Juan 25/03/2022:tensor/discretisation.f90, Cray GPU opt, exchange do seq/do...

Juan 25/03/2022:tensor/discretisation.f90, Cray GPU opt, exchange do seq/do concurrent loop & 3 nest do conccurrent -> mush faster !!!

src/ZSOLVER/tensorproductmultigrid_Source/discretisation.f90

@@ -802,11 +802,12 @@ end subroutine construct_vertical_coeff
                                               +zb_k(iz)          *Tij * zu_st(ii,ij,iz+1)  )
        end do
        !
-       do iz=2,ize-1
+!!$       !$acc loop seq
+!!$       do iz=2,ize-1 
 #ifdef MNH_COMPILER_NVHPC          
           !$acc loop independent  collapse(2)
 #endif          
-          do concurrent(ii=iib:iie,ij=ijb:ije)
+       do concurrent(ii=iib:iie,ij=ijb:ije,iz=2:ize-1)
                 zv_st(ii,ij,iz) = zd_k(iz)* ( ((-zb_k(iz)-zc_k(iz))*Tij - 4.0_rl ) * zu_st(ii,ij,iz)    &
                                                 +zb_k(iz)          *Tij            * zu_st(ii,ij,iz+1)  &
                                                          +zc_k(iz) *Tij            * zu_st(ii,ij,iz-1)  &
@@ -815,8 +816,8 @@ end subroutine construct_vertical_coeff
                                            +                                         zu_st(ii,ij+1,iz) &
                                            +                                         zu_st(ii,ij-1,iz) &
                                        )
-             end do
-          end do
+       end do
+!!$          end do
        !
        iz=ize
 #ifdef MNH_COMPILER_NVHPC       
@@ -2093,8 +2094,9 @@ end subroutine line_Jacobi_mnh
 #ifdef MNH_COMPILER_NVHPC          
        !$acc loop independent collapse(2)
 #endif
-       do iz=2,nz-1
-          do concurrent(ii=iib:iie,ij=ijb:ije)
+       do concurrent(ii=iib:iie,ij=ijb:ije)
+          !$acc loop seq
+          do iz=2,nz-1
              zSr_st(ii,ij,iz) = zb_st(ii,ij,iz) - zd_k(iz) * ( &
                   zSu_in_st(ii+1,ij,iz) + &
                   zSu_in_st(ii-1,ij,iz) + &
@@ -2110,12 +2112,12 @@ end subroutine line_Jacobi_mnh
        iz = 1     
        zSu_out_st(iib:iie,ijb:ije,iz) = zSr_st(iib:iie,ijb:ije,iz) / (tmp_k(iz)*Tijs*zd_k(iz))
        !
-       !$acc loop seq
-       do iz=2,nz-1
 #ifdef MNH_COMPILER_NVHPC          
-          !$acc loop independent collapse(2)
+       !$acc loop independent collapse(2)
 #endif          
-          do concurrent(ii=iib:iie,ij=ijb:ije)
+       do concurrent(ii=iib:iie,ij=ijb:ije)
+          !$acc loop seq
+          do iz=2,nz-1
              zSu_out_st(ii,ij,iz) = (zSr_st(ii,ij,iz) / (Tijs*zd_k(iz)) & 
                                      - zSu_out_st(ii,ij,iz-1)*zc_k(iz)) / tmp_k(iz)
           end do
@@ -2126,12 +2128,12 @@ end subroutine line_Jacobi_mnh
                                      -  zSu_out_st(iib:iie,ijb:ije,iz-1)*zc_k(iz)) / tmp_k(iz)
        
        ! STEP 2: back substitution
-       !$acc loop seq
-       do iz=nz-1,1,-1
 #ifdef MNH_COMPILER_NVHPC          
-          !$acc loop independent collapse(2)
+       !$acc loop independent collapse(2)
 #endif          
-          do concurrent(ii=iib:iie,ij=ijb:ije)
+       do concurrent(ii=iib:iie,ij=ijb:ije)
+       !$acc loop seq
+          do iz=nz-1,1,-1             
              zSu_out_st(ii,ij,iz) = zSu_out_st(ii,ij,iz) & 
                                         - c_k(iz) * zSu_out_st(ii,ij,iz+1)
           end do