Skip to content

Optimize ACC port of atm_compute_dyn_tend_work:6652 #1491

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
abishekg7 wants to merge 3 commits into
base: develop
Choose a base branch
from
+23 −22
Draft

Optimize ACC port of atm_compute_dyn_tend_work:6652 #1491

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e820014
Add timer dyn_tend_work_6652
abishekg7 Jun 23, 2026
4338a08
Opt 1 - Swap loops and replace flux_arr with scalar
abishekg7 Jun 23, 2026
80ce4e9
Opt 2
abishekg7 Jun 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
45 changes: 23 additions & 22 deletions src/core_atmosphere/dynamics/mpas_atm_time_integration.F
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6294,7 +6294,7 @@ subroutine atm_compute_dyn_tend_work(nCells, nEdges, nVertices, nVertLevels_dumm

real (kind=RKIND), dimension( nVertLevels ) :: rayleigh_damp_coef

real (kind=RKIND) :: flux3, flux4
real (kind=RKIND) :: flux3, flux4, flux_arr_local
real (kind=RKIND) :: q_im2, q_im1, q_i, q_ip1, ua, coef3

logical, parameter :: perturbation_coriolis = .true.
Expand Down Expand Up @@ -6648,9 +6648,9 @@ subroutine atm_compute_dyn_tend_work(nCells, nEdges, nVertices, nVertLevels_dumm
!
! horizontal advection for w
!

call mpas_timer_start('atm_dyn_tend_work_6652')
!$acc parallel default(present)
!$acc loop gang worker private(ru_edge_w, flux_arr)
!$acc loop gang worker private(ru_edge_w)
do iCell=cellSolveStart,cellSolveEnd ! Technically updating fewer cells than before...

!$acc loop vector
Expand All @@ -6669,34 +6669,35 @@ subroutine atm_compute_dyn_tend_work(nCells, nEdges, nVertices, nVertLevels_dumm
ru_edge_w(k) = fzm(k)*ru(k,iEdge) + fzp(k)*ru(k-1,iEdge)
end do

!$acc loop vector
do k=1,nVertLevels
flux_arr(k) = 0.0_RKIND
end do

! flux_arr stores the value of w at the cell edge used in the horizontal transport

!$acc loop seq
do j=1,nAdvCellsForEdge(iEdge)
iAdvCell = advCellsForEdge(j,iEdge)

!$acc loop vector
do k=2,nVertLevels
scalar_weight = adv_coefs(j,iEdge) + sign(1.0_RKIND,ru_edge_w(k)) * adv_coefs_3rd(j,iEdge)
flux_arr(k) = flux_arr(k) + scalar_weight * w(k,iAdvCell)
end do
end do

!DIR$ IVDEP
!$acc loop vector
do k=2,nVertLevels
tend_w(k,iCell) = tend_w(k,iCell) - edgesOnCell_sign(i,iCell) * ru_edge_w(k)*flux_arr(k)
flux_arr_local = 0.0_RKIND

select case(nAdvCellsForEdge(iEdge))
case (10)
!$acc loop seq
do j=1,nAdvCellsForEdge(iEdge)
iAdvCell = advCellsForEdge(j,iEdge)
scalar_weight = adv_coefs(j,iEdge) + sign(1.0_RKIND,ru_edge_w(k)) * adv_coefs_3rd(j,iEdge)
flux_arr_local = flux_arr_local + scalar_weight * w(k,iAdvCell)
end do
case default
!$acc loop seq
do j=1,nAdvCellsForEdge(iEdge)
iAdvCell = advCellsForEdge(j,iEdge)
scalar_weight = adv_coefs(j,iEdge) + sign(1.0_RKIND,ru_edge_w(k)) * adv_coefs_3rd(j,iEdge)
flux_arr_local = flux_arr_local + scalar_weight * w(k,iAdvCell)
end do
end select
tend_w(k,iCell) = tend_w(k,iCell) - edgesOnCell_sign(i,iCell) * ru_edge_w(k)*flux_arr_local
end do

end do
end do
!$acc end parallel

call mpas_timer_stop('atm_dyn_tend_work_6652')
#ifdef CURVATURE
!$acc parallel default(present)
!$acc loop gang worker
Expand Down