This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Fix PR39207, bogus strict-aliasing warning, PR39074, wrong PTA
On Wed, Feb 18, 2009 at 12:42 PM, Richard Guenther <rguenther@suse.de> wrote:
> On Wed, 18 Feb 2009, Daniel Berlin wrote:
>
>> On Wed, Feb 18, 2009 at 8:47 AM, Richard Guenther <rguenther@suse.de> wrote:
>> >
>> > This "backports" the fix for PR39074 from alias-improvements branch to
>> > trunk, thereby fixing PR39207, bogus strict-aliasing warnings from
>> > libstdc++ headers. On the branch I needed the second patch appended
>> > below to avoid performance regressions, so I will backport this as well.
>> >
>> > Bootstrapped and tested on x86_64-unknown-linux-gnu. I have put the
>> > patches on x86_64 and ia64 SPEC2000 testing (I hope we're not going
>> > the 4.3 way of very late performance regressions due to bug fixes...).
>> >
>> > I plan to apply this tomorrow after testresults have arrived.
>> >
>> > Richard.
>> >
>> > 2009-02-18 Richard Guenther <rguenther@suse.de>
>> >
>> > PR tree-optimization/39207
>> > PR tree-optimization/39074
>> > * tree-ssa-structalias.c (storedanything_id, var_storedanything,
>> > storedanything_tree): New.
>> > (do_ds_constraint): Simplify ANYTHING shortcutting. Update
>> > the STOREDANYTHING solution if the lhs solution contains
>> > ANYTHING.
>> > (build_pred_graph): Add edges from STOREDANYTHING to all
>> > non-direct nodes.
>> You mean build_succ_graph.
>
> Thanks, fixed.
>
>> Also, you have the comment and the description wrong.
>> It's really adding edges *to* STOREDANYTHING *from* all non-direct nodes.
>
> Sure? We want to propagate from STOREDANYTHING to all non-direct nodes,
> so with
Uh, well then i'm not clear on what you are doing.
Being an indirect node simply means the node may be referenced other
than by it's name.
This does not mean it points to anything. So why receive an edge from
STOREDANYTHING.
It will certainly make it point to more things, but i don't see why it
is correct at all.