TODO

- FIXME: the conversion of vec<f16_t, _N> to integer vectors is broken
- more AVX512 code: v32u16, v32s16, v64s8, v64u8
- fixed_vec_lookup_table specializations for v4f32 and v8f32 for table length
  2? how about a template solution for all vector lengths and types?
- documentation!!!

- lower precision functions 8, 4 ulps?
  4 ulps allow 2 errornous bits: 51, 22 bits precision
  8 ulps allow 3 errornous bits  50, 20 bits precision
  what about 48 bit precision for double?


double
fmod(double x, double y)
{
    double i, f;
    if (y == 0.0) {
        errno = EDOM;
        return 0.0;
    }

    /* return f such that x = i*y + f for some integer i
       such that |f| < |y| and f has the same sign as x */

    i = floor(x/y);
    f = x - i*y;
    if ((x < 0.0) != (y < 0.0))
        f = f-y;
    return f;
}