Re: [PATCH v3 0/1] MR3458: ddraw/tests: Fix overflow bugs in compare_float.

[Zebediah Figura (＠zfigura)]

> I'm going with a solution that
> - uses `compare_uint`,
> - ensures the things that should *definitively* return false *will* return false (e.g. 2 vs -2), and
> - other results stay the same as before (in the 0.0/-0.0/subnormal range), even if of dubious value.

I think Henri is right, though; we don't want positive and negative denormals to compare as equal.