Can we do that faster? Maybe via pre-computing and then loading or via Just-In-Time calculation (this might be better for linear interpolation)