! source file: /Users/oschlies/UVIC/master/source/mom/congrad.F subroutine congr (npt, variable, bc_symm &, guess, dpsi, forc, res &, cf &, max_iterations, iterations, epsilon &, imask, iperm, jperm, iofs, nisle, nippts &, converged &, estimated_error & ) !======================================================================= ! C O N G R A D ! solve: ! A * dpsi = forc ! for "dpsi" with dirichlet boundary conditions (dpsi=const on ! each component of the boundary) by a preconditioned conjugate ! gradient algorithm. ! inputs: ! npt = 5 or 9 (active coefficients) ! variable = character string identifying solution variable ! bc_symm = equatorial symmetry type (used only when the ! symmetry option is on. otherwise ignore it) ! guess = initial approximation to solution ! A = linear operator (assumed symmetric) ! typically A is grad{(1/h)*grad(dpsi)} - ! 2dt*acor*{grad(f/h) x grad(dpsi)} ! using 5 or 9 pt discretizations ! cf = imt x jmt x 3 x 3 array of coefficients of A ! forc = the sum of all terms evaluated at times tau ! or tau-1 ! epsilon = convergence criterion ! max_iterations = maximum number of iterations ! imask = shows which land masses have perimeter equations ! iperm = i coordinate of island perimeter points ! jperm = j coordinate of island perimeter points ! iofs = offset in iperm, jperm for start of perimeter ! of land_mass(isle) ! nisle = actual number of land_masses ! nippts = number of perimeter ocean points for a land_mass ! output: ! dpsi = answer ! iterations = actual number of iterations performed ! converged = logical value ! estimated_error = estimated maximum error in solution ! based on step sizes and convergence rate ! based on code by: C. Goldberg ! based on the preconditioned conjugate gradient algorithm given ! in: ! A Reformulation and Implementation of the Bryan-Cox-Semtner ! Ocean Model on the Connection Machine ! J.K. Dukowicz, R.D. Smith, and R.C. Malone ! Journal of Atmospheric and Oceanic Technology ! Vol 10. No. 2 April 1993 !======================================================================= ! more specifically, the equations to be solved are ! sum (A(ij,i'j') * dpsi(i'j')) = forc(ij) ! where the subscripts ij and i'j' range over all "free ocean" ! T cells ij=(i,j) that are not adjacent to land T cells, ! and one ij=isle for each boundary component of the ocean. ! with this choice of variables, in the absence of coriolis terms ! (acor=0), the operator A is symmetric, i.e., ! A(ij,i'j') = A(i'j',ij) ! the algorithm (essentially executable in Fortran 90) is... ! subroutine congrad (A, guess, forc, dpsi, iterations) ! use matrix_module ! intent (in) :: A, guess, forc ! intent (out) :: dpsi, iterations ! type(dpsi_type) :: guess, dpsi, Zres, s ! type(res_type) :: res, As, forc ! type(operator) :: A ! type(inv_op) :: Z ! dimension (0:max_iterations) :: dpsi, res, s, As, beta, alpha ! dpsi(0) = guess ! res(0) = forc - A * dpsi(0) ! beta(0) = 1 ! s(0) = zerovector() ! do k = 1 to max_iterations ! Zres(k-1) = Z * res(k-1) ! beta(k) = res(k-1) * Zres(k-1) ! s(k) = Zres(k-1) + (beta(k)/beta(k-1)) * s(k-1) ! As(k) = A * s(k) ! alpha(k) = beta(k) / (s(k) * As(k)) ! dpsi(k) = dpsi(k-1) + alpha(k) * s(k) ! res(k) = res(k-1) - alpha(k) * As(k) ! estimated_error = err_est(k, alpha(k), s(k)) ! if (estimated_error) < epsilon) exit ! enddo ! if (k > max_iterations) then ! print *, 'did not converge in ',k,' iterations' ! stop '=>congrad' ! endif ! iterations = k ! dpsi = dpsi(k) ! end ! where... ! the "vector" and "operator" types used in conjugate gradient ! are mapped to ordinary 2-dimensional fortran arrays as follows: ! type(dpsi_type) :: guess, dpsi, Zres, s ! if ij=(i,j) is a mid-ocean point, map dpsi(ij)-->dpsi(i,j) ! if ij=isle is an ocean boundary subscript, replicate the ! value dpsi(isle) in dpsi(i,j) for each (i,j) in the ocean ! perimeter of land_mass(isle). the arrays iperm(isle) and ! jperm(isle), along with iofs(isle) locate these ocean ! perimeter T cells. ! type(res_type) :: res, As, forc ! if ij=(i,j) is a mid-ocean point, res(ij)-->res(i,j) ! if ij=isle is an ocean boundary subscript, the value of ! res(isle) = sum (res(i,j)) ! where the sum is taken over all (i,j) in the ocean perimeter ! of land_mass(isle). sometimes, the computed values ! res(i,j) represent contributions of T cell (i,j) to the ! component res(isle), and sometimes the values are balanced ! so that res(i,j)=res(isle)/nippts(isle). note that, even ! when balanced, the relation between type(res_type) variables ! res(isle) and res(i,j) differs from that of type(dpsi_type) ! variables dpsi(isle) and dpsi(i,j) on T cells in the ocean ! perimeter. ! type(operator) :: A ! the nearly diagonal quality of the operators used ! permits a representation as a small collection of ! 2-dimensional arrays. ! the diagonal, A(ij,ij), is stored in an array cfdiag(i,j) ! as follows: ! if ij=(i,j) is a mid-ocean point, A(ij,ij) = cfdiag(i,j) ! if ij=isle is an ocean boundary subscript, ! A(isle, isle) = sum (cfdiag(i,j)) ! where the sum is taken over all (i,j) in the ocean perimeter ! of land_mass(isle). each cfdiag(i,j) represents the contribution ! of T cell (i,j) to the island variable diagonal coefficient. ! the off-diagonal terms A(ij,i`j`) are stored in 4 arrays ! cfn, cfs, cfe, and cfw if A is a 5-point operator, and in ! these and 4 additional arrays, cfne, cfnw, cfse, cfsw, if ! A is a 9-point operator. For example, if i`=i and j`=j+1, ! then A(ij,i`j`) is stored in cfn(i,j). ! if ij=(i,j) is a mid-ocean point and i`j`=isle` is and ocean ! perimeter subscript, with i`=i and j`=j+1, then ! cfn(i,j)=A(ij,isle`) is the coefficient of the island ! variable dpsi(isle`) in the equation for mid-ocean point ! dpsi(ij)=dpsi(i,j). ! if ij=isle is an ocean perimeter point and i`j`=(i`,j`) is ! a mid-ocean point, with i`=i and j`=j-1, then ! cfs(i,j)=A(isle,i`j`) is the coefficient of the mid-ocean ! variable dpsi(i`j`)=dpsi(i,j) in the equation for the island ! variable dpsi(isle). note that equations for island ! variables dpsi(isle) are "non-local" in the sense that ! they usually contain more than 5 or 9 terms, some of which ! involve values dpsi(i`j`) outside of a compact 5-point ! or 9-point neighborhood. ! type(inv_op) :: Z ! the approximate inverse operator Z used at present is a ! diagonal operator Z(ij,ij) = 1/A(ij,ij). ! if ij=(i,j) is a mid-ocean point, ! then Z(i,j)=Z(ij)=1/A(ij)=1/cfdiag(i,j) ! if ij=isle is an ocean perimeter point, then ! Z(isle) is replicated at each ocean perimeter T cell ! bordering land_mass(isle). ! Z(i,j)=Z(isle)=1/A(isle)=1/sum(A(i,j)) !======================================================================= include "size.h" include "task_on.h" logical imask, converged, diverging character(16) :: variable character(*) :: bc_symm dimension guess(imt,jmt), dpsi(imt,jmt), Zres(imt,jmt) dimension s(imt,jmt) dimension res(imt,jmt), As(imt,jmt), forc(imt,jmt) dimension cf(imt,jmt,-1:1,-1:1) dimension Z(imt,jmt) dimension imask(-mnisle:mnisle) dimension iperm(maxipp), jperm(maxipp), iofs(mnisle) dimension nippts(mnisle) !----------------------------------------------------------------------- ! impose boundary conditions on guess ! dpsi(0) = guess !----------------------------------------------------------------------- call border(guess, bc_symm) do i=1,imt do j=1,jmt dpsi(i,j) = guess(i,j) enddo enddo !----------------------------------------------------------------------- ! make approximate inverse operator Z (always even symmetry) !----------------------------------------------------------------------- call make_inv (cf, Z, & imask, iperm, jperm, iofs, nisle, nippts) call border(Z, 't even') !----------------------------------------------------------------------- ! res(0) = forc - A * dpsi(0) ! impose 1 and/or symmetry conditions on res(i,j) !----------------------------------------------------------------------- if (npt .eq. 5) then call op5_vec(cf, dpsi, res) else call op9_vec(cf, dpsi, res) endif do i=2,imt-1 do j=2,jmt-1 res(i,j) = forc(i,j) - res(i,j) enddo enddo call border(res, bc_symm) !----------------------------------------------------------------------- ! Zres(k-1) = Z * res(k-1) ! see if guess is a solution, bail out to avoid division by zero !----------------------------------------------------------------------- k = 0 call inv_op(Z, res, Zres, & imask, iperm, jperm, iofs, nisle, nippts) ! set borders of Zres using 1/symmetry, if defined. call border(Zres, bc_symm) Zresmax = absmax(Zres) diverging = .false. ! Assume convergence rate of 0.99 to extrapolate error if (100.0 * Zresmax .lt. epsilon) then estimated_error = 100.0 * Zresmax goto 101 endif !----------------------------------------------------------------------- ! beta(0) = 1 ! s(0) = zerovector() !----------------------------------------------------------------------- betakm1 = 1.0 call zero_vec(s) !----------------------------------------------------------------------- ! begin iteration loop !----------------------------------------------------------------------- do k = 1,max_iterations !----------------------------------------------------------------------- ! Zres(k-1) = Z * res(k-1) !----------------------------------------------------------------------- call inv_op(Z, res, Zres, & imask, iperm, jperm, iofs, nisle, nippts) ! set borders of Zres using 1/symmetry, if defined. call border(Zres, bc_symm) !----------------------------------------------------------------------- ! beta(k) = res(k-1) * Zres(k-1) !----------------------------------------------------------------------- betak = dot2(Zres, res) if (k .eq. 1) then betak_min = abs(betak) elseif (k .gt. 2) then betak_min = min(betak_min, abs(betak)) if (abs(betak) .gt. 100.0*betak_min) then write (*,'(/2(a/))') & 'WARNING: conjugate gradient solver terminated because' &, ' correction steps are diverging.' write (*,'(/7(a/))') & 'PROBABLE CAUSES:' &, ' 1. convergence criterion is too tight...' &, ' roundoff error prevents convergence' &, ' or 2. the solution is beginning to blow up...' &, ' if so, it is extremely unlikely that usable' &, ' results can be obtained in subsequent time' &, ' steps.' write (*,'(/3(a/))') & 'ERROR: It is assumed that the solution is blowing up.' &, ' It is extremely unlikely that usable results can' &, ' be obtained in subsequent time steps.' if (variable .ne. 'surfpres') then stop '==>congrad' endif diverging = .true. smax = absmax(s) step = abs(alpha) * smax estimated_error=step*convergence_rate/(1.0-convergence_rate) go to 101 endif endif !----------------------------------------------------------------------- ! s(k) = Zres(k-1) + (beta(k)/beta(k-1)) * s(k-1) !----------------------------------------------------------------------- betaquot = betak/betakm1 do i=1,imt do j=1,jmt s(i,j) = Zres(i,j) + betaquot * s(i,j) enddo enddo !----------------------------------------------------------------------- ! As(k) = A * s(k) !----------------------------------------------------------------------- if (npt .eq. 5) then call op5_vec(cf, s, As) else call op9_vec(cf, s, As) endif call border(As, bc_symm) !----------------------------------------------------------------------- ! If s=0 then the division for alpha(k) gives a float exception. ! Assume convergence rate of 0.99 to extrapolate error. ! Also assume alpha(k) ~ 1. !----------------------------------------------------------------------- s_dot_As = dot2(s, As) if (abs(s_dot_As) .lt. abs(betak)*1.e-10) then smax = absmax(s) estimated_error = 100.0 * smax goto 101 endif !----------------------------------------------------------------------- ! alpha(k) = beta(k) / (s(k) * As(k)) !----------------------------------------------------------------------- alpha = betak / s_dot_As !----------------------------------------------------------------------- ! update values: ! dpsi(k) = dpsi(k-1) + alpha(k) * s(k) ! res(k) = res(k-1) - alpha(k) * As(k) !----------------------------------------------------------------------- do i=1,imt do j=1,jmt dpsi (i,j) = dpsi(i,j) + alpha * s(i,j) res (i,j) = res (i,j) - alpha * As(i,j) enddo enddo call avg_dist (res, & imask, iperm, jperm, iofs, nisle, nippts) call border(res, bc_symm) smax = absmax(s) !----------------------------------------------------------------------- ! test for convergence ! if (estimated_error) < epsilon) exit !----------------------------------------------------------------------- step = abs(alpha) * smax if (k .eq. 1) then step1 = step estimated_error = step if (step .lt. epsilon) goto 101 elseif (step .lt. epsilon) then cfactor = log(step/step1) convergence_rate = exp(cfactor/(k-1)) estimated_error = step*convergence_rate/(1.0-convergence_rate) if (estimated_error .lt. epsilon) goto 101 endif betakm1 = betak enddo !----------------------------------------------------------------------- ! end of iteration loop !----------------------------------------------------------------------- 101 continue if (k .gt. max_iterations) then cfactor = log(step/step1) convergence_rate = exp(cfactor/(k-1)) estimated_error = step*convergence_rate/(1.0-convergence_rate) converged = .false. else if (diverging) then converged = .false. else converged = .true. endif endif iterations = k !----------------------------------------------------------------------- ! return the last increment of dpsi in the argument res !----------------------------------------------------------------------- if (iterations .eq. 0) then do i=1,imt do j=1,jmt res(i,j) = Zres(i,j) enddo enddo else do i=1,imt do j=1,jmt res(i,j) = alpha * s(i,j) enddo enddo endif return end !======================================================================= ! M A T R I X M O D U L E F O R C O N G R A D !======================================================================= subroutine zero_vec (v) include "size.h" include "task_on.h" dimension v(imt,jmt) do i=1,imt do j=1,jmt v(i,j) = 0.0 enddo enddo return end subroutine add_vec (v,w,vpw) include "size.h" include "task_on.h" dimension v(imt,jmt), w(imt,jmt), vpw(imt,jmt) do i=1,imt do j=1,jmt vpw(i,j) = v(i,j) + w(i,j) enddo enddo return end subroutine sub_vec (v,w,vmw) include "size.h" include "task_on.h" dimension v(imt,jmt), w(imt,jmt), vmw(imt,jmt) do i=1,imt do j=1,jmt vmw(i,j) = v(i,j) - w(i,j) enddo enddo return end subroutine mult_vec(v,w,vtw) include "size.h" include "task_on.h" dimension v(imt,jmt), w(imt,jmt), vtw(imt,jmt) do i=1,imt do j=1,jmt vtw(i,j) = v(i,j) * w(i,j) enddo enddo return end subroutine div_vec(v,w,vdw) include "size.h" include "task_on.h" dimension v(imt,jmt), w(imt,jmt), vdw(imt,jmt) do i=1,imt do j=1,jmt if (w(i,j) .ne. 0) then vdw(i,j) = v(i,j) / w(i,j) else vdw(i,j) = 0.0 endif enddo enddo return end subroutine scalar_vec (scalar,w,sw) include "size.h" include "task_on.h" dimension w(imt,jmt), sw(imt,jmt) do i=1,imt do j=1,jmt sw(i,j) = scalar * w(i,j) enddo enddo return end subroutine neg_vec (v) include "size.h" include "task_on.h" dimension v(imt,jmt) do i=1,imt do j=1,jmt v(i,j) = -v(i,j) enddo enddo return end function dot2 (dp_vec, res_vec) ! this dot product produces the correct answers because for ! ocean perimeter subscripts, ij=isle, the value on a ! type(dpsi_type) vector, dp_vec(isle)=dp_vec(i,j), i.e., the true ! value is replicated, and for a type(res_type) vector, ! res_vec(isle) = sum (res_vec(i,j)), i.e., the true value is the ! accumulation of the distributed values. include "size.h" dimension dp_vec(imt,jmt), res_vec(imt,jmt) dimension rowsum (jmt) do j=2,jmt-1 rowsum(j) = 0.0 do i=2,imt-1 rowsum(j) = rowsum(j) + dp_vec(i,j) * res_vec(i,j) enddo enddo dot2 = 0.0 do j=2,jmt-1 dot2 = dot2 + rowsum(j) enddo return end subroutine op5_vec(cf, dpsi, res) ! res = A * dpsi ! this subroutine does not collect the terms of the true value ! of res(isle) = sum (res(i,j)). the contributions to the sum ! remain distributed among the T cells (i,j) that form the ! ocean perimeter of land_mass(isle). ! at present, borders are not computed [i=1 or imt] [j=1 or jmt] include "size.h" include "task_on.h" dimension cf(imt,jmt,-1:1,-1:1) dimension dpsi(imt,jmt) dimension res(imt,jmt) do j=2,jmt-1 do i=2,imt-1 res(i,j) = cf(i,j, 0, 0) * dpsi(i,j) + & cf(i,j, 0, 1) * dpsi(i,j+1) + & cf(i,j, 0,-1) * dpsi(i,j-1) + & cf(i,j, 1, 0) * dpsi(i+1,j) + & cf(i,j,-1, 0) * dpsi(i-1,j) enddo enddo return end subroutine op9_vec(cf, dpsi, res) ! res = A * dpsi ! this subroutine does not collect the terms of the true value ! of res(isle) = sum (res(i,j)). the contributions to the sum ! remain distributed among the T cells (i,j) that form the ! ocean perimeter of land_mass(isle). ! at present, borders are not computed [i=1 or imt] [j=1 or jmt] include "size.h" include "task_on.h" dimension cf(imt,jmt,-1:1,-1:1) dimension dpsi(imt,jmt) dimension res(imt,jmt) do j=2,jmt-1 do i=2,imt-1 res(i,j) = cf(i,j, 0, 0) * dpsi(i ,j ) + & cf(i,j, 0, 1) * dpsi(i ,j+1) + & cf(i,j, 0,-1) * dpsi(i ,j-1) + & cf(i,j, 1, 0) * dpsi(i+1,j ) + & cf(i,j,-1, 0) * dpsi(i-1,j ) + & cf(i,j, 1, 1) * dpsi(i+1,j+1) + & cf(i,j,-1, 1) * dpsi(i-1,j+1) + & cf(i,j, 1,-1) * dpsi(i+1,j-1) + & cf(i,j,-1,-1) * dpsi(i-1,j-1) enddo enddo return end subroutine subset (a, b, nerror) ! verifies that the set of subscripts for which a(i,j) .ne. 0.0 ! is a subset of the set of subscripts for which b(i,j) .ne. 0.0 include "size.h" include "task_on.h" dimension a(imt,jmt), b(imt,jmt) nerror = 0 do i=2,imt-1 do j=2,jmt-1 if (a(i,j) .ne. 0.0 .and. b(i,j) .eq. 0.0) then nerror = nerror + 1 print '(a,i3,a,i3,a,a)', '(',i,',',j,')' & ,' forcing is reset to zero' ! set forcing (i.e., a(i,j)) to zero a(i,j) = 0.0 endif enddo enddo return end subroutine inv_op(Z, res, Zres, & imask, iperm, jperm, iofs, nisle, nippts) ! apply and approximate inverse Z or the operator A ! res is type(res_type), i.e., perimeter values res(isle) ! are the sum of the distributed contributions res(i,j) ! Zres is type(dpsi_type), i.e., perimeter values Zres(isle) ! must be replicated at each perimeter point Zres(i,j) ! borders of Zres [i=1 or imt] [j=1 or jmt] must be defined ! and must satisfy 1 and/or symmetry, if defined. ! currently, Z is diagonal: Z(ij) = 1/A(ij) ! and is stored in type(dpsi_type) format, i.e., Z(isle) is ! replicated and stored in each Z(i,j) in the perimeter of ! land_mass(isle). include "size.h" include "task_on.h" logical imask dimension imask (-mnisle:mnisle) dimension iperm(maxipp), jperm(maxipp) dimension iofs (mnisle) dimension nippts(mnisle) dimension Z(imt,jmt) dimension res(imt,jmt) dimension Zres(imt,jmt) do i=1,imt do j=1,jmt Zres(i,j) = Z(i,j) * res(i,j) enddo enddo ! sum contributions to Zres(isle) ! distribute Zres(isle) to all perimeter points call sum_dist (Zres, & imask, iperm, jperm, iofs, nisle, nippts) return end function absvecmax(res, imax, jmax) include "size.h" dimension res(imt,jmt) absvecmax = 0.0 do i=2,imt-1 do j=2,jmt-1 if (abs(res(i,j)) .gt. absvecmax) then absvecmax = abs(res(i,j)) imax = i jmax = j endif enddo enddo return end function absmax (f) include "size.h" dimension f(imt,jmt) amax = 0.0 do i=1,imt do j=1,jmt amax = max(amax, abs(f(i,j))) enddo enddo absmax = amax return end function absmin (f) include "size.h" dimension f(imt,jmt) amin = 1.0e37 do i=1,imt do j=1,jmt if (f(i,j) .ne. 0 .and. abs(f(i,j)) .lt. amin) then amin = abs(f(i,j)) endif enddo enddo absmin = amin return end subroutine make_inv (cf, Z, & imask, iperm, jperm, iofs, nisle, nippts) ! construct an approximate inverse Z to A ! Z will be diagonal: Z(ij) = 1/A(ij) ! and values for ocean perimeter entries Z(isle) will be replicated ! at all T cells Z(i,j) in the ocean perimeter of land_mass(isle). ! T cells (i,j) for which there is no diagonal coefficient ! i.e., A(ij)=A(i,j)=0, are masked off by assigning Z(i,j)=0. ! there are effectively no equations and no variables dpsi(i,j) ! at these points. include "size.h" include "task_on.h" logical imask dimension imask (-mnisle:mnisle) dimension iperm(maxipp), jperm(maxipp) dimension iofs (mnisle) dimension nippts(mnisle) dimension cf(imt,jmt,-1:1,-1:1) dimension Z(imt,jmt) ! copy diagonal coefficients of A to Z do i=2,imt-1 do j=2,jmt-1 Z(i,j) = cf(i,j,0,0) enddo enddo ! for each land_mass(isle), ! sum the contributions to cfdiag(isle)=A(isle,isle) ! now stored in Z(i,j) at ocean perimeter T cells and replicate ! the sum in all Z(i,j) for which (i,j) is in ocean perimeter ! of land_mass(isle). call sum_dist (Z, & imask, iperm, jperm, iofs, nisle, nippts) ! now invert Z do i=2,imt-1 do j=2,jmt-1 if (Z(i,j) .ne. 0.0) then Z(i,j) = 1/Z(i,j) else Z(i,j) = 0.0 endif enddo enddo ! make inverse zero on island perimeters that are not integrated do isle=1,nisle if (.not. imask(isle)) then do n=1,nippts(isle) i = iperm(iofs(isle)+n) j = jperm(iofs(isle)+n) Z(i,j) = 0.0 enddo endif enddo return end subroutine sum_dist (Zres, & imask, iperm, jperm, iofs, nisle, nippts) ! sum contributions to Zres(isle) ! distribute Zres(isle) to all perimeter points ! this subroutine converts a type(res_type) vector with ! distributed contributions to perimeter values ! Zres(isle) = sum (Zres(i,j)) ! into a type (dpsi_type) vector with replicated values ! for land_mass perimeters ! Zres(isle) = Zres(i,j) ! for all (i,j) in the ocean perimeter of land_mass(isle). include "size.h" include "task_on.h" logical imask dimension imask (-mnisle:mnisle) dimension iperm(maxipp), jperm(maxipp) dimension iofs (mnisle) dimension nippts(mnisle) dimension Zres(imt,jmt) dimension Zresisle(mnisle) ! sum contributions to Zres(isle) do isle=1,nisle if (imask(isle)) then Zresisle(isle) = 0.0 do n=1,nippts(isle) i = iperm(iofs(isle)+n) j = jperm(iofs(isle)+n) Zresisle(isle) = Zresisle(isle) + Zres(i,j) enddo endif enddo ! distribute Zres(isle) to all perimeter points do isle=1,nisle if (imask(isle)) then do n=1,nippts(isle) i = iperm(iofs(isle)+n) j = jperm(iofs(isle)+n) Zres(i,j) = Zresisle(isle) enddo endif enddo return end subroutine avg_dist (Zres, & imask, iperm, jperm, iofs, nisle, nippts) ! avg contributions to Zres(isle) ! distribute Zres(isle) to all perimeter points ! this subroutine converts a type(res_type) vector with ! distributed contributions to perimeter values ! Zres(isle) = avg (Zres(i,j)) ! into a type (dpsi_type) vector with replicated values ! for land_mass perimeters ! Zres(isle) = Zres(i,j) ! for all (i,j) in the ocean perimeter of land_mass(isle). include "size.h" include "task_on.h" logical imask dimension imask (-mnisle:mnisle) dimension iperm(maxipp), jperm(maxipp) dimension iofs (mnisle) dimension nippts(mnisle) dimension Zres(imt,jmt) dimension Zresisle(mnisle) ! avg contributions to Zres(isle) do isle=1,nisle if (imask(isle)) then ! print *,'isle=',isle,' nisle=',nisle Zresisle(isle) = 0.0 do n=1,nippts(isle) i = iperm(iofs(isle)+n) j = jperm(iofs(isle)+n) Zresisle(isle) = Zresisle(isle) + Zres(i,j) enddo endif enddo ! distribute Zres(isle) to all perimeter points do isle=1,nisle if (imask(isle)) then do n=1,nippts(isle) i = iperm(iofs(isle)+n) j = jperm(iofs(isle)+n) Zres(i,j) = Zresisle(isle)/nippts(isle) enddo endif enddo return end