Fastest Fourier Transform in the West

Some of you may have noticed that I've posted a request to bring FFTW into CodeCogs.

FFTW for those you haven't seen it before is developed by MIT as it probably the fastest implementation of Fourier Transforms around. Its used by virtually everyone now, including the giants MATLAB (though it took them 5 years to cotton on).

The speed in the library is achieved using some clever optimization according to your system and a lot of unwrapping of loops etc.

The library is free to download, so simply putting a copy on CodeCogs adds little value. However, I what I've asked for in the request is for someone to produce an FFTW-lite, which we put onto CodeCogs. As such people download this code through CodeCogs will get the benefit of the granular system we have (i.e. get only the code you need and nothing else); however by adhering strictly to their interface, anyone wanting to subsequently link to the real think (compiled and optimized to their machine), can.

Does anyone else have any thoughts on this. Since its and important library and its important we get this right.

I agree it would be good to have a set of multiprecision number theory algorithms that would form the base of all other numerical components on Code Cogs. This would have the advantage of providing an arbitrary number of precision digits in the evaluation of several mathematical functions, at the cost of execution speed.

_____________________________ swe