SSE/MMX moves optimization

[Jan Hubicka]

> > 	* i386.md (movsi_1): Cleanup constraints;
> 
> Could you explain this change? Why not put it on the 3.3 branch?

Because it is not fix, just optimization and cleanup.
The reason of the change was that K8 dislike moves that transfer data
between unit, so we have to claim GCC that those don't exist at all.
> 
> > Index: i386.md
> > ===================================================================
> > RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.md,v
> > retrieving revision 1.421
> > diff -c -3 -p -r1.421 i386.md
> > *** i386.md	3 Feb 2003 20:20:36 -0000	1.421
> > --- i386.md	4 Feb 2003 19:27:17 -0000
> > ***************
> > *** 1166,1173 ****
> >
> >   (define_insn "*movsi_1"
> >     [(set (match_operand:SI 0 "nonimmediate_operand"
> > "=*?a,r,*?a,m,!*y,!rm,!*y,!*Y,!rm,!*Y") ! 	(match_operand:SI 1
> > "general_operand" "im,rinm,rinm,rin,rm,*y,*y,rm,*Y,*Y"))] !   "GET_CODE
> > (operands[0]) != MEM || GET_CODE (operands[1]) != MEM" {
> >     switch (get_attr_type (insn))
> >       {
> > --- 1166,1214 ----
> >
> >   (define_insn "*movsi_1"
> >     [(set (match_operand:SI 0 "nonimmediate_operand"
> > "=*?a,r,*?a,m,!*y,!rm,!*y,!*Y,!rm,!*Y") ! 	(match_operand:SI 1
> > "general_operand" "im,rinm,rinm,rin,*y,*y,rm,*Y,*Y,rm"))] !  
> 
> Why this circular permutation of constraints for operand 1?

This is to match the order used by the other patterns so we have them in
sync.

Honza
> 
> > ***************
> > *** 1201,1207 ****
> >   	   ]
> >   	   (const_string "imov")))
> >      (set_attr "modrm" "0,*,0,*,*,*,*,*,*,*")
> > !    (set_attr "mode" "SI,SI,SI,SI,SI,SI,DI,TI,SI,SI")])
> >
> >   ;; Stores and loads of ax to arbitrary constant address.
> >   ;; We fake an second form of instruction to force reload to load address
> > --- 1242,1248 ----
> >   	   ]
> >   	   (const_string "imov")))
> >      (set_attr "modrm" "0,*,0,*,*,*,*,*,*,*")
> > !    (set_attr "mode" "SI,SI,SI,SI,DI,SI,SI,TI,SI,SI")])
> >
> >   ;; Stores and loads of ax to arbitrary constant address.
> >   ;; We fake an second form of instruction to force reload to load address
> 
> Same question.
> 
> -- 
> Eric Botcazou