symildl::solver< el_type, mat_type > Class Template Reference

Set of tools that facilitates conversion between different matrix formats. Also contains solver methods for matrices using a common interface. More...

#include <solver.h>

Collaboration diagram for symildl::solver< el_type, mat_type >:

Public Types
typedef mat_type::pivot_type	pivot_type

Public Member Functions
	solver ()
	Solver constructor, initializes default reordering scheme.
void	load (std::string filename)
	Loads the matrix A into solver. A must be of matrix market format. More...
void	load (const std::vector< int > &ptr, const std::vector< int > &row, const std::vector< el_type > &val)
	Loads the matrix A into solver. A must be of CSC format.
void	load (const int *ptr, const int *row, const el_type *val, int dim)
	Loads the matrix A into solver. A must be of CSC format.
void	set_rhs (vector< el_type > b)
	Loads a right hand side b into the solver. More...
void	set_reorder_scheme (const char *ordering)
	Sets the reordering scheme for the solver.
void	set_equil (const char *equil)
	Decides whether we should use equilibriation on the matrix or not.
void	set_solver (const char *solver)
	Decides whether we perform a full solve or not.
void	set_message_level (const char *msg)
	Controls how much information gets printed to stdout.
void	set_inplace (bool inplace)
	Decides whether we perform the factorization inplace or not.
void	set_pivot (const char *pivot)
	Decides the kind of partial pivoting we should use.
void	solve (double fill_factor, double tol, double pp_tol, int max_iter=-1, double minres_tol=1e-6, double shift=0.0)
	Factors the matrix A into P' * S * A * S * P = LDL' in addition to printing some timing data to screen. More...
void	minres (int max_iter=1000, double stop_tol=1e-6, double shift=0.0)
	Applies minres on A, preconditioning with factors L and D. More...
void	sqmr (int max_iter=1000, double stop_tol=1e-6)
	Applies SMQR on A, preconditioning with factors L and D. More...
void	save ()
	Save results of factorization (automatically saved into the output_matrices folder). More...
void	display ()
	Prints the L and D factors to stdout.

Public Attributes
mat_type	A
	The matrix to be factored.
mat_type	L
	The lower triangular factor of A.
vector< int >	perm
	A permutation vector containing all permutations on A.
block_diag_matrix< el_type >	D
	The diagonal factor of A.
int	reorder_type
	Set to to 0 for AMD, 1 for RCM, 2 for no reordering.
int	piv_type
	Set to 0 for rook, 1 for bunch.
int	equil_type
	The equilibration method used. Set to 1 for max-norm equilibriation.
int	msg_lvl
	Controls the amount of output to stdout.
int	solve_type
bool	has_rhs
	Set to true if we have a right hand side that we expect to solve.
bool	perform_inplace
	Set to true if we are factoring the matrix A inplace.
bool	save_sol
	Set to true if we want to save the solution to a file.
vector< el_type >	rhs
	The right hand side we'll solve for.
vector< el_type >	sol_vec
	The solution vector.

Detailed Description

template<class el_type, class mat_type = lilc_matrix<el_type>>
class symildl::solver< el_type, mat_type >

Set of tools that facilitates conversion between different matrix formats. Also contains solver methods for matrices using a common interface.

Currently, the only matrix type accepted is the lilc_matrix (as no other matrix type has been created yet).

Member Function Documentation

template<class el_type , class mat_type = lilc_matrix<el_type>>

void symildl::solver< el_type, mat_type >::load ( std::string  filename )

inline

Loads the matrix A into solver. A must be of matrix market format.

Parameters

filename

the filename of the matrix.

template<class el_type , class mat_type >

void solver::minres	(	int	max_iter = 1000,
		double	stop_tol = 1e-6,
		double	shift = 0.0
	)

Applies minres on A, preconditioning with factors L and D.

Parameters

max_iter	the maximum number of minres iterations.
stop_tol	the stopping tolerance of minres. i.e. we stop as soon as the residual goes below stop_tol.
shift	shifts A by shift*(identity matrix) to make it more positive definite. This sometimes helps.

template<class el_type , class mat_type = lilc_matrix<el_type>>

void symildl::solver< el_type, mat_type >::save ( )

inline

Save results of factorization (automatically saved into the output_matrices folder).

The names of the output matrices follow the format out{}.mtx, where {} describes what the file contains (i.e. A, L, or D).

template<class el_type , class mat_type = lilc_matrix<el_type>>

void symildl::solver< el_type, mat_type >::set_rhs ( vector< el_type >  b )

inline

Loads a right hand side b into the solver.

Parameters

b	a vector of the right hand side.

template<class el_type , class mat_type = lilc_matrix<el_type>>

void symildl::solver< el_type, mat_type >::solve ( double  fill_factor, double  tol, double  pp_tol, int  max_iter = -1, double  minres_tol = 1e-6, double  shift = 0.0  )

inline

Factors the matrix A into P' * S * A * S * P = LDL' in addition to printing some timing data to screen.

More information about the parameters can be found in the documentation for the ildl() function.

Parameters

fill_factor	a factor controling memory usage of factorization.
tol	a factor controling accuracy of factorization.
pp_tol	a factor controling the aggresiveness of Bunch-Kaufman pivoting.
max_iter	the maximum number of iterations for minres (ignored if no right hand side).

template<class el_type , class mat_type >

void solver::sqmr	(	int	max_iter = 1000,
		double	stop_tol = 1e-6
	)

Applies SMQR on A, preconditioning with factors L and D.

Parameters

max_iter	the maximum number of minres iterations.
stop_tol	the stopping tolerance of minres. i.e. we stop as soon as the residual goes below stop_tol.

---

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

• source/solver.h
• source/solver_minres.h
• source/solver_sqmr.h