|
AbstractLinAlgPack: C++ Interfaces For Vectors, Matrices And Related Linear Algebra Objects
Version of the Day
|
All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Groups Pages
|
AbstractLinAlgPack: C++ Interfaces For Vectors, Matrices And Related Linear Algebra Objects
Version of the Day
|
Interface for a collection of non-mutable vectors (multi-vector, matrix). More...
#include <AbstractLinAlgPack_MultiVector.hpp>
Public Types | |
| enum | |
| typedef int | access_by_t |
| typedef Teuchos::RCP< const Vector > | vec_ptr_t |
| typedef Teuchos::RCP< const MultiVector > | multi_vec_ptr_t |
 Public Types inherited from AbstractLinAlgPack::MatrixOp | |
| enum | EMatNormType { MAT_NORM_INF, MAT_NORM_2, MAT_NORM_1, MAT_NORM_FORB } |
| Type of matrix norm. More... | |
Friends | |
| void | apply_op (EApplyBy apply_by, const RTOpPack::RTOp &primary_op, const size_t num_multi_vecs, const MultiVector *multi_vecs[], const size_t num_targ_multi_vecs, MultiVectorMutable *targ_multi_vecs[], RTOpPack::ReductTarget *reduct_objs[], const index_type primary_first_ele, const index_type primary_sub_dim, const index_type primary_global_offset, const index_type secondary_first_ele, const index_type secondary_sub_dim) |
| void | apply_op (EApplyBy apply_by, const RTOpPack::RTOp &primary_op, const RTOpPack::RTOp &secondary_op, const size_t num_multi_vecs, const MultiVector *multi_vecs[], const size_t num_targ_multi_vecs, MultiVectorMutable *targ_multi_vecs[], RTOpPack::ReductTarget *reduct_obj, const index_type primary_first_ele, const index_type primary_sub_dim, const index_type primary_global_offset, const index_type secondary_first_ele, const index_type secondary_sub_dim) |
Clone | |
| virtual multi_vec_ptr_t | mv_clone () const |
| Clone the non-const multi-vector object. More... | |
Provide row, column and diagonal access as non-mutable vectors | |
| virtual access_by_t | access_by () const =0 |
| Return a bit field for the types of access that are the most convenient. More... | |
| virtual vec_ptr_t | col (index_type j) const =0 |
| Get a non-mutable column vector. More... | |
| virtual vec_ptr_t | row (index_type i) const =0 |
| Get a non-mutable row vector. More... | |
| virtual vec_ptr_t | diag (int k) const =0 |
| Get a non-mutable diagonal vector. More... | |
Sub-view methods | |
| virtual multi_vec_ptr_t | mv_sub_view (const Range1D &row_rng, const Range1D &col_rng) const |
| Returns a sub-view of the multi vector. More... | |
| multi_vec_ptr_t | mv_sub_view (const index_type &rl, const index_type &ru, const index_type &cl, const index_type &cu) const |
Inlined implementation calls this->mv_sub_view(Range1D(rl,ru),Range1D(cl,cu)). More... | |
Collective apply_op() methods | |
| virtual void | apply_op (EApplyBy apply_by, const RTOpPack::RTOp &primary_op, const size_t num_multi_vecs, const MultiVector *multi_vecs[], const size_t num_targ_multi_vecs, MultiVectorMutable *targ_multi_vecs[], RTOpPack::ReductTarget *reduct_objs[], const index_type primary_first_ele, const index_type primary_sub_dim, const index_type primary_global_offset, const index_type secondary_first_ele, const index_type secondary_sub_dim) const |
| Apply a reduction/transformation operator row by row, or column by column and return an array of the reduction objects. More... | |
| virtual void | apply_op (EApplyBy apply_by, const RTOpPack::RTOp &primary_op, const RTOpPack::RTOp &secondary_op, const size_t num_multi_vecs, const MultiVector *multi_vecs[], const size_t num_targ_multi_vecs, MultiVectorMutable *targ_multi_vecs[], RTOpPack::ReductTarget *reduct_obj, const index_type primary_first_ele, const index_type primary_sub_dim, const index_type primary_global_offset, const index_type secondary_first_ele, const index_type secondary_sub_dim) const |
| Apply a reduction/transformation operator row by row, or column by column and reduce the intermediate reduction objects into one reduction object. More... | |
Overridden from MatrixOp | |
| mat_ptr_t | clone () const |
Returns this->mv_clone() More... | |
| mat_ptr_t | sub_view (const Range1D &row_rng, const Range1D &col_rng) const |
Returns this->mv_sub_view(row_rng,col_rng) casted to a MatrixOp. More... | |
| bool | Mp_StMtM (MatrixOp *mwo_lhs, value_type alpha, const MatrixOp &mwo_rhs1, BLAS_Cpp::Transp trans_rhs1, BLAS_Cpp::Transp trans_rhs2, value_type beta) const |
Provides a specialized implementation for mwo_rhs1 of type MatrixSymDiag. More... | |
| bool | Mp_StMtM (MatrixOp *mwo_lhs, value_type alpha, BLAS_Cpp::Transp trans_rhs1, const MatrixOp &mwo_rhs2, BLAS_Cpp::Transp trans_rhs2, value_type beta) const |
Provides a specialized implementation for mwo_rhs2 of type MatrixSymDiag. More... | |
Additional Inherited Members | |
| Public Member Functions inherited from AbstractLinAlgPack::MatrixOp | |
| virtual void | zero_out () |
| M_lhs = 0 : Zero out the matrix. More... | |
| virtual void | Mt_S (value_type alpha) |
| M_lhs *= alpha : Multiply a matrix by a scalar. More... | |
| virtual MatrixOp & | operator= (const MatrixOp &mwo_rhs) |
| M_lhs = mwo_rhs : Virtual assignment operator. More... | |
| virtual mat_mut_ptr_t | clone () |
| Clone the non-const matrix object (if supported). More... | |
| virtual std::ostream & | output (std::ostream &out) const |
| Virtual output function. More... | |
| const MatNorm | calc_norm (EMatNormType requested_norm_type=MAT_NORM_1, bool allow_replacement=false) const |
| Compute a norm of this matrix. More... | |
| mat_ptr_t | sub_view (const index_type &rl, const index_type &ru, const index_type &cl, const index_type &cu) const |
Inlined implementation calls this->sub_view(Range1D(rl,ru),Range1D(cl,cu)). More... | |
| virtual mat_ptr_t | perm_view (const Permutation *P_row, const index_type row_part[], int num_row_part, const Permutation *P_col, const index_type col_part[], int num_col_part) const |
Create a permuted view: M_perm = P_row' * M * P_col. More... | |
| virtual mat_ptr_t | perm_view_update (const Permutation *P_row, const index_type row_part[], int num_row_part, const Permutation *P_col, const index_type col_part[], int num_col_part, const mat_ptr_t &perm_view) const |
Reinitialize a permuted view: M_perm = P_row' * M * P_col. More... | |
| Public Member Functions inherited from AbstractLinAlgPack::MatrixBase | |
| virtual | ~MatrixBase () |
| Virtual destructor. More... | |
| virtual const VectorSpace & | space_cols () const =0 |
| Vector space for vectors that are compatible with the columns of the matrix. More... | |
| virtual const VectorSpace & | space_rows () const =0 |
| Vector space for vectors that are compatible with the rows of the matrix. More... | |
| virtual size_type | rows () const |
| Return the number of rows in the matrix. More... | |
| virtual size_type | cols () const |
| Return the number of columns in the matrix. More... | |
| virtual size_type | nz () const |
| Return the number of nonzero elements in the matrix. More... | |
| Protected Member Functions inherited from AbstractLinAlgPack::MatrixOp | |
| virtual bool | Mp_StM (MatrixOp *mwo_lhs, value_type alpha, BLAS_Cpp::Transp trans_rhs) const |
| mwo_lhs += alpha * op(M_rhs) (BLAS xAXPY). More... | |
| virtual bool | Mp_StM (value_type alpha, const MatrixOp &M_rhs, BLAS_Cpp::Transp trans_rhs) |
| M_lhs += alpha * op(mwo_rhs) (BLAS xAXPY). More... | |
| virtual bool | Mp_StMtP (MatrixOp *mwo_lhs, value_type alpha, BLAS_Cpp::Transp M_trans, const GenPermMatrixSlice &P_rhs, BLAS_Cpp::Transp P_rhs_trans) const |
| mwo_lhs += alpha * op(M_rhs) * op(P_rhs). More... | |
| virtual bool | Mp_StMtP (value_type alpha, const MatrixOp &mwo_rhs, BLAS_Cpp::Transp M_trans, const GenPermMatrixSlice &P_rhs, BLAS_Cpp::Transp P_rhs_trans) |
| M_lhs += alpha * op(mwo_rhs) * op(P_rhs). More... | |
| virtual bool | Mp_StPtM (MatrixOp *mwo_lhs, value_type alpha, const GenPermMatrixSlice &P_rhs, BLAS_Cpp::Transp P_rhs_trans, BLAS_Cpp::Transp M_trans) const |
| mwo_lhs += alpha * op(P_rhs) * op(M_rhs). More... | |
| virtual bool | Mp_StPtM (value_type alpha, const GenPermMatrixSlice &P_rhs, BLAS_Cpp::Transp P_rhs_trans, const MatrixOp &mwo_rhs, BLAS_Cpp::Transp M_trans) |
| M_lhs += alpha * op(P_rhs) * op(mwo_rhs). More... | |
| virtual bool | Mp_StPtMtP (MatrixOp *mwo_lhs, value_type alpha, const GenPermMatrixSlice &P_rhs1, BLAS_Cpp::Transp P_rhs1_trans, BLAS_Cpp::Transp M_trans, const GenPermMatrixSlice &P_rhs2, BLAS_Cpp::Transp P_rhs2_trans) const |
| mwo_lhs += alpha * op(P_rhs1) * op(M_rhs) * op(P_rhs2). More... | |
| virtual bool | Mp_StPtMtP (value_type alpha, const GenPermMatrixSlice &P_rhs1, BLAS_Cpp::Transp P_rhs1_trans, const MatrixOp &mwo_rhs, BLAS_Cpp::Transp M_trans, const GenPermMatrixSlice &P_rhs2, BLAS_Cpp::Transp P_rhs2_trans) |
| M_lhs += alpha * op(P_rhs1) * op(mwo_rhs) * op(P_rhs2). More... | |
| virtual void | Vp_StMtV (VectorMutable *v_lhs, value_type alpha, BLAS_Cpp::Transp trans_rhs1, const Vector &v_rhs2, value_type beta) const =0 |
| v_lhs = alpha * op(M_rhs1) * v_rhs2 + beta * v_lhs (BLAS xGEMV) More... | |
| virtual void | Vp_StMtV (VectorMutable *v_lhs, value_type alpha, BLAS_Cpp::Transp trans_rhs1, const SpVectorSlice &sv_rhs2, value_type beta) const |
| v_lhs = alpha * op(M_rhs1) * sv_rhs2 + beta * v_lhs (BLAS xGEMV) More... | |
| virtual void | Vp_StPtMtV (VectorMutable *v_lhs, value_type alpha, const GenPermMatrixSlice &P_rhs1, BLAS_Cpp::Transp P_rhs1_trans, BLAS_Cpp::Transp M_rhs2_trans, const Vector &v_rhs3, value_type beta) const |
| v_lhs = alpha * op(P_rhs1) * op(M_rhs2) * v_rhs3 + beta * v_rhs More... | |
| virtual void | Vp_StPtMtV (VectorMutable *v_lhs, value_type alpha, const GenPermMatrixSlice &P_rhs1, BLAS_Cpp::Transp P_rhs1_trans, BLAS_Cpp::Transp M_rhs2_trans, const SpVectorSlice &sv_rhs3, value_type beta) const |
| v_lhs = alpha * op(P_rhs1) * op(M_rhs2) * sv_rhs3 + beta * v_rhs More... | |
| virtual value_type | transVtMtV (const Vector &v_rhs1, BLAS_Cpp::Transp trans_rhs2, const Vector &v_rhs3) const |
| result = v_rhs1' * op(M_rhs2) * v_rhs3 More... | |
| virtual value_type | transVtMtV (const SpVectorSlice &sv_rhs1, BLAS_Cpp::Transp trans_rhs2, const SpVectorSlice &sv_rhs3) const |
| result = sv_rhs1' * op(M_rhs2) * sv_rhs3 More... | |
| virtual void | syr2k (BLAS_Cpp::Transp M_trans, value_type alpha, const GenPermMatrixSlice &P1, BLAS_Cpp::Transp P1_trans, const GenPermMatrixSlice &P2, BLAS_Cpp::Transp P2_trans, value_type beta, MatrixSymOp *symwo_lhs) const |
| Perform a specialized rank-2k update of a dense symmetric matrix of the form: More... | |
| virtual bool | Mp_StMtM (value_type alpha, const MatrixOp &mwo_rhs1, BLAS_Cpp::Transp trans_rhs1, const MatrixOp &mwo_rhs2, BLAS_Cpp::Transp trans_rhs2, value_type beta) |
| M_lhs = alpha * op(mwo_rhs1) * op(mwo_rhs2) + beta * mwo_lhs (left) (xGEMM) More... | |
| virtual bool | syrk (BLAS_Cpp::Transp M_trans, value_type alpha, value_type beta, MatrixSymOp *sym_lhs) const |
| Perform a rank-k update of a symmetric matrix of the form: More... | |
| virtual bool | syrk (const MatrixOp &mwo_rhs, BLAS_Cpp::Transp M_trans, value_type alpha, value_type beta) |
| Perform a rank-k update of a symmetric matrix of the form: More... | |
Interface for a collection of non-mutable vectors (multi-vector, matrix).
This interface is quite restrictive in that it allows a client to access a matrix by accessing rows, columns and/or diagonals. The vector objects returned from the provided access methods row(), col() and diag() are abstract vectors so there is still good implementation flexibility but many MatrixOp implementations will not be able to support this interface.
The primary purpose for this interface is to allow for convienent aggregations of column vectors. Such an orderly arrangement allows for better optimized linear algebra operations such as matrix-matrix multiplication and the solution of linear systems for multiple right hand sides. Every application area (serial parallel, out-of-core etc.) should be able to define at least one reasonbly efficient implementation of a MultiVector (or a MultiVectorMutable) subclass.
The MultiVector interface is derived from the MatrixOp interface and therefore a MultiVector can be considered as a matrix which has some interesting implications. As an extended matrix interface, this is somewhat of a "last resort" interface that allows many matrix operations to have default implementations based on vector operations. None of the linear algebra methods in MatrixOp or any of the other matrix interfaces have methods that directly accept MultiVector objects. However, since MultiVector is derived from MatrixOp, a MultiVector object can be used anywere a MatrixOp object is accepted. In fact, many of the default implementations for the linear algebra methods in MatrixOp test to see if the matrix arguments support MultiVector (or MultiVectorMutable) and will fail if these interfaces are not supported.
Note that only certain kinds of access may be preferred and it is allowed for subclasses to return NULL vector objects for some types of access. For example, a matrix may be naturally oriented by column (for the primary role of as a multi-vector) but row or diagonal access may be very inefficient. For this reason, the client should call the access_by() method which returns a bit field that the client can compare to the constants ROW_ACCESS, COL_ACCESS and DIAG_ACCESS. The method access_by() only returns the types of access that are guarrentted to be efficient, but does not necessarily imply that a type of access is not supported. For example, (this->access_by() & ROW_ACCESS) == false this does not mean that this->row(i) will return NULL, it only means that row access will be inefficient. To determine if a certain type of access is even possible, check the return for row(), col() and/or diag(). For example, if this->rows(1) returns NULL, then this is a flag that row access for every row is not supported. Diagonal access may be a different story. For some matrix subclasses, only the center diagonal my be easily accessable in which case diag(0) may return != NULL but diag(k) for all k != 0 may return NULL.
Note that since, this interface is derived from MatrixOp that it must support the methods space_rows() and space_cols(). This does not imply however that either of the access methods row() or col() must return non-NULL.
Examples of matrix implementations that can support this interface are a dense BLAS compatible matrix (ROW_ACCESS, COL_ACCESS and DIAG_ACCESS), a compressed column sparse matrix (COL_ACCESS only), a compressed row sparse matrix (ROW_ACCESS only), a unordered sparse matrix with the diagonal explicity stored (diag(0) != NULL) etc.
Another very powerfull feature of this interface is the ability to apply reduction/transformation operators over a sub-set of rows and columns in a set of multi-vector objects. The behavior is identical as if the client extracted the rows or columns in a set of multi-vectors and called Vector::apply_op() itself. However, the advantage of using the multi-vector methods is that there may be greater opertunity to exploit parallelism. Also, the intermediate reduction objects over a set of rows or columns can be reduced by a secondary reduction object.
ToDo: Finish documentation!
Definition at line 170 of file AbstractLinAlgPack_MultiVector.hpp.
| typedef int AbstractLinAlgPack::MultiVector::access_by_t |
Definition at line 179 of file AbstractLinAlgPack_MultiVector.hpp.
| typedef Teuchos::RCP<const Vector> AbstractLinAlgPack::MultiVector::vec_ptr_t |
Definition at line 187 of file AbstractLinAlgPack_MultiVector.hpp.
Definition at line 189 of file AbstractLinAlgPack_MultiVector.hpp.
| anonymous enum |
Definition at line 181 of file AbstractLinAlgPack_MultiVector.hpp.
|
virtual |
Clone the non-const multi-vector object.
The default implementation returns return.get()==NULL.
Reimplemented in AbstractLinAlgPack::MultiVectorMutable.
Definition at line 162 of file AbstractLinAlgPack_MultiVector.cpp.
|
pure virtual |
Return a bit field for the types of access that are the most convenient.
Postconditions:
return & COL_ACCESS || return & ROW_ACCESS || return & DIAG_ACCESS Implemented in AbstractLinAlgPack::MultiVectorMutableDense, and AbstractLinAlgPack::MultiVectorMutableCols.
|
pure virtual |
Get a non-mutable column vector.
Postconditions:
this->access_by() & COL_ACCESS] return.get() != NULL return.get() != NULL] space_cols().is_compatible(return->space()) == true Implemented in AbstractLinAlgPack::MultiVectorMutable.
|
pure virtual |
Get a non-mutable row vector.
Postconditions:
this->access_by() & ROW_ACCESS] return.get() != NULL return.get() != NULL] space_rows().is_compatible(return->space()) == true Implemented in AbstractLinAlgPack::MultiVectorMutable.
|
pure virtual |
Get a non-mutable diagonal vector.
Postconditions:
this->access_by() & DIAG_ACCESS] return.get() != NULL Implemented in AbstractLinAlgPack::MultiVectorMutable.
|
virtual |
Returns a sub-view of the multi vector.
ToDo: Finish documentation!
The default implementation returns a MultiVectorSubView object for any valid arbitary sub-view.
Reimplemented in AbstractLinAlgPack::MultiVectorMutable.
Definition at line 168 of file AbstractLinAlgPack_MultiVector.cpp.
|
inline |
Inlined implementation calls this->mv_sub_view(Range1D(rl,ru),Range1D(cl,cu)).
Definition at line 418 of file AbstractLinAlgPack_MultiVector.hpp.
|
protectedvirtual |
Apply a reduction/transformation operator row by row, or column by column and return an array of the reduction objects.
Preconditions:
apply_by == APPLY_BY_COL] (this->access_by() & COL_ACCESS) == true) (throw ???) apply_by == APPLY_BY_ROW] (this->access_by() & ROW_ACCESS) == true) (throw ???) The default implementation calls this->apply_op().
ToDo: Finish documentation!
Definition at line 176 of file AbstractLinAlgPack_MultiVector.cpp.
|
protectedvirtual |
Apply a reduction/transformation operator row by row, or column by column and reduce the intermediate reduction objects into one reduction object.
Preconditions:
apply_by == APPLY_BY_COL] (this->access_by() & COL_ACCESS) == true) (throw ???) apply_by == APPLY_BY_ROW] (this->access_by() & ROW_ACCESS) == true) (throw ???) The default implementation calls this->apply_op().
ToDo: Finish documentation!
Definition at line 235 of file AbstractLinAlgPack_MultiVector.cpp.
|
virtual |
Returns this->mv_clone().
Reimplemented from AbstractLinAlgPack::MatrixOp.
Reimplemented in AbstractLinAlgPack::MultiVectorMutableCols.
Definition at line 285 of file AbstractLinAlgPack_MultiVector.cpp.
|
virtual |
Returns this->mv_sub_view(row_rng,col_rng) casted to a MatrixOp.
Reimplemented from AbstractLinAlgPack::MatrixOp.
Definition at line 291 of file AbstractLinAlgPack_MultiVector.cpp.
|
virtual |
Provides a specialized implementation for mwo_rhs1 of type MatrixSymDiag.
true and implements the operation if dynamic_cast<MatrixSymDiag>(&mwo_rhs1) != NULL && op(<em>this).access_by() =& MultiVector::COL_ACCESS && (mvm_lhs = dynamic_cast<MultiVectorMutable>(&mwo_lhs)) != NULL && mvm_lhs->access_by() & MultiVector::COL_ACCESS. Otherwise, this function returns false and does not implement the operation. or dynamic_cast<const MatrixSymDiag>(&mwo_rhs1) != NULL.The default implementation relies on column access of op(*this) and mwo_lhs to implement this method.
Reimplemented from AbstractLinAlgPack::MatrixOp.
Reimplemented in AbstractLinAlgPack::MultiVectorMutableDense.
Definition at line 296 of file AbstractLinAlgPack_MultiVector.cpp.
|
virtual |
Provides a specialized implementation for mwo_rhs2 of type MatrixSymDiag.
true and implements the operation if dynamic_cast<MatrixSymDiag>(&mwo_rhs1) != NULL && op(<em>this).access_by() =& MultiVector::ROW_ACCESS && (mvm_lhs = dynamic_cast<MultiVectorMutable>(&mwo_lhs)) != NULL && mvm_lhs->access_by() & MultiVector::ROW_ACCESS. Otherwise, this function returns false and does not implement the operation. or dynamic_cast<const MatrixSymDiag>(&mwo_rhs1) != NULL.The default implementation relies on row access of op(*this) and mwo_lhs to implement this method.
Reimplemented from AbstractLinAlgPack::MatrixOp.
Reimplemented in AbstractLinAlgPack::MultiVectorMutableDense.
Definition at line 311 of file AbstractLinAlgPack_MultiVector.cpp.
|
friend |
ToDo: Finish documentation!
|
friend |
ToDo: Finish documentation!
1.8.6