GL

avoid expensive matrix update

profiling showed that update_matrices has a huge impact on performance. It's better to avoid updating the forward and backward transformations and computing them on the fly instead.

See merge request !30