MODULE solsor !!====================================================================== !! *** MODULE solsor *** !! Ocean solver : Successive Over-Relaxation solver !!===================================================================== !! History : OPA ! 1990-10 (G. Madec) Original code !! 7.1 ! 1993-04 (G. Madec) time filter !! ! 1996-05 (G. Madec) merge sor and pcg formulations !! ! 1996-11 (A. Weaver) correction to preconditioning !! NEMO 1.0 ! 2003-04 (C. Deltel, G. Madec) Red-Black SOR in free form !! 2.0 ! 2005-09 (R. Benshila, G. Madec) MPI optimization !!---------------------------------------------------------------------- !!---------------------------------------------------------------------- !! sol_sor : Red-Black Successive Over-Relaxation solver !!---------------------------------------------------------------------- USE oce ! ocean dynamics and tracers variables USE dom_oce ! ocean space and time domain variables USE zdf_oce ! ocean vertical physics variables USE sol_oce ! solver variables USE in_out_manager ! I/O manager USE lib_mpp ! distributed memory computing USE lbclnk ! ocean lateral boundary conditions (or mpp link) USE lib_fortran ! Fortran routines library USE wrk_nemo ! Memory allocation USE timing ! Timing IMPLICIT NONE PRIVATE PUBLIC sol_sor ! !!---------------------------------------------------------------------- !! NEMO/OPA 3.3 , NEMO Consortium (2010) !! $Id$ !! Software governed by the CeCILL licence (NEMOGCM/NEMO_CeCILL.txt) !!---------------------------------------------------------------------- CONTAINS SUBROUTINE sol_sor( kindic ) !!---------------------------------------------------------------------- !! *** ROUTINE sol_sor *** !! !! ** Purpose : Solve the ellipic equation for the transport !! divergence system using a red-black successive-over- !! relaxation method. !! This routine provides a MPI optimization to the existing solsor !! by reducing the number of call to lbc. !! !! ** Method : Successive-over-relaxation method using the red-black !! technique. The former technique used was not compatible with !! the north-fold boundary condition used in orca configurations. !! Compared to the classical sol_sor, this routine provides a !! mpp optimization by reducing the number of calls to lnc_lnk !! The solution is computed on a larger area and the boudary !! conditions only when the inside domain is reached. !! !! References : Madec et al. 1988, Ocean Modelling, issue 78, 1-6. !! Beare and Stevens 1997 Ann. Geophysicae 15, 1369-1377 !!---------------------------------------------------------------------- !! INTEGER, INTENT(inout) :: kindic ! solver indicator, < 0 if the convergence is not reached: ! ! the model is stopped in step (set to zero before the call of solsor) !! INTEGER :: ji, jj, jn ! dummy loop indices INTEGER :: ishift, icount, ijmppodd, ijmppeven, ijpr2d ! local integers REAL(wp) :: ztmp, zres, zres2 ! local scalars REAL(wp), POINTER, DIMENSION(:,:) :: ztab ! 2D workspace !!---------------------------------------------------------------------- ! IF( nn_timing == 1 ) CALL timing_start('sol_sor') ! CALL wrk_alloc( jpi, jpj, ztab ) ! ijmppeven = MOD( nimpp+njmpp+jpr2di+jpr2dj , 2 ) ijmppodd = MOD( nimpp+njmpp+jpr2di+jpr2dj+1 , 2 ) ijpr2d = MAX( jpr2di , jpr2dj ) icount = 0 ! ! ============== DO jn = 1, nn_nmax ! Iterative loop ! ! ============== IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e( gcx, c_solver_pt, 1., jpr2di, jpr2dj ) ! lateral boundary conditions ! Residus ! ------- ! Guess black update DO jj = 2-jpr2dj, nlcj-1+jpr2dj ishift = MOD( jj-ijmppodd-jpr2dj, 2 ) DO ji = 2-jpr2di+ishift, nlci-1+jpr2di, 2 ztmp = gcb(ji ,jj ) & & - gcp(ji,jj,1) * gcx(ji ,jj-1) & & - gcp(ji,jj,2) * gcx(ji-1,jj ) & & - gcp(ji,jj,3) * gcx(ji+1,jj ) & & - gcp(ji,jj,4) * gcx(ji ,jj+1) ! Estimate of the residual zres = ztmp - gcx(ji,jj) gcr(ji,jj) = zres * gcdmat(ji,jj) * zres ! Guess update gcx(ji,jj) = rn_sor * ztmp + (1-rn_sor) * gcx(ji,jj) END DO END DO icount = icount + 1 IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e( gcx, c_solver_pt, 1., jpr2di, jpr2dj ) ! lateral boundary conditions ! Guess red update DO jj = 2-jpr2dj, nlcj-1+jpr2dj ishift = MOD( jj-ijmppeven-jpr2dj, 2 ) DO ji = 2-jpr2di+ishift, nlci-1+jpr2di, 2 ztmp = gcb(ji ,jj ) & & - gcp(ji,jj,1) * gcx(ji ,jj-1) & & - gcp(ji,jj,2) * gcx(ji-1,jj ) & & - gcp(ji,jj,3) * gcx(ji+1,jj ) & & - gcp(ji,jj,4) * gcx(ji ,jj+1) ! Estimate of the residual zres = ztmp - gcx(ji,jj) gcr(ji,jj) = zres * gcdmat(ji,jj) * zres ! Guess update gcx(ji,jj) = rn_sor * ztmp + (1-rn_sor) * gcx(ji,jj) END DO END DO icount = icount + 1 ! test of convergence IF ( jn > nn_nmin .AND. MOD( jn-nn_nmin, nn_nmod ) == 0 ) THEN SELECT CASE ( nn_sol_arp ) CASE ( 0 ) ! absolute precision (maximum value of the residual) zres2 = MAXVAL( gcr(2:nlci-1,2:nlcj-1) ) IF( lk_mpp ) CALL mpp_max( zres2 ) ! max over the global domain ! test of convergence IF( zres2 < rn_resmax .OR. jn == nn_nmax ) THEN res = SQRT( zres2 ) niter = jn ncut = 999 ENDIF CASE ( 1 ) ! relative precision ztab = 0. ztab(:,:) = gcr(2:nlci-1,2:nlcj-1) rnorme = glob_sum( ztab) ! sum over the global domain ! test of convergence IF( rnorme < epsr .OR. jn == nn_nmax ) THEN res = SQRT( rnorme ) niter = jn ncut = 999 ENDIF END SELECT !**** ! IF(lwp)WRITE(numsol,9300) jn, res, sqrt( epsr ) / eps 9300 FORMAT(' niter :',i4,' res :',e20.10,' b :',e20.10) !**** ENDIF ! indicator of non-convergence or explosion IF( jn == nn_nmax .OR. SQRT(epsr)/eps > 1.e+20 ) kindic = -2 IF( ncut == 999 ) GOTO 999 ! ! ===================== END DO ! END of iterative loop ! ! ===================== 999 CONTINUE ! Output in gcx ! ------------- CALL lbc_lnk_e( gcx, c_solver_pt, 1._wp, jpr2di, jpr2dj ) ! boundary conditions ! CALL wrk_dealloc( jpi, jpj, ztab ) ! IF( nn_timing == 1 ) CALL timing_stop('sol_sor') ! END SUBROUTINE sol_sor !!===================================================================== END MODULE solsor