Reduce memory usage in FFTs
Opened this issue · 1 comments
djps commented
Is your feature request related to a problem? Please describe.
Post-processing currently always auto-casts to double precision for methods that use scipy.fftpack, e.g. extract_amp_and_phase
. This can lead to OOM needless errors e.g. there have been cases where the simulation has run, but I have been unable to take the Fourier transform as I run out of memory.
Describe the solution you'd like
- use
scipy.fft
instead ofscipy.fftpack
ornumpy.fft
as this does not automatically cast to double precision - ensure that
get_win
andextract_amp_phase
keep the same precision as the data which is supplied.
Describe alternatives you've considered
The new numpy 2.0 will have this capability, as it also upcasts, see issue.
waltsims commented