Zoltan2
|
MatrixAdapter defines the adapter interface for matrices. More...
#include <Zoltan2_MatrixAdapter.hpp>
Public Member Functions | |
enum BaseAdapterType | adapterType () const |
Returns the type of adapter. More... | |
MatrixAdapter () | |
virtual | ~MatrixAdapter () |
Destructor. More... | |
virtual size_t | getLocalNumRows () const =0 |
Returns the number of rows on this process. More... | |
virtual size_t | getLocalNumColumns () const =0 |
Returns the number of columns on this process. More... | |
virtual size_t | getLocalNumEntries () const =0 |
Returns the number of nonzeros on this process. More... | |
virtual bool | CRSViewAvailable () const |
Indicates whether the MatrixAdapter implements a view of the matrix in compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual void | getRowIDsView (const gno_t *&rowIds) const |
Sets pointer to this process' rows' global IDs. More... | |
virtual void | getCRSView (const offset_t *&offsets, const gno_t *&colIds) const |
Sets pointers to this process' matrix entries using compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual void | getCRSView (const offset_t *&offsets, const gno_t *&colIds, const scalar_t *&values) const |
Sets pointers to this process' matrix entries and their values using compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual int | getNumWeightsPerRow () const |
Returns the number of weights per row (0 or greater). Row weights may be used when partitioning matrix rows. More... | |
virtual void | getRowWeightsView (const scalar_t *&weights, int &stride, int idx=0) const |
Provide a pointer to the row weights, if any. More... | |
virtual bool | useNumNonzerosAsRowWeight (int idx) const |
Indicate whether row weight with index idx should be the global number of nonzeros in the row. More... | |
virtual bool | CCSViewAvailable () const |
Indicates whether the MatrixAdapter implements a view of the matrix in compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual void | getColumnIDsView (const gno_t *&colIds) const |
Sets pointer to this process' columns' global IDs. More... | |
virtual void | getCCSView (const offset_t *&offsets, const gno_t *&rowIds) const |
Sets pointers to this process' matrix entries using compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual void | getCCSView (const offset_t *&offsets, const gno_t *&rowIds, const scalar_t *&values) const |
Sets pointers to this process' matrix entries and their values using compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required. More... | |
virtual int | getNumWeightsPerColumn () const |
Returns the number of weights per column (0 or greater). Column weights may be used when partitioning matrix columns. More... | |
virtual void | getColumnWeightsView (const scalar_t *&weights, int &stride, int idx=0) const |
Provide a pointer to the column weights, if any. More... | |
virtual bool | useNumNonzerosAsColumnWeight (int idx) const |
Indicate whether column weight with index idx should be the global number of nonzeros in the column. More... | |
void | setCoordinateInput (VectorAdapter< UserCoord > *coordData) |
Allow user to provide additional data that contains coordinate info associated with the MatrixAdapter's primaryEntityType. Associated data must have the same parallel distribution and ordering of entries as the primaryEntityType. More... | |
bool | coordinatesAvailable () const |
Indicate whether coordinate information has been set for this MatrixAdapter. More... | |
VectorAdapter< UserCoord > * | getCoordinateInput () const |
Obtain the coordinate data registered by the user. More... | |
enum MatrixEntityType | getPrimaryEntityType () const |
Returns the entity to be partitioned, ordered, colored, etc. Valid values are MATRIX_ROW, MATRIX_COLUMN, MATRIX_NONZERO. More... | |
void | setPrimaryEntityType (std::string typestr) |
Sets the primary entity type. Called by algorithm based on parameter value in parameter list from application. Also sets to adjacencyEntityType to something reasonable: opposite of primaryEntityType. More... | |
size_t | getLocalNumIDs () const |
Returns the number of objects on this process. More... | |
void | getIDsView (const gno_t *&Ids) const |
Provide a pointer to this process' identifiers. More... | |
int | getNumWeightsPerID () const |
Returns the number of weights per object. Number of weights per object should be zero or greater. If zero, then it is assumed that all objects are equally weighted. Default is zero weights per ID. More... | |
void | getWeightsView (const scalar_t *&wgt, int &stride, int idx=0) const |
Provide pointer to a weight array with stride. More... | |
bool | useDegreeAsWeight (int idx) const |
Public Member Functions inherited from Zoltan2::BaseAdapter< User > | |
virtual | ~BaseAdapter () |
Destructor. More... | |
virtual void | getIDsKokkosView (Kokkos::View< gno_t * > &) const |
Provide a pointer to this process' identifiers. More... | |
virtual void | getWeightsKokkosView (Kokkos::View< scalar_t * > &, int=0) const |
Provide pointer to a weight View. More... | |
void | getPartsView (const part_t *&inputPart) const |
Provide pointer to an array containing the input part assignment for each ID. The input part information may be used for re-partitioning to reduce data movement, or for mapping parts to processes. Adapters may return NULL for this pointer (the default behavior); if NULL is returned, algorithms will assume the rank. More... | |
template<typename Adapter > | |
void | applyPartitioningSolution (const User &in, User *&out, const PartitioningSolution< Adapter > &solution) const |
Apply a PartitioningSolution to an input. More... | |
Public Member Functions inherited from Zoltan2::BaseAdapterRoot | |
virtual | ~BaseAdapterRoot () |
Additional Inherited Members | |
Public Types inherited from Zoltan2::BaseAdapter< User > | |
typedef InputTraits< User >::lno_t | lno_t |
typedef InputTraits< User >::gno_t | gno_t |
typedef InputTraits< User > ::scalar_t | scalar_t |
typedef InputTraits< User >::part_t | part_t |
typedef InputTraits< User > ::offset_t | offset_t |
Protected Member Functions inherited from Zoltan2::BaseAdapter< User > | |
void | generateWeightFileOnly (const char *fileprefix, const Teuchos::Comm< int > &comm) const |
MatrixAdapter defines the adapter interface for matrices.
Adapter objects provide access for Zoltan2 to the user's data. Many built-in adapters are already defined for common data structures, such as Tpetra and Epetra objects and C-language pointers to arrays.
Data types:
scalar_t
row, column or non-zero weights lno_t
local indices and local counts gno_t
global indices and global counts node_t
is a Kokkos Node typeThe Kokkos node type can be safely ignored.
The template parameter User
is a user-defined data type which, through a traits mechanism, provides the actual data types with which the Zoltan2 library will be compiled. User
may be the actual class or structure used by application to represent a vector, or it may be the helper class BasicUserTypes. See InputTraits for more information.
The scalar_t
type, representing use data such as matrix values, is used by Zoltan2 for weights, coordinates, part sizes and quality metrics. Some User types (like Tpetra::CrsMatrix) have an inherent scalar type, and some (like Tpetra::CrsGraph) do not. For such objects, the scalar type is set by Zoltan2 to float
. If you wish to change it to double, set the second template parameter to double
.
Create BasicCrsMatrixAdapter subclass
Do we want to require adapters to give us the global number of rows, columns etc? We can figure that out.
This is a row-oriented matrix. Do we need a column-oriented matrix? In particular - we assumed coordinates are for rows.
If the user can tell us there are no diagonal entries in a square matrix, it can save time if we have to remove them for the algorithm. Should we have a set method in subclasses for setMatrixHasDiagonalEntries yes, no and maybe?
Definition at line 106 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Definition at line 129 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Destructor.
Definition at line 135 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Returns the type of adapter.
Implements Zoltan2::BaseAdapter< User >.
Definition at line 126 of file Zoltan2_MatrixAdapter.hpp.
|
pure virtual |
Returns the number of rows on this process.
Implemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
|
pure virtual |
Returns the number of columns on this process.
Implemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
|
pure virtual |
Returns the number of nonzeros on this process.
Implemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
|
inlinevirtual |
Indicates whether the MatrixAdapter implements a view of the matrix in compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
Reimplemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
Definition at line 156 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointer to this process' rows' global IDs.
rowIds | will on return a pointer to row global Ids |
Reimplemented in Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
Definition at line 161 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointers to this process' matrix entries using compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
offsets | is an array of size numRows + 1. The column Ids for rowIds[i] (returned by getRowIDsView) begin at colIds[offsets[i]]. The last element of offsets is the size of the colIds array. |
colIds | on return will point to the global column Ids for the non-zeros for each row. |
Definition at line 178 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointers to this process' matrix entries and their values using compressed sparse row (CRS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
offsets | is an array of size numRows + 1. The column Ids for rowIds[i] (returned by getRowIDsView) begin at colIds[offsets[i]]. The last element of offsets is the size of the colIds array. |
colIds | on return will point to the global column Ids for the non-zeros for each row. |
values | on return will point to the values stored in the non-zeros for each row. |
Definition at line 200 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Returns the number of weights per row (0 or greater). Row weights may be used when partitioning matrix rows.
Reimplemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
Definition at line 214 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Provide a pointer to the row weights, if any.
weights | is the list of weights with a given index for the rows returned in getRowIDsView(). |
stride | The k'th weight is located at weights[stride*k] |
idx | ranges from zero to one less than getNumWeightsPerRow(). |
Definition at line 222 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Indicate whether row weight with index idx should be the global number of nonzeros in the row.
Reimplemented in Zoltan2::XpetraCrsMatrixAdapter< User, UserCoord >, and Zoltan2::TpetraRowMatrixAdapter< User, UserCoord >.
Definition at line 234 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Indicates whether the MatrixAdapter implements a view of the matrix in compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
Definition at line 244 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointer to this process' columns' global IDs.
colIds | will on return a pointer to column global Ids |
Definition at line 249 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointers to this process' matrix entries using compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
offsets | is an array of size numCols + 1. The row Ids for colIds[i] (returned by getColumnIDsView) begin at rowIds[offsets[i]]. The last element of offsets is the size of the rowIds array. |
rowIds | on return will point to the global row Ids for the non-zeros for each column. |
Definition at line 266 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Sets pointers to this process' matrix entries and their values using compressed sparse column (CCS) format. All matrix adapters must implement either getCRSView or getCCSView, but implementation of both is not required.
offsets | is an array of size numCols + 1. The row Ids for colIds[i] (returned by getColumnIDsView) begin at rowIds[offsets[i]]. The last element of offsets is the size of the rowIds array. |
rowIds | on return will point to the global row Ids for the non-zeros for each column. |
values | on return will point to the values stored in the non-zeros for each column. |
Definition at line 289 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Returns the number of weights per column (0 or greater). Column weights may be used when partitioning matrix columns.
Definition at line 303 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Provide a pointer to the column weights, if any.
weights | is the list of weights with a given index for the columns returned in getColumnIDsView(). |
stride | The k'th weight is located at weights[stride*k] |
idx | ranges from zero to one less than getNumWeightsPerColumn(). |
Definition at line 311 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Indicate whether column weight with index idx should be the global number of nonzeros in the column.
Definition at line 323 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Allow user to provide additional data that contains coordinate info associated with the MatrixAdapter's primaryEntityType. Associated data must have the same parallel distribution and ordering of entries as the primaryEntityType.
coordData | is a pointer to a VectorAdapter with the user's coordinate data. |
Definition at line 341 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Indicate whether coordinate information has been set for this MatrixAdapter.
Definition at line 350 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Obtain the coordinate data registered by the user.
Definition at line 355 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Returns the entity to be partitioned, ordered, colored, etc. Valid values are MATRIX_ROW, MATRIX_COLUMN, MATRIX_NONZERO.
Definition at line 366 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Sets the primary entity type. Called by algorithm based on parameter value in parameter list from application. Also sets to adjacencyEntityType to something reasonable: opposite of primaryEntityType.
Definition at line 376 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Returns the number of objects on this process.
Objects may be coordinates, graph vertices, matrix rows, etc. They are the objects to be partitioned, ordered, or colored.
Implements Zoltan2::BaseAdapterRoot.
Definition at line 397 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Provide a pointer to this process' identifiers.
ids | will on return point to the list of the global Ids for this process. |
Reimplemented from Zoltan2::BaseAdapter< User >.
Definition at line 411 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Returns the number of weights per object. Number of weights per object should be zero or greater. If zero, then it is assumed that all objects are equally weighted. Default is zero weights per ID.
Reimplemented from Zoltan2::BaseAdapterRoot.
Definition at line 434 of file Zoltan2_MatrixAdapter.hpp.
|
inlinevirtual |
Provide pointer to a weight array with stride.
wgt | on return a pointer to the weights for this idx |
stride | on return, the value such that the nth weight should be found at wgt[n*stride] . |
idx | the weight index, zero or greater This function must be implemented in derived adapter if getNumWeightsPerID > 0. This function should not be called if getNumWeightsPerID is zero. |
Reimplemented from Zoltan2::BaseAdapter< User >.
Definition at line 448 of file Zoltan2_MatrixAdapter.hpp.
|
inline |
Definition at line 472 of file Zoltan2_MatrixAdapter.hpp.