FNFT
FNFT copied to clipboard
FastNFT
Fast numerical computation of (inverse) nonlinear Fourier transforms
In: 0.5 release / master branch when compiled, the kdvv c-example (`examples/fnft_kdvv_example.c`) fails: ``` FNFT Error: Invalid argument opts->grid_spacing. in kdvv_compute_boundstates(593)-0.5.0 FNFT Error: Subroutine failure. in fnft_kdvv_base(480)-0.5.0 FNFT Error: Subroutine...
> > > * Unless there is a very good reason too keep it, I would propose to revert commit [d67c4f5](https://github.com/FastNFT/FNFT/commit/d67c4f5ad7f4a60fbe252f1baede49ba5827c9f3) in which code specific to the KdV is added...
> > * commit [dbc8896](https://github.com/FastNFT/FNFT/commit/dbc88963788373d2d45e42cf504bca5c31af9a76): `misc_mat_mul_proto` was not indented to be called directly; instead the idea is to create a new function of the form `fnft__misc_matrix_mult_MxN` (with M,N actual numbers)...
Taking a deeper look into the code, I noticed that eps_t is initialized as https://github.com/FastNFT/FNFT/blob/bcd8767c99088d13c34f7b152ddea5447ee141c7/src/private/fnft__nsep_testcases.c#L187 However, eps_t seems not to be used afterwards.
There are several functions, where variables are declared, assigned a few lines later and not reassigned later on. E.g. the variable `eps_t` in `refine_roots_newton`: ```C REAL re_bound_val, im_bound_val, eps_t; UINT...
This would make calling mex files more efficient as the conversion to the complex array layout where real and imaginary parts are separated could be skipped. https://nl.mathworks.com/help/matlab/matlab_external/matlab-support-for-interleaved-complex.html https://nl.mathworks.com/help/matlab/matlab_external/upgrade-mex-files-to-use-interleaved-complex.html
Make changing FNFT_REAL to single and quad precision types work. Update tests where necessary.