numcxx::TSparseMatrix< T > Class Template Reference

Description

template<typename T>
class numcxx::TSparseMatrix< T >

Sparse matrix class using CRS storage scheme.

Examples:

15-numcxx-umfpack.cxx, 40-stationary-heat-fe.cxx, 41-stationary-heat-fv.cxx, 42-convtest-fem-sin.cxx, 43-convtest-fvm-sin.cxx, 44-transient-heat-fe.cxx, 45-convdiff1d.cxx, and 46-nonlin-fvm.cxx.

Definition at line 23 of file tsparsematrix.hxx.

Inheritance diagram for numcxx::TSparseMatrix< T >:

Collaboration diagram for numcxx::TSparseMatrix< T >:

Public Types
typedef T	value_type

Public Member Functions
	TSparseMatrix ()
	TSparseMatrix (index n1, index n2)
	Create an empty sparse matrix representing an n1 x n2 system of linear equations (currently, n1 must be equal to n2) More...
	TSparseMatrix (const std::initializer_list< std::initializer_list< T >> &il)
	Create an empty sparse matrix representing an n x n system of linear equations. More...
T &	operator() (int i, int j)
	Access operator. More...
void	flush ()
	Re-create the internal data structure in order to accomodated all newly created elements. More...
void	clear ()
void	apply (const TArray< T > &U, TArray< T > &V) const
	Apply sparse matrix to vector. More...
std::shared_ptr< TMatrix< T > >	copy_as_dense ()
	Create a dense matrix from the sparse matrix. More...
index	shape (int idim) const
	Return the shape od the matrix. More...
	TSparseMatrix (const TSparseMatrix< T > &A)=delete
	Copy constructor is deleted. More...
std::shared_ptr< TMatrix< T > >	calculate_inverse ()
	Calculate inverse of sparse matrix whuch is dense. More...
bool	pattern_changed () const
	Check if pattern has changed after last solver update. More...

Static Public Member Functions
static std::shared_ptr< TSparseMatrix< T > >	create (index n1, index n2)
	Static wrapper around corresponding constructor. More...
static std::shared_ptr< TSparseMatrix< T > >	create (const std::initializer_list< std::initializer_list< T >> &il)
	Static wrapper around corresponding constructor. More...

Public Attributes
std::shared_ptr< TArray1< int > >	pIA
	Row pointers. More...
std::shared_ptr< TArray1< int > >	pJA
	Column indices. More...
std::shared_ptr< TArray1< T > >	pA
	Entries. More...

Private Member Functions
void	pattern_changed (bool chg)
bool	empty () const

Private Attributes
bool	_pattern_changed =false
const index	n
int	maxrow =0
bool	_first_flush_done =false
std::shared_ptr< Extension >	pExt =nullptr

Friends
class	TSolverUMFPACK< T >
class	TPreconJacobi< T >
class	TPreconILU< T >

Member Typedef Documentation

template<typename T>

typedef T numcxx::TSparseMatrix< T >::value_type

Definition at line 29 of file tsparsematrix.hxx.

Constructor & Destructor Documentation

template<typename T>

numcxx::TSparseMatrix< T >::TSparseMatrix ( )

inline

Definition at line 31 of file tsparsematrix.hxx.

:n(0) {};

template<typename T >

numcxx::TSparseMatrix< T >::TSparseMatrix ( index  n1, index  n2  )

inline

Create an empty sparse matrix representing an n1 x n2 system of linear equations (currently, n1 must be equal to n2)

Definition at line 63 of file tsparsematrix.ixx.

                                                          :
    n(n1),
    pA(std::make_shared<TArray1<T>>(n1)),
    pJA(std::make_shared<TArray1<int>>(n1)),
    pIA(std::make_shared<TArray1<int>>(n1+1))
  {
    if (n1!=n2)
    {
      char errormsg[80];
      snprintf(errormsg,80,"numcxx::TSparseMatrix::TSparseMatrix unexpected non-equal matrix dimensions\n");
      throw std::length_error(errormsg);
    }

        
    auto &IA=*pIA;
    auto &JA=*pJA;
    auto &A=*pA;
    IA=0;
    JA=0;
    A=0.0;
  }

template<typename T >

numcxx::TSparseMatrix< T >::TSparseMatrix ( const std::initializer_list< std::initializer_list< T >> &  il )

inline

Create an empty sparse matrix representing an n x n system of linear equations.

Definition at line 86 of file tsparsematrix.ixx.

                                                                                              :
    TSparseMatrix(il.size(), il.begin()->size())
  {
    index i=0;
    for (auto jl = il.begin() ; jl != il.end(); jl++,i++)
    {
      index j=0;
      for (auto x = jl->begin() ; x != jl->end(); x++,j++) 
        this->operator()(i,j)= *x;

    }
    flush();
  }

Here is the call graph for this function:

template<typename T>

numcxx::TSparseMatrix< T >::TSparseMatrix ( const TSparseMatrix< T > &  A )

delete

Copy constructor is deleted.

Member Function Documentation

template<typename T>

static std::shared_ptr<TSparseMatrix <T> > numcxx::TSparseMatrix< T >::create ( index  n1, index  n2  )

inlinestatic

Static wrapper around corresponding constructor.

Examples:

16-numcxx-umfpack-sharedptr.cxx.

Definition at line 40 of file tsparsematrix.hxx.

{return std::make_shared<TSparseMatrix<T>>(n1,n2);}

Here is the call graph for this function:

template<typename T >

std::shared_ptr< TSparseMatrix< T > > numcxx::TSparseMatrix< T >::create ( const std::initializer_list< std::initializer_list< T >> &  il )

inlinestatic

Static wrapper around corresponding constructor.

Definition at line 101 of file tsparsematrix.ixx.

  {
    return std::make_shared<TSparseMatrix <T>>(il);
  }

template<typename T >

T & numcxx::TSparseMatrix< T >::operator() ( int  i, int  j  )

inline

Access operator.

This operator accesses the sparse matrix element at position i,j. If it is not existent, it is created. All freshly created elements are collected in an extension list which is joined with the main data structure using the flush() method.

Definition at line 226 of file tsparsematrix.ixx.

  {
    auto &IA=*pIA;
    auto &JA=*pJA;
    auto &A=*pA;
#ifdef NUMCXX_CHECK_BOUNDS
    if (i<0||j<0||i>=n||j>=n) 
    {
      char errormsg[80];
      snprintf(errormsg,80,"numcxx::TSparseMatrix::op(): shape=(%d %d) but using index (%d %d)\n",n,n,i,j);
      throw std::out_of_range(errormsg);
    }
#endif
    for (int k=IA[i];k<IA[i+1];k++)
      if (JA[k]==j) 
        return A[k];
        
    if (pExt==nullptr) pExt=std::make_shared<Extension>(n);
    pattern_changed(true);
    return pExt->entry(i,j);
  }

Here is the call graph for this function:

template<typename T >

void numcxx::TSparseMatrix< T >::flush ( )

inline

Re-create the internal data structure in order to accomodated all newly created elements.

Examples:

15-numcxx-umfpack.cxx, and 45-convdiff1d.cxx.

Definition at line 107 of file tsparsematrix.ixx.

  {
    if (pExt==nullptr) return;
    if (pExt->empty()) return;
    int nJA_old=0;
    if (!empty())
      nJA_old=pJA->size();
    int next=pExt->next;
    int nJA_new=nJA_old+next;
    
    auto pNew_IA=TArray1<int>::create(n+1);
    auto pNew_JA=TArray1<int>::create(nJA_new);
    auto pNew_A=TArray1<T>::create(nJA_new);
    
    auto &New_IA=*pNew_IA;
    auto &New_JA=*pNew_JA;
    auto &New_A=*pNew_A;
    
    auto &Old_IA=*pIA;
    auto &Old_JA=*pJA;
    auto &Old_A=*pA;
    
    auto &Ext_IA=*pExt->pIA;
    auto &Ext_JA=*pExt->pJA;
    auto &Ext_A=*pExt->pA;
    
    int maxrow_ext=0;
    for (index i=0;i<n;i++)
    {
      int k0=i;
      int lrow=1;
      for (int k=k0; k>=0;k0=k, k=Ext_IA[k]) lrow++;
      maxrow_ext=std::max(lrow,maxrow_ext);
    }
    
    std::vector<RowEntry> row(maxrow+maxrow_ext+10);
    int New_maxrow=0;
    int j=0;
    int i;
    for(i=0;i<n;i++)
    {
      // put extension entries into row and sort them
      int lxrow=0;
      int k0=i;
      for (int k=k0; k>=0;k0=k, k=Ext_IA[k])
        if (Ext_JA[k]>=0)
        {
          row[lxrow].i=Ext_JA[k];
          row[lxrow].a=Ext_A[k];
          lxrow++;
        }
        
        
      std::sort(row.begin(),row.begin()+lxrow, 
                [](const RowEntry&e1, const RowEntry &e2)-> bool
                {
                  return  (e1.i<e2.i);
                });
        
      // jointly sort Old and New entries into New_JA
      int j0;
      j0=j;
      New_IA[i]=j;
      int irow=0;
      int k=Old_IA[i];
        
      for(;;)
      {
        if (k<Old_IA[i+1] && ((irow>lxrow) ||Old_JA[k]<row[irow].i))
        {
          New_JA[j]=Old_JA[k];
          New_A[j]=Old_A[k];
          k++;
          j++;
          continue;
        }
            
        if (irow<lxrow)
        {
          New_JA[j]=row[irow].i;
          New_A[j]=row[irow].a;
          irow++;
          j++;
          continue;
        }
        break;
      }     
      New_maxrow=std::max(New_maxrow,j-j0);
    }
    New_IA[i]=j;
    
    pIA=pNew_IA;
    pJA=pNew_JA;
    pA=pNew_A;
    maxrow=New_maxrow;
    _pattern_changed=false;
    _first_flush_done=true;
    pExt=nullptr;
  }

Here is the call graph for this function:

template<typename T>

void numcxx::TSparseMatrix< T >::clear ( )

inline

Definition at line 63 of file tsparsematrix.hxx.

{  flush(); (*pIA)=0.0;}

Here is the call graph for this function:

template<typename T >

void numcxx::TSparseMatrix< T >::apply ( const TArray< T > &  U, TArray< T > &  V  ) const

inlinevirtual

Apply sparse matrix to vector.

Reimplemented from numcxx::TLinOperator< T >.

Definition at line 209 of file tsparsematrix.ixx.

  {
    v.resize(u.size());
    auto &IA=*pIA;
    auto &JA=*pJA;
    auto &A=*pA;
    for (int i=0;i<n;i++)
    {
      v[i]=0;
      for (int j=IA[i];j<IA[i+1];j++)
        v[i]+=A[j]*u[JA[j]];
    }
    if (pExt!=nullptr)
      pExt->apply(u,v);
  }

Here is the call graph for this function:

template<typename T >

std::shared_ptr< TMatrix< T > > numcxx::TSparseMatrix< T >::copy_as_dense ( )

inline

Create a dense matrix from the sparse matrix.

Definition at line 249 of file tsparsematrix.ixx.

  {
    flush();
    auto pM=TMatrix<T>::create(n,n);
    
    auto &M=*pM;
    auto &IA=*pIA;
    auto &JA=*pJA;
    auto &A=*pA;
    M=0.0;
    for (int i=0;i<n;i++)
      for (int j=IA[i];j<IA[i+1];j++)
        M(i,JA[j])=A[j];
    return pM;
  }

Here is the call graph for this function:

template<typename T>

index numcxx::TSparseMatrix< T >::shape ( int  idim ) const

inline

Return the shape od the matrix.

Definition at line 72 of file tsparsematrix.hxx.

{return n;}

Here is the call graph for this function:

template<typename T >

std::shared_ptr< TMatrix< T > > numcxx::TSparseMatrix< T >::calculate_inverse ( )

inline

Calculate inverse of sparse matrix whuch is dense.

Definition at line 403 of file tsparsematrix.ixx.

  {
    auto pA=copy_as_dense();  // essentially superfluous, but we seldomly use this
    auto pLU=TSolverLapackLU<T>::create(pA);
    return pLU->calculate_inverse();
  }

Here is the call graph for this function:

template<typename T>

bool numcxx::TSparseMatrix< T >::pattern_changed ( ) const

inline

Check if pattern has changed after last solver update.

Definition at line 100 of file tsparsematrix.hxx.

{return _pattern_changed;}

template<typename T>

void numcxx::TSparseMatrix< T >::pattern_changed ( bool  chg )

inlineprivate

Definition at line 107 of file tsparsematrix.hxx.

{_pattern_changed=chg;};

template<typename T>

bool numcxx::TSparseMatrix< T >::empty ( ) const

inlineprivate

Definition at line 115 of file tsparsematrix.hxx.

{ return !_first_flush_done;};

Friends And Related Function Documentation

template<typename T>

friend class TSolverUMFPACK< T >

friend

Definition at line 25 of file tsparsematrix.hxx.

template<typename T>

friend class TPreconJacobi< T >

friend

Definition at line 26 of file tsparsematrix.hxx.

template<typename T>

friend class TPreconILU< T >

friend

Definition at line 27 of file tsparsematrix.hxx.

Member Data Documentation

template<typename T>

std::shared_ptr<TArray1<int> > numcxx::TSparseMatrix< T >::pIA

Row pointers.

Definition at line 87 of file tsparsematrix.hxx.

template<typename T>

std::shared_ptr<TArray1<int> > numcxx::TSparseMatrix< T >::pJA

Column indices.

Definition at line 90 of file tsparsematrix.hxx.

template<typename T>

std::shared_ptr<TArray1 <T> > numcxx::TSparseMatrix< T >::pA

Entries.

Definition at line 93 of file tsparsematrix.hxx.

template<typename T>

bool numcxx::TSparseMatrix< T >::_pattern_changed =false

private

Definition at line 110 of file tsparsematrix.hxx.

template<typename T>

const index numcxx::TSparseMatrix< T >::n

private

Definition at line 111 of file tsparsematrix.hxx.

template<typename T>

int numcxx::TSparseMatrix< T >::maxrow =0

private

Definition at line 112 of file tsparsematrix.hxx.

template<typename T>

bool numcxx::TSparseMatrix< T >::_first_flush_done =false

private

Definition at line 114 of file tsparsematrix.hxx.

template<typename T>

std::shared_ptr<Extension> numcxx::TSparseMatrix< T >::pExt =nullptr

private

Definition at line 122 of file tsparsematrix.hxx.

---

The documentation for this class was generated from the following files:

• tsparsematrix.hxx
• tsparsematrix.ixx