Inner-loop optimization regression from 3.3 to 3.4

[Jan Hubicka]

> On Sat, Oct 18, 2003 at 10:37:27PM +0200, Jan Hubicka wrote:
> > Or you mean just check whether the proposed sequence is cheaper than
> > sequence of two setcc patterns and logical?
> 
> Or one setcc instead of the proposed operation, yes.
> It may be very cheap to get -1 via emit_store_flag.

Hmm, yes.  There are many cases one should consider, but probably too
many.  I will try to improve that code somewhat so it at least doesn't
assume that shifts are cheap.
I've got first SPEC runs and the difference is in the noise (which is
quite strange as it didn't used to be), so I will try to figure out
whether I can improve that somewhat :)

Honza
> 
> 
> r~