~ubuntu-branches/ubuntu/intrepid/gstreamer0.10-ffmpeg/intrepid

/*

 * MPEG Audio decoder

 * Copyright (c) 2001, 2002 Fabrice Bellard.

 *

 * This library is free software; you can redistribute it and/or

 * modify it under the terms of the GNU Lesser General Public

 * License as published by the Free Software Foundation; either

 * version 2 of the License, or (at your option) any later version.

 *

 * This library is distributed in the hope that it will be useful,

 * but WITHOUT ANY WARRANTY; without even the implied warranty of

 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

 * Lesser General Public License for more details.

 *

 * You should have received a copy of the GNU Lesser General Public

 * License along with this library; if not, write to the Free Software

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

 */

/**

 * @file mpegaudiodec.c

 * MPEG Audio decoder.

 */

//#define DEBUG

#include "avcodec.h"

#include "bitstream.h"

#include "dsputil.h"

/*

 * TODO:

 *  - in low precision mode, use more 16 bit multiplies in synth filter

 *  - test lsf / mpeg25 extensively.

 */

/* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg

   audio decoder */

#ifdef CONFIG_MPEGAUDIO_HP

#define USE_HIGHPRECISION

#endif

#include "mpegaudio.h"

#define FRAC_ONE    (1 << FRAC_BITS)

#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)

#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))

#define FIX(a)   ((int)((a) * FRAC_ONE))

/* WARNING: only correct for posititive numbers */

#define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))

#define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)

#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))

//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this

static always_inline int MULH(int a, int b){

    return ((int64_t)(a) * (int64_t)(b))>>32;

}

/****************/

#define HEADER_SIZE 4

#define BACKSTEP_SIZE 512

struct GranuleDef;

typedef struct MPADecodeContext {

    uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];        /* input buffer */

    int inbuf_index;

    uint8_t *inbuf_ptr, *inbuf;

    int frame_size;

    int free_format_frame_size; /* frame size in case of free format

                                   (zero if currently unknown) */

    /* next header (used in free format parsing) */

    uint32_t free_format_next_header;

    int error_protection;

    int layer;

    int sample_rate;

    int sample_rate_index; /* between 0 and 8 */

    int bit_rate;

    int old_frame_size;

    GetBitContext gb;

    int nb_channels;

    int mode;

    int mode_ext;

    int lsf;

    MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));

    int synth_buf_offset[MPA_MAX_CHANNELS];

    int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));

    int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */

#ifdef DEBUG

    int frame_count;

#endif

    void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);

    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3

    unsigned int dither_state;

} MPADecodeContext;

/**

 * Context for MP3On4 decoder

 */

typedef struct MP3On4DecodeContext {

    int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)

    int chan_cfg; ///< channel config number

    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

} MP3On4DecodeContext;

/* layer 3 "granule" */

typedef struct GranuleDef {

    uint8_t scfsi;

    int part2_3_length;

    int big_values;

    int global_gain;

    int scalefac_compress;

    uint8_t block_type;

    uint8_t switch_point;

    int table_select[3];

    int subblock_gain[3];

    uint8_t scalefac_scale;

    uint8_t count1table_select;

    int region_size[3]; /* number of huffman codes in each region */

    int preflag;

    int short_start, long_end; /* long/short band indexes */

    uint8_t scale_factors[40];

    int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */

} GranuleDef;

#define MODE_EXT_MS_STEREO 2

#define MODE_EXT_I_STEREO  1

/* layer 3 huffman tables */

typedef struct HuffTable {

    int xsize;

    const uint8_t *bits;

    const uint16_t *codes;

} HuffTable;

#include "mpegaudiodectab.h"

static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);

static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);

/* vlc structure for decoding layer 3 huffman tables */

static VLC huff_vlc[16];

static uint8_t *huff_code_table[16];

static VLC huff_quad_vlc[2];

/* computed from band_size_long */

static uint16_t band_index_long[9][23];

/* XXX: free when all decoders are closed */

#define TABLE_4_3_SIZE (8191 + 16)*4

static int8_t  *table_4_3_exp;

static uint32_t *table_4_3_value;

/* intensity stereo coef table */

static int32_t is_table[2][16];

static int32_t is_table_lsf[2][2][16];

static int32_t csa_table[8][4];

static float csa_table_float[8][4];

static int32_t mdct_win[8][36];

/* lower 2 bits: modulo 3, higher bits: shift */

static uint16_t scale_factor_modshift[64];

/* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */

static int32_t scale_factor_mult[15][3];

/* mult table for layer 2 group quantization */

#define SCALE_GEN(v) \

{ FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }

static const int32_t scale_factor_mult2[3][3] = {

    SCALE_GEN(4.0 / 3.0), /* 3 steps */

    SCALE_GEN(4.0 / 5.0), /* 5 steps */

    SCALE_GEN(4.0 / 9.0), /* 9 steps */

};

void ff_mpa_synth_init(MPA_INT *window);

static MPA_INT window[512] __attribute__((aligned(16)));

/* layer 1 unscaling */

/* n = number of bits of the mantissa minus 1 */

static inline int l1_unscale(int n, int mant, int scale_factor)

{

    int shift, mod;

    int64_t val;

    shift = scale_factor_modshift[scale_factor];

    mod = shift & 3;

    shift >>= 2;

    val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);

    shift += n;

    /* NOTE: at this point, 1 <= shift >= 21 + 15 */

    return (int)((val + (1LL << (shift - 1))) >> shift);

}

static inline int l2_unscale_group(int steps, int mant, int scale_factor)

{

    int shift, mod, val;

    shift = scale_factor_modshift[scale_factor];

    mod = shift & 3;

    shift >>= 2;

    val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];

    /* NOTE: at this point, 0 <= shift <= 21 */

    if (shift > 0)

        val = (val + (1 << (shift - 1))) >> shift;

    return val;

}

/* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */

static inline int l3_unscale(int value, int exponent)

{

    unsigned int m;

    int e;

    e = table_4_3_exp  [4*value + (exponent&3)];

    m = table_4_3_value[4*value + (exponent&3)];

    e -= (exponent >> 2);

    assert(e>=1);

    if (e > 31)

        return 0;

    m = (m + (1 << (e-1))) >> e;

    return m;

}

/* all integer n^(4/3) computation code */

#define DEV_ORDER 13

#define POW_FRAC_BITS 24

#define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

#define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))

#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)

static int dev_4_3_coefs[DEV_ORDER];

#if 0 /* unused */

static int pow_mult3[3] = {

    POW_FIX(1.0),

    POW_FIX(1.25992104989487316476),

    POW_FIX(1.58740105196819947474),

};

#endif

static void int_pow_init(void)

{

    int i, a;

    a = POW_FIX(1.0);

    for(i=0;i<DEV_ORDER;i++) {

        a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);

        dev_4_3_coefs[i] = a;

    }

}

#if 0 /* unused, remove? */

/* return the mantissa and the binary exponent */

static int int_pow(int i, int *exp_ptr)

{

    int e, er, eq, j;

    int a, a1;

    /* renormalize */

    a = i;

    e = POW_FRAC_BITS;

    while (a < (1 << (POW_FRAC_BITS - 1))) {

        a = a << 1;

        e--;

    }

    a -= (1 << POW_FRAC_BITS);

    a1 = 0;

    for(j = DEV_ORDER - 1; j >= 0; j--)

        a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);

    a = (1 << POW_FRAC_BITS) + a1;

    /* exponent compute (exact) */

    e = e * 4;

    er = e % 3;

    eq = e / 3;

    a = POW_MULL(a, pow_mult3[er]);

    while (a >= 2 * POW_FRAC_ONE) {

        a = a >> 1;

        eq++;

    }

    /* convert to float */

    while (a < POW_FRAC_ONE) {

        a = a << 1;

        eq--;

    }

    /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */

#if POW_FRAC_BITS > FRAC_BITS

    a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);

    /* correct overflow */

    if (a >= 2 * (1 << FRAC_BITS)) {

        a = a >> 1;

        eq++;

    }

#endif

    *exp_ptr = eq;

    return a;

}

#endif

static int decode_init(AVCodecContext * avctx)

{

    MPADecodeContext *s = avctx->priv_data;

    static int init=0;

    int i, j, k;

#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)

    avctx->sample_fmt= SAMPLE_FMT_S32;

#else

    avctx->sample_fmt= SAMPLE_FMT_S16;

#endif

    if(avctx->antialias_algo != FF_AA_FLOAT)

        s->compute_antialias= compute_antialias_integer;

    else

        s->compute_antialias= compute_antialias_float;

    if (!init && !avctx->parse_only) {

        /* scale factors table for layer 1/2 */

        for(i=0;i<64;i++) {

            int shift, mod;

            /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

            shift = (i / 3);

            mod = i % 3;

            scale_factor_modshift[i] = mod | (shift << 2);

        }

        /* scale factor multiply for layer 1 */

        for(i=0;i<15;i++) {

            int n, norm;

            n = i + 2;

            norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);

            scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);

            scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);

            scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);

            dprintf("%d: norm=%x s=%x %x %x\n",

                    i, norm,

                    scale_factor_mult[i][0],

                    scale_factor_mult[i][1],

                    scale_factor_mult[i][2]);

        }

        ff_mpa_synth_init(window);

        /* huffman decode tables */

        huff_code_table[0] = NULL;

        for(i=1;i<16;i++) {

            const HuffTable *h = &mpa_huff_tables[i];

            int xsize, x, y;

            unsigned int n;

            uint8_t *code_table;

            xsize = h->xsize;

            n = xsize * xsize;

            /* XXX: fail test */

            init_vlc(&huff_vlc[i], 8, n,

                     h->bits, 1, 1, h->codes, 2, 2, 1);

            code_table = av_mallocz(n);

            j = 0;

            for(x=0;x<xsize;x++) {

                for(y=0;y<xsize;y++)

                    code_table[j++] = (x << 4) | y;

            }

            huff_code_table[i] = code_table;

        }

        for(i=0;i<2;i++) {

            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,

                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);

        }

        for(i=0;i<9;i++) {

            k = 0;

            for(j=0;j<22;j++) {

                band_index_long[i][j] = k;

                k += band_size_long[i][j];

            }

            band_index_long[i][22] = k;

        }

        /* compute n ^ (4/3) and store it in mantissa/exp format */

        table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));

        if(!table_4_3_exp)

            return -1;

        table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));

        if(!table_4_3_value)

            return -1;

        int_pow_init();

        for(i=1;i<TABLE_4_3_SIZE;i++) {

            double f, fm;

            int e, m;

            f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);

            fm = frexp(f, &e);

            m = (uint32_t)(fm*(1LL<<31) + 0.5);

            e+= FRAC_BITS - 31 + 5;

            /* normalized to FRAC_BITS */

            table_4_3_value[i] = m;

//            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));

            table_4_3_exp[i] = -e;

        }

        for(i=0;i<7;i++) {

            float f;

            int v;

            if (i != 6) {

                f = tan((double)i * M_PI / 12.0);

                v = FIXR(f / (1.0 + f));

            } else {

                v = FIXR(1.0);

            }

            is_table[0][i] = v;

            is_table[1][6 - i] = v;

        }

        /* invalid values */

        for(i=7;i<16;i++)

            is_table[0][i] = is_table[1][i] = 0.0;

        for(i=0;i<16;i++) {

            double f;

            int e, k;

            for(j=0;j<2;j++) {

                e = -(j + 1) * ((i + 1) >> 1);

                f = pow(2.0, e / 4.0);

                k = i & 1;

                is_table_lsf[j][k ^ 1][i] = FIXR(f);

                is_table_lsf[j][k][i] = FIXR(1.0);

                dprintf("is_table_lsf %d %d: %x %x\n",

                        i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);

            }

        }

        for(i=0;i<8;i++) {

            float ci, cs, ca;

            ci = ci_table[i];

            cs = 1.0 / sqrt(1.0 + ci * ci);

            ca = cs * ci;

            csa_table[i][0] = FIXHR(cs/4);

            csa_table[i][1] = FIXHR(ca/4);

            csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);

            csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);

            csa_table_float[i][0] = cs;

            csa_table_float[i][1] = ca;

            csa_table_float[i][2] = ca + cs;

            csa_table_float[i][3] = ca - cs;

//            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));

//            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);

        }

        /* compute mdct windows */

        for(i=0;i<36;i++) {

            for(j=0; j<4; j++){

                double d;

                if(j==2 && i%3 != 1)

                    continue;

                d= sin(M_PI * (i + 0.5) / 36.0);

                if(j==1){

                    if     (i>=30) d= 0;

                    else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);

                    else if(i>=18) d= 1;

                }else if(j==3){

                    if     (i<  6) d= 0;

                    else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);

                    else if(i< 18) d= 1;

                }

                //merge last stage of imdct into the window coefficients

                d*= 0.5 / cos(M_PI*(2*i + 19)/72);

                if(j==2)

                    mdct_win[j][i/3] = FIXHR((d / (1<<5)));

                else

                    mdct_win[j][i  ] = FIXHR((d / (1<<5)));

//                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));

            }

        }

        /* NOTE: we do frequency inversion adter the MDCT by changing

           the sign of the right window coefs */

        for(j=0;j<4;j++) {

            for(i=0;i<36;i+=2) {

                mdct_win[j + 4][i] = mdct_win[j][i];

                mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];

            }

        }

#if defined(DEBUG)

        for(j=0;j<8;j++) {

            printf("win%d=\n", j);

            for(i=0;i<36;i++)

                printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);

            printf("\n");

        }

#endif

        init = 1;

    }

    s->inbuf_index = 0;

    s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

    s->inbuf_ptr = s->inbuf;

#ifdef DEBUG

    s->frame_count = 0;

#endif

    if (avctx->codec_id == CODEC_ID_MP3ADU)

        s->adu_mode = 1;

    return 0;

}

/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */

/* cos(i*pi/64) */

#define COS0_0  FIXR(0.50060299823519630134)

#define COS0_1  FIXR(0.50547095989754365998)

#define COS0_2  FIXR(0.51544730992262454697)

#define COS0_3  FIXR(0.53104259108978417447)

#define COS0_4  FIXR(0.55310389603444452782)

#define COS0_5  FIXR(0.58293496820613387367)

#define COS0_6  FIXR(0.62250412303566481615)

#define COS0_7  FIXR(0.67480834145500574602)

#define COS0_8  FIXR(0.74453627100229844977)

#define COS0_9  FIXR(0.83934964541552703873)

#define COS0_10 FIXR(0.97256823786196069369)

#define COS0_11 FIXR(1.16943993343288495515)

#define COS0_12 FIXR(1.48416461631416627724)

#define COS0_13 FIXR(2.05778100995341155085)

#define COS0_14 FIXR(3.40760841846871878570)

#define COS0_15 FIXR(10.19000812354805681150)

#define COS1_0 FIXR(0.50241928618815570551)

#define COS1_1 FIXR(0.52249861493968888062)

#define COS1_2 FIXR(0.56694403481635770368)

#define COS1_3 FIXR(0.64682178335999012954)

#define COS1_4 FIXR(0.78815462345125022473)

#define COS1_5 FIXR(1.06067768599034747134)

#define COS1_6 FIXR(1.72244709823833392782)

#define COS1_7 FIXR(5.10114861868916385802)

#define COS2_0 FIXR(0.50979557910415916894)

#define COS2_1 FIXR(0.60134488693504528054)

#define COS2_2 FIXR(0.89997622313641570463)

#define COS2_3 FIXR(2.56291544774150617881)

#define COS3_0 FIXR(0.54119610014619698439)

#define COS3_1 FIXR(1.30656296487637652785)

#define COS4_0 FIXR(0.70710678118654752439)

/* butterfly operator */

#define BF(a, b, c)\

{\

    tmp0 = tab[a] + tab[b];\

    tmp1 = tab[a] - tab[b];\

    tab[a] = tmp0;\

    tab[b] = MULL(tmp1, c);\

}

#define BF1(a, b, c, d)\

{\

    BF(a, b, COS4_0);\

    BF(c, d, -COS4_0);\

    tab[c] += tab[d];\

}

#define BF2(a, b, c, d)\

{\

    BF(a, b, COS4_0);\

    BF(c, d, -COS4_0);\

    tab[c] += tab[d];\

    tab[a] += tab[c];\

    tab[c] += tab[b];\

    tab[b] += tab[d];\

}

#define ADD(a, b) tab[a] += tab[b]

/* DCT32 without 1/sqrt(2) coef zero scaling. */

static void dct32(int32_t *out, int32_t *tab)

{

    int tmp0, tmp1;

    /* pass 1 */

    BF(0, 31, COS0_0);

    BF(1, 30, COS0_1);

    BF(2, 29, COS0_2);

    BF(3, 28, COS0_3);

    BF(4, 27, COS0_4);

    BF(5, 26, COS0_5);

    BF(6, 25, COS0_6);

    BF(7, 24, COS0_7);

    BF(8, 23, COS0_8);

    BF(9, 22, COS0_9);

    BF(10, 21, COS0_10);

    BF(11, 20, COS0_11);

    BF(12, 19, COS0_12);

    BF(13, 18, COS0_13);

    BF(14, 17, COS0_14);

    BF(15, 16, COS0_15);

    /* pass 2 */

    BF(0, 15, COS1_0);

    BF(1, 14, COS1_1);

    BF(2, 13, COS1_2);

    BF(3, 12, COS1_3);

    BF(4, 11, COS1_4);

    BF(5, 10, COS1_5);

    BF(6,  9, COS1_6);

    BF(7,  8, COS1_7);

    BF(16, 31, -COS1_0);

    BF(17, 30, -COS1_1);

    BF(18, 29, -COS1_2);

    BF(19, 28, -COS1_3);

    BF(20, 27, -COS1_4);

    BF(21, 26, -COS1_5);

    BF(22, 25, -COS1_6);

    BF(23, 24, -COS1_7);

    /* pass 3 */

    BF(0, 7, COS2_0);

    BF(1, 6, COS2_1);

    BF(2, 5, COS2_2);

    BF(3, 4, COS2_3);

    BF(8, 15, -COS2_0);

    BF(9, 14, -COS2_1);

    BF(10, 13, -COS2_2);

    BF(11, 12, -COS2_3);

    BF(16, 23, COS2_0);

    BF(17, 22, COS2_1);

    BF(18, 21, COS2_2);

    BF(19, 20, COS2_3);

    BF(24, 31, -COS2_0);

    BF(25, 30, -COS2_1);

    BF(26, 29, -COS2_2);

    BF(27, 28, -COS2_3);

    /* pass 4 */

    BF(0, 3, COS3_0);

    BF(1, 2, COS3_1);

    BF(4, 7, -COS3_0);

    BF(5, 6, -COS3_1);

    BF(8, 11, COS3_0);

    BF(9, 10, COS3_1);

    BF(12, 15, -COS3_0);

    BF(13, 14, -COS3_1);

    BF(16, 19, COS3_0);

    BF(17, 18, COS3_1);

    BF(20, 23, -COS3_0);

    BF(21, 22, -COS3_1);

    BF(24, 27, COS3_0);

    BF(25, 26, COS3_1);

    BF(28, 31, -COS3_0);

    BF(29, 30, -COS3_1);

    /* pass 5 */

    BF1(0, 1, 2, 3);

    BF2(4, 5, 6, 7);

    BF1(8, 9, 10, 11);

    BF2(12, 13, 14, 15);

    BF1(16, 17, 18, 19);

    BF2(20, 21, 22, 23);

    BF1(24, 25, 26, 27);

    BF2(28, 29, 30, 31);

    /* pass 6 */

    ADD( 8, 12);

    ADD(12, 10);

    ADD(10, 14);

    ADD(14,  9);

    ADD( 9, 13);

    ADD(13, 11);

    ADD(11, 15);

    out[ 0] = tab[0];

    out[16] = tab[1];

    out[ 8] = tab[2];

    out[24] = tab[3];

    out[ 4] = tab[4];

    out[20] = tab[5];

    out[12] = tab[6];

    out[28] = tab[7];

    out[ 2] = tab[8];

    out[18] = tab[9];

    out[10] = tab[10];

    out[26] = tab[11];

    out[ 6] = tab[12];

    out[22] = tab[13];

    out[14] = tab[14];

    out[30] = tab[15];

    ADD(24, 28);

    ADD(28, 26);

    ADD(26, 30);

    ADD(30, 25);

    ADD(25, 29);

    ADD(29, 27);