This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [rfc] PR tree-optimization/52633 - ICE due to vectorizer pattern detection collision
- From: "Ulrich Weigand" <uweigand at de dot ibm dot com>
- To: jakub at redhat dot com
- Cc: rguenther at suse dot de (Richard Guenther), gcc-patches at sourceware dot org
- Date: Wed, 25 Apr 2012 13:44:27 +0200 (CEST)
- Subject: Re: [rfc] PR tree-optimization/52633 - ICE due to vectorizer pattern detection collision
Jakub Jelinek wrote:
> On Wed, Apr 25, 2012 at 10:29:36AM +0200, Richard Guenther wrote:
> > > Does this look reasonable? Any comments or suggestions appreciated!
> >
> > Yes, getting rid of this fragile interaction by doing more work in
> > vect_recog_widen_shift_pattern sounds like the correct thing to do.
>
> Or give up when seeing already pattern recognized stmts when detecting
> different pattern, unless the current pattern recognizer is prepared to
> handle them (and in that case tweak everything as necessary).
>
> E.g. several pattern recognizers already start with
> if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
> return NULL;
Yes, we should do that, and it'll fix the ICE for sure, but -as I said
in the original mail- doing *only* this will cause regressions in some
cases because we no longer get the combined fix-over-widening-*and*-
use-widening-shift effect ...
Bye,
Ulrich
--
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE
Ulrich.Weigand@de.ibm.com