coriolis_adv_apply_tendencies Subroutine

public subroutine coriolis_adv_apply_tendencies(this, ms, dt, no_wait)

Per-layer forward-Euler velocity update. no_wait (optional, default .false.): when .true. the apply DC loops are issued on OpenACC queue 1 and the routine returns WITHOUT syncing, so a batched caller (run_stage_split velocity-apply chain) can pipeline the whole additive apply sequence and !$acc wait(1) ONCE. Default ⇒ self-contained blocking apply (historical, safe for non-batched callers — e.g. the unsplit run_stage). Not pure because of the async/wait directives; still functionally pure.

Arguments

Type IntentOptional Attributes Name
type(coriolis_adv_t), intent(in) :: this
type(multilayer_state_t), intent(inout) :: ms
real(kind=wp), intent(in) :: dt
logical, intent(in), optional :: no_wait

Called by

proc~~coriolis_adv_apply_tendencies~~CalledByGraph proc~coriolis_adv_apply_tendencies coriolis_adv_apply_tendencies proc~run_stage run_stage proc~run_stage->proc~coriolis_adv_apply_tendencies proc~run_stage_split run_stage_split proc~run_stage_split->proc~coriolis_adv_apply_tendencies proc~ocean_dyn_step ocean_dyn_step proc~ocean_dyn_step->proc~run_stage proc~ocean_dyn_step_split ocean_dyn_step_split proc~ocean_dyn_step_split->proc~run_stage_split proc~engine_step engine_step proc~engine_step->proc~ocean_dyn_step proc~engine_step->proc~ocean_dyn_step_split proc~driver_run_ocean driver_run_ocean proc~driver_run_ocean->proc~engine_step proc~rdb_ocean_step rdb_ocean_step proc~rdb_ocean_step->proc~engine_step proc~driver_run driver_run proc~driver_run->proc~driver_run_ocean

Variables

Type Visibility Attributes Name Initial
integer, private :: i
integer, private :: j
integer, private :: k
logical, private :: lwait
integer, private :: nx_face
integer, private :: nx_vface
integer, private :: ny_face
integer, private :: ny_uface
integer, private :: nz

Source Code

   subroutine coriolis_adv_apply_tendencies(this, ms, dt, no_wait)
      !! Per-layer forward-Euler velocity update.
      !! `no_wait` (optional, default .false.): when .true. the apply DC
      !! loops are issued on OpenACC queue 1 and the routine returns WITHOUT
      !! syncing, so a batched caller (`run_stage_split` velocity-apply chain)
      !! can pipeline the whole additive apply sequence and `!$acc wait(1)`
      !! ONCE.  Default ⇒ self-contained blocking apply (historical, safe for
      !! non-batched callers — e.g. the unsplit `run_stage`).  Not `pure`
      !! because of the async/wait directives; still functionally pure.
      type(coriolis_adv_t), intent(in) :: this
      type(multilayer_state_t), intent(inout) :: ms
      real(wp), intent(in) :: dt
      logical, intent(in), optional :: no_wait
      integer :: i, j, k, nx_face, ny_uface, nx_vface, ny_face, nz
      logical :: lwait

      lwait = .true.
      if (present(no_wait)) lwait = .not. no_wait

      nx_face = size(ms%u_face_x_layer, 1)
      ny_uface = size(ms%u_face_x_layer, 2)
      nx_vface = size(ms%v_face_y_layer, 1)
      ny_face = size(ms%v_face_y_layer, 2)
      nz = ms%nz_ml

      !$acc kernels async(1)
      do concurrent(k=1:nz, j=1:ny_uface, i=1:nx_face)
         ms%u_face_x_layer(i, j, k) = ms%u_face_x_layer(i, j, k) + &
                                      dt*this%pv_flux_x%data(i, j, k)
      end do
      do concurrent(k=1:nz, j=1:ny_face, i=1:nx_vface)
         ms%v_face_y_layer(i, j, k) = ms%v_face_y_layer(i, j, k) + &
                                      dt*this%pv_flux_y%data(i, j, k)
      end do
      !$acc end kernels
      if (lwait) then
         !$acc wait(1)
      end if
   end subroutine coriolis_adv_apply_tendencies