This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tuples][patch] Convert pass_ccp and pass_store_ccp to tuples
On Tue, Mar 4, 2008 at 6:43 AM, Diego Novillo <dnovillo@google.com> wrote:
> > + /* Substitute operands with their values and try to fold. */
> > + replace_uses_in (stmt, NULL, const_val);
> > + /* FIXME tuples. */
> > +#if 0
> > + retval = fold_call_expr (rhs, false);
> > +#else
> > + retval = NULL_TREE;
>
> This should call fold_gimple_call, but that is a destructive folder, so
> we need to figure out something different here. For now, let's just
> remove this handler, since as you note below, it is never invoked.
I will remove it if you feel strongly, but I'd rather preserve the code as
much as possible. I think the fact that all calls are filtered out by
likely_value is highly dubious, and that we will want to handle this
case in the future. (In fact, I'd like to try an experiment lobotomizing
likely_value temporarily as a torture test for ccp, as I believe it is purely
a a performance enhancement that is not essential for correctness,
and results in less "work" for the rest of ccp). I have to deal with
fold_call_expr anyhow in ccp_fold_builtin as well. It if proves easier to
handle that case in a way that is not applicable to the case here
in ccp_fold, then I'd be willing to drop the handler in ccp_fold for the
sake of expediency, but it seems a bit premature.
--Bill