Changeset 1257 for palm/trunk/SOURCE/prandtl_fluxes.f90

Timestamp:

Nov 8, 2013 3:18:40 PM (10 years ago)

Author:

raasch

Message:

New:
---

openACC porting of timestep calculation
(modules, timestep, time_integration)

Changed:

---

openACC loop directives and vector clauses removed (because they do not give any performance improvement with PGI
compiler versions > 13.6)
(advec_ws, buoyancy, coriolis, diffusion_e, diffusion_s, diffusion_u, diffusion_v, diffusion_w, diffusivities, exchange_horiz, fft_xy, pres, production_e, transpose, tridia_solver, wall_fluxes)

openACC loop independent clauses added
(boundary_conds, prandtl_fluxes, pres)

openACC declare create statements moved after FORTRAN declaration statement
(diffusion_u, diffusion_v, diffusion_w, fft_xy, poisfft, production_e, tridia_solver)

openACC end parallel replaced by end parallel loop
(flow_statistics, pres)

openACC "kernels do" replaced by "kernels loop"
(prandtl_fluxes)

output format for theta* changed to avoid output of *
(run_control)

Errors:

---

bugfix for calculation of advective timestep (old version may cause wrong timesteps in case of
vertixcally stretched grids)
Attention: standard run-control output has changed!
(timestep)

File:

• palm/trunk/SOURCE/prandtl_fluxes.f90 (modified) (13 diffs)

palm/trunk/SOURCE/prandtl_fluxes.f90

@@ -20 +20 @@
 ! Current revisions:
 ! -----------------
-!
+! openACC "kernels do" replaced by "kernels loop", "loop independent" added
 !
 ! Former revisions:
@@ -101 +101 @@
 !--    for u* use the value from the previous time step
        !$OMP PARALLEL DO
-       !$acc kernels do
+       !$acc kernels loop
        DO  i = nxlg, nxrg
           DO  j = nysg, nyng
@@ -118 +118 @@
 !--    (the Richardson number is still the one from the previous time step)
        !$OMP PARALLEL DO PRIVATE( a, b, k, z_p )
-       !$acc kernels do
+       !$acc kernels loop
        DO  i = nxlg, nxrg
           DO  j = nysg, nyng
@@ -151 +151 @@
     IF ( .NOT. humidity )  THEN
        !$OMP PARALLEL DO PRIVATE( k, z_p )
-       !$acc kernels do
+       !$acc kernels loop
        DO  i = nxlg, nxrg
           DO  j = nysg, nyng
@@ -170 +170 @@
     ELSE
        !$OMP PARALLEL DO PRIVATE( k, z_p )
-       !$acc kernels do
+       !$acc kernels loop
        DO  i = nxlg, nxrg
           DO  j = nysg, nyng
@@ -193 +193 @@
 !-- Compute u* at the scalars' grid points
     !$OMP PARALLEL DO PRIVATE( a, b, k, uv_total, z_p )
-    !$acc kernels do
+    !$acc kernels loop
     DO  i = nxl, nxr
        DO  j = nys, nyn
@@ -243 +243 @@
 !-- First compute the corresponding component of u* and square it.
     !$OMP PARALLEL DO PRIVATE( a, b, k, rifm, z_p )
-    !$acc kernels do
+    !$acc kernels loop
     DO  i = nxl, nxr
        DO  j = nys, nyn
@@ -281 +281 @@
 !-- First compute the corresponding component of u* and square it.
     !$OMP PARALLEL DO PRIVATE( a, b, k, rifm, z_p )
-    !$acc kernels do
+    !$acc kernels loop
     DO  i = nxl, nxr
        DO  j = nys, nyn
@@ -322 +322 @@
 !--       For a given water flux in the Prandtl layer:
           !$OMP PARALLEL DO
-          !$acc kernels do
+          !$acc kernels loop
           DO  i = nxlg, nxrg
              DO  j = nysg, nyng
@@ -332 +332 @@
           coupled_run = ( coupling_mode == 'atmosphere_to_ocean' .AND. run_coupled )
           !$OMP PARALLEL DO PRIVATE( a, b, k, z_p )
-          !$acc kernels do
+          !$acc kernels loop independent
           DO  i = nxlg, nxrg
+             !$acc loop independent
              DO  j = nysg, nyng
 
@@ -387 +388 @@
     IF ( .NOT. constant_heatflux )  THEN
        !$OMP PARALLEL DO
-       !$acc kernels do
-       DO  i = nxlg, nxrg
+       !$acc kernels loop independent
+       DO  i = nxlg, nxrg
+          !$acc loop independent
           DO  j = nysg, nyng
              shf(j,i) = -ts(j,i) * us(j,i)
@@ -399 +401 @@
     IF ( .NOT. constant_waterflux .AND. ( humidity .OR. passive_scalar ) ) THEN
        !$OMP PARALLEL DO
-       !$acc kernels do
-       DO  i = nxlg, nxrg
+       !$acc kernels loop independent
+       DO  i = nxlg, nxrg
+          !$acc loop independent
           DO  j = nysg, nyng
              qsws(j,i) = -qs(j,i) * us(j,i)
@@ -411 +414 @@
     IF ( ibc_e_b == 2 )  THEN
        !$OMP PARALLEL DO
-       !$acc kernels do
-       DO  i = nxlg, nxrg
+       !$acc kernels loop independent
+       DO  i = nxlg, nxrg
+          !$acc loop independent
           DO  j = nysg, nyng
              e(nzb_s_inner(j,i)+1,j,i) = ( us(j,i) / 0.1 )**2