SparsityPattern.h

Note

The documentation on this page was automatically extracted from the DOLFIN C++ code and may need to be edited or expanded.

class SparsityPattern

This class implements a sparsity pattern data structure. It is used by most linear algebra backends.

enum class Type

Whether SparsityPattern is sorted

SparsityPattern(MPI_Comm comm, std::size_t primary_dim)

Create empty sparsity pattern

SparsityPattern(MPI_Comm comm, std::vector<std::shared_ptr<const IndexMap>> index_maps, std::size_t primary_dim)

Create sparsity pattern for a generic tensor

void init(std::vector<std::shared_ptr<const IndexMap>> index_maps)

Initialize sparsity pattern for a generic tensor

void insert_global(dolfin::la_index i, dolfin::la_index j)

Insert a global entry - will be fixed by apply()

void insert_global(const std::vector<ArrayView<const dolfin::la_index>> &entries)

Insert non-zero entries using global indices

void insert_local(const std::vector<ArrayView<const dolfin::la_index>> &entries)

Insert non-zero entries using local (process-wise) indices

void insert_local_global(const std::vector<ArrayView<const dolfin::la_index>> &entries)

Insert non-zero entries using local (process-wise) indices for the primary dimension and global indices for the co-dimension

void insert_full_rows_local(const std::vector<std::size_t> &rows)

Insert full rows (or columns, according to primary dimension) using local (process-wise) indices. This must be called before any other sparse insertion occurs to avoid quadratic complexity of dense rows insertion

std::size_t rank() const

Return rank

std::size_t primary_dim() const

Return primary dimension (e.g., 0=row partition, 1=column partition)

std::pair<std::size_t, std::size_t> local_range(std::size_t dim) const

Return local range for dimension dim

std::size_t num_nonzeros() const

Return number of local nonzeros

void num_nonzeros_diagonal(std::vector<std::size_t> &num_nonzeros) const

Fill array with number of nonzeros for diagonal block in local_range for dimension 0. For matrices, fill array with number of nonzeros per local row for diagonal block

void num_nonzeros_off_diagonal(std::vector<std::size_t> &num_nonzeros) const

Fill array with number of nonzeros for off-diagonal block in local_range for dimension 0. For matrices, fill array with number of nonzeros per local row for off-diagonal block. If there is no off-diagonal pattern, the vector is resized to zero-length

void num_local_nonzeros(std::vector<std::size_t> &num_nonzeros) const

Fill vector with number of nonzeros in local_range for dimension 0

void apply()

Finalize sparsity pattern

MPI_Comm mpi_comm() const

Return MPI communicator

std::string str(bool verbose) const

Return informal string representation (pretty-print)

std::vector<std::vector<std::size_t>> diagonal_pattern(Type type) const

Return underlying sparsity pattern (diagonal). Options are ‘sorted’ and ‘unsorted’.

std::vector<std::vector<std::size_t>> off_diagonal_pattern(Type type) const

Return underlying sparsity pattern (off-diagonal). Options are ‘sorted’ and ‘unsorted’. Empty vector is returned if there is no off-diagonal contribution.