mbtempest.cpp

Go to the documentation of this file.

/*
 * Usage: MOAB-Tempest tool
 *
 * Generate a Cubed-Sphere mesh: ./mbtempest -t 0 -res 25 -f cubed_sphere_mesh.exo
 * Generate a RLL mesh: ./mbtempest -t 1 -res 25 -f rll_mesh.exo
 * Generate a Icosahedral-Sphere mesh: ./mbtempest -t 2 -res 25 <-dual> -f icosa_mesh.exo
 *
 * Now you can compute the intersections between the meshes too!
 *
 * Generate the overlap mesh: ./mbtempest -t 3 -l cubed_sphere_mesh.exo -l rll_mesh.exo -f
 * overlap_mesh.exo
 *
 */

#include <iostream>
#include <iomanip>
#include <cstdlib>
#include <vector>
#include <string>
#include <sstream>
#include <cassert>

#include "moab/Core.hpp"
#include "moab/IntxMesh/IntxUtils.hpp"
#include "moab/Remapping/TempestRemapper.hpp"
#include "moab/Remapping/TempestOnlineMap.hpp"
#include "moab/ProgOptions.hpp"
#include "moab/CpuTimer.hpp"
#include "DebugOutput.hpp"

//#ifndef MOAB_HAVE_MPI
//    #error mbtempest tool requires MPI configuration
//#endif

#ifdef MOAB_HAVE_MPI
// MPI includes
#include "moab_mpi.h"
#include "moab/ParallelComm.hpp"
#include "MBParallelConventions.h"
#endif

struct ToolContext
{
    moab::Interface* mbcore;
#ifdef MOAB_HAVE_MPI
    moab::ParallelComm* pcomm;
#endif
    const int proc_id, n_procs;
    moab::DebugOutput outputFormatter;
    int blockSize;
    std::vector< std::string > inFilenames;
    std::vector< Mesh* > meshes;
    std::vector< moab::EntityHandle > meshsets;
    std::vector< int > disc_orders;
    std::vector< std::string > disc_methods;
    std::vector< std::string > doftag_names;
    std::string outFilename;
    std::string intxFilename;
    std::string baselineFile;
    moab::TempestRemapper::TempestMeshType meshType;
    bool computeDual;
    bool computeWeights;
    bool verifyWeights;
    bool enforceConvexity;
    int ensureMonotonicity;
    bool rrmGrids;
    bool kdtreeSearch;
    bool fCheck;
    bool fVolumetric;
    bool fInverseDistanceMap;
    GenerateOfflineMapAlgorithmOptions mapOptions;

#ifdef MOAB_HAVE_MPI
    ToolContext( moab::Interface* icore, moab::ParallelComm* p_pcomm )
        : mbcore( icore ), pcomm( p_pcomm ), proc_id( pcomm->rank() ), n_procs( pcomm->size() ),
          outputFormatter( std::cout, pcomm->rank(), 0 ),
#else
    ToolContext( moab::Interface* icore )
        : mbcore( icore ), proc_id( 0 ), n_procs( 1 ), outputFormatter( std::cout, 0, 0 ),
#endif
          blockSize( 5 ), outFilename( "outputFile.nc" ), intxFilename( "intxFile.h5m" ), baselineFile( "" ),
          meshType( moab::TempestRemapper::DEFAULT ), computeDual( false ), computeWeights( false ),
          verifyWeights( false ), enforceConvexity( false ), ensureMonotonicity( 0 ), rrmGrids( false ),
          kdtreeSearch( true ), fCheck( n_procs > 1 ? false : true ), fVolumetric( false ), fInverseDistanceMap( false )
    {
        inFilenames.resize( 2 );
        doftag_names.resize( 2 );
        timer = new moab::CpuTimer();

        outputFormatter.set_prefix( "[MBTempest]: " );
    }

    ~ToolContext()
    {
        // for (unsigned i=0; i < meshes.size(); ++i) delete meshes[i];
        meshes.clear();
        inFilenames.clear();
        disc_orders.clear();
        disc_methods.clear();
        doftag_names.clear();
        outFilename.clear();
        intxFilename.clear();
        baselineFile.clear();
        meshsets.clear();
        delete timer;
    }

    void timer_push( std::string operation )
    {
        timer_ops = timer->time_since_birth();
        opName    = operation;
    }

    void timer_pop()
    {
        double locElapsed = timer->time_since_birth() - timer_ops, avgElapsed = 0, maxElapsed = 0;
#ifdef MOAB_HAVE_MPI
        MPI_Reduce( &locElapsed, &maxElapsed, 1, MPI_DOUBLE, MPI_MAX, 0, pcomm->comm() );
        MPI_Reduce( &locElapsed, &avgElapsed, 1, MPI_DOUBLE, MPI_SUM, 0, pcomm->comm() );
#else
        maxElapsed = locElapsed;
        avgElapsed = locElapsed;
#endif
        if( !proc_id )
        {
            avgElapsed /= n_procs;
            std::cout << "[LOG] Time taken to " << opName.c_str() << ": max = " << maxElapsed
                      << ", avg = " << avgElapsed << "\n";
        }
        // std::cout << "\n[LOG" << proc_id << "] Time taken to " << opName << " = " <<
        // timer->time_since_birth() - timer_ops << std::endl;
        opName.clear();
    }

    void ParseCLOptions( int argc, char* argv[] )
    {
        ProgOptions opts;
        int imeshType                  = 0;
        std::string expectedFName      = "output.exo";
        std::string expectedMethod     = "fv";
        std::string expectedDofTagName = "GLOBAL_ID";
        int expectedOrder              = 1;

        if( !proc_id )
        {
            std::cout << "Command line options provided to mbtempest:\n  ";
            for( int iarg = 0; iarg < argc; ++iarg )
                std::cout << argv[iarg] << " ";
            std::cout << std::endl << std::endl;
        }

        opts.addOpt< int >( "type,t",
                            "Type of mesh (default=CS; Choose from [CS=0, RLL=1, ICO=2, OVERLAP_FILES=3, "
                            "OVERLAP_MEMORY=4, OVERLAP_MOAB=5])",
                            &imeshType );

        opts.addOpt< int >( "res,r", "Resolution of the mesh (default=5)", &blockSize );

        opts.addOpt< void >( "dual,d", "Output the dual of the mesh (relevant only for ICO mesh type)", &computeDual );

        opts.addOpt< std::string >( "file,f", "Output computed mesh or remapping weights to specified filename",
                                    &outFilename );
        opts.addOpt< std::string >(
            "load,l", "Input mesh filenames for source and target meshes. (relevant only when computing weights)",
            &expectedFName );

        opts.addOpt< void >( "advfront,a",
                             "Use the advancing front intersection instead of the Kd-tree based algorithm "
                             "to compute mesh intersections. (relevant only when computing weights)" );

        opts.addOpt< std::string >( "intx,i", "Output TempestRemap intersection mesh filename", &intxFilename );

        opts.addOpt< void >( "weights,w",
                             "Compute and output the weights using the overlap mesh (generally "
                             "relevant only for OVERLAP mesh)",
                             &computeWeights );

        opts.addOpt< std::string >( "method,m", "Discretization method for the source and target solution fields",
                                    &expectedMethod );

        opts.addOpt< int >( "order,o", "Discretization orders for the source and target solution fields",
                            &expectedOrder );

        opts.addOpt< std::string >( "global_id,g",
                                    "Tag name that contains the global DoF IDs for source and target solution fields",
                                    &expectedDofTagName );

        opts.addOpt< void >( "noconserve",
                             "Do not apply conservation to the resultant weights (relevant only "
                             "when computing weights)",
                             &mapOptions.fNoConservation );

        opts.addOpt< void >( "volumetric",
                             "Apply a volumetric projection to compute the weights (relevant only "
                             "when computing weights)",
                             &fVolumetric );

        opts.addOpt< void >( "invdist",
                             "Apply a inverse distance weight projection to compute the weights (relevant only "
                             "when computing FV-FV weights)",
                             &fInverseDistanceMap );

        opts.addOpt< int >( "monotonicity", "Ensure monotonicity in the weight generation. Options=[0,1,2,3]",
                            &ensureMonotonicity );

        opts.addOpt< void >( "enforce_convexity", "check convexity of input meshes to compute mesh intersections",
                             &enforceConvexity );

        opts.addOpt< void >( "nobubble", "do not use bubble on interior of spectral element nodes",
                             &mapOptions.fNoBubble );

        opts.addOpt< void >( "sparseconstraints", "do not use bubble on interior of spectral element nodes",
                             &mapOptions.fSparseConstraints );

        opts.addOpt< void >( "rrmgrids",
                             "At least one of the meshes is a regionally refined grid (relevant to "
                             "accelerate intersection computation)",
                             &rrmGrids );

        opts.addOpt< void >( "checkmap", "Check the generated map for conservation and consistency", &fCheck );

        opts.addOpt< void >( "verify",
                             "Verify the accuracy of the maps by projecting analytical functions "
                             "from source to target "
                             "grid by applying the maps",
                             &verifyWeights );

        opts.addOpt< std::string >( "baseline", "Output baseline file", &baselineFile );

        opts.parseCommandLine( argc, argv );

        // By default - use Kd-tree based search; if user asks for advancing front, disable Kd-tree
        // algorithm
        kdtreeSearch = opts.numOptSet( "advfront,a" ) == 0;

        switch( imeshType )
        {
            case 0:
                meshType = moab::TempestRemapper::CS;
                break;

            case 1:
                meshType = moab::TempestRemapper::RLL;
                break;

            case 2:
                meshType = moab::TempestRemapper::ICO;
                break;

            case 3:
                meshType = moab::TempestRemapper::OVERLAP_FILES;
                break;

            case 4:
                meshType = moab::TempestRemapper::OVERLAP_MEMORY;
                break;

            case 5:
                meshType = moab::TempestRemapper::OVERLAP_MOAB;
                break;

            default:
                meshType = moab::TempestRemapper::DEFAULT;
                break;
        }

        if( meshType > moab::TempestRemapper::ICO ) // compute overlap mesh and maps possibly
        {
            opts.getOptAllArgs( "load,l", inFilenames );
            opts.getOptAllArgs( "order,o", disc_orders );
            opts.getOptAllArgs( "method,m", disc_methods );
            opts.getOptAllArgs( "global_id,i", doftag_names );

            assert( inFilenames.size() == 2 );
            assert( disc_orders.size() == 2 );
            assert( disc_methods.size() == 2 );
            assert( ensureMonotonicity >= 0 && ensureMonotonicity <= 3 );

            // get discretization order parameters
            if( disc_orders.size() == 0 ) disc_orders.resize( 2, 1 );
            if( disc_orders.size() == 1 ) disc_orders.push_back( 1 );

            // get discretization method parameters
            if( disc_methods.size() == 0 ) disc_methods.resize( 2, "fv" );
            if( disc_methods.size() == 1 ) disc_methods.push_back( "fv" );

            // get DoF tagname parameters
            if( doftag_names.size() == 0 ) doftag_names.resize( 2, "GLOBAL_ID" );
            if( doftag_names.size() == 1 ) doftag_names.push_back( "GLOBAL_ID" );

            // for computing maps and overlaps, set discretization orders
            mapOptions.nPin           = disc_orders[0];
            mapOptions.nPout          = disc_orders[1];
            mapOptions.fSourceConcave = false;
            mapOptions.fTargetConcave = false;

            mapOptions.strMethod = "";
            switch( ensureMonotonicity )
            {
                case 0:
                    mapOptions.fMonotone = false;
                    break;
                case 3:
                    mapOptions.strMethod += "mono3;";
                    break;
                case 2:
                    mapOptions.strMethod += "mono2;";
                    break;
                default:
                    mapOptions.fMonotone = true;
            }
            mapOptions.fNoCorrectAreas = false;
            mapOptions.fNoCheck        = !fCheck;

            //assert( fVolumetric && fInverseDistanceMap == false );  // both options cannot be active
            if( fVolumetric ) mapOptions.strMethod += "volumetric;";
            if( fInverseDistanceMap ) mapOptions.strMethod += "invdist;";
        }

        // clear temporary string name
        expectedFName.clear();

        mapOptions.strOutputMapFile = outFilename;
        mapOptions.strOutputFormat  = "Netcdf4";
    }

  private:
    moab::CpuTimer* timer;
    double timer_ops;
    std::string opName;
};

// Forward declare some methods
static moab::ErrorCode CreateTempestMesh( ToolContext&, moab::TempestRemapper& remapper, Mesh* );
inline double sample_slow_harmonic( double dLon, double dLat );
inline double sample_fast_harmonic( double dLon, double dLat );
inline double sample_constant( double dLon, double dLat );
inline double sample_stationary_vortex( double dLon, double dLat );

int main( int argc, char* argv[] )
{
    moab::ErrorCode rval;
    NcError error( NcError::verbose_nonfatal );
    std::stringstream sstr;

    int proc_id = 0, nprocs = 1;
#ifdef MOAB_HAVE_MPI
    MPI_Init( &argc, &argv );
    MPI_Comm_rank( MPI_COMM_WORLD, &proc_id );
    MPI_Comm_size( MPI_COMM_WORLD, &nprocs );
#endif

    moab::Interface* mbCore = new( std::nothrow ) moab::Core;

    if( NULL == mbCore )
    {
        return 1;
    }

    ToolContext* runCtx;
#ifdef MOAB_HAVE_MPI
    moab::ParallelComm* pcomm = new moab::ParallelComm( mbCore, MPI_COMM_WORLD, 0 );

    runCtx                   = new ToolContext( mbCore, pcomm );
    const char* writeOptions = ( nprocs > 1 ? "PARALLEL=WRITE_PART" : "" );
#else
    runCtx                   = new ToolContext( mbCore );
    const char* writeOptions = "";
#endif
    runCtx->ParseCLOptions( argc, argv );

    const double radius_src  = 1.0 /*2.0*acos(-1.0)*/;
    const double radius_dest = 1.0 /*2.0*acos(-1.0)*/;

    moab::DebugOutput& outputFormatter = runCtx->outputFormatter;

#ifdef MOAB_HAVE_MPI
    moab::TempestRemapper remapper( mbCore, pcomm );
#else
    moab::TempestRemapper remapper( mbCore );
#endif
    remapper.meshValidate     = true;
    remapper.constructEdgeMap = false;
    remapper.initialize();

    // Default area_method = lHuiller; Options: Girard, GaussQuadrature (if TR is available)
#ifdef MOAB_HAVE_TEMPESTREMAP
    moab::IntxAreaUtils areaAdaptor( moab::IntxAreaUtils::GaussQuadrature );
#else
    moab::IntxAreaUtils areaAdaptor( moab::IntxAreaUtils::lHuiller );
#endif

    Mesh* tempest_mesh = new Mesh();
    runCtx->timer_push( "create Tempest mesh" );
    rval = CreateTempestMesh( *runCtx, remapper, tempest_mesh );MB_CHK_ERR( rval );
    runCtx->timer_pop();

    const double epsrel = ReferenceTolerance;  // ReferenceTolerance is defined in Defines.h in tempestremap;
                                               // Defines.h is included in SparseMatrix.h
                                               // SparseMatrix.h is included in OfflineMap.h
                                               // OfflineMap.h is included in TempestOnlineMap.hpp
                                               // TempestOnlineMap.hpp is included in this file, and is part of MOAB
    // Some constant parameters

    const double boxeps = 1e-6;

    if( runCtx->meshType == moab::TempestRemapper::OVERLAP_MEMORY )
    {
        // Compute intersections with MOAB
        // For the overlap method, choose between: "fuzzy", "exact" or "mixed"
        assert( runCtx->meshes.size() == 3 );

#ifdef MOAB_HAVE_MPI
        rval = pcomm->check_all_shared_handles();MB_CHK_ERR( rval );
#endif

        // Load the meshes and validate
        rval = remapper.ConvertTempestMesh( moab::Remapper::SourceMesh );MB_CHK_ERR( rval );
        rval = remapper.ConvertTempestMesh( moab::Remapper::TargetMesh );MB_CHK_ERR( rval );
        remapper.SetMeshType( moab::Remapper::OverlapMesh, moab::TempestRemapper::OVERLAP_FILES );
        rval = remapper.ConvertTempestMesh( moab::Remapper::OverlapMesh );MB_CHK_ERR( rval );
        rval = mbCore->write_mesh( "tempest_intersection.h5m", &runCtx->meshsets[2], 1 );MB_CHK_ERR( rval );

        // print verbosely about the problem setting
        {
            moab::Range rintxverts, rintxelems;
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[0], 0, rintxverts );MB_CHK_ERR( rval );
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[0], 2, rintxelems );MB_CHK_ERR( rval );
            outputFormatter.printf( 0, "The source set contains %lu vertices and %lu elements \n", rintxverts.size(),
                                    rintxelems.size() );

            moab::Range bintxverts, bintxelems;
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[1], 0, bintxverts );MB_CHK_ERR( rval );
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[1], 2, bintxelems );MB_CHK_ERR( rval );
            outputFormatter.printf( 0, "The target set contains %lu vertices and %lu elements \n", bintxverts.size(),
                                    bintxelems.size() );
        }

        moab::EntityHandle intxset;  // == remapper.GetMeshSet(moab::Remapper::OverlapMesh);

        // Compute intersections with MOAB
        {
            // Create the intersection on the sphere object
            runCtx->timer_push( "setup the intersector" );

            moab::Intx2MeshOnSphere* mbintx = new moab::Intx2MeshOnSphere( mbCore );
            mbintx->set_error_tolerance( epsrel );
            mbintx->set_box_error( boxeps );
            mbintx->set_radius_source_mesh( radius_src );
            mbintx->set_radius_destination_mesh( radius_dest );
#ifdef MOAB_HAVE_MPI
            mbintx->set_parallel_comm( pcomm );
#endif
            rval = mbintx->FindMaxEdges( runCtx->meshsets[0], runCtx->meshsets[1] );MB_CHK_ERR( rval );

#ifdef MOAB_HAVE_MPI
            moab::Range local_verts;
            rval = mbintx->build_processor_euler_boxes( runCtx->meshsets[1], local_verts );MB_CHK_ERR( rval );

            runCtx->timer_pop();

            moab::EntityHandle covering_set;
            runCtx->timer_push( "communicate the mesh" );
            rval = mbintx->construct_covering_set( runCtx->meshsets[0], covering_set );MB_CHK_ERR( rval );  // lots of communication if mesh is distributed very differently
            runCtx->timer_pop();
#else
            moab::EntityHandle covering_set = runCtx->meshsets[0];
#endif
            // Now let's invoke the MOAB intersection algorithm in parallel with a
            // source and target mesh set representing two different decompositions
            runCtx->timer_push( "compute intersections with MOAB" );
            rval = mbCore->create_meshset( moab::MESHSET_SET, intxset );MB_CHK_SET_ERR( rval, "Can't create new set" );
            rval = mbintx->intersect_meshes( covering_set, runCtx->meshsets[1], intxset );MB_CHK_SET_ERR( rval, "Can't compute the intersection of meshes on the sphere" );
            runCtx->timer_pop();

            // free the memory
            delete mbintx;
        }

        {
            moab::Range intxelems, intxverts;
            rval = mbCore->get_entities_by_dimension( intxset, 2, intxelems );MB_CHK_ERR( rval );
            rval = mbCore->get_entities_by_dimension( intxset, 0, intxverts, true );MB_CHK_ERR( rval );
            outputFormatter.printf( 0, "The intersection set contains %lu elements and %lu vertices \n",
                                    intxelems.size(), intxverts.size() );

            double initial_sarea =
                areaAdaptor.area_on_sphere( mbCore, runCtx->meshsets[0],
                                            radius_src );  // use the target to compute the initial area
            double initial_tarea =
                areaAdaptor.area_on_sphere( mbCore, runCtx->meshsets[1],
                                            radius_dest );  // use the target to compute the initial area
            double intx_area = areaAdaptor.area_on_sphere( mbCore, intxset, radius_src );

            outputFormatter.printf( 0, "mesh areas: source = %12.10f, target = %12.10f, intersection = %12.10f \n",
                                    initial_sarea, initial_tarea, intx_area );
            outputFormatter.printf( 0, "relative error w.r.t source = %12.10e, target = %12.10e \n",
                                    fabs( intx_area - initial_sarea ) / initial_sarea,
                                    fabs( intx_area - initial_tarea ) / initial_tarea );
        }

        // Write out our computed intersection file
        rval = mbCore->write_mesh( "moab_intersection.h5m", &intxset, 1 );MB_CHK_ERR( rval );

        if( runCtx->computeWeights )
        {
            runCtx->timer_push( "compute weights with the Tempest meshes" );
            // Call to generate an offline map with the tempest meshes
            OfflineMap weightMap;
            int err = GenerateOfflineMapWithMeshes( *runCtx->meshes[0], *runCtx->meshes[1], *runCtx->meshes[2],
                                                    runCtx->disc_methods[0],  // std::string strInputType
                                                    runCtx->disc_methods[1],  // std::string strOutputType,
                                                    runCtx->mapOptions, weightMap );
            runCtx->timer_pop();

            std::map< std::string, std::string > mapAttributes;
            if( err )
            {
                rval = moab::MB_FAILURE;
            }
            else
            {
                weightMap.Write( "outWeights.nc", mapAttributes );
            }
        }
    }
    else if( runCtx->meshType == moab::TempestRemapper::OVERLAP_MOAB )
    {
        // Usage: mpiexec -n 2 tools/mbtempest -t 5 -l mycs_2.h5m -l myico_2.h5m -f myoverlap_2.h5m
#ifdef MOAB_HAVE_MPI
        rval = pcomm->check_all_shared_handles();MB_CHK_ERR( rval );
#endif
        // print verbosely about the problem setting
        {
            moab::Range rintxverts, rintxelems;
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[0], 0, rintxverts );MB_CHK_ERR( rval );
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[0], 2, rintxelems );MB_CHK_ERR( rval );
            rval = moab::IntxUtils::fix_degenerate_quads( mbCore, runCtx->meshsets[0] );MB_CHK_ERR( rval );
            if( runCtx->enforceConvexity )
            {
                rval = moab::IntxUtils::enforce_convexity( mbCore, runCtx->meshsets[0], proc_id );MB_CHK_ERR( rval );
            }
            rval = areaAdaptor.positive_orientation( mbCore, runCtx->meshsets[0], radius_src );MB_CHK_ERR( rval );
            if( !proc_id )
                outputFormatter.printf( 0, "The source set contains %lu vertices and %lu elements \n",
                                        rintxverts.size(), rintxelems.size() );

            moab::Range bintxverts, bintxelems;
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[1], 0, bintxverts );MB_CHK_ERR( rval );
            rval = mbCore->get_entities_by_dimension( runCtx->meshsets[1], 2, bintxelems );MB_CHK_ERR( rval );
            rval = moab::IntxUtils::fix_degenerate_quads( mbCore, runCtx->meshsets[1] );MB_CHK_ERR( rval );
            if( runCtx->enforceConvexity )
            {
                rval = moab::IntxUtils::enforce_convexity( mbCore, runCtx->meshsets[1], proc_id );MB_CHK_ERR( rval );
            }
            rval = areaAdaptor.positive_orientation( mbCore, runCtx->meshsets[1], radius_dest );MB_CHK_ERR( rval );
            if( !proc_id )
                outputFormatter.printf( 0, "The target set contains %lu vertices and %lu elements \n",
                                        bintxverts.size(), bintxelems.size() );
        }

        // First compute the covering set such that the target elements are fully covered by the
        // lcoal source grid
        runCtx->timer_push( "construct covering set for intersection" );
        rval = remapper.ConstructCoveringSet( epsrel, 1.0, 1.0, boxeps, runCtx->rrmGrids );MB_CHK_ERR( rval );
        runCtx->timer_pop();

        // Compute intersections with MOAB with either the Kd-tree or the advancing front algorithm
        runCtx->timer_push( "setup and compute mesh intersections" );
        rval = remapper.ComputeOverlapMesh( runCtx->kdtreeSearch, false );MB_CHK_ERR( rval );
        runCtx->timer_pop();

        // print some diagnostic checks to see if the overlap grid resolved the input meshes
        // correctly
        {
            double local_areas[3],
                global_areas[3];  // Array for Initial area, and through Method 1 and Method 2
            // local_areas[0] = area_on_sphere_lHuiller ( mbCore, runCtx->meshsets[1], radius_src );
            local_areas[0] = areaAdaptor.area_on_sphere( mbCore, runCtx->meshsets[0], radius_src );
            local_areas[1] = areaAdaptor.area_on_sphere( mbCore, runCtx->meshsets[1], radius_dest );
            local_areas[2] = areaAdaptor.area_on_sphere( mbCore, runCtx->meshsets[2], radius_src );

#ifdef MOAB_HAVE_MPI
            MPI_Allreduce( &local_areas[0], &global_areas[0], 3, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD );
#else
            global_areas[0]                 = local_areas[0];
            global_areas[1]                 = local_areas[1];
            global_areas[2]                 = local_areas[2];
#endif
            if( !proc_id )
            {
                outputFormatter.printf( 0,
                                        "initial area: source mesh = %12.14f, target mesh = "
                                        "%12.14f, overlap mesh = %12.14f\n",
                                        global_areas[0], global_areas[1], global_areas[2] );
                // outputFormatter.printf ( 0, " area with l'Huiller: %12.14f with Girard:
                // %12.14f\n", global_areas[2], global_areas[3] ); outputFormatter.printf ( 0, "
                // relative difference areas = %12.10e\n", fabs ( global_areas[2] - global_areas[3]
                // ) / global_areas[2] );
                outputFormatter.printf( 0, "relative error w.r.t source = %12.14e, and target = %12.14e\n",
                                        fabs( global_areas[0] - global_areas[2] ) / global_areas[0],
                                        fabs( global_areas[1] - global_areas[2] ) / global_areas[1] );
            }
        }

        if( runCtx->intxFilename.size() )
        {
            moab::EntityHandle writableOverlapSet;
            rval = mbCore->create_meshset( moab::MESHSET_SET, writableOverlapSet );MB_CHK_SET_ERR( rval, "Can't create new set" );
            moab::EntityHandle meshOverlapSet = remapper.GetMeshSet( moab::Remapper::OverlapMesh );
            moab::Range ovEnts;
            rval = mbCore->get_entities_by_dimension( meshOverlapSet, 2, ovEnts );MB_CHK_SET_ERR( rval, "Can't create new set" );
            rval = mbCore->get_entities_by_dimension( meshOverlapSet, 0, ovEnts );MB_CHK_SET_ERR( rval, "Can't create new set" );

#ifdef MOAB_HAVE_MPI
            // Do not remove ghosted entities if we still haven't computed weights
            // Remove ghosted entities from overlap set before writing the new mesh set to file
            if( nprocs > 1 )
            {
                moab::Range ghostedEnts;
                rval = remapper.GetOverlapAugmentedEntities( ghostedEnts );MB_CHK_ERR( rval );
                ovEnts = moab::subtract( ovEnts, ghostedEnts );
            }
#endif
            rval = mbCore->add_entities( writableOverlapSet, ovEnts );MB_CHK_SET_ERR( rval, "Deleting ghosted entities failed" );

            size_t lastindex = runCtx->intxFilename.find_last_of( "." );
            sstr.str( "" );
            sstr << runCtx->intxFilename.substr( 0, lastindex ) << ".h5m";
            if( !runCtx->proc_id )
                std::cout << "Writing out the MOAB intersection mesh file to " << sstr.str() << std::endl;

            // Write out our computed intersection file
            rval = mbCore->write_file( sstr.str().c_str(), NULL, writeOptions, &writableOverlapSet, 1 );MB_CHK_ERR( rval );
        }

        if( runCtx->computeWeights )
        {
            runCtx->meshes[2] = remapper.GetMesh( moab::Remapper::OverlapMesh );
            if( !runCtx->proc_id ) std::cout << std::endl;

            runCtx->timer_push( "setup computation of weights" );
            // Call to generate the remapping weights with the tempest meshes
            moab::TempestOnlineMap* weightMap = new moab::TempestOnlineMap( &remapper );
            runCtx->timer_pop();

            runCtx->timer_push( "compute weights with TempestRemap" );
            rval = weightMap->GenerateRemappingWeights(
                runCtx->disc_methods[0],  // std::string strInputType
                runCtx->disc_methods[1],  // std::string strOutputType,
                runCtx->mapOptions,       // const GenerateOfflineMapAlgorithmOptions& options
                runCtx->doftag_names[0],  // const std::string& source_tag_name
                runCtx->doftag_names[1]   // const std::string& target_tag_name
            );MB_CHK_ERR( rval );
            runCtx->timer_pop();

            // Invoke the CheckMap routine on the TempestRemap serial interface directly, if running
            // on a single process
            if( nprocs == 1 )
            {
                const double dNormalTolerance = 1.0E-8;
                const double dStrictTolerance = 1.0E-12;
                weightMap->CheckMap( runCtx->fCheck, runCtx->fCheck, runCtx->fCheck && ( runCtx->ensureMonotonicity ),
                                     dNormalTolerance, dStrictTolerance );
            }

            if( runCtx->outFilename.size() )
            {
                // Write the map file to disk in parallel using either HDF5 or SCRIP interface
                rval = weightMap->WriteParallelMap( runCtx->outFilename.c_str() );MB_CHK_ERR( rval );

                // Write out the metadata information for the map file
                if( proc_id == 0 )
                {
                    size_t lastindex = runCtx->outFilename.find_last_of( "." );
                    sstr.str( "" );
                    sstr << runCtx->outFilename.substr( 0, lastindex ) << ".meta";

                    std::ofstream metafile( sstr.str() );
                    metafile << "Generator = MOAB-TempestRemap (mbtempest) Offline Regridding "
                                "Weight Generator"
                             << std::endl;
                    metafile << "domain_a = " << runCtx->inFilenames[0] << std::endl;
                    metafile << "domain_b = " << runCtx->inFilenames[1] << std::endl;
                    metafile << "domain_aNb = "
                             << ( runCtx->intxFilename.size() ? runCtx->intxFilename : "outOverlap.h5m" ) << std::endl;
                    metafile << "map_aPb = " << runCtx->outFilename << std::endl;
                    metafile << "type_src = " << runCtx->disc_methods[0] << std::endl;
                    metafile << "np_src = " << runCtx->disc_orders[0] << std::endl;
                    metafile << "concave_src = " << ( runCtx->mapOptions.fSourceConcave ? "true" : "false" )
                             << std::endl;
                    metafile << "type_dst = " << runCtx->disc_methods[1] << std::endl;
                    metafile << "np_dst = " << runCtx->disc_orders[1] << std::endl;
                    metafile << "concave_dst = " << ( runCtx->mapOptions.fTargetConcave ? "true" : "false" )
                             << std::endl;
                    metafile << "mono_type = " << runCtx->ensureMonotonicity << std::endl;
                    metafile << "bubble = " << ( runCtx->mapOptions.fNoBubble ? "false" : "true" ) << std::endl;
                    metafile << "version = "
                             << "MOAB v5.1.0+" << std::endl;
                    metafile.close();
                }
            }

            if( runCtx->verifyWeights )
            {
                // Let us pick a sampling test function for solution evaluation
                moab::TempestOnlineMap::sample_function testFunction =
                    &sample_stationary_vortex;  // &sample_slow_harmonic;

                runCtx->timer_push( "describe a solution on source grid" );
                moab::Tag srcAnalyticalFunction;
                rval = weightMap->DefineAnalyticalSolution( srcAnalyticalFunction, "AnalyticalSolnSrcExact",
                                                            moab::Remapper::SourceMesh, testFunction );MB_CHK_ERR( rval );
                runCtx->timer_pop();
                // rval = mbCore->write_file ( "srcWithSolnTag.h5m", NULL, writeOptions,
                // &runCtx->meshsets[0], 1 ); MB_CHK_ERR ( rval );

                runCtx->timer_push( "describe a solution on target grid" );
                moab::Tag tgtAnalyticalFunction;
                moab::Tag tgtProjectedFunction;
                rval = weightMap->DefineAnalyticalSolution( tgtAnalyticalFunction, "AnalyticalSolnTgtExact",
                                                            moab::Remapper::TargetMesh, testFunction,
                                                            &tgtProjectedFunction, "ProjectedSolnTgt" );MB_CHK_ERR( rval );
                // rval = mbCore->write_file ( "tgtWithSolnTag.h5m", NULL, writeOptions,
                // &runCtx->meshsets[1], 1 ); MB_CHK_ERR ( rval );
                runCtx->timer_pop();

                runCtx->timer_push( "compute solution projection on target grid" );
                rval = weightMap->ApplyWeights( srcAnalyticalFunction, tgtProjectedFunction );MB_CHK_ERR( rval );
                runCtx->timer_pop();
                rval = mbCore->write_file( "tgtWithSolnTag2.h5m", NULL, writeOptions, &runCtx->meshsets[1], 1 );MB_CHK_ERR( rval );

                if( nprocs == 1 && runCtx->baselineFile.size() )
                {
                    // save the field from tgtWithSolnTag2 in a text file, and global ids for cells
                    moab::Range tgtCells;
                    rval = mbCore->get_entities_by_dimension( runCtx->meshsets[1], 2, tgtCells );MB_CHK_ERR( rval );
                    std::vector< int > globIds;
                    globIds.resize( tgtCells.size() );
                    std::vector< double > vals;
                    vals.resize( tgtCells.size() );
                    moab::Tag projTag;
                    rval = mbCore->tag_get_handle( "ProjectedSolnTgt", projTag );MB_CHK_ERR( rval );
                    moab::Tag gid = mbCore->globalId_tag();
                    rval          = mbCore->tag_get_data( gid, tgtCells, &globIds[0] );MB_CHK_ERR( rval );
                    rval = mbCore->tag_get_data( projTag, tgtCells, &vals[0] );MB_CHK_ERR( rval );
                    std::fstream fs;
                    fs.open( runCtx->baselineFile.c_str(), std::fstream::out );
                    fs << std::setprecision( 15 );  // maximum precision for doubles
                    for( size_t i = 0; i < tgtCells.size(); i++ )
                        fs << globIds[i] << " " << vals[i] << "\n";
                    fs.close();
                    // for good measure, save the source file too, with the tag AnalyticalSolnSrcExact
                    // it will be used later to test, along with a target file
                    rval = mbCore->write_file( "srcWithSolnTag.h5m", NULL, writeOptions, &runCtx->meshsets[0], 1 );MB_CHK_ERR( rval );
                }
                runCtx->timer_push( "compute error metrics against analytical solution on target grid" );
                std::map< std::string, double > errMetrics;
                rval = weightMap->ComputeMetrics( moab::Remapper::TargetMesh, tgtAnalyticalFunction,
                                                  tgtProjectedFunction, errMetrics, true );MB_CHK_ERR( rval );
                runCtx->timer_pop();
            }

            delete weightMap;
        }
    }

    // Clean up
    remapper.clear();
    delete runCtx;
    delete mbCore;

#ifdef MOAB_HAVE_MPI
    MPI_Finalize();
#endif
    exit( 0 );
}

static moab::ErrorCode CreateTempestMesh( ToolContext& ctx, moab::TempestRemapper& remapper, Mesh* tempest_mesh )
{
    moab::ErrorCode rval = moab::MB_SUCCESS;
    int err;
    moab::DebugOutput& outputFormatter = ctx.outputFormatter;

    if( !ctx.proc_id )
    {
        outputFormatter.printf( 0, "Creating TempestRemap Mesh object ...\n" );
    }

    if( ctx.meshType == moab::TempestRemapper::OVERLAP_FILES )
    {
        // For the overlap method, choose between: "fuzzy", "exact" or "mixed"
        err = GenerateOverlapMesh( ctx.inFilenames[0], ctx.inFilenames[1], *tempest_mesh, ctx.outFilename, "NetCDF4",
                                   "exact", true );

        if( err )
        {
            rval = moab::MB_FAILURE;
        }
        else
        {
            ctx.meshes.push_back( tempest_mesh );
        }
    }
    else if( ctx.meshType == moab::TempestRemapper::OVERLAP_MEMORY )
    {
        // Load the meshes and validate
        ctx.meshsets.resize( 3 );
        ctx.meshes.resize( 3 );
        ctx.meshsets[0] = remapper.GetMeshSet( moab::Remapper::SourceMesh );
        ctx.meshsets[1] = remapper.GetMeshSet( moab::Remapper::TargetMesh );
        ctx.meshsets[2] = remapper.GetMeshSet( moab::Remapper::OverlapMesh );

        // First the source
        rval = remapper.LoadMesh( moab::Remapper::SourceMesh, ctx.inFilenames[0], moab::TempestRemapper::DEFAULT );MB_CHK_ERR( rval );
        ctx.meshes[0] = remapper.GetMesh( moab::Remapper::SourceMesh );

        // Next the target
        rval = remapper.LoadMesh( moab::Remapper::TargetMesh, ctx.inFilenames[1], moab::TempestRemapper::DEFAULT );MB_CHK_ERR( rval );
        ctx.meshes[1] = remapper.GetMesh( moab::Remapper::TargetMesh );

        // Now let us construct the overlap mesh, by calling TempestRemap interface directly
        // For the overlap method, choose between: "fuzzy", "exact" or "mixed"
        err = GenerateOverlapWithMeshes( *ctx.meshes[0], *ctx.meshes[1], *tempest_mesh, "" /*ctx.outFilename*/,
                                         "NetCDF4", "exact", false );

        if( err )
        {
            rval = moab::MB_FAILURE;
        }
        else
        {
            remapper.SetMesh( moab::Remapper::OverlapMesh, tempest_mesh );
            ctx.meshes[2] = remapper.GetMesh( moab::Remapper::OverlapMesh );
            // ctx.meshes.push_back(*tempest_mesh);
        }
    }
    else if( ctx.meshType == moab::TempestRemapper::OVERLAP_MOAB )
    {
        ctx.meshsets.resize( 3 );
        ctx.meshes.resize( 3 );
        ctx.meshsets[0] = remapper.GetMeshSet( moab::Remapper::SourceMesh );
        ctx.meshsets[1] = remapper.GetMeshSet( moab::Remapper::TargetMesh );
        ctx.meshsets[2] = remapper.GetMeshSet( moab::Remapper::OverlapMesh );

        const double radius_src  = 1.0 /*2.0*acos(-1.0)*/;
        const double radius_dest = 1.0 /*2.0*acos(-1.0)*/;

        const char* additional_read_opts = ( ctx.n_procs > 1 ? "NO_SET_CONTAINING_PARENTS;" : "" );
        std::vector< int > smetadata, tmetadata;
        // Load the source mesh and validate
        rval = remapper.LoadNativeMesh( ctx.inFilenames[0], ctx.meshsets[0], smetadata, additional_read_opts );MB_CHK_ERR( rval );
        if( smetadata.size() )
        {
            // remapper.SetMeshType( moab::Remapper::SourceMesh,
            //                       static_cast< moab::TempestRemapper::TempestMeshType >( smetadata[0] ), &smetadata
            //                       );
            remapper.SetMeshType( moab::Remapper::SourceMesh,
                                  static_cast< moab::TempestRemapper::TempestMeshType >( smetadata[0] ) );
        }
        // Rescale the radius of both to compute the intersection
        rval = moab::IntxUtils::ScaleToRadius( ctx.mbcore, ctx.meshsets[0], radius_src );MB_CHK_ERR( rval );
        rval = remapper.ConvertMeshToTempest( moab::Remapper::SourceMesh );MB_CHK_ERR( rval );
        ctx.meshes[0] = remapper.GetMesh( moab::Remapper::SourceMesh );

        // Load the target mesh and validate
        rval = remapper.LoadNativeMesh( ctx.inFilenames[1], ctx.meshsets[1], tmetadata, additional_read_opts );MB_CHK_ERR( rval );
        if( tmetadata.size() )
        {
            // remapper.SetMeshType( moab::Remapper::TargetMesh,
            //                       static_cast< moab::TempestRemapper::TempestMeshType >( tmetadata[0] ), &tmetadata
            //                       );
            remapper.SetMeshType( moab::Remapper::TargetMesh,
                                  static_cast< moab::TempestRemapper::TempestMeshType >( tmetadata[0] ) );
        }
        rval = moab::IntxUtils::ScaleToRadius( ctx.mbcore, ctx.meshsets[1], radius_dest );MB_CHK_ERR( rval );
        rval = remapper.ConvertMeshToTempest( moab::Remapper::TargetMesh );MB_CHK_ERR( rval );
        ctx.meshes[1] = remapper.GetMesh( moab::Remapper::TargetMesh );
    }
    else if( ctx.meshType == moab::TempestRemapper::ICO )
    {
        err = GenerateICOMesh( *tempest_mesh, ctx.blockSize, ctx.computeDual, ctx.outFilename, "NetCDF4" );

        if( err )
        {
            rval = moab::MB_FAILURE;
        }
        else
        {
            ctx.meshes.push_back( tempest_mesh );
        }
    }
    else if( ctx.meshType == moab::TempestRemapper::RLL )
    {
        err = GenerateRLLMesh( *tempest_mesh,                     // Mesh& meshOut,
                               ctx.blockSize * 2, ctx.blockSize,  // int nLongitudes, int nLatitudes,
                               0.0, 360.0,                        // double dLonBegin, double dLonEnd,
                               -90.0, 90.0,                       // double dLatBegin, double dLatEnd,
                               false, false, false,  // bool fGlobalCap, bool fFlipLatLon, bool fForceGlobal,
                               "" /*ctx.inFilename*/, "",
                               "",  // std::string strInputFile, std::string strInputFileLonName, std::string
                                    // strInputFileLatName,
                               ctx.outFilename, "NetCDF4",  // std::string strOutputFile, std::string strOutputFormat
                               true                         // bool fVerbose
        );

        if( err )
        {
            rval = moab::MB_FAILURE;
        }
        else
        {
            ctx.meshes.push_back( tempest_mesh );
        }
    }
    else  // default
    {
        err = GenerateCSMesh( *tempest_mesh, ctx.blockSize, ctx.outFilename, "NetCDF4" );

        if( err )
        {
            rval = moab::MB_FAILURE;
        }
        else
        {
            ctx.meshes.push_back( tempest_mesh );
        }
    }

    if( ctx.meshType != moab::TempestRemapper::OVERLAP_MOAB && !tempest_mesh )
    {
        std::cout << "Tempest Mesh is not a complete object; Quitting...";
        exit( -1 );
    }

    return rval;
}

///////////////////////////////////////////////
//         Test functions

double sample_slow_harmonic( double dLon, double dLat )
{
    return ( 2.0 + cos( dLat ) * cos( dLat ) * cos( 2.0 * dLon ) );
}

double sample_fast_harmonic( double dLon, double dLat )
{
    return ( 2.0 + pow( sin( 2.0 * dLat ), 16.0 ) * cos( 16.0 * dLon ) );
    // return (2.0 + pow(cos(2.0 * dLat), 16.0) * cos(16.0 * dLon));
}

double sample_constant( double /*dLon*/, double /*dLat*/ )
{
    return 1.0;
}

double sample_stationary_vortex( double dLon, double dLat )
{
    const double dLon0 = 0.0;
    const double dLat0 = 0.6;
    const double dR0   = 3.0;
    const double dD    = 5.0;
    const double dT    = 6.0;

    ///     Find the rotated longitude and latitude of a point on a sphere
    ///     with pole at (dLonC, dLatC).
    {
        double dSinC = sin( dLat0 );
        double dCosC = cos( dLat0 );
        double dCosT = cos( dLat );
        double dSinT = sin( dLat );

        double dTrm = dCosT * cos( dLon - dLon0 );
        double dX   = dSinC * dTrm - dCosC * dSinT;
        double dY   = dCosT * sin( dLon - dLon0 );
        double dZ   = dSinC * dSinT + dCosC * dTrm;

        dLon = atan2( dY, dX );
        if( dLon < 0.0 )
        {
            dLon += 2.0 * M_PI;
        }
        dLat = asin( dZ );
    }

    double dRho = dR0 * cos( dLat );
    double dVt  = 3.0 * sqrt( 3.0 ) / 2.0 / cosh( dRho ) / cosh( dRho ) * tanh( dRho );

    double dOmega;
    if( dRho == 0.0 )
    {
        dOmega = 0.0;
    }
    else
    {
        dOmega = dVt / dRho;
    }

    return ( 1.0 - tanh( dRho / dD * sin( dLon - dOmega * dT ) ) );
}

///////////////////////////////////////////////