home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!copper!aspen.craycos.com!jrbd
- From: jrbd@craycos.com (James Davies)
- Newsgroups: comp.lang.fortran
- Subject: Re: Compiler groups working on real apps?
- Message-ID: <1993Jan25.171544.14680@craycos.com>
- Date: 25 Jan 93 17:15:44 GMT
- References: <C19wHr.GF3@news.cso.uiuc.edu> <1993Jan23.003639.13681@craycos.com> <C1Cnq5.1zt@news.cso.uiuc.edu>
- Organization: Cray Computer Corporation
- Lines: 32
-
- In article <C1Cnq5.1zt@news.cso.uiuc.edu> ercolessi@uimrl3.mrl.uiuc.edu (furio ercolessi) writes:
- >In article <1993Jan23.003639.13681@craycos.com>, jrbd@craycos.com (James Davies) writes:
- >|>In article <C19wHr.GF3@news.cso.uiuc.edu> ercolessi@uimrl3.mrl.uiuc.edu (furio ercolessi) writes:
- >|>
- >|>>This is probably too much to ask. But I am wondering. Maybe any one of
- >|>>us (users) could spend some time in putting together some "typical" code,
- >|>>containing a simplified version of the loops where our programs spend
- >|>>most of the time.
- >|>
- >|>You've just described the Livermore Loops.
- >
- >After seeing this comment, I fired up xnetlib, downloaded the Livermore
- >loops and looked at them. I must admit I was not familiar with them.
- >While there is certainly an interesting variety of codes in there (there are 24
- >different 'kernels'), at a first glance I was not able to locate something that
- >I feel represents the kind of codes I am dealing with.
- >
- >My applications involve classical molecular dynamics (i.e., interacting atoms
- >treated like classical masses and moved by time integration of F=ma).
-
- My previous answer was a bit flip, I suppose, but basically there have been
- lots of benchmarks floating around for decades now, and they tend to be
- used primarily for marketting ("New XYZ Workstation Achieves 250 SPECmarks!!")
- or as argument fodder ("Yeah, but they're using SPEC88, which was too easy
- to cook by using a smart compiler! They only get 127 on the SPEC93 suite,
- but our new ABC9000 gets 172 SPEC93marks!"). For what it's worth, the
- Perfect Club suite is a collection of application programs that was supposed
- to avoid compiler cheating, and also help advance compiler technology by
- encouraging sharing of techniques, but since the results don't boil down to a
- single number that can be easily quoted by marketroids most everyone basically
- ignores it. Perfect includes one code that is a molecular-dynamics
- simulation (MDG).
-