NCHelperMPAS.cpp

Go to the documentation of this file.

#include "NCHelperMPAS.hpp"
#include "moab/ReadUtilIface.hpp"
#include "moab/FileOptions.hpp"
#include "moab/SpectralMeshTool.hpp"
#include "MBTagConventions.hpp"

#ifdef MOAB_HAVE_ZOLTAN
#include "moab/ZoltanPartitioner.hpp"
#endif

#include 

namespace moab
{

const int DEFAULT_MAX_EDGES_PER_CELL = 10;

NCHelperMPAS::NCHelperMPAS( ReadNC* readNC, int fileId, const FileOptions& opts, EntityHandle fileSet )
    : UcdNCHelper( readNC, fileId, opts, fileSet ), maxEdgesPerCell( DEFAULT_MAX_EDGES_PER_CELL ), numCellGroups( 0 ),
      createGatherSet( false )
{
    // Ignore variables containing topological information
    ignoredVarNames.insert( "nEdgesOnEdge" );
    ignoredVarNames.insert( "nEdgesOnCell" );
    ignoredVarNames.insert( "edgesOnVertex" );
    ignoredVarNames.insert( "cellsOnVertex" );
    ignoredVarNames.insert( "verticesOnEdge" );
    ignoredVarNames.insert( "edgesOnEdge" );
    ignoredVarNames.insert( "cellsOnEdge" );
    ignoredVarNames.insert( "verticesOnCell" );
    ignoredVarNames.insert( "edgesOnCell" );
    ignoredVarNames.insert( "cellsOnCell" );
    // ignore variables containing mesh related info, that are currently saved as variable tags on
    // file set ; if needed, these should be saved as dense tags, and read accordingly, in parallel
    ignoredVarNames.insert( "weightsOnEdge" );
    ignoredVarNames.insert( "angleEdge" );
    ignoredVarNames.insert( "areaCell" );
    ignoredVarNames.insert( "areaTriangle" );
    ignoredVarNames.insert( "dcEdge" );
    ignoredVarNames.insert( "dv1Edge" );
    ignoredVarNames.insert( "dv2Edge" );
    ignoredVarNames.insert( "fEdge" );
    ignoredVarNames.insert( "fVertex" );
    ignoredVarNames.insert( "h_s" );
    ignoredVarNames.insert( "kiteAreasOnVertex" );
    ignoredVarNames.insert( "latCell" );
    ignoredVarNames.insert( "latEdge" );
    ignoredVarNames.insert( "latVertex" );
    ignoredVarNames.insert( "lonCell" );
    ignoredVarNames.insert( "lonEdge" );
    ignoredVarNames.insert( "lonVertex" );
    ignoredVarNames.insert( "meshDensity" );
    ignoredVarNames.insert( "xCell" );
    ignoredVarNames.insert( "xEdge" );
    ignoredVarNames.insert( "xVertex" );
    ignoredVarNames.insert( "yCell" );
    ignoredVarNames.insert( "yEdge" );
    ignoredVarNames.insert( "yVertex" );
    ignoredVarNames.insert( "zCell" );
    ignoredVarNames.insert( "zEdge" );
    ignoredVarNames.insert( "zVertex" );
    // Ignore variables for index conversion
    ignoredVarNames.insert( "indexToVertexID" );
    ignoredVarNames.insert( "indexToEdgeID" );
    ignoredVarNames.insert( "indexToCellID" );
}

bool NCHelperMPAS::can_read_file( ReadNC* readNC )
{
    std::vector< std::string >& dimNames = readNC->dimNames;

    // If dimension name "vertexDegree" exists then it should be the MPAS grid
    if( std::find( dimNames.begin(), dimNames.end(), std::string( "vertexDegree" ) ) != dimNames.end() ) return true;

    return false;
}

ErrorCode NCHelperMPAS::init_mesh_vals()
{
    std::vector< std::string >& dimNames              = _readNC->dimNames;
    std::vector< int >& dimLens                       = _readNC->dimLens;
    std::map< std::string, ReadNC::VarData >& varInfo = _readNC->varInfo;

    ErrorCode rval;
    unsigned int idx;
    std::vector< std::string >::iterator vit;

    // Get max edges per cell reported in the MPAS file header
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "maxEdges" ) ) != dimNames.end() )
    {
        idx             = vit - dimNames.begin();
        maxEdgesPerCell = dimLens[idx];
        if( maxEdgesPerCell > DEFAULT_MAX_EDGES_PER_CELL )
        {
            MB_SET_ERR( MB_INVALID_SIZE,
                        "maxEdgesPerCell read from the MPAS file header has exceeded " << DEFAULT_MAX_EDGES_PER_CELL );
        }
    }

    // Look for time dimension
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "Time" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else if( ( vit = std::find( dimNames.begin(), dimNames.end(), "time" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else
    {
        MB_SET_ERR( MB_FAILURE, "Couldn't find 'Time' or 'time' dimension" );
    }
    tDim       = idx;
    nTimeSteps = dimLens[idx];

    // Get number of cells
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nCells" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else
    {
        MB_SET_ERR( MB_FAILURE, "Couldn't find 'nCells' dimension" );
    }
    cDim   = idx;
    nCells = dimLens[idx];

    // Get number of edges
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nEdges" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else
    {
        MB_SET_ERR( MB_FAILURE, "Couldn't find 'nEdges' dimension" );
    }
    eDim   = idx;
    nEdges = dimLens[idx];

    // Get number of vertices
    vDim = -1;
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nVertices" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else
    {
        MB_SET_ERR( MB_FAILURE, "Couldn't find 'nVertices' dimension" );
    }
    vDim      = idx;
    nVertices = dimLens[idx];

    // Get number of vertex levels
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nVertLevels" ) ) != dimNames.end() )
        idx = vit - dimNames.begin();
    else
    {
        std::cerr << "Warning: dimension nVertLevels not found in header.\nThe file may contain "
                     "just the mesh"
                  << std::endl;
    }
    levDim  = idx;
    nLevels = dimLens[idx];

    // Dimension indices for other optional levels
    std::vector< unsigned int > opt_lev_dims;

    // Get index of vertex levels P1
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nVertLevelsP1" ) ) != dimNames.end() )
    {
        idx = vit - dimNames.begin();
        opt_lev_dims.push_back( idx );
    }

    // Get index of vertex levels P2
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nVertLevelsP2" ) ) != dimNames.end() )
    {
        idx = vit - dimNames.begin();
        opt_lev_dims.push_back( idx );
    }

    // Get index of soil levels
    if( ( vit = std::find( dimNames.begin(), dimNames.end(), "nSoilLevels" ) ) != dimNames.end() )
    {
        idx = vit - dimNames.begin();
        opt_lev_dims.push_back( idx );
    }

    std::map< std::string, ReadNC::VarData >::iterator vmit;

    // Store time coordinate values in tVals
    if( nTimeSteps > 0 )
    {
        // Note, two possible types for xtime variable: double(Time) or char(Time, StrLen)
        if( ( vmit = varInfo.find( "xtime" ) ) != varInfo.end() && ( *vmit ).second.varDims.size() == 1 )
        {
            // If xtime variable is double type, read time coordinate values to tVals
            rval = read_coordinate( "xtime", 0, nTimeSteps - 1, tVals );MB_CHK_SET_ERR( rval, "Trouble reading 'xtime' variable" );
        }
        else
        {
            // If xtime variable does not exist, or it is string type, set dummy values to tVals
            for( int t = 0; t < nTimeSteps; t++ )
                tVals.push_back( (double)t );
        }
    }

    // For each variable, determine the entity location type and number of levels
    for( vmit = varInfo.begin(); vmit != varInfo.end(); ++vmit )
    {
        ReadNC::VarData& vd = ( *vmit ).second;

        // Default entLoc is ENTLOCSET
        if( std::find( vd.varDims.begin(), vd.varDims.end(), tDim ) != vd.varDims.end() )
        {
            if( std::find( vd.varDims.begin(), vd.varDims.end(), vDim ) != vd.varDims.end() )
                vd.entLoc = ReadNC::ENTLOCVERT;
            else if( std::find( vd.varDims.begin(), vd.varDims.end(), eDim ) != vd.varDims.end() )
                vd.entLoc = ReadNC::ENTLOCEDGE;
            else if( std::find( vd.varDims.begin(), vd.varDims.end(), cDim ) != vd.varDims.end() )
                vd.entLoc = ReadNC::ENTLOCFACE;
        }

        // Default numLev is 0
        if( std::find( vd.varDims.begin(), vd.varDims.end(), levDim ) != vd.varDims.end() )
            vd.numLev = nLevels;
        else
        {
            // If nVertLevels dimension is not found, try other optional levels such as
            // nVertLevelsP1
            for( unsigned int i = 0; i < opt_lev_dims.size(); i++ )
            {
                if( std::find( vd.varDims.begin(), vd.varDims.end(), opt_lev_dims[i] ) != vd.varDims.end() )
                {
                    vd.numLev = dimLens[opt_lev_dims[i]];
                    break;
                }
            }
        }

        // Hack: ignore variables with more than 3 dimensions, e.g. tracers(Time, nCells,
        // nVertLevels, nTracers)
        if( vd.varDims.size() > 3 ) ignoredVarNames.insert( vd.varName );
    }

    // Hack: create dummy variables for dimensions (like nCells) with no corresponding coordinate
    // variables
    rval = create_dummy_variables();MB_CHK_SET_ERR( rval, "Failed to create dummy variables" );

    return MB_SUCCESS;
}

// When noMesh option is used on this read, the old ReadNC class instance for last read can get out
// of scope (and deleted). The old instance initialized some variables properly when the mesh was
// created, but they are now lost. The new instance (will not create the mesh with noMesh option)
// has to restore them based on the existing mesh from last read
ErrorCode NCHelperMPAS::check_existing_mesh()
{
    Interface*& mbImpl = _readNC->mbImpl;
    Tag& mGlobalIdTag  = _readNC->mGlobalIdTag;
    bool& noMesh       = _readNC->noMesh;

    if( noMesh )
    {
        ErrorCode rval;

        // Restore numCellGroups
        if( 0 == numCellGroups )
        {
            Tag numCellGroupsTag;
            rval = mbImpl->tag_get_handle( "__NUM_CELL_GROUPS", 1, MB_TYPE_INTEGER, numCellGroupsTag );MB_CHK_SET_ERR( rval, "Trouble getting __NUM_CELL_GROUPS tag" );
            if( MB_SUCCESS == rval ) rval = mbImpl->tag_get_data( numCellGroupsTag, &_fileSet, 1, &numCellGroups );MB_CHK_SET_ERR( rval, "Trouble getting data of __NUM_CELL_GROUPS tag" );
        }

        if( localGidVerts.empty() )
        {
            // Get all vertices from current file set (it is the input set in no_mesh scenario)
            Range local_verts;
            rval = mbImpl->get_entities_by_dimension( _fileSet, 0, local_verts );MB_CHK_SET_ERR( rval, "Trouble getting local vertices in current file set" );

            if( !local_verts.empty() )
            {
                std::vector< int > gids( local_verts.size() );

                // !IMPORTANT : this has to be the GLOBAL_ID tag
                rval = mbImpl->tag_get_data( mGlobalIdTag, local_verts, &gids[0] );MB_CHK_SET_ERR( rval, "Trouble getting local gid values of vertices" );

                // Restore localGidVerts
                std::copy( gids.rbegin(), gids.rend(), range_inserter( localGidVerts ) );
                nLocalVertices = localGidVerts.size();
            }
        }

        if( localGidEdges.empty() )
        {
            // Get all edges from current file set (it is the input set in no_mesh scenario)
            Range local_edges;
            rval = mbImpl->get_entities_by_dimension( _fileSet, 1, local_edges );MB_CHK_SET_ERR( rval, "Trouble getting local edges in current file set" );

            if( !local_edges.empty() )
            {
                std::vector< int > gids( local_edges.size() );

                // !IMPORTANT : this has to be the GLOBAL_ID tag
                rval = mbImpl->tag_get_data( mGlobalIdTag, local_edges, &gids[0] );MB_CHK_SET_ERR( rval, "Trouble getting local gid values of edges" );

                // Restore localGidEdges
                std::copy( gids.rbegin(), gids.rend(), range_inserter( localGidEdges ) );
                nLocalEdges = localGidEdges.size();
            }
        }

        if( localGidCells.empty() )
        {
            // Get all cells from current file set (it is the input set in no_mesh scenario)
            Range local_cells;
            rval = mbImpl->get_entities_by_dimension( _fileSet, 2, local_cells );MB_CHK_SET_ERR( rval, "Trouble getting local cells in current file set" );

            if( !local_cells.empty() )
            {
                std::vector< int > gids( local_cells.size() );

                // !IMPORTANT : this has to be the GLOBAL_ID tag
                rval = mbImpl->tag_get_data( mGlobalIdTag, local_cells, &gids[0] );MB_CHK_SET_ERR( rval, "Trouble getting local gid values of cells" );

                // Restore localGidCells
                std::copy( gids.rbegin(), gids.rend(), range_inserter( localGidCells ) );
                nLocalCells = localGidCells.size();

                if( numCellGroups > 1 )
                {
                    // Restore cellHandleToGlobalID map
                    Range::const_iterator rit;
                    int i;
                    for( rit = local_cells.begin(), i = 0; rit != local_cells.end(); ++rit, i++ )
                        cellHandleToGlobalID[*rit] = gids[i];
                }
            }
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_mesh( Range& faces )
{
    Interface*& mbImpl    = _readNC->mbImpl;
    bool& noMixedElements = _readNC->noMixedElements;
    bool& noEdges         = _readNC->noEdges;
    DebugOutput& dbgOut   = _readNC->dbgOut;

#ifdef MOAB_HAVE_MPI
    int rank         = 0;
    int procs        = 1;
    bool& isParallel = _readNC->isParallel;
    ParallelComm* myPcomm = NULL;
    if( isParallel )
    {
        myPcomm  = _readNC->myPcomm;
        rank                   = myPcomm->proc_config().proc_rank();
        procs                  = myPcomm->proc_config().proc_size();
    }

    // Need to know whether we'll be creating gather mesh
    if( rank == _readNC->gatherSetRank ) createGatherSet = true;

    if( procs >= 2 )
    {
        // Shift rank to obtain a rotated trivial partition
        int shifted_rank           = rank;
        int& trivialPartitionShift = _readNC->trivialPartitionShift;
        if( trivialPartitionShift > 0 ) shifted_rank = ( rank + trivialPartitionShift ) % procs;

        // Compute the number of local cells on this proc
        nLocalCells = int( std::floor( 1.0 * nCells / procs ) );

        // The starting global cell index in the MPAS file for this proc
        int start_cell_idx = shifted_rank * nLocalCells;

        // Number of extra cells after equal split over procs
        int iextra = nCells % procs;

        // Allocate extra cells over procs
        if( shifted_rank < iextra ) nLocalCells++;
        start_cell_idx += std::min( shifted_rank, iextra );

        start_cell_idx++;  // 0 based -> 1 based

        // Redistribute local cells after trivial partition (e.g. apply Zoltan partition)
        ErrorCode rval = redistribute_local_cells( start_cell_idx, myPcomm );MB_CHK_SET_ERR( rval, "Failed to redistribute local cells after trivial partition" );
    }
    else
    {
        nLocalCells = nCells;
        localGidCells.insert( 1, nLocalCells );
    }
#else
    nLocalCells = nCells;
    localGidCells.insert( 1, nLocalCells );
#endif
    dbgOut.tprintf( 1, " localGidCells.psize() = %d\n", (int)localGidCells.psize() );
    dbgOut.tprintf( 1, " localGidCells.size() = %d\n", (int)localGidCells.size() );

    // Read number of edges on each local cell, to calculate actual maxEdgesPerCell
    int nEdgesOnCellVarId;
    int success = NCFUNC( inq_varid )( _fileId, "nEdgesOnCell", &nEdgesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of nEdgesOnCell" );
    std::vector< int > num_edges_on_local_cells( nLocalCells );
#ifdef MOAB_HAVE_PNETCDF
    size_t nb_reads = localGidCells.psize();
    std::vector< int > requests( nb_reads );
    std::vector< int > statuss( nb_reads );
    size_t idxReq = 0;
#endif
    size_t indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidCells.pair_begin(); pair_iter != localGidCells.pair_end();
         ++pair_iter )
    {
        EntityHandle starth  = pair_iter->first;
        EntityHandle endh    = pair_iter->second;
        NCDF_SIZE read_start = (NCDF_SIZE)( starth - 1 );
        NCDF_SIZE read_count = (NCDF_SIZE)( endh - starth + 1 );

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count,
                                           &( num_edges_on_local_cells[indexInArray] ), &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count,
                                         &( num_edges_on_local_cells[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read nEdgesOnCell data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 );
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Get local maxEdgesPerCell on this proc
    int local_max_edges_per_cell =
        *( std::max_element( num_edges_on_local_cells.begin(), num_edges_on_local_cells.end() ) );
    maxEdgesPerCell = local_max_edges_per_cell;

    // If parallel, do a MPI_Allreduce to get global maxEdgesPerCell across all procs
#ifdef MOAB_HAVE_MPI
    if( procs >= 2 )
    {
        int global_max_edges_per_cell;
        ParallelComm*& myPcomm = _readNC->myPcomm;
        MPI_Allreduce( &local_max_edges_per_cell, &global_max_edges_per_cell, 1, MPI_INT, MPI_MAX,
                       myPcomm->proc_config().proc_comm() );
        assert( local_max_edges_per_cell <= global_max_edges_per_cell );
        maxEdgesPerCell = global_max_edges_per_cell;
        if( 0 == rank ) dbgOut.tprintf( 1, "  global_max_edges_per_cell = %d\n", global_max_edges_per_cell );
    }
#endif

    // Read vertices on each local cell, to get localGidVerts and cell connectivity later
    int verticesOnCellVarId;
    success = NCFUNC( inq_varid )( _fileId, "verticesOnCell", &verticesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of verticesOnCell" );
    std::vector< int > vertices_on_local_cells( nLocalCells * maxEdgesPerCell );
#ifdef MOAB_HAVE_PNETCDF
    idxReq = 0;
#endif
    indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidCells.pair_begin(); pair_iter != localGidCells.pair_end();
         ++pair_iter )
    {
        EntityHandle starth      = pair_iter->first;
        EntityHandle endh        = pair_iter->second;
        NCDF_SIZE read_starts[2] = { static_cast< NCDF_SIZE >( starth - 1 ), 0 };
        NCDF_SIZE read_counts[2] = { static_cast< NCDF_SIZE >( endh - starth + 1 ),
                                     static_cast< NCDF_SIZE >( maxEdgesPerCell ) };

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts,
                                           &( vertices_on_local_cells[indexInArray] ), &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts,
                                         &( vertices_on_local_cells[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnCell data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 ) * maxEdgesPerCell;
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Correct local cell vertices array, replace the padded vertices with the last vertices
    // in the corresponding cells; sometimes the padded vertices are 0, sometimes a large
    // vertex id. Make sure they are consistent to our padded option
    for( int local_cell_idx = 0; local_cell_idx < nLocalCells; local_cell_idx++ )
    {
        int num_edges             = num_edges_on_local_cells[local_cell_idx];
        int idx_in_local_vert_arr = local_cell_idx * maxEdgesPerCell;
        int last_vert_idx         = vertices_on_local_cells[idx_in_local_vert_arr + num_edges - 1];
        for( int i = num_edges; i < maxEdgesPerCell; i++ )
            vertices_on_local_cells[idx_in_local_vert_arr + i] = last_vert_idx;
    }

    // Create local vertices
    EntityHandle start_vertex;
    ErrorCode rval = create_local_vertices( vertices_on_local_cells, start_vertex );MB_CHK_SET_ERR( rval, "Failed to create local vertices for MPAS mesh" );

    // Create local edges (unless NO_EDGES read option is set)
    if( !noEdges )
    {
        rval = create_local_edges( start_vertex, num_edges_on_local_cells );MB_CHK_SET_ERR( rval, "Failed to create local edges for MPAS mesh" );
    }

    // Create local cells, either unpadded or padded
    if( noMixedElements )
    {
        rval = create_padded_local_cells( vertices_on_local_cells, start_vertex, faces );MB_CHK_SET_ERR( rval, "Failed to create padded local cells for MPAS mesh" );
    }
    else
    {
        rval = create_local_cells( vertices_on_local_cells, num_edges_on_local_cells, start_vertex, faces );MB_CHK_SET_ERR( rval, "Failed to create local cells for MPAS mesh" );
    }

    // Set tag for numCellGroups
    Tag numCellGroupsTag = 0;
    rval                 = mbImpl->tag_get_handle( "__NUM_CELL_GROUPS", 1, MB_TYPE_INTEGER, numCellGroupsTag,
                                                   MB_TAG_SPARSE | MB_TAG_CREAT );MB_CHK_SET_ERR( rval, "Trouble creating __NUM_CELL_GROUPS tag" );
    rval = mbImpl->tag_set_data( numCellGroupsTag, &_fileSet, 1, &numCellGroups );MB_CHK_SET_ERR( rval, "Trouble setting data to __NUM_CELL_GROUPS tag" );

    if( createGatherSet )
    {
        EntityHandle gather_set;
        rval = _readNC->readMeshIface->create_gather_set( gather_set );MB_CHK_SET_ERR( rval, "Failed to create gather set" );

        // Create gather set vertices
        EntityHandle start_gather_set_vertex;
        rval = create_gather_set_vertices( gather_set, start_gather_set_vertex );MB_CHK_SET_ERR( rval, "Failed to create gather set vertices for MPAS mesh" );

        // Create gather set edges (unless NO_EDGES read option is set)
        if( !noEdges )
        {
            rval = create_gather_set_edges( gather_set, start_gather_set_vertex );MB_CHK_SET_ERR( rval, "Failed to create gather set edges for MPAS mesh" );
        }

        // Create gather set cells, either unpadded or padded
        if( noMixedElements )
        {
            rval = create_padded_gather_set_cells( gather_set, start_gather_set_vertex );MB_CHK_SET_ERR( rval, "Failed to create padded gather set cells for MPAS mesh" );
        }
        else
        {
            rval = create_gather_set_cells( gather_set, start_gather_set_vertex );MB_CHK_SET_ERR( rval, "Failed to create gather set cells for MPAS mesh" );
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::read_ucd_variables_to_nonset_allocate( std::vector< ReadNC::VarData >& vdatas,
                                                               std::vector< int >& tstep_nums )
{
    Interface*& mbImpl          = _readNC->mbImpl;
    std::vector< int >& dimLens = _readNC->dimLens;
    bool& noEdges               = _readNC->noEdges;
    DebugOutput& dbgOut         = _readNC->dbgOut;

    Range* range = NULL;

    // Get vertices
    Range verts;
    ErrorCode rval = mbImpl->get_entities_by_dimension( _fileSet, 0, verts );MB_CHK_SET_ERR( rval, "Trouble getting vertices in current file set" );
    assert( "Should only have a single vertex subrange, since they were read in one shot" && verts.psize() == 1 );

    // Get edges
    Range edges;
    rval = mbImpl->get_entities_by_dimension( _fileSet, 1, edges );MB_CHK_SET_ERR( rval, "Trouble getting edges in current file set" );

    // Get faces
    Range faces;
    rval = mbImpl->get_entities_by_dimension( _fileSet, 2, faces );MB_CHK_SET_ERR( rval, "Trouble getting faces in current file set" );
    // Note, for MPAS faces.psize() can be more than 1

#ifdef MOAB_HAVE_MPI
    bool& isParallel = _readNC->isParallel;
    if( isParallel )
    {
        ParallelComm*& myPcomm = _readNC->myPcomm;
        rval                   = myPcomm->filter_pstatus( faces, PSTATUS_NOT_OWNED, PSTATUS_NOT, -1, &facesOwned );MB_CHK_SET_ERR( rval, "Trouble getting owned faces in current file set" );
    }
    else
        facesOwned = faces;  // not running in parallel, but still with MPI
#else
    facesOwned = faces;
#endif

    for( unsigned int i = 0; i < vdatas.size(); i++ )
    {
        // Skip edge variables, if specified by the read options
        if( noEdges && vdatas[i].entLoc == ReadNC::ENTLOCEDGE ) continue;

        // Support non-set variables with 3 dimensions like (Time, nCells, nVertLevels), or
        // 2 dimensions like (Time, nCells)
        assert( 3 == vdatas[i].varDims.size() || 2 == vdatas[i].varDims.size() );

        // For a non-set variable, time should be the first dimension
        assert( tDim == vdatas[i].varDims[0] );

        // Set up readStarts and readCounts
        vdatas[i].readStarts.resize( 3 );
        vdatas[i].readCounts.resize( 3 );

        // First: Time
        vdatas[i].readStarts[0] = 0;  // This value is timestep dependent, will be set later
        vdatas[i].readCounts[0] = 1;

        // Next: nVertices / nCells / nEdges
        switch( vdatas[i].entLoc )
        {
            case ReadNC::ENTLOCVERT:
                // Vertices
                // Start from the first localGidVerts
                // Actually, this will be reset later on in a loop
                vdatas[i].readStarts[1] = localGidVerts[0] - 1;
                vdatas[i].readCounts[1] = nLocalVertices;
                range                   = &verts;
                break;
            case ReadNC::ENTLOCFACE:
                // Faces
                // Start from the first localGidCells
                // Actually, this will be reset later on in a loop
                vdatas[i].readStarts[1] = localGidCells[0] - 1;
                vdatas[i].readCounts[1] = nLocalCells;
                range                   = &facesOwned;
                break;
            case ReadNC::ENTLOCEDGE:
                // Edges
                // Start from the first localGidEdges
                // Actually, this will be reset later on in a loop
                vdatas[i].readStarts[1] = localGidEdges[0] - 1;
                vdatas[i].readCounts[1] = nLocalEdges;
                range                   = &edges;
                break;
            default:
                MB_SET_ERR( MB_FAILURE, "Unexpected entity location type for variable " << vdatas[i].varName );
        }

        // Finally: nVertLevels or other optional levels, it is possible that there is no
        // level dimension (numLev is 0) for this non-set variable, e.g. (Time, nCells)
        if( vdatas[i].numLev < 1 ) vdatas[i].numLev = 1;
        vdatas[i].readStarts[2] = 0;
        vdatas[i].readCounts[2] = vdatas[i].numLev;

        // Get variable size
        vdatas[i].sz = 1;
        for( std::size_t idx = 0; idx != 3; idx++ )
            vdatas[i].sz *= vdatas[i].readCounts[idx];

        for( unsigned int t = 0; t < tstep_nums.size(); t++ )
        {
            dbgOut.tprintf( 2, "Reading variable %s, time step %d\n", vdatas[i].varName.c_str(), tstep_nums[t] );

            if( tstep_nums[t] >= dimLens[tDim] )
            {
                MB_SET_ERR( MB_INDEX_OUT_OF_RANGE, "Wrong value for timestep number " << tstep_nums[t] );
            }

            // Get the tag to read into
            if( !vdatas[i].varTags[t] )
            {
                rval = get_tag_to_nonset( vdatas[i], tstep_nums[t], vdatas[i].varTags[t], vdatas[i].numLev );MB_CHK_SET_ERR( rval, "Trouble getting tag for variable " << vdatas[i].varName );
            }

            // Get ptr to tag space
            if( vdatas[i].entLoc == ReadNC::ENTLOCFACE && numCellGroups > 1 )
            {
                // For a cell variable that is NOT on one contiguous chunk of faces, defer its tag
                // space allocation
                vdatas[i].varDatas[t] = NULL;
            }
            else
            {
                assert( 1 == range->psize() );
                void* data;
                int count;
                rval = mbImpl->tag_iterate( vdatas[i].varTags[t], range->begin(), range->end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate tag for variable " << vdatas[i].varName );
                assert( (unsigned)count == range->size() );
                vdatas[i].varDatas[t] = data;
            }
        }
    }

    return rval;
}

#ifdef MOAB_HAVE_PNETCDF
ErrorCode NCHelperMPAS::read_ucd_variables_to_nonset_async( std::vector< ReadNC::VarData >& vdatas,
                                                            std::vector< int >& tstep_nums )
{
    Interface*& mbImpl  = _readNC->mbImpl;
    bool& noEdges       = _readNC->noEdges;
    DebugOutput& dbgOut = _readNC->dbgOut;

    ErrorCode rval = read_ucd_variables_to_nonset_allocate( vdatas, tstep_nums );MB_CHK_SET_ERR( rval, "Trouble allocating space to read non-set variables" );

    // Finally, read into that space
    int success;
    Range* pLocalGid = NULL;

    for( unsigned int i = 0; i < vdatas.size(); i++ )
    {
        // Skip edge variables, if specified by the read options
        if( noEdges && vdatas[i].entLoc == ReadNC::ENTLOCEDGE ) continue;

        switch( vdatas[i].entLoc )
        {
            case ReadNC::ENTLOCVERT:
                pLocalGid = &localGidVerts;
                break;
            case ReadNC::ENTLOCFACE:
                pLocalGid = &localGidCells;
                break;
            case ReadNC::ENTLOCEDGE:
                pLocalGid = &localGidEdges;
                break;
            default:
                MB_SET_ERR( MB_FAILURE, "Unexpected entity location type for variable " << vdatas[i].varName );
        }

        std::size_t sz = vdatas[i].sz;

        for( unsigned int t = 0; t < tstep_nums.size(); t++ )
        {
            // We will synchronize all these reads with the other processors,
            // so the wait will be inside this double loop; is it too much?
            size_t nb_reads = pLocalGid->psize();
            std::vector< int > requests( nb_reads ), statuss( nb_reads );
            size_t idxReq = 0;

            // Set readStart for each timestep along time dimension
            vdatas[i].readStarts[0] = tstep_nums[t];

            switch( vdatas[i].varDataType )
            {
                case NC_FLOAT:
                case NC_DOUBLE: {
                    // Read float as double
                    std::vector< double > tmpdoubledata( sz );

                    // In the case of ucd mesh, and on multiple proc,
                    // we need to read as many times as subranges we have in the
                    // localGid range;
                    // basically, we have to give a different point
                    // for data to start, for every subrange :(
                    size_t indexInDoubleArray = 0;
                    size_t ic                 = 0;
                    for( Range::pair_iterator pair_iter = pLocalGid->pair_begin(); pair_iter != pLocalGid->pair_end();
                         ++pair_iter, ic++ )
                    {
                        EntityHandle starth     = pair_iter->first;
                        EntityHandle endh       = pair_iter->second;  // Inclusive
                        vdatas[i].readStarts[1] = (NCDF_SIZE)( starth - 1 );
                        vdatas[i].readCounts[1] = (NCDF_SIZE)( endh - starth + 1 );

                        // Do a partial read, in each subrange
                        // Wait outside this loop
                        success =
                            NCFUNCREQG( _vara_double )( _fileId, vdatas[i].varId, &( vdatas[i].readStarts[0] ),
                                                        &( vdatas[i].readCounts[0] ),
                                                        &( tmpdoubledata[indexInDoubleArray] ), &requests[idxReq++] );
                        if( success )
                            MB_SET_ERR( MB_FAILURE,
                                        "Failed to read double data in a loop for variable " << vdatas[i].varName );
                        // We need to increment the index in double array for the
                        // next subrange
                        indexInDoubleArray += ( endh - starth + 1 ) * 1 * vdatas[i].numLev;
                    }
                    assert( ic == pLocalGid->psize() );

                    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
                    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );

                    if( vdatas[i].entLoc == ReadNC::ENTLOCFACE && numCellGroups > 1 )
                    {
                        // For a cell variable that is NOT on one contiguous chunk of faces,
                        // allocate tag space for each cell group, and utilize cellHandleToGlobalID
                        // map to read tag data
                        Range::iterator iter = facesOwned.begin();
                        while( iter != facesOwned.end() )
                        {
                            int count;
                            void* ptr;
                            rval = mbImpl->tag_iterate( vdatas[i].varTags[t], iter, facesOwned.end(), count, ptr );MB_CHK_SET_ERR( rval, "Failed to iterate tag on owned faces" );

                            for( int j = 0; j < count; j++ )
                            {
                                int global_cell_idx =
                                    cellHandleToGlobalID[*( iter + j )];  // Global cell index, 1 based
                                int local_cell_idx =
                                    localGidCells.index( global_cell_idx );  // Local cell index, 0 based
                                assert( local_cell_idx != -1 );
                                for( int level = 0; level < vdatas[i].numLev; level++ )
                                    ( (double*)ptr )[j * vdatas[i].numLev + level] =
                                        tmpdoubledata[local_cell_idx * vdatas[i].numLev + level];
                            }

                            iter += count;
                        }
                    }
                    else
                    {
                        void* data = vdatas[i].varDatas[t];
                        for( std::size_t idx = 0; idx != tmpdoubledata.size(); idx++ )
                            ( (double*)data )[idx] = tmpdoubledata[idx];
                    }

                    break;
                }
                default:
                    MB_SET_ERR( MB_FAILURE, "Unexpected data type for variable " << vdatas[i].varName );
            }
        }
    }

    // Debug output, if requested
    if( 1 == dbgOut.get_verbosity() )
    {
        dbgOut.printf( 1, "Read variables: %s", vdatas.begin()->varName.c_str() );
        for( unsigned int i = 1; i < vdatas.size(); i++ )
            dbgOut.printf( 1, ", %s ", vdatas[i].varName.c_str() );
        dbgOut.tprintf( 1, "\n" );
    }

    return rval;
}
#else
ErrorCode NCHelperMPAS::read_ucd_variables_to_nonset( std::vector< ReadNC::VarData >& vdatas,
                                                      std::vector< int >& tstep_nums )
{
    Interface*& mbImpl = _readNC->mbImpl;
    bool& noEdges = _readNC->noEdges;
    DebugOutput& dbgOut = _readNC->dbgOut;

    ErrorCode rval = read_ucd_variables_to_nonset_allocate( vdatas, tstep_nums );MB_CHK_SET_ERR( rval, "Trouble allocating space to read non-set variables" );

    // Finally, read into that space
    int success;
    Range* pLocalGid = NULL;

    for( unsigned int i = 0; i < vdatas.size(); i++ )
    {
        // Skip edge variables, if specified by the read options
        if( noEdges && vdatas[i].entLoc == ReadNC::ENTLOCEDGE ) continue;

        switch( vdatas[i].entLoc )
        {
            case ReadNC::ENTLOCVERT:
                pLocalGid = &localGidVerts;
                break;
            case ReadNC::ENTLOCFACE:
                pLocalGid = &localGidCells;
                break;
            case ReadNC::ENTLOCEDGE:
                pLocalGid = &localGidEdges;
                break;
            default:
                MB_SET_ERR( MB_FAILURE, "Unexpected entity location type for variable " << vdatas[i].varName );
        }

        std::size_t sz = vdatas[i].sz;

        for( unsigned int t = 0; t < tstep_nums.size(); t++ )
        {
            // Set readStart for each timestep along time dimension
            vdatas[i].readStarts[0] = tstep_nums[t];

            switch( vdatas[i].varDataType )
            {
                case NC_FLOAT:
                case NC_DOUBLE: {
                    // Read float as double
                    std::vector< double > tmpdoubledata( sz );

                    // In the case of ucd mesh, and on multiple proc,
                    // we need to read as many times as subranges we have in the
                    // localGid range;
                    // basically, we have to give a different point
                    // for data to start, for every subrange :(
                    size_t indexInDoubleArray = 0;
                    size_t ic = 0;
                    for( Range::pair_iterator pair_iter = pLocalGid->pair_begin(); pair_iter != pLocalGid->pair_end();
                         ++pair_iter, ic++ )
                    {
                        EntityHandle starth = pair_iter->first;
                        EntityHandle endh = pair_iter->second;  // Inclusive
                        vdatas[i].readStarts[1] = (NCDF_SIZE)( starth - 1 );
                        vdatas[i].readCounts[1] = (NCDF_SIZE)( endh - starth + 1 );

                        success = NCFUNCAG( _vara_double )( _fileId, vdatas[i].varId, &( vdatas[i].readStarts[0] ),
                                                            &( vdatas[i].readCounts[0] ),
                                                            &( tmpdoubledata[indexInDoubleArray] ) );
                        if( success )
                            MB_SET_ERR( MB_FAILURE,
                                        "Failed to read double data in a loop for variable " << vdatas[i].varName );
                        // We need to increment the index in double array for the
                        // next subrange
                        indexInDoubleArray += ( endh - starth + 1 ) * 1 * vdatas[i].numLev;
                    }
                    assert( ic == pLocalGid->psize() );

                    if( vdatas[i].entLoc == ReadNC::ENTLOCFACE && numCellGroups > 1 )
                    {
                        // For a cell variable that is NOT on one contiguous chunk of faces,
                        // allocate tag space for each cell group, and utilize cellHandleToGlobalID
                        // map to read tag data
                        Range::iterator iter = facesOwned.begin();
                        while( iter != facesOwned.end() )
                        {
                            int count;
                            void* ptr;
                            rval = mbImpl->tag_iterate( vdatas[i].varTags[t], iter, facesOwned.end(), count, ptr );MB_CHK_SET_ERR( rval, "Failed to iterate tag on owned faces" );

                            for( int j = 0; j < count; j++ )
                            {
                                int global_cell_idx =
                                    cellHandleToGlobalID[*( iter + j )];  // Global cell index, 1 based
                                int local_cell_idx =
                                    localGidCells.index( global_cell_idx );  // Local cell index, 0 based
                                assert( local_cell_idx != -1 );
                                for( int level = 0; level < vdatas[i].numLev; level++ )
                                    ( (double*)ptr )[j * vdatas[i].numLev + level] =
                                        tmpdoubledata[local_cell_idx * vdatas[i].numLev + level];
                            }

                            iter += count;
                        }
                    }
                    else
                    {
                        void* data = vdatas[i].varDatas[t];
                        for( std::size_t idx = 0; idx != tmpdoubledata.size(); idx++ )
                            ( (double*)data )[idx] = tmpdoubledata[idx];
                    }

                    break;
                }
                default:
                    MB_SET_ERR( MB_FAILURE, "Unexpected data type for variable " << vdatas[i].varName );
            }
        }
    }

    // Debug output, if requested
    if( 1 == dbgOut.get_verbosity() )
    {
        dbgOut.printf( 1, "Read variables: %s", vdatas.begin()->varName.c_str() );
        for( unsigned int i = 1; i < vdatas.size(); i++ )
            dbgOut.printf( 1, ", %s ", vdatas[i].varName.c_str() );
        dbgOut.tprintf( 1, "\n" );
    }

    return rval;
}
#endif

#ifdef MOAB_HAVE_MPI
ErrorCode NCHelperMPAS::redistribute_local_cells( int start_cell_idx, ParallelComm * pco )
{
    // If possible, apply Zoltan partition
#ifdef MOAB_HAVE_ZOLTAN
    if( ScdParData::RCBZOLTAN == _readNC->partMethod )
    {
        // Read x coordinates of cell centers
        int xCellVarId;
        int success = NCFUNC( inq_varid )( _fileId, "xCell", &xCellVarId );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of xCell" );
        std::vector< double > xCell( nLocalCells );
        NCDF_SIZE read_start = static_cast< NCDF_SIZE >( start_cell_idx - 1 );
        NCDF_SIZE read_count = static_cast< NCDF_SIZE >( nLocalCells );
        success              = NCFUNCAG( _vara_double )( _fileId, xCellVarId, &read_start, &read_count, &xCell[0] );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read xCell data" );

        // Read y coordinates of cell centers
        int yCellVarId;
        success = NCFUNC( inq_varid )( _fileId, "yCell", &yCellVarId );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of yCell" );
        std::vector< double > yCell( nLocalCells );
        success = NCFUNCAG( _vara_double )( _fileId, yCellVarId, &read_start, &read_count, &yCell[0] );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read yCell data" );

        // Read z coordinates of cell centers
        int zCellVarId;
        success = NCFUNC( inq_varid )( _fileId, "zCell", &zCellVarId );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of zCell" );
        std::vector< double > zCell( nLocalCells );
        success = NCFUNCAG( _vara_double )( _fileId, zCellVarId, &read_start, &read_count, &zCell[0] );
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read zCell data" );

        // Zoltan partition using RCB; maybe more studies would be good, as to which partition
        // is better
        Interface*& mbImpl         = _readNC->mbImpl;
        DebugOutput& dbgOut        = _readNC->dbgOut;
        ZoltanPartitioner* mbZTool = new ZoltanPartitioner( mbImpl, pco, false, 0, NULL );
        ErrorCode rval             = mbZTool->repartition( xCell, yCell, zCell, start_cell_idx, "RCB", localGidCells );MB_CHK_SET_ERR( rval, "Error in Zoltan partitioning" );
        delete mbZTool;

        dbgOut.tprintf( 1, "After Zoltan partitioning, localGidCells.psize() = %d\n", (int)localGidCells.psize() );
        dbgOut.tprintf( 1, "                           localGidCells.size() = %d\n", (int)localGidCells.size() );

        // This is important: local cells are now redistributed, so nLocalCells might be different!
        nLocalCells = localGidCells.size();

        return MB_SUCCESS;
    }
#endif

    // By default, apply trivial partition
    localGidCells.insert( start_cell_idx, start_cell_idx + nLocalCells - 1 );

    return MB_SUCCESS;
}
#endif

ErrorCode NCHelperMPAS::create_local_vertices( const std::vector< int >& vertices_on_local_cells,
                                               EntityHandle& start_vertex )
{
    Interface*& mbImpl      = _readNC->mbImpl;
    Tag& mGlobalIdTag       = _readNC->mGlobalIdTag;
    const Tag*& mpFileIdTag = _readNC->mpFileIdTag;
    DebugOutput& dbgOut     = _readNC->dbgOut;

    // Make a copy of vertices_on_local_cells for sorting (keep original one to set cell
    // connectivity later)
    std::vector< int > vertices_on_local_cells_sorted( vertices_on_local_cells );
    std::sort( vertices_on_local_cells_sorted.begin(), vertices_on_local_cells_sorted.end() );
    std::copy( vertices_on_local_cells_sorted.rbegin(), vertices_on_local_cells_sorted.rend(),
               range_inserter( localGidVerts ) );
    nLocalVertices = localGidVerts.size();

    dbgOut.tprintf( 1, "   localGidVerts.psize() = %d\n", (int)localGidVerts.psize() );
    dbgOut.tprintf( 1, "   localGidVerts.size() = %d\n", (int)localGidVerts.size() );

    // Create local vertices
    std::vector< double* > arrays;
    ErrorCode rval =
        _readNC->readMeshIface->get_node_coords( 3, nLocalVertices, 0, start_vertex, arrays,
                                                 // Might have to create gather mesh later
                                                 ( createGatherSet ? nLocalVertices + nVertices : nLocalVertices ) );MB_CHK_SET_ERR( rval, "Failed to create local vertices" );

    // Add local vertices to current file set
    Range local_verts_range( start_vertex, start_vertex + nLocalVertices - 1 );
    rval = _readNC->mbImpl->add_entities( _fileSet, local_verts_range );MB_CHK_SET_ERR( rval, "Failed to add local vertices to current file set" );

    // Get ptr to GID memory for local vertices
    int count  = 0;
    void* data = NULL;
    rval       = mbImpl->tag_iterate( mGlobalIdTag, local_verts_range.begin(), local_verts_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate global id tag on local vertices" );
    assert( count == nLocalVertices );
    int* gid_data = (int*)data;
    std::copy( localGidVerts.begin(), localGidVerts.end(), gid_data );

    // Duplicate GID data, which will be used to resolve sharing
    if( mpFileIdTag )
    {
        rval = mbImpl->tag_iterate( *mpFileIdTag, local_verts_range.begin(), local_verts_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate file id tag on local vertices" );
        assert( count == nLocalVertices );
        int bytes_per_tag = 4;
        rval              = mbImpl->tag_get_bytes( *mpFileIdTag, bytes_per_tag );MB_CHK_SET_ERR( rval, "Can't get number of bytes for file id tag" );
        if( 4 == bytes_per_tag )
        {
            gid_data = (int*)data;
            std::copy( localGidVerts.begin(), localGidVerts.end(), gid_data );
        }
        else if( 8 == bytes_per_tag )
        {  // Should be a handle tag on 64 bit machine?
            long* handle_tag_data = (long*)data;
            std::copy( localGidVerts.begin(), localGidVerts.end(), handle_tag_data );
        }
    }

#ifdef MOAB_HAVE_PNETCDF
    size_t nb_reads = localGidVerts.psize();
    std::vector< int > requests( nb_reads );
    std::vector< int > statuss( nb_reads );
    size_t idxReq = 0;
#endif

    // Read x coordinates for local vertices
    double* xptr = arrays[0];
    int xVertexVarId;
    int success = NCFUNC( inq_varid )( _fileId, "xVertex", &xVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of xVertex" );
    size_t indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidVerts.pair_begin(); pair_iter != localGidVerts.pair_end();
         ++pair_iter )
    {
        EntityHandle starth  = pair_iter->first;
        EntityHandle endh    = pair_iter->second;
        NCDF_SIZE read_start = (NCDF_SIZE)( starth - 1 );
        NCDF_SIZE read_count = (NCDF_SIZE)( endh - starth + 1 );

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_double )( _fileId, xVertexVarId, &read_start, &read_count, &( xptr[indexInArray] ),
                                              &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_double )( _fileId, xVertexVarId, &read_start, &read_count, &( xptr[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read xVertex data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 );
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Read y coordinates for local vertices
    double* yptr = arrays[1];
    int yVertexVarId;
    success = NCFUNC( inq_varid )( _fileId, "yVertex", &yVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of yVertex" );
#ifdef MOAB_HAVE_PNETCDF
    idxReq = 0;
#endif
    indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidVerts.pair_begin(); pair_iter != localGidVerts.pair_end();
         ++pair_iter )
    {
        EntityHandle starth  = pair_iter->first;
        EntityHandle endh    = pair_iter->second;
        NCDF_SIZE read_start = (NCDF_SIZE)( starth - 1 );
        NCDF_SIZE read_count = (NCDF_SIZE)( endh - starth + 1 );

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_double )( _fileId, yVertexVarId, &read_start, &read_count, &( yptr[indexInArray] ),
                                              &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_double )( _fileId, yVertexVarId, &read_start, &read_count, &( yptr[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read yVertex data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 );
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Read z coordinates for local vertices
    double* zptr = arrays[2];
    int zVertexVarId;
    success = NCFUNC( inq_varid )( _fileId, "zVertex", &zVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of zVertex" );
#ifdef MOAB_HAVE_PNETCDF
    idxReq = 0;
#endif
    indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidVerts.pair_begin(); pair_iter != localGidVerts.pair_end();
         ++pair_iter )
    {
        EntityHandle starth  = pair_iter->first;
        EntityHandle endh    = pair_iter->second;
        NCDF_SIZE read_start = (NCDF_SIZE)( starth - 1 );
        NCDF_SIZE read_count = (NCDF_SIZE)( endh - starth + 1 );

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_double )( _fileId, zVertexVarId, &read_start, &read_count, &( zptr[indexInArray] ),
                                              &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_double )( _fileId, zVertexVarId, &read_start, &read_count, &( zptr[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read zVertex data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 );
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_local_edges( EntityHandle start_vertex,
                                            const std::vector< int >& num_edges_on_local_cells )
{
    Interface*& mbImpl  = _readNC->mbImpl;
    Tag& mGlobalIdTag   = _readNC->mGlobalIdTag;
    DebugOutput& dbgOut = _readNC->dbgOut;

    // Read edges on each local cell, to get localGidEdges
    int edgesOnCellVarId;
    int success = NCFUNC( inq_varid )( _fileId, "edgesOnCell", &edgesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of edgesOnCell" );

    std::vector< int > edges_on_local_cells( nLocalCells * maxEdgesPerCell );
    dbgOut.tprintf( 1, "   edges_on_local_cells.size() = %d\n", (int)edges_on_local_cells.size() );

#ifdef MOAB_HAVE_PNETCDF
    size_t nb_reads = localGidCells.psize();
    std::vector< int > requests( nb_reads );
    std::vector< int > statuss( nb_reads );
    size_t idxReq = 0;
#endif
    size_t indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidCells.pair_begin(); pair_iter != localGidCells.pair_end();
         ++pair_iter )
    {
        EntityHandle starth      = pair_iter->first;
        EntityHandle endh        = pair_iter->second;
        NCDF_SIZE read_starts[2] = { static_cast< NCDF_SIZE >( starth - 1 ), 0 };
        NCDF_SIZE read_counts[2] = { static_cast< NCDF_SIZE >( endh - starth + 1 ),
                                     static_cast< NCDF_SIZE >( maxEdgesPerCell ) };

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_int )( _fileId, edgesOnCellVarId, read_starts, read_counts,
                                           &( edges_on_local_cells[indexInArray] ), &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_int )( _fileId, edgesOnCellVarId, read_starts, read_counts,
                                         &( edges_on_local_cells[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read edgesOnCell data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 ) * maxEdgesPerCell;
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Correct local cell edges array in the same way as local cell vertices array, replace the
    // padded edges with the last edges in the corresponding cells
    for( int local_cell_idx = 0; local_cell_idx < nLocalCells; local_cell_idx++ )
    {
        int num_edges             = num_edges_on_local_cells[local_cell_idx];
        int idx_in_local_edge_arr = local_cell_idx * maxEdgesPerCell;
        int last_edge_idx         = edges_on_local_cells[idx_in_local_edge_arr + num_edges - 1];
        for( int i = num_edges; i < maxEdgesPerCell; i++ )
            edges_on_local_cells[idx_in_local_edge_arr + i] = last_edge_idx;
    }

    // Collect local edges
    std::sort( edges_on_local_cells.begin(), edges_on_local_cells.end() );
    std::copy( edges_on_local_cells.rbegin(), edges_on_local_cells.rend(), range_inserter( localGidEdges ) );
    nLocalEdges = localGidEdges.size();

    dbgOut.tprintf( 1, "   localGidEdges.psize() = %d\n", (int)localGidEdges.psize() );
    dbgOut.tprintf( 1, "   localGidEdges.size() = %d\n", (int)localGidEdges.size() );

    // Create local edges
    EntityHandle start_edge;
    EntityHandle* conn_arr_edges = NULL;
    ErrorCode rval =
        _readNC->readMeshIface->get_element_connect( nLocalEdges, 2, MBEDGE, 0, start_edge, conn_arr_edges,
                                                     // Might have to create gather mesh later
                                                     ( createGatherSet ? nLocalEdges + nEdges : nLocalEdges ) );MB_CHK_SET_ERR( rval, "Failed to create local edges" );

    // Add local edges to current file set
    Range local_edges_range( start_edge, start_edge + nLocalEdges - 1 );
    rval = _readNC->mbImpl->add_entities( _fileSet, local_edges_range );MB_CHK_SET_ERR( rval, "Failed to add local edges to current file set" );

    // Get ptr to GID memory for edges
    int count  = 0;
    void* data = NULL;
    rval       = mbImpl->tag_iterate( mGlobalIdTag, local_edges_range.begin(), local_edges_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate global id tag on local edges" );
    assert( count == nLocalEdges );
    int* gid_data = (int*)data;
    std::copy( localGidEdges.begin(), localGidEdges.end(), gid_data );

    int verticesOnEdgeVarId;

    // Read vertices on each local edge, to get edge connectivity
    success = NCFUNC( inq_varid )( _fileId, "verticesOnEdge", &verticesOnEdgeVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of verticesOnEdge" );
    // Utilize the memory storage pointed by conn_arr_edges
    int* vertices_on_local_edges = (int*)conn_arr_edges;
#ifdef MOAB_HAVE_PNETCDF
    nb_reads = localGidEdges.psize();
    requests.resize( nb_reads );
    statuss.resize( nb_reads );
    idxReq = 0;
#endif
    indexInArray = 0;
    for( Range::pair_iterator pair_iter = localGidEdges.pair_begin(); pair_iter != localGidEdges.pair_end();
         ++pair_iter )
    {
        EntityHandle starth      = pair_iter->first;
        EntityHandle endh        = pair_iter->second;
        NCDF_SIZE read_starts[2] = { static_cast< NCDF_SIZE >( starth - 1 ), 0 };
        NCDF_SIZE read_counts[2] = { static_cast< NCDF_SIZE >( endh - starth + 1 ), 2 };

        // Do a partial read in each subrange
#ifdef MOAB_HAVE_PNETCDF
        success = NCFUNCREQG( _vara_int )( _fileId, verticesOnEdgeVarId, read_starts, read_counts,
                                           &( vertices_on_local_edges[indexInArray] ), &requests[idxReq++] );
#else
        success = NCFUNCAG( _vara_int )( _fileId, verticesOnEdgeVarId, read_starts, read_counts,
                                         &( vertices_on_local_edges[indexInArray] ) );
#endif
        if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnEdge data in a loop" );

        // Increment the index for next subrange
        indexInArray += ( endh - starth + 1 ) * 2;
    }

#ifdef MOAB_HAVE_PNETCDF
    // Wait outside the loop
    success = NCFUNC( wait_all )( _fileId, requests.size(), &requests[0], &statuss[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed on wait_all" );
#endif

    // Populate connectivity data for local edges
    // Convert in-place from int (stored in the first half) to EntityHandle
    // Reading backward is the trick
    for( int edge_vert = nLocalEdges * 2 - 1; edge_vert >= 0; edge_vert-- )
    {
        int global_vert_idx = vertices_on_local_edges[edge_vert];      // Global vertex index, 1 based
        int local_vert_idx  = localGidVerts.index( global_vert_idx );  // Local vertex index, 0 based
        assert( local_vert_idx != -1 );
        conn_arr_edges[edge_vert] = start_vertex + local_vert_idx;
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_local_cells( const std::vector< int >& vertices_on_local_cells,
                                            const std::vector< int >& num_edges_on_local_cells,
                                            EntityHandle start_vertex,
                                            Range& faces )
{
    Interface*& mbImpl = _readNC->mbImpl;
    Tag& mGlobalIdTag  = _readNC->mGlobalIdTag;

    // Divide local cells into groups based on the number of edges
    Range local_cells_with_n_edges[DEFAULT_MAX_EDGES_PER_CELL + 1];
    // Insert larger values before smaller ones to increase efficiency
    for( int i = nLocalCells - 1; i >= 0; i-- )
    {
        int num_edges = num_edges_on_local_cells[i];
        local_cells_with_n_edges[num_edges].insert( localGidCells[i] );  // Global cell index
    }

    std::vector< int > num_edges_on_cell_groups;
    for( int i = 3; i <= maxEdgesPerCell; i++ )
    {
        if( local_cells_with_n_edges[i].size() > 0 ) num_edges_on_cell_groups.push_back( i );
    }
    numCellGroups = num_edges_on_cell_groups.size();

    EntityHandle* conn_arr_local_cells_with_n_edges[DEFAULT_MAX_EDGES_PER_CELL + 1];
    for( int i = 0; i < numCellGroups; i++ )
    {
        int num_edges_per_cell = num_edges_on_cell_groups[i];
        int num_group_cells    = (int)local_cells_with_n_edges[num_edges_per_cell].size();

        // Create local cells for each non-empty cell group
        EntityHandle start_element;
        ErrorCode rval = _readNC->readMeshIface->get_element_connect(
            num_group_cells, num_edges_per_cell, MBPOLYGON, 0, start_element,
            conn_arr_local_cells_with_n_edges[num_edges_per_cell], num_group_cells );MB_CHK_SET_ERR( rval, "Failed to create local cells" );
        faces.insert( start_element, start_element + num_group_cells - 1 );

        // Add local cells to current file set
        Range local_cells_range( start_element, start_element + num_group_cells - 1 );
        rval = _readNC->mbImpl->add_entities( _fileSet, local_cells_range );MB_CHK_SET_ERR( rval, "Failed to add local cells to current file set" );

        // Get ptr to gid memory for local cells
        int count  = 0;
        void* data = NULL;
        rval = mbImpl->tag_iterate( mGlobalIdTag, local_cells_range.begin(), local_cells_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate global id tag on local cells" );
        assert( count == num_group_cells );
        int* gid_data = (int*)data;
        std::copy( local_cells_with_n_edges[num_edges_per_cell].begin(),
                   local_cells_with_n_edges[num_edges_per_cell].end(), gid_data );

        // Set connectivity array with proper local vertices handles
        for( int j = 0; j < num_group_cells; j++ )
        {
            EntityHandle global_cell_idx =
                local_cells_with_n_edges[num_edges_per_cell][j];          // Global cell index, 1 based
            int local_cell_idx = localGidCells.index( global_cell_idx );  // Local cell index, 0 based
            assert( local_cell_idx != -1 );

            if( numCellGroups > 1 )
            {
                // Populate cellHandleToGlobalID map to read cell variables
                cellHandleToGlobalID[start_element + j] = global_cell_idx;
            }

            for( int k = 0; k < num_edges_per_cell; k++ )
            {
                EntityHandle global_vert_idx =
                    vertices_on_local_cells[local_cell_idx * maxEdgesPerCell + k];  // Global vertex index, 1 based
                int local_vert_idx = localGidVerts.index( global_vert_idx );        // Local vertex index, 0 based
                assert( local_vert_idx != -1 );
                conn_arr_local_cells_with_n_edges[num_edges_per_cell][j * num_edges_per_cell + k] =
                    start_vertex + local_vert_idx;
            }
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_padded_local_cells( const std::vector< int >& vertices_on_local_cells,
                                                   EntityHandle start_vertex,
                                                   Range& faces )
{
    Interface*& mbImpl = _readNC->mbImpl;
    Tag& mGlobalIdTag  = _readNC->mGlobalIdTag;

    // Only one group of cells (each cell is represented by a polygon with maxEdgesPerCell edges)
    numCellGroups = 1;

    // Create cells for this cell group
    EntityHandle start_element;
    EntityHandle* conn_arr_local_cells = NULL;
    ErrorCode rval =
        _readNC->readMeshIface->get_element_connect( nLocalCells, maxEdgesPerCell, MBPOLYGON, 0, start_element,
                                                     conn_arr_local_cells,
                                                     // Might have to create gather mesh later
                                                     ( createGatherSet ? nLocalCells + nCells : nLocalCells ) );MB_CHK_SET_ERR( rval, "Failed to create local cells" );
    faces.insert( start_element, start_element + nLocalCells - 1 );

    // Add local cells to current file set
    Range local_cells_range( start_element, start_element + nLocalCells - 1 );
    rval = _readNC->mbImpl->add_entities( _fileSet, local_cells_range );MB_CHK_SET_ERR( rval, "Failed to add local cells to current file set" );

    // Get ptr to GID memory for local cells
    int count  = 0;
    void* data = NULL;
    rval       = mbImpl->tag_iterate( mGlobalIdTag, local_cells_range.begin(), local_cells_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate global id tag on local cells" );
    assert( count == nLocalCells );
    int* gid_data = (int*)data;
    std::copy( localGidCells.begin(), localGidCells.end(), gid_data );

    // Set connectivity array with proper local vertices handles
    // vertices_on_local_cells array was already corrected to have the last vertices padded
    // no need for extra checks considering
    for( int local_cell_idx = 0; local_cell_idx < nLocalCells; local_cell_idx++ )
    {
        for( int i = 0; i < maxEdgesPerCell; i++ )
        {
            EntityHandle global_vert_idx =
                vertices_on_local_cells[local_cell_idx * maxEdgesPerCell + i];  // Global vertex index, 1 based
            int local_vert_idx = localGidVerts.index( global_vert_idx );        // Local vertex index, 0 based
            assert( local_vert_idx != -1 );
            conn_arr_local_cells[local_cell_idx * maxEdgesPerCell + i] = start_vertex + local_vert_idx;
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_gather_set_vertices( EntityHandle gather_set, EntityHandle& gather_set_start_vertex )
{
    Interface*& mbImpl      = _readNC->mbImpl;
    Tag& mGlobalIdTag       = _readNC->mGlobalIdTag;
    const Tag*& mpFileIdTag = _readNC->mpFileIdTag;

    // Create gather set vertices
    std::vector< double* > arrays;
    // Don't need to specify allocation number here, because we know enough vertices were created
    // before
    ErrorCode rval = _readNC->readMeshIface->get_node_coords( 3, nVertices, 0, gather_set_start_vertex, arrays );MB_CHK_SET_ERR( rval, "Failed to create gather set vertices" );

    // Add vertices to the gather set
    Range gather_set_verts_range( gather_set_start_vertex, gather_set_start_vertex + nVertices - 1 );
    rval = mbImpl->add_entities( gather_set, gather_set_verts_range );MB_CHK_SET_ERR( rval, "Failed to add vertices to the gather set" );

    // Read x coordinates for gather set vertices
    double* xptr = arrays[0];
    int xVertexVarId;
    int success = NCFUNC( inq_varid )( _fileId, "xVertex", &xVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of xVertex" );
    NCDF_SIZE read_start = 0;
    NCDF_SIZE read_count = static_cast< NCDF_SIZE >( nVertices );
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success = NCFUNCG( _vara_double )( _fileId, xVertexVarId, &read_start, &read_count, xptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read xVertex data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success = NCFUNCG( _vara_double )( _fileId, xVertexVarId, &read_start, &read_count, xptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read xVertex data" );
#endif

    // Read y coordinates for gather set vertices
    double* yptr = arrays[1];
    int yVertexVarId;
    success = NCFUNC( inq_varid )( _fileId, "yVertex", &yVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of yVertex" );
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success = NCFUNCG( _vara_double )( _fileId, yVertexVarId, &read_start, &read_count, yptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read yVertex data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success = NCFUNCG( _vara_double )( _fileId, yVertexVarId, &read_start, &read_count, yptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read yVertex data" );
#endif

    // Read z coordinates for gather set vertices
    double* zptr = arrays[2];
    int zVertexVarId;
    success = NCFUNC( inq_varid )( _fileId, "zVertex", &zVertexVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of zVertex" );
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success = NCFUNCG( _vara_double )( _fileId, zVertexVarId, &read_start, &read_count, zptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read zVertex data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success = NCFUNCG( _vara_double )( _fileId, zVertexVarId, &read_start, &read_count, zptr );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read zVertex data" );
#endif

    // Get ptr to GID memory for gather set vertices
    int count  = 0;
    void* data = NULL;
    rval =
        mbImpl->tag_iterate( mGlobalIdTag, gather_set_verts_range.begin(), gather_set_verts_range.end(), count, data );MB_CHK_SET_ERR( rval, "Failed to iterate global id tag on gather set vertices" );
    assert( count == nVertices );
    int* gid_data = (int*)data;
    for( int j = 1; j <= nVertices; j++ )
        gid_data[j - 1] = j;

    // Set the file id tag too, it should be bigger something not interfering with global id
    if( mpFileIdTag )
    {
        rval = mbImpl->tag_iterate( *mpFileIdTag, gather_set_verts_range.begin(), gather_set_verts_range.end(), count,
                                    data );MB_CHK_SET_ERR( rval, "Failed to iterate file id tag on gather set vertices" );
        assert( count == nVertices );
        int bytes_per_tag = 4;
        rval              = mbImpl->tag_get_bytes( *mpFileIdTag, bytes_per_tag );MB_CHK_SET_ERR( rval, "Can't get number of bytes for file id tag" );
        if( 4 == bytes_per_tag )
        {
            gid_data = (int*)data;
            for( int j = 1; j <= nVertices; j++ )
                gid_data[j - 1] = nVertices + j;  // Bigger than global id tag
        }
        else if( 8 == bytes_per_tag )
        {  // Should be a handle tag on 64 bit machine?
            long* handle_tag_data = (long*)data;
            for( int j = 1; j <= nVertices; j++ )
                handle_tag_data[j - 1] = nVertices + j;  // Bigger than global id tag
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_gather_set_edges( EntityHandle gather_set, EntityHandle gather_set_start_vertex )
{
    Interface*& mbImpl = _readNC->mbImpl;

    // Create gather set edges
    EntityHandle start_edge;
    EntityHandle* conn_arr_gather_set_edges = NULL;
    // Don't need to specify allocation number here, because we know enough edges were created
    // before
    ErrorCode rval =
        _readNC->readMeshIface->get_element_connect( nEdges, 2, MBEDGE, 0, start_edge, conn_arr_gather_set_edges );MB_CHK_SET_ERR( rval, "Failed to create gather set edges" );

    // Add edges to the gather set
    Range gather_set_edges_range( start_edge, start_edge + nEdges - 1 );
    rval = mbImpl->add_entities( gather_set, gather_set_edges_range );MB_CHK_SET_ERR( rval, "Failed to add edges to the gather set" );

    // Read vertices on each edge
    int verticesOnEdgeVarId;
    int success = NCFUNC( inq_varid )( _fileId, "verticesOnEdge", &verticesOnEdgeVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of verticesOnEdge" );
    // Utilize the memory storage pointed by conn_arr_gather_set_edges
    int* vertices_on_gather_set_edges = (int*)conn_arr_gather_set_edges;
    NCDF_SIZE read_starts[2]          = { 0, 0 };
    NCDF_SIZE read_counts[2]          = { static_cast< NCDF_SIZE >( nEdges ), 2 };
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success =
        NCFUNCG( _vara_int )( _fileId, verticesOnEdgeVarId, read_starts, read_counts, vertices_on_gather_set_edges );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnEdge data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success =
        NCFUNCG( _vara_int )( _fileId, verticesOnEdgeVarId, read_starts, read_counts, vertices_on_gather_set_edges );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnEdge data" );
#endif

    // Populate connectivity data for gather set edges
    // Convert in-place from int (stored in the first half) to EntityHandle
    // Reading backward is the trick
    for( int edge_vert = nEdges * 2 - 1; edge_vert >= 0; edge_vert-- )
    {
        int gather_set_vert_idx = vertices_on_gather_set_edges[edge_vert];  // Global vertex index, 1 based
        gather_set_vert_idx--;                                              // 1 based -> 0 based
        // Connectivity array is shifted by where the gather set vertices start
        conn_arr_gather_set_edges[edge_vert] = gather_set_start_vertex + gather_set_vert_idx;
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_gather_set_cells( EntityHandle gather_set, EntityHandle gather_set_start_vertex )
{
    Interface*& mbImpl = _readNC->mbImpl;

    // Read number of edges on each gather set cell
    int nEdgesOnCellVarId;
    int success = NCFUNC( inq_varid )( _fileId, "nEdgesOnCell", &nEdgesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of nEdgesOnCell" );
    std::vector< int > num_edges_on_gather_set_cells( nCells );
    NCDF_SIZE read_start = 0;
    NCDF_SIZE read_count = static_cast< NCDF_SIZE >( nCells );
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success =
        NCFUNCG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count, &num_edges_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read nEdgesOnCell data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success =
        NCFUNCG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count, &num_edges_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read nEdgesOnCell data" );
#endif

    // Read vertices on each gather set cell (connectivity)
    int verticesOnCellVarId;
    success = NCFUNC( inq_varid )( _fileId, "verticesOnCell", &verticesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of verticesOnCell" );
    std::vector< int > vertices_on_gather_set_cells( nCells * maxEdgesPerCell );
    NCDF_SIZE read_starts[2] = { 0, 0 };
    NCDF_SIZE read_counts[2] = { static_cast< NCDF_SIZE >( nCells ), static_cast< NCDF_SIZE >( maxEdgesPerCell ) };
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success = NCFUNCG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts,
                                    &vertices_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnCell data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success = NCFUNCG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts,
                                    &vertices_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnCell data" );
#endif

    // Divide gather set cells into groups based on the number of edges
    Range gather_set_cells_with_n_edges[DEFAULT_MAX_EDGES_PER_CELL + 1];
    // Insert larger values before smaller values to increase efficiency
    for( int i = nCells - 1; i >= 0; i-- )
    {
        int num_edges = num_edges_on_gather_set_cells[i];
        gather_set_cells_with_n_edges[num_edges].insert( i + 1 );  // 0 based -> 1 based
    }

    // Create gather set cells
    EntityHandle* conn_arr_gather_set_cells_with_n_edges[DEFAULT_MAX_EDGES_PER_CELL + 1];
    for( int num_edges_per_cell = 3; num_edges_per_cell <= maxEdgesPerCell; num_edges_per_cell++ )
    {
        int num_group_cells = gather_set_cells_with_n_edges[num_edges_per_cell].size();
        if( num_group_cells > 0 )
        {
            EntityHandle start_element;
            ErrorCode rval = _readNC->readMeshIface->get_element_connect(
                num_group_cells, num_edges_per_cell, MBPOLYGON, 0, start_element,
                conn_arr_gather_set_cells_with_n_edges[num_edges_per_cell], num_group_cells );MB_CHK_SET_ERR( rval, "Failed to create gather set cells" );

            // Add cells to the gather set
            Range gather_set_cells_range( start_element, start_element + num_group_cells - 1 );
            rval = mbImpl->add_entities( gather_set, gather_set_cells_range );MB_CHK_SET_ERR( rval, "Failed to add cells to the gather set" );

            for( int j = 0; j < num_group_cells; j++ )
            {
                int gather_set_cell_idx =
                    gather_set_cells_with_n_edges[num_edges_per_cell][j];  // Global cell index, 1 based
                gather_set_cell_idx--;                                     // 1 based -> 0 based

                for( int k = 0; k < num_edges_per_cell; k++ )
                {
                    EntityHandle gather_set_vert_idx =
                        vertices_on_gather_set_cells[gather_set_cell_idx * maxEdgesPerCell +
                                                     k];  // Global vertex index, 1 based
                    gather_set_vert_idx--;                // 1 based -> 0 based

                    // Connectivity array is shifted by where the gather set vertices start
                    conn_arr_gather_set_cells_with_n_edges[num_edges_per_cell][j * num_edges_per_cell + k] =
                        gather_set_start_vertex + gather_set_vert_idx;
                }
            }
        }
    }

    return MB_SUCCESS;
}

ErrorCode NCHelperMPAS::create_padded_gather_set_cells( EntityHandle gather_set, EntityHandle gather_set_start_vertex )
{
    Interface*& mbImpl = _readNC->mbImpl;

    // Read number of edges on each gather set cell
    int nEdgesOnCellVarId;
    int success = NCFUNC( inq_varid )( _fileId, "nEdgesOnCell", &nEdgesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of nEdgesOnCell" );
    std::vector< int > num_edges_on_gather_set_cells( nCells );
    NCDF_SIZE read_start = 0;
    NCDF_SIZE read_count = static_cast< NCDF_SIZE >( nCells );
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success =
        NCFUNCG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count, &num_edges_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read nEdgesOnCell data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success =
        NCFUNCG( _vara_int )( _fileId, nEdgesOnCellVarId, &read_start, &read_count, &num_edges_on_gather_set_cells[0] );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read nEdgesOnCell data" );
#endif

    // Create gather set cells
    EntityHandle start_element;
    EntityHandle* conn_arr_gather_set_cells = NULL;
    // Don't need to specify allocation number here, because we know enough cells were created
    // before
    ErrorCode rval = _readNC->readMeshIface->get_element_connect( nCells, maxEdgesPerCell, MBPOLYGON, 0, start_element,
                                                                  conn_arr_gather_set_cells );MB_CHK_SET_ERR( rval, "Failed to create gather set cells" );

    // Add cells to the gather set
    Range gather_set_cells_range( start_element, start_element + nCells - 1 );
    rval = mbImpl->add_entities( gather_set, gather_set_cells_range );MB_CHK_SET_ERR( rval, "Failed to add cells to the gather set" );

    // Read vertices on each gather set cell (connectivity)
    int verticesOnCellVarId;
    success = NCFUNC( inq_varid )( _fileId, "verticesOnCell", &verticesOnCellVarId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to get variable id of verticesOnCell" );
    // Utilize the memory storage pointed by conn_arr_gather_set_cells
    int* vertices_on_gather_set_cells = (int*)conn_arr_gather_set_cells;
    NCDF_SIZE read_starts[2]          = { 0, 0 };
    NCDF_SIZE read_counts[2] = { static_cast< NCDF_SIZE >( nCells ), static_cast< NCDF_SIZE >( maxEdgesPerCell ) };
#ifdef MOAB_HAVE_PNETCDF
    // Enter independent I/O mode, since this read is only for the gather processor
    success = NCFUNC( begin_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to begin independent I/O mode" );
    success =
        NCFUNCG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts, vertices_on_gather_set_cells );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnCell data" );
    success = NCFUNC( end_indep_data )( _fileId );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to end independent I/O mode" );
#else
    success =
        NCFUNCG( _vara_int )( _fileId, verticesOnCellVarId, read_starts, read_counts, vertices_on_gather_set_cells );
    if( success ) MB_SET_ERR( MB_FAILURE, "Failed to read verticesOnCell data" );
#endif

    // Correct gather set cell vertices array in the same way as local cell vertices array,
    // replace the padded vertices with the last vertices in the corresponding cells
    for( int gather_set_cell_idx = 0; gather_set_cell_idx < nCells; gather_set_cell_idx++ )
    {
        int num_edges                  = num_edges_on_gather_set_cells[gather_set_cell_idx];
        int idx_in_gather_set_vert_arr = gather_set_cell_idx * maxEdgesPerCell;
        int last_vert_idx              = vertices_on_gather_set_cells[idx_in_gather_set_vert_arr + num_edges - 1];
        for( int i = num_edges; i < maxEdgesPerCell; i++ )
            vertices_on_gather_set_cells[idx_in_gather_set_vert_arr + i] = last_vert_idx;
    }

    // Populate connectivity data for gather set cells
    // Convert in-place from int (stored in the first half) to EntityHandle
    // Reading backward is the trick
    for( int cell_vert = nCells * maxEdgesPerCell - 1; cell_vert >= 0; cell_vert-- )
    {
        int gather_set_vert_idx = vertices_on_gather_set_cells[cell_vert];  // Global vertex index, 1 based
        gather_set_vert_idx--;                                              // 1 based -> 0 based
        // Connectivity array is shifted by where the gather set vertices start
        conn_arr_gather_set_cells[cell_vert] = gather_set_start_vertex + gather_set_vert_idx;
    }

    return MB_SUCCESS;
}

}  // namespace moab