MOAB: Mesh Oriented datABase  (version 5.4.1)
imoab_coupler.cpp File Reference
#include "moab/Core.hpp"
#include "moab_mpi.h"
#include "moab/ParallelComm.hpp"
#include "MBParallelConventions.h"
#include "moab/iMOAB.h"
#include "TestUtil.hpp"
#include "moab/CpuTimer.hpp"
#include "moab/ProgOptions.hpp"
#include <iostream>
#include <sstream>
#include "imoab_coupler_utils.hpp"
+ Include dependency graph for imoab_coupler.cpp:

Go to the source code of this file.

Defines

#define ENABLE_ATMOCN_COUPLING
#define ENABLE_ATMLND_COUPLING

Functions

int main (int argc, char *argv[])

Define Documentation

Definition at line 42 of file imoab_coupler.cpp.

Definition at line 41 of file imoab_coupler.cpp.


Function Documentation

int main ( int  argc,
char *  argv[] 
)

we will use the element global id, which should uniquely identify the element

Definition at line 48 of file imoab_coupler.cpp.

References ProgOptions::addOpt(), atmFilename, CHECKIERR, cmpatm, context, create_group_and_comm(), create_joint_comm_group(), DENSE_DOUBLE, DENSE_INTEGER, endG1, endG2, fileWriteOptions(), groupTasks, ierr, iMOAB_AppID, iMOAB_DefineTagStorage(), iMOAB_DeregisterApplication(), iMOAB_Finalize(), iMOAB_GetDoubleTagStorage(), iMOAB_GetIntTagStorage(), iMOAB_GetMeshInfo(), iMOAB_Initialize(), iMOAB_RegisterApplication(), iMOAB_SetDoubleTagStorage(), iMOAB_WriteLocalMesh(), iMOAB_WriteMesh(), jgroup, MPI_COMM_WORLD, nghlay, numProcesses, ProgOptions::parseCommandLine(), POP_TIMER, PUSH_TIMER, rankInAtmComm, rankInGlobalComm, readopts(), setup_component_coupler_meshes(), startG1, and startG2.

{
    int ierr;
    int rankInGlobalComm, numProcesses;
    MPI_Group jgroup;
    std::string readopts( "PARALLEL=READ_PART;PARTITION=PARALLEL_PARTITION;PARALLEL_RESOLVE_SHARED_ENTS" );
    std::string readoptsLnd( "PARALLEL=READ_PART;PARTITION=PARALLEL_PARTITION" );

    // Timer data
    moab::CpuTimer timer;
    double timer_ops;
    std::string opName;

    int repartitioner_scheme = 0;
#ifdef MOAB_HAVE_ZOLTAN
    repartitioner_scheme = 2;  // use the graph partitioner in that caseS
#endif

    MPI_Init( &argc, &argv );
    MPI_Comm_rank( MPI_COMM_WORLD, &rankInGlobalComm );
    MPI_Comm_size( MPI_COMM_WORLD, &numProcesses );

    MPI_Comm_group( MPI_COMM_WORLD, &jgroup );  // all processes in jgroup

    std::string atmFilename = TestDir + "unittest/wholeATM_T.h5m";
    // on a regular case,  5 ATM, 6 CPLATM (ATMX), 17 OCN     , 18 CPLOCN (OCNX)  ;
    // intx atm/ocn is not in e3sm yet, give a number
    //   6 * 100+ 18 = 618 : atmocnid
    // 9 LND, 10 CPLLND
    //   6 * 100 + 10 = 610  atmlndid:
    // cmpatm is for atm on atm pes
    // cmpocn is for ocean, on ocean pe
    // cplatm is for atm on coupler pes
    // cplocn is for ocean on coupelr pes
    // atmocnid is for intx atm / ocn on coupler pes
    //
    int rankInAtmComm = -1;
    int cmpatm        = 5,
        cplatm        = 6;  // component ids are unique over all pes, and established in advance;
#ifdef ENABLE_ATMOCN_COUPLING
    std::string ocnFilename = TestDir + "unittest/recMeshOcn.h5m";
    std::string baseline    = TestDir + "unittest/baseline1.txt";
    int rankInOcnComm       = -1;
    int cmpocn = 17, cplocn = 18,
        atmocnid = 618;  // component ids are unique over all pes, and established in advance;
#endif

#ifdef ENABLE_ATMLND_COUPLING
    std::string lndFilename = TestDir + "unittest/wholeLnd.h5m";
    int cpllnd = 10, cmplnd = 9,
        atmlndid = 610;  // component ids are unique over all pes, and established in advance;
#endif

    int rankInCouComm = -1;

    int nghlay = 0;  // number of ghost layers for loading the file
    std::vector< int > groupTasks;
    int startG1 = 0, startG2 = 0, endG1 = numProcesses - 1, endG2 = numProcesses - 1, startG3 = startG1,
        endG3   = endG1;                   // Support launch of imoab_coupler test on any combo of 2*x processes
    int startG4 = startG1, endG4 = endG1;  // these are for coupler layout
    int context_id = -1;                   // used now for freeing buffers

    // default: load atm on 2 proc, ocean on 2, land on 2; migrate to 2 procs, then compute intx
    // later, we need to compute weight matrix with tempestremap

    ProgOptions opts;
    opts.addOpt< std::string >( "atmosphere,t", "atm mesh filename (source)", &atmFilename );
#ifdef ENABLE_ATMOCN_COUPLING
    opts.addOpt< std::string >( "ocean,m", "ocean mesh filename (target)", &ocnFilename );
#endif
#ifdef ENABLE_ATMLND_COUPLING
    opts.addOpt< std::string >( "land,l", "land mesh filename (target)", &lndFilename );
#endif
    opts.addOpt< int >( "startAtm,a", "start task for atmosphere layout", &startG1 );
    opts.addOpt< int >( "endAtm,b", "end task for atmosphere layout", &endG1 );
#ifdef ENABLE_ATMOCN_COUPLING
    opts.addOpt< int >( "startOcn,c", "start task for ocean layout", &startG2 );
    opts.addOpt< int >( "endOcn,d", "end task for ocean layout", &endG2 );
#endif
#ifdef ENABLE_ATMLND_COUPLING
    opts.addOpt< int >( "startLnd,e", "start task for land layout", &startG3 );
    opts.addOpt< int >( "endLnd,f", "end task for land layout", &endG3 );
#endif

    opts.addOpt< int >( "startCoupler,g", "start task for coupler layout", &startG4 );
    opts.addOpt< int >( "endCoupler,j", "end task for coupler layout", &endG4 );

    opts.addOpt< int >( "partitioning,p", "partitioning option for migration", &repartitioner_scheme );

    int n = 1;  // number of send/receive / project / send back cycles
    opts.addOpt< int >( "iterations,n", "number of iterations for coupler", &n );

    bool no_regression_test = false;
    opts.addOpt< void >( "no_regression,r", "do not do regression test against baseline 1", &no_regression_test );
    opts.parseCommandLine( argc, argv );

    char fileWriteOptions[] = "PARALLEL=WRITE_PART";

    if( !rankInGlobalComm )
    {
        std::cout << " atm file: " << atmFilename << "\n   on tasks : " << startG1 << ":" << endG1 <<
#ifdef ENABLE_ATMOCN_COUPLING
            "\n ocn file: " << ocnFilename << "\n     on tasks : " << startG2 << ":" << endG2 <<
#endif
#ifdef ENABLE_ATMLND_COUPLING
            "\n lnd file: " << lndFilename << "\n     on tasks : " << startG3 << ":" << endG3 <<
#endif
            "\n  partitioning (0 trivial, 1 graph, 2 geometry) " << repartitioner_scheme << "\n  ";
    }

    // load files on 3 different communicators, groups
    // first groups has task 0, second group tasks 0 and 1
    // coupler will be on joint tasks, will be on a third group (0 and 1, again)
    // first groups has task 0, second group tasks 0 and 1
    // coupler will be on joint tasks, will be on a third group (0 and 1, again)
    MPI_Group atmPEGroup;
    MPI_Comm atmComm;
    ierr = create_group_and_comm( startG1, endG1, jgroup, &atmPEGroup, &atmComm );
    CHECKIERR( ierr, "Cannot create atm MPI group and communicator " )

#ifdef ENABLE_ATMOCN_COUPLING
    MPI_Group ocnPEGroup;
    MPI_Comm ocnComm;
    ierr = create_group_and_comm( startG2, endG2, jgroup, &ocnPEGroup, &ocnComm );
    CHECKIERR( ierr, "Cannot create ocn MPI group and communicator " )
#endif

#ifdef ENABLE_ATMLND_COUPLING
    MPI_Group lndPEGroup;
    MPI_Comm lndComm;
    ierr = create_group_and_comm( startG3, endG3, jgroup, &lndPEGroup, &lndComm );
    CHECKIERR( ierr, "Cannot create lnd MPI group and communicator " )
#endif

    // we will always have a coupler
    MPI_Group couPEGroup;
    MPI_Comm couComm;
    ierr = create_group_and_comm( startG4, endG4, jgroup, &couPEGroup, &couComm );
    CHECKIERR( ierr, "Cannot create cpl MPI group and communicator " )

    // atm_coupler
    MPI_Group joinAtmCouGroup;
    MPI_Comm atmCouComm;
    ierr = create_joint_comm_group( atmPEGroup, couPEGroup, &joinAtmCouGroup, &atmCouComm );
    CHECKIERR( ierr, "Cannot create joint atm cou communicator" )

#ifdef ENABLE_ATMOCN_COUPLING
    // ocn_coupler
    MPI_Group joinOcnCouGroup;
    MPI_Comm ocnCouComm;
    ierr = create_joint_comm_group( ocnPEGroup, couPEGroup, &joinOcnCouGroup, &ocnCouComm );
    CHECKIERR( ierr, "Cannot create joint ocn cou communicator" )
#endif

#ifdef ENABLE_ATMLND_COUPLING
    // lnd_coupler
    MPI_Group joinLndCouGroup;
    MPI_Comm lndCouComm;
    ierr = create_joint_comm_group( lndPEGroup, couPEGroup, &joinLndCouGroup, &lndCouComm );
    CHECKIERR( ierr, "Cannot create joint ocn cou communicator" )
#endif

    ierr = iMOAB_Initialize( argc, argv );  // not really needed anything from argc, argv, yet; maybe we should
    CHECKIERR( ierr, "Cannot initialize iMOAB" )

    int cmpAtmAppID       = -1;
    iMOAB_AppID cmpAtmPID = &cmpAtmAppID;  // atm
    int cplAtmAppID       = -1;            // -1 means it is not initialized
    iMOAB_AppID cplAtmPID = &cplAtmAppID;  // atm on coupler PEs
#ifdef ENABLE_ATMOCN_COUPLING
    int cmpOcnAppID       = -1;
    iMOAB_AppID cmpOcnPID = &cmpOcnAppID;        // ocn
    int cplOcnAppID = -1, cplAtmOcnAppID = -1;   // -1 means it is not initialized
    iMOAB_AppID cplOcnPID    = &cplOcnAppID;     // ocn on coupler PEs
    iMOAB_AppID cplAtmOcnPID = &cplAtmOcnAppID;  // intx atm -ocn on coupler PEs
#endif

#ifdef ENABLE_ATMLND_COUPLING
    int cmpLndAppID       = -1;
    iMOAB_AppID cmpLndPID = &cmpLndAppID;        // lnd
    int cplLndAppID = -1, cplAtmLndAppID = -1;   // -1 means it is not initialized
    iMOAB_AppID cplLndPID    = &cplLndAppID;     // land on coupler PEs
    iMOAB_AppID cplAtmLndPID = &cplAtmLndAppID;  // intx atm - lnd on coupler PEs
#endif

    if( couComm != MPI_COMM_NULL )
    {
        MPI_Comm_rank( couComm, &rankInCouComm );
        // Register all the applications on the coupler PEs
        ierr = iMOAB_RegisterApplication( "ATMX", &couComm, &cplatm,
                                          cplAtmPID );  // atm on coupler pes
        CHECKIERR( ierr, "Cannot register ATM over coupler PEs" )
#ifdef ENABLE_ATMOCN_COUPLING
        ierr = iMOAB_RegisterApplication( "OCNX", &couComm, &cplocn,
                                          cplOcnPID );  // ocn on coupler pes
        CHECKIERR( ierr, "Cannot register OCN over coupler PEs" )
#endif
#ifdef ENABLE_ATMLND_COUPLING
        ierr = iMOAB_RegisterApplication( "LNDX", &couComm, &cpllnd,
                                          cplLndPID );  // lnd on coupler pes
        CHECKIERR( ierr, "Cannot register LND over coupler PEs" )
#endif
    }

    if( atmComm != MPI_COMM_NULL )
    {
        MPI_Comm_rank( atmComm, &rankInAtmComm );
        ierr = iMOAB_RegisterApplication( "ATM1", &atmComm, &cmpatm, cmpAtmPID );
        CHECKIERR( ierr, "Cannot register ATM App" )
    }

#ifdef ENABLE_ATMOCN_COUPLING
    if( ocnComm != MPI_COMM_NULL )
    {
        MPI_Comm_rank( ocnComm, &rankInOcnComm );
        ierr = iMOAB_RegisterApplication( "OCN1", &ocnComm, &cmpocn, cmpOcnPID );
        CHECKIERR( ierr, "Cannot register OCN App" )
    }
#endif

    // atm
    ierr =
        setup_component_coupler_meshes( cmpAtmPID, cmpatm, cplAtmPID, cplatm, &atmComm, &atmPEGroup, &couComm,
                                        &couPEGroup, &atmCouComm, atmFilename, readopts, nghlay, repartitioner_scheme );
    CHECKIERR( ierr, "Cannot load and migrate atm mesh" )
#ifdef VERBOSE
    if( couComm != MPI_COMM_NULL && 1 == n )
    {  // write only for n==1 case
        char outputFileTgt3[] = "recvAtm.h5m";
        ierr                  = iMOAB_WriteMesh( cplAtmPID, outputFileTgt3, fileWriteOptions, strlen( outputFileTgt3 ),
                                                 strlen( fileWriteOptions ) );
        CHECKIERR( ierr, "cannot write atm mesh after receiving" )
    }
#endif
#ifdef GRAPH_INFO
    if( atmComm != MPI_COMM_NULL )
    {

        int is_sender = 1;
        int context   = -1;
        iMOAB_DumpCommGraph( cmpAtmPID, &context, &is_sender, "AtmMigS" );
    }
    if( couComm != MPI_COMM_NULL )
    {
        int is_sender = 0;
        int context   = -1;
        iMOAB_DumpCommGraph( cplAtmPID, &context, &is_sender, "AtmMigR" );
    }
#endif
    MPI_Barrier( MPI_COMM_WORLD );

#ifdef ENABLE_ATMOCN_COUPLING
    // ocean
    ierr =
        setup_component_coupler_meshes( cmpOcnPID, cmpocn, cplOcnPID, cplocn, &ocnComm, &ocnPEGroup, &couComm,
                                        &couPEGroup, &ocnCouComm, ocnFilename, readopts, nghlay, repartitioner_scheme );
    CHECKIERR( ierr, "Cannot load and migrate ocn mesh" )

    MPI_Barrier( MPI_COMM_WORLD );

#ifdef VERBOSE
    if( couComm != MPI_COMM_NULL && 1 == n )
    {  // write only for n==1 case
        char outputFileTgt3[] = "recvOcn.h5m";
        ierr                  = iMOAB_WriteMesh( cplOcnPID, outputFileTgt3, fileWriteOptions );
        CHECKIERR( ierr, "cannot write ocn mesh after receiving" )
    }
#endif
#endif  // #ifdef ENABLE_ATMOCN_COUPLING

#ifdef ENABLE_ATMLND_COUPLING
    // land
    if( lndComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_RegisterApplication( "LND1", &lndComm, &cmplnd, cmpLndPID );
        CHECKIERR( ierr, "Cannot register LND App " )
    }
    ierr = setup_component_coupler_meshes( cmpLndPID, cmplnd, cplLndPID, cpllnd, &lndComm, &lndPEGroup, &couComm,
                                           &couPEGroup, &lndCouComm, lndFilename, readoptsLnd, nghlay,
                                           repartitioner_scheme );

    if( couComm != MPI_COMM_NULL && 1 == n )
    {  // write only for n==1 case
        char outputFileLnd[] = "recvLnd.h5m";
        ierr                 = iMOAB_WriteMesh( cplLndPID, outputFileLnd, fileWriteOptions );
        CHECKIERR( ierr, "cannot write lnd mesh after receiving" )
    }

#endif  // #ifdef ENABLE_ATMLND_COUPLING

    MPI_Barrier( MPI_COMM_WORLD );

#ifdef ENABLE_ATMOCN_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        // now compute intersection between OCNx and ATMx on coupler PEs
        ierr = iMOAB_RegisterApplication( "ATMOCN", &couComm, &atmocnid, cplAtmOcnPID );
        CHECKIERR( ierr, "Cannot register ocn_atm intx over coupler pes " )
    }
#endif
#ifdef ENABLE_ATMLND_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        // now compute intersection between LNDx and ATMx on coupler PEs
        ierr = iMOAB_RegisterApplication( "ATMLND", &couComm, &atmlndid, cplAtmLndPID );
        CHECKIERR( ierr, "Cannot register ocn_atm intx over coupler pes " )
    }
#endif

    int disc_orders[3]                       = { 4, 1, 1 };
    const std::string weights_identifiers[2] = { "scalar", "scalar-pc" };
    const std::string disc_methods[3]        = { "cgll", "fv", "pcloud" };
    const std::string dof_tag_names[3]       = { "GLOBAL_DOFS", "GLOBAL_ID", "GLOBAL_ID" };
#ifdef ENABLE_ATMOCN_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        PUSH_TIMER( "Compute ATM-OCN mesh intersection" )
        ierr = iMOAB_ComputeMeshIntersectionOnSphere(
            cplAtmPID, cplOcnPID,
            cplAtmOcnPID );  // coverage mesh was computed here, for cplAtmPID, atm on coupler pes
        // basically, atm was redistributed according to target (ocean) partition, to "cover" the
        // ocean partitions check if intx valid, write some h5m intx file
        CHECKIERR( ierr, "cannot compute intersection" )
        POP_TIMER( couComm, rankInCouComm )
#ifdef VERBOSE
        char prefix[] = "intx_atmocn";
        ierr          = iMOAB_WriteLocalMesh( cplAtmOcnPID, prefix, strlen( prefix ) );
        CHECKIERR( ierr, "failed to write local intx mesh" );
#endif
    }

    if( atmCouComm != MPI_COMM_NULL )
    {

        // the new graph will be for sending data from atm comp to coverage mesh;
        // it involves initial atm app; cmpAtmPID; also migrate atm mesh on coupler pes, cplAtmPID
        // results are in cplAtmOcnPID, intx mesh; remapper also has some info about coverage mesh
        // after this, the sending of tags from atm pes to coupler pes will use the new par comm
        // graph, that has more precise info about what to send for ocean cover ; every time, we
        // will
        //  use the element global id, which should uniquely identify the element
        PUSH_TIMER( "Compute OCN coverage graph for ATM mesh" )
        ierr = iMOAB_CoverageGraph( &atmCouComm, cmpAtmPID, cplAtmPID, cplAtmOcnPID, &cmpatm, &cplatm,
                                    &cplocn );  // it happens over joint communicator
        CHECKIERR( ierr, "cannot recompute direct coverage graph for ocean" )
        POP_TIMER( atmCouComm, rankInAtmComm )  // hijack this rank
    }
#endif

#ifdef ENABLE_ATMLND_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        PUSH_TIMER( "Compute ATM-LND mesh intersection" )
        ierr = iMOAB_ComputePointDoFIntersection( cplAtmPID, cplLndPID, cplAtmLndPID );
        CHECKIERR( ierr, "failed to compute point-cloud mapping" );
        POP_TIMER( couComm, rankInCouComm )
    }
    if( atmCouComm != MPI_COMM_NULL )
    {
        // the new graph will be for sending data from atm comp to coverage mesh for land mesh;
        // it involves initial atm app; cmpAtmPID; also migrate atm mesh on coupler pes, cplAtmPID
        // results are in cplAtmLndPID, intx mesh; remapper also has some info about coverage mesh
        // after this, the sending of tags from atm pes to coupler pes will use the new par comm
        // graph, that has more precise info about what to send (specifically for land cover); every
        // time,
        /// we will use the element global id, which should uniquely identify the element
        PUSH_TIMER( "Compute LND coverage graph for ATM mesh" )
        ierr = iMOAB_CoverageGraph( &atmCouComm, cmpAtmPID, cplAtmPID, cplAtmLndPID, &cmpatm, &cplatm,
                                    &cpllnd );  // it happens over joint communicator
        CHECKIERR( ierr, "cannot recompute direct coverage graph for land" )
        POP_TIMER( atmCouComm, rankInAtmComm )  // hijack this rank
    }
#endif

    MPI_Barrier( MPI_COMM_WORLD );

    int fMonotoneTypeID = 0, fVolumetric = 0, fValidate = 1, fNoConserve = 0, fNoBubble = 1, fInverseDistanceMap = 0;

#ifdef ENABLE_ATMOCN_COUPLING
#ifdef VERBOSE
    if( couComm != MPI_COMM_NULL && 1 == n )
    {                                    // write only for n==1 case
        char serialWriteOptions[] = "";  // for writing in serial
        std::stringstream outf;
        outf << "intxAtmOcn_" << rankInCouComm << ".h5m";
        std::string intxfile = outf.str();  // write in serial the intx file, for debugging
        ierr                 = iMOAB_WriteMesh( cplAtmOcnPID, intxfile.c_str(), serialWriteOptions );
        CHECKIERR( ierr, "cannot write intx file result" )
    }
#endif

    if( couComm != MPI_COMM_NULL )
    {
        PUSH_TIMER( "Compute the projection weights with TempestRemap" )
        ierr =
            iMOAB_ComputeScalarProjectionWeights( cplAtmOcnPID, weights_identifiers[0].c_str(), disc_methods[0].c_str(),
                                                  &disc_orders[0], disc_methods[1].c_str(), &disc_orders[1], &fNoBubble,
                                                  &fMonotoneTypeID, &fVolumetric, &fInverseDistanceMap, &fNoConserve,
                                                  &fValidate, dof_tag_names[0].c_str(), dof_tag_names[1].c_str() );
        CHECKIERR( ierr, "cannot compute scalar projection weights" )
        POP_TIMER( couComm, rankInCouComm )

        // Let us now write the map file to disk and then read it back to test the I/O API in iMOAB
#ifdef MOAB_HAVE_NETCDF
        {
            const std::string atmocn_map_file_name = "atm_ocn_map.nc";
            ierr = iMOAB_WriteMappingWeightsToFile( cplAtmOcnPID, weights_identifiers[0].c_str(),
                                                    atmocn_map_file_name.c_str() );
            CHECKIERR( ierr, "failed to write map file to disk" );

            const std::string intx_from_file_identifier = "map-from-file";
            int dummyCpl                                = -1;
            int dummy_rowcol                            = -1;
            int dummyType                               = 0;
            ierr = iMOAB_LoadMappingWeightsFromFile( cplAtmOcnPID, &dummyCpl, &dummy_rowcol, &dummyType,
                                                     intx_from_file_identifier.c_str(), atmocn_map_file_name.c_str() );
            CHECKIERR( ierr, "failed to load map file from disk" );
        }
#endif
    }

#endif

    MPI_Barrier( MPI_COMM_WORLD );

#ifdef ENABLE_ATMLND_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        fValidate = 0;
        /* Compute the weights to preoject the solution from ATM component to LND compoenent */
        PUSH_TIMER( "Compute ATM-LND remapping weights" )
        ierr =
            iMOAB_ComputeScalarProjectionWeights( cplAtmLndPID, weights_identifiers[1].c_str(), disc_methods[0].c_str(),
                                                  &disc_orders[0], disc_methods[2].c_str(), &disc_orders[2], &fNoBubble,
                                                  &fMonotoneTypeID, &fVolumetric, &fInverseDistanceMap, &fNoConserve,
                                                  &fValidate, dof_tag_names[0].c_str(), dof_tag_names[2].c_str() );
        CHECKIERR( ierr, "failed to compute remapping projection weights for ATM-LND scalar "
                         "non-conservative field" );
        POP_TIMER( couComm, rankInCouComm )

        // Let us now write the map file to disk and then read it back to test the I/O API in iMOAB
        // VSM: TODO: This does not work since the LND model is a point cloud and we do not initilize
        // data correctly in TempestOnlineMap::WriteParallelWeightsToFile routine.
        // {
        //     const char* atmlnd_map_file_name = "atm_lnd_map.nc";
        //     ierr = iMOAB_WriteMappingWeightsToFile( cplAtmLndPID, weights_identifiers[1], atmlnd_map_file_name );
        //     CHECKIERR( ierr, "failed to write map file to disk" );

        //     const char* intx_from_file_identifier = "map-from-file";
        //     ierr = iMOAB_LoadMappingWeightsFromFile( cplAtmLndPID, intx_from_file_identifier, atmlnd_map_file_name,
        //                                              NULL, NULL, NULL,
        //                                             );
        //     CHECKIERR( ierr, "failed to load map file from disk" );
        // }
    }
#endif

    int tagIndex[2];
    int tagTypes[2]  = { DENSE_DOUBLE, DENSE_DOUBLE };
    int atmCompNDoFs = disc_orders[0] * disc_orders[0], ocnCompNDoFs = 1 /*FV*/;

    const char* bottomFields          = "a2oTbot:a2oUbot:a2oVbot";
    const char* bottomProjectedFields = "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj";

    if( couComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DefineTagStorage( cplAtmPID, bottomFields, &tagTypes[0], &atmCompNDoFs, &tagIndex[0] );
        CHECKIERR( ierr, "failed to define the field tag a2oTbot" );
#ifdef ENABLE_ATMOCN_COUPLING
        ierr = iMOAB_DefineTagStorage( cplOcnPID, bottomProjectedFields, &tagTypes[1], &ocnCompNDoFs, &tagIndex[1] );
        CHECKIERR( ierr, "failed to define the field tag a2oTbot_proj" );
#endif
    }

    // need to make sure that the coverage mesh (created during intx method) received the tag that
    // need to be projected to target so far, the coverage mesh has only the ids and global dofs;
    // need to change the migrate method to accommodate any GLL tag
    // now send a tag from original atmosphere (cmpAtmPID) towards migrated coverage mesh
    // (cplAtmPID), using the new coverage graph communicator

    // make the tag 0, to check we are actually sending needed data
    {
        if( cplAtmAppID >= 0 )
        {
            int nverts[3], nelem[3], nblocks[3], nsbc[3], ndbc[3];
            /*
             * Each process in the communicator will have access to a local mesh instance, which
             * will contain the original cells in the local partition and ghost entities. Number of
             * vertices, primary cells, visible blocks, number of sidesets and nodesets boundary
             * conditions will be returned in numProcesses 3 arrays, for local, ghost and total
             * numbers.
             */
            ierr = iMOAB_GetMeshInfo( cplAtmPID, nverts, nelem, nblocks, nsbc, ndbc );
            CHECKIERR( ierr, "failed to get num primary elems" );
            int numAllElem = nelem[2];
            std::vector< double > vals;
            int storLeng = atmCompNDoFs * numAllElem * 3;  // 3 tags
            int eetype   = 1;

            vals.resize( storLeng );
            for( int k = 0; k < storLeng; k++ )
                vals[k] = 0.;

            ierr = iMOAB_SetDoubleTagStorage( cplAtmPID, bottomFields, &storLeng, &eetype, &vals[0] );
            CHECKIERR( ierr, "cannot make tag nul" )
            // set the tag to 0
        }
    }

    const char* concat_fieldname  = "a2oTbot:a2oUbot:a2oVbot";
    const char* concat_fieldnameT = "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj";

    // start a virtual loop for number of iterations
    for( int iters = 0; iters < n; iters++ )
    {
#ifdef ENABLE_ATMOCN_COUPLING
        PUSH_TIMER( "Send/receive data from atm component to coupler in ocn context" )
        if( atmComm != MPI_COMM_NULL )
        {
            // as always, use nonblocking sends
            // this is for projection to ocean:
            ierr = iMOAB_SendElementTag( cmpAtmPID, "a2oTbot:a2oUbot:a2oVbot", &atmCouComm, &cplocn );
            CHECKIERR( ierr, "cannot send tag values" )
#ifdef GRAPH_INFO
            int is_sender = 1;
            int context   = cplocn;
            iMOAB_DumpCommGraph( cmpAtmPID, &context, &is_sender, "AtmCovOcnS" );
#endif
        }
        if( couComm != MPI_COMM_NULL )
        {
            // receive on atm on coupler pes, that was redistributed according to coverage
            ierr = iMOAB_ReceiveElementTag( cplAtmPID, "a2oTbot:a2oUbot:a2oVbot", &atmCouComm, &cplocn );
            CHECKIERR( ierr, "cannot receive tag values" )
#ifdef GRAPH_INFO
            int is_sender = 0;
            int context   = cplocn;  // the same context, cplocn
            iMOAB_DumpCommGraph( cmpAtmPID, &context, &is_sender, "AtmCovOcnR" );
#endif
        }
        POP_TIMER( MPI_COMM_WORLD, rankInGlobalComm )

        // we can now free the sender buffers
        if( atmComm != MPI_COMM_NULL )
        {
            ierr = iMOAB_FreeSenderBuffers( cmpAtmPID, &cplocn );  // context is for ocean
            CHECKIERR( ierr, "cannot free buffers used to resend atm tag towards the coverage mesh" )
        }
#ifdef VERBOSE
        if( couComm != MPI_COMM_NULL && 1 == n )
        {
            // write only for n==1 case
            char outputFileRecvd[] = "recvAtmCoupOcn.h5m";
            ierr                   = iMOAB_WriteMesh( cplAtmPID, outputFileRecvd, fileWriteOptions );
            CHECKIERR( ierr, "could not write recvAtmCoupOcn.h5m to disk" )
        }
#endif

        if( couComm != MPI_COMM_NULL )
        {
            /* We have the remapping weights now. Let us apply the weights onto the tag we defined
               on the source mesh and get the projection on the target mesh */
            PUSH_TIMER( "Apply Scalar projection weights" )
            ierr = iMOAB_ApplyScalarProjectionWeights( cplAtmOcnPID, weights_identifiers[0].c_str(), concat_fieldname,
                                                       concat_fieldnameT );
            CHECKIERR( ierr, "failed to compute projection weight application" );
            POP_TIMER( couComm, rankInCouComm )
            if( 1 == n )  // write only for n==1 case
            {
                char outputFileTgt[] = "fOcnOnCpl.h5m";
                ierr                 = iMOAB_WriteMesh( cplOcnPID, outputFileTgt, fileWriteOptions );
                CHECKIERR( ierr, "could not write fOcnOnCpl.h5m to disk" )
            }
        }

        // send the projected tag back to ocean pes, with send/receive tag
        if( ocnComm != MPI_COMM_NULL )
        {
            int tagIndexIn2;
            ierr =
                iMOAB_DefineTagStorage( cmpOcnPID, bottomProjectedFields, &tagTypes[1], &ocnCompNDoFs, &tagIndexIn2 );
            CHECKIERR( ierr, "failed to define the field tag for receiving back the tags "
                             "a2oTbot_proj, a2oUbot_proj, a2oVbot_proj on ocn pes" );
        }
        // send the tag to ocean pes, from ocean mesh on coupler pes
        //   from couComm, using common joint comm ocn_coupler
        // as always, use nonblocking sends
        // original graph (context is -1_
        if( couComm != MPI_COMM_NULL )
        {
            // need to use ocean comp id for context
            context_id = cmpocn;  // id for ocean on comp
            ierr =
                iMOAB_SendElementTag( cplOcnPID, "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj", &ocnCouComm, &context_id );
            CHECKIERR( ierr, "cannot send tag values back to ocean pes" )
        }

        // receive on component 2, ocean
        if( ocnComm != MPI_COMM_NULL )
        {
            context_id = cplocn;  // id for ocean on coupler
            ierr       = iMOAB_ReceiveElementTag( cmpOcnPID, "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj", &ocnCouComm,
                                                  &context_id );
            CHECKIERR( ierr, "cannot receive tag values from ocean mesh on coupler pes" )
        }

        MPI_Barrier( MPI_COMM_WORLD );

        if( couComm != MPI_COMM_NULL )
        {
            context_id = cmpocn;
            ierr       = iMOAB_FreeSenderBuffers( cplOcnPID, &context_id );
            CHECKIERR( ierr, "cannot free send/receive buffers for OCN context" )
        }
        if( ocnComm != MPI_COMM_NULL && 1 == n )  // write only for n==1 case
        {
            char outputFileOcn[] = "OcnWithProj.h5m";
            ierr                 = iMOAB_WriteMesh( cmpOcnPID, outputFileOcn, fileWriteOptions );
            CHECKIERR( ierr, "could not write OcnWithProj.h5m to disk" )
            // test results only for n == 1, for bottomTempProjectedField
            if( !no_regression_test )
            {
                // the same as remap test
                // get temp field on ocean, from conservative, the global ids, and dump to the baseline file
                // first get GlobalIds from ocn, and fields:
                int nverts[3], nelem[3];
                ierr = iMOAB_GetMeshInfo( cmpOcnPID, nverts, nelem, 0, 0, 0 );
                CHECKIERR( ierr, "failed to get ocn mesh info" );
                std::vector< int > gidElems;
                gidElems.resize( nelem[2] );
                std::vector< double > tempElems;
                tempElems.resize( nelem[2] );
                // get global id storage
                const std::string GidStr = "GLOBAL_ID";  // hard coded too
                int tag_type = DENSE_INTEGER, ncomp = 1, tagInd = 0;
                ierr = iMOAB_DefineTagStorage( cmpOcnPID, GidStr.c_str(), &tag_type, &ncomp, &tagInd );
                CHECKIERR( ierr, "failed to define global id tag" );

                int ent_type = 1;
                ierr         = iMOAB_GetIntTagStorage( cmpOcnPID, GidStr.c_str(), &nelem[2], &ent_type, &gidElems[0] );
                CHECKIERR( ierr, "failed to get global ids" );
                ierr = iMOAB_GetDoubleTagStorage( cmpOcnPID, "a2oTbot_proj", &nelem[2], &ent_type, &tempElems[0] );
                CHECKIERR( ierr, "failed to get temperature field" );
                int err_code = 1;
                check_baseline_file( baseline, gidElems, tempElems, 1.e-9, err_code );
                if( 0 == err_code )
                    std::cout << " passed baseline test atm2ocn on ocean task " << rankInOcnComm << "\n";
            }
        }
#endif

#ifdef ENABLE_ATMLND_COUPLING
        // start land proj:
        PUSH_TIMER( "Send/receive data from component atm to coupler, in land context" )
        if( atmComm != MPI_COMM_NULL )
        {
            // as always, use nonblocking sends
            // this is for projection to land:
            ierr = iMOAB_SendElementTag( cmpAtmPID, "a2oTbot:a2oUbot:a2oVbot", &atmCouComm, &cpllnd );
            CHECKIERR( ierr, "cannot send tag values" )
        }
        if( couComm != MPI_COMM_NULL )
        {
            // receive on atm on coupler pes, that was redistributed according to coverage, for land
            // context
            ierr = iMOAB_ReceiveElementTag( cplAtmPID, "a2oTbot:a2oUbot:a2oVbot", &atmCouComm, &cpllnd );
            CHECKIERR( ierr, "cannot receive tag values" )
        }
        POP_TIMER( MPI_COMM_WORLD, rankInGlobalComm )

        // we can now free the sender buffers
        if( atmComm != MPI_COMM_NULL )
        {
            ierr = iMOAB_FreeSenderBuffers( cmpAtmPID, &cpllnd );
            CHECKIERR( ierr, "cannot free buffers used to resend atm tag towards the coverage mesh "
                             "for land context" )
        }
#ifdef VERBOSE
        if( couComm != MPI_COMM_NULL && 1 == n )
        {  // write only for n==1 case
            char outputFileRecvd[] = "recvAtmCoupLnd.h5m";
            ierr                   = iMOAB_WriteMesh( cplAtmPID, outputFileRecvd, fileWriteOptions );
            CHECKIERR( ierr, "could not write recvAtmCoupLnd.h5m to disk" )
        }
#endif

        /* We have the remapping weights now. Let us apply the weights onto the tag we defined
           on the source mesh and get the projection on the target mesh */
        if( couComm != MPI_COMM_NULL )
        {
            PUSH_TIMER( "Apply Scalar projection weights for land" )
            ierr = iMOAB_ApplyScalarProjectionWeights( cplAtmLndPID, weights_identifiers[1].c_str(), concat_fieldname,
                                                       concat_fieldnameT );
            CHECKIERR( ierr, "failed to compute projection weight application" );
            POP_TIMER( couComm, rankInCouComm )
        }

#ifdef VERBOSE
        if( couComm != MPI_COMM_NULL && 1 == n )
        {  // write only for n==1 case
            char outputFileTgtLnd[] = "fLndOnCpl.h5m";
            ierr                    = iMOAB_WriteMesh( cplLndPID, outputFileTgtLnd, fileWriteOptions );
            CHECKIERR( ierr, "could not write fLndOnCpl.h5m to disk" )
        }
#endif

        // end land proj
        // send the tags back to land pes, from land mesh on coupler pes
        // send from cplLndPID to cmpLndPID, using common joint comm
        // as always, use nonblocking sends
        // original graph
        // int context_id = -1;
        // the land might not have these tags yet; it should be a different name for land
        // in e3sm we do have different names
        if( lndComm != MPI_COMM_NULL )
        {
            int tagIndexIn2;
            ierr =
                iMOAB_DefineTagStorage( cmpLndPID, bottomProjectedFields, &tagTypes[1], &ocnCompNDoFs, &tagIndexIn2 );
            CHECKIERR( ierr, "failed to define the field tag for receiving back the tag "
                             "a2oTbot_proj, a2oUbot_proj, a2oVbot_proj  on lnd pes" );
        }
        if( couComm != MPI_COMM_NULL )
        {
            context_id = cmplnd;  // land comp id
            ierr =
                iMOAB_SendElementTag( cplLndPID, "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj", &lndCouComm, &context_id );
            CHECKIERR( ierr, "cannot send tag values back to land pes" )
        }
        // receive on component 3, land
        if( lndComm != MPI_COMM_NULL )
        {
            context_id = cpllnd;  // land on coupler id
            ierr       = iMOAB_ReceiveElementTag( cmpLndPID, "a2oTbot_proj:a2oUbot_proj:a2oVbot_proj", &lndCouComm,
                                                  &context_id );
            CHECKIERR( ierr, "cannot receive tag values from land mesh on coupler pes" )
        }

        MPI_Barrier( MPI_COMM_WORLD );
        if( couComm != MPI_COMM_NULL )
        {
            context_id = cmplnd;
            ierr       = iMOAB_FreeSenderBuffers( cplLndPID, &context_id );
            CHECKIERR( ierr, "cannot free buffers used to resend land tag towards the coverage mesh "
                             "for atm context" )
        }
        if( lndComm != MPI_COMM_NULL && 1 == n )  // write only for n==1 case
        {
            char outputFileLnd[] = "LndWithProj.h5m";
            ierr                 = iMOAB_WriteMesh( cmpLndPID, outputFileLnd, fileWriteOptions );
            CHECKIERR( ierr, "could not write LndWithProj.h5m to disk" )
        }
#endif  // ENABLE_ATMLND_COUPLING

    }  // end loop iterations n
#ifdef ENABLE_ATMLND_COUPLING
    if( lndComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cmpLndPID );
        CHECKIERR( ierr, "cannot deregister app LND1" )
    }
#endif  // ENABLE_ATMLND_COUPLING

#ifdef ENABLE_ATMOCN_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cplAtmOcnPID );
        CHECKIERR( ierr, "cannot deregister app intx AO" )
    }
    if( ocnComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cmpOcnPID );
        CHECKIERR( ierr, "cannot deregister app OCN1" )
    }
#endif  // ENABLE_ATMOCN_COUPLING

    if( atmComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cmpAtmPID );
        CHECKIERR( ierr, "cannot deregister app ATM1" )
    }

#ifdef ENABLE_ATMLND_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cplLndPID );
        CHECKIERR( ierr, "cannot deregister app LNDX" )
    }
#endif  // ENABLE_ATMLND_COUPLING

#ifdef ENABLE_ATMOCN_COUPLING
    if( couComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cplOcnPID );
        CHECKIERR( ierr, "cannot deregister app OCNX" )
    }
#endif  // ENABLE_ATMOCN_COUPLING

    if( couComm != MPI_COMM_NULL )
    {
        ierr = iMOAB_DeregisterApplication( cplAtmPID );
        CHECKIERR( ierr, "cannot deregister app ATMX" )
    }

    //#endif
    ierr = iMOAB_Finalize();
    CHECKIERR( ierr, "did not finalize iMOAB" )

    // free atm coupler group and comm
    if( MPI_COMM_NULL != atmCouComm ) MPI_Comm_free( &atmCouComm );
    MPI_Group_free( &joinAtmCouGroup );
    if( MPI_COMM_NULL != atmComm ) MPI_Comm_free( &atmComm );

#ifdef ENABLE_ATMOCN_COUPLING
    if( MPI_COMM_NULL != ocnComm ) MPI_Comm_free( &ocnComm );
    // free ocn - coupler group and comm
    if( MPI_COMM_NULL != ocnCouComm ) MPI_Comm_free( &ocnCouComm );
    MPI_Group_free( &joinOcnCouGroup );
#endif

#ifdef ENABLE_ATMLND_COUPLING
    if( MPI_COMM_NULL != lndComm ) MPI_Comm_free( &lndComm );
    // free land - coupler group and comm
    if( MPI_COMM_NULL != lndCouComm ) MPI_Comm_free( &lndCouComm );
    MPI_Group_free( &joinLndCouGroup );
#endif

    if( MPI_COMM_NULL != couComm ) MPI_Comm_free( &couComm );

    MPI_Group_free( &atmPEGroup );
#ifdef ENABLE_ATMOCN_COUPLING
    MPI_Group_free( &ocnPEGroup );
#endif
#ifdef ENABLE_ATMLND_COUPLING
    MPI_Group_free( &lndPEGroup );
#endif
    MPI_Group_free( &couPEGroup );
    MPI_Group_free( &jgroup );

    MPI_Finalize();

    return 0;
}
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines