This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: pr14627
On Wed, Apr 13, 2005 at 08:49:42AM -0600, Jeffrey A Law wrote:
> On Wed, 2005-04-13 at 10:31 -0400, Diego Novillo wrote:
>
> > Back to the original problem. If by preventing constants in PHI
> > arguments we save ourselves this pass and don't increase memory
> > consumption nor compile times, then what useful purpose do they
> > serve? They certainly get in the way, we are constantly
> > traversing PHI arguments with kid gloves.
> Simply disallowing constants would be a mistake -- we would need
> to do the legwork to fix the existing optimizers.
>
Oh, absolutely. Places that before would simply lift a constant
from the argument list, would have to consult SSA_NAME_VALUE, the
value vector in CCP and/or the equivalency table in DOM.
The bar for the replacement is identical code generation between
the patched and unpatched compiler. So it shouldn't be hard to
evaluate.
Diego.