diff --git a/src/General/memory.f90 b/src/General/memory.f90
index 2af71f75..e8497bd9 100644
--- a/src/General/memory.f90
+++ b/src/General/memory.f90
@@ -516,7 +516,7 @@ subroutine alloc_tot()
       gphiv    = huge(gphiv(1,1))
       allocate(gphif(jpj,jpi))          
       gphif    = huge(gphif(1,1))
-      allocate(e1t(jpj,jpi))            
+      allocate(e1t(jpj,jpi))
       e1t      = huge(e1t(1,1))
       allocate(e1u(jpj,jpi))            
       e1u      = huge(e1u(1,1))
@@ -790,10 +790,14 @@ subroutine alloc_tot()
         allocate(DAY_LENGTH(jpj,jpi))   
        DAY_LENGTH = huge(DAY_LENGTH(1,1))
        forcing_phys_initialized = .false.
+
+       !$acc enter data create(e1t,e2t,e3t,e3w,e3t_back,tra,trb,tmask,avt,&
+       !$acc& e1u,e2u,e3u,e1v,e2v,e3v,un,vn,wn,trn)
+
 #ifdef Mem_Monitor
       mem_all=get_mem(err) - aux_mem
 #endif
-  
+
         END subroutine alloc_tot
 
 
@@ -802,6 +806,8 @@ subroutine clean_memory()
 
             ! myalloc (memory.f90)
 
+            !$acc exit data delete(e1t,e2t,e3t,e3w,e3t_back,tra,trb,tmask,avt)
+
 #ifdef key_mpp
 
             !$acc exit data delete(te_send, tw_send, tn_send, ts_send) finalize
@@ -953,7 +959,7 @@ subroutine clean_memory()
             deallocate(tra_DIA_2d_IO_HIGH)
             deallocate(tra_PHYS_2d_IO)
             deallocate(tra_PHYS_2d_IO_HIGH)
-            
+
 
             if(lwp) then
                 deallocate(tottrn)
@@ -995,6 +1001,8 @@ subroutine clean_memory()
             deallocate(highfreq_table_dia)
             deallocate(highfreq_table_dia2d)
 
+            !$acc exit data delete(trn, e1u, e2u, e3u, e1v, e2v, e3v, un, vn, wn)
+
         end subroutine clean_memory
 
     INTEGER FUNCTION find_index_var(string)
diff --git a/src/General/step.f90 b/src/General/step.f90
index 80c8ef31..e776aa14 100644
--- a/src/General/step.f90
+++ b/src/General/step.f90
@@ -265,6 +265,10 @@ SUBROUTINE trcstp
 !         with surface boundary condition
 !         with IMPLICIT vertical diffusion
 
+      ! XXX: to be removed
+      use DIA_mem, only: diaflx
+      use myalloc, only: tra,trb,e1t,e3t_back,e2t,e3t,e3w,tmask,avt
+
        IMPLICIT NONE
       integer jn,jk,ji,jj
       trcstpparttime = MPI_WTIME() ! cronometer-start
@@ -296,7 +300,9 @@ SUBROUTINE trcstp
 
       IF (lbfm) CALL trcsms
 
+      !$acc update device(e1t,diaflx,e3t_back,e2t,trb,tmask,e3t,tra,avt,e3w) if (lzdf)
       IF (lzdf) CALL trczdf ! tracers: vertical diffusion
+      !$acc update host(diaflx,tra) if (lzdf)
 
       IF (lsnu) CALL snutel
 
diff --git a/src/IO/DIA_mem.f90 b/src/IO/DIA_mem.f90
index e385f32e..4b4a896a 100644
--- a/src/IO/DIA_mem.f90
+++ b/src/IO/DIA_mem.f90
@@ -50,6 +50,7 @@ SUBROUTINE alloc_DIA_local_flx()
        INDflxDUMP = huge(INDflxDUMP(1))
        allocate(diaflx        (7, Fsize, jptra   ))
        diaflx = 0
+       !$acc enter data create(flx_ridxt,diaflx)
       END SUBROUTINE alloc_DIA_local_flx
 
 
@@ -95,6 +96,7 @@ subroutine clean_memory_dia()
 
           if (allocated(diaflx)) then
               deallocate(diaflx)
+              !$acc exit data delete(diaflx)
           endif
 
           if (allocated(INDflxDUMPZERO)) then
diff --git a/src/MPI/ogstm_mpi.f90 b/src/MPI/ogstm_mpi.f90
index b6c0bcd4..f25a754a 100644
--- a/src/MPI/ogstm_mpi.f90
+++ b/src/MPI/ogstm_mpi.f90
@@ -413,6 +413,7 @@ SUBROUTINE mpplnk_my_openacc(ptab,gpu)
       !$acc kernels default(present) if(use_gpu)
       IF(nbondj.eq.0.or.nbondj.eq.1) THEN ! All but south boundary, we received from south
 
+         !$acc loop independent
          DO jw=1,SOUTH_count_recv
               ji = SOUTHpoints_recv(1,jw)
               jk = SOUTHpoints_recv(2,jw)
@@ -423,6 +424,7 @@ SUBROUTINE mpplnk_my_openacc(ptab,gpu)
 
       IF(nbondj.eq.-1.or.nbondj.eq.0) THEN ! All but north boundary, we received from north
 
+        !$acc loop independent
         DO jw=1,NORTH_count_recv
               ji = NORTHpoints_recv(1,jw)
               jk = NORTHpoints_recv(2,jw)
diff --git a/src/PHYS/ADV_mem.f90 b/src/PHYS/ADV_mem.f90
index 9d201de7..d1bba762 100644
--- a/src/PHYS/ADV_mem.f90
+++ b/src/PHYS/ADV_mem.f90
@@ -114,6 +114,9 @@ subroutine myalloc_ADV()
       mem_all=get_mem(err) - aux_mem
 #endif
 
+      !$acc enter data create(advmask,zaa,zbb,zcc,inv_eu,inv_ev,inv_et,&
+      !$acc& big_fact_zaa,big_fact_zbb,big_fact_zcc,zbtr_arr)
+
       END subroutine myalloc_ADV
 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!1
 
@@ -133,6 +136,9 @@ subroutine clean_memory_adv()
           deallocate(big_fact_zcc)
           deallocate(zbtr_arr)
 
+          !$acc exit data delete(advmask,zaa,zbb,zcc,inv_eu,inv_ev,inv_et,&
+          !$acc& big_fact_zaa,big_fact_zbb,big_fact_zcc,zbtr_arr)
+
       end subroutine clean_memory_adv
 
 
diff --git a/src/PHYS/ZDF_mem.f90 b/src/PHYS/ZDF_mem.f90
index cb3fe00b..265e7004 100644
--- a/src/PHYS/ZDF_mem.f90
+++ b/src/PHYS/ZDF_mem.f90
@@ -46,6 +46,9 @@ subroutine myalloc_ZDF()
         jarr_zdf     = huge(jarr_zdf(1,1))
        allocate(jarr_zdf_flx(jpi*jpj,jpk)) 
         jarr_zdf_flx = huge(jarr_zdf_flx(1,1))
+        !$acc enter data create(jarr_zdf,jarr_zdf_flx)
+        !$acc update device(jarr_zdf,jarr_zdf_flx)
+#ifndef _OPENACC
        allocate(zwd(jpk, ntids))           
         zwd          = huge(zwd(1,1))
        allocate(zws(jpk, ntids))           
@@ -60,6 +63,7 @@ subroutine myalloc_ZDF()
         zwz          = huge(zwz(1,1))
        allocate(zwt(jpk, ntids))           
         zwt          = huge(zwt(1,1))
+#endif
 
 #ifdef Mem_Monitor
       mem_all=get_mem(err) - aux_mem
@@ -67,7 +71,28 @@ subroutine myalloc_ZDF()
 
 
       END subroutine myalloc_ZDF
-      
+
+#ifdef _OPENACC
+      subroutine myalloc_ZDF_gpu()
+        allocate(zwd(jpk, dimen_jvzdf))
+        zwd          = huge(zwd(1,1))
+        allocate(zws(jpk, dimen_jvzdf))
+        zws          = huge(zws(1,1))
+        allocate(zwi(jpk, dimen_jvzdf))
+        zwi          = huge(zwi(1,1))
+        allocate(zwx(jpk, dimen_jvzdf))
+        zwx          = huge(zwx(1,1))
+        allocate(zwy(jpk, dimen_jvzdf))
+        zwy          = huge(zwy(1,1))
+        allocate(zwz(jpk, dimen_jvzdf))
+        zwz          = huge(zwz(1,1))
+        allocate(zwt(jpk, dimen_jvzdf))
+        zwt          = huge(zwt(1,1))
+
+        !$acc enter data create(zwd,zwi,zwx,zws,zwz,zwy,zwt)
+        !$acc update device(zwd,zwi,zwx,zws,zwz,zwy,zwt)
+      END subroutine myalloc_ZDF_gpu
+#endif
       
       
       subroutine clean_memory_zdf()
@@ -82,6 +107,8 @@ subroutine clean_memory_zdf()
           deallocate(zwz)
           deallocate(zwt)
 
+          !$acc exit data delete(jarr_zdf,jarr_zdf_flx,zwd,zwi,zwx,zws,zwz,zwy,zwt)
+
       end subroutine clean_memory_zdf
 
 
diff --git a/src/PHYS/trcadv.f90 b/src/PHYS/trcadv.f90
index 129613be..28e638c9 100644
--- a/src/PHYS/trcadv.f90
+++ b/src/PHYS/trcadv.f90
@@ -1,3 +1,8 @@
+#ifdef _OPENACC
+! BUG?: the fsx routine causes additional H2D copies
+#define fsx(pfx1, pfx2, pfu) ((((pfu) + abs(pfu)) * (pfx1) + ((pfu) - abs(pfu)) * (pfx2)) * 0.5)
+#endif
+
 SUBROUTINE trcadv
 !!!---------------------------------------------------------------------
 !!!
@@ -68,7 +73,7 @@ SUBROUTINE trcadv
   !!                               monthly weather review, pp 479-486
   !!
   LOGICAL :: MPI_CHECK,l1,l2,l3
-  INTEGER :: jk,jj,ji,jt,jn,jf,ju
+  INTEGER :: jk,jj,ji,jt,jn,jf,ju,queue
   double precision :: zbtr,zdt
   double precision :: junk, junki, junkj, junkk
   double precision :: timer
@@ -78,10 +83,16 @@ SUBROUTINE trcadv
   double precision, allocatable,dimension(:,:,:) :: zx,zy,zz,zbuf
   double precision, allocatable,dimension(:,:,:) :: zkx,zky,zkz
   logical :: use_gpu
+
+  queue=1
   
   trcadvparttime = MPI_WTIME()
 
+#ifdef _OPENACC
   use_gpu=.true.
+#else
+  use_gpu=.false.
+#endif
 
   !-------------------------------------------------------------------
 
@@ -159,8 +170,22 @@ SUBROUTINE trcadv
 
 
      adv_initialized=.true.
+
   endif
 
+  !!OpenMP compatibility broken. Possibility to use ifndef OpenMP + rename the file in trcadv.F90 to keep it
+  allocate(zy(jpk,jpj,jpi))
+  allocate(zx(jpk,jpj,jpi))
+  allocate(zz(jpk,jpj,jpi))
+  allocate(ztj(jpk,jpj,jpi))
+  allocate(zti(jpk,jpj,jpi))
+  allocate(zkx(jpk,jpj,jpi))
+  allocate(zky(jpk,jpj,jpi))
+  allocate(zkz(jpk,jpj,jpi))
+  allocate(zbuf(jpk,jpj,jpi))
+
+  !$acc enter data create(zy,zx,zz,ztj,zti,zkx,zky,zkz,zbuf)
+
 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! end initialization phase
 
   jk=1
@@ -168,30 +193,24 @@ SUBROUTINE trcadv
   zdt = rdt*ndttrc
   !$OMP TASK private(ji,jj) firstprivate(jpim1,jpjm1) shared(zbtr_arr,e1t,e2t,e3t) default(none)
 
+  !$acc update device( zaa(1:jpk,1:jpj,1:jpi), zbb(1:jpk,1:jpj,1:jpi), zcc(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( inv_eu(1:jpk,1:jpj,1:jpi), inv_ev(1:jpk,1:jpj,1:jpi), inv_et(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( big_fact_zaa (1:jpk,1:jpj,1:jpi), big_fact_zbb(1:jpk,1:jpj,1:jpi), big_fact_zcc(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( zbtr_arr(1:jpk,1:jpj,1:jpi) )
 
-  !$acc enter data create( zaa(1:jpk,1:jpj,1:jpi), zbb(1:jpk,1:jpj,1:jpi), zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( inv_eu(1:jpk,1:jpj,1:jpi), inv_ev(1:jpk,1:jpj,1:jpi), inv_et(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( big_fact_zaa (1:jpk,1:jpj,1:jpi), big_fact_zbb(1:jpk,1:jpj,1:jpi), big_fact_zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( zbtr_arr(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-
-  !$acc enter data create( e1t(1:jpj,1:jpi), e2t(1:jpj,1:jpi), e3t(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( e1u(1:jpj,1:jpi), e2u(1:jpj,1:jpi), e3u(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( e1v(1:jpj,1:jpi), e2v(1:jpj,1:jpi), e3v(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( e3w(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( un(1:jpk,1:jpj,1:jpi), vn(1:jpk,1:jpj,1:jpi), wn(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-
-  !$acc update device( zaa(1:jpk,1:jpj,1:jpi), zbb(1:jpk,1:jpj,1:jpi), zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( inv_eu(1:jpk,1:jpj,1:jpi), inv_ev(1:jpk,1:jpj,1:jpi), inv_et(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( big_fact_zaa (1:jpk,1:jpj,1:jpi), big_fact_zbb(1:jpk,1:jpj,1:jpi), big_fact_zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( zbtr_arr(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
+  !$acc update device( e1t(1:jpj,1:jpi), e2t(1:jpj,1:jpi), e3t(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( e1u(1:jpj,1:jpi), e2u(1:jpj,1:jpi), e3u(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( e1v(1:jpj,1:jpi), e2v(1:jpj,1:jpi), e3v(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( e3w(1:jpk,1:jpj,1:jpi) )
+  !$acc update device( un(1:jpk,1:jpj,1:jpi), vn(1:jpk,1:jpj,1:jpi), wn(1:jpk,1:jpj,1:jpi) )
 
-  !$acc update device( e1t(1:jpj,1:jpi), e2t(1:jpj,1:jpi), e3t(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( e1u(1:jpj,1:jpi), e2u(1:jpj,1:jpi), e3u(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( e1v(1:jpj,1:jpi), e2v(1:jpj,1:jpi), e3v(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( e3w(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update device( un(1:jpk,1:jpj,1:jpi), vn(1:jpk,1:jpj,1:jpi), wn(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
+  !$acc update device(tra(1:jpk,1:jpj,1:jpi,1:jptra))
+  !$acc update device(trn(1:jpk,1:jpj,1:jpi,1:jptra))
+  !$acc update device(advmask(1:jpk,1:jpj,1:jpi))
+  !$acc update device(flx_ridxt(1:Fsize,1:4))
+  !$acc update device( diaflx(1:7, 1:Fsize, 1:jptra))
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -206,7 +225,7 @@ SUBROUTINE trcadv
   !$OMP TASK private(ji,jj) firstprivate(jpim1,jpjm1,jpi,jpj,jpk) default(none) &
   !$OMP shared(zdt,zaa,inv_eu,e1u,e2u,e3u,un,big_fact_zaa)
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -218,7 +237,7 @@ SUBROUTINE trcadv
   !$acc end kernels
 
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -229,7 +248,7 @@ SUBROUTINE trcadv
   END DO
   !$acc end kernels
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -246,7 +265,7 @@ SUBROUTINE trcadv
   !$OMP TASK private(ji,jj) firstprivate(jpim1,jpjm1,jpi,jpj,jpk)  default(none) &
   !$OMP shared(inv_ev,e1v,e2v,e3v,vn,zdt,zbb,big_fact_zbb)
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -257,7 +276,7 @@ SUBROUTINE trcadv
   END DO
   !$acc end kernels
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -268,7 +287,7 @@ SUBROUTINE trcadv
   END DO
   !$acc end kernels
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -283,7 +302,7 @@ SUBROUTINE trcadv
   !$OMP TASK private(ji,jj) firstprivate(jpim1,jpjm1,jpi,jpj,jpk) default(none) &
   !$OMP shared(inv_et,e1t,e2t,e3w,wn,zcc,zdt,big_fact_zcc)
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -294,7 +313,7 @@ SUBROUTINE trcadv
   END DO
   !$acc end kernels
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -305,7 +324,7 @@ SUBROUTINE trcadv
   END DO
   !$acc end kernels
 
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1,jpi
      DO jj = 1,jpj
         !dir$ vector aligned
@@ -319,57 +338,10 @@ SUBROUTINE trcadv
 
   !$OMP TASKWAIT
 
-  !$acc update host( zaa(1:jpk,1:jpj,1:jpi), zbb(1:jpk,1:jpj,1:jpi), zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( inv_eu(1:jpk,1:jpj,1:jpi), inv_ev(1:jpk,1:jpj,1:jpi), inv_et(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( big_fact_zaa (1:jpk,1:jpj,1:jpi), big_fact_zbb(1:jpk,1:jpj,1:jpi), big_fact_zcc(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( zbtr_arr(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-
-
   !!     tracer loop parallelized (macrotasking)
   !!     =======================================
 
-
-
-  !!OpenMP compatibility broken. Possibility to use ifndef OpenMP + rename the file in trcadv.F90 to keep it
-  allocate(zy(jpk,jpj,jpi))
-  allocate(zx(jpk,jpj,jpi))
-  allocate(zz(jpk,jpj,jpi))
-  allocate(ztj(jpk,jpj,jpi))
-  allocate(zti(jpk,jpj,jpi))
-  allocate(zkx(jpk,jpj,jpi))
-  allocate(zky(jpk,jpj,jpi))
-  allocate(zkz(jpk,jpj,jpi))
-  allocate(zbuf(jpk,jpj,jpi))
-
-  zy(:,:,:) = 0
-  zz(:,:,:) = 0
-  zx(:,:,:) = 0
-  ztj(:,:,:)= 0
-  zti(:,:,:)= 0
-  zbuf(:,:,:) = 0.
-  zkx(:,:,:)=0.
-  zky(:,:,:)=0.
-  zkz(:,:,:)=0.
-
-  !!trn could be allocate earlier
-  !$acc enter data create(trn(1:jpk,1:jpj,1:jpi,1:jptra)) if(use_gpu)
-  !$acc enter data create(tra(1:jpk,1:jpj,1:jpi,1:jptra)) if(use_gpu)
-  !$acc enter data create(advmask(1:jpk,1:jpj,1:jpi)) if(use_gpu)
-  !$acc enter data create(flx_ridxt(1:Fsize,1:4)) if(use_gpu)
-  !$acc enter data create( diaflx(1:7, 1:Fsize, 1:jptra)) if(use_gpu)
-
-  !$acc enter data create( zy(1:jpk,1:jpj,1:jpi), zx(1:jpk,1:jpj,1:jpi), zz(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( ztj(1:jpk,1:jpj,1:jpi), zti(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( zkx(1:jpk,1:jpj,1:jpi), zky(1:jpk,1:jpj,1:jpi), zkz(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc enter data create( zbuf(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-
-  !$acc update device(tra(1:jpk,1:jpj,1:jpi,1:jptra)) if(use_gpu)
-  !$acc update device(trn(1:jpk,1:jpj,1:jpi,1:jptra)) if(use_gpu)
-  !$acc update device(advmask(1:jpk,1:jpj,1:jpi)) if(use_gpu)
-  !$acc update device(flx_ridxt(1:Fsize,1:4)) if(use_gpu)
-  !$acc update device( diaflx(1:7, 1:Fsize, 1:jptra)) if(use_gpu)
-
-  !$acc kernels default(present) if(use_gpu)
+  !$acc kernels default(present) async(queue)
   DO ji = 1, jpi
      DO jj = 1, jpj
         DO jk = 1, jpk
@@ -386,6 +358,7 @@ SUBROUTINE trcadv
      ENDDO
   ENDDO
   !$acc end kernels
+  !$acc wait(queue)
 
   !$omp taskloop default(none) private(jf,junk,junki,junkj,junkk,zbtr) &
   !$omp private(zkx,zky,zkz,zti,ztj,zx,zy,zz,zbuf) shared(diaflx,jarrt,tra,zdt) &
@@ -404,8 +377,8 @@ SUBROUTINE trcadv
      !!           and mass fluxes calculated above
      !!       calcul of tracer flux in the i and j direction
 
-     !$acc kernels async default(present) if(use_gpu)
-     !$acc loop independent
+
+     !$acc kernels default(present) async(queue)
      DO ji = 2,jpim1
         !dir$ vector aligned
         DO jj = 2,jpjm1
@@ -414,7 +387,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO ji = 2,jpim1
         !dir$ vector aligned
         !$acc loop independent
@@ -424,7 +397,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO ji = 1,jpi
         !dir$ vector aligned
         !$acc loop independent
@@ -434,7 +407,7 @@ SUBROUTINE trcadv
      ENDDO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO ji = 1,jpi
         !dir$ vector aligned
         !$acc loop independent
@@ -444,7 +417,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
      ! loop unfusion
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO jj = 2,jpjm1
         !dir$ vector aligned
         !$acc loop independent
@@ -454,7 +427,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO jj = 2,jpjm1
         !dir$ vector aligned
         !$acc loop independent
@@ -464,7 +437,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      !$acc loop independent
      DO  ji = 2,jpim1
         DO jj = 2,jpjm1
@@ -476,7 +449,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc kernels async default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO  ji = 2,jpim1
         !$acc loop independent
         DO jj = 2,jpjm1
@@ -488,7 +461,7 @@ SUBROUTINE trcadv
      END DO
      !$acc end kernels
 
-     !$acc parallel loop async collapse(3) gang vector default(present) if(use_gpu)
+     !$acc parallel loop collapse(3) gang vector default(present) async(queue)
      DO  ji = 2,jpim1
         DO jj = 2,jpjm1
            !dir$ vector aligned
@@ -498,7 +471,7 @@ SUBROUTINE trcadv
         END DO
      END DO
      !$acc end parallel loop
-     !$acc wait
+     !$acc wait(queue)
 
      ! ... Lateral boundary conditions on zk[xy]
 #ifdef key_mpp
@@ -526,7 +499,7 @@ SUBROUTINE trcadv
      !! 2. calcul of after field using an upstream advection scheme
      !! -----------------------------------------------------------
 
-     !$acc kernels default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO ji =2,jpim1
         DO jj =2,jpjm1
            DO jk =1,jpkm1
@@ -539,7 +512,7 @@ SUBROUTINE trcadv
      ENDDO
      !$acc end kernels
 
-     !$acc kernels default(present) if(use_gpu)
+     !$acc kernels default(present) async(queue)
      DO jf=1,Fsize
         jk = flx_ridxt(jf,2)
         jj = flx_ridxt(jf,3)
@@ -565,7 +538,7 @@ SUBROUTINE trcadv
         if(jt .EQ. 1) then
 
            if(ncor .EQ. 1) then
-              !$acc kernels default(present) if(use_gpu)
+              !$acc kernels default(present)
               DO ji = 2,jpim1
                  DO jj = 2,jpjm1
                     !dir$ vector aligned
@@ -577,7 +550,7 @@ SUBROUTINE trcadv
               !$acc end kernels
 
            else
-              !$acc kernels default(present) if(use_gpu)
+              !$acc kernels default(present) async(queue)
               DO ji = 2,jpim1
                  DO jj = 2,jpjm1
                     !dir$ vector aligned
@@ -589,7 +562,7 @@ SUBROUTINE trcadv
               END DO
               !$acc end kernels
 
-              !$acc kernels default(present) if(use_gpu)
+              !$acc kernels default(present) async(queue)
               DO ji = 2,jpim1
                  DO jj = 2,jpjm1
                     !dir$ vector aligned
@@ -604,7 +577,7 @@ SUBROUTINE trcadv
            endif
 
         else
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO ji = 2,jpim1
               DO jj = 2,jpjm1
                  !dir$ vector aligned
@@ -615,7 +588,7 @@ SUBROUTINE trcadv
            END DO
            !$acc end kernels
 
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO ji = 2,jpim1
               DO jj = 2,jpjm1
                  !dir$ vector aligned
@@ -627,8 +600,7 @@ SUBROUTINE trcadv
            !$acc end kernels
         endif
 
-
-
+        !$acc wait(queue)
 
         !! ... Lateral boundary conditions on zti
 #ifdef key_mpp
@@ -648,7 +620,7 @@ SUBROUTINE trcadv
 
         !jk = 1
         !          DO jk = 1,jpkm1
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO ji = 2,jpim1
            DO jj = 2,jpjm1
               junk  = zti(1,jj,ji )
@@ -661,7 +633,7 @@ SUBROUTINE trcadv
         !$acc end kernels
 
         !DO ju=1, dimen_jarr2
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO ji = 2,jpim1
            DO jj = 2,jpjm1
               !dir$ vector aligned
@@ -687,6 +659,7 @@ SUBROUTINE trcadv
         !$acc end kernels
         !                 endif
 
+        !$acc wait(queue)
 
         ! ... Lateral boundary conditions on z[xyz]
 #ifdef key_mpp
@@ -713,7 +686,7 @@ SUBROUTINE trcadv
         !!            2.5 calcul of the final field:
         !!                advection by antidiffusive mass fluxes and an upstream scheme
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         !$acc loop independent
         DO ji = 2,jpim1
            !dir$ vector aligned
@@ -723,7 +696,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO ji = 2,jpim1
            !dir$ vector aligned
            !$acc loop independent
@@ -733,7 +706,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO ji = 1,jpi
            !dir$ vector aligned
            !$acc loop independent
@@ -743,7 +716,7 @@ SUBROUTINE trcadv
         ENDDO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO ji = 1,jpi
            !dir$ vector aligned
            !$acc loop independent
@@ -753,7 +726,7 @@ SUBROUTINE trcadv
         ENDDO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO jj = 2,jpjm1
            !dir$ vector aligned
            !$acc loop independent
@@ -763,7 +736,7 @@ SUBROUTINE trcadv
         ENDDO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO jj = 2,jpjm1
            !dir$ vector aligned
            !$acc loop independent
@@ -773,7 +746,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         !$acc loop independent
         DO  ji = 2,jpim1
            DO jj = 2,jpjm1
@@ -787,7 +760,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO  ji = 2,jpim1
            !$acc loop independent
            DO jj = 2,jpjm1
@@ -801,7 +774,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         DO  ji = 2,jpim1
            DO jj = 2,jpjm1
               !dir$ vector aligned
@@ -815,6 +788,7 @@ SUBROUTINE trcadv
         END DO
         !$acc end kernels
 
+        !$acc wait(queue)
 
         !... Lateral boundary conditions on zk[xy]
 #ifdef key_mpp
@@ -837,7 +811,7 @@ SUBROUTINE trcadv
 
 
         if(ncor .EQ. 1) then
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO ji =2,jpim1
               DO jj =2,jpjm1
                  DO jk =1,jpkm1
@@ -848,7 +822,7 @@ SUBROUTINE trcadv
            ENDDO
            !$acc end kernels
 
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO jf=1,Fsize
               jk = flx_ridxt(jf,2)
               jj = flx_ridxt(jf,3)
@@ -863,7 +837,7 @@ SUBROUTINE trcadv
 
         else
 
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO ji =2,jpim1
               DO jj =2,jpjm1
                  DO jk =1,jpkm1
@@ -874,7 +848,7 @@ SUBROUTINE trcadv
            ENDDO
            !$acc end kernels
 
-           !$acc kernels default(present) if(use_gpu)
+           !$acc kernels default(present) async(queue)
            DO jf=1,Fsize
               jk = flx_ridxt(jf,2)
               jj = flx_ridxt(jf,3)
@@ -886,7 +860,6 @@ SUBROUTINE trcadv
            ENDDO
            !$acc end kernels
 
-
         endif
 
      ENDDO ANTIDIFF_CORR
@@ -898,7 +871,7 @@ SUBROUTINE trcadv
 
      if(ncor .EQ. 1) then
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         do ji=1,jpi
            do jj=1,jpj
               do jk=1,jpk
@@ -910,7 +883,7 @@ SUBROUTINE trcadv
 
      else
 
-        !$acc kernels default(present) if(use_gpu)
+        !$acc kernels default(present) async(queue)
         do ji=1,jpi
            do jj=1,jpj
               do jk=1,jpk
@@ -936,21 +909,24 @@ SUBROUTINE trcadv
 
 
   END DO TRACER_LOOP
+  !$acc wait(queue)
+
   !$OMP end taskloop
 
-  !$acc update host( diaflx(1:7, 1:Fsize, 1:jptra) ) if(use_gpu)
-  !$acc update host( tra(1:jpk,1:jpj,1:jpi,1:jptra) ) if(use_gpu)
+  !$acc update host( zaa(1:jpk,1:jpj,1:jpi), zbb(1:jpk,1:jpj,1:jpi), zcc(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( inv_eu(1:jpk,1:jpj,1:jpi), inv_ev(1:jpk,1:jpj,1:jpi), inv_et(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( big_fact_zaa (1:jpk,1:jpj,1:jpi), big_fact_zbb(1:jpk,1:jpj,1:jpi), big_fact_zcc(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( zbtr_arr(1:jpk,1:jpj,1:jpi) )
 
-  !$acc update host( zy(1:jpk,1:jpj,1:jpi), zx(1:jpk,1:jpj,1:jpi), zz(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( ztj(1:jpk,1:jpj,1:jpi), zti(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( zkx(1:jpk,1:jpj,1:jpi), zky(1:jpk,1:jpj,1:jpi), zkz(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
-  !$acc update host( zbuf(1:jpk,1:jpj,1:jpi) ) if(use_gpu)
+  !$acc update host( diaflx(1:7, 1:Fsize, 1:jptra) )
+  !$acc update host( tra(1:jpk,1:jpj,1:jpi,1:jptra) )
 
-  !$acc exit data delete( tra) finalize if(use_gpu)
-  !$acc exit data delete( trn, advmask ) finalize if(use_gpu)
-  !$acc exit data delete( flx_ridxt, diaflx ) finalize if(use_gpu)
-  !$acc exit data delete( zy, zx, zz, ztj, zti, zkx, zky, zkz, zbuf ) finalize if(use_gpu)
+  !$acc update host( zy(1:jpk,1:jpj,1:jpi), zx(1:jpk,1:jpj,1:jpi), zz(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( ztj(1:jpk,1:jpj,1:jpi), zti(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( zkx(1:jpk,1:jpj,1:jpi), zky(1:jpk,1:jpj,1:jpi), zkz(1:jpk,1:jpj,1:jpi) )
+  !$acc update host( zbuf(1:jpk,1:jpj,1:jpi) )
 
+  !$acc exit data delete( zy, zx, zz, ztj, zti, zkx, zky, zkz, zbuf ) finalize
   !!OpenMP compatibility broken. Possibility to use ifndef OpenMP + rename the file in trcadv.F90 to keep it
   deallocate(zy )
   deallocate(zx )
@@ -962,24 +938,22 @@ SUBROUTINE trcadv
   deallocate(zkz )
   deallocate(zbuf )
 
-  !$acc exit data delete( zaa, zbb, zcc, inv_eu, inv_ev, inv_et, big_fact_zaa , big_fact_zbb, big_fact_zcc, zbtr_arr ) finalize if(use_gpu)
-  !$acc exit data delete( e1t, e2t, e3t, e1u, e2u, e3u, e1v, e2v, e3v, e3w, un, vn, wn ) finalize if(use_gpu)
-
   trcadvparttime = MPI_WTIME() - trcadvparttime
   trcadvtottime = trcadvtottime + trcadvparttime
 !!!!
 
 contains
 
+#ifndef _OPENACC
   double precision function fsx(pfx1, pfx2, pfu)
     !dir$ attributes vector :: fsx
-    !$acc routine seq
     IMPLICIT NONE
     double precision, INTENT(IN) :: pfx1, pfx2, pfu
     double precision :: abspfu
     abspfu = abs(pfu)
     fsx = ((pfu + abspfu) * pfx1 + (pfu - abspfu) * pfx2) * 0.5
   end function fsx
+#endif
 
 
 END SUBROUTINE trcadv
diff --git a/src/PHYS/trczdf.f90 b/src/PHYS/trczdf.f90
index 0a02d892..a1b02ee3 100644
--- a/src/PHYS/trczdf.f90
+++ b/src/PHYS/trczdf.f90
@@ -69,7 +69,7 @@ SUBROUTINE trczdf
 
 
       LOGICAL :: l1,l2,l3
-      INTEGER :: jk,jj,ji, jn, jv, jf
+      INTEGER :: jk,jj,ji, jn, jv, jf, ntx
 ! omp variables
       
 
@@ -113,7 +113,11 @@ SUBROUTINE trczdf
                END DO
             END DO
          END DO
-
+         !$acc enter data create(delta_tra,int_tra)
+         !$acc update device(jarr_zdf,jarr_zdf_flx)
+#ifdef _OPENACC
+         call myalloc_ZDF_gpu()
+#endif
       ENDIF
 
 
@@ -129,12 +133,20 @@ SUBROUTINE trczdf
         ztavg = 0.e0
 !! vertical slab
 
+        ! NOTE: kernel is too big, should be split
+        !$acc parallel loop gang vector default(present) async vector_length(32)
         DO jv = 1, dimen_jvzdf
 
            ji  = jarr_zdf(2,jv)
            jj  = jarr_zdf(1,jv)
            Aij = e1t(jj,ji) * e2t(jj,ji)
 
+#ifdef _OPENACC
+           ntx=jv
+#else
+           ntx=1
+#endif
+
 !! I. Vertical trends associated with lateral mixing
 !! -------------------------------------------------
 !!    (excluding the vertical flux proportional to dk[t] )
@@ -155,14 +167,14 @@ SUBROUTINE trczdf
 !!   ... Euler time stepping when starting from rest
         DO jk = 1, jpkm1
           z2dtt = zdt * rdt
-          zwi(jk, 1) = - z2dtt * avt(jk,jj,ji  )/( e3t(jk,jj,ji) * e3w(jk,jj,ji  ) )
-          zws(jk, 1) = - z2dtt * avt(jk+1,jj,ji)/( e3t(jk,jj,ji) * e3w(jk+1,jj,ji) )
-          zwd(jk, 1) = 1. - zwi(jk, 1) - zws(jk, 1)
+          zwi(jk, ntx) = - z2dtt * avt(jk,jj,ji  )/( e3t(jk,jj,ji) * e3w(jk,jj,ji  ) )
+          zws(jk, ntx) = - z2dtt * avt(jk+1,jj,ji)/( e3t(jk,jj,ji) * e3w(jk+1,jj,ji) )
+          zwd(jk, ntx) = 1. - zwi(jk, ntx) - zws(jk, ntx)
         END DO
 
 !! Surface boundary conditions
-          zwi(1,1) = 0.e0
-          zwd(1,1) = 1. - zws(1,1)
+          zwi(1,ntx) = 0.e0
+          zwd(1,ntx) = 1. - zws(1,ntx)
 
 !! II.1. Vertical diffusion on tr
 !! ------------------------------
@@ -170,7 +182,7 @@ SUBROUTINE trczdf
 !!   ... Euler time stepping when starting from rest
         DO jk = 1, jpkm1
           z2dtt = zdt * rdt
-          zwy(jk,1) = trb(jk,jj,ji,jn)*e3t_back(jk,jj,ji)/e3t(jk,jj,ji) + z2dtt * tra(jk,jj,ji,jn)
+          zwy(jk,ntx) = trb(jk,jj,ji,jn)*e3t_back(jk,jj,ji)/e3t(jk,jj,ji) + z2dtt * tra(jk,jj,ji,jn)
         END DO
 
 !! Matrix inversion from the first level
@@ -208,22 +220,22 @@ SUBROUTINE trczdf
         ikstp1=ikst+1
         ikenm2=jpk-2
 
-        zwt(ikst,1)=zwd(ikst,1)
+        zwt(ikst,ntx)=zwd(ikst,ntx)
 
         DO jk=ikstp1,jpkm1
-            zwt(jk,1)=zwd(jk,1)-zwi(jk,1)*zws(jk-1,1)/zwt(jk-1,1)
+            zwt(jk,ntx)=zwd(jk,ntx)-zwi(jk,ntx)*zws(jk-1,ntx)/zwt(jk-1,ntx)
         END DO
 
-          zwz(ikst,1)=zwy(ikst,1)
+          zwz(ikst,ntx)=zwy(ikst,ntx)
 
         DO jk=ikstp1,jpkm1
-            zwz(jk,1)=zwy(jk,1)-zwi(jk, 1)/zwt(jk-1, 1)*zwz(jk-1, 1)
+            zwz(jk,ntx)=zwy(jk,ntx)-zwi(jk, ntx)/zwt(jk-1, ntx)*zwz(jk-1, ntx)
         END DO
 
-        zwx(jpkm1, 1)=zwz(jpkm1, 1)/zwt(jpkm1, 1)
+        zwx(jpkm1, ntx)=zwz(jpkm1, ntx)/zwt(jpkm1, ntx)
 
         DO jk=ikenm2,ikst,-1
-            zwx(jk, 1)=( zwz(jk, 1)-zws(jk, 1)*zwx(jk+1, 1) )/zwt(jk, 1)
+            zwx(jk, ntx)=( zwz(jk, ntx)-zws(jk, ntx)*zwx(jk+1, ntx) )/zwt(jk, ntx)
         END DO
 
 ! calculate flux due to vertical diffusion (on top face of the grid cell jk)
@@ -232,7 +244,7 @@ SUBROUTINE trczdf
       DO jk=1,jpkm1
 
          z2dtt = zdt * rdt
-         delta_tra(jk) = ( zwx(jk,1) - zwy(jk,1) ) / z2dtt * Aij * e3t(jk,jj,ji)! or trn(jk,jj,ji,jn+mytid)
+         delta_tra(jk) = ( zwx(jk,ntx) - zwy(jk,ntx) ) / z2dtt * Aij * e3t(jk,jj,ji)! or trn(jk,jj,ji,jn+mytid)
 
          IF (jk .EQ. 1) THEN
              int_tra(1)  = 0
@@ -258,16 +270,18 @@ SUBROUTINE trczdf
 !! (c a u t i o n:  tracer not its trend, Leap-frog scheme done
 !!                  it will not be done in trcnxt)
          DO jk = 1, jpkm1
-            tra(jk,jj,ji,jn) = zwx(jk,1) * tmask(jk,jj,ji)
+            tra(jk,jj,ji,jn) = zwx(jk,ntx) * tmask(jk,jj,ji)
          END DO
 
         END DO ! jv
+        !$acc end parallel loop
 
 !      end if
 
        END DO TRACER_LOOP
 !!!$omp    end parallel do
 
+       !$acc wait
 
 
        trczdfparttime = MPI_WTIME() - trczdfparttime   !cronometer-stop