[SciPy-Dev] Accuracy of single-precision FFT

[Sturla Molden]

Den 24.06.2010 19:41, skrev Anne Archibald:
> We can't use FFTW, unfortunately, but it's clear that prime-size FFTs
> can do much better than we are.


FFTW and FFTPACK are not the only options.

We could use Intel MKL, not just for FFTs but also BLAS. It's not worse 
than asking Intel to sponsor NumPy/SciPy with an MKL license. That would 
allow redistribution of a binary build linked against MKL, similar to 
what Cristoph Gohlke and Enthought are doing:

http://www.lfd.uci.edu/~gohlke/pythonlibs/ 
<http://www.lfd.uci.edu/%7Egohlke/pythonlibs/>
http://enthought.com/

We can also use OpenCL for those that have GPUs available:

https://developer.apple.com/mac/library/samplecode/OpenCL_FFT/Introduction/Intro.html
http://www.cs.berkeley.edu/~kubitron/courses/cs258-S08/projects/reports/project6_report.pdf 
<http://www.cs.berkeley.edu/%7Ekubitron/courses/cs258-S08/projects/reports/project6_report.pdf>
http://forums.nvidia.com/index.php?showtopic=102635


Sturla