Benjamin von Ardenne

The only single thing that I don't understand here is the problem that after initialization of gsl-shell the very first num.fft transform gives zero output and transforms the input to...

Excellent points. So I guess we should make a "standalone" version of the fft functions with FFTW_ESTIMATE which would then also operates with the same performance as the original gsl...

Hehe, sure, just didn't want to work on in if you are reworking the code at the same time. So my understanding is, for normal fft calls we use ESTIMATE...

In the current implementation `fftw_execute_dft` (as a new array execute function) is already used when calling the `fft_plan` functions. So its not that I haven't thought of reusing the plans...

- `fftw_alloc` - agreed and **check**. - caching: **agreed** to cache used plans and reuse if size and dimension fits. Also distinguish between forward and backward in addition to r2c,...

Hey Francesco, I have rewritten the whole fftw interface and reduced it down to three fundamental functions `dft`, `rdft` and `rdftinv`. The `dft`-function is the most general, handling only complex...

Remark to the fact that the output location can be given as a paramter: This significantly increases performance: ``` bash time gsl-shell -e "test=matrix.cnew(256,1,|i,j|i-1); for i = 1,1e6 do bla=num.fft(test)...

Hey, I will look into it. I did not fix the fftw_malloc yet because we wanted to integrate it into the matrix.lua but maybe that is the source of the...

Hey Francesco, I am happy you like it. So I looked into FFTW Advanced Interface which allows to give the stride as well. Now I am giving `tda` of the...

So it is strange: ``` lua c=matrix.cnew(4,4,|i|1) num.fft(c:slice(1,1,4,1), c:slice(1,1,4,1)) ``` gives no error when exiting gsl-shell and shows cool inplace fft transformation while ``` lua c=matrix.cnew(4,4,|i|1) num.fft(c:slice(1,1,4,1)) ``` gives the...