This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [COMMITTED] hppa: Regenerate libm-testulps.
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Carlos O'Donell <carlos at redhat dot com>
- Cc: GNU C Library <libc-alpha at sourceware dot org>
- Date: Mon, 3 Feb 2014 23:09:09 +0000
- Subject: Re: [COMMITTED] hppa: Regenerate libm-testulps.
- Authentication-results: sourceware.org; auth=none
- References: <52F00569 dot 9080003 at redhat dot com>
On Mon, 3 Feb 2014, Carlos O'Donell wrote:
> I've updated the hppa libm-test-ulps from scratch.
>
> There are still two failures in ctanh_downard with ULPs as
> hight as 5 and 6 and I do not consider those acceptable when
> x86-64 has lower. I'll look into those later though.
x86_64 has a different long double format; those 5 and 6 are the same as
other ldbl-128 architectures. I think of a rule of thumb as being that
ulps should be below 10. (Actually, except for IBM long double in
directed rounding modes, where we still need resolution on either putting
suitable versions of the libgcc functions in glibc or adding *_round /
*_ieee functions to libgcc that are enabled by suitable options, and
except for m68k where probably some long double functions need their
accuracy improved, all ulps are 7 or less.)
--
Joseph S. Myers
joseph@codesourcery.com