Message-Id:	<m0yw5Xm-000S3zC@inti.gov.ar>
Comments:	Authenticated sender is <salvador AT natacha DOT inti DOT gov DOT ar>
From:	"Salvador Eduardo Tropea (SET)" <salvador AT inti DOT gov DOT ar>
Organization:	INTI
To:	horst DOT kraemer AT snafu DOT de (Horst Kraemer), djgpp AT delorie DOT com
Date:	Tue, 14 Jul 1998 11:03:06 +0000
MIME-Version:	1.0
Subject:	Re: DJGPP division optimisations
In-reply-to:	<35ab1f95.79373671@news.snafu.de>

horst DOT kraemer AT snafu DOT de (Horst Kraemer) wrote:
> On Mon, 13 Jul 1998 20:32:05 -0400, Vic <tudor AT cam DOT org> wrote:
> 
> 
> >it's true, I checked this. I put "unsigned" in front of the variables
> >and the 2 loops performed the same (2.74 seconds)
> 
> And the reason is clear. The operations x/2 and x>>1 are different if
> x is a negative number. As you declared x to be a signed type the
> compiler may not "pessimize" /2 to >>1. It doesn't know when compiling
> that x will never be negative.
> 
>         int i = -1;
> 	printf("%d %d\n", i/2 , i>>1 );

More or less, but both are shifts, the point is that 
integer=integer/2 is a little bit tricky:

#include <stdio.h>
//#define TYPE unsigned
#define TYPE int

int main(int argc, char *argv[])
{
 TYPE x=argc/2;
 TYPE y=argc>>1;
 TYPE w=((unsigned)(argc))/2;
 TYPE z=((unsigned)(argc))>>1;
 printf("%d %d %d %d\n",x,y,w,z);
 return 0;
}
 
And it generates the following code:

	.file	"p.c"
gcc2_compiled.:
___gnu_compiled_c:
.text

LC0:
	.ascii "%d %d %d %d\12\0"
	.p2align 2
	.def	_main;	.val	_main;	.scl	2;	.type	044;	.endef
.globl _main
_main:

	pushl %ebp
	movl %esp,%ebp
	movl 8(%ebp),%eax

<<<<<<<<<<<<<<< Line 2 here comes the /2
	.ln	2
	movl %eax,%edx
	shrl $31,%edx
	addl %eax,%edx
	sarl $1,%edx

<<<<<<<<<<<<<<< Line 3 is the >>1
	.ln	3
	movl %eax,%ecx
	sarl $1,%ecx

<<<<<<<<<<<<<<< Using integers that's the same, so GCC calculates 1 
<<<<<<<<<<<<<<< and ...
	.ln	4
	shrl $1,%eax

	.ln	6
<<<<<<<<<<<<<<< ... puts it twice in the stack
	pushl %eax
	pushl %eax
	pushl %ecx
	pushl %edx
	pushl $LC0
	call _printf
	.ln	7
	xorl %eax,%eax
	.ln	8
	leave
	ret


Greetings, SET
------------------------------------ 0 --------------------------------
Visit my home page: http://set-soft.home.ml.org/
or
http://www.geocities.com/SiliconValley/Vista/6552/
Salvador Eduardo Tropea (SET). (Electronics Engineer)
Alternative e-mail: set-soft AT usa DOT net set AT computer DOT org
ICQ: 2951574
Address: Curapaligue 2124, Caseros, 3 de Febrero
Buenos Aires, (1678), ARGENTINA
TE: +(541) 759 0013

- Raw text -