Re: Fastcode MM: category weights

I agree that no single MM is the fastest in every situation. However, the
results of the MM challenge are presented as if there was only one winner of
the race. This is exactly the reason why I thought it could be useful to
try to figure out broad categories where different programs show their
strenghts (if such a characterization was possible) or at least broad
categories of applications (single and multi threaded is, in my opinion, a
good way to separate various routines).

> Is it simple enough to be included as a benchmark?

It is simple enough. A single small unit and a few lines of codes that
simulates the indexing of 10,000 documents or words and the classification
of a few hundred ones based on the computation of the similarity of each
document to all those indexed document. However, since it will be part of a
commercial application, we cannot release the full source code easily. I
already offered this unit, but with a condition that it would be made
avaibled in a compiled DCU form. (I can provide a D7 unit and the source to
someone so that he can compile it for different version version of Delphi
and I would have no objection to let the authors of different MM take a peek
at it, but I don't think it would be possible to obtain the authorization to
have its source releases. My boss already contacted Dennis about this
possibility and was told that it would not be possible to include a compiled
unit without the source.

I tried to compile the benchmark on my D7 but was unable because of some
missing parts, so I could not familiarize myself it. I am also not familiar
with the Usage Replay Log, but will check this. Actually, I think I used it
before and end up with a huge file for a single run.

Richard


"Robert Houdart" <robert@xxxxxxxxxxxxxx> wrote in message
news:427cdacd$1@xxxxxxxxxxxxxxxxxxxxxxxxx
> Hi Richard,
>
> No single MM is the fastest in every situation, and it's rather difficult
> to predict which MM will be fastest on a given application. The only way
> to tell is to plug in the MM and see for yourself. By all means you should
> get very good results with all the top performers in the challenge.
>
> BTW, your document indexing routine would be a very welcome addition to
> the B&V tool. Is it simple enough to be included as a benchmark? If not,
> could you send us a usage replay log?
>
> Regards,
> Robert
>


.

Relevant Pages

  • Re: VIA C3
    ... do you have a suggestion how to benchmark it? ... Other than that, time a buildworld/buildkernel. ... Doing the buildworld/buildkernel will probably be more intensive -- compile ... it with k6-3 flags, reboot, then do another compile with the same flags. ...
    (freebsd-current)
  • Re: dual G5 vs AMD Sempron
    ... >I cobled together a harness to an MPE benchmark file so it would compile ... >Ran the benchmark code with gForth-fast on Windows XT on my cheap Sempron ... For Gforth running code that fits in the ...
    (comp.lang.forth)
  • Re: crypto benchmark results with 2.6.0-test6
    ... >>would you like to benchmark ... one had to use a special compiling environment to compile a kernel? ... Probably you can one of IBM's compilers to compile the cipher of your ... send the line "unsubscribe linux-kernel" in ...
    (Linux-Kernel)
  • Re: Code Optimization - AnsiStringReplace
    ... >would not compile because of an asm syntax error, ... I have a modified version of the benchmark that compiles in D5 - had to ... Gallery: http://w1.161.telia.com/~u16122508/gallery/index.htm ...
    (borland.public.delphi.language.basm)
  • Re: Nothing to do with PG [was: The Purl Gurl problem [was: Re: Many]]
    ... Donna at least) and I responded that I didn't compile the list and ... I was idly googling one of the phrases from the "History" I posted on ... one Richard Lederer of St Pauls School, Concord, N. H.. ... receiving the occasional jewel of a student blooper in an essay. ...
    (alt.usage.english)