void InverseOperator::Reshape(Ifpack_Preconditioner* prec, const Operator& Op,
const bool ownership)
{
ResetTimer();
StackPush();
Op_ = Op;
RCPRowMatrix_ = Op.GetRCPRowMatrix();
RCPData_ = Teuchos::rcp(prec,ownership);
StackPop();
UpdateTime();
}
void InverseOperator::Reshape(const Operator& Op, const string Type,
Teuchos::ParameterList& List, Teuchos::ParameterList* pushlist)
{
ResetTimer();
StackPush();
Op_ = Op;
RCPRowMatrix_ = Op.GetRCPRowMatrix();
// FIXME: to add overlap and level-of-fill
int NumSweeps = List.get("smoother: sweeps", 1);
double Damping = List.get("smoother: damping factor", 0.67);
int LOF_ilu = List.get("smoother: ilu fill", 0);
double LOF_ict = List.get("smoother: ilut fill", 1.0);
double LOF_ilut = List.get("smoother: ict fill", 1.0);
string reorder = List.get("schwarz: reordering type","rcm");
Teuchos::ParameterList IFPACKList;
// any parameters from the main list List are overwritten by the pushlist
IFPACKList.set("relaxation: sweeps", NumSweeps);
IFPACKList.set("relaxation: damping factor", Damping);
IFPACKList.set("fact: level-of-fill", LOF_ilu);
IFPACKList.set("fact: ict level-of-fill", LOF_ict);
IFPACKList.set("fact: ilut level-of-fill", LOF_ilut);
IFPACKList.set("relaxation: zero starting solution", false);
IFPACKList.set("schwarz: reordering type",reorder);
IFPACKList.set("fact: relative threshold",1.0);
// if present, the pushlist is assumed to be a preconstructed ifpack list
// that is copied straight to the ifpack list here
// entries in pushlist overwrite previous list entries
if (pushlist)
IFPACKList.setParameters(*pushlist);
bool verbose = false; //(GetMyPID() == 0 && GetPrintLevel() > 5);
// the ML smoother
RCPMLPrec_ = Teuchos::null;
// The Ifpack smoother
Ifpack_Preconditioner* Prec = NULL;
if (Type == "Jacobi") {
if (verbose) {
cout << "Damping factor = " << Damping
<< ", sweeps = " << NumSweeps << endl;
cout << endl;
}
IFPACKList.set("relaxation: type", "Jacobi");
Prec = new Ifpack_PointRelaxation(RowMatrix());
}
else if (Type == "Gauss-Seidel") {
if (verbose) {
cout << "Damping factor = " << Damping
<< ", sweeps = " << NumSweeps << endl;
cout << endl;
}
IFPACKList.set("relaxation: type", "Gauss-Seidel");
Prec = new Ifpack_PointRelaxation(RowMatrix());
}
else if (Type == "symmetric Gauss-Seidel") {
if (verbose) {
cout << "Damping factor = " << Damping
<< ", sweeps = " << NumSweeps << endl;
cout << endl;
}
IFPACKList.set("relaxation: type", "symmetric Gauss-Seidel");
Prec = new Ifpack_PointRelaxation(RowMatrix());
}
else if (Type == "ILU") {
if (verbose) {
cout << "ILU factorization, ov = 0, no reordering, LOF = "
<< LOF_ilu << endl;
cout << endl;
}
// use the Additive Schwarz class because it does reordering
Prec = new Ifpack_AdditiveSchwarz<Ifpack_ILU>(RowMatrix());
}
else if (Type == "ILUT") {
if (verbose) {
cout << "ILUT factorization, ov = 0, no reordering, LOF = "
<< LOF_ilu << endl;
cout << endl;
}
Prec = new Ifpack_ILUT(RowMatrix());
}
else if (Type == "IC") {
if (verbose) {
cout << "IC factorization, ov = 0, no reordering, LOF = "
<< LOF_ilu << endl;
cout << endl;
}
Prec = new Ifpack_IC(RowMatrix());
}
else if (Type == "ICT") {
if (verbose) {
cout << "ICT factorization, ov = 0, no reordering, LOF = "
<< LOF_ilu << endl;
cout << endl;
}
Prec = new Ifpack_ICT(RowMatrix());
}
else if (Type == "LU") {
if (verbose) {
cout << "LU factorization, ov = 0, local solver = KLU" << endl;
cout << endl;
}
Prec = new Ifpack_AdditiveSchwarz<Ifpack_Amesos>(RowMatrix());
}
else if (Type == "Amesos" || Type == "Amesos-KLU") {
if (verbose) {
cout << "Amesos-KLU direct solver" << endl;
cout << endl;
}
Prec = new Ifpack_Amesos(RowMatrix());
}
else if (Type == "MLS" || Type == "ML MLS" ||
Type == "ML symmetric Gauss-Seidel" ||
Type == "ML Gauss-Seidel")
{
if (verbose) {
cout << "ML's MLS smoother" << endl;
cout << endl;
}
Teuchos::ParameterList mlparams;
ML_Epetra::SetDefaults("SA",mlparams);
int output = List.get("ML output",-47);
if (output == -47) output = List.get("output",-1);
if (output != -1) mlparams.set("ML output",output);
mlparams.set("max levels",1);
int sweeps = List.get("smoother: sweeps",1);
mlparams.set("coarse: sweeps",sweeps);
double damp = List.get("smoother: damping factor",0.67);
mlparams.set("coarse: damping factor",damp);
mlparams.set("zero starting solution", false);
if (Type == "MLS" || Type == "ML MLS")
{
mlparams.set("coarse: type","MLS"); // MLS symmetric Gauss-Seidel Amesos-KLU
int poly = List.get("smoother: MLS polynomial order",3);
mlparams.set("coarse: MLS polynomial order",poly);
}
else if (Type == "ML symmetric Gauss-Seidel")
mlparams.set("coarse: type","symmetric Gauss-Seidel"); // MLS symmetric Gauss-Seidel Amesos-KLU
else if (Type == "ML Gauss-Seidel")
mlparams.set("coarse: type","Gauss-Seidel");
else if (Type == "ML Jacobi")
mlparams.set("coarse: type","Jacobi");
else
ML_THROW("Requested type (" + Type + ") not recognized", -1);
RCPMLPrec_ = Teuchos::rcp(new ML_Epetra::MultiLevelPreconditioner(*RowMatrix(),mlparams,true));
}
else
ML_THROW("Requested type (" + Type + ") not recognized", -1);
if (Prec)
{
RCPData_ = Teuchos::rcp(Prec);
RCPData_->SetParameters(IFPACKList);
RCPData_->Initialize();
RCPData_->Compute();
UpdateFlops(RCPData_->InitializeFlops());
UpdateFlops(RCPData_->ComputeFlops());
}
else
RCPData_ = Teuchos::null;
StackPop();
UpdateTime();
}