Re: [Help-gsl] Requesting feedback on the FFT interface

[Dimitris Papavasiliou]

Brian Gough wrote:
> Dimitris Papavasiliou writes:
>  > 3) Well I suppose there's a good reason for this but the halfcomplex 
>  > arrays produced by radix2 and mixed-radix routines are different. 
>
> The Radix-2 ffts are designed to operate inplace without using any
> extra memory, so the layout is a consequence of that constraint and
> the algorithm.  
>
> If memory is not an issue it is fine to use the Mixed Radix routines
> for size N=2^k to get the same layout.  There is no requirement to use
> radix 2 routines in that case
Well as it turns out that permutation is a major bottleneck anyway.  
Well I should be ashamed to think I could get away with O(N^2) 
complexity because they were "just swaps" (although the swaps were made 
using gsl_vector_swap_elements which probably means two multiplications 
and additions per swap; it could be optimized more).  That was a minor 
point anyway.