FFT gets very expensive when the number of reference points grows too much, but between 5 to 50 reference points its doable on >current cpus. (not sure about those numbers, could be a lot more/less)
did my homework. Its N*log(N), and is significantly faster than what I thought - the fortran lib takes 1us per 64bit precision unit, 100 us per 2048 bit precision on a 2ghz pentium IV. All things considered, we could use it even if we had 512 references per node :)
So after the current NG code is out of experimental, I'll make an implementation that uses FFT, and lets see what happens :))
_______________________________________________ devl mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
