[racket-dev] expected timing difference between fft-complex-forward and fft-complex-radix2-forward?

From: John Clements (clements at brinckerhoff.org)
Date: Thu Oct 14 18:34:50 EDT 2010

On a vector of length 8192 (a power of 2, natch), fft-complex-radix2-forward takes about 1/8 of a second (lost in the noise, essentially), but fft-complex-forward takes more than 24 seconds, a difference of about 200x.  They do produce the same answers, up to differences of 1e-4 in the magnitude (didn't check the phase).

1) Is this expected? I thought the non-radix2 one was still fairly clever about subdividing when the number of points is divisible by 2.
2) If so, would it make sense to test for powers of 2?

John Clements

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4669 bytes
Desc: not available
URL: <http://lists.racket-lang.org/dev/archive/attachments/20101014/c35d4b77/attachment.p7s>

Posted on the dev mailing list.