Oust HPPA PIC_OFFSET_TABLE_REGNUM_SAVED

[Alan Modra]

On Mon, 15 Jan 2001, Jeffrey A Law wrote:

> First, if an ASM trashes a fixed register, then it is the ASM's responsibility
> to restore it.  So that part of this patch is wrong.

OK, that's fair enough.

> The rest of it looks pretty reasonable and can be installed.

There is possibly a small problem.  The test at the end of
hppa_expand_prologue:

  if (flag_pic && HARD_REGISTER_P (PIC_OFFSET_TABLE_SAVE_RTX))

might be wrong.  I tested HARD_REGISTER_P because we don't want to emit
the instruction as part of the prologue if PIC_OFFSET_TABLE_SAVE_RTX is
dead.  What happens if register pressure causes the pseudo to be allocated
a stack slot?

Alan Modra
-- 
Linuxcare.  Support for the Revolution.