Performance of FMU implementation

[FabiJa]

Performance of FMU implementation

[jaeandersson]

As of right now, the FMI interface in CasADi is really an advanced prototype and essentially unsupported if something goes wrong. In addition to general code immaturity, one fundamental limitation is that it currently doesn't support FMI 3, which means that the adjoint derivatives are calculated quite inefficiently. In general, you will probably always see a large overhead for very small dimensions (such as the bvsm.mo example), but for large-scale, that overhead shouldn't really matter. Being able to calculate multiple derivatives at once (the issue you referenced) would be nice and could give a significant speedup, but it's also not crucial - there are plenty of other ways to get efficiency improvements.

Unfortunately. as of now, the FMI interface is is in a bit of a standstill pending clarity of the funding situation. Hopefully that will be clarified in the next few weeks. There is a chance that the FMI interface will be removed from CasADi and moved into a separate library with a more restrictive licensing for industrial use. Otherwise, I don't see how to justify continuing the development. In any case, I would like to find a way to get the FMI interface mature and more efficient.