Changeset 696 for palm/trunk/SOURCE

Timestamp:

Mar 18, 2011 7:03:49 AM (13 years ago)

Author:

raasch

Message:

adjustments for openmp usage on ibmkisti (mrun, subjob); OpenMP-bugfixes: work_fftx removed from PRIVATE clauses in fftx_tr_xy and tr_yx_fftx (poisfft); Bugfix: Summation of Wicker-Skamarock scheme fluxes and variances for all threads (flow_statistics)

Location:

palm/trunk/SOURCE

Files:

• flow_statistics.f90 (modified) (3 diffs)
• poisfft.f90 (modified) (5 diffs)

palm/trunk/SOURCE/flow_statistics.f90

@@ -4 +4 @@
 ! Current revisions:
 ! -----------------
-! 
+! Bugfix: Summation of Wicker-Skamarock scheme fluxes and variances for all
+! threads
 !
 ! Former revisions:
@@ -155 +156 @@
 !--    the local array sums_l() for further computations
        IF ( ws_scheme_mom )  THEN
+
 !       
 !--       According to the Neumann bc for the horizontal velocity components,
@@ -162 +164 @@
              sums_vs2_ws_l(nzt+1,sr) = sums_vs2_ws_l(nzt,sr)    
           ENDIF
+
+          DO  i = 0, threads_per_task-1
 !          
-!--       Swap the turbulent quantities evaluated in advec_ws.
-          sums_l(:,13,0) = sums_wsus_ws_l(:,sr)       ! w*u* 
-          sums_l(:,15,0) = sums_wsvs_ws_l(:,sr)       ! w*v* 
-          sums_l(:,30,0) = sums_us2_ws_l(:,sr)        ! u*2 
-          sums_l(:,31,0) = sums_vs2_ws_l(:,sr)        ! v*2 
-          sums_l(:,32,0) = sums_ws2_ws_l(:,sr)        ! w*2 
-          sums_l(:,34,0) = sums_l(:,34,0) + 0.5 *                             & 
-                (sums_us2_ws_l(:,sr) + sums_vs2_ws_l(:,sr)                    &
-                + sums_ws2_ws_l(:,sr))                      ! e*
-          DO  k = nzb, nzt
-             sums_l(nzb+5,pr_palm,0) = sums_l(nzb+5,pr_palm,0) + 0.5 * (      &
-                sums_us2_ws_l(k,sr) + sums_vs2_ws_l(k,sr) +                   &
-                sums_ws2_ws_l(k,sr))
-          ENDDO
-       ENDIF
+!--          Swap the turbulent quantities evaluated in advec_ws.
+             sums_l(:,13,i) = sums_wsus_ws_l(:,sr)       ! w*u* 
+             sums_l(:,15,i) = sums_wsvs_ws_l(:,sr)       ! w*v* 
+             sums_l(:,30,i) = sums_us2_ws_l(:,sr)        ! u*2 
+             sums_l(:,31,i) = sums_vs2_ws_l(:,sr)        ! v*2 
+             sums_l(:,32,i) = sums_ws2_ws_l(:,sr)        ! w*2 
+             sums_l(:,34,i) = sums_l(:,34,i) + 0.5 *                        & 
+                              ( sums_us2_ws_l(:,sr) + sums_vs2_ws_l(:,sr) + &
+                                sums_ws2_ws_l(:,sr) )    ! e*
+             DO  k = nzb, nzt
+                sums_l(nzb+5,pr_palm,i) = sums_l(nzb+5,pr_palm,i) + 0.5 * (  &
+                                                      sums_us2_ws_l(k,sr) +  &
+                                                      sums_vs2_ws_l(k,sr) +  &
+                                                      sums_ws2_ws_l(k,sr) )
+             ENDDO
+          ENDDO
+
+       ENDIF
+
        IF ( ws_scheme_sca )  THEN
-          sums_l(:,17,0) = sums_wspts_ws_l(:,sr)      ! w*pt* from advec_s_ws
-          IF ( ocean ) sums_l(:,66,0) = sums_wssas_ws_l(:,sr) ! w*sa*
-          IF ( humidity  .OR.  passive_scalar ) sums_l(:,49,0) =              &
-                                                   sums_wsqs_ws_l(:,sr) !w*q* 
+
+          DO  i = 0, threads_per_task-1
+             sums_l(:,17,i) = sums_wspts_ws_l(:,sr)      ! w*pt* from advec_s_ws
+             IF ( ocean ) sums_l(:,66,i) = sums_wssas_ws_l(:,sr) ! w*sa*
+             IF ( humidity .OR. passive_scalar ) sums_l(:,49,i) =              &
+                                                   sums_wsqs_ws_l(:,sr) !w*q*
+          ENDDO
+
        ENDIF
 ! 

palm/trunk/SOURCE/poisfft.f90

@@ -4 +4 @@
 ! Current revisions:
 ! -----------------
-! 
+! work_fftx removed from PRIVATE clauses in fftx_tr_xy and tr_yx_fftx
 !
 ! Former revisions:
@@ -1070 +1070 @@
 !--       Cache optimized code (there might be still a potential for better
 !--       optimization).
-!$OMP     PARALLEL PRIVATE (i,j,k,work_fftx)
+!$OMP     PARALLEL PRIVATE (i,j,k)
 !$OMP     DO
           DO  i = 0, nx
@@ -1181 +1181 @@
 !--       Cache optimized code (there might be still a potential for better
 !--       optimization).
-!$OMP     PARALLEL PRIVATE (i,j,k,work_fftx)
+!$OMP     PARALLEL PRIVATE (i,j,k)
 !$OMP     DO
           DO  j = nys, nyn
@@ -1245 +1245 @@
 
        tn = 0           ! Default thread number in case of one thread
-!$OMP  PARALLEL PRIVATE ( i, j, k, m, n, tn, work_ffty, work_triy )
-!$OMP  DO
+!$OMP  PARALLEL DO PRIVATE ( i, j, k, m, n, tn, work_ffty, work_triy )
        DO  i = nxl_y, nxr_y
 
@@ -1336 +1335 @@
 
        ENDDO
-!$OMP  END PARALLEL
 
        DEALLOCATE( tri )