Possible way to speed up the Spline interpolation

[leouieda]

Just came across this paper: https://doi.org/10.1016/0898-1221(92)90167-G

It claims to reduce the forward modelling cost by a lot and allows doing matrix vector products using only O(N) storage. But it doesn't seem trivial to implement and would require a complete rewrite of much of our spline code.

Worth investigating, though!