This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Update x86_64 ULPs
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: OndÅej BÃlka <neleai at seznam dot cz>
- Cc: Andreas Jaeger <aj at suse dot com>, Markus Trippelsdorf <markus at trippelsdorf dot de>, <libc-alpha at sourceware dot org>
- Date: Thu, 2 May 2013 18:49:35 +0000
- Subject: Re: [PATCH] Update x86_64 ULPs
- References: <20130426055835 dot GA644 at x4> <517A2CF9 dot 20504 at suse dot com> <20130502160937 dot GA8564 at domone dot kolej dot mff dot cuni dot cz>
On Thu, 2 May 2013, Ondrej Bilka wrote:
> Could we compute ulp directly with mpfr or I am missing some roadblock?
The whole point of ulps baselines is to record the errors produced by the
actual glibc implementation, compared to an exact expected result (which
quite likely was computed with MPFR). Those baselines can, on x86 and
x86_64, depend on the processor implementation of the x87 instructions for
transcendental operations (so on whether an Intel or AMD processor is in
use, for example), as well as, on 32-bit x86, on the details of where the
compiler uses excess precision.
I've been thinking lately about how to better automate computation of the
*expected results* with MPFR and MPC, but that won't remove the need for
ulps values for the glibc implementations used on different platforms.
--
Joseph S. Myers
joseph@codesourcery.com