BelosThyraAdapter.hpp

Go to the documentation of this file.

// @HEADER
// ***********************************************************************
//
//         Stratimikos: Thyra-based strategies for linear solvers
//                Copyright (2006) Sandia Corporation
//
// Under terms of Contract DE-AC04-94AL85000, there is a non-exclusive
// license for use of this work by or on behalf of the U.S. Government.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// 1. Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// 2. Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// 3. Neither the name of the Corporation nor the names of the
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Questions? Contact Roscoe A. Bartlett (rabartl@sandia.gov)
//
// ***********************************************************************
// @HEADER

#ifndef BELOS_THYRA_ADAPTER_HPP
#define BELOS_THYRA_ADAPTER_HPP

#include "BelosConfigDefs.hpp"
#include "BelosMultiVecTraits.hpp"
#include "BelosOperatorTraits.hpp"

#include <Thyra_DetachedMultiVectorView.hpp>
#include <Thyra_MultiVectorBase.hpp>
#include <Thyra_MultiVectorStdOps.hpp>
#ifdef HAVE_BELOS_TSQR
#  include <Thyra_TsqrAdaptor.hpp>
#endif // HAVE_BELOS_TSQR

namespace Belos {

  //
  // Implementation of the Belos::MultiVecTraits for Thyra::MultiVectorBase
  //

  template<class ScalarType>
  class MultiVecTraits< ScalarType, Thyra::MultiVectorBase<ScalarType> >
  {
  private:
    typedef Thyra::MultiVectorBase<ScalarType> TMVB;
    typedef Teuchos::ScalarTraits<ScalarType> ST;
    typedef typename ST::magnitudeType magType;

  public:


    static Teuchos::RCP<TMVB> Clone( const TMVB& mv, const int numvecs )
    {
      Teuchos::RCP<TMVB> c = Thyra::createMembers( mv.range(), numvecs );
      return c;
    }

    static Teuchos::RCP<TMVB> CloneCopy( const TMVB& mv )
    {
      int numvecs = mv.domain()->dim();
      // create the new multivector
      Teuchos::RCP< TMVB > cc = Thyra::createMembers( mv.range(), numvecs );
      // copy the data from the source multivector to the new multivector
      Thyra::assign(cc.ptr(), mv);
      return cc;
    }

    static Teuchos::RCP<TMVB> CloneCopy( const TMVB& mv, const std::vector<int>& index )
    {
      int numvecs = index.size();
      // create the new multivector
      Teuchos::RCP<TMVB> cc = Thyra::createMembers( mv.range(), numvecs );
      // create a view to the relevant part of the source multivector
      Teuchos::RCP<const TMVB> view = mv.subView(index);
      // copy the data from the relevant view to the new multivector
      Thyra::assign(cc.ptr(), *view);
      return cc;
    }

    static Teuchos::RCP<TMVB>
    CloneCopy (const TMVB& mv, const Teuchos::Range1D& index)
    {
      const int numVecs = index.size();
      // Create the new multivector
      Teuchos::RCP<TMVB> cc = Thyra::createMembers (mv.range(), numVecs);
      // Create a view to the relevant part of the source multivector
      Teuchos::RCP<const TMVB> view = mv.subView (index);
      // Copy the data from the view to the new multivector.
      Thyra::assign (cc.ptr(), *view);
      return cc;
    }

    static Teuchos::RCP<TMVB> CloneViewNonConst( TMVB& mv, const std::vector<int>& index )
    {
      int numvecs = index.size();

      // We do not assume that the indices are sorted, nor do we check that
      // index.size() > 0. This code is fail-safe, in the sense that a zero
      // length index std::vector will pass the error on the Thyra.

      // Thyra has two ways to create an indexed View:
      // * contiguous (via a range of columns)
      // * indexed (via a std::vector of column indices)
      // The former is significantly more efficient than the latter, in terms of
      // computations performed with/against the created view.
      // We will therefore check to see if the given indices are contiguous, and
      // if so, we will use the contiguous view creation method.

      int lb = index[0];
      bool contig = true;
      for (int i=0; i<numvecs; i++) {
        if (lb+i != index[i]) contig = false;
      }

      Teuchos::RCP< TMVB > cc;
      if (contig) {
        const Thyra::Range1D rng(lb,lb+numvecs-1);
        // create a contiguous view to the relevant part of the source multivector
        cc = mv.subView(rng);
      }
      else {
        // create an indexed view to the relevant part of the source multivector
        cc = mv.subView(index);
      }
      return cc;
    }

    static Teuchos::RCP<TMVB>
    CloneViewNonConst (TMVB& mv, const Teuchos::Range1D& index)
    {
      // We let Thyra be responsible for checking that the index range
      // is nonempty.
      //
      // Create and return a contiguous view to the relevant part of
      // the source multivector.
      return mv.subView (index);
    }


    static Teuchos::RCP<const TMVB> CloneView( const TMVB& mv, const std::vector<int>& index )
    {
      int numvecs = index.size();

      // We do not assume that the indices are sorted, nor do we check that
      // index.size() > 0. This code is fail-safe, in the sense that a zero
      // length index std::vector will pass the error on the Thyra.

      // Thyra has two ways to create an indexed View:
      // * contiguous (via a range of columns)
      // * indexed (via a std::vector of column indices)
      // The former is significantly more efficient than the latter, in terms of
      // computations performed with/against the created view.
      // We will therefore check to see if the given indices are contiguous, and
      // if so, we will use the contiguous view creation method.

      int lb = index[0];
      bool contig = true;
      for (int i=0; i<numvecs; i++) {
        if (lb+i != index[i]) contig = false;
      }

      Teuchos::RCP< const TMVB > cc;
      if (contig) {
        const Thyra::Range1D rng(lb,lb+numvecs-1);
        // create a contiguous view to the relevant part of the source multivector
        cc = mv.subView(rng);
      }
      else {
        // create an indexed view to the relevant part of the source multivector
        cc = mv.subView(index);
      }
      return cc;
    }

    static Teuchos::RCP<const TMVB>
    CloneView (const TMVB& mv, const Teuchos::Range1D& index)
    {
      // We let Thyra be responsible for checking that the index range
      // is nonempty.
      //
      // Create and return a contiguous view to the relevant part of
      // the source multivector.
      return mv.subView (index);
    }



    static ptrdiff_t GetGlobalLength( const TMVB& mv ) {
      return Teuchos::as<ptrdiff_t>(mv.range()->dim());
    }

    static int GetNumberVecs( const TMVB& mv )
    { return mv.domain()->dim(); }



    static void MvTimesMatAddMv( const ScalarType alpha, const TMVB& A,
         const Teuchos::SerialDenseMatrix<int,ScalarType>& B,
         const ScalarType beta, TMVB& mv )
    {
      using Teuchos::arrayView; using Teuchos::arcpFromArrayView;
      const int m = B.numRows();
      const int n = B.numCols();
      // Create a view of the B object!
      Teuchos::RCP< const TMVB >
        B_thyra = Thyra::createMembersView(
          A.domain(),
          RTOpPack::ConstSubMultiVectorView<ScalarType>(
            0, m, 0, n,
            arcpFromArrayView(arrayView(&B(0,0), B.stride()*B.numCols())), B.stride()
            )
          );
      // perform the operation via A: mv <- alpha*A*B_thyra + beta*mv
      Thyra::apply<ScalarType>(A, Thyra::NOTRANS, *B_thyra, Teuchos::outArg(mv), alpha, beta);
    }

    static void MvAddMv( const ScalarType alpha, const TMVB& A,
                         const ScalarType beta,  const TMVB& B, TMVB& mv )
    {
      typedef Teuchos::ScalarTraits<ScalarType> ST;
      using Teuchos::tuple; using Teuchos::ptrInArg; using Teuchos::inoutArg;

      Thyra::linear_combination<ScalarType>(
        tuple(alpha, beta)(), tuple(ptrInArg(A), ptrInArg(B))(), ST::zero(), inoutArg(mv));
    }

    static void MvScale ( TMVB& mv, const ScalarType alpha )
      { Thyra::scale(alpha, Teuchos::inoutArg(mv)); }

    static void MvScale (TMVB& mv, const std::vector<ScalarType>& alpha)
    {
      for (unsigned int i=0; i<alpha.size(); i++) {
        Thyra::scale<ScalarType> (alpha[i], mv.col(i).ptr());
      }
    }

    static void MvTransMv( const ScalarType alpha, const TMVB& A, const TMVB& mv,
      Teuchos::SerialDenseMatrix<int,ScalarType>& B )
    {
      using Teuchos::arrayView; using Teuchos::arcpFromArrayView;
      // Create a multivector to hold the result (m by n)
      int m = A.domain()->dim();
      int n = mv.domain()->dim();
      // Create a view of the B object!
      Teuchos::RCP< TMVB >
        B_thyra = Thyra::createMembersView(
          A.domain(),
          RTOpPack::SubMultiVectorView<ScalarType>(
            0, m, 0, n,
            arcpFromArrayView(arrayView(&B(0,0), B.stride()*B.numCols())), B.stride()
            )
          );
      Thyra::apply<ScalarType>(A, Thyra::CONJTRANS, mv, B_thyra.ptr(), alpha);
    }

    static void MvDot( const TMVB& mv, const TMVB& A, std::vector<ScalarType>& b )
      { Thyra::dots(mv, A, Teuchos::arrayViewFromVector(b)); }



    static void MvNorm( const TMVB& mv, std::vector<magType>& normvec,
      NormType type = TwoNorm ) {
      if(type == TwoNorm)
        Thyra::norms_2(mv, Teuchos::arrayViewFromVector(normvec));
      else if(type == OneNorm)
        Thyra::norms_1(mv, Teuchos::arrayViewFromVector(normvec));
      else if(type == InfNorm)
        Thyra::norms_inf(mv, Teuchos::arrayViewFromVector(normvec));
      else
        TEUCHOS_TEST_FOR_EXCEPTION(true, std::invalid_argument,
                           "Belos::MultiVecTraits::MvNorm (Thyra specialization): "
                           "invalid norm type. Must be either TwoNorm, OneNorm or InfNorm");
    }



    static void SetBlock( const TMVB& A, const std::vector<int>& index, TMVB& mv )
    {
      // Extract the "numvecs" columns of mv indicated by the index std::vector.
      int numvecs = index.size();
      std::vector<int> indexA(numvecs);
      int numAcols = A.domain()->dim();
      for (int i=0; i<numvecs; i++) {
        indexA[i] = i;
      }
      // Thyra::assign requires that both arguments have the same number of
      // vectors. Enforce this, by shrinking one to match the other.
      if ( numAcols < numvecs ) {
        // A does not have enough columns to satisfy index_plus. Shrink
        // index_plus.
        numvecs = numAcols;
      }
      else if ( numAcols > numvecs ) {
        numAcols = numvecs;
        indexA.resize( numAcols );
      }
      // create a view to the relevant part of the source multivector
      Teuchos::RCP< const TMVB > relsource = A.subView(indexA);
      // create a view to the relevant part of the destination multivector
      Teuchos::RCP< TMVB > reldest = mv.subView(index);
      // copy the data to the destination multivector subview
      Thyra::assign(reldest.ptr(), *relsource);
    }

    static void
    SetBlock (const TMVB& A, const Teuchos::Range1D& index, TMVB& mv)
    {
      const int numColsA = A.domain()->dim();
      const int numColsMv = mv.domain()->dim();
      // 'index' indexes into mv; it's the index set of the target.
      const bool validIndex = index.lbound() >= 0 && index.ubound() < numColsMv;
      // We can't take more columns out of A than A has.
      const bool validSource = index.size() <= numColsA;

      if (! validIndex || ! validSource)
        {
          std::ostringstream os;
          os << "Belos::MultiVecTraits<Scalar, Thyra::MultiVectorBase<Scalar> "
            ">::SetBlock(A, [" << index.lbound() << ", " << index.ubound()
             << "], mv): ";
          TEUCHOS_TEST_FOR_EXCEPTION(index.lbound() < 0, std::invalid_argument,
                             os.str() << "Range lower bound must be nonnegative.");
          TEUCHOS_TEST_FOR_EXCEPTION(index.ubound() >= numColsMv, std::invalid_argument,
                             os.str() << "Range upper bound must be less than "
                             "the number of columns " << numColsA << " in the "
                             "'mv' output argument.");
          TEUCHOS_TEST_FOR_EXCEPTION(index.size() > numColsA, std::invalid_argument,
                             os.str() << "Range must have no more elements than"
                             " the number of columns " << numColsA << " in the "
                             "'A' input argument.");
          TEUCHOS_TEST_FOR_EXCEPTION(true, std::logic_error, "Should never get here!");
        }

      // View of the relevant column(s) of the target multivector mv.
      // We avoid view creation overhead by only creating a view if
      // the index range is different than [0, (# columns in mv) - 1].
      Teuchos::RCP<TMVB> mv_view;
      if (index.lbound() == 0 && index.ubound()+1 == numColsMv)
        mv_view = Teuchos::rcpFromRef (mv); // Non-const, non-owning RCP
      else
        mv_view = mv.subView (index);

      // View of the relevant column(s) of the source multivector A.
      // If A has fewer columns than mv_view, then create a view of
      // the first index.size() columns of A.
      Teuchos::RCP<const TMVB> A_view;
      if (index.size() == numColsA)
        A_view = Teuchos::rcpFromRef (A); // Const, non-owning RCP
      else
        A_view = A.subView (Teuchos::Range1D(0, index.size()-1));

      // Copy the data to the destination multivector.
      Thyra::assign(mv_view.ptr(), *A_view);
    }

    static void
    Assign (const TMVB& A, TMVB& mv)
    {
      const int numColsA = A.domain()->dim();
      const int numColsMv = mv.domain()->dim();
      if (numColsA > numColsMv)
        {
          std::ostringstream os;
          os << "Belos::MultiVecTraits<Scalar, Thyra::MultiVectorBase<Scalar>"
            " >::Assign(A, mv): ";
          TEUCHOS_TEST_FOR_EXCEPTION(numColsA > numColsMv, std::invalid_argument,
                             os.str() << "Input multivector 'A' has "
                             << numColsA << " columns, but output multivector "
                             "'mv' has only " << numColsMv << " columns.");
          TEUCHOS_TEST_FOR_EXCEPTION(true, std::logic_error, "Should never get here!");
        }
      // Copy the data to the destination multivector.
      if (numColsA == numColsMv) {
        Thyra::assign (Teuchos::outArg (mv), A);
      } else {
        Teuchos::RCP<TMVB> mv_view =
          CloneViewNonConst (mv, Teuchos::Range1D(0, numColsA-1));
        Thyra::assign (mv_view.ptr(), A);
      }
    }

    static void MvRandom( TMVB& mv )
    {
      // Thyra::randomize generates via a uniform distribution on [l,u]
      // We will use this to generate on [-1,1]
      Thyra::randomize<ScalarType>(
        -Teuchos::ScalarTraits<ScalarType>::one(),
        Teuchos::ScalarTraits<ScalarType>::one(),
        Teuchos::outArg(mv));
    }

    static void
    MvInit (TMVB& mv, ScalarType alpha = Teuchos::ScalarTraits<ScalarType>::zero())
    {
      Thyra::assign (Teuchos::outArg (mv), alpha);
    }



    static void MvPrint( const TMVB& mv, std::ostream& os )
      { os << describe(mv,Teuchos::VERB_EXTREME); }


#ifdef HAVE_BELOS_TSQR
    typedef Thyra::TsqrAdaptor< ScalarType > tsqr_adaptor_type;
#endif // HAVE_BELOS_TSQR
  };

  //
  // Implementation of the Belos::OperatorTraits for Thyra::LinearOpBase
  //

  template<class ScalarType>
  class OperatorTraits <ScalarType,
                        Thyra::MultiVectorBase<ScalarType>,
                        Thyra::LinearOpBase<ScalarType> >
  {
  private:
    typedef Thyra::MultiVectorBase<ScalarType> TMVB;
    typedef Thyra::LinearOpBase<ScalarType>    TLOB;

  public:
    static void
    Apply (const TLOB& Op,
           const TMVB& x,
           TMVB& y,
           ETrans trans = NOTRANS)
    {
      Thyra::EOpTransp whichOp;

      // We don't check here whether the operator implements the
      // requested operation.  Call HasApplyTranspose() to check.
      // Thyra::LinearOpBase implementations are not required to
      // implement NOTRANS.  However, Belos needs NOTRANS
      // (obviously!), so we assume that Op implements NOTRANS.
      if (trans == NOTRANS)
        whichOp = Thyra::NOTRANS;
      else if (trans == TRANS)
        whichOp = Thyra::TRANS;
      else if (trans == CONJTRANS)
        whichOp = Thyra::CONJTRANS;
      else
        TEUCHOS_TEST_FOR_EXCEPTION(true, std::invalid_argument,
                           "Belos::OperatorTraits::Apply (Thyra specialization): "
                           "'trans' argument must be neither NOTRANS=" << NOTRANS
                           << ", TRANS=" << TRANS << ", or CONJTRANS=" << CONJTRANS
                           << ", but instead has an invalid value of " << trans << ".");
      Thyra::apply<ScalarType>(Op, whichOp, x, Teuchos::outArg(y));
    }

    static bool HasApplyTranspose (const TLOB& Op)
    {
      typedef Teuchos::ScalarTraits<ScalarType> STS;

      // Thyra::LinearOpBase's interface lets you check whether the
      // operator implements any of all four possible combinations of
      // conjugation and transpose.  Belos only needs transpose
      // (TRANS) if the operator is real; in that case, Apply() does
      // the same thing with trans = CONJTRANS or TRANS.  If the
      // operator is complex, Belos needs both transpose and conjugate
      // transpose (CONJTRANS) if the operator is complex.
      return Op.opSupported (Thyra::TRANS) &&
        (! STS::isComplex || Op.opSupported (Thyra::CONJTRANS));
    }
  };

} // end of Belos namespace

#endif
// end of file BELOS_THYRA_ADAPTER_HPP