I actually was thinking about this. I could see how to easily
implement the multiplication of two matrices by say breaking up each
matrix into sub-matrices and then letting different nodes multiply the
sub-matrices together and then combining them at the end. However, to
do matrix inversion (or Gaussian Elimination like the backslash
operator) I cannot see how it would be that simple. I do not believe
you can split up a matrix to invert it as easily, but perhaps I am
wrong. Maybe I am just confused about using mpitb, it seems like you
must find a way to efficiently break up a specific operation.
Thanks, Mike
Date: Mon, 15 Oct 2007 09:41:02 +0200 From: address@hidden To: address@hidden Subject: Re: Matrix Multiplication in Parallel in Octave CC: address@hidden
It's true that MPITB doesn't automatically parallelize things for you. It gives Octave access to MPI functions. Using MPITB, it would be easy to write a function like c=p_multiply(a,b) where a and b are matrices to be multiplied.
Michael
On 10/12/07, Peter A. Gustafson <address@hidden> wrote:
On Friday 12 October 2007 02:26:01 pm mrober01 wrote: > Hi. > My goal is to do matrix multiplication / inversion using Octave in
> Parallel. I actually ended up installing mpitb, but have learned that it > does not actually parallelize operations, it must be done manually. I have > been told that perhaps ScaLAPACK might work to parallelize these matrix
> multiplication / inversion operations. Any ideas? > > Thanks, > Mike
I have not experience with it but this was in my bookmarks
http://atc.ugr.es/javier-bin/mpitb
I hope it helps, Pete _______________________________________________ Help-octave mailing list address@hidden
https://www.cae.wisc.edu/mailman/listinfo/help-octave
Windows Live Hotmail and Microsoft Office Outlook – together at last. Get it now!
|