[Bug 294719] lib/msun: Added fmaximum_mag_num

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D294719

--- Comment #9 from Robert Clausecker <[email protected]> ---
(In reply to Steve Kargl from comment #8)

The problem is that our signbit() macro resolves to a __signbit() function,
which gcc and clang do not recognise as being the signbit() macro.  This is
probably fixable by checking if the compiler has a built-in implementation =
and
then resolving signbit() to __builtin_signbit(), but let's do that another
time.

> I haven't figured out what Jesus is trying to do
> with the force_except.  Reading the description of
> fmaximum_num etc, I would try something like

The idea is that we force the compiler to perform an addition on the NaN
values.
If either is a signalling NaN, this raises FE_INVALID as usual, avoiding a
costly branch and sign-bit check.

We do similar things in other variants of the fmax/fmin family.  Whenever
return (x + y) is used to return a NaN, that too raises FE_INVALID if eithe=
r is
a signalling NaN.

> Need to some bit twiddling to do the equivalent.

I've checked the assembly of a the functions and the bitfield-union approac=
h we
use is surprisingly inefficient.  It's probably a good idea to try to avoid
doing this sort of stuff if we can help it.

--=20
You are receiving this mail because:
You are the assignee for the bug.=