Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!ucbvax!hplabs!hp-pcd!hplsla!jima
From: jima@hplsla.HP.COM (Jim Adcock)
Newsgroups: comp.lang.c
Subject: Re: Optimal for loop on the 68020.
Message-ID: <5260014@hplsla.HP.COM>
Date: 5 Jun 89 19:18:17 GMT
References: <11993@well.UUCP>
Organization: HP Lake Stevens, WA
Lines: 210

Unfortunately, the "favorite loop constuct" is going to be very dependent on
the particular compiler used, and exactly *what* you do inside the loop.
Below is your examples, redone with a null loop body, plus *my* favorite
loop construct: for(i=COUNT; i--;).  I used gcc and hp's cc 6.5 compiler
with -O -S and in gnu -O -S -fstrength-reduce -fcombine-regs -fforce-mem
-fforce-addr.  The point is that these things are very unpredictable,
and optimizing compilers work well on linear code segments, not over
branch points in programs [including loops].  The "optimizing" compilers
I've seen do about 20% better than non-optimizing compilers, but are still
far from "optimal" by our human standards.


#define COUNT 100
main()
{
    int i;
    for ( i = 0; i < COUNT; i++ ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = 0; i < COUNT; ++i ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = 0; ++i <= COUNT; ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = 0; i++ < COUNT; ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = COUNT; i > 0; i-- ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = COUNT; i > 0; --i ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = COUNT; --i >= 0; ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = COUNT; i-- > 0; ); 
xxxxxxxxxxxxxxxxxxx();
    for ( i = COUNT; i--; ); 
xxxxxxxxxxxxxxxxxxx();
}


#NO_APP
gcc_compiled.:
.text
	.even
.globl _main
_main:
	link a6,#0


	moveq #99,d0
L4:
	dbra d0,L4
	clrw d0
	subql #1,d0
	jcc L4

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #99,d0
L8:
	dbra d0,L8
	clrw d0
	subql #1,d0
	jcc L8

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L10:
	dbra d0,L10
	clrw d0
	subql #1,d0
	jcc L10

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L14:
	dbra d0,L14
	clrw d0
	subql #1,d0
	jcc L14

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L20:
	subql #1,d0
	tstl d0
	jgt L20

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L24:
	subql #1,d0
	tstl d0
	jgt L24

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L26:
	dbra d0,L26
	clrw d0
	subql #1,d0
	jcc L26

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L30:
	subql #1,d0
	moveq #-1,d1
	cmpl d0,d1
	jlt L30

	jbsr _xxxxxxxxxxxxxxxxxxx

	moveq #100,d0
L34:
	dbra d0,L34
	clrw d0
	subql #1,d0
	jcc L34

	jbsr _xxxxxxxxxxxxxxxxxxx

	unlk a6
	rts

[hp cc 6.5:] ==============================================

	global	_main
_main:
	link.w	%a6,&-8
	mov.l	%d7,-(%sp)

	movq	&99,%d7
L12:
	dbf	%d7,L12

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&99,%d7
L16:
	dbf	%d7,L16

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&0,%d7
L21:
	addq.l	&1,%d7
	movq	&100,%d0
	cmp.l	%d7,%d0
	ble.b	L21

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&0,%d7
L24:
	mov.l	%d7,%d0
	addq.l	&1,%d7
	movq	&100,%d1
	cmp.l	%d0,%d1
	blt.b	L24

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&100,%d7
L20001:
	subq.l	&1,%d7
	tst.l	%d7
	bgt.b	L20001

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&100,%d7
L20003:
	subq.l	&1,%d7
	tst.l	%d7
	bgt.b	L20003

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&100,%d7
L33:
	subq.l	&1,%d7
	bge.b	L33

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&100,%d7
L36:
	mov.l	%d7,%d0
	subq.l	&1,%d7
	tst.l	%d0
	bgt.b	L36

	jsr	_xxxxxxxxxxxxxxxxxxx

	movq	&100,%d7
L39:
	mov.l	%d7,%d0
	subq.l	&1,%d7
	tst.l	%d0
	bne.b	L39

	jsr	_xxxxxxxxxxxxxxxxxxx

	mov.l	(%sp)+,%d7
	unlk	%a6
	rts