mn10300/AM33: Simplify movhi patterns

[Jeffrey A Law]

  In message < 20000426012012.B6814@cygnus.com >you write:
  > On Wed, Apr 26, 2000 at 04:39:27AM -0300, Alexandre Oliva wrote:
  > > -  [(set (match_operand:HI 0 "general_operand" "=dx,a,dx,a,dx,a,dx,a,dxa,
  > m")
  > > -	(match_operand:HI 1 "general_operand" "0,0,I,I,a,dx,dxi,ia,m,dxa"))]
  > > +  [(set (match_operand:HI 0 "general_operand" "=d*x*a,d*x*a,d*x*a,d*x*a,
  > m")
  > > +	(match_operand:HI 1 "general_operand" "0,I,d*x*ai,m,d*x*a"))]
  > [...]
  > >      case 0:
  > > -    case 1:
  > >        return \"nop\";
  > 
  > Question.  Why does this port define this alternative?  Surely
  > this never gets matched, since delete_trivially_dead_insns should
  > kill them if nothing else.
It did happen at one time, possibly in a non-optimizing compile (it's only
been 4+ years since I did the work).  The mn102/mn103 can't encode a mov with
an identical source/dest, so we emit a nop instead.

jeff