Classes and function to represent and solve large distributed sparse linear systems. More...

Classes
class	CPreconditioner< ScalarType >
	Abstract base class for defining a preconditioning operation. More...

class	CIdentityPreconditioner< ScalarType >
	No-op preconditioner used when Krylov solvers run without preconditioning. More...

class	CJacobiPreconditioner< ScalarType >
	Specialization of preconditioner that uses CSysMatrix class. More...

class	CILUPreconditioner< ScalarType >
	Specialization of preconditioner that uses CSysMatrix class. More...

class	CLU_SGSPreconditioner< ScalarType >
	Specialization of preconditioner that uses CSysMatrix class. More...

class	CLineletPreconditioner< ScalarType >
	Specialization of preconditioner that uses CSysMatrix class. More...

class	CPastixPreconditioner< ScalarType >
	Specialization of preconditioner that uses PaStiX to factorize a CSysMatrix. More...

class	CAbstractPreconditioner< ScalarType >
	Applies a std::function as the preconditioning operation. More...

struct	CSysMatrixComms
	Helper to communicate distributed vectors. More...

class	CMatrixVectorProduct< ScalarType >
	Abstract base class for defining matrix-vector products. More...

class	CSysMatrixVectorProduct< ScalarType >
	Specialization of matrix-vector product that uses CSysMatrix class. More...

class	CPastixWrapper< ScalarType >
	Wrapper class that converts between SU2 sparse format and PaStiX format and simplifies calls to the external solver. More...

class	CSysMatrix< ScalarType >
	Main class for defining block-compressed-row-storage sparse matrices. More...

class	CSysSolve< ScalarType >
	Class for solving linear systems using classical and Krylov-subspace iterative methods. More...

class	CSysVector< ScalarType >
	Class for holding and manipulating vectors needed by linear solvers. More...

Enumerations
enum class	FgcrodrMode { FgcrodrMode::NORMAL , FgcrodrMode::SAME_MAT }
	Modes of using FGCRODR. More...

Functions
template<class ScalarType , class Apply >
void	ApplyPreconditionerOnHost (const CSysVector< ScalarType > &u, CSysVector< ScalarType > &v, Apply &&apply)
	Applies a preconditioner that only has a host implementation to vectors that live on the device: bring the input down, apply, put the result back.

template<class ScalarType , class Apply >
void	ApplyPreconditionerOnDevice (const CSysVector< ScalarType > &u, CSysVector< ScalarType > &v, bool useCuda, Apply &&apply)
	Mirror of ApplyPreconditionerOnHost: applies a device preconditioner to host vectors.

virtual	CPreconditioner< ScalarType >::~CPreconditioner ()=0
	Destructor of the class.

static CPreconditioner *	CPreconditioner< ScalarType >::Create (ENUM_LINEAR_SOLVER_PREC kind, CSysMatrix< ScalarType > &jacobian, CGeometry geometry, const CConfig config)
	Factory method.

Detailed Description

Classes and function to represent and solve large distributed sparse linear systems.

Enumeration Type Documentation

◆ FgcrodrMode

enum class FgcrodrMode

strong

Modes of using FGCRODR.

Enumerator
NORMAL	Solve the linear system.
SAME_MAT	"NORMAL" but knowing the matrix did not change.

Function Documentation

◆ ApplyPreconditionerOnDevice()

template<class ScalarType , class Apply >

void ApplyPreconditionerOnDevice	(	const CSysVector< ScalarType > &	u,
		CSysVector< ScalarType > &	v,
		bool	useCuda,
		Apply &&	apply
	)

inline

Mirror of ApplyPreconditionerOnHost: applies a device preconditioner to host vectors.

Note: For callers that drive the Krylov solvers themselves and so never went through CSysSolve::Solve, which is what normally leaves the vectors on the device (Newton-Krylov). Device expressions are on for the duration so that a nested solve also uses the device copies. Only u is uploaded, v is always overwritten by the apply.

◆ ApplyPreconditionerOnHost()

template<class ScalarType , class Apply >

void ApplyPreconditionerOnHost	(	const CSysVector< ScalarType > &	u,
		CSysVector< ScalarType > &	v,
		Apply &&	apply
	)

inline

Applies a preconditioner that only has a host implementation to vectors that live on the device: bring the input down, apply, put the result back.

Note: This is what keeps LU-SGS, Linelet and PaStiX usable on the GPU path. The transfers are issued by one thread with the team synchronized around them, the apply itself is the normal OpenMP parallel host code.

◆ Create()

template<class ScalarType >

CPreconditioner< ScalarType > * CPreconditioner< ScalarType >::Create	(	ENUM_LINEAR_SOLVER_PREC	kind,
		CSysMatrix< ScalarType > &	jacobian,
		CGeometry *	geometry,
		const CConfig *	config
	)

static

Factory method.

◆ ~CPreconditioner()

template<class ScalarType >

CPreconditioner< ScalarType >::~CPreconditioner ( )

pure virtual

Destructor of the class.

Classes

Enumerations

Functions