Changeset 9176 for branches/UKMO/dev_r5518_GO6_package_OMP/NEMOGCM/NEMO/OPA_SRC/SOL/solpcg.F90

Timestamp:

2018-01-04T13:30:03+01:00 (6 years ago)

Author:

andmirek

Message:

#2001: OMP directives

File:

• branches/UKMO/dev_r5518_GO6_package_OMP/NEMOGCM/NEMO/OPA_SRC/SOL/solpcg.F90 (modified) (8 diffs)

branches/UKMO/dev_r5518_GO6_package_OMP/NEMOGCM/NEMO/OPA_SRC/SOL/solpcg.F90

@@ -94 +94 @@
       REAL(wp), DIMENSION(2) ::   zsum
       REAL(wp), POINTER, DIMENSION(:,:) ::   zgcr
+      REAL(wp),     DIMENSION(jpi, jpj) ::   tmp1, tmp2
       !!----------------------------------------------------------------------
       !
@@ -109 +110 @@
       ! gcr   = gcb-a.gcx
       ! gcdes = gcr
+!$OMP PARALLEL DO PRIVATE(zgcad)
       DO jj = 2, jpjm1
          DO ji = fs_2, fs_jpim1   ! vector opt.
@@ -120 +122 @@
          END DO
       END DO
+!$OMP END PARALLEL DO 
 
       ! rnorme = (gcr,gcr)
-      rnorme = glob_sum(  gcr(:,:) * gcdmat(:,:) * gcr(:,:)  )
-
+      tmp1 = 0.
+!$OMP PARALLEL DO 
+      DO jj = 2, jpjm1
+         DO ji = fs_2, fs_jpim1   ! vector opt.
+            tmp1(ji, jj) = gcr(ji, jj) * gcdmat(ji, jj) * gcr(ji, jj)
+         END DO
+      END DO      
+!$OMP END PARALLEL DO
+      rnorme = glob_sum(  tmp1(:,:)  )
       CALL lbc_lnk( gcdes, c_solver_pt, 1. )   ! lateral boundary condition
 
       ! gccd = matrix . gcdes
+      gccd = 0.
+!$OMP PARALLEL DO 
       DO jj = 2, jpjm1
          DO ji = fs_2, fs_jpim1   ! vector opt.
@@ -134 +146 @@
          END DO
       END DO 
-
+!$OMP END PARALLEL DO
       ! alph = (gcr,gcr)/(gcdes,gccd)
-      radd = glob_sum(  gcdes(:,:) * gcdmat(:,:) * gccd(:,:)  )
+!$OMP PARALLEL DO 
+      DO jj = 1, jpj
+         DO ji = 1, jpi
+            tmp1(ji, jj) = gcdes(ji, jj) * gcdmat(ji, jj) * gccd(ji, jj)
+         END DO
+      END DO      
+!$OMP END PARALLEL DO
+      radd = glob_sum(  tmp1  )
       alph = rnorme /radd
 
       ! gcx = gcx + alph * gcdes
       ! gcr = gcr - alph * gccd
+!$OMP PARALLEL DO
       DO jj = 2, jpjm1
          DO ji = fs_2, fs_jpim1   ! vector opt.
@@ -147 +167 @@
          END DO
       END DO
-
+!$OMP END PARALLEL DO
       ! Algorithm wtih Eijkhout rearrangement
       ! -------------------------------------
@@ -158 +178 @@
 
          ! zgcr = matrix . gcr
+!$OMP PARALLEL DO
          DO jj = 2, jpjm1
             DO ji = fs_2, fs_jpim1   ! vector opt.
@@ -168 +189 @@
          ! rnorme = (gcr,gcr)
          rr = rnorme
-
-         ! zgcad = (zgcr,gcr) 
-         zsum(1) = glob_sum(gcr(:,:) * gcdmat(:,:) * gcr(:,:))
-         zsum(2) = glob_sum(gcr(:,:) * gcdmat(:,:) * zgcr(:,:) * bmask(:,:))
+         
+         ! zgcad = (zgcr,gcr)
+      tmp2 = 0.
+!$OMP PARALLEL
+!$OMP DO 
+      DO jj = 1, jpj
+         DO ji = 1, jpi
+            tmp2(ji, jj) = gcr(ji, jj) * gcdmat(ji, jj)
+            tmp1(ji, jj) = tmp2(ji, jj) * gcr(ji, jj)
+         END DO
+      END DO      
+!$OMP END DO
+!$OMP DO 
+!DIR$ IVDEP
+      DO jj = 1, jpj
+!DIR$ IVDEP
+         DO ji = 1, jpi
+            tmp2(ji, jj) = tmp2(ji, jj) * zgcr(ji, jj) * bmask(ji, jj)
+         END DO
+      END DO      
+!$OMP END DO
+!$OMP END PARALLEL
+ 
+!        zsum(1) = glob_sum(gcr(:,:) * gcdmat(:,:) * gcr(:,:))
+!        zsum(2) = glob_sum(gcr(:,:) * gcdmat(:,:) * zgcr(:,:) * bmask(:,:))
+         zsum = glob_asum_2d(tmp1, tmp2)
 
          !!RB we should gather the 2 glob_sum
@@ -190 +233 @@
          ! gcx = gcx + alph * gcdes
          ! gcr = gcr - alph * gccd
+!$OMP PARALLEL DO
          DO jj = 2, jpjm1
             DO ji = fs_2, fs_jpim1   ! vector opt.