mpegaudiodec.c

/*
 * MPEG Audio decoder
 * Copyright (c) 2001, 2002 Fabrice Bellard
 *
 * This file is part of Libav.
 *
 * Libav 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 2.1 of the License, or (at your option) any later version.
 *
 * Libav 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 Libav; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * MPEG Audio decoder.
 */

#include "libavutil/audioconvert.h"
#include "avcodec.h"
#include "get_bits.h"
#include "dsputil.h"
#include "mathops.h"
#include "dct32.h"

/*
 * TODO:
 *  - test lsf / mpeg25 extensively.
 */

#include "mpegaudio.h"
#include "mpegaudiodecheader.h"

#if CONFIG_FLOAT
#   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
#   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
#   define FIXR(x)        ((float)(x))
#   define FIXHR(x)       ((float)(x))
#   define MULH3(x, y, s) ((s)*(y)*(x))
#   define MULLx(x, y, s) ((y)*(x))
#   define RENAME(a) a ## _float
#   define OUT_FMT AV_SAMPLE_FMT_FLT
#else
#   define SHR(a,b)       ((a)>>(b))
/* WARNING: only correct for posititive numbers */
#   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
#   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
#   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
#   define MULH3(x, y, s) MULH((s)*(x), y)
#   define MULLx(x, y, s) MULL(x,y,s)
#   define RENAME(a)      a ## _fixed
#   define OUT_FMT AV_SAMPLE_FMT_S16
#endif

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

#define HEADER_SIZE 4

#include "mpegaudiodata.h"
#include "mpegaudiodectab.h"

static void RENAME(compute_antialias)(MPADecodeContext *s, GranuleDef *g);
static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
                               int *dither_state, OUT_INT *samples, int incr);

/* vlc structure for decoding layer 3 huffman tables */
static VLC huff_vlc[16];
static VLC_TYPE huff_vlc_tables[
  0+128+128+128+130+128+154+166+
  142+204+190+170+542+460+662+414
  ][2];
static const int huff_vlc_tables_sizes[16] = {
  0, 128, 128, 128, 130, 128, 154, 166,
  142, 204, 190, 170, 542, 460, 662, 414
};
static VLC huff_quad_vlc[2];
static VLC_TYPE huff_quad_vlc_tables[128+16][2];
static const int huff_quad_vlc_tables_sizes[2] = {
  128, 16
};
/* computed from band_size_long */
static uint16_t band_index_long[9][23];
#include "mpegaudio_tablegen.h"
/* intensity stereo coef table */
static INTFLOAT is_table[2][16];
static INTFLOAT is_table_lsf[2][2][16];
static int32_t csa_table[8][4];
static float csa_table_float[8][4];
static INTFLOAT mdct_win[8][36];

static int16_t division_tab3[1<<6 ];
static int16_t division_tab5[1<<8 ];
static int16_t division_tab9[1<<11];

static int16_t * const division_tabs[4] = {
    division_tab3, division_tab5, NULL, division_tab9
};

/* 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_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(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 */
};

DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];

/**
 * Convert region offsets to region sizes and truncate
 * size to big_values.
 */
static void ff_region_offset2size(GranuleDef *g){
    int i, k, j=0;
    g->region_size[2] = (576 / 2);
    for(i=0;i<3;i++) {
        k = FFMIN(g->region_size[i], g->big_values);
        g->region_size[i] = k - j;
        j = k;
    }
}

static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
    if (g->block_type == 2)
        g->region_size[0] = (36 / 2);
    else {
        if (s->sample_rate_index <= 2)
            g->region_size[0] = (36 / 2);
        else if (s->sample_rate_index != 8)
            g->region_size[0] = (54 / 2);
        else
            g->region_size[0] = (108 / 2);
    }
    g->region_size[1] = (576 / 2);
}

static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
    int l;
    g->region_size[0] =
        band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
    /* should not overflow */
    l = FFMIN(ra1 + ra2 + 2, 22);
    g->region_size[1] =
        band_index_long[s->sample_rate_index][l] >> 1;
}

static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
    if (g->block_type == 2) {
        if (g->switch_point) {
            /* if switched mode, we handle the 36 first samples as
                long blocks.  For 8000Hz, we handle the 48 first
                exponents as long blocks (XXX: check this!) */
            if (s->sample_rate_index <= 2)
                g->long_end = 8;
            else if (s->sample_rate_index != 8)
                g->long_end = 6;
            else
                g->long_end = 4; /* 8000 Hz */

            g->short_start = 2 + (s->sample_rate_index != 8);
        } else {
            g->long_end = 0;
            g->short_start = 0;
        }
    } else {
        g->short_start = 13;
        g->long_end = 22;
    }
}

/* 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];

static av_cold 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;
    }
}

static av_cold int decode_init(AVCodecContext * avctx)
{
    MPADecodeContext *s = avctx->priv_data;
    static int init=0;
    int i, j, k;

    s->avctx = avctx;
    s->apply_window_mp3 = apply_window_mp3_c;
#if HAVE_MMX && CONFIG_FLOAT
    ff_mpegaudiodec_init_mmx(s);
#endif
#if CONFIG_FLOAT
    ff_dct_init(&s->dct, 5, DCT_II);
#endif
    if (HAVE_ALTIVEC && CONFIG_FLOAT) ff_mpegaudiodec_init_altivec(s);

    avctx->sample_fmt= OUT_FMT;
    s->error_recognition= avctx->error_recognition;

    if (!init && !avctx->parse_only) {
        int offset;

        /* 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_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
            scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0          * 2.0), FRAC_BITS);
            scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
            scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
            av_dlog(avctx, "%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]);
        }

        RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));

        /* huffman decode tables */
        offset = 0;
        for(i=1;i<16;i++) {
            const HuffTable *h = &mpa_huff_tables[i];
            int xsize, x, y;
            uint8_t  tmp_bits [512];
            uint16_t tmp_codes[512];

            memset(tmp_bits , 0, sizeof(tmp_bits ));
            memset(tmp_codes, 0, sizeof(tmp_codes));

            xsize = h->xsize;

            j = 0;
            for(x=0;x<xsize;x++) {
                for(y=0;y<xsize;y++){
                    tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
                    tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
                }
            }

            /* XXX: fail test */
            huff_vlc[i].table = huff_vlc_tables+offset;
            huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
            init_vlc(&huff_vlc[i], 7, 512,
                     tmp_bits, 1, 1, tmp_codes, 2, 2,
                     INIT_VLC_USE_NEW_STATIC);
            offset += huff_vlc_tables_sizes[i];
        }
        assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));

        offset = 0;
        for(i=0;i<2;i++) {
            huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
            huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
            init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
                     mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
                     INIT_VLC_USE_NEW_STATIC);
            offset += huff_quad_vlc_tables_sizes[i];
        }
        assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));

        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 */

        int_pow_init();
        mpegaudio_tableinit();

        for (i = 0; i < 4; i++)
            if (ff_mpa_quant_bits[i] < 0)
                for (j = 0; j < (1<<(-ff_mpa_quant_bits[i]+1)); j++) {
                    int val1, val2, val3, steps;
                    int val = j;
                    steps  = ff_mpa_quant_steps[i];
                    val1 = val % steps;
                    val /= steps;
                    val2 = val % steps;
                    val3 = val / steps;
                    division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
                }


        for(i=0;i<7;i++) {
            float f;
            INTFLOAT 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);
                av_dlog(avctx, "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;
        }

        /* 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)));
            }
        }

        /* 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];
            }
        }

        init = 1;
    }

    if (avctx->codec_id == CODEC_ID_MP3ADU)
        s->adu_mode = 1;
    return 0;
}


#if CONFIG_FLOAT
static inline float round_sample(float *sum)
{
    float sum1=*sum;
    *sum = 0;
    return sum1;
}

/* signed 16x16 -> 32 multiply add accumulate */
#define MACS(rt, ra, rb) rt+=(ra)*(rb)

/* signed 16x16 -> 32 multiply */
#define MULS(ra, rb) ((ra)*(rb))

#define MLSS(rt, ra, rb) rt-=(ra)*(rb)

#else

static inline int round_sample(int64_t *sum)
{
    int sum1;
    sum1 = (int)((*sum) >> OUT_SHIFT);
    *sum &= (1<<OUT_SHIFT)-1;
    return av_clip_int16(sum1);
}

#   define MULS(ra, rb) MUL64(ra, rb)
#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
#endif

#define SUM8(op, sum, w, p)               \
{                                         \
    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
}

#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
{                                               \
    INTFLOAT tmp;\
    tmp = p[0 * 64];\
    op1(sum1, (w1)[0 * 64], tmp);\
    op2(sum2, (w2)[0 * 64], tmp);\
    tmp = p[1 * 64];\
    op1(sum1, (w1)[1 * 64], tmp);\
    op2(sum2, (w2)[1 * 64], tmp);\
    tmp = p[2 * 64];\
    op1(sum1, (w1)[2 * 64], tmp);\
    op2(sum2, (w2)[2 * 64], tmp);\
    tmp = p[3 * 64];\
    op1(sum1, (w1)[3 * 64], tmp);\
    op2(sum2, (w2)[3 * 64], tmp);\
    tmp = p[4 * 64];\
    op1(sum1, (w1)[4 * 64], tmp);\
    op2(sum2, (w2)[4 * 64], tmp);\
    tmp = p[5 * 64];\
    op1(sum1, (w1)[5 * 64], tmp);\
    op2(sum2, (w2)[5 * 64], tmp);\
    tmp = p[6 * 64];\
    op1(sum1, (w1)[6 * 64], tmp);\
    op2(sum2, (w2)[6 * 64], tmp);\
    tmp = p[7 * 64];\
    op1(sum1, (w1)[7 * 64], tmp);\
    op2(sum2, (w2)[7 * 64], tmp);\
}

void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
{
    int i, j;

    /* max = 18760, max sum over all 16 coefs : 44736 */
    for(i=0;i<257;i++) {
        INTFLOAT v;
        v = ff_mpa_enwindow[i];
#if CONFIG_FLOAT
        v *= 1.0 / (1LL<<(16 + FRAC_BITS));
#endif
        window[i] = v;
        if ((i & 63) != 0)
            v = -v;
        if (i != 0)
            window[512 - i] = v;
    }

    // Needed for avoiding shuffles in ASM implementations
    for(i=0; i < 8; i++)
        for(j=0; j < 16; j++)
            window[512+16*i+j] = window[64*i+32-j];

    for(i=0; i < 8; i++)
        for(j=0; j < 16; j++)
            window[512+128+16*i+j] = window[64*i+48-j];
}

static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
                               int *dither_state, OUT_INT *samples, int incr)
{
    register const MPA_INT *w, *w2, *p;
    int j;
    OUT_INT *samples2;
#if CONFIG_FLOAT
    float sum, sum2;
#else
    int64_t sum, sum2;
#endif

    /* copy to avoid wrap */
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));

    samples2 = samples + 31 * incr;
    w = window;
    w2 = window + 31;

    sum = *dither_state;
    p = synth_buf + 16;
    SUM8(MACS, sum, w, p);
    p = synth_buf + 48;
    SUM8(MLSS, sum, w + 32, p);
    *samples = round_sample(&sum);
    samples += incr;
    w++;

    /* we calculate two samples at the same time to avoid one memory
       access per two sample */
    for(j=1;j<16;j++) {
        sum2 = 0;
        p = synth_buf + 16 + j;
        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
        p = synth_buf + 48 - j;
        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);

        *samples = round_sample(&sum);
        samples += incr;
        sum += sum2;
        *samples2 = round_sample(&sum);
        samples2 -= incr;
        w++;
        w2--;
    }

    p = synth_buf + 32;
    SUM8(MLSS, sum, w + 32, p);
    *samples = round_sample(&sum);
    *dither_state= sum;
}


/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
   32 samples. */
/* XXX: optimize by avoiding ring buffer usage */
#if !CONFIG_FLOAT
void ff_mpa_synth_filter_fixed(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
                         MPA_INT *window, int *dither_state,
                         OUT_INT *samples, int incr,
                         INTFLOAT sb_samples[SBLIMIT])
{
    register MPA_INT *synth_buf;
    int offset;

    offset = *synth_buf_offset;
    synth_buf = synth_buf_ptr + offset;

    ff_dct32_fixed(synth_buf, sb_samples);
    apply_window_mp3_c(synth_buf, window, dither_state, samples, incr);

    offset = (offset - 32) & 511;
    *synth_buf_offset = offset;
}
#endif

#define C3 FIXHR(0.86602540378443864676/2)

/* 0.5 / cos(pi*(2*i+1)/36) */
static const INTFLOAT icos36[9] = {
    FIXR(0.50190991877167369479),
    FIXR(0.51763809020504152469), //0
    FIXR(0.55168895948124587824),
    FIXR(0.61038729438072803416),
    FIXR(0.70710678118654752439), //1
    FIXR(0.87172339781054900991),
    FIXR(1.18310079157624925896),
    FIXR(1.93185165257813657349), //2
    FIXR(5.73685662283492756461),
};

/* 0.5 / cos(pi*(2*i+1)/36) */
static const INTFLOAT icos36h[9] = {
    FIXHR(0.50190991877167369479/2),
    FIXHR(0.51763809020504152469/2), //0
    FIXHR(0.55168895948124587824/2),
    FIXHR(0.61038729438072803416/2),
    FIXHR(0.70710678118654752439/2), //1
    FIXHR(0.87172339781054900991/2),
    FIXHR(1.18310079157624925896/4),
    FIXHR(1.93185165257813657349/4), //2
//    FIXHR(5.73685662283492756461),
};

/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
   cases. */
static void imdct12(INTFLOAT *out, INTFLOAT *in)
{
    INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;

    in0= in[0*3];
    in1= in[1*3] + in[0*3];
    in2= in[2*3] + in[1*3];
    in3= in[3*3] + in[2*3];
    in4= in[4*3] + in[3*3];
    in5= in[5*3] + in[4*3];
    in5 += in3;
    in3 += in1;

    in2= MULH3(in2, C3, 2);
    in3= MULH3(in3, C3, 4);

    t1 = in0 - in4;
    t2 = MULH3(in1 - in5, icos36h[4], 2);

    out[ 7]=
    out[10]= t1 + t2;
    out[ 1]=
    out[ 4]= t1 - t2;

    in0 += SHR(in4, 1);
    in4 = in0 + in2;
    in5 += 2*in1;
    in1 = MULH3(in5 + in3, icos36h[1], 1);
    out[ 8]=
    out[ 9]= in4 + in1;
    out[ 2]=
    out[ 3]= in4 - in1;

    in0 -= in2;
    in5 = MULH3(in5 - in3, icos36h[7], 2);
    out[ 0]=
    out[ 5]= in0 - in5;
    out[ 6]=
    out[11]= in0 + in5;
}

/* cos(pi*i/18) */
#define C1 FIXHR(0.98480775301220805936/2)
#define C2 FIXHR(0.93969262078590838405/2)
#define C3 FIXHR(0.86602540378443864676/2)
#define C4 FIXHR(0.76604444311897803520/2)
#define C5 FIXHR(0.64278760968653932632/2)
#define C6 FIXHR(0.5/2)
#define C7 FIXHR(0.34202014332566873304/2)
#define C8 FIXHR(0.17364817766693034885/2)


/* using Lee like decomposition followed by hand coded 9 points DCT */
static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
{
    int i, j;
    INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
    INTFLOAT tmp[18], *tmp1, *in1;

    for(i=17;i>=1;i--)
        in[i] += in[i-1];
    for(i=17;i>=3;i-=2)
        in[i] += in[i-2];

    for(j=0;j<2;j++) {
        tmp1 = tmp + j;
        in1 = in + j;

        t2 = in1[2*4] + in1[2*8] - in1[2*2];

        t3 = in1[2*0] + SHR(in1[2*6],1);
        t1 = in1[2*0] - in1[2*6];
        tmp1[ 6] = t1 - SHR(t2,1);
        tmp1[16] = t1 + t2;

        t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
        t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
        t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);

        tmp1[10] = t3 - t0 - t2;
        tmp1[ 2] = t3 + t0 + t1;
        tmp1[14] = t3 + t2 - t1;

        tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
        t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
        t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
        t0 = MULH3(in1[2*3], C3, 2);

        t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);

        tmp1[ 0] = t2 + t3 + t0;
        tmp1[12] = t2 + t1 - t0;
        tmp1[ 8] = t3 - t1 - t0;
    }

    i = 0;
    for(j=0;j<4;j++) {
        t0 = tmp[i];
        t1 = tmp[i + 2];
        s0 = t1 + t0;
        s2 = t1 - t0;

        t2 = tmp[i + 1];
        t3 = tmp[i + 3];
        s1 = MULH3(t3 + t2, icos36h[j], 2);
        s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);

        t0 = s0 + s1;
        t1 = s0 - s1;
        out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];
        out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];
        buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);
        buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);

        t0 = s2 + s3;
        t1 = s2 - s3;
        out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
        out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];
        buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);
        buf[      + j] = MULH3(t0, win[18         + j], 1);
        i += 4;
    }

    s0 = tmp[16];
    s1 = MULH3(tmp[17], icos36h[4], 2);
    t0 = s0 + s1;
    t1 = s0 - s1;
    out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
    out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
    buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);
    buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);
}

/* return the number of decoded frames */
static int mp_decode_layer1(MPADecodeContext *s)
{
    int bound, i, v, n, ch, j, mant;
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

    if (s->mode == MPA_JSTEREO)
        bound = (s->mode_ext + 1) * 4;
    else
        bound = SBLIMIT;

    /* allocation bits */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            allocation[ch][i] = get_bits(&s->gb, 4);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        allocation[0][i] = get_bits(&s->gb, 4);
    }

    /* scale factors */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (allocation[ch][i])
                scale_factors[ch][i] = get_bits(&s->gb, 6);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        if (allocation[0][i]) {
            scale_factors[0][i] = get_bits(&s->gb, 6);
            scale_factors[1][i] = get_bits(&s->gb, 6);
        }
    }

    /* compute samples */
    for(j=0;j<12;j++) {
        for(i=0;i<bound;i++) {
            for(ch=0;ch<s->nb_channels;ch++) {
                n = allocation[ch][i];
                if (n) {
                    mant = get_bits(&s->gb, n + 1);
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
                } else {
                    v = 0;
                }
                s->sb_samples[ch][j][i] = v;
            }
        }
        for(i=bound;i<SBLIMIT;i++) {
            n = allocation[0][i];
            if (n) {
                mant = get_bits(&s->gb, n + 1);
                v = l1_unscale(n, mant, scale_factors[0][i]);
                s->sb_samples[0][j][i] = v;
                v = l1_unscale(n, mant, scale_factors[1][i]);
                s->sb_samples[1][j][i] = v;
            } else {
                s->sb_samples[0][j][i] = 0;
                s->sb_samples[1][j][i] = 0;
            }
        }
    }
    return 12;
}

static int mp_decode_layer2(MPADecodeContext *s)
{
    int sblimit; /* number of used subbands */
    const unsigned char *alloc_table;
    int table, bit_alloc_bits, i, j, ch, bound, v;
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
    int scale, qindex, bits, steps, k, l, m, b;

    /* select decoding table */
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
                            s->sample_rate, s->lsf);
    sblimit = ff_mpa_sblimit_table[table];
    alloc_table = ff_mpa_alloc_tables[table];

    if (s->mode == MPA_JSTEREO)
        bound = (s->mode_ext + 1) * 4;
    else
        bound = sblimit;

    av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

    /* sanity check */
    if( bound > sblimit ) bound = sblimit;

    /* parse bit allocation */
    j = 0;
    for(i=0;i<bound;i++) {
        bit_alloc_bits = alloc_table[j];
        for(ch=0;ch<s->nb_channels;ch++) {
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
        }
        j += 1 << bit_alloc_bits;
    }
    for(i=bound;i<sblimit;i++) {
        bit_alloc_bits = alloc_table[j];
        v = get_bits(&s->gb, bit_alloc_bits);
        bit_alloc[0][i] = v;
        bit_alloc[1][i] = v;
        j += 1 << bit_alloc_bits;
    }

    /* scale codes */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (bit_alloc[ch][i])
                scale_code[ch][i] = get_bits(&s->gb, 2);
        }
    }

    /* scale factors */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (bit_alloc[ch][i]) {
                sf = scale_factors[ch][i];
                switch(scale_code[ch][i]) {
                default:
                case 0:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    break;
                case 2:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    sf[2] = sf[0];
                    break;
                case 1:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    break;
                case 3:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[2];
                    break;
                }
            }
        }
    }

    /* samples */
    for(k=0;k<3;k++) {
        for(l=0;l<12;l+=3) {
            j = 0;
            for(i=0;i<bound;i++) {
                bit_alloc_bits = alloc_table[j];
                for(ch=0;ch<s->nb_channels;ch++) {
                    b = bit_alloc[ch][i];
                    if (b) {
                        scale = scale_factors[ch][i][k];
                        qindex = alloc_table[j+b];
                        bits = ff_mpa_quant_bits[qindex];
                        if (bits < 0) {
                            int v2;
                            /* 3 values at the same time */
                            v = get_bits(&s->gb, -bits);
                            v2 = division_tabs[qindex][v];
                            steps  = ff_mpa_quant_steps[qindex];

                            s->sb_samples[ch][k * 12 + l + 0][i] =
                                l2_unscale_group(steps, v2        & 15, scale);
                            s->sb_samples[ch][k * 12 + l + 1][i] =
                                l2_unscale_group(steps, (v2 >> 4) & 15, scale);
                            s->sb_samples[ch][k * 12 + l + 2][i] =
                                l2_unscale_group(steps,  v2 >> 8      , scale);
                        } else {
                            for(m=0;m<3;m++) {
                                v = get_bits(&s->gb, bits);
                                v = l1_unscale(bits - 1, v, scale);
                                s->sb_samples[ch][k * 12 + l + m][i] = v;
                            }
                        }
                    } else {
                        s->sb_samples[ch][k * 12 + l + 0][i] = 0;
                        s->sb_samples[ch][k * 12 + l + 1][i] = 0;
                        s->sb_samples[ch][k * 12 + l + 2][i] = 0;
                    }
                }
                /* next subband in alloc table */
                j += 1 << bit_alloc_bits;
            }
            /* XXX: find a way to avoid this duplication of code */
            for(i=bound;i<sblimit;i++) {
                bit_alloc_bits = alloc_table[j];
                b = bit_alloc[0][i];
                if (b) {
                    int mant, scale0, scale1;