This is the mail archive of the
mailing list for the GCC project.
Re: [patch] Fix PR43519: Bootstrap with Graphite enabled fails in Java libs
- From: Richard Guenther <rguenther at suse dot de>
- To: Sebastian Pop <sebpop at gmail dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Wed, 7 Apr 2010 10:23:46 +0200 (CEST)
- Subject: Re: [patch] Fix PR43519: Bootstrap with Graphite enabled fails in Java libs
- References: <email@example.com>
On Tue, 6 Apr 2010, Sebastian Pop wrote:
> The attached patches are fixing the bootstrap problem that we had in
> the Graphite branch, by computing the induction variable types based
> on the ranges the IVs can have after transforms. The old scheme that
> was inferring the IV types from the old-IV types wouldn't have worked
> as soon as the polyhedral transforms would have been more complex: for
> example loop skewing for negative factors would have broke as the
> transformed ranges of the IV do not correspond to the old type ranges.
> These patches passed bootstrap and test on the Graphite branch and
> they passed the SPEC 2006 benchmarks with no errors: see the recent
> test results at http://groups.google.com/group/gcc-graphite-test
> I will commit these patche to trunk.
As said in the PR you should be using build_nonstandard_integer_type,
not lang_hooks.types.type_for_size (which can return NULL).
> Richi, are these changes ok for 4.5 as well?
0001 is quite intrusive and IMHO not suitable at this point.
0002 looks simple (but you need to make sure 'one' is of
sizetype when using POINTER_PLUS_EXPR, so I'm not sure how
the patch could work. if gmp_one is 1, why not use
build_int_cst (type, 1) and/or size_one_node?).
0003 has a changelog that doesn't exactly match what is going on.