[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]
Re: Impossible
- From: "Adam C. Powell, IV" <adam powell nist gov>
- To: axp-list redhat com
- Subject: Re: Impossible
- Date: Wed, 09 Sep 1998 10:50:43 -0400
Mark Dohojda wrote:
> Hello everyone
>
> I have an Impassable situation here (or so I believe). I have a
> customer here who just installed Linux on alpha LX600mhz, 512mb of RAM.
> He run C program in NT and in Linux (on the same box), and NT proved to
> be 30% faster. Now I don't know much about the code, however I know
> that he used the same sources on both of the operating systems. Also
> the source (to the best of my knowledge) was in no way optimized for NT.
>
> My question is, how can NT beat Linux????? Is that possible? And what to
> do, to make sure that Linux is not going to look bad.
>
> He has RedHat 5.1 no special editions or patches other then the man
> pages patch.
What kind of program is it? If it has a lot of math it might be sped up
quite a bit using the free fast math library libffm of Wesner and Goto at
http://people.frankfurt.netsurf.de/Joachim.Wesner/ . A friend of mine had
a program which was heavy in exp and sqrt, and this library sped up its
runtime from 380 seconds to 98!! That's an extreme example, but just shows
it can make a substantial difference.
One tip- if that library causes floating exceptions even with -mieee (due
to underflow in exp), try removing exp from libffm, building libfm, and
linking -lffm -lfm -lm. (I had to do this for my friend's program in the
above example.)
Zeen,
-Adam `Cold Fusion' Powell, IV _/\_ http://www.ctcms.nist.gov/~powell/
Call with Netscape Conference! \/ http://dls.netscape.com/ ____
USDoC, National Institute of Standards & Technology (NIST) |\ ||< |
Center for Theoretical and Computational Materials Science | \||_> |
[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]
[]