Re: implement a SOR like routine efficiently

[Miroslaw Kwasniak]

On Tue, Nov 16, 2004 at 10:40:17AM -0500, Przemek Klosowski wrote:
> 
> Well, as written, the loop is not vectorizable; but the expression is
> suspiciously close to an 'averaging' kernel. I just wanted you to
> confirm that a convolution of a and [0 1 0; 1 0 1; 0 1 0] would not be
> an acceptable solution:
>               
>       a(2:n-1,2:n-1) = conv2(a, [0 1 0; 1 0 1; 0 1 0],'valid');

I though yesterday about this solution but Ganesh operator is strange
diagonal symmetric. 

In http://www.zind.ikem.pwr.wroc.pl/arch/octave/sor/ you can see my try to
visualise Ganesh `sor' for 9x9 case: 

- animated image sor99.gif
- script sor.m wich did image

Each frame represent walk trough indices i,j=1:9 of matrix SOR(i,j,9,9) of
operator coefficients. 

Asumming that b=sor(a) in my representations b(i,j) = sum(sum(SOR(i,j,:,:).*a))

Mirek



-------------------------------------------------------------
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
-------------------------------------------------------------