dolfin team mailing list archive

[Robert C. Kirby <kirby@xxxxxxxxxxxx>]

On Nov 10, 2005, at 9:38 AM, Johan Jansson wrote:

> On Thu, Nov 10, 2005 at 09:05:57AM +0100, Johan Hoffman wrote:
>
> ...
>
> > Interesting. How come that assembling the matrix is 3-10 times as 
> > slow as
> > "assembling" the action? Is it only due to the overhead of managing 
> > the
> > sparse structure?
> >
> > /Johan
>
> Yes. The only real difference is the destination of the computed
> values. In the matrix case they are put into a sparse matrix and in
> the action case the appropriate values are summed together and put
> into a vector (together with coefficient values for the action
> operand).

Anytime you work with assembled sparse matrices, you are 
bandwidth-limited.  However, it seems many popular and effective 
preconditioners require the matrix.  If it weren't for this, nobody 
would ever use matrices.  Matt & Barry are on a crusade to eliminate 
them.

>   Johan
>
>
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/cgi-bin/mailman/listinfo/dolfin-dev