Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!rutgers!ames!ucbcad!ucbvax!amdcad!bcase
From: bcase@amdcad.UUCP
Newsgroups: comp.arch
Subject: Re: 01/31/87 Dhrystone Results and Source
Message-ID: <14782@amdcad.UUCP>
Date: Mon, 16-Feb-87 13:37:07 EST
Article-I.D.: amdcad.14782
Posted: Mon Feb 16 13:37:07 1987
Date-Received: Tue, 17-Feb-87 06:12:28 EST
References: <2348@homxb.UUCP> <15203@onfcanim.UUCP> <293@ames.UUCP> <1224@husc6.UUCP> <2387@homxb.UUCP>
Reply-To: bcase@amdcad.UUCP (Brian Case)
Organization: Advanced Micro Devices, Sunnyvale, California
Lines: 28
Keywords: Benchmark, performance measurement

In article <2387@homxb.UUCP> gemini@homxb.UUCP (Rick Richardson) writes:
>In article <1224@husc6.UUCP>, reiter@endor.harvard.edu (Ehud Reiter) writes:
>> Has anyone actually tried to evaluate the Dhrystone (and other benchmarks)
>> by seeing how well it predicts performance on real applications?  It would
>> seem straightforward to take ten random applications running on specific test
>> data, measure their performance on some target machine/compiler combinations,
>> and statistically analyze how much of the peformance differences had been
>> predicted by the Dhrystone figures.
>
>Since Dhrystone does no I/O during the benchmark, I doubt that Dhrystone
>predicts performance of *ANY* application.  Except perhaps other integer
>benchmarks!  It's probably useful in predicting the performance of,
>say, qsort(3), UNIX's in core sort routine, and stuff like that.  We've
>found it most useful in evaluating compiler code generation technology.

Ahhh, now you're talking.  This is one good use of the Dhrystone benchmark.
I (we) have found it tremendously useful in evaluating the effects of
little compiler tweeks.  It is a short program, but it has most of the
constructs that will occur commonly in real programs; thus, an optimization
that affects the run time of dhrystone will probably affect the run time
of a real program in a similar way.  This does not mean that dhrystone
will predict the absolute performance of that real program, but it can
predict the relative performance of the two versions (one with the tweek
one without) of that real program.

    bcase
-----------
Watch this space!