#define LHFILTER(src)    ((((src)[0]+(src)[1])*19 >> 1)-((src)[-1]+(src)[2  ])*2+(((src)[-2  ]+(src)[3  ])>>1)+8>>4)
#define LVFILTER(src, i) ((((src)[0]+(src)[i])*19 >> 1)-((src)[-i]+(src)[2*i])*2+(((src)[-2*i]+(src)[3*i])>>1)+8>>4)

static void bink2f_y_mc(Bink2Context *c, int x, int y,
                        uint8_t *dst, int stride,
                        uint8_t *src, int sstride,
                        int width, int height,
                        int mv_x, int mv_y, int mode)
{
    uint8_t *msrc;

    msrc = src + mv_x + mv_y * sstride;

    ff_dlog(NULL, "%d: %dx%d | %dx%d\n", mode, x, y, mv_x, mv_y);
    if (mode == 0) {
        copy_block16(dst, msrc, stride, sstride, 16);
    } else if (mode == 1) {
        for (int j = 0; j < 16; j++) {
            for (int i = 0; i < 16; i++)
                dst[i] = av_clip_uint8(LHFILTER(msrc + i));
            dst  += stride;
            msrc += sstride;
        }
    } else if (mode == 2) {
        for (int j = 0; j < 16; j++) {
            for (int i = 0; i < 16; i++)
                dst[i*stride] = av_clip_uint8(LVFILTER(msrc + i*sstride, sstride));
            dst  += 1;
            msrc += 1;
        }
    } else if (mode == 3) {
        uint8_t temph[21*21] = { 0 };

        for (int j = 0; j < 16; j++) {
            for (int i = 0; i < 21; i++)
                temph[j*21+i+42] = av_clip_uint8(LHFILTER(msrc + i));
            msrc += sstride;
        }

        for (int j = 0; j < 16; j++) {
            for (int i = 0; i < 16; i++)
                dst[i*stride] = av_clip_uint8(LVFILTER(temph + i*21+42, 21));
            dst  += 1;
        }
    }
}