[2/3] vcomp: Implement _vcomp_reduction_r4 and add tests.

[Henri Verbeet]

On 8 September 2016 at 08:21, Sebastian Lackner <sebastian at fds-team.de> wrote:
> Yes, there are really bool_and and bool_or reduction operations for float/double.
> I decided to explicitly write down the comparison with 0.0 in this case because
> it otherwise looks like a mistake. According to my tests all values != 0.0 are
> seen as true (without any small epsilon to avoid rounding errors... well).
>
I'd expect it to just look at the bit pattern, but just for laughs,
what about -0.0f?