Re: [eigen] A complex FFT for Eigen

[ Thread Index | Date Index | More Archives ]


I think we now know where we are going, so we can start adding an
experimental FFT module to Eigen.

Let's summarize the goals:
- generalization to large/dynamic sizes, with an optimized fixed-side kernel
- allow optional FFTW backend. This must be thought of early in the
process as it impacts the API
- replace for loops by Eigen expressions as much as possible

Tim, are you willing to do this yourself? You could use an existing
Eigen module, like Regression or LU, as a model. We could help you
with any Eigen-specific aspect. One important thing is that the API,
the naming of variables, the coding style are uniform with the rest of
Eigen. If you do it yourself, send your first patch to this list
(generated by svn diff) and then you'll be able to apply for your own
KDE SVN account. I'd like the above 3 goals to be met before I merge
this new module.

Otherwise, i'll do it when I have time, but that may be not soon as
there are other things to do.


2008/11/30 Tim Molteno <tim@xxxxxxxxxxxxxxxxxxx>:
> On Sun, 30 Nov 2008, Benoit Jacob wrote:
>> Hehe yeah, it's a very strange aspect of C++ syntax that we too
>> struggled with when we first encountered it...
>> it the faulty line
>> decimation.apply(data.segment<N>(0),-1);
>> the problem is that data itself is of a templated type, so when you
>> call the templated method segment<int>(int) on it, you need to add the
>> "template" keyword, like this:
>> decimation.apply(data.template segment<N>(0),-1);
>> Otherwise the C++ compiler doesn't understand that the < sign means
>> template parameter, and tries to interprete it as operator<.
> Excellent! and thank you for your help. I have it working now and have
> attached the most recent version to this email. I have a small standalone
> testharness here as well.
> Cheers
> Tim


Mail converted by MHonArc 2.6.19+