TrilinosWrappers::MPI::BlockVector Class Reference
[Vector classesTrilinosWrappers]

Inheritance diagram for TrilinosWrappers::MPI::BlockVector:
Inheritance graph
[legend]

List of all members.

Classes

class  ExcIteratorRangeDoesNotMatchVectorSize
class  ExcNonMatchingBlockVectors

Public Types

typedef BlockVectorBase< VectorBaseClass
typedef BaseClass::BlockType BlockType
typedef BaseClass::value_type value_type
typedef BaseClass::pointer pointer
typedef BaseClass::const_pointer const_pointer
typedef BaseClass::reference reference
typedef BaseClass::const_reference const_reference
typedef BaseClass::size_type size_type
typedef BaseClass::iterator iterator
typedef BaseClass::const_iterator const_iterator

Public Member Functions

 BlockVector ()
 BlockVector (const std::vector< Epetra_Map > &InputMaps)
 BlockVector (const BlockVector &V)
 BlockVector (const unsigned int num_blocks)
 ~BlockVector ()
BlockVectoroperator= (const value_type s)
BlockVectoroperator= (const BlockVector &V)
BlockVectoroperator= (const ::::TrilinosWrappers::BlockVector &V)
template<typename Number >
BlockVectoroperator= (const ::::BlockVector< Number > &V)
void reinit (const std::vector< Epetra_Map > &input_maps, const bool fast=false)
void reinit (const BlockVector &V, const bool fast=false)
void reinit (const unsigned int num_blocks)
void import_nonlocal_data_for_fe (const TrilinosWrappers::BlockSparseMatrix &m, const BlockVector &v)
void compress ()
bool is_compressed () const
void swap (BlockVector &v)

Detailed Description

An implementation of block vectors based on the vector class implemented in TrilinosWrappers. While the base class provides for most of the interface, this class handles the actual allocation of vectors and provides functions that are specific to the underlying vector type.

The model of distribution of data is such that each of the blocks is distributed across all MPI processes named in the MPI communicator. I.e. we don't just distribute the whole vector, but each component. In the constructors and reinit() functions, one therefore not only has to specify the sizes of the individual blocks, but also the number of elements of each of these blocks to be stored on the local process.

Author:
Martin Kronbichler, Wolfgang Bangerth, 2008

Member Typedef Documentation

Typedef the base class for simpler access to its own typedefs.

Typedef the type of the underlying vector.

Reimplemented from BlockVectorBase< Vector >.

Import the typedefs from the base class.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.

Reimplemented from BlockVectorBase< Vector >.


Constructor & Destructor Documentation

BlockVector< Number >::BlockVector (  )  [inline]

Default constructor. Generate an empty vector without any blocks.

BlockVector< Number >::BlockVector ( const std::vector< Epetra_Map > &  InputMaps  )  [inline]

Constructor. Generate a block vector with as many blocks as there are entries in Input_Maps. Each Epetra_Map already knows the distribution of data among the MPI processes.

References reinit().

BlockVector< Number >::BlockVector ( const BlockVector V  )  [inline]
BlockVector< Number >::BlockVector ( const unsigned int  num_blocks  )  [inline]

Creates a block vector consisting of num_blocks components, but there is no content in the individual components and the user has to fill appropriate data using a reinit of the blocks.

References reinit().

TrilinosWrappers::MPI::BlockVector::~BlockVector (  ) 

Destructor. Clears memory


Member Function Documentation

BlockVector& TrilinosWrappers::MPI::BlockVector::operator= ( const value_type  s  ) 

Copy operator: fill all components of the vector that are locally stored with the given scalar value.

BlockVector& TrilinosWrappers::MPI::BlockVector::operator= ( const BlockVector V  ) 

Copy operator for arguments of the same type.

Reimplemented from Subscriptor.

BlockVector& TrilinosWrappers::MPI::BlockVector::operator= ( const ::::TrilinosWrappers::BlockVector V  ) 

Copy operator for arguments of the localized Trilinos vector type.

template<typename Number >
BlockVector& TrilinosWrappers::MPI::BlockVector::operator= ( const ::::BlockVector< Number > &  V  )  [inline]

Another copy function. This one takes a deal.II block vector and copies it into a TrilinosWrappers block vector. Note that the number of blocks has to be the same in the vector as in the input vector. Use the reinit() command for resizing the BlockVector or for changing the internal structure of the block components.

Since Trilinos only works on doubles, this function is limited to accept only one possible number type in the deal.II vector.

void TrilinosWrappers::MPI::BlockVector::reinit ( const std::vector< Epetra_Map > &  input_maps,
const bool  fast = false 
)

Reinitialize the BlockVector to contain as many blocks as there are Epetra_Maps given in the input argument, according to the parallel distribution of the individual components described in the maps.

If fast==false, the vector is filled with zeros.

Referenced by BlockVector().

void TrilinosWrappers::MPI::BlockVector::reinit ( const BlockVector V,
const bool  fast = false 
)

Change the dimension to that of the vector V. The same applies as for the other reinit() function.

The elements of V are not copied, i.e. this function is the same as calling reinit (V.size(), fast).

Note that you must call this (or the other reinit() functions) function, rather than calling the reinit() functions of an individual block, to allow the block vector to update its caches of vector sizes. If you call reinit() on one of the blocks, then subsequent actions on this object may yield unpredictable results since they may be routed to the wrong block.

void TrilinosWrappers::MPI::BlockVector::reinit ( const unsigned int  num_blocks  ) 

Change the number of blocks to num_blocks. The individual blocks will get initialized with zero size, so it is assumed that the user resizes the individual blocks by herself in an appropriate way, and calls collect_sizes afterwards.

void TrilinosWrappers::MPI::BlockVector::import_nonlocal_data_for_fe ( const TrilinosWrappers::BlockSparseMatrix m,
const BlockVector v 
)

This reinit function is meant to be used for parallel calculations where some non-local data has to be used. The typical situation where one needs this function is the call of the FEValues<dim>::get_function_values function (or of some derivatives) in parallel. Since it is usually faster to retrieve the data in advance, this function can be called before the assembly forks out to the different processors. What this function does is the following: It takes the information in the columns of the given matrix and looks which data couples between the different processors. That data is then queried from the input vector. Note that you should not write to the resulting vector any more, since the some data can be stored several times on different processors, leading to unpredictable results. In particular, such a vector cannot be used for matrix-vector products as for example done during the solution of linear systems.

void TrilinosWrappers::MPI::BlockVector::compress (  ) 

Compresses all the components after assembling together all elements.

Reimplemented from BlockVectorBase< Vector >.

bool BlockVector< Number >::is_compressed (  )  const [inline]

Returns the state of the vector, i.e., whether compress() needs to be called after an operation requiring data exchange. Does only return non-true values when used in debug mode, since it is quite expensive to keep track of all operations that lead to the need for compress().

References BlockVectorBase< Vector >::block(), and BlockVectorBase< Vector >::n_blocks().

void BlockVector< Number >::swap ( BlockVector v  )  [inline]

Swap the contents of this vector and the other vector v. One could do this operation with a temporary variable and copying over the data elements, but this function is significantly more efficient since it only swaps the pointers to the data of the two vectors and therefore does not need to allocate temporary storage and move data around.

Limitation: right now this function only works if both vectors have the same number of blocks. If needed, the numbers of blocks should be exchanged, too.

This function is analog to the the swap() function of all C++ standard containers. Also, there is a global function swap(u,v) that simply calls u.swap(v), again in analogy to standard functions.

References Assert, BlockVectorBase< VectorType >::block(), BlockVectorBase< Vector >::block(), BlockVectorBase< VectorType >::n_blocks(), BlockVectorBase< Vector >::n_blocks(), and Vector< Number >::swap().


The documentation for this class was generated from the following file:

deal.II documentation generated on Mon Nov 23 22:58:41 2009 by doxygen 1.6.1