Passes on GPUs

Author: Benjamin Wilfong

src/simulation/m_monopole.f90

@@ -32,6 +32,9 @@ module m_monopole
     real(kind(0d0)), allocatable, dimension(:) :: mag, length, npulse, dir, delay
     !$acc declare create(mag, length, npulse, dir, delay)
 
+    real(kind(0d0)), allocatable, dimension(:) :: gammas, pi_infs
+    !$acc declare create(gammas, pi_infs)
+
 contains
 
     subroutine s_initialize_monopole_module()
@@ -54,10 +57,18 @@ subroutine s_initialize_monopole_module()
             end do
         end do
         !$acc update device(mag, support, length, npulse, pulse, dir, delay, foc_length, aperture, loc_mono)
+
+        allocate (gammas(1:num_fluids), pi_infs(1:num_fluids))
+
+        do i = 1, num_fluids
+            gammas(i) = fluid_pp(i)%gamma
+            pi_infs(i) = fluid_pp(i)%pi_inf
+        end do
+        !$acc update device(gammas, pi_infs)
     end subroutine
 
     subroutine s_monopole_calculations(mono_mass_src, mono_mom_src, mono_e_src, myalpha_rho, myalpha, q_cons_vf, &
-                                         q_prim_vf, t_step, id, rhs_vf, gammas, pi_infs)
+                                         q_prim_vf, t_step, id, rhs_vf)
 
         type(scalar_field), dimension(sys_size), intent(inout) :: q_cons_vf !<
         !! This variable contains the WENO-reconstructed values of the cell-average
@@ -75,8 +86,6 @@ subroutine s_monopole_calculations(mono_mass_src, mono_mom_src, mono_e_src, myal
         real(kind(0d0)) :: myR, myV, alf, myP, myRho, R2Vav
         integer, intent(IN) :: t_step
 
-        real(kind(0d0)), dimension(1:num_fluids) :: gammas, pi_infs
-
         integer :: i, j, k, l, q, ii, id !< generic loop variables
         integer :: term_index
 

src/simulation/m_rhs.f90

@@ -1325,7 +1325,7 @@ subroutine s_compute_rhs(q_cons_vf, q_prim_vf, rhs_vf, t_step) ! -------
                 if (monopole) then
                     call s_monopole_calculations(mono_mass_src, mono_mom_src, mono_e_src, myalpha_rho, myalpha,&
                                              q_cons_qp%vf(1:sys_size), q_prim_qp%vf(1:sys_size), t_step, id, &
-                                            rhs_vf, gammas, pi_infs)
+                                            rhs_vf)
                 end if
 
                 if (model_eqns == 3) then
@@ -1647,7 +1647,7 @@ subroutine s_compute_rhs(q_cons_vf, q_prim_vf, rhs_vf, t_step) ! -------
                 if (monopole) then
                     call s_monopole_calculations(mono_mass_src, mono_mom_src, mono_e_src, myalpha_rho, myalpha,&
                                              q_cons_qp%vf(1:sys_size), q_prim_qp%vf(1:sys_size), t_step, id, &
-                                             rhs_vf, gammas, pi_infs)
+                                             rhs_vf)
                 end if
 
                 if (model_eqns == 3) then
@@ -2170,7 +2170,7 @@ subroutine s_compute_rhs(q_cons_vf, q_prim_vf, rhs_vf, t_step) ! -------
                 if (monopole) then
                     call s_monopole_calculations(mono_mass_src, mono_mom_src, mono_e_src, myalpha_rho, myalpha,&
                                              q_cons_qp%vf(1:sys_size), q_prim_qp%vf(1:sys_size), t_step, id, &
-                                             rhs_vf, gammas, pi_infs)
+                                             rhs_vf)
                 end if
 
                 call nvtxEndRange()

src/simulation/m_viscous.f90

@@ -91,6 +91,8 @@ subroutine s_compute_viscous_stress_tensor(q_prim_vf, grad_x_vf, grad_y_vf, grad
 
         type(int_bounds_info) :: ix, iy, iz
 
+    !$acc update device(ix, iy, iz)
+
     !$acc parallel loop collapse(3) gang vector default(present)
         do l = iz%beg, iz%end
             do k = iy%beg, iy%end
@@ -107,7 +109,7 @@ subroutine s_compute_viscous_stress_tensor(q_prim_vf, grad_x_vf, grad_y_vf, grad
             do l = iz%beg, iz%end
                 do k = -1, 1
                     do j = ix%beg, ix%end
-                        print*, j, k, l
+
     !$acc loop seq
                         do i = 1, num_fluids
                             alpha_rho_visc(i) = q_prim_vf(i)%sf(j, k, l)