postprocessing_mmx.c

来自「VLC媒体播放程序」· C语言 代码 · 共 890 行 · 第 1/3 页

    MMXEXT_GET_LMINMAX( (%%eax, %3),    %%mm0, %%mm1, %%mm7 )
    MMXEXT_GET_LMINMAX( (%%eax, %3,2), %%mm0, %%mm1, %%mm7 )
    MMXEXT_GET_LMINMAX( (%2, %3, 4),   %%mm0, %%mm1, %%mm7 )
    "leal   (%%eax,%3,4),  %%eax       \n"

    MMXEXT_GET_LMINMAX( (%%eax),        %%mm0, %%mm1, %%mm7 )
    MMXEXT_GET_LMINMAX( (%%eax, %3),    %%mm0, %%mm1, %%mm7 )
    MMXEXT_GET_LMINMAX( (%%eax, %3,2), %%mm0, %%mm1, %%mm7 )
    MMXEXT_GET_PMIN( %%mm0, %%mm7 )
    MMXEXT_GET_PMAX( %%mm1, %%mm7 )
    "movd   %%mm0,  %%eax               \n"
    "andl   $255,   %%eax               \n"
    "movl   %%eax,  (%0)                \n"
    "movd   %%mm1,  %%eax               \n"
    "andl   $255,   %%eax               \n"
    "movl   %%eax,  (%1)                \n"

     : : "r"(pi_min), "r"(pi_max), "r"(p_block), "r"(i_stride) : "%eax", "memory" );
#endif

    i_min = 255; i_max = 0;

    for( y = 0; y < 8; y++ )
    {
        for( x = 0; x < 8; x++ )
        {
            if( i_min > p_block[x] ) i_min = p_block[x];
            if( i_max < p_block[x] ) i_max = p_block[x];
        }
        p_block += i_stride;
    }

    *pi_min = i_min;
    *pi_max = i_max;
}


static inline void pp_dering_BinIndex( uint8_t  *p_block, int i_stride,
                                       int i_thr, uint32_t *p_bin )
{
    int x, y;
    uint32_t i_bin;

    for( y = 0; y < 10; y++ )
    {
        i_bin = 0;
        for( x = 0; x < 10; x++ )
        {
            if( p_block[x] > i_thr )
            {
                i_bin |= 1 << x;
            }
        }
        i_bin |= (~i_bin) << 16;  /* for detect also three 0 */
        *p_bin = i_bin&( i_bin >> 1 )&( i_bin << 1 );

        p_block += i_stride;
        p_bin++;
    }
}

static inline void pp_dering_Filter( uint8_t  *p_block, int i_stride,
                                     uint32_t *p_bin,
                                     int i_QP )
{
    int x, y;
    uint32_t i_bin;
    int i_flt[8][8];
    int i_f;
    uint8_t *p_sav;
    int i_QP_2;

    p_sav = p_block;
    i_QP_2 = i_QP >> 1;

    for( y = 0; y < 8; y++ )
    {
        i_bin = p_bin[y] & p_bin[y+1] & p_bin[y+2]; /* To be optimised */
        i_bin |= i_bin >> 16; /* detect 0 or 1 */

        for( x = 0; x < 8; x++ )
        {
            if( i_bin&0x02 ) /* 0x02 since 10 index but want 1-9 */
            {
                /* apply dering */
                /* 1 2 1
                   2 4 2   + (8)
                   1 2 1 */
                i_f =   p_block[x - i_stride - 1] +
                      ( p_block[x - i_stride    ] << 1)+
                        p_block[x - i_stride + 1] +

                      ( p_block[x - 1] << 1 )+
                      ( p_block[x    ] << 2 )+
                      ( p_block[x + 1] << 1 )+

                        p_block[x + i_stride - 1] +
                      ( p_block[x + i_stride    ] << 1 ) +
                        p_block[x + i_stride + 1];

                i_f = ( 8 + i_f ) >> 4;

                /* Clamp this value */

                if( i_f - p_block[x] > ( i_QP_2 ) )
                {
                    i_flt[y][x] = p_block[x] + i_QP_2;
                }
                else
                if( i_f - p_block[x] < -i_QP_2 )
                {
                    i_flt[y][x] = p_block[x] - i_QP_2;
                }
                else
                {
                    i_flt[y][x] = i_f ;
                }
            }
            else
            {
                i_flt[y][x] = p_block[x];
            }
            i_bin >>= 1;

        }
        p_block += i_stride;
    }
    for( y = 0; y < 8; y++ )
    {
        for( x = 0; x < 8; x++ )
        {
            p_sav[x] = i_flt[y][x];
        }
        p_sav+= i_stride;
    }
}


/*****************************************************************************/
/*---------------------------------------------------------------------------*/
/*                                                                           */
/*     ----------------- Dering filter on Y and C blocks -----------------   */
/*                                                                           */
/*---------------------------------------------------------------------------*/
/*****************************************************************************/

void E_( pp_dering_Y )( uint8_t *p_plane,
                        int i_width, int i_height, int i_stride,
                        QT_STORE_T *p_QP_store, int i_QP_stride )
{
    int x, y, k;
    int i_max[4], i_min[4], i_range[4];
    int i_thr[4];
    int i_max_range, i_kmax;
    uint32_t i_bin[4][10];
    uint8_t  *p_block[4];
    QT_STORE_T *p_QP;

    /* We process 4 blocks/loop*/
    for( y = 8; y < i_height-8; y += 16 )
    {
        /* +---+
           |0|1|
           +-+-+   :))
           |2|3|
           +-+-+ */

        p_block[0] = p_plane + y * i_stride + 8;
        p_block[1] = p_block[0] + 8;
        p_block[2] = p_block[0] + ( i_stride << 3 );
        p_block[3] = p_block[2] + 8;

        for( x = 8; x < i_width-8; x += 16 )
        {
            /* 1: Calculate threshold */
            /* Calculate max/min for each block */
            pp_dering_MinMax( p_block[0], i_stride, &i_min[0], &i_max[0] );
            pp_dering_MinMax( p_block[1], i_stride, &i_min[1], &i_max[1] );
            pp_dering_MinMax( p_block[2], i_stride, &i_min[2], &i_max[2] );
            pp_dering_MinMax( p_block[3], i_stride, &i_min[3], &i_max[3] );
            /* Calculate range, max_range and thr */
            i_max_range = 0; i_kmax = 0;
            for( k = 0; k <= 4; k++ )
            {
                i_range[k] = i_max[k] - i_min[k];
                i_thr[k] = ( i_max[k] + i_min[k] + 1 )/2;
                if( i_max_range < i_max[k])
                {
                    i_max_range = i_max[k];
                    i_kmax = k;
                }
            }
            /* Now rearrange thr */
            if(  i_max_range > 64 )
            {
                for( k = 1; k < 5; k++ )
                {
                    if( i_range[k] < 16 )
                    {
                        i_thr[k] = 0;
                    }
                    else
                    if( i_range[k] < 32 )
                    {
                        i_thr[k] = i_thr[i_kmax];
                    }
                }
            }
            else
            {
                for( k = 1; k < 5; k++ )
                {
                    if( i_range[k] < 16 )
                    {
                        i_thr[k] = 0;
                    }
                }
            }
            /* 2: Index acquisition 10x10 ! so " -i_stride - 1"*/
            pp_dering_BinIndex( p_block[0] - i_stride - 1, i_stride,
                                i_thr[0], i_bin[0] );
            pp_dering_BinIndex( p_block[1] - i_stride - 1, i_stride,
                                i_thr[1], i_bin[1] );
            pp_dering_BinIndex( p_block[2] - i_stride - 1, i_stride,
                                i_thr[2], i_bin[2] );
            pp_dering_BinIndex( p_block[3] - i_stride - 1, i_stride,
                                i_thr[3], i_bin[3] );


            /* 3: adaptive smoothing */
            /* since we begin at (8,8) QP can be different for each block */
            p_QP = &( p_QP_store[( y >> 4) * i_QP_stride + (x >> 4)] );

            pp_dering_Filter( p_block[0], i_stride,
                              i_bin[0], p_QP[0] );

            pp_dering_Filter( p_block[1], i_stride,
                              i_bin[1], p_QP[1] );

            pp_dering_Filter( p_block[2], i_stride,
                              i_bin[2], p_QP[i_QP_stride] );

            pp_dering_Filter( p_block[3], i_stride,
                              i_bin[3], p_QP[i_QP_stride+1] );

            p_block[0] += 8;
            p_block[1] += 8;
            p_block[2] += 8;
            p_block[3] += 8;
        }
    }

}

void E_( pp_dering_C )( uint8_t *p_plane,
                        int i_width, int i_height, int i_stride,
                        QT_STORE_T *p_QP_store, int i_QP_stride )
{
    int x, y;
    int i_max, i_min;
    int i_thr;
    uint32_t i_bin[10];

    uint8_t *p_block;


    for( y = 8; y < i_height-8; y += 8 )
    {

        p_block = p_plane + y * i_stride + 8;
        for( x = 8; x < i_width-8; x += 8 )
        {

            /* 1: Calculate threshold */
            /* Calculate max/min for each block */
            pp_dering_MinMax( p_block, i_stride,
                              &i_min, &i_max );
            /* Calculate thr*/
            i_thr = ( i_max + i_min + 1 )/2;

            /* 2: Index acquisition 10x10 */
            /* point on 10x10 in wich we have our 8x8 block */
            pp_dering_BinIndex( p_block - i_stride -1, i_stride,
                                i_thr,
                                i_bin );

            /* 3: adaptive smoothing */
            pp_dering_Filter( p_block, i_stride,
                              i_bin,
                              p_QP_store[(y>>5)*i_QP_stride+ (x>>5)]);
            p_block += 8;
        }
    }

}