This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [patch] Remove sysdeps/ieee754/ldbl-128/bits/huge_vall.h
On Thu, 27 Sep 2012, Carlos O'Donell wrote:
> I'm always kinda paranoid so I'd be happier if you also did a build test
> that showed that x86 or x86_64 didn't break. I don't see how they would
> break but I also didn't look deeply at the implies ordering for header
> inclusion.
x86 and x86_64 don't use ldbl-128, so this change seems completely
irrelevant to them.
(libquadmath in GCC uses the ldbl-128 code to provide functions for
__float128 for those architectures, but that's separately updated from
glibc by Jakub. If in future we wish to implement Part 3 (not yet
written) of the in-progress ISO Technical Specification for C bindings to
IEEE 754-2008 (Floating-point extensions for C - Part 3: Interchange and
extended types) in glibc, then it may make sense to subsume the
libquadmath functionality directly in glibc's libm in some way, with the
ldbl-128 code used to provide *f128 functions using _Float128, not long
double, as the type in question. That is, the functionality of ldbl-128
as providing functions for IEEE binary128 would need to be separate from
the functionality as providing the "long double" functions, likely meaning
a split into two separate sysdeps directories, one for "_Float128 is
supported" and one for "long double is _Float128" - with the further
complication that _Float64x (a type with greater range and precision than
_Float64, where supported) might be some form of ldbl-96 on some platforms
supporting _Float128, but _Float128 on other platforms. But the header
would need to define HUGE_VALF128 for those bindings; HUGE_VALL would
still be for the existing long double type.)
--
Joseph S. Myers
joseph@codesourcery.com