MOAB: Mesh Oriented datABase
(version 5.4.1)
|
#include "moab/ParallelComm.hpp"
#include "moab/Core.hpp"
#include "moab_mpi.h"
#include "moab/ParallelMergeMesh.hpp"
#include "MBParallelConventions.h"
#include "TestUtil.hpp"
#include <iostream>
Go to the source code of this file.
Functions | |
int | main (int argc, char *argv[]) |
int main | ( | int | argc, |
char * | argv[] | ||
) |
Definition at line 11 of file parmerge_test.cpp.
References ErrorCode, moab::ParallelComm::filter_pstatus(), moab::Core::get_coords(), moab::Core::get_entities_by_type(), moab::ParallelComm::get_shared_entities(), moab::Core::load_file(), mb, MB_CHK_ERR, MB_SUCCESS, MBVERTEX, moab::ParallelMergeMesh::merge(), MPI_COMM_WORLD, moab::Range::num_of_dimension(), PSTATUS_NOT, PSTATUS_NOT_OWNED, rank, moab::Core::set_coords(), moab::Range::size(), and moab::Core::write_file().
{ MPI_Init( &argc, &argv ); int nproc, rank; MPI_Comm_size( MPI_COMM_WORLD, &nproc ); MPI_Comm_rank( MPI_COMM_WORLD, &rank ); std::string filename0 = TestDir + "unittest/brick1.vtk"; std::string filename1 = TestDir + "unittest/brick2.vtk"; moab::Core* mb = new moab::Core(); moab::ParallelComm* pc = new moab::ParallelComm( mb, MPI_COMM_WORLD ); ErrorCode rval = MB_SUCCESS; if( 0 == rank%2 ) rval = mb->load_file( filename0.c_str() ); else rval = mb->load_file( filename1.c_str() ); if( rval != MB_SUCCESS ) { std::cout << "fail to load file\n"; delete pc; delete mb; MPI_Finalize(); return 1; } // if rank > 2, translate in z direction, with the distance ( (rank)/2 ) * 10 if (rank >= 2) { Range verts; rval = mb->get_entities_by_type(0, MBVERTEX, verts);MB_CHK_ERR( rval ); int num_verts=(int)verts.size(); std::vector<double> coords; coords.resize(num_verts*3); rval = mb->get_coords(verts, &coords[0]);MB_CHK_ERR( rval ); int steps=rank/2; double z_translate = steps*10.;// the 2 bricks are size 10 for (int i=0; i<num_verts; i++) coords[3*i+2] += z_translate; rval = mb->set_coords(verts, &coords[0]);MB_CHK_ERR( rval ); } ParallelMergeMesh pm( pc, 0.001 ); rval = pm.merge(); if( rval != MB_SUCCESS ) { std::cout << "fail to merge in parallel \n"; delete pc; delete mb; MPI_Finalize(); return 1; } if (nproc == 2) { // check number of shared entities Range shared_ents; // Get entities shared with all other processors rval = pc->get_shared_entities( -1, shared_ents ); if( rval != MB_SUCCESS ) { delete pc; delete mb; MPI_Finalize(); return 1; } // there should be exactly 9 vertices, 12 edges, 4 faces unsigned numV = shared_ents.num_of_dimension( 0 ); unsigned numE = shared_ents.num_of_dimension( 1 ); unsigned numF = shared_ents.num_of_dimension( 2 ); if( numV != 9 || numE != 12 || numF != 4 ) { std::cout << " wrong number of shared entities on proc " << rank << " v:" << numV << " e:" << numE << " f:" << numF << "\n"; delete pc; delete mb; MPI_Finalize(); return 1; } } Range verts, verts_owned; rval = mb->get_entities_by_type( 0, MBVERTEX, verts );MB_CHK_ERR( rval ); // Get local owned vertices rval = pc->filter_pstatus( verts, PSTATUS_NOT_OWNED, PSTATUS_NOT, -1, &verts_owned );MB_CHK_ERR( rval ); int num_owned_verts = (int) verts_owned.size(); int num_total_verts = 0; MPI_Reduce(&num_owned_verts, &num_total_verts, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD); if (0==rank) std::cout << "total vertex number: " << num_total_verts << "\n"; // each brick has 27 vertices; // first 2 share 9, total number of vertices should be 27*2 -9 = 45 // for each additional row of 2 bricks, we will have 3*5* 2 = 30 more vertices int correct = 45 + (nproc/2 - 1) * 30; if (nproc%2 == 1) correct += 18; // odd cases have extra 18 nodes (3 * 3 * 2) if (nproc >= 2 && 0==rank) { if (correct != num_total_verts) { std::cout << "incorrect number of vertices, expected: " << correct << "\n"; delete pc; delete mb; MPI_Finalize(); return 1; } } rval = mb->write_file( "testpm.h5m", 0, "PARALLEL=WRITE_PART" );MB_CHK_ERR( rval ); if( rval != MB_SUCCESS ) { std::cout << "fail to write output file \n"; delete pc; delete mb; MPI_Finalize(); return 1; } delete pc; delete mb; MPI_Finalize(); return 0; }