[Insight-developers] Linear solvers

[kent williams]

For FFTW, we have gone with the method that Luis described: You make the
external library an optional dependency, and only use it if it is installed.
You can look at the CMakeLists.txt files for FFT as a template.  A VXL
version of the FFT is always supported, but it is much slower than FFTW

This makes the licensing issue (mostly) go away, and leaves it to the
library user to decide if they want to go the extra mile to install and
build against the external library.  It also keeps ITK from bloating up with
external packages for which we need to track and integrate changes.  It adds
a layer of complexity to the configuration and build process, but it is
reasonably straightforward.

ITK is as straightforward a resource to use as it can be, given the
complexity inherent in Image Processing.  It's not too much to ask of users
to embrace some of that complexity -- if they can't handle configuring and
building ITK, there's not much chance they'd be able to do any useful work
with it anyway.


On 10/7/07 9:48 PM, "Alexandre GOUAILLARD" <hanfei at caltech.edu> wrote:

> Hi guys,
> 
> We have implemented several mesh parameterisation algorithms. Those
> algorithm requires solving quite big linear systems, and we have been
> playing wth different solvers out there: VNL, TAUCS, MUMPS, of which only
> VNL is included in ITK. Unfortunatly, that's also the slower.
> We have designed the code to be solver-templated. By defualt the solver
> traits will be the VNL,
> 
> 
> but we are wondering if it would be too much of an effort to include TAUCS
> in ITK? The performances are really better. It sems that the licence would
> allow it, and it seems to be cross plateform enough:
> http://www.tau.ac.il/~stoledo/taucs/
> 
> Alex. 
> 
> 
> _______________________________________________
> Insight-developers mailing list
> Insight-developers at itk.org
> http://www.itk.org/mailman/listinfo/insight-developers