allow recursive peepholing

[Jan Hubicka]

> On Fri, Oct 05, 2001 at 08:43:49PM +0200, Jan Hubicka wrote:
> > for the setcc peepholer it is needed to peep2 recursivly - first
> > to create new sequence and then to convert movl $0, reg to xor.
> 
> I'm not so sure this is a great idea.  I'd rather handle the
> mov/xor thing directly for now.  If someone else comes up with
> a better example of why re-scanning peepholes is helpful, I'll
> reconsider then.

Another idea from i386 peep2s is the splitting of long instructions.
I believe we split out the load of immediate to register and the immediate
can be zero, so we are hitting same situation.
I need to checks peep2s in i386.md. I've always believed that we do
recursive peepholing, but obviously we don't :)

I am week offline so I will do so then.  Thanks for caring the second
change.

Honza
> 
> 
> r~