This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Re: target/2273: [Alpha/Tru64 UNIX V5.1] SEGV in long double code
- To: Jeff Knaggs <jknaggs at cygnus dot com>
- Subject: Re: [PATCH] Re: target/2273: [Alpha/Tru64 UNIX V5.1] SEGV in long double code
- From: Rainer Orth <ro at TechFak dot Uni-Bielefeld dot DE>
- Date: Fri, 6 Apr 2001 00:52:12 +0200 (MEST)
- Cc: gcc-patches at gcc dot gnu dot org
- References: <20010405121427.11461@cse.cygnus.com>
Jeff Knaggs writes:
> 2001-04-05 Jeff Knaggs <jknaggs@redhat.com>
>
> * alpha/alpha.md (movtf_internal): Changed to
> reg_overlap_mentioned_p to catch early clobber
> of register used by memory reference.
excellent, thanks alot.
I've just re-bootstrapped on V4.0F (without 16-byte long double) and V5.1
to check if this fixes all 5.1 regressions due to the broken long double
support. We're pretty much there, with only three regressions from V4.0F
to V5.1:
* gcc.c-torture/execute/va-arg-5.c aborts in line 28:
if (va_arg (args, long double) != 3.141592L)
abort ();
* gcc.c-torture/execute/va-arg-6.c aborts in line 16:
if (va_arg (args, long double) != 3.14L)
abort ();
* gcc.c-torture/execute/ieee/fp-cmp-3.c aborts in line 27, even when -mieee
is added manually:
if (dnan != x)
x = 1.0;
else
abort ();
I haven't analyzed those failures yet.
Anyway, this patch allows us to close both PR target/2273 and fixes two of
the additional failures in PR c/2128.
Rainer
-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University
Email: ro@TechFak.Uni-Bielefeld.DE