static void FUNC(put_unweighted_pred)(uint8_t *_dst, ptrdiff_t _dststride,
                                      int16_t *src, ptrdiff_t srcstride,
                                      int width, int height)
{
    int x, y;
    pixel *dst          = (pixel *)_dst;
    ptrdiff_t dststride = _dststride / sizeof(pixel);

    int shift = 14 - BIT_DEPTH;
#if BIT_DEPTH < 14
    int offset = 1 << (shift - 1);
#else
    int offset = 0;
#endif
    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x++)
            dst[x] = av_clip_pixel((src[x] + offset) >> shift);
        dst += dststride;
        src += srcstride;
    }
}

static void FUNC(put_weighted_pred_avg)(uint8_t *_dst, ptrdiff_t _dststride,
                                        int16_t *src1, int16_t *src2,
                                        ptrdiff_t srcstride,
                                        int width, int height)
{
    int x, y;
    pixel *dst          = (pixel *)_dst;
    ptrdiff_t dststride = _dststride / sizeof(pixel);

    int shift = 14 + 1 - BIT_DEPTH;
#if BIT_DEPTH < 14
    int offset = 1 << (shift - 1);
#else
    int offset = 0;
#endif

    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x++)
            dst[x] = av_clip_pixel((src1[x] + src2[x] + offset) >> shift);
        dst  += dststride;
        src1 += srcstride;
        src2 += srcstride;
    }
}

static void FUNC(weighted_pred)(uint8_t denom, int16_t wlxFlag, int16_t olxFlag,
                                uint8_t *_dst, ptrdiff_t _dststride,
                                int16_t *src, ptrdiff_t srcstride,
                                int width, int height)
{
    int shift, wx, ox, x, y, offset;
    pixel *dst          = (pixel *)_dst;
    ptrdiff_t dststride = _dststride / sizeof(pixel);

    shift  = denom + 14 - BIT_DEPTH;
    if (shift >= 1) {
        offset = 1 << (shift - 1);
    } else {
        shift  = 0;
        offset = 0;
    }
    wx     = wlxFlag;
    ox     = olxFlag * (1 << (BIT_DEPTH - 8));

    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x++) {
            dst[x] = av_clip_pixel(((src[x] * wx + offset) >> shift) + ox);
        }
        dst += dststride;
        src += srcstride;
    }
}

static void FUNC(weighted_pred_avg)(uint8_t denom,
                                    int16_t wl0Flag, int16_t wl1Flag,
                                    int16_t ol0Flag, int16_t ol1Flag,
                                    uint8_t *_dst, ptrdiff_t _dststride,
                                    int16_t *src1, int16_t *src2,
                                    ptrdiff_t srcstride,
                                    int width, int height)
{
    int shift, log2Wd, w0, w1, o0, o1, x, y;
    pixel *dst = (pixel *)_dst;
    ptrdiff_t dststride = _dststride / sizeof(pixel);

    shift  = 14 - BIT_DEPTH;
    log2Wd = denom + shift;
    w0     = wl0Flag;
    w1     = wl1Flag;
    o0     = ol0Flag * (1 << (BIT_DEPTH - 8));
    o1     = ol1Flag * (1 << (BIT_DEPTH - 8));

    for (y = 0; y < height; y++) {
        for (x = 0; x < width; x++)
            dst[x] = av_clip_pixel((src1[x] * w0 + src2[x] * w1 +
                                    ((o0 + o1 + 1) << log2Wd)) >> (log2Wd + 1));
        dst  += dststride;
        src1 += srcstride;
        src2 += srcstride;
    }
}