Changeset 1221 for palm/trunk/SOURCE/init_grid.f90

Timestamp:

Sep 10, 2013 8:59:13 AM (11 years ago)

Author:

raasch

Message:

New:

---

openACC porting of reduction operations
additional 3D-flag arrays for replacing the 2D-index arrays nzb_s_inner and nzb_diff_s_inner
(flow_statistics, init_grid, init_3d_model, modules, palm, pres, time_integration)

Changed:

---

for PGI/openACC performance reasons set default compile options for openACC to "-ta=nocache",
and set environment variable PGI_ACC_SYNCHRONOUS=1
(MAKE.inc.pgi.openacc, palm_simple_run)

wall_flags_0 changed to 32bit INTEGER, additional array wall_flags_00 introduced to hold
bits 32-63
(advec_ws, init_grid, modules, palm)

Errors:

---

dummy argument tri in 1d-routines replaced by tri_for_1d because of name
conflict with arry tri in module arrays_3d
(tridia_solver)

File:

• palm/trunk/SOURCE/init_grid.f90 (modified) (7 diffs)

palm/trunk/SOURCE/init_grid.f90

@@ -20 +20 @@
 ! Current revisions:
 ! -----------------
-! 
+! wall_flags_00 introduced to hold bits 32-63,
+! additional 3D-flag arrays for replacing the 2D-index array nzb_s_inner in
+! loops optimized for openACC (pres + flow_statistics)
 !
 ! Former revisions:
@@ -387 +389 @@
               nzb_diff_v(nysg:nyng,nxlg:nxrg),                              &
               nzb_2d(nysg:nyng,nxlg:nxrg),                                  &
+              rflags_s_inner(nzb:nzt+2,nysg:nyng,nxlg:nxrg),                &
+              rflags_invers(nysg:nyng,nxlg:nxrg,nzb:nzt+2),                 &
               wall_e_x(nysg:nyng,nxlg:nxrg),                                &
               wall_e_y(nysg:nyng,nxlg:nxrg),                                &
@@ -403 +407 @@
     nzb_v_inner = nzb;  nzb_v_outer = nzb
     nzb_w_inner = nzb;  nzb_w_outer = nzb
+
+    rflags_s_inner = 1.0
+    rflags_invers  = 1.0
 
 !
@@ -854 +861 @@
           
        ENDIF
+!
+!--    Set flag arrays to be used for masking of grid points
+       DO  i = nxlg, nxrg
+          DO  j = nysg, nyng
+             DO  k = nzb, nzt+1
+                IF ( k <= nzb_s_inner(j,i) )  rflags_s_inner(k,j,i) = 0.0
+                IF ( k <= nzb_s_inner(j,i) )  rflags_invers(j,i,k)  = 0.0
+             ENDDO
+          ENDDO
+       ENDDO
 #endif
     ENDIF
@@ -1107 +1124 @@
 !
 !-- Allocate flags needed for masking walls.
-    ALLOCATE( wall_flags_0(nzb:nzt,nys:nyn,nxl:nxr) )
-    wall_flags_0 = 0
+    ALLOCATE( wall_flags_0(nzb:nzt,nys:nyn,nxl:nxr), &
+              wall_flags_00(nzb:nzt,nys:nyn,nxl:nxr) )
+    wall_flags_0  = 0
+    wall_flags_00 = 0
 
     IF ( scalar_advec == 'ws-scheme' )  THEN
@@ -1311 +1330 @@
                    wall_flags_0(k,j,i) = IBSET( wall_flags_0(k,j,i), 31 )
                 ELSE
-                   wall_flags_0(k,j,i) = IBSET( wall_flags_0(k,j,i), 32 )
+                   wall_flags_00(k,j,i) = IBSET( wall_flags_00(k,j,i), 0 )
                 ENDIF
 !
@@ -1325 +1344 @@
 !--                because flux_t(nzb_w_inner(j,i)) is used for the tendency
 !--                at k == nzb_w_inner(j,i)+1.
-                   wall_flags_0(k,j,i) = IBSET( wall_flags_0(k,j,i), 33 )
+                   wall_flags_00(k,j,i) = IBSET( wall_flags_00(k,j,i), 1 )
                    flag_set = .TRUE.
                 ELSEIF ( k == nzb_w_inner(j,i) + 2 .OR. k == nzt - 1 )  THEN
-                   wall_flags_0(k,j,i) = IBSET( wall_flags_0(k,j,i), 34 )
+                   wall_flags_00(k,j,i) = IBSET( wall_flags_00(k,j,i), 2 )
                    flag_set = .TRUE.
                 ELSEIF ( k > nzb_w_inner(j,i) .AND. .NOT. flag_set )  THEN
-                   wall_flags_0(k,j,i) = IBSET( wall_flags_0(k,j,i), 35 )
+                   wall_flags_00(k,j,i) = IBSET( wall_flags_00(k,j,i), 3 )
                 ENDIF