Binds a Krylov solver to an SpMatrix. More...

#include <AMReX_KrylovMV.H>

Public Types
using	RT = T

using	VEC = AlgVector< T >

using	MAT = SpMatrix< T >

using	Solver = S< VEC, KrylovMV< S, T > >

using	GM = Solver

using	PC = std::function< void(VEC &, VEC const &)>

Public Member Functions
	KrylovMV (MAT const *a_mat)
	Bind the solver to a sparse matrix described by `a_mat`.

	~KrylovMV ()=default

	KrylovMV (const KrylovMV &)=delete

	KrylovMV (KrylovMV &&)=delete

KrylovMV &	operator= (const KrylovMV &)=delete

KrylovMV &	operator= (KrylovMV &&)=delete

void	setPrecond (PC a_pc)
	Supply an optional right-preconditioner functor.

void	solve (VEC &a_sol, VEC const &a_rhs, T a_tol_rel, T a_tol_abs)
	Solve the linear system.

void	setVerbose (int v)
	Set verbosity level `v` for the underlying solver.

Solver &	getSolver ()
	Access the underlying solver for additional settings and results.

Solver &	getGMRES ()
	Same as getSolver(), for GMRES_MV only.

VEC	makeVecRHS () const
	Return an AlgVector that mirrors the matrix partition for RHS storage.

VEC	makeVecLHS () const
	Return another AlgVector with the same partition for LHS storage.

void	apply (VEC &lhs, VEC &rhs) const
	Apply the sparse operator: `lhs = A rhs`.

void	precond (VEC &lhs, VEC const &rhs) const
	Apply the optional preconditioner (or copy if none is provided).

Static Public Member Functions
static T	norm2 (VEC const &vec)
	Euclidean norm of `vec`.

static void	scale (VEC &vec, T scale_factor)
	Scale `vec` in place by `scale_factor`.

static T	dotProduct (VEC const &vec1, VEC const &vec2)
	Dot product between `vec1` and `vec2`.

static void	setToZero (VEC &lhs)
	Reset `lhs` to zero.

static void	assign (VEC &lhs, VEC const &rhs)
	Copy `rhs` into `lhs`.

static void	increment (VEC &lhs, VEC const &rhs, T a)
	Accumulate `lhs += a * rhs`.

static void	linComb (VEC &lhs, T a, VEC const &rhs_a, T b, VEC const &rhs_b)
	Form the linear combination `lhs = a * rhs_a + b * rhs_b`.

Detailed Description

template<template< typename, typename > class S, typename T>
class amrex::KrylovMV< S, T >

Binds a Krylov solver to an SpMatrix.

Template Parameters

S	Krylov solver class template with the GMRES operator interface, e.g., GMRES, BiCGStab or PCG.
T	Value type of the matrix and vectors (double or float).

Use the aliases GMRES_MV, BiCGStab_MV and PCG_MV.

Member Typedef Documentation

◆ GM

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::GM = Solver

◆ MAT

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::MAT = SpMatrix<T>

◆ PC

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::PC = std::function<void(VEC&,VEC const&)>

◆ RT

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::RT = T

◆ Solver

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::Solver = S<VEC,KrylovMV<S,T> >

◆ VEC

template<template< typename, typename > class S, typename T >

using amrex::KrylovMV< S, T >::VEC = AlgVector<T>

Constructor & Destructor Documentation

◆ KrylovMV() [1/3]

template<template< typename, typename > class S, typename T >

amrex::KrylovMV< S, T >::KrylovMV ( MAT const * a_mat )

Bind the solver to a sparse matrix described by a_mat.

Parameters

a_mat Matrix supplying the SpMV operation and partition.

◆ ~KrylovMV()

template<template< typename, typename > class S, typename T >

amrex::KrylovMV< S, T >::~KrylovMV ( )

default

◆ KrylovMV() [2/3]

template<template< typename, typename > class S, typename T >

amrex::KrylovMV< S, T >::KrylovMV ( const KrylovMV< S, T > & )

delete

◆ KrylovMV() [3/3]

template<template< typename, typename > class S, typename T >

amrex::KrylovMV< S, T >::KrylovMV ( KrylovMV< S, T > && )

delete

Member Function Documentation

◆ apply()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::apply	(	VEC &	lhs,
		VEC &	rhs
	)		const

Apply the sparse operator: lhs = A rhs.

rhs is non-const to satisfy the solver interface.

◆ assign()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::assign	(	VEC &	lhs,
		VEC const &	rhs
	)

static

Copy rhs into lhs.

◆ dotProduct()

template<template< typename, typename > class S, typename T >

T amrex::KrylovMV< S, T >::dotProduct	(	VEC const &	vec1,
		VEC const &	vec2
	)

static

Dot product between vec1 and vec2.

◆ getGMRES()

template<template< typename, typename > class S, typename T >

Solver & amrex::KrylovMV< S, T >::getGMRES ( )

inline

Same as getSolver(), for GMRES_MV only.

◆ getSolver()

template<template< typename, typename > class S, typename T >

Solver & amrex::KrylovMV< S, T >::getSolver ( )

inline

Access the underlying solver for additional settings and results.

◆ increment()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::increment	(	VEC &	lhs,
		VEC const &	rhs,
		T	a
	)

static

Accumulate lhs += a * rhs.

◆ linComb()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::linComb	(	VEC &	lhs,
		T	a,
		VEC const &	rhs_a,
		T	b,
		VEC const &	rhs_b
	)

static

Form the linear combination lhs = a * rhs_a + b * rhs_b.

◆ makeVecLHS()

template<template< typename, typename > class S, typename T >

auto amrex::KrylovMV< S, T >::makeVecLHS ( ) const

Return another AlgVector with the same partition for LHS storage.

◆ makeVecRHS()

template<template< typename, typename > class S, typename T >

auto amrex::KrylovMV< S, T >::makeVecRHS ( ) const

Return an AlgVector that mirrors the matrix partition for RHS storage.

◆ norm2()

template<template< typename, typename > class S, typename T >

T amrex::KrylovMV< S, T >::norm2 ( VEC const & vec )

static

Euclidean norm of vec.

◆ operator=() [1/2]

template<template< typename, typename > class S, typename T >

KrylovMV & amrex::KrylovMV< S, T >::operator= ( const KrylovMV< S, T > & )

delete

◆ operator=() [2/2]

template<template< typename, typename > class S, typename T >

KrylovMV & amrex::KrylovMV< S, T >::operator= ( KrylovMV< S, T > && )

delete

◆ precond()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::precond	(	VEC &	lhs,
		VEC const &	rhs
	)		const

Apply the optional preconditioner (or copy if none is provided).

◆ scale()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::scale	(	VEC &	vec,
		T	scale_factor
	)

static

Scale vec in place by scale_factor.

◆ setPrecond()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::setPrecond ( PC a_pc )

inline

Supply an optional right-preconditioner functor.

The functor must follow the signature void(VEC& lhs, VEC const& rhs).

Parameters

a_pc	Callable that computes `lhs = P^{-1}(rhs)`.

◆ setToZero()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::setToZero ( VEC & lhs )

static

Reset lhs to zero.

◆ setVerbose()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::setVerbose ( int v )

inline

Set verbosity level v for the underlying solver.

◆ solve()

template<template< typename, typename > class S, typename T >

void amrex::KrylovMV< S, T >::solve	(	VEC &	a_sol,
		VEC const &	a_rhs,
		T	a_tol_rel,
		T	a_tol_abs
	)

Solve the linear system.

Parameters

a_sol	unknowns, i.e., x in A x = b.
a_rhs	RHS, i.e., b in A x = b.
a_tol_rel	relative tolerance.
a_tol_abs	absolute tolerance.

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

/home/runner/work/amrex/amrex/Src/LinearSolvers/AMReX_KrylovMV.H

Public Types

Public Member Functions

Static Public Member Functions

Detailed Description

Member Typedef Documentation

◆ GM

◆ MAT

◆ PC

◆ RT

◆ Solver

◆ VEC

Constructor & Destructor Documentation

◆ KrylovMV() [1/3]

◆ ~KrylovMV()

◆ KrylovMV() [2/3]

◆ KrylovMV() [3/3]

Member Function Documentation

◆ apply()

◆ assign()

◆ dotProduct()

◆ getGMRES()

◆ getSolver()

◆ increment()

◆ linComb()

◆ makeVecLHS()

◆ makeVecRHS()

◆ norm2()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ precond()

◆ scale()

◆ setPrecond()

◆ setToZero()

◆ setVerbose()

◆ solve()