x264_me_search_ref函数分析

void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, int *p_halfpel_thresh )

{
    const int bw = x264_pixel_size[m->i_pixel].w;
    const int bh = x264_pixel_size[m->i_pixel].h;
    const int i_pixel = m->i_pixel;
    const int stride = m->i_stride[0];
    int i_me_range = h->param.analyse.i_me_range;
    int bmx, bmy, bcost;
    int bpred_mx = 0, bpred_my = 0, bpred_cost = COST_MAX;
    int omx, omy, pmx, pmy;
    pixel *p_fenc = m->p_fenc[0];
    pixel *p_fref_w = m->p_fref_w;
    ALIGNED_ARRAY_16( pixel, pix,[16*16] );

    int costs[16];

    int mv_x_min = h->mb.mv_min_fpel[0];
    int mv_y_min = h->mb.mv_min_fpel[1];  
    int mv_x_max = h->mb.mv_max_fpel[0]; 
    int mv_y_max = h->mb.mv_max_fpel[1];  //以上是整像素点
    int mv_x_min_qpel = mv_x_min << 2;
    int mv_y_min_qpel = mv_y_min << 2;
    int mv_x_max_qpel = mv_x_max << 2;
    int mv_y_max_qpel = mv_y_max << 2; //左移两位,变成四分之一像素点
/* Special version of pack to allow shortcuts in CHECK_MVRANGE */
//使mx在32位中的高16位,my在低16位
#define pack16to32_mask2(mx,my) ((mx<<16)|(my&0x7FFF))
    uint32_t mv_min = pack16to32_mask2( -mv_x_min, -mv_y_min );
    uint32_t mv_max = pack16to32_mask2( mv_x_max, mv_y_max )|0x8000;
  //如果mx,my超过了边界mv_min或者mv_max , 在最高位即符号位会为1，取反后变为0，与0x80004000与变为0
#define CHECK_MVRANGE(mx,my) (!(((pack16to32_mask2(mx,my) + mv_min) | (mv_max - pack16to32_mask2(mx,my))) & 0x80004000))

    const uint16_t *p_cost_mvx = m->p_cost_mv - m->mvp[0];
    const uint16_t *p_cost_mvy = m->p_cost_mv - m->mvp[1];//之前计算的mvp为四分之一像素的

    uint32_t pmv;
    bmx = x264_clip3( m->mvp[0], mv_x_min_qpel, mv_x_max_qpel );
    bmy = x264_clip3( m->mvp[1], mv_y_min_qpel, mv_y_max_qpel );
    pmx = ( bmx + 2 ) >> 2;//将bmy,bmy 变为整数精度矢量
    pmy = ( bmy + 2 ) >> 2;
    bcost = COST_MAX;

    /* try extra predictors if provided */
    if( h->mb.i_subpel_refine >= 3 )//如果精度为1/4
    {
        pmv = pack16to32_mask(bmx,bmy);
        if( i_mvc ) //计算mx,my的cost代价是否比bpred_cost小，如果小，就bpred_mx=mx ,bpred_my=my
            COST_MV_HPEL( bmx, bmy );//此函数先找到1/4像素的mv对应的位置，然后算satd 
        for( int i = 0; i < i_mvc; i++ )//对每一个mvc 计算代价
        {
            if( M32( mvc[i] ) && (pmv != M32( mvc[i] )) )
            {