MODULE pmc_particle_interface !------------------------------------------------------------------------------! ! This file is part of PALM. ! ! PALM is free software: you can redistribute it and/or modify it under the ! terms of the GNU General Public License as published by the Free Software ! Foundation, either version 3 of the License, or (at your option) any later ! version. ! ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY ! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR ! A PARTICULAR PURPOSE. See the GNU General Public License for more details. ! ! You should have received a copy of the GNU General Public License along with ! PALM. If not, see . ! ! Copyright 1997-2019 Leibniz Universitaet Hannover !------------------------------------------------------------------------------! ! ! Current revisions: ! ------------------ ! ! ! Former revisions: ! -----------------! ! $Id: pmc_particle_interface.f90 3883 2019-04-10 12:51:50Z forkel $ ! Function get_number_of_childs renamed to get_number_of_children and cg ! renamed to pg according to their definitions in pmc_interface_mod ! ! 3655 2019-01-07 16:51:22Z knoop ! unused variables removed ! ! 3123 2018-07-12 16:21:53Z suehring ! Correct working precision for REAL numbers ! ! 3049 2018-05-29 13:52:36Z Giersch ! Error messages revised ! ! 2967 2018-04-13 11:22:08Z Giersch ! bugfix: missing parallel cpp-directives added ! ! 2884 2018-03-14 08:33:20Z scharf ! B: corrected KIND of variable "parsize" for some MPI calls ! ! 2841 2018-02-27 15:02:57Z knoop ! Bugfix: wrong placement of include 'mpif.h' corrected ! ! 2809 2018-02-15 09:55:58Z schwenkel ! Bugfix for gfortran: Replace the function C_SIZEOF with STORAGE_SIZE. ! ! ! Initial Version (by Klaus Ketelsen) ! ! ! Description: ! ------------ ! Introduce particle transfer in nested models. ! Interface to palm lpm model to handel particle transfer between parent and ! child model. !------------------------------------------------------------------------------! USE, INTRINSIC :: ISO_C_BINDING #if defined( __parallel ) && !defined( __mpifh ) USE MPI #endif USE kinds USE pegrid, & ONLY: myidx,myidy USE indices, & ONLY: nx, ny, nxl, nxr, nys, nyn, nzb, nzt, nxlg, nxrg, nysg, nyng, nbgp USE grid_variables, & ONLY: dx, dy USE arrays_3d, & ONLY: zw USE control_parameters, & ONLY: message_string USE particle_attributes, & ONLY: prt_count, particles, grid_particles, & particle_type, number_of_particles, zero_particle, & ibc_par_t, ibc_par_lr, ibc_par_ns, alloc_factor USE lpm_pack_and_sort_mod USE lpm_exchange_horiz_mod, & ONLY: realloc_particles_array #if defined( __parallel ) USE pmc_general, & ONLY: pedef USE pmc_parent, & ONLY: children, pmc_s_fillbuffer, pmc_s_getdata_from_buffer, & pmc_s_get_child_npes USE pmc_child, & ONLY: me, pmc_c_getbuffer, pmc_c_putbuffer USE pmc_interface, & ONLY: cpl_id, get_number_of_children, nr_part, part_adr, nested_run, & get_childid, get_child_edges, nr_partc, part_adrc, & coarse_bound, coord_x, coord_y, pg, get_child_gridspacing, & lower_left_coord_x, lower_left_coord_y USE pmc_handle_communicator, & ONLY: pmc_parent_for_child USE pmc_mpi_wrapper, & ONLY: pmc_send_to_parent, pmc_recv_from_child #endif IMPLICIT NONE #if defined( __parallel ) && defined( __mpifh ) INCLUDE "mpif.h" #endif PRIVATE SAVE TYPE coarse_particle_def INTEGER(iwp) :: nr_particle TYPE(particle_type),ALLOCATABLE,DIMENSION(:) :: parent_particles END TYPE coarse_particle_def INTEGER(iwp),PARAMETER :: Min_particles_per_column = 100 !< INTEGER(iwp),PARAMETER :: max_nr_particle_in_rma_win = 100000 !< INTEGER(iwp) :: nr_fine_in_coarse !< Number of fine grid cells in coarse grid (one direction) INTEGER(iwp) :: particle_win_child !< INTEGER(iwp),ALLOCATABLE,DIMENSION(:) :: particle_win_parent !< TYPE(C_PTR), ALLOCATABLE,DIMENSION(:) :: buf_ptr !< TYPE(particle_type), DIMENSION(:),POINTER :: particle_in_win !< TYPE(coarse_particle_def),ALLOCATABLE,DIMENSION(:,:) :: coarse_particles !< INTERFACE pmcp_g_init MODULE PROCEDURE pmcp_g_init END INTERFACE pmcp_g_init INTERFACE pmcp_g_alloc_win MODULE PROCEDURE pmcp_g_alloc_win END INTERFACE pmcp_g_alloc_win INTERFACE pmcp_c_get_particle_from_parent MODULE PROCEDURE pmcp_c_get_particle_from_parent END INTERFACE pmcp_c_get_particle_from_parent INTERFACE pmcp_c_send_particle_to_parent MODULE PROCEDURE pmcp_c_send_particle_to_parent END INTERFACE pmcp_c_send_particle_to_parent INTERFACE pmcp_p_fill_particle_win MODULE PROCEDURE pmcp_p_fill_particle_win END INTERFACE pmcp_p_fill_particle_win INTERFACE pmcp_p_empty_particle_win MODULE PROCEDURE pmcp_p_empty_particle_win END INTERFACE pmcp_p_empty_particle_win INTERFACE pmcp_g_print_number_of_particles MODULE PROCEDURE pmcp_g_print_number_of_particles END INTERFACE pmcp_g_print_number_of_particles INTERFACE pmcp_p_delete_particles_in_fine_grid_area MODULE PROCEDURE pmcp_p_delete_particles_in_fine_grid_area END INTERFACE pmcp_p_delete_particles_in_fine_grid_area PUBLIC pmcp_g_init, pmcp_g_alloc_win, pmcp_c_get_particle_from_parent PUBLIC pmcp_c_send_particle_to_parent, pmcp_p_fill_particle_win, pmcp_g_print_number_of_particles PUBLIC pmcp_p_empty_particle_win, pmcp_p_delete_particles_in_fine_grid_area CONTAINS !------------------------------------------------------------------------------! ! Description: ! ------------ !> general routine: !> Initializing actions of the particle interface !> check particle boundary conditions for the child models !------------------------------------------------------------------------------! SUBROUTINE pmcp_g_init IMPLICIT NONE INTEGER(iwp) :: nr_childs !< Number of child models of the current model #if defined( __parallel ) nr_childs = get_number_of_children() ! !-- Check if the current model has child models IF ( nr_childs > 0 ) THEN ALLOCATE( nr_part(nysg:nyng, nxlg:nxrg) ) ALLOCATE( part_adr(nysg:nyng, nxlg:nxrg) ) nr_part = 0 part_adr = 0 ENDIF ! !-- Set the boundary conditions to nested for all non root (i.e child) models IF ( cpl_id > 1 ) THEN IF ( ibc_par_t /= 3 ) THEN ibc_par_t = 3 message_string = 'In Child model: ibc_par_t is automatically set to nested ' CALL message( 'pmcp_g_init ', 'PA0477', 0, 1, 0, 6, 0 ) ENDIF IF ( ibc_par_lr /= 3 ) THEN ibc_par_lr = 3 message_string = 'In Child model: ibc_par_lr is automatically set to nested ' CALL message( 'pmcp_g_init ', 'PA0478', 0, 1, 0, 6, 0 ) ENDIF IF ( ibc_par_ns /= 3 ) THEN ibc_par_ns = 3 message_string = 'In Child model: ibc_par_ns is automatically set to nested ' CALL message( 'pmcp_g_init ', 'PA0479', 0, 1, 0, 6, 0 ) ENDIF ENDIF #endif END SUBROUTINE pmcp_g_init !------------------------------------------------------------------------------! ! Description: ! ------------ !> general routine: !> allocate the MPI windows !------------------------------------------------------------------------------! SUBROUTINE pmcp_g_alloc_win IMPLICIT NONE INTEGER(iwp) :: m !< loop index INTEGER(iwp) :: ierr !< error code INTEGER(iwp) :: icl !< left boundary in coarse(parent) index space INTEGER(iwp) :: icr !< right boundary in coarse(parent) index space INTEGER(iwp) :: jcs !< south boundary in coarse(parent) index space INTEGER(iwp) :: jcn !< north boundary in coarse(parent) index space INTEGER(iwp) :: child_id !< Id of a child model INTEGER(iwp) :: nr_childs !< Number of child models of the current model INTEGER :: parsize !< TYPE(C_PTR), SAVE :: ptr !< TYPE(particle_type),DIMENSION(:),POINTER :: win_buffer !< INTEGER(iwp),DIMENSION(1) :: buf_shape !< #if defined( __parallel ) INTEGER(KIND=MPI_ADDRESS_KIND) :: parsize_mpi_address_kind !< INTEGER(KIND=MPI_ADDRESS_KIND) :: winsize !< ! !-- If the model has a parent model prepare the structures for transfer IF ( cpl_id > 1 ) THEN parsize_mpi_address_kind = STORAGE_SIZE(zero_particle)/8 CALL MPI_ALLOC_MEM( parsize_mpi_address_kind , MPI_INFO_NULL, ptr, ierr ) parsize = parsize_mpi_address_kind buf_shape(1) = 1 CALL C_F_POINTER( ptr, win_buffer, buf_shape ) CALL MPI_WIN_CREATE( win_buffer, parsize_mpi_address_kind, parsize, & MPI_INFO_NULL, me%intra_comm, particle_win_child, & ierr ) ! !-- Child domain boundaries in the parent index space icl = coarse_bound(1) icr = coarse_bound(2) jcs = coarse_bound(3) jcn = coarse_bound(4) ALLOCATE( coarse_particles(jcs:jcn, icl:icr) ) coarse_particles(:,:)%nr_particle = 0 ENDIF ! !-- If the model has child models prepare the structures for transfer nr_childs = get_number_of_children() IF ( nr_childs > 0 ) THEN ALLOCATE( particle_win_parent(nr_childs) ) ALLOCATE( buf_ptr(nr_childs) ) DO m = 1, nr_childs child_id = get_childid(m) parsize_mpi_address_kind = STORAGE_SIZE(zero_particle)/8 parsize = parsize_mpi_address_kind winsize = max_nr_particle_in_rma_win * parsize_mpi_address_kind CALL MPI_ALLOC_MEM( winsize , MPI_INFO_NULL, buf_ptr(m), ierr ) buf_shape(1) = max_nr_particle_in_rma_win CALL C_F_POINTER( buf_ptr(m), win_buffer, buf_shape ) CALL MPI_WIN_CREATE( win_buffer, winsize, parsize, MPI_INFO_NULL, & children(child_id)%intra_comm, & particle_win_parent(m), ierr ) ENDDO ENDIF #endif END SUBROUTINE pmcp_g_alloc_win !------------------------------------------------------------------------------! ! Description: ! ------------ !> child routine: !> Read/get particles out of the parent MPI window !------------------------------------------------------------------------------! SUBROUTINE pmcp_c_get_particle_from_parent IMPLICIT NONE INTEGER(iwp) :: i !< x grid index INTEGER(iwp) :: icl !< left boundary in coarse(parent) index space INTEGER(iwp) :: ierr !< error code INTEGER(iwp) :: ij !< combined xy index for the buffer array INTEGER(iwp) :: ip !< loop index (child PEs) INTEGER(iwp) :: j !< y grid index INTEGER(iwp) :: jcs !< south boundary in coarse(parent) index space INTEGER(iwp) :: nr !< number of particles to receive from a parent box INTEGER :: parsize !< #if defined( __parallel ) TYPE(pedef), POINTER :: ape !< TO_DO Klaus: give a description and better name of the variable INTEGER(KIND=MPI_ADDRESS_KIND) :: parsize_mpi_address_kind !< INTEGER(KIND=MPI_ADDRESS_KIND) :: target_disp !< IF ( cpl_id > 1 ) THEN CALL pmc_c_getbuffer( particle_transfer = .TRUE. ) !Get number of particle/column and offset in RMA window xx ! !-- Wait for buffer to fill. ! !-- The parent side (in pmc_s_fillbuffer) is filling the buffer in the MPI RMA window !-- When the filling is complete, a MPI_BARRIER is called. !-- The child is not allowd to access the parent-buffer before it is completely filled !-- Synchronization is done implicitely in pmc_c_getbuffer and pmc_s_fillbuffer on the parent side icl = coarse_bound(1) jcs = coarse_bound(3) DO ip = 1, me%inter_npes ape => me%pes(ip) DO ij = 1, ape%nrele j = ape%locind(ij)%j + jcs - 1 i = ape%locind(ij)%i + icl - 1 nr = nr_partc(j,i) IF ( nr > 0 ) THEN CALL check_and_alloc_coarse_particle (i, j, nr) parsize_mpi_address_kind = STORAGE_SIZE(zero_particle)/8 parsize = parsize_mpi_address_kind target_disp = part_adrc(j,i) - 1 CALL MPI_WIN_LOCK( MPI_LOCK_SHARED , ip - 1, 0, & particle_win_child, ierr ) CALL MPI_GET( coarse_particles(j,i)%parent_particles, & nr * parsize, MPI_BYTE, ip - 1, target_disp, & nr * parsize, MPI_BYTE, particle_win_child, ierr ) CALL MPI_WIN_UNLOCK( ip - 1, particle_win_child, ierr ) ENDIF coarse_particles(j,i)%nr_particle = nr ENDDO ENDDO CALL c_copy_particle_to_child_grid ENDIF #endif END SUBROUTINE pmcp_c_get_particle_from_parent !------------------------------------------------------------------------------! ! Description: ! ------------ !> child routine: !> Write/put particles into the parent MPI window !------------------------------------------------------------------------------! SUBROUTINE pmcp_c_send_particle_to_parent IMPLICIT NONE INTEGER(iwp) :: disp_offset !< INTEGER(iwp) :: i !< x loop index INTEGER(iwp) :: icl !< left boundary in coarse(parent) index space INTEGER(iwp) :: icr !< right boundary in coarse(parent) index space INTEGER(iwp) :: ierr !< error code INTEGER(iwp) :: ij !< combined xy index for the buffer array INTEGER(iwp) :: ip !< loop index (child PEs) INTEGER(iwp) :: j !< y loop index INTEGER(iwp) :: jcs !< south boundary in coarse(parent) index space INTEGER(iwp) :: jcn !< north boundary in coarse(parent) index space INTEGER(iwp) :: max_nr_particle_per_pe !< maximum number of particles per PE (depending on grid apect ratio) INTEGER(iwp) :: n !< shorter variable name for nr_fine_in_coarse INTEGER(iwp) :: nr !< shorter variabel name for nr_partc INTEGER(iwp) :: pe_offset !< offset index of the current PE INTEGER :: parsize !< REAL(wp) :: eps=0.00001 !< used in calculations to avoid rounding errors REAL(wp) :: xx !< number of fine grid cells inside a coarse grid cell in x-direction REAL(wp) :: yy !< number of fine grid cells inside a coarse grid cell in y-direction ! TYPE(particle_type) :: dummy_part !< dummy particle (needed for size calculations) #if defined( __parallel ) TYPE(pedef), POINTER :: ape !< TO_DO Klaus: give a description and better name of the variable INTEGER(KIND=MPI_ADDRESS_KIND) :: parsize_mpi_address_kind !< INTEGER(KIND=MPI_ADDRESS_KIND) :: target_disp !< IF ( cpl_id > 1 ) THEN CALL c_copy_particle_to_coarse_grid ! !-- Child domain boundaries in the parent index space icl = coarse_bound(1) icr = coarse_bound(2) jcs = coarse_bound(3) jcn = coarse_bound(4) nr_partc = 0 DO i = icl, icr DO j = jcs, jcn nr_partc(j,i) = coarse_particles(j,i)%nr_particle ENDDO ENDDO part_adrc = 0 ! !-- compute number of fine grid cells in coarse grid (one direction) xx = ( pg%dx + eps ) / dx ! +eps to avoid rounding error yy = ( pg%dy + eps ) / dy nr_fine_in_coarse = MAX( INT(xx), INT(yy) ) IF ( MOD( coord_x(0), pg%dx ) /= 0.0 .OR. MOD( coord_y(0), pg%dy ) /= 0.0 ) THEN nr_fine_in_coarse = nr_fine_in_coarse + 1 ENDIF ! !-- Assign a number to my child PE to select different areas in the RMA window on server side !-- With this number a square of child PEs is defined which share the same coarse grid cells n = nr_fine_in_coarse ! local variable n to make folloing statements shorter pe_offset = MOD( myidx, n ) * n + MOD( myidy, n) max_nr_particle_per_pe = max_nr_particle_in_rma_win / ( n * n ) disp_offset = pe_offset * max_nr_particle_per_pe parsize_mpi_address_kind = STORAGE_SIZE(zero_particle)/8 parsize = parsize_mpi_address_kind DO ip = 1, me%inter_npes ape => me%pes(ip) target_disp = disp_offset DO ij = 1, ape%nrele j = ape%locind(ij)%j + jcs - 1 i = ape%locind(ij)%i + icl - 1 nr = nr_partc(j,i) IF( nr > 0 ) THEN IF ( target_disp + nr - disp_offset >= max_nr_particle_per_pe ) THEN WRITE(9,*) 'RMA window too small on child ', & target_disp + nr - disp_offset, & max_nr_particle_per_pe, max_nr_particle_in_rma_win message_string = 'RMA window too small on child' CALL message( 'pmci_create_child_arrays', 'PA0480', 3, 2, 0, 6, 0 ) ENDIF CALL MPI_WIN_LOCK( MPI_LOCK_SHARED , ip - 1, 0, & particle_win_child, ierr ) CALL MPI_PUT( coarse_particles(j,i)%parent_particles, & nr * parsize, MPI_BYTE, ip - 1, target_disp, & nr * parsize, MPI_BYTE, particle_win_child, ierr ) CALL MPI_WIN_UNLOCK( ip - 1, particle_win_child, ierr ) part_adrc(j,i) = target_disp + 1 target_disp = target_disp + nr ENDIF ENDDO ENDDO CALL pmc_c_putbuffer ( particle_transfer = .TRUE. ) !Send new number of particle/column and offset to parent ENDIF #endif END SUBROUTINE pmcp_c_send_particle_to_parent !------------------------------------------------------------------------------! ! Description: ! ------------ !> parent routine: !> write particles into the MPI window !------------------------------------------------------------------------------! SUBROUTINE pmcp_p_fill_particle_win IMPLICIT NONE LOGICAL :: active_particle !< Particles located in the fine/child grid area are marked as active (to be transferred) LOGICAL,SAVE :: lfirst = .TRUE. !< INTEGER(iwp) :: child_id !< id of the child model INTEGER(iwp) :: i !< x grid box index INTEGER(iwp) :: ij !< combined xy index for the buffer array INTEGER(iwp) :: ip !< loop index (child PEs) INTEGER(iwp) :: j !< y grid box index INTEGER(iwp) :: k !< z grid box index INTEGER(iwp) :: m !< loop index (number of childs) INTEGER(iwp) :: n !< loop index (number of particles) INTEGER(iwp) :: nr_part_col !< Number of particles to transfer per column INTEGER(iwp) :: number_of_particles !< INTEGER(iwp) :: pindex !< INTEGER(iwp) :: tot_particle_count !< Total number of particles per child REAL(wp) :: dx_child !< child grid spacing REAL(wp) :: dy_child !< child grid spacing REAL(wp) :: dz_child !< child grid spacing REAL(wp) :: ny_coord !< north coordinate of child grid REAL(wp) :: ny_coord_b !< north coordinate of child grid boundary REAL(wp) :: lx_coord !< left coordinate of child grid REAL(wp) :: lx_coord_b !< left coordinate of child grid boundary REAL(wp) :: rx_coord !< right coordinate of child grid REAL(wp) :: rx_coord_b !< right coordinate of child grid boundary REAL(wp) :: sy_coord !< south coordinate of child grid REAL(wp) :: sy_coord_b !< south coordinate of child grid boundary REAL(wp) :: uz_coord !< upper coordinate of child grid REAL(wp) :: uz_coord_b !< upper coordinate of child grid boundary REAL(wp) :: x !< particle position REAL(wp) :: y !< particle position REAL(wp) :: z !< particle position INTEGER(iwp),DIMENSION(1) :: buf_shape !< #if defined( __parallel ) TYPE(pedef), POINTER :: ape !< TO_DO Klaus: give a description and better name of the variable DO m = 1, get_number_of_children() child_id = get_childid(m) CALL get_child_edges( m, lx_coord, lx_coord_b, rx_coord, rx_coord_b, & sy_coord, sy_coord_b, ny_coord, ny_coord_b, & uz_coord, uz_coord_b) CALL get_child_gridspacing( m, dx_child, dy_child, dz_child ) IF ( lfirst ) THEN WRITE(9,'(a,5f10.2)') 'edges ',lx_coord,rx_coord,sy_coord,ny_coord,uz_coord WRITE(9,'(a,5f10.2)') 'edges boundary ',lx_coord_b,rx_coord_b,sy_coord_b,ny_coord_b,uz_coord_b WRITE(9,'(a,5f10.2)') 'child spacing ',dx_child, dy_child, dz_child,lower_left_coord_x,lower_left_coord_y ENDIF ! !-- reset values for every child tot_particle_count = 0 nr_part = 0 part_adr = 0 pindex = 1 buf_shape(1) = max_nr_particle_in_rma_win CALL C_F_POINTER( buf_ptr(m), particle_in_win , buf_shape ) DO ip = 1, children(child_id)%inter_npes ape => children(child_id)%pes(ip) nr_part_col = 0 DO ij = 1, ape%nrele ! !-- Inside the PMC adressing of 3d arrays starts with 1 i = ape%locind(ij)%i + nxl - nbgp - 1 j = ape%locind(ij)%j + nys - nbgp - 1 nr_part_col = 0 ! Number of particles to transfer per column part_adr(j,i) = pindex DO k = nzb + 1, nzt number_of_particles = prt_count(k,j,i) IF ( number_of_particles <= 0 ) CYCLE particles => grid_particles(k,j,i)%particles(1:number_of_particles) ! Select particles within boundary area DO n = 1, number_of_particles x = particles(n)%x y = particles(n)%y z = particles(n)%z ! !-- check if the particle is located in the fine grid area active_particle = ((x > lx_coord .AND. x < rx_coord) .AND. & (y > sy_coord .AND. y < ny_coord) .AND. & (z > 0.000000001 .AND. z < uz_coord)) IF ( active_particle .AND. particles(n)%particle_mask ) THEN particle_in_win(pindex) = particles(n) ! !-- Change particle positions and origin relative to global origin particle_in_win(pindex)%x = particle_in_win(pindex)%x + lower_left_coord_x particle_in_win(pindex)%y = particle_in_win(pindex)%y + lower_left_coord_y particle_in_win(pindex)%origin_x = particle_in_win(pindex)%origin_x + lower_left_coord_x particle_in_win(pindex)%origin_y = particle_in_win(pindex)%origin_y + lower_left_coord_y tot_particle_count = tot_particle_count + 1 nr_part_col = nr_part_col + 1 pindex = pindex + 1 IF ( pindex > max_nr_particle_in_rma_win ) THEN WRITE(9,*) 'RMA window too small on parent ',pindex, max_nr_particle_in_rma_win message_string = 'RMA window too small on parent' CALL message( 'pmci_create_child_arrays', 'PA0481', 3, 2, 0, 6, 0 ) ! PA number has to be adjusted ENDIF END IF ENDDO ENDDO nr_part(j,i) = nr_part_col ENDDO ENDDO CALL pmc_s_fillbuffer( child_id, particle_transfer = .TRUE. ) ENDDO lfirst = .FALSE. #endif END SUBROUTINE pmcp_p_fill_particle_win !------------------------------------------------------------------------------! ! Description: ! ------------ !> parent routine: !> delete particles from the MPI window !------------------------------------------------------------------------------! SUBROUTINE pmcp_p_empty_particle_win IMPLICIT NONE INTEGER(iwp) :: child_id !< model id of the child INTEGER(iwp) :: ip !< loop index (child PEs) INTEGER(iwp) :: m !< loop index (number of childs) INTEGER(iwp),DIMENSION(1) :: buf_shape !< #if defined( __parallel ) DO m = 1, get_number_of_children() child_id = get_childid(m) buf_shape(1) = max_nr_particle_in_rma_win CALL C_F_POINTER( buf_ptr(m), particle_in_win , buf_shape ) ! !-- In some cells of the coarse grid, there are contributions from more than !-- one child process. Therefore p_copy_particle_to_org_grid is done for one !-- child process per call DO ip = 1, pmc_s_get_child_npes( child_id ) nr_part = 0 part_adr = 0 CALL pmc_s_getdata_from_buffer( child_id, particle_transfer = .TRUE.,& child_process_nr = ip ) CALL p_copy_particle_to_org_grid( m ) ENDDO ENDDO #endif END SUBROUTINE pmcp_p_empty_particle_win !------------------------------------------------------------------------------! ! Description: ! ------------ !> parent routine: !> After the transfer mark all parent particles that are still inside on of the !> child areas for deletion. !------------------------------------------------------------------------------! SUBROUTINE pmcp_p_delete_particles_in_fine_grid_area IMPLICIT NONE LOGICAL :: to_delete !< particles outside of model domain are marked as to_delete INTEGER(iwp) :: i !< loop index (x grid) INTEGER(iwp) :: j !< loop index (y grid) INTEGER(iwp) :: k !< loop index (z grid) INTEGER(iwp) :: m !< loop index (number of particles) INTEGER(iwp) :: n !< loop index (number of childs) REAL(wp) :: dx_child !< child grid spacing REAL(wp) :: dy_child !< child grid spacing REAL(wp) :: dz_child !< child grid spacing REAL(wp) :: ny_coord !< north coordinate of child grid REAL(wp) :: ny_coord_b !< north coordinate of child grid boundary REAL(wp) :: lx_coord !< left coordinate of child grid REAL(wp) :: lx_coord_b !< left coordinate of child grid boundary REAL(wp) :: rx_coord !< right coordinate of child grid REAL(wp) :: rx_coord_b !< right coordinate of child grid boundary REAL(wp) :: sy_coord !< south coordinate of child grid REAL(wp) :: sy_coord_b !< south coordinate of child grid boundary REAL(wp) :: uz_coord !< upper coordinate of child grid REAL(wp) :: uz_coord_b !< upper coordinate of child grid boundary REAL(wp) :: x !< particle position REAL(wp) :: y !< particle position REAL(wp) :: z !< particle position #if defined( __parallel ) DO m = 1, get_number_of_children() CALL get_child_edges( m, lx_coord, lx_coord_b, rx_coord, rx_coord_b, & sy_coord, sy_coord_b, ny_coord, ny_coord_b, & uz_coord, uz_coord_b ) CALL get_child_gridspacing( m, dx_child, dy_child, dz_child ) DO i = nxl, nxr DO j = nys, nyn DO k = nzb, nzt number_of_particles = prt_count(k,j,i) IF ( number_of_particles == 0 ) CYCLE particles => grid_particles(k,j,i)%particles(1:number_of_particles) DO n = 1, number_of_particles x = particles(n)%x y = particles(n)%y z = particles(n)%z to_delete = ((x > lx_coord .AND. x < rx_coord) .AND. & (y > sy_coord .AND. y < ny_coord) .AND. & (z > 0.000000001 .AND. z < uz_coord)) IF ( to_delete ) THEN particles(n)%particle_mask = .FALSE. ENDIF ENDDO ENDDO ENDDO ENDDO ENDDO #endif END SUBROUTINE pmcp_p_delete_particles_in_fine_grid_area !------------------------------------------------------------------------------! ! Description: ! ------------ !> general routine: !> print the total number of of the current model and its child models into a file !------------------------------------------------------------------------------! SUBROUTINE pmcp_g_print_number_of_particles (my_time,local_nr_particles) USE pegrid, & ONLY: myid IMPLICIT NONE INTEGER(iwp),INTENT(IN) :: local_nr_particles !< REAL(wp),INTENT(IN) :: my_time !< LOGICAL, SAVE :: is_file_open=.FALSE. !< INTEGER(iwp) :: child_id !< INTEGER(iwp) :: child_nr_particles !< total number of particles in all child models INTEGER(iwp) :: ierr !< error code INTEGER(iwp) :: m !< loop index (number of childs) CHARACTER(LEN=16) :: fname !< filename INTEGER(iwp),DIMENSION(2) :: ivalr !< integer value to be received INTEGER(iwp),DIMENSION(2) :: ivals !< integer value to be send #if defined( __parallel ) child_nr_particles = 0 IF ( myid == 0 ) THEN IF ( cpl_id > 1 ) THEN ivals(1) = local_nr_particles CALL pmc_send_to_parent( ivals, 1, 0, 400, ierr ) ENDIF DO m = 1, SIZE( pmc_parent_for_child ) - 1 child_id = pmc_parent_for_child(m) CALL pmc_recv_from_child( child_id, ivalr, 1, 0, 400, ierr ) child_nr_particles = child_nr_particles + ivalr(1) ENDDO IF ( SIZE( pmc_parent_for_child ) > 1 ) THEN IF ( .NOT. is_file_open ) THEN !kk muss noch auf file_open umgestellt werden WRITE(fname,'(a,i2.2)') 'nr_particles_',cpl_id OPEN (333,file = fname) is_file_open = .true. ENDIF WRITE(333,'(f12.2,3i10)') my_time,local_nr_particles + child_nr_particles,local_nr_particles,child_nr_particles ENDIF ENDIF #endif END SUBROUTINE pmcp_g_print_number_of_particles !------------------------------------------------------------------------------! !------------------------------------------------------------------------------! ! Private subroutines !------------------------------------------------------------------------------! !------------------------------------------------------------------------------! !------------------------------------------------------------------------------! ! Description: ! ------------ !> child routine !> update the size of the local buffer (coarse_particles) !------------------------------------------------------------------------------! SUBROUTINE check_and_alloc_coarse_particle (ic,jc,nr,with_copy) IMPLICIT NONE INTEGER(iwp),INTENT(IN) :: ic !< coarse x grid index INTEGER(iwp),INTENT(IN) :: jc !< coarse y grid index INTEGER(iwp),INTENT(IN) :: nr !< number of particles to be transferred/stored in local buffer LOGICAL,INTENT(IN),OPTIONAL :: with_copy !< copy particles in buffer? or reallocate empty buffer LOGICAL :: with_copy_lo !< local variable of with copy INTEGER(iwp) :: new_size !< new size of the local buffer INTEGER(iwp) :: old_size !< old size of the local buffer TYPE(particle_type), DIMENSION(:), ALLOCATABLE :: tmp_particles_d !< #if defined( __parallel ) with_copy_lo = .FALSE. IF ( PRESENT( with_copy ) ) with_copy_lo = with_copy IF ( .NOT. ALLOCATED( coarse_particles(jc,ic)%parent_particles ) ) THEN new_size = MAX( nr, Min_particles_per_column ) ALLOCATE( coarse_particles(jc,ic)%parent_particles(new_size) ) ELSEIF ( nr > SIZE( coarse_particles(jc,ic)%parent_particles ) ) THEN old_size = SIZE( coarse_particles(jc,ic)%parent_particles ) new_size = old_size * ( 1.0_wp + alloc_factor / 100.0_wp ) new_size = MAX( nr, new_size, old_size + Min_particles_per_column ) ! !-- Copy existing particles to new particle buffer IF ( with_copy_lo ) THEN ALLOCATE( tmp_particles_d(old_size) ) tmp_particles_d(1:old_size) = coarse_particles(jc,ic)%parent_particles DEALLOCATE( coarse_particles(jc,ic)%parent_particles ) ALLOCATE( coarse_particles(jc,ic)%parent_particles(new_size) ) coarse_particles(jc,ic)%parent_particles(1:old_size) = tmp_particles_d(1:old_size) coarse_particles(jc,ic)%parent_particles(old_size+1:new_size) = zero_particle DEALLOCATE( tmp_particles_d ) ! !-- allocate or reallocate an empty buffer ELSE DEALLOCATE( coarse_particles(jc,ic)%parent_particles ) ALLOCATE( coarse_particles(jc,ic)%parent_particles(new_size) ) ENDIF ENDIF #endif END SUBROUTINE check_and_alloc_coarse_particle !------------------------------------------------------------------------------! ! Description: ! ------------ !> child routine: !> copy/sort particles out of the local buffer into the respective grid boxes !------------------------------------------------------------------------------! SUBROUTINE c_copy_particle_to_child_grid IMPLICIT NONE INTEGER(iwp) :: ic !< coarse x grid index INTEGER(iwp) :: icl !< left boundary in coarse(parent) index space INTEGER(iwp) :: icr !< right boundary in coarse(parent) index space INTEGER(iwp) :: ip !< x grid index INTEGER(iwp) :: jc !< coarse y grid index INTEGER(iwp) :: jcn !< north boundary in coarse(parent) index space INTEGER(iwp) :: jcs !< south boundary in coarse(parent) index space INTEGER(iwp) :: jp !< y grid index INTEGER(iwp) :: kp !< z grid index INTEGER(iwp) :: n !< loop index (number of particles) INTEGER(iwp) :: nr !< short variable for name number or particles REAL(wp) :: xc !< child x coordinate REAL(wp) :: xoc !< child x origin REAL(wp) :: yc !< child y coordinate REAL(wp) :: yoc !< child y origin REAL(wp) :: zc !< child z coordinate #if defined( __parallel ) ! !-- Child domain boundaries in the parent index space icl = coarse_bound(1) icr = coarse_bound(2) jcs = coarse_bound(3) jcn = coarse_bound(4) DO ic = icl, icr DO jc = jcs, jcn nr = coarse_particles(jc,ic)%nr_particle IF ( nr > 0 ) THEN DO n = 1, nr xc = coarse_particles(jc,ic)%parent_particles(n)%x-lower_left_coord_x yc = coarse_particles(jc,ic)%parent_particles(n)%y-lower_left_coord_y zc = coarse_particles(jc,ic)%parent_particles(n)%z xoc = coarse_particles(jc,ic)%parent_particles(n)%origin_x-lower_left_coord_x yoc = coarse_particles(jc,ic)%parent_particles(n)%origin_y-lower_left_coord_y ip = xc / dx jp = yc / dy kp = nzt DO WHILE ( zw(kp-1) > zc .AND. kp > nzb + 1 ) ! kk search loop has to be optimzed !!! kp = kp - 1 ENDDO IF ( ip >= nxl .AND. ip <= nxr .AND. jp >= nys .AND. jp <= nyn ) THEN prt_count(kp,jp,ip) = prt_count(kp,jp,ip) + 1 IF ( prt_count(kp,jp,ip) > SIZE( grid_particles(kp,jp,ip)%particles ) ) THEN CALL realloc_particles_array( ip, jp, kp ) ENDIF coarse_particles(jc,ic)%parent_particles(n)%x = xc ! Adjust coordinates to child grid coarse_particles(jc,ic)%parent_particles(n)%y = yc coarse_particles(jc,ic)%parent_particles(n)%origin_x = xoc ! Adjust origins to child grid coarse_particles(jc,ic)%parent_particles(n)%origin_y = yoc grid_particles(kp,jp,ip)%particles(prt_count(kp,jp,ip)) = coarse_particles(jc,ic)%parent_particles(n) ENDIF ENDDO ENDIF ENDDO ENDDO #endif END SUBROUTINE c_copy_particle_to_child_grid !------------------------------------------------------------------------------! ! Description: ! ------------ !> child routine: !> copy particles which left the model area into the local buffer !------------------------------------------------------------------------------! SUBROUTINE c_copy_particle_to_coarse_grid IMPLICIT NONE LOGICAL :: boundary_particle !< INTEGER(iwp) :: i !< loop index (x grid) INTEGER(iwp) :: ic !< loop index (coarse x grid) INTEGER(iwp) :: icl !< left boundary in coarse(parent) index space INTEGER(iwp) :: icr !< left boundary in coarse(parent) index space INTEGER(iwp) :: ierr !< error code INTEGER(iwp) :: j !< loop index (y grid) INTEGER(iwp) :: jc !< loop index (coarse y grid) INTEGER(iwp) :: jcs !< south boundary in coarse(parent) index space INTEGER(iwp) :: jcn !< north boundary in coarse(parent) index space INTEGER(iwp) :: k !< loop index (z grid) INTEGER(iwp) :: n !< loop index (number of particles) REAL(wp) :: x !< x coordinate REAL(wp) :: xo !< x origin REAL(wp) :: x_left !< absolute left boundary REAL(wp) :: x_right !< absolute right boundary REAL(wp) :: y !< left boundary in coarse(parent) index space REAL(wp) :: yo !< y origin REAL(wp) :: y_north !< absolute north boundary REAL(wp) :: y_south !< absoulte south boundary REAL(wp) :: z !< z coordinate #if defined( __parallel ) ! !-- Child domain boundaries in the parent index space icl = coarse_bound(1) icr = coarse_bound(2) jcs = coarse_bound(3) jcn = coarse_bound(4) ! !-- absolute coordinates x_left = coord_x(0) x_right = coord_x(nx) + dx y_south = coord_y(0) y_north = coord_y(ny) + dy ! Clear Particle Buffer DO ic = icl, icr DO jc = jcs, jcn coarse_particles(jc,ic)%nr_particle = 0 ENDDO ENDDO ! !-- Determine particles which leave the inner area in east or west dirextion !-- Compute only first (nxl) and last (nxr) loop iterration. DO i = nxl, nxr DO j = nys, nyn DO k = nzb + 1, nzt number_of_particles = prt_count(k,j,i) IF ( number_of_particles <= 0 ) CYCLE particles => grid_particles(k,j,i)%particles(1:number_of_particles) DO n = 1, number_of_particles IF ( particles(n)%particle_mask ) THEN x = particles(n)%x+ lower_left_coord_x y = particles(n)%y+ lower_left_coord_y xo = particles(n)%origin_x + lower_left_coord_x yo = particles(n)%origin_y + lower_left_coord_y z = particles(n)%z boundary_particle = .FALSE. boundary_particle = boundary_particle .OR. (x < x_left .OR. x > x_right) boundary_particle = boundary_particle .OR. (y < y_south .OR. y > y_north) boundary_particle = boundary_particle .OR. (z > zw(nzt)) IF ( boundary_particle ) THEN ic = x / pg%dx !TODO anpassen auf Mehrfachnesting jc = y / pg%dy IF ( ic >= icl .AND. ic <= icr .AND. jc >= jcs .AND. jc <= jcn ) THEN coarse_particles(jc,ic)%nr_particle = coarse_particles(jc,ic)%nr_particle + 1 CALL check_and_alloc_coarse_particle( ic, jc, coarse_particles(jc,ic)%nr_particle, with_copy=.TRUE. ) coarse_particles(jc,ic)%parent_particles(coarse_particles(jc,ic)%nr_particle) = particles(n) coarse_particles(jc,ic)%parent_particles(coarse_particles(jc,ic)%nr_particle)%x = x !adapt to global coordinates coarse_particles(jc,ic)%parent_particles(coarse_particles(jc,ic)%nr_particle)%y = y coarse_particles(jc,ic)%parent_particles(coarse_particles(jc,ic)%nr_particle)%origin_x = xo coarse_particles(jc,ic)%parent_particles(coarse_particles(jc,ic)%nr_particle)%origin_y = yo !-- Mark particle as deleted after copying it to the transfer buffer grid_particles(k,j,i)%particles(n)%particle_mask = .FALSE. ELSE WRITE(9,'(a,10i6)') 'This should not happen ',i,j,k,ic,jc,icl,icr,jcs,jcn CALL MPI_Abort( MPI_COMM_WORLD, 9999, ierr ) ENDIF ENDIF ENDIF ENDDO ENDDO ENDDO ENDDO ! !- Pack particles (eliminate those marked for deletion), !- determine new number of particles CALL lpm_sort_in_subboxes #endif END SUBROUTINE c_copy_particle_to_coarse_grid !------------------------------------------------------------------------------! ! Description: ! ------------ !> parent routine: !> copy/sort particles from the MPI window into the respective grid boxes !------------------------------------------------------------------------------! SUBROUTINE p_copy_particle_to_org_grid( m ) IMPLICIT NONE INTEGER(iwp),INTENT(IN) :: m !< INTEGER(iwp) :: i !< loop index (x grid) INTEGER(iwp) :: j !< loop index (y grid) INTEGER(iwp) :: k !< loop index (z grid) INTEGER(iwp) :: n !< loop index (nr part) INTEGER(iwp) :: nr !< short variable name for nr_part INTEGER(iwp) :: pindex !< short variable name part_adr REAL(wp) :: x !< x coordinate REAL(wp) :: xo !< x origin REAL(wp) :: y !< y coordinate REAL(wp) :: yo !< y origin REAL(wp) :: z !< z coordinate INTEGER(iwp),DIMENSION(1) :: buf_shape !< #if defined( __parallel ) buf_shape(1) = max_nr_particle_in_rma_win CALL C_F_POINTER( buf_ptr(m), particle_in_win , buf_shape ) DO i = nxl, nxr DO j = nys, nyn nr = nr_part(j,i) IF ( nr > 0 ) THEN pindex = part_adr(j,i) DO n = 1, nr x = particle_in_win(pindex)%x-lower_left_coord_x y = particle_in_win(pindex)%y-lower_left_coord_y z = particle_in_win(pindex)%z xo = particle_in_win(pindex)%origin_x-lower_left_coord_x yo = particle_in_win(pindex)%origin_y-lower_left_coord_y k = nzt + 1 DO WHILE ( zw(k-1) > z .AND. k > nzb + 1 ) ! kk search loop has to be optimzed !!! k = k - 1 END DO prt_count(k,j,i) = prt_count(k,j,i) + 1 IF ( prt_count(k,j,i) > SIZE( grid_particles(k,j,i)%particles ) ) THEN CALL realloc_particles_array( i, j, k ) ENDIF grid_particles(k,j,i)%particles(prt_count(k,j,i)) = particle_in_win(pindex) ! !-- Update particle positions and origins relative to parent domain grid_particles(k,j,i)%particles(prt_count(k,j,i))%x = x grid_particles(k,j,i)%particles(prt_count(k,j,i))%y = y grid_particles(k,j,i)%particles(prt_count(k,j,i))%origin_x = xo grid_particles(k,j,i)%particles(prt_count(k,j,i))%origin_y = yo pindex = pindex + 1 ENDDO ENDIF ENDDO ENDDO #endif END SUBROUTINE p_copy_particle_to_org_grid END MODULE pmc_particle_interface