Jason Stratos Papadopoulos wrote:

> Hey everybody. Now that our Fermat testing is starting to wind down
> I've decided to make available the source I've written for it.

Cool!

> www.glue.umd.edu/~jasonp/f24v131.zip
> 
> The code there is heavily optimized for the UltraSPARC processor,
> and includes gobs and gobs of sparc assembly language. Squaring
> times are as follows:
> 
>                 ultra-1   ultra-1  ultra-2i
> number   bits    143MHz    167MHz   300MHz
> 
> F22      4.2M    .229     .196     .167
> F23      8.4M    .465     .404     .322
> F24     16.8M    1.08     .885     .700
> F25     33.6M                      1.58
> 
> The GIMPS speed page places the Ultra-2i times in the neighborhood
> of a PII-300 (a Pepin test takes about the same time as an LL test
> of similar size, maybe a little less).
> 
> So, a few questions:
> 
> How do these times compare with MacLucasUnix on an Ultra, or with
>     Prime95 for the really big sizes?

Today I posted some times - I was getting approx 0.4s per iteration for
a 256K element FFT (testing in the 4.6M range) and 0.22s for a 128K FFT
on a 200MHz Ultra.

> How fast does the program run on a really big Ultra, i.e. a 440MHz
>     server with piles of RAM and a 4MB cache?

I've only got up-to-200MHz Ultras :-(

> Are there enough Ultras out there to justify making a fast LL test out
>     of this code? I'd be willing to try, but only if there's sufficient
>     interest; the code won't run on normal sparcs, my time is very
>     limited I don't want to expend a considerable amount of
>     energy on something no one will use. Would anyone like to help?

I'd be interested in helping - I think I can see the gist of what
you are doing at first glance.  A whiles back I wrote a very simple
implementation of the LL test using FFTW, so I should be able to use
that as a framework.  It's at
        ftp://melanoma.cs.rmit.edu.au/pub/simonb/fftwll.tar.gz
if anyone's interested...

While on the subject of FFTW, I tried to use a multithreaded FFTW but
came up with the wrong results (M1279 wasn't prime :-( ).  The aim of
the exercise was a fast double-checker - imagine throwing a 64 CPU Sparc
E10000 at a double checking exercise next time we want to double check
a number quickly!  Has anyone used FFTW's threads?

Simon.
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to