Closed JakobAsslaender closed 1 year ago
Base: 66.68% // Head: 66.68% // No change to project coverage :thumbsup:
Coverage data is based on head (
e1c28e7
) compared to base (8d19c14
). Patch coverage: 100.00% of modified lines in pull request are covered.
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
Oh this is ugly but I am pragmatic here. The code still makes sense and it basically says: Ignore all dimensions before the spatial ones. So fine for me.
Hi, it seems that FFTW with the MKL backend does not support calls like
fft!(x, 3:ndims(x))
: https://github.com/JuliaMath/FFTW.jl/issues/252I implemented a hack to circumvent this problem by doing a reshape and then calling
fft!(x, 2:ndims(x)-1)
. If found virtually no speed regression when using the FFTW backend and, well, no error when using the MKL backend ;). (I tested it speed wise with the small example in the test and replace@time
with@btime
for reliable benchmarking.