AbstractLinAlgPack_MatrixSparseCOORSerial.cpp

Go to the documentation of this file.

// @HEADER
// ***********************************************************************
// 
// Moocho: Multi-functional Object-Oriented arCHitecture for Optimization
//                  Copyright (2003) 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

#include <assert.h>

#include "AbstractLinAlgPack_MatrixSparseCOORSerial.hpp"
#include "AbstractLinAlgPack_MatrixCOORTmplItfc.hpp"
#include "AbstractLinAlgPack_COOMatrixTmplOp.hpp"
#include "AbstractLinAlgPack_VectorDenseEncap.hpp"
#include "AbstractLinAlgPack_AssertOp.hpp"
#include "AbstractLinAlgPack_LinAlgOpPackHack.hpp"
#include "Teuchos_Assert.hpp"
#include "Teuchos_dyn_cast.hpp"

namespace AbstractLinAlgPack {

MatrixSparseCOORSerial::ReleaseValRowColArrays::~ReleaseValRowColArrays()
{
  if(owns_mem_) {
    if(val_)   delete [] val_;
    if(row_i_) delete [] row_i_;
    if(col_j_) delete [] col_j_;
  }
}

bool MatrixSparseCOORSerial::ReleaseValRowColArrays::resource_is_bound() const
{
  return val_ != NULL;
}

// static members

MatrixSparseCOORSerial::release_resource_ptr_t
MatrixSparseCOORSerial::release_resource_null_ = Teuchos::null;

// Constructors / initializers

MatrixSparseCOORSerial::MatrixSparseCOORSerial()
  :rows_(0)
  ,cols_(0)
  ,max_nz_(0)
  ,nz_(0)
  ,val_(NULL)
  ,row_i_(NULL)
  ,col_j_(NULL)
  ,self_allocate_(true)
{}

void MatrixSparseCOORSerial::set_buffers(
  size_type                      max_nz
  ,value_type                    *val
  ,index_type                    *row_i
  ,index_type                    *col_j
  ,const release_resource_ptr_t  &release_resource
  ,size_type                     rows
  ,size_type                     cols
  ,size_type                     nz
  ,bool                          check_input
  )
{
#ifdef TEUCHOS_DEBUG
  const char msg_err[] = "MatrixSparseCOORSerial::set_buffer(...) : Error,!";
  TEUCHOS_TEST_FOR_EXCEPTION( max_nz <= 0, std::invalid_argument, msg_err );
  TEUCHOS_TEST_FOR_EXCEPTION( val == NULL || row_i == NULL || col_j == NULL, std::invalid_argument, msg_err );
  TEUCHOS_TEST_FOR_EXCEPTION( rows > 0 && cols <= 0 , std::invalid_argument, msg_err );
  TEUCHOS_TEST_FOR_EXCEPTION( rows > 0 && (nz < 0 || nz > max_nz), std::invalid_argument, msg_err );
#endif
  max_nz_           = max_nz;
  val_              = val;
  row_i_            = row_i;
  col_j_            = col_j;
  release_resource_ = release_resource;
  self_allocate_    = false;
  if(rows) {
    rows_ = rows;
    cols_ = cols;
    nz_   = nz;
    space_cols_.initialize(rows);
    space_rows_.initialize(cols);
    if( nz && check_input ) {
      TEUCHOS_TEST_FOR_EXCEPT(true); // Todo: Check that row_i[] and col_j[] are in bounds
    }
  }
  else {
    rows_ = 0;
    cols_ = 0;
    nz_   = 0;
    space_cols_.initialize(0);
    space_rows_.initialize(0);
  }
}

void MatrixSparseCOORSerial::set_uninitialized()
{
  max_nz_           = 0;
  val_              = NULL;
  row_i_            = NULL;
  col_j_            = NULL;
  release_resource_ = Teuchos::null;
  self_allocate_    = true;
  rows_             = 0;
  cols_             = 0;
  nz_               = 0;
  space_cols_.initialize(0);
  space_rows_.initialize(0);
}

// Overridden from MatrixBase

size_type MatrixSparseCOORSerial::rows() const
{
  return rows_;
}

size_type MatrixSparseCOORSerial::cols() const
{
  return cols_;
}

size_type MatrixSparseCOORSerial::nz() const
{
  return nz_;
}

// Overridden from MatrixOp

const VectorSpace& MatrixSparseCOORSerial::space_cols() const
{
  return space_cols_;
}

const VectorSpace& MatrixSparseCOORSerial::space_rows() const
{
  return space_rows_;
}

MatrixOp& MatrixSparseCOORSerial::operator=(const MatrixOp& M)
{
  using Teuchos::dyn_cast;
  const MatrixSparseCOORSerial
    &Mc = dyn_cast<const MatrixSparseCOORSerial>(M);
  if( this == &Mc )
    return *this; // assignment to self
  // A shallow copy is fine as long as we are carefull.
  max_nz_           = Mc.max_nz_;
  val_              = Mc.val_;
  row_i_            = Mc.row_i_;
  col_j_            = Mc.col_j_;
  release_resource_ = Mc.release_resource_;
  self_allocate_    = Mc.self_allocate_;
  rows_             = Mc.rows_;
  cols_             = Mc.cols_;
  nz_               = Mc.nz_;
  space_cols_.initialize(rows_);
  space_rows_.initialize(cols_);
  return *this;
}

std::ostream& MatrixSparseCOORSerial::output(std::ostream& out) const
{
  const size_type
    rows = this->rows(),
    cols = this->cols(),
    nz   = this->nz();
  out
    << "Sparse " << rows << " x " << cols << " matrix with "
    << nz << " nonzero entries:\n";
  const value_type
    *itr_val      = &val_[0],
    *itr_val_end  = itr_val + nz_;
  const index_type
    *itr_ivect    = &row_i_[0],
    *itr_jvect    = &col_j_[0];
  for(; itr_val != itr_val_end; ++itr_val, ++itr_ivect, ++itr_jvect)
    out << *itr_val << ":" << *itr_ivect << ":" << *itr_jvect << " ";
  out << "\n";
  
  if( rows * cols <= 400 ) {
    out << "Converted to dense =\n";
    MatrixOp::output(out);
  }

  return out;
}

void MatrixSparseCOORSerial::Vp_StMtV(
  VectorMutable* y, value_type a, BLAS_Cpp::Transp M_trans
  , const Vector& x, value_type b
  ) const
{
  AbstractLinAlgPack::Vp_MtV_assert_compatibility( y, *this, M_trans, x );
  VectorDenseMutableEncap   y_d(*y);
  VectorDenseEncap          x_d(x);
  DenseLinAlgPack::Vt_S(&y_d(),b);
  Vp_StCOOMtV(
    &y_d(),a,MatrixCOORTmplItfc<value_type,index_type>(
      rows_,cols_,nz_,0,0,val_,row_i_,col_j_ )
    ,M_trans, x_d()
    );
}

// Overridden from MatrixLoadSparseElements

void MatrixSparseCOORSerial::reinitialize(
  size_type                  rows
  ,size_type                 cols
  ,size_type                 max_nz
  ,EAssumeElementUniqueness  element_uniqueness
  )
{
  namespace rcp = MemMngPack;
#ifdef TEUCHOS_DEBUG
  const char msg_err_head[] = "MatrixSparseCOORSerial::reinitialize(...) : Error";
  TEUCHOS_TEST_FOR_EXCEPTION( max_nz <= 0, std::invalid_argument, msg_err_head<<"!" );
  TEUCHOS_TEST_FOR_EXCEPTION( rows <= 0 || cols <= 0 , std::invalid_argument, msg_err_head<<"!" );
#endif
  rows_               = rows;
  cols_               = cols; 
  element_uniqueness_ = element_uniqueness;
  if( self_allocate_ ) {
    if(max_nz_ < max_nz) {
      release_resource_ = Teuchos::rcp(
        new ReleaseValRowColArrays(
          val_    = new value_type[max_nz]
          ,row_i_ = new index_type[max_nz]
          ,col_j_ = new index_type[max_nz]
          )
        );
      max_nz_ = max_nz;
    }

  }
  else {
#ifdef TEUCHOS_DEBUG
    TEUCHOS_TEST_FOR_EXCEPTION(
      max_nz <= max_nz_, std::invalid_argument
      ,msg_err_head << "Buffers set up by client in set_buffers() only allows storage for "
      "max_nz_ = " << max_nz_ << " nonzero entries while client requests storage for "
      "max_nz = " << max_nz << " nonzero entries!" );
#endif
  }
  reload_val_only_         = false;
  reload_val_only_nz_last_ = 0;
  nz_                      = 0;
  max_nz_load_             = 0;
}

void MatrixSparseCOORSerial::reset_to_load_values()
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPTION(
    rows_ == 0 || cols_ == 0, std::invalid_argument
    ,"MatrixSparseCOORSerial::reset_to_load_values(...) : Error, "
    "this matrix is not initialized so it can't be rest to load "
    "new values for nonzero entries." );
#endif
  reload_val_only_         = true;
  reload_val_only_nz_last_ = nz_;
  nz_                      = 0;
  max_nz_load_             = 0;
}

void MatrixSparseCOORSerial::get_load_nonzeros_buffers(
  size_type      max_nz_load
  ,value_type    **val
  ,index_type    **row_i
  ,index_type    **col_j
  )
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPTION(
    max_nz_load_ != 0 , std::logic_error
    ,"MatrixSparseCOORSerial::get_load_nonzeros_buffers(...) : Error, "
    "You must call commit_load_nonzeros_buffers() between calls to this method!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    max_nz_load <= 0 || max_nz_load > max_nz_ - nz_, std::invalid_argument
    ,"MatrixSparseCOORSerial::get_load_nonzeros_buffers(...) : Error, "
    "The number of nonzeros to load max_nz_load = " << max_nz_load << " can not "
    "be greater than max_nz - nz = " << max_nz_ << " - " << nz_ << " = " << (max_nz_-nz_) <<
    " entries!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    reload_val_only_ && (row_i != NULL || col_j != NULL), std::invalid_argument
    ,"MatrixSparseCOORSerial::get_load_nonzeros_buffers(...) : Error, "
    "reset_to_load_values() was called and therefore the structure of the matrix "
    "can not be set!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    !reload_val_only_  && (row_i == NULL || col_j == NULL), std::invalid_argument
    ,"MatrixSparseCOORSerial::get_load_nonzeros_buffers(...) : Error, "
    "both *row_i and *col_j must be non-NULL since reinitialize() was called" );
#endif
  max_nz_load_ = max_nz_load;
  *val   = val_   + nz_;
  if(!reload_val_only_)
    *row_i = row_i_ + nz_;
  if(!reload_val_only_)
    *col_j = col_j_ + nz_;
}

void MatrixSparseCOORSerial::commit_load_nonzeros_buffers(
  size_type      nz_commit
  ,value_type    **val
  ,index_type    **row_i
  ,index_type    **col_j
  )
{
#ifdef TEUCHOS_DEBUG
  TEUCHOS_TEST_FOR_EXCEPTION(
    max_nz_load_ == 0 , std::logic_error
    ,"MatrixSparseCOORSerial::commit_load_nonzeros_buffers(...) : Error, "
    "You must call get_load_nonzeros_buffers() before calling this method!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    nz_commit > max_nz_load_ , std::logic_error
    ,"MatrixSparseCOORSerial::commit_load_nonzeros_buffers(...) : Error, "
    "You can not commit more nonzero entries than you requested buffer space for in "
    "get_load_nonzeros_buffers(...)!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    *val != val_ + nz_
    , std::logic_error
    ,"MatrixSparseCOORSerial::commit_load_nonzeros_buffers(...) : Error, "
    "This is not the buffer I give you in get_load_nonzeros_buffers(...)!" );
  TEUCHOS_TEST_FOR_EXCEPTION(
    reload_val_only_ && (row_i != NULL || col_j != NULL), std::invalid_argument
    ,"MatrixSparseCOORSerial::commit_load_nonzeros_buffers(...) : Error, "
    "reset_to_load_values() was called and therefore the structure of the matrix "
    "can not be set!" );
#endif
  nz_            += nz_commit;
  max_nz_load_    = 0;
}

void MatrixSparseCOORSerial::finish_construction( bool test_setup )
{
  TEUCHOS_TEST_FOR_EXCEPTION(
    reload_val_only_ == true && reload_val_only_nz_last_ != nz_, std::logic_error
    ,"MatrixSparseCOORSerial::finish_construction() : Error, the number of nonzeros on"
    " the initial load with row and column indexes was = " << reload_val_only_nz_last_ <<
    " and does not agree with the number of nonzero values = " << nz_ << " loaded this time!" );
  if( test_setup ) {
    for( size_type k = 0; k < nz_; ++k ) {
      const index_type
        i = row_i_[k],
        j = col_j_[k];
      TEUCHOS_TEST_FOR_EXCEPTION(
        i < 1 || rows_ < i, std::logic_error
        ,"MatrixSparseCOORSerial::finish_construction(true) : Error, "
        "row_i[" << k << "] = " << i << " is not in the range [1,rows] = [1,"<<rows_<<"]!" );
      TEUCHOS_TEST_FOR_EXCEPTION(
        j < 1 || cols_ < j, std::logic_error
        ,"MatrixSparseCOORSerial::finish_construction(true) : Error, "
        "col_j[" << k << "] = " << j << " is not in the range [1,cols] = [1,"<<cols_<<"]!" );
    }
  }
  space_cols_.initialize(rows_);
  space_rows_.initialize(cols_);
}

// Overridden from MatrixExtractSparseElements

#ifdef TEUCHOS_DEBUG
#define VALIDATE_ROW_COL_IN_RANGE() \
TEUCHOS_TEST_FOR_EXCEPTION( \
  i < 1 || rows_ < i, std::invalid_argument \
  ,err_msg_head<<", i = inv_row_perm[(row_i["<<k<<"]=="<<*row_i<<")-1] = "<<i<<" > rows = "<<rows_ ); \
TEUCHOS_TEST_FOR_EXCEPTION( \
  j < 1 || cols_ < j, std::invalid_argument \
  ,err_msg_head<<", j = inv_col_perm[(col_j["<<k<<"]=="<<*col_j<<")-1] = "<<j<<" > rows = "<<cols_ );
#else
#define VALIDATE_ROW_COL_IN_RANGE()
#endif

index_type MatrixSparseCOORSerial::count_nonzeros(
  EElementUniqueness    element_uniqueness
  ,const index_type     inv_row_perm[]
  ,const index_type     inv_col_perm[]
  ,const Range1D        &row_rng_in
  ,const Range1D        &col_rng_in
  ,index_type           dl
  ,index_type           du
  ) const
{
#ifdef TEUCHOS_DEBUG
  const char err_msg_head[] = "MatrixSparseCOORSerial::count_nonzeros(...): Error";
  TEUCHOS_TEST_FOR_EXCEPTION(
    element_uniqueness_ == ELEMENTS_ASSUME_DUPLICATES_SUM && element_uniqueness == ELEMENTS_FORCE_UNIQUE
    ,std::logic_error
    ,err_msg_head << ", the client requests a count for unique "
    "elements but this sparse matrix object is not allowed to assume this!" );
#endif
  const Range1D
    row_rng = RangePack::full_range(row_rng_in,1,rows_),
    col_rng = RangePack::full_range(col_rng_in,1,rows_),
    row_rng_full(1,rows_),
    col_rng_full(1,cols_);
  const index_type
    *row_i    = row_i_,
    *col_j    = col_j_;
  index_type
    cnt_nz    = 0,
    k         = 0;
  if( dl == -row_rng.ubound() + col_rng.lbound() && du == +col_rng.ubound() - row_rng.lbound() ) {
    // The diagonals are not limiting so we can ignore them
    if( row_rng == row_rng_full && col_rng == col_rng_full ) {
      // The row and column ranges are not limiting either
      cnt_nz = nz_; // Just return the count of all the elements!
    }
    else {
      // The row or column range is limiting
      if( inv_row_perm == NULL && inv_col_perm == NULL ) {
        // Neither the rows nor columns are permuted
        for( k = 0; k < nz_; ++row_i, ++col_j, ++k ) {
          const index_type
            i = (*row_i),
            j = (*col_j);
          VALIDATE_ROW_COL_IN_RANGE();
          cnt_nz += row_rng.in_range(i) && col_rng.in_range(j) ? 1 : 0;
        }
      }
      else if ( inv_row_perm != NULL && inv_col_perm == NULL ) {
        // Only the rows are permuted 
        for( k = 0; k < nz_; ++row_i, ++col_j, ++k ) {
          const index_type
            i = inv_row_perm[(*row_i)-1],
            j = (*col_j);
          VALIDATE_ROW_COL_IN_RANGE();
          cnt_nz += row_rng.in_range(i) && col_rng.in_range(j) ? 1 : 0;
        }
      }
      else if ( inv_row_perm == NULL && inv_col_perm != NULL ) {
        // Only the columns are permuted 
        for( k = 0; k < nz_; ++row_i, ++col_j, ++k ) {
          const index_type
            i = (*row_i),
            j = inv_col_perm[(*col_j)-1];
          VALIDATE_ROW_COL_IN_RANGE();
          cnt_nz += row_rng.in_range(i) && col_rng.in_range(j) ? 1 : 0;
        }
      }
      else {
        // Both the rows and columns are permuted!
        for( k = 0; k < nz_; ++row_i, ++col_j, ++k ) {
          const index_type
            i = inv_row_perm[(*row_i)-1],
            j = inv_col_perm[(*col_j)-1];
          VALIDATE_ROW_COL_IN_RANGE();
          cnt_nz += row_rng.in_range(i) && col_rng.in_range(j) ? 1 : 0;
        }
      }
    }
  }
  else {
    // We have to consider the diagonals dl and du
    TEUCHOS_TEST_FOR_EXCEPT(true); // ToDo: Implement!
  }
  return cnt_nz;
}

void MatrixSparseCOORSerial::coor_extract_nonzeros(
  EElementUniqueness    element_uniqueness
  ,const index_type     inv_row_perm[]
  ,const index_type     inv_col_perm[]
  ,const Range1D        &row_rng_in
  ,const Range1D        &col_rng_in
  ,index_type           dl
  ,index_type           du
  ,value_type           alpha
  ,const index_type     len_Aval
  ,value_type           Aval[]
  ,const index_type     len_Aij
  ,index_type           Arow[]
  ,index_type           Acol[]
  ,const index_type     row_offset
  ,const index_type     col_offset
  ) const
{
#ifdef TEUCHOS_DEBUG
  const char err_msg_head[] = "MatrixSparseCOORSerial::count_nonzeros(...): Error";
  TEUCHOS_TEST_FOR_EXCEPTION(
    element_uniqueness_ == ELEMENTS_ASSUME_DUPLICATES_SUM && element_uniqueness == ELEMENTS_FORCE_UNIQUE
    ,std::logic_error
    ,err_msg_head << ", the client requests extraction of unique "
    "elements but this sparse matrix object can not guarantee this!" );
#endif
  const Range1D
    row_rng = RangePack::full_range(row_rng_in,1,rows_),
    col_rng = RangePack::full_range(col_rng_in,1,rows_),
    row_rng_full(1,rows_),
    col_rng_full(1,cols_);
  value_type
    *val      = val_;
  const index_type
    *row_i    = row_i_,
    *col_j    = col_j_;
  index_type
    cnt_nz    = 0,
    k         = 0;
  if( dl == -row_rng.ubound() + col_rng.lbound() && du == +col_rng.ubound() - row_rng.lbound() ) {
    // The diagonals are not limiting so we can ignore them
    if( row_rng == row_rng_full && col_rng == col_rng_full ) {
      // The row and column ranges are not limiting either
      if( inv_row_perm == NULL && inv_col_perm == NULL ) {
        // We are just extracting the whole, unpermuted matrix
        for( k = 0; k < nz_; ++val, ++row_i, ++col_j, ++k ) {
          ++cnt_nz;
          if( len_Aval )
            *Aval++ = *val;           // ToDo: Split into different loops (no inner if())
          if( len_Aij ) {
            *Arow++ = *row_i + row_offset;
            *Acol++ = *col_j + col_offset;
          }
        }
      }
      else {
        TEUCHOS_TEST_FOR_EXCEPT(true); // ToDo: Implement!
      }
    }
    else {
      // The row or column range is limiting
      if( inv_row_perm == NULL && inv_col_perm == NULL ) {
        // There are no permutations to consider
        for( k = 0; k < nz_; ++val, ++row_i, ++col_j, ++k ) {
          const index_type
            i = (*row_i),
            j = (*col_j);
          VALIDATE_ROW_COL_IN_RANGE();
          if( row_rng.in_range(i) && col_rng.in_range(j) ) {
            ++cnt_nz;
            if( len_Aval )
              *Aval++ = *val;           // ToDo: Split into different loops (no inner if())
            if( len_Aij ) {
              *Arow++ = i + row_offset;
              *Acol++ = j + col_offset;
            }
          }
        }
      }
      else if( inv_row_perm != NULL && inv_col_perm == NULL ) {
        // We must consider only row permutations
        for( k = 0; k < nz_; ++val, ++row_i, ++col_j, ++k ) {
          const index_type
            i = inv_row_perm[(*row_i)-1],
            j = (*col_j);
          VALIDATE_ROW_COL_IN_RANGE();
          if( row_rng.in_range(i) && col_rng.in_range(j) ) {
            ++cnt_nz;
            if( len_Aval )
              *Aval++ = *val;           // ToDo: Split into different loops (no inner if())
            if( len_Aij ) {
              *Arow++ = i + row_offset;
              *Acol++ = j + col_offset;
            }
          }
        }
      }
      else if( inv_row_perm == NULL && inv_col_perm != NULL ) {
        // We must consider only column permutations
        for( k = 0; k < nz_; ++val, ++row_i, ++col_j, ++k ) {
          const index_type
            i = (*row_i),
            j = inv_col_perm[(*col_j)-1];
          VALIDATE_ROW_COL_IN_RANGE();
          if( row_rng.in_range(i) && col_rng.in_range(j) ) {
            ++cnt_nz;
            if( len_Aval )
              *Aval++ = *val;           // ToDo: Split into different loops (no inner if())
            if( len_Aij ) {
              *Arow++ = i + row_offset;
              *Acol++ = j + col_offset;
            }
          }
        }
      }
      else {
        // We must consider row and column permutations
        for( k = 0; k < nz_; ++val, ++row_i, ++col_j, ++k ) {
          const index_type
            i = inv_row_perm[(*row_i)-1],
            j = inv_col_perm[(*col_j)-1];
          VALIDATE_ROW_COL_IN_RANGE();
          if( row_rng.in_range(i) && col_rng.in_range(j) ) {
            ++cnt_nz;
            if( len_Aval )
              *Aval++ = *val;           // ToDo: Split into different loops (no inner if())
            if( len_Aij ) {
              *Arow++ = i + row_offset;
              *Acol++ = j + col_offset;
            }
          }
        }
      }
    }
  }
  else {
    // We have to consider the diagonals dl and du
    TEUCHOS_TEST_FOR_EXCEPT(true); // ToDo: Implement!
  }
  TEUCHOS_TEST_FOR_EXCEPT( !(  len_Aval == 0 || len_Aval == cnt_nz  ) );
  TEUCHOS_TEST_FOR_EXCEPT( !(  len_Aij == 0  || len_Aij  == cnt_nz  ) );
}

// private

void MatrixSparseCOORSerial::make_storage_unique()
{
  if( release_resource_.total_count() > 1 ) {
    TEUCHOS_TEST_FOR_EXCEPT(true); // ToDo: Allocate new storage and copy this memory.
    self_allocate_ = true;
  }
}

} // end namespace AbstractLinAlgPack