Dear GIMPSers:

I am pleased to announce Mlucas 2.7b (source code and binaries for
selected platforms) is available. See

ftp://hogranch.com/pub/mayer/README.html
ftp://hogranch.com/pub/mayer/gimps_timings.html

(768kbps capacity) or the mirror at

ftp://209.133.33.182/pub/mayer/README.html
ftp://209.133.33.182/pub/mayer/gimps_timings.html

(160kbps capacity).

The major differences from v2.7a are:

1) New FFT radices, including ones of the form (11,13,15)*2^n. Thus
every interval between adjacent power-of-2 runlengths can be split
into roughly eight equal pieces. Folks testing exponents > 12.59M
and double-checking exponents > 6.39M will (on all but a few of the
older platforms) benefit from the new radix-11 capability. There are
also new larger power-of-2 radices (32 and 64), which allow even very
large runlengths (up to 2^23) to be handled with as few as 6 passes
through the data.

2) Runtime performance optimization for the user's platform. The
program now allows the user to choose from a reasonable set of possible
radices which can be used at each runlength by changing a single entry
in a configuration file. This can be done without interrupting the
program - the file is reread each time a 2000-iteration checkpoint
is reached, and if the prescribed radix set is different from the one
currently being used, the code switches over and prints the actual
new radices being used to the .stat file for the exponent in question.
The elapsed time written to the .stat file at the next checkpoint
can then be used to find the optimal radix set for the current
runlength. Preoptimized configuration files for most major hardware
setups on which the program has been tested accompany the precompiled
binaries, so folks who don't want to fiddle around with this feature
don't have to. Folks who do the runtime tuning will often be rewarded
with a healthy speedup.


If you peruse the timings page, you'll also see that Mlucas now has
competition for the hearts and minds (well, mainly the CPU cycles :)
of the RISC crowd. Guillermo Ballester Valor's excellent C program,
Glucas, is showing excellent performance on most platforms, and even
beats Mlucas on some. Mac/PowerPC users will be especially interested
in Glucas, since it outperforms MacLucasUNIX except for a few narrow
ranges of exponents, and (IMO more importantly) is being actively
maintained and improved. Right now Glucas has only been compiled
using the CodeWarrior C compiler (which I believe has the Gnu C
compiler at its core - correct me if I'm wrong), and we're looking
for folks to compile and do timings on various generations of the
PPC chip, and also who have other compilers, such as the Absoft
suite (whose Fortran-90 compiler could also be used to compile
Mlucas for Mac.)

Anyway, the above links have all the details. Right now I'm really
bushed, so please excuse me while I go sleep for about 12 hours.

Happy hunting,
-Ernst


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

Reply via email to