motion_est.c

ffmpeg移植到symbian的全部源代码

    *mx_ptr = 16 * s->mb_x;
    *my_ptr = 16 * s->mb_y;
}

#define Z_THRESHOLD 256

#define CHECK_SAD_HALF_MV(suffix, x, y) \
{\
    d= s->dsp.pix_abs[size][(x?1:0)+(y?2:0)](NULL, pix, ptr+((x)>>1), stride, h);\
    d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
    COPY3_IF_LT(dminh, d, dx, x, dy, y)\
}

static inline int sad_hpel_motion_search(MpegEncContext * s,
                                  int *mx_ptr, int *my_ptr, int dmin,
                                  int src_index, int ref_index,
                                  int size, int h)
{
    MotionEstContext * const c= &s->me;
    const int penalty_factor= c->sub_penalty_factor;
    int mx, my, dminh;
    uint8_t *pix, *ptr;
    int stride= c->stride;
    const int flags= c->sub_flags;
    LOAD_COMMON

    assert(flags == 0);

    if(c->skip){
//    printf("S");
        *mx_ptr = 0;
        *my_ptr = 0;
        return dmin;
    }
//    printf("N");

    pix = c->src[src_index][0];

    mx = *mx_ptr;
    my = *my_ptr;
    ptr = c->ref[ref_index][0] + (my * stride) + mx;

    dminh = dmin;

    if (mx > xmin && mx < xmax &&
        my > ymin && my < ymax) {
        int dx=0, dy=0;
        int d, pen_x, pen_y;
        const int index= (my<<ME_MAP_SHIFT) + mx;
        const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
        const int l= score_map[(index- 1               )&(ME_MAP_SIZE-1)];
        const int r= score_map[(index+ 1               )&(ME_MAP_SIZE-1)];
        const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
        mx<<=1;
        my<<=1;


        pen_x= pred_x + mx;
        pen_y= pred_y + my;

        ptr-= stride;
        if(t<=b){
            CHECK_SAD_HALF_MV(y2 , 0, -1)
            if(l<=r){
                CHECK_SAD_HALF_MV(xy2, -1, -1)
                if(t+r<=b+l){
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
                    ptr+= stride;
                }else{
                    ptr+= stride;
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
                }
                CHECK_SAD_HALF_MV(x2 , -1,  0)
            }else{
                CHECK_SAD_HALF_MV(xy2, +1, -1)
                if(t+l<=b+r){
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
                    ptr+= stride;
                }else{
                    ptr+= stride;
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
                }
                CHECK_SAD_HALF_MV(x2 , +1,  0)
            }
        }else{
            if(l<=r){
                if(t+l<=b+r){
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
                    ptr+= stride;
                }else{
                    ptr+= stride;
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
                }
                CHECK_SAD_HALF_MV(x2 , -1,  0)
                CHECK_SAD_HALF_MV(xy2, -1, +1)
            }else{
                if(t+r<=b+l){
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
                    ptr+= stride;
                }else{
                    ptr+= stride;
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
                }
                CHECK_SAD_HALF_MV(x2 , +1,  0)
                CHECK_SAD_HALF_MV(xy2, +1, +1)
            }
            CHECK_SAD_HALF_MV(y2 ,  0, +1)
        }
        mx+=dx;
        my+=dy;

    }else{
        mx<<=1;
        my<<=1;
    }

    *mx_ptr = mx;
    *my_ptr = my;
    return dminh;
}

static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
{
    const int xy= s->mb_x + s->mb_y*s->mb_stride;

    s->p_mv_table[xy][0] = mx;
    s->p_mv_table[xy][1] = my;

    /* has already been set to the 4 MV if 4MV is done */
    if(mv4){
        int mot_xy= s->block_index[0];

        s->current_picture.motion_val[0][mot_xy  ][0]= mx;
        s->current_picture.motion_val[0][mot_xy  ][1]= my;
        s->current_picture.motion_val[0][mot_xy+1][0]= mx;
        s->current_picture.motion_val[0][mot_xy+1][1]= my;

        mot_xy += s->b8_stride;
        s->current_picture.motion_val[0][mot_xy  ][0]= mx;
        s->current_picture.motion_val[0][mot_xy  ][1]= my;
        s->current_picture.motion_val[0][mot_xy+1][0]= mx;
        s->current_picture.motion_val[0][mot_xy+1][1]= my;
    }
}

/**
 * get fullpel ME search limits.
 */
static inline void get_limits(MpegEncContext *s, int x, int y)
{
    MotionEstContext * const c= &s->me;
    int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
/*
    if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
    else                   c->range= 16;
*/
    if (s->unrestricted_mv) {
        c->xmin = - x - 16;
        c->ymin = - y - 16;
        c->xmax = - x + s->mb_width *16;
        c->ymax = - y + s->mb_height*16;
    } else if (s->out_format == FMT_H261){
        // Search range of H261 is different from other codec standards
        c->xmin = (x > 15) ? - 15 : 0;
        c->ymin = (y > 15) ? - 15 : 0;
        c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
        c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
    } else {
        c->xmin = - x;
        c->ymin = - y;
        c->xmax = - x + s->mb_width *16 - 16;
        c->ymax = - y + s->mb_height*16 - 16;
    }
    if(range){
        c->xmin = FFMAX(c->xmin,-range);
        c->xmax = FFMIN(c->xmax, range);
        c->ymin = FFMAX(c->ymin,-range);
        c->ymax = FFMIN(c->ymax, range);
    }
}

static inline void init_mv4_ref(MotionEstContext *c){
    const int stride= c->stride;

    c->ref[1][0] = c->ref[0][0] + 8;
    c->ref[2][0] = c->ref[0][0] + 8*stride;
    c->ref[3][0] = c->ref[2][0] + 8;
    c->src[1][0] = c->src[0][0] + 8;
    c->src[2][0] = c->src[0][0] + 8*stride;
    c->src[3][0] = c->src[2][0] + 8;
}

static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
{
    MotionEstContext * const c= &s->me;
    const int size= 1;
    const int h=8;
    int block;
    int P[10][2];
    int dmin_sum=0, mx4_sum=0, my4_sum=0;
    int same=1;
    const int stride= c->stride;
    uint8_t *mv_penalty= c->current_mv_penalty;

    init_mv4_ref(c);

    for(block=0; block<4; block++){
        int mx4, my4;
        int pred_x4, pred_y4;
        int dmin4;
        static const int off[4]= {2, 1, 1, -1};
        const int mot_stride = s->b8_stride;
        const int mot_xy = s->block_index[block];

        P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
        P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];

        if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);

        /* special case for first line */
        if (s->first_slice_line && block<2) {
            c->pred_x= pred_x4= P_LEFT[0];
            c->pred_y= pred_y4= P_LEFT[1];
        } else {
            P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][0];
            P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][1];
            P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
            P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
            if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
            if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
            if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
            if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);

            P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
            P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

            c->pred_x= pred_x4 = P_MEDIAN[0];
            c->pred_y= pred_y4 = P_MEDIAN[1];
        }
        P_MV1[0]= mx;
        P_MV1[1]= my;

        dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);

        dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);

        if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
            int dxy;
            const int offset= ((block&1) + (block>>1)*stride)*8;
            uint8_t *dest_y = c->scratchpad + offset;
            if(s->quarter_sample){
                uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
                dxy = ((my4 & 3) << 2) | (mx4 & 3);

                if(s->no_rounding)
                    s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y   , ref    , stride);
                else
                    s->dsp.put_qpel_pixels_tab       [1][dxy](dest_y   , ref    , stride);
            }else{
                uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
                dxy = ((my4 & 1) << 1) | (mx4 & 1);

                if(s->no_rounding)
                    s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y    , ref    , stride, h);
                else
                    s->dsp.put_pixels_tab       [1][dxy](dest_y    , ref    , stride, h);
            }
            dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
        }else
            dmin_sum+= dmin4;

        if(s->quarter_sample){
            mx4_sum+= mx4/2;
            my4_sum+= my4/2;
        }else{
            mx4_sum+= mx4;
            my4_sum+= my4;
        }

        s->current_picture.motion_val[0][ s->block_index[block] ][0]= mx4;
        s->current_picture.motion_val[0][ s->block_index[block] ][1]= my4;

        if(mx4 != mx || my4 != my) same=0;
    }

    if(same)
        return INT_MAX;

    if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
        dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
    }

    if(c->avctx->mb_cmp&FF_CMP_CHROMA){
        int dxy;
        int mx, my;
        int offset;

        mx= ff_h263_round_chroma(mx4_sum);
        my= ff_h263_round_chroma(my4_sum);
        dxy = ((my & 1) << 1) | (mx & 1);

        offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;

        if(s->no_rounding){
            s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);
            s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);
        }else{
            s->dsp.put_pixels_tab       [1][dxy](c->scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);
            s->dsp.put_pixels_tab       [1][dxy](c->scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);
        }

        dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad  , s->uvlinesize, 8);
        dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);
    }

    c->pred_x= mx;
    c->pred_y= my;

    switch(c->avctx->mb_cmp&0xFF){
    /*case FF_CMP_SSE:
        return dmin_sum+ 32*s->qscale*s->qscale;*/
    case FF_CMP_RD:
        return dmin_sum;
    default:
        return dmin_sum+ 11*c->mb_penalty_factor;
    }
}

static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
    MotionEstContext * const c= &s->me;

    c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
    c->src[1][0] = c->src[0][0] + s->linesize;
    if(c->flags & FLAG_CHROMA){
        c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
        c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
        c->src[1][1] = c->src[0][1] + s->uvlinesize;
        c->src[1][2] = c->src[0][2] + s->uvlinesize;
    }
}

static int interlaced_search(MpegEncContext *s, int ref_index,
                             int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
{
    MotionEstContext * const c= &s->me;