TEUCHOS_UNIT_TEST(NOX_Thyra_RightScaling, CTOR2)
{
// Create a communicator for Epetra objects
#ifdef HAVE_MPI
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
// Check we have only one processor since this problem doesn't work
// for more than one proc
TEST_ASSERT(Comm.NumProc() == 1);
// Create the model evaluator object
double d = 1.0;
double p0 = 1.0e6;
double p1 = -1.0e-6;
double x00 = 0.0;
double x01 = 0.0;
Teuchos::RCP<ModelEvaluator2DSim<double> > thyraModel =
Teuchos::rcp(new ModelEvaluator2DSim<double>(Teuchos::rcp(&Comm,false),
d,p0,p1,x00,x01));
::Stratimikos::DefaultLinearSolverBuilder builder;
Teuchos::RCP<Teuchos::ParameterList> p =
Teuchos::rcp(new Teuchos::ParameterList);
{
p->set("Linear Solver Type", "AztecOO");
//p->set("Preconditioner Type", "Ifpack");
p->set("Preconditioner Type", "None");
Teuchos::ParameterList& az = p->sublist("Linear Solver Types").sublist("AztecOO");
az.sublist("Forward Solve").sublist("AztecOO Settings").set("Output Frequency", 1);
az.sublist("VerboseObject").set("Verbosity Level", "high");
Teuchos::ParameterList& ip = p->sublist("Preconditioner Types").sublist("Ifpack");
ip.sublist("VerboseObject").set("Verbosity Level", "high");
}
builder.setParameterList(p);
Teuchos::RCP< ::Thyra::LinearOpWithSolveFactoryBase<double> >
lowsFactory = builder.createLinearSolveStrategy("");
thyraModel->set_W_factory(lowsFactory);
// Create nox parameter list
Teuchos::RCP<Teuchos::ParameterList> nl_params =
Teuchos::rcp(new Teuchos::ParameterList);
nl_params->set("Nonlinear Solver", "Line Search Based");
nl_params->sublist("Line Search").set("Method", "Full Step");
Teuchos::ParameterList& printParams = nl_params->sublist("Printing");
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Parameters +
NOX::Utils::Details +
NOX::Utils::Warning +
NOX::Utils::Debug +
NOX::Utils::TestDetails +
NOX::Utils::Error);
nl_params->sublist("Solver Options").set("Status Test Check Type", "Complete");
// Enable row sum scaling
nl_params->sublist("Thyra Group Options").set("Function Scaling", "None");
// Create right scaling vector
Teuchos::RCP<Thyra::VectorBase<double> > values = Thyra::createMember(thyraModel->get_x_space());
Thyra::put_scalar(0.0,values.ptr());
// Values chosen such that WRMS convergence is achieved in 1 step
// Unscaled will not converge in less than 20 steps
Thyra::set_ele(0,1.0e14,values.ptr());
Thyra::set_ele(1,1.0e2,values.ptr());
Teuchos::RCP<NOX::Abstract::Vector > right_scaling = Teuchos::rcp<NOX::Abstract::Vector>(new NOX::Thyra::Vector(values));
// Create Status Tests
Teuchos::RCP<NOX::StatusTest::Generic> status_tests;
{
Teuchos::RCP<Teuchos::ParameterList> st =
Teuchos::parameterList("Status Tests");
st->set("Test Type", "Combo");
st->set("Combo Type", "OR");
st->set("Number of Tests", 3);
{
Teuchos::ParameterList& normWRMS = st->sublist("Test 0");
normWRMS.set("Test Type", "NormWRMS");
normWRMS.set("Absolute Tolerance", 1.0e-8);
normWRMS.set("Relative Tolerance", 1.0e-5);
normWRMS.set("Tolerance", 1.0);
normWRMS.set("BDF Multiplier", 1.0);
normWRMS.set("Alpha", 1.0);
normWRMS.set("Beta", 0.5);
normWRMS.set("Disable Implicit Weighting", true);
}
{
Teuchos::ParameterList& fv = st->sublist("Test 1");
fv.set("Test Type", "FiniteValue");
fv.set("Vector Type", "F Vector");
fv.set("Norm Type", "Two Norm");
}
{
Teuchos::ParameterList& maxiters = st->sublist("Test 2");
maxiters.set("Test Type", "MaxIters");
maxiters.set("Maximum Iterations", 20);
}
// Create a print class for controlling output below
nl_params->sublist("Printing").set("MyPID", Comm.MyPID());
nl_params->sublist("Printing").set("Output Precision", 3);
nl_params->sublist("Printing").set("Output Processor", 0);
NOX::Utils printing(nl_params->sublist("Printing"));
NOX::StatusTest::Factory st_factory;
status_tests = st_factory.buildStatusTests(*st,printing);
}
Teuchos::RCP< ::Thyra::VectorBase<double> >
initial_guess = thyraModel->getNominalValues().get_x()->clone_v();
Teuchos::RCP<NOX::Thyra::Vector> noxThyraRightScaleVec =
Teuchos::rcp_dynamic_cast<NOX::Thyra::Vector>(right_scaling);
Teuchos::RCP<NOX::Thyra::Group> nox_group =
Teuchos::rcp(new NOX::Thyra::Group(*initial_guess,
thyraModel,
thyraModel->create_W_op(),
lowsFactory,
thyraModel->create_W_prec(),
lowsFactory->getPreconditionerFactory(),
Teuchos::null,
noxThyraRightScaleVec->getThyraRCPVector()
));
Teuchos::RCP<NOX::Solver::Generic> solver =
NOX::Solver::buildSolver(nox_group, status_tests, nl_params);
NOX::StatusTest::StatusType solveStatus = solver->solve();
TEST_ASSERT(solveStatus == NOX::StatusTest::Converged);
// Test total num iterations
{
// Problem converges in >20 nonlinear iterations with NO scaling
// Problem converges in 1 nonlinear iterations with scaling
TEST_EQUALITY(solver->getNumIterations(), 1);
}
}
int main(int argc, char *argv[])
{
bool boolret;
int MyPID;
#ifdef HAVE_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
MyPID = Comm.MyPID();
bool testFailed;
bool verbose = false;
bool debug = false;
bool shortrun = false;
bool insitu = false;
std::string which("LM");
CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("insitu","exsitu",&insitu,"Perform in situ restarting.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SM or LM).");
cmdp.setOption("shortrun","longrun",&shortrun,"Allow only a small number of iterations.");
if (cmdp.parse(argc,argv) != CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
if (debug) verbose = true;
typedef double ScalarType;
typedef ScalarTraits<ScalarType> SCT;
typedef SCT::magnitudeType MagnitudeType;
typedef Anasazi::MultiVec<ScalarType> MV;
typedef Anasazi::Operator<ScalarType> OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVT;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OPT;
const ScalarType ONE = SCT::one();
if (verbose && MyPID == 0) {
cout << Anasazi::Anasazi_Version() << endl << endl;
}
// Problem information
int space_dim = 1;
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
std::vector<int> elements( space_dim );
elements[0] = 100;
// Create problem
RCP<ModalProblem> testCase = rcp( new ModeLaplace1DQ1(Comm, brick_dim[0], elements[0]) );
//
// Get the stiffness and mass matrices
RCP<Epetra_CrsMatrix> K = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
RCP<Epetra_CrsMatrix> M = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
//
// Create solver for mass matrix
const int maxIterCG = 100;
const double tolCG = 1e-7;
RCP<BlockPCGSolver> opStiffness = rcp( new BlockPCGSolver(Comm, M.get(), tolCG, maxIterCG, 0) );
opStiffness->setPreconditioner( 0 );
RCP<const Anasazi::EpetraGenOp> InverseOp = rcp( new Anasazi::EpetraGenOp( opStiffness, K ) );
// Create the initial vectors
int blockSize = 3;
RCP<Anasazi::EpetraOpMultiVec> ivec = rcp( new Anasazi::EpetraOpMultiVec(K, K->OperatorDomainMap(), blockSize) );
ivec->MvRandom();
// Create eigenproblem
const int nev = 5;
RCP<Anasazi::BasicEigenproblem<ScalarType,MV,OP> > problem =
rcp( new Anasazi::BasicEigenproblem<ScalarType,MV,OP>(InverseOp, ivec) );
//
// Inform the eigenproblem that the operator InverseOp is Hermitian under an M inner-product
problem->setHermitian(true);
//
// Set the number of eigenvalues requested
problem->setNEV( nev );
//
// Inform the eigenproblem that you are done passing it information
boolret = problem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::SetProblem() returned with error." << endl
<< "End Result: TEST FAILED" << endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
// Eigensolver parameters
int numBlocks;
int maxRestarts;
if (shortrun) {
maxRestarts = 25;
numBlocks = 5;
}
else {
maxRestarts = 50;
numBlocks = 10;
}
int stepSize = numBlocks*maxRestarts;
MagnitudeType tol = tolCG * 10.0;
// Create a sort manager to pass into the block Krylov-Schur solver manager
// --> Make sure the reference-counted pointer is of type Anasazi::SortManager<>
// --> The block Krylov-Schur solver manager uses Anasazi::BasicSort<> by default,
// so you can also pass in the parameter "Which", instead of a sort manager.
RCP<Anasazi::SortManager<MagnitudeType> > MySort =
rcp( new Anasazi::BasicSort<MagnitudeType>( which ) );
//
// Create parameter list to pass into the solver manager
ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Sort Manager", MySort );
//MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Num Blocks", numBlocks );
MyPL.set( "Maximum Restarts", maxRestarts );
MyPL.set( "Step Size", stepSize );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "In Situ Restarting", insitu );
//
// Create the solver manager
Anasazi::BlockKrylovSchurSolMgr<ScalarType,MV,OP> MySolverMgr(problem, MyPL);
//
// Check that the parameters were all consumed
if (MyPL.getEntryPtr("Verbosity")->isUsed() == false ||
MyPL.getEntryPtr("Block Size")->isUsed() == false ||
MyPL.getEntryPtr("Num Blocks")->isUsed() == false ||
MyPL.getEntryPtr("Maximum Restarts")->isUsed() == false ||
MyPL.getEntryPtr("Step Size")->isUsed() == false ||
MyPL.getEntryPtr("In Situ Restarting")->isUsed() == false ||
MyPL.getEntryPtr("Convergence Tolerance")->isUsed() == false) {
if (verbose && MyPID==0) {
cout << "Failure! Unused parameters: " << endl;
MyPL.unused(cout);
}
}
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
testFailed = false;
if (returnCode != Anasazi::Converged && shortrun==false) {
testFailed = true;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ScalarType,MV> sol = problem->getSolution();
std::vector<Anasazi::Value<ScalarType> > evals = sol.Evals;
RCP<MV> evecs = sol.Evecs;
int numev = sol.numVecs;
if (numev > 0) {
std::ostringstream os;
os.setf(std::ios::scientific, std::ios::floatfield);
os.precision(6);
// Compute the direct residual
std::vector<ScalarType> normV( numev );
SerialDenseMatrix<int,ScalarType> T(numev,numev);
for (int i=0; i<numev; i++) {
T(i,i) = evals[i].realpart;
}
RCP<OP> KOp = rcp( new Anasazi::EpetraOp( K ));
RCP<OP> MOp = rcp( new Anasazi::EpetraOp( M ));
RCP<MV> Mvecs = MVT::Clone( *evecs, numev ),
Kvecs = MVT::Clone( *evecs, numev );
OPT::Apply( *KOp, *evecs, *Kvecs );
OPT::Apply( *MOp, *evecs, *Mvecs );
MVT::MvTimesMatAddMv( -ONE, *Mvecs, T, ONE, *Kvecs );
// compute 2-norm of residuals
std::vector<MagnitudeType> resnorm(numev);
MVT::MvNorm( *Kvecs, resnorm );
os << "Number of iterations performed in BlockKrylovSchur_test.exe: " << MySolverMgr.getNumIters() << endl
<< "Direct residual norms computed in BlockKrylovSchur_test.exe" << endl
<< std::setw(20) << "Eigenvalue" << std::setw(20) << "Residual" << endl
<< "----------------------------------------" << endl;
for (int i=0; i<numev; i++) {
if ( SCT::magnitude(evals[i].realpart) != SCT::zero() ) {
resnorm[i] = SCT::magnitude( SCT::squareroot( resnorm[i] ) / evals[i].realpart );
}
else {
resnorm[i] = SCT::magnitude( SCT::squareroot( resnorm[i] ) );
}
os << std::setw(20) << evals[i].realpart << std::setw(20) << resnorm[i] << endl;
if ( resnorm[i] > tol ) {
testFailed = true;
}
}
if (verbose && MyPID==0) {
cout << endl << os.str() << endl;
}
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
if (testFailed) {
if (verbose && MyPID==0) {
cout << "End Result: TEST FAILED" << endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
cout << "End Result: TEST PASSED" << endl;
}
return 0;
}
int main(int argc, char *argv[]) {
using std::cout;
using std::endl;
int i;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
int MyPID = Comm.MyPID();
// Number of dimension of the domain
int space_dim = 2;
// Size of each of the dimensions of the domain
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
brick_dim[1] = 1.0;
// Number of elements in each of the dimensions of the domain
std::vector<int> elements( space_dim );
elements[0] = 10;
elements[1] = 10;
// Create problem
Teuchos::RCP<ModalProblem> testCase = Teuchos::rcp( new ModeLaplace2DQ2(Comm, brick_dim[0], elements[0], brick_dim[1], elements[1]) );
// Get the stiffness and mass matrices
Teuchos::RCP<Epetra_CrsMatrix> K = Teuchos::rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
Teuchos::RCP<Epetra_CrsMatrix> M = Teuchos::rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
//
// ************Construct preconditioner*************
//
Teuchos::ParameterList ifpackList;
// allocates an IFPACK factory. No data is associated
// to this object (only method Create()).
Ifpack Factory;
// create the preconditioner. For valid PrecType values,
// please check the documentation
std::string PrecType = "ICT"; // incomplete Cholesky
int OverlapLevel = 0; // must be >= 0. If Comm.NumProc() == 1,
// it is ignored.
Teuchos::RCP<Ifpack_Preconditioner> Prec = Teuchos::rcp( Factory.Create(PrecType, &*K, OverlapLevel) );
assert(Prec != Teuchos::null);
// specify parameters for ICT
ifpackList.set("fact: drop tolerance", 1e-4);
ifpackList.set("fact: ict level-of-fill", 0.);
// the combine mode is on the following:
// "Add", "Zero", "Insert", "InsertAdd", "Average", "AbsMax"
// Their meaning is as defined in file Epetra_CombineMode.h
ifpackList.set("schwarz: combine mode", "Add");
// sets the parameters
IFPACK_CHK_ERR(Prec->SetParameters(ifpackList));
// initialize the preconditioner. At this point the matrix must
// have been FillComplete()'d, but actual values are ignored.
IFPACK_CHK_ERR(Prec->Initialize());
// Builds the preconditioners, by looking for the values of
// the matrix.
IFPACK_CHK_ERR(Prec->Compute());
//
//*******************************************************/
// Set up Belos Block CG operator for inner iteration
//*******************************************************/
//
int blockSize = 3; // block size used by linear solver and eigensolver [ not required to be the same ]
int maxits = K->NumGlobalRows(); // maximum number of iterations to run
//
// Create the Belos::LinearProblem
//
Teuchos::RCP<Belos::LinearProblem<double,Epetra_MultiVector,Epetra_Operator> >
My_LP = Teuchos::rcp( new Belos::LinearProblem<double,Epetra_MultiVector,Epetra_Operator>() );
My_LP->setOperator( K );
// Create the Belos preconditioned operator from the Ifpack preconditioner.
// NOTE: This is necessary because Belos expects an operator to apply the
// preconditioner with Apply() NOT ApplyInverse().
Teuchos::RCP<Epetra_Operator> belosPrec = Teuchos::rcp( new Epetra_InvOperator( Prec.get() ) );
My_LP->setLeftPrec( belosPrec );
//
// Create the ParameterList for the Belos Operator
//
Teuchos::RCP<Teuchos::ParameterList> My_List = Teuchos::rcp( new Teuchos::ParameterList() );
My_List->set( "Solver", "BlockCG" );
My_List->set( "Maximum Iterations", maxits );
My_List->set( "Block Size", 1 );
My_List->set( "Convergence Tolerance", 1e-12 );
//
// Create the Belos::EpetraOperator
//
Teuchos::RCP<Belos::EpetraOperator> BelosOp =
Teuchos::rcp( new Belos::EpetraOperator( My_LP, My_List ));
//
// ************************************
// Start the block Arnoldi iteration
// ************************************
//
// Variables used for the Block Arnoldi Method
//
double tol = 1.0e-8;
int nev = 10;
int numBlocks = 3*nev/blockSize;
int maxRestarts = 5;
//int step = 5;
std::string which = "LM";
int verbosity = Anasazi::Errors + Anasazi::Warnings + Anasazi::FinalSummary;
//
// Create parameter list to pass into solver
//
Teuchos::ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Num Blocks", numBlocks );
MyPL.set( "Maximum Restarts", maxRestarts );
MyPL.set( "Convergence Tolerance", tol );
//MyPL.set( "Step Size", step );
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<double, MV> MVT;
typedef Anasazi::OperatorTraits<double, MV, OP> OPT;
// Create an Epetra_MultiVector for an initial vector to start the solver.
// Note: This needs to have the same number of columns as the blocksize.
Teuchos::RCP<Epetra_MultiVector> ivec = Teuchos::rcp( new Epetra_MultiVector(K->Map(), blockSize) );
MVT::MvRandom( *ivec );
// Call the ctor that calls the petra ctor for a matrix
Teuchos::RCP<Anasazi::EpetraGenOp> Aop = Teuchos::rcp( new Anasazi::EpetraGenOp(BelosOp, M, false) );
Teuchos::RCP<Anasazi::BasicEigenproblem<double,MV,OP> > MyProblem =
Teuchos::rcp( new Anasazi::BasicEigenproblem<double,MV,OP>(Aop, M, ivec) );
// Inform the eigenproblem that the matrix pencil (K,M) is symmetric
MyProblem->setHermitian(true);
// Set the number of eigenvalues requested
MyProblem->setNEV( nev );
// Inform the eigenproblem that you are finished passing it information
bool boolret = MyProblem->setProblem();
if (boolret != true) {
if (MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::setProblem() returned with error." << endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Initialize the Block Arnoldi solver
Anasazi::BlockKrylovSchurSolMgr<double, MV, OP> MySolverMgr(MyProblem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
if (returnCode != Anasazi::Converged && MyPID==0) {
cout << "Anasazi::EigensolverMgr::solve() returned unconverged." << endl;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<double,MV> sol = MyProblem->getSolution();
std::vector<Anasazi::Value<double> > evals = sol.Evals;
Teuchos::RCP<MV> evecs = sol.Evecs;
int numev = sol.numVecs;
if (numev > 0) {
Teuchos::SerialDenseMatrix<int,double> dmatr(numev,numev);
Epetra_MultiVector tempvec(K->Map(), MVT::GetNumberVecs( *evecs ));
OPT::Apply( *K, *evecs, tempvec );
MVT::MvTransMv( 1.0, tempvec, *evecs, dmatr );
if (MyPID==0) {
double compeval = 0.0;
cout.setf(std::ios_base::right, std::ios_base::adjustfield);
cout<<"Actual Eigenvalues (obtained by Rayleigh quotient) : "<<endl;
cout<<"------------------------------------------------------"<<endl;
cout<<std::setw(16)<<"Real Part"
<<std::setw(16)<<"Rayleigh Error"<<endl;
cout<<"------------------------------------------------------"<<endl;
for (i=0; i<numev; i++) {
compeval = dmatr(i,i);
cout<<std::setw(16)<<compeval
<<std::setw(16)<<Teuchos::ScalarTraits<double>::magnitude(compeval-1.0/evals[i].realpart)
<<endl;
}
cout<<"------------------------------------------------------"<<endl;
}
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return 0;
}
int main(int argc, char *argv[])
{
using Teuchos::rcp_implicit_cast;
using std::cout;
using std::endl;
bool boolret;
int MyPID;
#ifdef HAVE_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
MyPID = Comm.MyPID();
bool testFailed = false;
bool verbose = false;
bool debug = false;
std::string filename("mhd1280b.cua");
std::string which("LR");
bool skinny = true;
bool success = true;
try {
CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SR or LR).");
cmdp.setOption("skinny","hefty",&skinny,"Use a skinny (low-mem) or hefty (higher-mem) implementation of IRTR.");
if (cmdp.parse(argc,argv) != CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
#ifndef HAVE_EPETRA_THYRA
if (verbose && MyPid == 0) {
cout << "Please configure Anasazi with:" << endl;
cout << "--enable-epetra-thyra" << endl;
cout << "--enable-anasazi-thyra" << endl;
}
return 0;
#endif
typedef double ScalarType;
typedef ScalarTraits<ScalarType> SCT;
typedef SCT::magnitudeType MagnitudeType;
typedef Thyra::MultiVectorBase<double> MV;
typedef Thyra::LinearOpBase<double> OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVT;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OPT;
const ScalarType ONE = SCT::one();
if (verbose && MyPID == 0) {
cout << Anasazi::Anasazi_Version() << endl << endl;
}
// Problem information
int space_dim = 1;
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
std::vector<int> elements( space_dim );
elements[0] = 100;
// Create problem
RCP<ModalProblem> testCase = rcp( new ModeLaplace1DQ1(Comm, brick_dim[0], elements[0]) );
//
// Get the stiffness and mass matrices
RCP<Epetra_CrsMatrix> K = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
RCP<Epetra_CrsMatrix> M = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
//
// Create the initial vectors
int blockSize = 5;
//
// Get a pointer to the Epetra_Map
RCP<const Epetra_Map> Map = rcp( &K->OperatorDomainMap(), false );
//
// create an epetra multivector
RCP<Epetra_MultiVector> ivec =
rcp( new Epetra_MultiVector(K->OperatorDomainMap(), blockSize) );
ivec->Random();
// create a Thyra::VectorSpaceBase
RCP<const Thyra::VectorSpaceBase<double> > epetra_vs =
Thyra::create_VectorSpace(Map);
// create a MultiVectorBase (from the Epetra_MultiVector)
RCP<Thyra::MultiVectorBase<double> > thyra_ivec =
Thyra::create_MultiVector(ivec, epetra_vs);
// Create Thyra LinearOpBase objects from the Epetra_Operator objects
RCP<const Thyra::LinearOpBase<double> > thyra_K =
Thyra::epetraLinearOp(K);
RCP<const Thyra::LinearOpBase<double> > thyra_M =
Thyra::epetraLinearOp(M);
// Create eigenproblem
const int nev = 5;
RCP<Anasazi::BasicEigenproblem<ScalarType,MV,OP> > problem =
rcp( new Anasazi::BasicEigenproblem<ScalarType,MV,OP>(thyra_K,thyra_M,thyra_ivec) );
//
// Inform the eigenproblem that the operator K is symmetric
problem->setHermitian(true);
//
// Set the number of eigenvalues requested
problem->setNEV( nev );
//
// Inform the eigenproblem that you are done passing it information
boolret = problem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::SetProblem() returned with error." << endl
<< "End Result: TEST FAILED" << endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
// Eigensolver parameters
int maxIters = 450;
MagnitudeType tol = 1.0e-6;
//
// Create parameter list to pass into the solver manager
ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Maximum Iterations", maxIters );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Skinny Solver", skinny);
//
// Create the solver manager
Anasazi::RTRSolMgr<ScalarType,MV,OP> MySolverMan(problem, MyPL);
//
// Check that the parameters were all consumed
if (MyPL.getEntryPtr("Verbosity")->isUsed() == false ||
MyPL.getEntryPtr("Which")->isUsed() == false ||
MyPL.getEntryPtr("Block Size")->isUsed() == false ||
MyPL.getEntryPtr("Maximum Iterations")->isUsed() == false ||
MyPL.getEntryPtr("Convergence Tolerance")->isUsed() == false ||
MyPL.getEntryPtr("Skinny Solver")->isUsed() == false) {
if (verbose && MyPID==0) {
cout << "Failure! Unused parameters: " << endl;
MyPL.unused(cout);
}
}
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMan.solve();
if (returnCode != Anasazi::Converged) {
testFailed = true;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ScalarType,MV> sol = problem->getSolution();
std::vector<Anasazi::Value<ScalarType> > evals = sol.Evals;
RCP<MV> evecs = sol.Evecs;
int numev = sol.numVecs;
if (numev > 0) {
std::ostringstream os;
os.setf(std::ios::scientific, std::ios::floatfield);
os.precision(6);
// Compute the direct residual
std::vector<ScalarType> normV( numev );
SerialDenseMatrix<int,ScalarType> T(numev,numev);
for (int i=0; i<numev; i++) {
T(i,i) = evals[i].realpart;
}
RCP<MV> Mvecs = MVT::Clone( *evecs, numev ),
Kvecs = MVT::Clone( *evecs, numev );
OPT::Apply( *thyra_K, *evecs, *Kvecs );
OPT::Apply( *thyra_M, *evecs, *Mvecs );
MVT::MvTimesMatAddMv( -ONE, *Mvecs, T, ONE, *Kvecs );
// compute M-norm of residuals
OPT::Apply( *thyra_M, *Kvecs, *Mvecs );
MVT::MvDot( *Mvecs, *Kvecs, normV );
os << "Direct residual norms computed in LOBPCGThyra_test.exe" << endl
<< std::setw(20) << "Eigenvalue" << std::setw(20) << "Residual(M)" << endl
<< "----------------------------------------" << endl;
for (int i=0; i<numev; i++) {
if ( SCT::magnitude(evals[i].realpart) != SCT::zero() ) {
normV[i] = SCT::magnitude( SCT::squareroot( normV[i] ) / evals[i].realpart );
}
else {
normV[i] = SCT::magnitude( SCT::squareroot( normV[i] ) );
}
os << std::setw(20) << evals[i].realpart << std::setw(20) << normV[i] << endl;
if ( normV[i] > tol ) {
testFailed = true;
}
}
if (verbose && MyPID==0) {
cout << endl << os.str() << endl;
}
}
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(true,cout,success);
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
if (testFailed || success==false) {
if (verbose && MyPID==0) {
cout << "End Result: TEST FAILED" << endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
cout << "End Result: TEST PASSED" << endl;
}
return 0;
}
int main(int argc, char *argv[])
{
int ierr = 0;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
int rank; // My process ID
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
int rank = 0;
Epetra_SerialComm Comm;
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
int verbose_int = verbose ? 1 : 0;
Comm.Broadcast(&verbose_int, 1, 0);
verbose = verbose_int==1 ? true : false;
// char tmp;
// if (rank==0) cout << "Press any key to continue..."<< std::endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if(verbose && MyPID==0)
cout << Epetra_Version() << std::endl << std::endl;
if (verbose) cout << "Processor "<<MyPID<<" of "<< NumProc
<< " is alive."<<endl;
//bool verbose1 = verbose; // unused
// Redefine verbose to only print on PE 0
if(verbose && rank!=0) verbose = false;
int NumMyEquations = 1;
long long NumGlobalEquations = NumProc;
// Get update list and number of local equations from newly created Map
long long* MyGlobalElementsLL = new long long[NumMyEquations];
MyGlobalElementsLL[0] = 2000000000+MyPID;
// Construct a Map that puts approximately the same Number of equations on each processor
Epetra_Map MapLL(NumGlobalEquations, NumMyEquations, MyGlobalElementsLL, 0, Comm);
EPETRA_TEST_ERR(MapLL.GlobalIndicesInt(),ierr);
EPETRA_TEST_ERR(!(MapLL.GlobalIndicesLongLong()),ierr);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation on this processor
int* NumNzLL = new int[NumMyEquations];
NumNzLL[0] = 0;
// Create int types meant to add to long long matrix for test of failure
int* MyIntGlobalElementsLL = new int[NumMyEquations];
MyIntGlobalElementsLL[0] = 20000+MyPID;
// Create a long long Epetra_Matrix
Epetra_CrsMatrix A_LL(Copy, MapLL, NumNzLL);
EPETRA_TEST_ERR(A_LL.IndicesAreGlobal(),ierr);
EPETRA_TEST_ERR(A_LL.IndicesAreLocal(),ierr);
// Insert values
double one = 1.0;
#ifndef EPETRA_NO_32BIT_GLOBAL_INDICES
// Try to add ints which should fail and be caught as an int
try {
A_LL.InsertGlobalValues(MyIntGlobalElementsLL[0], 1, &one, MyIntGlobalElementsLL+0);
} catch(int i) {
EPETRA_TEST_ERR(!(i==-1),ierr);
}
#endif
// Add long longs which should succeed
EPETRA_TEST_ERR(!(A_LL.InsertGlobalValues(MyGlobalElementsLL[0], 1, &one, MyGlobalElementsLL+0)==0),ierr);
EPETRA_TEST_ERR(!(A_LL.IndicesAreGlobal()),ierr);
EPETRA_TEST_ERR(!(A_LL.FillComplete(false)==0),ierr);
EPETRA_TEST_ERR(!(A_LL.IndicesAreLocal()),ierr);
// Get update list and number of local equations from newly created Map
int* MyGlobalElementsInt = new int[NumMyEquations];
MyGlobalElementsInt[0] = 2000+MyPID;
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation on this processor
int* NumNzInt = new int[NumMyEquations];
NumNzInt[0] = 0;
// Create int types meant to add to long long matrix for test of failure
long long* MyLLGlobalElementsInt = new long long[NumMyEquations];
MyLLGlobalElementsInt[0] = 2000000000+MyPID;
#ifndef EPETRA_NO_32BIT_GLOBAL_INDICES
// Construct a Map that puts approximately the same Number of equations on each processor
Epetra_Map MapInt(NumGlobalEquations, NumMyEquations, MyGlobalElementsInt, 0LL, Comm);
EPETRA_TEST_ERR(!(MapInt.GlobalIndicesInt()),ierr);
EPETRA_TEST_ERR(MapInt.GlobalIndicesLongLong(),ierr);
// Create a int Epetra_Matrix
Epetra_CrsMatrix A_Int(Copy, MapInt, NumNzInt);
EPETRA_TEST_ERR(A_Int.IndicesAreGlobal(),ierr);
EPETRA_TEST_ERR(A_Int.IndicesAreLocal(),ierr);
// Insert values
try {
A_Int.InsertGlobalValues(MyLLGlobalElementsInt[0], 1, &one, MyLLGlobalElementsInt+0);
} catch(int i) {
EPETRA_TEST_ERR(!(i==-1),ierr);
}
// Add long longs which should succeed
EPETRA_TEST_ERR(!(A_Int.InsertGlobalValues(MyGlobalElementsInt[0], 1, &one, MyGlobalElementsInt+0)==0),ierr);
EPETRA_TEST_ERR(!(A_Int.IndicesAreGlobal()),ierr);
EPETRA_TEST_ERR(!(A_Int.FillComplete(false)==0),ierr);
EPETRA_TEST_ERR(!(A_Int.IndicesAreLocal()),ierr);
#endif
delete [] MyGlobalElementsLL;
delete [] NumNzLL;
delete [] MyIntGlobalElementsLL;
delete [] MyGlobalElementsInt;
delete [] NumNzInt;
delete [] MyLLGlobalElementsInt;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
int main(int argc, char *argv[])
{
int ierr = 0;
double elapsed_time;
double total_flops;
double MFLOPs;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm comm;
#endif
bool verbose = false;
bool summary = false;
// Check if we should print verbose results to standard out
if (argc>6) if (argv[6][0]=='-' && argv[6][1]=='v') verbose = true;
// Check if we should print verbose results to standard out
if (argc>6) if (argv[6][0]=='-' && argv[6][1]=='s') summary = true;
if(argc < 6) {
cerr << "Usage: " << argv[0]
<< " NumNodesX NumNodesY NumProcX NumProcY NumPoints [-v|-s]" << endl
<< "where:" << endl
<< "NumNodesX - Number of mesh nodes in X direction per processor" << endl
<< "NumNodesY - Number of mesh nodes in Y direction per processor" << endl
<< "NumProcX - Number of processors to use in X direction" << endl
<< "NumProcY - Number of processors to use in Y direction" << endl
<< "NumPoints - Number of points to use in stencil (5, 9 or 25 only)" << endl
<< "-v|-s - (Optional) Run in verbose mode if -v present or summary mode if -s present" << endl
<< " NOTES: NumProcX*NumProcY must equal the number of processors used to run the problem." << endl << endl
<< " Serial example:" << endl
<< argv[0] << " 16 12 1 1 25 -v" << endl
<< " Run this program in verbose mode on 1 processor using a 16 X 12 grid with a 25 point stencil."<< endl <<endl
<< " MPI example:" << endl
<< "mpirun -np 32 " << argv[0] << " 10 12 4 8 9 -v" << endl
<< " Run this program in verbose mode on 32 processors putting a 10 X 12 subgrid on each processor using 4 processors "<< endl
<< " in the X direction and 8 in the Y direction. Total grid size is 40 points in X and 96 in Y with a 9 point stencil."<< endl
<< endl;
return(1);
}
//char tmp;
//if (comm.MyPID()==0) cout << "Press any key to continue..."<< endl;
//if (comm.MyPID()==0) cin >> tmp;
//comm.Barrier();
comm.SetTracebackMode(0); // This should shut down any error traceback reporting
if (verbose && comm.MyPID()==0)
cout << Epetra_Version() << endl << endl;
if (summary && comm.MyPID()==0) {
if (comm.NumProc()==1)
cout << Epetra_Version() << endl << endl;
else
cout << endl << endl; // Print two blank line to keep output columns lined up
}
if (verbose) cout << comm <<endl;
// Redefine verbose to only print on PE 0
if (verbose && comm.MyPID()!=0) verbose = false;
if (summary && comm.MyPID()!=0) summary = false;
int numNodesX = atoi(argv[1]);
int numNodesY = atoi(argv[2]);
int numProcsX = atoi(argv[3]);
int numProcsY = atoi(argv[4]);
int numPoints = atoi(argv[5]);
if (verbose || (summary && comm.NumProc()==1)) {
cout << " Number of local nodes in X direction = " << numNodesX << endl
<< " Number of local nodes in Y direction = " << numNodesY << endl
<< " Number of global nodes in X direction = " << numNodesX*numProcsX << endl
<< " Number of global nodes in Y direction = " << numNodesY*numProcsY << endl
<< " Number of local nonzero entries = " << numNodesX*numNodesY*numPoints << endl
<< " Number of global nonzero entries = " << numNodesX*numNodesY*numPoints*numProcsX*numProcsY << endl
<< " Number of Processors in X direction = " << numProcsX << endl
<< " Number of Processors in Y direction = " << numProcsY << endl
<< " Number of Points in stencil = " << numPoints << endl << endl;
}
// Print blank line to keep output columns lined up
if (summary && comm.NumProc()>1)
cout << endl << endl << endl << endl << endl << endl << endl << endl<< endl << endl;
if (numProcsX*numProcsY!=comm.NumProc()) {
cerr << "Number of processors = " << comm.NumProc() << endl
<< " is not the product of " << numProcsX << " and " << numProcsY << endl << endl;
return(1);
}
if (numPoints!=5 && numPoints!=9 && numPoints!=25) {
cerr << "Number of points specified = " << numPoints << endl
<< " is not 5, 9, 25" << endl << endl;
return(1);
}
if (numNodesX*numNodesY<=0) {
cerr << "Product of number of nodes is <= zero" << endl << endl;
return(1);
}
Epetra_IntSerialDenseVector Xoff, XLoff, XUoff;
Epetra_IntSerialDenseVector Yoff, YLoff, YUoff;
if (numPoints==5) {
// Generate a 5-point 2D Finite Difference matrix
Xoff.Size(5);
Yoff.Size(5);
Xoff[0] = -1;
Xoff[1] = 1;
Xoff[2] = 0;
Xoff[3] = 0;
Xoff[4] = 0;
Yoff[0] = 0;
Yoff[1] = 0;
Yoff[2] = 0;
Yoff[3] = -1;
Yoff[4] = 1;
// Generate a 2-point 2D Lower triangular Finite Difference matrix
XLoff.Size(2);
YLoff.Size(2);
XLoff[0] = -1;
XLoff[1] = 0;
YLoff[0] = 0;
YLoff[1] = -1;
// Generate a 3-point 2D upper triangular Finite Difference matrix
XUoff.Size(3);
YUoff.Size(3);
XUoff[0] = 0;
XUoff[1] = 1;
XUoff[2] = 0;
YUoff[0] = 0;
YUoff[1] = 0;
YUoff[2] = 1;
}
else if (numPoints==9) {
// Generate a 9-point 2D Finite Difference matrix
Xoff.Size(9);
Yoff.Size(9);
Xoff[0] = -1;
Xoff[1] = 0;
Xoff[2] = 1;
Yoff[0] = -1;
Yoff[1] = -1;
Yoff[2] = -1;
Xoff[3] = -1;
Xoff[4] = 0;
Xoff[5] = 1;
Yoff[3] = 0;
Yoff[4] = 0;
Yoff[5] = 0;
Xoff[6] = -1;
Xoff[7] = 0;
Xoff[8] = 1;
Yoff[6] = 1;
Yoff[7] = 1;
Yoff[8] = 1;
// Generate a 5-point lower triangular 2D Finite Difference matrix
XLoff.Size(5);
YLoff.Size(5);
XLoff[0] = -1;
XLoff[1] = 0;
Xoff[2] = 1;
YLoff[0] = -1;
YLoff[1] = -1;
Yoff[2] = -1;
XLoff[3] = -1;
XLoff[4] = 0;
YLoff[3] = 0;
YLoff[4] = 0;
// Generate a 4-point upper triangular 2D Finite Difference matrix
XUoff.Size(4);
YUoff.Size(4);
XUoff[0] = 1;
YUoff[0] = 0;
XUoff[1] = -1;
XUoff[2] = 0;
XUoff[3] = 1;
YUoff[1] = 1;
YUoff[2] = 1;
YUoff[3] = 1;
}
else {
// Generate a 25-point 2D Finite Difference matrix
Xoff.Size(25);
Yoff.Size(25);
int xi = 0, yi = 0;
int xo = -2, yo = -2;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Xoff[xi++] = xo++;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
Yoff[yi++] = yo ;
// Generate a 13-point lower triangular 2D Finite Difference matrix
XLoff.Size(13);
YLoff.Size(13);
xi = 0, yi = 0;
xo = -2, yo = -2;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
xo = -2, yo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
xo = -2, yo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
XLoff[xi++] = xo++;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
YLoff[yi++] = yo ;
// Generate a 13-point upper triangular 2D Finite Difference matrix
XUoff.Size(13);
YUoff.Size(13);
xi = 0, yi = 0;
xo = 0, yo = 0;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
xo = -2, yo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
xo = -2, yo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
XUoff[xi++] = xo++;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
YUoff[yi++] = yo ;
}
Epetra_Map * map;
Epetra_Map * mapL;
Epetra_Map * mapU;
Epetra_CrsMatrix * A;
Epetra_CrsMatrix * L;
Epetra_CrsMatrix * U;
Epetra_MultiVector * b;
Epetra_MultiVector * bt;
Epetra_MultiVector * xexact;
Epetra_MultiVector * bL;
Epetra_MultiVector * btL;
Epetra_MultiVector * xexactL;
Epetra_MultiVector * bU;
Epetra_MultiVector * btU;
Epetra_MultiVector * xexactU;
Epetra_SerialDenseVector resvec(0);
//Timings
Epetra_Flops flopcounter;
Epetra_Time timer(comm);
#ifdef EPETRA_VERY_SHORT_PERFTEST
int jstop = 1;
#elif EPETRA_SHORT_PERFTEST
int jstop = 1;
#else
int jstop = 2;
#endif
for (int j=0; j<jstop; j++) {
for (int k=1; k<17; k++) {
#ifdef EPETRA_VERY_SHORT_PERFTEST
if (k<3 || (k%4==0 && k<9)) {
#elif EPETRA_SHORT_PERFTEST
if (k<6 || k%4==0) {
#else
if (k<7 || k%2==0) {
#endif
int nrhs=k;
if (verbose) cout << "\n*************** Results for " << nrhs << " RHS with ";
bool StaticProfile = (j!=0);
if (verbose) {
if (StaticProfile) cout << " static profile\n";
else cout << " dynamic profile\n";
}
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, numPoints,
Xoff.Values(), Yoff.Values(), nrhs, comm, verbose, summary,
map, A, b, bt, xexact, StaticProfile, false);
#ifdef EPETRA_HAVE_JADMATRIX
timer.ResetStartTime();
Epetra_JadMatrix JA(*A);
elapsed_time = timer.ElapsedTime();
if (verbose) cout << "Time to create Jagged diagonal matrix = " << elapsed_time << endl;
//cout << "A = " << *A << endl;
//cout << "JA = " << JA << endl;
runJadMatrixTests(&JA, b, bt, xexact, StaticProfile, verbose, summary);
#endif
runMatrixTests(A, b, bt, xexact, StaticProfile, verbose, summary);
delete A;
delete b;
delete bt;
delete xexact;
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, XLoff.Length(),
XLoff.Values(), YLoff.Values(), nrhs, comm, verbose, summary,
mapL, L, bL, btL, xexactL, StaticProfile, true);
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, XUoff.Length(),
XUoff.Values(), YUoff.Values(), nrhs, comm, verbose, summary,
mapU, U, bU, btU, xexactU, StaticProfile, true);
runLUMatrixTests(L, bL, btL, xexactL, U, bU, btU, xexactU, StaticProfile, verbose, summary);
delete L;
delete bL;
delete btL;
delete xexactL;
delete mapL;
delete U;
delete bU;
delete btU;
delete xexactU;
delete mapU;
Epetra_MultiVector q(*map, nrhs);
Epetra_MultiVector z(q);
Epetra_MultiVector r(q);
delete map;
q.SetFlopCounter(flopcounter);
z.SetFlopCounter(q);
r.SetFlopCounter(q);
resvec.Resize(nrhs);
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 norms
for( int i = 0; i < 10; ++i )
q.Norm2( resvec.Values() );
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\nTotal MFLOPs for 10 Norm2's= " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "Norm2" << '\t';
cout << MFLOPs << endl;
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 dot's
for( int i = 0; i < 10; ++i )
q.Dot(z, resvec.Values());
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 Dot's = " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "DotProd" << '\t';
cout << MFLOPs << endl;
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 dot's
for( int i = 0; i < 10; ++i )
q.Update(1.0, z, 1.0, r, 0.0);
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 Updates= " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "Update" << '\t';
cout << MFLOPs << endl;
}
}
}
}
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return ierr ;
}
// Constructs a 2D PDE finite difference matrix using the list of x and y offsets.
//
// nx (In) - number of grid points in x direction
// ny (In) - number of grid points in y direction
// The total number of equations will be nx*ny ordered such that the x direction changes
// most rapidly:
// First equation is at point (0,0)
// Second at (1,0)
// ...
// nx equation at (nx-1,0)
// nx+1st equation at (0,1)
// numPoints (In) - number of points in finite difference stencil
// xoff (In) - stencil offsets in x direction (of length numPoints)
// yoff (In) - stencil offsets in y direction (of length numPoints)
// A standard 5-point finite difference stencil would be described as:
// numPoints = 5
// xoff = [-1, 1, 0, 0, 0]
// yoff = [ 0, 0, 0, -1, 1]
// nrhs - Number of rhs to generate. (First interface produces vectors, so nrhs is not needed
// comm (In) - an Epetra_Comm object describing the parallel machine (numProcs and my proc ID)
// map (Out) - Epetra_Map describing distribution of matrix and vectors/multivectors
// A (Out) - Epetra_CrsMatrix constructed for nx by ny grid using prescribed stencil
// Off-diagonal values are random between 0 and 1. If diagonal is part of stencil,
// diagonal will be slightly diag dominant.
// b (Out) - Generated RHS. Values satisfy b = A*xexact
// bt (Out) - Generated RHS. Values satisfy b = A'*xexact
// xexact (Out) - Generated exact solution to Ax = b and b' = A'xexact
// Note: Caller of this function is responsible for deleting all output objects.
void GenerateCrsProblem(int numNodesX, int numNodesY, int numProcsX, int numProcsY, int numPoints,
int * xoff, int * yoff,
const Epetra_Comm &comm, bool verbose, bool summary,
Epetra_Map *& map,
Epetra_CrsMatrix *& A,
Epetra_Vector *& b,
Epetra_Vector *& bt,
Epetra_Vector *&xexact, bool StaticProfile, bool MakeLocalOnly) {
Epetra_MultiVector * b1, * bt1, * xexact1;
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, numPoints,
xoff, yoff, 1, comm, verbose, summary,
map, A, b1, bt1, xexact1, StaticProfile, MakeLocalOnly);
b = dynamic_cast<Epetra_Vector *>(b1);
bt = dynamic_cast<Epetra_Vector *>(bt1);
xexact = dynamic_cast<Epetra_Vector *>(xexact1);
return;
}
void GenerateCrsProblem(int numNodesX, int numNodesY, int numProcsX, int numProcsY, int numPoints,
int * xoff, int * yoff, int nrhs,
const Epetra_Comm &comm, bool verbose, bool summary,
Epetra_Map *& map,
Epetra_CrsMatrix *& A,
Epetra_MultiVector *& b,
Epetra_MultiVector *& bt,
Epetra_MultiVector *&xexact, bool StaticProfile, bool MakeLocalOnly) {
Epetra_Time timer(comm);
// Determine my global IDs
long long * myGlobalElements;
GenerateMyGlobalElements(numNodesX, numNodesY, numProcsX, numProcsY, comm.MyPID(), myGlobalElements);
int numMyEquations = numNodesX*numNodesY;
map = new Epetra_Map((long long)-1, numMyEquations, myGlobalElements, 0, comm); // Create map with 2D block partitioning.
delete [] myGlobalElements;
long long numGlobalEquations = map->NumGlobalElements64();
int profile = 0;
if (StaticProfile) profile = numPoints;
#ifdef EPETRA_HAVE_STATICPROFILE
if (MakeLocalOnly)
A = new Epetra_CrsMatrix(Copy, *map, *map, profile, StaticProfile); // Construct matrix with rowmap=colmap
else
A = new Epetra_CrsMatrix(Copy, *map, profile, StaticProfile); // Construct matrix
#else
if (MakeLocalOnly)
A = new Epetra_CrsMatrix(Copy, *map, *map, profile); // Construct matrix with rowmap=colmap
else
A = new Epetra_CrsMatrix(Copy, *map, profile); // Construct matrix
#endif
long long * indices = new long long[numPoints];
double * values = new double[numPoints];
double dnumPoints = (double) numPoints;
int nx = numNodesX*numProcsX;
for (int i=0; i<numMyEquations; i++) {
long long rowID = map->GID64(i);
int numIndices = 0;
for (int j=0; j<numPoints; j++) {
long long colID = rowID + xoff[j] + nx*yoff[j]; // Compute column ID based on stencil offsets
if (colID>-1 && colID<numGlobalEquations) {
indices[numIndices] = colID;
double value = - ((double) rand())/ ((double) RAND_MAX);
if (colID==rowID)
values[numIndices++] = dnumPoints - value; // Make diagonal dominant
else
values[numIndices++] = value;
}
}
//cout << "Building row " << rowID << endl;
A->InsertGlobalValues(rowID, numIndices, values, indices);
}
delete [] indices;
delete [] values;
double insertTime = timer.ElapsedTime();
timer.ResetStartTime();
A->FillComplete(false);
double fillCompleteTime = timer.ElapsedTime();
if (verbose)
cout << "Time to insert matrix values = " << insertTime << endl
<< "Time to complete fill = " << fillCompleteTime << endl;
if (summary) {
if (comm.NumProc()==1) cout << "InsertTime" << '\t';
cout << insertTime << endl;
if (comm.NumProc()==1) cout << "FillCompleteTime" << '\t';
cout << fillCompleteTime << endl;
}
if (nrhs<=1) {
b = new Epetra_Vector(*map);
bt = new Epetra_Vector(*map);
xexact = new Epetra_Vector(*map);
}
else {
b = new Epetra_MultiVector(*map, nrhs);
bt = new Epetra_MultiVector(*map, nrhs);
xexact = new Epetra_MultiVector(*map, nrhs);
}
xexact->Random(); // Fill xexact with random values
A->Multiply(false, *xexact, *b);
A->Multiply(true, *xexact, *bt);
return;
}
int main(int argc, char *argv[])
{
bool boolret;
int MyPID;
#ifdef HAVE_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
MyPID = Comm.MyPID();
bool testFailed = false;
bool verbose = false;
bool debug = false;
string filename("mhd1280b.cua");
string which("LR");
bool skinny = true;
bool success = true;
try {
CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("skinny","hefty",&skinny,"Use a skinny (low-mem) or hefty (higher-mem) implementation of IRTR.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SR or LR).");
if (cmdp.parse(argc,argv) != CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
if (debug) verbose = true;
typedef double ScalarType;
typedef ScalarTraits<ScalarType> SCT;
typedef SCT::magnitudeType MagnitudeType;
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVT;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OPT;
if (verbose && MyPID == 0) {
cout << Anasazi::Anasazi_Version() << endl << endl;
}
// Problem information
int space_dim = 1;
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
std::vector<int> elements( space_dim );
elements[0] = 100;
// Create problem
RCP<ModalProblem> testCase = rcp( new ModeLaplace1DQ1(Comm, brick_dim[0], elements[0]) );
//
// Get the stiffness and mass matrices
RCP<const Epetra_CrsMatrix> K = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
RCP<const Epetra_CrsMatrix> M = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
const int FIRST_BS = 5;
const int SECOND_BS = 5;
const int THIRD_BS = 5;
const int NEV = FIRST_BS+SECOND_BS+THIRD_BS;
////////////////////////////////////////////////////////////////////////////////
//
// Shared parameters
RCP<Anasazi::BasicEigenproblem<ScalarType,MV,OP> > problem;
Anasazi::Eigensolution<ScalarType,MV> sol1, sol21, sol22, sol23;
RCP<const MV> cpoint;
RCP<MV> ev2 = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), FIRST_BS+SECOND_BS+THIRD_BS) );
Anasazi::ReturnType returnCode;
//
// Verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (debug) {
verbosity += Anasazi::Debug;
}
//
// Eigensolver parameters
int maxIters = 450;
MagnitudeType tol = 1.0e-8;
//
// Create parameter list to pass into the solver managers
ParameterList MyPL;
MyPL.set( "Skinny Solver", skinny);
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Maximum Iterations", maxIters );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Use Locking", true );
MyPL.set( "Locking Tolerance", tol/10 );
try {
////////////////////////////////////////////////////////////////////////////////
//
// Build the first eigenproblem
{
RCP<Epetra_MultiVector> ivec = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), FIRST_BS+SECOND_BS+THIRD_BS) );
ivec->Random();
problem = rcp( new Anasazi::BasicEigenproblem<ScalarType,MV,OP>(K,M,ivec) );
problem->setHermitian(true);
problem->setNEV( FIRST_BS+SECOND_BS+THIRD_BS );
boolret = problem->setProblem();
TEST_FOR_EXCEPTION(boolret != true,get_out,"Anasazi::BasicEigenproblem::SetProblem() returned with error.");
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the first solver manager
{
MyPL.set( "Block Size", FIRST_BS+SECOND_BS+THIRD_BS );
Anasazi::RTRSolMgr<ScalarType,MV,OP> solverman1(problem, MyPL);
returnCode = solverman1.solve();
TEST_FOR_EXCEPTION(returnCode != Anasazi::Converged, get_out, "First problem was not fully solved.");
sol1 = problem->getSolution();
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/1 eigenproblem
{
RCP<Epetra_MultiVector> ivec = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), FIRST_BS ) );
ivec->Random();
problem->setInitVec(ivec);
problem->setNEV( FIRST_BS );
boolret = problem->setProblem();
TEST_FOR_EXCEPTION(boolret != true, get_out, "Anasazi::BasicEigenproblem::SetProblem() returned with error.");
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/1 solver manager
{
MyPL.set( "Block Size", FIRST_BS );
Anasazi::RTRSolMgr<ScalarType,MV,OP> solverman21(problem, MyPL);
returnCode = solverman21.solve();
TEST_FOR_EXCEPTION(returnCode != Anasazi::Converged, get_out, "Second/1 problem was not fully solved.");
sol21 = problem->getSolution();
std::vector<int> bsind1(FIRST_BS);
for (int i=0; i<FIRST_BS; i++) bsind1[i] = i;
MVT::SetBlock(*sol21.Evecs,bsind1,*ev2);
cpoint = MVT::CloneView(*ev2,bsind1);
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/2 eigenproblem
{
RCP<Epetra_MultiVector> ivec = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), SECOND_BS ) );
ivec->Random();
problem->setAuxVecs(cpoint);
problem->setInitVec(ivec);
problem->setNEV( SECOND_BS );
boolret = problem->setProblem();
TEST_FOR_EXCEPTION(boolret != true, get_out, "Anasazi::BasicEigenproblem::SetProblem() returned with error.");
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/2 solver manager
{
MyPL.set( "Block Size", SECOND_BS );
Anasazi::RTRSolMgr<ScalarType,MV,OP> solverman22(problem, MyPL);
returnCode = solverman22.solve();
TEST_FOR_EXCEPTION(returnCode != Anasazi::Converged, get_out, "Second/2 problem was not fully solved." );
sol22 = problem->getSolution();
std::vector<int> bsind2(SECOND_BS);
for (int i=0; i<SECOND_BS; i++) bsind2[i] = FIRST_BS+i;
MVT::SetBlock(*sol22.Evecs,bsind2,*ev2);
std::vector<int> bsind12(FIRST_BS+SECOND_BS);
for (int i=0; i<FIRST_BS+SECOND_BS; i++) bsind12[i] = i;
cpoint = MVT::CloneView(*ev2,bsind12);
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/3 eigenproblem
{
RCP<Epetra_MultiVector> ivec = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), THIRD_BS ) );
ivec->Random();
problem->setAuxVecs(cpoint);
problem->setInitVec(ivec);
problem->setNEV( THIRD_BS );
boolret = problem->setProblem();
TEST_FOR_EXCEPTION(boolret != true, get_out, "Anasazi::BasicEigenproblem::SetProblem() returned with error.");
}
////////////////////////////////////////////////////////////////////////////////
//
// Build the second/3 solver manager
{
MyPL.set( "Block Size", THIRD_BS );
Anasazi::RTRSolMgr<ScalarType,MV,OP> solverman23(problem, MyPL);
returnCode = solverman23.solve();
TEST_FOR_EXCEPTION(returnCode != Anasazi::Converged, get_out, "Second/3 problem was not fully solved." );
sol23 = problem->getSolution();
std::vector<int> bsind3(THIRD_BS);
for (int i=0; i<THIRD_BS; i++) bsind3[i] = FIRST_BS+SECOND_BS+i;
MVT::SetBlock(*sol23.Evecs,bsind3,*ev2);
cpoint = Teuchos::null;
}
}
catch (const get_out &go) {
if (verbose && MyPID==0) {
cout << go.what() << endl;
}
testFailed = true;
}
if (testFailed == false) {
cout.setf(std::ios::scientific, std::ios::floatfield);
cout.precision(6);
//
// check the checkpointed solution against the non-checkpointed solution
//
// First, check the eigenvalues
//
// unless we altogether missed an eigenvalue, then
// {sol21.Evals sol22.Evals sol23.Evals} == {sol1.Evals}
std::vector<Anasazi::Value<double> > Evals1 = sol1.Evals;
std::vector<Anasazi::Value<double> > Evals2 = sol21.Evals;
Evals2.insert(Evals2.end(),sol22.Evals.begin(),sol22.Evals.end());
Evals2.insert(Evals2.end(),sol23.Evals.begin(),sol23.Evals.end());
// compare the differences
double maxd = 0;
if (verbose && MyPID==0) {
cout << std::setw(40) << "Computed Eigenvalues" << endl;
cout << std::setw(20) << "Without c/p" << std::setw(20) << "With c/p" << std::setw(20) << "Rel. error" << endl;
cout << "============================================================" << endl;
}
for (int i=0; i<NEV; i++) {
double tmpd = SCT::magnitude((Evals1[i].realpart - Evals2[i].realpart)/Evals1[i].realpart);
maxd = (tmpd > maxd ? tmpd : maxd);
if (verbose && MyPID==0) {
cout << std::setw(20) << Evals1[i].realpart << std::setw(20) << Evals2[i].realpart << std::setw(20) << tmpd << endl;
}
}
if (maxd > tol) {
testFailed = true;
}
if (verbose && MyPID==0) {
cout << endl;
}
//
// Second, check the eigenspaces
RCP<MV> Mev1;
if (M != null) {
Mev1 = MVT::Clone(*sol1.Evecs,NEV);
OPT::Apply(*M,*sol1.Evecs,*Mev1);
}
else {
Mev1 = sol1.Evecs;
}
SerialDenseMatrix<int,double> vtv(NEV,NEV);
MVT::MvTransMv(1.0,*Mev1,*ev2,vtv);
for (int i=0; i<NEV; i++) vtv(i,i) = SCT::magnitude(vtv(i,i)) - 1.0;
maxd = vtv.normFrobenius();
if (verbose && MyPID==0) {
cout << std::setw(20) << "|| EV1^H M EV2 - I ||_F" << endl;
cout << std::setw(20) << maxd << endl;
cout << endl;
}
if (maxd > SCT::squareroot(tol)*10) {
testFailed = true;
}
}
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(true,cout,success);
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
if (testFailed || success==false) {
if (verbose && MyPID==0) {
cout << "End Result: TEST FAILED" << endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
cout << "End Result: TEST PASSED" << endl;
}
return 0;
}
int main(int argc, char *argv[]) {
int ierr = 0;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// Comm.SetTracebackMode(0); // This should shut down any error tracing
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
#ifdef EPETRA_MPI
int localverbose = verbose ? 1 : 0;
int globalverbose=0;
MPI_Allreduce(&localverbose, &globalverbose, 1, MPI_INT, MPI_SUM,
MPI_COMM_WORLD);
verbose = (globalverbose>0);
#endif
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if (verbose && MyPID==0)
cout << Epetra_Version() << endl << endl;
if (verbose) cout << Comm <<endl;
int NumMyElements = 4;
long long NumGlobalElements = NumMyElements*NumProc;
int IndexBase = 0;
Epetra_Map Map(NumGlobalElements, NumMyElements, IndexBase, Comm);
EPETRA_TEST_ERR( Drumm1(Map, verbose),ierr);
EPETRA_TEST_ERR( Drumm2(Map, verbose),ierr);
bool preconstruct_graph = false;
EPETRA_TEST_ERR( four_quads(Comm, preconstruct_graph, verbose), ierr);
preconstruct_graph = true;
EPETRA_TEST_ERR( four_quads(Comm, preconstruct_graph, verbose), ierr);
EPETRA_TEST_ERR( rectangular(Comm, verbose), ierr);
EPETRA_TEST_ERR( Young1(Comm, verbose), ierr);
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return ierr;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, j, k;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
if(verbose && Comm.MyPID()==0)
std::cout << Epetra_Version() << std::endl << std::endl;
int rank = Comm.MyPID();
// char tmp;
// if (rank==0) std::cout << "Press any key to continue..."<< std::endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
if (verbose) std::cout << Comm << std::endl;
// bool verbose1 = verbose;
// Redefine verbose to only print on PE 0
if (verbose && rank!=0) verbose = false;
int N = 20;
int NRHS = 4;
double * A = new double[N*N];
double * A1 = new double[N*N];
double * X = new double[(N+1)*NRHS];
double * X1 = new double[(N+1)*NRHS];
int LDX = N+1;
int LDX1 = N+1;
double * B = new double[N*NRHS];
double * B1 = new double[N*NRHS];
int LDB = N;
int LDB1 = N;
int LDA = N;
int LDA1 = LDA;
double OneNorm1;
bool Upper = false;
Epetra_SerialSpdDenseSolver solver;
Epetra_SerialSymDenseMatrix * Matrix;
for (int kk=0; kk<2; kk++) {
for (i=1; i<=N; i++) {
GenerateHilbert(A, LDA, i);
OneNorm1 = 0.0;
for (j=1; j<=i; j++) OneNorm1 += 1.0/((double) j); // 1-Norm = 1 + 1/2 + ...+1/n
if (kk==0) {
Matrix = new Epetra_SerialSymDenseMatrix(View, A, LDA, i);
LDA1 = LDA;
}
else {
Matrix = new Epetra_SerialSymDenseMatrix(Copy, A, LDA, i);
LDA1 = i;
}
GenerateHilbert(A1, LDA1, i);
if (kk==1) {
solver.FactorWithEquilibration(true);
Matrix->SetUpper();
Upper = true;
solver.SolveToRefinedSolution(false);
}
for (k=0; k<NRHS; k++)
for (j=0; j<i; j++) {
B[j+k*LDB] = 1.0/((double) (k+3)*(j+3));
B1[j+k*LDB1] = B[j+k*LDB1];
}
Epetra_SerialDenseMatrix Epetra_B(View, B, LDB, i, NRHS);
Epetra_SerialDenseMatrix Epetra_X(View, X, LDX, i, NRHS);
solver.SetMatrix(*Matrix);
solver.SetVectors(Epetra_X, Epetra_B);
ierr = check(solver, A1, LDA1, i, NRHS, OneNorm1, B1, LDB1, X1, LDX1, Upper, verbose);
assert (ierr>-1);
delete Matrix;
if (ierr!=0) {
if (verbose) std::cout << "Factorization failed due to bad conditioning. This is normal if SCOND is small."
<< std::endl;
break;
}
}
}
delete [] A;
delete [] A1;
delete [] X;
delete [] X1;
delete [] B;
delete [] B1;
/////////////////////////////////////////////////////////////////////
// Now test norms and scaling functions
/////////////////////////////////////////////////////////////////////
Epetra_SerialSymDenseMatrix D;
double ScalarA = 2.0;
int DM = 10;
int DN = 10;
D.Shape(DM);
for (j=0; j<DN; j++)
for (i=0; i<DM; i++) D[j][i] = (double) (1+i+j*DM) ;
//std::cout << D << std::endl;
double NormInfD_ref = (double)(DM*(DN*(DN+1))/2);
double NormOneD_ref = NormInfD_ref;
double NormInfD = D.NormInf();
double NormOneD = D.NormOne();
if (verbose) {
std::cout << " *** Before scaling *** " << std::endl
<< " Computed one-norm of test matrix = " << NormOneD << std::endl
<< " Expected one-norm = " << NormOneD_ref << std::endl
<< " Computed inf-norm of test matrix = " << NormInfD << std::endl
<< " Expected inf-norm = " << NormInfD_ref << std::endl;
}
D.Scale(ScalarA); // Scale entire D matrix by this value
//std::cout << D << std::endl;
NormInfD = D.NormInf();
NormOneD = D.NormOne();
if (verbose) {
std::cout << " *** After scaling *** " << std::endl
<< " Computed one-norm of test matrix = " << NormOneD << std::endl
<< " Expected one-norm = " << NormOneD_ref*ScalarA << std::endl
<< " Computed inf-norm of test matrix = " << NormInfD << std::endl
<< " Expected inf-norm = " << NormInfD_ref*ScalarA << std::endl;
}
/////////////////////////////////////////////////////////////////////
// Now test for larger system, both correctness and performance.
/////////////////////////////////////////////////////////////////////
N = 2000;
NRHS = 5;
LDA = N;
LDB = N;
LDX = N;
if (verbose) std::cout << "\n\nComputing factor of an " << N << " x " << N << " SPD matrix...Please wait.\n\n" << std::endl;
// Define A and X
A = new double[LDA*N];
X = new double[LDB*NRHS];
for (j=0; j<N; j++) {
for (k=0; k<NRHS; k++) X[j+k*LDX] = 1.0/((double) (j+5+k));
for (i=0; i<N; i++) {
if (i==j) A[i+j*LDA] = 100.0 + i;
else A[i+j*LDA] = -1.0/((double) (i+10)*(j+10));
}
}
// Define Epetra_SerialDenseMatrix object
Epetra_SerialSymDenseMatrix BigMatrix(Copy, A, LDA, N);
Epetra_SerialSymDenseMatrix OrigBigMatrix(View, A, LDA, N);
Epetra_SerialSpdDenseSolver BigSolver;
BigSolver.FactorWithEquilibration(true);
BigSolver.SetMatrix(BigMatrix);
// Time factorization
Epetra_Flops counter;
BigSolver.SetFlopCounter(counter);
Epetra_Time Timer(Comm);
double tstart = Timer.ElapsedTime();
ierr = BigSolver.Factor();
if (ierr!=0 && verbose) std::cout << "Error in factorization = "<<ierr<< std::endl;
assert(ierr==0);
double time = Timer.ElapsedTime() - tstart;
double FLOPS = counter.Flops();
double MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS for Factorization = " << MFLOPS << std::endl;
// Define Left hand side and right hand side
Epetra_SerialDenseMatrix LHS(View, X, LDX, N, NRHS);
Epetra_SerialDenseMatrix RHS;
RHS.Shape(N,NRHS); // Allocate RHS
// Compute RHS from A and X
Epetra_Flops RHS_counter;
RHS.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
RHS.Multiply('L', 1.0, OrigBigMatrix, LHS, 0.0); // Symmetric Matrix-multiply
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseMatrix OrigRHS = RHS;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to build RHS (NRHS = " << NRHS <<") = " << MFLOPS << std::endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(LHS, RHS);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
if (ierr==1 && verbose) std::cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << std::endl;
else if (ierr!=0 && verbose) std::cout << "Error in solve = "<<ierr<< std::endl;
assert(ierr>=0);
time = Timer.ElapsedTime() - tstart;
FLOPS = BigSolver.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS for Solve (NRHS = " << NRHS <<") = " << MFLOPS << std::endl;
double * resid = new double[NRHS];
bool OK = Residual(N, NRHS, A, LDA, BigSolver.X(), BigSolver.LDX(),
OrigRHS.A(), OrigRHS.LDA(), resid);
if (verbose) {
if (!OK) std::cout << "************* Residual do not meet tolerance *************" << std::endl;
for (i=0; i<NRHS; i++)
std::cout << "Residual[" << i <<"] = "<< resid[i] << std::endl;
std::cout << std::endl;
}
// Solve again using the Epetra_SerialDenseVector class for LHS and RHS
Epetra_SerialDenseVector X2;
Epetra_SerialDenseVector B2;
X2.Size(BigMatrix.N());
B2.Size(BigMatrix.M());
int length = BigMatrix.N();
{for (int kk=0; kk<length; kk++) X2[kk] = ((double ) kk)/ ((double) length);} // Define entries of X2
RHS_counter.ResetFlops();
B2.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
B2.Multiply('N', 'N', 1.0, OrigBigMatrix, X2, 0.0); // Define B2 = A*X2
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseVector OrigB2 = B2;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to build single RHS = " << MFLOPS << std::endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(X2, B2);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
time = Timer.ElapsedTime() - tstart;
if (ierr==1 && verbose) std::cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << std::endl;
else if (ierr!=0 && verbose) std::cout << "Error in solve = "<<ierr<< std::endl;
assert(ierr>=0);
FLOPS = counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to solve single RHS = " << MFLOPS << std::endl;
OK = Residual(N, 1, A, LDA, BigSolver.X(), BigSolver.LDX(), OrigB2.A(),
OrigB2.LDA(), resid);
if (verbose) {
if (!OK) std::cout << "************* Residual do not meet tolerance *************" << std::endl;
std::cout << "Residual = "<< resid[0] << std::endl;
}
delete [] resid;
delete [] A;
delete [] X;
///////////////////////////////////////////////////
// Now test default constructor and index operators
///////////////////////////////////////////////////
N = 5;
Epetra_SerialSymDenseMatrix C; // Implicit call to default constructor, should not need to call destructor
C.Shape(5); // Make it 5 by 5
double * C1 = new double[N*N];
GenerateHilbert(C1, N, N); // Generate Hilber matrix
C1[1+2*N] = 1000.0; // Make matrix nonsymmetric
// Fill values of C with Hilbert values
for (i=0; i<N; i++)
for (j=0; j<N; j++)
C(i,j) = C1[i+j*N];
// Test if values are correctly written and read
for (i=0; i<N; i++)
for (j=0; j<N; j++) {
assert(C(i,j) == C1[i+j*N]);
assert(C(i,j) == C[j][i]);
}
if (verbose)
std::cout << "Default constructor and index operator check OK. Values of Hilbert matrix = "
<< std::endl << C << std::endl
<< "Values should be 1/(i+j+1), except value (1,2) should be 1000" << std::endl;
delete [] C1;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
void solvetriadmatrixwithtrilinos(int& nnz, int& order, int* row,
int* col, double* val, double* rhs, double* solution) {
#else
void solvetriadmatrixwithtrilinos_(int& nnz, int& order, int* row,
int* col, double* val, double* rhs, double* solution) {
#endif
try{
#ifdef _MPI
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
int i, j, ierr;
int MyPID = Comm.MyPID();
bool verbose = (MyPID == 0);
Epetra_Map RowMap(order, 0, Comm);
int NumMyElements = RowMap.NumMyElements();
int *MyGlobalElements = new int[NumMyElements];
RowMap.MyGlobalElements(&MyGlobalElements[0]);
#ifdef _MPI
int nPEs;
MPI_Comm_size(MPI_COMM_WORLD, &nPEs);
#endif
int anEst = nnz / order + 1;
Epetra_CrsMatrix A(Copy, RowMap, anEst);
for (j=0; j<nnz; ++j) {
if (RowMap.MyGID(row[j]) ) {
ierr = A.InsertGlobalValues(row[j], 1, &(val[j]), &(col[j]) );
assert(ierr >= 0);
}
}
ierr = A.FillComplete();
assert(ierr == 0);
//-------------------------------------------------------------------------
// RN_20091221: Taking care of the rhs
//-------------------------------------------------------------------------
Epetra_Vector b(RowMap);
// Inserting values into the rhs
double *MyGlobalValues = new double[NumMyElements];
for (j=0; j<NumMyElements; ++j) {
MyGlobalValues[j] = rhs[MyGlobalElements[j] ];
}
ierr = b.ReplaceGlobalValues(NumMyElements, &MyGlobalValues[0],
&MyGlobalElements[0]);
//-------------------------------------------------------------------------
// RN_20091221: Taking care of the solution
//-------------------------------------------------------------------------
Epetra_Vector x(RowMap);
Teuchos::ParameterList paramList;
Teuchos::RCP<Teuchos::ParameterList>
paramList1 = Teuchos::rcp(¶mList, false);
Teuchos::updateParametersFromXmlFile("./strat1.xml", paramList1.get() );
Teuchos::RCP<Teuchos::FancyOStream>
out = Teuchos::VerboseObjectBase::getDefaultOStream();
Stratimikos::DefaultLinearSolverBuilder linearSolverBuilder;
Teuchos::RCP<Epetra_CrsMatrix> epetraOper = Teuchos::rcp(&A, false);
Teuchos::RCP<Epetra_Vector> epetraRhs = Teuchos::rcp(&b, false);
Teuchos::RCP<Epetra_Vector> epetraSol = Teuchos::rcp(&x, false);
Teuchos::RCP<const Thyra::LinearOpBase<double> >
thyraOper = Thyra::epetraLinearOp(epetraOper);
Teuchos::RCP<Thyra::VectorBase<double> >
thyraRhs = Thyra::create_Vector(epetraRhs, thyraOper->range() );
Teuchos::RCP<Thyra::VectorBase<double> >
thyraSol = Thyra::create_Vector(epetraSol, thyraOper->domain() );
linearSolverBuilder.setParameterList(Teuchos::rcp(¶mList, false) );
Teuchos::RCP<Thyra::LinearOpWithSolveFactoryBase<double> >
lowsFactory = linearSolverBuilder.createLinearSolveStrategy("");
lowsFactory->setOStream(out);
lowsFactory->setVerbLevel(Teuchos::VERB_LOW);
Teuchos::RCP<Thyra::LinearOpWithSolveBase<double> >
lows = Thyra::linearOpWithSolve(*lowsFactory, thyraOper);
Thyra::SolveStatus<double>
status = Thyra::solve(*lows, Thyra::NOTRANS, *thyraRhs, &*thyraSol);
thyraSol = Teuchos::null;
// For debugging =)
// cout << "A: " << A << endl;
// cout << "b: " << b << endl;
// cout << "x: " << x << endl;
// Epetra_Vector temp(RowMap);
// double residualNorm;
// A.Multiply(false, x, temp);
// temp.Update(-1, b, 1);
// temp.Norm2(&residualNorm);
Epetra_LocalMap localMap(order, 0, Comm);
Epetra_Vector xExtra(localMap); // local vector in each processor
// RN_20091218: Create an import map and then import the data to the
// local vector.
Epetra_Import import(localMap, RowMap);
xExtra.Import(x, import, Add);
xExtra.ExtractCopy(solution);
delete[] MyGlobalElements;
delete[] MyGlobalValues;
}
catch (std::exception& e) {
cout << e.what() << endl;
}
catch (string& s) {
cout << s << endl;
}
catch (char *s) {
cout << s << endl;
}
catch (...) {
cout << "Caught unknown exception!" << endl;
}
}
} // extern "C"
int main(int argc, char *argv[])
{
// Initialize MPI
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
#endif
// Create a communicator for Epetra objects
#ifdef HAVE_MPI
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
// Get the process ID and the total number of processors
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
// Check verbosity level
bool verbose = false;
if (argc > 1)
if (argv[1][0]=='-' && argv[1][1]=='v')
verbose = true;
// Get the number of elements from the command line
int NumGlobalElements = 0;
if ((argc > 2) && (verbose))
NumGlobalElements = atoi(argv[2]) + 1;
else if ((argc > 1) && (!verbose))
NumGlobalElements = atoi(argv[1]) + 1;
else
NumGlobalElements = 101;
// The number of unknowns must be at least equal to the
// number of processors.
if (NumGlobalElements < NumProc) {
cout << "numGlobalBlocks = " << NumGlobalElements
<< " cannot be < number of processors = " << NumProc << endl;
cout << "Test failed!" << endl;
throw "NOX Error";
}
// Create the interface between NOX and the application
// This object is derived from NOX::Epetra::Interface
Teuchos::RCP<Interface> interface =
Teuchos::rcp(new Interface(NumGlobalElements, Comm));
// Set the PDE factor (for nonlinear forcing term). This could be specified
// via user input.
interface->setPDEfactor(1000.0);
// Begin Nonlinear Solver ************************************
// Create the top level parameter list
Teuchos::RCP<Teuchos::ParameterList> IfpackParamsPtr =
Teuchos::rcp(new Teuchos::ParameterList);
// Set the printing parameters in the "Printing" sublist
Teuchos::ParameterList printParams;
printParams.set("MyPID", MyPID);
printParams.set("Output Precision", 3);
printParams.set("Output Processor", 0);
if (verbose)
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Parameters +
NOX::Utils::Details +
NOX::Utils::Warning +
NOX::Utils::Debug +
NOX::Utils::TestDetails +
NOX::Utils::Error);
else
printParams.set("Output Information", NOX::Utils::Error +
NOX::Utils::TestDetails);
// Create a print class for controlling output below
NOX::Utils p(printParams);
// *******************************
// Setup Test Objects
// *******************************
// Create Linear Objects
// Get the vector from the Problem
if (verbose)
p.out() << "Creating Vectors and Matrices" << endl;
Teuchos::RCP<Epetra_Vector> solution_vec =
interface->getSolution();
Teuchos::RCP<Epetra_Vector> rhs_vec =
Teuchos::rcp(new Epetra_Vector(*solution_vec));
Teuchos::RCP<Epetra_Vector> lhs_vec =
Teuchos::rcp(new Epetra_Vector(*solution_vec));
Teuchos::RCP<Epetra_CrsMatrix> jacobian_matrix =
interface->getJacobian();
if (verbose)
p.out() << "Evaluating F and J" << endl;
solution_vec->PutScalar(1.0);
interface->computeF(*solution_vec, *rhs_vec);
rhs_vec->Scale(-1.0);
interface->computeJacobian(*solution_vec, *jacobian_matrix);
double norm =0.0;
rhs_vec->Norm2(&norm);
if (verbose)
p.out() << "Step 0, ||F|| = " << norm << endl;
if (verbose)
p.out() << "Creating Ifpack preconditioner" << endl;
Ifpack Factory;
Teuchos::RCP<Ifpack_Preconditioner> PreconditionerPtr;
PreconditionerPtr = Teuchos::rcp(Factory.Create("ILU",
jacobian_matrix.get(),0));
Teuchos::ParameterList teuchosParams;
PreconditionerPtr->SetParameters(teuchosParams);
PreconditionerPtr->Initialize();
PreconditionerPtr->Compute();
if (verbose)
p.out() << "Creating Aztec Solver" << endl;
Teuchos::RCP<AztecOO> aztecSolverPtr = Teuchos::rcp(new AztecOO());
if (verbose)
aztecSolverPtr->SetAztecOption(AZ_output, AZ_last);
else
aztecSolverPtr->SetAztecOption(AZ_output, AZ_none);
// *******************************
// Recompute Test
// *******************************
if (verbose) {
p.out() << "**********************************************" << endl;
p.out() << "Testing Newton solve with prec reuse" << endl;
p.out() << "**********************************************" << endl;
}
int step_number = 0;
int max_steps = 20;
bool converged = false;
int total_linear_iterations = 0;
while (norm > 1.0e-8 && step_number < max_steps) {
step_number++;
if (verbose)
p.out() << "Step " << step_number << ", ||F|| = " << norm << endl;
if (step_number != 1) {
if (verbose)
p.out() << "Recomputing Preconditioner (reusing graph)" << endl;
PreconditionerPtr->Compute();
}
aztecSolverPtr->SetUserMatrix(jacobian_matrix.get(), false);
aztecSolverPtr->SetPrecOperator(PreconditionerPtr.get());
aztecSolverPtr->SetRHS(rhs_vec.get());
aztecSolverPtr->SetLHS(lhs_vec.get());
aztecSolverPtr->Iterate(400, 1.0e-4);
solution_vec->Update(1.0, *lhs_vec, 1.0);
interface->computeF(*solution_vec, *rhs_vec);
rhs_vec->Scale(-1.0);
interface->computeJacobian(*solution_vec, *jacobian_matrix);
rhs_vec->Norm2(&norm);
total_linear_iterations += aztecSolverPtr->NumIters();
if (norm < 1.0e-8)
converged = true;
}
if (verbose) {
p.out() << "Final Step " << step_number << ", ||F|| = " << norm << endl;
if (converged)
p.out() << "Converged!!" << endl;
else
p.out() << "Failed!!" << endl;
}
// Tests
int status = 0; // Converged
if (verbose)
p.out() << "Total Number of Linear Iterations = "
<< total_linear_iterations << endl;
if (Comm.NumProc() == 1 && total_linear_iterations != 10)
status = 1;
if (!converged)
status = 2;
// Summarize test results
if (converged && status == 0)
p.out() << "Test passed!" << endl;
else
p.out() << "Test failed!" << endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
if (verbose)
p.out() << "Status = " << status << endl;
// Final return value (0 = successfull, non-zero = failure)
return status;
}
//==============================================================================
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// only one process
if (Comm.NumProc() != 1) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
if (Comm.MyPID() == 0)
cout << "Please run this test with one process only" << endl;
// return success not to break the tests
exit(EXIT_SUCCESS);
}
// ======================================================== //
// now create the famous "upper arrow" matrix, which //
// should be reordered as a "lower arrow". Sparsity pattern //
// will be printed on screen. //
// ======================================================== //
int NumPoints = 16;
#if !defined(EPETRA_NO_32BIT_GLOBAL_INDICES) || !defined(EPETRA_NO_64BIT_GLOBAL_INDICES)
Epetra_Map Map(-1,NumPoints,0,Comm);
#else
Epetra_Map Map;
#endif
std::vector<int> Indices(NumPoints);
std::vector<double> Values(NumPoints);
Teuchos::RefCountPtr<Epetra_CrsMatrix> A = Teuchos::rcp( new Epetra_CrsMatrix(Copy,Map,0) );
for (int i = 0 ; i < NumPoints ; ++i) {
int NumEntries;
if (i == 0) {
NumEntries = NumPoints;
for (int j = 0 ; j < NumPoints ; ++j) {
Indices[j] = j;
Values[j] = 1.0;
}
}
else {
NumEntries = 2;
Indices[0] = 0;
Indices[1] = i;
Values[0] = 1.0;
Values[1] = 1.0;
}
#if !defined(EPETRA_NO_32BIT_GLOBAL_INDICES) || !defined(EPETRA_NO_64BIT_GLOBAL_INDICES)
A->InsertGlobalValues(i, NumEntries, &Values[0], &Indices[0]);
#endif
}
A->FillComplete();
// print the sparsity to file, postscript format
////Ifpack_PrintSparsity(A,"OrigA.ps");
// create the reordering...
Teuchos::RefCountPtr<Ifpack_RCMReordering> Reorder = Teuchos::rcp( new Ifpack_RCMReordering() );
// and compute is on A
IFPACK_CHK_ERR(Reorder->Compute(*A));
// cout information
cout << *Reorder;
// create a reordered matrix
Ifpack_ReorderFilter ReordA(A, Reorder);
// print the sparsity to file, postscript format
////Ifpack_PrintSparsity(ReordA,"ReordA.ps");
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
}
int main( int argc, char **argv )
{
// check for parallel computation
#ifdef HAVE_MPI
MPI_Init(&argc, &argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// define main parameters
double c = 0.9999; // continuation parameter
int N = 50; // number of grid points
int maxNewtonIters = 20; // max number of Newton iterations
int maxSteps = 50; // max number of continuation steps taken
int ilocal, iglobal; // counter variables used for loops:
// ilocal = counter for local elements on this processor;
// iglobal = counter to signify global position across all procs
int Myele; // holds the number of elements on the processor
// Set flag for whether the computations will be Matrix-free (true) or will use a computed
// Jacobian (false)
bool doMatFree = false;
// Create output file to save solutions
ofstream outFile("Heq5.dat");
outFile.setf(ios::scientific, ios::floatfield);
outFile.precision(10);
// Define the problem class
HeqProblem Problem(N,&Comm,outFile);
// Build initial guess. The initial guess should be a solution vector x close to the
// bifurcation point.
// Create the initial guess vector
Epetra_Vector InitialGuess(Problem.GetMap());
// Get the number of elements on this processor
Myele = Problem.GetMap().NumMyElements();
// Compute the initial guess. For this example, it is a line from (0,1) to (1,8/3)
for (ilocal=0; ilocal<Myele; ilocal++) {
iglobal=Problem.GetMap().GID(ilocal);
InitialGuess[ilocal]= 1.0 + (5.0*iglobal)/(3.0*(N-1));
}
// Create the null vector for the Jacobian (ie, J*v=0, used to solve the system of equations
// f(x,p)=0; J*v=0; v0*v=1. The solution of the system is [x*,v*,p*]. )
Teuchos::RCP<NOX::Abstract::Vector> nullVec =
Teuchos::rcp(new NOX::Epetra::Vector(InitialGuess));
// Initialize to all ones
nullVec->init(1.0);
// NOTE: init is a function within the NOX::Abstract:Vector class which initializes every
// value of the vector to the value within the parentheses (must be in 'double' format)
// Create the top level parameter list
Teuchos::RCP<Teuchos::ParameterList> ParamList = Teuchos::rcp(new Teuchos::ParameterList);
// Create LOCA sublist
Teuchos::ParameterList& locaParamsList = ParamList->sublist("LOCA");
// Create the sublist for continuation and set the stepper parameters
Teuchos::ParameterList& stepperList = locaParamsList.sublist("Stepper");
//stepperList.set("Continuation Method", "Arc Length");// Default
stepperList.set("Continuation Method", "Natural");
stepperList.set("Continuation Parameter", "dummy"); // Must set
stepperList.set("Initial Value", 999.0); // Must set
stepperList.set("Max Value", 50.0e4); // Must set
stepperList.set("Min Value", 0.0); // Must set
stepperList.set("Max Steps", maxSteps); // Should set
stepperList.set("Max Nonlinear Iterations", maxNewtonIters); // Should set
stepperList.set("Bordered Solver Method", "Bordering");
// Teuchos::ParameterList& nestedList =
// stepperList.sublist("Nested Bordered Solver");
// nestedList.set("Bordered Solver Method", "Householder");
// nestedList.set("Include UV In Preconditioner", true);
// //nestedList.set("Use P For Preconditioner", true);
// nestedList.set("Preconditioner Method", "SMW");
// Set up parameters to compute Eigenvalues
#ifdef HAVE_LOCA_ANASAZI
// Create Anasazi Eigensolver sublist (needs --with-loca-anasazi)
stepperList.set("Compute Eigenvalues",true);
Teuchos::ParameterList& aList = stepperList.sublist("Eigensolver");
aList.set("Method", "Anasazi");
aList.set("Block Size", 1); // Size of blocks
aList.set("Num Blocks", 20); // Size of Arnoldi factorization
aList.set("Num Eigenvalues", 5); // Number of eigenvalues
// aList.set("Sorting Order", "SR");
aList.set("Convergence Tolerance", 2.0e-7); // Tolerance
aList.set("Step Size", 1); // How often to check convergence
aList.set("Maximum Restarts",2); // Maximum number of restarts
aList.set("Verbosity",
Anasazi::Errors +
Anasazi::Warnings +
Anasazi::FinalSummary); // Verbosity
#else
stepperList.set("Compute Eigenvalues",false);
#endif
// Create bifurcation sublist. Note that for turning point continuation, the "type"
// is set to "Turning Point". If not doing TP, type should be "None".
Teuchos::ParameterList& bifurcationList = locaParamsList.sublist("Bifurcation");
bifurcationList.set("Type", "Turning Point");
bifurcationList.set("Bifurcation Parameter", "c");
// bifurcationList.set("Formulation", "Minimally Augmented");
bifurcationList.set("Symmetric Jacobian", false);
bifurcationList.set("Update Null Vectors Every Continuation Step", true);
bifurcationList.set("Update Null Vectors Every Nonlinear Iteration", false);
bifurcationList.set("Transpose Solver Method","Explicit Transpose");
// bifurcationList.set("Transpose Solver Method","Transpose Preconditioner");
// bifurcationList.set("Transpose Solver Method","Left Preconditioning");
bifurcationList.set("Initial Null Vector Computation", "Solve df/dp");
// bifurcationList.set("Initial A Vector", nullVec); // minimally augmented
// bifurcationList.set("Initial B Vector", nullVec); //minimally augmented
// bifurcationList.set("Bordered Solver Method", "Householder");
// bifurcationList.set("Include UV In Preconditioner", true);
// //bifurcationList.set("Use P For Preconditioner", true);
// bifurcationList.set("Preconditioner Method", "SMW");
bifurcationList.set("Formulation", "Moore-Spence");
bifurcationList.set("Solver Method", "Phipps Bordering"); // better for nearly singular matrices
// bifurcationList.set("Solver Method", "Salinger Bordering");
bifurcationList.set("Initial Null Vector", nullVec);
bifurcationList.set("Length Normalization Vector", nullVec);
// Create the sublist for the predictor
Teuchos::ParameterList& predictorList = locaParamsList.sublist("Predictor");
predictorList.set("Method", "Secant"); // Default
// predictorList.set("Method", "Constant"); // Other options
// predictorList.set("Method", "Tangent"); // Other options
// Create step size sublist
Teuchos::ParameterList& stepSizeList = locaParamsList.sublist("Step Size");
stepSizeList.set("Method", "Adaptive"); // Default
stepSizeList.set("Initial Step Size", 0.1); // Should set
stepSizeList.set("Min Step Size", 1.0e-6); // Should set
stepSizeList.set("Max Step Size", 1.0); // Should set
stepSizeList.set("Aggressiveness", 0.1);
// Set up NOX info
Teuchos::ParameterList& nlParams = ParamList->sublist("NOX");
// Set the nonlinear solver method
nlParams.set("Nonlinear Solver", "Line Search Based");
// Set the printing parameters in the "Printing" sublist. This list determines how much
// of the NOX information is output
Teuchos::ParameterList& printParams = nlParams.sublist("Printing");
printParams.set("MyPID", Comm.MyPID());
printParams.set("Output Precision", 5);
printParams.set("Output Processor", 0);
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Parameters +
NOX::Utils::Details +
NOX::Utils::Warning +
NOX::Utils::StepperIteration +
NOX::Utils::StepperDetails +
NOX::Utils::StepperParameters);
// NOX parameters - Sublist for line search
Teuchos::ParameterList& searchParams = nlParams.sublist("Line Search");
searchParams.set("Method", "Backtrack");
// searchParams.set("Method", "Full Step");
// Sublist for direction
Teuchos::ParameterList& dirParams = nlParams.sublist("Direction");
dirParams.set("Method", "Newton");
Teuchos::ParameterList& newtonParams = dirParams.sublist("Newton");
newtonParams.set("Forcing Term Method", "Constant");
// Sublist for linear solver for the Newton method
Teuchos::ParameterList& lsParams = newtonParams.sublist("Linear Solver");
lsParams.set("Aztec Solver", "GMRES");
lsParams.set("Max Iterations", 800);
lsParams.set("Tolerance", 1e-8);
lsParams.set("Output Frequency", 1);
lsParams.set("Preconditioner", "None");
// lsParams.set("Preconditioner", "AztecOO");
// lsParams.set("Aztec Preconditioner", "ilu");
// lsParams.set("Scaling", "None");
// lsParams.set("Scaling", "Row Sum");
lsParams.set("Compute Scaling Manually", false);
// lsParams.set("Preconditioner", "Ifpack");
// lsParams.set("Ifpack Preconditioner", "ILU");
// lsParams.set("Preconditioner", "New Ifpack");
// Teuchos::ParameterList& ifpackParams = lsParams.sublist("Ifpack");
// ifpackParams.set("fact: level-of-fill", 1);
// Set up the continuation parameter vector
LOCA::ParameterVector p;
p.addParameter("c",c);
p.addParameter("dummy",999.0);
// Create the problem interface
Teuchos::RCP<SimpleProblemInterface> interface =
Teuchos::rcp(new SimpleProblemInterface(&Problem,c) );
Teuchos::RCP<LOCA::Epetra::Interface::Required> iReq = interface;
// Create the operator to hold either the Jacobian matrix or the Matrix-free operator
Teuchos::RCP<Epetra_Operator> A;
// Teuchos::RCP<Epetra_RowMatrix> A;
Teuchos::RCP<NOX::Epetra::Interface::Jacobian> iJac;
// Need a NOX::Epetra::Vector for constructor
// This becomes the initial guess vector that is used for the nonlinear solves
NOX::Epetra::Vector noxInitGuess(InitialGuess, NOX::DeepCopy);
if (doMatFree) {
// Matrix Free application (Epetra Operator):
Teuchos::RCP<NOX::Epetra::MatrixFree> MF =
Teuchos::rcp(new NOX::Epetra::MatrixFree(printParams, interface, noxInitGuess));
A = MF;
iJac = MF;
}
else { // Computed Jacobian application
A = Teuchos::rcp( Problem.GetMatrix(), false );
iJac = interface;
}
// Create scaling object
Teuchos::RCP<NOX::Epetra::Scaling> scaling = Teuchos::null;
// scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> scalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// //scaling->addRowSumScaling(NOX::Epetra::Scaling::Left, scalingVector);
// scaling->addColSumScaling(NOX::Epetra::Scaling::Right, scalingVector);
// Create transpose scaling object
Teuchos::RCP<NOX::Epetra::Scaling> trans_scaling = Teuchos::null;
// trans_scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> transScalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// trans_scaling->addRowSumScaling(NOX::Epetra::Scaling::Right,
// transScalingVector);
// trans_scaling->addColSumScaling(NOX::Epetra::Scaling::Left,
// transScalingVector);
//bifurcationList.set("Transpose Scaling", trans_scaling);
// Build the linear system solver
Teuchos::RCP<NOX::Epetra::LinearSystemAztecOO> linSys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(printParams, lsParams,
iReq,
iJac, A,
noxInitGuess, scaling)); // use if scaling
// noxInitGuess)); // use if no scaling
// Create the Loca (continuation) vector
NOX::Epetra::Vector locaSoln(noxInitGuess);
// Create Epetra Factory
Teuchos::RCP<LOCA::Abstract::Factory> epetraFactory = Teuchos::rcp(new LOCA::Epetra::Factory);
// Create global data object
Teuchos::RCP<LOCA::GlobalData> globalData = LOCA::createGlobalData(ParamList, epetraFactory);
// Create the Group - must be LOCA group
Teuchos::RCP<LOCA::Epetra::Group> grpPtr =
Teuchos::rcp(new LOCA::Epetra::Group(globalData, printParams,
iReq, locaSoln,
linSys, p));
// Calculate the first F(x0) as a starting point. This is only needed for
// certain status tests, to ensure that an initial residual (|r0|) is calculated
grpPtr->computeF();
// Set up the status tests to check for convergence
// Determines the error tolerance for the Newton solves
Teuchos::RCP<NOX::StatusTest::NormF> testNormF =
Teuchos::rcp(new NOX::StatusTest::NormF(1.0e-4));
// Sets the max number of nonlinear (Newton) iterations that will be taken. If this is not
// already set, it will default to the '20' given
Teuchos::RCP<NOX::StatusTest::MaxIters> testMaxIters =
Teuchos::rcp(new NOX::StatusTest::MaxIters(stepperList.get("Max Nonlinear Iterations", 20)));
// This combination of tests will be used by NOX to determine whether the step converged
Teuchos::RCP<NOX::StatusTest::Combo> combo =
Teuchos::rcp(new NOX::StatusTest::Combo(NOX::StatusTest::Combo::OR,
testNormF, testMaxIters));
// This is sample code to write and read parameters to/from a file. Currently not activated!
// To use, change the 'XXXHAVE_TEUCHOS_EXTENDED' TO 'HAVE_TEUCHOS_EXTENDED'
#ifdef XXXHAVE_TEUCHOS_EXTENDED
// Write the parameter list to a file
cout << "Writing parameter list to \"input.xml\"" << endl;
Teuchos::writeParameterListToXmlFile(*ParamList, "input.xml");
// Read in the parameter list from a file
cout << "Reading parameter list from \"input.xml\"" << endl;
Teuchos::RCP<Teuchos::ParameterList> paramList2 =
Teuchos::rcp(new Teuchos::ParameterList);
Teuchos::updateParametersFromXmlFile("input.xml", paramList2.get());
ParamList = paramList2;
#endif
// Create the stepper
LOCA::Stepper stepper(globalData, grpPtr, combo, ParamList);
LOCA::Abstract::Iterator::IteratorStatus status = stepper.run();
// Check if the stepper completed
if (status == LOCA::Abstract::Iterator::Finished)
globalData->locaUtils->out() << "\nAll tests passed!" << endl;
else
if (globalData->locaUtils->isPrintType(NOX::Utils::Error))
globalData->locaUtils->out() << "\nStepper failed to converge!" << endl;
// Output the stepper parameter list info
if (globalData->locaUtils->isPrintType(NOX::Utils::StepperParameters)) {
globalData->locaUtils->out() << endl << "Final Parameters" << endl
<< "*******************" << endl;
stepper.getList()->print(globalData->locaUtils->out());
globalData->locaUtils->out() << endl;
}
// Make sure all processors are done and close the output file
Comm.Barrier();
outFile.close();
// Deallocate memory
LOCA::destroyGlobalData(globalData);
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
} // DONE!!
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
CommandLineProcessor CLP;
int n = 10;
int m = (int)pow((double)Comm.NumProc(), 0.3334);
double DampingFactor = 1.333;
std::string AggregationScheme = "Uncoupled";
int NPA = 16;
int MaxLevels = 5;
CLP.setOption("l", &MaxLevels, "number of levels");
CLP.setOption("n", &n, "number of nodes along each axis");
CLP.setOption("damp", &DampingFactor, "prolongator damping factor");
CLP.setOption("aggr", &AggregationScheme, "aggregation scheme");
CLP.setOption("npa", &NPA, "nodes per aggregate (if supported by the aggregation scheme)");
CLP.throwExceptions(false);
CLP.parse(argc,argv);
if (m * m * m != Comm.NumProc()) {
if (Comm.MyPID() == 0)
{
std::cout << "Number of processes must be a perfect cube." << std::endl;
std::cout << "Please re-run with --help option for details." << std::endl;
}
#ifdef HAVE_MPI
MPI_Finalize();
#endif
exit(EXIT_SUCCESS);
}
n *= m;
Laplace3D A(Comm, n, n, n, m, m, m, true);
Epetra_Vector LHS(A.OperatorDomainMap());
Epetra_Vector RHS(A.OperatorDomainMap());
LHS.Random();
RHS.PutScalar(0.0);
Epetra_LinearProblem Problem(&A, &LHS, &RHS);
// Construct a solver object for this problem
AztecOO solver(Problem);
// create a parameter list for ML options
ParameterList MLList;
// set defaults for classic smoothed aggregation
ML_Epetra::SetDefaults("SA",MLList);
// overwrite some parameters. Please refer to the user's guide
// for more information
MLList.set("max levels",MaxLevels);
MLList.set("increasing or decreasing","increasing");
MLList.set("aggregation: type", AggregationScheme);
MLList.set("aggregation: damping factor", DampingFactor);
MLList.set("aggregation: nodes per aggregate", NPA);
MLList.set("smoother: type","symmetric Gauss-Seidel");
MLList.set("smoother: pre or post", "both");
MLList.set("coarse: max size", 512);
MLList.set("coarse: type","Amesos-KLU");
MLList.set("analyze memory", true);
MLList.set("repartition: enable", true);
MLList.set("repartition: max min ratio", 1.1);
MLList.set("repartition: min per proc", 512);
MLList.set("low memory usage", true);
MLList.set("x-coordinates", (double*) A.XCoord());
MLList.set("y-coordinates", (double*) A.YCoord());
MLList.set("z-coordinates", (double*) A.ZCoord());
// create the preconditioner object and compute hierarchy
MultiLevelPreconditioner* MLPrec = new MultiLevelPreconditioner(A, MLList);
// tell AztecOO to use this preconditioner, then solve
solver.SetPrecOperator(MLPrec);
solver.SetAztecOption(AZ_solver, AZ_cg_condnum);
solver.SetAztecOption(AZ_output, 1);
solver.Iterate(500, 1e-10);
delete MLPrec;
double norm;
LHS.Norm2(&norm);
if (Comm.MyPID() == 0)
std::cout << "Norm of the error = " << norm << std::endl;
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
exit(EXIT_SUCCESS);
}
int main(int argc, char *argv[]) {
int i;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
int MyPID = Comm.MyPID();
// Number of dimension of the domain
int space_dim = 2;
// Size of each of the dimensions of the domain
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
brick_dim[1] = 1.0;
// Number of elements in each of the dimensions of the domain
std::vector<int> elements( space_dim );
elements[0] = 10;
elements[1] = 10;
// Create problem
Teuchos::RCP<ModalProblem> testCase = Teuchos::rcp( new ModeLaplace2DQ2(Comm, brick_dim[0], elements[0], brick_dim[1], elements[1]) );
// Get the stiffness and mass matrices
Teuchos::RCP<Epetra_CrsMatrix> K = Teuchos::rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
Teuchos::RCP<Epetra_CrsMatrix> M = Teuchos::rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
//
// *******************************************************
// Set up Amesos direct solver for inner iteration
// *******************************************************
//
// Create the shifted system K - sigma * M.
// For the buckling transformation, this shift must be nonzero.
double sigma = 1.0;
Epetra_CrsMatrix Kcopy( *K );
int addErr = EpetraExt::MatrixMatrix::Add( *M, false, -sigma, Kcopy, 1.0 );
if (addErr != 0) {
if (MyPID == 0) {
std::cout << "EpetraExt::MatrixMatrix::Add returned with error: " << addErr << std::endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Create Epetra linear problem class to solve "x = b"
Epetra_LinearProblem AmesosProblem;
AmesosProblem.SetOperator(&Kcopy);
// Create Amesos factory and solver for solving "(K - sigma*M)x = b" using a direct factorization
Amesos amesosFactory;
Teuchos::RCP<Amesos_BaseSolver> AmesosSolver =
Teuchos::rcp( amesosFactory.Create( "Klu", AmesosProblem ) );
// The AmesosBucklingOp class assumes that the symbolic/numeric factorizations have already
// been performed on the linear problem.
AmesosSolver->SymbolicFactorization();
AmesosSolver->NumericFactorization();
//
// ************************************
// Start the block Arnoldi iteration
// ************************************
//
// Variables used for the Block Arnoldi Method
//
int nev = 10;
int blockSize = 3;
int numBlocks = 3*nev/blockSize;
int maxRestarts = 5;
//int step = 5;
double tol = 1.0e-8;
std::string which = "LM";
int verbosity = Anasazi::Errors + Anasazi::Warnings + Anasazi::FinalSummary;
//
// Create parameter list to pass into solver
//
Teuchos::ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Num Blocks", numBlocks );
MyPL.set( "Maximum Restarts", maxRestarts );
MyPL.set( "Convergence Tolerance", tol );
//MyPL.set( "Step Size", step );
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<double, MV> MVT;
typedef Anasazi::OperatorTraits<double, MV, OP> OPT;
// Create an Epetra_MultiVector for an initial vector to start the solver.
// Note: This needs to have the same number of columns as the blocksize.
Teuchos::RCP<Epetra_MultiVector> ivec = Teuchos::rcp( new Epetra_MultiVector(K->Map(), blockSize) );
MVT::MvRandom( *ivec );
// Create the Epetra_Operator for the buckling transformation using the Amesos direct solver.
Teuchos::RCP<AmesosBucklingOp> BucklingOp
= Teuchos::rcp( new AmesosBucklingOp(AmesosProblem, AmesosSolver, K) );
Teuchos::RCP<Anasazi::BasicEigenproblem<double,MV,OP> > MyProblem =
Teuchos::rcp( new Anasazi::BasicEigenproblem<double,MV,OP>(BucklingOp, K, ivec) );
// Inform the eigenproblem that the matrix pencil (K,M) is symmetric
MyProblem->setHermitian(true);
// Set the number of eigenvalues requested
MyProblem->setNEV( nev );
// Inform the eigenproblem that you are finished passing it information
bool boolret = MyProblem->setProblem();
if (boolret != true) {
if (MyPID == 0) {
std::cout << "Anasazi::BasicEigenproblem::setProblem() returned with error." << std::endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Initialize the Block Arnoldi solver
Anasazi::BlockKrylovSchurSolMgr<double, MV, OP> MySolverMgr(MyProblem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
if (returnCode != Anasazi::Converged && MyPID==0) {
std::cout << "Anasazi::EigensolverMgr::solve() returned unconverged." << std::endl;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<double,MV> sol = MyProblem->getSolution();
std::vector<Anasazi::Value<double> > evals = sol.Evals;
Teuchos::RCP<MV> evecs = sol.Evecs;
int numev = sol.numVecs;
if (numev > 0) {
// Undo buckling transformation; computed eigenvalues are real
std::vector<double> compEvals(numev);
for (i=0; i<numev; ++i) {
compEvals[i] = sigma*evals[i].realpart/(evals[i].realpart-1.0);
}
// Remember, eigenvectors are constructed K-orthogonal to preserve symmetry,
// so numerator of the Rayleigh quotient is 1.0.
Teuchos::SerialDenseMatrix<int,double> dmatr(numev,numev);
Epetra_MultiVector tempvec(M->Map(), MVT::GetNumberVecs( *evecs ));
OPT::Apply( *M, *evecs, tempvec );
MVT::MvTransMv( 1.0, tempvec, *evecs, dmatr );
if (MyPID==0) {
double rq_eval = 0.0;
std::cout.setf(std::ios_base::right, std::ios_base::adjustfield);
std::cout<<"Actual Eigenvalues (obtained by Rayleigh quotient) : "<<std::endl;
std::cout<<"------------------------------------------------------"<<std::endl;
std::cout<<std::setw(16)<<"Real Part"
<<std::setw(16)<<"Rayleigh Error"<<std::endl;
std::cout<<"------------------------------------------------------"<<std::endl;
for (i=0; i<numev; i++) {
rq_eval = 1.0 / dmatr(i,i);
std::cout<<std::setw(16)<<rq_eval
<<std::setw(16)<<Teuchos::ScalarTraits<double>::magnitude(rq_eval-compEvals[i])
<<std::endl;
}
std::cout<<"------------------------------------------------------"<<std::endl;
}
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return 0;
}
int main(int argc, char *argv[]) {
int ierr=0, returnierr=0;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
if (!verbose) {
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
}
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if (verbose && MyPID==0)
cout << Epetra_Version() << endl << endl;
if (verbose) cout << Comm << endl;
bool verbose1 = verbose;
if (verbose) verbose = (MyPID==0);
int NumMyElements = 10000;
int NumMyElements1 = NumMyElements; // Used for local map
int NumGlobalElements = NumMyElements*NumProc+EPETRA_MIN(NumProc,3);
if (MyPID < 3) NumMyElements++;
int IndexBase = 0;
bool DistributedGlobal = (NumGlobalElements>NumMyElements);
Epetra_Map* Map;
// Test exceptions
if (verbose)
cout << "*******************************************************************************************" << endl
<< " Testing Exceptions (Expect error messages if EPETRA_NO_ERROR_REPORTS is not defined" << endl
<< "*******************************************************************************************" << endl
<< endl << endl;
try {
if (verbose) cout << "Checking Epetra_Map(-2, IndexBase, Comm)" << endl;
Map = new Epetra_Map(-2, IndexBase, Comm);
}
catch (int Error) {
if (Error!=-1) {
if (Error!=0) {
EPETRA_TEST_ERR(Error,returnierr);
if (verbose) cout << "Error code should be -1" << endl;
}
else {
cout << "Error code = " << Error << "Should be -1" << endl;
returnierr+=1;
}
}
else if (verbose) cout << "Checked OK\n\n" << endl;
}
try {
if (verbose) cout << "Checking Epetra_Map(2, 3, IndexBase, Comm)" << endl;
Map = new Epetra_Map(2, 3, IndexBase, Comm);
}
catch (int Error) {
if (Error!=-4) {
if (Error!=0) {
EPETRA_TEST_ERR(Error,returnierr);
if (verbose) cout << "Error code should be -4" << endl;
}
else {
cout << "Error code = " << Error << "Should be -4" << endl;
returnierr+=1;
}
}
else if (verbose) cout << "Checked OK\n\n" << endl;
}
if (verbose) cerr << flush;
if (verbose) cout << flush;
Comm.Barrier();
if (verbose)
cout << endl << endl
<< "*******************************************************************************************" << endl
<< " Testing valid constructor now......................................................" << endl
<< "*******************************************************************************************" << endl
<< endl << endl;
// Test Epetra-defined uniform linear distribution constructor
Map = new Epetra_Map(NumGlobalElements, IndexBase, Comm);
if (verbose) cout << "Checking Epetra_Map(NumGlobalElements, IndexBase, Comm)" << endl;
ierr = checkmap(*Map, NumGlobalElements, NumMyElements, 0,
IndexBase, Comm, DistributedGlobal);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
delete Map;
// Test User-defined linear distribution constructor
Map = new Epetra_Map(NumGlobalElements, NumMyElements, IndexBase, Comm);
if (verbose) cout << "Checking Epetra_Map(NumGlobalElements, NumMyElements, IndexBase, Comm)" << endl;
ierr = checkmap(*Map, NumGlobalElements, NumMyElements, 0,
IndexBase, Comm, DistributedGlobal);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
delete Map;
// Test User-defined arbitrary distribution constructor
// Generate Global Element List. Do in reverse for fun!
int * MyGlobalElements = new int[NumMyElements];
int MaxMyGID = (Comm.MyPID()+1)*NumMyElements-1+IndexBase;
if (Comm.MyPID()>2) MaxMyGID+=3;
for (int i = 0; i<NumMyElements; i++) MyGlobalElements[i] = MaxMyGID-i;
Map = new Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements,
IndexBase, Comm);
if (verbose) cout << "Checking Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements, IndexBase, Comm)" << endl;
ierr = checkmap(*Map, NumGlobalElements, NumMyElements, MyGlobalElements,
IndexBase, Comm, DistributedGlobal);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
// Test Copy constructor
Epetra_Map* Map1 = new Epetra_Map(*Map);
// Test SameAs() method
bool same = Map1->SameAs(*Map);
EPETRA_TEST_ERR(!(same==true),ierr);// should return true since Map1 is a copy of Map
Epetra_BlockMap* Map2 = new Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements, IndexBase, Comm);
same = Map2->SameAs(*Map);
EPETRA_TEST_ERR(!(same==true),ierr); // Map and Map2 were created with the same sets of parameters
delete Map2;
// now test SameAs() on a map that is different
Map2 = new Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements, IndexBase-1, Comm);
same = Map2->SameAs(*Map);
EPETRA_TEST_ERR(!(same==false),ierr); // IndexBases are different
delete Map2;
// Back to testing copy constructor
if (verbose) cout << "Checking Epetra_Map(*Map)" << endl;
ierr = checkmap(*Map1, NumGlobalElements, NumMyElements, MyGlobalElements,
IndexBase, Comm, DistributedGlobal);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
Epetra_Map* SmallMap = 0;
if (verbose1) {
// Build a small map for test cout. Use 10 elements from current map
int* MyEls = Map->MyGlobalElements();
int IndBase = Map->IndexBase();
int MyLen = EPETRA_MIN(10+Comm.MyPID(),Map->NumMyElements());
SmallMap = new Epetra_Map(-1, MyLen, MyEls, IndBase, Comm);
}
delete [] MyGlobalElements;
delete Map;
delete Map1;
// Test reference-counting in Epetra_Map
if (verbose) cout << "Checking Epetra_Map reference counting" << endl;
ierr = checkMapDataClass(Comm, verbose);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
// Test LocalMap constructor
Epetra_LocalMap* LocalMap = new Epetra_LocalMap(NumMyElements1, IndexBase, Comm);
if (verbose) cout << "Checking Epetra_LocalMap(NumMyElements1, IndexBase, Comm)" << endl;
ierr = checkmap(*LocalMap, NumMyElements1, NumMyElements1, 0, IndexBase, Comm, false);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
// Test Copy constructor
Epetra_LocalMap* LocalMap1 = new Epetra_LocalMap(*LocalMap);
if (verbose) cout << "Checking Epetra_LocalMap(*LocalMap)" << endl;
ierr = checkmap(*LocalMap1, NumMyElements1, NumMyElements1, 0, IndexBase, Comm, false);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
delete LocalMap1;
delete LocalMap;
// Test reference-counting in Epetra_LocalMap
if (verbose) cout << "Checking Epetra_LocalMap reference counting" << endl;
ierr = checkLocalMapDataClass(Comm, verbose);
EPETRA_TEST_ERR(ierr,returnierr);
if (verbose && ierr==0) cout << "Checked OK\n\n" <<endl;
// Test output
if (verbose1) {
if (verbose) cout << "Test ostream << operator" << endl << flush;
cout << *SmallMap;
delete SmallMap;
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return returnierr;
}
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
Teuchos::CommandLineProcessor clp(false);
clp.setDocString("This is the canonical ML scaling example");
//Problem
std::string optMatrixType = "Laplace2D"; clp.setOption("matrixType", &optMatrixType, "matrix type ('Laplace2D', 'Laplace3D')");
int optNx = 100; clp.setOption("nx", &optNx, "mesh size in x direction");
int optNy = -1; clp.setOption("ny", &optNy, "mesh size in y direction");
int optNz = -1; clp.setOption("nz", &optNz, "mesh size in z direction");
//Smoothers
//std::string optSmooType = "Chebshev"; clp.setOption("smooType", &optSmooType, "smoother type ('l1-sgs', 'sgs 'or 'cheby')");
int optSweeps = 3; clp.setOption("sweeps", &optSweeps, "Chebyshev degreee (or SGS sweeps)");
double optAlpha = 7; clp.setOption("alpha", &optAlpha, "Chebyshev eigenvalue ratio (recommend 7 in 2D, 20 in 3D)");
//Coarsening
int optMaxCoarseSize = 500; clp.setOption("maxcoarse", &optMaxCoarseSize, "Size of coarsest grid when coarsening should stop");
int optMaxLevels = 10; clp.setOption("maxlevels", &optMaxLevels, "Maximum number of levels");
//Krylov solver
double optTol = 1e-12; clp.setOption("tol", &optTol, "stopping tolerance for Krylov method");
int optMaxIts = 500; clp.setOption("maxits", &optMaxIts, "maximum iterations for Krylov method");
//XML file with additional options
std::string xmlFile = ""; clp.setOption("xml", &xmlFile, "XML file containing ML options. [OPTIONAL]");
//Debugging
int optWriteMatrices = -2; clp.setOption("write", &optWriteMatrices, "write matrices to file (-1 means all; i>=0 means level i)");
switch (clp.parse(argc, argv)) {
case Teuchos::CommandLineProcessor::PARSE_HELP_PRINTED: return EXIT_SUCCESS; break;
case Teuchos::CommandLineProcessor::PARSE_ERROR:
case Teuchos::CommandLineProcessor::PARSE_UNRECOGNIZED_OPTION: return EXIT_FAILURE; break;
case Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL: break;
}
#ifdef ML_SCALING
const int ntimers=4;
enum {total, probBuild, precBuild, solve};
ml_DblLoc timeVec[ntimers], maxTime[ntimers], minTime[ntimers];
for (int i=0; i<ntimers; i++) timeVec[i].rank = Comm.MyPID();
timeVec[total].value = MPI_Wtime();
#endif
// Creates the linear problem using the Galeri package.
// Several matrix examples are supported; please refer to the
// Galeri documentation for more details.
// Most of the examples using the ML_Epetra::MultiLevelPreconditioner
// class are based on Epetra_CrsMatrix. Example
// `ml_EpetraVbr.cpp' shows how to define a Epetra_VbrMatrix.
// `Laplace2D' is a symmetric matrix; an example of non-symmetric
// matrices is `Recirc2D' (advection-diffusion in a box, with
// recirculating flow). The grid has optNx x optNy nodes, divided into
// mx x my subdomains, each assigned to a different processor.
if (optNy == -1) optNy = optNx;
if (optNz == -1) optNz = optNx;
ParameterList GaleriList;
GaleriList.set("nx", optNx);
GaleriList.set("ny", optNy);
GaleriList.set("nz", optNz);
//GaleriList.set("mx", 1);
//GaleriList.set("my", Comm.NumProc());
#ifdef ML_SCALING
timeVec[probBuild].value = MPI_Wtime();
#endif
Epetra_Map* Map;
Epetra_CrsMatrix* A;
Epetra_MultiVector* Coord;
if (optMatrixType == "Laplace2D") {
Map = CreateMap("Cartesian2D", Comm, GaleriList);
A = CreateCrsMatrix("Laplace2D", Map, GaleriList);
Coord = CreateCartesianCoordinates("2D", &(A->Map()), GaleriList);
} else if (optMatrixType == "Laplace3D") {
Map = CreateMap("Cartesian3D", Comm, GaleriList);
A = CreateCrsMatrix("Laplace3D", Map, GaleriList);
Coord = CreateCartesianCoordinates("3D", &(A->Map()), GaleriList);
} else {
throw(std::runtime_error("Bad matrix type"));
}
//EpetraExt::RowMatrixToMatlabFile("A.m",*A);
double *x_coord = (*Coord)[0];
double *y_coord = (*Coord)[1];
double* z_coord=NULL;
if (optMatrixType == "Laplace3D") z_coord = (*Coord)[2];
//EpetraExt::MultiVectorToMatrixMarketFile("mlcoords.m",*Coord);
if( Comm.MyPID()==0 ) {
std::cout << "========================================================" << std::endl;
std::cout << " Matrix type: " << optMatrixType << std::endl;
if (optMatrixType == "Laplace2D")
std::cout << " Problem size: " << optNx*optNy << " (" << optNx << "x" << optNy << ")" << std::endl;
else if (optMatrixType == "Laplace3D")
std::cout << " Problem size: " << optNx*optNy*optNz << " (" << optNx << "x" << optNy << "x" << optNz << ")" << std::endl;
int mx = GaleriList.get("mx", -1);
int my = GaleriList.get("my", -1);
int mz = GaleriList.get("my", -1);
std::cout << " Processor subdomains in x direction: " << mx << std::endl
<< " Processor subdomains in y direction: " << my << std::endl;
if (optMatrixType == "Laplace3D")
std::cout << " Processor subdomains in z direction: " << mz << std::endl;
std::cout << "========================================================" << std::endl;
}
// Build a linear system with trivial solution, using a random vector
// as starting solution.
Epetra_Vector LHS(*Map); LHS.Random();
Epetra_Vector RHS(*Map); RHS.PutScalar(0.0);
Epetra_LinearProblem Problem(A, &LHS, &RHS);
// As we wish to use AztecOO, we need to construct a solver object
// for this problem
AztecOO solver(Problem);
#ifdef ML_SCALING
timeVec[probBuild].value = MPI_Wtime() - timeVec[probBuild].value;
#endif
// =========================== begin of ML part ===========================
#ifdef ML_SCALING
timeVec[precBuild].value = MPI_Wtime();
#endif
// create a parameter list for ML options
ParameterList MLList;
// Sets default parameters for classic smoothed aggregation. After this
// call, MLList contains the default values for the ML parameters,
// as required by typical smoothed aggregation for symmetric systems.
// Other sets of parameters are available for non-symmetric systems
// ("DD" and "DD-ML"), and for the Maxwell equations ("maxwell").
ML_Epetra::SetDefaults("SA",MLList);
// overwrite some parameters. Please refer to the user's guide
// for more information
// some of the parameters do not differ from their default value,
// and they are here reported for the sake of clarity
// output level, 0 being silent and 10 verbose
MLList.set("ML output", 10);
// maximum number of levels
MLList.set("max levels",optMaxLevels);
// set finest level to 0
MLList.set("increasing or decreasing","increasing");
MLList.set("coarse: max size",optMaxCoarseSize);
// use Uncoupled scheme to create the aggregate
MLList.set("aggregation: type", "Uncoupled");
// smoother is Chebyshev. Example file
// `ml/examples/TwoLevelDD/ml_2level_DD.cpp' shows how to use
// AZTEC's preconditioners as smoothers
MLList.set("smoother: type","Chebyshev");
MLList.set("smoother: Chebyshev alpha",optAlpha);
MLList.set("smoother: sweeps",optSweeps);
// use both pre and post smoothing
MLList.set("smoother: pre or post", "both");
#ifdef HAVE_ML_AMESOS
// solve with serial direct solver KLU
MLList.set("coarse: type","Amesos-KLU");
#else
// this is for testing purposes only, you should have
// a direct solver for the coarse problem (either Amesos, or the SuperLU/
// SuperLU_DIST interface of ML)
MLList.set("coarse: type","Jacobi");
#endif
MLList.set("repartition: enable",1);
MLList.set("repartition: start level",1);
MLList.set("repartition: max min ratio",1.1);
MLList.set("repartition: min per proc",800);
MLList.set("repartition: partitioner","Zoltan");
MLList.set("repartition: put on single proc",1);
MLList.set("x-coordinates", x_coord);
MLList.set("y-coordinates", y_coord);
if (optMatrixType == "Laplace2D") {
MLList.set("repartition: Zoltan dimensions",2);
} else if (optMatrixType == "Laplace3D") {
MLList.set("repartition: Zoltan dimensions",3);
MLList.set("z-coordinates", z_coord);
}
MLList.set("print hierarchy",optWriteMatrices);
//MLList.set("aggregation: damping factor",0.);
// Read in XML options
if (xmlFile != "")
ML_Epetra::ReadXML(xmlFile,MLList,Comm);
// Creates the preconditioning object. We suggest to use `new' and
// `delete' because the destructor contains some calls to MPI (as
// required by ML and possibly Amesos). This is an issue only if the
// destructor is called **after** MPI_Finalize().
ML_Epetra::MultiLevelPreconditioner* MLPrec =
new ML_Epetra::MultiLevelPreconditioner(*A, MLList);
// verify unused parameters on process 0 (put -1 to print on all
// processes)
MLPrec->PrintUnused(0);
#ifdef ML_SCALING
timeVec[precBuild].value = MPI_Wtime() - timeVec[precBuild].value;
#endif
// ML allows the user to cheaply recompute the preconditioner. You can
// simply uncomment the following line:
//
// MLPrec->ReComputePreconditioner();
//
// It is supposed that the linear system matrix has different values, but
// **exactly** the same structure and layout. The code re-built the
// hierarchy and re-setup the smoothers and the coarse solver using
// already available information on the hierarchy. A particular
// care is required to use ReComputePreconditioner() with nonzero
// threshold.
// =========================== end of ML part =============================
// tell AztecOO to use the ML preconditioner, specify the solver
// and the output, then solve with 500 maximum iterations and 1e-12
// of tolerance (see AztecOO's user guide for more details)
#ifdef ML_SCALING
timeVec[solve].value = MPI_Wtime();
#endif
solver.SetPrecOperator(MLPrec);
solver.SetAztecOption(AZ_solver, AZ_cg);
solver.SetAztecOption(AZ_output, 32);
solver.Iterate(optMaxIts, optTol);
#ifdef ML_SCALING
timeVec[solve].value = MPI_Wtime() - timeVec[solve].value;
#endif
// destroy the preconditioner
delete MLPrec;
// compute the real residual
double residual;
LHS.Norm2(&residual);
if( Comm.MyPID()==0 ) {
cout << "||b-Ax||_2 = " << residual << endl;
}
// for testing purposes
if (residual > 1e-5)
exit(EXIT_FAILURE);
delete A;
delete Map;
#ifdef ML_SCALING
timeVec[total].value = MPI_Wtime() - timeVec[total].value;
//avg
double dupTime[ntimers],avgTime[ntimers];
for (int i=0; i<ntimers; i++) dupTime[i] = timeVec[i].value;
MPI_Reduce(dupTime,avgTime,ntimers,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD);
for (int i=0; i<ntimers; i++) avgTime[i] = avgTime[i]/Comm.NumProc();
//min
MPI_Reduce(timeVec,minTime,ntimers,MPI_DOUBLE_INT,MPI_MINLOC,0,MPI_COMM_WORLD);
//max
MPI_Reduce(timeVec,maxTime,ntimers,MPI_DOUBLE_INT,MPI_MAXLOC,0,MPI_COMM_WORLD);
if (Comm.MyPID() == 0) {
printf("timing : max (pid) min (pid) avg\n");
printf("Problem build : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[probBuild].value,maxTime[probBuild].rank,
minTime[probBuild].value,minTime[probBuild].rank,
avgTime[probBuild]);
printf("Preconditioner build : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[precBuild].value,maxTime[precBuild].rank,
minTime[precBuild].value,minTime[precBuild].rank,
avgTime[precBuild]);
printf("Solve : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[solve].value,maxTime[solve].rank,
minTime[solve].value,minTime[solve].rank,
avgTime[solve]);
printf("Total : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[total].value,maxTime[total].rank,
minTime[total].value,minTime[total].rank,
avgTime[total]);
}
#endif
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
}
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc, &argv);
// define an Epetra communicator
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// check number of processes
if (Comm.NumProc() != 1) {
if (Comm.MyPID() == 0)
cerr << "*ERR* can be used only with one process" << endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
exit(EXIT_SUCCESS);
}
// process 0 will read an HB matrix, and store it
// in the MSR format given by the arrays bindx and val
int N_global;
int N_nonzeros;
double * val = NULL;
int * bindx = NULL;
double * x = NULL, * b = NULL, * xexact = NULL;
FILE* fp = fopen("../HBMatrices/fidap005.rua", "r");
if (fp == 0)
{
cerr << "Matrix file not available" << endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
exit(EXIT_SUCCESS);
}
fclose(fp);
Trilinos_Util_read_hb("../HBMatrices/fidap005.rua", 0,
&N_global, &N_nonzeros,
&val, &bindx,
&x, &b, &xexact);
// assign all the elements to process 0
// (this code can run ONLY with one process, extensions to more
// processes will require functions to handle update of ghost nodes)
Epetra_Map Map(N_global,0,Comm);
MSRMatrix A(Map,bindx,val);
// define two vectors
Epetra_Vector xxx(Map);
Epetra_Vector yyy(Map);
xxx.Random();
A.Apply(xxx,yyy);
cout << yyy;
double norm2;
yyy.Norm2(&norm2);
cout << norm2 << endl;
// free memory allocated by Trilinos_Util_read_hb
if (val != NULL) free((void*)val);
if (bindx != NULL) free((void*)bindx);
if (x != NULL) free((void*)x);
if (b != NULL) free((void*)b);
if (xexact != NULL) free((void*)xexact);;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
} /* main */
int main(int argc, char *argv[]) {
//
bool haveM = false;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
int MyPID = Comm.MyPID();
int nev = 5;
int blockSize = 5;
int maxIterations = 1000;
double tol = 1.0e-8;
bool verbose=false, locking=false, fullOrtho=true;
std::string k_filename = "";
std::string m_filename = "";
std::string which = "SM";
Teuchos::CommandLineProcessor cmdp(false,true);
cmdp.setOption("nev",&nev,"Number of eigenvalues to compute.");
cmdp.setOption("blocksize",&blockSize,"Block size used in LOBPCG.");
cmdp.setOption("maxiters",&maxIterations,"Maximum number of iterations used in LOBPCG.");
cmdp.setOption("tol",&tol,"Convergence tolerance requested for computed eigenvalues.");
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("locking","nolocking",&locking,"Use locking of converged eigenvalues.");
cmdp.setOption("fullortho","nofullortho",&fullOrtho,"Use full orthogonalization.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SM,LM,SR,or LR).");
cmdp.setOption("K-filename",&k_filename,"Filename and path of the stiffness matrix.");
cmdp.setOption("M-filename",&m_filename,"Filename and path of the mass matrix.");
if (cmdp.parse(argc,argv) != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
if (k_filename=="") {
cout << "The matrix K must be supplied through an input file!!!" << endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
if (m_filename!="") {
haveM = true;
}
//
//**********************************************************************
//******************Set up the problem to be solved*********************
//**********************************************************************
//
// *****Read in matrix from file******
//
Teuchos::RCP<Epetra_Map> Map;
Teuchos::RCP<Epetra_CrsMatrix> K, M;
EpetraExt::readEpetraLinearSystem( k_filename, Comm, &K, &Map );
if (haveM) {
EpetraExt::readEpetraLinearSystem( m_filename, Comm, &M, &Map );
}
//
//************************************
// Start the block Davidson iteration
//***********************************
//
// Variables used for the LOBPCG Method
//
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
//
// Create parameter list to pass into solver
//
Teuchos::ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Maximum Iterations", maxIterations );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Use Locking", locking );
MyPL.set( "Locking Tolerance", tol/10 );
MyPL.set( "Full Ortho", fullOrtho );
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<double, MV> MVT;
typedef Anasazi::OperatorTraits<double, MV, OP> OPT;
//
// Create the eigenproblem to be solved.
//
Teuchos::RCP<Epetra_MultiVector> ivec = Teuchos::rcp( new Epetra_MultiVector(*Map, blockSize) );
ivec->Random();
Teuchos::RCP<Anasazi::BasicEigenproblem<double, MV, OP> > MyProblem;
if (haveM) {
MyProblem = Teuchos::rcp( new Anasazi::BasicEigenproblem<double, MV, OP>( K, M, ivec ) );
}
else {
MyProblem = Teuchos::rcp( new Anasazi::BasicEigenproblem<double, MV, OP>( K, ivec ) );
}
// Inform the eigenproblem that (K,M) is Hermitian
MyProblem->setHermitian(true);
// Set the number of eigenvalues requested
MyProblem->setNEV( nev );
// Inform the eigenproblem that you are finished passing it information
bool boolret = MyProblem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::setProblem() returned with error." << endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Initialize the LOBPCG solver
Anasazi::LOBPCGSolMgr<double, MV, OP> MySolverMgr(MyProblem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
if (returnCode != Anasazi::Converged && MyPID==0 && verbose) {
cout << "Anasazi::EigensolverMgr::solve() returned unconverged." << endl;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<double,MV> sol = MyProblem->getSolution();
std::vector<Anasazi::Value<double> > evals = sol.Evals;
Teuchos::RCP<MV> evecs = sol.Evecs;
std::vector<int> index = sol.index;
int numev = sol.numVecs;
if (numev > 0) {
// Compute residuals.
Teuchos::LAPACK<int,double> lapack;
std::vector<double> normEV(numev);
// Get storage
Teuchos::RCP<Epetra_MultiVector> Kevecs, Mevecs;
Teuchos::SerialDenseMatrix<int,double> B(numev,numev);
B.putScalar(0.0);
for (int i=0; i<numev; i++) {B(i,i) = evals[i].realpart;}
// Compute K*evecs
Kevecs = Teuchos::rcp(new Epetra_MultiVector(*Map,numev) );
OPT::Apply( *K, *evecs, *Kevecs );
// Compute M*evecs
if (haveM) {
Mevecs = Teuchos::rcp(new Epetra_MultiVector(*Map,numev) );
OPT::Apply( *M, *evecs, *Mevecs );
}
else {
Mevecs = evecs;
}
// Compute K*evecs - lambda*M*evecs and its norm
MVT::MvTimesMatAddMv( -1.0, *Mevecs, B, 1.0, *Kevecs );
MVT::MvNorm( *Kevecs, normEV );
// Scale the norms by the eigenvalue
for (int i=0; i<numev; i++) {
normEV[i] /= Teuchos::ScalarTraits<double>::magnitude( evals[i].realpart );
}
// Output computed eigenvalues and their direct residuals
if (verbose && MyPID==0) {
cout.setf(std::ios_base::right, std::ios_base::adjustfield);
cout<<endl<< "Actual Residuals"<<endl;
cout<< std::setw(16) << "Real Part"
<< std::setw(20) << "Direct Residual"<< endl;
cout<<"-----------------------------------------------------------"<<endl;
for (int i=0; i<numev; i++) {
cout<< std::setw(16) << evals[i].realpart
<< std::setw(20) << normEV[i] << endl;
}
cout<<"-----------------------------------------------------------"<<endl;
}
}
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return 0;
} // end LOBPCGEpetraExFile.cpp
int main(int argc, char *argv[]) {
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool testFailed;
bool boolret;
int MyPID = Comm.MyPID();
bool verbose = true;
bool debug = false;
std::string which("SM");
Teuchos::CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SM,LM,SR,LR,SI,or LI).");
if (cmdp.parse(argc,argv) != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
typedef double ScalarType;
typedef Teuchos::ScalarTraits<ScalarType> ScalarTypeTraits;
typedef ScalarTypeTraits::magnitudeType MagnitudeType;
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVTraits;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OpTraits;
// Dimension of the matrix
int nx = 10; // Discretization points in any one direction.
int NumGlobalElements = nx*nx; // Size of matrix nx*nx
// Construct a Map that puts approximately the same number of
// equations on each processor.
Epetra_Map Map(NumGlobalElements, 0, Comm);
// Get update list and number of local equations from newly created Map.
int NumMyElements = Map.NumMyElements();
std::vector<int> MyGlobalElements(NumMyElements);
Map.MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation
// on this processor
std::vector<int> NumNz(NumMyElements);
/* We are building a matrix of block structure:
| T -I |
|-I T -I |
| -I T |
| ... -I|
| -I T|
where each block is dimension nx by nx and the matrix is on the order of
nx*nx. The block T is a tridiagonal matrix.
*/
for (int i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i] == 0 || MyGlobalElements[i] == NumGlobalElements-1 ||
MyGlobalElements[i] == nx-1 || MyGlobalElements[i] == nx*(nx-1) ) {
NumNz[i] = 3;
}
else if (MyGlobalElements[i] < nx || MyGlobalElements[i] > nx*(nx-1) ||
MyGlobalElements[i]%nx == 0 || (MyGlobalElements[i]+1)%nx == 0) {
NumNz[i] = 4;
}
else {
NumNz[i] = 5;
}
}
// Create an Epetra_Matrix
Teuchos::RCP<Epetra_CrsMatrix> A = Teuchos::rcp( new Epetra_CrsMatrix(Copy, Map, &NumNz[0]) );
// Diffusion coefficient, can be set by user.
// When rho*h/2 <= 1, the discrete convection-diffusion operator has real eigenvalues.
// When rho*h/2 > 1, the operator has complex eigenvalues.
double rho = 2*(nx+1);
// Compute coefficients for discrete convection-diffution operator
const double one = 1.0;
std::vector<double> Values(4);
std::vector<int> Indices(4);
double h = one /(nx+1);
double h2 = h*h;
double c = 5.0e-01*rho/ h;
Values[0] = -one/h2 - c; Values[1] = -one/h2 + c; Values[2] = -one/h2; Values[3]= -one/h2;
double diag = 4.0 / h2;
int NumEntries, info;
for (int i=0; i<NumMyElements; i++)
{
if (MyGlobalElements[i]==0)
{
Indices[0] = 1;
Indices[1] = nx;
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == nx*(nx-1))
{
Indices[0] = nx*(nx-1)+1;
Indices[1] = nx*(nx-2);
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == nx-1)
{
Indices[0] = nx-2;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
Indices[0] = 2*nx-1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == NumGlobalElements-1)
{
Indices[0] = NumGlobalElements-2;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
Indices[0] = nx*(nx-1)-1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] < nx)
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]+nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] > nx*(nx-1))
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]-nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i]%nx == 0)
{
Indices[0] = MyGlobalElements[i]+1;
Indices[1] = MyGlobalElements[i]-nx;
Indices[2] = MyGlobalElements[i]+nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if ((MyGlobalElements[i]+1)%nx == 0)
{
Indices[0] = MyGlobalElements[i]-nx;
Indices[1] = MyGlobalElements[i]+nx;
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
Indices[0] = MyGlobalElements[i]-1;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]-nx;
Indices[3] = MyGlobalElements[i]+nx;
NumEntries = 4;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
// Put in the diagonal entry
info = A->InsertGlobalValues(MyGlobalElements[i], 1, &diag, &MyGlobalElements[i]);
assert( info==0 );
}
// Finish up
info = A->FillComplete();
assert( info==0 );
A->SetTracebackMode(1); // Shutdown Epetra Warning tracebacks
//************************************
// Start the block Davidson iteration
//***********************************
//
// Variables used for the Generalized Davidson Method
//
int nev = 4;
int blockSize = 1;
int maxDim = 50;
int restartDim = 10;
int maxRestarts = 500;
double tol = 1e-10;
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
//
// Create parameter list to pass into solver manager
//
Teuchos::ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Maximum Subspace Dimension", maxDim);
MyPL.set( "Restart Dimension", restartDim);
MyPL.set( "Maximum Restarts", maxRestarts );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Relative Convergence Tolerance", true );
MyPL.set( "Initial Guess", "User" );
// Create an Epetra_MultiVector for an initial vector to start the solver.
// Note: This needs to have the same number of columns as the blocksize.
Teuchos::RCP<Epetra_MultiVector> ivec = Teuchos::rcp( new Epetra_MultiVector(Map, blockSize) );
ivec->Random();
// Create the eigenproblem.
Teuchos::RCP<Anasazi::BasicEigenproblem<double, MV, OP> > MyProblem = Teuchos::rcp(
new Anasazi::BasicEigenproblem<double,MV,OP>() );
MyProblem->setA(A);
MyProblem->setInitVec(ivec);
// Inform the eigenproblem that the operator A is non-Hermitian
MyProblem->setHermitian(false);
// Set the number of eigenvalues requested
MyProblem->setNEV( nev );
// Inform the eigenproblem that you are finishing passing it information
boolret = MyProblem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
std::cout << "Anasazi::BasicEigenproblem::setProblem() returned with error." << std::endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Initialize the Block Arnoldi solver
Anasazi::GeneralizedDavidsonSolMgr<double, MV, OP> MySolverMgr(MyProblem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
testFailed = false;
if (returnCode != Anasazi::Converged && MyPID==0 && verbose) {
testFailed = true;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ScalarType,MV> sol = MyProblem->getSolution();
std::vector<Anasazi::Value<ScalarType> > evals = sol.Evals;
Teuchos::RCP<MV> evecs = sol.Evecs;
std::vector<int> index = sol.index;
int numev = sol.numVecs;
// Output computed eigenvalues and their direct residuals
if (verbose && MyPID==0) {
int numritz = (int)evals.size();
std::cout.setf(std::ios_base::right, std::ios_base::adjustfield);
std::cout<<std::endl<< "Computed Ritz Values"<< std::endl;
std::cout<< std::setw(16) << "Real Part"
<< std::setw(16) << "Imag Part"
<< std::endl;
std::cout<<"-----------------------------------------------------------"<<std::endl;
for (int i=0; i<numritz; i++) {
std::cout<< std::setw(16) << evals[i].realpart
<< std::setw(16) << evals[i].imagpart
<< std::endl;
}
std::cout<<"-----------------------------------------------------------"<<std::endl;
}
if (numev > 0) {
// Compute residuals.
Teuchos::LAPACK<int,double> lapack;
std::vector<double> normA(numev);
// The problem is non-Hermitian.
int i=0;
std::vector<int> curind(1);
std::vector<double> resnorm(1), tempnrm(1);
Teuchos::RCP<MV> tempAevec;
Teuchos::RCP<const MV> evecr, eveci;
Epetra_MultiVector Aevec(Map,numev);
// Compute A*evecs
OpTraits::Apply( *A, *evecs, Aevec );
Teuchos::SerialDenseMatrix<int,double> Breal(1,1), Bimag(1,1);
while (i<numev) {
if (index[i]==0) {
// Get a view of the current eigenvector (evecr)
curind[0] = i;
evecr = MVTraits::CloneView( *evecs, curind );
// Get a copy of A*evecr
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Compute A*evecr - lambda*evecr
Breal(0,0) = evals[i].realpart;
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Breal, 1.0, *tempAevec );
// Compute the norm of the residual and increment counter
MVTraits::MvNorm( *tempAevec, resnorm );
normA[i] = resnorm[0] / Teuchos::ScalarTraits<MagnitudeType>::magnitude( evals[i].realpart );
i++;
} else {
// Get a view of the real part of the eigenvector (evecr)
curind[0] = i;
evecr = MVTraits::CloneView( *evecs, curind );
// Get a copy of A*evecr
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Get a view of the imaginary part of the eigenvector (eveci)
curind[0] = i+1;
eveci = MVTraits::CloneView( *evecs, curind );
// Set the eigenvalue into Breal and Bimag
Breal(0,0) = evals[i].realpart;
Bimag(0,0) = evals[i].imagpart;
// Compute A*evecr - evecr*lambdar + eveci*lambdai
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Breal, 1.0, *tempAevec );
MVTraits::MvTimesMatAddMv( 1.0, *eveci, Bimag, 1.0, *tempAevec );
MVTraits::MvNorm( *tempAevec, tempnrm );
// Get a copy of A*eveci
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Compute A*eveci - eveci*lambdar - evecr*lambdai
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Bimag, 1.0, *tempAevec );
MVTraits::MvTimesMatAddMv( -1.0, *eveci, Breal, 1.0, *tempAevec );
MVTraits::MvNorm( *tempAevec, resnorm );
// Compute the norms and scale by magnitude of eigenvalue
normA[i] = lapack.LAPY2( tempnrm[0], resnorm[0] ) /
lapack.LAPY2( evals[i].realpart, evals[i].imagpart );
normA[i+1] = normA[i];
i=i+2;
}
}
// Output computed eigenvalues and their direct residuals
if (verbose && MyPID==0) {
std::cout.setf(std::ios_base::right, std::ios_base::adjustfield);
std::cout<<std::endl<< "Actual Residuals"<<std::endl;
std::cout<< std::setw(16) << "Real Part"
<< std::setw(16) << "Imag Part"
<< std::setw(20) << "Direct Residual"<< std::endl;
std::cout<<"-----------------------------------------------------------"<<std::endl;
for (int j=0; j<numev; j++) {
std::cout<< std::setw(16) << evals[j].realpart
<< std::setw(16) << evals[j].imagpart
<< std::setw(20) << normA[j] << std::endl;
if ( normA[j] > tol ) {
testFailed = true;
}
}
std::cout<<"-----------------------------------------------------------"<<std::endl;
}
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
if (testFailed) {
if (verbose && MyPID==0) {
std::cout << "End Result: TEST FAILED" << std::endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
std::cout << "End Result: TEST PASSED" << std::endl;
}
return 0;
}
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
#define ML_SCALING
#ifdef ML_SCALING
const int ntimers=4;
enum {total, probBuild, precBuild, solve};
ml_DblLoc timeVec[ntimers], maxTime[ntimers], minTime[ntimers];
for (int i=0; i<ntimers; i++) timeVec[i].rank = Comm.MyPID();
timeVec[total].value = MPI_Wtime();
#endif
int nx;
if (argc > 1) nx = (int) strtol(argv[1],NULL,10);
else nx = 256;
if (nx < 1) nx = 256; // input a nonpositive integer if you want to specify
// the XML input file name.
nx = nx*(int)sqrt((double)Comm.NumProc());
int ny = nx;
printf("nx = %d\nny = %d\n",nx,ny);
fflush(stdout);
char xmlFile[80];
bool readXML=false;
if (argc > 2) {strcpy(xmlFile,argv[2]); readXML = true;}
else sprintf(xmlFile,"%s","params.xml");
ParameterList GaleriList;
GaleriList.set("nx", nx);
GaleriList.set("ny", ny);
#ifdef ML_SCALING
timeVec[probBuild].value = MPI_Wtime();
#endif
Epetra_Map* Map = CreateMap("Cartesian2D", Comm, GaleriList);
Epetra_CrsMatrix* A = CreateCrsMatrix("Laplace2D", Map, GaleriList);
if (!Comm.MyPID()) printf("nx = %d, ny = %d, mx = %d, my = %d\n",nx,ny,GaleriList.get("mx",-1),GaleriList.get("my",-1));
fflush(stdout);
//avoid potential overflow
double numMyRows = A->NumMyRows();
double numGlobalRows;
Comm.SumAll(&numMyRows,&numGlobalRows,1);
if (!Comm.MyPID()) printf("# global rows = %1.0f\n",numGlobalRows);
//printf("pid %d: #rows = %d\n",Comm.MyPID(),A->NumMyRows());
fflush(stdout);
Epetra_MultiVector *coords = CreateCartesianCoordinates("2D", Map,GaleriList);
double *x_coord=0,*y_coord=0,*z_coord=0;
double **ttt;
if (!coords->ExtractView(&ttt)) {
x_coord = ttt[0];
y_coord = ttt[1];
} else {
if (!Comm.MyPID()) printf("Error extracting coordinate vectors\n");
MPI_Finalize();
exit(EXIT_FAILURE);
}
Epetra_Vector LHS(*Map); LHS.Random();
Epetra_Vector RHS(*Map); RHS.PutScalar(0.0);
Epetra_LinearProblem Problem(A, &LHS, &RHS);
AztecOO solver(Problem);
#ifdef ML_SCALING
timeVec[probBuild].value = MPI_Wtime() - timeVec[probBuild].value;
#endif
// =========================== begin of ML part ===========================
#ifdef ML_SCALING
timeVec[precBuild].value = MPI_Wtime();
#endif
ParameterList MLList;
if (readXML) {
MLList.set("read XML",true);
MLList.set("XML input file",xmlFile);
}
else {
cout << "here" << endl;
ML_Epetra::SetDefaults("SA",MLList);
MLList.set("smoother: type","Chebyshev");
MLList.set("smoother: sweeps",3);
MLList.set("coarse: max size",1);
}
MLList.set("x-coordinates", x_coord);
MLList.set("y-coordinates", y_coord);
MLList.set("z-coordinates", z_coord);
/*
RCP<std::vector<int> >
m_smootherAztecOptions = rcp(new std::vector<int>(AZ_OPTIONS_SIZE));
RCP<std::vector<double> >
m_smootherAztecParams = rcp(new std::vector<double>(AZ_PARAMS_SIZE));
//int m_smootherAztecOptions[AZ_OPTIONS_SIZE];
//double m_smootherAztecParams[AZ_PARAMS_SIZE];
std::string smootherType("Aztec");
AZ_defaults(&(*m_smootherAztecOptions)[0],&(*m_smootherAztecParams)[0]);
(*m_smootherAztecOptions)[AZ_precond] = AZ_dom_decomp;
(*m_smootherAztecOptions)[AZ_subdomain_solve] = AZ_icc;
bool smootherAztecAsSolver = true;
*/
MLList.set("ML output",10);
MLList.set("repartition: enable",1);
MLList.set("repartition: max min ratio",1.3);
MLList.set("repartition: min per proc",200);
MLList.set("repartition: partitioner","Zoltan");
MLList.set("repartition: Zoltan dimensions",2);
MLList.set("repartition: put on single proc",1);
MLList.set("repartition: Zoltan type","hypergraph");
MLList.set("repartition: estimated iterations",13);
/*
MLList.set("smoother: Aztec options",m_smootherAztecOptions);
MLList.set("smoother: Aztec params",m_smootherAztecParams);
MLList.set("smoother: type",smootherType.c_str());
MLList.set("smoother: Aztec as solver", smootherAztecAsSolver);
MLList.set("ML print initial list", 0);
MLList.set("ML print final list", 0);
*/
ML_Epetra::MultiLevelPreconditioner* MLPrec =
new ML_Epetra::MultiLevelPreconditioner(*A, MLList);
// verify unused parameters on process 0 (put -1 to print on all
// processes)
MLPrec->PrintUnused(0);
#ifdef ML_SCALING
timeVec[precBuild].value = MPI_Wtime() - timeVec[precBuild].value;
#endif
// =========================== end of ML part =============================
#ifdef ML_SCALING
timeVec[solve].value = MPI_Wtime();
#endif
solver.SetPrecOperator(MLPrec);
solver.SetAztecOption(AZ_solver, AZ_cg);
solver.SetAztecOption(AZ_output, 1);
solver.Iterate(500, 1e-12);
#ifdef ML_SCALING
timeVec[solve].value = MPI_Wtime() - timeVec[solve].value;
#endif
// destroy the preconditioner
delete MLPrec;
// compute the real residual
double residual;
LHS.Norm2(&residual);
if( Comm.MyPID()==0 ) {
cout << "||b-Ax||_2 = " << residual << endl;
}
// for testing purposes
if (residual > 1e-5)
exit(EXIT_FAILURE);
delete A;
delete Map;
delete coords;
#ifdef ML_SCALING
timeVec[total].value = MPI_Wtime() - timeVec[total].value;
//avg
double dupTime[ntimers],avgTime[ntimers];
for (int i=0; i<ntimers; i++) dupTime[i] = timeVec[i].value;
MPI_Reduce(dupTime,avgTime,ntimers,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD);
for (int i=0; i<ntimers; i++) avgTime[i] = avgTime[i]/Comm.NumProc();
//min
MPI_Reduce(timeVec,minTime,ntimers,MPI_DOUBLE_INT,MPI_MINLOC,0,MPI_COMM_WORLD);
//max
MPI_Reduce(timeVec,maxTime,ntimers,MPI_DOUBLE_INT,MPI_MAXLOC,0,MPI_COMM_WORLD);
if (Comm.MyPID() == 0) {
printf("timing : max (pid) min (pid) avg\n");
printf("Problem build : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[probBuild].value,maxTime[probBuild].rank,
minTime[probBuild].value,minTime[probBuild].rank,
avgTime[probBuild]);
printf("Preconditioner build : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[precBuild].value,maxTime[precBuild].rank,
minTime[precBuild].value,minTime[precBuild].rank,
avgTime[precBuild]);
printf("Solve : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[solve].value,maxTime[solve].rank,
minTime[solve].value,minTime[solve].rank,
avgTime[solve]);
printf("Total : %2.3e (%d) %2.3e (%d) %2.3e \n",
maxTime[total].value,maxTime[total].rank,
minTime[total].value,minTime[total].rank,
avgTime[total]);
}
#endif
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
}
int main(int argc, char *argv[])
{
int pid = -1;
Teuchos::GlobalMPISession session(&argc, &argv, NULL);
#ifdef EPETRA_MPI
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool verbose = false;
bool success = true;
try {
pid = Comm.MyPID();
int n(10);
int numRHS=1;
Epetra_Map Map = Epetra_Map(n, 0, Comm);
Epetra_MultiVector X(Map, numRHS, false), Y(Map, numRHS, false);
X.PutScalar( 1.0 );
// Inner computes inv(D2)*y
Teuchos::RCP<Diagonal_Operator_2> D2 = Teuchos::rcp(new Diagonal_Operator_2(n, 1.0));
Iterative_Inverse_Operator A2(n, 1, D2, "Belos (inv(D2))", true);
// should return x=(1, 1/2, 1/3, ..., 1/10)
A2(X,Y);
if (pid==0) {
std::cout << "Vector Y should have all entries [1, 1/2, 1/3, ..., 1/10]" << std::endl;
}
Y.Print(std::cout);
// Inner computes inv(D)*x
Teuchos::RCP<Diagonal_Operator> D = Teuchos::rcp(new Diagonal_Operator(n, 4.0));
Teuchos::RCP<Iterative_Inverse_Operator> Inner =
Teuchos::rcp(new Iterative_Inverse_Operator(n, 1, D, "Belos (inv(D))", false));
// Composed_Operator computed inv(D)*B*x
Teuchos::RCP<Diagonal_Operator> B = Teuchos::rcp(new Diagonal_Operator(n, 4.0));
Teuchos::RCP<Composed_Operator> C = Teuchos::rcp(new Composed_Operator(n, Inner, B));
// Outer computes inv(C) = inv(inv(D)*B)*x = inv(B)*D*x = x
Teuchos::RCP<Iterative_Inverse_Operator> Outer =
Teuchos::rcp(new Iterative_Inverse_Operator(n, 1, C, "Belos (inv(C)=inv(inv(D)*B))", true));
// should return x=1/4
(*Inner)(X,Y);
if (pid==0) {
std::cout << std::endl << "Vector Y should have all entries [1/4, 1/4, 1/4, ..., 1/4]" << std::endl;
}
Y.Print(std::cout);
// should return x=1
(*Outer)(X,Y);
if (pid==0) {
std::cout << "Vector Y should have all entries [1, 1, 1, ..., 1]" << std::endl;
}
Y.Print(std::cout);
// Compute the norm of Y - 1.0
std::vector<double> norm_Y(Y.NumVectors());
Y.Update(-1.0, X, 1.0);
Y.Norm2(&norm_Y[0]);
if (pid==0)
std::cout << "Two-norm of std::vector (Y-1.0) : "<< norm_Y[0] << std::endl;
success = (norm_Y[0] < 1e-10 && !Teuchos::ScalarTraits<double>::isnaninf( norm_Y[0] ) );
if (success) {
if (pid==0)
std::cout << "End Result: TEST PASSED" << std::endl;
} else {
if (pid==0)
std::cout << "End Result: TEST FAILED" << std::endl;
}
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(verbose,std::cerr,success);
return success ? EXIT_SUCCESS : EXIT_FAILURE;
}
int main(int argc, char *argv[])
{
int *update; /* vector elements updated on this node. */
int *bindx;
double *val;
double *xguess, *b, *xexact;
int n_nonzeros;
int N_update; /* # of block unknowns updated on this node */
int numLocalEquations;
/* Number scalar equations on this node */
int numGlobalEquations; /* Total number of equations */
int *numNz, *ColInds;
int row, *col_inds, numEntries;
double *row_vals;
int i;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm comm;
#endif
cout << comm << endl;
if(argc != 2) cerr << "error: enter name of data file on command line" << endl;
/* Set exact solution to NULL */
xexact = NULL;
/* Read matrix file and distribute among processors.
Returns with this processor's set of rows */
Trilinos_Util_read_hb(argv[1], comm.MyPID(), &numGlobalEquations, &n_nonzeros,
&val, &bindx, &xguess, &b, &xexact);
Trilinos_Util_distrib_msr_matrix(comm, &numGlobalEquations, &n_nonzeros, &N_update,
&update, &val, &bindx, &xguess, &b, &xexact);
numLocalEquations = N_update;
/* Make numNzBlks - number of block entries in each block row */
numNz = new int[numLocalEquations];
for (i=0; i<numLocalEquations; i++) numNz[i] = bindx[i+1] - bindx[i] + 1;
/* Make ColInds - Exactly bindx, offset by diag (just copy pointer) */
ColInds = bindx+numLocalEquations+1;
Epetra_Map map(numGlobalEquations, numLocalEquations, update, 0, comm);
Epetra_CrsMatrix A(Copy, map, numNz);
/* Add rows one-at-a-time */
for (row=0; row<numLocalEquations; row++) {
row_vals = val + bindx[row];
col_inds = bindx + bindx[row];
numEntries = bindx[row+1] - bindx[row];
assert(A.InsertGlobalValues(update[row], numEntries, row_vals, col_inds)==0);
assert(A.InsertGlobalValues(update[row], 1, val+row, update+row)==0);
}
assert(A.FillComplete()==0);
Epetra_Vector xx(Copy, map, xexact);
Epetra_Vector bb(Copy, map, b);
// Construct a Petra Linear Problem
Epetra_Vector x(map);
Epetra_LinearProblem problem(&A, &x, &bb);
// Construct a solver object for this problem
AztecOO solver(problem);
// Assert symmetric
// problem->AssertSymmetric();
// Set Problem Difficulty Level
//problem->SetPDL(easy);
//solver.SetAztecOption(AZ_precond, AZ_none);
solver.SetAztecOption(AZ_precond, AZ_dom_decomp);
//solver.SetAztecOption(AZ_precond, AZ_ls);
//solver.SetAztecOption(AZ_scaling, 8);
solver.SetAztecOption(AZ_subdomain_solve, AZ_ilut);
//solver.SetAztecOption(AZ_subdomain_solve, AZ_bilu_ifp);
bool bilu = false;
//solver.SetAztecOption(AZ_output, 0);
//solver.SetAztecOption(AZ_graph_fill, 2);
solver.SetAztecOption(AZ_overlap, 0);
//solver.SetAztecOption(AZ_reorder, 0);
//solver.SetAztecOption(AZ_poly_ord, 9);
solver.SetAztecParam(AZ_ilut_fill, 1.0);
solver.SetAztecParam(AZ_drop, 0.0);
//double rthresh = 1.01;
//cout << "Rel threshold = " << rthresh << endl;
//solver.SetAztecParam(AZ_rthresh, rthresh);
//double athresh = 1.0e-2;
//cout << "Abs threshold = " << athresh << endl;
//solver.SetAztecParam(AZ_athresh, athresh);
//solver.SetAztecOption(AZ_/conv, AZ_noscaled);
//solver.SetAztecParam(AZ_ill_cond_thresh, 1.0e12);
int Niters = 400;
solver.SetAztecOption(AZ_kspace, Niters);
double norminf = A.NormInf();
double normone = A.NormOne();
if (comm.MyPID()==0)
cout << "\n Inf-norm of A before scaling = " << norminf
<< "\n One-norm of A before scaling = " << normone<< endl << endl;
if (bilu) {
int NumTrials = 3;
double athresholds[] = {0.0, 1.0E-14, 1.0E-3};
double rthresholds[] = {0.0, 1.0E-14, 1.0E-3};
double condestThreshold = 1.0E16;
double maxFill = 4.0;
int maxKspace = 4*Niters;
solver.SetAdaptiveParams(NumTrials, athresholds, rthresholds, condestThreshold, maxFill, maxKspace);
}
else {
int NumTrials = 7;
double athresholds[] = {0.0, 1.0E-12, 1.0E-12, 1.0E-5, 1.0E-5, 1.0E-2, 1.0E-2};
double rthresholds[] = {1.0, 1.0, 1.01, 1.0, 1.01, 1.01, 1.1 };
double condestThreshold = 1.0E16;
double maxFill = 4.0;
int maxKspace = 4*Niters;
solver.SetAdaptiveParams(NumTrials, athresholds, rthresholds, condestThreshold, maxFill, maxKspace);
}
Epetra_Time timer(comm);
solver.AdaptiveIterate(Niters, 1, 5.0e-7);
double atime = timer.ElapsedTime();
if (comm.MyPID()==0) cout << "AdaptiveIterate total time = " << atime << endl;
norminf = A.NormInf();
normone = A.NormOne();
if (comm.MyPID()==0)
cout << "\n Inf-norm of A after scaling = " << norminf
<< "\n One-norm of A after scaling = " << normone << endl << endl;
Epetra_Vector bcomp(map);
assert(A.Multiply(false, x, bcomp)==0);
Epetra_Vector resid(map);
assert(resid.Update(1.0, bb, -1.0, bcomp, 0.0)==0);
double residual;
assert(resid.Norm2(&residual)==0);
if (comm.MyPID()==0) cout << "Residual = " << residual << endl;
assert(resid.Update(1.0, xx, -1.0, x, 0.0)==0);
assert(resid.Norm2(&residual)==0);
if (comm.MyPID()==0)
cout << "2-norm of difference between computed and exact solution = " << residual << endl;
if (residual>1.0e-5) {
cout << "Difference between computed and exact solution is large..." << endl
<< "Computing norm of A times this difference. If this norm is small, then matrix is singular"
<< endl;
assert(A.Multiply(false, resid, bcomp)==0);
assert(bcomp.Norm2(&residual)==0);
if (comm.MyPID()==0)
cout << "2-norm of A times difference between computed and exact solution = " << residual << endl;
}
free ((void *) xguess);
free ((void *) b);
free ((void *) xexact);
free ((void *) val);
free ((void *) bindx);
free ((void *) update);
delete [] numNz;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return 0 ;
}
int main(int argc, char *argv[])
{
// Initialize MPI
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
#endif
// Create a communicator for Epetra objects
#ifdef HAVE_MPI
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
int * testInt = new int[100];
delete [] testInt;
bool verbose = false;
if (argc > 1)
if (argv[1][0]=='-' && argv[1][1]=='v')
verbose = true;
// Get the process ID and the total number of processors
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
// Set up theolver options parameter list
Teuchos::RCP<Teuchos::ParameterList> noxParamsPtr = Teuchos::rcp(new Teuchos::ParameterList);
Teuchos::ParameterList & noxParams = *(noxParamsPtr.get());
// Set up the printing utilities
// Only print output if the "-v" flag is set on the command line
Teuchos::ParameterList& printParams = noxParams.sublist("Printing");
printParams.set("MyPID", MyPID);
printParams.set("Output Precision", 5);
printParams.set("Output Processor", 0);
if( verbose )
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::Parameters +
NOX::Utils::Details +
NOX::Utils::Warning +
NOX::Utils::TestDetails);
else
printParams.set("Output Information", NOX::Utils::Error +
NOX::Utils::TestDetails);
Teuchos::RCP<NOX::Utils> printing = Teuchos::rcp( new NOX::Utils(printParams) );
// Identify the test problem
if (printing->isPrintType(NOX::Utils::TestDetails))
printing->out() << "Starting epetra/NOX_Operators/NOX_BroydenOp.exe" << endl;
// Identify processor information
#ifdef HAVE_MPI
if (printing->isPrintType(NOX::Utils::TestDetails))
{
printing->out() << "Parallel Run" << endl;
printing->out() << "Number of processors = " << NumProc << endl;
printing->out() << "Print Process = " << MyPID << endl;
}
Comm.Barrier();
if (printing->isPrintType(NOX::Utils::TestDetails))
printing->out() << "Process " << MyPID << " is alive!" << endl;
Comm.Barrier();
#else
if (printing->isPrintType(NOX::Utils::TestDetails))
printing->out() << "Serial Run" << endl;
#endif
int status = 0;
// Create a TestCompare class
NOX::Epetra::TestCompare tester( printing->out(), *printing);
double abstol = 1.e-4;
double reltol = 1.e-4 ;
// Test NOX::Epetra::BroydenOperator
int numGlobalElems = 3 * NumProc;
Epetra_Map broydenRowMap ( numGlobalElems, 0, Comm );
Epetra_Vector broydenWorkVec ( broydenRowMap );
Epetra_CrsGraph broydenWorkGraph( Copy, broydenRowMap, 0 );
vector<int> globalIndices(3);
for( int lcol = 0; lcol < 3; ++lcol )
globalIndices[lcol] = 3 * MyPID + lcol;
vector<int> myGlobalIndices(2);
// Row 1 structure
myGlobalIndices[0] = globalIndices[0];
myGlobalIndices[1] = globalIndices[2];
broydenWorkGraph.InsertGlobalIndices( globalIndices[0], 2, &myGlobalIndices[0] );
// Row 2 structure
myGlobalIndices[0] = globalIndices[0];
myGlobalIndices[1] = globalIndices[1];
broydenWorkGraph.InsertGlobalIndices( globalIndices[1], 2, &myGlobalIndices[0] );
// Row 3 structure
myGlobalIndices[0] = globalIndices[1];
myGlobalIndices[1] = globalIndices[2];
broydenWorkGraph.InsertGlobalIndices( globalIndices[2], 2, &myGlobalIndices[0] );
broydenWorkGraph.FillComplete();
Teuchos::RCP<Epetra_CrsMatrix> broydenWorkMatrix =
Teuchos::rcp( new Epetra_CrsMatrix( Copy, broydenWorkGraph ) );
// Create an identity matrix
broydenWorkVec.PutScalar(1.0);
broydenWorkMatrix->ReplaceDiagonalValues(broydenWorkVec);
NOX::Epetra::BroydenOperator broydenOp( noxParams, printing, broydenWorkVec, broydenWorkMatrix, true );
broydenWorkVec[0] = 1.0;
broydenWorkVec[1] = -1.0;
broydenWorkVec[2] = 2.0;
broydenOp.setStepVector( broydenWorkVec );
broydenWorkVec[0] = 2.0;
broydenWorkVec[1] = 1.0;
broydenWorkVec[2] = 3.0;
broydenOp.setYieldVector( broydenWorkVec );
broydenOp.computeSparseBroydenUpdate();
// Create the gold matrix for comparison
Teuchos::RCP<Epetra_CrsMatrix> goldMatrix = Teuchos::rcp( new Epetra_CrsMatrix( Copy, broydenWorkGraph ) );
int numCols ;
double * values ;
// Row 1 answers
goldMatrix->ExtractMyRowView( 0, numCols, values );
values[0] = 6.0 ;
values[1] = 2.0 ;
// Row 2 answers
goldMatrix->ExtractMyRowView( 1, numCols, values );
values[0] = 5.0 ;
values[1] = 0.0 ;
// Row 3 structure
goldMatrix->ExtractMyRowView( 2, numCols, values );
values[0] = -1.0 ;
values[1] = 7.0 ;
goldMatrix->Scale(0.2);
status += tester.testCrsMatrices( broydenOp.getBroydenMatrix(), *goldMatrix, reltol, abstol,
"Broyden Sparse Operator Update Test" );
// Now try a dense Broyden Update
Epetra_CrsGraph broydenWorkGraph2( Copy, broydenRowMap, 0 );
myGlobalIndices.resize(3);
// All Rowsstructure
myGlobalIndices[0] = globalIndices[0];
myGlobalIndices[1] = globalIndices[1];
myGlobalIndices[2] = globalIndices[2];
broydenWorkGraph2.InsertGlobalIndices( globalIndices[0], 3, &myGlobalIndices[0] );
broydenWorkGraph2.InsertGlobalIndices( globalIndices[1], 3, &myGlobalIndices[0] );
broydenWorkGraph2.InsertGlobalIndices( globalIndices[2], 3, &myGlobalIndices[0] );
broydenWorkGraph2.FillComplete();
Teuchos::RCP<Epetra_CrsMatrix> broydenWorkMatrix2 = Teuchos::rcp( new Epetra_CrsMatrix( Copy, broydenWorkGraph2 ) );
// Create an identity matrix
broydenWorkVec.PutScalar(1.0);
broydenWorkMatrix2->ReplaceDiagonalValues(broydenWorkVec);
NOX::Epetra::BroydenOperator broydenOp2( noxParams, printing, broydenWorkVec, broydenWorkMatrix2, true );
broydenWorkVec[0] = 1.0;
broydenWorkVec[1] = -1.0;
broydenWorkVec[2] = 2.0;
broydenOp2.setStepVector( broydenWorkVec );
broydenWorkVec[0] = 2.0;
broydenWorkVec[1] = 1.0;
broydenWorkVec[2] = 3.0;
broydenOp2.setYieldVector( broydenWorkVec );
broydenOp2.computeSparseBroydenUpdate();
// Create the gold matrix for comparison
Teuchos::RCP<Epetra_CrsMatrix> goldMatrix2 = Teuchos::rcp( new Epetra_CrsMatrix( Copy, broydenWorkGraph2 ) );
// Row 1 answers
goldMatrix2->ExtractMyRowView( 0, numCols, values );
values[0] = 7.0 ;
values[1] = -1.0 ;
values[2] = 2.0 ;
// Row 2 answers
goldMatrix2->ExtractMyRowView( 1, numCols, values );
values[0] = 2.0 ;
values[1] = 4.0 ;
values[2] = 4.0 ;
// Row 3 structure
goldMatrix2->ExtractMyRowView( 2, numCols, values );
values[0] = 1.0 ;
values[1] = -1.0 ;
values[2] = 8.0 ;
double scaleF = 1.0 / 6.0;
goldMatrix2->Scale( scaleF );
status += tester.testCrsMatrices( broydenOp2.getBroydenMatrix(), *goldMatrix2, reltol, abstol,
"Broyden Sparse Operator Update Test (Dense)" );
// Now test the ability to remove active entries in the Broyden update
Epetra_CrsGraph inactiveGraph( Copy, broydenRowMap, 0 );
// Row 1 structure
inactiveGraph.InsertGlobalIndices( globalIndices[0], 1, &myGlobalIndices[1] );
// Row 2 structure
inactiveGraph.InsertGlobalIndices( globalIndices[1], 1, &myGlobalIndices[2] );
// Row 3 structure
inactiveGraph.InsertGlobalIndices( globalIndices[2], 1, &myGlobalIndices[0] );
inactiveGraph.FillComplete();
// Inactivate entries in dense matrix to arrive again at the original sparse structure
broydenOp2.removeEntriesFromBroydenUpdate( inactiveGraph );
#ifdef HAVE_NOX_DEBUG
if( verbose )
broydenOp2.outputActiveEntries();
#endif
// Reset to the identity matrix
broydenOp2.resetBroydenMatrix( *broydenWorkMatrix2 );
// Step and Yield vectors are already set
broydenOp2.computeSparseBroydenUpdate();
status += tester.testCrsMatrices( broydenOp2.getBroydenMatrix(), *goldMatrix, reltol, abstol,
"Broyden Sparse Operator Update Test (Entry Removal)", false );
// Summarize test results
if( status == 0 )
printing->out() << "Test passed!" << endl;
else
printing->out() << "Test failed!" << endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
// Final return value (0 = successfull, non-zero = failure)
return status;
}
int main(int argc, char *argv[])
{
using std::cout;
using std::endl;
bool boolret;
int MyPID;
#ifdef HAVE_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
MyPID = Comm.MyPID();
bool testFailed = false;
bool verbose = false;
bool debug = false;
bool shortrun = false;
bool skinny = true;
bool useSA = false;
std::string which("SR");
int numElements = 100;
std::string prec("none");
int user_verbosity = 0;
int numicgs = 2;
bool success = true;
try {
CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SR or LR).");
cmdp.setOption("shortrun","longrun",&shortrun,"Allow only a small number of iterations.");
cmdp.setOption("skinny","hefty",&skinny,"Use a skinny (low-mem) or hefty (higher-mem) implementation of IRTR.");
cmdp.setOption("numElements",&numElements,"Number of elements in discretization.");
cmdp.setOption("prec",&prec,"Preconditioning type (""none"", ""olsen"", ""simple""");
cmdp.setOption("useSA","noSA",&useSA,"Use subspace acceleration.");
cmdp.setOption("verbosity",&user_verbosity,"Additional verbosity falgs.");
cmdp.setOption("numICGS",&numicgs,"Num ICGS iterations");
if (cmdp.parse(argc,argv) != CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
if (debug) verbose = true;
if (prec != "olsen" && prec != "simple") {
prec = "none";
}
typedef double ScalarType;
typedef ScalarTraits<ScalarType> SCT;
typedef SCT::magnitudeType MagnitudeType;
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVT;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OPT;
const ScalarType ONE = SCT::one();
if (verbose && MyPID == 0) {
cout << Anasazi::Anasazi_Version() << endl << endl;
}
// Problem information
int space_dim = 1;
std::vector<double> brick_dim( space_dim );
brick_dim[0] = 1.0;
std::vector<int> elements( space_dim );
elements[0] = numElements;
// Create problem
RCP<ModalProblem> testCase = rcp( new ModeLaplace1DQ1(Comm, brick_dim[0], elements[0]) );
//
// Get the stiffness and mass matrices
RCP<const Epetra_CrsMatrix> K = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getStiffness()), false );
RCP<const Epetra_CrsMatrix> M = rcp( const_cast<Epetra_CrsMatrix *>(testCase->getMass()), false );
RCP<const Epetra_CrsMatrix> P;
if (prec != "none") {
// construct a diagonal preconditioner
Epetra_Vector D(K->RowMap(),false);
K->ExtractDiagonalCopy(D);
for (int i=0; i<D.MyLength(); ++i) {
D[i] = 1.0 / D[i];
}
RCP<Epetra_CrsMatrix> ncP = rcp( new Epetra_CrsMatrix(Epetra_DataAccess::Copy,K->RowMap(),K->RowMap(),1,true) );
for (int i=0; i<D.MyLength(); ++i) {
ncP->InsertMyValues(i,1,&D[i],&i);
}
ncP->FillComplete(true);
P = ncP;
}
//
// Create the initial vectors
int blockSize = 5;
RCP<Epetra_MultiVector> ivec = rcp( new Epetra_MultiVector(K->OperatorDomainMap(), blockSize) );
ivec->Random();
// Create eigenproblem
const int nev = 5;
RCP<Anasazi::BasicEigenproblem<ScalarType,MV,OP> > problem =
rcp( new Anasazi::BasicEigenproblem<ScalarType,MV,OP>(K,M,ivec) );
//
// Inform the eigenproblem that the operator K is symmetric
problem->setHermitian(true);
//
// Set the number of eigenvalues requested
problem->setNEV( nev );
//
// Set the preconditioner, if using one. Also, indicates the type.
if (prec != "none") {
problem->setPrec(P);
}
//
// Inform the eigenproblem that you are done passing it information
boolret = problem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::SetProblem() returned with error." << endl
<< "End Result: TEST FAILED" << endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
verbosity |= user_verbosity;
// Eigensolver parameters
int maxIters;
if (shortrun) {
maxIters = 50;
}
else {
maxIters = 450;
}
MagnitudeType tol = 1.0e-6;
//
// Create parameter list to pass into the solver manager
ParameterList MyPL;
MyPL.set( "Skinny Solver", skinny);
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Maximum Iterations", maxIters );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Use SA", useSA );
MyPL.set( "Num ICGS", numicgs );
if (prec == "olsen") {
MyPL.set( "Olsen Prec", true );
}
else if (prec == "simple") {
MyPL.set( "Olsen Prec", false );
}
//
// Create the solver manager
Anasazi::RTRSolMgr<ScalarType,MV,OP> MySolverMan(problem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMan.solve();
testFailed = false;
if (returnCode != Anasazi::Converged) {
if (shortrun==false) {
testFailed = true;
}
}
/*
//
// Check that the parameters were all consumed
if (MyPL.getEntryPtr("Verbosity")->isUsed() == false ||
MyPL.getEntryPtr("Which")->isUsed() == false ||
MyPL.getEntryPtr("Skinny Solver")->isUsed() == false ||
MyPL.getEntryPtr("Block Size")->isUsed() == false ||
MyPL.getEntryPtr("Maximum Iterations")->isUsed() == false ||
MyPL.getEntryPtr("Convergence Tolerance")->isUsed() == false) {
if (verbose && MyPID==0) {
cout << "Failure! Unused parameters: " << endl;
MyPL.unused(cout);
}
}
*/
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ScalarType,MV> sol = problem->getSolution();
std::vector<Anasazi::Value<ScalarType> > evals = sol.Evals;
RCP<MV> evecs = sol.Evecs;
int numev = sol.numVecs;
if (numev > 0) {
std::ostringstream os;
os.setf(std::ios::scientific, std::ios::floatfield);
os.precision(6);
// Check the problem against the analytical solutions
if (verbose) {
double *revals = new double[numev];
for (int i=0; i<numev; i++) {
revals[i] = evals[i].realpart;
}
bool smallest = false;
if (which == "SR") {
smallest = true;
}
testCase->eigenCheck( *evecs, revals, 0, smallest );
delete [] revals;
}
// Compute the direct residual
std::vector<ScalarType> normV( numev );
SerialDenseMatrix<int,ScalarType> T(numev,numev);
for (int i=0; i<numev; i++) {
T(i,i) = evals[i].realpart;
}
RCP<MV> Mvecs = MVT::Clone( *evecs, numev ),
Kvecs = MVT::Clone( *evecs, numev );
OPT::Apply( *K, *evecs, *Kvecs );
OPT::Apply( *M, *evecs, *Mvecs );
MVT::MvTimesMatAddMv( -ONE, *Mvecs, T, ONE, *Kvecs );
// compute M-norm of residuals
OPT::Apply( *M, *Kvecs, *Mvecs );
MVT::MvDot( *Mvecs, *Kvecs, normV );
os << "Number of iterations performed in RTR_test.exe: " << MySolverMan.getNumIters() << endl
<< "Direct residual norms computed in RTR_test.exe" << endl
<< std::setw(20) << "Eigenvalue" << std::setw(20) << "Residual(M)" << endl
<< "----------------------------------------" << endl;
for (int i=0; i<numev; i++) {
if ( SCT::magnitude(evals[i].realpart) != SCT::zero() ) {
normV[i] = SCT::magnitude( SCT::squareroot( normV[i] ) / evals[i].realpart );
}
else {
normV[i] = SCT::magnitude( SCT::squareroot( normV[i] ) );
}
os << std::setw(20) << evals[i].realpart << std::setw(20) << normV[i] << endl;
if ( normV[i] > tol ) {
testFailed = true;
}
}
if (verbose && MyPID==0) {
cout << endl << os.str() << endl;
}
}
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(true,cout,success);
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
if (testFailed || success==false) {
if (verbose && MyPID==0) {
cout << "End Result: TEST FAILED" << endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
cout << "End Result: TEST PASSED" << endl;
}
return 0;
}
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc, &argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool verbose = (Comm.MyPID() == 0);
double TotalResidual = 0.0;
// Create the Map, defined as a grid, of size nx x ny x nz,
// subdivided into mx x my x mz cubes, each assigned to a
// different processor.
#if 0
ParameterList GaleriList;
GaleriList.set("nx", 4);
GaleriList.set("ny", 4);
GaleriList.set("nz", 4 * Comm.NumProc());
GaleriList.set("mx", 1);
GaleriList.set("my", 1);
GaleriList.set("mz", Comm.NumProc());
Epetra_Map* Map = CreateMap("Cartesian3D", Comm, GaleriList);
// Create a matrix, in this case corresponding to a 3D Laplacian
// discretized using a classical 7-point stencil. Please refer to
// the Galeri documentation for an overview of available matrices.
//
// NOTE: matrix must be symmetric if DSCPACK is used.
Epetra_CrsMatrix* Matrix = CreateCrsMatrix("Laplace3D", Map, GaleriList);
#else
bool transpose = false ;
bool distribute = false ;
bool symmetric ;
Epetra_CrsMatrix *Matrix = 0 ;
Epetra_Map *Map = 0 ;
CreateCrsMatrix( "ibm.triU", Comm, Map, transpose, distribute, &symmetric, Matrix ) ;
#endif
// build vectors, in this case with 1 vector
Epetra_MultiVector LHS(*Map, 1);
Epetra_MultiVector RHS(*Map, 1);
// create a linear problem object
Epetra_LinearProblem Problem(Matrix, &LHS, &RHS);
// use this list to set up parameters, now it is required
// to use all the available processes (if supported by the
// underlying solver). Uncomment the following two lines
// to let Amesos print out some timing and status information.
ParameterList List;
List.set("PrintTiming",true);
List.set("PrintStatus",true);
List.set("MaxProcs",Comm.NumProc());
std::vector<std::string> SolverType;
SolverType.push_back("Amesos_Paraklete");
Epetra_Time Time(Comm);
// this is the Amesos factory object that will create
// a specific Amesos solver.
Amesos Factory;
// Cycle over all solvers.
// Only installed solvers will be tested.
for (unsigned int i = 0 ; i < SolverType.size() ; ++i)
{
// Check whether the solver is available or not
if (Factory.Query(SolverType[i]))
{
// 1.- set exact solution (constant vector)
LHS.PutScalar(1.0);
// 2.- create corresponding rhs
Matrix->Multiply(false, LHS, RHS);
// 3.- randomize solution vector
LHS.Random();
// 4.- create the amesos solver object
Amesos_BaseSolver* Solver = Factory.Create(SolverType[i], Problem);
assert (Solver != 0);
Solver->SetParameters(List);
// 5.- factorize and solve
Time.ResetStartTime();
AMESOS_CHK_ERR(Solver->SymbolicFactorization());
// 7.- delete the object
delete Solver;
}
}
delete Matrix;
delete Map;
if (TotalResidual > 1e-9)
exit(EXIT_FAILURE);
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
} // end of main()
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// Creates the linear problem using the Galeri package.
// Several matrix examples are supported; please refer to the
// Galeri documentation for more details.
// Most of the examples using the ML_Epetra::MultiLevelPreconditioner
// class are based on Epetra_CrsMatrix. Example
// `ml_EpetraVbr.cpp' shows how to define a Epetra_VbrMatrix.
// `Laplace2D' is a symmetric matrix; an example of non-symmetric
// matrices is `Recirc2D' (advection-diffusion in a box, with
// recirculating flow). The grid has nx x ny nodes, divided into
// mx x my subdomains, each assigned to a different processor.
int nx;
if (argc > 1)
nx = (int) strtol(argv[1],NULL,10);
else
nx = 8;
int ny = nx * Comm.NumProc(); // each subdomain is a square
ParameterList GaleriList;
GaleriList.set("nx", nx);
GaleriList.set("ny", ny);
GaleriList.set("mx", 1);
GaleriList.set("my", Comm.NumProc());
Epetra_Map* Map = CreateMap("Cartesian2D", Comm, GaleriList);
Epetra_CrsMatrix* A = CreateCrsMatrix("Laplace2D", Map, GaleriList);
// Build a linear system with trivial solution, using a random vector
// as starting solution.
Epetra_Vector LHS(*Map); LHS.Random();
Epetra_Vector RHS(*Map); RHS.PutScalar(0.0);
Epetra_LinearProblem Problem(A, &LHS, &RHS);
// As we wish to use AztecOO, we need to construct a solver object
// for this problem
AztecOO solver(Problem);
// =========================== begin of ML part ===========================
// create a parameter list for ML options
ParameterList MLList;
// Sets default parameters for classic smoothed aggregation. After this
// call, MLList contains the default values for the ML parameters,
// as required by typical smoothed aggregation for symmetric systems.
// Other sets of parameters are available for non-symmetric systems
// ("DD" and "DD-ML"), and for the Maxwell equations ("maxwell").
ML_Epetra::SetDefaults("SA",MLList);
// overwrite some parameters. Please refer to the user's guide
// for more information
// some of the parameters do not differ from their default value,
// and they are here reported for the sake of clarity
// output level, 0 being silent and 10 verbose
MLList.set("ML output", 10);
// maximum number of levels
MLList.set("max levels",5);
// set finest level to 0
MLList.set("increasing or decreasing","increasing");
// use Uncoupled scheme to create the aggregate
MLList.set("aggregation: type", "Uncoupled");
// smoother is symmetric Gauss-Seidel. Example file
// `ml/examples/TwoLevelDD/ml_2level_DD.cpp' shows how to use
// AZTEC's preconditioners as smoothers
//MLList.set("smoother: type","symmetric Gauss-Seidel");
// use both pre and post smoothing
MLList.set("smoother: pre or post", "both");
MLList.set("smoother: type", "user-defined");
//The function pointer to the user-defined smoother.
MLList.set("smoother: user-defined function",userSmoother);
//(optional) The label for the user-defined smoother.
MLList.set("smoother: user-defined name","myJacobi");
//Use this smoother on the finest level, 0.
MLList.set("smoother: type (level 0)", "user-defined");
MLList.set("smoother: sweeps (level 0)", 1);
//Use symmetric Gauss-Seidel on all subsequent levels.
MLList.set("smoother: type (level 1)", "symmetric Gauss-Seidel");
//user-defined smoothing for a 1-level method only
//to use this, uncomment the next 3 lines and comment out the direct-solver
//settings further below
//MLList.set("coarse: type", "user-defined");
//MLList.set("coarse: user-defined function",userSmoother);
//MLList.set("coarse: user-defined name","myJacobi");
#ifdef HAVE_ML_AMESOS
// solve with serial direct solver KLU
MLList.set("coarse: type","Amesos-KLU");
#else
// this is for testing purposes only, you should have
// a direct solver for the coarse problem (either Amesos, or the SuperLU/
// SuperLU_DIST interface of ML)
MLList.set("aggregation: type", "MIS");
MLList.set("smoother: type","Jacobi");
MLList.set("coarse: type","Jacobi");
#endif
MLList.set("read XML", false); // skip XML file in this directory
// Creates the preconditioning object. We suggest to use `new' and
// `delete' because the destructor contains some calls to MPI (as
// required by ML and possibly Amesos). This is an issue only if the
// destructor is called **after** MPI_Finalize().
ML_Epetra::MultiLevelPreconditioner* MLPrec =
new ML_Epetra::MultiLevelPreconditioner(*A, MLList);
// verify unused parameters on process 0 (put -1 to print on all
// processes)
MLPrec->PrintUnused(0);
// =========================== end of ML part =============================
// tell AztecOO to use the ML preconditioner, specify the solver
// and the output, then solve with 500 maximum iterations and 1e-12
// of tolerance (see AztecOO's user guide for more details)
solver.SetPrecOperator(MLPrec);
solver.SetAztecOption(AZ_solver, AZ_gmres);
solver.SetAztecOption(AZ_output, 32);
solver.Iterate(500, 1e-12);
// destroy the preconditioner
delete MLPrec;
// compute the real residual
double residual;
LHS.Norm2(&residual);
if( Comm.MyPID()==0 ) {
cout << "||b-Ax||_2 = " << residual << endl;
}
// for testing purposes
if (residual > 1e-5)
exit(EXIT_FAILURE);
delete A;
delete Map;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
} //main
int main(int argc, char *argv[]) {
int ierr = 0, i;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
int rank; // My process ID
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
int rank = 0;
Epetra_SerialComm Comm;
#endif
#ifdef HAVE_EPETRA_TEUCHOS
Teuchos::RCP<Teuchos::FancyOStream>
fancyOut = Teuchos::VerboseObjectBase::getDefaultOStream();
if (Comm.NumProc() > 1 ) {
fancyOut->setShowProcRank(true);
fancyOut->setOutputToRootOnly(-1);
}
std::ostream &out = *fancyOut;
#else
std::ostream &out = std::cout;
#endif
Comm.SetTracebackMode(0); // This should shut down any error tracing
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
// char tmp;
// if (rank==0) out << "Press any key to continue..."<< endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if (verbose && MyPID==0)
out << Epetra_Version() << endl << endl;
if (verbose) out << Comm <<endl;
bool verbose1 = verbose;
// Redefine verbose to only print on PE 0
if (verbose && rank!=0) verbose = false;
int NumMyElements = 10000;
int NumMyElements1 = NumMyElements; // Needed for localmap
int NumGlobalElements = NumMyElements*NumProc+EPETRA_MIN(NumProc,3);
if (MyPID < 3) NumMyElements++;
int IndexBase = 0;
int ElementSize = 7;
// Test LocalMap constructor
// and Petra-defined uniform linear distribution constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_LocalMap(NumMyElements1, IndexBase, Comm)" << endl;
if (verbose) out << " and Epetra_BlockMap(NumGlobalElements, ElementSize, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
Epetra_LocalMap *LocalMap = new Epetra_LocalMap(NumMyElements1, IndexBase,
Comm);
Epetra_BlockMap * BlockMap = new Epetra_BlockMap(NumGlobalElements, ElementSize, IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*BlockMap, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*BlockMap, *LocalMap, verbose),ierr);
delete BlockMap;
// Test User-defined linear distribution constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_BlockMap(NumGlobalElements, NumMyElements, ElementSize, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
BlockMap = new Epetra_BlockMap(NumGlobalElements, NumMyElements, ElementSize, IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*BlockMap, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*BlockMap, *LocalMap, verbose),ierr);
delete BlockMap;
// Test User-defined arbitrary distribution constructor
// Generate Global Element List. Do in reverse for fun!
int * MyGlobalElements = new int[NumMyElements];
int MaxMyGID = (Comm.MyPID()+1)*NumMyElements-1+IndexBase;
if (Comm.MyPID()>2) MaxMyGID+=3;
for (i = 0; i<NumMyElements; i++) MyGlobalElements[i] = MaxMyGID-i;
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_BlockMap(NumGlobalElements, NumMyElements, MyGlobalElements, ElementSize, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
BlockMap = new Epetra_BlockMap(NumGlobalElements, NumMyElements, MyGlobalElements, ElementSize,
IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*BlockMap, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*BlockMap, *LocalMap, verbose),ierr);
delete BlockMap;
int * ElementSizeList = new int[NumMyElements];
int NumMyEquations = 0;
int NumGlobalEquations = 0;
for (i = 0; i<NumMyElements; i++)
{
ElementSizeList[i] = i%6+2; // blocksizes go from 2 to 7
NumMyEquations += ElementSizeList[i];
}
ElementSize = 7; // Set to maximum for use in checkmap
NumGlobalEquations = Comm.NumProc()*NumMyEquations;
// Adjust NumGlobalEquations based on processor ID
if (Comm.NumProc() > 3)
{
if (Comm.MyPID()>2)
NumGlobalEquations += 3*((NumMyElements)%6+2);
else
NumGlobalEquations -= (Comm.NumProc()-3)*((NumMyElements-1)%6+2);
}
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_BlockMap(NumGlobalElements, NumMyElements, MyGlobalElements, ElementSizeList, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
BlockMap = new Epetra_BlockMap(NumGlobalElements, NumMyElements, MyGlobalElements, ElementSizeList,
IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*BlockMap, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*BlockMap, *LocalMap, verbose),ierr);
// Test Copy constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_BlockMap(*BlockMap)" << endl;
if (verbose) out << "*********************************************************" << endl;
Epetra_BlockMap * BlockMap1 = new Epetra_BlockMap(*BlockMap);
EPETRA_TEST_ERR(VectorTests(*BlockMap, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*BlockMap, *LocalMap, verbose),ierr);
delete [] ElementSizeList;
delete [] MyGlobalElements;
delete BlockMap;
delete BlockMap1;
// Test Petra-defined uniform linear distribution constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_Map(NumGlobalElements, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
Epetra_Map * Map = new Epetra_Map(NumGlobalElements, IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*Map, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*Map, *LocalMap, verbose),ierr);
delete Map;
// Test User-defined linear distribution constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_Map(NumGlobalElements, NumMyElements, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
Map = new Epetra_Map(NumGlobalElements, NumMyElements, IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*Map, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*Map, *LocalMap, verbose),ierr);
delete Map;
// Test User-defined arbitrary distribution constructor
// Generate Global Element List. Do in reverse for fun!
MyGlobalElements = new int[NumMyElements];
MaxMyGID = (Comm.MyPID()+1)*NumMyElements-1+IndexBase;
if (Comm.MyPID()>2) MaxMyGID+=3;
for (i = 0; i<NumMyElements; i++) MyGlobalElements[i] = MaxMyGID-i;
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements, IndexBase, Comm)" << endl;
if (verbose) out << "*********************************************************" << endl;
Map = new Epetra_Map(NumGlobalElements, NumMyElements, MyGlobalElements,
IndexBase, Comm);
EPETRA_TEST_ERR(VectorTests(*Map, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*Map, *LocalMap, verbose),ierr);
// Test Copy constructor
if (verbose) out << "\n*********************************************************" << endl;
if (verbose) out << "Checking Epetra_Map(*Map)" << endl;
if (verbose) out << "*********************************************************" << endl;
Epetra_Map Map1(*Map);
EPETRA_TEST_ERR(VectorTests(*Map, verbose),ierr);
EPETRA_TEST_ERR(MatrixTests(*Map, *LocalMap, verbose),ierr);
delete [] MyGlobalElements;
delete Map;
if (verbose1)
{
// Test Vector MFLOPS for 2D Dot Product
int M = 1;
int K = 1000000;
Epetra_Map Map2(-1, K, IndexBase, Comm);
Epetra_LocalMap Map3(M, IndexBase, Comm);
Epetra_Vector A(Map2);A.Random();
Epetra_Vector B(Map2);B.Random();
Epetra_Vector C(Map3);C.Random();
// Test Epetra_Vector label
const char* VecLabel = A.Label();
const char* VecLabel1 = "Epetra::Vector";
if (verbose) out << endl << endl <<"This should say " << VecLabel1 << ": " << VecLabel << endl << endl << endl;
EPETRA_TEST_ERR(strcmp(VecLabel1,VecLabel),ierr);
if (verbose) out << "Testing Assignment operator" << endl;
double tmp1 = 1.00001* (double) (MyPID+1);
double tmp2 = tmp1;
A[1] = tmp1;
tmp2 = A[1];
out << "On PE "<< MyPID << " A[1] should equal = " << tmp1;
if (tmp1==tmp2) out << " and it does!" << endl;
else out << " but it equals " << tmp2;
Comm.Barrier();
if (verbose) out << endl << endl << "Testing MFLOPs" << endl;
Epetra_Flops counter;
C.SetFlopCounter(counter);
Epetra_Time mytimer(Comm);
C.Multiply('T', 'N', 0.5, A, B, 0.0);
double Multiply_time = mytimer.ElapsedTime();
double Multiply_flops = C.Flops();
if (verbose) out << "\n\nTotal FLOPs = " << Multiply_flops << endl;
if (verbose) out << "Total Time = " << Multiply_time << endl;
if (verbose) out << "MFLOPs = " << Multiply_flops/Multiply_time/1000000.0 << endl;
Comm.Barrier();
// Test Vector ostream operator with Petra-defined uniform linear distribution constructor
// and a small vector
Epetra_Map Map4(100, IndexBase, Comm);
double * Dp = new double[100];
for (i=0; i<100; i++)
Dp[i] = i;
Epetra_Vector D(View, Map4,Dp);
if (verbose) out << "\n\nTesting ostream operator: Multivector should be 100-by-2 and print i,j indices"
<< endl << endl;
out << D << endl;
if (verbose) out << "Traceback Mode value = " << D.GetTracebackMode() << endl;
delete [] Dp;
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return ierr;
}
int main(int argc, char *argv[])
{
int ierr = 0;
double nonlinear_factor = 1.0;
double left_bc = 0.0;
double right_bc = 0.40;
// Initialize MPI
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
#endif
// Create a communicator for Epetra objects
#ifdef HAVE_MPI
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
// Get the process ID and the total number of processors
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
// Get the number of elements from the command line
int NumGlobalElements = 100 + 1;
// The number of unknowns must be at least equal to the
// number of processors.
if (NumGlobalElements < NumProc) {
std::cout << "numGlobalBlocks = " << NumGlobalElements
<< " cannot be < number of processors = " << NumProc << std::endl;
exit(1);
}
// Create the FiniteElementProblem class. This creates all required
// Epetra objects for the problem and allows calls to the
// function (RHS) and Jacobian evaluation routines.
FiniteElementProblem Problem(NumGlobalElements, Comm);
// Get the vector from the Problem
Epetra_Vector& soln = Problem.getSolution();
// Initialize Solution
soln.PutScalar(0.1);
// Create initial guess for the null vector of jacobian
Teuchos::RCP<NOX::Abstract::Vector> solnTwo =
Teuchos::rcp(new NOX::Epetra::Vector(soln));
solnTwo->init(2.5); // initial value 1.0
// Begin LOCA Solver ************************************
// Create parameter list
Teuchos::RCP<Teuchos::ParameterList> paramList =
Teuchos::rcp(new Teuchos::ParameterList);
// Create LOCA sublist
Teuchos::ParameterList& locaParamsList = paramList->sublist("LOCA");
// Create the stepper sublist and set the stepper parameters
Teuchos::ParameterList& locaStepperList = locaParamsList.sublist("Stepper");
locaStepperList.set("Continuation Method", "Natural");
//locaStepperList.set("Bordered Solver Method", "Nested");
//locaStepperList.set("Bordered Solver Method", "Householder");
locaStepperList.set("Continuation Parameter", "Nonlinear Factor");
locaStepperList.set("Initial Value", nonlinear_factor);
locaStepperList.set("Max Value", 1.6);
locaStepperList.set("Min Value", 0.00);
locaStepperList.set("Max Steps", 20);
locaStepperList.set("Max Nonlinear Iterations", 15);
// Create bifurcation sublist
Teuchos::ParameterList& bifurcationList =
locaParamsList.sublist("Bifurcation");
bifurcationList.set("Type", "Phase Transition");
bifurcationList.set("Bifurcation Parameter", "Right BC");
bifurcationList.set("Second Solution Vector", solnTwo);
// Create predictor sublist
Teuchos::ParameterList& predictorList = locaParamsList.sublist("Predictor");
predictorList.set("Method", "Secant");
// Create step size sublist
Teuchos::ParameterList& stepSizeList = locaParamsList.sublist("Step Size");
stepSizeList.set("Method", "Constant");
stepSizeList.set("Initial Step Size", 0.1);
stepSizeList.set("Min Step Size", 1.0e-3);
stepSizeList.set("Max Step Size", 2000.0);
stepSizeList.set("Aggressiveness", 0.1);
// Create the "Solver" parameters sublist to be used with NOX Solvers
Teuchos::ParameterList& nlParams = paramList->sublist("NOX");
// Create the NOX printing parameter list
Teuchos::ParameterList& nlPrintParams = nlParams.sublist("Printing");
nlPrintParams.set("MyPID", MyPID);
nlPrintParams.set("Output Precision", 6);
nlPrintParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::Details +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Warning +
NOX::Utils::StepperIteration +
NOX::Utils::StepperDetails +
NOX::Utils::StepperParameters);
// Create the "Linear Solver" sublist for Newton's method
Teuchos::ParameterList& dirParams = nlParams.sublist("Direction");
Teuchos::ParameterList& newParams = dirParams.sublist("Newton");
Teuchos::ParameterList& lsParams = newParams.sublist("Linear Solver");
lsParams.set("Aztec Solver", "GMRES");
lsParams.set("Max Iterations", 200);
lsParams.set("Tolerance", 1e-6);
lsParams.set("Output Frequency", 50);
//lsParams.set("Scaling", "None");
//lsParams.set("Scaling", "Row Sum");
lsParams.set("Compute Scaling Manually", false);
lsParams.set("Preconditioner", "Ifpack");
lsParams.set("Ifpack Preconditioner", "ILU");
//lsParams.set("Preconditioner", "New Ifpack");
//Teuchos::ParameterList& ifpackParams = lsParams.sublist("Ifpack");
//ifpackParams.set("fact: level-of-fill", 1);
// Create and initialize the parameter vector
LOCA::ParameterVector pVector;
pVector.addParameter("Nonlinear Factor",nonlinear_factor);
pVector.addParameter("Left BC", left_bc);
pVector.addParameter("Right BC", right_bc);
// Create the interface between the test problem and the nonlinear solver
// This is created by the user using inheritance of the abstract base class:
Teuchos::RCP<Problem_Interface> interface =
Teuchos::rcp(new Problem_Interface(Problem));
Teuchos::RCP<LOCA::Epetra::Interface::TimeDependent> iReq = interface;
Teuchos::RCP<NOX::Epetra::Interface::Jacobian> iJac = interface;
// Create the Epetra_RowMatrixfor the Jacobian/Preconditioner
Teuchos::RCP<Epetra_RowMatrix> Amat =
Teuchos::rcp(&Problem.getJacobian(),false);
// Create scaling object
Teuchos::RCP<NOX::Epetra::Scaling> scaling = Teuchos::null;
// scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> scalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// //scaling->addRowSumScaling(NOX::Epetra::Scaling::Left, scalingVector);
// scaling->addColSumScaling(NOX::Epetra::Scaling::Right, scalingVector);
// Create transpose scaling object
Teuchos::RCP<NOX::Epetra::Scaling> trans_scaling = Teuchos::null;
// trans_scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> transScalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// trans_scaling->addRowSumScaling(NOX::Epetra::Scaling::Right,
// transScalingVector);
// trans_scaling->addColSumScaling(NOX::Epetra::Scaling::Left,
// transScalingVector);
//bifurcationList.set("Transpose Scaling", trans_scaling);
// Create the linear systems
Teuchos::RCP<NOX::Epetra::LinearSystemAztecOO> linsys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(nlPrintParams, lsParams,
iReq, iJac, Amat, soln,
scaling));
// Create the loca vector
NOX::Epetra::Vector locaSoln(soln);
// Create Epetra factory
Teuchos::RCP<LOCA::Abstract::Factory> epetraFactory =
Teuchos::rcp(new LOCA::Epetra::Factory);
// Create global data object
Teuchos::RCP<LOCA::GlobalData> globalData =
LOCA::createGlobalData(paramList, epetraFactory);
// Create the Group
Teuchos::RCP<LOCA::Epetra::Group> grp =
Teuchos::rcp(new LOCA::Epetra::Group(globalData, nlPrintParams, iReq,
locaSoln, linsys, linsys,
pVector));
// Inject FreeEnergy interface into the group
Teuchos::RCP<LOCA::Epetra::Interface::FreeEnergy> iFE = interface;
grp->setFreeEnergyInterface(iFE);
grp->computeF();
// Create the Solver convergence test
//NOX::StatusTest::NormWRMS wrms(1.0e-2, 1.0e-8);
Teuchos::RCP<NOX::StatusTest::NormF> wrms =
Teuchos::rcp(new NOX::StatusTest::NormF(1.0e-12));
Teuchos::RCP<NOX::StatusTest::MaxIters> maxiters =
Teuchos::rcp(new NOX::StatusTest::MaxIters(locaStepperList.get("Max Nonlinear Iterations", 10)));
Teuchos::RCP<NOX::StatusTest::Combo> combo =
Teuchos::rcp(new NOX::StatusTest::Combo(NOX::StatusTest::Combo::OR));
combo->addStatusTest(wrms);
combo->addStatusTest(maxiters);
// Create the stepper
LOCA::Stepper stepper(globalData, grp, combo, paramList);
LOCA::Abstract::Iterator::IteratorStatus status = stepper.run();
if (status == LOCA::Abstract::Iterator::Finished)
globalData->locaUtils->out() << "All tests passed" << std::endl;
else {
if (globalData->locaUtils->isPrintType(NOX::Utils::Error))
globalData->locaUtils->out()
<< "Stepper failed to converge!" << std::endl;
}
// Output the parameter list
if (globalData->locaUtils->isPrintType(NOX::Utils::StepperParameters)) {
globalData->locaUtils->out()
<< std::endl << "Final Parameters" << std::endl
<< "****************" << std::endl;
stepper.getList()->print(globalData->locaUtils->out());
globalData->locaUtils->out() << std::endl;
}
LOCA::destroyGlobalData(globalData);
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
int main (int argc, char **argv)
{
#ifdef HAVE_MPI
MPI_Init(&argc, &argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// define some Epetra objects
int n = Comm.NumProc() * 4;
Epetra_Map Map(n, 0, Comm);
Epetra_MultiVector x(Map, 2); x.Random();
Epetra_MultiVector b(Map, 2); x.Random();
Epetra_CrsMatrix Matrix(Copy, Map, 0);
// diagonal matrix
for (int i = 0; i < Map.NumMyElements(); ++i)
{
int ii = Map.GID(i);
double one = 1.0;
Matrix.InsertGlobalValues(ii, 1, &one, &ii);
}
Matrix.FillComplete();
Teuchos::ParameterList List;
List.set("int parameter", 10);
List.set("double parameter", 10.0);
List.set("std::string parameter", "std::string");
// ========================= //
// Part I: generate XML file //
// ========================= //
EpetraExt::XMLWriter XMLWriter(Comm, "data.xml");
std::vector<std::string> Content;
Content.push_back("This is an example of description");
Content.push_back("The description is as long as desired,");
Content.push_back("just put it in a std::vector of strings.");
XMLWriter.Create("MyProblem");
XMLWriter.Write("Author", "myself and others");
XMLWriter.Write("Date", "May 2006");
XMLWriter.Write("MyMap", Map);
XMLWriter.Write("MyMatrix", Matrix);
XMLWriter.Write("MyLHS", x);
XMLWriter.Write("MyRHS", b);
XMLWriter.Write("MyContent", Content);
XMLWriter.Write("MyParameters", List);
XMLWriter.Close();
// ================== //
// Part II: read data //
// ================== //
EpetraExt::XMLReader XMLReader(Comm, "data.xml");
Epetra_Map* MyMap;
Epetra_CrsMatrix* MyMatrix;
Epetra_MultiVector* MyLHS;
Epetra_MultiVector* MyRHS;
Teuchos::ParameterList MyParameters;
std::vector<std::string> Author;
std::vector<std::string> Date;
std::vector<std::string> MyContent;
XMLReader.Read("Author", Author);
XMLReader.Read("Date", Date);
XMLReader.Read("MyMap", MyMap);
XMLReader.Read("MyMatrix", MyMatrix);
XMLReader.Read("MyLHS", MyLHS);
XMLReader.Read("MyRHS", MyRHS);
XMLReader.Read("MyContent", MyContent);
XMLReader.Read("MyParameters", MyParameters);
std::cout << *MyMap;
std::cout << *MyMatrix;
std::cout << *MyLHS;
std::cout << *MyRHS;
if (Comm.MyPID() == 0)
{
int Msize = (int) MyContent.size();
for (int i = 0; i < Msize; ++i)
std::cout << MyContent[i] << std::endl;
std::cout << MyParameters;
std::cout << "Author = " << Author[0] << std::endl;
std::cout << "Date = " << Date[0] << std::endl;
}
delete MyMap;
delete MyMatrix;
delete MyLHS;
delete MyRHS;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return(EXIT_SUCCESS);
}
