[Shootout-list] Organisation for numerical tests

[Jon Harrop]

On Wednesday 04 May 2005 12:24, Sebastien Loisel wrote:
> > >    5. LU - computes the LU factorization of a dense N x N matrix
> >
> > Again, I have the practical concern that nobody would want to implement
> > these I have the theoretical concern that the performance measurements
> > from these
>
> I do it all the time. It's not always easy to find an LU code that
> works on interval arithmetic. The performance is meaningful.

Ok.

> > > * Conjugate Gradient
> >
> > It would be interesting to see this coded in different languages but, in
> > terms of performance, you'd either be measuring the time taken to
> > evaluate the function(s) or measuring the difference in performance of
> > different algorithms.
>
> I meant as a linear solver. All you need is fast matrix-vector product
> and vector operations. These are used mainly for solving linear PDEs
> like the Laplace problem above.

What would the benchmark be measuring the performance of?

> > > * Automatic differentiation
> >
> > What exactly do you mean by this?
>
> http://www.autodiff.org/
> http://www.math.mcgill.ca/loisel/ad-matlab.pdf

I like the paper and this certainly sounds very interesting but I'm not sure 
how this can be made into a shootout benchmark?

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
Objective CAML for Scientists
http://www.ffconsultancy.com/products/ocaml_for_scientists