On Thu, 2016-06-30 at 17:55 +0000, Joseph Myers wrote:
> On Thu, 30 Jun 2016, Bill Schmidt wrote:
>
> > +# Return 1 if the target supports __float128 at run time,
> > +# 0 otherwise.
> > +
> > +proc check_effective_target___float128_runnable { } {
>
> I'd think you should have an effective-target for this that's shared
> between _Float128 and __float128, possibly one that gets anded with the
> one for the relevant type being available at all.
Well, as is it can be used for either _Float128 or __float128, so I
suppose we could just change the name.
check_effective_target_quadfloat_runnable?
To provide something minimal to and with the effective-target __float128
check would get the somewhat silly:
# Return 1 if the target supports any special run-time requirements
# for __float128 or _Float128,
# 0 otherwise.
proc check_effective_target_base_quadfloat_support { } {
if { [istarget powerpc*-*-*] } {
return [check_vsx_hw_available]
}
return 1
}
so I think what I have (with a new name) is probably better.
> > +# Return 1 if the *-*-*gnu* target supports __float128 at run time,
> > +# 0 otherwise.
>
> *-*-*gnu* here is actually a proxy for feenableexcept trap enablement
> being available; if you're adding an effective-target, I'd think it should
> be one for feenableexcept with an actual test of function availability,
> which gets anded with the others.
>
Hm, ok. Perhaps for now I will just ditch changing this test and let
that be a follow-on opportunity for me or someone else when we aren't
under quite such a time crunch.
Thanks,
Bill
