Infrastructure to let ipcp to remove unused arguments from calls
Jan Hubicka
hubicka@ucw.cz
Fri Aug 29 22:15:00 GMT 2008
> > On Wed, Aug 27, 2008 at 11:39, Jan Hubicka <jh@suse.cz> wrote:
> >
> > > +
> > > + /* This is a new type, not a copy of an old type. Need to reassociate
> > > + variants. We can handle everything except the main variant lazily. */
> > > + t = TYPE_MAIN_VARIANT (orig_type);
> >
> > Why not use build_variant_type_copy() here?
>
> I didn't noticed existence of this function so far ;). The variant handling
> is copied from type duplication in tree-inline.
>
> In meantime I however noticed problem with java, where constant
> propagating THIS pointer to method (that happens quite often in libjava
> build with clonning) confuse debug output as METHOD_TYPE is supposed to
> always have THIS pointer as first operand.
>
> I think I need to translate the type to FUNCTION_TYPE and I guess only
> way is to build type from scratch via build_function_type.
>
> This is patch I tested in meantime (+the renaming), I guess
> build_variant_type_copy does not give us much benefit here...
Forgot to add, bootstrapped/regtested x86_64-linux with wnd without ipcp
patch, OK? :)
Honza
More information about the Gcc-patches
mailing list