/*!
* \brief Solve.
*/
void JacobiSolver::iterate( const int max_iters, const double tolerance )
{
// Extract the linear problem.
Epetra_CrsMatrix *A =
dynamic_cast<Epetra_CrsMatrix*>( d_linear_problem->GetMatrix() );
Epetra_Vector *x =
dynamic_cast<Epetra_Vector*>( d_linear_problem->GetLHS() );
const Epetra_Vector *b =
dynamic_cast<Epetra_Vector*>( d_linear_problem->GetRHS() );
// Setup the residual.
Epetra_Map row_map = A->RowMap();
Epetra_Vector residual( row_map );
// Iterate.
Epetra_CrsMatrix H = buildH( A );
Epetra_Vector temp_vec( row_map );
d_num_iters = 0;
double residual_norm = 1.0;
double b_norm;
b->NormInf( &b_norm );
double conv_crit = b_norm*tolerance;
while ( residual_norm > conv_crit && d_num_iters < max_iters )
{
H.Apply( *x, temp_vec );
x->Update( 1.0, temp_vec, 1.0, *b, 0.0 );
A->Apply( *x, temp_vec );
residual.Update( 1.0, *b, -1.0, temp_vec, 0.0 );
residual.NormInf( &residual_norm );
++d_num_iters;
}
}
int power_method(bool TransA, Epetra_RowMatrix& A, Epetra_Vector& q, Epetra_Vector& z0,
Epetra_Vector& resid, double* lambda, int niters, double tolerance, bool verbose)
{
// Fill z with random Numbers
Epetra_Vector z(z0);
// variable needed for iteration
double normz, residual;
int ierr = 1;
for(int iter = 0; iter < niters; iter++) {
z.Norm2(&normz); // Compute 2-norm of z
q.Scale(1.0/normz, z);
A.Multiply(TransA, q, z); // Compute z = A*q // SEGFAULT HAPPENS HERE
q.Dot(z, lambda); // Approximate maximum eigenvaluE
if(iter%100==0 || iter+1==niters) {
resid.Update(1.0, z, -(*lambda), q, 0.0); // Compute A*q - lambda*q
resid.Norm2(&residual);
if(verbose) cout << "Iter = " << iter << " Lambda = " << *lambda
<< " Residual of A*q - lambda*q = " << residual << endl;
}
if(residual < tolerance) {
ierr = 0;
break;
}
}
return(ierr);
}
/*----------------------------------------------------------------------*
| evaluate nonlinear function (public, derived) m.gee 3/06|
*----------------------------------------------------------------------*/
bool NLNML::NLNML_CoarseLevelNoxInterface::computeF(
const Epetra_Vector& x, Epetra_Vector& F,
const FillType fillFlag)
{
bool err;
if (!Level())
{
err = fineinterface_->computeF(x,F,fillFlag);
if (err==false)
{
cout << "**ERR**: NLNML::NLNML_CoarseLevelNoxInterface::computeF:\n"
<< "**ERR**: call to fine-userinterface returned false on level " << level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
}
else
{
RefCountPtr<Epetra_Vector> Ffine = rcp(new Epetra_Vector(fineinterface_->getGraph()->RowMap(),false));
RefCountPtr<Epetra_Vector> xfine = rcp(prolong_this_to_fine(x));
err = fineinterface_->computeF(*xfine,*Ffine,fillFlag);
if (err==false)
{
cout << "**ERR**: NLNML::NLNML_CoarseLevelNoxInterface::computeF:\n"
<< "**ERR**: call to fine-userinterface returned false on level " << level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
RefCountPtr<Epetra_Vector> Fcoarse = rcp(restrict_fine_to_this(*Ffine));
F.Scale(1.0,*Fcoarse);
}
if (isFAS())
F.Update(-1.0,*fxbar_,1.0,*fbar_,1.0);
return err;
}
int
powerMethod (double & lambda,
Epetra_CrsMatrix& A,
const int niters,
const double tolerance,
const bool verbose)
{
// In the power iteration, z = A*q. Thus, q must be in the domain
// of A, and z must be in the range of A. The residual vector is of
// course in the range of A.
Epetra_Vector q (A.OperatorDomainMap ());
Epetra_Vector z (A.OperatorRangeMap ());
Epetra_Vector resid (A.OperatorRangeMap ());
Epetra_Flops* counter = A.GetFlopCounter();
if (counter != 0) {
q.SetFlopCounter(A);
z.SetFlopCounter(A);
resid.SetFlopCounter(A);
}
// Initialize the starting vector z with random data.
z.Random();
double normz, residual;
int ierr = 1;
for (int iter = 0; iter < niters; ++iter)
{
z.Norm2 (&normz); // normz := ||z||_2
q.Scale (1.0/normz, z); // q := z / normz
A.Multiply(false, q, z); // z := A * q
q.Dot(z, &lambda); // lambda := dot (q, z)
// Compute the residual vector and display status output every
// 100 iterations, or if we have reached the maximum number of
// iterations.
if (iter % 100 == 0 || iter + 1 == niters)
{
resid.Update (1.0, z, -lambda, q, 0.0); // resid := A*q - lambda*q
resid.Norm2 (&residual); // residual := ||resid||_2
if (verbose)
cout << "Iter = " << iter << " Lambda = " << lambda
<< " Residual of A*q - lambda*q = " << residual << endl;
}
if (residual < tolerance) { // We've converged!
ierr = 0;
break;
}
}
return ierr;
}
void
Stokhos::ProductEpetraVector::
assignToBlockVector(Epetra_Vector& v) const
{
if (this->size() > 0) {
if (bv != Teuchos::null)
v.Update(1.0, *bv, 0.0);
else {
EpetraExt::BlockVector block_v(View, *coeff_map, v);
for (int i=0; i<this->size(); i++)
*(block_v.GetBlock(i)) = *(coeff_[i]);
}
}
}
/*----------------------------------------------------------------------*
| Constructor (public) m.gee 12/04|
| IMPORTANT: |
| No matter on which level we are here, the vector xfine is ALWAYS |
| a fine grid vector here! |
*----------------------------------------------------------------------*/
ML_NOX::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel(int level, int nlevel, int plevel,
ML* ml, ML_Aggregate* ag, Epetra_CrsMatrix** P,
ML_NOX::Ml_Nox_Fineinterface& interface, const Epetra_Comm& comm,
const Epetra_Vector& xfine, double fd_alpha, double fd_beta,
bool fd_centered, bool isDiagonalOnly, int bsize)
: fineinterface_(interface),
comm_(comm)
{
level_ = level;
nlevel_ = nlevel;
ml_printlevel_ = plevel;
ml_ = ml;
ag_ = ag;
fd_alpha_ = fd_alpha;
fd_beta_ = fd_beta;
fd_centered_ = fd_centered;
isDiagonalOnly_ = isDiagonalOnly;
A_ = 0;
coarseinterface_= 0;
bsize_ = bsize;
// we need the graph of the operator on this level. On the fine grid we can just ask the
// fineinterface for it, on the coarser levels it has to be extracted from the ML-hierachy
if (level_==0)
{
// the Epetra_CrsGraph-copy-constructor shares data with the original one.
// We want a really deep copy here so we cannot use it
// graph_ will be given to the FiniteDifferencing class and will be destroyed by it
graph_ = ML_NOX::deepcopy_graph(interface.getGraph());
}
else
{
// Note that ML has no understanding of global indices, so it makes up new GIDs
// (This also holds for the Prolongators P)
Epetra_CrsMatrix* tmpMat = 0;
int maxnnz = 0;
double cputime = 0.0;
ML_Operator2EpetraCrsMatrix(&(ml_->Amat[level_]), tmpMat, maxnnz, false, cputime);
// copy the graph
double t0 = GetClock();
graph_ = ML_NOX::deepcopy_graph(&(tmpMat->Graph()));
// delete the copy of the Epetra_CrsMatrix
if (tmpMat) delete tmpMat; tmpMat = 0;
double t1 = GetClock();
if (ml_printlevel_ > 0 && 0 == comm_.MyPID())
cout << "matrixfreeML (level " << level_ << "): extraction/copy of Graph in " << cputime+t1-t0 << " sec\n"
<< " max-nonzeros in Graph: " << maxnnz << "\n";
}
// create this levels coarse interface
coarseinterface_ = new ML_NOX::Nox_CoarseProblem_Interface(fineinterface_,level_,ml_printlevel_,
P,&(graph_->RowMap()),nlevel_);
// restrict the xfine-vector to this level
Epetra_Vector* xthis = coarseinterface_->restrict_fine_to_this(xfine);
if (!xthis)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: ML_Epetra::Nox_CoarseProblem_Interface::restrict_fine_to_this returned NULL on level "
<< level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
Epetra_Vector* xc = new Epetra_Vector(graph_->RowMap(),false);
// FIXME: after intesive testing, this test might be obsolet
#if 0
bool samemap = xc->Map().PointSameAs(xthis->Map());
if (samemap)
{
#endif
xc->Update(1.0,*xthis,0.0);
#if 0
}
else
{
cout << "**WRN** Maps are not equal in\n"
<< "**WRN** file/line: " << __FILE__ << "/" << __LINE__ << "\n";
// import the xthis vector in the Map that ML produced for graph_
Epetra_Import* importer = new Epetra_Import(graph_->RowMap(),xthis->Map());
int ierr = xc->Import(*xthis,*importer,Insert);
if (ierr)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: export from xthis to xc returned err=" << ierr <<"\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
if (importer) delete importer; importer = 0;
}
#endif
if (xthis) delete xthis; xthis = 0;
// create the coloring of the graph
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Entering Coloring on level " << level_ << "\n";
fflush(stdout);
}
double t0 = GetClock();
colorMap_ = ML_NOX::ML_Nox_collapsedcoloring(graph_,bsize_,isDiagonalOnly,ml_printlevel_);
if (!colorMap_) colorMap_ = ML_NOX::ML_Nox_standardcoloring(graph_,isDiagonalOnly);
colorMapIndex_ = new EpetraExt::CrsGraph_MapColoringIndex(*colorMap_);
colorcolumns_ = &(*colorMapIndex_)(*graph_);
double t1 = GetClock();
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Proc " << comm_.MyPID() <<" Coloring time is " << (t1-t0) << " sec\n";
fflush(stdout);
}
// construct the FiniteDifferenceColoring-Matrix
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Entering Construction FD-Operator on level " << level_ << "\n";
fflush(stdout);
}
t0 = GetClock();
#if 1 // FD-operator with coloring
FD_ = new NOX::EpetraNew::FiniteDifferenceColoring(*coarseinterface_,
*xc,
*graph_,
*colorMap_,
*colorcolumns_,
true,
isDiagonalOnly_,
fd_beta_,fd_alpha_);
#else // FD-operator without coloring
FD_ = new NOX::EpetraNew::FiniteDifference(*coarseinterface_,
*xc,
*graph_,
fd_beta_,fd_alpha_);
#endif
// set differencing method
if (fd_centered_)
FD_->setDifferenceMethod(NOX::EpetraNew::FiniteDifferenceColoring::Centered);
bool err = FD_->computeJacobian(*xc);
if (err==false)
{
cout << "**ERR**: ML_NOX::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: NOX::Epetra::FiniteDifferenceColoring returned an error on level "
<< level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
// print number of calls to the coarse interface
if (ml_printlevel_>0 && comm_.MyPID()==0)
cout << "matrixfreeML (level " << level_ << "): Calls to coarse-computeF in FD-Operator: "
<< coarseinterface_->numcallscomputeF() << "\n";
t1 = GetClock();
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Proc " << comm_.MyPID() <<" colored Finite Differencing time is " << (t1-t0) << " sec\n";
fflush(stdout);
}
// get ref to computed Epetra_CrsMatrix
A_ = dynamic_cast<Epetra_CrsMatrix*>(&(FD_->getUnderlyingMatrix()));
// set counter for number of calls to the coarseinterface_->computeF back to zero
coarseinterface_->resetnumcallscomputeF();
// tidy up
if (xc) delete xc; xc = 0;
return;
}
/*----------------------------------------------------------------------*
| recreate this level (public) m.gee 01/05|
| this function assumes, that the graph of the fine level problem has |
| not changed since call to the constructor and therefore |
| the graph and it's coloring do not have to be recomputed |
| IMPORTANT: |
| No matter on which level we are here, the vector xfine is ALWAYS |
| a fine grid vector here! |
*----------------------------------------------------------------------*/
bool ML_NOX::ML_Nox_MatrixfreeLevel::recreateLevel(int level, int nlevel, int plevel,
ML* ml, ML_Aggregate* ag, Epetra_CrsMatrix** P,
ML_NOX::Ml_Nox_Fineinterface& interface,
const Epetra_Comm& comm, const Epetra_Vector& xfine)
{
// make some tests
if (level != level_)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::recreateLevel:\n"
<< "**ERR**: level_ " << level_ << " not equal level " << level << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
if (nlevel != nlevel_)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::recreateLevel:\n"
<< "**ERR**: nlevel_ " << nlevel_ << " not equal nlevel " << nlevel << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
// printlevel might have changed
ml_printlevel_ = plevel;
ml_ = ml;
ag_ = ag;
destroyP(); // safer to use the new Ps
setP(NULL);
// we need the graph of the operator on this level. On the fine grid we can just ask the
// fineinterface for it, on the coarser levels it has to be extracted from the ML-hierachy
bool same;
if (level_==0)
{
const Epetra_CrsGraph* graph = interface.getGraph();
// check whether the old graph matches the new one
same = compare_graphs(graph,graph_);
destroyFD(); // we are here to recompute the FD-operator (this destroys graph_)
graph_ = ML_NOX::deepcopy_graph(graph);
}
else
{
// Note that ML has no understanding of global indices, so it makes up new GIDs
// (This also holds for the Prolongators P)
Epetra_CrsMatrix* tmpMat = 0;
int maxnnz = 0;
double cputime = 0.0;
ML_Operator2EpetraCrsMatrix(&(ml_->Amat[level_]), tmpMat, maxnnz, false, cputime);
// get a view from the graph
const Epetra_CrsGraph& graph = tmpMat->Graph();
// compare the graph to the existing one
same = compare_graphs(&graph,graph_);
destroyFD(); // we are here to recompute the FD-operator (this destroys graph_)
double t0 = GetClock();
graph_ = ML_NOX::deepcopy_graph(&graph);
// delete the copy of the Epetra_CrsMatrix
if (tmpMat) delete tmpMat; tmpMat = 0;
double t1 = GetClock();
if (ml_printlevel_ > 0 && 0 == comm_.MyPID())
cout << "matrixfreeML (level " << level_ << "): extraction/copy of Graph in " << cputime+t1-t0 << " sec\n"
<< " max-nonzeros in Graph: " << maxnnz << "\n";
}
// recreate this levels coarse interface
if (same)
coarseinterface_->recreate(ml_printlevel_,P,&(graph_->RowMap()));
else
{
delete coarseinterface_;
coarseinterface_ = new ML_NOX::Nox_CoarseProblem_Interface(fineinterface_,level_,ml_printlevel_,
P,&(graph_->RowMap()),nlevel_);
}
// restrict the xfine-vector to this level
Epetra_Vector* xthis = coarseinterface_->restrict_fine_to_this(xfine);
if (!xthis)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: ML_Epetra::Nox_CoarseProblem_Interface::restrict_fine_to_this returned NULL on level "
<< level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
Epetra_Vector* xc = new Epetra_Vector(graph_->RowMap(),false);
// FIXME: after intesive testing, this test might be obsolet
#if 0
bool samemap = xc->Map().PointSameAs(xthis->Map());
if (samemap)
{
#endif
xc->Update(1.0,*xthis,0.0);
#if 0
}
else
{
cout << "**WRN** Maps are not equal in\n"
<< "**WRN** file/line: " << __FILE__ << "/" << __LINE__ << "\n";
// import the xthis vector in the Map that ML produced for graph_
Epetra_Import* importer = new Epetra_Import(graph_->RowMap(),xthis->Map());
int ierr = xc->Import(*xthis,*importer,Insert);
if (ierr)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: export from xthis to xc returned err=" << ierr <<"\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
if (importer) delete importer; importer = 0;
}
#endif
if (xthis) delete xthis; xthis = 0;
// create the coloring of the graph
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Entering Recoloring on level " << level_ << "\n";
fflush(stdout);
}
double t0 = GetClock();
if (!same) // te graph has obviously changed, so we need to recolor
{
if (colorMap_) delete colorMap_; colorMap_ = 0;
if (colorMapIndex_) delete colorMapIndex_; colorMapIndex_ = 0;
if (colorcolumns_) delete colorcolumns_; colorcolumns_ = 0;
colorMap_ = ML_NOX::ML_Nox_collapsedcoloring(graph_,bsize_,isDiagonalOnly_,ml_printlevel_);
if (!colorMap_) colorMap_ = ML_NOX::ML_Nox_standardcoloring(graph_,isDiagonalOnly_);
colorMapIndex_ = new EpetraExt::CrsGraph_MapColoringIndex(*colorMap_);
colorcolumns_ = &(*colorMapIndex_)(*graph_);
}
else if (ml_printlevel_>0 && comm_.MyPID()==0)
cout << "matrixfreeML (level " << level_ << "): Reusing existing Coloring on level " << level_ << "\n";
double t1 = GetClock();
if (ml_printlevel_>5)
{
cout << "matrixfreeML (level " << level_ << "): Proc " << comm_.MyPID() <<" (Re)Coloring time is " << (t1-t0) << " sec\n";
fflush(stdout);
}
#if 0
// print the colorMap_
if (comm_.MyPID()==0)
cout << "colorMap_\n";
cout << *colorMap_;
for (int i=0; i<colorcolumns_->size(); i++)
{
if (comm_.MyPID()==0)
cout << "the " << i << " th colorcolumn_ - vector\n";
cout << colorcolumns_->operator[](i);
}
#endif
// construct the FiniteDifferenceColoring-Matrix
if (ml_printlevel_>0 && comm_.MyPID()==0)
{
cout << "matrixfreeML (level " << level_ << "): Entering Construction FD-Operator on level " << level_ << "\n";
fflush(stdout);
}
t0 = GetClock();
#if 1 // FD-operator with coloring (see the #if 1 in ml_nox_matrixfreelevel.H as well!)
FD_ = new NOX::EpetraNew::FiniteDifferenceColoring(*coarseinterface_,
*xc,
*graph_,
*colorMap_,
*colorcolumns_,
true,
isDiagonalOnly_,
fd_beta_,fd_alpha_);
#else // FD-operator without coloring
FD_ = new NOX::EpetraNew::FiniteDifference(*coarseinterface_,
*xc,
*graph_,
fd_beta_,fd_alpha_);
#endif
// set differencing method
if (fd_centered_)
{
FD_->setDifferenceMethod(NOX::EpetraNew::FiniteDifferenceColoring::Centered);
}
bool err = FD_->computeJacobian(*xc);
if (err==false)
{
cout << "**ERR**: ML_Epetra::ML_Nox_MatrixfreeLevel::ML_Nox_MatrixfreeLevel:\n"
<< "**ERR**: NOX::Epetra::FiniteDifferenceColoring returned an error on level "
<< level_ << "\n"
<< "**ERR**: file/line: " << __FILE__ << "/" << __LINE__ << "\n"; throw -1;
}
t1 = GetClock();
if (ml_printlevel_>5)
cout << "matrixfreeML (level " << level_ << "): Proc " << comm_.MyPID()
<<" Finite Differencing operator constr. in " << (t1-t0) << " sec\n";
// get ref to computed Epetra_CrsMatrix
A_ = dynamic_cast<Epetra_CrsMatrix*>(&(FD_->getUnderlyingMatrix()));
// print number of calls to the coarse interface
if (ml_printlevel_>5 && comm_.MyPID()==0)
cout << "matrixfreeML (level " << level_ << "): Calls to coarse-computeF in FD-Operator: "
<< coarseinterface_->numcallscomputeF() << "\n";
// set counter for number of calls to the coarseinterface_->computeF back to zero
coarseinterface_->resetnumcallscomputeF();
// tidy up
if (xc) delete xc; xc = 0;
return true;
}
void BiCGSTAB(Epetra_CrsMatrix &A,
Epetra_Vector &x,
Epetra_Vector &b,
Ifpack_CrsRiluk *M,
int Maxiter,
double Tolerance,
double *residual, bool verbose) {
// Allocate vectors needed for iterations
Epetra_Vector phat(x.Map()); phat.SetFlopCounter(x);
Epetra_Vector p(x.Map()); p.SetFlopCounter(x);
Epetra_Vector shat(x.Map()); shat.SetFlopCounter(x);
Epetra_Vector s(x.Map()); s.SetFlopCounter(x);
Epetra_Vector r(x.Map()); r.SetFlopCounter(x);
Epetra_Vector rtilde(x.Map()); rtilde.Random(); rtilde.SetFlopCounter(x);
Epetra_Vector v(x.Map()); v.SetFlopCounter(x);
A.Multiply(false, x, r); // r = A*x
r.Update(1.0, b, -1.0); // r = b - A*x
double r_norm, b_norm, scaled_r_norm, rhon, rhonm1 = 1.0;
double alpha = 1.0, omega = 1.0, sigma;
double omega_num, omega_den;
r.Norm2(&r_norm);
b.Norm2(&b_norm);
scaled_r_norm = r_norm/b_norm;
r.Dot(rtilde,&rhon);
if (verbose) cout << "Initial residual = " << r_norm
<< " Scaled residual = " << scaled_r_norm << endl;
for (int i=0; i<Maxiter; i++) { // Main iteration loop
double beta = (rhon/rhonm1) * (alpha/omega);
rhonm1 = rhon;
/* p = r + beta*(p - omega*v) */
/* phat = M^-1 p */
/* v = A phat */
double dtemp = - beta*omega;
p.Update(1.0, r, dtemp, v, beta);
if (M==0)
phat.Scale(1.0, p);
else
M->Solve(false, p, phat);
A.Multiply(false, phat, v);
rtilde.Dot(v,&sigma);
alpha = rhon/sigma;
/* s = r - alpha*v */
/* shat = M^-1 s */
/* r = A shat (r is a tmp here for t ) */
s.Update(-alpha, v, 1.0, r, 0.0);
if (M==0)
shat.Scale(1.0, s);
else
M->Solve(false, s, shat);
A.Multiply(false, shat, r);
r.Dot(s, &omega_num);
r.Dot(r, &omega_den);
omega = omega_num / omega_den;
/* x = x + alpha*phat + omega*shat */
/* r = s - omega*r */
x.Update(alpha, phat, omega, shat, 1.0);
r.Update(1.0, s, -omega);
r.Norm2(&r_norm);
scaled_r_norm = r_norm/b_norm;
r.Dot(rtilde,&rhon);
if (verbose) cout << "Iter "<< i << " residual = " << r_norm
<< " Scaled residual = " << scaled_r_norm << endl;
if (scaled_r_norm < Tolerance) break;
}
return;
}
int TestOneMatrix( std::string HBname, std::string MMname, std::string TRIname, Epetra_Comm &Comm, bool verbose ) {
if ( Comm.MyPID() != 0 ) verbose = false ;
Epetra_Map * readMap = 0;
Epetra_CrsMatrix * HbA = 0;
Epetra_Vector * Hbx = 0;
Epetra_Vector * Hbb = 0;
Epetra_Vector * Hbxexact = 0;
Epetra_CrsMatrix * TriplesA = 0;
Epetra_Vector * Triplesx = 0;
Epetra_Vector * Triplesb = 0;
Epetra_Vector * Triplesxexact = 0;
Epetra_CrsMatrix * MatrixMarketA = 0;
Epetra_Vector * MatrixMarketx = 0;
Epetra_Vector * MatrixMarketb = 0;
Epetra_Vector * MatrixMarketxexact = 0;
int TRI_Size = TRIname.size() ;
std::string LastFiveBytes = TRIname.substr( EPETRA_MAX(0,TRI_Size-5), TRI_Size );
if ( LastFiveBytes == ".TimD" ) {
// Call routine to read in a file with a Tim Davis header and zero-based indexing
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], false, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, true, true ) );
delete readMap;
} else {
if ( LastFiveBytes == ".triU" ) {
// Call routine to read in unsymmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], false, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, false ) );
delete readMap;
} else {
if ( LastFiveBytes == ".triS" ) {
// Call routine to read in symmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], true, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, false ) );
delete readMap;
} else {
assert( false ) ;
}
}
}
EPETRA_CHK_ERR( Trilinos_Util_ReadMatrixMarket2Epetra64( &MMname[0], Comm, readMap,
MatrixMarketA, MatrixMarketx,
MatrixMarketb, MatrixMarketxexact) );
delete readMap;
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra64( &HBname[0], Comm, readMap, HbA, Hbx,
Hbb, Hbxexact) ;
#if 0
std::cout << " HbA " ;
HbA->Print( std::cout ) ;
std::cout << std::endl ;
std::cout << " MatrixMarketA " ;
MatrixMarketA->Print( std::cout ) ;
std::cout << std::endl ;
std::cout << " TriplesA " ;
TriplesA->Print( std::cout ) ;
std::cout << std::endl ;
#endif
int TripleErr = 0 ;
int MMerr = 0 ;
for ( int i = 0 ; i < 10 ; i++ )
{
double resid_Hb_Triples;
double resid_Hb_Matrix_Market;
double norm_A ;
Hbx->Random();
//
// Set the output vectors to different values:
//
Triplesb->PutScalar(1.1);
Hbb->PutScalar(1.2);
MatrixMarketb->PutScalar(1.3);
HbA->Multiply( false, *Hbx, *Hbb );
norm_A = HbA->NormOne( ) ;
TriplesA->Multiply( false, *Hbx, *Triplesb );
Triplesb->Update( 1.0, *Hbb, -1.0 ) ;
MatrixMarketA->Multiply( false, *Hbx, *MatrixMarketb );
MatrixMarketb->Update( 1.0, *Hbb, -1.0 ) ;
Triplesb->Norm1( &resid_Hb_Triples ) ;
MatrixMarketb->Norm1( &resid_Hb_Matrix_Market ) ;
TripleErr += ( resid_Hb_Triples > 1e-11 * norm_A ) ;
MMerr += ( resid_Hb_Matrix_Market > 1e-11 * norm_A ) ;
if ( verbose && resid_Hb_Triples > 1e-11 * norm_A )
std::cout << " resid_Hb_Triples = " << resid_Hb_Triples
<< " norm_A = " << norm_A << std::endl ;
if ( verbose && resid_Hb_Matrix_Market > 1e-11 * norm_A )
std::cout << " resid_Hb_Matrix_Market = " << resid_Hb_Matrix_Market
<< " norm_A = " << norm_A << std::endl ;
}
if ( verbose ) {
if ( TripleErr ) std::cout << " Error in reading " << HBname << " or " << TRIname << std::endl ;
if ( MMerr ) std::cout << " Error in reading " << HBname << " or " << MMname << std::endl ;
}
delete HbA;
delete Hbx;
delete Hbb;
delete Hbxexact;
delete TriplesA;
delete Triplesx;
delete Triplesb;
delete Triplesxexact;
delete MatrixMarketA;
delete MatrixMarketx;
delete MatrixMarketb;
delete MatrixMarketxexact;
delete readMap;
return TripleErr+MMerr ;
}
//! Update vector
static void update(Epetra_Vector& vec, double a, const Epetra_Vector& x) {
vec.Update(a,x,1.0);
}
// ================================================ ====== ==== ==== == =
// the tentative null space is in input because the user
// has to remember to allocate and fill it, and then to delete
// it after calling this method.
int ML_Epetra::MultiLevelPreconditioner::
ComputeAdaptivePreconditioner(int TentativeNullSpaceSize,
double* TentativeNullSpace)
{
if ((TentativeNullSpaceSize == 0) || (TentativeNullSpace == 0))
ML_CHK_ERR(-1);
// ================================== //
// get parameters from the input list //
// ================================== //
// maximum number of relaxation sweeps
int MaxSweeps = List_.get("adaptive: max sweeps", 10);
// number of std::vector to be added to the tentative null space
int NumAdaptiveVectors = List_.get("adaptive: num vectors", 1);
if (verbose_) {
std::cout << PrintMsg_ << "*** Adaptive Smoother Aggregation setup ***" << std::endl;
std::cout << PrintMsg_ << " Maximum relaxation sweeps = " << MaxSweeps << std::endl;
std::cout << PrintMsg_ << " Additional vectors to compute = " << NumAdaptiveVectors << std::endl;
}
// ==================================================== //
// compute the preconditioner, set null space from user //
// (who will have to delete std::vector TentativeNullSpace) //
// ==================================================== //
double* NewNullSpace = 0;
double* OldNullSpace = TentativeNullSpace;
int OldNullSpaceSize = TentativeNullSpaceSize;
// need some work otherwise matvec() with Epetra_Vbr fails.
// Also, don't differentiate between range and domain here
// as ML will not work if range != domain
const Epetra_VbrMatrix* VbrA = NULL;
VbrA = dynamic_cast<const Epetra_VbrMatrix*>(RowMatrix_);
Epetra_Vector* LHS = 0;
Epetra_Vector* RHS = 0;
if (VbrA != 0) {
LHS = new Epetra_Vector(VbrA->DomainMap());
RHS = new Epetra_Vector(VbrA->DomainMap());
} else {
LHS = new Epetra_Vector(RowMatrix_->OperatorDomainMap());
RHS = new Epetra_Vector(RowMatrix_->OperatorDomainMap());
}
// destroy what we may already have
if (IsComputePreconditionerOK_ == true) {
DestroyPreconditioner();
}
// build the preconditioner for the first time
List_.set("null space: type", "pre-computed");
List_.set("null space: dimension", OldNullSpaceSize);
List_.set("null space: vectors", OldNullSpace);
ComputePreconditioner();
// ====================== //
// add one std::vector at time //
// ====================== //
for (int istep = 0 ; istep < NumAdaptiveVectors ; ++istep) {
if (verbose_) {
std::cout << PrintMsg_ << "\tAdaptation step " << istep << std::endl;
std::cout << PrintMsg_ << "\t---------------" << std::endl;
}
// ==================== //
// look for "bad" modes //
// ==================== //
// note: should an error occur, ML_CHK_ERR will return,
// and LHS and RHS will *not* be delete'd (--> memory leak).
// Anyway, this means that something wrong happened in the code
// and should be fixed by the user.
LHS->Random();
double Norm2;
for (int i = 0 ; i < MaxSweeps ; ++i) {
// RHS = (I - ML^{-1} A) LHS
ML_CHK_ERR(RowMatrix_->Multiply(false,*LHS,*RHS));
// FIXME: can do something slightly better here
ML_CHK_ERR(ApplyInverse(*RHS,*RHS));
ML_CHK_ERR(LHS->Update(-1.0,*RHS,1.0));
LHS->Norm2(&Norm2);
if (verbose_) {
std::cout << PrintMsg_ << "\titer " << i << ", ||x||_2 = ";
std::cout << Norm2 << std::endl;
}
}
// scaling vectors
double NormInf;
LHS->NormInf(&NormInf);
LHS->Scale(1.0 / NormInf);
// ========================================================= //
// copy tentative and computed null space into NewNullSpace, //
// which now becomes the standard null space //
// ========================================================= //
int NewNullSpaceSize = OldNullSpaceSize + 1;
NewNullSpace = new double[NumMyRows() * NewNullSpaceSize];
assert (NewNullSpace != 0);
int itmp = OldNullSpaceSize * NumMyRows();
for (int i = 0 ; i < itmp ; ++i) {
NewNullSpace[i] = OldNullSpace[i];
}
for (int j = 0 ; j < NumMyRows() ; ++j) {
NewNullSpace[itmp + j] = (*LHS)[j];
}
// =============== //
// visualize modes //
// =============== //
if (List_.get("adaptive: visualize", false)) {
double* x_coord = List_.get("viz: x-coordinates", (double*)0);
double* y_coord = List_.get("viz: y-coordinates", (double*)0);
double* z_coord = List_.get("viz: z-coordinates", (double*)0);
assert (x_coord != 0);
std::vector<double> plot_me(NumMyRows()/NumPDEEqns_);
ML_Aggregate_Viz_Stats info;
info.Amatrix = &(ml_->Amat[LevelID_[0]]);
info.x = x_coord;
info.y = y_coord;
info.z = z_coord;
info.Nlocal = NumMyRows() / NumPDEEqns_;
info.Naggregates = 1;
ML_Operator_AmalgamateAndDropWeak(&(ml_->Amat[LevelID_[0]]),
NumPDEEqns_, 0.0);
for (int ieqn = 0 ; ieqn < NumPDEEqns_ ; ++ieqn) {
for (int j = 0 ; j < NumMyRows() ; j+=NumPDEEqns_) {
plot_me[j / NumPDEEqns_] = (*LHS)[j + ieqn];
}
char FileName[80];
sprintf(FileName,"nullspace-mode%d-eq%d.xyz", istep, ieqn);
if (verbose_)
std::cout << PrintMsg_ << "writing file " << FileName << "..." << std::endl;
ML_Aggregate_VisualizeXYZ(info,FileName,
ml_->comm,&plot_me[0]);
}
ML_Operator_UnAmalgamateAndDropWeak(&(ml_->Amat[LevelID_[0]]),
NumPDEEqns_, 0.0);
}
// Destroy the old preconditioner
DestroyPreconditioner();
// ==================================================== //
// build the new preconditioner with the new null space //
// ==================================================== //
List_.set("null space: type", "pre-computed");
List_.set("null space: dimension", NewNullSpaceSize);
List_.set("null space: vectors", NewNullSpace);
ML_CHK_ERR(ComputePreconditioner());
if (istep && (istep != NumAdaptiveVectors))
delete OldNullSpace;
OldNullSpace = NewNullSpace;
OldNullSpaceSize = NewNullSpaceSize;
}
// keep trace of this pointer, it will be delete'd later
NullSpaceToFree_ = NewNullSpace;
delete LHS;
delete RHS;
return(0);
}
void
exampleRoutine (const Epetra_Comm& comm,
std::ostream& out)
{
using std::endl;
// Print out the Epetra software version.
if (comm.MyPID () == 0) {
out << Epetra_Version () << endl << endl;
}
// The type of global indices. You could just set this to int,
// but we want the example to work for Epetra64 as well.
#ifdef EPETRA_NO_32BIT_GLOBAL_INDICES
// Epetra was compiled only with 64-bit global index support, so use
// 64-bit global indices.
typedef long long global_ordinal_type;
#else
// Epetra was compiled with 32-bit global index support. If
// EPETRA_NO_64BIT_GLOBAL_INDICES is defined, it does not also
// support 64-bit indices.
typedef int global_ordinal_type;
#endif // EPETRA_NO_32BIT_GLOBAL_INDICES
//////////////////////////////////////////////////////////////////////
// Create some Epetra_Map objects
//////////////////////////////////////////////////////////////////////
//
// Epetra has local and global Maps. Local maps describe objects
// that are replicated over all participating MPI processes. Global
// maps describe distributed objects. You can do imports and
// exports between local and global maps; this is how you would turn
// locally replicated objects into distributed objects and vice
// versa.
//
// The total (global, i.e., over all MPI processes) number of
// entries in the Map. This has the same type as that of global
// indices, so it can represent very large values if Epetra was
// built with 64-bit global index support.
//
// For this example, we scale the global number of entries in the
// Map with the number of MPI processes. That way, you can run this
// example with any number of MPI processes and every process will
// still have a positive number of entries.
const global_ordinal_type numGlobalEntries = comm.NumProc () * 5;
// Tpetra can index the entries of a Map starting with 0 (C style),
// 1 (Fortran style), or any base you want. 1-based indexing is
// handy when interfacing with Fortran. We choose 0-based indexing
// here. This also has the same type as that of global indices.
const global_ordinal_type indexBase = 0;
// Construct a Map that puts the same number of equations on each
// (MPI) process. The Epetra_Comm is passed in by value, but that's
// OK, because Epetra_Comm has shallow copy semantics. (Its copy
// constructor and assignment operator do not call MPI_Comm_dup;
// they just pass along the MPI_Comm.)
Epetra_Map contigMap (numGlobalEntries, indexBase, comm);
// contigMap is contiguous by construction.
if (! contigMap.LinearMap ()) {
throw std::logic_error ("The supposedly contiguous Map isn't contiguous.");
}
// Let's create a second Map. It will have the same number of
// global entries per process, but will distribute them differently,
// in round-robin (1-D cyclic) fashion instead of contiguously.
// We'll use the version of the Map constructor that takes, on each
// MPI process, a list of the global indices in the Map belonging to
// that process. You can use this constructor to construct an
// overlapping (also called "not 1-to-1") Map, in which one or more
// entries are owned by multiple processes. We don't do that here;
// we make a nonoverlapping (also called "1-to-1") Map.
const int numGblIndsPerProc = 5;
global_ordinal_type* gblIndList = new global_ordinal_type [numGblIndsPerProc];
const int numProcs = comm.NumProc ();
const int myRank = comm.MyPID ();
for (int k = 0; k < numGblIndsPerProc; ++k) {
gblIndList[k] = myRank + k*numProcs;
}
Epetra_Map cyclicMap (numGlobalEntries, numGblIndsPerProc,
gblIndList, indexBase, comm);
// The above constructor makes a deep copy of the input index list,
// so it's safe to deallocate that list after this constructor
// completes.
if (gblIndList != NULL) {
delete [] gblIndList;
gblIndList = NULL;
}
// If there's more than one MPI process in the communicator,
// then cyclicMap is definitely NOT contiguous.
if (comm.NumProc () > 1 && cyclicMap.LinearMap ()) {
throw std::logic_error ("The cyclic Map claims to be contiguous.");
}
// contigMap and cyclicMap should always be compatible. However, if
// the communicator contains more than 1 process, then contigMap and
// cyclicMap are NOT the same.
// if (! contigMap.isCompatible (*cyclicMap)) {
// throw std::logic_error ("contigMap should be compatible with cyclicMap, "
// "but it's not.");
// }
if (comm.NumProc () > 1 && contigMap.SameAs (cyclicMap)) {
throw std::logic_error ("contigMap should not be the same as cyclicMap.");
}
//////////////////////////////////////////////////////////////////////
// We have maps now, so we can create vectors.
//////////////////////////////////////////////////////////////////////
// Create an Epetra_Vector with the contiguous Map we created above.
// This version of the constructor will fill the vector with zeros.
// The Vector constructor takes a Map by value, but that's OK,
// because Epetra_Map has shallow copy semantics. It uses reference
// counting internally to avoid copying data unnecessarily.
Epetra_Vector x (contigMap);
// The copy constructor performs a deep copy.
// x and y have the same Map.
Epetra_Vector y (x);
// Create a Vector with the 1-D cyclic Map. Calling the constructor
// with false for the second argument leaves the data uninitialized,
// so that you can fill it later without paying the cost of
// initially filling it with zeros.
Epetra_Vector z (cyclicMap, false);
// Set the entries of z to (pseudo)random numbers. Please don't
// consider this a good parallel pseudorandom number generator.
(void) z.Random ();
// Set the entries of x to all ones.
(void) x.PutScalar (1.0);
// Define some constants for use below.
const double alpha = 3.14159;
const double beta = 2.71828;
const double gamma = -10.0;
// x = beta*x + alpha*z
//
// This is a legal operation! Even though the Maps of x and z are
// not the same, their Maps are compatible. Whether it makes sense
// or not depends on your application.
(void) x.Update (alpha, z, beta);
(void) y.PutScalar (42.0); // Set all entries of y to 42.0
// y = gamma*y + alpha*x + beta*z
y.Update (alpha, x, beta, z, gamma);
// Compute the 2-norm of y.
//
// The norm may have a different type than scalar_type.
// For example, if scalar_type is complex, then the norm is real.
// The ScalarTraits "traits class" gives us the type of the norm.
double theNorm = 0.0;
(void) y.Norm2 (&theNorm);
// Print the norm of y on Proc 0.
out << "Norm of y: " << theNorm << endl;
}
int Amesos_TestSolver( Epetra_Comm &Comm, char *matrix_file,
SparseSolverType SparseSolver,
bool transpose,
int special, AMESOS_MatrixType matrix_type ) {
Epetra_Map * readMap;
Epetra_CrsMatrix * readA;
Epetra_Vector * readx;
Epetra_Vector * readb;
Epetra_Vector * readxexact;
std::string FileName = matrix_file ;
int FN_Size = FileName.size() ;
std::string LastFiveBytes = FileName.substr( EPETRA_MAX(0,FN_Size-5), FN_Size );
std::string LastFourBytes = FileName.substr( EPETRA_MAX(0,FN_Size-4), FN_Size );
bool NonContiguousMap = false;
if ( LastFiveBytes == ".triU" ) {
// Call routine to read in unsymmetric Triplet matrix
NonContiguousMap = true;
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra( matrix_file, false, Comm, readMap, readA, readx,
readb, readxexact, NonContiguousMap ) );
} else {
if ( LastFiveBytes == ".triS" ) {
NonContiguousMap = true;
// Call routine to read in symmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra( matrix_file, true, Comm, readMap, readA, readx,
readb, readxexact, NonContiguousMap ) );
} else {
if ( LastFourBytes == ".mtx" ) {
EPETRA_CHK_ERR( Trilinos_Util_ReadMatrixMarket2Epetra( matrix_file, Comm, readMap,
readA, readx, readb, readxexact) );
} else {
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra( matrix_file, Comm, readMap, readA, readx,
readb, readxexact) ;
}
}
}
Epetra_CrsMatrix transposeA(Copy, *readMap, 0);
Epetra_CrsMatrix *serialA ;
if ( transpose ) {
assert( CrsMatrixTranspose( readA, &transposeA ) == 0 );
serialA = &transposeA ;
} else {
serialA = readA ;
}
Epetra_RowMatrix * passA = 0;
Epetra_Vector * passx = 0;
Epetra_Vector * passb = 0;
Epetra_Vector * passxexact = 0;
Epetra_Vector * passresid = 0;
Epetra_Vector * passtmp = 0;
// Create uniform distributed map
Epetra_Map map(readMap->NumGlobalElements(), 0, Comm);
Epetra_Map* map_;
if( NonContiguousMap ) {
//
// map gives us NumMyElements and MyFirstElement;
//
int NumGlobalElements = readMap->NumGlobalElements();
int NumMyElements = map.NumMyElements();
int MyFirstElement = map.MinMyGID();
std::vector<int> MapMap_( NumGlobalElements );
readMap->MyGlobalElements( &MapMap_[0] ) ;
Comm.Broadcast( &MapMap_[0], NumGlobalElements, 0 ) ;
map_ = new Epetra_Map( NumGlobalElements, NumMyElements, &MapMap_[MyFirstElement], 0, Comm);
} else {
map_ = new Epetra_Map( map ) ;
}
Epetra_CrsMatrix A(Copy, *map_, 0);
const Epetra_Map &OriginalMap = serialA->RowMatrixRowMap() ;
assert( OriginalMap.SameAs(*readMap) );
Epetra_Export exporter(OriginalMap, *map_);
Epetra_Export exporter2(OriginalMap, *map_);
Epetra_Export MatrixExporter(OriginalMap, *map_);
Epetra_CrsMatrix AwithDiag(Copy, *map_, 0);
Epetra_Vector x(*map_);
Epetra_Vector b(*map_);
Epetra_Vector xexact(*map_);
Epetra_Vector resid(*map_);
Epetra_Vector readresid(*readMap);
Epetra_Vector tmp(*map_);
Epetra_Vector readtmp(*readMap);
// Epetra_Vector xcomp(*map_); // X as computed by the solver
bool distribute_matrix = ( matrix_type == AMESOS_Distributed ) ;
if ( distribute_matrix ) {
// Create Exporter to distribute read-in matrix and vectors
//
// Initialize x, b and xexact to the values read in from the file
//
x.Export(*readx, exporter, Add);
b.Export(*readb, exporter, Add);
xexact.Export(*readxexact, exporter, Add);
Comm.Barrier();
A.Export(*serialA, exporter, Add);
assert(A.FillComplete()==0);
Comm.Barrier();
passA = &A;
passx = &x;
passb = &b;
passxexact = &xexact;
passresid = &resid;
passtmp = &tmp;
} else {
passA = serialA;
passx = readx;
passb = readb;
passxexact = readxexact;
passresid = &readresid;
passtmp = &readtmp;
}
Epetra_MultiVector CopyB( *passb ) ;
double Anorm = passA->NormInf() ;
SparseDirectTimingVars::SS_Result.Set_Anorm(Anorm) ;
Epetra_LinearProblem Problem( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb );
for ( int i = 0; i < 1+special ; i++ ) {
Epetra_Time TotalTime( Comm ) ;
if ( false ) {
// TEST_UMFPACK is never set by configure
#ifdef HAVE_AMESOS_SUPERLUDIST
} else if ( SparseSolver == SUPERLUDIST ) {
Teuchos::ParameterList ParamList ;
ParamList.set( "MaxProcs", -3 );
Amesos_Superludist A_Superludist( Problem ) ;
//ParamList.set( "Redistribute", true );
//ParamList.set( "AddZeroToDiag", true );
Teuchos::ParameterList& SuperludistParams = ParamList.sublist("Superludist") ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_Superludist.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_Superludist.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_Superludist.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_Superludist.NumericFactorization( ) );
EPETRA_CHK_ERR( A_Superludist.Solve( ) );
#endif
#ifdef HAVE_AMESOS_DSCPACK
} else if ( SparseSolver == DSCPACK ) {
Teuchos::ParameterList ParamList ;
ParamList.set( "MaxProcs", -3 );
Amesos_Dscpack A_dscpack( Problem ) ;
EPETRA_CHK_ERR( A_dscpack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_dscpack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_dscpack.NumericFactorization( ) );
EPETRA_CHK_ERR( A_dscpack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SCALAPACK
} else if ( SparseSolver == SCALAPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Scalapack A_scalapack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_scalapack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_scalapack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_scalapack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_scalapack.NumericFactorization( ) );
EPETRA_CHK_ERR( A_scalapack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_TAUCS
} else if ( SparseSolver == TAUCS ) {
Teuchos::ParameterList ParamList ;
Amesos_Taucs A_taucs( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_taucs.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_taucs.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_taucs.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_taucs.NumericFactorization( ) );
EPETRA_CHK_ERR( A_taucs.Solve( ) );
#endif
#ifdef HAVE_AMESOS_PARDISO
} else if ( SparseSolver == PARDISO ) {
Teuchos::ParameterList ParamList ;
Amesos_Pardiso A_pardiso( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_pardiso.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_pardiso.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_pardiso.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_pardiso.NumericFactorization( ) );
EPETRA_CHK_ERR( A_pardiso.Solve( ) );
#endif
#ifdef HAVE_AMESOS_PARAKLETE
} else if ( SparseSolver == PARAKLETE ) {
Teuchos::ParameterList ParamList ;
Amesos_Paraklete A_paraklete( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_paraklete.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_paraklete.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_paraklete.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_paraklete.NumericFactorization( ) );
EPETRA_CHK_ERR( A_paraklete.Solve( ) );
#endif
#ifdef HAVE_AMESOS_MUMPS
} else if ( SparseSolver == MUMPS ) {
Teuchos::ParameterList ParamList ;
Amesos_Mumps A_mumps( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_mumps.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_mumps.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_mumps.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_mumps.NumericFactorization( ) );
EPETRA_CHK_ERR( A_mumps.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SUPERLU
} else if ( SparseSolver == SUPERLU ) {
Teuchos::ParameterList ParamList ;
Amesos_Superlu A_superlu( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_superlu.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_superlu.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_superlu.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_superlu.NumericFactorization( ) );
EPETRA_CHK_ERR( A_superlu.Solve( ) );
#endif
#ifdef HAVE_AMESOS_LAPACK
} else if ( SparseSolver == LAPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Lapack A_lapack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_lapack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_lapack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_lapack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_lapack.NumericFactorization( ) );
EPETRA_CHK_ERR( A_lapack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_UMFPACK
} else if ( SparseSolver == UMFPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Umfpack A_umfpack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( A_umfpack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( A_umfpack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( A_umfpack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( A_umfpack.NumericFactorization( ) );
EPETRA_CHK_ERR( A_umfpack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_KLU
} else if ( SparseSolver == KLU ) {
using namespace Teuchos;
Amesos_Time AT;
int setupTimePtr = -1, symTimePtr = -1, numTimePtr = -1, refacTimePtr = -1, solveTimePtr = -1;
AT.CreateTimer(Comm, 2);
AT.ResetTimer(0);
Teuchos::ParameterList ParamList ;
// ParamList.set("OutputLevel",2);
Amesos_Klu A_klu( Problem );
ParamList.set( "MaxProcs", -3 );
ParamList.set( "TrustMe", false );
// ParamList.set( "Refactorize", true );
EPETRA_CHK_ERR( A_klu.SetParameters( ParamList ) ) ;
EPETRA_CHK_ERR( A_klu.SetUseTranspose( transpose ) );
setupTimePtr = AT.AddTime("Setup", setupTimePtr, 0);
EPETRA_CHK_ERR( A_klu.SymbolicFactorization( ) );
symTimePtr = AT.AddTime("Symbolic", symTimePtr, 0);
EPETRA_CHK_ERR( A_klu.NumericFactorization( ) );
numTimePtr = AT.AddTime("Numeric", numTimePtr, 0);
EPETRA_CHK_ERR( A_klu.NumericFactorization( ) );
refacTimePtr = AT.AddTime("Refactor", refacTimePtr, 0);
// for ( int i=0; i<100000 ; i++ )
EPETRA_CHK_ERR( A_klu.Solve( ) );
solveTimePtr = AT.AddTime("Solve", solveTimePtr, 0);
double SetupTime = AT.GetTime(setupTimePtr);
double SymbolicTime = AT.GetTime(symTimePtr);
double NumericTime = AT.GetTime(numTimePtr);
double RefactorTime = AT.GetTime(refacTimePtr);
double SolveTime = AT.GetTime(solveTimePtr);
std::cout << __FILE__ << "::" << __LINE__ << " SetupTime = " << SetupTime << std::endl ;
std::cout << __FILE__ << "::" << __LINE__ << " SymbolicTime = " << SymbolicTime - SetupTime << std::endl ;
std::cout << __FILE__ << "::" << __LINE__ << " NumericTime = " << NumericTime - SymbolicTime<< std::endl ;
std::cout << __FILE__ << "::" << __LINE__ << " RefactorTime = " << RefactorTime - NumericTime << std::endl ;
std::cout << __FILE__ << "::" << __LINE__ << " SolveTime = " << SolveTime - RefactorTime << std::endl ;
#endif
} else {
SparseDirectTimingVars::log_file << "Solver not implemented yet" << std::endl ;
std::cerr << "\n\n#################### Requested solver not available on this platform ##################### ATS\n" << std::endl ;
std::cout << " SparseSolver = " << SparseSolver << std::endl ;
std::cerr << " SparseSolver = " << SparseSolver << std::endl ;
}
SparseDirectTimingVars::SS_Result.Set_Total_Time( TotalTime.ElapsedTime() );
} // end for (int i=0; i<special; i++ )
//
// Compute the error = norm(xcomp - xexact )
//
double error;
passresid->Update(1.0, *passx, -1.0, *passxexact, 0.0);
passresid->Norm2(&error);
SparseDirectTimingVars::SS_Result.Set_Error(error) ;
// passxexact->Norm2(&error ) ;
// passx->Norm2(&error ) ;
//
// Compute the residual = norm(Ax - b)
//
double residual ;
passA->Multiply( transpose, *passx, *passtmp);
passresid->Update(1.0, *passtmp, -1.0, *passb, 0.0);
// passresid->Update(1.0, *passtmp, -1.0, CopyB, 0.0);
passresid->Norm2(&residual);
SparseDirectTimingVars::SS_Result.Set_Residual(residual) ;
double bnorm;
passb->Norm2( &bnorm ) ;
SparseDirectTimingVars::SS_Result.Set_Bnorm(bnorm) ;
double xnorm;
passx->Norm2( &xnorm ) ;
SparseDirectTimingVars::SS_Result.Set_Xnorm(xnorm) ;
delete readA;
delete readx;
delete readb;
delete readxexact;
delete readMap;
delete map_;
Comm.Barrier();
return 0;
}