Newsgroups: comp.arch
Path: utzoo!henry
From: henry@zoo.toronto.edu (Henry Spencer)
Subject: denormalized numbers
Message-ID: <1990Nov28.173640.7185@zoo.toronto.edu>
Organization: U of Toronto Zoology
References: <897@sibyl.eleceng.ua.OZ> <13342@encore.Encore.COM> <46866@apple.Apple.COM> <KHB.90Nov27131731@chiba.Eng.Sun.COM>
Date: Wed, 28 Nov 90 17:36:40 GMT

In article <KHB.90Nov27131731@chiba.Eng.Sun.COM> khb@chiba.Eng.Sun.COM (chiba) writes:
>Those who have codes like this, usually think everything is OK, worked
>fine on their ----- (fill in CDC, Honeywell, Univac, VAX, etc.) and
>are perfectly happy with flush to zero semantics.

Of course, sometimes this is "worked fine" for suitable values of "worked"!

Long ago, we had a system with software emulation of hardware floating-point.
Then we got the hardware version.  The one noticeable difference in behavior
was that the hardware trapped division by zero and the software hadn't.  You
would be amazed at how many users came in complaining that their "fully
debugged" programs had stopped working.  I hope none of the pre-changeover
results found their way into anybody's paper or thesis...
-- 
"The average pointer, statistically,    |Henry Spencer at U of Toronto Zoology
points somewhere in X." -Hugh Redelmeier| henry@zoo.toronto.edu   utzoo!henry
