This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: rs6000 long double extenddftf patch
- From: Alan Modra <amodra at bigpond dot net dot au>
- To: Geoffrey Keating <gkeating at apple dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Fri, 9 Jan 2004 10:39:52 +1030
- Subject: Re: rs6000 long double extenddftf patch
- References: <20040108224155.13CF6CC0E4A@geoffk4.apple.com>
On Thu, Jan 08, 2004 at 02:41:55PM -0800, Geoffrey Keating wrote:
> darwin-ldouble.c. Unfortunately, real.c and darwin-ldouble.c are also
> inconsistent with each other, so I am not sure what to do. For
> instance, real.c represents -Inf as (-Inf, -Inf); darwin-ldouble.c
> represents it as (-Inf, -0.0); and extenddftf now represents it as
> (-Inf, +0.0). One of those is probably right, but which one?
I believe (-Inf, -0.0) is the canonical one. The general rule in
the extended double "a + aa", is that of all possible choices of
doubles for "a" and "aa", that "aa" should be the double nearest 0.0.
In the case where "aa" == 0.0, it should have the same sign as "a".
--
Alan Modra
IBM OzLabs - Linux Technology Centre