i386 PIC prologues/epilogues changes (and replacement for CONST fix)

Jan Hubicka jh@suse.cz
Thu Aug 3 12:18:00 GMT 2000


> On Thu, Aug 03, 2000 at 06:06:29PM +0200, Jan Hubicka wrote:
> > Thats was my original plan too - problem is that the fusage seems to be
> > attached to the pattern after it is generated, so I can't control it :(
> 
> Sure you can.  In the expander, you emit an insn, then attach
> stuff to it.
After some experimenting, this seems to be dangerous to me.  First of all,
the FUSAGE seems to be lost somewhere, I will figure out where, but more
importantly, when non-constant call insn gets converted to normal call
insn, it will not have that use.
So I think it is safer to go with the existing pattern with test
added that prohibit non-local calls in the non-ebx patterns.
Sounds resonable?
> 
> > BTW any hints about how to teach reload to use the EBX?
> 
> Nothing comes to mind.
> 
> 
> r~


More information about the Gcc-patches mailing list