fftw is buit-in and included with the Matlab distribution,
ls -al /Applications/MATLAB_R2023b.app/bin/maca64 | grep fftw
which includes single and double precision copies of the library. We can use this to perform substantially faster computations.
So assuming we switch to half-complex, we will drop half the memory requirements.
fftw is buit-in and included with the Matlab distribution,
ls -al /Applications/MATLAB_R2023b.app/bin/maca64 | grep fftw
which includes single and double precision copies of the library. We can use this to perform substantially faster computations.So assuming we switch to half-complex, we will drop half the memory requirements.
We can then also drop to single-precision floating point and drop another half of the memory requirements.
I think this will be the right thing moving forward.