Re: [RFA] mips-n32 reg_struct_has_addr

[Michael Snyder <msnyder at redhat dot com>]

Andrew Cagney wrote:
> 
> > + static int
> > + mips_newabi_reg_struct_has_addr (int gcc_p, struct type *type)
> > + {
> > +   enum type_code typecode = TYPE_CODE (check_typedef (type));
> > +   int len = TYPE_LENGTH (check_typedef (type));
> > +
> > +   if (typecode == TYPE_CODE_STRUCT   || typecode == TYPE_CODE_UNION)
> > +     return (len > MIPS_SAVED_REGSIZE && len % MIPS_SAVED_REGSIZE != 0);
> > +
> > +   return 0;
> > + }
> 
> No, I don't think this is right.  As best I can tell, there is no cap on
> the size of a struct that is passed by register.
> 
> http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi/srch3@o32/0650/bks/SGI_Developer/books/Mpro_n32_ABI/sgi_html/ch02.html#id80931
> 
> I.e., o32 and n32/n64 are the same -> never pass structures by reference.

OK, patch withdrawn, I will resubmit it using this assumption, 
just to get the gdbarch-ification checked in, and then we can
work on resolving how N32/64 should work.


> 
> > *************** mips_push_arguments (int nargs,
> > *** 2428,2448 ****
> >                           "mips_push_arguments: %d len=%d type=%d",
> >                           argnum + 1, len, (int) typecode);
> >
> > !       /* The EABI passes structures that do not fit in a register by
> 
> Per other e-mail, rather than modify the existing mips_push_argument()
> can you please created a dedicated mips_n32n64_push-arguments() function
> and hack on that.  That way, I don't have to worry about the other ABIs.
> 
> > +       /* Set up reg_struct_has_addr.  */
> > +       set_gdbarch_reg_struct_has_addr (gdbarch,
> 
> BTW, the comment here is redundant.
> 
> enjoy,
> Andrew