MOAB: Mesh Oriented datABase
(version 5.4.1)
|
#include <MeshGeneration.hpp>
Classes | |
struct | BrickOpts |
Public Member Functions | |
MeshGeneration (Interface *mbi, EntityHandle rset=0) | |
virtual | ~MeshGeneration () |
ErrorCode | BrickInstance (BrickOpts &opts) |
Private Attributes | |
Interface * | mb |
EntityHandle | cset |
Definition at line 18 of file MeshGeneration.hpp.
moab::MeshGeneration::MeshGeneration | ( | Interface * | mbi, |
EntityHandle | rset = 0 |
||
) |
Definition at line 30 of file MeshGeneration.cpp.
References moab::ParallelComm::get_pcomm(), and mb.
: mb( impl ), #ifdef MOAB_HAVE_MPI pc( comm ), #endif cset( rset ) { // ErrorCode error; #ifdef MOAB_HAVE_MPI // Get the Parallel Comm instance to prepare all new sets to work in parallel // in case the user did not provide any arguments if( !comm ) pc = moab::ParallelComm::get_pcomm( mb, 0 ); #endif }
moab::MeshGeneration::~MeshGeneration | ( | ) | [virtual] |
Definition at line 50 of file MeshGeneration.cpp.
{}
Definition at line 52 of file MeshGeneration.cpp.
References moab::MeshGeneration::BrickOpts::A, moab::Interface::add_entities(), moab::MeshGeneration::BrickOpts::adjEnts, moab::MeshGeneration::BrickOpts::B, moab::Range::begin(), moab::MeshGeneration::BrickOpts::blockSize, moab::MeshGeneration::BrickOpts::C, CC, moab::Interface::create_meshset(), cset, moab::E, moab::Range::end(), ErrorCode, moab::F, moab::Interface::get_adjacencies(), moab::ReadUtilIface::get_element_connect(), moab::Interface::get_entities_by_dimension(), moab::ReadUtilIface::get_node_coords(), moab::MeshGeneration::BrickOpts::GL, ie, iface, moab::Range::insert(), moab::MeshGeneration::BrickOpts::K, moab::MeshGeneration::BrickOpts::keep_skins, moab::MeshGeneration::BrickOpts::M, mb, MB_CHK_ERR, MB_CHK_SET_ERR, MB_SUCCESS, MB_TAG_CREAT, MB_TAG_DENSE, MB_TAG_SPARSE, MB_TYPE_INTEGER, MB_TYPE_OPAQUE, MBHEX, MBTET, moab::ParallelMergeMesh::merge(), moab::Range::merge(), moab::MergeMesh::merge_entities(), moab::MergeMesh::merge_using_integer_tag(), MESHSET_SET, moab::MeshGeneration::BrickOpts::N, moab::MeshGeneration::BrickOpts::newMergeMethod, moab::MeshGeneration::BrickOpts::parmerge, moab::Range::print(), moab::MeshGeneration::BrickOpts::quadratic, moab::Interface::query_interface(), rank, moab::Interface::remove_entities(), size, moab::Range::size(), moab::Interface::tag_delete(), moab::Interface::tag_get_handle(), moab::Interface::tag_set_data(), moab::MeshGeneration::BrickOpts::tetra, moab::Interface::UNION, moab::ReadUtilIface::update_adjacencies(), moab::MeshGeneration::BrickOpts::xsize, moab::MeshGeneration::BrickOpts::ysize, and moab::MeshGeneration::BrickOpts::zsize.
Referenced by main().
{ int A = opts.A, B = opts.B, C = opts.C, M = opts.M, N = opts.N, K = opts.K; int blockSize = opts.blockSize; double xsize = opts.xsize, ysize = opts.ysize, zsize = opts.zsize; // The size of the region bool newMergeMethod = opts.newMergeMethod; bool quadratic = opts.quadratic; bool tetra = opts.tetra; bool adjEnts = opts.adjEnts; bool parmerge = opts.parmerge; #ifdef MOAB_HAVE_MPI int GL = opts.GL; // number of ghost layers bool keep_skins = opts.keep_skins; #endif int rank = 0, size = 1; #ifndef _MSC_VER /* windows */ clock_t tt = clock(); #endif #ifdef MOAB_HAVE_MPI rank = pc->rank(); size = pc->size(); #endif if( M * N * K != size ) { if( 0 == rank ) std::cout << "M*N*K = " << M * N * K << " != size = " << size << "\n"; return MB_FAILURE; } // Determine m, n, k for processor rank int m, n, k; k = rank / ( M * N ); int leftover = rank % ( M * N ); n = leftover / M; m = leftover % M; // Used for nodes increments int q = ( quadratic ) ? 2 : 1; // Used for element increments int factor = ( tetra ) ? 6 : 1; double dx = xsize / ( A * M * blockSize * q ); // Distance between 2 nodes in x direction double dy = ysize / ( B * N * blockSize * q ); // Distance between 2 nodes in y direction double dz = zsize / ( C * K * blockSize * q ); // Distance between 2 nodes in z direction int NX = ( q * M * A * blockSize + 1 ); int NY = ( q * N * B * blockSize + 1 ); int nex = M * A * blockSize; // Number of elements in x direction, used for global id on element int ney = N * B * blockSize; // Number of elements in y direction ... // int NZ = (K * C * blockSize + 1); // Not used int blockSize1 = q * blockSize + 1; // Used for vertices long num_total_verts = (long)NX * NY * ( K * C * blockSize + 1 ); if( 0 == rank ) { std::cout << "Generate mesh on " << size << " processors \n"; std::cout << "Total number of vertices: " << num_total_verts << "\n"; } // int xstride = 1; int ystride = blockSize1; int zstride = blockSize1 * blockSize1; // Generate the block at (a, b, c); it will represent a partition, it will get a partition tag ReadUtilIface* iface; ErrorCode rval = mb->query_interface( iface );MB_CHK_SET_ERR( rval, "Can't get reader interface" ); Tag global_id_tag; rval = mb->tag_get_handle( "GLOBAL_ID", 1, MB_TYPE_INTEGER, global_id_tag );MB_CHK_SET_ERR( rval, "Can't get global id tag" ); // set global ids Tag new_id_tag; if( !parmerge ) { rval = mb->tag_get_handle( "HANDLEID", sizeof( long ), MB_TYPE_OPAQUE, new_id_tag, MB_TAG_CREAT | MB_TAG_DENSE );MB_CHK_SET_ERR( rval, "Can't get handle id tag" ); } Tag part_tag; int dum_id = -1; rval = mb->tag_get_handle( "PARALLEL_PARTITION", 1, MB_TYPE_INTEGER, part_tag, MB_TAG_CREAT | MB_TAG_SPARSE, &dum_id );MB_CHK_SET_ERR( rval, "Can't get parallel partition tag" ); Range wsets; // write only part sets Range localVerts; Range all3dcells; for( int a = 0; a < A; a++ ) { for( int b = 0; b < B; b++ ) { for( int c = 0; c < C; c++ ) { // We will generate (q*block + 1)^3 vertices, and block^3 hexas; q is 1 for linear, // 2 for quadratic the global id of the vertices will come from m, n, k, a, b, c x // will vary from m*A*q*block + a*q*block to m*A*q*block + (a+1)*q*block etc; int num_nodes = blockSize1 * blockSize1 * blockSize1; vector< double* > arrays; EntityHandle startv; rval = iface->get_node_coords( 3, num_nodes, 0, startv, arrays );MB_CHK_SET_ERR( rval, "Can't get node coords" ); // Will start with the lower corner: int x = m * A * q * blockSize + a * q * blockSize; int y = n * B * q * blockSize + b * q * blockSize; int z = k * C * q * blockSize + c * q * blockSize; int ix = 0; vector< int > gids( num_nodes ); vector< long > lgids( num_nodes ); Range verts( startv, startv + num_nodes - 1 ); for( int kk = 0; kk < blockSize1; kk++ ) { for( int jj = 0; jj < blockSize1; jj++ ) { for( int ii = 0; ii < blockSize1; ii++ ) { arrays[0][ix] = ( x + ii ) * dx; arrays[1][ix] = ( y + jj ) * dy; arrays[2][ix] = ( z + kk ) * dz; gids[ix] = 1 + ( x + ii ) + ( y + jj ) * NX + ( z + kk ) * ( NX * NY ); if( !parmerge ) lgids[ix] = 1 + ( x + ii ) + ( y + jj ) * NX + (long)( z + kk ) * ( NX * NY ); // Set int tags, some nice values? ix++; } } } rval = mb->tag_set_data( global_id_tag, verts, &gids[0] );MB_CHK_SET_ERR( rval, "Can't set global ids to vertices" ); if( !parmerge ) { rval = mb->tag_set_data( new_id_tag, verts, &lgids[0] );MB_CHK_SET_ERR( rval, "Can't set the new handle id tags" ); } localVerts.merge( verts ); int num_hexas = blockSize * blockSize * blockSize; int num_el = num_hexas * factor; EntityHandle starte; // Connectivity EntityHandle* conn; int num_v_per_elem = 8; if( quadratic ) { num_v_per_elem = 27; rval = iface->get_element_connect( num_el, 27, MBHEX, 0, starte, conn );MB_CHK_SET_ERR( rval, "Can't get element connectivity" ); } else if( tetra ) { num_v_per_elem = 4; rval = iface->get_element_connect( num_el, 4, MBTET, 0, starte, conn );MB_CHK_SET_ERR( rval, "Can't get element connectivity" ); } else { rval = iface->get_element_connect( num_el, 8, MBHEX, 0, starte, conn );MB_CHK_SET_ERR( rval, "Can't get element connectivity" ); } Range cells( starte, starte + num_el - 1 ); // Should be elements // Fill cells ix = 0; // Identify the elements at the lower corner, for their global ids int xe = m * A * blockSize + a * blockSize; int ye = n * B * blockSize + b * blockSize; int ze = k * C * blockSize + c * blockSize; gids.resize( num_el ); lgids.resize( num_el ); int ie = 0; // Index now in the elements, for global ids for( int kk = 0; kk < blockSize; kk++ ) { for( int jj = 0; jj < blockSize; jj++ ) { for( int ii = 0; ii < blockSize; ii++ ) { EntityHandle corner = startv + q * ii + q * jj * ystride + q * kk * zstride; // These could overflow for large numbers gids[ie] = 1 + ( ( xe + ii ) + ( ye + jj ) * nex + ( ze + kk ) * ( nex * ney ) ) * factor; // 6 more for tetra lgids[ie] = 1 + ( ( xe + ii ) + ( ye + jj ) * nex + (long)( ze + kk ) * ( nex * ney ) ) * factor; // 6 more for tetra // EntityHandle eh = starte + ie; ie++; if( quadratic ) { // 4 ----- 19 ----- 7 // . | . | // 16 25 18 | // . | . | // 5 ----- 17 ----- 6 | // | 12 | 23 15 // | | | // | 20 | 26 | 22 | // | | | // 13 21 | 14 | // | 0 ----- 11 ----- 3 // | . | . // | 8 24 | 10 // | . | . // 1 ----- 9 ----- 2 // conn[ix] = corner; conn[ix + 1] = corner + 2; conn[ix + 2] = corner + 2 + 2 * ystride; conn[ix + 3] = corner + 2 * ystride; conn[ix + 4] = corner + 2 * zstride; conn[ix + 5] = corner + 2 + 2 * zstride; conn[ix + 6] = corner + 2 + 2 * ystride + 2 * zstride; conn[ix + 7] = corner + 2 * ystride + 2 * zstride; conn[ix + 8] = corner + 1; // 0-1 conn[ix + 9] = corner + 2 + ystride; // 1-2 conn[ix + 10] = corner + 1 + 2 * ystride; // 2-3 conn[ix + 11] = corner + ystride; // 3-0 conn[ix + 12] = corner + zstride; // 0-4 conn[ix + 13] = corner + 2 + zstride; // 1-5 conn[ix + 14] = corner + 2 + 2 * ystride + zstride; // 2-6 conn[ix + 15] = corner + 2 * ystride + zstride; // 3-7 conn[ix + 16] = corner + 1 + 2 * zstride; // 4-5 conn[ix + 17] = corner + 2 + ystride + 2 * zstride; // 5-6 conn[ix + 18] = corner + 1 + 2 * ystride + 2 * zstride; // 6-7 conn[ix + 19] = corner + ystride + 2 * zstride; // 4-7 conn[ix + 20] = corner + 1 + zstride; // 0154 conn[ix + 21] = corner + 2 + ystride + zstride; // 1265 conn[ix + 22] = corner + 1 + 2 * ystride + zstride; // 2376 conn[ix + 23] = corner + ystride + zstride; // 0374 conn[ix + 24] = corner + 1 + ystride; // 0123 conn[ix + 25] = corner + 1 + ystride + 2 * zstride; // 4567 conn[ix + 26] = corner + 1 + ystride + zstride; // center ix += 27; } else if( tetra ) { // E H // F G // // A D // B C EntityHandle AA = corner; EntityHandle BB = corner + 1; EntityHandle CC = corner + 1 + ystride; EntityHandle D = corner + ystride; EntityHandle E = corner + zstride; EntityHandle F = corner + 1 + zstride; EntityHandle G = corner + 1 + ystride + zstride; EntityHandle H = corner + ystride + zstride; // tet EDHG conn[ix] = E; conn[ix + 1] = D; conn[ix + 2] = H; conn[ix + 3] = G; // tet ABCF conn[ix + 4] = AA; conn[ix + 5] = BB; conn[ix + 6] = CC; conn[ix + 7] = F; // tet ADEF conn[ix + 8] = AA; conn[ix + 9] = D; conn[ix + 10] = E; conn[ix + 11] = F; // tet CGDF conn[ix + 12] = CC; conn[ix + 13] = G; conn[ix + 14] = D; conn[ix + 15] = F; // tet ACDF conn[ix + 16] = AA; conn[ix + 17] = CC; conn[ix + 18] = D; conn[ix + 19] = F; // tet DGEF conn[ix + 20] = D; conn[ix + 21] = G; conn[ix + 22] = E; conn[ix + 23] = F; ix += 24; for( int ff = 0; ff < factor - 1; ff++ ) { gids[ie] = gids[ie - 1] + 1; // 6 more for tetra // eh = starte + ie; ie++; } } else { // Linear hex conn[ix] = corner; conn[ix + 1] = corner + 1; conn[ix + 2] = corner + 1 + ystride; conn[ix + 3] = corner + ystride; conn[ix + 4] = corner + zstride; conn[ix + 5] = corner + 1 + zstride; conn[ix + 6] = corner + 1 + ystride + zstride; conn[ix + 7] = corner + ystride + zstride; ix += 8; } } } } EntityHandle part_set; rval = mb->create_meshset( MESHSET_SET, part_set );MB_CHK_SET_ERR( rval, "Can't create mesh set" ); rval = mb->add_entities( part_set, cells );MB_CHK_SET_ERR( rval, "Can't add entities to set" ); all3dcells.merge( cells ); // update adjacencies now, because some elements are new; rval = iface->update_adjacencies( starte, num_el, num_v_per_elem, conn );MB_CHK_SET_ERR( rval, "Can't update adjacencies" ); // If needed, add all edges and faces if( adjEnts ) { // Generate all adj entities dimension 1 and 2 (edges and faces/ tri or qua) Range edges, faces; rval = mb->get_adjacencies( cells, 1, true, edges, Interface::UNION );MB_CHK_SET_ERR( rval, "Can't get edges" ); rval = mb->get_adjacencies( cells, 2, true, faces, Interface::UNION );MB_CHK_SET_ERR( rval, "Can't get faces" ); // rval = mb->add_entities(part_set, edges);MB_CHK_SET_ERR(rval, "Can't add // edges to partition set"); rval = mb->add_entities(part_set, // faces);MB_CHK_SET_ERR(rval, "Can't add faces to partition set"); } rval = mb->tag_set_data( global_id_tag, cells, &gids[0] );MB_CHK_SET_ERR( rval, "Can't set global ids to elements" ); if( !parmerge ) { rval = mb->tag_set_data( new_id_tag, cells, &lgids[0] );MB_CHK_SET_ERR( rval, "Can't set new ids to elements" ); } int part_num = a + m * A + ( b + n * B ) * ( M * A ) + ( c + k * C ) * ( M * A * N * B ); rval = mb->tag_set_data( part_tag, &part_set, 1, &part_num );MB_CHK_SET_ERR( rval, "Can't set part tag on set" ); wsets.insert( part_set ); } } } mb->add_entities( cset, all3dcells ); rval = mb->add_entities( cset, wsets );MB_CHK_SET_ERR( rval, "Can't add entity sets" ); #ifdef MOAB_HAVE_MPI pc->partition_sets() = wsets; #endif /* // Before merge locally rval = mb->write_file("test0.h5m", 0, ";;PARALLEL=WRITE_PART");MB_CHK_SET_ERR(rval, "Can't write in parallel, before merging"); */ // After the mesh is generated on each proc, merge the vertices MergeMesh mm( mb ); // rval = mb->get_entities_by_dimension(0, 3, all3dcells);MB_CHK_SET_ERR(rval, "Can't get all 3d // cells elements"); if( 0 == rank ) { #ifndef _MSC_VER /* windows */ std::cout << "generate local mesh: " << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); #endif std::cout << "number of elements on rank 0: " << all3dcells.size() << endl; std::cout << "Total number of elements " << all3dcells.size() * size << endl; std::cout << "Element type: " << ( tetra ? "MBTET" : "MBHEX" ) << " order:" << ( quadratic ? "quadratic" : "linear" ) << endl; } if( A * B * C != 1 ) { // Merge needed if( newMergeMethod ) { rval = mm.merge_using_integer_tag( localVerts, global_id_tag );MB_CHK_SET_ERR( rval, "Can't merge" ); } else { rval = mm.merge_entities( all3dcells, 0.0001 );MB_CHK_SET_ERR( rval, "Can't merge" ); } #ifndef _MSC_VER /* windows */ if( 0 == rank ) { std::cout << "merge locally: " << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); } #endif } // if adjEnts, add now to each set if( adjEnts ) { for( Range::iterator wsit = wsets.begin(); wsit != wsets.end(); ++wsit ) { EntityHandle ws = *wsit; // write set Range cells, edges, faces; rval = mb->get_entities_by_dimension( ws, 3, cells );MB_CHK_SET_ERR( rval, "Can't get cells" ); rval = mb->get_adjacencies( cells, 1, false, edges, Interface::UNION );MB_CHK_SET_ERR( rval, "Can't get edges" ); rval = mb->get_adjacencies( cells, 2, false, faces, Interface::UNION );MB_CHK_SET_ERR( rval, "Can't get faces" ); rval = mb->add_entities( ws, edges );MB_CHK_SET_ERR( rval, "Can't add edges to partition set" ); rval = mb->add_entities( ws, faces );MB_CHK_SET_ERR( rval, "Can't add faces to partition set" ); } } #ifdef MOAB_HAVE_MPI if( size > 1 ) { // rval = mb->create_meshset(MESHSET_SET, mesh_set);MB_CHK_SET_ERR(rval, "Can't create new // set"); if( parmerge ) { ParallelMergeMesh pm( pc, 0.00001 ); rval = pm.merge();MB_CHK_SET_ERR( rval, "Can't resolve shared ents" ); #ifndef _MSC_VER /* windows */ if( 0 == rank ) { std::cout << "parallel mesh merge: " << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); } #endif } else { rval = pc->resolve_shared_ents( cset, -1, -1, &new_id_tag );MB_CHK_SET_ERR( rval, "Can't resolve shared ents" ); #ifndef _MSC_VER /* windows */ if( 0 == rank ) { std::cout << "resolve shared entities: " << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); } #endif } if( !keep_skins ) { // Default is to delete the 1- and 2-dimensional entities // Delete all quads and edges Range toDelete; rval = mb->get_entities_by_dimension( cset, 1, toDelete );MB_CHK_SET_ERR( rval, "Can't get edges" ); rval = mb->get_entities_by_dimension( cset, 2, toDelete );MB_CHK_SET_ERR( rval, "Can't get faces" ); rval = pc->delete_entities( toDelete );MB_CHK_SET_ERR( rval, "Can't delete entities" ); rval = mb->remove_entities( cset, toDelete );MB_CHK_SET_ERR( rval, "Can't remove entities from base set" ); if( 0 == rank ) { std::cout << "delete edges and faces \n"; toDelete.print( std::cout ); #ifndef _MSC_VER /* windows */ std::cout << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); #endif } } // do some ghosting if required if( GL > 0 ) { rval = pc->exchange_ghost_cells( 3, // int ghost_dim 0, // int bridge_dim GL, // int num_layers 0, // int addl_ents true );MB_CHK_ERR( rval ); // bool store_remote_handles #ifndef _MSC_VER /* windows */ if( 0 == rank ) { std::cout << "exchange " << GL << " ghost layer(s) :" << ( clock() - tt ) / (double)CLOCKS_PER_SEC << " seconds" << endl; tt = clock(); } #endif } } #endif if( !parmerge ) { rval = mb->tag_delete( new_id_tag );MB_CHK_SET_ERR( rval, "Can't delete new ID tag" ); } return MB_SUCCESS; }
EntityHandle moab::MeshGeneration::cset [private] |
Definition at line 53 of file MeshGeneration.hpp.
Referenced by BrickInstance().
Interface* moab::MeshGeneration::mb [private] |
Definition at line 49 of file MeshGeneration.hpp.
Referenced by BrickInstance(), and MeshGeneration().