[SciPy-User] Single precision FFT insufficiently accurate.

Pauli Virtanen pav@iki...
Mon Jun 28 03:29:29 CDT 2010

Mon, 28 Jun 2010 11:30:55 +0900, David Cournapeau wrote:
> Maybe it is better to remove it altogether in 0.8.0 - I will try to
> implement Bluestein algo during euroscipy,

Yes, I guess removing the support altogether is the second alternative 
for 0.8.x. It's not really clear to me which is more useful for the user 
-- a partially functional feature, or none at all.

Ok, and the point that the casting can cause problems because of large-
sized axes orthogonal to the FFT direction is taken, I didn't think about 
that. If we want to evaluate the FFT in blocks in such cases, that 
probably can be done but is a bit fiddly to get right.

Pauli Virtanen

More information about the SciPy-User mailing list