This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PR middle-end/30191 (was: [PATCH] Fix PR middle-end/28690, indexed load/store performance + reload bug)
- From: Peter Bergner <bergner at vnet dot ibm dot com>
- To: Ulrich Weigand <uweigand at de dot ibm dot com>
- Cc: Michael Matz <matz at suse dot de>, Andrew Pinski <pinskia at gmail dot com>, David Edelsohn <dje at makai dot watson dot ibm dot com>, gcc-patches at gcc dot gnu dot org
- Date: Thu, 14 Dec 2006 19:56:26 -0600
- Subject: Re: PR middle-end/30191 (was: [PATCH] Fix PR middle-end/28690, indexed load/store performance + reload bug)
- References: <200612132125.kBDLPpBY016275@d12av02.megacenter.de.ibm.com> <1166122367.7419.27.camel@otta>
On Thu, 2006-12-14 at 12:52 -0600, Peter Bergner wrote:
> On Wed, 2006-12-13 at 22:25 +0100, Ulrich Weigand wrote:
> > The following patch attempts to work both in situations where we should
> > re-recognize in order to change the insn code, and in situations where we
> > cannot re-recognize because the intermediate form of the insn is not
> > accepted by the target predicate.
> [snip]
> > Peter, could you try whether it indeed fixes your original problem?
>
> I bootstrapped and regtested this patch on powerpc64-linux with and
> without the indexed load/store patch that exposed this. They both
> passed with no errors/regressions. However, looking into the RTL
> dumps for the compiler with the indexed load/store patch, we no longer
> have the REG_UNUSED note attached to the PARALLEL. It might be due to
> me updating my sources. I'll rerun the bootstrap and regtest with an
> older version of mainline I know generated the REG_UNUSED note to make
> sure your change indeed handles that case.
Ok, using a slightly older version of mainline along with your patch
and the indexed load/store patch, I now see the PARALLEL insn with the
REG_UNUSED note and it compiles with no problem and eliminates the
unused/dead load portion of the PARALLEL. So it looks good!
Peter