Re: [PATCH] Fix -m64 on powerpc-darwin for pic code

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Nov 2, 2004, at 1:23 AM, Stan Shebs wrote:

> Andrew Pinski wrote:
>
> > Even for the following simple code we ICE:
> > int i = 0;
> > void f(void)
> > {
> >  i = 0;
> > }
> Although the ideas of the patch seem generally sensible, I'm curious
> how it is you're getting an ICE; I'm trying various combos of flags,
> can't get my sources-updated-this-morning build to choke on this bit.
>
> Also, if you're going to do this sort of churn, might as well take the
> opportunity to move the macho patterns to darwin.md.
From a clean tree updated last night:
[zhivago:gcc/objdir/gcc] pinskia% ./cc1 -fPIC t.c -m64
 f

t.c: In function 'f':
t.c:5: error: unrecognizable insn:
(insn 29 28 30 (set (reg:DI 65 lr)
        (unspec:SI [
                (symbol_ref:DI ("<pic base>") [flags 0x180])
            ] 15)) -1 (nil)
    (expr_list:REG_MAYBE_DEAD (const_int 0 [0x0])
        (nil)))
t.c:5: internal compiler error: in insn_default_length, at 
insn-attrtab.c:767
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

We get the unrecognizable instruction because we only call 
load_macho_picbase
instead of checking TARGET_64BIT and the appropriate function.  This is 
around
rs6000.c:14022 where the problem is.

Thanks,
Andrew Pinski