These are chat archives for Fortran-FOSS-Programmers/General-Discussion

Mar 2017
Stefano Zaghi
Mar 13 2017 11:00


Dear all, this is an help request for expert readers of Fortran Standard (like @zbeekman @rouson ...) : if you have the time, can you let me know if the following snippet is standard?

   subroutine solve(self, eos_left, state_left, eos_right, state_right, normal, fluxes, pattern)
   !< Solve Riemann Problem.
   !< Approximate Riemann Solver based on (local) Lax-Friedrichs (known also as Rusanov) algorithm.
   class(riemann_solver_llf),     intent(inout)         :: self         !< Solver.
   class(eos_object),             intent(in)            :: eos_left     !< Equation of state for left state.
   class(conservative_object),    intent(in)            :: state_left   !< Left Riemann state.
   class(eos_object),             intent(in)            :: eos_right    !< Equation of state for right state.
   class(conservative_object),    intent(in)            :: state_right  !< Right Riemann state.
   type(vector),                  intent(in)            :: normal       !< Normal (versor) of face where fluxes are given.
   class(conservative_object),    intent(inout)         :: fluxes       !< Fluxes of the Riemann Problem solution.
   class(riemann_pattern_object), intent(out), optional :: pattern      !< Riemann pattern.
   type(conservative_compressible), pointer             :: state_left_  !< Left Riemann state, local variable.
   type(conservative_compressible), pointer             :: state_right_ !< Right Riemann state, local variable.
   type(conservative_compressible)                      :: fluxes_left  !< Fluxes of left state.
   type(conservative_compressible)                      :: fluxes_right !< Fluxes of right state.
   type(conservative_compressible)                      :: fluxes_      !< Fluxes, local variable.
   type(riemann_pattern_compressible_pvl)               :: pattern_     !< Riemann pattern.
   real(R8P)                                            :: lmax         !< Maximum wave speed estimation.

   call pattern_%compute(eos_left=eos_left, state_left=state_left, eos_right=eos_right, state_right=state_right, normal=normal)
   lmax = maxval(abs(pattern_%waves_extrema()))
   call state_left%compute_fluxes(eos=eos_left, normal=normal, fluxes=fluxes_left)
   call state_right%compute_fluxes(eos=eos_right, normal=normal, fluxes=fluxes_right)
   state_left_ => conservative_compressible_pointer(to=state_left)
   state_right_ => conservative_compressible_pointer(to=state_right)
   select type(fluxes)
   type is(conservative_compressible)
#ifdef __GFORTRAN__
      fluxes = 0._R8P * (fluxes_left + fluxes_right - (lmax * (state_right_ - state_left_)))
      ! Intel Fortran has issue in resolving the equation with multiple operators... it must be split
      fluxes_ = state_right_ - state_left_
      fluxes_ = lmax * fluxes_
      fluxes = fluxes_left + fluxes_right
      fluxes = fluxes - fluxes_
      fluxes = 0._R8P * fluxes
   if (present(pattern)) then
      select type(pattern)
      type is(riemann_pattern_compressible_pvl)
         pattern = pattern_
   endsubroutine solve

The full story can be read here on google group, but in few words, Intel Fortran does not accept the simple equation and I have to split it.

I suspect that this is a compiler bug, but I like to know your opinion.