On 29-lug-10, at 00:01, dsimcha wrote:
== Quote from jtravs ([email protected])'s article
dsimcha Wrote:
I'm going to need an FFT library to perform some convolutions at
some point
soon. Two absolute, non-negotiable requirements are that it be
written in
pure D and that it be Boost or compatibly (i.e. zlib or public
domain)
licensed. I also prefer "simple and good enough" over "has every
micro-optimization in the book but a PITA to maintain/modify/use",
as long as
it's at least a true fft as opposed to an O(N^2) DFT. A few
questions:
1. Does anyone already have such a lib?
2. If noone has one I'll probably either write my own from
scratch or port
some code from C if I can find code that's under a suitable
license and
written with a "simple and good enough" philosophy rather than an
"every tiny
optimization in the book" philosophy. Could anyone recommend one
to port?
You should have a look at kissfft. It is very small (one short c
file),
appropriately licensed and supports arbitrary sizes quite efficiently.
I have recently ported this to the go programming language in just
a few hours
work. I'm willing to help with a D version too, if you want, as I
might also need
this soon.
kissfft can be found at: http://kissfft.sourceforge.net
J
This looks great except that the license requires binary
attribution. I've
emailed the author and asked him to waive the binary attribution
requirement and
let me use it under the Boost license for the purpose of a D port.
If he goes for
it, I'll probably just port this code to D for std.numeric, and then
use it where
I need to in dstats. Otherwise, it's back to square 1.
otherwise, as I already said http://www.netlib.org/fftpack/fft.c is
public domain, has reasonable performance, and supports all sizes.
It might not be the always beautiful, but it is stable and works well...
Fawzi
