moving 20010114-2.c to ieee directory

[Geoff Keating]

Franz Sirl <Franz.Sirl-kernel@lauterbach.com> writes:

> At 19:44 20.07.2001, Richard Henderson wrote:
> >On Fri, Jul 20, 2001 at 04:17:25PM +0200, Jan Hubicka wrote:
> > > Fails on i386, because it depends on the floats being really floats, not
> > > long doubles.  -ffloat-store solves the failure.  Would be OK to move the
> > > testcase to subdirectory ieee, where the flag is set?
> >
> >No.  Search the archives for discussion on this issue.
> 
> Uhm, really no? I thought the move was OK as long we replace the global 
> -ffloat-store with with a target specific one in the ieee.exp.
> How can I find out which targets default to IEEE conforming floats (like 
> PPC which passes all tests in ieee even without -ffloat-store) or which 
> targets have bad floating point behaviour like x86? I'll happily update 
> ieee.exp then.

Yes, this is OK so long as ieee.exp is modified so that -ffloat-store
is only passed on ix86.  [There are some cases in which some m68k
variants---like the original 68000 but not newer chips---need it, but
I think we can leave that for later.]

The original test case passes on powerpc with -ffloat-store, but fails
without it, but -ffloat-store is hiding the bug not fixing it.

Also, the similar flag for Alpha should be moved in to ieee.exp, for
consistency.

I submitted a patch to Rob Savoye, the dejagnu maintainer, to put the
flag for i386 in dejagnu, as is currently done for Alpha, but it was
rejected, and Rob said that such things would be better in individual
testsuites; he agreed that it followed from this that the flag for
Alpha should go in individual testsuites too.

-- 
- Geoffrey Keating <geoffk@geoffk.org>