discouraging register use

[DJ Delorie]

> >   [(set (match_operand:SI 0 "" "c,c,?m,?m")
> >         (plus:SI (match_operand:SI 1 "" "0,0,0,0")
> > 		 (match_operand:SI 1 "" "c,?m,c,?m")))]
> 
> Have you tried using '*' to hide 'm' from register preferencing?

Yes.  That makes it worse, because gcc doesn't use the "m" registers
at all (it seems) in some cases, and ignores the preference in other
cases (reload).  I don't recall the specifics, but it didn't work.