Re: MPI (was Re: BIG libraries with self-compiled version)

[James R. Phillips]

--- Michael Creel  wrote:

> Peter Soendergaard wrote:
> 
> > It is possible to compile ATLAS exactly for use on multi-cpu shared 
> > memory machines. The code uses pthreads. I have not done it myself, 
> > but it should say in the documentation how to do it.
> >
> > This should be a very easy way to use more CPUs. I think that simply 
> > linking Octave to a pthreads enabled ATLAS library would be 
> > sufficient. The interfaces should be the exact same as the single 
> > processor library.
> >
> > Cheers,
> >
> > Peter.
> >
> Hello all, of the suggestions made, which all sound pretty reasonable, 
> I'm planning on trying this one. Before I do though, has anyone built 
> Octave using a parallel ATLAS?  Does parallel ATLAS break operations 
> into a number of threads that depends on the size of the problem, or is 
> the same number of threads always used? Well, I hope to have some 
> results to report in a week or so. M.
> 

Well, atlas does take a long time to compile, but probably not a week!

Seriously though, I have configured, compiled, and installed a 2-cpu enabled
atlas on octave-2.1.71/cygwin, and it worked fine on a 2 cpu system.

The lapack source package on cygwin contains atlas, as well as instructions for
building and installing an optimized lapack/blas.

I believe you could do something similar with the Debian atlas source package.

Have fun!

Jim Phillips
volunteer octave (and lapack) cygwin maintainer




-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------