home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!sun-barr!olivea!spool.mu.edu!yale.edu!qt.cs.utexas.edu!cs.utexas.edu!convex!egsner!adaptex!sdf!bolan
- From: bolan@sdf.lonestar.org (Bolan Meek)
- Newsgroups: comp.os.coherent
- Subject: Re: Poor performance.
- Keywords: Dhrystones, 486, 386.
- Message-ID: <1992Sep5.071531.687@sdf.lonestar.org>
- Date: 5 Sep 92 07:15:31 GMT
- References: <01GOD0KWMK4Y90N81N@delphi.com> <1992Sep4.081336.5972@iitmax.iit.edu>
- Organization: sdf Public Access UNIX, Dallas--unrestricted free shell access
- Lines: 34
-
- In article <1992Sep4.081336.5972@iitmax.iit.edu> pred@iitmax.iit.edu (Predrag S. Bundalo) writes:
- >In article <01GOD0KWMK4Y90N81N@delphi.com> "Coherent operating system" <COHERENT@indycms.bitnet> writes:
- >>Pred Bundalo writes:
- >>
- >>> When I run a dhrystone program on the DOS side, I get 25000 dhrystones/s
- >>> on average. When I run the same program on the COHERENT side, I get a
- >>> horrible 2200 dhrystones/s--consistantly.
- >>
- >>> I'm perplexed. Why would the performance be so poor on the COHERENT side?
- >>> Also, it appears the problem is specific to my system/motherboard since I'm
- >>> somewhat sure I had at least 2000 dhrystones/sec when I used my old 386-20
- >>> motherboard.
- >>
- >>You have to be very careful with benchmarks. One thing I've noticed over
- >>the years is that dhrystone is wonderful for benchmarking compilers, but
- >>it doesn't tell you squat about the speed of the computer. For example,
- [stuff deleted]
- >Agreed. But when I said I had at least 2000 dhrystones with my 386/20,
- >it was under COHERENT v3.1. I used the very same binary under v4.0.1 and
- >I get about 2200 with a 486/33. Which means the performance is about the
- >same regardless of the motherboard. On the DOS side, there was no
- >comparison--the 386 did maybe 5000 dhrystones, while the 486 does 25000.
- >
-
- What's the chance that what you compiled under DOS used the FPU resident in
- the '486, while Coherent's libraries do not? I am betting that if you
- flagged the compiler to produce software emulation of the '87 only, you'd
- get just about the same Dhrystones on the '486 as on a fast '386- with the
- DOS compiler I mean.
-
- So, has anyone built a NPX-accessing math library yet? I have wondered if
- MWC did not supply
- that in their libraries to avoid having to either wrestle with NPX context-
- switches in the kernel or craft an NPX-controlling device driver.
-