Patch to fix recent local-alloc IA-64 breakage (version 2)

John Wehle john@feith.com
Mon Oct 23 14:01:00 GMT 2000


> On Mon, Oct 23, 2000 at 01:19:42PM -0400, John Wehle wrote:
> > +     case REG:
> > +       return REGNO (x) < FIRST_PSEUDO_REGISTER && call_used_regs[REGNO (x)];
> 
> This is, unfortunately, too simplistic.  All fixed registers are
> also marked call_used.  But obviously this should not apply to the
> stack pointer, arg pointer etc.
> 
> For years I've been wanting to fix the mistake that documents
> CALL_USED must include FIXED.  But adjusting all the targets
> always seemed daunting.

Umm ... are you saying that there's currently no way to tell what
registers are actually clobbered by a call?  Major bummer .. in
that case I'm not sure what to due in the short term other than
the patch you came up with for rtx_varies_p and rtx_unstable_p.
I suppose another option is to have update_equiv_regs not use
rtx_varies_p, or to have special case code in update_equiv_regs
which handles PIC_OFFSET_REG_CALL_CLOBBERED.

I'll think on this some more.  Any idea what direction I should
go in?

-- John
-------------------------------------------------------------------------
|   Feith Systems  |   Voice: 1-215-646-8000  |  Email: john@feith.com  |
|    John Wehle    |     Fax: 1-215-540-5495  |                         |
-------------------------------------------------------------------------



More information about the Gcc-patches mailing list