Floating Point Verification

[Moderator, John S. Quarterman]

From: Dominic Dunlop <uunet!mcvax!sphinx.co.uk!domo>
Cc: bryson <bryson at tahoe.unr.edu>

In article <208 at longway.TIC.COM> Derry Bryson writes:
>I am searching for a standard to check floating point processors and 
>emulators against.  Basically what I would like to find is a program 
>that performs several operations and evaluates the accuracy of the answers
>based upon some kind of accepted standard.
>...
>Volume-Number: Volume 14, Number 28

;login:, Volume 11, Number 2, March/April, 1986, pp 31-56

A Report on the Accuracy of Some Floating Point Math Functions on Selected
Computers

Technical Report GIT-ICS 85/06

Eugene H Spafford
John C Flaspohler

School of Information and Computer Science
Georgia Institute of Technology
Atlanta, Georgia 30332-0280
(404) 894-3152

Software Engineering Research center
Georgia Institute of Technology
Atlanta, Georgia 30332-0280
(404) 894-3180

Abstract
The UNIX operating system and the C programming language have gained a
large folloing in recent years, especially in research and academic
settings.  C and UNIX-like environments are available on a wide variety of
machines from personal computer to mainframe computers; however, few, if
any, of these implementations provide accurate floating point libraries,
although users tend to believe they do.  This paper presents the results of
running a set of accuracy test on more than a dozen different computer
systems under various versions of UNIX and UNIX-like environments.

Funding for the original printing of this report was obtained from the
School of Information and Computer Science, and from the Software
Engineering Research Center.

Copies of this report may be requested by sending US mail to the authors at
the address given on the title page or from
{akgua,decvax,hplabs,ihnp4,seismo}!gatech!tech-reports

Well, the path's a history lesson,

[ Spafford is now at Purdue, and can be reached as spaf at cs.purdue.edu;
as Dominic alludes, ihnp4 and seismo are effectively dead, but gatech is
still quite active, and can be reached from most of the known world.  -mod ]

but the paper's well worth getting hold of.  It pulls no punches in
showing some magnificently awful results.  No re-usable test suite is
mentioned -- although I dare say Georgia Institute of Technology may
still have one.  The tests used were carefully coded from ``random
accuracy tests for various standard mathematical functions'' given in
_Software Manual for Elementary Functions_, William J Cody Jr. &
William Waite, Computational Mathematics, Prentice-Hall, 1980.

It came as a surprise to me that the 3B2 had the most accurate math
package at the time.  It was also as slow as hell, but that's been fixed
now.  Wonder how accurate the new math package is...
-- 
Dominic Dunlop
domo at sphinx.co.uk  domo at riddle.uucp

Volume-Number: Volume 14, Number 33