Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ames!ucbcad!ucbvax!decvax!decwrl!pyramid!prls!mips!earl
From: earl@mips.UUCP (Earl Killian)
Newsgroups: comp.arch,comp.sys.nsc.32k
Subject: Re: "Unoptimizing" Dhrystone
Message-ID: <333@gumby.UUCP>
Date: Wed, 22-Apr-87 19:18:05 EST
Article-I.D.: gumby.333
Posted: Wed Apr 22 19:18:05 1987
Date-Received: Fri, 24-Apr-87 23:43:13 EST
References: <4190@nsc.nsc.com> <951@moscom.UUCP> <2577@intelca.UUCP> <312@gumby.UUCP>
Lines: 18
Summary: there's no such thing as whetstone
Xref: mnetor comp.arch:1063 comp.sys.nsc.32k:108

David Hough pointed out to me that most of the nice benchmarking
features I attributed to whetstone are not generic whetstone.  The
original version was written in Algol, and lacked the features I
described.

The one I found lying around here runs the whole program twice, once
with the loop multiplier set to an input parameter, and once run with
that parameter plus ten.  The difference of the times for these two
runs is a measure of the time for 10 loops with various overhead
canceled (for example, the timing calls and result printout outside of
the loops).

I just sent away and received a version from Argonne's netlib service,
and found that it does not do this.  There's probably a lot of
different whetstones out there.

I have heard that NBS is setting up a benchmark service.  Perhaps
they'll straighten up some of this.  Anyone know any more about it?