Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!bbn!gatech!psuvax1!schwartz@shire.cs.psu.edu
From: schwartz@shire.cs.psu.edu (Scott Schwartz)
Newsgroups: comp.arch
Subject: Re: Questions on SparcStation 1 performance
Message-ID: <4566@psuvax1.cs.psu.edu>
Date: 10 May 89 04:34:53 GMT
References: <6008@brunix.UUCP>
Sender: news@psuvax1.cs.psu.edu
Reply-To: schwartz@shire.cs.psu.edu (Scott Schwartz)
Organization: Pennsylvania State University, Computer Science
Lines: 26
In-reply-to: jps@cs.brown.edu (John Shewchuk)

In article <6008@brunix.UUCP>, jps@cs (John Shewchuk) writes:
>   x = (a * y)/b;  [100,000 iterations]

>To my surprise I found that the floating point version returned
>quickly - it used about 1.6 seconds of user time - while the integer
>version took much longer- about 43 seconds.

Could you give us some more information?  I just tried the following
on a Sun4/260, sans -O; it took 0.4 user, 0.0 sys. 

main()
{
	int x = 1, y = 9, a = 3, b = 4;
	int i = 100000;

	&x; &y; &a; &b;
	while (--i) {
		x = (a * y)/b;
		&x;
	}
}

Hey, the Sun4/110 does fp via kernel emulation: maybe the SparcStations
do integer that way!  :-) :-) 
-- 
Scott Schwartz		<schwartz@shire.cs.psu.edu>