[C++ PATCH] Rewrite convert_nontype_argument

[Andrew Pinski]

> 
> 
> On Tue, 20 Jul 2004, Mark Mitchell wrote:
> > Could well be.  I do think someone should measure first, though.  There
> > look to be some code paths through fold_convert that involve making
> > calls back to types_compatible_p, which has been showing up in my
> > profiling as not insignificant.
> 
> Interestingly, the call to types_compatible_p in fold_convert is
> just a temporary work-around for a bug in the C++ front-end:
> http://gcc.gnu.org/ml/gcc-patches/2004-06/msg01237.html
> 
> My hope is that once pointer member functions are given the correct
> TYPE_MAIN_VARIANT, that this call can be removed.  The optimizations
> performed by this point in the middle-end should be independent of the
> front-end's type system.  Indeed this was one of the motivations of
> a middle-end fold_convert that was independent of "convert".

Well it also used for IMA reasons so it cannot be removed after fixing
the C++ front-end.

Andrew