quant[subband_vlc_index[i]];
308 
309  /* inverse quantize the coefficients */
310  for (j = 0; first < last; first++, j++)
311  output[first] = mantissas[j] * scale_factor;
312  } else {
313  /* this subband was not coded, so zero the entire subband */
314  memset(output + first, 0, subband_size * sizeof(*output));
315  }
316  }
317 
318  /* clear the subbands that were not coded */
319  first = subband_tab[i];
320  memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
321  return num_subbands;
322 }
323 
324 /**
325  * Restore the quantized tonal components
326  *
327  * @param components tonal components
328  * @param num_bands number of coded bands
329  */
331  TonalComponent *components, int num_bands)
332 {
333  int i, b, c, m;
334  int nb_components, coding_mode_selector, coding_mode;
335  int band_flags[4], mantissa[8];
336  int component_count = 0;
337 
338  nb_components = get_bits(gb, 5);
339 
340  /* no tonal components */
341  if (nb_components == 0)
342  return 0;
343 
344  coding_mode_selector = get_bits(gb, 2);
345  if (coding_mode_selector == 2)
346  return AVERROR_INVALIDDATA;
347 
348  coding_mode = coding_mode_selector & 1;
349 
350  for (i = 0; i < nb_components; i++) {
351  int coded_values_per_component, quant_step_index;
352 
353  for (b = 0; b <= num_bands; b++)
354  band_flags[b] = get_bits1(gb);
355 
356  coded_values_per_component = get_bits(gb, 3);
357 
358  quant_step_index = get_bits(gb, 3);
359  if (quant_step_index <= 1)
360  return AVERROR_INVALIDDATA;
361 
362  if (coding_mode_selector == 3)
363  coding_mode = get_bits1(gb);
364 
365  for (b = 0; b < (num_bands + 1) * 4; b++) {
366  int coded_components;
367 
368  if (band_flags[b >> 2] == 0)
369  continue;
370 
371  coded_components = get_bits(gb, 3);
372 
373  for (c = 0; c < coded_components; c++) {
374  TonalComponent *cmp = &components[component_count];
375  int sf_index, coded_values, max_coded_values;
376  float scale_factor;
377 
378  sf_index = get_bits(gb, 6);
379  if (component_count >= 64)
380  return AVERROR_INVALIDDATA;
381 
382  cmp->pos = b * 64 + get_bits(gb, 6);
383 
384  max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
385  coded_values = coded_values_per_component + 1;
386  coded_values = FFMIN(max_coded_values, coded_values);
387 
388  scale_factor = ff_atrac_sf_table[sf_index] *
389  inv_max_quant[quant_step_index];
390 
391  read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
392  mantissa, coded_values);
393 
394  cmp->num_coefs = coded_values;
395 
396  /* inverse quant */
397  for (m = 0; m < coded_values; m++)
398  cmp->coef[m] = mantissa[m] * scale_factor;
399 
400  component_count++;
401  }
402  }
403  }
404 
405  return component_count;
406 }
407 
408 /**
409  * Decode gain parameters for the coded bands
410  *
411  * @param block the gainblock for the current band
412  * @param num_bands amount of coded bands
413  */
415  int num_bands)
416 {
417  int i, cf, num_data;
418  int *level, *loc;
419 
420  GainInfo *gain = block->g_block;
421 
422  for (i = 0; i <= num_bands; i++) {
423  num_data = get_bits(gb, 3);
424  gain[i].num_gain_data = num_data;
425  level = gain[i].lev_code;
426  loc = gain[i].loc_code;
427 
428  for (cf = 0; cf < gain[i].num_gain_data; cf++) {
429  level[cf] = get_bits(gb, 4);
430  loc [cf] = get_bits(gb, 5);
431  if (cf && loc[cf] <= loc[cf - 1])
432  return AVERROR_INVALIDDATA;
433  }
434  }
435 
436  /* Clear the unused blocks. */
437  for (; i < 4 ; i++)
438  gain[i].num_gain_data = 0;
439 
440  return 0;
441 }
442 
443 /**
444  * Apply gain parameters and perform the MDCT overlapping part
445  *
446  * @param input input buffer
447  * @param prev previous buffer to perform overlap against
448  * @param output output buffer
449  * @param gain1 current band gain info
450  * @param gain2 next band gain info
451  */
452 static void gain_compensate_and_overlap(float *input, float *prev,
453  float *output, GainInfo *gain1,
454  GainInfo *gain2)
455 {
456  float g1, g2, gain_inc;
457  int i, j, num_data, start_loc, end_loc;
458 
459 
460  if (gain2->num_gain_data == 0)
461  g1 = 1.0;
462  else
463  g1 = gain_tab1[gain2->lev_code[0]];
464 
465  if (gain1->num_gain_data == 0) {
466  for (i = 0; i < 256; i++)
467  output[i] = input[i] * g1 + prev[i];
468  } else {
469  num_data = gain1->num_gain_data;
470  gain1->loc_code[num_data] = 32;
471  gain1->lev_code[num_data] = 4;
472 
473  for (i = 0, j = 0; i < num_data; i++) {
474  start_loc = gain1->loc_code[i] * 8;
475  end_loc = start_loc + 8;
476 
477  g2 = gain_tab1[gain1->lev_code[i]];
478  gain_inc = gain_tab2[gain1->lev_code[i + 1] -
479  gain1->lev_code[i ] + 15];
480 
481  /* interpolate */
482  for (; j < start_loc; j++)
483  output[j] = (input[j] * g1 + prev[j]) * g2;
484 
485  /* interpolation is done over eight samples */
486  for (; j < end_loc; j++) {
487  output[j] = (input[j] * g1 + prev[j]) * g2;
488  g2 *= gain_inc;
489  }
490  }
491 
492  for (; j < 256; j++)
493  output[j] = input[j] * g1 + prev[j];
494  }
495 
496  /* Delay for the overlapping part. */
497  memcpy(prev, &input[256], 256 * sizeof(*prev));
498 }
499 
500 /**
501  * Combine the tonal band spectrum and regular band spectrum
502  *
503  * @param spectrum output spectrum buffer
504  * @param num_components number of tonal components
505  * @param components tonal components for this band
506  * @return position of the last tonal coefficient
507  */
508 static int add_tonal_components(float *spectrum, int num_components,
509  TonalComponent *components)
510 {
511  int i, j, last_pos = -1;
512  float *input, *output;
513 
514  for (i = 0; i < num_components; i++) {
515  last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
516  input = components[i].coef;
517  output = &spectrum[components[i].pos];
518 
519  for (j = 0; j < components[i].num_coefs; j++)
520  output[j] += input[j];
521  }
522 
523  return last_pos;
524 }
525 
526 #define INTERPOLATE(old, new, nsample) \
527  ((old) + (nsample) * 0.125 * ((new) - (old)))
528 
529 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
530  int *curr_code)
531 {
532  int i, nsample, band;
533  float mc1_l, mc1_r, mc2_l, mc2_r;
534 
535  for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
536  int s1 = prev_code[i];
537  int s2 = curr_code[i];
538  nsample = band;
539 
540  if (s1 != s2) {
541  /* Selector value changed, interpolation needed. */
542  mc1_l = matrix_coeffs[s1 * 2 ];
543  mc1_r = matrix_coeffs[s1 * 2 + 1];
544  mc2_l = matrix_coeffs[s2 * 2 ];
545  mc2_r = matrix_coeffs[s2 * 2 + 1];
546 
547  /* Interpolation is done over the first eight samples. */
548  for (; nsample < band + 8; nsample++) {
549  float c1 = su1[nsample];
550  float c2 = su2[nsample];
551  c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
552  c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
553  su1[nsample] = c2;
554  su2[nsample] = c1 * 2.0 - c2;
555  }
556  }
557 
558  /* Apply the matrix without interpolation. */
559  switch (s2) {
560  case 0: /* M/S decoding */
561  for (; nsample < band + 256; nsample++) {
562  float c1 = su1[nsample];
563  float c2 = su2[nsample];
564  su1[nsample] = c2 * 2.0;
565  su2[nsample] = (c1 - c2) * 2.0;
566  }
567  break;
568  case 1:
569  for (; nsample < band + 256; nsample++) {
570  float c1 = su1[nsample];
571  float c2 = su2[nsample];
572  su1[nsample] = (c1 + c2) * 2.0;
573  su2[nsample] = c2 * -2.0;
574  }
575  break;
576  case 2:
577  case 3:
578  for (; nsample < band + 256; nsample++) {
579  float c1 = su1[nsample];
580  float c2 = su2[nsample];
581  su1[nsample] = c1 + c2;
582  su2[nsample] = c1 - c2;
583  }
584  break;
585  default:
586  av_assert1(0);
587  }
588  }
589 }
590 
591 static void get_channel_weights(int index, int flag, float ch[2])
592 {
593  if (index == 7) {
594  ch[0] = 1.0;
595  ch[1] = 1.0;
596  } else {
597  ch[0] = (index & 7) / 7.0;
598  ch[1] = sqrt(2 - ch[0] * ch[0]);
599  if (flag)
600  FFSWAP(float, ch[0], ch[1]);
601  }
602 }
603 
604 static void channel_weighting(float *su1, float *su2, int *p3)
605 {
606  int band, nsample;
607  /* w[x][y] y=0 is left y=1 is right */
608  float w[2][2];
609 
610  if (p3[1] != 7 || p3[3] != 7) {
611  get_channel_weights(p3[1], p3[0], w[0]);
612  get_channel_weights(p3[3], p3[2], w[1]);
613 
614  for (band = 256; band < 4 * 256; band += 256) {
615  for (nsample = band; nsample < band + 8; nsample++) {
616  su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
617  su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
618  }
619  for(; nsample < band + 256; nsample++) {
620  su1[nsample] *= w[1][0];
621  su2[nsample] *= w[1][1];
622  }
623  }
624  }
625 }
626 
627 /**
628  * Decode a Sound Unit
629  *
630  * @param snd the channel unit to be used
631  * @param output the decoded samples before IQMF in float representation
632  * @param channel_num channel number
633  * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
634  */
636  ChannelUnit *snd, float *output,
637  int channel_num, int coding_mode)
638 {
639  int band, ret, num_subbands, last_tonal, num_bands;
640  GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
641  GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
642 
643  if (coding_mode == JOINT_STEREO && channel_num == 1) {
644  if (get_bits(gb, 2) != 3) {
645  av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
646  return AVERROR_INVALIDDATA;
647  }
648  } else {
649  if (get_bits(gb, 6) != 0x28) {
650  av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
651  return AVERROR_INVALIDDATA;
652  }
653  }
654 
655  /* number of coded QMF bands */
656  snd->bands_coded = get_bits(gb, 2);
657 
658  ret = decode_gain_control(gb, gain2, snd->bands_coded);
659  if (ret)
660  return ret;
661 
663  snd->bands_coded);
664  if (snd->num_components == -1)
665  return -1;
666 
667  num_subbands = decode_spectrum(gb, snd->spectrum);
668 
669  /* Merge the decoded spectrum and tonal components. */
670  last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
671  snd->components);
672 
673 
674  /* calculate number of used MLT/QMF bands according to the amount of coded
675  spectral lines */
676  num_bands = (subband_tab[num_subbands] - 1) >> 8;
677  if (last_tonal >= 0)
678  num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
679 
680 
681  /* Reconstruct time domain samples. */
682  for (band = 0; band < 4; band++) {
683  /* Perform the IMDCT step without overlapping. */
684  if (band <= num_bands)
685  imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
686  else
687  memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
688 
689  /* gain compensation and overlapping */
691  &snd->prev_frame[band * 256],
692  &output[band * 256],
693  &gain1->g_block[band],
694  &gain2->g_block[band]);
695  }
696 
697  /* Swap the gain control buffers for the next frame. */
698  snd->gc_blk_switch ^= 1;
699 
700  return 0;
701 }
702 
703 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
704  float **out_samples)
705 {
706  ATRAC3Context *q = avctx->priv_data;
707  int ret, i;
708  uint8_t *ptr1;
709 
710  if (q->coding_mode == JOINT_STEREO) {
711  /* channel coupling mode */
712  /* decode Sound Unit 1 */
713  init_get_bits(&q->gb, databuf, avctx->block_align * 8);
714 
715  ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
716  JOINT_STEREO);
717  if (ret != 0)
718  return ret;
719 
720  /* Framedata of the su2 in the joint-stereo mode is encoded in
721  * reverse byte order so we need to swap it first. */
722  if (databuf == q->decoded_bytes_buffer) {
723  uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
724  ptr1 = q->decoded_bytes_buffer;
725  for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
726  FFSWAP(uint8_t, *ptr1, *ptr2);
727  } else {
728  const uint8_t *ptr2 = databuf + avctx->block_align - 1;
729  for (i = 0; i < avctx->block_align; i++)
730  q->decoded_bytes_buffer[i] = *ptr2--;
731  }
732 
733  /* Skip the sync codes (0xF8). */
734  ptr1 = q->decoded_bytes_buffer;
735  for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
736  if (i >= avctx->block_align)
737  return AVERROR_INVALIDDATA;
738  }
739 
740 
741  /* set the bitstream reader at the start of the second Sound Unit*/
742  init_get_bits8(&q->gb, ptr1, q->decoded_bytes_buffer + avctx->block_align - ptr1);
743 
744  /* Fill the Weighting coeffs delay buffer */
745  memmove(q->weighting_delay, &q->weighting_delay[2],
746  4 * sizeof(*q->weighting_delay));
747  q->weighting_delay[4] = get_bits1(&q->gb);
748  q->weighting_delay[5] = get_bits(&q->gb, 3);
749 
750  for (i = 0; i < 4; i++) {
753  q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
754  }
755 
756  /* Decode Sound Unit 2. */
757  ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
758  out_samples[1], 1, JOINT_STEREO);
759  if (ret != 0)
760  return ret;
761 
762  /* Reconstruct the channel coefficients. */
763  reverse_matrixing(out_samples[0], out_samples[1],
766 
767  channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
768  } else {
769  /* normal stereo mode or mono */
770  /* Decode the channel sound units. */
771  for (i = 0; i < avctx->channels; i++) {
772  /* Set the bitstream reader at the start of a channel sound unit. */
773  init_get_bits(&q->gb,
774  databuf + i * avctx->block_align / avctx->channels,
775  avctx->block_align * 8 / avctx->channels);
776 
777  ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
778  out_samples[i], i, q->coding_mode);
779  if (ret != 0)
780  return ret;
781  }
782  }
783 
784  /* Apply the iQMF synthesis filter. */
785  for (i = 0; i < avctx->channels; i++) {
786  float *p1 = out_samples[i];
787  float *p2 = p1 + 256;
788  float *p3 = p2 + 256;
789  float *p4 = p3 + 256;
790  ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
791  ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
792  ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
793  }
794 
795  return 0;
796 }
797 
798 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
799  int *got_frame_ptr, AVPacket *avpkt)
800 {
801  AVFrame *frame = data;
802  const uint8_t *buf = avpkt->data;
803  int buf_size = avpkt->size;
804  ATRAC3Context *q = avctx->priv_data;
805  int ret;
806  const uint8_t *databuf;
807 
808  if (buf_size < avctx->block_align) {
809  av_log(avctx, AV_LOG_ERROR,
810  "Frame too small (%d bytes). Truncated file?\n", buf_size);
811  return AVERROR_INVALIDDATA;
812  }
813 
814  /* get output buffer */
815  frame->nb_samples = SAMPLES_PER_FRAME;
816  if ((ret = ff_get_buffer(avctx, frame)) < 0) {
817  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
818  return ret;
819  }
820 
821  /* Check if we need to descramble and what buffer to pass on. */
822  if (q->scrambled_stream) {
824  databuf = q->decoded_bytes_buffer;
825  } else {
826  databuf = buf;
827  }
828 
829  ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
830  if (ret) {
831  av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
832  return ret;
833  }
834 
835  *got_frame_ptr = 1;
836 
837  return avctx->block_align;
838 }
839 
840 static void atrac3_init_static_data(void)
841 {
842  int i;
843 
846 
847  /* Initialize the VLC tables. */
848  for (i = 0; i < 7; i++) {
849  spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
850  spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
851  atrac3_vlc_offs[i ];
852  init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
853  huff_bits[i], 1, 1,
855  }
856 
857  /* Generate gain tables */
858  for (i = 0; i < 16; i++)
859  gain_tab1[i] = exp2f (4 - i);
860 
861  for (i = -15; i < 16; i++)
862  gain_tab2[i + 15] = exp2f (i * -0.125);
863 }
864 
866 {
867  static int static_init_done;
868  int i, ret;
869  int version, delay, samples_per_frame, frame_factor;
870  const uint8_t *edata_ptr = avctx->extradata;
871  ATRAC3Context *q = avctx->priv_data;
872 
873  if (avctx->channels <= 0 || avctx->channels > 2) {
874  av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
875  return AVERROR(EINVAL);
876  }
877 
878  if (!static_init_done)
880  static_init_done = 1;
881 
882  /* Take care of the codec-specific extradata. */
883  if (avctx->extradata_size == 14) {
884  /* Parse the extradata, WAV format */
885  av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
886  bytestream_get_le16(&edata_ptr)); // Unknown value always 1
887  edata_ptr += 4; // samples per channel
888  q->coding_mode = bytestream_get_le16(&edata_ptr);
889  av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
890  bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
891  frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
892  av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
893  bytestream_get_le16(&edata_ptr)); // Unknown always 0
894 
895  /* setup */
896  samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
897  version = 4;
898  delay = 0x88E;
900  q->scrambled_stream = 0;
901 
902  if (avctx->block_align != 96 * avctx->channels * frame_factor &&
903  avctx->block_align != 152 * avctx->channels * frame_factor &&
904  avctx->block_align != 192 * avctx->channels * frame_factor) {
905  av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
906  "configuration %d/%d/%d\n", avctx->block_align,
907  avctx->channels, frame_factor);
908  return AVERROR_INVALIDDATA;
909  }
910  } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
911  /* Parse the extradata, RM format. */
912  version = bytestream_get_be32(&edata_ptr);
913  samples_per_frame = bytestream_get_be16(&edata_ptr);
914  delay = bytestream_get_be16(&edata_ptr);
915  q->coding_mode = bytestream_get_be16(&edata_ptr);
916  q->scrambled_stream = 1;
917 
918  } else {
919  av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
920  avctx->extradata_size);
921  return AVERROR(EINVAL);
922  }
923 
924  if (q->coding_mode == JOINT_STEREO && avctx->channels < 2) {
925  av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n");
926  return AVERROR_INVALIDDATA;
927  }
928 
929  /* Check the extradata */
930 
931  if (version != 4) {
932  av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
933  return AVERROR_INVALIDDATA;
934  }
935 
936  if (samples_per_frame != SAMPLES_PER_FRAME &&
937  samples_per_frame != SAMPLES_PER_FRAME * 2) {
938  av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
939  samples_per_frame);
940  return AVERROR_INVALIDDATA;
941  }
942 
943  if (delay != 0x88E) {
944  av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
945  delay);
946  return AVERROR_INVALIDDATA;
947  }
948 
949  if (q->coding_mode == STEREO)
950  av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
951  else if (q->coding_mode == JOINT_STEREO)
952  av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
953  else {
954  av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
955  q->coding_mode);
956  return AVERROR_INVALIDDATA;
957  }
958 
959  if (avctx->block_align >= UINT_MAX / 2)
960  return AVERROR(EINVAL);
961 
964  if (q->decoded_bytes_buffer == NULL)
965  return AVERROR(ENOMEM);
966 
968 
969  /* initialize the MDCT transform */
970  if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
971  av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
973  return ret;
974  }
975 
976  /* init the joint-stereo decoding data */
977  q->weighting_delay[0] = 0;
978  q->weighting_delay[1] = 7;
979  q->weighting_delay[2] = 0;
980  q->weighting_delay[3] = 7;
981  q->weighting_delay[4] = 0;
982  q->weighting_delay[5] = 7;
983 
984  for (i = 0; i < 4; i++) {
985  q->matrix_coeff_index_prev[i] = 3;
986  q->matrix_coeff_index_now[i] = 3;
987  q->matrix_coeff_index_next[i] = 3;
988  }
989 
991  ff_fmt_convert_init(&q->fmt_conv, avctx);
992 
993  q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
994  if (!q->units) {
995  atrac3_decode_close(avctx);
996  return AVERROR(ENOMEM);
997  }
998 
999  return 0;
1000 }
1001 
1003  .name = "atrac3",
1004  .type = AVMEDIA_TYPE_AUDIO,
1005  .id = AV_CODEC_ID_ATRAC3,
1006  .priv_data_size = sizeof(ATRAC3Context),
1010  .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1011  .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1012  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1014 };