Assistance sought with mondo numerical app

[Fernando Pérez]

D. Michael McFarland wrote:

> Dear Fellow Pythonistas,
> 
>     A friend of mine is in the throes of finishing his
> Ph.D. dissertation, comparing experimental and numerical results for
> material damage caused by grinding.  For the numerical part, he's
> using a finite element library I wrote in Python.  This library made
> programming the tensor operations much easier than they might have
> been in, say, Fortran, but the end result takes rather longer to run
> than one could hope--about a day per problem case on a PII-450.  He
> has two machines of this class available, and I have two, and that
> might just be enough to let him make his deadline, but it occurred to
> me where to find more machines: this newsgroup.
> 
>     In short, we want your flops.

Have you looked at the possibility of rewriting some of the tight inner loops 
in C? These days using weave (http://scipy.org/) it's perfectly reasonable to 
get large speedups with very little work. I don't know your code so I don't 
know how realistic this is (and you guys seem to be under tight deadlines), 
but you can look at this pdf: 
http://www-hep.colorado.edu/~fperez/python_misc/python-c.pdf
for a simple example where the speedup is a factor of 25.

Just an idea. Sorry but the cycles on our machines are tied up 24/7 by our own 
numerical codes, so I can't help on that front.

Cheers,

f.