> Hi,
>
> > -----Original Message-----
> > From: kifer-EX0cT3Az47bauI2f2gSDlQ@xxxxxxxxxxxxxxxx
> > [mailto:kifer-EX0cT3Az47bauI2f2gSDlQ@xxxxxxxxxxxxxxxx]
> > Sent: Sunday, October 07, 2007 12:51 PM
> > To: PhiHo Hoang
> > Cc: 'Terrance Swift';
> > xsb-development-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@xxxxxxxxxxxxxxxx; flora-
> > users-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@xxxxxxxxxxxxxxxx
> > Subject: Re: [Flora-users] Benchmarks
> >
> >
> > > Hi,
> > >
> > > >
> > > > The overhead for HiLog in the loss of some indexing.
> > >
> > > Thanks for the insight.
> > >
> > > > If XSB supported multi-argument indexing for compiled code
> > > > then the overhead would have been minimal.
> > > >
> > > > --michael
> > >
> > > What would be the price that non-hilog XSB programs must pay for this
> > support?
> >
> > I do not understand what you are asking here.
> > If you get somebody to implement multi-argument indexing then that would be
> > the price :-)
> >
> >
>
> Yes, and that would be a high price to be paid for a feature that is used by
> less than a handful of users(???) that can be counted with the fingers (of 1
> hand ;-)
>
> HiLog users can be content with Flora-2, trading execution performance for
> convenience and expressiveness and praying for bigger successes from AMD and
> Intel in their innovations on their multi-core chips.(I used to think that
> it's
> cheaper to innovate in software than in hardware ;-)
Well, you would be surprised. But it works quite well already. It is much
more important to be able to index facts than rules for us.
> > > > David and Kostis Sagonis wrote a paper on Hilog computation for the
> > > > 1995 ICLP (I think), and they might have some measurements in there.
> > > >
> > > > Terry
> > >
> > > If anyone has this paper, please allow me to peek at.
> >
> > http://citeseer.ist.psu.edu/sagonas95efficient.html
> >
> >
>
> Actually, my real question was that what are Flora-2 users paying for the
> price
> of HiLog feature and many thanks for the link to that paper.
>
> According to the above cited paper, "specialised" HiLog and prolog programs
> execute at the same speed.
>
> To quote, "In XSB, HiLog using specialisation clearly outperforms the set of
> generic predicates; HiLog programs run between 7 and 8 times than generic
> Prolog
> programs. Furthermore, compared to the specialised generic HiLog predicates,
> the
> non-generic Prolog predicates execute only 10-13% faster."
>
> This 10-13% is more than compensated for with efforts from AMD and Intel ;-)
>
> I wholeheartedly agree with the conclusion from the authors of that paper:
>
> "... HiLog, Prolog programmers would rightfully demand future logic
> programming
> systems extended to support HiLog functionality."
We did not find that specialization works for us. Perhaps for some
programs, but not for our benchmarks.
In fact, for the combination of F-logic and HiLog the gain should be
minimal. So, Flora, in fact, disables specialization due to some other
negative considerations.
>
> Cheers,
>
> PhiHo.
>
>
> P.S: and this was written more than a decade ago in the last millennium.
>
>
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems? Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/
|
