Re: msp430 port

[nick clifton <nickc at redhat dot com>]

Hi DJ, Hi Richard,

> > > +;; The next two patterns are here to support a "feature" of how GCC implements
> > > +;; varargs.  When a function uses varargs and the *second* to last named
> > > +;; argument is split between argument registers and the stack, gcc expects the
> > > +;; callee to allocate space on the stack that can contain the register-based
> > > +;; part of the argument.  This space *has* to be just before the remaining
> > > +;; arguments (ie the ones that are fully on the stack).
> >
> > Ug.  We ought to have been able to convince the compiler to copy the other
> > direction.  I.e. create storage for that argument as if it were a regular local
> > variable, spill the register portion into it, and also copy in the fragment
> > from the stack.
> >
> > That sort of solution seems better than fiddling the location of the return
> > address in both directions.

I agree - but I just do not have the GCC-fu skills to do this.  Hence I 
created the two patterns.


> > > +static enum machine_mode
> > > +msp430_get_raw_arg_mode (int regno)
> > > +{
> > > +  return (regno == ARG_POINTER_REGNUM) ? VOIDmode : Pmode;
> > > +}
> >
> > How can the raw mode of AP be VOID?

It cannot.  This was a bit of debugging that I put in whilst trying to 
track down problems with __builtin_apply_args().  Sorry - it should have 
been removed.

Cheers
  Nick