me.c

mpeg4 video codec mpeg4 video codec

            const int dh = x264_pixel_size[i_pixel].h * stride;
            static uint8_t zero[16*16] = {0,};
            const int enc_dc = h->pixf.sad[i_pixel]( m->p_fenc[0], stride, zero, 16 );
            const uint16_t *integral_base = &m->integral[ -1 - 1*stride ];

            if( h->pixf.sad_pde[i_pixel] )
            {
                for( my = min_y; my <= max_y; my++ )
                    for( mx = min_x; mx <= max_x; mx++ )
                    {
                        const uint16_t *integral = &integral_base[ mx + my * stride ];
                        const uint16_t ref_dc = integral[  0 ] + integral[ dh + dw ]
                                              - integral[ dw ] - integral[ dh ];
                        const int bsad = bcost - p_cost_mvx[ (mx)<<2 ] - p_cost_mvy[ (my)<<2 ];
                        if( abs( ref_dc - enc_dc ) < bsad )
                            COST_MV_PDE( mx, my );
                    }
            }
            else
            {
                for( my = min_y; my <= max_y; my++ )
                    for( mx = min_x; mx <= max_x; mx++ )
                    {
                        const uint16_t *integral = &integral_base[ mx + my * stride ];
                        const uint16_t ref_dc = integral[  0 ] + integral[ dh + dw ]
                                              - integral[ dw ] - integral[ dh ];
                        const int bsad = bcost - p_cost_mvx[ (mx)<<2 ] - p_cost_mvy[ (my)<<2 ];
                        if( abs( ref_dc - enc_dc ) < bsad )
                            COST_MV( mx, my );
                    }
            }
#endif
        }
        break;
    }

    /* -> qpel mv */
    m->mv[0] = bmx << 2;
    m->mv[1] = bmy << 2;

    /* compute the real cost */
    m->cost_mv = p_cost_mvx[ m->mv[0] ] + p_cost_mvy[ m->mv[1] ];
    m->cost = bcost;
    if( bmx == pmx && bmy == pmy )
        m->cost += m->cost_mv;
    
    /* subpel refine */
    if( h->mb.i_subpel_refine >= 2 )
    {
        int hpel = subpel_iterations[h->mb.i_subpel_refine][2];
        int qpel = subpel_iterations[h->mb.i_subpel_refine][3];
        refine_subpel( h, m, hpel, qpel, p_halfpel_thresh, 0 );
    }
}
#undef COST_MV

void x264_me_refine_qpel( x264_t *h, x264_me_t *m )
{
    int hpel = subpel_iterations[h->mb.i_subpel_refine][0];
    int qpel = subpel_iterations[h->mb.i_subpel_refine][1];

    if( m->i_pixel <= PIXEL_8x8 && h->sh.i_type == SLICE_TYPE_P )
        m->cost -= m->i_ref_cost;
	
    refine_subpel( h, m, hpel, qpel, NULL, 1 );
}

#define COST_MV_SAD( mx, my, dir ) \
if( b_refine_qpel || (dir^1) != odir ) \
{ \
    int stride = 16; \
    uint8_t *src = h->mc.get_ref( m->p_fref, m->i_stride[0], pix, &stride, mx, my, bw, bh ); \
    int cost = h->pixf.sad[i_pixel]( m->p_fenc[0], m->i_stride[0], src, stride ) \
             + p_cost_mvx[ mx ] + p_cost_mvy[ my ]; \
    if( cost < bcost ) \
    {                  \
        bcost = cost;  \
        bmx = mx;      \
        bmy = my;      \
        bdir = dir;    \
    } \
}

#define COST_MV_SATD( mx, my, dir ) \
if( b_refine_qpel || (dir^1) != odir ) \
{ \
    int stride = 16; \
    uint8_t *src = h->mc.get_ref( m->p_fref, m->i_stride[0], pix, &stride, mx, my, bw, bh ); \
    int cost = h->pixf.mbcmp[i_pixel]( m->p_fenc[0], m->i_stride[0], src, stride ) \
             + p_cost_mvx[ mx ] + p_cost_mvy[ my ]; \
    if( b_chroma_me && cost < bcost ) \
    { \
        h->mc.mc_chroma( m->p_fref[4], m->i_stride[1], pix, 8, mx, my, bw/2, bh/2 ); \
        cost += h->pixf.mbcmp[i_pixel+3]( m->p_fenc[1], m->i_stride[1], pix, 8 ); \
        if( cost < bcost ) \
        { \
            h->mc.mc_chroma( m->p_fref[5], m->i_stride[1], pix, 8, mx, my, bw/2, bh/2 ); \
            cost += h->pixf.mbcmp[i_pixel+3]( m->p_fenc[2], m->i_stride[1], pix, 8 ); \
        } \
    } \
    if( cost < bcost ) \
    {                  \
        bcost = cost;  \
        bmx = mx;      \
        bmy = my;      \
        bdir = dir;    \
    } \
}

static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_iters, int *p_halfpel_thresh, int b_refine_qpel )
{
    const int bw = x264_pixel_size[m->i_pixel].w;
    const int bh = x264_pixel_size[m->i_pixel].h;
    const int16_t *p_cost_mvx = m->p_cost_mv - m->mvp[0];
    const int16_t *p_cost_mvy = m->p_cost_mv - m->mvp[1];
    const int i_pixel = m->i_pixel;
    const int b_chroma_me = h->mb.b_chroma_me && i_pixel <= PIXEL_8x8;

    DECLARE_ALIGNED( uint8_t, pix[16*16], 16 );
    int omx, omy;
    int i;

    int bmx = m->mv[0];
    int bmy = m->mv[1];
    int bcost = m->cost;
    int odir = -1, bdir;


    /* try the subpel component of the predicted mv */
    if( hpel_iters )
    {
        int mx = x264_clip3( m->mvp[0], h->mb.mv_min[0], h->mb.mv_max[0] );
        int my = x264_clip3( m->mvp[1], h->mb.mv_min[1], h->mb.mv_max[1] );
        if( mx != bmx || my != bmy )
            COST_MV_SAD( mx, my, -1 );
    }
    
    /* hpel search */
    bdir = -1;
    for( i = hpel_iters; i > 0; i-- )
    {
        odir = bdir;
        omx = bmx;
        omy = bmy;
        COST_MV_SAD( omx, omy - 2, 0 );
        COST_MV_SAD( omx, omy + 2, 1 );
        COST_MV_SAD( omx - 2, omy, 2 );
        COST_MV_SAD( omx + 2, omy, 3 );
        if( bmx == omx && bmy == omy )
            break;
    }
    
    if( !b_refine_qpel )
    {
        bcost = COST_MAX;
        COST_MV_SATD( bmx, bmy, -1 );
    }
    
    /* early termination when examining multiple reference frames */
    if( p_halfpel_thresh )
    {
        if( (bcost*7)>>3 > *p_halfpel_thresh )
        {
            m->cost = bcost;
            m->mv[0] = bmx;
            m->mv[1] = bmy;
            // don't need cost_mv
            return;
        }
        else if( bcost < *p_halfpel_thresh )
            *p_halfpel_thresh = bcost;
    }

    /* qpel search */
    bdir = -1;
    for( i = qpel_iters; i > 0; i-- )
    {
        odir = bdir;
        omx = bmx;
        omy = bmy;
        COST_MV_SATD( omx, omy - 1, 0 );
        COST_MV_SATD( omx, omy + 1, 1 );
        COST_MV_SATD( omx - 1, omy, 2 );
        COST_MV_SATD( omx + 1, omy, 3 );
        if( bmx == omx && bmy == omy )
            break;
    }

    m->cost = bcost;
    m->mv[0] = bmx;
    m->mv[1] = bmy;
    m->cost_mv = p_cost_mvx[ bmx ] + p_cost_mvy[ bmy ];
}

#define BIME_CACHE( dx, dy ) \
{ \
    int i = 4 + 3*dx + dy; \
    h->mc.mc_luma( m0->p_fref, m0->i_stride[0], pix0[i], bw, om0x+dx, om0y+dy, bw, bh ); \
    h->mc.mc_luma( m1->p_fref, m1->i_stride[0], pix1[i], bw, om1x+dx, om1y+dy, bw, bh ); \
}

#define BIME_CACHE2(a,b) \
    BIME_CACHE(a,b) \
    BIME_CACHE(-(a),-(b))

#define COST_BIMV_SATD( m0x, m0y, m1x, m1y ) \
if( pass == 0 || !visited[(m0x)&7][(m0y)&7][(m1x)&7][(m1y)&7] ) \
{ \
    int cost; \
    int i0 = 4 + 3*(m0x-om0x) + (m0y-om0y); \
    int i1 = 4 + 3*(m1x-om1x) + (m1y-om1y); \
    visited[(m0x)&7][(m0y)&7][(m1x)&7][(m1y)&7] = 1; \
    memcpy( pix, pix0[i0], bs ); \
    if( i_weight == 32 ) \
        h->mc.avg[i_pixel]( pix, bw, pix1[i1], bw ); \
    else \
        h->mc.avg_weight[i_pixel]( pix, bw, pix1[i1], bw, i_weight ); \
    cost = h->pixf.mbcmp[i_pixel]( m0->p_fenc[0], m0->i_stride[0], pix, bw ) \
         + p_cost_m0x[ m0x ] + p_cost_m0y[ m0y ] \
         + p_cost_m1x[ m1x ] + p_cost_m1y[ m1y ]; \
    if( cost < bcost ) \
    {                  \
        bcost = cost;  \
        bm0x = m0x;    \
        bm0y = m0y;    \
        bm1x = m1x;    \
        bm1y = m1y;    \
    } \
}

#define CHECK_BIDIR(a,b,c,d) \
    COST_BIMV_SATD(om0x+a, om0y+b, om1x+c, om1y+d)

#define CHECK_BIDIR2(a,b,c,d) \
    CHECK_BIDIR(a,b,c,d) \
    CHECK_BIDIR(-(a),-(b),-(c),-(d))

#define CHECK_BIDIR8(a,b,c,d) \
    CHECK_BIDIR2(a,b,c,d) \
    CHECK_BIDIR2(b,c,d,a) \
    CHECK_BIDIR2(c,d,a,b) \
    CHECK_BIDIR2(d,a,b,c)

int x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight )
{
    const int i_pixel = m0->i_pixel;
    const int bw = x264_pixel_size[i_pixel].w;
    const int bh = x264_pixel_size[i_pixel].h;
    const int bs = bw*bh;
    const int16_t *p_cost_m0x = m0->p_cost_mv - x264_clip3( m0->mvp[0], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int16_t *p_cost_m0y = m0->p_cost_mv - x264_clip3( m0->mvp[1], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int16_t *p_cost_m1x = m1->p_cost_mv - x264_clip3( m1->mvp[0], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int16_t *p_cost_m1y = m1->p_cost_mv - x264_clip3( m1->mvp[1], h->mb.mv_min[0], h->mb.mv_max[0] );
    DECLARE_ALIGNED( uint8_t, pix0[9][16*16], 16 );
    DECLARE_ALIGNED( uint8_t, pix1[9][16*16], 16 );
    DECLARE_ALIGNED( uint8_t, pix[16*16], 16 );
    int bm0x = m0->mv[0], om0x = bm0x;
    int bm0y = m0->mv[1], om0y = bm0y;
    int bm1x = m1->mv[0], om1x = bm1x;
    int bm1y = m1->mv[1], om1y = bm1y;
    int bcost = COST_MAX;
    int pass = 0;
    uint8_t visited[8][8][8][8];
    memset( visited, 0, sizeof(visited) );

    BIME_CACHE( 0, 0 );
    CHECK_BIDIR( 0, 0, 0, 0 );

    for( pass = 0; pass < 8; pass++ )
    {
        /* check all mv pairs that differ in at most 2 components from the current mvs. */
        /* doesn't do chroma ME. this probably doesn't matter, as the gains
         * from bidir ME are the same with and without chroma ME. */

        BIME_CACHE2( 1, 0 );
        BIME_CACHE2( 0, 1 );
        BIME_CACHE2( 1, 1 );
        BIME_CACHE2( 1,-1 );

        CHECK_BIDIR8( 0, 0, 0, 1 );
        CHECK_BIDIR8( 0, 0, 1, 1 );
        CHECK_BIDIR2( 0, 1, 0, 1 );
        CHECK_BIDIR2( 1, 0, 1, 0 );
        CHECK_BIDIR8( 0, 0,-1, 1 );
        CHECK_BIDIR2( 0,-1, 0, 1 );
        CHECK_BIDIR2(-1, 0, 1, 0 );

        if( om0x == bm0x && om0y == bm0y && om1x == bm1x && om1y == bm1y )
            break;

        om0x = bm0x;
        om0y = bm0y;
        om1x = bm1x;
        om1y = bm1y;
        BIME_CACHE( 0, 0 );
    }

    m0->mv[0] = bm0x;
    m0->mv[1] = bm0y;
    m1->mv[0] = bm1x;
    m1->mv[1] = bm1y;
    return bcost;
}