[Haskell-cafe] What is a functional approach to linear algebra routines? I see REPA warns that one must use fusion or risk slow routines.