Fast Fourier Transform (SSE3)

[baltoro]

...eigenvectors and eigenvalues,...

:green Good thing you warned us...I was just getting juiced up for a long, incredibly annoying rant about the unconstrained usage of eigenstuff,... :green
For people like me,...any transformation beyond what you would use to balance your checkbook, is inherently destabilizing (in respect to the functionality of my cerebtal cortex).

[vanjast]

I started implementing a radix-4 FFT and inverse FFT to compute some crazy correlation function before I realized that for the particular thing the correlation function was in turn being used to compute, it really only needed: bit-shift, XOR, sum the bits, repeat a bunch of times.  That ended up being over 10x faster, whereas the vectorized FFT would've only been 4x faster.

I was thinking along these lines.. but i see you've beaten me to it.
After looking at the FFT algorithm's implementation(not yours Red), I always thought that it was a bit cumbersome, and could be improved...
I'll tweak later.
:bg

[vanjast]

You could speed it up by making the Bit Reversal section seperate from all the FFTs..  :green2

[vanjast]

I've finished a version of FFT's based on FPU instructions (non mmx, SSE).
All these are direct translations of Steve Smiths 'Basic' Algorithms.. I'll post these later  :bg

[vanjast]

Here we go...
I haven't checked, optimised or run this yet, it's currently part of a bigger project so I'll only test this later.
There's probably a few things still to do..

[vanjast]

I was going over this project today... and picked up a few mistakes with the FFT module, just a few incorrect store instructions that should be FIST(P) instead of FST(P) when saving to integer formats.. plus a few other funnies. Fixit later.
:8)