Re: Missed optimization in m68k

[Alan Lehotsky <apl at alum dot mit dot edu>]

At 1:48 PM -0600 6/14/02, law@redhat.com wrote:
> In message <200206141923.g5EJNeK22444@hyper.wm.sps.mot.com>, Peter 
> Barada write
> s:
>  >
>  > I'm compiling the small snippet:
>  .....
>  > The problem is that the tst.b is not necessary since the move.b
>  > updated the condition codes as part of the move.  How can I best
>  > convince gcc that it should generate the follwing instead:
>  >
>  > copy:
>  >	move.l 4(%sp),%a1
>  >	move.l 8(%sp),%a0
>  >	move.l %a0,%d0
>  > .L2:
>  >	move.b (%a1)+,(%a0)+
>  >	jbne .L2
>  >	rts
>  >
>  > Thanx...
> I believe you'll be hacking in NOTICE_UPDATE_CC.
>
> jeff

In this case, I suspect that a DEFINE_PEEPHOLE is probably better!

Since I'm guessing that the RTL for the generated code has the 
POST_INC on the source side
of the  (SET (CC) ...) the 'move.b' instruction is already output 
into the ".s" file, so you can't
get the CC_UPDATE logic to generate a

	move.b	(%a1)+,(%a0)+

but if you recognized  the sequence of instructions and emitted the 
optimal loop....


--
		    Quality Software Management
		http://home.earthlink.net/~qsmgmt
		          apl@alum.mit.edu
		          (978)287-0435 Voice
		          (978)808-6836 Cell

	Software Process Improvement / Management Consulting
	     Language Design / Compiler Implementation