This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PR 14776
- From: Jan Hubicka <jh at suse dot cz>
- To: Jan Hubicka <jh at suse dot cz>
- Cc: Richard Henderson <rth at redhat dot com>, Michael Matz <matz at suse dot de>,gcc-patches at gcc dot gnu dot org, mark at codesourcery dot com
- Date: Thu, 1 Apr 2004 16:03:09 +0200
- Subject: Re: PR 14776
- References: <20040331100803.GP2539@kam.mff.cuni.cz> <Pine.LNX.4.58.0403311500460.18453@wotan.suse.de> <20040331140031.GZ1542@kam.mff.cuni.cz> <20040401003211.GB25777@redhat.com> <20040401073009.GU2539@kam.mff.cuni.cz>
> > On Wed, Mar 31, 2004 at 04:00:31PM +0200, Jan Hubicka wrote:
> > > * emit-rtl.c (try_split): Preserve REG_EQUIV/EQUAL notes in
> > > trivial cases.
> >
> > Eh. Maybe.
> >
> > In this particular case, our silliness is in splitting the insn
> > so early. With the following patch we produce a nice
> >
> > xorpd .LC0, %xmm0
>
> The idea has been to allow CSE/invariant motion of the constant operand
> as at least of K8 the 128bit loads are expensive (and we can use 64bit
> load here). Not sure what alternative works better, I can give it a
> SPEC run to see if it shows something.
I should've read the patch first. Of course splitting the move late is
not preserving any CSE. (i tought you was inventing the constant for
fabs late).
One problem is medium model that needs temporary integer register for
accessing constpool, however I am already having patch in testing that
adds small data area. Having that we should be safe until we start
poking around large model but I don't see real need to implement it.
Honza
>
> Honza