Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * ATRAC3 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.  * ATRAC3 compatible decoder.
  26.  * This decoder handles Sony's ATRAC3 data.
  27.  *
  28.  * Container formats used to store ATRAC3 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/attributes.h"
  40. #include "libavutil/float_dsp.h"
  41. #include "libavutil/libm.h"
  42. #include "avcodec.h"
  43. #include "bytestream.h"
  44. #include "fft.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 GainBlock {
  58.     AtracGainInfo g_block[4];
  59. } GainBlock;
  60.  
  61. typedef struct TonalComponent {
  62.     int pos;
  63.     int num_coefs;
  64.     float coef[8];
  65. } TonalComponent;
  66.  
  67. typedef struct ChannelUnit {
  68.     int            bands_coded;
  69.     int            num_components;
  70.     float          prev_frame[SAMPLES_PER_FRAME];
  71.     int            gc_blk_switch;
  72.     TonalComponent components[64];
  73.     GainBlock      gain_block[2];
  74.  
  75.     DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
  76.     DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
  77.  
  78.     float          delay_buf1[46]; ///<qmf delay buffers
  79.     float          delay_buf2[46];
  80.     float          delay_buf3[46];
  81. } ChannelUnit;
  82.  
  83. typedef struct ATRAC3Context {
  84.     GetBitContext gb;
  85.     //@{
  86.     /** stream data */
  87.     int coding_mode;
  88.  
  89.     ChannelUnit *units;
  90.     //@}
  91.     //@{
  92.     /** joint-stereo related variables */
  93.     int matrix_coeff_index_prev[4];
  94.     int matrix_coeff_index_now[4];
  95.     int matrix_coeff_index_next[4];
  96.     int weighting_delay[6];
  97.     //@}
  98.     //@{
  99.     /** data buffers */
  100.     uint8_t *decoded_bytes_buffer;
  101.     float temp_buf[1070];
  102.     //@}
  103.     //@{
  104.     /** extradata */
  105.     int scrambled_stream;
  106.     //@}
  107.  
  108.     AtracGCContext    gainc_ctx;
  109.     FFTContext        mdct_ctx;
  110.     AVFloatDSPContext *fdsp;
  111. } ATRAC3Context;
  112.  
  113. static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
  114. static VLC_TYPE atrac3_vlc_table[4096][2];
  115. static VLC   spectral_coeff_tab[7];
  116.  
  117. /**
  118.  * Regular 512 points IMDCT without overlapping, with the exception of the
  119.  * swapping of odd bands caused by the reverse spectra of the QMF.
  120.  *
  121.  * @param odd_band  1 if the band is an odd band
  122.  */
  123. static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
  124. {
  125.     int i;
  126.  
  127.     if (odd_band) {
  128.         /**
  129.          * Reverse the odd bands before IMDCT, this is an effect of the QMF
  130.          * transform or it gives better compression to do it this way.
  131.          * FIXME: It should be possible to handle this in imdct_calc
  132.          * for that to happen a modification of the prerotation step of
  133.          * all SIMD code and C code is needed.
  134.          * Or fix the functions before so they generate a pre reversed spectrum.
  135.          */
  136.         for (i = 0; i < 128; i++)
  137.             FFSWAP(float, input[i], input[255 - i]);
  138.     }
  139.  
  140.     q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
  141.  
  142.     /* Perform windowing on the output. */
  143.     q->fdsp->vector_fmul(output, output, mdct_window, MDCT_SIZE);
  144. }
  145.  
  146. /*
  147.  * indata descrambling, only used for data coming from the rm container
  148.  */
  149. static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
  150. {
  151.     int i, off;
  152.     uint32_t c;
  153.     const uint32_t *buf;
  154.     uint32_t *output = (uint32_t *)out;
  155.  
  156.     off = (intptr_t)input & 3;
  157.     buf = (const uint32_t *)(input - off);
  158.     if (off)
  159.         c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
  160.     else
  161.         c = av_be2ne32(0x537F6103U);
  162.     bytes += 3 + off;
  163.     for (i = 0; i < bytes / 4; i++)
  164.         output[i] = c ^ buf[i];
  165.  
  166.     if (off)
  167.         avpriv_request_sample(NULL, "Offset of %d", off);
  168.  
  169.     return off;
  170. }
  171.  
  172. static av_cold void init_imdct_window(void)
  173. {
  174.     int i, j;
  175.  
  176.     /* generate the mdct window, for details see
  177.      * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
  178.     for (i = 0, j = 255; i < 128; i++, j--) {
  179.         float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
  180.         float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
  181.         float w  = 0.5 * (wi * wi + wj * wj);
  182.         mdct_window[i] = mdct_window[511 - i] = wi / w;
  183.         mdct_window[j] = mdct_window[511 - j] = wj / w;
  184.     }
  185. }
  186.  
  187. static av_cold int atrac3_decode_close(AVCodecContext *avctx)
  188. {
  189.     ATRAC3Context *q = avctx->priv_data;
  190.  
  191.     av_freep(&q->units);
  192.     av_freep(&q->decoded_bytes_buffer);
  193.     av_freep(&q->fdsp);
  194.  
  195.     ff_mdct_end(&q->mdct_ctx);
  196.  
  197.     return 0;
  198. }
  199.  
  200. /**
  201.  * Mantissa decoding
  202.  *
  203.  * @param selector     which table the output values are coded with
  204.  * @param coding_flag  constant length coding or variable length coding
  205.  * @param mantissas    mantissa output table
  206.  * @param num_codes    number of values to get
  207.  */
  208. static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
  209.                                        int coding_flag, int *mantissas,
  210.                                        int num_codes)
  211. {
  212.     int i, code, huff_symb;
  213.  
  214.     if (selector == 1)
  215.         num_codes /= 2;
  216.  
  217.     if (coding_flag != 0) {
  218.         /* constant length coding (CLC) */
  219.         int num_bits = clc_length_tab[selector];
  220.  
  221.         if (selector > 1) {
  222.             for (i = 0; i < num_codes; i++) {
  223.                 if (num_bits)
  224.                     code = get_sbits(gb, num_bits);
  225.                 else
  226.                     code = 0;
  227.                 mantissas[i] = code;
  228.             }
  229.         } else {
  230.             for (i = 0; i < num_codes; i++) {
  231.                 if (num_bits)
  232.                     code = get_bits(gb, num_bits); // num_bits is always 4 in this case
  233.                 else
  234.                     code = 0;
  235.                 mantissas[i * 2    ] = mantissa_clc_tab[code >> 2];
  236.                 mantissas[i * 2 + 1] = mantissa_clc_tab[code &  3];
  237.             }
  238.         }
  239.     } else {
  240.         /* variable length coding (VLC) */
  241.         if (selector != 1) {
  242.             for (i = 0; i < num_codes; i++) {
  243.                 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
  244.                                      spectral_coeff_tab[selector-1].bits, 3);
  245.                 huff_symb += 1;
  246.                 code = huff_symb >> 1;
  247.                 if (huff_symb & 1)
  248.                     code = -code;
  249.                 mantissas[i] = code;
  250.             }
  251.         } else {
  252.             for (i = 0; i < num_codes; i++) {
  253.                 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
  254.                                      spectral_coeff_tab[selector - 1].bits, 3);
  255.                 mantissas[i * 2    ] = mantissa_vlc_tab[huff_symb * 2    ];
  256.                 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
  257.             }
  258.         }
  259.     }
  260. }
  261.  
  262. /**
  263.  * Restore the quantized band spectrum coefficients
  264.  *
  265.  * @return subband count, fix for broken specification/files
  266.  */
  267. static int decode_spectrum(GetBitContext *gb, float *output)
  268. {
  269.     int num_subbands, coding_mode, i, j, first, last, subband_size;
  270.     int subband_vlc_index[32], sf_index[32];
  271.     int mantissas[128];
  272.     float scale_factor;
  273.  
  274.     num_subbands = get_bits(gb, 5);  // number of coded subbands
  275.     coding_mode  = get_bits1(gb);    // coding Mode: 0 - VLC/ 1-CLC
  276.  
  277.     /* get the VLC selector table for the subbands, 0 means not coded */
  278.     for (i = 0; i <= num_subbands; i++)
  279.         subband_vlc_index[i] = get_bits(gb, 3);
  280.  
  281.     /* read the scale factor indexes from the stream */
  282.     for (i = 0; i <= num_subbands; i++) {
  283.         if (subband_vlc_index[i] != 0)
  284.             sf_index[i] = get_bits(gb, 6);
  285.     }
  286.  
  287.     for (i = 0; i <= num_subbands; i++) {
  288.         first = subband_tab[i    ];
  289.         last  = subband_tab[i + 1];
  290.  
  291.         subband_size = last - first;
  292.  
  293.         if (subband_vlc_index[i] != 0) {
  294.             /* decode spectral coefficients for this subband */
  295.             /* TODO: This can be done faster is several blocks share the
  296.              * same VLC selector (subband_vlc_index) */
  297.             read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
  298.                                        mantissas, subband_size);
  299.  
  300.             /* decode the scale factor for this subband */
  301.             scale_factor = ff_atrac_sf_table[sf_index[i]] *
  302.                            inv_max_quant[subband_vlc_index[i]];
  303.  
  304.             /* inverse quantize the coefficients */
  305.             for (j = 0; first < last; first++, j++)
  306.                 output[first] = mantissas[j] * scale_factor;
  307.         } else {
  308.             /* this subband was not coded, so zero the entire subband */
  309.             memset(output + first, 0, subband_size * sizeof(*output));
  310.         }
  311.     }
  312.  
  313.     /* clear the subbands that were not coded */
  314.     first = subband_tab[i];
  315.     memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
  316.     return num_subbands;
  317. }
  318.  
  319. /**
  320.  * Restore the quantized tonal components
  321.  *
  322.  * @param components tonal components
  323.  * @param num_bands  number of coded bands
  324.  */
  325. static int decode_tonal_components(GetBitContext *gb,
  326.                                    TonalComponent *components, int num_bands)
  327. {
  328.     int i, b, c, m;
  329.     int nb_components, coding_mode_selector, coding_mode;
  330.     int band_flags[4], mantissa[8];
  331.     int component_count = 0;
  332.  
  333.     nb_components = get_bits(gb, 5);
  334.  
  335.     /* no tonal components */
  336.     if (nb_components == 0)
  337.         return 0;
  338.  
  339.     coding_mode_selector = get_bits(gb, 2);
  340.     if (coding_mode_selector == 2)
  341.         return AVERROR_INVALIDDATA;
  342.  
  343.     coding_mode = coding_mode_selector & 1;
  344.  
  345.     for (i = 0; i < nb_components; i++) {
  346.         int coded_values_per_component, quant_step_index;
  347.  
  348.         for (b = 0; b <= num_bands; b++)
  349.             band_flags[b] = get_bits1(gb);
  350.  
  351.         coded_values_per_component = get_bits(gb, 3);
  352.  
  353.         quant_step_index = get_bits(gb, 3);
  354.         if (quant_step_index <= 1)
  355.             return AVERROR_INVALIDDATA;
  356.  
  357.         if (coding_mode_selector == 3)
  358.             coding_mode = get_bits1(gb);
  359.  
  360.         for (b = 0; b < (num_bands + 1) * 4; b++) {
  361.             int coded_components;
  362.  
  363.             if (band_flags[b >> 2] == 0)
  364.                 continue;
  365.  
  366.             coded_components = get_bits(gb, 3);
  367.  
  368.             for (c = 0; c < coded_components; c++) {
  369.                 TonalComponent *cmp = &components[component_count];
  370.                 int sf_index, coded_values, max_coded_values;
  371.                 float scale_factor;
  372.  
  373.                 sf_index = get_bits(gb, 6);
  374.                 if (component_count >= 64)
  375.                     return AVERROR_INVALIDDATA;
  376.  
  377.                 cmp->pos = b * 64 + get_bits(gb, 6);
  378.  
  379.                 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
  380.                 coded_values     = coded_values_per_component + 1;
  381.                 coded_values     = FFMIN(max_coded_values, coded_values);
  382.  
  383.                 scale_factor = ff_atrac_sf_table[sf_index] *
  384.                                inv_max_quant[quant_step_index];
  385.  
  386.                 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
  387.                                            mantissa, coded_values);
  388.  
  389.                 cmp->num_coefs = coded_values;
  390.  
  391.                 /* inverse quant */
  392.                 for (m = 0; m < coded_values; m++)
  393.                     cmp->coef[m] = mantissa[m] * scale_factor;
  394.  
  395.                 component_count++;
  396.             }
  397.         }
  398.     }
  399.  
  400.     return component_count;
  401. }
  402.  
  403. /**
  404.  * Decode gain parameters for the coded bands
  405.  *
  406.  * @param block      the gainblock for the current band
  407.  * @param num_bands  amount of coded bands
  408.  */
  409. static int decode_gain_control(GetBitContext *gb, GainBlock *block,
  410.                                int num_bands)
  411. {
  412.     int b, j;
  413.     int *level, *loc;
  414.  
  415.     AtracGainInfo *gain = block->g_block;
  416.  
  417.     for (b = 0; b <= num_bands; b++) {
  418.         gain[b].num_points = get_bits(gb, 3);
  419.         level              = gain[b].lev_code;
  420.         loc                = gain[b].loc_code;
  421.  
  422.         for (j = 0; j < gain[b].num_points; j++) {
  423.             level[j] = get_bits(gb, 4);
  424.             loc[j]   = get_bits(gb, 5);
  425.             if (j && loc[j] <= loc[j - 1])
  426.                 return AVERROR_INVALIDDATA;
  427.         }
  428.     }
  429.  
  430.     /* Clear the unused blocks. */
  431.     for (; b < 4 ; b++)
  432.         gain[b].num_points = 0;
  433.  
  434.     return 0;
  435. }
  436.  
  437. /**
  438.  * Combine the tonal band spectrum and regular band spectrum
  439.  *
  440.  * @param spectrum        output spectrum buffer
  441.  * @param num_components  number of tonal components
  442.  * @param components      tonal components for this band
  443.  * @return                position of the last tonal coefficient
  444.  */
  445. static int add_tonal_components(float *spectrum, int num_components,
  446.                                 TonalComponent *components)
  447. {
  448.     int i, j, last_pos = -1;
  449.     float *input, *output;
  450.  
  451.     for (i = 0; i < num_components; i++) {
  452.         last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
  453.         input    = components[i].coef;
  454.         output   = &spectrum[components[i].pos];
  455.  
  456.         for (j = 0; j < components[i].num_coefs; j++)
  457.             output[j] += input[j];
  458.     }
  459.  
  460.     return last_pos;
  461. }
  462.  
  463. #define INTERPOLATE(old, new, nsample) \
  464.     ((old) + (nsample) * 0.125 * ((new) - (old)))
  465.  
  466. static void reverse_matrixing(float *su1, float *su2, int *prev_code,
  467.                               int *curr_code)
  468. {
  469.     int i, nsample, band;
  470.     float mc1_l, mc1_r, mc2_l, mc2_r;
  471.  
  472.     for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
  473.         int s1 = prev_code[i];
  474.         int s2 = curr_code[i];
  475.         nsample = band;
  476.  
  477.         if (s1 != s2) {
  478.             /* Selector value changed, interpolation needed. */
  479.             mc1_l = matrix_coeffs[s1 * 2    ];
  480.             mc1_r = matrix_coeffs[s1 * 2 + 1];
  481.             mc2_l = matrix_coeffs[s2 * 2    ];
  482.             mc2_r = matrix_coeffs[s2 * 2 + 1];
  483.  
  484.             /* Interpolation is done over the first eight samples. */
  485.             for (; nsample < band + 8; nsample++) {
  486.                 float c1 = su1[nsample];
  487.                 float c2 = su2[nsample];
  488.                 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
  489.                      c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
  490.                 su1[nsample] = c2;
  491.                 su2[nsample] = c1 * 2.0 - c2;
  492.             }
  493.         }
  494.  
  495.         /* Apply the matrix without interpolation. */
  496.         switch (s2) {
  497.         case 0:     /* M/S decoding */
  498.             for (; nsample < band + 256; nsample++) {
  499.                 float c1 = su1[nsample];
  500.                 float c2 = su2[nsample];
  501.                 su1[nsample] =  c2       * 2.0;
  502.                 su2[nsample] = (c1 - c2) * 2.0;
  503.             }
  504.             break;
  505.         case 1:
  506.             for (; nsample < band + 256; nsample++) {
  507.                 float c1 = su1[nsample];
  508.                 float c2 = su2[nsample];
  509.                 su1[nsample] = (c1 + c2) *  2.0;
  510.                 su2[nsample] =  c2       * -2.0;
  511.             }
  512.             break;
  513.         case 2:
  514.         case 3:
  515.             for (; nsample < band + 256; nsample++) {
  516.                 float c1 = su1[nsample];
  517.                 float c2 = su2[nsample];
  518.                 su1[nsample] = c1 + c2;
  519.                 su2[nsample] = c1 - c2;
  520.             }
  521.             break;
  522.         default:
  523.             av_assert1(0);
  524.         }
  525.     }
  526. }
  527.  
  528. static void get_channel_weights(int index, int flag, float ch[2])
  529. {
  530.     if (index == 7) {
  531.         ch[0] = 1.0;
  532.         ch[1] = 1.0;
  533.     } else {
  534.         ch[0] = (index & 7) / 7.0;
  535.         ch[1] = sqrt(2 - ch[0] * ch[0]);
  536.         if (flag)
  537.             FFSWAP(float, ch[0], ch[1]);
  538.     }
  539. }
  540.  
  541. static void channel_weighting(float *su1, float *su2, int *p3)
  542. {
  543.     int band, nsample;
  544.     /* w[x][y] y=0 is left y=1 is right */
  545.     float w[2][2];
  546.  
  547.     if (p3[1] != 7 || p3[3] != 7) {
  548.         get_channel_weights(p3[1], p3[0], w[0]);
  549.         get_channel_weights(p3[3], p3[2], w[1]);
  550.  
  551.         for (band = 256; band < 4 * 256; band += 256) {
  552.             for (nsample = band; nsample < band + 8; nsample++) {
  553.                 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
  554.                 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
  555.             }
  556.             for(; nsample < band + 256; nsample++) {
  557.                 su1[nsample] *= w[1][0];
  558.                 su2[nsample] *= w[1][1];
  559.             }
  560.         }
  561.     }
  562. }
  563.  
  564. /**
  565.  * Decode a Sound Unit
  566.  *
  567.  * @param snd           the channel unit to be used
  568.  * @param output        the decoded samples before IQMF in float representation
  569.  * @param channel_num   channel number
  570.  * @param coding_mode   the coding mode (JOINT_STEREO or regular stereo/mono)
  571.  */
  572. static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
  573.                                      ChannelUnit *snd, float *output,
  574.                                      int channel_num, int coding_mode)
  575. {
  576.     int band, ret, num_subbands, last_tonal, num_bands;
  577.     GainBlock *gain1 = &snd->gain_block[    snd->gc_blk_switch];
  578.     GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
  579.  
  580.     if (coding_mode == JOINT_STEREO && channel_num == 1) {
  581.         if (get_bits(gb, 2) != 3) {
  582.             av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
  583.             return AVERROR_INVALIDDATA;
  584.         }
  585.     } else {
  586.         if (get_bits(gb, 6) != 0x28) {
  587.             av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
  588.             return AVERROR_INVALIDDATA;
  589.         }
  590.     }
  591.  
  592.     /* number of coded QMF bands */
  593.     snd->bands_coded = get_bits(gb, 2);
  594.  
  595.     ret = decode_gain_control(gb, gain2, snd->bands_coded);
  596.     if (ret)
  597.         return ret;
  598.  
  599.     snd->num_components = decode_tonal_components(gb, snd->components,
  600.                                                   snd->bands_coded);
  601.     if (snd->num_components < 0)
  602.         return snd->num_components;
  603.  
  604.     num_subbands = decode_spectrum(gb, snd->spectrum);
  605.  
  606.     /* Merge the decoded spectrum and tonal components. */
  607.     last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
  608.                                       snd->components);
  609.  
  610.  
  611.     /* calculate number of used MLT/QMF bands according to the amount of coded
  612.        spectral lines */
  613.     num_bands = (subband_tab[num_subbands] - 1) >> 8;
  614.     if (last_tonal >= 0)
  615.         num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
  616.  
  617.  
  618.     /* Reconstruct time domain samples. */
  619.     for (band = 0; band < 4; band++) {
  620.         /* Perform the IMDCT step without overlapping. */
  621.         if (band <= num_bands)
  622.             imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
  623.         else
  624.             memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
  625.  
  626.         /* gain compensation and overlapping */
  627.         ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
  628.                                    &snd->prev_frame[band * 256],
  629.                                    &gain1->g_block[band], &gain2->g_block[band],
  630.                                    256, &output[band * 256]);
  631.     }
  632.  
  633.     /* Swap the gain control buffers for the next frame. */
  634.     snd->gc_blk_switch ^= 1;
  635.  
  636.     return 0;
  637. }
  638.  
  639. static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
  640.                         float **out_samples)
  641. {
  642.     ATRAC3Context *q = avctx->priv_data;
  643.     int ret, i;
  644.     uint8_t *ptr1;
  645.  
  646.     if (q->coding_mode == JOINT_STEREO) {
  647.         /* channel coupling mode */
  648.         /* decode Sound Unit 1 */
  649.         init_get_bits(&q->gb, databuf, avctx->block_align * 8);
  650.  
  651.         ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
  652.                                         JOINT_STEREO);
  653.         if (ret != 0)
  654.             return ret;
  655.  
  656.         /* Framedata of the su2 in the joint-stereo mode is encoded in
  657.          * reverse byte order so we need to swap it first. */
  658.         if (databuf == q->decoded_bytes_buffer) {
  659.             uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
  660.             ptr1          = q->decoded_bytes_buffer;
  661.             for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
  662.                 FFSWAP(uint8_t, *ptr1, *ptr2);
  663.         } else {
  664.             const uint8_t *ptr2 = databuf + avctx->block_align - 1;
  665.             for (i = 0; i < avctx->block_align; i++)
  666.                 q->decoded_bytes_buffer[i] = *ptr2--;
  667.         }
  668.  
  669.         /* Skip the sync codes (0xF8). */
  670.         ptr1 = q->decoded_bytes_buffer;
  671.         for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
  672.             if (i >= avctx->block_align)
  673.                 return AVERROR_INVALIDDATA;
  674.         }
  675.  
  676.  
  677.         /* set the bitstream reader at the start of the second Sound Unit*/
  678.         init_get_bits8(&q->gb, ptr1, q->decoded_bytes_buffer + avctx->block_align - ptr1);
  679.  
  680.         /* Fill the Weighting coeffs delay buffer */
  681.         memmove(q->weighting_delay, &q->weighting_delay[2],
  682.                 4 * sizeof(*q->weighting_delay));
  683.         q->weighting_delay[4] = get_bits1(&q->gb);
  684.         q->weighting_delay[5] = get_bits(&q->gb, 3);
  685.  
  686.         for (i = 0; i < 4; i++) {
  687.             q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
  688.             q->matrix_coeff_index_now[i]  = q->matrix_coeff_index_next[i];
  689.             q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
  690.         }
  691.  
  692.         /* Decode Sound Unit 2. */
  693.         ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
  694.                                         out_samples[1], 1, JOINT_STEREO);
  695.         if (ret != 0)
  696.             return ret;
  697.  
  698.         /* Reconstruct the channel coefficients. */
  699.         reverse_matrixing(out_samples[0], out_samples[1],
  700.                           q->matrix_coeff_index_prev,
  701.                           q->matrix_coeff_index_now);
  702.  
  703.         channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
  704.     } else {
  705.         /* normal stereo mode or mono */
  706.         /* Decode the channel sound units. */
  707.         for (i = 0; i < avctx->channels; i++) {
  708.             /* Set the bitstream reader at the start of a channel sound unit. */
  709.             init_get_bits(&q->gb,
  710.                           databuf + i * avctx->block_align / avctx->channels,
  711.                           avctx->block_align * 8 / avctx->channels);
  712.  
  713.             ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
  714.                                             out_samples[i], i, q->coding_mode);
  715.             if (ret != 0)
  716.                 return ret;
  717.         }
  718.     }
  719.  
  720.     /* Apply the iQMF synthesis filter. */
  721.     for (i = 0; i < avctx->channels; i++) {
  722.         float *p1 = out_samples[i];
  723.         float *p2 = p1 + 256;
  724.         float *p3 = p2 + 256;
  725.         float *p4 = p3 + 256;
  726.         ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
  727.         ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
  728.         ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
  729.     }
  730.  
  731.     return 0;
  732. }
  733.  
  734. static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
  735.                                int *got_frame_ptr, AVPacket *avpkt)
  736. {
  737.     AVFrame *frame     = data;
  738.     const uint8_t *buf = avpkt->data;
  739.     int buf_size = avpkt->size;
  740.     ATRAC3Context *q = avctx->priv_data;
  741.     int ret;
  742.     const uint8_t *databuf;
  743.  
  744.     if (buf_size < avctx->block_align) {
  745.         av_log(avctx, AV_LOG_ERROR,
  746.                "Frame too small (%d bytes). Truncated file?\n", buf_size);
  747.         return AVERROR_INVALIDDATA;
  748.     }
  749.  
  750.     /* get output buffer */
  751.     frame->nb_samples = SAMPLES_PER_FRAME;
  752.     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
  753.         return ret;
  754.  
  755.     /* Check if we need to descramble and what buffer to pass on. */
  756.     if (q->scrambled_stream) {
  757.         decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
  758.         databuf = q->decoded_bytes_buffer;
  759.     } else {
  760.         databuf = buf;
  761.     }
  762.  
  763.     ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
  764.     if (ret) {
  765.         av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
  766.         return ret;
  767.     }
  768.  
  769.     *got_frame_ptr = 1;
  770.  
  771.     return avctx->block_align;
  772. }
  773.  
  774. static av_cold void atrac3_init_static_data(void)
  775. {
  776.     int i;
  777.  
  778.     init_imdct_window();
  779.     ff_atrac_generate_tables();
  780.  
  781.     /* Initialize the VLC tables. */
  782.     for (i = 0; i < 7; i++) {
  783.         spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
  784.         spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
  785.                                                 atrac3_vlc_offs[i    ];
  786.         init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
  787.                  huff_bits[i],  1, 1,
  788.                  huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
  789.     }
  790. }
  791.  
  792. static av_cold int atrac3_decode_init(AVCodecContext *avctx)
  793. {
  794.     static int static_init_done;
  795.     int i, ret;
  796.     int version, delay, samples_per_frame, frame_factor;
  797.     const uint8_t *edata_ptr = avctx->extradata;
  798.     ATRAC3Context *q = avctx->priv_data;
  799.  
  800.     if (avctx->channels <= 0 || avctx->channels > 2) {
  801.         av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
  802.         return AVERROR(EINVAL);
  803.     }
  804.  
  805.     if (!static_init_done)
  806.         atrac3_init_static_data();
  807.     static_init_done = 1;
  808.  
  809.     /* Take care of the codec-specific extradata. */
  810.     if (avctx->extradata_size == 14) {
  811.         /* Parse the extradata, WAV format */
  812.         av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
  813.                bytestream_get_le16(&edata_ptr));  // Unknown value always 1
  814.         edata_ptr += 4;                             // samples per channel
  815.         q->coding_mode = bytestream_get_le16(&edata_ptr);
  816.         av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
  817.                bytestream_get_le16(&edata_ptr));  //Dupe of coding mode
  818.         frame_factor = bytestream_get_le16(&edata_ptr);  // Unknown always 1
  819.         av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
  820.                bytestream_get_le16(&edata_ptr));  // Unknown always 0
  821.  
  822.         /* setup */
  823.         samples_per_frame    = SAMPLES_PER_FRAME * avctx->channels;
  824.         version              = 4;
  825.         delay                = 0x88E;
  826.         q->coding_mode       = q->coding_mode ? JOINT_STEREO : STEREO;
  827.         q->scrambled_stream  = 0;
  828.  
  829.         if (avctx->block_align !=  96 * avctx->channels * frame_factor &&
  830.             avctx->block_align != 152 * avctx->channels * frame_factor &&
  831.             avctx->block_align != 192 * avctx->channels * frame_factor) {
  832.             av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
  833.                    "configuration %d/%d/%d\n", avctx->block_align,
  834.                    avctx->channels, frame_factor);
  835.             return AVERROR_INVALIDDATA;
  836.         }
  837.     } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
  838.         /* Parse the extradata, RM format. */
  839.         version                = bytestream_get_be32(&edata_ptr);
  840.         samples_per_frame      = bytestream_get_be16(&edata_ptr);
  841.         delay                  = bytestream_get_be16(&edata_ptr);
  842.         q->coding_mode         = bytestream_get_be16(&edata_ptr);
  843.         q->scrambled_stream    = 1;
  844.  
  845.     } else {
  846.         av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
  847.                avctx->extradata_size);
  848.         return AVERROR(EINVAL);
  849.     }
  850.  
  851.     /* Check the extradata */
  852.  
  853.     if (version != 4) {
  854.         av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
  855.         return AVERROR_INVALIDDATA;
  856.     }
  857.  
  858.     if (samples_per_frame != SAMPLES_PER_FRAME &&
  859.         samples_per_frame != SAMPLES_PER_FRAME * 2) {
  860.         av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
  861.                samples_per_frame);
  862.         return AVERROR_INVALIDDATA;
  863.     }
  864.  
  865.     if (delay != 0x88E) {
  866.         av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
  867.                delay);
  868.         return AVERROR_INVALIDDATA;
  869.     }
  870.  
  871.     if (q->coding_mode == STEREO)
  872.         av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
  873.     else if (q->coding_mode == JOINT_STEREO) {
  874.         if (avctx->channels != 2) {
  875.             av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n");
  876.             return AVERROR_INVALIDDATA;
  877.         }
  878.         av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
  879.     } else {
  880.         av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
  881.                q->coding_mode);
  882.         return AVERROR_INVALIDDATA;
  883.     }
  884.  
  885.     if (avctx->block_align >= UINT_MAX / 2)
  886.         return AVERROR(EINVAL);
  887.  
  888.     q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
  889.                                          AV_INPUT_BUFFER_PADDING_SIZE);
  890.     if (!q->decoded_bytes_buffer)
  891.         return AVERROR(ENOMEM);
  892.  
  893.     avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
  894.  
  895.     /* initialize the MDCT transform */
  896.     if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
  897.         av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
  898.         av_freep(&q->decoded_bytes_buffer);
  899.         return ret;
  900.     }
  901.  
  902.     /* init the joint-stereo decoding data */
  903.     q->weighting_delay[0] = 0;
  904.     q->weighting_delay[1] = 7;
  905.     q->weighting_delay[2] = 0;
  906.     q->weighting_delay[3] = 7;
  907.     q->weighting_delay[4] = 0;
  908.     q->weighting_delay[5] = 7;
  909.  
  910.     for (i = 0; i < 4; i++) {
  911.         q->matrix_coeff_index_prev[i] = 3;
  912.         q->matrix_coeff_index_now[i]  = 3;
  913.         q->matrix_coeff_index_next[i] = 3;
  914.     }
  915.  
  916.     ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
  917.     q->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
  918.  
  919.     q->units = av_mallocz_array(avctx->channels, sizeof(*q->units));
  920.     if (!q->units || !q->fdsp) {
  921.         atrac3_decode_close(avctx);
  922.         return AVERROR(ENOMEM);
  923.     }
  924.  
  925.     return 0;
  926. }
  927.  
  928. AVCodec ff_atrac3_decoder = {
  929.     .name             = "atrac3",
  930.     .long_name        = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
  931.     .type             = AVMEDIA_TYPE_AUDIO,
  932.     .id               = AV_CODEC_ID_ATRAC3,
  933.     .priv_data_size   = sizeof(ATRAC3Context),
  934.     .init             = atrac3_decode_init,
  935.     .close            = atrac3_decode_close,
  936.     .decode           = atrac3_decode_frame,
  937.     .capabilities     = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
  938.     .sample_fmts      = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
  939.                                                         AV_SAMPLE_FMT_NONE },
  940. };
  941.