To write efficient df1b2variable code for moderate to large size matrices M with y=solve(M,x) one needs to write the df1b2variable equivalent of adjoint code for the LU decomposiiton. This is not too difficult but not the most interesting exercise in the world.