comparison vc1.c @ 3430:d0e85690841d libavcodec

Improve chroma MC: correct case for FASTUVMC=1, use slower but correct /2, and always use halfpel MC. Not quite there yet but a bit better picture on some files.

3429:fb2a69de8ded

    uvmy = (my + ((my & 3) == 3)) >> 1;
    srcY = s->last_picture.data[0];
    srcU = s->last_picture.data[1];
    srcV = s->last_picture.data[2];

    if(v->fastuvmc) { // XXX: 8.3.5.4.5 specifies something different
        uvmx = (uvmx + 1) & ~1;
        uvmy = (uvmy + 1) & ~1;
    }

    src_x = s->mb_x * 16 + (mx >> 2);
    src_y = s->mb_y * 16 + (my >> 2);
    uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
    uvsrc_y = s->mb_y * 8 + (uvmy >> 2);

                src2 += s->uvlinesize;
            }
        }
    }

    if(!s->quarter_sample) { // hpel mc
        mx >>= 1;
        my >>= 1;
        dxy = ((my & 1) << 1) | (mx & 1);
        uvdxy = 0;

        dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
    } else {
        dxy = ((my & 3) << 2) | (mx & 3);
        uvdxy = ((uvmy & 1) << 1) | (uvmx & 1);

        dsp->put_no_rnd_qpel_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize);
    }
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize, 8);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize, 8);
//    dsp->put_mspel_pixels_tab[uvdxy](s->dest[1], srcU, s->uvlinesize);
//    dsp->put_mspel_pixels_tab[uvdxy](s->dest[2], srcV, s->uvlinesize);
}

}

static inline int median4(int a, int b, int c, int d)
{
    if(a < b) {
        if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) >> 1;
        else      return (FFMIN(b, c) + FFMAX(a, d)) >> 1;
    } else {
        if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) >> 1;
        else      return (FFMIN(a, c) + FFMAX(b, d)) >> 1;
    }
}


/** Do motion compensation for 4-MV macroblock - both chroma blocks

        }
    } else if(count[idx] == 2) {
        int t1 = 0, t2 = 0;
        for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
        for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
        tx = (mvx[t1] + mvx[t2]) >> 1;
        ty = (mvy[t1] + mvy[t2]) >> 1;
    } else
        return; //no need to do MC for inter blocks

    uvmx = (tx + ((tx&3) == 3)) >> 1;
    uvmy = (ty + ((ty&3) == 3)) >> 1;

                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        srcU = s->edge_emu_buffer;
        srcV = s->edge_emu_buffer + 16;
    }

    if(!s->quarter_sample) // hpel mc
        uvdxy = 0;
    else
        uvdxy = ((uvmy & 1) << 1) | (uvmx & 1);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize, 8);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize, 8);
}

/**

3430:d0e85690841d

    uvmy = (my + ((my & 3) == 3)) >> 1;
    srcY = s->last_picture.data[0];
    srcU = s->last_picture.data[1];
    srcV = s->last_picture.data[2];

    src_x = s->mb_x * 16 + (mx >> 2);
    src_y = s->mb_y * 16 + (my >> 2);
    uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
    uvsrc_y = s->mb_y * 8 + (uvmy >> 2);

                src2 += s->uvlinesize;
            }
        }
    }

    if(v->fastuvmc) {
        uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
        uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
    }

    if(!s->quarter_sample) { // hpel mc
        mx >>= 1;
        my >>= 1;
        dxy = ((my & 1) << 1) | (mx & 1);

        dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
    } else {
        dxy = ((my & 3) << 2) | (mx & 3);

        dsp->put_no_rnd_qpel_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize);
    }
    uvmx >>= 1;
    uvmy >>= 1;
    uvdxy = ((uvmy & 1) << 1) | (uvmx & 1);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize, 8);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize, 8);
//    dsp->put_mspel_pixels_tab[uvdxy](s->dest[1], srcU, s->uvlinesize);
//    dsp->put_mspel_pixels_tab[uvdxy](s->dest[2], srcV, s->uvlinesize);
}

}

static inline int median4(int a, int b, int c, int d)
{
    if(a < b) {
        if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
        else      return (FFMIN(b, c) + FFMAX(a, d)) / 2;
    } else {
        if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
        else      return (FFMIN(a, c) + FFMAX(b, d)) / 2;
    }
}


/** Do motion compensation for 4-MV macroblock - both chroma blocks

        }
    } else if(count[idx] == 2) {
        int t1 = 0, t2 = 0;
        for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
        for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
        tx = (mvx[t1] + mvx[t2]) / 2;
        ty = (mvy[t1] + mvy[t2]) / 2;
    } else
        return; //no need to do MC for inter blocks

    uvmx = (tx + ((tx&3) == 3)) >> 1;
    uvmy = (ty + ((ty&3) == 3)) >> 1;

                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        srcU = s->edge_emu_buffer;
        srcV = s->edge_emu_buffer + 16;
    }

    if(v->fastuvmc) {
        uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
        uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
    }

    uvmx >>= 1;
    uvmy >>= 1;
    uvdxy = ((uvmy & 1) << 1) | (uvmx & 1);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize, 8);
    dsp->put_no_rnd_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize, 8);
}

/**