//-----------------------------------------------------------------------------
void MultiMeshAssembler::assemble(GenericTensor& A, const MultiMeshForm& a)
{
// Developer note: This implementation does not yet handle
// - subdomains
// - interior facets
// - exterior facets
begin(PROGRESS, "Assembling tensor over multimesh function space.");
// Initialize global tensor
_init_global_tensor(A, a);
// Assemble over uncut cells
_assemble_uncut_cells(A, a);
// Assemble over cut cells
_assemble_cut_cells(A, a);
// Assemble over interface
_assemble_interface(A, a);
// Assemble over overlap
_assemble_overlap(A, a);
// Finalize assembly of global tensor
if (finalize_tensor)
A.apply("add");
// Lock any remaining inactive dofs
if (A.rank() == 2)
static_cast<GenericMatrix&>(A).ident_zeros();
end();
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::assemble(GenericTensor& A, const MultiMeshForm& a)
{
// Developer note: This implementation does not yet handle
// - subdomains
// - interior facets
// - exterior facets
begin(PROGRESS, "Assembling tensor over multimesh function space.");
// Initialize global tensor
_init_global_tensor(A, a);
// Assemble over uncut cells
_assemble_uncut_cells(A, a);
// Assemble over exterior facets
_assemble_uncut_exterior_facets(A, a);
// Assemble over cut cells
_assemble_cut_cells(A, a);
// Assemble over interface
_assemble_interface(A, a);
// Assemble over overlap
_assemble_overlap(A, a);
// Finalize assembly of global tensor
if (finalize_tensor)
A.apply("add");
end();
}
//----------------------------------------------------------------------------
void OpenMpAssembler::assemble(GenericTensor& A, const Form& a)
{
// Get mesh
const Mesh& mesh = a.mesh();
if (MPI::size(mesh.mpi_comm()) > 1)
{
dolfin_error("OpenMPAssembler.cpp",
"perform multithreaded assembly using OpenMP assembler",
"The OpenMp assembler has not been tested in combination with MPI");
}
dolfin_assert(a.ufc_form());
// All assembler functions above end up calling this function, which
// in turn calls the assembler functions below to assemble over
// cells, exterior and interior facets. Note the importance of
// treating empty mesh functions as null pointers for the PyDOLFIN
// interface.
// Get cell domains
std::shared_ptr<const MeshFunction<std::size_t>> cell_domains
= a.cell_domains();
// Get exterior facet domains
std::shared_ptr<const MeshFunction<std::size_t>> exterior_facet_domains
= a.exterior_facet_domains();
// Get interior facet domains
std::shared_ptr<const MeshFunction<std::size_t>> interior_facet_domains
= a.interior_facet_domains();
// Check form
AssemblerBase::check(a);
// Create data structure for local assembly data
UFC ufc(a);
// Initialize global tensor
init_global_tensor(A, a);
// FIXME: The below selections should be made robust
if (a.ufc_form()->has_interior_facet_integrals())
assemble_interior_facets(A, a, ufc, interior_facet_domains, cell_domains, 0);
if (a.ufc_form()->has_exterior_facet_integrals())
{
assemble_cells_and_exterior_facets(A, a, ufc, cell_domains,
exterior_facet_domains, 0);
}
else
assemble_cells(A, a, ufc, cell_domains, 0);
// Finalize assembly of global tensor
if (finalize_tensor)
A.apply("add");
}
//-----------------------------------------------------------------------------
void Assembler::assemble_interior_facets(
GenericTensor& A,
const Form& a,
UFC& ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains,
std::shared_ptr<const MeshFunction<std::size_t>> cell_domains,
std::vector<double>* values)
{
// Skip assembly if there are no interior facet integrals
if (!ufc.form.has_interior_facet_integrals())
return;
// Set timer
Timer timer("Assemble interior facets");
// Extract mesh and coefficients
const Mesh& mesh = a.mesh();
// MPI rank
const int my_mpi_rank = MPI::rank(mesh.mpi_comm());
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dofs for cells, and a vector holding pointers to same
std::vector<std::vector<dolfin::la_index>> macro_dofs(form_rank);
std::vector<ArrayView<const dolfin::la_index>> macro_dof_ptrs(form_rank);
// Interior facet integral
const ufc::interior_facet_integral* integral
= ufc.default_interior_facet_integral.get();
// Check whether integral is domain-dependent
bool use_domains = domains && !domains->empty();
bool use_cell_domains = cell_domains && !cell_domains->empty();
// Compute facets and facet - cell connectivity if not already computed
const std::size_t D = mesh.topology().dim();
mesh.init(D - 1);
mesh.init(D - 1, D);
dolfin_assert(mesh.ordered());
// Assemble over interior facets (the facets of the mesh)
ufc::cell ufc_cell[2];
std::vector<double> coordinate_dofs[2];
Progress p(AssemblerBase::progress_message(A.rank(), "interior facets"),
mesh.num_facets());
for (FacetIterator facet(mesh); !facet.end(); ++facet)
{
if (facet->num_entities(D) == 1)
continue;
// Check that facet is not a ghost
dolfin_assert(!facet->is_ghost());
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_interior_facet_integral((*domains)[*facet]);
// Skip integral if zero
if (!integral)
continue;
// Get cells incident with facet (which is 0 and 1 here is arbitrary)
dolfin_assert(facet->num_entities(D) == 2);
std::size_t cell_index_plus = facet->entities(D)[0];
std::size_t cell_index_minus = facet->entities(D)[1];
if (use_cell_domains && (*cell_domains)[cell_index_plus]
< (*cell_domains)[cell_index_minus])
{
std::swap(cell_index_plus, cell_index_minus);
}
// The convention '+' = 0, '-' = 1 is from ffc
const Cell cell0(mesh, cell_index_plus);
const Cell cell1(mesh, cell_index_minus);
// Get local index of facet with respect to each cell
std::size_t local_facet0 = cell0.index(*facet);
std::size_t local_facet1 = cell1.index(*facet);
// Update to current pair of cells
cell0.get_cell_data(ufc_cell[0], local_facet0);
cell0.get_coordinate_dofs(coordinate_dofs[0]);
cell1.get_cell_data(ufc_cell[1], local_facet1);
cell1.get_coordinate_dofs(coordinate_dofs[1]);
ufc.update(cell0, coordinate_dofs[0], ufc_cell[0],
cell1, coordinate_dofs[1], ufc_cell[1],
integral->enabled_coefficients());
// Tabulate dofs for each dimension on macro element
for (std::size_t i = 0; i < form_rank; i++)
{
// Get dofs for each cell
const ArrayView<const dolfin::la_index> cell_dofs0
= dofmaps[i]->cell_dofs(cell0.index());
const ArrayView<const dolfin::la_index> cell_dofs1
= dofmaps[i]->cell_dofs(cell1.index());
// Create space in macro dof vector
macro_dofs[i].resize(cell_dofs0.size() + cell_dofs1.size());
// Copy cell dofs into macro dof vector
std::copy(cell_dofs0.data(), cell_dofs0.data() + cell_dofs0.size(),
macro_dofs[i].begin());
std::copy(cell_dofs1.data(), cell_dofs1.data() + cell_dofs1.size(),
macro_dofs[i].begin() + cell_dofs0.size());
macro_dof_ptrs[i].set(macro_dofs[i]);
}
// Tabulate interior facet tensor on macro element
integral->tabulate_tensor(ufc.macro_A.data(),
ufc.macro_w(),
coordinate_dofs[0].data(),
coordinate_dofs[1].data(),
local_facet0,
local_facet1,
ufc_cell[0].orientation,
ufc_cell[1].orientation);
if (cell0.is_ghost() != cell1.is_ghost())
{
int ghost_rank = -1;
if (cell0.is_ghost())
ghost_rank = cell0.owner();
else
ghost_rank = cell1.owner();
dolfin_assert(my_mpi_rank != ghost_rank);
dolfin_assert(ghost_rank != -1);
if (ghost_rank < my_mpi_rank)
continue;
}
// Add entries to global tensor
A.add_local(ufc.macro_A.data(), macro_dof_ptrs);
p++;
}
}
//-----------------------------------------------------------------------------
void Assembler::assemble_exterior_facets(
GenericTensor& A,
const Form& a,
UFC& ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains,
std::vector<double>* values)
{
// Skip assembly if there are no exterior facet integrals
if (!ufc.form.has_exterior_facet_integrals())
return;
// Set timer
Timer timer("Assemble exterior facets");
// Extract mesh
const Mesh& mesh = a.mesh();
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Exterior facet integral
const ufc::exterior_facet_integral* integral
= ufc.default_exterior_facet_integral.get();
// Check whether integral is domain-dependent
bool use_domains = domains && !domains->empty();
// Compute facets and facet - cell connectivity if not already computed
const std::size_t D = mesh.topology().dim();
mesh.init(D - 1);
mesh.init(D - 1, D);
dolfin_assert(mesh.ordered());
// Assemble over exterior facets (the cells of the boundary)
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
Progress p(AssemblerBase::progress_message(A.rank(), "exterior facets"),
mesh.num_facets());
for (FacetIterator facet(mesh); !facet.end(); ++facet)
{
// Only consider exterior facets
if (!facet->exterior())
{
p++;
continue;
}
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_exterior_facet_integral((*domains)[*facet]);
// Skip integral if zero
if (!integral)
continue;
// Get mesh cell to which mesh facet belongs (pick first, there is
// only one)
dolfin_assert(facet->num_entities(D) == 1);
Cell mesh_cell(mesh, facet->entities(D)[0]);
// Check that cell is not a ghost
dolfin_assert(!mesh_cell.is_ghost());
// Get local index of facet with respect to the cell
const std::size_t local_facet = mesh_cell.index(*facet);
// Update UFC cell
mesh_cell.get_cell_data(ufc_cell, local_facet);
mesh_cell.get_coordinate_dofs(coordinate_dofs);
// Update UFC object
ufc.update(mesh_cell, coordinate_dofs, ufc_cell,
integral->enabled_coefficients());
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
dofs[i] = dofmaps[i]->cell_dofs(mesh_cell.index());
// Tabulate exterior facet tensor
integral->tabulate_tensor(ufc.A.data(),
ufc.w(),
coordinate_dofs.data(),
local_facet,
ufc_cell.orientation);
// Add entries to global tensor
A.add_local(ufc.A.data(), dofs);
p++;
}
}
//-----------------------------------------------------------------------------
void Assembler::assemble_cells(
GenericTensor& A,
const Form& a,
UFC& ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains,
std::vector<double>* values)
{
// Skip assembly if there are no cell integrals
if (!ufc.form.has_cell_integrals())
return;
// Set timer
Timer timer("Assemble cells");
// Extract mesh
const Mesh& mesh = a.mesh();
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Check if form is a functional
const bool is_cell_functional = (values && form_rank == 0) ? true : false;
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Cell integral
ufc::cell_integral* integral = ufc.default_cell_integral.get();
// Check whether integral is domain-dependent
bool use_domains = domains && !domains->empty();
// Assemble over cells
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
Progress p(AssemblerBase::progress_message(A.rank(), "cells"),
mesh.num_cells());
for (CellIterator cell(mesh); !cell.end(); ++cell)
{
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_cell_integral((*domains)[*cell]);
// Skip if no integral on current domain
if (!integral)
continue;
// Check that cell is not a ghost
dolfin_assert(!cell->is_ghost());
// Update to current cell
cell->get_cell_data(ufc_cell);
cell->get_coordinate_dofs(coordinate_dofs);
ufc.update(*cell, coordinate_dofs, ufc_cell,
integral->enabled_coefficients());
// Get local-to-global dof maps for cell
bool empty_dofmap = false;
for (std::size_t i = 0; i < form_rank; ++i)
{
dofs[i] = dofmaps[i]->cell_dofs(cell->index());
empty_dofmap = empty_dofmap || dofs[i].size() == 0;
}
// Skip if at least one dofmap is empty
if (empty_dofmap)
continue;
// Tabulate cell tensor
integral->tabulate_tensor(ufc.A.data(), ufc.w(),
coordinate_dofs.data(),
ufc_cell.orientation);
// Add entries to global tensor. Either store values cell-by-cell
// (currently only available for functionals)
if (is_cell_functional)
(*values)[cell->index()] = ufc.A[0];
else
A.add_local(ufc.A.data(), dofs);
p++;
}
}
//-----------------------------------------------------------------------------
void OpenMpAssembler::assemble_interior_facets(
GenericTensor& A,
const Form& a, UFC& _ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains,
std::shared_ptr<const MeshFunction<std::size_t>> cell_domains,
std::vector<double>* values)
{
warning("OpenMpAssembler::assemble_interior_facets is untested.");
// Extract mesh
const Mesh& mesh = a.mesh();
// Topological dimension
const std::size_t D = mesh.topology().dim();
dolfin_assert(!values);
// Skip assembly if there are no interior facet integrals
if (!_ufc.form.has_interior_facet_integrals())
return;
Timer timer("Assemble interior facets");
// Set number of OpenMP threads (from parameter systems)
omp_set_num_threads(parameters["num_threads"]);
// Get integral for sub domain (if any)
bool use_domains = domains && !domains->empty();
bool use_cell_domains = cell_domains && !cell_domains->empty();
if (use_domains)
{
dolfin_error("OpenMPAssembler.cpp",
"perform multithreaded assembly using OpenMP assembler",
"Subdomains are not yet handled");
}
// Color mesh
std::vector<std::size_t> coloring_type = a.coloring(D - 1);
mesh.color(coloring_type);
// Dummy UFC object since each thread needs to created its own UFC object
UFC ufc(_ufc);
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dofs for cells
std::vector<std::vector<dolfin::la_index>> macro_dofs(form_rank);
// Interior facet integral
const ufc::interior_facet_integral* integral
= ufc.default_interior_facet_integral.get();
// Compute facets and facet - cell connectivity if not already computed
mesh.init(D - 1);
mesh.init(D - 1, D);
dolfin_assert(mesh.ordered());
// Get coloring data
std::map<const std::vector<std::size_t>,
std::pair<std::vector<std::size_t>,
std::vector<std::vector<std::size_t>>>>::const_iterator
mesh_coloring;
mesh_coloring = mesh.topology().coloring.find(coloring_type);
// Check that requested coloring has been computed
if (mesh_coloring == mesh.topology().coloring.end())
{
dolfin_error("OpenMPAssembler.cpp",
"perform multithreaded assembly using OpenMP assembler",
"Requested mesh coloring has not been computed");
}
// Get coloring data
const std::vector<std::vector<std::size_t>>& entities_of_color
= mesh_coloring->second.second;
// UFC cells and vertex coordinates
ufc::cell ufc_cell0, ufc_cell1;
std::vector<double> vertex_coordinates0, vertex_coordinates1;
// Assemble over interior facets (loop over colours, then cells of same color)
const std::size_t num_colors = entities_of_color.size();
for (std::size_t color = 0; color < num_colors; ++color)
{
// Get the array of facet indices of current color
const std::vector<std::size_t>& colored_facets = entities_of_color[color];
// Number of facets of current color
const int num_facets = colored_facets.size();
// OpenMP test loop over cells of the same color
Progress p(AssemblerBase::progress_message(A.rank(), "interior facets"),
mesh.num_facets());
#pragma omp parallel for schedule(guided, 20) firstprivate(ufc, ufc_cell0, ufc_cell1, vertex_coordinates0, vertex_coordinates1, macro_dofs, integral)
for (int facet_index = 0; facet_index < num_facets; ++facet_index)
{
// Facet index
const std::size_t index = colored_facets[facet_index];
// Create cell
const Facet facet(mesh, index);
// Only consider interior facets
if (facet.exterior())
{
p++;
continue;
}
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_interior_facet_integral((*domains)[facet]);
// Skip integral if zero
if (!integral)
continue;
// Get cells incident with facet (which is 0 and 1 here is arbitrary)
dolfin_assert(facet.num_entities(D) == 2);
std::size_t cell_index_plus = facet.entities(D)[0];
std::size_t cell_index_minus = facet.entities(D)[1];
if (use_cell_domains && (*cell_domains)[cell_index_plus] < (*cell_domains)[cell_index_minus])
std::swap(cell_index_plus, cell_index_minus);
// The convention '+' = 0, '-' = 1 is from ffc
const Cell cell0(mesh, cell_index_plus);
const Cell cell1(mesh, cell_index_minus);
// Get local index of facet with respect to each cell
const std::size_t local_facet0 = cell0.index(facet);
const std::size_t local_facet1 = cell1.index(facet);
// Update UFC cell
cell0.get_vertex_coordinates(vertex_coordinates0);
cell0.get_cell_data(ufc_cell0, local_facet0);
cell1.get_vertex_coordinates(vertex_coordinates1);
cell1.get_cell_data(ufc_cell1, local_facet1);
// Update to current pair of cells
ufc.update(cell0, vertex_coordinates0, ufc_cell0,
cell1, vertex_coordinates1, ufc_cell1,
integral->enabled_coefficients());
// Tabulate dofs for each dimension on macro element
for (std::size_t i = 0; i < form_rank; i++)
{
// Get dofs for each cell
const ArrayView<const dolfin::la_index> cell_dofs0
= dofmaps[i]->cell_dofs(cell0.index());
const ArrayView<const dolfin::la_index> cell_dofs1
= dofmaps[i]->cell_dofs(cell1.index());
// Create space in macro dof vector
macro_dofs[i].resize(cell_dofs0.size() + cell_dofs1.size());
// Copy cell dofs into macro dof vector
std::copy(cell_dofs0.begin(), cell_dofs0.end(), macro_dofs[i].begin());
std::copy(cell_dofs1.begin(), cell_dofs1.end(),
macro_dofs[i].begin() + cell_dofs0.size());
}
// Tabulate exterior interior facet tensor on macro element
integral->tabulate_tensor(ufc.macro_A.data(),
ufc.macro_w(),
vertex_coordinates0.data(),
vertex_coordinates1.data(),
local_facet0,
local_facet1,
ufc_cell0.orientation,
ufc_cell1.orientation);
// Add entries to global tensor
std::vector<ArrayView<const la_index>>
macro_dofs_p(macro_dofs.size());
for (std::size_t i = 0; i < macro_dofs.size(); ++i)
macro_dofs_p[i].set(macro_dofs[i]);
A.add_local(ufc.macro_A.data(), macro_dofs_p);
p++;
}
}
}
//-----------------------------------------------------------------------------
void AssemblerBase::init_global_tensor(GenericTensor& A, const Form& a)
{
dolfin_assert(a.ufc_form());
// Get dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < a.rank(); ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
if (A.size(0) == 0)
{
Timer t0("Build sparsity");
// Create layout for initialising tensor
std::shared_ptr<TensorLayout> tensor_layout;
tensor_layout = A.factory().create_layout(a.rank());
dolfin_assert(tensor_layout);
// Get dimensions
std::vector<std::size_t> global_dimensions;
std::vector<std::pair<std::size_t, std::size_t> > local_range;
std::vector<std::size_t> block_sizes;
for (std::size_t i = 0; i < a.rank(); i++)
{
dolfin_assert(dofmaps[i]);
global_dimensions.push_back(dofmaps[i]->global_dimension());
local_range.push_back(dofmaps[i]->ownership_range());
block_sizes.push_back(dofmaps[i]->block_size);
}
// Set block size for sparsity graphs
std::size_t block_size = 1;
if (a.rank() == 2)
{
const std::vector<std::size_t> _bs(a.rank(), dofmaps[0]->block_size);
block_size = (block_sizes == _bs) ? dofmaps[0]->block_size : 1;
}
// Initialise tensor layout
tensor_layout->init(a.mesh().mpi_comm(), global_dimensions, block_size,
local_range);
// Build sparsity pattern if required
if (tensor_layout->sparsity_pattern())
{
GenericSparsityPattern& pattern = *tensor_layout->sparsity_pattern();
SparsityPatternBuilder::build(pattern,
a.mesh(), dofmaps,
a.ufc_form()->has_cell_integrals(),
a.ufc_form()->has_interior_facet_integrals(),
a.ufc_form()->has_exterior_facet_integrals(),
keep_diagonal);
}
t0.stop();
// Initialize tensor
Timer t1("Init tensor");
A.init(*tensor_layout);
t1.stop();
// Insert zeros on the diagonal as diagonal entries may be prematurely
// optimised away by the linear algebra backend when calling
// GenericMatrix::apply, e.g. PETSc does this then errors when matrices
// have no diagonal entry inserted.
if (A.rank() == 2 && keep_diagonal)
{
// Down cast to GenericMatrix
GenericMatrix& _A = A.down_cast<GenericMatrix>();
// Loop over rows and insert 0.0 on the diagonal
const double block = 0.0;
const std::pair<std::size_t, std::size_t> row_range = A.local_range(0);
const std::size_t range = std::min(row_range.second, A.size(1));
for (std::size_t i = row_range.first; i < range; i++)
{
dolfin::la_index _i = i;
_A.set(&block, 1, &_i, 1, &_i);
}
A.apply("flush");
}
// Delete sparsity pattern
Timer t2("Delete sparsity");
t2.stop();
}
else
{
// If tensor is not reset, check that dimensions are correct
for (std::size_t i = 0; i < a.rank(); ++i)
{
if (A.size(i) != dofmaps[i]->global_dimension())
{
dolfin_error("AssemblerBase.cpp",
"assemble form",
"Reset of tensor in assembly not requested, but dim %d of tensor does not match form", i);
}
}
}
if (!add_values)
A.zero();
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_assemble_uncut_cells(GenericTensor& A,
const MultiMeshForm& a)
{
// Get form rank
const std::size_t form_rank = a.rank();
// Extract multimesh
std::shared_ptr<const MultiMesh> multimesh = a.multimesh();
// Collect pointers to dof maps
std::vector<const MultiMeshDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; i++)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Initialize variables that will be reused throughout assembly
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
// Iterate over parts
for (std::size_t part = 0; part < a.num_parts(); part++)
{
log(PROGRESS, "Assembling multimesh form over uncut cells on part %d.", part);
// Get form for current part
const Form& a_part = *a.part(part);
// Create data structure for local assembly data
UFC ufc_part(a_part);
// Extract mesh
const Mesh& mesh_part = a_part.mesh();
// FIXME: Handle subdomains
// Get integral
ufc::cell_integral* integral = ufc_part.default_cell_integral.get();
// Skip if we don't have a cell integral
if (!integral) continue;
// Get uncut cells
const std::vector<unsigned int>& uncut_cells = multimesh->uncut_cells(part);
// Iterate over uncut cells
for (auto it = uncut_cells.begin(); it != uncut_cells.end(); ++it)
{
// Create cell
Cell cell(mesh_part, *it);
// Update to current cell
cell.get_cell_data(ufc_cell);
cell.get_coordinate_dofs(coordinate_dofs);
ufc_part.update(cell, coordinate_dofs, ufc_cell);
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
{
const auto dofmap = a.function_space(i)->dofmap()->part(part);
dofs[i] = dofmap->cell_dofs(cell.index());
}
// Tabulate cell tensor
integral->tabulate_tensor(ufc_part.A.data(),
ufc_part.w(),
coordinate_dofs.data(),
ufc_cell.orientation);
// Add entries to global tensor
A.add(ufc_part.A.data(), dofs);
}
}
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_init_global_tensor(GenericTensor& A,
const MultiMeshForm& a)
{
log(PROGRESS, "Initializing global tensor.");
// This function initializes the big system matrix corresponding to
// all dofs (including inactive dofs) on all parts of the MultiMesh
// function space.
// Create layout for initializing tensor
std::shared_ptr<TensorLayout> tensor_layout;
tensor_layout = A.factory().create_layout(a.rank());
dolfin_assert(tensor_layout);
// Get dimensions
std::vector<std::shared_ptr<const IndexMap>> index_maps;
for (std::size_t i = 0; i < a.rank(); i++)
{
std::shared_ptr<const MultiMeshFunctionSpace> V = a.function_space(i);
dolfin_assert(V);
index_maps.push_back(std::shared_ptr<const IndexMap>
(new IndexMap(MPI_COMM_WORLD, V->dim(), 1)));
}
// Initialise tensor layout
tensor_layout->init(MPI_COMM_WORLD, index_maps,
TensorLayout::Ghosts::UNGHOSTED);
// Build sparsity pattern if required
if (tensor_layout->sparsity_pattern())
{
GenericSparsityPattern& pattern = *tensor_layout->sparsity_pattern();
SparsityPatternBuilder::build_multimesh_sparsity_pattern(pattern, a);
}
// Initialize tensor
A.init(*tensor_layout);
// Insert zeros on the diagonal as diagonal entries may be prematurely
// optimised away by the linear algebra backend when calling
// GenericMatrix::apply, e.g. PETSc does this then errors when matrices
// have no diagonal entry inserted.
if (A.rank() == 2)
{
// Down cast to GenericMatrix
GenericMatrix& _matA = A.down_cast<GenericMatrix>();
// Loop over rows and insert 0.0 on the diagonal
const double block = 0.0;
const std::pair<std::size_t, std::size_t> row_range = A.local_range(0);
const std::size_t range = std::min(row_range.second, A.size(1));
for (std::size_t i = row_range.first; i < range; i++)
{
dolfin::la_index _i = i;
_matA.set(&block, 1, &_i, 1, &_i);
}
A.apply("flush");
}
// Set tensor to zero
A.zero();
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_assemble_overlap(GenericTensor& A,
const MultiMeshForm& a)
{
// FIXME: This function and assemble_interface are very similar.
// FIXME: Refactor to improve code reuse.
// Extract multimesh
std::shared_ptr<const MultiMesh> multimesh = a.multimesh();
// Get form rank
const std::size_t form_rank = a.rank();
// Collect pointers to dof maps
std::vector<const MultiMeshDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; i++)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<const std::vector<dolfin::la_index>* > dofs(form_rank);
// Initialize variables that will be reused throughout assembly
ufc::cell ufc_cell[2];
std::vector<double> coordinate_dofs[2];
std::vector<double> macro_coordinate_dofs;
// Vector to hold dofs for cells, and a vector holding pointers to same
std::vector<ArrayView<const dolfin::la_index>> macro_dof_ptrs(form_rank);
std::vector<std::vector<dolfin::la_index>> macro_dofs(form_rank);
// Iterate over parts
for (std::size_t part = 0; part < a.num_parts(); part++)
{
log(PROGRESS, "Assembling multimesh form over overlap on part %d.", part);
// Get form for current part
const Form& a_part = *a.part(part);
// Create data structure for local assembly data
UFC ufc_part(a_part);
// FIXME: Handle subdomains
// Get integral
ufc::overlap_integral* integral = ufc_part.default_overlap_integral.get();
// Skip if we don't have an overlap integral
if (!integral) continue;
// Get quadrature rules
const auto& quadrature_rules = multimesh->quadrature_rule_overlap(part);
// Get collision map
const auto& cmap = multimesh->collision_map_cut_cells(part);
// Iterate over all cut cells in collision map
for (auto it = cmap.begin(); it != cmap.end(); ++it)
{
// Get cut cell
const unsigned int cut_cell_index = it->first;
const Cell cut_cell(*multimesh->part(part), cut_cell_index);
// Iterate over cutting cells
const auto& cutting_cells = it->second;
for (auto jt = cutting_cells.begin(); jt != cutting_cells.end(); jt++)
{
// Get cutting part and cutting cell
const std::size_t cutting_part = jt->first;
const std::size_t cutting_cell_index = jt->second;
const Cell cutting_cell(*multimesh->part(cutting_part), cutting_cell_index);
// Get quadrature rule for interface part defined by
// intersection of the cut and cutting cells
const std::size_t k = jt - cutting_cells.begin();
dolfin_assert(k < quadrature_rules.at(cut_cell_index).size());
const auto& qr = quadrature_rules.at(cut_cell_index)[k];
// FIXME: There might be quite a few cases when we skip cutting
// FIXME: cells because there are no quadrature points. Perhaps
// FIXME: we can rewrite this inner loop to avoid unnecessary
// FIXME: iterations.
// Skip if there are no quadrature points
const std::size_t num_quadrature_points = qr.second.size();
if (num_quadrature_points == 0)
continue;
// Create aliases for cells to simplify notation
const Cell& cell_0 = cut_cell;
const Cell& cell_1 = cutting_cell;
// Update to current pair of cells
cell_0.get_cell_data(ufc_cell[0], 0);
cell_1.get_cell_data(ufc_cell[1], 0);
cell_0.get_coordinate_dofs(coordinate_dofs[0]);
cell_1.get_coordinate_dofs(coordinate_dofs[1]);
ufc_part.update(cell_0, coordinate_dofs[0], ufc_cell[0],
cell_1, coordinate_dofs[1], ufc_cell[1]);
// Collect vertex coordinates
macro_coordinate_dofs.resize(coordinate_dofs[0].size() +
coordinate_dofs[0].size());
std::copy(coordinate_dofs[0].begin(),
coordinate_dofs[0].end(),
macro_coordinate_dofs.begin());
std::copy(coordinate_dofs[1].begin(),
coordinate_dofs[1].end(),
macro_coordinate_dofs.begin() + coordinate_dofs[0].size());
// Tabulate dofs for each dimension on macro element
for (std::size_t i = 0; i < form_rank; i++)
{
// Get dofs for cut mesh
const auto dofmap_0 = a.function_space(i)->dofmap()->part(part);
const auto dofs_0 = dofmap_0->cell_dofs(cell_0.index());
// Get dofs for cutting mesh
const auto dofmap_1 = a.function_space(i)->dofmap()->part(cutting_part);
const auto dofs_1 = dofmap_1->cell_dofs(cell_1.index());
// Create space in macro dof vector
macro_dofs[i].resize(dofs_0.size() + dofs_1.size());
// Copy cell dofs into macro dof vector
std::copy(dofs_0.begin(), dofs_0.end(),
macro_dofs[i].begin());
std::copy(dofs_1.begin(), dofs_1.end(),
macro_dofs[i].begin() + dofs_0.size());
// Update array view
macro_dof_ptrs[i]
= ArrayView<const dolfin::la_index>(macro_dofs[i].size(),
macro_dofs[i].data());
}
// FIXME: Cell orientation not supported
const int cell_orientation = ufc_cell[0].orientation;
// Tabulate overlap tensor on macro element
integral->tabulate_tensor(ufc_part.macro_A.data(),
ufc_part.macro_w(),
macro_coordinate_dofs.data(),
num_quadrature_points,
qr.first.data(),
qr.second.data(),
cell_orientation);
// Add entries to global tensor
A.add(ufc_part.macro_A.data(), macro_dof_ptrs);
}
}
}
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_assemble_cut_cells(GenericTensor& A,
const MultiMeshForm& a)
{
// Get form rank
const std::size_t form_rank = a.rank();
// Extract multimesh
std::shared_ptr<const MultiMesh> multimesh = a.multimesh();
// Collect pointers to dof maps
std::vector<const MultiMeshDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; i++)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Initialize variables that will be reused throughout assembly
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
// Iterate over parts
for (std::size_t part = 0; part < a.num_parts(); part++)
{
log(PROGRESS, "Assembling multimesh form over cut cells on part %d.", part);
// Get form for current part
const Form& a_part = *a.part(part);
// Create data structure for local assembly data
UFC ufc_part(a_part);
// Extract mesh
const Mesh& mesh_part = a_part.mesh();
// FIXME: Handle subdomains
// Get integral
ufc::cutcell_integral* integral = ufc_part.default_cutcell_integral.get();
// Skip if we don't have a cutcell integral
if (!integral) continue;
// Get cut cells and quadrature rules
const std::vector<unsigned int>& cut_cells = multimesh->cut_cells(part);
const auto& quadrature_rules = multimesh->quadrature_rule_cut_cells(part);
// Iterate over cut cells
for (auto it = cut_cells.begin(); it != cut_cells.end(); ++it)
{
// Create cell
Cell cell(mesh_part, *it);
// Update to current cell
cell.get_cell_data(ufc_cell);
cell.get_coordinate_dofs(coordinate_dofs);
ufc_part.update(cell, coordinate_dofs, ufc_cell);
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
{
const auto dofmap = a.function_space(i)->dofmap()->part(part);
dofs[i] = dofmap->cell_dofs(cell.index());
}
// Get quadrature rule for cut cell
const auto& qr = quadrature_rules.at(*it);
// Skip if there are no quadrature points
std::size_t num_quadrature_points = qr.second.size();
if (num_quadrature_points == 0)
continue;
// FIXME: Handle this inside the quadrature point generation,
// FIXME: perhaps by storing three different sets of points,
// FIXME: including cut cell, overlap and the whole cell.
// Include only quadrature points with positive weight if
// integration should be extended on cut cells
std::pair<std::vector<double>, std::vector<double>> pr;
if (extend_cut_cell_integration)
{
const std::size_t gdim = mesh_part.geometry().dim();
for (std::size_t i = 0; i < num_quadrature_points; i++)
{
if (qr.second[i] > 0.0)
{
pr.second.push_back(qr.second[i]);
for (std::size_t j = i*gdim; j < (i + 1)*gdim; j++)
pr.first.push_back(qr.first[j]);
}
}
num_quadrature_points = pr.second.size();
}
else
{
pr = qr;
}
// Tabulate cell tensor
integral->tabulate_tensor(ufc_part.A.data(),
ufc_part.w(),
coordinate_dofs.data(),
num_quadrature_points,
pr.first.data(),
pr.second.data(),
ufc_cell.orientation);
// Add entries to global tensor
A.add(ufc_part.A.data(), dofs);
}
}
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_assemble_uncut_exterior_facets(GenericTensor& A,
const MultiMeshForm& a)
{
// FIXME: This implementation assumes that there is one background mesh
// that contains the entire exterior facet.
// Get form rank
const std::size_t form_rank = a.rank();
// Extract multimesh
std::shared_ptr<const MultiMesh> multimesh = a.multimesh();
// Collect pointers to dof maps
std::vector<const MultiMeshDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; i++)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Initialize variables that will be reused throughout assembly
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
// Assembly exterior uncut facets on mesh 0, the background mesh
int part = 0;
log(PROGRESS, "Assembling multimesh form over uncut facets on part %d.", part);
// Get form for current part
const Form& a_part = *a.part(part);
// Create data structure for local assembly data
UFC ufc_part(a_part);
// Extract mesh
dolfin_assert(a_part.mesh());
const Mesh& mesh_part = *(a_part.mesh());
// FIXME: Handle subdomains
// Exterior facet integral
const ufc::exterior_facet_integral* integral = ufc_part.default_exterior_facet_integral.get();
// Skip if we don't have a facet integral
if (!integral) return;
// Iterate over uncut cells
for (FacetIterator facet(mesh_part); !facet.end(); ++facet)
{
// Only consider exterior facets
if (!facet->exterior())
{
continue;
}
const std::size_t D = mesh_part.topology().dim();
// Get mesh cell to which mesh facet belongs (pick first, there is
// only one)
dolfin_assert(facet->num_entities(D) == 1);
Cell mesh_cell(mesh_part, facet->entities(D)[0]);
// Check that cell is not a ghost
dolfin_assert(!mesh_cell.is_ghost());
// Get local index of facet with respect to the cell
const std::size_t local_facet = mesh_cell.index(*facet);
// Update UFC cell
mesh_cell.get_cell_data(ufc_cell, local_facet);
mesh_cell.get_coordinate_dofs(coordinate_dofs);
// Update UFC object
ufc_part.update(mesh_cell, coordinate_dofs, ufc_cell,
integral->enabled_coefficients());
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
{
const auto dofmap = a.function_space(i)->dofmap()->part(part);
const auto dmap = dofmap->cell_dofs(mesh_cell.index());
dofs[i] = ArrayView<const dolfin::la_index>(dmap.size(), dmap.data());
}
// Tabulate cell tensor
integral->tabulate_tensor(ufc_part.A.data(),
ufc_part.w(),
coordinate_dofs.data(),
local_facet,
ufc_cell.orientation);
// Add entries to global tensor
A.add(ufc_part.A.data(), dofs);
}
}
//-----------------------------------------------------------------------------
void Assembler::assemble_vertices(
GenericTensor& A,
const Form& a,
UFC& ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains)
{
// Skip assembly if there are no point integrals
if (!ufc.form.has_vertex_integrals())
return;
// Set timer
Timer timer("Assemble vertices");
// Extract mesh
const Mesh& mesh = a.mesh();
// Compute cell and vertex - cell connectivity if not already
// computed
const std::size_t D = mesh.topology().dim();
mesh.init(0);
mesh.init(0, D);
dolfin_assert(mesh.ordered());
// Logics for shared vertices
const bool has_shared_vertices = mesh.topology().have_shared_entities(0);
const std::map<unsigned int, std::set<unsigned int>>&
shared_vertices = mesh.topology().shared_entities(0);
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps(form_rank);
// Create a vector for storying local to local map for vertex entity
// dofs
std::vector<std::vector<std::size_t>> local_to_local_dofs(form_rank);
// Create a values vector to be used to fan out local tabulated
// values to the global tensor
std::vector<double> local_values(1);
// Vector to hold local dof map for a vertex
std::vector<std::vector<dolfin::la_index>> global_dofs(form_rank);
std::vector<ArrayView<const dolfin::la_index>> global_dofs_p(form_rank);
std::vector<dolfin::la_index> local_dof_size(form_rank);
for (std::size_t i = 0; i < form_rank; ++i)
{
dofmaps[i] = a.function_space(i)->dofmap().get();
// Check that the test and trial space as dofs on the vertices
if (dofmaps[i]->num_entity_dofs(0) == 0)
{
dolfin_error("Assembler.cpp",
"assemble form over vertices",
"Expecting test and trial spaces to have dofs on "\
"vertices for point integrals");
}
// Check that the test and trial spaces do not have dofs other
// than on vertices
for (std::size_t j = 1; j <= D; j++)
{
if (dofmaps[i]->num_entity_dofs(j)!=0)
{
dolfin_error("Assembler.cpp",
"assemble form over vertices",
"Expecting test and trial spaces to only have dofs on " \
"vertices for point integrals");
}
}
// Resize local values so it can hold dofs on one vertex
local_values.resize(local_values.size()*dofmaps[i]->num_entity_dofs(0));
// Resize local to local map according to the number of vertex
// entities dofs
local_to_local_dofs[i].resize(dofmaps[i]->num_entity_dofs(0));
// Resize local dof map vector
global_dofs[i].resize(dofmaps[i]->num_entity_dofs(0));
// Get size of local dofs
local_dof_size[i] = dofmaps[i]->ownership_range().second
- dofmaps[i]->ownership_range().first;
// Get pointer to global dofs
global_dofs_p[i].set(global_dofs[i]);
}
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Exterior point integral
const ufc::vertex_integral* integral
= ufc.default_vertex_integral.get();
// Check whether integral is domain-dependent
bool use_domains = domains && !domains->empty();
// MPI rank
const unsigned int my_mpi_rank = MPI::rank(mesh.mpi_comm());
// Assemble over vertices
ufc::cell ufc_cell;
std::vector<double> coordinate_dofs;
Progress p(AssemblerBase::progress_message(A.rank(), "vertices"),
mesh.num_vertices());
for (VertexIterator vert(mesh); !vert.end(); ++vert)
{
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_vertex_integral((*domains)[*vert]);
// Skip integral if zero
if (!integral)
continue;
// Check if assembling a scalar and a vertex is shared
if (form_rank == 0 && has_shared_vertices)
{
// Find shared processes for this global vertex
std::map<unsigned int, std::set<unsigned int>>::const_iterator e;
e = shared_vertices.find(vert->index());
// If vertex is shared and this rank is not the lowest do not
// include the contribution from this vertex to scalar sum
if (e != shared_vertices.end())
{
bool skip_vertex = false;
std::set<unsigned int>::const_iterator it;
for (it = e->second.begin(); it != e->second.end(); it++)
{
// Check if a shared vertex has a lower process rank
if (*it < my_mpi_rank)
{
skip_vertex = true;
break;
}
}
if (skip_vertex)
continue;
}
}
// Get mesh cell to which mesh vertex belongs (pick first)
Cell mesh_cell(mesh, vert->entities(D)[0]);
// Check that cell is not a ghost
dolfin_assert(!mesh_cell.is_ghost());
// Get local index of vertex with respect to the cell
const std::size_t local_vertex = mesh_cell.index(*vert);
// Update UFC cell
mesh_cell.get_cell_data(ufc_cell);
mesh_cell.get_coordinate_dofs(coordinate_dofs);
// Update UFC object
ufc.update(mesh_cell, coordinate_dofs, ufc_cell,
integral->enabled_coefficients());
// Tabulate vertex tensor
integral->tabulate_tensor(ufc.A.data(),
ufc.w(),
coordinate_dofs.data(),
local_vertex,
ufc_cell.orientation);
// For rank 1 and 2 tensors we need to check if tabulated dofs for
// the test space is within the local range
bool owns_all_dofs = true;
for (std::size_t i = 0; i < form_rank; ++i)
{
// Get local-to-global dof maps for cell
dofs[i] = dofmaps[i]->cell_dofs(mesh_cell.index());
// Get local dofs of the local vertex
dofmaps[i]->tabulate_entity_dofs(local_to_local_dofs[i], 0, local_vertex);
// Copy cell dofs to local dofs and check owner ship range
for (std::size_t j = 0; j < local_to_local_dofs[i].size(); ++j)
{
global_dofs[i][j] = dofs[i][local_to_local_dofs[i][j]];
// It is the dofs for the test space that determines if a dof
// is owned by a process, therefore i==0
if (i == 0 && global_dofs[i][j] >= local_dof_size[i])
{
owns_all_dofs = false;
break;
}
}
}
// If not owning all dofs
if (!owns_all_dofs)
continue;
// Scalar
if (form_rank == 0)
{
// Add entries to global tensor
A.add_local(ufc.A.data(), dofs);
}
else if (form_rank == 1)
{
// Copy tabulated tensor to local value vector
for (std::size_t i = 0; i < local_to_local_dofs[0].size(); ++i)
local_values[i] = ufc.A[local_to_local_dofs[0][i]];
// Add local entries to global tensor
A.add_local(local_values.data(), global_dofs_p);
}
else
{
// Copy tabulated tensor to local value vector
const std::size_t num_cols = dofs[1].size();
for (std::size_t i = 0; i < local_to_local_dofs[0].size(); ++i)
{
for (std::size_t j = 0; j < local_to_local_dofs[1].size(); ++j)
{
local_values[i*local_to_local_dofs[1].size() + j]
= ufc.A[local_to_local_dofs[0][i]*num_cols
+ local_to_local_dofs[1][j]];
}
}
// Add local entries to global tensor
A.add_local(local_values.data(), global_dofs_p);
}
p++;
}
}
//-----------------------------------------------------------------------------
void OpenMpAssembler::assemble_cells(
GenericTensor& A, const Form& a,
UFC& _ufc,
std::shared_ptr<const MeshFunction<std::size_t>> domains,
std::vector<double>* values)
{
// Skip assembly if there are no cell integrals
if (!_ufc.form.has_cell_integrals())
return;
Timer timer("Assemble cells");
// Set number of OpenMP threads (from parameter systems)
const std::size_t num_threads = parameters["num_threads"];
omp_set_num_threads(num_threads);
// Extract mesh
const Mesh& mesh = a.mesh();
// FIXME: Check that UFC copy constructor is dealing with copying
// pointers correctly
// Dummy UFC object since each thread needs to created its own UFC object
UFC ufc(_ufc);
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Cell integral
const ufc::cell_integral* integral = ufc.default_cell_integral.get();
// Check whether integral is domain-dependent
bool use_domains = domains && !domains->empty();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// Color mesh
std::vector<std::size_t> coloring_type = a.coloring(mesh.topology().dim());
mesh.color(coloring_type);
// Get coloring data
std::map<const std::vector<std::size_t>,
std::pair<std::vector<std::size_t>,
std::vector<std::vector<std::size_t>>>>::const_iterator
mesh_coloring;
mesh_coloring = mesh.topology().coloring.find(coloring_type);
if (mesh_coloring == mesh.topology().coloring.end())
{
dolfin_error("OpenMPAssembler.cpp",
"perform multithreaded assembly using OpenMP assembler",
"Requested mesh coloring has not been computed");
}
// Get coloring data
const std::vector<std::vector<std::size_t>>& entities_of_color
= mesh_coloring->second.second;
// If assembling a scalar we need to ensure each threads assemble
// its own scalar
std::vector<double> scalars(num_threads, 0.0);
// Assemble over cells (loop over colours, then cells of same color)
const std::size_t num_colors = entities_of_color.size();
Progress p("Assembling cells (threaded)", num_colors);
for (std::size_t color = 0; color < num_colors; ++color)
{
// Get the array of cell indices of current color
const std::vector<std::size_t>& colored_cells = entities_of_color[color];
// Number of cells of current color
const int num_cells = colored_cells.size();
ufc::cell ufc_cell;
std::vector<double> vertex_coordinates;
// OpenMP test loop over cells of the same color
#pragma omp parallel for schedule(guided, 20) firstprivate(ufc, ufc_cell, vertex_coordinates, dofs, integral)
for (int cell_index = 0; cell_index < num_cells; ++cell_index)
{
// Cell index
const std::size_t index = colored_cells[cell_index];
// Create cell
const Cell cell(mesh, index);
// Get integral for sub domain (if any)
if (use_domains)
integral = ufc.get_cell_integral((*domains)[cell]);
// Skip integral if zero
if (!integral)
continue;
// Update to current cell
cell.get_cell_data(ufc_cell);
cell.get_vertex_coordinates(vertex_coordinates);
ufc.update(cell, vertex_coordinates, ufc_cell,
integral->enabled_coefficients());
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
dofs[i] = dofmaps[i]->cell_dofs(index);
// Tabulate cell tensor
integral->tabulate_tensor(ufc.A.data(),
ufc.w(),
vertex_coordinates.data(),
ufc_cell.orientation);
// Add entries to global tensor
if (values && form_rank == 0)
(*values)[cell_index] = ufc.A[0];
else if (form_rank == 0)
scalars[omp_get_thread_num()] += ufc.A[0];
else
A.add_local(ufc.A.data(), dofs);
}
p++;
}
// If we assemble a scalar we need to sum the contributions from each thread
if (form_rank == 0)
{
const double scalar_sum = std::accumulate(scalars.begin(), scalars.end(),
0.0);
A.add_local(&scalar_sum, dofs);
}
}
//----------------------------------------------------------------------------
void Assembler::assemble(GenericTensor& A, const Form& a)
{
// All assembler functions above end up calling this function, which
// in turn calls the assembler functions below to assemble over
// cells, exterior and interior facets.
// Check whether we should call the multi-core assembler
#ifdef HAS_OPENMP
const std::size_t num_threads = parameters["num_threads"];
if (num_threads > 0)
{
OpenMpAssembler assembler;
assembler.add_values = add_values;
assembler.finalize_tensor = finalize_tensor;
assembler.keep_diagonal = keep_diagonal;
assembler.assemble(A, a);
return;
}
#endif
// Get cell domains
std::shared_ptr<const MeshFunction<std::size_t>>
cell_domains = a.cell_domains();
// Get exterior facet domains
std::shared_ptr<const MeshFunction<std::size_t>> exterior_facet_domains
= a.exterior_facet_domains();
// Get interior facet domains
std::shared_ptr<const MeshFunction<std::size_t>> interior_facet_domains
= a.interior_facet_domains();
// Get vertex domains
std::shared_ptr<const MeshFunction<std::size_t>> vertex_domains
= a.vertex_domains();
// Check form
AssemblerBase::check(a);
// Create data structure for local assembly data
UFC ufc(a);
// Update off-process coefficients
const std::vector<std::shared_ptr<const GenericFunction>>
coefficients = a.coefficients();
// Initialize global tensor
init_global_tensor(A, a);
// Assemble over cells
assemble_cells(A, a, ufc, cell_domains, NULL);
// Assemble over exterior facets
assemble_exterior_facets(A, a, ufc, exterior_facet_domains, NULL);
// Assemble over interior facets
assemble_interior_facets(A, a, ufc, interior_facet_domains,
cell_domains, NULL);
// Assemble over vertices
assemble_vertices(A, a, ufc, vertex_domains);
// Finalize assembly of global tensor
if (finalize_tensor)
A.apply("add");
}
//-----------------------------------------------------------------------------
void OpenMpAssembler::assemble_cells_and_exterior_facets(
GenericTensor& A,
const Form& a, UFC& _ufc,
std::shared_ptr<const MeshFunction<std::size_t>> cell_domains,
std::shared_ptr<const MeshFunction<std::size_t>> exterior_facet_domains,
std::vector<double>* values)
{
Timer timer("Assemble cells and exterior facets");
// Set number of OpenMP threads (from parameter systems)
const int num_threads = parameters["num_threads"];
omp_set_num_threads(num_threads);
// Extract mesh
const Mesh& mesh = a.mesh();
// Compute facets and facet - cell connectivity if not already computed
const std::size_t D = mesh.topology().dim();
mesh.init(D - 1);
mesh.init(D - 1, D);
dolfin_assert(mesh.ordered());
// Get connectivity
const MeshConnectivity& connectivity = mesh.topology()(D, D - 1);
dolfin_assert(!connectivity.empty());
// Dummy UFC object since each thread needs to created its own UFC object
UFC ufc(_ufc);
// Form rank
const std::size_t form_rank = ufc.form.rank();
// Cell and facet integrals
ufc::cell_integral* cell_integral = ufc.default_cell_integral.get();
ufc::exterior_facet_integral* facet_integral
= ufc.default_exterior_facet_integral.get();
// Check whether integrals are domain-dependent
bool use_cell_domains = cell_domains && !cell_domains->empty();
bool use_exterior_facet_domains
= exterior_facet_domains && !exterior_facet_domains->empty();
// Collect pointers to dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof maps for a cell
std::vector<ArrayView<const dolfin::la_index>> dofs(form_rank);
// FIXME: Pass or determine coloring type
// Define graph type
std::vector<std::size_t> coloring_type = a.coloring(mesh.topology().dim());
mesh.color(coloring_type);
// Get coloring data
std::map<const std::vector<std::size_t>,
std::pair<std::vector<std::size_t>,
std::vector<std::vector<std::size_t>>>>::const_iterator
mesh_coloring;
mesh_coloring = mesh.topology().coloring.find(coloring_type);
if (mesh_coloring == mesh.topology().coloring.end())
{
dolfin_error("OpenMPAssembler.cpp",
"perform multithreaded assembly using OpenMP assembler",
"Requested mesh coloring has not been computed");
}
// Get coloring data
const std::vector<std::vector<std::size_t>>& entities_of_color
= mesh_coloring->second.second;
// If assembling a scalar we need to ensure each threads assemble
// its own scalar
std::vector<double> scalars(num_threads, 0.0);
// UFC cell and vertex coordinates
ufc::cell ufc_cell;
std::vector<double> vertex_coordinates;
// Assemble over cells (loop over colors, then cells of same color)
const std::size_t num_colors = entities_of_color.size();
for (std::size_t color = 0; color < num_colors; ++color)
{
// Get the array of cell indices of current color
const std::vector<std::size_t>& colored_cells = entities_of_color[color];
// Number of cells of current color
const int num_cell_in_color = colored_cells.size();
// OpenMP test loop over cells of the same color
Progress p(AssemblerBase::progress_message(A.rank(), "cells"), num_colors);
#pragma omp parallel for schedule(guided, 20) firstprivate(ufc, ufc_cell, vertex_coordinates, dofs, cell_integral, facet_integral)
for (int index = 0; index < num_cell_in_color; ++index)
{
// Cell index
const std::size_t cell_index = colored_cells[index];
// Create cell
const Cell cell(mesh, cell_index);
// Get integral for sub domain (if any)
if (use_cell_domains)
cell_integral = ufc.get_cell_integral((*cell_domains)[cell_index]);
// Update to current cell
cell.get_cell_data(ufc_cell);
cell.get_vertex_coordinates(vertex_coordinates);
// Get local-to-global dof maps for cell
for (std::size_t i = 0; i < form_rank; ++i)
dofs[i] = dofmaps[i]->cell_dofs(cell_index);
// Get number of entries in cell tensor
std::size_t dim = 1;
for (std::size_t i = 0; i < form_rank; ++i)
dim *= dofs[i].size();
// Tabulate cell tensor if we have a cell_integral
if (cell_integral)
{
ufc.update(cell, vertex_coordinates, ufc_cell,
cell_integral->enabled_coefficients());
cell_integral->tabulate_tensor(ufc.A.data(),
ufc.w(),
vertex_coordinates.data(),
ufc_cell.orientation);
}
else
std::fill(ufc.A.begin(), ufc.A.end(), 0.0);
// Assemble over external facet
for (FacetIterator facet(cell); !facet.end(); ++facet)
{
// Only consider exterior facets
if (!facet->exterior())
{
p++;
continue;
}
// Get local facet index
const std::size_t local_facet = cell.index(*facet);
// Get integral for sub domain (if any)
if (use_exterior_facet_domains)
{
// Get global facet index
const std::size_t facet_index = connectivity(cell_index)[local_facet];
facet_integral = ufc.get_exterior_facet_integral((*exterior_facet_domains)[facet_index]);
}
// Skip integral if zero
if (!facet_integral)
continue;
// FIXME: Do we really need an update version with the local
// facet index?
// Update UFC object
ufc_cell.local_facet = local_facet;
ufc.update(cell, vertex_coordinates, ufc_cell,
facet_integral->enabled_coefficients());
// Tabulate tensor
facet_integral->tabulate_tensor(ufc.A_facet.data(),
ufc.w(),
vertex_coordinates.data(),
local_facet,
ufc_cell.orientation);
// Add facet contribution
for (std::size_t i = 0; i < dim; ++i)
ufc.A[i] += ufc.A_facet[i];
}
// Add entries to global tensor
if (values && form_rank == 0)
(*values)[cell_index] = ufc.A[0];
else if (form_rank == 0)
scalars[omp_get_thread_num()] += ufc.A[0];
else
A.add_local(&ufc.A[0], dofs);
}
p++;
}
// If we assemble a scalar we need to sum the contributions from each thread
if (form_rank == 0)
{
const double scalar_sum = std::accumulate(scalars.begin(),
scalars.end(), 0.0);
A.add_local(&scalar_sum, dofs);
}
}
//-----------------------------------------------------------------------------
void AssemblerBase::init_global_tensor(GenericTensor& A, const Form& a)
{
dolfin_assert(a.ufc_form());
// Get dof maps
std::vector<const GenericDofMap*> dofmaps;
for (std::size_t i = 0; i < a.rank(); ++i)
dofmaps.push_back(a.function_space(i)->dofmap().get());
if (A.empty())
{
Timer t0("Build sparsity");
// Create layout for initialising tensor
std::shared_ptr<TensorLayout> tensor_layout;
tensor_layout = A.factory().create_layout(a.rank());
dolfin_assert(tensor_layout);
// Get dimensions and mapping across processes for each dimension
std::vector<std::shared_ptr<const IndexMap> > index_maps;
for (std::size_t i = 0; i < a.rank(); i++)
{
dolfin_assert(dofmaps[i]);
index_maps.push_back(dofmaps[i]->index_map());
}
// Initialise tensor layout
// FIXME: somewhere need to check block sizes are same on both axes
// NOTE: Jan: that will be done on the backend side; IndexMap will
// provide tabulate functions with arbitrary block size;
// moreover the functions will tabulate directly using a
// correct int type
tensor_layout->init(a.mesh().mpi_comm(), index_maps,
TensorLayout::Ghosts::UNGHOSTED);
// Build sparsity pattern if required
if (tensor_layout->sparsity_pattern())
{
GenericSparsityPattern& pattern = *tensor_layout->sparsity_pattern();
SparsityPatternBuilder::build(pattern,
a.mesh(), dofmaps,
a.ufc_form()->has_cell_integrals(),
a.ufc_form()->has_interior_facet_integrals(),
a.ufc_form()->has_exterior_facet_integrals(),
a.ufc_form()->has_vertex_integrals(),
keep_diagonal);
}
t0.stop();
// Initialize tensor
Timer t1("Init tensor");
A.init(*tensor_layout);
t1.stop();
// Insert zeros on the diagonal as diagonal entries may be
// prematurely optimised away by the linear algebra backend when
// calling GenericMatrix::apply, e.g. PETSc does this then errors
// when matrices have no diagonal entry inserted.
if (A.rank() == 2 && keep_diagonal)
{
// Down cast to GenericMatrix
GenericMatrix& _matA = A.down_cast<GenericMatrix>();
// Loop over rows and insert 0.0 on the diagonal
const double block = 0.0;
const std::pair<std::size_t, std::size_t> row_range = A.local_range(0);
const std::size_t range = std::min(row_range.second, A.size(1));
for (std::size_t i = row_range.first; i < range; i++)
{
dolfin::la_index _i = i;
_matA.set(&block, 1, &_i, 1, &_i);
}
A.apply("flush");
}
// Delete sparsity pattern
Timer t2("Delete sparsity");
t2.stop();
}
else
{
// If tensor is not reset, check that dimensions are correct
for (std::size_t i = 0; i < a.rank(); ++i)
{
if (A.size(i) != dofmaps[i]->global_dimension())
{
dolfin_error("AssemblerBase.cpp",
"assemble form",
"Dim %d of tensor does not match form", i);
}
}
}
if (!add_values)
A.zero();
}
//-----------------------------------------------------------------------------
void MultiMeshAssembler::_assemble_interface(GenericTensor& A,
const MultiMeshForm& a)
{
// Extract multimesh
std::shared_ptr<const MultiMesh> multimesh = a.multimesh();
// Get form rank
const std::size_t form_rank = a.rank();
// Get multimesh coefficients
// These are updated in within this assembly loop
std::map<std::size_t, std::shared_ptr<const MultiMeshFunction> >
multimesh_coefficients = a.multimesh_coefficients();
// Identify the coefficents that are not MultiMeshFunction
// These will be updated by UFC
// It is assumed that the coefficents are the same for all parts
std::vector<bool> ufc_enabled_coefficients;
for (std::size_t i = 0; i < a.part(0)->coefficients().size(); i++)
{
bool ufc_update = multimesh_coefficients.find(i) == multimesh_coefficients.end();
ufc_enabled_coefficients.push_back(ufc_update);
}
// Collect pointers to dof maps
std::vector<const MultiMeshDofMap*> dofmaps;
for (std::size_t i = 0; i < form_rank; i++)
dofmaps.push_back(a.function_space(i)->dofmap().get());
// Vector to hold dof map for a cell
std::vector<const std::vector<dolfin::la_index>* > dofs(form_rank);
// Initialize variables that will be reused throughout assembly
ufc::cell ufc_cell[2];
std::vector<double> coordinate_dofs[2];
std::vector<double> macro_coordinate_dofs;
// Vector to hold dofs for cells, and a vector holding pointers to same
std::vector<ArrayView<const dolfin::la_index>> macro_dof_ptrs(form_rank);
std::vector<std::vector<dolfin::la_index>> macro_dofs(form_rank);
// Iterate over parts
for (std::size_t part = 0; part < a.num_parts(); part++)
{
log(PROGRESS, "Assembling multimesh form over interface on part %d.",
part);
// Get form for current part
const Form& a_part = *a.part(part);
// Create data structure for local assembly data
UFC ufc_part(a_part);
// FIXME: Handle subdomains
// Get integral
ufc::interface_integral* integral = ufc_part.default_interface_integral.get();
// Skip if we don't have an interface integral
if (!integral) continue;
// Get quadrature rules
const auto& quadrature_rules = multimesh->quadrature_rules_interface(part);
// Get collision map
const auto& cmap = multimesh->collision_map_cut_cells(part);
// Get facet normals
const auto& facet_normals = multimesh->facet_normals(part);
// Iterate over all cut cells in collision map
for (auto it = cmap.begin(); it != cmap.end(); ++it)
{
// Get cut cell
const unsigned int cut_cell_index = it->first;
const Cell cut_cell(*multimesh->part(part), cut_cell_index);
// Iterate over cutting cells
const auto& cutting_cells = it->second;
for (auto jt = cutting_cells.begin(); jt != cutting_cells.end(); jt++)
{
// Get cutting part and cutting cell
const std::size_t cutting_part = jt->first;
const std::size_t cutting_cell_index = jt->second;
const Cell cutting_cell(*multimesh->part(cutting_part),
cutting_cell_index);
// Get quadrature rule for interface part defined by
// intersection of the cut and cutting cells
const std::size_t k = jt - cutting_cells.begin();
dolfin_assert(k < quadrature_rules.at(cut_cell_index).size());
const auto& qr = quadrature_rules.at(cut_cell_index)[k];
// FIXME: There might be quite a few cases when we skip cutting
// FIXME: cells because there are no quadrature points. Perhaps
// FIXME: we can rewrite this inner loop to avoid unnecessary
// FIXME: iterations.
// Skip if there are no quadrature points
const std::size_t num_quadrature_points = qr.second.size();
if (num_quadrature_points == 0)
continue;
// Create aliases for cells to simplify notation
const std::size_t& part_1 = cutting_part;
const std::size_t& part_0 = part;
const Cell& cell_1 = cutting_cell;
const Cell& cell_0 = cut_cell;
// Update to current pair of cells
// Let UFC update the coefficients that are not MultiMeshFunction
cell_0.get_cell_data(ufc_cell[0], 0);
cell_1.get_cell_data(ufc_cell[1], 0);
cell_0.get_coordinate_dofs(coordinate_dofs[0]);
cell_1.get_coordinate_dofs(coordinate_dofs[1]);
ufc_part.update(cell_0, coordinate_dofs[0], ufc_cell[0],
cell_1, coordinate_dofs[1], ufc_cell[1],
ufc_enabled_coefficients);
// Manually update multimesh coefficients
for (auto it : multimesh_coefficients)
{
std::size_t coefficient_number = it.first;
std::shared_ptr<const MultiMeshFunction> coefficient = it.second;
double** macro_w = ufc_part.macro_w();
const FiniteElement& element = *coefficient->function_space()->part(part_0)->element();
std::size_t offset = element.space_dimension();
double * w_0 = macro_w[coefficient_number];
double * w_1 = macro_w[coefficient_number] + offset;
coefficient->restrict(w_0, element,
part_0, cell_0, coordinate_dofs[0].data(), ufc_cell[0]);
coefficient->restrict(w_1, element,
part_1, cell_1, coordinate_dofs[1].data(), ufc_cell[1]);
}
// Collect vertex coordinates
macro_coordinate_dofs.resize(coordinate_dofs[0].size() +
coordinate_dofs[1].size());
std::copy(coordinate_dofs[0].begin(), coordinate_dofs[0].end(),
macro_coordinate_dofs.begin());
std::copy(coordinate_dofs[1].begin(), coordinate_dofs[1].end(),
macro_coordinate_dofs.begin() + coordinate_dofs[0].size());
// Tabulate dofs for each dimension on macro element
for (std::size_t i = 0; i < form_rank; i++)
{
// Get dofs for cut mesh
const auto dofmap_0 = a.function_space(i)->dofmap()->part(part_0);
const auto dofs_0 = dofmap_0->cell_dofs(cell_0.index());
// Get dofs for cutting mesh
const auto dofmap_1 = a.function_space(i)->dofmap()->part(part_1);
const auto dofs_1 = dofmap_1->cell_dofs(cell_1.index());
// Create space in macro dof vector
macro_dofs[i].resize(dofs_0.size() + dofs_1.size());
// Copy cell dofs into macro dof vector
std::copy(dofs_0.data(), dofs_0.data() + dofs_0.size(),
macro_dofs[i].begin());
std::copy(dofs_1.data(), dofs_1.data() + dofs_1.size(),
macro_dofs[i].begin() + dofs_0.size());
// Update array view
macro_dof_ptrs[i]
= ArrayView<const dolfin::la_index>(macro_dofs[i].size(),
macro_dofs[i].data());
}
// Get facet normals
const auto& n = facet_normals.at(cut_cell_index)[k];
// FIXME: We would like to use this assertion (but it fails
// for 2 meshes)
dolfin_assert(n.size() == a_part.mesh()->geometry().dim()*num_quadrature_points);
// FIXME: For now, use this assertion (which fails for 3 meshes)
//dolfin_assert(n.size() > 0);
// FIXME: Cell orientation not supported
const int cell_orientation = ufc_cell[0].orientation;
// Tabulate interface tensor on macro element
integral->tabulate_tensor(ufc_part.macro_A.data(),
ufc_part.macro_w(),
macro_coordinate_dofs.data(),
num_quadrature_points,
qr.first.data(),
qr.second.data(),
n.data(),
cell_orientation);
// Add entries to global tensor
A.add(ufc_part.macro_A.data(), macro_dof_ptrs);
}
}
}
}
