This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [WIP] Fixing ulp near zero.
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Rich Felker <dalias at aerifal dot cx>
- Cc: Carlos O'Donell <carlos at redhat dot com>, GNU C Library <libc-alpha at sourceware dot org>, Andreas Schwab <schwab at suse dot de>
- Date: Thu, 25 Apr 2013 21:27:24 +0000
- Subject: Re: [WIP] Fixing ulp near zero.
- References: <5178AC1D dot 90001 at redhat dot com> <Pine dot LNX dot 4 dot 64 dot 1304251455170 dot 8948 at digraph dot polyomino dot org dot uk> <517962E0 dot 4030108 at redhat dot com> <20130425204928 dot GO20323 at brightrain dot aerifal dot cx>
On Thu, 25 Apr 2013, Rich Felker wrote:
> On Thu, Apr 25, 2013 at 01:07:44PM -0400, Carlos O'Donell wrote:
> > Is it a bug that nextafter behaves as-if the mantissa were fixed
> > precision on IBM long double?
>
> I would say it's a bug that IBM long double is supported at all. Archs
> using 128-bit long double should be fixed, like ppc was, to use IEEE
> quad.
I have no great liking for IBM long double - its non-IEEE nature often
requires special cases in libm code - but the ABI for Power GNU/Linux is
what it is, and for whatever reason when 128-bit long double support was
added to the ABI, it was added as IBM long double rather than as IEEE
binary128 which was the type in the 1995 ABI for PowerPC Solaris on which
the GNU/Linux ABI was distantly based.
--
Joseph S. Myers
joseph@codesourcery.com