void IfpackSmoother::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const {
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::IfpackSmoother::Apply(): Setup() has not been called");
// Forward the InitialGuessIsZero option to Ifpack
Teuchos::ParameterList paramList;
if (type_ == "Chebyshev")
paramList.set("chebyshev: zero starting solution", InitialGuessIsZero);
else if (type_ == "point relaxation stand-alone")
paramList.set("relaxation: zero starting solution", InitialGuessIsZero);
SetPrecParameters(paramList);
// Apply
if (InitialGuessIsZero) {
Epetra_MultiVector& epX = Utils::MV2NonConstEpetraMV(X);
const Epetra_MultiVector& epB = Utils::MV2EpetraMV(B);
prec_->ApplyInverse(epB, epX);
} else {
RCP<MultiVector> Residual = Utils::Residual(*A_,X,B);
RCP<MultiVector> Correction = MultiVectorFactory::Build(A_->getDomainMap(), X.getNumVectors());
Epetra_MultiVector& epX = Utils::MV2NonConstEpetraMV(*Correction);
const Epetra_MultiVector& epB = Utils::MV2EpetraMV(*Residual);
prec_->ApplyInverse(epB, epX);
X.update(1.0, *Correction, 1.0);
}
}
void ZoltanInterface<LocalOrdinal, GlobalOrdinal, Node, LocalMatOps>::
GetProblemGeometry(void *data, int numGIDEntries, int numLIDEntries, int numObjectIDs,
ZOLTAN_ID_PTR gids, ZOLTAN_ID_PTR lids, int dim, double *coordinates, int *ierr)
{
if (data == NULL) {
*ierr = ZOLTAN_FATAL;
return;
}
MultiVector *Coords = (MultiVector*) data;
if (dim != Teuchos::as<int>(Coords->getNumVectors())) {
//FIXME I'm assuming dim should be 1, 2, or 3 coming in?!
*ierr = ZOLTAN_FATAL;
return;
}
TEUCHOS_TEST_FOR_EXCEPTION(numObjectIDs != Teuchos::as<int>(Coords->getLocalLength()), Exceptions::Incompatible, "Length of coordinates must be the same as the number of objects");
ArrayRCP<ArrayRCP<const SC> > CoordsData(dim);
for (int j = 0; j < dim; ++j)
CoordsData[j] = Coords->getData(j);
size_t numElements = Coords->getLocalLength();
for (size_t i = 0; i < numElements; ++i)
for (int j = 0; j < dim; ++j)
coordinates[i*dim+j] = (double) CoordsData[j][i];
*ierr = ZOLTAN_OK;
} //GetProblemGeometry
void IndefBlockedDiagonalSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node, LocalMatOps>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const
{
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::IndefBlockedDiagonalSmoother::Apply(): Setup() has not been called");
Teuchos::RCP<Teuchos::FancyOStream> fos = Teuchos::getFancyOStream(Teuchos::rcpFromRef(std::cout));
SC zero = Teuchos::ScalarTraits<SC>::zero(), one = Teuchos::ScalarTraits<SC>::one();
// extract parameters from internal parameter list
const ParameterList & pL = Factory::GetParameterList();
LocalOrdinal nSweeps = pL.get<LocalOrdinal>("Sweeps");
Scalar omega = pL.get<Scalar>("Damping factor");
// wrap current solution vector in RCP
RCP<MultiVector> rcpX = Teuchos::rcpFromRef(X);
// create residual vector
// contains current residual of current solution X with rhs B
RCP<MultiVector> residual = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
// incrementally improve solution vector X
for (LocalOrdinal run = 0; run < nSweeps; ++run) {
// 1) calculate current residual
residual->update(one,B,zero); // residual = B
A_->apply(*rcpX, *residual, Teuchos::NO_TRANS, -one, one);
// split residual vector
Teuchos::RCP<MultiVector> r1 = rangeMapExtractor_->ExtractVector(residual, 0);
Teuchos::RCP<MultiVector> r2 = rangeMapExtractor_->ExtractVector(residual, 1);
// 2) solve F * \Delta \tilde{x}_1 = r_1
// start with zero guess \Delta \tilde{x}_1
RCP<MultiVector> xtilde1 = MultiVectorFactory::Build(F_->getRowMap(),1);
xtilde1->putScalar(zero);
velPredictSmoo_->Apply(*xtilde1,*r1);
// 3) solve SchurComp equation
// start with zero guess \Delta \tilde{x}_2
RCP<MultiVector> xtilde2 = MultiVectorFactory::Build(Z_->getRowMap(),1);
xtilde2->putScalar(zero);
schurCompSmoo_->Apply(*xtilde2,*r2);
// 4) extract parts of solution vector X
Teuchos::RCP<MultiVector> x1 = domainMapExtractor_->ExtractVector(rcpX, 0);
Teuchos::RCP<MultiVector> x2 = domainMapExtractor_->ExtractVector(rcpX, 1);
// 5) update solution vector with increments xhat1 and xhat2
// rescale increment for x2 with omega_
x1->update(omega,*xtilde1,one); // x1 = x1_old + omega xtilde1
x2->update(omega,*xtilde2,one); // x2 = x2_old + omega xtilde2
// write back solution in global vector X
domainMapExtractor_->InsertVector(x1, 0, rcpX);
domainMapExtractor_->InsertVector(x2, 1, rcpX);
}
}
void
CrsMatrixSolveOp<OpScalar,MatScalar,LocalOrdinal,GlobalOrdinal,Node,LocalMatOps>::apply(
const MultiVector<OpScalar,LocalOrdinal,GlobalOrdinal,Node> & X,
MultiVector<OpScalar,LocalOrdinal,GlobalOrdinal,Node> & Y,
Teuchos::ETransp mode, OpScalar alpha, OpScalar beta) const
{
TEUCHOS_TEST_FOR_EXCEPTION(!matrix_->isFillComplete(), std::runtime_error,
Teuchos::typeName(*this) << "::apply(): underlying matrix is not fill-complete.");
TEUCHOS_TEST_FOR_EXCEPTION(X.getNumVectors() != Y.getNumVectors(), std::runtime_error,
Teuchos::typeName(*this) << "::apply(X,Y): X and Y must have the same number of vectors.");
TEUCHOS_TEST_FOR_EXCEPTION(matrix_->isLowerTriangular() == false && matrix_->isUpperTriangular() == false, std::runtime_error,
Teuchos::typeName(*this) << "::apply() requires either upper or lower triangular structure in underlying matrix.");
TEUCHOS_TEST_FOR_EXCEPTION( alpha != Teuchos::ScalarTraits<OpScalar>::one() || beta != Teuchos::ScalarTraits<OpScalar>::zero(), std::runtime_error,
Teuchos::typeName(*this) << "::apply(): non-trivial alpha,beta not supported at this time.");
if (mode == Teuchos::NO_TRANS) {
applyNonTranspose(X,Y);
}
else {
applyTranspose(X,Y);
}
}
/// \brief Compute \f$Y = \beta Y + \alpha B X\f$, where \f$B X\f$
/// represents the result of the local triangular solve.
void
apply (const MultiVector<Scalar,LocalOrdinal,GlobalOrdinal,Node> & X,
MultiVector<Scalar,LocalOrdinal,GlobalOrdinal,Node> &Y,
Teuchos::ETransp mode = Teuchos::NO_TRANS,
Scalar alpha = Teuchos::ScalarTraits<Scalar>::one (),
Scalar beta = Teuchos::ScalarTraits<Scalar>::zero ()) const
{
typedef Teuchos::ScalarTraits<Scalar> STOS;
const char prefix[] = "Tpetra::CrsMatrixSolveOp::apply: ";
TEUCHOS_TEST_FOR_EXCEPTION
(! matrix_->isFillComplete (), std::runtime_error,
prefix << "Underlying matrix is not fill complete.");
TEUCHOS_TEST_FOR_EXCEPTION
(! matrix_->isLowerTriangular () && ! matrix_->isUpperTriangular (),
std::runtime_error, prefix << "The matrix is neither lower nor upper "
"triangular. Remember that in Tpetra, triangular-ness is a local "
"(per MPI process) property.");
TEUCHOS_TEST_FOR_EXCEPTION
(X.getNumVectors () != Y.getNumVectors (), std::invalid_argument,
prefix << "X and Y must have the same number of columns (vectors). "
"X.getNumVectors() = " << X.getNumVectors ()
<< " != Y.getNumVectors() = " << Y.getNumVectors () << ".");
TEUCHOS_TEST_FOR_EXCEPTION
(alpha != STOS::one () || beta != STOS::zero (), std::logic_error,
prefix << "The case alpha != 1 or beta != 0 has not yet been "
"implemented. Please speak with the Tpetra developers.");
if (mode == Teuchos::NO_TRANS) {
applyNonTranspose (X,Y);
} else if (mode == Teuchos::TRANS || mode == Teuchos::CONJ_TRANS) {
applyTranspose (X, Y, mode);
} else {
TEUCHOS_TEST_FOR_EXCEPTION
(true, std::invalid_argument, prefix << "The 'mode' argument has an "
"invalid value " << mode << ". Valid values are Teuchos::NO_TRANS="
<< Teuchos::NO_TRANS << ", Teuchos::TRANS=" << Teuchos::TRANS << ", "
"and Teuchos::CONJ_TRANS=" << Teuchos::CONJ_TRANS << ".");
}
}
void
apply (const MultiVector<S,LO,GO,Node> &X,
MultiVector<S,LO,GO,Node> &Y,
Teuchos::ETransp mode = Teuchos::NO_TRANS,
S alpha = Teuchos::ScalarTraits<S>::one (),
S beta = Teuchos::ScalarTraits<S>::zero ()) const
{
const size_t numVectors = X.getNumVectors ();
RCP<MultiVector<S,LO,GO,Node> > mvec_inout;
RCP<const MultiVector<S,LO,GO,Node> > mvec_in2;
if (_importer != null) {
if (_importMV != null && _importMV->getNumVectors () != numVectors) {
_importMV = null;
}
if (_importMV == null) {
_importMV = createMultiVector<S> (_importer->getTargetMap (), numVectors);
}
_importMV->doImport (X, *_importer, INSERT);
mvec_in2 = _importMV;
}
else {
mvec_in2 = rcpFromRef(X);
}
if (_exporter != null) {
if (_exportMV != null && _exportMV->getNumVectors () != numVectors) {
_exportMV = null;
}
if (_exportMV == null) {
_exportMV = createMultiVector<S> (_exporter->getSourceMap (), numVectors);
}
mvec_inout = _exportMV;
}
else {
mvec_inout = rcpFromRef (Y);
}
_kernel.setAlphaBeta (alpha, beta);
//
for (size_t j=0; j < numVectors; ++j) {
RCP< Vector<S,LO,GO,Node> > vec_inout = mvec_inout->getVectorNonConst(j);
RCP< const Vector<S,LO,GO,Node> > vec_in2 = mvec_in2->getVector(j);
Tpetra::RTI::detail::binary_transform( *vec_inout, *vec_in2, _kernel );
}
// export
if (_exporter != null) {
Y.doExport (*_exportMV, *_exporter, ADD);
}
}
void Amesos2Smoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const {
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::Amesos2Smoother::Apply(): Setup() has not been called");
RCP<Tpetra_MultiVector> tX, tB;
if (!useTransformation_) {
tX = Utilities::MV2NonConstTpetraMV2(X);
tB = Utilities::MV2NonConstTpetraMV2(const_cast<MultiVector&>(B));
} else {
// Copy data of the original vectors into the transformed ones
size_t numVectors = X.getNumVectors();
size_t length = X.getLocalLength();
TEUCHOS_TEST_FOR_EXCEPTION(numVectors > 1, Exceptions::RuntimeError,
"MueLu::Amesos2Smoother::Apply: Fixing coarse matrix for Amesos2 for multivectors has not been implemented yet.");
ArrayRCP<const SC> Xdata = X. getData(0), Bdata = B. getData(0);
ArrayRCP<SC> X_data = X_->getDataNonConst(0), B_data = B_->getDataNonConst(0);
for (size_t i = 0; i < length; i++) {
X_data[i] = Xdata[i];
B_data[i] = Bdata[i];
}
tX = Utilities::MV2NonConstTpetraMV2(*X_);
tB = Utilities::MV2NonConstTpetraMV2(*B_);
}
prec_->setX(tX);
prec_->setB(tB);
prec_->solve();
prec_->setX(Teuchos::null);
prec_->setB(Teuchos::null);
if (useTransformation_) {
// Copy data from the transformed vectors into the original ones
size_t length = X.getLocalLength();
ArrayRCP<SC> Xdata = X. getDataNonConst(0);
ArrayRCP<const SC> X_data = X_->getData(0);
for (size_t i = 0; i < length; i++)
Xdata[i] = X_data[i];
}
}
void Constraint<Scalar, LocalOrdinal, GlobalOrdinal, Node, LocalMatOps>::Setup(const MultiVector& B, const MultiVector& Bc, RCP<const CrsGraph> Ppattern) {
Ppattern_ = Ppattern;
const RCP<const Map> uniqueMap = Ppattern_->getDomainMap();
const RCP<const Map> nonUniqueMap = Ppattern_->getColMap();
RCP<const Import> importer = ImportFactory::Build(uniqueMap, nonUniqueMap);
const size_t NSDim = Bc.getNumVectors();
X_ = MultiVectorFactory::Build(nonUniqueMap, NSDim);
X_->doImport(Bc, *importer, Xpetra::INSERT);
size_t numRows = Ppattern_->getNodeNumRows();
XXtInv_.resize(numRows);
Teuchos::SerialDenseVector<LO,SC> BcRow(NSDim, false);
for (size_t i = 0; i < numRows; i++) {
Teuchos::ArrayView<const LO> indices;
Ppattern_->getLocalRowView(i, indices);
size_t nnz = indices.size();
Teuchos::SerialDenseMatrix<LO,SC> locX(NSDim, nnz, false);
for (size_t j = 0; j < nnz; j++) {
for (size_t k = 0; k < NSDim; k++)
BcRow[k] = X_->getData(k)[indices[j]];
Teuchos::setCol(BcRow, (LO)j, locX);
}
XXtInv_[i] = Teuchos::SerialDenseMatrix<LO,SC>(NSDim, NSDim, false);
Teuchos::BLAS<LO,SC> blas;
blas.GEMM(Teuchos::NO_TRANS, Teuchos::CONJ_TRANS, NSDim, NSDim, nnz,
Teuchos::ScalarTraits<SC>::one(), locX.values(), locX.stride(),
locX.values(), locX.stride(), Teuchos::ScalarTraits<SC>::zero(),
XXtInv_[i].values(), XXtInv_[i].stride());
Teuchos::LAPACK<LO,SC> lapack;
LO info, lwork = 3*NSDim;
ArrayRCP<LO> IPIV(NSDim);
ArrayRCP<SC> WORK(lwork);
lapack.GETRF(NSDim, NSDim, XXtInv_[i].values(), XXtInv_[i].stride(), IPIV.get(), &info);
lapack.GETRI(NSDim, XXtInv_[i].values(), XXtInv_[i].stride(), IPIV.get(), WORK.get(), lwork, &info);
}
}
void BraessSarazinSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const {
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError,
"MueLu::BraessSarazinSmoother::Apply(): Setup() has not been called");
RCP<MultiVector> rcpX = rcpFromRef(X);
RCP<MultiVector> deltaX0 = MultiVectorFactory::Build(A00_->getRowMap(), 1);
RCP<MultiVector> deltaX1 = MultiVectorFactory::Build(A10_->getRowMap(), 1);
RCP<MultiVector> Rtmp = MultiVectorFactory::Build(A10_->getRowMap(), 1);
typedef Teuchos::ScalarTraits<SC> STS;
SC one = STS::one(), zero = STS::zero();
// extract parameters from internal parameter list
const ParameterList& pL = Factory::GetParameterList();
LO nSweeps = pL.get<LO>("Sweeps");
RCP<MultiVector> R;
if (InitialGuessIsZero) {
R = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
R->update(one, B, zero);
} else {
R = Utilities::Residual(*A_, X, B);
}
for (LO run = 0; run < nSweeps; ++run) {
// Extract corresponding subvectors from X and R
RCP<MultiVector> R0 = rangeMapExtractor_ ->ExtractVector(R, 0);
RCP<MultiVector> R1 = rangeMapExtractor_ ->ExtractVector(R, 1);
RCP<MultiVector> X0 = domainMapExtractor_->ExtractVector(rcpX, 0);
RCP<MultiVector> X1 = domainMapExtractor_->ExtractVector(rcpX, 1);
// Calculate Rtmp = R1 - D * deltaX0 (equation 8.14)
deltaX0->putScalar(zero);
deltaX0->elementWiseMultiply(one, *D_, *R0, zero); // deltaX0 = D * R0 (equation 8.13)
A10_->apply(*deltaX0, *Rtmp); // Rtmp = A10*D*deltaX0 (intermediate step)
Rtmp->update(one, *R1, -one); // Rtmp = R1 - A10*D*deltaX0
// Compute deltaX1 (pressure correction)
// We use user provided preconditioner
deltaX1->putScalar(zero); // just for safety
smoo_->Apply(*deltaX1, *Rtmp);
// Compute deltaX0
deltaX0->putScalar(zero); // just for safety
A01_->apply(*deltaX1, *deltaX0); // deltaX0 = A01*deltaX1
deltaX0->update(one, *R0, -one); // deltaX0 = R0 - A01*deltaX1
R0.swap(deltaX0);
deltaX0->elementWiseMultiply(one, *D_, *R0, zero); // deltaX0 = D*(R0 - A01*deltaX1)
// Update solution
X0->update(one, *deltaX0, one);
X1->update(one, *deltaX1, one);
domainMapExtractor_->InsertVector(X0, 0, rcpX);
domainMapExtractor_->InsertVector(X1, 1, rcpX);
if (run < nSweeps-1)
R = Utilities::Residual(*A_, X, B);
}
}
void Constraint<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Setup(const MultiVector& B, const MultiVector& Bc, RCP<const CrsGraph> Ppattern) {
const size_t NSDim = Bc.getNumVectors();
Ppattern_ = Ppattern;
size_t numRows = Ppattern_->getNodeNumRows();
XXtInv_.resize(numRows);
RCP<const Import> importer = Ppattern_->getImporter();
X_ = MultiVectorFactory::Build(Ppattern_->getColMap(), NSDim);
if (!importer.is_null())
X_->doImport(Bc, *importer, Xpetra::INSERT);
else
*X_ = Bc;
std::vector<const SC*> Xval(NSDim);
for (size_t j = 0; j < NSDim; j++)
Xval[j] = X_->getData(j).get();
SC zero = Teuchos::ScalarTraits<SC>::zero();
SC one = Teuchos::ScalarTraits<SC>::one();
Teuchos::BLAS <LO,SC> blas;
Teuchos::LAPACK<LO,SC> lapack;
LO lwork = 3*NSDim;
ArrayRCP<LO> IPIV(NSDim);
ArrayRCP<SC> WORK(lwork);
for (size_t i = 0; i < numRows; i++) {
Teuchos::ArrayView<const LO> indices;
Ppattern_->getLocalRowView(i, indices);
size_t nnz = indices.size();
XXtInv_[i] = Teuchos::SerialDenseMatrix<LO,SC>(NSDim, NSDim, false/*zeroOut*/);
Teuchos::SerialDenseMatrix<LO,SC>& XXtInv = XXtInv_[i];
if (NSDim == 1) {
SC d = zero;
for (size_t j = 0; j < nnz; j++)
d += Xval[0][indices[j]] * Xval[0][indices[j]];
XXtInv(0,0) = one/d;
} else {
Teuchos::SerialDenseMatrix<LO,SC> locX(NSDim, nnz, false/*zeroOut*/);
for (size_t j = 0; j < nnz; j++)
for (size_t k = 0; k < NSDim; k++)
locX(k,j) = Xval[k][indices[j]];
// XXtInv_ = (locX*locX^T)^{-1}
blas.GEMM(Teuchos::NO_TRANS, Teuchos::CONJ_TRANS, NSDim, NSDim, nnz,
one, locX.values(), locX.stride(),
locX.values(), locX.stride(),
zero, XXtInv.values(), XXtInv.stride());
LO info;
// Compute LU factorization using partial pivoting with row exchanges
lapack.GETRF(NSDim, NSDim, XXtInv.values(), XXtInv.stride(), IPIV.get(), &info);
// Use the computed factorization to compute the inverse
lapack.GETRI(NSDim, XXtInv.values(), XXtInv.stride(), IPIV.get(), WORK.get(), lwork, &info);
}
}
}
void BlockedGaussSeidelSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector &X, const MultiVector& B, bool InitialGuessIsZero) const
{
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::BlockedGaussSeidelSmoother::Apply(): Setup() has not been called");
RCP<Xpetra::MultiVector<Scalar, LocalOrdinal, GlobalOrdinal, Node> > residual = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
RCP<Xpetra::MultiVector<Scalar, LocalOrdinal, GlobalOrdinal, Node> > tempres = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
RCP<MultiVector> rcpX = Teuchos::rcpFromRef(X);
//Teuchos::RCP<Teuchos::FancyOStream> fos = Teuchos::getFancyOStream(Teuchos::rcpFromRef(std::cout));
// extract parameters from internal parameter list
const ParameterList & pL = Factory::GetParameterList();
LocalOrdinal nSweeps = pL.get<LocalOrdinal>("Sweeps");
Scalar omega = pL.get<Scalar>("Damping factor");
// outer Richardson loop
for (LocalOrdinal run = 0; run < nSweeps; ++run) {
// one BGS sweep
// loop over all block rows
for(size_t i = 0; i<Inverse_.size(); i++) {
// calculate block residual r = B-A*X
// note: A_ is the full blocked operator
residual->update(1.0,B,0.0); // r = B
A_->apply(X, *residual, Teuchos::NO_TRANS, -1.0, 1.0);
// extract corresponding subvectors from X and residual
size_t blockRowIndex = at(bgsOrderingIndex2blockRowIndex_, i); // == bgsOrderingIndex2blockRowIndex_.at(i) (only available since C++11)
Teuchos::RCP<MultiVector> Xi = domainMapExtractor_->ExtractVector(rcpX, blockRowIndex);
Teuchos::RCP<MultiVector> ri = rangeMapExtractor_->ExtractVector(residual, blockRowIndex);
Teuchos::RCP<MultiVector> tXi = domainMapExtractor_->getVector(blockRowIndex, X.getNumVectors());
// apply solver/smoother
Inverse_.at(i)->Apply(*tXi, *ri, false);
// update vector
Xi->update(omega,*tXi,1.0); // X_{i+1} = X_i + omega \Delta X_i
// update corresponding part of rhs and lhs
domainMapExtractor_->InsertVector(Xi, blockRowIndex, rcpX); // TODO wrong! fix me
}
}
}
void
CrsMatrixSolveOp<OpScalar,MatScalar,LocalOrdinal,GlobalOrdinal,Node,LocalMatOps>::applyNonTranspose(
const MultiVector<OpScalar,LocalOrdinal,GlobalOrdinal,Node> & X_in,
MultiVector<OpScalar,LocalOrdinal,GlobalOrdinal,Node> & Y_in) const
{
// Solve U X = Y or L X = Y
// X belongs to domain map, while Y belongs to range map
typedef Teuchos::ScalarTraits<OpScalar> ST;
using Teuchos::null;
const size_t numVectors = X_in.getNumVectors();
Teuchos::RCP<const Import<LocalOrdinal,GlobalOrdinal,Node> > importer = matrix_->getGraph()->getImporter();
Teuchos::RCP<const Export<LocalOrdinal,GlobalOrdinal,Node> > exporter = matrix_->getGraph()->getExporter();
Teuchos::RCP<const MV> X;
// it is okay if X and Y reference the same data, because we can perform a triangular solve in-situ.
// however, we require that column access to each is strided.
// set up import/export temporary multivectors
if (importer != null) {
if (importMV_ != null && importMV_->getNumVectors() != numVectors) importMV_ = null;
if (importMV_ == null) {
importMV_ = Teuchos::rcp( new MV(matrix_->getColMap(),numVectors) );
}
}
if (exporter != null) {
if (exportMV_ != null && exportMV_->getNumVectors() != numVectors) exportMV_ = null;
if (exportMV_ == null) {
exportMV_ = Teuchos::rcp( new MV(matrix_->getRowMap(),numVectors) );
}
}
// solve(NO_TRANS): RangeMap -> DomainMap
// lclMatSolve_: RowMap -> ColMap
// importer: DomainMap -> ColMap
// exporter: RowMap -> RangeMap
//
// solve = reverse(exporter) o lclMatSolve_ o reverse(importer)
// RangeMap -> RowMap -> ColMap -> DomainMap
// If we have a non-trivial exporter, we must import elements that are permuted or are on other processors
if (exporter != null) {
exportMV_->doImport(X_in, *exporter, INSERT);
X = exportMV_;
}
else if (X_in.isConstantStride() == false) {
// cannot handle non-constant stride right now
// generate a copy of X_in
X = Teuchos::rcp(new MV(X_in));
}
else {
// just temporary, so this non-owning RCP is okay
X = Teuchos::rcp( &X_in, false );
}
// If we have a non-trivial importer, we must export elements that are permuted or belong to other processors
// We will compute solution into the to-be-exported MV
if (importer != null) {
matrix_->template localSolve<OpScalar,OpScalar>(*X,*importMV_,Teuchos::NO_TRANS);
// Make sure target is zero: necessary because we are adding.
Y_in.putScalar(ST::zero());
Y_in.doExport(*importMV_, *importer, ADD);
}
// otherwise, solve into Y
else {
// can't solve into non-strided multivector
if (Y_in.isConstantStride() == false) {
// generate a strided copy of Y
MV Y(Y_in);
matrix_->template localSolve<OpScalar,OpScalar>(*X,Y,Teuchos::NO_TRANS);
Y_in = Y;
}
else {
matrix_->template localSolve<OpScalar,OpScalar>(*X,Y_in,Teuchos::NO_TRANS);
}
}
}
void Ifpack2Smoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const {
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::IfpackSmoother::Apply(): Setup() has not been called");
// Forward the InitialGuessIsZero option to Ifpack2
// TODO: It might be nice to switch back the internal
// "zero starting solution" option of the ifpack2 object prec_ to his
// initial value at the end but there is no way right now to get
// the current value of the "zero starting solution" in ifpack2.
// It's not really an issue, as prec_ can only be used by this method.
// TODO: When https://software.sandia.gov/bugzilla/show_bug.cgi?id=5283#c2 is done
// we should remove the if/else/elseif and just test if this
// option is supported by current ifpack2 preconditioner
Teuchos::ParameterList paramList;
bool supportInitialGuess = false;
if (type_ == "CHEBYSHEV") {
paramList.set("chebyshev: zero starting solution", InitialGuessIsZero);
SetPrecParameters(paramList);
supportInitialGuess = true;
} else if (type_ == "RELAXATION") {
paramList.set("relaxation: zero starting solution", InitialGuessIsZero);
SetPrecParameters(paramList);
supportInitialGuess = true;
} else if (type_ == "KRYLOV") {
paramList.set("krylov: zero starting solution", InitialGuessIsZero);
SetPrecParameters(paramList);
supportInitialGuess = true;
} else if (type_ == "SCHWARZ") {
paramList.set("schwarz: zero starting solution", InitialGuessIsZero);
//Because additive Schwarz has "delta" semantics, it's sufficient to
//toggle only the zero initial guess flag, and not pass in already
//set parameters. If we call SetPrecParameters, the subdomain solver
//will be destroyed.
prec_->setParameters(paramList);
supportInitialGuess = true;
}
//TODO JJH 30Apr2014 Calling SetPrecParameters(paramList) when the smoother
//is Ifpack2::AdditiveSchwarz::setParameterList() will destroy the subdomain
//(aka inner) solver. This behavior is documented but a departure from what
//it previously did, and what other Ifpack2 solvers currently do. So I have
//moved SetPrecParameters(paramList) into the if-else block above.
// Apply
if (InitialGuessIsZero || supportInitialGuess) {
Tpetra::MultiVector<SC,LO,GO,NO>& tpX = Utilities::MV2NonConstTpetraMV(X);
const Tpetra::MultiVector<SC,LO,GO,NO>& tpB = Utilities::MV2TpetraMV(B);
prec_->apply(tpB, tpX);
} else {
typedef Teuchos::ScalarTraits<Scalar> TST;
RCP<MultiVector> Residual = Utilities::Residual(*A_, X, B);
RCP<MultiVector> Correction = MultiVectorFactory::Build(A_->getDomainMap(), X.getNumVectors());
Tpetra::MultiVector<SC,LO,GO,NO>& tpX = Utilities::MV2NonConstTpetraMV(*Correction);
const Tpetra::MultiVector<SC,LO,GO,NO>& tpB = Utilities::MV2TpetraMV(*Residual);
prec_->apply(tpB, tpX);
X.update(TST::one(), *Correction, TST::one());
}
}
void IndefBlockedDiagonalSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const
{
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::IndefBlockedDiagonalSmoother::Apply(): Setup() has not been called");
Teuchos::RCP<Teuchos::FancyOStream> fos = Teuchos::getFancyOStream(Teuchos::rcpFromRef(std::cout));
SC zero = Teuchos::ScalarTraits<SC>::zero(), one = Teuchos::ScalarTraits<SC>::one();
// The following boolean flags catch the case where we need special transformation
// for the GIDs when calling the subsmoothers.
RCP<BlockedCrsMatrix> bA00 = Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(F_);
RCP<BlockedCrsMatrix> bA11 = Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(Z_);
bool bA00ThyraSpecialTreatment = false;
bool bA11ThyraSpecialTreatment = false;
if (bA00 != Teuchos::null) {
if(bA00->Rows() == 1 && bA00->Cols() == 1 && rangeMapExtractor_->getThyraMode() == true) bA00ThyraSpecialTreatment = true;
}
if (bA11 != Teuchos::null) {
if(bA11->Rows() == 1 && bA11->Cols() == 1 && rangeMapExtractor_->getThyraMode() == true) bA11ThyraSpecialTreatment = true;
}
// extract parameters from internal parameter list
const ParameterList & pL = Factory::GetParameterList();
LocalOrdinal nSweeps = pL.get<LocalOrdinal>("Sweeps");
Scalar omega = pL.get<Scalar>("Damping factor");
// wrap current solution vector in RCP
RCP<MultiVector> rcpX = Teuchos::rcpFromRef(X);
// create residual vector
// contains current residual of current solution X with rhs B
RCP<MultiVector> residual = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
// incrementally improve solution vector X
for (LocalOrdinal run = 0; run < nSweeps; ++run) {
// 1) calculate current residual
residual->update(one,B,zero); // residual = B
A_->apply(*rcpX, *residual, Teuchos::NO_TRANS, -one, one);
// split residual vector
Teuchos::RCP<MultiVector> r1 = rangeMapExtractor_->ExtractVector(residual, 0);
Teuchos::RCP<MultiVector> r2 = rangeMapExtractor_->ExtractVector(residual, 1);
// 2) solve F * \Delta \tilde{x}_1 = r_1
// start with zero guess \Delta \tilde{x}_1
RCP<MultiVector> xtilde1 = MultiVectorFactory::Build(F_->getRowMap(),X.getNumVectors(),true);
// Special handling if SchurComplement operator was a 1x1 blocked operator in Thyra mode
// Then, we have to translate the Xpetra offset GIDs to plain Thyra GIDs and vice versa
if(bA00ThyraSpecialTreatment == true) {
RCP<MultiVector> xtilde1_thyra = domainMapExtractor_->getVector(0, X.getNumVectors(), true);
RCP<MultiVector> r1_thyra = rangeMapExtractor_->getVector(0, B.getNumVectors(), true);
// transform vector
for(size_t k=0; k < r1->getNumVectors(); k++) {
Teuchos::ArrayRCP<const Scalar> xpetraVecData = r1->getData(k);
Teuchos::ArrayRCP<Scalar> thyraVecData = r1_thyra->getDataNonConst(k);
for(size_t i=0; i < r1->getLocalLength(); i++) {
thyraVecData[i] = xpetraVecData[i];
}
}
velPredictSmoo_->Apply(*xtilde1_thyra,*r1_thyra);
for(size_t k=0; k < xtilde1_thyra->getNumVectors(); k++) {
Teuchos::ArrayRCP<Scalar> xpetraVecData = xtilde1->getDataNonConst(k);
Teuchos::ArrayRCP<const Scalar> thyraVecData = xtilde1_thyra->getData(k);
for(size_t i=0; i < xtilde1_thyra->getLocalLength(); i++) {
xpetraVecData[i] = thyraVecData[i];
}
}
} else {
velPredictSmoo_->Apply(*xtilde1,*r1);
}
// 3) solve SchurComp equation
// start with zero guess \Delta \tilde{x}_2
RCP<MultiVector> xtilde2 = MultiVectorFactory::Build(Z_->getRowMap(),X.getNumVectors(),true);
// Special handling if SchurComplement operator was a 1x1 blocked operator in Thyra mode
// Then, we have to translate the Xpetra offset GIDs to plain Thyra GIDs and vice versa
if(bA11ThyraSpecialTreatment == true) {
RCP<MultiVector> xtilde2_thyra = domainMapExtractor_->getVector(1, X.getNumVectors(), true);
RCP<MultiVector> r2_thyra = rangeMapExtractor_->getVector(1, B.getNumVectors(), true);
// transform vector
for(size_t k=0; k < r2->getNumVectors(); k++) {
Teuchos::ArrayRCP<const Scalar> xpetraVecData = r2->getData(k);
Teuchos::ArrayRCP<Scalar> thyraVecData = r2_thyra->getDataNonConst(k);
for(size_t i=0; i < r2->getLocalLength(); i++) {
thyraVecData[i] = xpetraVecData[i];
}
}
schurCompSmoo_->Apply(*xtilde2_thyra,*r2_thyra);
for(size_t k=0; k < xtilde2_thyra->getNumVectors(); k++) {
Teuchos::ArrayRCP<Scalar> xpetraVecData = xtilde2->getDataNonConst(k);
Teuchos::ArrayRCP<const Scalar> thyraVecData = xtilde2_thyra->getData(k);
for(size_t i=0; i < xtilde2_thyra->getLocalLength(); i++) {
xpetraVecData[i] = thyraVecData[i];
}
}
} else {
schurCompSmoo_->Apply(*xtilde2,*r2);
}
// 4) extract parts of solution vector X
Teuchos::RCP<MultiVector> x1 = domainMapExtractor_->ExtractVector(rcpX, 0);
Teuchos::RCP<MultiVector> x2 = domainMapExtractor_->ExtractVector(rcpX, 1);
// 5) update solution vector with increments xhat1 and xhat2
// rescale increment for x2 with omega_
x1->update(omega,*xtilde1,one); // x1 = x1_old + omega xtilde1
x2->update(omega,*xtilde2,one); // x2 = x2_old + omega xtilde2
// write back solution in global vector X
domainMapExtractor_->InsertVector(x1, 0, rcpX);
domainMapExtractor_->InsertVector(x2, 1, rcpX);
}
}
void SimpleSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const
{
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError, "MueLu::SimpleSmoother::Apply(): Setup() has not been called");
#ifdef HAVE_MUELU_DEBUG
TEUCHOS_TEST_FOR_EXCEPTION(A_->getRangeMap()->isSameAs(*(B.getMap())) == false, Exceptions::RuntimeError, "MueLu::SimpleSmoother::Apply(): The map of RHS vector B is not the same as range map of the blocked operator A. Please check the map of B and A.");
TEUCHOS_TEST_FOR_EXCEPTION(A_->getDomainMap()->isSameAs(*(X.getMap())) == false, Exceptions::RuntimeError, "MueLu::SimpleSmoother::Apply(): The map of the solution vector X is not the same as domain map of the blocked operator A. Please check the map of X and A.");
#endif
Teuchos::RCP<Teuchos::FancyOStream> fos = Teuchos::getFancyOStream(Teuchos::rcpFromRef(std::cout));
SC zero = Teuchos::ScalarTraits<SC>::zero(), one = Teuchos::ScalarTraits<SC>::one();
// extract parameters from internal parameter list
const ParameterList & pL = Factory::GetParameterList();
LocalOrdinal nSweeps = pL.get<LocalOrdinal>("Sweeps");
Scalar omega = pL.get<Scalar>("Damping factor");
// The boolean flags check whether we use Thyra or Xpetra style GIDs
// However, assuming that SIMPLE always only works for 2x2 blocked operators, we
// most often have to use the ReorderedBlockedCrsOperator as input. If either the
// F or Z (or SchurComplement block S) are 1x1 blocked operators with Thyra style
// GIDs we need an extra transformation of vectors
// In this case, we use the Xpetra (offset) GIDs for all operations and only transform
// the input/output vectors before and after the subsolver calls!
bool bRangeThyraModePredict = rangeMapExtractor_->getThyraMode() && (Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(F_) == Teuchos::null);
bool bDomainThyraModePredict = domainMapExtractor_->getThyraMode() && (Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(F_) == Teuchos::null);
bool bRangeThyraModeSchur = rangeMapExtractor_->getThyraMode() && (Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(Z_) == Teuchos::null);
bool bDomainThyraModeSchur = domainMapExtractor_->getThyraMode() && (Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(Z_) == Teuchos::null);
// The following boolean flags catch the case where we need special transformation
// for the GIDs when calling the subsmoothers.
RCP<BlockedCrsMatrix> bF = Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(F_);
RCP<BlockedCrsMatrix> bZ = Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(Z_);
bool bFThyraSpecialTreatment = false;
bool bZThyraSpecialTreatment = false;
if (bF != Teuchos::null) {
if(bF->Rows() == 1 && bF->Cols() == 1 && rangeMapExtractor_->getThyraMode() == true) bFThyraSpecialTreatment = true;
}
if (bZ != Teuchos::null) {
if(bZ->Rows() == 1 && bZ->Cols() == 1 && rangeMapExtractor_->getThyraMode() == true) bZThyraSpecialTreatment = true;
}
#if 1// new implementation
// create a new vector for storing the current residual in a blocked multi vector
RCP<MultiVector> res = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
RCP<BlockedMultiVector> residual = Teuchos::rcp(new BlockedMultiVector(rangeMapExtractor_,res));
// create a new solution vector as a blocked multi vector
RCP<MultiVector> rcpX = Teuchos::rcpFromRef(X);
RCP<BlockedMultiVector> bX = Teuchos::rcp(new BlockedMultiVector(domainMapExtractor_,rcpX));
// create a blocked rhs vector
RCP<const MultiVector> rcpB = Teuchos::rcpFromRef(B);
RCP<const BlockedMultiVector> bB = Teuchos::rcp(new const BlockedMultiVector(rangeMapExtractor_,rcpB));
// incrementally improve solution vector X
for (LocalOrdinal run = 0; run < nSweeps; ++run) {
// 1) calculate current residual
residual->update(one,*bB,zero); // r = B
A_->apply(*bX, *residual, Teuchos::NO_TRANS, -one, one);
// split residual vector
Teuchos::RCP<MultiVector> r1 = rangeMapExtractor_->ExtractVector(residual, 0, bRangeThyraModePredict);
Teuchos::RCP<MultiVector> r2 = rangeMapExtractor_->ExtractVector(residual, 1, bRangeThyraModeSchur);
// 2) solve F * \Delta \tilde{x}_1 = r_1
// start with zero guess \Delta \tilde{x}_1
RCP<MultiVector> xtilde1 = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
xtilde1->putScalar(zero);
if(bFThyraSpecialTreatment == true) {
xtilde1->replaceMap(domainMapExtractor_->getMap(0,true));
r1->replaceMap(rangeMapExtractor_->getMap(0,true));
velPredictSmoo_->Apply(*xtilde1,*r1);
xtilde1->replaceMap(domainMapExtractor_->getMap(0,false));
} else {
velPredictSmoo_->Apply(*xtilde1,*r1);
}
// 3) calculate rhs for SchurComp equation
// r_2 - D \Delta \tilde{x}_1
RCP<MultiVector> schurCompRHS = rangeMapExtractor_->getVector(1, B.getNumVectors(), bRangeThyraModeSchur);
D_->apply(*xtilde1,*schurCompRHS);
schurCompRHS->update(one,*r2,-one);
// 4) solve SchurComp equation
// start with zero guess \Delta \tilde{x}_2
RCP<MultiVector> xtilde2 = domainMapExtractor_->getVector(1, X.getNumVectors(), bDomainThyraModeSchur);
xtilde2->putScalar(zero);
// Special handling if SchurComplement operator was a 1x1 blocked operator in Thyra mode
// Then, we have to translate the Xpetra offset GIDs to plain Thyra GIDs and vice versa
if(bZThyraSpecialTreatment == true) {
xtilde2->replaceMap(domainMapExtractor_->getMap(1,true));
schurCompRHS->replaceMap(rangeMapExtractor_->getMap(1,true));
schurCompSmoo_->Apply(*xtilde2,*schurCompRHS);
xtilde2->replaceMap(domainMapExtractor_->getMap(1,false));
} else {
schurCompSmoo_->Apply(*xtilde2,*schurCompRHS);
}
// 5) scale xtilde2 with omega
// store this in xhat2
RCP<MultiVector> xhat2 = domainMapExtractor_->getVector(1, X.getNumVectors(), bDomainThyraModeSchur);
xhat2->update(omega,*xtilde2,zero);
// 6) calculate xhat1
RCP<MultiVector> xhat1 = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
RCP<MultiVector> xhat1_temp = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
G_->apply(*xhat2,*xhat1_temp); // store result temporarely in xtilde1_temp
xhat1->elementWiseMultiply(one/*/omega*/,*diagFinv_,*xhat1_temp,zero);
xhat1->update(one,*xtilde1,-one);
// 7) extract parts of solution vector X
Teuchos::RCP<MultiVector> x1 = domainMapExtractor_->ExtractVector(bX, 0, bDomainThyraModePredict);
Teuchos::RCP<MultiVector> x2 = domainMapExtractor_->ExtractVector(bX, 1, bDomainThyraModeSchur);
// 8) update solution vector with increments xhat1 and xhat2
// rescale increment for x2 with omega_
x1->update(one,*xhat1,one); // x1 = x1_old + xhat1
x2->update(/*omega*/ one,*xhat2,one); // x2 = x2_old + omega xhat2
// write back solution in global vector X
domainMapExtractor_->InsertVector(x1, 0, bX, bDomainThyraModePredict);
domainMapExtractor_->InsertVector(x2, 1, bX, bDomainThyraModeSchur);
}
// write back solution
domainMapExtractor_->InsertVector(bX->getMultiVector(0,bDomainThyraModePredict), 0, rcpX, bDomainThyraModePredict);
domainMapExtractor_->InsertVector(bX->getMultiVector(1,bDomainThyraModeSchur), 1, rcpX, bDomainThyraModeSchur);
#else
// wrap current solution vector in RCP
RCP<MultiVector> rcpX = Teuchos::rcpFromRef(X);
// create residual vector
// contains current residual of current solution X with rhs B
RCP<MultiVector> residual = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
// incrementally improve solution vector X
for (LocalOrdinal run = 0; run < nSweeps; ++run) {
// 1) calculate current residual
residual->update(one,B,zero); // residual = B
A_->apply(*rcpX, *residual, Teuchos::NO_TRANS, -one, one);
// split residual vector
Teuchos::RCP<MultiVector> r1 = rangeMapExtractor_->ExtractVector(residual, 0, bRangeThyraModePredict);
Teuchos::RCP<MultiVector> r2 = rangeMapExtractor_->ExtractVector(residual, 1, bRangeThyraModeSchur);
// 2) solve F * \Delta \tilde{x}_1 = r_1
// start with zero guess \Delta \tilde{x}_1
RCP<MultiVector> xtilde1 = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
xtilde1->putScalar(zero);
// Special handling in case that F block is a 1x1 blocked operator in Thyra mode
// Then we have to feed the smoother with real Thyra-based vectors
if(bFThyraSpecialTreatment == true) {
// create empty solution vector based on Thyra GIDs
RCP<MultiVector> xtilde1_thyra = domainMapExtractor_->getVector(0, X.getNumVectors(), true);
// create new RHS vector based on Thyra GIDs
Teuchos::RCP<MultiVector> r1_thyra = rangeMapExtractor_->ExtractVector(residual, 0, true);
velPredictSmoo_->Apply(*xtilde1_thyra,*r1_thyra);
for(size_t k=0; k < xtilde1_thyra->getNumVectors(); k++) {
Teuchos::ArrayRCP<Scalar> xpetraVecData = xtilde1->getDataNonConst(k);
Teuchos::ArrayRCP<const Scalar> thyraVecData = xtilde1_thyra->getData(k);
for(size_t i=0; i < xtilde1_thyra->getLocalLength(); i++) {
xpetraVecData[i] = thyraVecData[i];
}
}
} else {
velPredictSmoo_->Apply(*xtilde1,*r1);
}
// 3) calculate rhs for SchurComp equation
// r_2 - D \Delta \tilde{x}_1
RCP<MultiVector> schurCompRHS = rangeMapExtractor_->getVector(1, B.getNumVectors(), bRangeThyraModeSchur);
D_->apply(*xtilde1,*schurCompRHS);
schurCompRHS->update(one,*r2,-one);
// 4) solve SchurComp equation
// start with zero guess \Delta \tilde{x}_2
RCP<MultiVector> xtilde2 = domainMapExtractor_->getVector(1, X.getNumVectors(), bDomainThyraModeSchur);
xtilde2->putScalar(zero);
// Special handling if SchurComplement operator was a 1x1 blocked operator in Thyra mode
// Then, we have to translate the Xpetra offset GIDs to plain Thyra GIDs and vice versa
if(bZThyraSpecialTreatment == true) {
// create empty solution vector based on Thyra GIDs
RCP<MultiVector> xtilde2_thyra = domainMapExtractor_->getVector(1, X.getNumVectors(), true);
// create new RHS vector based on Thyra GIDs
RCP<MultiVector> schurCompRHS_thyra = rangeMapExtractor_->getVector(1, B.getNumVectors(), true);
// transform vector
for(size_t k=0; k < schurCompRHS->getNumVectors(); k++) {
Teuchos::ArrayRCP<const Scalar> xpetraVecData = schurCompRHS->getData(k);
Teuchos::ArrayRCP<Scalar> thyraVecData = schurCompRHS_thyra->getDataNonConst(k);
for(size_t i=0; i < schurCompRHS->getLocalLength(); i++) {
thyraVecData[i] = xpetraVecData[i];
}
}
schurCompSmoo_->Apply(*xtilde2_thyra,*schurCompRHS_thyra);
for(size_t k=0; k < xtilde2_thyra->getNumVectors(); k++) {
Teuchos::ArrayRCP<Scalar> xpetraVecData = xtilde2->getDataNonConst(k);
Teuchos::ArrayRCP<const Scalar> thyraVecData = xtilde2_thyra->getData(k);
for(size_t i=0; i < xtilde2_thyra->getLocalLength(); i++) {
xpetraVecData[i] = thyraVecData[i];
}
}
} else {
schurCompSmoo_->Apply(*xtilde2,*schurCompRHS);
}
// 5) scale xtilde2 with omega
// store this in xhat2
RCP<MultiVector> xhat2 = domainMapExtractor_->getVector(1, X.getNumVectors(), bDomainThyraModeSchur);
xhat2->update(omega,*xtilde2,zero);
// 6) calculate xhat1
RCP<MultiVector> xhat1 = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
RCP<MultiVector> xhat1_temp = domainMapExtractor_->getVector(0, X.getNumVectors(), bDomainThyraModePredict);
G_->apply(*xhat2,*xhat1_temp); // store result temporarely in xtilde1_temp
xhat1->elementWiseMultiply(one/*/omega*/,*diagFinv_,*xhat1_temp,zero);
xhat1->update(one,*xtilde1,-one);
// 7) extract parts of solution vector X
Teuchos::RCP<MultiVector> x1 = domainMapExtractor_->ExtractVector(rcpX, 0, bDomainThyraModePredict);
Teuchos::RCP<MultiVector> x2 = domainMapExtractor_->ExtractVector(rcpX, 1, bDomainThyraModeSchur);
// 8) update solution vector with increments xhat1 and xhat2
// rescale increment for x2 with omega_
x1->update(one,*xhat1,one); // x1 = x1_old + xhat1
x2->update(/*omega*/ one,*xhat2,one); // x2 = x2_old + omega xhat2
// write back solution in global vector X
domainMapExtractor_->InsertVector(x1, 0, rcpX, bDomainThyraModePredict);
domainMapExtractor_->InsertVector(x2, 1, rcpX, bDomainThyraModeSchur);
}
#endif
}
void BraessSarazinSmoother<Scalar, LocalOrdinal, GlobalOrdinal, Node>::Apply(MultiVector& X, const MultiVector& B, bool InitialGuessIsZero) const {
TEUCHOS_TEST_FOR_EXCEPTION(SmootherPrototype::IsSetup() == false, Exceptions::RuntimeError,
"MueLu::BraessSarazinSmoother::Apply(): Setup() has not been called");
#ifdef HAVE_MUELU_DEBUG
TEUCHOS_TEST_FOR_EXCEPTION(A_->getRangeMap()->isSameAs(*(B.getMap())) == false, Exceptions::RuntimeError, "MueLu::BlockedGaussSeidelSmoother::Apply(): The map of RHS vector B is not the same as range map of the blocked operator A. Please check the map of B and A.");
TEUCHOS_TEST_FOR_EXCEPTION(A_->getDomainMap()->isSameAs(*(X.getMap())) == false, Exceptions::RuntimeError, "MueLu::BlockedGaussSeidelSmoother::Apply(): The map of the solution vector X is not the same as domain map of the blocked operator A. Please check the map of X and A.");
#endif
// The following boolean flags catch the case where we need special transformation
// for the GIDs when calling the subsmoothers.
RCP<BlockedCrsMatrix> bA11 = Teuchos::rcp_dynamic_cast<BlockedCrsMatrix>(A11_);
bool bA11ThyraSpecialTreatment = false;
if (bA11 != Teuchos::null) {
if(bA11->Rows() == 1 && bA11->Cols() == 1 && rangeMapExtractor_->getThyraMode() == true) bA11ThyraSpecialTreatment = true;
}
RCP<MultiVector> rcpX = rcpFromRef(X);
// use the GIDs of the sub blocks
// This is valid as the subblocks actually represent the GIDs (either Thyra, Xpetra or pseudo Xpetra)
RCP<MultiVector> deltaX0 = MultiVectorFactory::Build(A00_->getRowMap(), X.getNumVectors());
RCP<MultiVector> deltaX1 = MultiVectorFactory::Build(A10_->getRowMap(), X.getNumVectors());
RCP<MultiVector> Rtmp = MultiVectorFactory::Build(A10_->getRowMap(), B.getNumVectors());
typedef Teuchos::ScalarTraits<SC> STS;
SC one = STS::one(), zero = STS::zero();
// extract parameters from internal parameter list
const ParameterList& pL = Factory::GetParameterList();
LO nSweeps = pL.get<LO>("Sweeps");
RCP<MultiVector> R;
if (InitialGuessIsZero) {
R = MultiVectorFactory::Build(B.getMap(), B.getNumVectors());
R->update(one, B, zero);
} else {
R = Utilities::Residual(*A_, X, B);
}
// extract diagonal of Schur complement operator
RCP<Vector> diagSVector = VectorFactory::Build(S_->getRowMap());
S_->getLocalDiagCopy(*diagSVector);
ArrayRCP<SC> Sdiag = diagSVector->getDataNonConst(0);
for (LO run = 0; run < nSweeps; ++run) {
// Extract corresponding subvectors from X and R
// Automatically detect whether we use Thyra or Xpetra GIDs
// The GIDs should be always compatible with the GIDs of A00, A01, etc...
RCP<MultiVector> R0 = rangeMapExtractor_ ->ExtractVector(R, 0);
RCP<MultiVector> R1 = rangeMapExtractor_ ->ExtractVector(R, 1);
RCP<MultiVector> X0 = domainMapExtractor_->ExtractVector(rcpX, 0);
RCP<MultiVector> X1 = domainMapExtractor_->ExtractVector(rcpX, 1);
// Calculate Rtmp = R1 - D * deltaX0 (equation 8.14)
deltaX0->putScalar(zero);
deltaX0->elementWiseMultiply(one, *D_, *R0, zero); // deltaX0 = D * R0 (equation 8.13)
A10_->apply(*deltaX0, *Rtmp); // Rtmp = A10*D*deltaX0 (intermediate step)
Rtmp->update(one, *R1, -one); // Rtmp = R1 - A10*D*deltaX0
if (!pL.get<bool>("q2q1 mode")) {
deltaX1->putScalar(zero);
} else {
ArrayRCP<SC> deltaX1data = deltaX1->getDataNonConst(0);
ArrayRCP<SC> Rtmpdata = Rtmp->getDataNonConst(0);
for (GO row = 0; row < deltaX1data.size(); row++)
deltaX1data[row] = 1.1*Rtmpdata[row] / Sdiag[row];
}
// Special handling if SchurComplement operator was a 1x1 blocked operator in Thyra mode
// Then, we have to translate the Xpetra offset GIDs to plain Thyra GIDs and vice versa
if(bA11ThyraSpecialTreatment == true) {
RCP<MultiVector> deltaX1_thyra = domainMapExtractor_->getVector(1, X.getNumVectors(), true);
RCP<MultiVector> Rtmp_thyra = rangeMapExtractor_->getVector(1, B.getNumVectors(), true);
// transform vector
for(size_t k=0; k < Rtmp->getNumVectors(); k++) {
Teuchos::ArrayRCP<const Scalar> xpetraVecData = Rtmp->getData(k);
Teuchos::ArrayRCP<Scalar> thyraVecData = Rtmp_thyra->getDataNonConst(k);
for(size_t i=0; i < Rtmp->getLocalLength(); i++) {
thyraVecData[i] = xpetraVecData[i];
}
}
smoo_->Apply(*deltaX1_thyra,*Rtmp_thyra);
for(size_t k=0; k < deltaX1_thyra->getNumVectors(); k++) {
Teuchos::ArrayRCP<Scalar> xpetraVecData = deltaX1->getDataNonConst(k);
Teuchos::ArrayRCP<const Scalar> thyraVecData = deltaX1_thyra->getData(k);
for(size_t i=0; i < deltaX1_thyra->getLocalLength(); i++) {
xpetraVecData[i] = thyraVecData[i];
}
}
} else {
// Compute deltaX1 (pressure correction)
// We use user provided preconditioner
smoo_->Apply(*deltaX1,*Rtmp);
}
// Compute deltaX0
deltaX0->putScalar(zero); // just for safety
A01_->apply(*deltaX1, *deltaX0); // deltaX0 = A01*deltaX1
deltaX0->update(one, *R0, -one); // deltaX0 = R0 - A01*deltaX1
R0.swap(deltaX0);
deltaX0->elementWiseMultiply(one, *D_, *R0, zero); // deltaX0 = D*(R0 - A01*deltaX1)
// Update solution
X0->update(one, *deltaX0, one);
X1->update(one, *deltaX1, one);
domainMapExtractor_->InsertVector(X0, 0, rcpX);
domainMapExtractor_->InsertVector(X1, 1, rcpX);
if (run < nSweeps-1)
R = Utilities::Residual(*A_, X, B);
}
}
