- parallelizable through image slicing and joining only?
- L2 not parallelizable the same way?
- way of vectorization (along width, height and range dimension) irrelevant?
- fftw3 lib discards redundant part of domain -> size-problems?
- lib different functions for real and complex input -> convertible?
- d_i * d_i_conjugate -> flip imaginary sign for conjugate? (after Fourier Transform?)
- In Sherman Morrison: A == I ?
- Other way to test apart from sumsTo1 ?