Fix latent bug in __expand_buitin_nonlocal_goto

[Andrew Pinski]

On Sun, Jun 22, 2008 at 4:46 PM, Andrew Pinski <pinskia@gmail.com> wrote:
> On Tue, Jun 17, 2008 at 11:58 AM, Jan Hubicka <jh@suse.cz> wrote:
>> Hi,
>> with SSA at O0 the following testcase in testsuite:
>
> This causes a failure on powerpc*-*-* as the restore_stack_nonlocal
> pattern requires a memory address for the saved stack pointer.
> I think the x86 back-end should change to stabilize the saved pointer
> to a register.

Looking over the code for expand_builtin_nonlocal_goto, I think we
should stabilize r_save_area instead of r_sp.  This will cause us to
create a MEM of a REG instead of a MEM of a REG plus constant where
REG will most likely be the frame pointer.

Thanks,
Andrew Pinski