Class to test AMG solvers. More...

Detailed Description

Class to test AMG solvers.

Definition at line 42 of file petsc_solver_AMG_report.hpp.

#include <petsc_solver_AMG_report.hpp>

Public Member Functions
void	setTergetAMGAccuracy (double t_a)
	Set the target accuracy to score the AMG solver.

void	setMaxSweep (int max_sw)
	Set the target accuracy to score the AMG solver.

void	try_solve (SparseMatrix< double, int, PETSC_BASE > &A, const Vector< double, PETSC_BASE > &b)
	Try to use AMG pre-conditioner and check how they they perform.

Private Member Functions
void	benchmark (SparseMatrix< double, int, PETSC_BASE > &A, const Vector< double, PETSC_BASE > &b, petsc_solver< double > &solver, openfpm::vector< AMG_time_err_coars > &perf_amg)
	benchmark the selected setting for the preconditioner

void	test_coarsener (SparseMatrix< double, int, PETSC_BASE > &A, const Vector< double, PETSC_BASE > &b)
	test the corasener for this problem

double	score_solver (double t_solve, double t_m)
	Score the solver.

void	write_report_coars (GoogleChart &cg)
	Write the report for coarsening.

void	write_report_cycle (GoogleChart &cg, openfpm::vector< std::string > &coars)
	Write the report for coarsening.

void	write_report_cycle_asym (GoogleChart &cg, openfpm::vector< std::string > &coars)
	Write the report for coarsening.

void	test_cycle_type (SparseMatrix< double, int, PETSC_BASE > &A, const Vector< double, PETSC_BASE > &b, openfpm::vector< std::string > &coarsener_to_test)
	test the corasener for this problem

void	test_cycle_asym (SparseMatrix< double, int, PETSC_BASE > &A, const Vector< double, PETSC_BASE > &b, openfpm::vector< size_t > &ids_ts, openfpm::vector< std::string > &coarsener_to_test)
	test best asymmetric combination of sweeps

void	write_report (openfpm::vector< std::string > &coars)

void	best_coarsener (openfpm::vector< std::string > &coars)
	take the most accurate and the fastest AMG

void	best_score (openfpm::vector< AMG_time_err_coars > &perf, openfpm::vector< size_t > &sw_optimal, size_t nm)
	Return the best scoring solver.

Private Attributes
size_t	num_sweep_test = 10
	Number of sweeps for test.

double	target_accuracy = 1
	Target accuracy for the preconditioner.

openfpm::vector< AMG_time_err_coars >	perf_amg_coars
	It contains the performance of several AMG methods coarsener.

openfpm::vector< AMG_time_err_coars >	perf_amg_sweep_sym
	It contain the performance of several AMG methods sweep configuration.

openfpm::vector< AMG_time_err_coars >	perf_amg_sweep_asym
	It contain the performance of several AMG methods sweep configuration.

openfpm::vector< std::string >	method
	List of methods.

openfpm::vector< size_t >	v_cycle_tested_sym
	List of tested cycles.

openfpm::vector< std::pair< size_t, size_t > >	v_cycle_tested_asym
	List of tested cycles in case of asymmetric.

openfpm::vector< size_t >	asym_tests
	starting point of each coarsening test for the asymmetric test

Member Function Documentation

◆ benchmark()

void petsc_AMG_report::benchmark	(	SparseMatrix< double, int, PETSC_BASE > &	A,
		const Vector< double, PETSC_BASE > &	b,
		petsc_solver< double > &	solver,
		openfpm::vector< AMG_time_err_coars > &	perf_amg
	)

inlineprivate

benchmark the selected setting for the preconditioner

Parameters

A	sparse matrix
b	right-hand-side
solver	to use for benchmarking
perf_amg	performance of the AMG

Definition at line 79 of file petsc_solver_AMG_report.hpp.

◆ best_coarsener()

void petsc_AMG_report::best_coarsener ( openfpm::vector< std::string > & coars )

inlineprivate

take the most accurate and the fastest AMG

Parameters

coars vector with the best coarsener

Definition at line 481 of file petsc_solver_AMG_report.hpp.

◆ best_score()

void petsc_AMG_report::best_score	(	openfpm::vector< AMG_time_err_coars > &	perf,
		openfpm::vector< size_t > &	sw_optimal,
		size_t	nm
	)

inlineprivate

Return the best scoring solver.

Parameters

perf	where to search for the best score
optimal	number of sweeps for each tested method
mn	number of method tested

Definition at line 514 of file petsc_solver_AMG_report.hpp.

◆ score_solver()

double petsc_AMG_report::score_solver	(	double	t_solve,
		double	t_m
	)

inlineprivate

Score the solver.

Parameters

t_solve	time to solve
t_m	target accuracy the solver should reach

Definition at line 197 of file petsc_solver_AMG_report.hpp.

◆ setMaxSweep()

void petsc_AMG_report::setMaxSweep ( int max_sw )

inline

Set the target accuracy to score the AMG solver.

Set the maximum number of sweep for testing

Parameters

max_sw max number of sweep

Definition at line 562 of file petsc_solver_AMG_report.hpp.

◆ setTergetAMGAccuracy()

void petsc_AMG_report::setTergetAMGAccuracy ( double t_a )

inline

Set the target accuracy to score the AMG solver.

The score is calculated as \( \frac{1}{t_s} max(1,t_a/t_m) \)

where \( t_s \) is the time to solve the system \( t_a \) is the target accuracy \( t_m \) is the time of the method

Parameters

t_a	target accuracy

Definition at line 550 of file petsc_solver_AMG_report.hpp.

◆ test_coarsener()

void petsc_AMG_report::test_coarsener	(	SparseMatrix< double, int, PETSC_BASE > &	A,
		const Vector< double, PETSC_BASE > &	b
	)

inlineprivate

test the corasener for this problem

Parameters

A	matrix to invert
b	right-hand-side

Definition at line 124 of file petsc_solver_AMG_report.hpp.

◆ test_cycle_asym()

void petsc_AMG_report::test_cycle_asym	(	SparseMatrix< double, int, PETSC_BASE > &	A,
		const Vector< double, PETSC_BASE > &	b,
		openfpm::vector< size_t > &	ids_ts,
		openfpm::vector< std::string > &	coarsener_to_test
	)

inlineprivate

test best asymmetric combination of sweeps

Parameters

A	matrix to invert
b	right-hand-side
ids_ts	set of sweep configuration to test
coarsener_to_test	coarsener to test

Definition at line 404 of file petsc_solver_AMG_report.hpp.

◆ test_cycle_type()

void petsc_AMG_report::test_cycle_type	(	SparseMatrix< double, int, PETSC_BASE > &	A,
		const Vector< double, PETSC_BASE > &	b,
		openfpm::vector< std::string > &	coarsener_to_test
	)

inlineprivate

test the corasener for this problem

Parameters

A	matrix to invert
b	right-hand-side
coarsener_to_test	coarsener to test

Definition at line 350 of file petsc_solver_AMG_report.hpp.

◆ try_solve()

void petsc_AMG_report::try_solve	(	SparseMatrix< double, int, PETSC_BASE > &	A,
		const Vector< double, PETSC_BASE > &	b
	)

inline

Try to use AMG pre-conditioner and check how they they perform.

Parameters

A	Sparse matrix
b	right-hand side

Definition at line 575 of file petsc_solver_AMG_report.hpp.

◆ write_report()

void petsc_AMG_report::write_report ( openfpm::vector< std::string > & coars )

inlineprivate

Write the report on file

Parameters

coars set of coarsener tested

Definition at line 460 of file petsc_solver_AMG_report.hpp.

◆ write_report_coars()

void petsc_AMG_report::write_report_coars ( GoogleChart & cg )

inlineprivate

Write the report for coarsening.

Parameters

cg	Google chart

Definition at line 207 of file petsc_solver_AMG_report.hpp.

◆ write_report_cycle()

void petsc_AMG_report::write_report_cycle	(	GoogleChart &	cg,
		openfpm::vector< std::string > &	coars
	)

inlineprivate

Write the report for coarsening.

Parameters

cg	Google chart
coars	type of coarsening

Definition at line 250 of file petsc_solver_AMG_report.hpp.

◆ write_report_cycle_asym()

void petsc_AMG_report::write_report_cycle_asym	(	GoogleChart &	cg,
		openfpm::vector< std::string > &	coars
	)

inlineprivate

Write the report for coarsening.

Parameters

cg	Google chart
coars	type of coarsening

Definition at line 298 of file petsc_solver_AMG_report.hpp.

Field Documentation

◆ asym_tests

openfpm::vector<size_t> petsc_AMG_report::asym_tests

private

starting point of each coarsening test for the asymmetric test

Definition at line 69 of file petsc_solver_AMG_report.hpp.

◆ method

openfpm::vector<std::string> petsc_AMG_report::method

private

List of methods.

Definition at line 60 of file petsc_solver_AMG_report.hpp.

◆ num_sweep_test

size_t petsc_AMG_report::num_sweep_test = 10

private

Number of sweeps for test.

Definition at line 45 of file petsc_solver_AMG_report.hpp.

◆ perf_amg_coars

openfpm::vector<AMG_time_err_coars> petsc_AMG_report::perf_amg_coars

private

It contains the performance of several AMG methods coarsener.

Definition at line 51 of file petsc_solver_AMG_report.hpp.

◆ perf_amg_sweep_asym

openfpm::vector<AMG_time_err_coars> petsc_AMG_report::perf_amg_sweep_asym

private

It contain the performance of several AMG methods sweep configuration.

Definition at line 57 of file petsc_solver_AMG_report.hpp.

◆ perf_amg_sweep_sym

openfpm::vector<AMG_time_err_coars> petsc_AMG_report::perf_amg_sweep_sym

private

It contain the performance of several AMG methods sweep configuration.

Definition at line 54 of file petsc_solver_AMG_report.hpp.

◆ target_accuracy

double petsc_AMG_report::target_accuracy = 1

private

Target accuracy for the preconditioner.

Definition at line 48 of file petsc_solver_AMG_report.hpp.

◆ v_cycle_tested_asym

openfpm::vector<std::pair<size_t,size_t> > petsc_AMG_report::v_cycle_tested_asym

private

List of tested cycles in case of asymmetric.

Definition at line 66 of file petsc_solver_AMG_report.hpp.

◆ v_cycle_tested_sym

openfpm::vector<size_t> petsc_AMG_report::v_cycle_tested_sym

private

List of tested cycles.

Definition at line 63 of file petsc_solver_AMG_report.hpp.

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

openfpm_numerics/src/Solvers/petsc_solver_AMG_report.hpp

Detailed Description

Public Member Functions

Private Member Functions

Private Attributes

Member Function Documentation

◆ benchmark()

◆ best_coarsener()

◆ best_score()

◆ score_solver()

◆ setMaxSweep()

◆ setTergetAMGAccuracy()

◆ test_coarsener()

◆ test_cycle_asym()

◆ test_cycle_type()

◆ try_solve()

◆ write_report()

◆ write_report_coars()

◆ write_report_cycle()

◆ write_report_cycle_asym()

Field Documentation

◆ asym_tests

◆ method

◆ num_sweep_test

◆ perf_amg_coars

◆ perf_amg_sweep_asym

◆ perf_amg_sweep_sym

◆ target_accuracy

◆ v_cycle_tested_asym

◆ v_cycle_tested_sym