This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: [fortran patch] Don't use TREE_LISTs for storing arguments to intrinsic functions
- From: "Michael Meissner" <michael dot meissner at amd dot com>
- To: "Lee Millward" <lee dot millward at gmail dot com>
- Cc: fortran at gcc dot gnu dot org, gcc-patches at gcc dot gnu dot org
- Date: Fri, 6 Jul 2007 11:36:42 -0400
- Subject: Re: [fortran patch] Don't use TREE_LISTs for storing arguments to intrinsic functions
- References: <9784d3ab0705081537jcb70109g962f91caf5c6ca97@mail.gmail.com> <46411641.8070206@codesourcery.com> <9784d3ab0705101534x2c7849eeh93e03afc7bda71d7@mail.gmail.com> <4643A623.6000706@codesourcery.com> <9784d3ab0705271210l5c836055wf01b439ff2a2ac23@mail.gmail.com> <4663B7A0.8050802@codesourcery.com> <4.3.1.2.20070604112111.02f91f08@cits1.stanford.edu> <49F9D4F6-C730-4517-86D8-CED39CE5D6BC@gmail.com> <9784d3ab0706060421h1c88d5f4r411deee1f5b5eba9@mail.gmail.com> <9784d3ab0707051416g4312d36eteae7dde1fc5d4290@mail.gmail.com>
On Thu, Jul 05, 2007 at 10:16:02PM +0100, Lee Millward wrote:
> Hi,
>
> I've attached an updated version of the patch which fixes the 3 PRs
> opened against it in my absense. It passes bootstrap and regression
> testing but as shown with the last patch, this doesn't necessarily
> indicate that the patch is correct.
>
> In an attempt to minimise the potential of any further breakage, I've
> ran the patched compiler through the Polyhedron 2k5 benchmarks and the
> "all.f90" file from CP2K meta-bug PR29975 - all completed successfully
> although the CP2K code did trigger an different ICE that has now been
> fixed and a reduced testcase included for it.
>
> If anyone has any other benchmarks or tests they think would be
> suitable for this sort of patch then I'd be glad to make use of them
> before this gets approved/committed. Needless to say I'll be
> monitoring Bugzilla for a while after the patch gets committed to
> watch out for any new problems caused by the patch.
>
> Ok to apply?
When I was doing the infrastructure work to remove the knowledge that arguments
used TREE_LIST's in the backends, I noticed the overloaded machine specific
intrinsic support for C/C++ was another place to be changed. I was just about
to start hacking on the front ends to get them to use a common infrastructure
for arguments and use vector support. You might glance over to my patches to
see if the infrastructure support in tree.h is sufficient to your needs:
http://gcc.gnu.org/ml/gcc-patches/2007-07/msg00445.html
--
Michael Meissner, AMD
90 Central Street, MS 83-29, Boxborough, MA, 01719, USA
michael.meissner@amd.com