[SciPy-Dev] Accuracy of single-precision FFT

Pauli Virtanen pav@iki...
Fri Jun 25 04:02:02 CDT 2010


Fri, 25 Jun 2010 10:47:32 +0900, David wrote:
[clip]
> Yep, the bluestein algorithm. I thought fftpack in scipy did this, but
> it actually does not (I tested size 4096 vs 4093, one is 100 times
> slower than the other).
> 
> So the solution would be to implement this (for both numpy and scipy). I
> won't be able to do it for 0.8, though.

Ok, so it seems clear we can't fix the accuracy issue fast, and should 
not wait for it for 0.8.

Now the question is whether the accuracy problem is severe enough so that 
we should disable the single-precision transforms in this release.

If we are unsure that it's not severe, I think we should disable the 
support for single precision FFT in the 0.8.x branch, and remove the 
corresponding mention from the release notes.

Like so:

	http://github.com/pv/scipy-work/tree/disable-single-precision-fft

-- 
Pauli Virtanen



More information about the SciPy-Dev mailing list