Changeset 4740

Timestamp:

Oct 14, 2020 11:22:04 AM (4 years ago)

Author:

suehring

Message:

Bugfix in OpenMP directives - intel compiler do not allow reduction operations on array elements

File:

• palm/trunk/SOURCE/pres.f90 (modified) (2 diffs)

palm/trunk/SOURCE/pres.f90

@@ -25 +25 @@
 ! -----------------
 ! $Id$
+! Bugfix in OpenMP directives - intel compiler do not allow reduction operations on array elements
+!
+! 4719 2020-10-01 11:28:33Z pavelkrc
 ! Bugfix of previous commit (div_old discrepancy)
 !
@@ -674 +677 @@
 !-- Sum up the volume flow through the right and north boundary
     IF ( conserve_volume_flow  .AND.  bc_lr_cyc  .AND.  bc_ns_cyc  .AND.  nxr == nx )  THEN
-
-       !$OMP PARALLEL DO PRIVATE (j,k) REDUCTION (+:volume_flow_l(1))
+       threadsum = 0.0_wp
+!
+!--    Summation of the volume flow is done on threadsum rather than on volumen_flow itself.
+!--    This is because intel compiler, when compiled with openmp, do not allow reduction
+!--    operation on array elements.
+       !$OMP PARALLEL DO PRIVATE (j,k) REDUCTION (+:threadsum)
        DO  j = nys, nyn
           DO  k = nzb+1, nzt
-             volume_flow_l(1) = volume_flow_l(1) + u(k,j,nxr) * dzw(k)                             &
+             threadsum = threadsum + u(k,j,nxr) * dzw(k)                             &
                                 * MERGE( 1.0_wp, 0.0_wp, BTEST( wall_flags_total_0(k,j,nxr), 1 ) )
           ENDDO
        ENDDO
+       volume_flow_l(1) = threadsum
 
     ENDIF
 
     IF ( conserve_volume_flow  .AND.  bc_ns_cyc  .AND.  bc_lr_cyc  .AND. nyn == ny )  THEN
-
-       !$OMP PARALLEL DO PRIVATE (i,k) REDUCTION (+:volume_flow_l(2))
+       threadsum = 0.0_wp
+       !$OMP PARALLEL DO PRIVATE (j,k) REDUCTION (+:threadsum)
        DO  i = nxl, nxr
           DO  k = nzb+1, nzt
-             volume_flow_l(2) = volume_flow_l(2) + v(k,nyn,i) * dzw(k)                             &
+             threadsum = threadsum + v(k,nyn,i) * dzw(k)                             &
                                 * MERGE( 1.0_wp, 0.0_wp, BTEST( wall_flags_total_0(k,nyn,i), 2 ) )
            ENDDO
        ENDDO
+       volume_flow_l(2) = threadsum
 
     ENDIF