macroblock.c

H.264 source codes

        if( i_ref >= 0 )
        {
            const int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
            int x4, y4;

            x264_macroblock_cache_ref( h, x8, y8, 2, 2, 0, i_ref );

            for( y4 = y8; y4 < y8+2; y4++ )
                for( x4 = x8; x4 < x8+2; x4++ )
                {
                    const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + x4 + y4 * 4 * h->mb.i_mb_stride ];
                    int mv_l0[2];
                    mv_l0[0] = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
                    mv_l0[1] = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
                    x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, mv_l0[0], mv_l0[1] );
                    x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1] );
                }
        }
        else
        {
            /* the colocated ref isn't in the current list0 */
            /* FIXME: we might still be able to use direct_8x8 on some partitions */
            return 0;
        }
    }

    return 1;
}

static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
{
    int ref[2];
    int mv[2][2];
    int i_list;
    int i8, i4;
    const int8_t *l1ref = &h->fref1[0]->ref[0][ h->mb.i_b8_xy ];
    const int16_t (*l1mv)[2] = (const int16_t (*)[2]) &h->fref1[0]->mv[0][ h->mb.i_b4_xy ];

    for( i_list=0; i_list<2; i_list++ )
    {
        int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
        int i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
        int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
        if( i_refc == -2 )
            i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];

        ref[i_list] = i_refa;
        if( ref[i_list] < 0 || ( i_refb < ref[i_list] && i_refb >= 0 ))
            ref[i_list] = i_refb;
        if( ref[i_list] < 0 || ( i_refc < ref[i_list] && i_refc >= 0 ))
            ref[i_list] = i_refc;
        if( ref[i_list] < 0 )
            ref[i_list] = -1;
    }

    if( ref[0] < 0 && ref[1] < 0 )
    {
        ref[0] = 
        ref[1] = 0;
        mv[0][0] = 
        mv[0][1] = 
        mv[1][0] = 
        mv[1][1] = 0;
    }
    else
    {
        for( i_list=0; i_list<2; i_list++ )
        {
            if( ref[i_list] >= 0 )
                x264_mb_predict_mv_16x16( h, i_list, ref[i_list], mv[i_list] );
            else
                mv[i_list][0] = mv[i_list][1] = 0;
        }
    }

    x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, ref[0] );
    x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, ref[1] );
    x264_macroblock_cache_mv(  h, 0, 0, 4, 4, 0, mv[0][0], mv[0][1] );
    x264_macroblock_cache_mv(  h, 0, 0, 4, 4, 1, mv[1][0], mv[1][1] );

    /* col_zero_flag */
    for( i8=0; i8<4; i8++ )
    {
        const int x8 = i8%2;
        const int y8 = i8/2;
        if( l1ref[ x8 + y8 * h->mb.i_b8_stride ] == 0 )
        {
            for( i4=0; i4<4; i4++ )
            {
                const int x4 = i4%2 + 2*x8;
                const int y4 = i4/2 + 2*y8;
                const int16_t *mvcol = l1mv[x4 + y4 * h->mb.i_b4_stride];
                if( abs( mvcol[0] ) <= 1 && abs( mvcol[1] ) <= 1 )
                {
                    if( ref[0] == 0 )
                        x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, 0, 0 );
                    if( ref[1] == 0 )
                        x264_macroblock_cache_mv( h, x4, y4, 1, 1, 1, 0, 0 );
                }
            }
        }
    }

    return 1;
}

int x264_mb_predict_mv_direct16x16( x264_t *h )
{
    int b_available;
    if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_NONE )
        return 0;
    else if( h->sh.b_direct_spatial_mv_pred )
        b_available = x264_mb_predict_mv_direct16x16_spatial( h );
    else
        b_available = x264_mb_predict_mv_direct16x16_temporal( h );

    /* cache ref & mv */
    if( b_available )
    {
        int i, l;
        for( l = 0; l < 2; l++ )
            for( i = 0; i < 4; i++ )
                h->mb.cache.direct_ref[l][i] = h->mb.cache.ref[l][x264_scan8[i*4]];
        memcpy(h->mb.cache.direct_mv, h->mb.cache.mv, sizeof(h->mb.cache.mv));
    }

    return b_available;
}

void x264_mb_load_mv_direct8x8( x264_t *h, int idx )
{
    const int x = 2*(idx%2);
    const int y = 2*(idx/2);
    int l;
    x264_macroblock_cache_ref( h, x, y, 2, 2, 0, h->mb.cache.direct_ref[0][idx] );
    x264_macroblock_cache_ref( h, x, y, 2, 2, 1, h->mb.cache.direct_ref[1][idx] );
    for( l = 0; l < 2; l++ )
    {
        *(uint64_t*)h->mb.cache.mv[l][x264_scan8[idx*4]] =
        *(uint64_t*)h->mb.cache.direct_mv[l][x264_scan8[idx*4]];
        *(uint64_t*)h->mb.cache.mv[l][x264_scan8[idx*4]+8] =
        *(uint64_t*)h->mb.cache.direct_mv[l][x264_scan8[idx*4]+8];
    }
}

/* This just improves encoder performance, it's not part of the spec */
void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int mvc[5][2], int *i_mvc )
{
    int16_t (*mvr)[2] = h->mb.mvr[i_list][i_ref];
    int i = 0;

    /* temporal */
    if( h->sh.i_type == SLICE_TYPE_B )
    {
        if( h->mb.cache.ref[i_list][x264_scan8[12]] == i_ref )
        {
            /* FIXME: use direct_mv to be clearer? */
            int16_t *mvp = h->mb.cache.mv[i_list][x264_scan8[12]];
            mvc[i][0] = mvp[0];
            mvc[i][1] = mvp[1];
            i++;
        }
    }

    /* spatial */
    if( h->mb.i_mb_x > 0 )
    {
        int i_mb_l = h->mb.i_mb_xy - 1;
        /* skip MBs didn't go through the whole search process, so mvr is undefined */
        if( !IS_SKIP( h->mb.type[i_mb_l] ) )
        {
            mvc[i][0] = mvr[i_mb_l][0];
            mvc[i][1] = mvr[i_mb_l][1];
            i++;
        }
    }
    if( h->mb.i_mb_y > 0 )
    {
        int i_mb_t = h->mb.i_mb_xy - h->mb.i_mb_stride;
        if( !IS_SKIP( h->mb.type[i_mb_t] ) )
        {
            mvc[i][0] = mvr[i_mb_t][0];
            mvc[i][1] = mvr[i_mb_t][1];
            i++;
        }

        if( h->mb.i_mb_x > 0 && !IS_SKIP( h->mb.type[i_mb_t - 1] ) )
        {
            mvc[i][0] = mvr[i_mb_t - 1][0];
            mvc[i][1] = mvr[i_mb_t - 1][1];
            i++;
        }
        if( h->mb.i_mb_x < h->mb.i_mb_stride - 1 && !IS_SKIP( h->mb.type[i_mb_t + 1] ) )
        {
            mvc[i][0] = mvr[i_mb_t + 1][0];
            mvc[i][1] = mvr[i_mb_t + 1][1];
            i++;
        }
    }
    *i_mvc = i;
}

static inline void x264_mb_mc_0xywh( x264_t *h, int x, int y, int width, int height )
{
    const int i8 = x264_scan8[0]+x+8*y;
    const int i_ref = h->mb.cache.ref[0][i8];
    const int mvx   = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int mvy   = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );

    h->mc.mc_luma( h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
                    &h->mb.pic.p_fdec[0][4*y * h->mb.pic.i_stride[0]+4*x],           h->mb.pic.i_stride[0],
                    mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );

    h->mc.mc_chroma( &h->mb.pic.p_fref[0][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
                      &h->mb.pic.p_fdec[1][2*y*h->mb.pic.i_stride[1]+2*x],           h->mb.pic.i_stride[1],
                      mvx, mvy, 2*width, 2*height );

    h->mc.mc_chroma( &h->mb.pic.p_fref[0][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
                      &h->mb.pic.p_fdec[2][2*y*h->mb.pic.i_stride[2]+2*x],           h->mb.pic.i_stride[2],
                      mvx, mvy, 2*width, 2*height );
}
static inline void x264_mb_mc_1xywh( x264_t *h, int x, int y, int width, int height )
{
    const int i8 = x264_scan8[0]+x+8*y;
    const int i_ref = h->mb.cache.ref[1][i8];
    const int mvx   = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int mvy   = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );

    h->mc.mc_luma( h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
                    &h->mb.pic.p_fdec[0][4*y *h->mb.pic.i_stride[0]+4*x],            h->mb.pic.i_stride[0],
                    mvx + 4*4*x, mvy + 4*4*y, 4*width, 4*height );

    h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
                      &h->mb.pic.p_fdec[1][2*y*h->mb.pic.i_stride[1]+2*x],           h->mb.pic.i_stride[1],
                      mvx, mvy, 2*width, 2*height );

    h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
                      &h->mb.pic.p_fdec[2][2*y*h->mb.pic.i_stride[2]+2*x],           h->mb.pic.i_stride[2],
                      mvx, mvy, 2*width, 2*height );
}

static inline void x264_mb_mc_01xywh( x264_t *h, int x, int y, int width, int height )
{
    const int i8 = x264_scan8[0]+x+8*y;

    const int i_ref1 = h->mb.cache.ref[1][i8];
    const int mvx1   = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] );
    const int mvy1   = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] );
    DECLARE_ALIGNED( uint8_t, tmp[16*16], 16 );
    int i_mode = x264_size2pixel[height][width];

    x264_mb_mc_0xywh( h, x, y, width, height );

    h->mc.mc_luma( h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
                    tmp, 16, mvx1 + 4*4*x, mvy1 + 4*4*y, 4*width, 4*height );

    if( h->param.analyse.b_weighted_bipred )
    {
        const int i_ref0 = h->mb.cache.ref[0][i8];
        const int weight = h->mb.bipred_weight[i_ref0][i_ref1];

        h->pixf.avg_weight[i_mode]( &h->mb.pic.p_fdec[0][4*y *h->mb.pic.i_stride[0]+4*x], h->mb.pic.i_stride[0], tmp, 16, weight );

        h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
                          tmp, 16, mvx1, mvy1, 2*width, 2*height );
        h->pixf.avg_weight[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1], tmp, 16, weight );

        h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
                          tmp, 16, mvx1, mvy1, 2*width, 2*height );
        h->pixf.avg_weight[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2], tmp, 16, weight );
    }
    else
    {
        h->pixf.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y *h->mb.pic.i_stride[0]+4*x], h->mb.pic.i_stride[0], tmp, 16 );

        h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][4][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1],
                          tmp, 16, mvx1, mvy1, 2*width, 2*height );
        h->pixf.avg[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*h->mb.pic.i_stride[1]+2*x], h->mb.pic.i_stride[1], tmp, 16 );

        h->mc.mc_chroma( &h->mb.pic.p_fref[1][i_ref1][5][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2],
                          tmp, 16, mvx1, mvy1, 2*width, 2*height );
        h->pixf.avg[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*h->mb.pic.i_stride[2]+2*x], h->mb.pic.i_stride[2], tmp, 16 );
    }
}

static void x264_mb_mc_direct8x8( x264_t *h, int x, int y )
{
    const int i8 = x264_scan8[0] + x + 8*y;

    /* FIXME: optimize based on current block size, not global settings? */
    if( h->sps->b_direct8x8_inference )
    {
        if( h->mb.cache.ref[0][i8] >= 0 )
            if( h->mb.cache.ref[1][i8] >= 0 )
                x264_mb_mc_01xywh( h, x, y, 2, 2 );
            else
                x264_mb_mc_0xywh( h, x, y, 2, 2 );
        else
            x264_mb_mc_1xywh( h, x, y, 2, 2 );
    }
    else
    {
        if( h->mb.cache.ref[0][i8] >= 0 )
        {
            if( h->mb.cache.ref[1][i8] >= 0 )
            {
                x264_mb_mc_01xywh( h, x+0, y+0, 1, 1 );
                x264_mb_mc_01xywh( h, x+1, y+0, 1, 1 );
                x264_mb_mc_01xywh( h, x+0, y+1, 1, 1 );
                x264_mb_mc_01xywh( h, x+1, y+1, 1, 1 );
            }
            else
            {
                x264_mb_mc_0xywh( h, x+0, y+0, 1, 1 );
                x264_mb_mc_0xywh( h, x+1, y+0, 1, 1 );
                x264_mb_mc_0xywh( h, x+0, y+1, 1, 1 );
                x264_mb_mc_0xywh( h, x+1, y+1, 1, 1 );
            }
        }
        else
        {
            x264_mb_mc_1xywh( h, x+0, y+0, 1, 1 );
            x264_mb_mc_1xywh( h, x+1, y+0, 1, 1 );
            x264_mb_mc_1xywh( h, x+0, y+1, 1, 1 );
            x264_mb_mc_1xywh( h, x+1, y+1, 1, 1 );
        }
    }
}

void x264_mb_mc( x264_t *h )
{
    if( h->mb.i_type == P_L0 )
    {
        if( h->mb.i_partition == D_16x16 )
        {
            x264_mb_mc_0xywh( h, 0, 0, 4, 4 );
        }
        else if( h->mb.i_partition == D_16x8 )
        {
            x264_mb_mc_0xywh( h, 0, 0, 4, 2 );
            x264_mb_mc_0xywh( h, 0, 2, 4, 2 );
        }
        else if( h->mb.i_partition == D_8x16 )
        {
            x264_mb_mc_0xywh( h, 0, 0, 2, 4 );
            x264_mb_mc_0xywh( h, 2, 0, 2, 4 );
        }
    }
    else if( h->mb.i_type == P_8x8 || h->mb.i_type == B_8x8 )
    {
        int i;
        for( i = 0; i < 4; i++ )
        {
            const int x = 2*(i%2);
            const int y = 2*(i/2);
            switch( h->mb.i_sub_partition[i] )
            {
                case D_L0_8x8:
                    x264_mb_mc_0xywh( h, x, y, 2, 2 );