Skip to content
Prev 132431 / 398506 Next

strange timings in convolve(x,y,type="open")

Thanks Charles.  That must be it.  (Berwin also
noticed this.)

When convolve hit the wall, I switched over
to FFTW in C.  That is actually pretty nice
code which runs in n log(n) even for prime n
and takes special account of factors of n up
to about 19 or so.  So if the R team ever wants
to put in a new FFT, that looks like a good one.

But I think easier fix for me, or others in this boat,
would just be to write a new convolve(x,y) that
pads x and y with zeros out to length
2*max( length(x), length(y) ).  Then if x and y
have very composite lengths, especially powers of
2, the fft should go quickly.

-Art



Quoting "Charles C. Berry" <cberry at tajo.ucsd.edu>: