Re: divmodhi4

[Michael Meissner <meissner at cygnus dot com>]

On Thu, May 18, 2000 at 02:00:08PM -0700, jeff hammond wrote:
> I am creating a .md file for a processor we are
> developing and ran into a problem with divmodhi4.
> 
> The instruction that our processor uses for a 
> divmodhi4 always put the remainder in the
> register just after the quotient.  If the
> quotient goes in register N, the remainder always 
> goes in register N+1.  I have no idea how to 
> express this in rtx.  Can anyone suggest a good
> example or point to where this is documented?

I usually express it using a temporary type that is big enough to hold two
registers (SI in this case):

	[(set (match_operand:HI 0 "register_operand" "=r")
	      (div:HI (match_operand:HI 1 "register_operand" "r")
		      (match_operand:HI 2 "register_operand" "r")))
	 (set (match_operand:HI 3 "register_operand" "=r")
	      (mod:HI (match_dup 1)
		      (match_dup 2)))
	 (clobber (match_scratch:SI 4 "=&r"))]
	""
	"divmod %4,%1,%2\;move %4,%0\;move %L4,%1"

where print_operand code 'L' prints the second register in a 2 register type.
To get fancier, you can define just div/mod patterns to avoid a second useless
move where you only want a div or mod (and keeping divmodhi4 for cases where
you want both), and then split the pattern after reload into component
instructions.

-- 
Michael Meissner, Cygnus Solutions, a Red Hat company.
PMB 198, 174 Littleton Road #3, Westford, Massachusetts 01886, USA
Work:	  meissner@redhat.com		phone: +1 978-486-9304
Non-work: meissner@spectacle-pond.org	fax:   +1 978-692-4482