GinkgoWrappers::SolverFCG< ValueType, IndexType > Class Template Reference

#include <deal.II/lac/ginkgo_solver.h>

Inheritance diagram for GinkgoWrappers::SolverFCG< ValueType, IndexType >:

Classes
struct	AdditionalData

Public Member Functions
	SolverFCG (SolverControl &solver_control, const std::string &exec_type, const AdditionalData &data=AdditionalData())
	SolverFCG (SolverControl &solver_control, const std::string &exec_type, const std::shared_ptr< gko::LinOpFactory > &preconditioner, const AdditionalData &data=AdditionalData())
void	initialize (const SparseMatrix< double > &matrix)
void	apply (Vector< double > &solution, const Vector< double > &rhs)
void	solve (const SparseMatrix< double > &matrix, Vector< double > &solution, const Vector< double > &rhs)
SolverControl &	control () const

Protected Attributes
const AdditionalData	additional_data
SolverControl &	solver_control
std::shared_ptr< gko::LinOpFactory >	solver_gen
std::shared_ptr< gko::stop::ResidualNormReduction<>::Factory >	residual_criterion
std::shared_ptr< gko::log::Convergence<> >	convergence_logger
std::shared_ptr< gko::stop::Combined::Factory >	combined_factory
std::shared_ptr< gko::Executor >	executor

Private Member Functions
void	initialize_ginkgo_log ()

Private Attributes
std::shared_ptr< gko::matrix::Csr< double, int32_t > >	system_matrix
const std::string	exec_type

Detailed Description

template<typename ValueType = double, typename IndexType = int32_t>
class GinkgoWrappers::SolverFCG< ValueType, IndexType >

An implementation of the solver interface using the Ginkgo FCG solver.

FCG or the flexible conjugate gradient method is an iterative type Krylov subspace method which is suitable for symmetric positive definite methods.

Though this method performs very well for symmetric positive definite matrices, it is in general not suitable for general matrices.

In contrast to the standard CG based on the Polack-Ribiere formula, the flexible CG uses the Fletcher-Reeves formula for creating the orthonormal vectors spanning the Krylov subspace. This increases the computational cost of every Krylov solver iteration but allows for non-constant preconditioners.

Definition at line 385 of file ginkgo_solver.h.

Constructor & Destructor Documentation

SolverFCG() [1/2]

template<typename ValueType , typename IndexType >

GinkgoWrappers::SolverFCG< ValueType, IndexType >::SolverFCG	(	SolverControl &	solver_control,
		const std::string &	exec_type,
		const AdditionalData &	data = AdditionalData() )

Constructor.

Parameters

[in,out]	solver_control	The solver control object is then used to set the parameters and set up the FCG solver from the FCG factory which solves the linear system.
[in]	exec_type	The execution paradigm for the FCG solver.
[in]	data	The additional data required by the solver.

Definition at line 404 of file ginkgo_solver.cc.

SolverFCG() [2/2]

template<typename ValueType , typename IndexType >

GinkgoWrappers::SolverFCG< ValueType, IndexType >::SolverFCG	(	SolverControl &	solver_control,
		const std::string &	exec_type,
		const std::shared_ptr< gko::LinOpFactory > &	preconditioner,
		const AdditionalData &	data = AdditionalData() )

Constructor.

Parameters

[in,out]	solver_control	The solver control object is then used to set the parameters and set up the FCG solver from the FCG factory which solves the linear system.
[in]	exec_type	The execution paradigm for the FCG solver.
[in]	preconditioner	The preconditioner for the solver.
[in]	data	The additional data required by the solver.

Definition at line 418 of file ginkgo_solver.cc.

Member Function Documentation

initialize()

void SolverBase< double, int32_t >::initialize ( const SparseMatrix< double > & matrix )

inherited

Initialize the matrix and copy over its data to Ginkgo's data structures.

Definition at line 113 of file ginkgo_solver.cc.

apply()

void SolverBase< double, int32_t >::apply ( Vector< double > & solution, const Vector< double > & rhs )

inherited

Solve the linear system Ax=b. Dependent on the information provided by derived classes one of Ginkgo's linear solvers is chosen.

Definition at line 121 of file ginkgo_solver.cc.

solve()

void SolverBase< double, int32_t >::solve ( const SparseMatrix< double > & matrix, Vector< double > & solution, const Vector< double > & rhs )

inherited

Solve the linear system Ax=b. Dependent on the information provided by derived classes one of Ginkgo's linear solvers is chosen.

Definition at line 129 of file ginkgo_solver.cc.

control()

SolverControl & SolverBase< double, int32_t >::control ( ) const

inherited

Access to the object that controls convergence.

Definition at line 137 of file ginkgo_solver.cc.

initialize_ginkgo_log()

void SolverBase< double, int32_t >::initialize_ginkgo_log ( )

privateinherited

Initialize the Ginkgo logger object with event masks. Refer to Ginkgo's logging event masks.

Definition at line 186 of file ginkgo_solver.cc.

Member Data Documentation

additional_data

template<typename ValueType = double, typename IndexType = int32_t>

const AdditionalData GinkgoWrappers::SolverFCG< ValueType, IndexType >::additional_data

protected

Store a copy of the settings for this particular solver.

Definition at line 431 of file ginkgo_solver.h.

solver_control

SolverControl& GinkgoWrappers::SolverBase< double, int32_t >::solver_control

protectedinherited

Reference to the object that controls convergence of the iterative solvers.

Definition at line 145 of file ginkgo_solver.h.

solver_gen

std::shared_ptr<gko::LinOpFactory> GinkgoWrappers::SolverBase< double, int32_t >::solver_gen

protectedinherited

The Ginkgo generated solver factory object.

Definition at line 150 of file ginkgo_solver.h.

residual_criterion

std::shared_ptr<gko::stop::ResidualNormReduction<>::Factory> GinkgoWrappers::SolverBase< double, int32_t >::residual_criterion

protectedinherited

The residual criterion object that controls the reduction of the residual based on the tolerance set in the solver_control member.

Definition at line 157 of file ginkgo_solver.h.

convergence_logger

std::shared_ptr<gko::log::Convergence<> > GinkgoWrappers::SolverBase< double, int32_t >::convergence_logger

protectedinherited

The Ginkgo convergence logger used to check for convergence and other solver data if needed.

Definition at line 163 of file ginkgo_solver.h.

combined_factory

std::shared_ptr<gko::stop::Combined::Factory> GinkgoWrappers::SolverBase< double, int32_t >::combined_factory

protectedinherited

The Ginkgo combined factory object is used to create a combined stopping criterion to be passed to the solver.

Definition at line 169 of file ginkgo_solver.h.

executor

std::shared_ptr<gko::Executor> GinkgoWrappers::SolverBase< double, int32_t >::executor

protectedinherited

The execution paradigm in Ginkgo. The choices are between gko::OmpExecutor, gko::CudaExecutor and gko::ReferenceExecutor and more details can be found in Ginkgo's documentation.

Definition at line 176 of file ginkgo_solver.h.

system_matrix

std::shared_ptr<gko::matrix::Csr<double, int32_t> > GinkgoWrappers::SolverBase< double, int32_t >::system_matrix

privateinherited

Ginkgo matrix data structure. First template parameter is for storing the array of the non-zeros of the matrix. The second is for the row pointers and the column indices.

Todo

Templatize based on Matrix type.

Definition at line 195 of file ginkgo_solver.h.

exec_type

const std::string GinkgoWrappers::SolverBase< double, int32_t >::exec_type

privateinherited

The execution paradigm as a string to be set by the user. The choices are between omp, cuda and reference and more details can be found in Ginkgo's documentation.

Definition at line 202 of file ginkgo_solver.h.

---

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

• include/deal.II/lac/ginkgo_solver.h
• source/lac/ginkgo_solver.cc