MODULE pmc_parent !------------------------------------------------------------------------------! ! 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 1997-2016 Leibniz Universitaet Hannover !------------------------------------------------------------------------------! ! ! Current revisions: ! ------------------ ! ! ! Former revisions: ! ----------------- ! $Id: pmc_parent_mod.f90 2001 2016-08-20 18:41:22Z gronemeier $ ! ! 2000 2016-08-20 18:09:15Z knoop ! Forced header and separation lines into 80 columns ! ! 1938 2016-06-13 15:26:05Z hellstea ! Minor clean up. ! ! 1901 2016-05-04 15:39:38Z raasch ! Module renamed. Code clean up. The words server/client changed to parent/child. ! ! 1900 2016-05-04 15:27:53Z raasch ! re-formatted to match PALM style ! ! 1850 2016-04-08 13:29:27Z maronga ! Module renamed ! ! ! 1833 2016-04-07 14:23:03Z raasch ! gfortran requires pointer attributes for some array declarations, ! long line wrapped ! ! 1808 2016-04-05 19:44:00Z raasch ! MPI module used by default on all machines ! ! 1797 2016-03-21 16:50:28Z raasch ! introduction of different datatransfer modes ! ! 1791 2016-03-11 10:41:25Z raasch ! Debug write-statements commented out ! ! 1786 2016-03-08 05:49:27Z raasch ! change in child-parent data transfer: parent now gets data from child ! instead that child put's it to the parent ! ! 1779 2016-03-03 08:01:28Z raasch ! kind=dp replaced by wp, ! error messages removed or changed to PALM style, dim_order removed ! array management changed from linked list to sequential loop ! ! 1766 2016-02-29 08:37:15Z raasch ! modifications to allow for using PALM's pointer version ! +new routine pmc_s_set_active_data_array ! ! 1764 2016-02-28 12:45:19Z raasch ! cpp-statement added (nesting can only be used in parallel mode) ! ! 1762 2016-02-25 12:31:13Z hellstea ! Initial revision by K. Ketelsen ! ! Description: ! ------------ ! ! Parent part of Palm Model Coupler !-------------------------------------------------------------------------------! #if defined( __parallel ) USE, INTRINSIC :: ISO_C_BINDING #if defined( __mpifh ) INCLUDE "mpif.h" #else USE MPI #endif USE kinds USE pmc_general, & ONLY: arraydef, childdef, da_namedef, da_namelen, pedef, & pmc_g_setname, pmc_max_array, pmc_max_models, pmc_sort USE pmc_handle_communicator, & ONLY: m_model_comm,m_model_rank,m_model_npes, m_to_child_comm, & m_world_rank, pmc_parent_for_child USE pmc_mpi_wrapper, & ONLY: pmc_alloc_mem, pmc_bcast, pmc_time IMPLICIT NONE PRIVATE SAVE TYPE childindexdef INTEGER :: nrpoints !< INTEGER, DIMENSION(:,:), ALLOCATABLE :: index_list_2d !< END TYPE childindexdef TYPE(childdef), DIMENSION(pmc_max_models) :: children !< TYPE(childindexdef), DIMENSION(pmc_max_models) :: indchildren !< INTEGER :: next_array_in_list = 0 !< PUBLIC pmc_parent_for_child INTERFACE pmc_parentinit MODULE PROCEDURE pmc_parentinit END INTERFACE pmc_parentinit INTERFACE pmc_s_set_2d_index_list MODULE PROCEDURE pmc_s_set_2d_index_list END INTERFACE pmc_s_set_2d_index_list INTERFACE pmc_s_clear_next_array_list MODULE PROCEDURE pmc_s_clear_next_array_list END INTERFACE pmc_s_clear_next_array_list INTERFACE pmc_s_getnextarray MODULE PROCEDURE pmc_s_getnextarray END INTERFACE pmc_s_getnextarray INTERFACE pmc_s_set_dataarray MODULE PROCEDURE pmc_s_set_dataarray_2d MODULE PROCEDURE pmc_s_set_dataarray_3d END INTERFACE pmc_s_set_dataarray INTERFACE pmc_s_setind_and_allocmem MODULE PROCEDURE pmc_s_setind_and_allocmem END INTERFACE pmc_s_setind_and_allocmem INTERFACE pmc_s_fillbuffer MODULE PROCEDURE pmc_s_fillbuffer END INTERFACE pmc_s_fillbuffer INTERFACE pmc_s_getdata_from_buffer MODULE PROCEDURE pmc_s_getdata_from_buffer END INTERFACE pmc_s_getdata_from_buffer INTERFACE pmc_s_set_active_data_array MODULE PROCEDURE pmc_s_set_active_data_array END INTERFACE pmc_s_set_active_data_array PUBLIC pmc_parentinit, pmc_s_clear_next_array_list, pmc_s_fillbuffer, & pmc_s_getdata_from_buffer, pmc_s_getnextarray, & pmc_s_setind_and_allocmem, pmc_s_set_active_data_array, & pmc_s_set_dataarray, pmc_s_set_2d_index_list CONTAINS SUBROUTINE pmc_parentinit IMPLICIT NONE INTEGER :: childid !< INTEGER :: i !< INTEGER :: j !< INTEGER :: istat !< DO i = 1, SIZE( pmc_parent_for_child )-1 childid = pmc_parent_for_child( i ) children(childid)%model_comm = m_model_comm children(childid)%inter_comm = m_to_child_comm(childid) ! !-- Get rank and size CALL MPI_COMM_RANK( children(childid)%model_comm, & children(childid)%model_rank, istat ) CALL MPI_COMM_SIZE( children(childid)%model_comm, & children(childid)%model_npes, istat ) CALL MPI_COMM_REMOTE_SIZE( children(childid)%inter_comm, & children(childid)%inter_npes, istat ) ! !-- Intra communicater is used for MPI_GET CALL MPI_INTERCOMM_MERGE( children(childid)%inter_comm, .FALSE., & children(childid)%intra_comm, istat ) CALL MPI_COMM_RANK( children(childid)%intra_comm, & children(childid)%intra_rank, istat ) ALLOCATE( children(childid)%pes(children(childid)%inter_npes)) ! !-- Allocate array of TYPE arraydef for all child PEs to store information !-- of the transfer array DO j = 1, children(childid)%inter_npes ALLOCATE( children(childid)%pes(j)%array_list(pmc_max_array) ) ENDDO CALL get_da_names_from_child (childid) ENDDO END SUBROUTINE pmc_parentinit SUBROUTINE pmc_s_set_2d_index_list( childid, index_list ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< INTEGER, DIMENSION(:,:), INTENT(INOUT) :: index_list !< INTEGER :: ian !< INTEGER :: ic !< INTEGER :: ie !< INTEGER :: ip !< INTEGER :: is !< INTEGER :: istat !< INTEGER :: n !< IF ( m_model_rank == 0 ) THEN ! !-- Sort to ascending parent PE order CALL pmc_sort( index_list, 6 ) is = 1 DO ip = 0, m_model_npes-1 ! !-- Split into parent PEs ie = is - 1 ! !-- There may be no entry for this PE IF ( is <= SIZE( index_list,2 ) .AND. ie >= 0 ) THEN DO WHILE ( index_list(6,ie+1 ) == ip ) ie = ie + 1 IF ( ie == SIZE( index_list,2 ) ) EXIT ENDDO ian = ie - is + 1 ELSE is = -1 ie = -2 ian = 0 ENDIF ! !-- Send data to other parent PEs IF ( ip == 0 ) THEN indchildren(childid)%nrpoints = ian IF ( ian > 0) THEN ALLOCATE( indchildren(childid)%index_list_2d(6,ian) ) indchildren(childid)%index_list_2d(:,1:ian) = & index_list(:,is:ie) ENDIF ELSE CALL MPI_SEND( ian, 1, MPI_INTEGER, ip, 1000, m_model_comm, & istat ) IF ( ian > 0) THEN CALL MPI_SEND( index_list(1,is), 6*ian, MPI_INTEGER, ip, & 1001, m_model_comm, istat ) ENDIF ENDIF is = ie + 1 ENDDO ELSE CALL MPI_RECV( indchildren(childid)%nrpoints, 1, MPI_INTEGER, 0, 1000, & m_model_comm, MPI_STATUS_IGNORE, istat ) ian = indchildren(childid)%nrpoints IF ( ian > 0 ) THEN ALLOCATE( indchildren(childid)%index_list_2d(6,ian) ) CALL MPI_RECV( indchildren(childid)%index_list_2d, 6*ian, & MPI_INTEGER, 0, 1001, m_model_comm, & MPI_STATUS_IGNORE, istat) ENDIF ENDIF CALL set_pe_index_list( childid, children(childid), & indchildren(childid)%index_list_2d, & indchildren(childid)%nrpoints ) END SUBROUTINE pmc_s_set_2d_index_list SUBROUTINE pmc_s_clear_next_array_list IMPLICIT NONE next_array_in_list = 0 END SUBROUTINE pmc_s_clear_next_array_list LOGICAL FUNCTION pmc_s_getnextarray( childid, myname ) ! !-- List handling is still required to get minimal interaction with !-- pmc_interface !-- TODO: what does "still" mean? Is there a chance to change this! CHARACTER(LEN=*), INTENT(OUT) :: myname !< INTEGER(iwp), INTENT(IN) :: childid !< TYPE(arraydef), POINTER :: ar TYPE(pedef), POINTER :: ape next_array_in_list = next_array_in_list + 1 ! !-- Array names are the same on all children PEs, so take first PE to get the name ape => children(childid)%pes(1) IF ( next_array_in_list > ape%nr_arrays ) THEN ! !-- All arrays are done pmc_s_getnextarray = .FALSE. RETURN ENDIF ar => ape%array_list(next_array_in_list) myname = ar%name ! !-- Return true if legal array !-- TODO: what does this comment mean? Can there be non-legal arrays?? pmc_s_getnextarray = .TRUE. END FUNCTION pmc_s_getnextarray SUBROUTINE pmc_s_set_dataarray_2d( childid, array, array_2 ) IMPLICIT NONE INTEGER,INTENT(IN) :: childid !< REAL(wp), INTENT(IN), DIMENSION(:,:), POINTER :: array !< REAL(wp), INTENT(IN), DIMENSION(:,:), POINTER, OPTIONAL :: array_2 !< INTEGER :: nrdims !< INTEGER, DIMENSION(4) :: dims !< TYPE(C_PTR) :: array_adr !< TYPE(C_PTR) :: second_adr !< dims = 1 nrdims = 2 dims(1) = SIZE( array,1 ) dims(2) = SIZE( array,2 ) array_adr = C_LOC( array ) IF ( PRESENT( array_2 ) ) THEN second_adr = C_LOC(array_2) CALL pmc_s_setarray( childid, nrdims, dims, array_adr, & second_adr = second_adr) ELSE CALL pmc_s_setarray( childid, nrdims, dims, array_adr ) ENDIF END SUBROUTINE pmc_s_set_dataarray_2d SUBROUTINE pmc_s_set_dataarray_3d( childid, array, nz_cl, nz, array_2 ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< INTEGER, INTENT(IN) :: nz !< INTEGER, INTENT(IN) :: nz_cl !< REAL(wp), INTENT(IN), DIMENSION(:,:,:), POINTER :: array !< REAL(wp), INTENT(IN), DIMENSION(:,:,:), POINTER, OPTIONAL :: array_2 !< INTEGER :: nrdims !< INTEGER, DIMENSION(4) :: dims !< TYPE(C_PTR) :: array_adr !< TYPE(C_PTR) :: second_adr !< ! !-- TODO: the next assignment seems to be obsolete. Please check! dims = 1 dims = 0 nrdims = 3 dims(1) = SIZE( array,1 ) dims(2) = SIZE( array,2 ) dims(3) = SIZE( array,3 ) dims(4) = nz_cl+dims(1)-nz ! works for first dimension 1:nz and 0:nz+1 array_adr = C_LOC(array) ! !-- In PALM's pointer version, two indices have to be stored internally. !-- The active address of the data array is set in swap_timelevel. IF ( PRESENT( array_2 ) ) THEN second_adr = C_LOC( array_2 ) CALL pmc_s_setarray( childid, nrdims, dims, array_adr, & second_adr = second_adr) ELSE CALL pmc_s_setarray( childid, nrdims, dims, array_adr ) ENDIF END SUBROUTINE pmc_s_set_dataarray_3d SUBROUTINE pmc_s_setind_and_allocmem( childid ) USE control_parameters, & ONLY: message_string IMPLICIT NONE ! !-- Naming convention for appendices: _pc -> parent to child transfer !-- _cp -> child to parent transfer !-- send -> parent to child transfer !-- recv -> child to parent transfer INTEGER, INTENT(IN) :: childid !< INTEGER :: arlen !< INTEGER :: i !< INTEGER :: ierr !< INTEGER :: istat !< INTEGER :: j !< INTEGER :: myindex !< INTEGER :: rcount !< count MPI requests INTEGER :: tag !< INTEGER(idp) :: bufsize !< size of MPI data window INTEGER(KIND=MPI_ADDRESS_KIND) :: winsize !< INTEGER, DIMENSION(1024) :: req !< TYPE(C_PTR) :: base_ptr !< TYPE(pedef), POINTER :: ape !< TYPE(arraydef), POINTER :: ar !< REAL(wp),DIMENSION(:), POINTER, SAVE :: base_array_pc !< base array for parent to child transfer REAL(wp),DIMENSION(:), POINTER, SAVE :: base_array_cp !< base array for child to parent transfer ! !-- Parent to child direction myindex = 1 rcount = 0 bufsize = 8 ! !-- First stride: compute size and set index DO i = 1, children(childid)%inter_npes ape => children(childid)%pes(i) tag = 200 DO j = 1, ape%nr_arrays ar => ape%array_list(j) IF ( ar%nrdims == 2 ) THEN arlen = ape%nrele ELSEIF ( ar%nrdims == 3 ) THEN arlen = ape%nrele * ar%a_dim(4) ELSE arlen = -1 ENDIF ar%sendindex = myindex tag = tag + 1 rcount = rcount + 1 CALL MPI_ISEND( myindex, 1, MPI_INTEGER, i-1, tag, & children(childid)%inter_comm, req(rcount), ierr ) ! !-- Maximum of 1024 outstanding requests !-- TODO: what does this limit mean? IF ( rcount == 1024 ) THEN CALL MPI_WAITALL( rcount, req, MPI_STATUSES_IGNORE, ierr ) rcount = 0 ENDIF myindex = myindex + arlen bufsize = bufsize + arlen ar%sendsize = arlen ENDDO IF ( rcount > 0 ) THEN CALL MPI_WAITALL( rcount, req, MPI_STATUSES_IGNORE, ierr ) ENDIF ENDDO ! !-- Create RMA (One Sided Communication) window for data buffer parent to !-- child transfer. !-- The buffer of MPI_GET (counterpart of transfer) can be PE-local, i.e. !-- it can but must not be part of the MPI RMA window. Only one RMA window is !-- required to prepare the data for !-- parent -> child transfer on the parent side !-- and for !-- child -> parent transfer on the child side CALL pmc_alloc_mem( base_array_pc, bufsize ) children(childid)%totalbuffersize = bufsize * wp winsize = bufsize * wp CALL MPI_WIN_CREATE( base_array_pc, winsize, wp, MPI_INFO_NULL, & children(childid)%intra_comm, & children(childid)%win_parent_child, ierr ) ! !-- Open window to set data CALL MPI_WIN_FENCE( 0, children(childid)%win_parent_child, ierr ) ! !-- Second stride: set buffer pointer DO i = 1, children(childid)%inter_npes ape => children(childid)%pes(i) DO j = 1, ape%nr_arrays ar => ape%array_list(j) ar%sendbuf = C_LOC( base_array_pc(ar%sendindex) ) IF ( ar%sendindex + ar%sendsize > bufsize ) THEN WRITE( message_string, '(a,i4,4i7,1x,a)' ) & 'parent buffer too small ',i, & ar%sendindex,ar%sendsize,ar%sendindex+ar%sendsize, & bufsize,trim(ar%name) CALL message( 'pmc_s_setind_and_allocmem', 'PA0429', 3, 2, 0, 6, 0 ) ENDIF ENDDO ENDDO ! !-- Child to parent direction bufsize = 8 ! !-- First stride: compute size and set index DO i = 1, children(childid)%inter_npes ape => children(childid)%pes(i) tag = 300 DO j = 1, ape%nr_arrays ar => ape%array_list(j) ! !-- Receive index from child tag = tag + 1 CALL MPI_RECV( myindex, 1, MPI_INTEGER, i-1, tag, & children(childid)%inter_comm, MPI_STATUS_IGNORE, ierr ) IF ( ar%nrdims == 3 ) THEN bufsize = MAX( bufsize, ape%nrele * ar%a_dim(4) ) ELSE bufsize = MAX( bufsize, ape%nrele ) ENDIF ar%recvindex = myindex ENDDO ENDDO ! !-- Create RMA (one sided communication) data buffer. !-- The buffer for MPI_GET can be PE local, i.e. it can but must not be part of !-- the MPI RMA window CALL pmc_alloc_mem( base_array_cp, bufsize, base_ptr ) children(childid)%totalbuffersize = bufsize * wp CALL MPI_BARRIER( children(childid)%intra_comm, ierr ) ! !-- Second stride: set buffer pointer DO i = 1, children(childid)%inter_npes ape => children(childid)%pes(i) DO j = 1, ape%nr_arrays ar => ape%array_list(j) ar%recvbuf = base_ptr ENDDO ENDDO END SUBROUTINE pmc_s_setind_and_allocmem SUBROUTINE pmc_s_fillbuffer( childid, waittime ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< REAL(wp), INTENT(OUT), OPTIONAL :: waittime !< INTEGER :: ierr !< INTEGER :: ij !< INTEGER :: ip !< INTEGER :: istat !< INTEGER :: j !< INTEGER :: myindex !< INTEGER, DIMENSION(1) :: buf_shape REAL(wp) :: t1 !< REAL(wp) :: t2 !< REAL(wp), POINTER, DIMENSION(:) :: buf !< REAL(wp), POINTER, DIMENSION(:,:) :: data_2d !< REAL(wp), POINTER, DIMENSION(:,:,:) :: data_3d !< TYPE(pedef), POINTER :: ape !< TYPE(arraydef), POINTER :: ar !< ! !-- Synchronization of the model is done in pmci_synchronize. !-- Therefor the RMA window can be filled without !-- sychronization at this point and a barrier is not necessary. !-- Please note that waittime has to be set in pmc_s_fillbuffer AND !-- pmc_c_getbuffer IF ( PRESENT( waittime) ) THEN t1 = pmc_time() CALL MPI_BARRIER( children(childid)%intra_comm, ierr ) t2 = pmc_time() waittime = t2- t1 ENDIF DO ip = 1, children(childid)%inter_npes ape => children(childid)%pes(ip) DO j = 1, ape%nr_arrays ar => ape%array_list(j) myindex = 1 IF ( ar%nrdims == 2 ) THEN buf_shape(1) = ape%nrele CALL C_F_POINTER( ar%sendbuf, buf, buf_shape ) CALL C_F_POINTER( ar%data, data_2d, ar%a_dim(1:2) ) DO ij = 1, ape%nrele buf(myindex) = data_2d(ape%locind(ij)%j,ape%locind(ij)%i) myindex = myindex + 1 ENDDO ELSEIF ( ar%nrdims == 3 ) THEN buf_shape(1) = ape%nrele*ar%a_dim(4) CALL C_F_POINTER( ar%sendbuf, buf, buf_shape ) CALL C_F_POINTER( ar%data, data_3d, ar%a_dim(1:3) ) DO ij = 1, ape%nrele buf(myindex:myindex+ar%a_dim(4)-1) = & data_3d(1:ar%a_dim(4),ape%locind(ij)%j,ape%locind(ij)%i) myindex = myindex + ar%a_dim(4) ENDDO ENDIF ENDDO ENDDO ! !-- Buffer is filled CALL MPI_BARRIER( children(childid)%intra_comm, ierr ) END SUBROUTINE pmc_s_fillbuffer SUBROUTINE pmc_s_getdata_from_buffer( childid, waittime ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< REAL(wp), INTENT(OUT), OPTIONAL :: waittime !< INTEGER :: ierr !< INTEGER :: ij !< INTEGER :: ip !< INTEGER :: istat !< INTEGER :: j !< INTEGER :: myindex !< INTEGER :: nr !< INTEGER :: target_pe !< INTEGER(kind=MPI_ADDRESS_KIND) :: target_disp !< INTEGER, DIMENSION(1) :: buf_shape !< REAL(wp) :: t1 !< REAL(wp) :: t2 !< REAL(wp), POINTER, DIMENSION(:) :: buf !< REAL(wp), POINTER, DIMENSION(:,:) :: data_2d !< REAL(wp), POINTER, DIMENSION(:,:,:) :: data_3d !< TYPE(pedef), POINTER :: ape !< TYPE(arraydef), POINTER :: ar !< t1 = pmc_time() ! !-- Wait for child to fill buffer CALL MPI_BARRIER( children(childid)%intra_comm, ierr ) t2 = pmc_time() - t1 IF ( PRESENT( waittime ) ) waittime = t2 ! !-- TODO: check next statement !-- Fence might do it, test later !-- CALL MPI_WIN_FENCE( 0, children(childid)%win_parent_child, ierr) CALL MPI_BARRIER( children(childid)%intra_comm, ierr ) DO ip = 1, children(childid)%inter_npes ape => children(childid)%pes(ip) DO j = 1, ape%nr_arrays ar => ape%array_list(j) IF ( ar%recvindex < 0 ) CYCLE IF ( ar%nrdims == 2 ) THEN nr = ape%nrele ELSEIF ( ar%nrdims == 3 ) THEN nr = ape%nrele * ar%a_dim(4) ENDIF buf_shape(1) = nr CALL C_F_POINTER( ar%recvbuf, buf, buf_shape ) ! !-- MPI passive target RMA IF ( nr > 0 ) THEN target_disp = ar%recvindex - 1 ! !-- Child PEs are located behind parent PEs target_pe = ip - 1 + m_model_npes CALL MPI_WIN_LOCK( MPI_LOCK_SHARED, target_pe, 0, & children(childid)%win_parent_child, ierr ) CALL MPI_GET( buf, nr, MPI_REAL, target_pe, target_disp, nr, & MPI_REAL, children(childid)%win_parent_child, ierr ) CALL MPI_WIN_UNLOCK( target_pe, & children(childid)%win_parent_child, ierr ) ENDIF myindex = 1 IF ( ar%nrdims == 2 ) THEN CALL C_F_POINTER( ar%data, data_2d, ar%a_dim(1:2) ) DO ij = 1, ape%nrele data_2d(ape%locind(ij)%j,ape%locind(ij)%i) = buf(myindex) myindex = myindex + 1 ENDDO ELSEIF ( ar%nrdims == 3 ) THEN CALL C_F_POINTER( ar%data, data_3d, ar%a_dim(1:3)) DO ij = 1, ape%nrele data_3d(1:ar%a_dim(4),ape%locind(ij)%j,ape%locind(ij)%i) = & buf(myindex:myindex+ar%a_dim(4)-1) myindex = myindex + ar%a_dim(4) ENDDO ENDIF ENDDO ENDDO END SUBROUTINE pmc_s_getdata_from_buffer SUBROUTINE get_da_names_from_child( childid ) ! !-- Get data array description and name from child IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< TYPE(da_namedef) :: myname !< DO CALL pmc_bcast( myname%couple_index, 0, comm=m_to_child_comm(childid) ) IF ( myname%couple_index == -1 ) EXIT CALL pmc_bcast( myname%parentdesc, 0, comm=m_to_child_comm(childid) ) CALL pmc_bcast( myname%nameonparent, 0, comm=m_to_child_comm(childid) ) CALL pmc_bcast( myname%childdesc, 0, comm=m_to_child_comm(childid) ) CALL pmc_bcast( myname%nameonchild, 0, comm=m_to_child_comm(childid) ) CALL pmc_g_setname( children(childid), myname%couple_index, & myname%nameonparent ) ENDDO END SUBROUTINE get_da_names_from_child SUBROUTINE pmc_s_setarray(childid, nrdims, dims, array_adr, second_adr ) ! !-- Set array for child inter PE 0 IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< INTEGER, INTENT(IN) :: nrdims !< INTEGER, INTENT(IN), DIMENSION(:) :: dims !< TYPE(C_PTR), INTENT(IN) :: array_adr !< TYPE(C_PTR), INTENT(IN), OPTIONAL :: second_adr !< INTEGER :: i !< local counter TYPE(pedef), POINTER :: ape !< TYPE(arraydef), POINTER :: ar !< DO i = 1, children(childid)%inter_npes ape => children(childid)%pes(i) ar => ape%array_list(next_array_in_list) ar%nrdims = nrdims ar%a_dim = dims ar%data = array_adr IF ( PRESENT( second_adr ) ) THEN ar%po_data(1) = array_adr ar%po_data(2) = second_adr ELSE ar%po_data(1) = C_NULL_PTR ar%po_data(2) = C_NULL_PTR ENDIF ENDDO END SUBROUTINE pmc_s_setarray SUBROUTINE pmc_s_set_active_data_array( childid, iactive ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< INTEGER, INTENT(IN) :: iactive !< INTEGER :: i !< INTEGER :: ip !< INTEGER :: j !< TYPE(pedef), POINTER :: ape !< TYPE(arraydef), POINTER :: ar !< DO ip = 1, children(childid)%inter_npes ape => children(childid)%pes(ip) DO j = 1, ape%nr_arrays ar => ape%array_list(j) IF ( iactive == 1 .OR. iactive == 2 ) THEN ar%data = ar%po_data(iactive) ENDIF ENDDO ENDDO END SUBROUTINE pmc_s_set_active_data_array SUBROUTINE set_pe_index_list( childid, mychild, index_list, nrp ) IMPLICIT NONE INTEGER, INTENT(IN) :: childid !< INTEGER, INTENT(IN), DIMENSION(:,:) :: index_list !< INTEGER, INTENT(IN) :: nrp !< TYPE(childdef), INTENT(INOUT) :: mychild !< INTEGER :: i !< INTEGER :: ierr !< INTEGER :: ind !< INTEGER :: indwin !< INTEGER :: indwin2 !< INTEGER :: i2 !< INTEGER :: j !< INTEGER :: rempe !< INTEGER(KIND=MPI_ADDRESS_KIND) :: winsize !< INTEGER, DIMENSION(mychild%inter_npes) :: remind !< INTEGER, DIMENSION(:), POINTER :: remindw !< INTEGER, DIMENSION(:), POINTER :: rldef !< TYPE(pedef), POINTER :: ape !< ! !-- First, count entries for every remote child PE DO i = 1, mychild%inter_npes ape => mychild%pes(i) ape%nrele = 0 ENDDO ! !-- Loop over number of coarse grid cells DO j = 1, nrp rempe = index_list(5,j) + 1 ! PE number on remote PE ape => mychild%pes(rempe) ape%nrele = ape%nrele + 1 ! Increment number of elements for this child PE ENDDO DO i = 1, mychild%inter_npes ape => mychild%pes(i) ALLOCATE( ape%locind(ape%nrele) ) ENDDO remind = 0 ! !-- Second, create lists !-- Loop over number of coarse grid cells DO j = 1, nrp rempe = index_list(5,j) + 1 ape => mychild%pes(rempe) remind(rempe) = remind(rempe)+1 ind = remind(rempe) ape%locind(ind)%i = index_list(1,j) ape%locind(ind)%j = index_list(2,j) ENDDO ! !-- Prepare number of elements for children PEs CALL pmc_alloc_mem( rldef, mychild%inter_npes*2 ) ! !-- Number of child PEs * size of INTEGER (i just arbitrary INTEGER) winsize = mychild%inter_npes*c_sizeof(i)*2 CALL MPI_WIN_CREATE( rldef, winsize, iwp, MPI_INFO_NULL, & mychild%intra_comm, indwin, ierr ) ! !-- Open window to set data CALL MPI_WIN_FENCE( 0, indwin, ierr ) rldef(1) = 0 ! index on remote PE 0 rldef(2) = remind(1) ! number of elements on remote PE 0 ! !-- Reserve buffer for index array DO i = 2, mychild%inter_npes i2 = (i-1) * 2 + 1 rldef(i2) = rldef(i2-2) + rldef(i2-1) * 2 ! index on remote PE rldef(i2+1) = remind(i) ! number of elements on remote PE ENDDO ! !-- Close window to allow child to access data CALL MPI_WIN_FENCE( 0, indwin, ierr ) ! !-- Child has retrieved data CALL MPI_WIN_FENCE( 0, indwin, ierr ) i2 = 2 * mychild%inter_npes - 1 winsize = ( rldef(i2) + rldef(i2+1) ) * 2 ! !-- Make sure, MPI_ALLOC_MEM works winsize = MAX( winsize, 1 ) CALL pmc_alloc_mem( remindw, INT( winsize ) ) CALL MPI_BARRIER( m_model_comm, ierr ) CALL MPI_WIN_CREATE( remindw, winsize*c_sizeof(i), iwp, MPI_INFO_NULL, & mychild%intra_comm, indwin2, ierr ) ! !-- Open window to set data CALL MPI_WIN_FENCE( 0, indwin2, ierr ) ! !-- Create the 2D index list DO j = 1, nrp rempe = index_list(5,j) + 1 ! PE number on remote PE ape => mychild%pes(rempe) i2 = rempe * 2 - 1 ind = rldef(i2) + 1 remindw(ind) = index_list(3,j) remindw(ind+1) = index_list(4,j) rldef(i2) = rldef(i2)+2 ENDDO ! !-- All data are set CALL MPI_WIN_FENCE( 0, indwin2, ierr ) ! !-- Don't know why, but this barrier is necessary before windows can be freed !-- TODO: find out why this is required CALL MPI_BARRIER( mychild%intra_comm, ierr ) CALL MPI_WIN_FREE( indwin, ierr ) CALL MPI_WIN_FREE( indwin2, ierr ) ! !-- TODO: check if the following idea needs to be done !-- Sollte funktionieren, Problem mit MPI implementation !-- https://www.lrz.de/services/software/parallel/mpi/onesided !-- CALL MPI_Free_mem (remindw, ierr) END SUBROUTINE set_pe_index_list #endif END MODULE pmc_parent