void PrecPrecomputed::applyTempl(
const Tpetra::MultiVector<DomainScalar, Ordinal, Ordinal, Node>& X,
Tpetra::MultiVector<RangeScalar, Ordinal, Ordinal, Node>& Z,
Teuchos::ETransp mode,
RangeScalar alpha,
RangeScalar beta) const
{
using Teuchos::RCP;
using Teuchos::rcp;
typedef Tpetra::MultiVector<DomainScalar, Ordinal, Ordinal, Node> MV;
Teuchos::Time timer ("ILUT::apply");
{ // Timer scope for timing apply()
Teuchos::TimeMonitor timeMon (timer, true);
TEUCHOS_TEST_FOR_EXCEPTION(
! isComputed (), std::runtime_error,
"Ifpack2::ILUT::apply: You must call compute() to compute the incomplete "
"factorization, before calling apply().");
TEUCHOS_TEST_FOR_EXCEPTION(
X.getNumVectors() != Z.getNumVectors(), std::runtime_error,
"Ifpack2::ILUT::apply: X and Y must have the same number of columns. "
"X has " << X.getNumVectors () << " columns, but Y has "
<< Z.getNumVectors () << " columns.");
TEUCHOS_TEST_FOR_EXCEPTION(
beta != STS::zero (), std::logic_error,
"Ifpack2::ILUT::apply: This method does not currently work when beta != 0.");
// If X and Y are pointing to the same memory location,
// we need to create an auxiliary vector, Xcopy
RCP<const MV> Xcopy;
if (X.getLocalMV ().getValues () == Z.getLocalMV ().getValues ()) {
Xcopy = rcp (new MV (X));
}
else {
Xcopy = rcpFromRef (X);
//ZACH always taken
}
//if (mode == Teuchos::NO_TRANS) { // Solve L U Y = X
/*
L_->template localSolve<RangeScalar,DomainScalar> (*Xcopy, Y, Teuchos::NO_TRANS);
U_->template localSolve<RangeScalar,RangeScalar> (Y, Y, Teuchos::NO_TRANS);
*/
//dump(X, "##X");
//dump(*Xcopy, "##Xcopy");
//ZACH the problem is here: This actually needs to be a global triangular solve
L_->template localSolve<RangeScalar,DomainScalar> (*Xcopy, *Y_, Teuchos::NO_TRANS);
//dump(*Y_, "##Y");
U_->template localSolve<RangeScalar,RangeScalar> (*Y_, Z, Teuchos::NO_TRANS);
//U_->template localSolve<RangeScalar,RangeScalar> (*Xcopy, Z, Teuchos::NO_TRANS);
//dump(Z, "##Z");
//U_->template localSolve<RangeScalar,DomainScalar> (*Xcopy, Y, Teuchos::NO_TRANS);
//U_->template localSolve<RangeScalar,RangeScalar> (Y, Y, Teuchos::TRANS);
//}
//else { // Solve U^* L^* Y = X
/*
U_->template localSolve<RangeScalar,DomainScalar> (*Xcopy, Y, mode);
L_->template localSolve<RangeScalar,RangeScalar> (Y, Y, mode);
*/
//}
if (alpha != STS::one ()) {
Z.scale (alpha);
}
}
++NumApply_;
ApplyTime_ += timer.totalElapsedTime ();
}
void
SupportGraph<MatrixType>::
apply (const Tpetra::MultiVector<scalar_type,
local_ordinal_type,
global_ordinal_type,
node_type>& X,
Tpetra::MultiVector<scalar_type,
local_ordinal_type,
global_ordinal_type,
node_type>& Y,
Teuchos::ETransp mode,
scalar_type alpha,
scalar_type beta) const
{
using Teuchos::FancyOStream;
using Teuchos::getFancyOStream;
using Teuchos::RCP;
using Teuchos::rcp;
using Teuchos::rcpFromRef;
using Teuchos::Time;
using Teuchos::TimeMonitor;
typedef scalar_type DomainScalar;
typedef scalar_type RangeScalar;
typedef Tpetra::MultiVector<DomainScalar, local_ordinal_type,
global_ordinal_type, node_type> MV;
RCP<FancyOStream> out = getFancyOStream(rcpFromRef(std::cout));
// Create a timer for this method, if it doesn't exist already.
// TimeMonitor::getNewCounter registers the timer, so that
// TimeMonitor's class methods like summarize() will report the
// total time spent in successful calls to this method.
const std::string timerName ("Ifpack2::SupportGraph::apply");
RCP<Time> timer = TimeMonitor::lookupCounter(timerName);
if (timer.is_null()) {
timer = TimeMonitor::getNewCounter(timerName);
}
{ // Start timing here.
Teuchos::TimeMonitor timeMon (*timer);
TEUCHOS_TEST_FOR_EXCEPTION(
! isComputed(), std::runtime_error,
"Ifpack2::SupportGraph::apply: You must call compute() to compute the "
"incomplete factorization, before calling apply().");
TEUCHOS_TEST_FOR_EXCEPTION(
X.getNumVectors() != Y.getNumVectors(), std::runtime_error,
"Ifpack2::SupportGraph::apply: X and Y must have the same number of "
"columns. X has " << X.getNumVectors() << " columns, but Y has "
<< Y.getNumVectors() << " columns.");
TEUCHOS_TEST_FOR_EXCEPTION(
beta != STS::zero(), std::logic_error,
"Ifpack2::SupportGraph::apply: This method does not currently work when "
"beta != 0.");
// If X and Y are pointing to the same memory location,
// we need to create an auxiliary vector, Xcopy
RCP<const MV> Xcopy;
if (X.getLocalMV().getValues() == Y.getLocalMV().getValues()) {
Xcopy = rcp (new MV(X));
}
else {
Xcopy = rcpFromRef(X);
}
if (alpha != STS::one()) {
Y.scale(alpha);
}
RCP<MV> Ycopy = rcpFromRef(Y);
solver_->setB(Xcopy);
solver_->setX(Ycopy);
solver_->solve ();
} // Stop timing here.
++NumApply_;
// timer->totalElapsedTime() returns the total time over all timer
// calls. Thus, we use = instead of +=.
ApplyTime_ = timer->totalElapsedTime();
}