BelosBlockGmresIter.hpp

Go to the documentation of this file.

// @HEADER
// *****************************************************************************
//                 Belos: Block Linear Solvers Package
//
// Copyright 2004-2016 NTESS and the Belos contributors.
// SPDX-License-Identifier: BSD-3-Clause
// *****************************************************************************
// @HEADER
 
#ifndef BELOS_BLOCK_GMRES_ITER_HPP
#define BELOS_BLOCK_GMRES_ITER_HPP

 
#include "BelosConfigDefs.hpp"
#include "BelosTypes.hpp"
#include "BelosGmresIteration.hpp"
 
#include "BelosLinearProblem.hpp"
#include "BelosMatOrthoManager.hpp"
#include "BelosOutputManager.hpp"
#include "BelosStatusTest.hpp"
#include "BelosOperatorTraits.hpp"
#include "BelosMultiVecTraits.hpp"
#include "BelosDenseMatTraits.hpp"
 
#include "Teuchos_BLAS.hpp"
#include "Teuchos_LAPACK.hpp"
#include "Teuchos_ScalarTraits.hpp"
#include "Teuchos_ParameterList.hpp"
#include "Teuchos_TimeMonitor.hpp"
 
#include <vector>

 
namespace Belos {
 
template<class ScalarType, class MV, class OP, class DM = DefaultDenseMatrix<int,ScalarType>>
class BlockGmresIter : virtual public GmresIteration<ScalarType,MV,OP,DM> {
 
  public:
 
  //
  // Convenience typedefs
  //
  typedef MultiVecTraits<ScalarType,MV,DM> MVT;
  typedef DenseMatTraits<ScalarType,DM> DMT;
  typedef OperatorTraits<ScalarType,MV,OP> OPT;
  typedef Teuchos::ScalarTraits<ScalarType> SCT;
  typedef typename SCT::magnitudeType MagnitudeType;



  BlockGmresIter( const Teuchos::RCP<LinearProblem<ScalarType,MV,OP,DM> > &problem,
                  const Teuchos::RCP<OutputManager<ScalarType> > &printer,
                  const Teuchos::RCP<StatusTest<ScalarType,MV,OP,DM> > &tester,
                  const Teuchos::RCP<MatOrthoManager<ScalarType,MV,OP,DM> > &ortho,
                  Teuchos::ParameterList &params );

  virtual ~BlockGmresIter() {};
 



  void iterate();

  void initializeGmres(GmresIterationState<ScalarType,MV,DM>& newstate);

  void initialize()
  {
    GmresIterationState<ScalarType,MV,DM> empty;
    initializeGmres(empty);
  }

  GmresIterationState<ScalarType,MV,DM> getState() const {
    GmresIterationState<ScalarType,MV,DM> state;
    state.curDim = curDim_;
    state.V = V_;
    state.H = H_;
    state.R = R_;
    state.z = z_;
    return state;
  }

 



  int getNumIters() const { return iter_; }

  void resetNumIters( int iter = 0 ) { iter_ = iter; }

  Teuchos::RCP<const MV> getNativeResiduals( std::vector<MagnitudeType> *norms ) const;


  Teuchos::RCP<MV> getCurrentUpdate() const;


  void updateLSQR( int dim = -1 );

  int getCurSubspaceDim() const {
    if (!initialized_) return 0;
    return curDim_;
  };

  int getMaxSubspaceDim() const { return blockSize_*numBlocks_; }

 



  const LinearProblem<ScalarType,MV,OP,DM>& getProblem() const { return *lp_; }

  int getBlockSize() const { return blockSize_; }

  void setBlockSize(int blockSize) { setSize( blockSize, numBlocks_ ); }

  int getNumBlocks() const { return numBlocks_; }

  void setNumBlocks(int numBlocks) { setSize( blockSize_, numBlocks ); }

  void setSize(int blockSize, int numBlocks);

  bool isInitialized() { return initialized_; }

 
  private:
 
  //
  // Internal structs
  //
  struct CheckList {
      bool checkV;
      bool checkArn;
      CheckList() : checkV(false), checkArn(false) {};
  };
  //
  // Internal methods
  //
  std::string accuracyCheck(const CheckList &chk, const std::string &where) const;

  void setStateSize();
 
  //
  // Classes inputed through constructor that define the linear problem to be solved.
  //
  const Teuchos::RCP<LinearProblem<ScalarType,MV,OP,DM> >    lp_;
  const Teuchos::RCP<OutputManager<ScalarType> >          om_;
  const Teuchos::RCP<StatusTest<ScalarType,MV,OP,DM> >       stest_;
  const Teuchos::RCP<OrthoManager<ScalarType,MV,DM> >        ortho_;
 
  //
  // Algorithmic parameters
  //
  // blockSize_ is the solver block size.
  // It controls the number of vectors added to the basis on each iteration.
  int blockSize_;
  // numBlocks_ is the size of the allocated space for the Krylov basis, in blocks.
  int numBlocks_;
 
  // Storage for QR factorization of the least squares system.
  std::vector<ScalarType> beta, sn;
  std::vector<MagnitudeType> cs;
 
  //
  // Current solver state
  //
  // initialized_ specifies that the basis vectors have been initialized and the iterate() routine
  // is capable of running; _initialize is controlled  by the initialize() member method
  // For the implications of the state of initialized_, please see documentation for initialize()
  bool initialized_;
 
  // stateStorageInitialized_ specifies that the state storage has be initialized to the current
  // blockSize_ and numBlocks_.  This initialization may be postponed if the linear problem was
  // generated without the right-hand side or solution vectors.
  bool stateStorageInitialized_;
 
  // keepHessenberg_ specifies that the iteration must keep the Hessenberg matrix formed via the
  // Arnoldi factorization and the upper triangular matrix that is the Hessenberg matrix reduced via
  // QR factorization separate.
  bool keepHessenberg_;
 
  // initHessenberg_ specifies that the iteration should reinitialize the Hessenberg matrix by zeroing
  // out all entries before an iteration is started.
  bool initHessenberg_;
 
  // Current subspace dimension, and number of iterations performed.
  int curDim_, iter_;
 
  //
  // State Storage
  //
  Teuchos::RCP<MV> V_;
  //
  // Projected matrices
  // H_ : Projected matrix from the Krylov factorization AV = VH + FE^T
  //
  Teuchos::RCP<DM> H_;
  //
  // QR decomposition of Projected matrices for solving the least squares system HY = B.
  // R_: Upper triangular reduction of H
  // z_: Q applied to right-hand side of the least squares system
  Teuchos::RCP<DM> R_;
  Teuchos::RCP<DM> z_;
};

  // Constructor.
  template<class ScalarType, class MV, class OP, class DM>
  BlockGmresIter<ScalarType,MV,OP,DM>::BlockGmresIter(const Teuchos::RCP<LinearProblem<ScalarType,MV,OP,DM> > &problem,
                                                   const Teuchos::RCP<OutputManager<ScalarType> > &printer,
                                                   const Teuchos::RCP<StatusTest<ScalarType,MV,OP,DM> > &tester,
                                                   const Teuchos::RCP<MatOrthoManager<ScalarType,MV,OP,DM> > &ortho,
                                                   Teuchos::ParameterList &params ):
    lp_(problem),
    om_(printer),
    stest_(tester),
    ortho_(ortho),
    blockSize_(0),
    numBlocks_(0),
    initialized_(false),
    stateStorageInitialized_(false),
    keepHessenberg_(false),
    initHessenberg_(false),
    curDim_(0),
    iter_(0)
  {
    // Find out whether we are saving the Hessenberg matrix.
    if ( om_->isVerbosity( Debug ) )
      keepHessenberg_ = true;
    else
      keepHessenberg_ = params.get("Keep Hessenberg", false);
 
    // Find out whether we are initializing the Hessenberg matrix.
    initHessenberg_ = params.get("Initialize Hessenberg", false);
 
    // Get the maximum number of blocks allowed for this Krylov subspace
    TEUCHOS_TEST_FOR_EXCEPTION(!params.isParameter("Num Blocks"), std::invalid_argument,
                       "Belos::BlockGmresIter::constructor: mandatory parameter 'Num Blocks' is not specified.");
    int nb = Teuchos::getParameter<int>(params, "Num Blocks");
 
    // Set the block size and allocate data
    int bs = params.get("Block Size", 1);
    setSize( bs, nb );
  }

  // Set the block size and make necessary adjustments.
  template <class ScalarType, class MV, class OP, class DM>
  void BlockGmresIter<ScalarType,MV,OP,DM>::setSize (int blockSize, int numBlocks)
  {
    // This routine only allocates space; it doesn't not perform any computation
    // any change in size will invalidate the state of the solver.
 
    TEUCHOS_TEST_FOR_EXCEPTION(numBlocks <= 0 || blockSize <= 0, std::invalid_argument, "Belos::BlockGmresIter::setSize was passed a non-positive argument.");
    if (blockSize == blockSize_ && numBlocks == numBlocks_) {
      // do nothing
      return;
    }
 
    if (blockSize!=blockSize_ || numBlocks!=numBlocks_)
      stateStorageInitialized_ = false;
 
    blockSize_ = blockSize;
    numBlocks_ = numBlocks;
 
    initialized_ = false;
    curDim_ = 0;
 
    // Use the current blockSize_ and numBlocks_ to initialize the state storage.
    setStateSize();
 
  }

  // Setup the state storage.
  template <class ScalarType, class MV, class OP, class DM>
  void BlockGmresIter<ScalarType,MV,OP,DM>::setStateSize ()
  {
    if (!stateStorageInitialized_) {
 
      // Check if there is any multivector to clone from.
      Teuchos::RCP<const MV> lhsMV = lp_->getLHS();
      Teuchos::RCP<const MV> rhsMV = lp_->getRHS();
      if (lhsMV == Teuchos::null && rhsMV == Teuchos::null) {
        stateStorageInitialized_ = false;
        return;
      }
      else {

        // blockSize*numBlocks dependent
        //
        int newsd = blockSize_*(numBlocks_+1);
 
        if (blockSize_==1) {
          cs.resize( newsd );
          sn.resize( newsd );
        }
        else {
          beta.resize( newsd );
        }
 
        // Initialize the state storage
        TEUCHOS_TEST_FOR_EXCEPTION(blockSize_*static_cast<ptrdiff_t>(numBlocks_) > MVT::GetGlobalLength(*rhsMV),std::invalid_argument,
                           "Belos::BlockGmresIter::setStateSize(): Cannot generate a Krylov basis with dimension larger the operator!");
 
        // If the subspace has not be initialized before, generate it using the LHS or RHS from lp_.
        if (V_ == Teuchos::null) {
          // Get the multivector that is not null.
          Teuchos::RCP<const MV> tmp = ( (rhsMV!=Teuchos::null)? rhsMV: lhsMV );
          TEUCHOS_TEST_FOR_EXCEPTION(tmp == Teuchos::null,std::invalid_argument,
                                     "Belos::BlockGmresIter::setStateSize(): linear problem does not specify multivectors to clone from.");
          V_ = MVT::Clone( *tmp, newsd );
        }
        else {
          // Generate V_ by cloning itself ONLY if more space is needed.
          if (MVT::GetNumberVecs(*V_) < newsd) {
            Teuchos::RCP<const MV> tmp = V_;
            V_ = MVT::Clone( *tmp, newsd );
          }
        }
 
        // Generate R_ only if it doesn't exist, otherwise resize it.
        if (R_ == Teuchos::null) {
          R_ = DMT::Create();
        }
        if (initHessenberg_) {
          DMT::Reshape(*R_, newsd, newsd-blockSize_, true);
        }
        else {
          if (DMT::GetNumRows(*R_) < newsd || DMT::GetNumCols(*R_) < newsd-blockSize_) {
            DMT::Reshape(*R_, newsd, newsd-blockSize_, false);
          }
        }
 
        // Generate H_ only if it doesn't exist, and we are keeping the upper Hessenberg matrix.
        if (keepHessenberg_) {
          if (H_ == Teuchos::null) {
            H_ = DMT::Create();
          }
          if (initHessenberg_) {
            DMT::Reshape(*H_, newsd, newsd-blockSize_, true);
          }
          else {
            if (DMT::GetNumRows(*H_)< newsd || DMT::GetNumCols(*H_)< newsd-blockSize_) {
              DMT::Reshape(*H_, newsd, newsd-blockSize_, false);
            }
          }
        }
        else {
          // Point H_ and R_ at the same object.
          H_ = R_;
        }
 
        // Generate z_ only if it doesn't exist, otherwise resize it.
        if (z_ == Teuchos::null) {
          z_ = DMT::Create();
        }
        if (DMT::GetNumRows(*z_) < newsd || DMT::GetNumCols(*z_) < blockSize_) {
          DMT::Reshape(*z_, newsd, blockSize_);
        }
 
        // State storage has now been initialized.
        stateStorageInitialized_ = true;
      }
    }
  }

  // Get the current update from this subspace.
  template <class ScalarType, class MV, class OP, class DM>
  Teuchos::RCP<MV> BlockGmresIter<ScalarType,MV,OP,DM>::getCurrentUpdate() const
  {
    //
    // If this is the first iteration of the Arnoldi factorization,
    // there is no update, so return Teuchos::null.
    //
    Teuchos::RCP<MV> currentUpdate = Teuchos::null;
    if (curDim_==0) {
      return currentUpdate;
    } else {
      const ScalarType one  = SCT::one();
      const ScalarType zero = SCT::zero();
      Teuchos::BLAS<int,ScalarType> blas;
      currentUpdate = MVT::Clone( *V_, blockSize_ );
      //
      //  Make a view and then copy the RHS of the least squares problem.  DON'T OVERWRITE IT!
      //
      Teuchos::RCP<DM> y = DMT::SubviewCopy(*z_, curDim_, blockSize_);
      //
      //  Solve the least squares problem.
      //
      blas.TRSM( Teuchos::LEFT_SIDE, Teuchos::UPPER_TRI, Teuchos::NO_TRANS,
                 Teuchos::NON_UNIT_DIAG, curDim_, blockSize_, one,
                 DMT::GetRawHostPtr(*R_), DMT::GetStride(*R_), DMT::GetRawHostPtr(*y), DMT::GetStride(*y) );
      DMT::SyncHostToDevice(*y);
      //
      //  Compute the current update.
      //
      std::vector<int> index(curDim_);
      for ( int i=0; i<curDim_; i++ ) {
        index[i] = i;
      }
      Teuchos::RCP<const MV> Vjp1 = MVT::CloneView( *V_, index );
      MVT::MvTimesMatAddMv( one, *Vjp1, *y, zero, *currentUpdate );
    }
    return currentUpdate;
  }
 

  // Get the native residuals stored in this iteration.
  // Note:  No residual std::vector will be returned by Gmres.
  template <class ScalarType, class MV, class OP, class DM>
  Teuchos::RCP<const MV> BlockGmresIter<ScalarType,MV,OP,DM>::getNativeResiduals( std::vector<MagnitudeType> *norms ) const
  {
    //
    // NOTE: Make sure the incoming std::vector is the correct size!
    //
    if ( norms && (int)norms->size() < blockSize_ )
      norms->resize( blockSize_ );
 
    if (norms) {
      Teuchos::BLAS<int,ScalarType> blas;
      DMT::SyncDeviceToHost(*z_);
      for (int j=0; j<blockSize_; j++) {
        Teuchos::RCP<DM> z_j = DMT::Subview(*z_, blockSize_, 1, curDim_, j);
        (*norms)[j] = blas.NRM2( blockSize_, DMT::GetRawHostPtr(*z_j), 1);
      }
    }
    return Teuchos::null;
  }
 
 

  // Initialize this iteration object
  template <class ScalarType, class MV, class OP, class DM>
  void BlockGmresIter<ScalarType,MV,OP,DM>::initializeGmres(GmresIterationState<ScalarType,MV,DM>& newstate)
  {
    // Initialize the state storage if it isn't already.
    if (!stateStorageInitialized_)
      setStateSize();
 
    TEUCHOS_TEST_FOR_EXCEPTION(!stateStorageInitialized_,std::invalid_argument,
                               "Belos::BlockGmresIter::initialize(): Cannot initialize state storage!");
 
    // NOTE:  In BlockGmresIter, V and Z are required!!!
    // inconsitent multivectors widths and lengths will not be tolerated, and
    // will be treated with exceptions.
    //
    std::string errstr("Belos::BlockGmresIter::initialize(): Specified multivectors must have a consistent length and width.");
 
    if (newstate.V != Teuchos::null && newstate.z != Teuchos::null) {
 
      // initialize V_,z_, and curDim_
 
      TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.V) != MVT::GetGlobalLength(*V_),
                          std::invalid_argument, errstr );
      TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.V) < blockSize_,
                          std::invalid_argument, errstr );
      TEUCHOS_TEST_FOR_EXCEPTION( newstate.curDim > blockSize_*(numBlocks_+1),
                          std::invalid_argument, errstr );
 
      curDim_ = newstate.curDim;
      int lclDim = MVT::GetNumberVecs(*newstate.V);
 
      // check size of Z
      TEUCHOS_TEST_FOR_EXCEPTION(DMT::GetNumRows(*newstate.z) < curDim_ || DMT::GetNumCols(*newstate.z) < blockSize_, std::invalid_argument, errstr);
 
 
      // copy basis vectors from newstate into V
      if (newstate.V != V_) {
        // only copy over the first block and print a warning.
        if (curDim_ == 0 && lclDim > blockSize_) {
          om_->stream(Warnings) << "Belos::BlockGmresIter::initialize(): the solver was initialized with a kernel of " << lclDim << std::endl
                                                                         << "The block size however is only " << blockSize_ << std::endl
                                                                         << "The last " << lclDim - blockSize_ << " vectors will be discarded." << std::endl;
        }
        std::vector<int> nevind(curDim_+blockSize_);
        for (int i=0; i<curDim_+blockSize_; i++) nevind[i] = i;
        Teuchos::RCP<const MV> newV = MVT::CloneView( *newstate.V, nevind );
        Teuchos::RCP<MV> lclV = MVT::CloneViewNonConst( *V_, nevind );
        MVT::Assign( *newV, *lclV );
 
        // done with local pointers
        lclV = Teuchos::null;
      }
 
      // put data into z_, make sure old information is not still hanging around.
      if (newstate.z != z_) {
        DMT::PutScalar(*z_);
        //Note: Need a SubviewConst here because the z in GMRES Iteration State is defined as const.
        Teuchos::RCP<const DM> newZ = DMT::SubviewConst(*newstate.z,curDim_+blockSize_,blockSize_);
        Teuchos::RCP<DM> lclZ = DMT::Subview(*z_,curDim_+blockSize_,blockSize_);
        DMT::Assign(*lclZ,*newZ);
 
        // done with local pointers
        lclZ = Teuchos::null;
      }
 
    }
    else {
 
      TEUCHOS_TEST_FOR_EXCEPTION(newstate.V == Teuchos::null,std::invalid_argument,
                         "Belos::BlockGmresIter::initialize(): BlockGmresStateIterState does not have initial kernel V_0.");
 
      TEUCHOS_TEST_FOR_EXCEPTION(newstate.z == Teuchos::null,std::invalid_argument,
                         "Belos::BlockGmresIter::initialize(): BlockGmresStateIterState does not have initial norms z_0.");
    }
 
    // the solver is initialized
    initialized_ = true;
 
    if (om_->isVerbosity( Debug ) ) {
      // Check almost everything here
      CheckList chk;
      chk.checkV = true;
      chk.checkArn = true;
      om_->print( Debug, accuracyCheck(chk, ": after initialize()") );
    }
 
  }
 

  // Iterate until the status test informs us we should stop.
  template <class ScalarType, class MV, class OP, class DM>
  void BlockGmresIter<ScalarType,MV,OP,DM>::iterate()
  {
    //
    // Allocate/initialize data structures
    //
    if (initialized_ == false) {
      initialize();
    }
 
    // Compute the current search dimension.
    int searchDim = blockSize_*numBlocks_;

    // iterate until the status test tells us to stop.
    //
    // also break if our basis is full
    //
    while (stest_->checkStatus(this) != Passed && curDim_+blockSize_ <= searchDim) {
 
      iter_++;
 
      // F can be found at the curDim_ block, but the next block is at curDim_ + blockSize_.
      int lclDim = curDim_ + blockSize_;
 
      // Get the current part of the basis.
      std::vector<int> curind(blockSize_);
      for (int i=0; i<blockSize_; i++) { curind[i] = lclDim + i; }
      Teuchos::RCP<MV> Vnext = MVT::CloneViewNonConst(*V_,curind);
 
      // Get a view of the previous vectors.
      // This is used for orthogonalization and for computing V^H K H
      for (int i=0; i<blockSize_; i++) { curind[i] = curDim_ + i; }
      Teuchos::RCP<const MV> Vprev = MVT::CloneView(*V_,curind);
 
      // Compute the next std::vector in the Krylov basis:  Vnext = Op*Vprev
      lp_->apply(*Vprev,*Vnext);
      Vprev = Teuchos::null;
 
      // Remove all previous Krylov basis vectors from Vnext
      // Get a view of all the previous vectors
      std::vector<int> prevind(lclDim);
      for (int i=0; i<lclDim; i++) { prevind[i] = i; }
      Vprev = MVT::CloneView(*V_,prevind);
      Teuchos::Array<Teuchos::RCP<const MV> > AVprev(1, Vprev);
 
      // Get a view of the part of the Hessenberg matrix needed to hold the ortho coeffs.
      Teuchos::RCP<DM> subH = DMT::Subview(*H_,lclDim,blockSize_,0,curDim_ );
      Teuchos::Array<Teuchos::RCP<DM> > AsubH;
      AsubH.append( subH );
 
      // Get a view of the part of the Hessenberg matrix needed to hold the norm coeffs.
      Teuchos::RCP<DM> subH2 = DMT::Subview(*H_,blockSize_,blockSize_,lclDim,curDim_);
      DMT::PutScalar(*subH2);  // Initialize subdiagonal to zero
      
      // TODO
      // Make an abstract dense matrix that holds the data of subH2 (an RCP of serialDense.)
      // Do the same for AsubH. ????
      // subH needs to be a new rcp to an abstract dense. No, keep subH how it is.
      // Then make a subHAbstract that grabs the pointer from subH, making it an abstract dense guy. 
      // Then AsubH can be a Teuchos::Array
      // of the abstract dense guys. 
      int rank = ortho_->projectAndNormalize(*Vnext,AsubH,subH2,AVprev);
 
      // Copy over the coefficients if we are saving the upper Hessenberg matrix,
      // just in case we run into an error.
      if (keepHessenberg_) {
        // Copy over the orthogonalization coefficients.
        Teuchos::RCP<DM> subR = DMT::Subview(*R_,lclDim,blockSize_,0,curDim_ );
        DMT::Assign(*subR,*subH);
 
        // Copy over the lower diagonal block of the Hessenberg matrix.
        Teuchos::RCP<DM> subR2 = DMT::Subview(*R_,blockSize_,blockSize_,lclDim,curDim_ );
        DMT::Assign(*subR2,*subH2);
      }
 
      TEUCHOS_TEST_FOR_EXCEPTION(rank != blockSize_,GmresIterationOrthoFailure,
                                 "Belos::BlockGmresIter::iterate(): couldn't generate basis of full rank.");
      //
      // V has been extended, and H has been extended.
      //
      // Update the QR factorization of the upper Hessenberg matrix
      //
      updateLSQR();
      //
      // Update basis dim and release all pointers.
      //
      Vnext = Teuchos::null;
      curDim_ += blockSize_;
      //
      // When required, monitor some orthogonalities
      if (om_->isVerbosity( Debug ) ) {
        // Check almost everything here
        CheckList chk;
        chk.checkV = true;
        chk.checkArn = true;
        om_->print( Debug, accuracyCheck(chk, ": after local update") );
      }
      else if (om_->isVerbosity( OrthoDetails ) ) {
        CheckList chk;
        chk.checkV = true;
        om_->print( OrthoDetails, accuracyCheck(chk, ": after local update") );
      }
 
    } // end while (statusTest == false)
 
  }
 
 
  template<class ScalarType, class MV, class OP, class DM>
  void BlockGmresIter<ScalarType,MV,OP,DM>::updateLSQR( int dim )
  {
    int i, j, maxidx;
    ScalarType sigma, mu, vscale, maxelem;
    const ScalarType zero = SCT::zero();
 
    // Get correct dimension based on input "dim"
    // Remember that ortho failures result in an exit before updateLSQR() is called.
    // Therefore, it is possible that dim == curDim_.
    int curDim = curDim_;
    if (dim >= curDim_ && dim < getMaxSubspaceDim()) {
      curDim = dim;
    }
 
    Teuchos::BLAS<int, ScalarType> blas;
    //
    // Apply previous transformations and compute new transformation to reduce upper-Hessenberg
    // system to upper-triangular form.
    //
    DMT::SyncDeviceToHost(*R_);
    DMT::SyncDeviceToHost(*z_);
 
    if (blockSize_ == 1) {
      //
      // QR factorization of Least-Squares system with Givens rotations
      //
      for (i=0; i<curDim; i++) {
        //
        // Apply previous Givens rotations to new column of Hessenberg matrix
        //
        blas.ROT( 1, &DMT::Value(*R_,i,curDim), 1, &DMT::Value(*R_,i+1, curDim), 1, &cs[i], &sn[i] );
      }
      //
      // Calculate new Givens rotation
      //
      blas.ROTG( &DMT::Value(*R_,curDim,curDim), &DMT::Value(*R_,curDim+1,curDim), &cs[curDim], &sn[curDim] );
      DMT::Value(*R_,curDim+1,curDim) = zero;
      //
      // Update RHS w/ new transformation
      //
      blas.ROT( 1, &DMT::Value(*z_,curDim,0), 1, &DMT::Value(*z_,curDim+1,0), 1, &cs[curDim], &sn[curDim] );
    }
    else {
      //
      // QR factorization of Least-Squares system with Householder reflectors
      //
      for (j=0; j<blockSize_; j++) {
        //
        // Apply previous Householder reflectors to new block of Hessenberg matrix
        //
        for (i=0; i<curDim+j; i++) {
          sigma = blas.DOT( blockSize_, &DMT::Value(*R_,i+1,i), 1, &DMT::Value(*R_,i+1,curDim+j), 1);
          sigma += DMT::ValueConst(*R_,i,curDim+j);
          sigma *= SCT::conjugate(beta[i]);
          blas.AXPY(blockSize_, ScalarType(-sigma), &DMT::Value(*R_,i+1,i), 1, &DMT::Value(*R_,i+1,curDim+j), 1);
          DMT::Value(*R_,i,curDim+j) -= sigma;
        }
        //
        // Compute new Householder reflector
        //
        maxidx = blas.IAMAX( blockSize_+1, &DMT::Value(*R_,curDim+j,curDim+j), 1 );
        maxelem = SCT::magnitude(DMT::Value(*R_,curDim+j+maxidx-1,curDim+j));
        for (i=0; i<blockSize_+1; i++)
          DMT::Value(*R_,curDim+j+i,curDim+j) /= maxelem;
        sigma = blas.DOT( blockSize_, &DMT::Value(*R_,curDim+j+1,curDim+j), 1,
                          &DMT::Value(*R_,curDim+j+1,curDim+j), 1 );
        MagnitudeType sign_Rjj = -SCT::real(DMT::Value(*R_,curDim+j,curDim+j)) /
                 SCT::magnitude(SCT::real((DMT::Value(*R_,curDim+j,curDim+j))));
        if (sigma == zero) {
          beta[curDim + j] = zero;
        } else {
          mu = SCT::squareroot(SCT::conjugate(DMT::Value(*R_,curDim+j,curDim+j))*DMT::Value(*R_,curDim+j,curDim+j)+sigma);
          vscale = DMT::ValueConst(*R_,curDim+j,curDim+j) - Teuchos::as<ScalarType>(sign_Rjj)*mu;
          beta[curDim+j] = -Teuchos::as<ScalarType>(sign_Rjj) * vscale / mu;
          DMT::Value(*R_,curDim+j,curDim+j) = Teuchos::as<ScalarType>(sign_Rjj)*maxelem*mu;
          for (i=0; i<blockSize_; i++)
            DMT::Value(*R_,curDim+j+1+i,curDim+j) /= vscale;
        }
        //
        // Apply new Householder reflector to rhs
        //
        for (i=0; i<blockSize_; i++) {
          sigma = blas.DOT( blockSize_, &DMT::Value(*R_,curDim+j+1,curDim+j),
                            1, &DMT::Value(*z_,curDim+j+1,i), 1);
          sigma += DMT::ValueConst(*z_,curDim+j,i);
          sigma *= SCT::conjugate(beta[curDim+j]);
          blas.AXPY(blockSize_, ScalarType(-sigma), &DMT::Value(*R_,curDim+j+1,curDim+j),
                    1, &DMT::Value(*z_,curDim+j+1,i), 1);
          DMT::Value(*z_,curDim+j,i) -= sigma;
        }
      }
    } // end if (blockSize_ == 1)
 
    DMT::SyncHostToDevice(*z_); 
    DMT::SyncHostToDevice(*R_); 
 
    // If the least-squares problem is updated wrt "dim" then update the curDim_.
    if (dim >= curDim_ && dim < getMaxSubspaceDim()) {
      curDim_ = dim + blockSize_;
    }
  } // end updateLSQR()

  // Check accuracy, orthogonality, and other debugging stuff
  //
  // bools specify which tests we want to run (instead of running more than we actually care about)
  //
  // checkV : V orthonormal
  //
  // checkArn: check the Arnoldi factorization
  //
  // NOTE:  This method needs to check the current dimension of the subspace, since it is possible to
  //        call this method when curDim_ = 0 (after initialization).
  template <class ScalarType, class MV, class OP, class DM>
  std::string BlockGmresIter<ScalarType,MV,OP,DM>::accuracyCheck( const CheckList &chk, const std::string &where ) const
  {
    std::stringstream os;
    os.precision(2);
    os.setf(std::ios::scientific, std::ios::floatfield);
    MagnitudeType tmp;
 
    os << " Debugging checks: iteration " << iter_ << where << std::endl;
 
    // index vectors for V and F
    std::vector<int> lclind(curDim_);
    for (int i=0; i<curDim_; i++) lclind[i] = i;
    std::vector<int> bsind(blockSize_);
    for (int i=0; i<blockSize_; i++) { bsind[i] = curDim_ + i; }
 
    Teuchos::RCP<const MV> lclV,lclF;
    Teuchos::RCP<MV> lclAV;
    if (curDim_)
      lclV = MVT::CloneView(*V_,lclind);
    lclF = MVT::CloneView(*V_,bsind);
 
    if (chk.checkV) {
      if (curDim_) {
        tmp = ortho_->orthonormError(*lclV);
        os << " >> Error in V^H M V == I  : " << tmp << std::endl;
      }
      tmp = ortho_->orthonormError(*lclF);
      os << " >> Error in F^H M F == I  : " << tmp << std::endl;
      if (curDim_) {
        tmp = ortho_->orthogError(*lclV,*lclF);
        os << " >> Error in V^H M F == 0  : " << tmp << std::endl;
      }
    }
  
    if (chk.checkArn) {
 
      if (curDim_) {
        // Compute AV    
        lclAV = MVT::Clone(*V_,curDim_);
        lp_->apply(*lclV,*lclAV);
 
        // Compute AV - VH
        const ScalarType one  = Teuchos::ScalarTraits<ScalarType>::one();
        Teuchos::RCP<DM> subH = DMT::Subview(*H_,curDim_,curDim_);
        MVT::MvTimesMatAddMv( -one, *lclV, *subH, one, *lclAV );
 
        // Compute FB_k^T - (AV-VH)
        Teuchos::RCP<DM> curB = DMT::Subview(*H_,blockSize_,curDim_,curDim_);
        MVT::MvTimesMatAddMv( -one, *lclF, *curB, one, *lclAV );
 
        // Compute || FE_k^T - (AV-VH) ||
        std::vector<MagnitudeType> arnNorms( curDim_ );
        ortho_->norm( *lclAV, arnNorms );
 
        for (int i=0; i<curDim_; i++) {
        os << " >> Error in Krylov factorization (R = AV-VH-FB^H), ||R[" << i << "]|| : " << arnNorms[i] << std::endl;
        }
      }
    }
 
    os << std::endl;
 
    return os.str();
  }
 
} // end Belos namespace
 
#endif /* BELOS_BLOCK_GMRES_ITER_HPP */