Path: utzoo!attcan!uunet!husc6!cmcl2!adm!xadmx!rbj@dsys.ncsl.nist.gov
From: rbj@dsys.ncsl.nist.gov (Root Boy Jim)
Newsgroups: comp.unix.wizards
Subject: Optimal for loop on the 68020.
Message-ID: <19945@adm.BRL.MIL>
Date: 8 Jun 89 18:59:46 GMT
Sender: news@adm.BRL.MIL
Lines: 70

? From: Jef Poskanzer <pokey@well.uucp>

? In the referenced message, edler@cmcl2.UUCP (Jan Edler) wrote:
? }However, the compiler shouldn't do either of these when d0 is initialized
? }to a value < 65536.

? Not good enough, since the loop body might modify the index variable.
? If the compiler is smart enough to do a global semantic check for this
? sort of thing, then you can do this optimization; but we've already
? established that cc and gcc aren't that smart.

Hey, aren't most programs *I/O bound* anyway? And you can sometimes
unroll your loops.

I seem to remember Chris Torek stating something about using

main()
{
	register short n = 10;
	do {
		printf("n = %d\n");
	} while (--n != -1);
}

On a Sun 3/180 SunOS 3.5 compiled with `cc -S -O' generates:

	.data
	.text
LL0:
|#PROC# 04
	.data1
L18:
	.ascii	"n = %d\12\0"
	LF12	=	4
	LS12	=	128
	LFF12	=	0
	LSS12	=	0
	LP12	=	16
	.data
	.text
	.globl	_main
_main:
|#PROLOGUE# 0
	link	a6,#-4
	movl	d7,sp@
|#PROLOGUE# 1
	moveq	#10,d7
L16:
	movw	d7,d0
	extl	d0
	movl	d0,sp@-
	pea	L18
	jbsr	_printf
	addqw	#8,sp
	dbra	d7,L16
	movl	a6@(-4),d7
	unlk	a6
	rts

and prints "n = 10" thru "n = 0". I even tried initializing n to -5
and adding `n += 2' before the printf. Prints -3 thru 0.

? Jef

?    Jef Poskanzer   {ucbvax, lll-crg, sun!pacbell, apple, hplabs}!well!pokey
?  "A banker is a fellow who lends you his umbrella when the sun is shining and
?           wants it back the minute it begins to rain." -- Mark Twain

	Root Boy Jim is what I am
	Are you what you are or what?