Skip to content

Optimize ACC port of atm_advance_scalars_work:4760 #1492

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 6 commits into
base: develop
Choose a base branch
from
+19 −37
Draft

Optimize ACC port of atm_advance_scalars_work:4760 #1492

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
d15259e
Add timer adv_scalars_4760
abishekg7 Jun 22, 2026
20de296
Opt 1 - Rearrange and fuse loops in case default
abishekg7 Jun 22, 2026
0f1b1e6
Opt 2 - Remove ica loop, directly subst advCellsForEdge
abishekg7 Jun 22, 2026
5b83a01
Opt 3 - Case 10 Remove scalar_weight2 loop and inline it in next loop
abishekg7 Jun 22, 2026
102b6ad
fix to prev commit
abishekg7 Jun 23, 2026
755e0fc
Opt 4 - use local_sign
abishekg7 Jun 23, 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
56 changes: 19 additions & 37 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 @@ -4725,7 +4725,7 @@ subroutine atm_advance_scalars_work(nCells, num_scalars, dt, &

real (kind=RKIND) :: weight_time_old, weight_time_new
real (kind=RKIND), dimension(num_scalars,nVertLevels) :: scalar_tend_column ! local storage to accumulate tendency
real (kind=RKIND) :: u_direction, u_positive, u_negative
real (kind=RKIND) :: u_direction, u_positive, u_negative, local_sign

flux4(q_im2, q_im1, q_i, q_ip1, ua) = &
ua*( 7.*(q_i + q_im1) - (q_ip1 + q_im2) )/12.0
Expand Down Expand Up @@ -4755,10 +4755,10 @@ subroutine atm_advance_scalars_work(nCells, num_scalars, dt, &
if(rk_step == 3) weight_time_new = 1.
end if
weight_time_old = 1. - weight_time_new


local_sign = 1.0_RKIND
call mpas_timer_start('atm_advance_scalars_4760')
!$acc parallel async
!$acc loop gang worker private(scalar_weight2, ica)
!$acc loop gang worker
do iEdge=edgeStart,edgeEnd

if ((.not.config_apply_lbcs) &
Expand All @@ -4768,32 +4768,21 @@ subroutine atm_advance_scalars_work(nCells, num_scalars, dt, &

case(10)

!$acc loop vector collapse(2)
do j=1,10
do k=1,nVertLevels
scalar_weight2(k,j) = adv_coefs(j,iEdge) + sign(1.0_RKIND,uhAvg(k,iEdge))*adv_coefs_3rd(j,iEdge)
end do
end do

!$acc loop vector
do j=1,10
ica(j) = advCellsForEdge(j,iEdge)
end do

!$acc loop vector collapse(2)
do k = 1,nVertLevels
do iScalar = 1,num_scalars
local_sign = sign(1.0_RKIND,uhAvg(k,iEdge))
horiz_flux_arr(iScalar,k,iEdge) = &
scalar_weight2(k,1) * scalar_new(iScalar,k,ica(1)) + &
scalar_weight2(k,2) * scalar_new(iScalar,k,ica(2)) + &
scalar_weight2(k,3) * scalar_new(iScalar,k,ica(3)) + &
scalar_weight2(k,4) * scalar_new(iScalar,k,ica(4)) + &
scalar_weight2(k,5) * scalar_new(iScalar,k,ica(5)) + &
scalar_weight2(k,6) * scalar_new(iScalar,k,ica(6)) + &
scalar_weight2(k,7) * scalar_new(iScalar,k,ica(7)) + &
scalar_weight2(k,8) * scalar_new(iScalar,k,ica(8)) + &
scalar_weight2(k,9) * scalar_new(iScalar,k,ica(9)) + &
scalar_weight2(k,10) * scalar_new(iScalar,k,ica(10))
(adv_coefs(1,iEdge) + local_sign*adv_coefs_3rd(1,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(1,iEdge)) + &
(adv_coefs(2,iEdge) + local_sign*adv_coefs_3rd(2,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(2,iEdge)) + &
(adv_coefs(3,iEdge) + local_sign*adv_coefs_3rd(3,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(3,iEdge)) + &
(adv_coefs(4,iEdge) + local_sign*adv_coefs_3rd(4,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(4,iEdge)) + &
(adv_coefs(5,iEdge) + local_sign*adv_coefs_3rd(5,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(5,iEdge)) + &
(adv_coefs(6,iEdge) + local_sign*adv_coefs_3rd(6,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(6,iEdge)) + &
(adv_coefs(7,iEdge) + local_sign*adv_coefs_3rd(7,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(7,iEdge)) + &
(adv_coefs(8,iEdge) + local_sign*adv_coefs_3rd(8,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(8,iEdge)) + &
(adv_coefs(9,iEdge) + local_sign*adv_coefs_3rd(9,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(9,iEdge)) + &
(adv_coefs(10,iEdge) + local_sign*adv_coefs_3rd(10,iEdge)) * scalar_new(iScalar,k,advCellsForEdge(10,iEdge))
end do
end do

Expand All @@ -4803,16 +4792,9 @@ subroutine atm_advance_scalars_work(nCells, num_scalars, dt, &
do k=1,nVertLevels
do iScalar=1,num_scalars
horiz_flux_arr(iScalar,k,iEdge) = 0.0_RKIND
end do
end do

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

!$acc loop vector collapse(2)
do k=1,nVertLevels
do iScalar=1,num_scalars
!$acc loop seq
do j=1,nAdvCellsForEdge(iEdge)
iAdvCell = advCellsForEdge(j,iEdge)
scalar_weight = adv_coefs(j,iEdge) + sign(1.0_RKIND,uhAvg(k,iEdge))*adv_coefs_3rd(j,iEdge)
horiz_flux_arr(iScalar,k,iEdge) = horiz_flux_arr(iScalar,k,iEdge) &
+ scalar_weight * scalar_new(iScalar,k,iAdvCell)
Expand Down Expand Up @@ -4842,7 +4824,7 @@ subroutine atm_advance_scalars_work(nCells, num_scalars, dt, &
end if ! end of regional MPAS test
end do
!$acc end parallel

call mpas_timer_stop('atm_advance_scalars_4760')
!$OMP BARRIER

!
Expand Down