Fortran elemental functions vs elemental subroutines

Fortan allows elemental subroutines to have intent(inout) and intent(out) arguments, but elemental functions are only allowed intent(in).

Why is that? Is it just a stylistic convention, or is there something generically different about invoking functions and calling subroutines?

In other words,

Elemental Integer Function FOO(i)
  Integer, intent(in) :: i
    ...
  FOO=something
End Function

and

Elemental Subroutine FOO(i, v)
  Integer, intent(in)  :: i
  Integer, intent(out) :: v
    ...
  v=something
End Subroutine

— are these implementations of FOO equivalently efficient?