void
Albany::ResponseFactory::
createResponseFunction(
const std::string& name,
Teuchos::ParameterList& responseParams,
Teuchos::Array< Teuchos::RCP<AbstractResponseFunction> >& responses) const
{
using std::string;
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::ParameterList;
using Teuchos::Array;
RCP<const Epetra_Comm> comm = app->getComm();
if (name == "Solution Average") {
responses.push_back(rcp(new Albany::SolutionAverageResponseFunction(comm)));
}
else if (name == "Solution Two Norm") {
responses.push_back(rcp(new Albany::SolutionTwoNormResponseFunction(comm)));
}
else if (name == "Solution Values") {
responses.push_back(rcp(new Albany::SolutionValuesResponseFunction(app, responseParams)));
}
else if (name == "Solution Max Value") {
int eq = responseParams.get("Equation", 0);
int neq = responseParams.get("Num Equations", 1);
bool inor = responseParams.get("Interleaved Ordering", true);
responses.push_back(
rcp(new Albany::SolutionMaxValueResponseFunction(comm, neq, eq, inor)));
}
else if (name == "Solution Two Norm File") {
responses.push_back(
rcp(new Albany::SolutionFileResponseFunction<Albany::NormTwo>(comm)));
}
else if (name == "Solution Inf Norm File") {
responses.push_back(
rcp(new Albany::SolutionFileResponseFunction<Albany::NormInf>(comm)));
}
else if (name == "Aggregated") {
int num_aggregate_responses = responseParams.get<int>("Number");
Array< RCP<AbstractResponseFunction> > aggregated_responses;
Array< RCP<ScalarResponseFunction> > scalar_responses;
for (int i=0; i<num_aggregate_responses; i++) {
std::string id = Albany::strint("Response",i);
std::string name = responseParams.get<std::string>(id);
std::string sublist_name = Albany::strint("ResponseParams",i);
createResponseFunction(name, responseParams.sublist(sublist_name),
aggregated_responses);
}
scalar_responses.resize(aggregated_responses.size());
for (int i=0; i<aggregated_responses.size(); i++) {
TEUCHOS_TEST_FOR_EXCEPTION(
aggregated_responses[i]->isScalarResponse() != true, std::logic_error,
"Response function " << i << " is not a scalar response function." <<
std::endl <<
"The aggregated response can only aggregate scalar response " <<
"functions!");
scalar_responses[i] =
Teuchos::rcp_dynamic_cast<ScalarResponseFunction>(
aggregated_responses[i]);
}
responses.push_back(
rcp(new Albany::AggregateScalarResponseFunction(comm, scalar_responses)));
}
else if (name == "Field Integral" ||
name == "Field Value" ||
name == "Field Average" ||
name == "Surface Velocity Mismatch" ||
name == "Center Of Mass" ||
name == "Save Field" ||
name == "Region Boundary" ||
name == "Element Size Field" ||
name == "IP to Nodal Field" ||
name == "Save Nodal Fields" ||
name == "PHAL Field Integral") {
responseParams.set("Name", name);
for (int i=0; i<meshSpecs.size(); i++) {
responses.push_back(
rcp(new Albany::FieldManagerScalarResponseFunction(
app, prob, meshSpecs[i], stateMgr, responseParams)));
}
}
else if (name == "Solution") {
responses.push_back(
rcp(new Albany::SolutionResponseFunction(app, responseParams)));
}
else if (name == "KL") {
Array< RCP<AbstractResponseFunction> > base_responses;
std::string name = responseParams.get<std::string>("Response");
createResponseFunction(name, responseParams.sublist("ResponseParams"),
base_responses);
for (int i=0; i<base_responses.size(); i++)
responses.push_back(
rcp(new Albany::KLResponseFunction(base_responses[i], responseParams)));
}
#ifdef ALBANY_QCAD
else if (name == "Saddle Value") {
responseParams.set("Name", name);
for (int i=0; i<meshSpecs.size(); i++) {
responses.push_back(
rcp(new QCAD::SaddleValueResponseFunction(
app, prob, meshSpecs[i], stateMgr, responseParams)));
}
}
#endif
else {
TEUCHOS_TEST_FOR_EXCEPTION(
true, Teuchos::Exceptions::InvalidParameter,
std::endl << "Error! Unknown response function " << name <<
"!" << std::endl << "Supplied parameter list is " <<
std::endl << responseParams);
}
}
TEUCHOS_UNIT_TEST(gather_orientation, gather_constr)
{
const std::size_t workset_size = 4;
const std::string fieldName_q1 = "U";
const std::string fieldName_qedge1 = "V";
Teuchos::RCP<panzer_stk_classic::STK_Interface> mesh = buildMesh(2,2);
// build input physics block
Teuchos::RCP<panzer::PureBasis> basis_q1 = buildBasis(workset_size,"Q1");
Teuchos::RCP<panzer::PureBasis> basis_qedge1 = buildBasis(workset_size,"QEdge1");
Teuchos::RCP<Teuchos::ParameterList> ipb = Teuchos::parameterList();
testInitialization(ipb);
const int default_int_order = 1;
std::string eBlockID = "eblock-0_0";
Teuchos::RCP<user_app::MyFactory> eqset_factory = Teuchos::rcp(new user_app::MyFactory);
panzer::CellData cellData(workset_size,mesh->getCellTopology("eblock-0_0"));
Teuchos::RCP<panzer::GlobalData> gd = panzer::createGlobalData();
Teuchos::RCP<panzer::PhysicsBlock> physicsBlock =
Teuchos::rcp(new PhysicsBlock(ipb,eBlockID,default_int_order,cellData,eqset_factory,gd,false));
Teuchos::RCP<std::vector<panzer::Workset> > work_sets = panzer_stk_classic::buildWorksets(*mesh,*physicsBlock);
TEST_EQUALITY(work_sets->size(),1);
// build connection manager and field manager
const Teuchos::RCP<panzer::ConnManager<int,int> > conn_manager = Teuchos::rcp(new panzer_stk_classic::STKConnManager<int>(mesh));
RCP<panzer::DOFManager<int,int> > dofManager = Teuchos::rcp(new panzer::DOFManager<int,int>(conn_manager,MPI_COMM_WORLD));
dofManager->addField(fieldName_q1,Teuchos::rcp(new panzer::Intrepid2FieldPattern(basis_q1->getIntrepid2Basis())));
dofManager->addField(fieldName_qedge1,Teuchos::rcp(new panzer::Intrepid2FieldPattern(basis_qedge1->getIntrepid2Basis())));
dofManager->setOrientationsRequired(true);
dofManager->buildGlobalUnknowns();
// setup field manager, add evaluator under test
/////////////////////////////////////////////////////////////
PHX::FieldManager<panzer::Traits> fm;
Teuchos::RCP<PHX::FieldTag> evalField_q1, evalField_qedge1;
{
Teuchos::RCP<std::vector<std::string> > dofNames = Teuchos::rcp(new std::vector<std::string>);
dofNames->push_back(fieldName_q1);
Teuchos::ParameterList pl;
pl.set("Indexer Names",dofNames);
pl.set("DOF Names",dofNames);
pl.set("Basis",basis_q1);
Teuchos::RCP<PHX::Evaluator<panzer::Traits> > evaluator
= Teuchos::rcp(new panzer::GatherOrientation<panzer::Traits::Residual,panzer::Traits,int,int>(dofManager,pl));
TEST_EQUALITY(evaluator->evaluatedFields().size(),1);
evalField_q1 = evaluator->evaluatedFields()[0];
TEST_EQUALITY(evalField_q1->name(),basis_q1->name()+" Orientation");
TEST_EQUALITY(evalField_q1->dataLayout().dimension(0),basis_q1->functional->dimension(0));
TEST_EQUALITY(evalField_q1->dataLayout().dimension(1),basis_q1->functional->dimension(1));
fm.registerEvaluator<panzer::Traits::Residual>(evaluator);
fm.requireField<panzer::Traits::Residual>(*evaluator->evaluatedFields()[0]);
}
{
Teuchos::RCP<std::vector<std::string> > dofNames = Teuchos::rcp(new std::vector<std::string>);
dofNames->push_back(fieldName_qedge1);
Teuchos::ParameterList pl;
pl.set("Indexer Names",dofNames);
pl.set("DOF Names",dofNames);
pl.set("Basis",basis_qedge1);
Teuchos::RCP<PHX::Evaluator<panzer::Traits> > evaluator
= Teuchos::rcp(new panzer::GatherOrientation<panzer::Traits::Residual,panzer::Traits,int,int>(dofManager,pl));
TEST_EQUALITY(evaluator->evaluatedFields().size(),1);
evalField_qedge1 = evaluator->evaluatedFields()[0];
TEST_EQUALITY(evalField_qedge1->name(),basis_qedge1->name()+" Orientation");
TEST_EQUALITY(evalField_qedge1->dataLayout().dimension(0),basis_qedge1->functional->dimension(0));
TEST_EQUALITY(evalField_qedge1->dataLayout().dimension(1),basis_qedge1->functional->dimension(1));
fm.registerEvaluator<panzer::Traits::Residual>(evaluator);
fm.requireField<panzer::Traits::Residual>(*evaluator->evaluatedFields()[0]);
}
panzer::Traits::SetupData sd;
fm.postRegistrationSetup(sd);
// run tests
/////////////////////////////////////////////////////////////
panzer::Workset & workset = (*work_sets)[0];
workset.alpha = 0.0;
workset.beta = 0.0;
workset.time = 0.0;
workset.evaluate_transient_terms = false;
fm.evaluateFields<panzer::Traits::Residual>(workset);
// <cell,basis>
PHX::MDField<panzer::Traits::Residual::ScalarT>
fieldData_q1(evalField_q1->name(),basis_q1->functional);
// <cell,basis>
PHX::MDField<panzer::Traits::Residual::ScalarT>
fieldData_qedge1(evalField_qedge1->name(),basis_qedge1->functional);
fm.getFieldData<panzer::Traits::Residual::ScalarT,panzer::Traits::Residual>(fieldData_q1);
fm.getFieldData<panzer::Traits::Residual::ScalarT,panzer::Traits::Residual>(fieldData_qedge1);
for(int i=0;i<fieldData_q1.size();i++) {
TEST_EQUALITY(fieldData_q1[i],1);
}
for(int i=0;i<fieldData_qedge1.dimension(0);i++) {
TEST_EQUALITY(fieldData_qedge1(i,0), 1);
TEST_EQUALITY(fieldData_qedge1(i,1), 1);
TEST_EQUALITY(fieldData_qedge1(i,2),-1);
TEST_EQUALITY(fieldData_qedge1(i,3),-1);
}
}
const RCP<const FactoryBase> FactoryManager<Scalar, LocalOrdinal, GlobalOrdinal, Node, LocalMatOps>::GetDefaultFactory(const std::string & varName) const {
if (IsAvailable(varName, defaultFactoryTable_)) {
return defaultFactoryTable_[varName];
} else {
if (varName == "A") return SetAndReturnDefaultFactory(varName, rcp(new RAPFactory()));
if (varName == "RAP Pattern") return GetFactory("A");
if (varName == "AP Pattern") return GetFactory("A");
if (varName == "P") {
RCP<Factory> factory = rcp(new SaPFactory());
factory->SetFactory("P", GetFactory("Ptent")); // GetFactory("Ptent"): Use the same factory instance for both "P" and "Nullspace"
return SetAndReturnDefaultFactory(varName, factory);
}
if (varName == "R") return SetAndReturnDefaultFactory(varName, rcp(new TransPFactory()));
#if defined(HAVE_MUELU_ZOLTAN) && defined(HAVE_MPI)
if (varName == "Partition") {
return SetAndReturnDefaultFactory(varName, rcp(new ZoltanInterface()));
}
#endif //ifdef HAVE_MPI
if (varName == "Importer") {
#ifdef HAVE_MPI
return SetAndReturnDefaultFactory(varName, rcp(new RepartitionFactory()));
#else
return SetAndReturnDefaultFactory(varName, NoFactory::getRCP());
#endif
}
if (varName == "number of partitions") {
return GetFactory("Importer");
}
//JJH FIXME is this going to bite me in the backside?
// if (varName == "Coordinates") {
// return SetAndReturnDefaultFactory(varName, rcp(new MueLu::MultiVectorTransferFactory<SC,LO,GO,NO,LMO>(varName,"R")));
// }
if (varName == "Nullspace") {
RCP<Factory> factory = rcp(new NullspaceFactory());
factory->SetFactory("Nullspace", GetFactory("Ptent")); // GetFactory("Ptent"): Use the same factory instance for both "P" and "Nullspace"
return SetAndReturnDefaultFactory(varName, factory);
}
if (varName == "Graph") return SetAndReturnDefaultFactory(varName, rcp(new CoalesceDropFactory()));
if (varName == "UnAmalgamationInfo") return SetAndReturnDefaultFactory(varName, rcp(new AmalgamationFactory())); //GetFactory("Graph"));
if (varName == "Aggregates") return SetAndReturnDefaultFactory(varName, rcp(new CoupledAggregationFactory()));
if (varName == "CoarseMap") return SetAndReturnDefaultFactory(varName, rcp(new CoarseMapFactory()));
if (varName == "DofsPerNode") return GetFactory("Graph");
if (varName == "Filtering") return GetFactory("Graph");
// Same factory for both Pre and Post Smoother. Factory for key "Smoother" can be set by users.
if (varName == "PreSmoother") return GetFactory("Smoother");
if (varName == "PostSmoother") return GetFactory("Smoother");
#ifdef HAVE_MUELU_EXPERIMENTAL
if (varName == "Ppattern") {
RCP<PatternFactory> PpFact = rcp(new PatternFactory);
PpFact->SetFactory("P", GetFactory("Ptent"));
return SetAndReturnDefaultFactory(varName, PpFact);
}
if (varName == "Constraint") return SetAndReturnDefaultFactory(varName, rcp(new ConstraintFactory()));
#endif
//if (varName == "Smoother") return SetAndReturnDefaultFactory(varName, rcp(new SmootherFactory(rcp(new GaussSeidelSmoother()))));
if (varName == "Smoother") {
Teuchos::ParameterList smootherParamList;
smootherParamList.set("relaxation: type", "Symmetric Gauss-Seidel");
smootherParamList.set("relaxation: sweeps", (LO) 1);
smootherParamList.set("relaxation: damping factor", (Scalar) 1.0); //FIXME once Ifpack2's parameter list validator is fixed, change this back to Scalar
return SetAndReturnDefaultFactory(varName, rcp( new SmootherFactory(rcp(new TrilinosSmoother("RELAXATION", smootherParamList)))));
}
if (varName == "CoarseSolver") return SetAndReturnDefaultFactory(varName, rcp(new SmootherFactory(rcp(new DirectSolver()),Teuchos::null)));
if (varName == "Ptent") return SetAndReturnDefaultFactory(varName, rcp(new TentativePFactory())); // Use the same factory instance for both "P" and "Nullspace"
TEUCHOS_TEST_FOR_EXCEPTION(true, MueLu::Exceptions::RuntimeError, "MueLu::FactoryManager::GetDefaultFactory(): No default factory available for building '"+varName+"'.");
}
}
void MLParameterListInterpreter<Scalar, LocalOrdinal, GlobalOrdinal, Node, LocalMatOps>::SetParameterList(const Teuchos::ParameterList & paramList_in) {
Teuchos::ParameterList paramList = paramList_in;
RCP<Teuchos::FancyOStream> out = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout)); // TODO: use internal out (GetOStream())
//
// Read top-level of the parameter list
//
// hard-coded default values == ML defaults according to the manual
MUELU_READ_PARAM(paramList, "ML output", int, 0, verbosityLevel);
MUELU_READ_PARAM(paramList, "max levels", int, 10, maxLevels);
MUELU_READ_PARAM(paramList, "PDE equations", int, 1, nDofsPerNode);
MUELU_READ_PARAM(paramList, "coarse: max size", int, 128, maxCoarseSize);
MUELU_READ_PARAM(paramList, "aggregation: type", std::string, "Uncoupled", agg_type);
//MUELU_READ_PARAM(paramList, "aggregation: threshold", double, 0.0, agg_threshold);
MUELU_READ_PARAM(paramList, "aggregation: damping factor", double, (double)4/(double)3, agg_damping);
//MUELU_READ_PARAM(paramList, "aggregation: smoothing sweeps", int, 1, agg_smoothingsweeps);
MUELU_READ_PARAM(paramList, "aggregation: nodes per aggregate", int, 1, minPerAgg);
MUELU_READ_PARAM(paramList, "aggregation: keep Dirichlet bcs", bool, false, bKeepDirichletBcs); // This is a MueLu specific extension that does not exist in ML
MUELU_READ_PARAM(paramList, "aggregation: max neighbours already aggregated", int, 0, maxNbrAlreadySelected); // This is a MueLu specific extension that does not exist in ML
MUELU_READ_PARAM(paramList, "null space: type", std::string, "default vectors", nullspaceType);
MUELU_READ_PARAM(paramList, "null space: dimension", int, -1, nullspaceDim); // TODO: ML default not in documentation
MUELU_READ_PARAM(paramList, "null space: vectors", double*, NULL, nullspaceVec); // TODO: ML default not in documentation
MUELU_READ_PARAM(paramList, "energy minimization: enable", bool, false, bEnergyMinimization);
MUELU_READ_PARAM(paramList, "RAP: fix diagonal", bool, false, bFixDiagonal); // This is a MueLu specific extension that does not exist in ML
//
// Move smoothers/aggregation/coarse parameters to sublists
//
// ML allows to have level-specific smoothers/aggregation/coarse parameters at the top level of the list or/and defined in sublists:
// See also: ML Guide section 6.4.1, MueLu::CreateSublists, ML_CreateSublists
ParameterList paramListWithSubList;
MueLu::CreateSublists(paramList, paramListWithSubList);
paramList = paramListWithSubList; // swap
// std::cout << std::endl << "Parameter list after CreateSublists" << std::endl;
// std::cout << paramListWithSubList << std::endl;
//
// Validate parameter list
//
{
bool validate = paramList.get("ML validate parameter list", true); /* true = default in ML */
if (validate) {
#if defined(HAVE_MUELU_ML) && defined(HAVE_MUELU_EPETRA)
// Validate parameter list using ML validator
int depth = paramList.get("ML validate depth", 5); /* 5 = default in ML */
TEUCHOS_TEST_FOR_EXCEPTION(! ML_Epetra::ValidateMLPParameters(paramList, depth), Exceptions::RuntimeError,
"ERROR: ML's Teuchos::ParameterList contains incorrect parameter!");
#else
// If no validator available: issue a warning and set parameter value to false in the output list
*out << "Warning: MueLu_ENABLE_ML=OFF. The parameter list cannot be validated." << std::endl;
paramList.set("ML validate parameter list", false);
#endif // HAVE_MUELU_ML
} // if(validate)
} // scope
//
//
//
// Matrix option
blksize_ = nDofsPerNode;
// Translate verbosity parameter
// Translate verbosity parameter
MsgType eVerbLevel = None;
if (verbosityLevel == 0) eVerbLevel = None;
if (verbosityLevel > 0) eVerbLevel = Low;
if (verbosityLevel > 4) eVerbLevel = Medium;
if (verbosityLevel > 7) eVerbLevel = High;
if (verbosityLevel > 9) eVerbLevel = Extreme;
if (verbosityLevel > 9) eVerbLevel = Test;
this->verbosity_ = eVerbLevel;
TEUCHOS_TEST_FOR_EXCEPTION(agg_type != "Uncoupled" && agg_type != "Coupled", Exceptions::RuntimeError, "MueLu::MLParameterListInterpreter::Setup(): parameter \"aggregation: type\": only 'Uncoupled' or 'Coupled' aggregation is supported.");
// Create MueLu factories
RCP<CoalesceDropFactory> dropFact = rcp(new CoalesceDropFactory());
RCP<FactoryBase> CoupledAggFact = Teuchos::null;
if(agg_type == "Uncoupled") {
// Uncoupled aggregation
RCP<UncoupledAggregationFactory> CoupledAggFact2 = rcp(new UncoupledAggregationFactory());
/*CoupledAggFact2->SetMinNodesPerAggregate(minPerAgg); //TODO should increase if run anything other than 1D
CoupledAggFact2->SetMaxNeighAlreadySelected(maxNbrAlreadySelected);
CoupledAggFact2->SetOrdering("natural");*/
CoupledAggFact2->SetFactory("Graph", dropFact);
CoupledAggFact2->SetFactory("DofsPerNode", dropFact);
CoupledAggFact2->SetParameter("UsePreserveDirichletAggregationAlgorithm", Teuchos::ParameterEntry(bKeepDirichletBcs));
CoupledAggFact2->SetParameter("aggregation: ordering", Teuchos::ParameterEntry(std::string("natural")));
CoupledAggFact2->SetParameter("aggregation: max selected neighbors", Teuchos::ParameterEntry(maxNbrAlreadySelected));
CoupledAggFact2->SetParameter("aggregation: min agg size", Teuchos::ParameterEntry(minPerAgg));
CoupledAggFact = CoupledAggFact2;
} else {
// Coupled Aggregation (default)
RCP<CoupledAggregationFactory> CoupledAggFact2 = rcp(new CoupledAggregationFactory());
CoupledAggFact2->SetMinNodesPerAggregate(minPerAgg); //TODO should increase if run anything other than 1D
CoupledAggFact2->SetMaxNeighAlreadySelected(maxNbrAlreadySelected);
CoupledAggFact2->SetOrdering("natural");
CoupledAggFact2->SetPhase3AggCreation(0.5);
CoupledAggFact2->SetFactory("Graph", dropFact);
CoupledAggFact2->SetFactory("DofsPerNode", dropFact);
CoupledAggFact = CoupledAggFact2;
}
if (verbosityLevel > 3) { // TODO fix me: Setup is a static function: we cannot use GetOStream without an object...
*out << "========================= Aggregate option summary  =========================" << std::endl;
*out << "min Nodes per aggregate : Â Â Â Â Â Â Â " << minPerAgg << std::endl;
*out << "min # of root nbrs already aggregated : " << maxNbrAlreadySelected << std::endl;
*out << "aggregate ordering : Â Â Â Â Â Â Â Â Â Â natural" << std::endl;
*out << "=============================================================================" << std::endl;
}
RCP<Factory> PFact;
RCP<Factory> RFact;
RCP<Factory> PtentFact = rcp( new TentativePFactory() );
if (agg_damping == 0.0 && bEnergyMinimization == false) {
// tentative prolongation operator (PA-AMG)
PFact = PtentFact;
RFact = rcp( new TransPFactory() );
} else if (agg_damping != 0.0 && bEnergyMinimization == false) {
// smoothed aggregation (SA-AMG)
RCP<SaPFactory> SaPFact = rcp( new SaPFactory() );
SaPFact->SetDampingFactor(agg_damping);
PFact = SaPFact;
RFact = rcp( new TransPFactory() );
} else if (bEnergyMinimization == true) {
// Petrov Galerkin PG-AMG smoothed aggregation (energy minimization in ML)
PFact = rcp( new PgPFactory() );
RFact = rcp( new GenericRFactory() );
}
RCP<RAPFactory> AcFact = rcp( new RAPFactory() );
AcFact->SetParameter("RepairMainDiagonal", Teuchos::ParameterEntry(bFixDiagonal));
for (size_t i = 0; i<TransferFacts_.size(); i++) {
AcFact->AddTransferFactory(TransferFacts_[i]);
}
//
// introduce rebalancing
//
#if defined(HAVE_MUELU_ISORROPIA) && defined(HAVE_MPI)
Teuchos::RCP<Factory> RebalancedPFact = Teuchos::null;
Teuchos::RCP<Factory> RebalancedRFact = Teuchos::null;
Teuchos::RCP<Factory> RepartitionFact = Teuchos::null;
Teuchos::RCP<RebalanceAcFactory> RebalancedAFact = Teuchos::null;
MUELU_READ_PARAM(paramList, "repartition: enable", int, 0, bDoRepartition);
if (bDoRepartition == 1) {
// The Factory Manager will be configured to return the rebalanced versions of P, R, A by default.
// Everytime we want to use the non-rebalanced versions, we need to explicitly define the generating factory.
RFact->SetFactory("P", PFact);
//
AcFact->SetFactory("P", PFact);
AcFact->SetFactory("R", RFact);
MUELU_READ_PARAM(paramList, "repartition: max min ratio", double, 1.3, maxminratio);
MUELU_READ_PARAM(paramList, "repartition: min per proc", int, 512, minperproc);
// create "Partition"
Teuchos::RCP<MueLu::IsorropiaInterface<LO, GO, NO, LMO> > isoInterface = Teuchos::rcp(new MueLu::IsorropiaInterface<LO, GO, NO, LMO>());
isoInterface->SetFactory("A", AcFact);
// Repartitioning (creates "Importer" from "Partition")
RepartitionFact = Teuchos::rcp(new RepartitionFactory());
{
Teuchos::ParameterList paramListRepFact;
paramListRepFact.set("repartition: min rows per proc", minperproc);
paramListRepFact.set("repartition: max imbalance", maxminratio);
RepartitionFact->SetParameterList(paramListRepFact);
}
RepartitionFact->SetFactory("A", AcFact);
RepartitionFact->SetFactory("Partition", isoInterface);
// Reordering of the transfer operators
RebalancedPFact = Teuchos::rcp(new RebalanceTransferFactory());
RebalancedPFact->SetParameter("type", Teuchos::ParameterEntry(std::string("Interpolation")));
RebalancedPFact->SetFactory("P", PFact);
RebalancedRFact = Teuchos::rcp(new RebalanceTransferFactory());
RebalancedRFact->SetParameter("type", Teuchos::ParameterEntry(std::string("Restriction")));
RebalancedRFact->SetFactory("R", RFact);
RebalancedRFact->SetFactory("Nullspace", PtentFact);
// Compute Ac from rebalanced P and R
RebalancedAFact = Teuchos::rcp(new RebalanceAcFactory());
RebalancedAFact->SetFactory("A", AcFact);
}
int main(int argc, char *argv[])
{
#ifndef HAVE_TPETRA_COMPLEX_DOUBLE
# error "Anasazi: This test requires Scalar = std::complex<double> to be enabled in Tpetra."
#else
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::tuple;
using std::cout;
using std::endl;
typedef std::complex<double> ST;
typedef Teuchos::ScalarTraits<ST> SCT;
typedef SCT::magnitudeType MT;
typedef Tpetra::MultiVector<ST> MV;
typedef MV::global_ordinal_type GO;
typedef Tpetra::Operator<ST> OP;
typedef Anasazi::MultiVecTraits<ST,MV> MVT;
typedef Anasazi::OperatorTraits<ST,MV,OP> OPT;
Tpetra::ScopeGuard tpetraScope (&argc, &argv);
bool success = false;
const ST ONE = SCT::one ();
int info = 0;
RCP<const Teuchos::Comm<int> > comm = Tpetra::getDefaultComm ();
const int MyPID = comm->getRank ();
bool verbose = false;
bool debug = false;
bool insitu = false;
bool herm = false;
std::string which("LM");
std::string filename;
int nev = 4;
int blockSize = 4;
MT tol = 1.0e-6;
Teuchos::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("herm","nonherm",&herm,"Solve Hermitian or non-Hermitian problem.");
cmdp.setOption("filename",&filename,"Filename for Harwell-Boeing test matrix (assumes non-Hermitian unless specified otherwise).");
cmdp.setOption("nev",&nev,"Number of eigenvalues to compute.");
cmdp.setOption("blockSize",&blockSize,"Block size for the algorithm.");
cmdp.setOption("tol",&tol,"Tolerance for convergence.");
if (cmdp.parse(argc,argv) != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL) {
return -1;
}
if (debug) verbose = true;
if (filename == "") {
// get default based on herm
if (herm) {
filename = "mhd1280b.cua";
}
else {
filename = "mhd1280a.cua";
}
}
if (MyPID == 0) {
cout << Anasazi::Anasazi_Version() << endl << endl;
}
// Get the data from the HB file
int dim,dim2,nnz;
int rnnzmax;
double *dvals;
int *colptr,*rowind;
nnz = -1;
if (MyPID == 0) {
info = readHB_newmat_double(filename.c_str(),&dim,&dim2,&nnz,&colptr,&rowind,&dvals);
// find maximum NNZ over all rows
vector<int> rnnz(dim,0);
for (int *ri=rowind; ri<rowind+nnz; ++ri) {
++rnnz[*ri-1];
}
rnnzmax = *std::max_element(rnnz.begin(),rnnz.end());
}
else {
// address uninitialized data warnings
dvals = NULL;
colptr = NULL;
rowind = NULL;
}
Teuchos::broadcast(*comm,0,&info);
Teuchos::broadcast(*comm,0,&nnz);
Teuchos::broadcast(*comm,0,&dim);
Teuchos::broadcast(*comm,0,&rnnzmax);
if (info == 0 || nnz < 0) {
if (MyPID == 0) {
cout << "Error reading '" << filename << "'" << endl
<< "End Result: TEST FAILED" << endl;
}
return -1;
}
// create map
RCP<const Map<> > map = rcp (new Map<> (dim, 0, comm));
RCP<CrsMatrix<ST> > K = rcp (new CrsMatrix<ST> (map, rnnzmax));
if (MyPID == 0) {
// Convert interleaved doubles to complex values
// HB format is compressed column. CrsMatrix is compressed row.
const double *dptr = dvals;
const int *rptr = rowind;
for (int c=0; c<dim; ++c) {
for (int colnnz=0; colnnz < colptr[c+1]-colptr[c]; ++colnnz) {
K->insertGlobalValues (static_cast<GO> (*rptr++ - 1), tuple<GO> (c), tuple (ST (dptr[0], dptr[1])));
dptr += 2;
}
}
}
if (MyPID == 0) {
// Clean up.
free( dvals );
free( colptr );
free( rowind );
}
K->fillComplete();
// cout << *K << endl;
// Create initial vectors
RCP<MV> ivec = rcp( new MV(map,blockSize) );
ivec->randomize ();
// Create eigenproblem
RCP<Anasazi::BasicEigenproblem<ST,MV,OP> > problem =
rcp( new Anasazi::BasicEigenproblem<ST,MV,OP>(K,ivec) );
//
// Inform the eigenproblem that the operator K is symmetric
problem->setHermitian(herm);
//
// Set the number of eigenvalues requested
problem->setNEV( nev );
//
// Inform the eigenproblem that you are done passing it information
bool boolret = problem->setProblem();
if (boolret != true) {
if (MyPID == 0) {
cout << "Anasazi::BasicEigenproblem::SetProblem() returned with error." << endl
<< "End Result: TEST FAILED" << endl;
}
return -1;
}
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings + Anasazi::FinalSummary + Anasazi::TimingDetails;
if (verbose) {
verbosity += Anasazi::IterationDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
// Eigensolver parameters
int numBlocks = 8;
int maxRestarts = 10;
//
// Create parameter list to pass into the solver manager
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( "In Situ Restarting", insitu );
//
// Create the solver manager
Anasazi::BlockKrylovSchurSolMgr<ST,MV,OP> MySolverMgr(problem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
success = (returnCode == Anasazi::Converged);
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ST,MV> sol = problem->getSolution();
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<MT> normV( numev );
Teuchos::SerialDenseMatrix<int,ST> T (numev, numev);
for (int i=0; i<numev; i++) {
T(i,i) = ST(sol.Evals[i].realpart,sol.Evals[i].imagpart);
}
RCP<MV> Kvecs = MVT::Clone( *evecs, numev );
OPT::Apply( *K, *evecs, *Kvecs );
MVT::MvTimesMatAddMv( -ONE, *evecs, T, ONE, *Kvecs );
MVT::MvNorm( *Kvecs, normV );
os << "Direct residual norms computed in BlockKrylovSchurComplex_test.exe" << endl
<< std::setw(20) << "Eigenvalue" << std::setw(20) << "Residual " << endl
<< "----------------------------------------" << endl;
for (int i=0; i<numev; i++) {
if ( SCT::magnitude(T(i,i)) != SCT::zero() ) {
normV[i] = SCT::magnitude(normV[i]/T(i,i));
}
os << std::setw(20) << T(i,i) << std::setw(20) << normV[i] << endl;
success = (normV[i] < tol);
}
if (MyPID==0) {
cout << endl << os.str() << endl;
}
}
if (MyPID==0) {
if (success)
cout << "End Result: TEST PASSED" << endl;
else
cout << "End Result: TEST FAILED" << endl;
}
return ( success ? EXIT_SUCCESS : EXIT_FAILURE );
#endif // HAVE_TPETRA_COMPLEX_DOUBLE
}
int main(int argc, char *argv[]) {
#include "MueLu_UseShortNames.hpp"
using Teuchos::RCP;
using Teuchos::rcp;
//
// MPI initialization
//
Teuchos::oblackholestream blackhole;
Teuchos::GlobalMPISession mpiSession(&argc, &argv, &blackhole);
bool success = false;
bool verbose = true;
try {
RCP< const Teuchos::Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
//
// Process command line arguments
//
Teuchos::CommandLineProcessor clp(false);
Galeri::Xpetra::Parameters<GO> matrixParameters(clp, 81); // manage parameters of the test case
Xpetra::Parameters xpetraParameters(clp); // manage parameters of xpetra
switch (clp.parse(argc,argv)) {
case Teuchos::CommandLineProcessor::PARSE_HELP_PRINTED: return EXIT_SUCCESS;
case Teuchos::CommandLineProcessor::PARSE_ERROR:
case Teuchos::CommandLineProcessor::PARSE_UNRECOGNIZED_OPTION: return EXIT_FAILURE;
case Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL: break;
default:;
}
if (comm->getRank() == 0) std::cout << xpetraParameters << matrixParameters;
//
// Setup test case (Ax = b)
//
// Linear Algebra Library
Xpetra::UnderlyingLib lib = xpetraParameters.GetLib();
// Distribution
RCP<const Map> map = MapFactory::Build(lib, matrixParameters.GetNumGlobalElements(), 0, comm);
// Matrix
RCP<Galeri::Xpetra::Problem<Map,CrsMatrixWrap,MultiVector> > Pr =
Galeri::Xpetra::BuildProblem<SC, LO, GO, Map, CrsMatrixWrap, MultiVector>(matrixParameters.GetMatrixType(), map, matrixParameters.GetParameterList());
RCP<Matrix> A = Pr->BuildMatrix();
// User defined nullspace
RCP<MultiVector> nullSpace = VectorFactory::Build(map,1); nullSpace->putScalar((SC) 1.0);
// Define B
RCP<Vector> X = VectorFactory::Build(map,1);
RCP<Vector> B = VectorFactory::Build(map,1);
X->setSeed(846930886);
X->randomize();
A->apply(*X, *B, Teuchos::NO_TRANS, (SC)1.0, (SC)0.0);
// X = 0
X->putScalar((SC) 0.0);
//
// Create a multigrid configuration
//
// Transfer operators
RCP<TentativePFactory> TentativePFact = rcp( new TentativePFactory() );
RCP<SaPFactory> SaPFact = rcp( new SaPFactory() );
RCP<TransPFactory> RFact = rcp( new TransPFactory());
FactoryManager M;
M.SetFactory("Ptent", TentativePFact);
M.SetFactory("P", SaPFact);
M.SetFactory("R", RFact);
M.SetFactory("Smoother", Teuchos::null); //skips smoother setup
M.SetFactory("CoarseSolver", Teuchos::null); //skips coarsest solve setup
//
// Multigrid setup phase
//
int startLevel = 0;
int maxLevels = 10;
std::cout << "=============== Setup transfers only ====================" << std::endl;
Hierarchy H;
H.SetDefaultVerbLevel(MueLu::Medium);
RCP<Level> finestLevel = H.GetLevel();
finestLevel->Set("A", A);
finestLevel->Set("Nullspace", nullSpace);
// Indicate which Hierarchy operators we want to keep
H.Keep("P", SaPFact.get()); //SaPFact is the generating factory for P.
H.Keep("R", RFact.get()); //RFact is the generating factory for R.
H.Keep("Ptent", TentativePFact.get()); //SaPFact is the generating factory for P.
H.Setup(M,startLevel,maxLevels);
std::cout << "=============== Setup smoothers only ====================" << std::endl;
// Create a new A.
RCP<Matrix> newA = Pr->BuildMatrix();
finestLevel->Set("A", newA);
// Create Gauss-Seidel smoother.
std::string ifpackType = "RELAXATION";
Teuchos::ParameterList ifpackList;
ifpackList.set("relaxation: sweeps", (LO) 3);
ifpackList.set("relaxation: damping factor", (SC) 1.0);
RCP<SmootherPrototype> smootherPrototype = rcp(new TrilinosSmoother(ifpackType, ifpackList));
M.SetFactory("Smoother", rcp(new SmootherFactory(smootherPrototype)));
// Create coarsest solver.
RCP<SmootherPrototype> coarseSolverPrototype = rcp( new DirectSolver() );
RCP<SmootherFactory> coarseSolverFact = rcp( new SmootherFactory(coarseSolverPrototype, Teuchos::null) );
M.SetFactory("CoarseSolver", coarseSolverFact);
// Note that we pass the number of levels back in.
H.Setup(M,startLevel, H.GetNumLevels());
std::cout << "=============== Solve ====================" << std::endl;
//
// Solve Ax = B
//
LO nIts = 9;
H.Iterate(*B, *X, nIts);
//
// Print relative residual norm
//
Teuchos::ScalarTraits<SC>::magnitudeType residualNorms = Utilities::ResidualNorm(*A, *X, *B)[0];
if (comm->getRank() == 0) {
std::ios::fmtflags f(std::cout.flags());
std::cout << "||Residual|| = " << std::setiosflags(std::ios::fixed) << std::setprecision(20) << residualNorms << std::endl;
std::cout.flags(f);
}
success = true;
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(verbose, std::cerr, success);
return ( success ? EXIT_SUCCESS : EXIT_FAILURE );
}
bool CompareWithAztecOO(Epetra_LinearProblem& Problem, const std::string what,
int Overlap, int ival)
{
using std::cout;
using std::endl;
AztecOO AztecOOSolver(Problem);
AztecOOSolver.SetAztecOption(AZ_solver,AZ_gmres);
AztecOOSolver.SetAztecOption(AZ_output,AZ_none);
AztecOOSolver.SetAztecOption(AZ_overlap,Overlap);
AztecOOSolver.SetAztecOption(AZ_graph_fill,ival);
AztecOOSolver.SetAztecOption(AZ_poly_ord, ival);
AztecOOSolver.SetAztecParam(AZ_drop, 0.0);
AztecOOSolver.SetAztecParam(AZ_athresh, 0.0);
AztecOOSolver.SetAztecParam(AZ_rthresh, 0.0);
Epetra_MultiVector& RHS = *(Problem.GetRHS());
Epetra_MultiVector& LHS = *(Problem.GetLHS());
Teuchos::RefCountPtr<Epetra_RowMatrix> A = Teuchos::rcp(Problem.GetMatrix(), false);
LHS.Random();
A->Multiply(false,LHS,RHS);
Teuchos::ParameterList List;
List.set("fact: level-of-fill", ival);
List.set("relaxation: sweeps", ival);
List.set("relaxation: damping factor", 1.0);
List.set("relaxation: zero starting solution", true);
//default combine mode is as for AztecOO
List.set("schwarz: combine mode", Zero);
Epetra_Time Time(A->Comm());
Teuchos::RefCountPtr<Ifpack_Preconditioner> Prec;
if (what == "Jacobi") {
Prec = Teuchos::rcp( new Ifpack_PointRelaxation(&*A) );
List.set("relaxation: type", "Jacobi");
AztecOOSolver.SetAztecOption(AZ_precond,AZ_Jacobi);
AztecOOSolver.SetAztecOption(AZ_reorder,0);
}
else if (what == "IC no reord") {
Prec = Teuchos::rcp( new Ifpack_AdditiveSchwarz<Ifpack_IC>(&*A,Overlap) );
AztecOOSolver.SetAztecOption(AZ_precond,AZ_dom_decomp);
AztecOOSolver.SetAztecOption(AZ_subdomain_solve,AZ_icc);
AztecOOSolver.SetAztecOption(AZ_reorder,0);
}
else if (what == "IC reord") {
Prec = Teuchos::rcp( new Ifpack_AdditiveSchwarz<Ifpack_IC>(&*A,Overlap) );
List.set("schwarz: use reordering", true);
AztecOOSolver.SetAztecOption(AZ_precond,AZ_dom_decomp);
AztecOOSolver.SetAztecOption(AZ_subdomain_solve,AZ_icc);
AztecOOSolver.SetAztecOption(AZ_reorder,1);
}
else if (what == "ILU no reord") {
Prec = Teuchos::rcp( new Ifpack_AdditiveSchwarz<Ifpack_ILU>(&*A,Overlap) );
AztecOOSolver.SetAztecOption(AZ_precond,AZ_dom_decomp);
AztecOOSolver.SetAztecOption(AZ_subdomain_solve,AZ_ilu);
AztecOOSolver.SetAztecOption(AZ_reorder,0);
}
else if (what == "ILU reord") {
Prec = Teuchos::rcp( new Ifpack_AdditiveSchwarz<Ifpack_ILU>(&*A,Overlap) );
List.set("schwarz: use reordering", true);
AztecOOSolver.SetAztecOption(AZ_precond,AZ_dom_decomp);
AztecOOSolver.SetAztecOption(AZ_subdomain_solve,AZ_ilu);
AztecOOSolver.SetAztecOption(AZ_reorder,1);
}
#ifdef HAVE_IFPACK_AMESOS
else if (what == "LU") {
Prec = Teuchos::rcp( new Ifpack_AdditiveSchwarz<Ifpack_Amesos>(&*A,Overlap) );
List.set("amesos: solver type", "Klu");
AztecOOSolver.SetAztecOption(AZ_precond,AZ_dom_decomp);
AztecOOSolver.SetAztecOption(AZ_subdomain_solve,AZ_lu);
}
#endif
else {
cerr << "Option not recognized" << endl;
exit(EXIT_FAILURE);
}
// ==================================== //
// Solve with AztecOO's preconditioners //
// ==================================== //
LHS.PutScalar(0.0);
Time.ResetStartTime();
AztecOOSolver.Iterate(150,1e-5);
if (verbose) {
cout << endl;
cout << "==================================================" << endl;
cout << "Testing `" << what << "', Overlap = "
<< Overlap << ", ival = " << ival << endl;
cout << endl;
cout << "[AztecOO] Total time = " << Time.ElapsedTime() << " (s)" << endl;
cout << "[AztecOO] Residual = " << AztecOOSolver.TrueResidual() << " (s)" << endl;
cout << "[AztecOO] Iterations = " << AztecOOSolver.NumIters() << endl;
cout << endl;
}
int AztecOOPrecIters = AztecOOSolver.NumIters();
// =========================================== //
// Create the IFPACK preconditioner and solver //
// =========================================== //
Epetra_Time Time2(A->Comm());
assert(Prec != Teuchos::null);
IFPACK_CHK_ERR(Prec->SetParameters(List));
Time.ResetStartTime();
IFPACK_CHK_ERR(Prec->Initialize());
if (verbose)
cout << "[IFPACK] Time for Initialize() = "
<< Time.ElapsedTime() << " (s)" << endl;
Time.ResetStartTime();
IFPACK_CHK_ERR(Prec->Compute());
if (verbose)
cout << "[IFPACK] Time for Compute() = "
<< Time.ElapsedTime() << " (s)" << endl;
AztecOOSolver.SetPrecOperator(&*Prec);
LHS.PutScalar(0.0);
Time.ResetStartTime();
AztecOOSolver.Iterate(150,1e-5);
if (verbose) {
cout << "[IFPACK] Total time = " << Time2.ElapsedTime() << " (s)" << endl;
cout << "[IFPACK] Residual = " << AztecOOSolver.TrueResidual() << " (s)" << endl;
cout << "[IFPACK] Iterations = " << AztecOOSolver.NumIters() << endl;
cout << endl;
}
int IFPACKPrecIters = AztecOOSolver.NumIters();
if (IFPACK_ABS(AztecOOPrecIters - IFPACKPrecIters) > 3) {
cerr << "TEST FAILED (" << AztecOOPrecIters << " != "
<< IFPACKPrecIters << ")" << endl;
return(false);
}
else
return(true);
}
// The solve is done in the felix_driver_run function, and the solution is passed back to Glimmer-CISM
// IK, 12/3/13: time_inc_yr and cur_time_yr are not used here...
void felix_driver_run(FelixToGlimmer * ftg_ptr, double& cur_time_yr, double time_inc_yr)
{
//IK, 12/9/13: how come FancyOStream prints an all processors??
Teuchos::RCP<Teuchos::FancyOStream> out(Teuchos::VerboseObjectBase::getDefaultOStream());
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0) {
std::cout << "In felix_driver_run, cur_time, time_inc = " << cur_time_yr
<< " " << time_inc_yr << std::endl;
}
// ---------------------------------------------
// get u and v velocity solution from Glimmer-CISM
// IK, 11/26/13: need to concatenate these into a single solve for initial condition for Albany/FELIX solve
// IK, 3/14/14: moved this step to felix_driver_run from felix_driver init, since we still want to grab and u and v velocities for CISM if the mesh hasn't changed,
// in which case only felix_driver_run will be called, not felix_driver_init.
// ---------------------------------------------
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0)
std::cout << "In felix_driver_run: grabbing pointers to u and v velocities in CISM..." << std::endl;
uVel_ptr = ftg_ptr ->getDoubleVar("uvel", "velocity");
vVel_ptr = ftg_ptr ->getDoubleVar("vvel", "velocity");
// ---------------------------------------------
// Set restart solution to the one passed from CISM
// IK, 3/14/14: moved this from felix_driver_init to felix_driver_run.
// ---------------------------------------------
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0)
std::cout << "In felix_driver_run: setting initial condition from CISM..." << std::endl;
//Check what kind of ordering you have in the solution & create solutionField object.
interleavedOrdering = meshStruct->getInterleavedOrdering();
Albany::AbstractSTKFieldContainer::VectorFieldType* solutionField;
if(interleavedOrdering)
solutionField = Teuchos::rcp_dynamic_cast<Albany::OrdinarySTKFieldContainer<true> >(meshStruct->getFieldContainer())->getSolutionField();
else
solutionField = Teuchos::rcp_dynamic_cast<Albany::OrdinarySTKFieldContainer<false> >(meshStruct->getFieldContainer())->getSolutionField();
//Create vector used to renumber nodes on each processor from the Albany convention (horizontal levels first) to the CISM convention (vertical layers first)
nNodes2D = (global_ewn + 1)*(global_nsn+1); //number global nodes in the domain in 2D
nNodesProc2D = (nsn-2*nhalo+1)*(ewn-2*nhalo+1); //number of nodes on each processor in 2D
cismToAlbanyNodeNumberMap.resize(upn*nNodesProc2D);
for (int j=0; j<nsn-2*nhalo+1;j++) {
for (int i=0; i<ewn-2*nhalo+1; i++) {
for (int k=0; k<upn; k++) {
int index = k+upn*i + j*(ewn-2*nhalo+1)*upn;
cismToAlbanyNodeNumberMap[index] = k*nNodes2D + global_node_id_owned_map_Ptr[i+j*(ewn-2*nhalo+1)];
//if (mpiComm->MyPID() == 0)
// std::cout << "index: " << index << ", cismToAlbanyNodeNumberMap: " << cismToAlbanyNodeNumberMap[index] << std::endl;
}
}
}
//The way it worked out, uVel_ptr and vVel_ptr have more nodes than the nodes in the mesh passed to Albany/CISM for the solve. In particular,
//there is 1 row of halo elements in uVel_ptr and vVel_ptr. To account for this, we copy uVel_ptr and vVel_ptr into std::vectors, which do not have the halo elements.
std::vector<double> uvel_vec(upn*nNodesProc2D);
std::vector<double> vvel_vec(upn*nNodesProc2D);
int counter1 = 0;
int counter2 = 0;
int local_nodeID;
for (int j=0; j<nsn-1; j++) {
for (int i=0; i<ewn-1; i++) {
for (int k=0; k<upn; k++) {
if (j >= nhalo-1 & j < nsn-nhalo) {
if (i >= nhalo-1 & i < ewn-nhalo) {
#ifdef CISM_USE_EPETRA
local_nodeID = node_map->LID(cismToAlbanyNodeNumberMap[counter1]);
#else
local_nodeID = node_map->getLocalElement(cismToAlbanyNodeNumberMap[counter1]);
#endif
uvel_vec[counter1] = uVel_ptr[counter2];
vvel_vec[counter1] = vVel_ptr[counter2];
counter1++;
}
}
counter2++;
}
}
}
//Loop over all the elements to find which nodes are active. For the active nodes, copy uvel and vvel from CISM into Albany solution array to
//use as initial condition.
//NOTE: there is some inefficiency here by looping over all the elements. TO DO? pass only active nodes from Albany-CISM to improve this?
double velScale = seconds_per_year*vel_scaling_param;
for (int i=0; i<nElementsActive; i++) {
for (int j=0; j<8; j++) {
int node_GID = global_element_conn_active_Ptr[i + nElementsActive*j]; //node_GID is 1-based
#ifdef CISM_USE_EPETRA
int node_LID = node_map->LID(node_GID); //node_LID is 0-based
#else
int node_LID = node_map->getLocalElement(node_GID); //node_LID is 0-based
#endif
stk::mesh::Entity node = meshStruct->bulkData->get_entity(stk::topology::NODE_RANK, node_GID);
double* sol = stk::mesh::field_data(*solutionField, node);
//IK, 3/18/14: added division by velScale to convert uvel and vvel from dimensionless to having units of m/year (the Albany units)
sol[0] = uvel_vec[node_LID]/velScale;
sol[1] = vvel_vec[node_LID]/velScale;
}
}
// ---------------------------------------------------------------------------------------------------
// Solve
// ---------------------------------------------------------------------------------------------------
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0)
std::cout << "In felix_driver_run: starting the solve... " << std::endl;
//Need to set HasRestart solution such that uvel_Ptr and vvel_Ptr (u and v from Glimmer/CISM) are always set as initial condition?
meshStruct->setHasRestartSolution(!first_time_step);
//Turn off homotopy if we're not in the first time-step.
//NOTE - IMPORTANT: Glen's Law Homotopy parameter should be set to 1.0 in the parameter list for this logic to work!!!
if (!first_time_step)
{
meshStruct->setRestartDataTime(parameterList->sublist("Problem").get("Homotopy Restart Step", 1.));
double homotopy = parameterList->sublist("Problem").sublist("FELIX Viscosity").get("Glen's Law Homotopy Parameter", 1.0);
if(meshStruct->restartDataTime()== homotopy) {
parameterList->sublist("Problem").set("Solution Method", "Steady");
parameterList->sublist("Piro").set("Solver Type", "NOX");
}
}
albanyApp->createDiscretization();
//IK, 10/30/14: Check that # of elements from previous time step hasn't changed.
//If it has not, use previous solution as initial guess for current time step.
//Otherwise do not set initial solution. It's possible this can be improved so some part of the previous solution is used
//defined on the current mesh (if it receded, which likely it will in dynamic ice sheet simulations...).
if (nElementsActivePrevious != nElementsActive) previousSolution = Teuchos::null;
albanyApp->finalSetUp(parameterList, previousSolution);
//if (!first_time_step)
// std::cout << "previousSolution: " << *previousSolution << std::endl;
#ifdef CISM_USE_EPETRA
solver = slvrfctry->createThyraSolverAndGetAlbanyApp(albanyApp, mpiComm, mpiComm, Teuchos::null, false);
#else
solver = slvrfctry->createAndGetAlbanyAppT(albanyApp, mpiCommT, mpiCommT, Teuchos::null, false);
#endif
Teuchos::ParameterList solveParams;
solveParams.set("Compute Sensitivities", true);
Teuchos::Array<Teuchos::RCP<const Thyra::VectorBase<double> > > thyraResponses;
Teuchos::Array<Teuchos::Array<Teuchos::RCP<const Thyra::MultiVectorBase<double> > > > thyraSensitivities;
Piro::PerformSolveBase(*solver, solveParams, thyraResponses, thyraSensitivities);
#ifdef CISM_USE_EPETRA
const Epetra_Map& ownedMap(*albanyApp->getDiscretization()->getMap()); //owned map
const Epetra_Map& overlapMap(*albanyApp->getDiscretization()->getOverlapMap()); //overlap map
Epetra_Import import(overlapMap, ownedMap); //importer from ownedMap to overlapMap
Epetra_Vector solutionOverlap(overlapMap); //overlapped solution
solutionOverlap.Import(*albanyApp->getDiscretization()->getSolutionField(), import, Insert);
#else
Teuchos::RCP<const Tpetra_Map> ownedMap = albanyApp->getDiscretization()->getMapT(); //owned map
Teuchos::RCP<const Tpetra_Map> overlapMap = albanyApp->getDiscretization()->getOverlapMapT(); //overlap map
Teuchos::RCP<Tpetra_Import> import = Teuchos::rcp(new Tpetra_Import(ownedMap, overlapMap));
Teuchos::RCP<Tpetra_Vector> solutionOverlap = Teuchos::rcp(new Tpetra_Vector(overlapMap));
solutionOverlap->doImport(*albanyApp->getDiscretization()->getSolutionFieldT(), *import, Tpetra::INSERT);
Teuchos::ArrayRCP<const ST> solutionOverlap_constView = solutionOverlap->get1dView();
#endif
#ifdef WRITE_TO_MATRIX_MARKET
#ifdef CISM_USE_EPETRA
//For debug: write solution and maps to matrix market file
EpetraExt::BlockMapToMatrixMarketFile("node_map.mm", *node_map);
EpetraExt::BlockMapToMatrixMarketFile("map.mm", ownedMap);
EpetraExt::BlockMapToMatrixMarketFile("overlap_map.mm", overlapMap);
EpetraExt::MultiVectorToMatrixMarketFile("solution.mm", *albanyApp->getDiscretization()->getSolutionField());
#else
Tpetra_MatrixMarket_Writer::writeMapFile("node_map.mm", *node_map);
Tpetra_MatrixMarket_Writer::writeMapFile("map.mm", *ownedMap);
Tpetra_MatrixMarket_Writer::writeMapFile("overlap_map.mm", *overlapMap);
Tpetra_MatrixMarket_Writer::writeDenseFile("solution.mm", app->getDiscretization()->getSolutionFieldT());
#endif
#endif
//set previousSolution (used as initial guess for next time step) to final Albany solution.
previousSolution = Teuchos::rcp(new Tpetra_Vector(*albanyApp->getDiscretization()->getSolutionFieldT()));
nElementsActivePrevious = nElementsActive;
//std::cout << "Final solution: " << *albanyApp->getDiscretization()->getSolutionField() << std::endl;
// ---------------------------------------------------------------------------------------------------
// Compute sensitivies / responses and perform regression tests
// IK, 12/9/13: how come this is turned off in mpas branch?
// ---------------------------------------------------------------------------------------------------
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0)
std::cout << "Computing responses and sensitivities..." << std::endl;
int status=0; // 0 = pass, failures are incremented
#ifdef CISM_USE_EPETRA
Teuchos::Array<Teuchos::RCP<const Epetra_Vector> > responses;
Teuchos::Array<Teuchos::Array<Teuchos::RCP<const Epetra_MultiVector> > > sensitivities;
epetraFromThyra(mpiComm, thyraResponses, thyraSensitivities, responses, sensitivities);
#else
Teuchos::Array<Teuchos::RCP<const Tpetra_Vector> > responses;
Teuchos::Array<Teuchos::Array<Teuchos::RCP<const Tpetra_MultiVector> > > sensitivities;
tpetraFromThyra(thyraResponses, thyraSensitivities, responses, sensitivities);
#endif
const int num_p = solver->Np(); // Number of *vectors* of parameters
const int num_g = solver->Ng(); // Number of *vectors* of responses
if (debug_output_verbosity != 0) {
*out << "Finished eval of first model: Params, Responses "
<< std::setprecision(12) << std::endl;
}
const Thyra::ModelEvaluatorBase::InArgs<double> nominal = solver->getNominalValues();
if (debug_output_verbosity != 0) {
for (int i=0; i<num_p; i++) {
#ifdef CISM_USE_EPETRA
const Teuchos::RCP<const Epetra_Vector> p_init = epetraVectorFromThyra(mpiComm, nominal.get_p(i));
p_init->Print(*out << "\nParameter vector " << i << ":\n");
#else
Albany::printTpetraVector(*out << "\nParameter vector " << i << ":\n",
ConverterT::getConstTpetraVector(nominal.get_p(i)));
#endif
}
}
for (int i=0; i<num_g-1; i++) {
#ifdef CISM_USE_EPETRA
const Teuchos::RCP<const Epetra_Vector> g = responses[i];
#else
const Teuchos::RCP<const Tpetra_Vector> g = responses[i];
#endif
bool is_scalar = true;
if (albanyApp != Teuchos::null)
is_scalar = albanyApp->getResponse(i)->isScalarResponse();
if (is_scalar) {
if (debug_output_verbosity != 0) {
#ifdef CISM_USE_EPETRA
g->Print(*out << "\nResponse vector " << i << ":\n");
#else
Albany::printTpetraVector(*out << "\nResponse vector " << i << ":\n", g);
#endif
}
if (num_p == 0 && cur_time_yr == final_time) {
// Just calculate regression data -- only if in final time step
#ifdef CISM_USE_EPETRA
status += slvrfctry->checkSolveTestResults(i, 0, g.get(), NULL);
#else
status += slvrfctry->checkSolveTestResultsT(i, 0, g.get(), NULL);
#endif
} else {
for (int j=0; j<num_p; j++) {
#ifdef CISM_USE_EPETRA
const Teuchos::RCP<const Epetra_MultiVector> dgdp = sensitivities[i][j];
#else
const Teuchos::RCP<const Tpetra_MultiVector> dgdp = sensitivities[i][j];
#endif
if (debug_output_verbosity != 0) {
if (Teuchos::nonnull(dgdp)) {
#ifdef CISM_USE_EPETRA
dgdp->Print(*out << "\nSensitivities (" << i << "," << j << "):!\n");
#else
Albany::printTpetraVector(*out << "\nSensitivities (" << i << "," << j << "):!\n", dgdp);
#endif
}
}
if (cur_time_yr == final_time) {
#ifdef CISM_USE_EPETRA
status += slvrfctry->checkSolveTestResults(i, j, g.get(), dgdp.get());
#else
status += slvrfctry->checkSolveTestResultsT(i, j, g.get(), dgdp.get());
#endif
}
}
}
}
}
if (debug_output_verbosity != 0 && cur_time_yr == final_time) //only print regression test result if you're in the final time step
*out << "\nNumber of Failed Comparisons: " << status << std::endl;
//IK, 10/30/14: added the following line so that when you run ctest from CISM the test fails if there are some failed comparisons.
if (status > 0)
TEUCHOS_TEST_FOR_EXCEPTION(true, std::logic_error, "All regression comparisons did not pass!" << std::endl);
// ---------------------------------------------------------------------------------------------------
// Copy solution back to glimmer uvel and vvel arrays to be passed back
// ---------------------------------------------------------------------------------------------------
//std::cout << "overlapMap # global elements: " << overlapMap.NumGlobalElements() << std::endl;
//std::cout << "overlapMap # my elements: " << overlapMap.NumMyElements() << std::endl;
//std::cout << "overlapMap: " << overlapMap << std::endl;
//std::cout << "map # global elements: " << ownedMap.NumGlobalElements() << std::endl;
//std::cout << "map # my elements: " << ownedMap.NumMyElements() << std::endl;
//std::cout << "node_map # global elements: " << node_map->NumGlobalElements() << std::endl;
//std::cout << "node_map # my elements: " << node_map->NumMyElements() << std::endl;
//std::cout << "node_map: " << *node_map << std::endl;
if (debug_output_verbosity != 0 & mpiCommT->getRank() == 0)
std::cout << "In felix_driver_run: copying Albany solution to uvel and vvel to send back to CISM... " << std::endl;
#ifdef CISM_USE_EPETRA
//Epetra_Vectors to hold uvel and vvel to be passed to Glimmer/CISM
Epetra_Vector uvel(*node_map, true);
Epetra_Vector vvel(*node_map, true);
#else
//Tpetra_Vectors to hold uvel and vvel to be passed to Glimmer/CISM
Teuchos::RCP<Tpetra_Vector> uvel = Teuchos::rcp(new Tpetra_Vector(node_map, true));
Teuchos::RCP<Tpetra_Vector> vvel = Teuchos::rcp(new Tpetra_Vector(node_map, true));
#endif
#ifdef CISM_USE_EPETRA
if (interleavedOrdering == true) {
for (int i=0; i<overlapMap.NumMyElements(); i++) {
int global_dof = overlapMap.GID(i);
double sol_value = solutionOverlap[i];
int modulo = (global_dof % 2); //check if dof is for u or for v
int vel_global_dof, vel_local_dof;
if (modulo == 0) { //u dof
vel_global_dof = global_dof/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->LID(vel_global_dof); //look up local id corresponding to global id in node_map
//std::cout << "uvel: global_dof = " << global_dof << ", uvel_global_dof = " << vel_global_dof << ", uvel_local_dof = " << vel_local_dof << std::endl;
uvel.ReplaceMyValues(1, &sol_value, &vel_local_dof);
}
else { // v dof
vel_global_dof = (global_dof-1)/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->LID(vel_global_dof); //look up local id corresponding to global id in node_map
vvel.ReplaceMyValues(1, &sol_value, & vel_local_dof);
}
}
}
else { //note: the case with non-interleaved ordering has not been tested...
int numDofs = overlapMap.NumGlobalElements();
for (int i=0; i<overlapMap.NumMyElements(); i++) {
int global_dof = overlapMap.GID(i);
double sol_value = solutionOverlap[i];
int vel_global_dof, vel_local_dof;
if (global_dof < numDofs/2) { //u dof
vel_global_dof = global_dof+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->LID(vel_global_dof); //look up local id corresponding to global id in node_map
uvel.ReplaceMyValues(1, &sol_value, &vel_local_dof);
}
else { //v dofs
vel_global_dof = global_dof-numDofs/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->LID(vel_global_dof); //look up local id corresponding to global id in node_map
vvel.ReplaceMyValues(1, &sol_value, & vel_local_dof);
}
}
}
#else
if (interleavedOrdering == true) {
for (int i=0; i<overlapMap->getNodeNumElements(); i++) {
int global_dof = overlapMap->getGlobalElement(i);
double sol_value = solutionOverlap_constView[i];
int modulo = (global_dof % 2); //check if dof is for u or for v
int vel_global_dof, vel_local_dof;
if (modulo == 0) { //u dof
vel_global_dof = global_dof/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->getLocalElement(vel_global_dof); //look up local id corresponding to global id in node_map
//std::cout << "uvel: global_dof = " << global_dof << ", uvel_global_dof = " << vel_global_dof << ", uvel_local_dof = " << vel_local_dof << std::endl;
uvel->replaceLocalValue(vel_local_dof, sol_value);
}
else { // v dof
vel_global_dof = (global_dof-1)/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->getLocalElement(vel_global_dof); //look up local id corresponding to global id in node_map
vvel->replaceLocalValue(vel_local_dof, sol_value);
}
}
}
else { //note: the case with non-interleaved ordering has not been tested...
int numDofs = overlapMap->getGlobalNumElements();
for (int i=0; i<overlapMap->getNodeNumElements(); i++) {
int global_dof = overlapMap->getGlobalElement(i);
double sol_value = solutionOverlap_constView[i];
int vel_global_dof, vel_local_dof;
if (global_dof < numDofs/2) { //u dof
vel_global_dof = global_dof+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->getLocalElement(vel_global_dof); //look up local id corresponding to global id in node_map
uvel->replaceLocalValue(vel_local_dof, sol_value);
}
else { //v dofs
vel_global_dof = global_dof-numDofs/2+1; //add 1 because node_map is 1-based
vel_local_dof = node_map->getLocalElement(vel_global_dof); //look up local id corresponding to global id in node_map
vvel->replaceLocalValue(vel_local_dof, sol_value);
}
}
}
#endif
#ifdef WRITE_TO_MATRIX_MARKET
//For debug: write solution to matrix market file
#ifdef CISM_USE_EPETRA
EpetraExt::MultiVectorToMatrixMarketFile("uvel.mm", uvel);
EpetraExt::MultiVectorToMatrixMarketFile("vvel.mm", vvel);
#else
Tpetra_MatrixMarket_Writer::writeDenseFile("uvel.mm", uvel);
Tpetra_MatrixMarket_Writer::writeDenseFile("vvel.mm", vvel);
#endif
#endif
//Copy uvel and vvel into uVel_ptr and vVel_ptr respectively (the arrays passed back to CISM) according to the numbering consistent w/ CISM.
counter1 = 0;
counter2 = 0;
#ifdef CISM_USE_EPETRA
#else
Teuchos::ArrayRCP<const ST> uvel_constView = uvel->get1dView();
Teuchos::ArrayRCP<const ST> vvel_constView = vvel->get1dView();
#endif
local_nodeID = 0;
for (int j=0; j<nsn-1; j++) {
for (int i=0; i<ewn-1; i++) {
for (int k=0; k<upn; k++) {
if (j >= nhalo-1 & j < nsn-nhalo) {
if (i >= nhalo-1 & i < ewn-nhalo) {
#ifdef CISM_USE_EPETRA
local_nodeID = node_map->LID(cismToAlbanyNodeNumberMap[counter1]);
//if (mpiComm->MyPID() == 0)
//std::cout << "counter1:" << counter1 << ", cismToAlbanyNodeNumberMap[counter1]: " << cismToAlbanyNodeNumberMap[counter1] << ", local_nodeID: "
//<< local_nodeID << ", uvel: " << uvel[local_nodeID] << std::endl; //uvel[local_nodeID] << std::endl;
uVel_ptr[counter2] = uvel[local_nodeID];
vVel_ptr[counter2] = vvel[local_nodeID];
#else
local_nodeID = node_map->getLocalElement(cismToAlbanyNodeNumberMap[counter1]);
uVel_ptr[counter2] = uvel_constView[local_nodeID];
vVel_ptr[counter2] = vvel_constView[local_nodeID];
#endif
counter1++;
}
}
else {
uVel_ptr[counter2] = 0.0;
vVel_ptr[counter2] = 0.0;
}
counter2++;
}
}
}
first_time_step = false;
}
int main(int argc, char *argv[])
{
// MEMORY_CHECK(true, "Before initializing MPI");
Teuchos::GlobalMPISession session(&argc, &argv, NULL);
RCP<const Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
int rank = comm->getRank();
int nprocs = comm->getSize();
MEMORY_CHECK(rank==0, "After initializing MPI");
if (rank==0)
cout << "Number of processes: " << nprocs << endl;
// Default values
double numGlobalCoords = 1000;
int numTestCuts = 1;
int nWeights = 0;
string timingType("no_timers");
string debugLevel("basic_status");
string memoryOn("memoryOn");
string memoryOff("memoryOff");
string memoryProcs("0");
bool doMemory=false;
int numGlobalParts = nprocs;
CommandLineProcessor commandLine(false, true);
commandLine.setOption("size", &numGlobalCoords,
"Approximate number of global coordinates.");
commandLine.setOption("testCuts", &numTestCuts,
"Number of test cuts to make when looking for bisector.");
commandLine.setOption("numParts", &numGlobalParts,
"Number of parts (default is one per proc).");
commandLine.setOption("nWeights", &nWeights,
"Number of weights per coordinate, zero implies uniform weights.");
string balanceCount("balance_object_count");
string balanceWeight("balance_object_weight");
string mcnorm1("multicriteria_minimize_total_weight");
string mcnorm2("multicriteria_balance_total_maximum");
string mcnorm3("multicriteria_minimize_maximum_weight");
string objective(balanceWeight); // default
string doc(balanceCount); doc.append(": ignore weights\n");
doc.append(balanceWeight); doc.append(": balance on first weight\n");
doc.append(mcnorm1);
doc.append(": given multiple weights, balance their total.\n");
doc.append(mcnorm3);
doc.append(": given multiple weights, balance the maximum for each coordinate.\n");
doc.append(mcnorm2);
doc.append(": given multiple weights, balance the L2 norm of the weights.\n");
commandLine.setOption("objective", &objective, doc.c_str());
commandLine.setOption("timers", &timingType,
"no_timers, micro_timers, macro_timers, both_timers, test_timers");
commandLine.setOption("debug", &debugLevel,
"no_status, basic_status, detailed_status, verbose_detailed_status");
commandLine.setOption(memoryOn.c_str(), memoryOff.c_str(), &doMemory,
"do memory profiling");
commandLine.setOption("memoryProcs", &memoryProcs,
"list of processes that output memory usage");
CommandLineProcessor::EParseCommandLineReturn rc =
commandLine.parse(argc, argv);
if (rc != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL){
if (rc == Teuchos::CommandLineProcessor::PARSE_HELP_PRINTED){
if (rank==0)
cout << "PASS" << endl;
return 1;
}
else{
if (rank==0)
cout << "FAIL" << endl;
return 0;
}
}
//MEMORY_CHECK(doMemory && rank==0, "After processing parameters");
zgno_t globalSize = static_cast<zgno_t>(numGlobalCoords);
RCP<tMVector_t> coordinates = getMeshCoordinates(comm, globalSize);
size_t numLocalCoords = coordinates->getLocalLength();
#if 0
comm->barrier();
for (int p=0; p < nprocs; p++){
if (p==rank){
cout << "Rank " << rank << ", " << numLocalCoords << "coords" << endl;
const zscalar_t *x = coordinates->getData(0).getRawPtr();
const zscalar_t *y = coordinates->getData(1).getRawPtr();
const zscalar_t *z = coordinates->getData(2).getRawPtr();
for (zlno_t i=0; i < numLocalCoords; i++)
cout << " " << x[i] << " " << y[i] << " " << z[i] << endl;
}
cout.flush();
comm->barrier();
}
#endif
Array<ArrayRCP<zscalar_t> > weights(nWeights);
if (nWeights > 0){
int wt = 0;
zscalar_t scale = 1.0;
for (int i=0; i < nWeights; i++){
weights[i] =
makeWeights(comm, numLocalCoords, weightTypes(wt++), scale, rank);
if (wt == numWeightTypes){
wt = 0;
scale++;
}
}
}
MEMORY_CHECK(doMemory && rank==0, "After creating input");
// Create an input adapter.
const RCP<const tMap_t> &coordmap = coordinates->getMap();
ArrayView<const zgno_t> ids = coordmap->getNodeElementList();
const zgno_t *globalIds = ids.getRawPtr();
size_t localCount = coordinates->getLocalLength();
typedef Zoltan2::BasicVectorAdapter<tMVector_t> inputAdapter_t;
RCP<inputAdapter_t> ia;
if (nWeights == 0){
ia = rcp(new inputAdapter_t (localCount, globalIds,
coordinates->getData(0).getRawPtr(), coordinates->getData(1).getRawPtr(),
coordinates->getData(2).getRawPtr(), 1,1,1));
}
else{
vector<const zscalar_t *> values(3);
for (int i=0; i < 3; i++)
values[i] = coordinates->getData(i).getRawPtr();
vector<int> valueStrides(0); // implies stride is one
vector<const zscalar_t *> weightPtrs(nWeights);
for (int i=0; i < nWeights; i++)
weightPtrs[i] = weights[i].getRawPtr();
vector<int> weightStrides(0); // implies stride is one
ia = rcp(new inputAdapter_t (localCount, globalIds,
values, valueStrides, weightPtrs, weightStrides));
}
MEMORY_CHECK(doMemory && rank==0, "After creating input adapter");
// Parameters
Teuchos::ParameterList params;
if (timingType != "no_timers"){
params.set("timer_output_stream" , "std::cout");
params.set("timer_type" , timingType);
}
if (doMemory){
params.set("memory_output_stream" , "std::cout");
params.set("memory_procs" , memoryProcs);
}
params.set("debug_output_stream" , "std::cerr");
params.set("debug_procs" , "0");
if (debugLevel != "basic_status"){
params.set("debug_level" , debugLevel);
}
params.set("algorithm", "rcb");
params.set("partitioning_objective", objective);
double tolerance = 1.1;
params.set("imbalance_tolerance", tolerance );
if (numGlobalParts != nprocs)
params.set("num_global_parts" , numGlobalParts);
if (rank==0){
cout << "Number of parts: " << numGlobalParts << endl;
}
// Create a problem, solve it, and display the quality.
Zoltan2::PartitioningProblem<inputAdapter_t> problem(&(*ia), ¶ms);
problem.solve();
comm->barrier();
problem.printTimers();
comm->barrier();
if (rank == 0){
cout << "PASS" << endl;
}
return 0;
}
/* Find the DBBD form */
int shylu_symbolic_factor
(
Epetra_CrsMatrix *A, // i/p: A matrix
shylu_symbolic *ssym, // symbolic structure
shylu_data *data, // numeric structure, TODO: Required ?
shylu_config *config // i/p: library configuration
)
{
#ifdef TIMING_OUTPUT
Teuchos::Time symtime("symbolic time");
symtime.start();
#endif
int myPID = A->Comm().MyPID();
int n = A->NumGlobalRows();
int Dnr;
int Snr;
int *DRowElems;
int *SRowElems;
int sym = config->sym;
checkMaps(A);
// Get column map
Epetra_Map AColMap = A->ColMap();
int ncols = AColMap.NumMyElements();
int *cols = AColMap.MyGlobalElements();
// Get row map
Epetra_Map ARowMap = A->RowMap();
int nrows = ARowMap.NumMyElements();
int *rows = ARowMap.MyGlobalElements();
// Find all columns in this proc
int *gvals = new int[n]; // vector of size n, not ncols !
// gvals[local cols] = 1, gvals[shared cols] > 1.
int SNumGlobalCols;
findLocalColumns(A, gvals, SNumGlobalCols);
// See if you can shrink the separator by assigning more rows/columns to
// the block diagonals
// TODO: This is because of a bug in coloring remove the if once that is
// fixed
//if (config->schurApproxMethod == 2)
if (config->sep_type == 2)
findNarrowSeparator(A, gvals);
// 3. Assemble diagonal block and the border in convenient form [
/* In each processor, we have (in a permuted form)
* | D_i C_i |
* | R_i S_i |
* D_i - diagonal block, C_i - Column Separator, R_i - Row separator
* S_i - A22 block corresponding to Schur complement part of A
* Assemble all four blocks in local matrices. */
ostringstream ssmsg1;
ssmsg1 << "PID =" << myPID << " ";
string msg = ssmsg1.str();
ssmsg1.clear(); ssmsg1.str("");
// Find #cols in each block
int Dnc = 0; // #cols in diagonal block
int Snc = 0; // #cols in the col. separator
/* Looping on cols will work only for wide separator
* as for narrow sep there will be some sep cols with gvals[col] ==1
* */
/*for (int i = 0; i < ncols ; i++)
{
if (gvals[cols[i]] == 1)
Dnc++;
else
Snc++;
}
// Find #rows in each block
Dnr = Dnc; // #rows in square diagonal block
Snr = nrows - Dnr; // #rows in the row separator*/
// Find #rows in each block
Dnr = 0;
Snr = 0;
for (int i = 0; i < nrows ; i++)
{
if (gvals[rows[i]] == 1)
Dnr++;
else
Snr++;
}
Dnc = Dnr;
// TODO: Snc is no longer useful, should remove it
for (int i = 0; i < ncols ; i++)
{
if (gvals[cols[i]] != 1)
Snc++;
}
assert(Snc >= 0);
// TODO : The above assignment may not be correct in the unsymetric case
////config->dm.print(2, msg + " Mycols=");
cout << msg << " Mycols="<< ncols << "Myrows ="<< nrows << endl;
cout << msg << " #rows and #cols in diagonal blk ="<< Dnr << endl;
cout << msg << " #columns in S ="<< Snc << endl;
cout << msg << " #rows in S ="<< Snr << endl;
ostringstream pidstr;
pidstr << myPID ;
// Create a row map for the D and S blocks [
DRowElems = new int[Dnr];
SRowElems = new int[Snr];
int gid;
// Assemble row ids in two arrays (for D and R blocks)
if (sym)
{
findBlockElems(A, nrows, rows, gvals, Dnr, DRowElems, Snr, SRowElems,
"D"+pidstr.str()+"Rows", "S"+pidstr.str()+"Rows", false) ;
}
else
{
// SRowElems are not known until factorization, TODO
assert(0 == 1);
}
data->Dnr = Dnr;
data->Snr = Snr;
data->Dnc = Dnc;
data->DRowElems = DRowElems;
data->SRowElems = SRowElems;
// Create a column map for the D and S blocks [
int *DColElems = new int[Dnc]; // Elems in column map of D
int *SColElems = new int[Snc]; // Elems in column map of C TODO: Unused
// Assemble column ids in two arrays (for D and C blocks)
findBlockElems(A, ncols, cols, gvals, Dnc, DColElems, Snc, SColElems,
"D"+pidstr.str()+"Cols", "S"+pidstr.str()+"Cols", true) ;
data->DColElems = DColElems;
data->gvals = gvals;
for (int i = 0; i < Snr; i++)
{
// Epetra guarentees columns corresponding to local rows will be first
// in the column map.
assert(SRowElems[i] == SColElems[i]);
}
// ]
/*--Create the Epetra Matrices with the maps (does not insert values) --- */
create_matrices(A, ssym, data, config);
/*--Extract the Epetra Matrices and call fillComplete --- */
extract_matrices(A, ssym, data, config, true);
delete[] SColElems;
Amesos Factory;
const char* SolverType = config->diagonalBlockSolver.c_str();
bool IsAvailable = Factory.Query(SolverType);
assert(IsAvailable == true);
Teuchos::RCP<Epetra_LinearProblem> LP = Teuchos::RCP<Epetra_LinearProblem>
(new Epetra_LinearProblem());
LP->SetOperator((ssym->D).getRawPtr());
//LP->SetOperator((ssym->DT).getRawPtr()); // for transpose
// Create temp vectors
ssym->Dlhs = Teuchos::RCP<Epetra_MultiVector>
(new Epetra_MultiVector(ssym->D->RowMap(), 16));
ssym->Drhs = Teuchos::RCP<Epetra_MultiVector>
(new Epetra_MultiVector(ssym->D->RowMap(), 16));
ssym->Gvec = Teuchos::RCP<Epetra_MultiVector>
(new Epetra_MultiVector(ssym->G->RowMap(), 16));
LP->SetRHS(ssym->Drhs.getRawPtr());
LP->SetLHS(ssym->Dlhs.getRawPtr());
ssym->ReIdx_LP = Teuchos::RCP<
EpetraExt::ViewTransform<Epetra_LinearProblem> >
(new EpetraExt::LinearProblem_Reindex2(0));
ssym->LP = Teuchos::RCP<Epetra_LinearProblem>(&((*(ssym->ReIdx_LP))(*LP)),
false);
Teuchos::RCP<Amesos_BaseSolver> Solver = Teuchos::RCP<Amesos_BaseSolver>
(Factory.Create(SolverType, *(ssym->LP)));
//config->dm.print(5, "Created the diagonal solver");
#ifdef TIMING_OUTPUT
Teuchos::Time ftime("setup time");
ftime.start();
#endif
//Solver->SetUseTranspose(true); // for transpose
Teuchos::ParameterList aList;
aList.set("TrustMe", true);
Solver->SetParameters(aList);
Solver->SymbolicFactorization();
//config->dm.print(3, "Symbolic Factorization done");
#ifdef TIMING_OUTPUT
ftime.stop();
cout << "Symbolic Factorization Time" << ftime.totalElapsedTime() << endl;
ftime.reset();
#endif
ssym->OrigLP = LP;
//ssym->LP = LP;
ssym->Solver = Solver;
if (config->schurApproxMethod == 1)
{
Teuchos::ParameterList pList;
Teuchos::RCP<Isorropia::Epetra::Prober> prober =
Teuchos::RCP<Isorropia::Epetra::Prober> (new
Isorropia::Epetra::Prober((ssym->Sg).getRawPtr(),
pList, false));
//config->dm.print(3, "Doing Coloring");
#ifdef TIMING_OUTPUT
ftime.start();
#endif
prober->color();
#ifdef TIMING_OUTPUT
ftime.stop();
cout << "Time to color" << ftime.totalElapsedTime() << endl;
ftime.reset();
ftime.start();
#endif
ssym->prober = prober;
}
#ifdef TIMING_OUTPUT
symtime.stop();
cout << "Symbolic Time" << symtime.totalElapsedTime() << endl;
symtime.reset();
#endif
}
const RCP<const FactoryBase> FactoryManager<Scalar, LocalOrdinal, GlobalOrdinal, Node>::GetDefaultFactory(const std::string& varName) const {
if (defaultFactoryTable_.count(varName)) {
// The factory for this name was already created (possibly, for previous level, if we reuse factory manager)
return defaultFactoryTable_.find(varName)->second;
} else {
// No factory was created for this name, but we may know which one to create
if (varName == "A") return SetAndReturnDefaultFactory(varName, rcp(new RAPFactory()));
if (varName == "RAP Pattern") return GetFactory("A");
if (varName == "AP Pattern") return GetFactory("A");
if (varName == "Ptent") return SetAndReturnDefaultFactory(varName, rcp(new TentativePFactory()));
if (varName == "P") {
// GetFactory("Ptent"): we need to use the same factory instance for both "P" and "Nullspace"
RCP<Factory> factory = rcp(new SaPFactory());
factory->SetFactory("P", GetFactory("Ptent"));
return SetAndReturnDefaultFactory(varName, factory);
}
if (varName == "Nullspace") {
// GetFactory("Ptent"): we need to use the same factory instance for both "P" and "Nullspace"
RCP<Factory> factory = rcp(new NullspaceFactory());
factory->SetFactory("Nullspace", GetFactory("Ptent"));
return SetAndReturnDefaultFactory(varName, factory);
}
if (varName == "R") return SetAndReturnDefaultFactory(varName, rcp(new TransPFactory()));
#if defined(HAVE_MUELU_ZOLTAN) && defined(HAVE_MPI)
if (varName == "Partition") return SetAndReturnDefaultFactory(varName, rcp(new ZoltanInterface()));
#endif //ifdef HAVE_MPI
if (varName == "Importer") {
#ifdef HAVE_MPI
return SetAndReturnDefaultFactory(varName, rcp(new RepartitionFactory()));
#else
return SetAndReturnDefaultFactory(varName, NoFactory::getRCP());
#endif
}
if (varName == "Graph") return SetAndReturnDefaultFactory(varName, rcp(new CoalesceDropFactory()));
if (varName == "UnAmalgamationInfo") return SetAndReturnDefaultFactory(varName, rcp(new AmalgamationFactory())); //GetFactory("Graph"));
if (varName == "Aggregates") return SetAndReturnDefaultFactory(varName, rcp(new UncoupledAggregationFactory()));
if (varName == "CoarseMap") return SetAndReturnDefaultFactory(varName, rcp(new CoarseMapFactory()));
if (varName == "DofsPerNode") return GetFactory("Graph");
if (varName == "Filtering") return GetFactory("Graph");
if (varName == "LineDetection_VertLineIds") return SetAndReturnDefaultFactory(varName, rcp(new LineDetectionFactory()));
if (varName == "LineDetection_Layers") return GetFactory("LineDetection_VertLineIds");
if (varName == "CoarseNumZLayers") return GetFactory("LineDetection_VertLineIds");
// Same factory for both Pre and Post Smoother. Factory for key "Smoother" can be set by users.
if (varName == "PreSmoother") return GetFactory("Smoother");
if (varName == "PostSmoother") return GetFactory("Smoother");
if (varName == "Ppattern") {
RCP<PatternFactory> PpFact = rcp(new PatternFactory);
PpFact->SetFactory("P", GetFactory("Ptent"));
return SetAndReturnDefaultFactory(varName, PpFact);
}
if (varName == "Constraint") return SetAndReturnDefaultFactory(varName, rcp(new ConstraintFactory()));
if (varName == "Smoother") {
Teuchos::ParameterList smootherParamList;
smootherParamList.set("relaxation: type", "Symmetric Gauss-Seidel");
smootherParamList.set("relaxation: sweeps", Teuchos::OrdinalTraits<LO>::one());
smootherParamList.set("relaxation: damping factor", Teuchos::ScalarTraits<Scalar>::one());
return SetAndReturnDefaultFactory(varName, rcp(new SmootherFactory(rcp(new TrilinosSmoother("RELAXATION", smootherParamList)))));
}
if (varName == "CoarseSolver") return SetAndReturnDefaultFactory(varName, rcp(new SmootherFactory(rcp(new DirectSolver()), Teuchos::null)));
TEUCHOS_TEST_FOR_EXCEPTION(true, MueLu::Exceptions::RuntimeError, "MueLu::FactoryManager::GetDefaultFactory(): No default factory available for building '" + varName + "'.");
}
}
int main(int argc, char *argv[]) {
#include <MueLu_UseShortNames.hpp>
using Teuchos::RCP; // reference count pointers
using Teuchos::rcp;
using Teuchos::TimeMonitor;
// =========================================================================
// MPI initialization using Teuchos
// =========================================================================
Teuchos::GlobalMPISession mpiSession(&argc, &argv, NULL);
RCP< const Teuchos::Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
int MyPID = comm->getRank();
int NumProc = comm->getSize();
const Teuchos::RCP<Epetra_Comm> epComm = Teuchos::rcp_const_cast<Epetra_Comm>(Xpetra::toEpetra(comm));
// =========================================================================
// Convenient definitions
// =========================================================================
//SC zero = Teuchos::ScalarTraits<SC>::zero(), one = Teuchos::ScalarTraits<SC>::one();
// Instead of checking each time for rank, create a rank 0 stream
RCP<Teuchos::FancyOStream> fancy = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout));
Teuchos::FancyOStream& fancyout = *fancy;
fancyout.setOutputToRootOnly(0);
// =========================================================================
// Parameters initialization
// =========================================================================
Teuchos::CommandLineProcessor clp(false);
GO nx = 100, ny = 100;
clp.setOption("nx", &nx, "mesh size in x direction");
clp.setOption("ny", &ny, "mesh size in y direction");
std::string xmlFileName = "xml/s3a.xml"; clp.setOption("xml", &xmlFileName, "read parameters from a file. Otherwise, this example uses by default 'tutorial1a.xml'");
int mgridSweeps = 1; clp.setOption("mgridSweeps", &mgridSweeps, "number of multigrid sweeps within Multigrid solver.");
std::string printTimings = "no"; clp.setOption("timings", &printTimings, "print timings to screen [yes/no]");
double tol = 1e-12; clp.setOption("tol", &tol, "solver convergence tolerance");
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;
}
// =========================================================================
// Problem construction
// =========================================================================
RCP<TimeMonitor> globalTimeMonitor = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: S - Global Time"))), tm;
comm->barrier();
tm = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 1 - Matrix Build")));
Teuchos::ParameterList GaleriList;
GaleriList.set("nx", nx);
GaleriList.set("ny", ny);
GaleriList.set("mx", epComm->NumProc());
GaleriList.set("my", 1);
GaleriList.set("lx", 1.0); // length of x-axis
GaleriList.set("ly", 1.0); // length of y-axis
GaleriList.set("diff", 1e-5);
GaleriList.set("conv", 1.0);
// create map
Teuchos::RCP<Epetra_Map> epMap = Teuchos::rcp(Galeri::CreateMap("Cartesian2D", *epComm, GaleriList));
// create coordinates
Teuchos::RCP<Epetra_MultiVector> epCoord = Teuchos::rcp(Galeri::CreateCartesianCoordinates("2D", epMap.get(), GaleriList));
// create matrix
Teuchos::RCP<Epetra_CrsMatrix> epA = Teuchos::rcp(Galeri::CreateCrsMatrix("Recirc2D", epMap.get(), GaleriList));
// Epetra -> Xpetra
Teuchos::RCP<CrsMatrix> exA = Teuchos::rcp(new Xpetra::EpetraCrsMatrix(epA));
Teuchos::RCP<CrsMatrixWrap> exAWrap = Teuchos::rcp(new CrsMatrixWrap(exA));
RCP<Matrix> A = Teuchos::rcp_dynamic_cast<Matrix>(exAWrap);
int numPDEs = 1;
A->SetFixedBlockSize(numPDEs);
// set rhs and solution vector
RCP<Epetra_Vector> B = Teuchos::rcp(new Epetra_Vector(*epMap));
RCP<Epetra_Vector> X = Teuchos::rcp(new Epetra_Vector(*epMap));
B->PutScalar(1.0);
X->PutScalar(0.0);
// Epetra -> Xpetra
RCP<Vector> xB = Teuchos::rcp(new Xpetra::EpetraVector(B));
RCP<Vector> xX = Teuchos::rcp(new Xpetra::EpetraVector(X));
xX->setSeed(100);
xX->randomize();
// build null space vector
RCP<const Map> map = A->getRowMap();
RCP<MultiVector> nullspace = MultiVectorFactory::Build(map, numPDEs);
for (int i=0; i<numPDEs; ++i) {
Teuchos::ArrayRCP<Scalar> nsValues = nullspace->getDataNonConst(i);
int numBlocks = nsValues.size() / numPDEs;
for (int j=0; j< numBlocks; ++j) {
nsValues[j*numPDEs + i] = 1.0;
}
}
comm->barrier();
tm = Teuchos::null;
fancyout << "========================================================\nGaleri complete.\n========================================================" << std::endl;
// =========================================================================
// Preconditioner construction
// =========================================================================
comm->barrier();
tm = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 1.5 - MueLu read XML")));
ParameterListInterpreter mueLuFactory(xmlFileName, *comm);
comm->barrier();
tm = Teuchos::null;
tm = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 2 - MueLu Setup")));
RCP<Hierarchy> H = mueLuFactory.CreateHierarchy();
H->GetLevel(0)->Set("A", A);
H->GetLevel(0)->Set("Nullspace", nullspace);
mueLuFactory.SetupHierarchy(*H);
comm->barrier();
tm = Teuchos::null;
// =========================================================================
// System solution (Ax = b)
// =========================================================================
//
// generate exact solution using a direct solver
//
RCP<Epetra_Vector> exactLsgVec = rcp(new Epetra_Vector(X->Map()));
{
fancyout << "========================================================\nCalculate exact solution." << std::endl;
tm = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 3 - direct solve")));
exactLsgVec->PutScalar(0.0);
exactLsgVec->Update(1.0,*X,1.0);
Epetra_LinearProblem epetraProblem(epA.get(), exactLsgVec.get(), B.get());
Amesos amesosFactory;
RCP<Amesos_BaseSolver> rcp_directSolver = Teuchos::rcp(amesosFactory.Create("Amesos_Klu", epetraProblem));
rcp_directSolver->SymbolicFactorization();
rcp_directSolver->NumericFactorization();
rcp_directSolver->Solve();
comm->barrier();
tm = Teuchos::null;
}
//
// Solve Ax = b using AMG as a preconditioner in AztecOO
//
RCP<Epetra_Vector> precLsgVec = rcp(new Epetra_Vector(X->Map()));
{
fancyout << "========================================================\nUse multigrid hierarchy as preconditioner within CG." << std::endl;
tm = rcp(new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 4 - AMG as preconditioner")));
precLsgVec->PutScalar(0.0);
precLsgVec->Update(1.0,*X,1.0);
Epetra_LinearProblem epetraProblem(epA.get(), precLsgVec.get(), B.get());
AztecOO aztecSolver(epetraProblem);
aztecSolver.SetAztecOption(AZ_solver, AZ_gmres);
MueLu::EpetraOperator aztecPrec(H);
aztecSolver.SetPrecOperator(&aztecPrec);
int maxIts = 50;
aztecSolver.Iterate(maxIts, tol);
comm->barrier();
tm = Teuchos::null;
}
//////////////////
// use multigrid hierarchy as solver
RCP<Vector> mgridLsgVec = VectorFactory::Build(map);
mgridLsgVec->putScalar(0.0);
{
fancyout << "========================================================\nUse multigrid hierarchy as solver." << std::endl;
tm = rcp (new TimeMonitor(*TimeMonitor::getNewTimer("ScalingTest: 5 - Multigrid Solve")));
mgridLsgVec->update(1.0,*xX,1.0);
H->IsPreconditioner(false);
H->Iterate(*xB, mgridSweeps, *mgridLsgVec);
comm->barrier();
tm = Teuchos::null;
}
//////////////////
fancyout << "========================================================\nExport results.\n========================================================" << std::endl;
std::ofstream myfile;
std::stringstream ss; ss << "example" << MyPID << ".txt";
myfile.open (ss.str().c_str());
//////////////////
// loop over all procs
for (int iproc=0; iproc < NumProc; iproc++) {
if (MyPID==iproc) {
int NumVectors1 = 2;
int NumMyElements1 = epCoord->Map(). NumMyElements();
int MaxElementSize1 = epCoord->Map().MaxElementSize();
int * FirstPointInElementList1 = NULL;
if (MaxElementSize1!=1) FirstPointInElementList1 = epCoord->Map().FirstPointInElementList();
double ** A_Pointers = epCoord->Pointers();
if (MyPID==0) {
myfile.width(8);
myfile << "# MyPID"; myfile << " ";
myfile.width(12);
if (MaxElementSize1==1)
myfile << "GID ";
else
myfile << " GID/Point";
for (int j = 0; j < NumVectors1 ; j++)
{
myfile.width(20);
myfile << "Value ";
}
myfile << std::endl;
}
for (int i=0; i < NumMyElements1; i++) {
for (int ii=0; ii< epCoord->Map().ElementSize(i); ii++) {
int iii;
myfile.width(10);
myfile << MyPID; myfile << " ";
myfile.width(10);
if (MaxElementSize1==1) {
if(epCoord->Map().GlobalIndicesInt())
{
int * MyGlobalElements1 = epCoord->Map().MyGlobalElements();
myfile << MyGlobalElements1[i] << " ";
}
iii = i;
}
else {
if(epCoord->Map().GlobalIndicesInt())
{
int * MyGlobalElements1 = epCoord->Map().MyGlobalElements();
myfile << MyGlobalElements1[i]<< "/" << ii << " ";
}
iii = FirstPointInElementList1[i]+ii;
}
for (int j = 0; j < NumVectors1 ; j++)
{
myfile.width(20);
myfile << A_Pointers[j][iii];
}
myfile.precision(18); // set high precision for output
// add solution vector entry
myfile.width(25); myfile << (*exactLsgVec)[iii];
// add preconditioned solution vector entry
myfile.width(25); myfile << (*precLsgVec)[iii];
Teuchos::ArrayRCP<SC> mgridLsgVecData = mgridLsgVec->getDataNonConst(0);
myfile.width(25); myfile << mgridLsgVecData[iii];
myfile.precision(6); // set default precision
myfile << std::endl;
}
} // end loop over all lines on current proc
myfile << std::flush;
// syncronize procs
comm->barrier();
comm->barrier();
comm->barrier();
} // end myProc
}
////////////
myfile.close();
comm->barrier();
tm = Teuchos::null;
globalTimeMonitor = Teuchos::null;
if (printTimings == "yes") {
TimeMonitor::summarize(A->getRowMap()->getComm().ptr(), std::cout, false, true, false, Teuchos::Union);
}
return 0;
} //main
void testInitialzation(const Teuchos::RCP<Teuchos::ParameterList>& ipb,
std::vector<panzer::BC>& bcs)
{
// Physics block
Teuchos::ParameterList& physics_block = ipb->sublist("test physics");
{
Teuchos::ParameterList& p = physics_block.sublist("a");
p.set("Type","Energy");
p.set("Prefix","");
p.set("Model ID","solid");
p.set("Basis Type","HGrad");
p.set("Basis Order",2);
}
{
Teuchos::ParameterList& p = physics_block.sublist("b");
p.set("Type","Energy");
p.set("Prefix","ION_");
p.set("Model ID","solid");
p.set("Basis Type","HGrad");
p.set("Basis Order",1);
}
{
std::size_t bc_id = 0;
panzer::BCType neumann = BCT_Dirichlet;
std::string sideset_id = "left";
std::string element_block_id = "eblock-0_0";
std::string dof_name = "TEMPERATURE";
std::string strategy = "Constant";
double value = 5.0;
Teuchos::ParameterList p;
p.set("Value",value);
panzer::BC bc(bc_id, neumann, sideset_id, element_block_id, dof_name,
strategy, p);
bcs.push_back(bc);
}
{
std::size_t bc_id = 1;
panzer::BCType neumann = BCT_Dirichlet;
std::string sideset_id = "right";
std::string element_block_id = "eblock-1_0";
std::string dof_name = "TEMPERATURE";
std::string strategy = "Constant";
double value = 5.0;
Teuchos::ParameterList p;
p.set("Value",value);
panzer::BC bc(bc_id, neumann, sideset_id, element_block_id, dof_name,
strategy, p);
bcs.push_back(bc);
}
{
std::size_t bc_id = 2;
panzer::BCType neumann = BCT_Dirichlet;
std::string sideset_id = "top";
std::string element_block_id = "eblock-1_0";
std::string dof_name = "TEMPERATURE";
std::string strategy = "Constant";
double value = 5.0;
Teuchos::ParameterList p;
p.set("Value",value);
panzer::BC bc(bc_id, neumann, sideset_id, element_block_id, dof_name,
strategy, p);
bcs.push_back(bc);
}
}
TEUCHOS_UNIT_TEST(scatter_field_evaluators, cell_field)
{
const std::size_t workset_size = 5;
linBasis = buildLinearBasis(workset_size);
Teuchos::RCP<panzer_stk::STK_Interface> mesh = buildMesh(5,5,false);
RCP<Epetra_Comm> Comm = Teuchos::rcp(new Epetra_MpiComm(MPI_COMM_WORLD));
Teuchos::RCP<shards::CellTopology> topo =
Teuchos::rcp(new shards::CellTopology(shards::getCellTopologyData< shards::Quadrilateral<4> >()));
panzer::CellData cellData(workset_size,topo);
Teuchos::RCP<panzer::IntegrationRule> intRule = Teuchos::rcp(new panzer::IntegrationRule(1,cellData));
Teuchos::RCP<Teuchos::ParameterList> ipb = Teuchos::parameterList("Physics Blocks");
std::vector<panzer::BC> bcs;
testInitialzation(ipb, bcs);
Teuchos::RCP<PHX::FieldManager<panzer::Traits> > fm =
Teuchos::rcp(new PHX::FieldManager<panzer::Traits>);
{
Teuchos::ParameterList pl;
pl.set("Data Layout",linBasis->functional);
fm->registerEvaluator<panzer::Traits::Residual>(Teuchos::rcp(new XCoordinate<panzer::Traits::Residual,panzer::Traits>(pl)));
}
{
Teuchos::ParameterList pl;
pl.set("Nodes",1);
pl.set("Data Layout",intRule->dl_scalar);
fm->registerEvaluator<panzer::Traits::Residual>(Teuchos::rcp(new XCoordinate<panzer::Traits::Residual,panzer::Traits>(pl)));
}
{
Teuchos::RCP<std::vector<std::string> > fieldNames
= Teuchos::rcp(new std::vector<std::string>);
fieldNames->push_back("x-coord");
Teuchos::ParameterList pl;
pl.set("Mesh",mesh);
pl.set("IR",intRule);
pl.set("Field Names",fieldNames);
pl.set("Scatter Name", "xcoord-scatter-cell-residual");
Teuchos::RCP<PHX::Evaluator<panzer::Traits> > eval
= Teuchos::rcp(new panzer_stk::ScatterCellAvgQuantity<panzer::Traits::Residual,panzer::Traits>(pl));
fm->registerEvaluator<panzer::Traits::Residual>(eval);
fm->requireField<panzer::Traits::Residual>(*eval->evaluatedFields()[0]);
}
{
Teuchos::RCP<std::vector<std::string> > fieldNames
= Teuchos::rcp(new std::vector<std::string>);
fieldNames->push_back("x-coord");
std::string scatterName = "xcoord-scatter-residual";
Teuchos::RCP<PHX::Evaluator<panzer::Traits> > eval
= Teuchos::rcp(new panzer_stk::ScatterFields<panzer::Traits::Residual,panzer::Traits>(scatterName,mesh,linBasis,*fieldNames));
fm->registerEvaluator<panzer::Traits::Residual>(eval);
fm->requireField<panzer::Traits::Residual>(*eval->evaluatedFields()[0]);
}
// build physics blocks
//////////////////////////////////////////////////////////////
std::vector<Teuchos::RCP<panzer::PhysicsBlock> > physicsBlocks;
{
Teuchos::RCP<user_app::MyFactory> eqset_factory = Teuchos::rcp(new user_app::MyFactory);
user_app::BCFactory bc_factory;
const int default_integration_order = 1;
std::map<std::string,std::string> block_ids_to_physics_ids;
block_ids_to_physics_ids["eblock-0_0"] = "test physics";
std::map<std::string,Teuchos::RCP<const shards::CellTopology> > block_ids_to_cell_topo;
block_ids_to_cell_topo["eblock-0_0"] = mesh->getCellTopology("eblock-0_0");
Teuchos::RCP<panzer::GlobalData> gd = panzer::createGlobalData();
panzer::buildPhysicsBlocks(block_ids_to_physics_ids,
block_ids_to_cell_topo,
ipb,
default_integration_order,
workset_size,
eqset_factory,
gd,
false,
physicsBlocks);
}
// register jacobian size
//////////////////////////////////////////////////////////////
std::vector<PHX::index_size_type> derivative_dimensions;
derivative_dimensions.push_back(9+4);
fm->setKokkosExtendedDataTypeDimensions<panzer::Traits::Jacobian>(derivative_dimensions);
// build worksets
//////////////////////////////////////////////////////////////
Teuchos::RCP<panzer::PhysicsBlock> physics_block_one = panzer::findPhysicsBlock("eblock-0_0",physicsBlocks);
Teuchos::RCP<std::vector<panzer::Workset> > volume_worksets = panzer_stk::buildWorksets(*mesh,*physics_block_one);
panzer::Traits::SetupData sd;
sd.worksets_ = volume_worksets;
fm->postRegistrationSetupForType<panzer::Traits::Residual>(sd);
fm->writeGraphvizFile<panzer::Traits::Residual>("resi-eval-graph.dot");
std::vector<panzer::Workset> & worksets = *volume_worksets;
panzer::Traits::PreEvalData preEvalData;
fm->preEvaluate<panzer::Traits::Residual>(preEvalData);
for(std::size_t ws=0;ws<worksets.size();ws++) {
fm->evaluateFields<panzer::Traits::Residual>(worksets[ws]);
}
fm->postEvaluate<panzer::Traits::Residual>(0);
if(mesh->isWritable())
mesh->writeToExodus("x-coord-cell.exo");
}
QCAD::ResponseCenterOfMass<EvalT, Traits>::
ResponseCenterOfMass(Teuchos::ParameterList& p,
const Teuchos::RCP<Albany::Layouts>& dl) :
coordVec("Coord Vec", dl->qp_vector),
weights("Weights", dl->qp_scalar)
{
// get and validate Response parameter list
Teuchos::ParameterList* plist =
p.get<Teuchos::ParameterList*>("Parameter List");
Teuchos::RCP<const Teuchos::ParameterList> reflist =
this->getValidResponseParameters();
plist->validateParameters(*reflist,0);
// number of quad points per cell and dimension of space
Teuchos::RCP<PHX::DataLayout> scalar_dl = dl->qp_scalar;
Teuchos::RCP<PHX::DataLayout> vector_dl = dl->qp_vector;
std::vector<PHX::DataLayout::size_type> dims;
vector_dl->dimensions(dims);
numQPs = dims[1];
numDims = dims[2];
//! Get material DB from parameters passed down from problem (if given)
Teuchos::RCP<QCAD::MaterialDatabase> materialDB;
Teuchos::RCP<Teuchos::ParameterList> paramsFromProblem =
p.get< Teuchos::RCP<Teuchos::ParameterList> >("Parameters From Problem");
if(paramsFromProblem != Teuchos::null)
materialDB = paramsFromProblem->get< Teuchos::RCP<QCAD::MaterialDatabase> >("MaterialDB");
else materialDB = Teuchos::null;
// User-specified parameters
fieldName = plist->get<std::string>("Field Name");
opRegion = Teuchos::rcp( new QCAD::MeshRegion<EvalT, Traits>("Coord Vec","Weights",*plist,materialDB,dl) );
// setup field
PHX::MDField<ScalarT> f(fieldName, scalar_dl); field = f;
// add dependent fields
this->addDependentField(field);
this->addDependentField(coordVec);
this->addDependentField(weights);
opRegion->addDependentFields(this);
this->setName(fieldName+" Response Center of Mass" );
using PHX::MDALayout;
// Setup scatter evaluator
p.set("Stand-alone Evaluator", false);
std::string local_response_name =
fieldName + " Local Response Center of Mass";
std::string global_response_name =
fieldName + " Global Response Center of Mass";
int worksetSize = scalar_dl->dimension(0);
int responseSize = 4;
Teuchos::RCP<PHX::DataLayout> local_response_layout =
Teuchos::rcp(new MDALayout<Cell,Dim>(worksetSize, responseSize));
Teuchos::RCP<PHX::DataLayout> global_response_layout =
Teuchos::rcp(new MDALayout<Dim>(responseSize));
PHX::Tag<ScalarT> local_response_tag(local_response_name,
local_response_layout);
PHX::Tag<ScalarT> global_response_tag(global_response_name,
global_response_layout);
p.set("Local Response Field Tag", local_response_tag);
p.set("Global Response Field Tag", global_response_tag);
PHAL::SeparableScatterScalarResponse<EvalT,Traits>::setup(p,dl);
}
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) {
#if defined(HAVE_MPI) && defined(HAVE_EPETRA)
int numProcs = 1;
int localProc = 0;
//first, set up our MPI environment...
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &localProc);
MPI_Comm_size(MPI_COMM_WORLD, &numProcs);
// This program can only run on 3 processors, to make things
// work out easy.
//
if (numProcs != 3) {
std::cout << "num-procs="<<numProcs<<". This program can only "
<< "run on 3 procs. Exiting."<<std::endl;
MPI_Finalize();
return(0);
}
//Consider the following mesh of 4 2-D quad elements:
//
// *-------*-------*
// 8| 7| 6|
// | E2 | E3 |
// *-------*-------*
// 3| 2| 5|
// | E0 | E1 |
// *-------*-------*
// 0 1 4
//
// Node-ids are to the lower-left of each node (*).
//
// Mimicing a finite-element application, we will say that
// each node has 1 scalar degree-of-freedom, and assemble
// a matrix which would have 9 global rows and columns.
//
// Each processor will have 3 rows. We'll set up a strange
// initial map, where nodes are distributed as follows:
//
// proc 0: nodes 0,3,8,
// proc 1: nodes 1,2,7
// proc 2: nodes 4,5,6.
//
// After we assemble our matrix, we'll create another matrix
// and populate it with graph edge weights such that the
// partitioner repartitions the problem so that nodes are
// laid out as follows:
//
// proc 0: nodes 0, 1, 4
// proc 1: nodes 3, 2, 5
// proc 2: nodes 8, 7, 6
//
int nodesPerElem = 4;
int global_n = 9;
//First, set up the initial map:
std::vector<int> mynodes(3);
if (localProc == 0) {
mynodes[0] = 0; mynodes[1] = 3; mynodes[2] = 8;
}
if (localProc == 1) {
mynodes[0] = 1; mynodes[1] = 2; mynodes[2] = 7;
}
if (localProc == 2) {
mynodes[0] = 4; mynodes[1] = 5; mynodes[2] = 6;
}
Epetra_MpiComm comm(MPI_COMM_WORLD);
Epetra_Map origmap(global_n, 3, &mynodes[0], 0, comm);
Teuchos::RCP<Epetra_FECrsMatrix> matrix =
Teuchos::rcp(new Epetra_FECrsMatrix(Copy, origmap, 0));
//We'll assemble elements E0 and E1 on proc 0,
// element E2 or proc 1,
// element E3 on proc 2.
std::vector<int> indices(nodesPerElem);
std::vector<double> coefs(nodesPerElem*nodesPerElem,2.0);
if (localProc == 0) {
//element E0:
indices[0] = 0; indices[1] = 1; indices[2] = 2; indices[3] = 3;
matrix->InsertGlobalValues(nodesPerElem, &indices[0], &coefs[0]);
//element E1:
indices[0] = 1; indices[1] = 4; indices[2] = 5; indices[3] = 2;
matrix->InsertGlobalValues(nodesPerElem, &indices[0], &coefs[0]);
}
else if (localProc == 1) {
//element E2:
indices[0] = 3; indices[1] = 2; indices[2] = 7; indices[3] = 8;
matrix->InsertGlobalValues(nodesPerElem, &indices[0], &coefs[0]);
}
else { //localProc==2
//element E3:
indices[0] = 2; indices[1] = 5; indices[2] = 6; indices[3] = 7;
matrix->InsertGlobalValues(nodesPerElem, &indices[0], &coefs[0]);
}
int err = matrix->GlobalAssemble();
if (err != 0) {
std::cout << "err="<<err<<" returned from matrix->GlobalAssemble()"
<< std::endl;
}
// std::cout << "matrix: " << std::endl;
// std::cout << *matrix << std::endl;
//We'll need a Teuchos::ParameterList object to pass to the
//Isorropia::Epetra::Partitioner class.
Teuchos::ParameterList paramlist;
#ifdef HAVE_ISORROPIA_ZOLTAN
// If Zoltan is available, we'll specify that the Zoltan package be
// used for the partitioning operation, by creating a parameter
// sublist named "Zoltan".
// In the sublist, we'll set parameters that we want sent to Zoltan.
paramlist.set("PARTITIONING METHOD", "GRAPH");
paramlist.set("PRINT ZOLTAN METRICS", "2");
Teuchos::ParameterList& sublist = paramlist.sublist("Zoltan");
sublist.set("GRAPH_PACKAGE", "PHG");
//sublist.set("DEBUG_LEVEL", "1"); // Zoltan will print out parameters
//sublist.set("DEBUG_LEVEL", "5"); // proc 0 will trace Zoltan calls
//sublist.set("DEBUG_MEMORY", "2"); // Zoltan will trace alloc & free
#else
// If Zoltan is not available, a simple linear partitioner will be
// used to partition such that the number of nonzeros is equal (or
// close to equal) on each processor. No parameter is necessary to
// specify this.
#endif
Teuchos::RCP<Isorropia::Epetra::CostDescriber> costs =
Teuchos::rcp(new Isorropia::Epetra::CostDescriber);
//Next create a matrix which is a copy of the matrix we just
//assembled, but we'll replace the values with graph edge weights.
Teuchos::RCP<Epetra_FECrsMatrix> ge_weights =
Teuchos::rcp(new Epetra_FECrsMatrix(*matrix));
Teuchos::RCP<Epetra_CrsMatrix> crs_ge_weights;
crs_ge_weights = ge_weights;
//Fill the matrix with a "default" weight of 1.0.
crs_ge_weights->PutScalar(1.0);
//Now we'll put a "large" weight on edges that connect nodes
//0 and 1, 1 and 4,
//3 and 2, 2 and 5,
//8 and 7, 7 and 6.
double weight = 500.0;
if (localProc == 0) {
//row 0, edge 1
indices[0] = 1;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(0, 1, &coefs[0], &indices[0]);
//row 3, edge 2
indices[0] = 2;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(3, 1, &coefs[0], &indices[0]);
//row 8, edge 7
indices[0] = 7;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(8, 1, &coefs[0], &indices[0]);
}
if (localProc == 1) {
//row 1, edges 0 and 4
indices[0] = 0; indices[1] = 4;
coefs[0] = weight; coefs[1] = weight;
crs_ge_weights->ReplaceGlobalValues(1, 2, &coefs[0], &indices[0]);
//row 2, edges 3 and 5
indices[0] = 3; indices[1] = 5;
coefs[0] = weight; coefs[1] = weight;
crs_ge_weights->ReplaceGlobalValues(2, 2, &coefs[0], &indices[0]);
//row 7, edges 6 and 8
indices[0] = 6; indices[1] = 8;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(7, 2, &coefs[0], &indices[0]);
}
if (localProc == 2) {
//row 4, edge 1
indices[0] = 1;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(4, 1, &coefs[0], &indices[0]);
//row 5, edge 2
indices[0] = 2;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(5, 1, &coefs[0], &indices[0]);
//row 6, edge 7
indices[0] = 7;
coefs[0] = weight;
crs_ge_weights->ReplaceGlobalValues(6, 1, &coefs[0], &indices[0]);
}
// std::cout << "crs_ge_weights: " << std::endl
// << *crs_ge_weights << std::endl;
//Now give the graph edge weights to the CostDescriber:
costs->setGraphEdgeWeights(crs_ge_weights);
Teuchos::RCP<const Epetra_RowMatrix> rowmatrix;
rowmatrix = matrix;
//Now create the partitioner object using an Isorropia factory-like
//function...
Teuchos::RCP<Isorropia::Epetra::Partitioner> partitioner =
Teuchos::rcp(new Isorropia::Epetra::Partitioner(rowmatrix, costs, paramlist));
//Next create a Redistributor object and use it to create a
//repartitioned copy of the matrix
Isorropia::Epetra::Redistributor rd(partitioner);
Teuchos::RCP<Epetra_CrsMatrix> bal_matrix;
//Use a try-catch block because Isorropia will throw an exception
//if it encounters an error.
if (localProc == 0) {
std::cout << " calling Isorropia::Epetra::Redistributor::redistribute..."
<< std::endl;
}
try {
bal_matrix = rd.redistribute(*rowmatrix);
}
catch(std::exception& exc) {
std::cout << "linsys example: Isorropia::Epetra::Redistributor threw "
<< "exception '" << exc.what() << "' on proc "
<< localProc << std::endl;
MPI_Finalize();
return(-1);
}
// Results
double bal0, bal1, cutn0, cutn1, cutl0, cutl1, cutWgt0, cutWgt1;
int numCuts0, numCuts1;
#if 1
// Balance and cut quality before partitioning
double goalWeight = 1.0 / (double)numProcs;
ispatest::compute_graph_metrics(*rowmatrix, *costs, goalWeight,
bal0, numCuts0, cutWgt0, cutn0, cutl0);
// Balance and cut quality after partitioning
Teuchos::RCP<Epetra_CrsMatrix> new_weights = rd.redistribute(*crs_ge_weights);
Isorropia::Epetra::CostDescriber new_costs;
new_costs.setGraphEdgeWeights(new_weights);
ispatest::compute_graph_metrics(*bal_matrix, new_costs, goalWeight,
bal1, numCuts1, cutWgt1, cutn1, cutl1);
#else
std::vector<double> bal(2), cutwgt(2), cutn(2), cutl(2);
std::vector<int >ncuts(2);
Epetra_Import &importer = rd.get_importer();
costs->compareBeforeAndAfterGraph(*rowmatrix, *bal_matrix, importer,
bal, ncuts, cutwgt, cutn, cutl);
bal0 = bal[0]; cutn0 = cutn[0]; cutl0 = cutl[0]; cutWgt0 = cutwgt[0]; numCuts0 = ncuts[0];
bal1 = bal[1]; cutn1 = cutn[1]; cutl1 = cutl[1]; cutWgt1 = cutwgt[1]; numCuts1 = ncuts[1];
#endif
bal_matrix.release();
if (localProc == 0){
std::cout << "Before partitioning: Number of cuts " << numCuts0 << " Cut weight " << cutWgt0 << std::endl;
std::cout << " Balance " << bal0 << " cutN " << cutn0 << " cutL " << cutl0;
std::cout << std::endl;
std::cout << "After partitioning: Number of cuts " << numCuts1 << " Cut weight " << cutWgt1 << std::endl;
std::cout << " Balance " << bal1 << " cutN " << cutn1 << " cutL " << cutl1;
std::cout << std::endl;
}
MPI_Finalize();
#else
std::cout << "part_redist: must have both MPI and EPETRA. Make sure Trilinos "
<< "is configured with --enable-mpi and --enable-epetra." << std::endl;
#endif
return(0);
}
void
Albany::ResponseFactory::
createResponseFunction(
const std::string& name,
Teuchos::ParameterList& responseParams,
Teuchos::Array< Teuchos::RCP<AbstractResponseFunction> >& responses) const
{
using std::string;
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::ParameterList;
using Teuchos::Array;
RCP<const Teuchos_Comm> comm = app->getComm();
if (name == "Solution Average") {
responses.push_back(rcp(new Albany::SolutionAverageResponseFunction(comm)));
}
else if (name == "Solution Two Norm") {
responses.push_back(rcp(new Albany::SolutionTwoNormResponseFunction(comm)));
}
else if (name == "Solution Values") {
responses.push_back(rcp(new Albany::SolutionValuesResponseFunction(app, responseParams)));
}
else if (name == "Solution Max Value") {
int eq = responseParams.get("Equation", 0);
int neq = responseParams.get("Num Equations", 1);
bool inor = responseParams.get("Interleaved Ordering", true);
responses.push_back(
rcp(new Albany::SolutionMaxValueResponseFunction(comm, neq, eq, inor)));
}
else if (name == "Solution Two Norm File") {
responses.push_back(
rcp(new Albany::SolutionFileResponseFunction<Albany::NormTwo>(comm)));
}
else if (name == "Solution Inf Norm File") {
responses.push_back(
rcp(new Albany::SolutionFileResponseFunction<Albany::NormInf>(comm)));
}
else if (name == "OBC Functional") {
#ifdef ALBANY_PERIDIGM
#ifdef ALBANY_EPETRA
responses.push_back(rcp(new Albany::AlbanyPeridigmOBCFunctional(comm)));
#endif
#endif
}
else if (name == "Aggregated") {
int num_aggregate_responses = responseParams.get<int>("Number");
Array< RCP<AbstractResponseFunction> > aggregated_responses;
Array< RCP<ScalarResponseFunction> > scalar_responses;
for (int i=0; i<num_aggregate_responses; i++) {
std::string id = Albany::strint("Response",i);
std::string name = responseParams.get<std::string>(id);
std::string sublist_name = Albany::strint("ResponseParams",i);
createResponseFunction(name, responseParams.sublist(sublist_name),
aggregated_responses);
}
scalar_responses.resize(aggregated_responses.size());
for (int i=0; i<aggregated_responses.size(); i++) {
TEUCHOS_TEST_FOR_EXCEPTION(
aggregated_responses[i]->isScalarResponse() != true, std::logic_error,
"Response function " << i << " is not a scalar response function." <<
std::endl <<
"The aggregated response can only aggregate scalar response " <<
"functions!");
scalar_responses[i] =
Teuchos::rcp_dynamic_cast<ScalarResponseFunction>(
aggregated_responses[i]);
}
responses.push_back(
rcp(new Albany::AggregateScalarResponseFunction(comm, scalar_responses)));
}
else if (name == "Field Integral" ||
name == "Field Value" ||
name == "Field Average" ||
name == "Surface Velocity Mismatch" ||
name == "Aeras Shallow Water L2 Error" ||
name == "Aeras Total Volume" ||
name == "Center Of Mass" ||
name == "Save Field" ||
name == "Region Boundary" ||
name == "Element Size Field" ||
name == "Save Nodal Fields" ||
name == "Stiffness Objective" ||
name == "Internal Energy Objective" ||
name == "Tensor PNorm Objective" ||
name == "Homogenized Constants Response" ||
name == "Modal Objective" ||
name == "PHAL Field Integral" ||
name == "PHAL Field IntegralT") {
responseParams.set("Name", name);
for (int i=0; i<meshSpecs.size(); i++) {
#if defined(ALBANY_LCM)
// Skip if dealing with interface block
//if (meshSpecs[i]->ebName == "Surface Element") continue;
#endif
responses.push_back(
rcp(new Albany::FieldManagerScalarResponseFunction(
app, prob, meshSpecs[i], stateMgr, responseParams)));
}
}
else if (name == "IP to Nodal Field" ||
name == "Project IP to Nodal Field") {
responseParams.set("Name", name);
for (int i=0; i<meshSpecs.size(); i++) {
#if defined(ALBANY_LCM)
// Skip if dealing with interface block
//if (meshSpecs[i]->ebName == "Surface Element") continue;
#endif
// For these RFs, default to true for this parm.
const char* reb_parm = "Restrict to Element Block";
if ( ! responseParams.isType<bool>(reb_parm) &&
(name == "IP to Nodal Field" ||
name == "Project IP to Nodal Field"))
responseParams.set<bool>(reb_parm, true);
responses.push_back(
rcp(new Albany::FieldManagerResidualOnlyResponseFunction(
app, prob, meshSpecs[i], stateMgr, responseParams)));
}
}
else if (name == "Solution") {
#if defined(ALBANY_EPETRA)
responses.push_back(
rcp(new Albany::SolutionResponseFunction(app, responseParams)));
#endif
}
else if (name == "KL") {
Array< RCP<AbstractResponseFunction> > base_responses;
std::string name = responseParams.get<std::string>("Response");
createResponseFunction(name, responseParams.sublist("ResponseParams"),
base_responses);
for (int i=0; i<base_responses.size(); i++)
responses.push_back(
rcp(new Albany::KLResponseFunction(base_responses[i], responseParams)));
}
#ifdef ALBANY_QCAD
#if defined(ALBANY_EPETRA)
else if (name == "Saddle Value") {
responseParams.set("Name", name);
for (int i=0; i<meshSpecs.size(); i++) {
responses.push_back(
rcp(new QCAD::SaddleValueResponseFunction(
app, prob, meshSpecs[i], stateMgr, responseParams)));
}
}
#endif
#endif
else {
TEUCHOS_TEST_FOR_EXCEPTION(
true, Teuchos::Exceptions::InvalidParameter,
std::endl << "Error! Unknown response function " << name <<
"!" << std::endl << "Supplied parameter list is " <<
std::endl << responseParams);
}
}
// ======================================================================
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
int nx = 30;
Teuchos::ParameterList GaleriList;
GaleriList.set("n", nx * nx);
GaleriList.set("nx", nx);
GaleriList.set("ny", nx);
Teuchos::RefCountPtr<Epetra_Map> Map = Teuchos::rcp( Galeri::CreateMap("Linear", Comm, GaleriList) );
Teuchos::RefCountPtr<Epetra_RowMatrix> A = Teuchos::rcp( Galeri::CreateCrsMatrix("Laplace2D", &*Map, GaleriList) );
Epetra_Vector LHS(*Map);
Epetra_Vector RHS(*Map);
Epetra_LinearProblem Problem(&*A, &LHS, &RHS);
int TestPassed = true;
// Jacobi as in AztecOO (no overlap)
for (int ival = 1 ; ival < 10 ; ival += 3) {
TestPassed = TestPassed &&
CompareWithAztecOO(Problem,"Jacobi",0,ival);
}
#if 0
// AztecOO with IC and overlap complains, also with
// large fill-ins (in parallel)
TestPassed = TestPassed &&
CompareWithAztecOO(Problem,"IC no reord",0,0);
TestPassed = TestPassed &&
CompareWithAztecOO(Problem,"IC reord",0,0);
vector<std::string> Tests;
// now test solvers that accept overlap
Tests.push_back("ILU no reord");
Tests.push_back("ILU reord");
// following requires --enable-aztecoo-azlu
#ifdef HAVE_IFPACK_AMESOS
//Tests.push_back("LU");
#endif
for (unsigned int i = 0 ; i < Tests.size() ; ++i) {
for (int overlap = 0 ; overlap < 1 ; overlap += 2) {
for (int ival = 0 ; ival < 10 ; ival += 4)
TestPassed = TestPassed &&
CompareWithAztecOO(Problem,Tests[i],overlap,ival);
}
}
#endif
if (!TestPassed) {
cerr << "Test `CompareWithAztecOO.exe' FAILED!" << endl;
exit(EXIT_FAILURE);
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
cout << "Test `CompareWithAztecOO.exe' passed!" << endl;
exit(EXIT_SUCCESS);
}
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[]) {
#if defined(HAVE_MUELU_EPETRA) && defined(HAVE_MUELU_EPETRAEXT)
typedef double Scalar;
typedef int LocalOrdinal;
typedef int GlobalOrdinal;
typedef LocalOrdinal LO;
typedef GlobalOrdinal GO;
typedef Xpetra::EpetraNode Node;
#include "MueLu_UseShortNames.hpp"
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::rcpFromRef;
using namespace MueLuTests;
Teuchos::oblackholestream blackhole;
Teuchos::GlobalMPISession mpiSession(&argc,&argv,&blackhole);
bool success = false;
bool verbose = true;
try {
RCP<const Teuchos::Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
RCP<Teuchos::FancyOStream> out = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout));
out->setOutputToRootOnly(0);
*out << MueLu::MemUtils::PrintMemoryUsage() << std::endl;
// Timing
Teuchos::Time myTime("global");
Teuchos::TimeMonitor MM(myTime);
// read in some command line parameters
Teuchos::CommandLineProcessor clp(false);
int rebalanceBlocks = 1;
clp.setOption("rebalanceBlocks", &rebalanceBlocks, "rebalance blocks (1=yes, else=no)");
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;
}
#if defined(HAVE_MPI) && defined(HAVE_MUELU_ZOLTAN) && defined(HAVE_MUELU_ISORROPIA)
#ifndef HAVE_XPETRA_INT_LONG_LONG
*out << "Warning: scaling test was not compiled with long long int support" << std::endl;
// custom parameters
LocalOrdinal maxLevels = 3;
GlobalOrdinal maxCoarseSize=1; //FIXME clp doesn't like long long int
int globalNumDofs = 1500; // used for the maps
int nDofsPerNode = 3; // used for generating the fine level null-space
// build strided maps
// striding information: 2 velocity dofs and 1 pressure dof = 3 dofs per node
std::vector<size_t> stridingInfo;
stridingInfo.push_back(2);
stridingInfo.push_back(1);
/////////////////////////////////////// build strided maps
// build strided maps:
// xstridedfullmap: full map (velocity and pressure dof gids), continous
// xstridedvelmap: only velocity dof gid maps (i.e. 0,1,3,4,6,7...)
// xstridedpremap: only pressure dof gid maps (i.e. 2,5,8,...)
Xpetra::UnderlyingLib lib = Xpetra::UseEpetra;
RCP<const StridedMap> xstridedfullmap = StridedMapFactory::Build(lib,globalNumDofs,0,stridingInfo,comm,-1);
RCP<const StridedMap> xstridedvelmap = StridedMapFactory::Build(xstridedfullmap,0);
RCP<const StridedMap> xstridedpremap = StridedMapFactory::Build(xstridedfullmap,1);
/////////////////////////////////////// transform Xpetra::Map objects to Epetra
// this is needed for AztecOO
const RCP<const Epetra_Map> fullmap = rcpFromRef(Xpetra::toEpetra(*xstridedfullmap));
RCP<const Epetra_Map> velmap = rcpFromRef(Xpetra::toEpetra(*xstridedvelmap));
RCP<const Epetra_Map> premap = rcpFromRef(Xpetra::toEpetra(*xstridedpremap));
/////////////////////////////////////// import problem matrix and RHS from files (-> Epetra)
// read in problem
Epetra_CrsMatrix * ptrA = 0;
Epetra_Vector * ptrf = 0;
Epetra_MultiVector* ptrNS = 0;
*out << "Reading matrix market file" << std::endl;
EpetraExt::MatrixMarketFileToCrsMatrix("A_re1000_5932.txt",*fullmap,*fullmap,*fullmap,ptrA);
EpetraExt::MatrixMarketFileToVector("b_re1000_5932.txt",*fullmap,ptrf);
//EpetraExt::MatrixMarketFileToCrsMatrix("/home/tobias/promotion/trilinos/fc17-dyn/packages/muelu/test/navierstokes/A_re1000_5932.txt",*fullmap,*fullmap,*fullmap,ptrA);
//EpetraExt::MatrixMarketFileToVector("/home/tobias/promotion/trilinos/fc17-dyn/packages/muelu/test/navierstokes/b_re1000_5932.txt",*fullmap,ptrf);
RCP<Epetra_CrsMatrix> epA = Teuchos::rcp(ptrA);
RCP<Epetra_Vector> epv = Teuchos::rcp(ptrf);
RCP<Epetra_MultiVector> epNS = Teuchos::rcp(ptrNS);
/////////////////////////////////////// split system into 2x2 block system
*out << "Split matrix into 2x2 block matrix" << std::endl;
// split fullA into A11,..., A22
Teuchos::RCP<Epetra_CrsMatrix> A11;
Teuchos::RCP<Epetra_CrsMatrix> A12;
Teuchos::RCP<Epetra_CrsMatrix> A21;
Teuchos::RCP<Epetra_CrsMatrix> A22;
if(SplitMatrix2x2(epA,*velmap,*premap,A11,A12,A21,A22)==false)
*out << "Problem with splitting matrix"<< std::endl;
/////////////////////////////////////// transform Epetra objects to Xpetra (needed for MueLu)
// build Xpetra objects from Epetra_CrsMatrix objects
Teuchos::RCP<Xpetra::CrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node> > xA11 = Teuchos::rcp(new Xpetra::EpetraCrsMatrixT<GlobalOrdinal,Node>(A11));
Teuchos::RCP<Xpetra::CrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node> > xA12 = Teuchos::rcp(new Xpetra::EpetraCrsMatrixT<GlobalOrdinal,Node>(A12));
Teuchos::RCP<Xpetra::CrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node> > xA21 = Teuchos::rcp(new Xpetra::EpetraCrsMatrixT<GlobalOrdinal,Node>(A21));
Teuchos::RCP<Xpetra::CrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node> > xA22 = Teuchos::rcp(new Xpetra::EpetraCrsMatrixT<GlobalOrdinal,Node>(A22));
/////////////////////////////////////// generate MapExtractor object
std::vector<Teuchos::RCP<const Xpetra::Map<LocalOrdinal,GlobalOrdinal,Node> > > xmaps;
xmaps.push_back(xstridedvelmap);
xmaps.push_back(xstridedpremap);
Teuchos::RCP<const Xpetra::MapExtractor<Scalar,LocalOrdinal,GlobalOrdinal,Node> > map_extractor = Xpetra::MapExtractorFactory<Scalar,LocalOrdinal,GlobalOrdinal,Node>::Build(xstridedfullmap,xmaps);
/////////////////////////////////////// build blocked transfer operator
// using the map extractor
Teuchos::RCP<Xpetra::BlockedCrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node> > bOp = Teuchos::rcp(new Xpetra::BlockedCrsMatrix<Scalar,LocalOrdinal,GlobalOrdinal,Node>(map_extractor,map_extractor,10));
bOp->setMatrix(0,0,Teuchos::rcp(new Xpetra::CrsMatrixWrap<Scalar,LocalOrdinal,GlobalOrdinal,Node>(xA11)));
bOp->setMatrix(0,1,Teuchos::rcp(new Xpetra::CrsMatrixWrap<Scalar,LocalOrdinal,GlobalOrdinal,Node>(xA12)));
bOp->setMatrix(1,0,Teuchos::rcp(new Xpetra::CrsMatrixWrap<Scalar,LocalOrdinal,GlobalOrdinal,Node>(xA21)));
bOp->setMatrix(1,1,Teuchos::rcp(new Xpetra::CrsMatrixWrap<Scalar,LocalOrdinal,GlobalOrdinal,Node>(xA22)));
bOp->fillComplete();
//////////////////////////////////////////////////// create Hierarchy
RCP<Hierarchy> H = rcp ( new Hierarchy() );
H->setDefaultVerbLevel(Teuchos::VERB_HIGH);
//H->setDefaultVerbLevel(Teuchos::VERB_NONE);
H->SetMaxCoarseSize(maxCoarseSize);
//////////////////////////////////////////////////////// finest Level
RCP<MueLu::Level> Finest = H->GetLevel();
Finest->setDefaultVerbLevel(Teuchos::VERB_HIGH);
Finest->Set("A",Teuchos::rcp_dynamic_cast<Matrix>(bOp));
////////////////////////////////////////// prepare null space for A11
RCP<MultiVector> nullspace11 = MultiVectorFactory::Build(xstridedvelmap, 2); // this is a 2D standard null space
for (int i=0; i<nDofsPerNode-1; ++i) {
Teuchos::ArrayRCP<Scalar> nsValues = nullspace11->getDataNonConst(i);
int numBlocks = nsValues.size() / (nDofsPerNode - 1);
for (int j=0; j< numBlocks; ++j) {
nsValues[j*(nDofsPerNode - 1) + i] = 1.0;
}
}
Finest->Set("Nullspace1",nullspace11);
////////////////////////////////////////// prepare null space for A22
RCP<MultiVector> nullspace22 = MultiVectorFactory::Build(xstridedpremap, 1); // this is a 2D standard null space
Teuchos::ArrayRCP<Scalar> nsValues22 = nullspace22->getDataNonConst(0);
for (int j=0; j< nsValues22.size(); ++j) {
nsValues22[j] = 1.0;
}
Finest->Set("Nullspace2",nullspace22);
/////////////////////////////////////////// define rebalanced block AC factory
// This is the main factory for "A" and defines the input for
// - the SubBlockAFactory objects
// - the rebalanced block Ac factory
RCP<RebalanceBlockAcFactory> RebalancedAcFact = rcp(new RebalanceBlockAcFactory());
/////////////////////////////////////////// define non-rebalanced blocked transfer ops
RCP<BlockedPFactory> PFact = rcp(new BlockedPFactory()); // use row map index base from bOp
RCP<GenericRFactory> RFact = rcp(new GenericRFactory());
RFact->SetFactory("P", PFact);
// non-rebalanced block coarse matrix factory
// output is non-rebalanced coarse block matrix Ac
// used as input for rebalanced block coarse factory RebalancedAcFact
RCP<Factory> AcFact = rcp(new BlockedRAPFactory());
AcFact->SetFactory("A", MueLu::NoFactory::getRCP());
AcFact->SetFactory("P", PFact); // use non-rebalanced block prolongator as input
AcFact->SetFactory("R", RFact); // use non-rebalanced block restrictor as input
// Repartitioning (decides how many partitions are built)
RCP<Factory> RepartitionHeuristicFact = rcp(new RepartitionHeuristicFactory());
{
Teuchos::ParameterList paramList;
paramList.set("repartition: min rows per proc", 200);
paramList.set("repartition: max imbalance", 1.3);
if(rebalanceBlocks == 1)
paramList.set("repartition: start level",1);
else
paramList.set("repartition: start level",10); // supress rebalancing
RepartitionHeuristicFact->SetParameterList(paramList);
}
RepartitionHeuristicFact->SetFactory("A", AcFact);
// define matrix sub-blocks of possibly rebalanced block matrix A
// These are used as input for
// - the sub blocks of the transfer operators
RCP<SubBlockAFactory> A11Fact = Teuchos::rcp(new SubBlockAFactory());
A11Fact->SetFactory("A",MueLu::NoFactory::getRCP());
A11Fact->SetParameter("block row",Teuchos::ParameterEntry(0));
A11Fact->SetParameter("block col",Teuchos::ParameterEntry(0));
RCP<SubBlockAFactory> A22Fact = Teuchos::rcp(new SubBlockAFactory());
A22Fact->SetFactory("A",MueLu::NoFactory::getRCP());
A22Fact->SetParameter("block row",Teuchos::ParameterEntry(1));
A22Fact->SetParameter("block col",Teuchos::ParameterEntry(1));
/////////////////////////////////////////// define rebalancing factories
// define sub blocks of the coarse non-rebalanced block matrix Ac
// input is the block operator generated by AcFact
RCP<SubBlockAFactory> rebA11Fact = Teuchos::rcp(new SubBlockAFactory());
rebA11Fact->SetFactory("A",AcFact);
rebA11Fact->SetParameter("block row",Teuchos::ParameterEntry(0));
rebA11Fact->SetParameter("block col",Teuchos::ParameterEntry(0));
RCP<SubBlockAFactory> rebA22Fact = Teuchos::rcp(new SubBlockAFactory());
rebA22Fact->SetFactory("A",AcFact);
rebA22Fact->SetParameter("block row",Teuchos::ParameterEntry(1));
rebA22Fact->SetParameter("block col",Teuchos::ParameterEntry(1));
// define rebalancing factory for coarse block matrix A(1,1)
RCP<AmalgamationFactory> rebAmalgFact11 = rcp(new AmalgamationFactory());
rebAmalgFact11->SetFactory("A", rebA11Fact);
rebAmalgFact11->setDefaultVerbLevel(Teuchos::VERB_EXTREME);
RCP<MueLu::IsorropiaInterface<LO, GO, NO> > isoInterface1 = rcp(new MueLu::IsorropiaInterface<LO, GO, NO>());
isoInterface1->SetFactory("A", rebA11Fact);
isoInterface1->SetFactory("number of partitions", RepartitionHeuristicFact);
isoInterface1->SetFactory("UnAmalgamationInfo", rebAmalgFact11);
RCP<MueLu::RepartitionInterface<LO, GO, NO> > repInterface1 = rcp(new MueLu::RepartitionInterface<LO, GO, NO>());
repInterface1->SetFactory("A", rebA11Fact);
repInterface1->SetFactory("number of partitions", RepartitionHeuristicFact);
repInterface1->SetFactory("AmalgamatedPartition", isoInterface1);
// Repartitioning (creates "Importer" from "Partition")
RCP<Factory> RepartitionFact = rcp(new RepartitionFactory());
RepartitionFact->SetFactory("A", rebA11Fact);
RepartitionFact->SetFactory("number of partitions", RepartitionHeuristicFact);
RepartitionFact->SetFactory("Partition", repInterface1);
// define rebalancing factory for coarse block matrix A(1,1)
RCP<AmalgamationFactory> rebAmalgFact22 = rcp(new AmalgamationFactory());
rebAmalgFact22->SetFactory("A", rebA22Fact);
rebAmalgFact22->setDefaultVerbLevel(Teuchos::VERB_EXTREME);
RCP<MueLu::RepartitionInterface<LO, GO, NO> > repInterface2 = rcp(new MueLu::RepartitionInterface<LO, GO, NO>());
repInterface2->SetFactory("A", rebA22Fact);
repInterface2->SetFactory("number of partitions", RepartitionHeuristicFact);
repInterface2->SetFactory("AmalgamatedPartition", isoInterface1);
// second repartition factory
RCP<Factory> RepartitionFact2 = rcp(new RepartitionFactory());
RepartitionFact2->SetFactory("A", rebA22Fact);
RepartitionFact2->SetFactory("number of partitions", RepartitionHeuristicFact);
RepartitionFact2->SetFactory("Partition", repInterface2); // this is not valid
////////////////////////////////////////// build non-rebalanced matrix blocks
// build factories for transfer operator P(1,1) and R(1,1)
RCP<AmalgamationFactory> amalgFact11 = rcp(new AmalgamationFactory());
amalgFact11->SetFactory("A", A11Fact);
amalgFact11->setDefaultVerbLevel(Teuchos::VERB_EXTREME);
RCP<CoalesceDropFactory> dropFact11 = rcp(new CoalesceDropFactory());
dropFact11->SetFactory("A", A11Fact);
dropFact11->SetFactory("UnAmalgamationInfo", amalgFact11);
dropFact11->setDefaultVerbLevel(Teuchos::VERB_EXTREME);
RCP<UncoupledAggregationFactory> UncoupledAggFact11 = rcp(new UncoupledAggregationFactory());
UncoupledAggFact11->SetFactory("Graph", dropFact11);
UncoupledAggFact11->SetMinNodesPerAggregate(9);
UncoupledAggFact11->SetMaxNeighAlreadySelected(2);
UncoupledAggFact11->SetOrdering("natural");
RCP<CoarseMapFactory> coarseMapFact11 = Teuchos::rcp(new CoarseMapFactory());
coarseMapFact11->setStridingData(stridingInfo);
coarseMapFact11->setStridedBlockId(0);
RCP<TentativePFactory> P11Fact = rcp(new TentativePFactory());
RCP<TransPFactory> R11Fact = rcp(new TransPFactory());
Teuchos::RCP<NullspaceFactory> nspFact11 = Teuchos::rcp(new NullspaceFactory("Nullspace1"));
nspFact11->SetFactory("Nullspace1",P11Fact); // pick "Nullspace1" from Finest level
//////////////////////////////// define factory manager for (1,1) block
RCP<FactoryManager> M11 = rcp(new FactoryManager());
M11->SetFactory("A", A11Fact); // rebalanced fine-level block operator
M11->SetFactory("P", P11Fact); // non-rebalanced transfer operator block P(1,1)
M11->SetFactory("R", R11Fact); // non-rebalanced transfer operator block R(1,1)
M11->SetFactory("Aggregates", UncoupledAggFact11);
M11->SetFactory("Graph", dropFact11);
M11->SetFactory("DofsPerNode", dropFact11);
M11->SetFactory("UnAmalgamationInfo", amalgFact11);
M11->SetFactory("Nullspace", nspFact11); // TODO check me?
M11->SetFactory("CoarseMap", coarseMapFact11);
M11->SetIgnoreUserData(true); // always use data from factories defined in factory manager
////////////////////////////////////////// build non-rebalanced matrix blocks
// build factories for transfer operator P(2,2) and R(2,2)
RCP<AmalgamationFactory> amalgFact22 = rcp(new AmalgamationFactory());
RCP<TentativePFactory> P22Fact = rcp(new TentativePFactory());
RCP<TransPFactory> R22Fact = rcp(new TransPFactory());
// connect null space and tentative PFactory
Teuchos::RCP<NullspaceFactory> nspFact22 = Teuchos::rcp(new NullspaceFactory("Nullspace2"));
nspFact22->SetFactory("Nullspace2", P22Fact); // define null space generated by P22Fact as null space for coarse level (non-rebalanced)
RCP<CoarseMapFactory> coarseMapFact22 = Teuchos::rcp(new CoarseMapFactory());
coarseMapFact22->setStridingData(stridingInfo);
coarseMapFact22->setStridedBlockId(1);
//////////////////////////////// define factory manager for (2,2) block
RCP<FactoryManager> M22 = rcp(new FactoryManager());
M22->SetFactory("A", A22Fact); // rebalanced fine-level block operator
M22->SetFactory("P", P22Fact); // non-rebalanced transfer operator P(2,2)
M22->SetFactory("R", R22Fact); // non-rebalanced transfer operator R(2,2)
M22->SetFactory("Aggregates", UncoupledAggFact11); // aggregates from block (1,1)
M22->SetFactory("Nullspace", nspFact22);
M22->SetFactory("UnAmalgamationInfo", amalgFact22);
M22->SetFactory("Ptent", P22Fact);
M22->SetFactory("CoarseMap", coarseMapFact22);
M22->SetIgnoreUserData(true);
/////////////////////////////////////////// define rebalanced blocked transfer ops
//////////////////////////////// define factory manager for (1,1) block
RCP<FactoryManager> rebM11 = rcp(new FactoryManager());
rebM11->SetFactory("A", AcFact ); // important: must be a 2x2 block A Factory
rebM11->SetFactory("Importer", RepartitionFact);
rebM11->SetFactory("Nullspace", nspFact11);
//rebM11->SetIgnoreUserData(true);
RCP<FactoryManager> rebM22 = rcp(new FactoryManager());
rebM22->SetFactory("A", AcFact ); // important: must be a 2x2 block A Factory
rebM22->SetFactory("Importer", RepartitionFact2); // use dummy repartitioning factory
rebM22->SetFactory("Nullspace", nspFact22);
// Reordering of the transfer operators
RCP<RebalanceBlockInterpolationFactory> RebalancedBlockPFact = rcp(new RebalanceBlockInterpolationFactory());
RebalancedBlockPFact->SetFactory("P", PFact); // use non-rebalanced block P operator as input
RebalancedBlockPFact->AddFactoryManager(rebM11);
RebalancedBlockPFact->AddFactoryManager(rebM22);
RCP<RebalanceBlockRestrictionFactory> RebalancedBlockRFact = rcp(new RebalanceBlockRestrictionFactory());
//RebalancedBlockRFact->SetParameter("type", Teuchos::ParameterEntry(std::string("Restriction")));
RebalancedBlockRFact->SetFactory("R", RFact); // non-rebalanced block P operator
RebalancedBlockRFact->AddFactoryManager(rebM11);
RebalancedBlockRFact->AddFactoryManager(rebM22);
///////////////////////////////////////// initialize non-rebalanced block transfer operators
// output are the non-rebalanced block transfer operators used as input in AcFact to build
// the non-rebalanced coarse level block matrix Ac
PFact->AddFactoryManager(M11); // use non-rebalanced information from sub block factory manager M11
PFact->AddFactoryManager(M22); // use non-rebalanced information from sub block factory manager M22
///////////////////////////////////////// initialize rebalanced coarse block AC factory
RebalancedAcFact->SetFactory("A", AcFact); // use non-rebalanced block operator as input
RebalancedAcFact->AddFactoryManager(rebM11);
RebalancedAcFact->AddFactoryManager(rebM22);
//////////////////////////////////////////////////////////////////////
// Smoothers
//Another factory manager for braes sarazin smoother
//Schur Complement Factory, using the factory to generate AcFact
SC omega = 1.3;
RCP<SchurComplementFactory> SFact = Teuchos::rcp(new SchurComplementFactory());
SFact->SetParameter("omega", Teuchos::ParameterEntry(omega));
SFact->SetFactory("A", MueLu::NoFactory::getRCP()); // this finally be the rebalanced block operator!
//Smoother Factory, using SFact as a factory for A
std::string ifpackSCType;
Teuchos::ParameterList ifpackSCList;
ifpackSCList.set("relaxation: sweeps", (LocalOrdinal) 3);
ifpackSCList.set("relaxation: damping factor", (Scalar) 1.0);
ifpackSCType = "RELAXATION";
ifpackSCList.set("relaxation: type", "Gauss-Seidel");
RCP<SmootherPrototype> smoProtoSC = rcp( new TrilinosSmoother(ifpackSCType, ifpackSCList, 0) );
smoProtoSC->SetFactory("A", SFact);
RCP<SmootherFactory> SmooSCFact = rcp( new SmootherFactory(smoProtoSC) );
RCP<BraessSarazinSmoother> smootherPrototype = rcp( new BraessSarazinSmoother() );
smootherPrototype->SetParameter("Sweeps", Teuchos::ParameterEntry(3));
smootherPrototype->SetParameter("Damping factor", Teuchos::ParameterEntry(omega));
smootherPrototype->SetFactory("A",MueLu::NoFactory::getRCP());
RCP<SmootherFactory> smootherFact = rcp( new SmootherFactory(smootherPrototype) );
RCP<BraessSarazinSmoother> coarseSolverPrototype = rcp( new BraessSarazinSmoother() );
coarseSolverPrototype->SetParameter("Sweeps", Teuchos::ParameterEntry(3));
coarseSolverPrototype->SetParameter("Damping factor", Teuchos::ParameterEntry(omega));
coarseSolverPrototype->SetFactory("A",MueLu::NoFactory::getRCP());
RCP<SmootherFactory> coarseSolverFact = rcp( new SmootherFactory(coarseSolverPrototype, Teuchos::null) );
RCP<FactoryManager> MB = rcp(new FactoryManager());
MB->SetFactory("A", SFact);
MB->SetFactory("Smoother", SmooSCFact);
MB->SetIgnoreUserData(true); // always use data from factories defined in factory manager
smootherPrototype->AddFactoryManager(MB,0);
coarseSolverPrototype->AddFactoryManager(MB,0);
////////////////////////////////////////// define main factory manager
FactoryManager M;
M.SetFactory("A", RebalancedAcFact); // rebalance block AC Factory using importer
M.SetFactory("P", RebalancedBlockPFact); // rebalance prolongator using non-balanced Ac
M.SetFactory("R", RebalancedBlockRFact); // rebalance restrictor and null space using non-balanced Ac
M.SetFactory("Smoother", smootherFact);
M.SetFactory("PreSmoother", smootherFact);
M.SetFactory("PostSmoother", smootherFact);
M.SetFactory("CoarseSolver", coarseSolverFact);
H->Setup(M,0,maxLevels);
/**out << std::endl;
*out << "print content of multigrid levels:" << std::endl;
Finest->print(*out);
RCP<Level> coarseLevel = H->GetLevel(1);
coarseLevel->print(*out);
RCP<Level> coarseLevel2 = H->GetLevel(2);
coarseLevel2->print(*out);*/
RCP<MultiVector> xLsg = MultiVectorFactory::Build(xstridedfullmap,1);
// Use AMG directly as an iterative method
#if 0
{
xLsg->putScalar( (SC) 0.0);
// Epetra_Vector -> Xpetra::Vector
RCP<Vector> xRhs = Teuchos::rcp(new Xpetra::EpetraVector(epv));
// calculate initial (absolute) residual
Teuchos::Array<Teuchos::ScalarTraits<SC>::magnitudeType> norms(1);
xRhs->norm2(norms);
*out << "||x_0|| = " << norms[0] << std::endl;
// apply ten multigrid iterations
H->Iterate(*xRhs,*xLsg,100);
// calculate and print residual
RCP<MultiVector> xTmp = MultiVectorFactory::Build(xstridedfullmap,1);
bOp->apply(*xLsg,*xTmp,Teuchos::NO_TRANS,(SC)1.0,(SC)0.0);
xRhs->update((SC)-1.0,*xTmp,(SC)1.0);
xRhs->norm2(norms);
*out << "||x|| = " << norms[0] << std::endl;
}
#endif
//
// Solve Ax = b using AMG as a preconditioner in AztecOO
//
{
RCP<Epetra_Vector> X = rcp(new Epetra_Vector(epv->Map()));
X->PutScalar(0.0);
Epetra_LinearProblem epetraProblem(epA.get(), X.get(), epv.get());
AztecOO aztecSolver(epetraProblem);
aztecSolver.SetAztecOption(AZ_solver, AZ_gmres);
MueLu::EpetraOperator aztecPrec(H);
aztecSolver.SetPrecOperator(&aztecPrec);
int maxIts = 50;
double tol = 1e-8;
aztecSolver.Iterate(maxIts, tol);
}
#endif // end ifndef HAVE_LONG_LONG_INT
#endif // #if defined(HAVE_MPI) && defined(HAVE_MUELU_ZOLTAN) && defined(HAVE_MUELU_ISORROPIA)
success = true;
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(verbose, std::cerr, success);
return ( success ? EXIT_SUCCESS : EXIT_FAILURE );
#else
std::cout << "Epetra (and/or EpetraExt) are not available. Skip test." << std::endl;
return EXIT_SUCCESS;
#endif
}
Teuchos::ParameterList Ifpack_GetValidParameters()
{
Teuchos::ParameterList List; // empty list
// ============================================================ //
// Parameters are reported from each used file in IFPACK. Files //
// are listed in alphabetical order, first all *.cpp, then *.h. //
// Some options not very tested or documented anywhere //
// are not reported here. //
// ============================================================ //
// Ifpack_Amesos.cpp
List.set("amesos: solver type", "Amesos_Klu");
// Ifpack_IC.cpp
List.set("fact: level-of-fill", (int)1);
List.set("fact: absolute threshold", (double)0.0);
List.set("fact: relative threshold", (double)0.0);
List.set("fact: drop tolerance", (double)0.0);
// Ifpack_ICT.cpp
List.set("fact: ict level-of-fill", (double)1.0);
List.set("fact: absolute threshold", (double)0.0);
List.set("fact: relative threshold", (double)1.0);
List.set("fact: relax value", (double)0.0);
List.set("fact: drop tolerance", (double)0.0);
// Ifpack_ILU.cpp
List.set("fact: level-of-fill", (int)0);
List.set("fact: absolute threshold", (double)0.0);
List.set("fact: relative threshold", (double)1.0);
List.set("fact: relax value", (double)0.0);
// Ifpack_ILUT.cpp
List.set("fact: ilut level-of-fill", (double)1.0);
List.set("fact: absolute threshold", (double)0.0);
List.set("fact: relative threshold", (double)1.0);
List.set("fact: relax value", (double)0.0);
#ifdef HAVE_IFPACK_SUPERLU
// Ifpack_SILU.cpp
List.set("fact: drop tolerance",1e-4);
List.set("fact: zero pivot threshold",1e-2);
List.set("fact: maximum fill factor",10.0);
List.set("fact: silu drop rule",9);
#endif
// Ifpack_METISPartitioner.cpp
List.set("partitioner: local parts", (int)1);
List.set("partitioner: overlap", (int)0);
List.set("partitioner: print level", (int)0);
// Ifpack_PointRelaxation.cpp
List.set("relaxation: type", "Jacobi");
List.set("relaxation: sweeps", (int)1);
List.set("relaxation: damping factor", (double)1.0);
List.set("relaxation: min diagonal value", (double)1.0);
List.set("relaxation: zero starting solution", true);
List.set("relaxation: backward mode",false);
List.set("relaxation: use l1",false);
List.set("relaxation: l1 eta",(double)1.5);
// Ifpack_SPARSKIT.cpp
List.set("fact: sparskit: lfil", (int)0);
List.set("fact: sparskit: tol", (double)0.0);
List.set("fact: sparskit: droptol", (double)0.0);
List.set("fact: sparskit: permtol", (double)0.1);
List.set("fact: sparskit: alph", (double)0.0);
List.set("fact: sparskit: mbloc", (int)(-1));
List.set("fact: sparskit: type", ("ILUT"));
// Additive Schwarz preconditioner
List.set("schwarz: compute condest", true);
List.set("schwarz: combine mode", "Zero"); // use string mode for this
List.set("schwarz: reordering type", "none");
List.set("schwarz: filter singletons", false);
// Ifpack_BlockRelaxation.h
// List.set("relaxation: type", "Jacobi"); // already set
// List.set("relaxation: sweeps", 1); // already set
// List.get("relaxation: damping factor", 1.0); // already set
// List.get("relaxation: zero starting solution", true); // already set
List.set("partitioner: type", "greedy");
List.set("partitioner: local parts", (int)1);
List.set("partitioner: overlap", (int)0);
// Ifpack_METISPartitioner.h
List.set("partitioner: use symmetric graph", true);
// Krylov smoother
List.set("krylov: iterations",(int)5);
List.set("krylov: tolerance",(double)0.001);
List.set("krylov: solver",(int)1);
List.set("krylov: preconditioner",(int)0);
List.set("krylov: number of sweeps",(int)1);
List.set("krylov: block size",(int)1);
List.set("krylov: damping parameter",(double)1.0);
List.set("krylov: zero starting solution",true);
return(List);
}
int main(int argc, char *argv[]) {
int n = 32; // spatial discretization (per dimension)
int num_KL = 2; // number of KL terms
int p = 3; // polynomial order
double mu = 0.1; // mean of exponential random field
double s = 0.2; // std. dev. of exponential r.f.
bool nonlinear_expansion = false; // nonlinear expansion of diffusion coeff
// (e.g., log-normal)
bool matrix_free = true; // use matrix-free stochastic operator
bool symmetric = false; // use symmetric formulation
double g_mean_exp = 0.172988; // expected response mean
double g_std_dev_exp = 0.0380007; // expected response std. dev.
double g_tol = 1e-6; // tolerance on determining success
// Initialize MPI
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
#endif
int MyPID;
try {
{
TEUCHOS_FUNC_TIME_MONITOR("Total PCE Calculation Time");
// Create a communicator for Epetra objects
Teuchos::RCP<const Epetra_Comm> globalComm;
#ifdef HAVE_MPI
globalComm = Teuchos::rcp(new Epetra_MpiComm(MPI_COMM_WORLD));
#else
globalComm = Teuchos::rcp(new Epetra_SerialComm);
#endif
MyPID = globalComm->MyPID();
// Create Stochastic Galerkin basis and expansion
Teuchos::Array< Teuchos::RCP<const Stokhos::OneDOrthogPolyBasis<int,double> > > bases(num_KL);
for (int i=0; i<num_KL; i++)
bases[i] = Teuchos::rcp(new Stokhos::LegendreBasis<int,double>(p, true));
Teuchos::RCP<const Stokhos::CompletePolynomialBasis<int,double> > basis =
Teuchos::rcp(new Stokhos::CompletePolynomialBasis<int,double>(bases));
int sz = basis->size();
Teuchos::RCP<Stokhos::Sparse3Tensor<int,double> > Cijk;
if (nonlinear_expansion)
Cijk = basis->computeTripleProductTensor();
else
Cijk = basis->computeLinearTripleProductTensor();
Teuchos::RCP<Stokhos::OrthogPolyExpansion<int,double> > expansion =
Teuchos::rcp(new Stokhos::AlgebraicOrthogPolyExpansion<int,double>(basis,
Cijk));
if (MyPID == 0)
std::cout << "Stochastic Galerkin expansion size = " << sz << std::endl;
// Create stochastic parallel distribution
int num_spatial_procs = -1;
if (argc > 1)
num_spatial_procs = std::atoi(argv[1]);
Teuchos::ParameterList parallelParams;
parallelParams.set("Number of Spatial Processors", num_spatial_procs);
Teuchos::RCP<Stokhos::ParallelData> sg_parallel_data =
Teuchos::rcp(new Stokhos::ParallelData(basis, Cijk, globalComm,
parallelParams));
Teuchos::RCP<const EpetraExt::MultiComm> sg_comm =
sg_parallel_data->getMultiComm();
Teuchos::RCP<const Epetra_Comm> app_comm =
sg_parallel_data->getSpatialComm();
// Create application
Teuchos::RCP<twoD_diffusion_ME> model =
Teuchos::rcp(new twoD_diffusion_ME(app_comm, n, num_KL, mu, s, basis,
nonlinear_expansion, symmetric));
// Setup stochastic Galerkin algorithmic parameters
Teuchos::RCP<Teuchos::ParameterList> sgParams =
Teuchos::rcp(new Teuchos::ParameterList);
Teuchos::ParameterList& sgOpParams =
sgParams->sublist("SG Operator");
Teuchos::ParameterList& sgPrecParams =
sgParams->sublist("SG Preconditioner");
if (!nonlinear_expansion) {
sgParams->set("Parameter Expansion Type", "Linear");
sgParams->set("Jacobian Expansion Type", "Linear");
}
if (matrix_free) {
sgOpParams.set("Operator Method", "Matrix Free");
sgPrecParams.set("Preconditioner Method", "Approximate Gauss-Seidel");
sgPrecParams.set("Symmetric Gauss-Seidel", symmetric);
sgPrecParams.set("Mean Preconditioner Type", "ML");
Teuchos::ParameterList& precParams =
sgPrecParams.sublist("Mean Preconditioner Parameters");
precParams.set("default values", "SA");
precParams.set("ML output", 0);
precParams.set("max levels",5);
precParams.set("increasing or decreasing","increasing");
precParams.set("aggregation: type", "Uncoupled");
precParams.set("smoother: type","ML symmetric Gauss-Seidel");
precParams.set("smoother: sweeps",2);
precParams.set("smoother: pre or post", "both");
precParams.set("coarse: max size", 200);
#ifdef HAVE_ML_AMESOS
precParams.set("coarse: type","Amesos-KLU");
#else
precParams.set("coarse: type","Jacobi");
#endif
}
else {
sgOpParams.set("Operator Method", "Fully Assembled");
sgPrecParams.set("Preconditioner Method", "None");
}
// Create stochastic Galerkin model evaluator
Teuchos::RCP<Stokhos::SGModelEvaluator> sg_model =
Teuchos::rcp(new Stokhos::SGModelEvaluator(model, basis, Teuchos::null,
expansion, sg_parallel_data,
sgParams));
// Set up stochastic parameters
// The current implementation of the model doesn't actually use these
// values, but is hard-coded to certain uncertainty models
Teuchos::Array<double> point(num_KL, 1.0);
Teuchos::Array<double> basis_vals(sz);
basis->evaluateBases(point, basis_vals);
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> sg_p_init =
sg_model->create_p_sg(0);
for (int i=0; i<num_KL; i++) {
sg_p_init->term(i,0)[i] = 0.0;
sg_p_init->term(i,1)[i] = 1.0 / basis_vals[i+1];
}
sg_model->set_p_sg_init(0, *sg_p_init);
// Setup stochastic initial guess
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> sg_x_init =
sg_model->create_x_sg();
sg_x_init->init(0.0);
sg_model->set_x_sg_init(*sg_x_init);
// Set up NOX parameters
Teuchos::RCP<Teuchos::ParameterList> noxParams =
Teuchos::rcp(new Teuchos::ParameterList);
// Set the nonlinear solver method
noxParams->set("Nonlinear Solver", "Line Search Based");
// Set the printing parameters in the "Printing" sublist
Teuchos::ParameterList& printParams = noxParams->sublist("Printing");
printParams.set("MyPID", MyPID);
printParams.set("Output Precision", 3);
printParams.set("Output Processor", 0);
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Warning +
NOX::Utils::Error);
// Create printing utilities
NOX::Utils utils(printParams);
// Sublist for line search
Teuchos::ParameterList& searchParams = noxParams->sublist("Line Search");
searchParams.set("Method", "Full Step");
// Sublist for direction
Teuchos::ParameterList& dirParams = noxParams->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");
if (symmetric)
lsParams.set("Aztec Solver", "CG");
else
lsParams.set("Aztec Solver", "GMRES");
lsParams.set("Max Iterations", 1000);
lsParams.set("Size of Krylov Subspace", 100);
lsParams.set("Tolerance", 1e-12);
lsParams.set("Output Frequency", 1);
if (matrix_free)
lsParams.set("Preconditioner", "User Defined");
else {
lsParams.set("Preconditioner", "ML");
Teuchos::ParameterList& precParams =
lsParams.sublist("ML");
ML_Epetra::SetDefaults("DD", precParams);
lsParams.set("Write Linear System", false);
}
// Sublist for convergence tests
Teuchos::ParameterList& statusParams = noxParams->sublist("Status Tests");
statusParams.set("Test Type", "Combo");
statusParams.set("Number of Tests", 2);
statusParams.set("Combo Type", "OR");
Teuchos::ParameterList& normF = statusParams.sublist("Test 0");
normF.set("Test Type", "NormF");
normF.set("Tolerance", 1e-10);
normF.set("Scale Type", "Scaled");
Teuchos::ParameterList& maxIters = statusParams.sublist("Test 1");
maxIters.set("Test Type", "MaxIters");
maxIters.set("Maximum Iterations", 1);
// Create NOX interface
Teuchos::RCP<NOX::Epetra::ModelEvaluatorInterface> nox_interface =
Teuchos::rcp(new NOX::Epetra::ModelEvaluatorInterface(sg_model));
// Create NOX linear system object
Teuchos::RCP<const Epetra_Vector> u = sg_model->get_x_init();
Teuchos::RCP<Epetra_Operator> A = sg_model->create_W();
Teuchos::RCP<NOX::Epetra::Interface::Required> iReq = nox_interface;
Teuchos::RCP<NOX::Epetra::Interface::Jacobian> iJac = nox_interface;
Teuchos::RCP<NOX::Epetra::LinearSystemAztecOO> linsys;
if (matrix_free) {
Teuchos::RCP<Epetra_Operator> M = sg_model->create_WPrec()->PrecOp;
Teuchos::RCP<NOX::Epetra::Interface::Preconditioner> iPrec = nox_interface;
linsys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(printParams, lsParams,
iJac, A, iPrec, M,
*u));
}
else {
linsys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(printParams, lsParams,
iReq, iJac, A,
*u));
}
// Build NOX group
Teuchos::RCP<NOX::Epetra::Group> grp =
Teuchos::rcp(new NOX::Epetra::Group(printParams, iReq, *u, linsys));
// Create the Solver convergence test
Teuchos::RCP<NOX::StatusTest::Generic> statusTests =
NOX::StatusTest::buildStatusTests(statusParams, utils);
// Create the solver
Teuchos::RCP<NOX::Solver::Generic> solver =
NOX::Solver::buildSolver(grp, statusTests, noxParams);
// Solve the system
NOX::StatusTest::StatusType status = solver->solve();
// Get final solution
const NOX::Epetra::Group& finalGroup =
dynamic_cast<const NOX::Epetra::Group&>(solver->getSolutionGroup());
const Epetra_Vector& finalSolution =
(dynamic_cast<const NOX::Epetra::Vector&>(finalGroup.getX())).getEpetraVector();
// Save final solution to file
EpetraExt::VectorToMatrixMarketFile("nox_stochastic_solution.mm",
finalSolution);
// Save mean and variance to file
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> sg_x_poly =
sg_model->create_x_sg(View, &finalSolution);
Epetra_Vector mean(*(model->get_x_map()));
Epetra_Vector std_dev(*(model->get_x_map()));
sg_x_poly->computeMean(mean);
sg_x_poly->computeStandardDeviation(std_dev);
EpetraExt::VectorToMatrixMarketFile("mean_gal.mm", mean);
EpetraExt::VectorToMatrixMarketFile("std_dev_gal.mm", std_dev);
// Evaluate SG responses at SG parameters
EpetraExt::ModelEvaluator::InArgs sg_inArgs = sg_model->createInArgs();
EpetraExt::ModelEvaluator::OutArgs sg_outArgs =
sg_model->createOutArgs();
Teuchos::RCP<const Epetra_Vector> sg_p = sg_model->get_p_init(1);
Teuchos::RCP<Epetra_Vector> sg_g =
Teuchos::rcp(new Epetra_Vector(*(sg_model->get_g_map(0))));
sg_inArgs.set_p(1, sg_p);
sg_inArgs.set_x(Teuchos::rcp(&finalSolution,false));
sg_outArgs.set_g(0, sg_g);
sg_model->evalModel(sg_inArgs, sg_outArgs);
// Print mean and standard deviation of response
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> sg_g_poly =
sg_model->create_g_sg(0, View, sg_g.get());
Epetra_Vector g_mean(*(model->get_g_map(0)));
Epetra_Vector g_std_dev(*(model->get_g_map(0)));
sg_g_poly->computeMean(g_mean);
sg_g_poly->computeStandardDeviation(g_std_dev);
std::cout.precision(16);
// std::cout << "\nResponse Expansion = " << std::endl;
// std::cout.precision(12);
// sg_g_poly->print(std::cout);
std::cout << std::endl;
std::cout << "Response Mean = " << std::endl << g_mean << std::endl;
std::cout << "Response Std. Dev. = " << std::endl << g_std_dev << std::endl;
// Determine if example passed
bool passed = false;
if (status == NOX::StatusTest::Converged &&
std::abs(g_mean[0]-g_mean_exp) < g_tol &&
std::abs(g_std_dev[0]-g_std_dev_exp) < g_tol)
passed = true;
if (MyPID == 0) {
if (passed)
std::cout << "Example Passed!" << std::endl;
else
std::cout << "Example Failed!" << std::endl;
}
}
Teuchos::TimeMonitor::summarize(std::cout);
Teuchos::TimeMonitor::zeroOutTimers();
}
catch (std::exception& e) {
std::cout << e.what() << std::endl;
}
catch (string& s) {
std::cout << s << std::endl;
}
catch (char *s) {
std::cout << s << std::endl;
}
catch (...) {
std::cout << "Caught unknown exception!" <<std:: endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
}
int main(int argc, char *argv[]) {
// Initialize MPI
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
#endif
int MyPID;
try {
// Create a communicator for Epetra objects
Teuchos::RCP<const Epetra_Comm> globalComm;
#ifdef HAVE_MPI
globalComm = Teuchos::rcp(new Epetra_MpiComm(MPI_COMM_WORLD));
#else
globalComm = Teuchos::rcp(new Epetra_SerialComm);
#endif
MyPID = globalComm->MyPID();
// Create Stochastic Galerkin basis and expansion
int p = 5;
Teuchos::Array< Teuchos::RCP<const Stokhos::OneDOrthogPolyBasis<int,double> > > bases(1);
bases[0] = Teuchos::rcp(new Stokhos::LegendreBasis<int,double>(p));
Teuchos::RCP<const Stokhos::CompletePolynomialBasis<int,double> > basis =
Teuchos::rcp(new Stokhos::CompletePolynomialBasis<int,double>(bases));
int sz = basis->size();
Teuchos::RCP<Stokhos::Sparse3Tensor<int,double> > Cijk;
Cijk = basis->computeTripleProductTensor(sz);
Teuchos::RCP<Stokhos::OrthogPolyExpansion<int,double> > expansion =
Teuchos::rcp(new Stokhos::AlgebraicOrthogPolyExpansion<int,double>(basis,
Cijk));
if (MyPID == 0)
std::cout << "Stochastic Galerkin expansion size = " << sz << std::endl;
// Create stochastic parallel distribution
int num_spatial_procs = -1;
Teuchos::ParameterList parallelParams;
parallelParams.set("Number of Spatial Processors", num_spatial_procs);
Teuchos::RCP<Stokhos::ParallelData> sg_parallel_data =
Teuchos::rcp(new Stokhos::ParallelData(basis, Cijk, globalComm,
parallelParams));
Teuchos::RCP<const EpetraExt::MultiComm> sg_comm =
sg_parallel_data->getMultiComm();
Teuchos::RCP<const Epetra_Comm> app_comm =
sg_parallel_data->getSpatialComm();
// Create application model evaluator
Teuchos::RCP<EpetraExt::ModelEvaluator> model =
Teuchos::rcp(new SimpleME(app_comm));
// Setup stochastic Galerkin algorithmic parameters
Teuchos::RCP<Teuchos::ParameterList> sgParams =
Teuchos::rcp(new Teuchos::ParameterList);
sgParams->set("Jacobian Method", "Matrix Free");
sgParams->set("Mean Preconditioner Type", "Ifpack");
// Create stochastic Galerkin model evaluator
Teuchos::RCP<Stokhos::SGModelEvaluator> sg_model =
Teuchos::rcp(new Stokhos::SGModelEvaluator(model, basis, Teuchos::null,
expansion, sg_parallel_data,
sgParams));
// Stochastic Galerkin initial guess
// Set the mean to the deterministic initial guess, higher-order terms
// to zero
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> x_init_sg =
sg_model->create_x_sg();
x_init_sg->init(0.0);
(*x_init_sg)[0] = *(model->get_x_init());
sg_model->set_x_sg_init(*x_init_sg);
// Stochastic Galerkin parameters
// Linear expansion with the mean given by the deterministic initial
// parameter values, linear terms equal to 1, and higher order terms
// equal to zero.
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> p_init_sg =
sg_model->create_p_sg(0);
p_init_sg->init(0.0);
(*p_init_sg)[0] = *(model->get_p_init(0));
for (int i=0; i<model->get_p_map(0)->NumMyElements(); i++)
(*p_init_sg)[i+1][i] = 1.0;
sg_model->set_p_sg_init(0, *p_init_sg);
std::cout << "Stochatic Galerkin parameter expansion = " << std::endl
<< *p_init_sg << std::endl;
// Set up NOX parameters
Teuchos::RCP<Teuchos::ParameterList> noxParams =
Teuchos::rcp(new Teuchos::ParameterList);
// Set the nonlinear solver method
noxParams->set("Nonlinear Solver", "Line Search Based");
// Set the printing parameters in the "Printing" sublist
Teuchos::ParameterList& printParams = noxParams->sublist("Printing");
printParams.set("MyPID", MyPID);
printParams.set("Output Precision", 3);
printParams.set("Output Processor", 0);
printParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
//NOX::Utils::Parameters +
//NOX::Utils::Details +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Warning +
NOX::Utils::Error);
// Create printing utilities
NOX::Utils utils(printParams);
// Sublist for line search
Teuchos::ParameterList& searchParams = noxParams->sublist("Line Search");
searchParams.set("Method", "Full Step");
// Sublist for direction
Teuchos::ParameterList& dirParams = noxParams->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", 100);
lsParams.set("Size of Krylov Subspace", 100);
lsParams.set("Tolerance", 1e-4);
lsParams.set("Output Frequency", 10);
lsParams.set("Preconditioner", "Ifpack");
// Sublist for convergence tests
Teuchos::ParameterList& statusParams = noxParams->sublist("Status Tests");
statusParams.set("Test Type", "Combo");
statusParams.set("Number of Tests", 2);
statusParams.set("Combo Type", "OR");
Teuchos::ParameterList& normF = statusParams.sublist("Test 0");
normF.set("Test Type", "NormF");
normF.set("Tolerance", 1e-10);
normF.set("Scale Type", "Scaled");
Teuchos::ParameterList& maxIters = statusParams.sublist("Test 1");
maxIters.set("Test Type", "MaxIters");
maxIters.set("Maximum Iterations", 10);
// Create NOX interface
Teuchos::RCP<NOX::Epetra::ModelEvaluatorInterface> nox_interface =
Teuchos::rcp(new NOX::Epetra::ModelEvaluatorInterface(sg_model));
// Create NOX linear system object
Teuchos::RCP<const Epetra_Vector> u = sg_model->get_x_init();
Teuchos::RCP<Epetra_Operator> A = sg_model->create_W();
Teuchos::RCP<NOX::Epetra::Interface::Required> iReq = nox_interface;
Teuchos::RCP<NOX::Epetra::Interface::Jacobian> iJac = nox_interface;
Teuchos::RCP<NOX::Epetra::LinearSystemAztecOO> linsys;
Teuchos::RCP<Epetra_Operator> M = sg_model->create_WPrec()->PrecOp;
Teuchos::RCP<NOX::Epetra::Interface::Preconditioner> iPrec = nox_interface;
lsParams.set("Preconditioner", "User Defined");
linsys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(printParams, lsParams,
iJac, A, iPrec, M,
*u));
// linsys =
// Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(printParams, lsParams,
// iReq, iJac, A, *u));
// Build NOX group
Teuchos::RCP<NOX::Epetra::Group> grp =
Teuchos::rcp(new NOX::Epetra::Group(printParams, iReq, *u, linsys));
// Create the Solver convergence test
Teuchos::RCP<NOX::StatusTest::Generic> statusTests =
NOX::StatusTest::buildStatusTests(statusParams, utils);
// Create the solver
Teuchos::RCP<NOX::Solver::Generic> solver =
NOX::Solver::buildSolver(grp, statusTests, noxParams);
// Solve the system
NOX::StatusTest::StatusType status = solver->solve();
// Get final solution
const NOX::Epetra::Group& finalGroup =
dynamic_cast<const NOX::Epetra::Group&>(solver->getSolutionGroup());
const Epetra_Vector& finalSolution =
(dynamic_cast<const NOX::Epetra::Vector&>(finalGroup.getX())).getEpetraVector();
// Convert block Epetra_Vector to orthogonal polynomial of Epetra_Vector's
Teuchos::RCP<Stokhos::EpetraVectorOrthogPoly> x_sg =
sg_model->create_x_sg(View, &finalSolution);
utils.out() << "Final Solution (block vector) = " << std::endl;
std::cout << finalSolution << std::endl;
utils.out() << "Final Solution (polynomial) = " << std::endl;
std::cout << *x_sg << std::endl;
if (status == NOX::StatusTest::Converged && MyPID == 0)
utils.out() << "Example Passed!" << std::endl;
}
catch (std::exception& e) {
std::cout << e.what() << std::endl;
}
catch (string& s) {
std::cout << s << std::endl;
}
catch (char *s) {
std::cout << s << std::endl;
}
catch (...) {
std::cout << "Caught unknown exception!" <<std:: endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
}
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;
Teuchos::GlobalMPISession mpiSession(&argc, &argv, NULL);
RCP<const Teuchos::Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
// Read the --linAlgebra option
::Xpetra::UnderlyingLib lib;
{
// Note: use --help to list available options.
Teuchos::CommandLineProcessor clp(false);
::Xpetra::Parameters xpetraParameters(clp);
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;
}
if (comm->getRank() == 0) std::cout << xpetraParameters;
lib = xpetraParameters.GetLib();
}
{
// Creates an Xpetra::Map corresponding to a 2D Cartesian grid
// on the unit square. For parallel runs, the nodes are divided into
// strips, so that the total number of subdomains is comm->getSize() x 1.
// Type of the object
string mapType = "Cartesian2D";
// Container for parameters
Teuchos::ParameterList galeriList;
galeriList.set("nx", (GO) 2 * comm->getSize());
galeriList.set("ny", (GO) 2);
galeriList.set("mx", (GO) comm->getSize());
galeriList.set("my", (GO) 1);
// Creation of the map
RCP< ::Xpetra::Map<int, GO, KokkosClassic::DefaultNode::DefaultNodeType> > map = Galeri::Xpetra::CreateMap<int, GO, KokkosClassic::DefaultNode::DefaultNodeType>(lib, "Cartesian2D", comm, galeriList);
// Print out the parameters
cout << galeriList;
// Print out the map
RCP<Teuchos::FancyOStream> fos = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout));
map->describe(*fos, Teuchos::VERB_EXTREME);
}
{
// Creates an Xpetra::Map corresponding to a 3D Cartesian grid
// Type of the object
string mapType = "Cartesian3D";
// Container for parameters
Teuchos::ParameterList galeriList;
galeriList.set("nx", (GO) 2 * comm->getSize());
galeriList.set("ny", (GO) 2);
galeriList.set("nz", (GO) 2);
// Creation of the map
RCP< ::Xpetra::Map<int, GO, KokkosClassic::DefaultNode::DefaultNodeType> > map = Galeri::Xpetra::CreateMap<int, GO, KokkosClassic::DefaultNode::DefaultNodeType>(lib, "Cartesian3D", comm, galeriList);
// Print out the parameters
cout << galeriList;
// Print out the map
RCP<Teuchos::FancyOStream> fos = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout));
map->describe(*fos, Teuchos::VERB_EXTREME);
}
return EXIT_SUCCESS;
}
int main(int argc, char *argv[]) {
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm (MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
// The problem is defined on a 2D grid, global size is nx * nx.
int nx = 30;
Teuchos::ParameterList GaleriList;
GaleriList.set("nx", nx);
GaleriList.set("ny", nx * Comm.NumProc());
GaleriList.set("mx", 1);
GaleriList.set("my", Comm.NumProc());
Teuchos::RefCountPtr<Epetra_Map> Map = Teuchos::rcp( Galeri::CreateMap("Cartesian2D", Comm, GaleriList) );
Teuchos::RefCountPtr<Epetra_CrsMatrix> A = Teuchos::rcp( Galeri::CreateCrsMatrix("Laplace2D", &*Map, GaleriList) );
Teuchos::RefCountPtr<Epetra_MultiVector> LHS = Teuchos::rcp( new Epetra_MultiVector(*Map, 1) );
Teuchos::RefCountPtr<Epetra_MultiVector> RHS = Teuchos::rcp( new Epetra_MultiVector(*Map, 1) );
LHS->PutScalar(0.0); RHS->Random();
// ========================================= //
// Compare IC preconditioners to no precond. //
// ----------------------------------------- //
const double tol = 1e-5;
const int maxIter = 500;
// Baseline: No preconditioning
// Compute number of iterations, to compare to IC later.
// Here we create an AztecOO object
LHS->PutScalar(0.0);
AztecOO solver;
solver.SetUserMatrix(&*A);
solver.SetLHS(&*LHS);
solver.SetRHS(&*RHS);
solver.SetAztecOption(AZ_solver,AZ_cg);
//solver.SetPrecOperator(&*PrecDiag);
solver.SetAztecOption(AZ_output, 16);
solver.Iterate(maxIter, tol);
int Iters = solver.NumIters();
//cout << "No preconditioner iterations: " << Iters << endl;
#if 0
// Not sure how to use Ifpack_CrsRick - leave out for now.
//
// I wanna test funky values to be sure that they have the same
// influence on the algorithms, both old and new
int LevelFill = 2;
double DropTol = 0.3333;
double Condest;
Teuchos::RefCountPtr<Ifpack_CrsRick> IC;
Ifpack_IlukGraph mygraph (A->Graph(), 0, 0);
IC = Teuchos::rcp( new Ifpack_CrsRick(*A, mygraph) );
IC->SetAbsoluteThreshold(0.00123);
IC->SetRelativeThreshold(0.9876);
// Init values from A
IC->InitValues(*A);
// compute the factors
IC->Factor();
// and now estimate the condition number
IC->Condest(false,Condest);
if( Comm.MyPID() == 0 ) {
cout << "Condition number estimate (level-of-fill = "
<< LevelFill << ") = " << Condest << endl;
}
// Define label for printing out during the solve phase
std::string label = "Ifpack_CrsRick Preconditioner: LevelFill = " + toString(LevelFill) +
" Overlap = 0";
IC->SetLabel(label.c_str());
// Here we create an AztecOO object
LHS->PutScalar(0.0);
AztecOO solver;
solver.SetUserMatrix(&*A);
solver.SetLHS(&*LHS);
solver.SetRHS(&*RHS);
solver.SetAztecOption(AZ_solver,AZ_cg);
solver.SetPrecOperator(&*IC);
solver.SetAztecOption(AZ_output, 16);
solver.Iterate(maxIter, tol);
int RickIters = solver.NumIters();
//cout << "Ifpack_Rick iterations: " << RickIters << endl;
// Compare to no preconditioning
if (RickIters > Iters/2)
IFPACK_CHK_ERR(-1);
#endif
//////////////////////////////////////////////////////
// Same test with Ifpack_IC
// This is Crout threshold Cholesky, so different than IC(0)
Ifpack Factory;
Teuchos::RefCountPtr<Ifpack_Preconditioner> PrecIC = Teuchos::rcp( Factory.Create("IC", &*A) );
Teuchos::ParameterList List;
//List.get("fact: ict level-of-fill", 2.);
//List.get("fact: drop tolerance", 0.3333);
//List.get("fact: absolute threshold", 0.00123);
//List.get("fact: relative threshold", 0.9876);
//List.get("fact: relaxation value", 0.0);
IFPACK_CHK_ERR(PrecIC->SetParameters(List));
IFPACK_CHK_ERR(PrecIC->Compute());
// Here we create an AztecOO object
LHS->PutScalar(0.0);
//AztecOO solver;
solver.SetUserMatrix(&*A);
solver.SetLHS(&*LHS);
solver.SetRHS(&*RHS);
solver.SetAztecOption(AZ_solver,AZ_cg);
solver.SetPrecOperator(&*PrecIC);
solver.SetAztecOption(AZ_output, 16);
solver.Iterate(maxIter, tol);
int ICIters = solver.NumIters();
//cout << "Ifpack_IC iterations: " << ICIters << endl;
// Compare to no preconditioning
if (ICIters > Iters/2)
IFPACK_CHK_ERR(-1);
#if 0
//////////////////////////////////////////////////////
// Same test with Ifpack_ICT
// This is another threshold Cholesky
Teuchos::RefCountPtr<Ifpack_Preconditioner> PrecICT = Teuchos::rcp( Factory.Create("ICT", &*A) );
//Teuchos::ParameterList List;
//List.get("fact: level-of-fill", 2);
//List.get("fact: drop tolerance", 0.3333);
//List.get("fact: absolute threshold", 0.00123);
//List.get("fact: relative threshold", 0.9876);
//List.get("fact: relaxation value", 0.0);
IFPACK_CHK_ERR(PrecICT->SetParameters(List));
IFPACK_CHK_ERR(PrecICT->Compute());
// Here we create an AztecOO object
LHS->PutScalar(0.0);
solver.SetUserMatrix(&*A);
solver.SetLHS(&*LHS);
solver.SetRHS(&*RHS);
solver.SetAztecOption(AZ_solver,AZ_cg);
solver.SetPrecOperator(&*PrecICT);
solver.SetAztecOption(AZ_output, 16);
solver.Iterate(maxIter, tol);
int ICTIters = solver.NumIters();
//cout << "Ifpack_ICT iterations: " << ICTIters << endl;
// Compare to no preconditioning
if (ICTIters > Iters/2)
IFPACK_CHK_ERR(-1);
#endif
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return(EXIT_SUCCESS);
}
int main(int argc, char **argv)
{
Kokkos::initialize();
try {
// Create random field
int M = 10;
Teuchos::ParameterList solverParams;
solverParams.set("Number of KL Terms", M);
solverParams.set("Mean", 1.0);
solverParams.set("Standard Deviation", 0.1);
int ndim = 3;
Teuchos::Array<double> domain_upper(ndim), domain_lower(ndim),
correlation_length(ndim);
for (int i=0; i<ndim; i++) {
domain_upper[i] = 1.0;
domain_lower[i] = 0.0;
correlation_length[i] = 10.0;
}
solverParams.set("Domain Upper Bounds", domain_upper);
solverParams.set("Domain Lower Bounds", domain_lower);
solverParams.set("Correlation Lengths", correlation_length);
Stokhos::KL::ExponentialRandomField<double> rf(solverParams);
rf.print(std::cout);
// Evaluate random field at a point
Teuchos::Array<double> x(ndim);
for (int i=0; i<ndim; i++)
x[i] = (domain_upper[i] + domain_lower[i])/2.0 +
0.1*(domain_upper[i] - domain_lower[i])/2.0;
Teuchos::Array<double> rvar(M);
for (int i=0; i<M; i++)
rvar[i] = 1.5;
double result = rf.evaluate(x, rvar);
std::cout << "result (host) = " << result << std::endl;
// Evaluate random field in a functor on device
typedef Kokkos::View<double*> view_type;
typedef view_type::HostMirror host_view_type;
view_type x_view("x", ndim);
host_view_type host_x = Kokkos::create_mirror_view(x_view);
for (int i=0; i<ndim; i++)
host_x(i) = x[i];
Kokkos::deep_copy(x_view, host_x);
view_type rvar_view("rvar", M);
host_view_type host_rvar = Kokkos::create_mirror_view(rvar_view);
for (int i=0; i<M; i++)
host_rvar(i) = rvar[i];
Kokkos::deep_copy(rvar_view, host_rvar);
RF<double> rf_func(rf, x_view, rvar_view);
host_view_type host_y = Kokkos::create_mirror_view(rf_func.y);
Kokkos::deep_copy(host_y, rf_func.y);
double result2 = host_y(0);
std::cout << "result (device)= " << result2 << std::endl;
}
catch (std::exception& e) {
std::cout << e.what() << std::endl;
}
Kokkos::finalize();
}
// ======================================================================
bool TestContainer(string Type, const Teuchos::RefCountPtr<Epetra_RowMatrix>& A)
{
int NumVectors = 3;
int NumMyRows = A->NumMyRows();
Epetra_MultiVector LHS_exact(A->RowMatrixRowMap(), NumVectors);
Epetra_MultiVector LHS(A->RowMatrixRowMap(), NumVectors);
Epetra_MultiVector RHS(A->RowMatrixRowMap(), NumVectors);
LHS_exact.Random(); LHS.PutScalar(0.0);
A->Multiply(false, LHS_exact, RHS);
Epetra_LinearProblem Problem(&*A, &LHS, &RHS);
if (verbose) {
cout << "Container type = " << Type << endl;
cout << "NumMyRows = " << NumMyRows << ", NumVectors = " << NumVectors << endl;
}
LHS.PutScalar(0.0);
Teuchos::RefCountPtr<Ifpack_Container> Container;
if (Type == "dense")
Container = Teuchos::rcp( new Ifpack_DenseContainer(A->NumMyRows(), NumVectors) );
else
Container = Teuchos::rcp( new Ifpack_SparseContainer<Ifpack_Amesos>(A->NumMyRows(), NumVectors) );
assert (Container != Teuchos::null);
IFPACK_CHK_ERR(Container->Initialize());
// set as ID all the local rows of A
for (int i = 0 ; i < A->NumMyRows() ; ++i)
Container->ID(i) = i;
// extract submatrix (in this case, the entire matrix)
// and complete setup
IFPACK_CHK_ERR(Container->Compute(*A));
// set the RHS and LHS
for (int i = 0 ; i < A->NumMyRows() ; ++i)
for (int j = 0 ; j < NumVectors ; ++j) {
Container->RHS(i,j) = RHS[j][i];
Container->LHS(i,j) = LHS[j][i];
}
// set parameters (empty for dense containers)
Teuchos::ParameterList List;
List.set("amesos: solver type", Type);
IFPACK_CHK_ERR(Container->SetParameters(List));
// solve the linear system
IFPACK_CHK_ERR(Container->ApplyInverse());
// get the computed solution, store it in LHS
for (int i = 0 ; i < A->NumMyRows() ; ++i)
for (int j = 0 ; j < NumVectors ; ++j) {
LHS[j][i] = Container->LHS(i,j);
}
double residual = Galeri::ComputeNorm(&LHS, &LHS_exact);
if (A->Comm().MyPID() == 0 && verbose) {
cout << "||x_exact - x||_2 = " << residual << endl;
cout << *Container;
}
bool passed = false;
if (residual < 1e-5)
passed = true;
return(passed);
}
int main(int argc, char *argv[]) {
using Teuchos::RCP;
Teuchos::oblackholestream blackhole;
Teuchos::GlobalMPISession mpiSession(&argc,&argv,&blackhole);
RCP<const Teuchos::Comm<int> > comm = Teuchos::DefaultComm<int>::getComm();
RCP<Teuchos::FancyOStream> out = Teuchos::fancyOStream(Teuchos::rcpFromRef(std::cout));
out->setOutputToRootOnly(0);
*out << MueLu::MemUtils::PrintMemoryUsage() << std::endl;
// Timing
Teuchos::Time myTime("global");
Teuchos::TimeMonitor M(myTime);
#ifndef HAVE_TEUCHOS_LONG_LONG_INT
*out << "Warning: scaling test was not compiled with long long int support" << std::endl;
#endif
// custom parameters
LO maxLevels = 10;
LO its=40;
std::string smooType="sgs";
int sweeps=1;
int maxCoarseSize=1; //FIXME clp doesn't like long long int
std::string aggOrdering = "natural";
int minPerAgg=2;
int maxNbrAlreadySelected=0;
// read in problem
Epetra_Map emap(10201,0,*Xpetra::toEpetra(comm));
Epetra_CrsMatrix * ptrA = 0;
Epetra_Vector * ptrf = 0;
std::cout << "Reading matrix market file" << std::endl;
EpetraExt::MatrixMarketFileToCrsMatrix("Condif2Mat.mat",emap,emap,emap,ptrA);
EpetraExt::MatrixMarketFileToVector("Condif2Rhs.mat",emap,ptrf);
RCP<Epetra_CrsMatrix> epA = Teuchos::rcp(ptrA);
RCP<Epetra_Vector> epv = Teuchos::rcp(ptrf);
// Epetra_CrsMatrix -> Xpetra::Matrix
RCP<Xpetra::CrsMatrix<double, int, int> > exA = Teuchos::rcp(new Xpetra::EpetraCrsMatrix(epA));
RCP<Xpetra::CrsMatrixWrap<double, int, int> > crsOp = Teuchos::rcp(new Xpetra::CrsMatrixWrap<double, int, int>(exA));
RCP<Xpetra::Matrix<double, int, int> > Op = Teuchos::rcp_dynamic_cast<Xpetra::Matrix<double, int, int> >(crsOp);
// Epetra_Vector -> Xpetra::Vector
RCP<Xpetra::Vector<double,int,int> > xRhs = Teuchos::rcp(new Xpetra::EpetraVector(epv));
// Epetra_Map -> Xpetra::Map
const RCP< const Xpetra::Map<int, int> > map = Xpetra::toXpetra(emap);
// build nullspace
RCP<MultiVector> nullSpace = MultiVectorFactory::Build(map,1);
nullSpace->putScalar( (SC) 1.0);
/*for (size_t i=0; i<nullSpace->getLocalLength(); i++) {
Teuchos::ArrayRCP< Scalar > data0 = nullSpace->getDataNonConst(0);
Teuchos::ArrayRCP< Scalar > data1 = nullSpace->getDataNonConst(1);
GlobalOrdinal gid = map->getGlobalElement(Teuchos::as<LocalOrdinal>(i));
if(gid % 2 == 0) {
data0[i] = 1.0; data1[i] = 0.0;
}
else {
data0[i] = 0.0; data1[i] = 1.0;
}
}*/
RCP<MueLu::Hierarchy<SC,LO,GO,NO,LMO> > H = rcp ( new Hierarchy() );
H->setDefaultVerbLevel(Teuchos::VERB_HIGH);
H->SetMaxCoarseSize((GO) maxCoarseSize);;
// build finest Level
RCP<MueLu::Level> Finest = H->GetLevel();
Finest->setDefaultVerbLevel(Teuchos::VERB_HIGH);
Finest->Set("A",Op);
Finest->Set("Nullspace",nullSpace);
RCP<CoupledAggregationFactory> CoupledAggFact = rcp(new CoupledAggregationFactory());
*out << "========================= Aggregate option summary =========================" << std::endl;
*out << "min DOFs per aggregate : " << minPerAgg << std::endl;
*out << "min # of root nbrs already aggregated : " << maxNbrAlreadySelected << std::endl;
CoupledAggFact->SetMinNodesPerAggregate(minPerAgg); //TODO should increase if run anything other than 1D
CoupledAggFact->SetMaxNeighAlreadySelected(maxNbrAlreadySelected);
std::transform(aggOrdering.begin(), aggOrdering.end(), aggOrdering.begin(), ::tolower);
if (aggOrdering == "natural") {
*out << "aggregate ordering : NATURAL" << std::endl;
CoupledAggFact->SetOrdering(MueLu::AggOptions::NATURAL);
} else if (aggOrdering == "random") {
*out << "aggregate ordering : RANDOM" << std::endl;
CoupledAggFact->SetOrdering(MueLu::AggOptions::RANDOM);
} else if (aggOrdering == "graph") {
*out << "aggregate ordering : GRAPH" << std::endl;
CoupledAggFact->SetOrdering(MueLu::AggOptions::GRAPH);
} else {
std::string msg = "main: bad aggregation option """ + aggOrdering + """.";
throw(MueLu::Exceptions::RuntimeError(msg));
}
CoupledAggFact->SetPhase3AggCreation(0.5);
*out << "=============================================================================" << std::endl;
// build transfer operators
RCP<TentativePFactory> TentPFact = rcp(new TentativePFactory(CoupledAggFact));
*out << " afer TentativePFactory " << std::endl;
//RCP<TentativePFactory> Pfact = rcp(new TentativePFactory(CoupledAggFact));
//RCP<Factory> Rfact = rcp( new TransPFactory(Pfact));
//RCP<SaPFactory> Pfact = rcp( new SaPFactory(TentPFact) );
//RCP<Factory> Rfact = rcp( new TransPFactory(Pfact));
RCP<ThresholdAFilterFactory> Afiltered = rcp(new ThresholdAFilterFactory("A",NULL,0.0005));
RCP<PgPFactory> Pfact = rcp( new PgPFactory(TentPFact,Afiltered) );
RCP<Factory> Rfact = rcp( new GenericRFactory(Pfact));
RCP<RAPFactory> Acfact = rcp( new RAPFactory(Pfact, Rfact) );
Acfact->setVerbLevel(Teuchos::VERB_HIGH);
*out << " after ACFactory " << std::endl;
// build level smoothers
RCP<SmootherPrototype> smooProto;
std::string ifpackType;
Teuchos::ParameterList ifpackList;
ifpackList.set("relaxation: sweeps", (LO) sweeps);
ifpackList.set("relaxation: damping factor", (SC) 0.9); // 0.7
ifpackType = "RELAXATION";
ifpackList.set("relaxation: type", "Gauss-Seidel");
smooProto = Teuchos::rcp( new TrilinosSmoother(Xpetra::UseEpetra, ifpackType, ifpackList) );
RCP<SmootherFactory> SmooFact;
if (maxLevels > 1)
SmooFact = rcp( new SmootherFactory(smooProto) );
Teuchos::ParameterList status;
#if 0 // both variants are equivalent
// fill FactoryManager object by hand
FactoryManager Manager;
Manager.SetFactory("A", Acfact);
Manager.SetFactory("P", Pfact);
Manager.SetFactory("R", Rfact);
Manager.SetFactory("Smoother", SmooFact);
Manager.SetFactory("CoarseSolver", Teuchos::null);
status = H->Setup(Manager);
#else
// use FullPopulate (short, but outdated?)
status = H->FullPopulate(*Pfact,*Rfact,*Acfact,*SmooFact,0,maxLevels);
#endif
H->SetCoarsestSolver(*SmooFact,MueLu::PRE);
*out << "======================\n Multigrid statistics \n======================" << std::endl;
status.print(*out,Teuchos::ParameterList::PrintOptions().indent(2));
/**********************************************************************************/
/* SOLVE PROBLEM */
/**********************************************************************************/
RCP<MultiVector> x = MultiVectorFactory::Build(map,1);
RCP<Xpetra::MultiVector<double,int,int> > rhs = Teuchos::rcp_dynamic_cast<Xpetra::MultiVector<double,int,int> >(xRhs);//(new Xpetra::EpetraVector(epv));
// Use AMG directly as an iterative method
{
x->putScalar( (SC) 0.0);
H->Iterate(*rhs,its,*x);
//x->describe(*out,Teuchos::VERB_EXTREME);
}
RCP<Level> coarseLevel = H->GetLevel(1);
coarseLevel->print(*out);
RCP<Level> coarseLevel2 = H->GetLevel(1);
coarseLevel2->print(*out);
//M.summarize();
return EXIT_SUCCESS;
}