dolfin team mailing list archive

[Robert C.Kirby <kirby@xxxxxxxxxxxxxxx>]

On Oct 26, 2004, at 9:05 AM, Matthew Knepley wrote:

> "Robert C.Kirby" <kirby@xxxxxxxxxxxxxxx> writes:
> > Of course ferari << quadrature, but if even quadrature << solver, why 
> > bother optimizing it?
>         I wouldn't. I msut be getting confused with the discussion.

The whole point of the ferari paper is how quickly you can build local 
stiffness matrices.  I think the attitude I have detailed above (local 
computation/assembly << solve) is mainly true for:
1.) suboptimal solvers
2.) Low degree (say linear)
3.) Poisson

It's not as clear when you get to multilinear forms (e.g. 
Navier-Stokes) what the relative costs are.  Also, the better your 
solver is, the more likely you are to notice assembly time.

> > Both/and.  First, there is the mathematical issue (this is tantamount 
> > and has the biggest impact).  But even if you have
> > an optimal preconditioner, a crappy implementation of the 
> > matrix-vector product (e.g. (i,j) index storing by unsorted
> > lists), you will have a huge difference in performance among 
> > implementations.
>         I wasn't allowing for anything that crappy. Most common 
> implementations
> are fine, but someone could still screw it up. An excellent arugment 
> for using
> existing libraries.

My example was of course for the sake of argument and to make exactly 
your point that people have done the work already so we shouldn't worry 
about doing a crappy job.