Thyra_DefaultBlockedLinearOp_def.hpp

// @HEADER
// ***********************************************************************
//
//    Thyra: Interfaces and Support for Abstract Numerical Algorithms
//                 Copyright (2004) 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 (bartlettra@ornl.gov)
//
// ***********************************************************************
// @HEADER


#ifndef THYRA_DEFAULT_BLOCKED_LINEAR_OP_DEF_HPP
#define THYRA_DEFAULT_BLOCKED_LINEAR_OP_DEF_HPP


#include "Thyra_DefaultBlockedLinearOp_decl.hpp"
#include "Thyra_DefaultProductVectorSpace.hpp"
#include "Thyra_DefaultProductVector.hpp"
#include "Thyra_DefaultProductMultiVector.hpp"
#include "Thyra_MultiVectorStdOps.hpp"
#include "Thyra_VectorStdOps.hpp"
#include "Thyra_AssertOp.hpp"
#include "Thyra_ScaledAdjointLinearOpBase.hpp"


namespace Thyra {


// Constructors


template<class Scalar>
DefaultBlockedLinearOp<Scalar>::DefaultBlockedLinearOp()
  :numRowBlocks_(0), numColBlocks_(0), blockFillIsActive_(false)
{}


// Overridden from PhysicallyBlockedLinearOpBase


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::beginBlockFill()
{
  assertBlockFillIsActive(false);
  uninitialize();
  resetStorage(0,0);
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::beginBlockFill(
  const int numRowBlocks, const int numColBlocks
  )
{
  assertBlockFillIsActive(false);
  uninitialize();
  resetStorage(numRowBlocks,numColBlocks);
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::beginBlockFill(
  const RCP<const ProductVectorSpaceBase<Scalar> > &new_productRange
  ,const RCP<const ProductVectorSpaceBase<Scalar> > &new_productDomain
  )
{
  using Teuchos::rcp_dynamic_cast;
  assertBlockFillIsActive(false);
  uninitialize();
  productRange_ = new_productRange.assert_not_null();
  productDomain_ = new_productDomain.assert_not_null();
  defaultProductRange_ =
    rcp_dynamic_cast<const DefaultProductVectorSpace<Scalar> >(productRange_);
  defaultProductDomain_ =
    rcp_dynamic_cast<const DefaultProductVectorSpace<Scalar> >(productDomain_);
  // Note: the above spaces must be set before calling the next function!
  resetStorage(productRange_->numBlocks(), productDomain_->numBlocks());
}


template<class Scalar>
bool DefaultBlockedLinearOp<Scalar>::blockFillIsActive() const
{
  return blockFillIsActive_;
}


template<class Scalar>
bool DefaultBlockedLinearOp<Scalar>::acceptsBlock(
  const int i, const int j
  ) const
{
  assertBlockFillIsActive(true);
  assertBlockRowCol(i,j);
  return true;
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::setNonconstBlock(
  const int i, const int j
  ,const RCP<LinearOpBase<Scalar> > &block
  )
{
  setBlockImpl(i, j, block);
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::setBlock(
  const int i, const int j
  ,const RCP<const LinearOpBase<Scalar> > &block
  )
{
  setBlockImpl(i, j, block);
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::endBlockFill()
{

  using Teuchos::as;

  assertBlockFillIsActive(true);

  // 2009/05/06: rabartl: ToDo: When doing a flexible block fill
  // (Ops_stack_.size() > 0), we need to assert that all of the block rows and
  // columns have been filled in.  I don't think we do that here.

  // Get the number of block rows and columns
  if (nonnull(productRange_)) {
    numRowBlocks_ = productRange_->numBlocks();
    numColBlocks_ = productDomain_->numBlocks();
  }
  else {
    numRowBlocks_ = rangeBlocks_.size();
    numColBlocks_ = domainBlocks_.size();
    // NOTE: Above, whether doing a flexible fill or not, all of the blocks
    // must be set in order to have a valid filled operator so this
    // calculation should be correct.
  }

  // Assert that all of the block rows and columns have at least one entry if
  // the spaces were not given up front.
#ifdef TEUCHOS_DEBUG
  if (is_null(productRange_)) {
    for (int i = 0; i < numRowBlocks_; ++i) {
      TEUCHOS_TEST_FOR_EXCEPTION(
        !rangeBlocks_[i].get(), std::logic_error
        ,"DefaultBlockedLinearOp<Scalar>::endBlockFill():"
        " Error, no linear operator block for the i="<<i<<" block row was added"
        " and we can not complete the block fill!"
        );
    }
    for(int j = 0; j < numColBlocks_; ++j) {
      TEUCHOS_TEST_FOR_EXCEPTION(
        !domainBlocks_[j].get(), std::logic_error
        ,"DefaultBlockedLinearOp<Scalar>::endBlockFill():"
        " Error, no linear operator block for the j="
        <<j<<" block column was added"
        " and we can not complete the block fill!"
        );
    }
  }
#endif

  // Insert the block LOB objects if doing a flexible fill.
  if (Ops_stack_.size()) {
    Ops_.resize(numRowBlocks_*numColBlocks_);
    for ( int k = 0; k < as<int>(Ops_stack_.size()); ++k ) {
      const BlockEntry<Scalar> &block_i_j = Ops_stack_[k];
      Ops_[numRowBlocks_*block_i_j.j + block_i_j.i] = block_i_j.block;
    }
    Ops_stack_.resize(0);
  }

  // Set the product range and domain spaces if not already set
  if (is_null(productRange_)) {
    adjustBlockSpaces();
    defaultProductRange_ = productVectorSpace<Scalar>(rangeBlocks_());
    defaultProductDomain_ = productVectorSpace<Scalar>(domainBlocks_());
    productRange_ = defaultProductRange_;
    productDomain_ = defaultProductDomain_;
  }

  rangeBlocks_.resize(0);
  domainBlocks_.resize(0);

  blockFillIsActive_ = false;

}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::uninitialize()
{
  productRange_ = Teuchos::null;
  productDomain_ = Teuchos::null;
  numRowBlocks_ = 0;
  numColBlocks_ = 0;
  Ops_.resize(0);
  Ops_stack_.resize(0);
  rangeBlocks_.resize(0);
  domainBlocks_.resize(0);
  blockFillIsActive_ = false;
}


// Overridden from BlockedLinearOpBase


template<class Scalar>
RCP<const ProductVectorSpaceBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::productRange() const
{
  return productRange_;
}


template<class Scalar>
RCP<const ProductVectorSpaceBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::productDomain() const
{
  return productDomain_;
}


template<class Scalar>
bool DefaultBlockedLinearOp<Scalar>::blockExists(
  const int i, const int j
  ) const
{
  assertBlockFillIsActive(false);
  assertBlockRowCol(i,j);
  return true;
}


template<class Scalar>
bool DefaultBlockedLinearOp<Scalar>::blockIsConst(
  const int i, const int j
  ) const
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPT(!blockExists(i,j));
#endif
  assertBlockFillIsActive(false);
  assertBlockRowCol(i,j);
  return Ops_[numRowBlocks_*j+i].isConst();
}


template<class Scalar>
RCP<LinearOpBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::getNonconstBlock(const int i, const int j)
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPT(!blockExists(i,j));
#endif
  assertBlockFillIsActive(false);
  assertBlockRowCol(i,j);
  return Ops_[numRowBlocks_*j+i].getNonconstObj();
}


template<class Scalar>
RCP<const LinearOpBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::getBlock(const int i, const int j) const
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPT(!blockExists(i,j));
#endif
  assertBlockFillIsActive(false);
  assertBlockRowCol(i,j);
  return Ops_[numRowBlocks_*j+i];
}


// Overridden from LinearOpBase


template<class Scalar>
RCP< const VectorSpaceBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::range() const
{
  return productRange_;
}


template<class Scalar>
RCP< const VectorSpaceBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::domain() const
{
  return productDomain_;
}


template<class Scalar>
RCP<const LinearOpBase<Scalar> >
DefaultBlockedLinearOp<Scalar>::clone() const
{
  return Teuchos::null; // ToDo: Implement this when needed!
}


// Overridden from Teuchos::Describable


template<class Scalar>
std::string DefaultBlockedLinearOp<Scalar>::description() const
{
  assertBlockFillIsActive(false);
  std::ostringstream oss;
  oss
    << Teuchos::Describable::description() << "{"
    << "numRowBlocks="<<numRowBlocks_
    << ",numColBlocks="<<numColBlocks_
    << "}";
  return oss.str();
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::describe(
  Teuchos::FancyOStream &out_arg
  ,const Teuchos::EVerbosityLevel verbLevel
  ) const
{
  using Teuchos::rcpFromRef;
  using Teuchos::FancyOStream;
  using Teuchos::OSTab;
  assertBlockFillIsActive(false);
  RCP<FancyOStream> out = rcpFromRef(out_arg);
  OSTab tab1(out);
  switch(verbLevel) {
    case Teuchos::VERB_DEFAULT:
    case Teuchos::VERB_LOW:
      *out << this->description() << std::endl;
      break;
    case Teuchos::VERB_MEDIUM:
    case Teuchos::VERB_HIGH:
    case Teuchos::VERB_EXTREME:
    {
      *out
        << Teuchos::Describable::description() << "{"
        << "rangeDim=" << this->range()->dim()
        << ",domainDim=" << this->domain()->dim()
        << ",numRowBlocks=" << numRowBlocks_
        << ",numColBlocks=" << numColBlocks_
        << "}\n";
      OSTab tab2(out);
      *out
        << "Constituent LinearOpBase objects for M = [ Op[0,0] ..."
        << " ; ... ; ... Op[numRowBlocks-1,numColBlocks-1] ]:\n";
      tab2.incrTab();
      for( int i = 0; i < numRowBlocks_; ++i ) {
        for( int j = 0; j < numColBlocks_; ++j ) {
          *out << "Op["<<i<<","<<j<<"] = ";
          RCP<const LinearOpBase<Scalar> >
            block_i_j = getBlock(i,j);
          if(block_i_j.get())
            *out << Teuchos::describe(*getBlock(i,j),verbLevel);
          else
            *out << "NULL\n";
        }
      }
      break;
    }
    default:
      TEUCHOS_TEST_FOR_EXCEPT(true); // Should never get here!
  }
}


// protected


// Overridden from LinearOpBase


template<class Scalar>
bool DefaultBlockedLinearOp<Scalar>::opSupportedImpl(EOpTransp M_trans) const
{
  bool supported = true;
  for( int i = 0; i < numRowBlocks_; ++i ) {
    for( int j = 0; j < numColBlocks_; ++j ) {
      RCP<const LinearOpBase<Scalar> >
        block_i_j = getBlock(i,j);
      if( block_i_j.get() && !Thyra::opSupported(*block_i_j,M_trans) )
        supported = false;
    }
  }
  return supported;
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::applyImpl(
  const EOpTransp M_trans,
  const MultiVectorBase<Scalar> &X_in,
  const Ptr<MultiVectorBase<Scalar> > &Y_inout,
  const Scalar alpha,
  const Scalar beta
  ) const
{

  using Teuchos::rcpFromRef;
  typedef Teuchos::ScalarTraits<Scalar> ST;
  typedef RCP<MultiVectorBase<Scalar> > MultiVectorPtr;
  typedef RCP<const MultiVectorBase<Scalar> > ConstMultiVectorPtr;
  typedef RCP<const LinearOpBase<Scalar> > ConstLinearOpPtr;

#ifdef TEUCHOS_DEBUG
  THYRA_ASSERT_LINEAR_OP_MULTIVEC_APPLY_SPACES(
    "DefaultBlockedLinearOp<Scalar>::apply(...)", *this, M_trans, X_in, &*Y_inout
    );
#endif // TEUCHOS_DEBUG

  const bool
    struct_transp = (real_trans(M_trans)!=NOTRANS); // Structural transpose?
  const int
    opNumRowBlocks = ( !struct_transp ? numRowBlocks_ : numColBlocks_ ),
    opNumColBlocks = ( !struct_transp ? numColBlocks_ : numRowBlocks_ );

  //
  // Y = alpha * op(M) * X + beta*Y
  //
  // =>
  //
  // Y[i] = beta+Y[i] + sum(alpha*op(Op)[i,j]*X[j],j=0...opNumColBlocks-1)
  //
  // , for i=0...opNumRowBlocks-1
  //

  const RCP<const DefaultProductVectorSpace<Scalar> >
    defaultProductRange_op = ( real_trans(M_trans)==NOTRANS
      ? defaultProductRange_ : defaultProductDomain_ ),
    defaultProductDomain_op = ( real_trans(M_trans)==NOTRANS
      ? defaultProductDomain_ : defaultProductRange_ );

  const RCP<const ProductMultiVectorBase<Scalar> >
    X = castOrCreateSingleBlockProductMultiVector<Scalar>(
      defaultProductDomain_op, rcpFromRef(X_in));
  const RCP<ProductMultiVectorBase<Scalar> >
    Y = nonconstCastOrCreateSingleBlockProductMultiVector<Scalar>(
      defaultProductRange_op, rcpFromPtr(Y_inout));

  for( int i = 0; i < opNumRowBlocks; ++i ) {
    MultiVectorPtr Y_i = Y->getNonconstMultiVectorBlock(i);
    for( int j = 0; j < opNumColBlocks; ++j ) {
      ConstLinearOpPtr
        Op_i_j = ( !struct_transp ? getBlock(i,j) : getBlock(j,i) );
      ConstMultiVectorPtr
        X_j = X->getMultiVectorBlock(j);
      if(j==0) {
        if (nonnull(Op_i_j))
          Thyra::apply(*Op_i_j, M_trans,* X_j, Y_i.ptr(), alpha, beta);
        else
          scale(beta, Y_i.ptr());
      }
      else {
        if (nonnull(Op_i_j))
          Thyra::apply(*Op_i_j, M_trans, *X_j, Y_i.ptr(), alpha, ST::one());
      }
    }
  }
}

// Overridden from RowStatLinearOpBase

template<class Scalar>
bool
DefaultBlockedLinearOp<Scalar>::rowStatIsSupportedImpl(
    const RowStatLinearOpBaseUtils::ERowStat row_stat) const
{
  using Teuchos::rcpFromRef;
  using Teuchos::rcp_dynamic_cast;
  using Teuchos::dyn_cast;
  //typedef Teuchos::ScalarTraits<Scalar> ST; // unused
  typedef RowStatLinearOpBase<Scalar> RowStatOp;
  typedef RCP<const LinearOpBase<Scalar> > ConstLinearOpPtr;
  typedef const LinearOpBase<Scalar> ConstLinearOp;

  // Figure out what needs to be check for each sub block to compute
  // the require row stat
  RowStatLinearOpBaseUtils::ERowStat
    subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM,
    subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
  switch (row_stat) {
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_ROW_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM:
      subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM;
      subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
      break;
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_COL_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM:
      subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
      subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM;
      break;
    default:
      TEUCHOS_TEST_FOR_EXCEPT(true);
  }

  // check each sub block for compatibility (null means zero,
  // automatically compatible)
  for( int i = 0; i < numRowBlocks_; ++i ) {
    for( int j = 0; j < numColBlocks_; ++j ) {
      ConstLinearOpPtr Op_i_j = getBlock(i,j);

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        ConstLinearOp * Orig_i_j = 0;
        Scalar scalar = 1.0;
        EOpTransp transp = NOTRANS;
        Thyra::unwrap(*Op_i_j,&scalar,&transp,&Orig_i_j);

        const RowStatOp & row_stat_op = Teuchos::dyn_cast<const RowStatOp>(*Orig_i_j);

        // sub block must also support the required row stat operation
        RowStatLinearOpBaseUtils::ERowStat stat = subblk_stat;
        if(transp==NOTRANS || transp==CONJ)
          stat = subblk_stat;
        else if(transp==TRANS || transp==CONJTRANS)
          stat = subblk_trans_stat;

        if(!row_stat_op.rowStatIsSupported(stat))
          return false;
      }
      // else: sub block is null, "0" is good enough
    }
  }

  return true;
}

template<class Scalar>
void
DefaultBlockedLinearOp<Scalar>::getRowStatImpl(
    const RowStatLinearOpBaseUtils::ERowStat row_stat,
    const Teuchos::Ptr<VectorBase< Scalar> > &rowStatVec) const
{
  using Teuchos::rcpFromRef;
  using Teuchos::rcpFromPtr;
  using Teuchos::rcp_dynamic_cast;
  using Teuchos::dyn_cast;
  typedef Teuchos::ScalarTraits<Scalar> ST;
  typedef RowStatLinearOpBase<Scalar> RowStatOp;
  typedef RCP<const LinearOpBase<Scalar> > ConstLinearOpPtr;
  typedef const LinearOpBase<Scalar> ConstLinearOp;
  typedef RCP<VectorBase<Scalar> > VectorPtr;

  // Figure out what needs to be check for each sub block to compute
  // the require row stat
  RowStatLinearOpBaseUtils::ERowStat
    subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM,
    subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
  switch (row_stat) {
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_ROW_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM:
      subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM;
      subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
      break;
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_COL_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM:
      subblk_stat = RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM;
      subblk_trans_stat = RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM;
      break;
    default:
      TEUCHOS_TEST_FOR_EXCEPT(true);
  }

  const RCP<ProductVectorBase<Scalar> >
    Y = rcp_dynamic_cast<ProductVectorBase<Scalar> >(rcpFromPtr(rowStatVec));

  // using sub block row stat capability interrogate each for
  // its constribution
  for( int i = 0; i < numRowBlocks_; ++i ) {
    VectorPtr blk_vec = Y->getNonconstVectorBlock(i);
    put_scalar (ST::zero (), blk_vec.ptr ()); // clear vector just in case

    for( int j = 0; j < numColBlocks_; ++j ) {
      ConstLinearOpPtr Op_i_j = getBlock(i,j);

      VectorPtr tmp_vec = createMember(Op_i_j->range());

      put_scalar (ST::zero (), tmp_vec.ptr ()); // clear vector just in case

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        ConstLinearOp * Orig_i_j = 0;
        Scalar scalar = 1.0;
        EOpTransp transp = NOTRANS;
        Thyra::unwrap(*Op_i_j,&scalar,&transp,&Orig_i_j);

        const RowStatOp & row_stat_op = Teuchos::dyn_cast<const RowStatOp>(*Orig_i_j);

        // sub block must also support the required row stat operation
        if(transp==NOTRANS || transp==CONJ)
          row_stat_op.getRowStat(subblk_stat,tmp_vec.ptr());
        else if(transp==TRANS || transp==CONJTRANS)
          row_stat_op.getRowStat(subblk_trans_stat,tmp_vec.ptr());
        else
        { TEUCHOS_ASSERT(false); }

        // some the temporary into the block vector
        Vp_V(blk_vec.ptr(),*tmp_vec);
      }
    }
  }

  // do post processing as needed (take the inverse currently
  switch (row_stat) {
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_ROW_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_INV_COL_SUM:
      reciprocal(*rowStatVec,rowStatVec.ptr());
    case RowStatLinearOpBaseUtils::ROW_STAT_ROW_SUM:
    case RowStatLinearOpBaseUtils::ROW_STAT_COL_SUM:
      break;
    default:
      TEUCHOS_TEST_FOR_EXCEPT(true);
  }
}

// Overridden from ScaledLinearOpBase

template<class Scalar>
bool
DefaultBlockedLinearOp<Scalar>::supportsScaleLeftImpl() const
{
  using Teuchos::rcp_dynamic_cast;
  using Teuchos::dyn_cast;
  typedef RCP<const LinearOpBase<Scalar> > LinearOpPtr;
  typedef const LinearOpBase<Scalar> LinearOp;
  typedef const ScaledLinearOpBase<Scalar> ScaledOp;

  bool supported = true;
  for( int i = 0; i < numRowBlocks_; ++i ) {
    for( int j = 0; j < numColBlocks_; ++j ) {
      LinearOpPtr Op_i_j = getBlock(i,j);

      EOpTransp transp = NOTRANS;

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        LinearOp * Orig_i_j = 0;
        Scalar scalar = 1.0;
        Thyra::unwrap(*Op_i_j,&scalar,&transp,&Orig_i_j);

        ScaledOp & scaled_op = Teuchos::dyn_cast<ScaledOp>(*Orig_i_j);

        if(transp==NOTRANS || transp==CONJ)
          supported &= scaled_op.supportsScaleLeft();
        else if(transp==TRANS || transp==CONJTRANS)
          supported &= scaled_op.supportsScaleRight();
      }
    }
  }

  return supported;
}

template<class Scalar>
bool
DefaultBlockedLinearOp<Scalar>::supportsScaleRightImpl() const
{
  using Teuchos::rcp_dynamic_cast;
  using Teuchos::dyn_cast;
  typedef RCP<const LinearOpBase<Scalar> > LinearOpPtr;
  typedef const LinearOpBase<Scalar> LinearOp;
  typedef const ScaledLinearOpBase<Scalar> ScaledOp;

  bool supported = true;
  for( int i = 0; i < numRowBlocks_; ++i ) {
    for( int j = 0; j < numColBlocks_; ++j ) {
      LinearOpPtr Op_i_j = getBlock(i,j);

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        LinearOp * Orig_i_j = 0;
        Scalar scalar = 1.0;
        EOpTransp transp = NOTRANS;
        Thyra::unwrap(*Op_i_j,&scalar,&transp,&Orig_i_j);

        ScaledOp & scaled_op = Teuchos::dyn_cast<ScaledOp>(*Orig_i_j);

        if(transp==NOTRANS || transp==CONJ)
          supported &= scaled_op.supportsScaleRight();
        else if(transp==TRANS || transp==CONJTRANS)
          supported &= scaled_op.supportsScaleLeft();
      }
    }
  }

  return supported;
}

template<class Scalar>
void
DefaultBlockedLinearOp<Scalar>::scaleLeftImpl(
  const VectorBase< Scalar > &row_scaling)
{
  using Teuchos::dyn_cast;
  using Teuchos::rcp_dynamic_cast;
  typedef ScaledLinearOpBase<Scalar> ScaledOp;
  typedef RCP<LinearOpBase<Scalar> > LinearOpPtr;
  typedef RCP<const VectorBase<Scalar> > VectorPtr;

  const ProductVectorBase<Scalar> &
    Y = dyn_cast<const ProductVectorBase<Scalar> >(row_scaling);

  // using sub block row stat capability interrogate each for
  // its constribution
  for( int i = 0; i < numRowBlocks_; ++i ) {
    VectorPtr blk_vec = Y.getVectorBlock(i);

    for( int j = 0; j < numColBlocks_; ++j ) {
      LinearOpPtr Op_i_j = getNonconstBlock(i,j);

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        LinearOpPtr Orig_i_j;
        EOpTransp transp = NOTRANS;

        {
          RCP<ScaledAdjointLinearOpBase<Scalar> > saOp
              = rcp_dynamic_cast<ScaledAdjointLinearOpBase<Scalar> >(Op_i_j);
          if(nonnull(saOp)) {
            transp = saOp->overallTransp();
            Orig_i_j = saOp->getNonconstOrigOp();
          }
          else
            Orig_i_j = Op_i_j; // stick with the original
        }

        RCP<ScaledOp> scaled_op = rcp_dynamic_cast<ScaledOp>(Orig_i_j);

        // sub block must support a row stat operation
        TEUCHOS_ASSERT(scaled_op!=Teuchos::null); // guranteed by compatibility check

        // sub block must also support the required row stat operation
        if(transp==NOTRANS || transp==CONJ)
          scaled_op->scaleLeft(*blk_vec);
        else if(transp==TRANS || transp==CONJTRANS)
          scaled_op->scaleRight(*blk_vec);
      }
    }
  }
}

template<class Scalar>
void
DefaultBlockedLinearOp<Scalar>::scaleRightImpl(
  const VectorBase< Scalar > &col_scaling)
{
  using Teuchos::dyn_cast;
  using Teuchos::rcp_dynamic_cast;
  typedef ScaledLinearOpBase<Scalar> ScaledOp;
  typedef RCP<LinearOpBase<Scalar> > LinearOpPtr;
  typedef RCP<const VectorBase<Scalar> > VectorPtr;

  const ProductVectorBase<Scalar> &
    Y = dyn_cast<const ProductVectorBase<Scalar> >(col_scaling);

  // using sub block row stat capability interrogate each for
  // its constribution
  for( int j = 0; j < numColBlocks_; ++j ) {
    VectorPtr blk_vec = Y.getVectorBlock(j);

    for( int i = 0; i < numRowBlocks_; ++i ) {
      LinearOpPtr Op_i_j = getNonconstBlock(i,j);

      if (nonnull(Op_i_j)) {
        // pull out adjoint, and scaling
        LinearOpPtr Orig_i_j;
        EOpTransp transp = NOTRANS;

        {
          RCP<ScaledAdjointLinearOpBase<Scalar> > saOp
              = rcp_dynamic_cast<ScaledAdjointLinearOpBase<Scalar> >(Op_i_j);
          if(nonnull(saOp)) {
            transp = saOp->overallTransp();
            Orig_i_j = saOp->getNonconstOrigOp();
          }
          else
            Orig_i_j = Op_i_j; // stick with the original
        }

        RCP<ScaledOp> scaled_op = rcp_dynamic_cast<ScaledOp>(Orig_i_j);

        // sub block must support a row stat operation
        TEUCHOS_ASSERT(scaled_op!=Teuchos::null); // guranteed by compatibility check

        // sub block must also support the required row stat operation
        if(transp==NOTRANS || transp==CONJ)
          scaled_op->scaleRight(*blk_vec);
        else if(transp==TRANS || transp==CONJTRANS)
          scaled_op->scaleLeft(*blk_vec);
      }
    }
  }
}

// private


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::resetStorage(
  const int numRowBlocks, const int numColBlocks
  )
{
  numRowBlocks_ = numRowBlocks;
  numColBlocks_ = numColBlocks;
  Ops_.resize(numRowBlocks_*numColBlocks_);
  if (is_null(productRange_)) {
    rangeBlocks_.resize(numRowBlocks);
    domainBlocks_.resize(numColBlocks);
  }
  blockFillIsActive_ = true;
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::assertBlockFillIsActive(
  bool wantedValue
  ) const
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPT(!(blockFillIsActive_==wantedValue));
#else
  (void)wantedValue;
#endif
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::assertBlockRowCol(
  const int i, const int j
  ) const
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPTION(
    !( 0 <= i ), std::logic_error
    ,"Error, i="<<i<<" is invalid!"
    );
  TEUCHOS_TEST_FOR_EXCEPTION(
    !( 0 <= j ), std::logic_error
    ,"Error, j="<<j<<" is invalid!"
    );
  // Only validate upper range if the number of row and column blocks is
  // fixed!
  if(Ops_.size()) {
    TEUCHOS_TEST_FOR_EXCEPTION(
      !( 0 <= i && i < numRowBlocks_ ), std::logic_error
      ,"Error, i="<<i<<" does not fall in the range [0,"<<numRowBlocks_-1<<"]!"
      );
    TEUCHOS_TEST_FOR_EXCEPTION(
      !( 0 <= j && j < numColBlocks_ ), std::logic_error
      ,"Error, j="<<j<<" does not fall in the range [0,"<<numColBlocks_-1<<"]!"
      );
  }
#else
  (void)i;
  (void)j;
#endif
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::setBlockSpaces(
  const int i, const int j, const LinearOpBase<Scalar> &block
  )
{
  using Teuchos::toString;
  assertBlockFillIsActive(true);
  assertBlockRowCol(i,j);

  // Validate that if the vector space block is already set that it is
  // compatible with the block that is being set.
  if( i < numRowBlocks_ && j < numColBlocks_ ) {
#ifdef TEUCHOS_DEBUG
    RCP<const VectorSpaceBase<Scalar> >
      rangeBlock = (
        productRange_.get()
        ? productRange_->getBlock(i)
        : rangeBlocks_[i]
        ),
      domainBlock = (
        productDomain_.get()
        ? productDomain_->getBlock(j)
        : domainBlocks_[j]
        );
    if(rangeBlock.get()) {
      THYRA_ASSERT_VEC_SPACES_NAMES(
        "DefaultBlockedLinearOp<Scalar>::setBlockSpaces(i,j,block):\n\n"
        "Adding block: " + block.description(),
        *rangeBlock,("(*productRange->getBlock("+toString(i)+"))"),
        *block.range(),("(*block["+toString(i)+","+toString(j)+"].range())")
        );
    }
    if(domainBlock.get()) {
      THYRA_ASSERT_VEC_SPACES_NAMES(
        "DefaultBlockedLinearOp<Scalar>::setBlockSpaces(i,j,block):\n\n"
        "Adding block: " + block.description(),
        *domainBlock,("(*productDomain->getBlock("+toString(j)+"))"),
        *block.domain(),("(*block["+toString(i)+","+toString(j)+"].domain())")
        );
    }
#endif // TEUCHOS_DEBUG
  }

  // Add spaces missing range and domain space blocks if we are doing a
  // flexible fill (otherwise these loops will not be executed)
  for( int k = numRowBlocks_; k <= i; ++k )
    rangeBlocks_.push_back(Teuchos::null);
  for( int k = numColBlocks_; k <= j; ++k )
    domainBlocks_.push_back(Teuchos::null);

  // Set the incoming range and domain blocks if not already set
  if(!productRange_.get()) {
    if(!rangeBlocks_[i].get())
      rangeBlocks_[i] = block.range().assert_not_null();
    if(!domainBlocks_[j].get()) {
      domainBlocks_[j] = block.domain().assert_not_null();
    }
  }

  // Update the current number of row and columns blocks if doing a flexible
  // fill.
  if(!Ops_.size()) {
    numRowBlocks_ = rangeBlocks_.size();
    numColBlocks_ = domainBlocks_.size();
  }

}


template<class Scalar>
template<class LinearOpType>
void DefaultBlockedLinearOp<Scalar>::setBlockImpl(
  const int i, const int j,
  const RCP<LinearOpType> &block
  )
{
  setBlockSpaces(i, j, *block);
  if (Ops_.size()) {
    // We are doing a fill with a fixed number of row and column blocks so we
    // can just set this.
    Ops_[numRowBlocks_*j+i] = block;
  }
  else {
    // We are doing a flexible fill so add the block to the stack of blocks or
    // replace a block that already exists.
    bool foundBlock = false;
    for( unsigned int k = 0; k < Ops_stack_.size(); ++k ) {
      BlockEntry<Scalar> &block_i_j = Ops_stack_[k];
      if( block_i_j.i == i && block_i_j.j == j ) {
        block_i_j.block = block;
        foundBlock = true;
        break;
      }
    }
    if(!foundBlock)
      Ops_stack_.push_back(BlockEntry<Scalar>(i,j,block));
  }
}


template<class Scalar>
void DefaultBlockedLinearOp<Scalar>::adjustBlockSpaces()
{

#ifdef TEUCHOS_DEBUG
  TEUCHOS_ASSERT_INEQUALITY(Ops_.size(), !=, 0);
#endif

  //
  // Loop through the rows and columns looking for rows with mixed
  // single-space range and/or domain spaces on operators and set the single
  // spaces as the block space if it exists.
  //
  // NOTE: Once we get here, we can safely assume that all of the operators
  // are compatible w.r.t. their spaces so if there are rows and/or columns
  // with mixed product and single vector spaces that we can just pick the
  // single vector space for the whole row and/or column.
  //

  // Adjust blocks in the range space
  for (int i = 0; i < numRowBlocks_; ++i) {
    for (int j = 0; j < numColBlocks_; ++j) {
      const RCP<const LinearOpBase<Scalar> >
        op_i_j = Ops_[numRowBlocks_*j+i];
      if (is_null(op_i_j))
        continue;
      const RCP<const VectorSpaceBase<Scalar> > range_i_j = op_i_j->range();
      if (is_null(productVectorSpaceBase<Scalar>(range_i_j, false))) {
        rangeBlocks_[i] = range_i_j;
        break;
      }
    }
  }

  // Adjust blocks in the domain space
  for (int j = 0; j < numColBlocks_; ++j) {
    for (int i = 0; i < numRowBlocks_; ++i) {
      const RCP<const LinearOpBase<Scalar> >
        op_i_j = Ops_[numRowBlocks_*j+i];
      if (is_null(op_i_j))
        continue;
      const RCP<const VectorSpaceBase<Scalar> >
        domain_i_j = op_i_j->domain();
      if (is_null(productVectorSpaceBase<Scalar>(domain_i_j, false))) {
        domainBlocks_[j] = domain_i_j;
        break;
      }
    }
  }

}


} // namespace Thyra


template<class Scalar>
Teuchos::RCP<Thyra::DefaultBlockedLinearOp<Scalar> >
Thyra::defaultBlockedLinearOp()
{
  return Teuchos::rcp(new DefaultBlockedLinearOp<Scalar>());
}


template<class Scalar>
Teuchos::RCP<const Thyra::LinearOpBase<Scalar> >
Thyra::block1x1(
  const RCP<const LinearOpBase<Scalar> > &A00,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(1,1);
  M->setBlock(0, 0, A00);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<const Thyra::LinearOpBase<Scalar> >
Thyra::block1x2(
  const RCP<const LinearOpBase<Scalar> > &A00,
  const RCP<const LinearOpBase<Scalar> > &A01,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(1,2);
  if(A00.get()) M->setBlock(0,0,A00);
  if(A01.get()) M->setBlock(0,1,A01);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<const Thyra::LinearOpBase<Scalar> >
Thyra::block2x1(
  const RCP<const LinearOpBase<Scalar> > &A00,
  const RCP<const LinearOpBase<Scalar> > &A10,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(2,1);
  if(A00.get()) M->setBlock(0,0,A00);
  if(A10.get()) M->setBlock(1,0,A10);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<const Thyra::LinearOpBase<Scalar> >
Thyra::block2x2(
  const RCP<const LinearOpBase<Scalar> > &A00,
  const RCP<const LinearOpBase<Scalar> > &A01,
  const RCP<const LinearOpBase<Scalar> > &A10,
  const RCP<const LinearOpBase<Scalar> > &A11,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(2,2);
  if(A00.get()) M->setBlock(0,0,A00);
  if(A01.get()) M->setBlock(0,1,A01);
  if(A10.get()) M->setBlock(1,0,A10);
  if(A11.get()) M->setBlock(1,1,A11);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<Thyra::LinearOpBase<Scalar> >
Thyra::nonconstBlock1x1(
  const RCP<LinearOpBase<Scalar> > &A00,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(1, 1);
  M->setNonconstBlock(0, 0, A00);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<Thyra::LinearOpBase<Scalar> >
Thyra::nonconstBlock1x2(
  const RCP<LinearOpBase<Scalar> > &A00,
  const RCP<LinearOpBase<Scalar> > &A01,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(1,2);
  if(A00.get()) M->setNonconstBlock(0,0,A00);
  if(A01.get()) M->setNonconstBlock(0,1,A01);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<Thyra::LinearOpBase<Scalar> >
Thyra::nonconstBlock2x1(
  const RCP<LinearOpBase<Scalar> > &A00,
  const RCP<LinearOpBase<Scalar> > &A10,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(2,1);
  if(A00.get()) M->setNonconstBlock(0,0,A00);
  if(A10.get()) M->setNonconstBlock(1,0,A10);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


template<class Scalar>
Teuchos::RCP<Thyra::LinearOpBase<Scalar> >
Thyra::nonconstBlock2x2(
  const RCP<LinearOpBase<Scalar> > &A00,
  const RCP<LinearOpBase<Scalar> > &A01,
  const RCP<LinearOpBase<Scalar> > &A10,
  const RCP<LinearOpBase<Scalar> > &A11,
  const std::string &label
  )
{
  RCP<PhysicallyBlockedLinearOpBase<Scalar> >
    M = defaultBlockedLinearOp<Scalar>();
  M->beginBlockFill(2,2);
  if(A00.get()) M->setNonconstBlock(0,0,A00);
  if(A01.get()) M->setNonconstBlock(0,1,A01);
  if(A10.get()) M->setNonconstBlock(1,0,A10);
  if(A11.get()) M->setNonconstBlock(1,1,A11);
  M->endBlockFill();
  if (label.length())
    M->setObjectLabel(label);
  return M;
}


//
// Explicit instantiation macro
//
// Must be expanded from within the Thyra namespace!
//


#define THYRA_DEFAULT_BLOCKED_LINEAR_OP_INSTANT(SCALAR) \
  \
  template class DefaultBlockedLinearOp<SCALAR >; \
   \
  template RCP<DefaultBlockedLinearOp<SCALAR > > \
  defaultBlockedLinearOp<SCALAR >(); \
   \
  template RCP<const LinearOpBase<SCALAR > > \
  block1x1( \
    const RCP<const LinearOpBase<SCALAR > > &A00, \
    const std::string &label \
    ); \
   \
  template RCP<const LinearOpBase<SCALAR > > \
  block1x2( \
    const RCP<const LinearOpBase<SCALAR > > &A00, \
    const RCP<const LinearOpBase<SCALAR > > &A01, \
    const std::string &label \
    ); \
   \
  template RCP<const LinearOpBase<SCALAR > > \
  block2x1( \
    const RCP<const LinearOpBase<SCALAR > > &A00, \
    const RCP<const LinearOpBase<SCALAR > > &A10, \
    const std::string &label \
    ); \
   \
  template RCP<const LinearOpBase<SCALAR > > \
  block2x2( \
    const RCP<const LinearOpBase<SCALAR > > &A00, \
    const RCP<const LinearOpBase<SCALAR > > &A01, \
    const RCP<const LinearOpBase<SCALAR > > &A10, \
    const RCP<const LinearOpBase<SCALAR > > &A11, \
    const std::string &label \
    ); \
   \
  template RCP<LinearOpBase<SCALAR > > \
  nonconstBlock1x1( \
    const RCP<LinearOpBase<SCALAR > > &A00, \
    const std::string &label \
    ); \
   \
  template RCP<LinearOpBase<SCALAR > > \
  nonconstBlock1x2( \
    const RCP<LinearOpBase<SCALAR > > &A00, \
    const RCP<LinearOpBase<SCALAR > > &A01, \
    const std::string &label \
    ); \
   \
  template RCP<LinearOpBase<SCALAR > > \
  nonconstBlock2x1( \
    const RCP<LinearOpBase<SCALAR > > &A00, \
    const RCP<LinearOpBase<SCALAR > > &A10, \
    const std::string &label \
    ); \
   \
  template RCP<LinearOpBase<SCALAR > > \
  nonconstBlock2x2( \
    const RCP<LinearOpBase<SCALAR > > &A00, \
    const RCP<LinearOpBase<SCALAR > > &A01, \
    const RCP<LinearOpBase<SCALAR > > &A10, \
    const RCP<LinearOpBase<SCALAR > > &A11, \
    const std::string &label \
    );


#endif  // THYRA_DEFAULT_BLOCKED_LINEAR_OP_DEF_HPP