[SciPy-User] Can fftconvolve use a faster fft?
Alexander Eberspächer
alex.eberspaecher at gmail.com
Fri Jan 12 14:07:08 EST 2018
On 11.01.2018 22:04, Gregory Lee wrote:
> Also, note that the pyFFTW scipy interfaces default to "threads=1", so
> the monkeypatching as listed in the pyFFTW docs may not give a big speed
> improvement for all transform sizes. It is likely you will get further
> speedup if you monkey patch specific functions using functools.partial
> to change the default threads to a more appropriate value for your system.
Some time ago I've written a small tool [1] which takes care of creating
wrappers around the pyfftw routines (a "wrapper around a wrapper"). The
wrappers are created on module import and inject a number of threads
read from an environment variable.
Maybe you'll find it useful. Please note there's no distutils or
setuptools setup yet, instead a waf-based build described in the readme
is used.
Regards
Alex
[1]: https://github.com/aeberspaecher/transparent_pyfftw
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <http://mail.python.org/pipermail/scipy-user/attachments/20180112/5a45e9d9/attachment.sig>
More information about the SciPy-User
mailing list