mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Memory expansion and few bugs

Browse Source
This commit is contained in:
Desire NUENTSA 2012-06-06 18:23:39 +02:00
parent 4e5655cc03
commit 268ba3b521
4 changed files with 296 additions and 165 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
Eigen/src/SparseLU/SparseLU.h
View File

@ -412,15 +412,14 @@ void SparseLU::factorize(const MatrixType& matrix)
// Apply permutation to the L subscripts
LU_fixupL(min_mn, m_perm_r, m_Glu);
// Free work space and compress storage iwork and work
// ?? Should it be done automatically by C++
// Free work space iwork and work
//...
// Create supernode matrix L
m_Lstore.setInfos(m, min_mn, nnzL, Glu.lusup, Glu.xlusup, Glu.lsub, Glu.xlsub, Glu.supno; Glu.xsup);
// Create the column major upper sparse matrix U
// Could be great to have the SparseMatrix constructor accepting the CSC matrix pointers
// The Map class can do the job somehow
// ?? Use the MappedSparseMatrix class ??
new (&m_Ustore) Map<SparseMatrix<Scalar, ColumnMajor> > ( m, min_mn, nnzU, Glu.xusub.data(), Glu.usub.data(), Glu.ucol.data() );
m_info = Success;
m_factorizationIsOk = ok;
}

174
Eigen/src/SparseLU/SparseLU_Matrix.h
View File

@ -53,17 +53,21 @@ class SuperNodalMatrix
}
SuperNodalMatrix(Index m, Index n, Index nnz, Scalar *nzval, Index* nzval_colptr, Index* rowind,
Index* rowind_colptr, Index* col_to_sup, Index* sup_to_col ):m_row(m),m_col(n),m_nnz(nnz),
m_nzval(nzval),m_nzval_colptr(nzval_colptr),m_rowind(rowind),
m_rowind_colptr(rowind_colptr),m_col_to_sup(col_to_sup),m_sup_to_col(sup_to_col)
Index* rowind_colptr, Index* col_to_sup, Index* sup_to_col )
{
setInfos(m, n, nnz, nzval, nzval_colptr, rowind, rowind_colptr, col_to_sup, sup_to_col);
}
~SuperNodalMatrix()
{
}
/**
* Set appropriate pointers for the lower triangular supernodal matrix
* These infos are available at the end of the numerical factorization
* FIXME This class will be modified such that it can be use in the course
* of the factorization.
*/
void setInfos(Index m, Index n, Index nnz, Scalar *nzval, Index* nzval_colptr, Index* rowind,
Index* rowind_colptr, Index* col_to_sup, Index* sup_to_col )
{
@ -78,21 +82,80 @@ class SuperNodalMatrix
m_sup_to_col = sup_to_col;
}
SuperNodalMatrix(SparseMatrix& mat);
class InnerIterator
/**
* Number of rows
*/
int rows()
{
public:
protected:
}:
return m_row;
}
/**
* Number of columns
*/
int cols()
{
return m_col;
}
/**
* Return the array of nonzero values packed by column
*
* The size is nnz
*/
Scalar* valuePtr()
{
return m_nzval;
}
/**
* Return the pointers to the beginning of each column in \ref outerIndexPtr()
*/
Index* colIndexPtr()
{
return m_nzval_colptr;
}
/**
* Return the array of compressed row indices of all supernodes
*/
Index* rowIndex()
{
return m_rowind;
}
/**
* Return the location in \em rowvaluePtr() which starts each column
*/
Index* rowIndexPtr()
{
return m_rowind_colptr;
}
/**
* Return the array of column-to-supernode mapping
*/
Index colToSup()
{
return m_col_to_sup;
}
/**
* Return the array of supernode-to-column mapping
*/
Index supToCol()
{
return m_sup_to_col;
}
class InnerIterator;
class SuperNodeIterator;
protected:
Index m_row; // Number of rows
Index m_col; // Number of columns
Index m_nnz; // Number of nonzero values
Index m_nsupper; // Index of the last supernode
Scalar* m_nzval; //array of nonzero values packed by (supernode ??) column
Index m_nsuper; // Number of supernodes
Scalar* m_nzval; //array of nonzero values packed by column
Index* m_nzval_colptr; //nzval_colptr[j] Stores the location in nzval[] which starts column j
Index* m_rowind; // Array of compressed row indices of rectangular supernodes
Index* m_rowind_colptr; //rowind_colptr[j] stores the location in rowind[] which starts column j
@ -100,11 +163,90 @@ class SuperNodalMatrix
Index *m_sup_to_col; //sup_to_col[s] points to the starting column of the s-th supernode
private :
SuperNodalMatrix(SparseMatrix& ) {}
};
SuperNodalMatrix::SuperNodalMatrix(SparseMatrix& mat)
/**
* \brief InnerIterator class to iterate over nonzero values in the triangular supernodal matrix
*
*/
template<typename Scalar, typename Index>
class SuperNodalMatrix::InnerIterator
{
public:
InnerIterator(const SuperNodalMatrix& mat, Index outer)
: m_matrix(mat),
m_outer(outer),
m_idval(mat.colIndexPtr()[outer]),
m_startval(m_idval),
m_endval(mat.colIndexPtr()[outer+1])
m_idrow(mat.rowIndexPtr()[outer]),
m_startidrow(m_idrow),
m_endidrow(mat.rowIndexPtr()[outer+1])
{}
inline InnerIterator& operator++()
{
m_idval++;
m_idrow++ ;
return *this;
}
inline Scalar value() const { return m_matrix.valuePtr()[m_idval]; }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_idval]; }
inline Index index() const { return m_matrix.rowIndex()[m_idrow]; }
inline Index row() const { return index(); }
inline Index col() const { return m_outer; }
inline Index supIndex() const { return m_matrix.colToSup()[m_outer]; }
inline operator bool() const
{
return ( (m_idval < m_endval) && (m_idval > m_startval) &&
(m_idrow < m_endidrow) && (m_idrow > m_startidrow) );
}
protected:
const SuperNodalMatrix& m_matrix; // Supernodal lower triangular matrix
const Index m_outer; // Current column
Index m_idval; //Index to browse the values in the current column
const Index m_startval; // Start of the column value
const Index m_endval; // End of the column value
Index m_idrow; //Index to browse the row indices
const Index m_startidrow; // Start of the row indices of the current column value
const Index m_endidrow; // End of the row indices of the current column value
};
/**
* \brief Iterator class to iterate over nonzeros Supernodes in the triangular supernodal matrix
*
* The final goal is to use this class when dealing with supernodes during numerical factorization
*/
template<typename Scalar, typename Index>
class SuperNodalMatrix::SuperNodeIterator
{
public:
SuperNodeIterator(const SuperNodalMatrix& mat)
{
}
SuperNodeIterator(const SuperNodalMatrix& mat, Index supno)
{
}
/*
* Available Methods :
* Browse all supernodes (operator ++ )
* Number of supernodes
* Columns of the current supernode
* triangular matrix of the current supernode
* rectangular part of the current supernode
*/
protected:
const SuperNodalMatrix& m_matrix; // Supernodal lower triangular matrix
Index m_idsup; // Index to browse all supernodes
const Index m_nsuper; // Number of all supernodes
Index m_startidsup;
Index m_endidsup;
}
};
#endif

243
Eigen/src/SparseLU/SparseLU_Memory.h
View File

@ -52,169 +52,155 @@
+ (w + 1) * m * sizeof(Scalar)
namespace internal {
/* Allocate various working space needed in the numerical factorization phase.
* m_work : space fot the output data structures (lwork is the size)
* m_Glu: persistent data to facilitate multiple factors : is it really necessary ??
/**
* \brief Allocate various working space needed in the numerical factorization phase.
* \param m number of rows of the input matrix
* \param n number of columns
* \param annz number of initial nonzeros in the matrix
* \param work scalar working space needed by all factor routines
* \param iwork Integer working space
* \param lwork if lwork=-1, this routine returns an estimated size of the required memory
* \param Glu persistent data to facilitate multiple factors : will be deleted later ??
* \return an estimated size of the required memory if lwork = -1;
* FIXME should also return the size of actually allocated when memory allocation failed
* NOTE Unlike SuperLU, this routine does not allow the user to provide the size to allocate
* nor it return an estimated amount of space required.
*
* Useful variables :
* - m_fillratio : Ratio of fill expected
* - lwork = -1 : return an estimated size of the required memory
* = 0 : Estimate and allocate the memory
*/
template <typename Scalar,typename Index>
int SparseLU::LUMemInit(int lwork)
template <typename ScalarVector,typename IndexVector>
int SparseLU::LUMemInit(int m, int n, int annz, Scalar *work, Index *iwork, int lwork, int fillratio, GlobalLU_t& Glu)
{
int iword = sizeof(Index);
int dword = sizeof(Scalar);
int n = m_Glu.n = m_mat.cols();
int m = m_mat.rows();
m_Glu.num_expansions = 0; // No memory expansions so far ??
typedef typename ScalarVector::Scalar;
typedef typename IndexVector::Index;
Glu.num_expansions = 0; //No memory expansions so far
if (!Glu.expanders)
Glu.expanders = new ExpHeader(LU_NBR_MEMTYPE);
// Guess the size for L\U factors
int nzlmax, nzumax, nzlumax;
nzumax = nzlumax = m_fillratio * annz; // estimated number of nonzeros in U
nzlmax = std::max(1, m_fill_ratio/4.) * annz; // estimated nnz in L factor
// Return the estimated size to the user if necessary
int estimated_size;
if (!m_Glu.expanders)
m_Glu.expanders = new ExpHeader(NO_MEMTYPE);
if (m_fact_t != SamePattern_SameRowPerm) // Create space for a new factorization
{
// Guess the size for L\U factors
int annz = m_mat.nonZeros();
int nzlmax, nzumax, nzlumax;
nzumax = nzlumax = m_fillratio * annz; // ???
nzlmax = std::max(1, m_fill_ratio/4.) * annz; //???
// Return the estimated size to the user if necessary
if (lwork == IND_EMPTY)
{
estimated_size = LU_GluIntArray(n) * iword + LU_TempSpace(m, m_panel_size)
+ (nzlmax + nzumax) * iword + (nzlumax+nzumax) * dword + n);
return estimated_size;
}
// Setup the required space
// NOTE: In SuperLU, there is an option to let the user provide its own space.
// Allocate Integer pointers for L\U factors.resize(n+1);
m_Glu.supno.resize(n+1);
m_Glu.xlsub.resize(n+1);
m_Glu.xlusup.resize(n+1);
m_Glu.xusub.resize(n+1);
// Reserve memory for L/U factors
m_Glu.lusup = internal::expand<Scalar>(nzlumax, LUSUP, 0, 0, m_Glu);
m_Glu.ucol = internal::expand<Scalar>(nzumax, UCOL, 0, 0, m_Glu);
m_Glu.lsub = internal::expand<Index>(nzlmax, LSUB, 0, 0, m_Glu);
m_Glu.usub = internal::expand<Index>(nzumax, USUB, 0, 1, m_Glu);
// Check if the memory is correctly allocated,
while ( !m_Glu.lusup || !m_Glu.ucol || !m_Glu.lsub || !m_Glu.usub)
{
//otherwise reduce the estimated size and retry
delete [] m_Glu.lusup;
delete [] m_Glu.ucol;
delete [] m_Glu.lsub;
delete [] m_Glu.usub;
nzlumax /= 2;
nzumax /= 2;
nzlmax /= 2;
eigen_assert (nzlumax > annz && "Not enough memory to perform factorization");
m_Glu.lusup = internal::expand<Scalar>(nzlumax, LUSUP, 0, 0, m_Glu);
m_Glu.ucol = internal::expand<Scalar>(nzumax, UCOL, 0, 0, m_Glu);
m_Glu.lsub = internal::expand<Index>(nzlmax, LSUB, 0, 0, m_Glu);
m_Glu.usub = internal::expand<Index>(nzumax, USUB, 0, 1, m_Glu);
}
}
else // m_fact == SamePattern_SameRowPerm;
if (lwork == IND_EMPTY)
{
if (lwork == IND_EMPTY)
{
estimated_size = LU_GluIntArray(n) * iword + LU_TempSpace(m, m_panel_size)
+ (Glu.nzlmax + Glu.nzumax) * iword + (Glu.nzlumax+Glu.nzumax) * dword + n);
return estimated_size;
}
// Use existing space from previous factorization
// Unlike in SuperLU, this should not be necessary here since m_Glu is persistent as a member of the class
m_Glu.xsup = m_Lstore.sup_to_col;
m_Glu.supno = m_Lstore.col_to_sup;
m_Glu.xlsub = m_Lstore.rowind_colptr;
m_Glu.xlusup = m_Lstore.nzval_colptr;
xusub = m_Ustore.outerIndexPtr();
estimated_size = LU_GluIntArray(n) * sizeof(Index) + LU_TempSpace(m, m_panel_size)
+ (nzlmax + nzumax) * sizeof(Index) + (nzlumax+nzumax) * sizeof(Scalar) + n);
return estimated_size;
}
// Setup the required space
// NOTE: In SuperLU, there is an option to let the user provide its own space, unlike here.
// Allocate Integer pointers for L\U factors
Glu.supno = new IndexVector;
Glu.supno->resize(n+1);
Glu.xlsub = new IndexVector;
Glu.xlsub->resize(n+1);
Glu.xlusup = new IndexVector;
Glu.xlusup->resize(n+1);
Glu.xusub = new IndexVector;
Glu.xusub->resize(n+1);
// Reserve memory for L/U factors
Glu.lusup = new ScalarVector;
Glu.ucol = new ScalarVector;
Glu.lsub = new IndexVector;
Glu.usub = new IndexVector;
expand<ScalarVector>(Glu.lusup,nzlumax, LUSUP, 0, 0, Glu);
expand<ScalarVector>(Glu.ucol,nzumax, UCOL, 0, 0, Glu);
expand<IndexVector>(Glu.lsub,nzlmax, LSUB, 0, 0, Glu);
expand<IndexVector>(Glu.usub,nzumax, USUB, 0, 1, Glu);
// Check if the memory is correctly allocated,
// Should be a try... catch section here
while ( !Glu.lusup.size() || !Glu.ucol.size() || !Glu.lsub.size() || !Glu.usub.size())
{
//otherwise reduce the estimated size and retry
// delete [] Glu.lusup;
// delete [] Glu.ucol;
// delete [] Glu.lsub;
// delete [] Glu.usub;
//
nzlumax /= 2;
nzumax /= 2;
nzlmax /= 2;
//FIXME Should be an excpetion here
eigen_assert (nzlumax > annz && "Not enough memory to perform factorization");
m_Glu.expanders[LSUB].size = m_Glu.nzlmax; // Maximum value from previous factorization
m_Glu.expanders[LUSUP].size = m_Glu.nzlumax;
m_Glu.expanders[USUB].size = GLu.nzumax;
m_Glu.expanders[UCOL].size = m_Glu.nzumax;
m_Glu.lsub = GLu.expanders[LSUB].mem = m_Lstore.rowind;
m_Glu.lusup = GLu.expanders[LUSUP].mem = m_Lstore.nzval;
GLu.usub = m_Glu.expanders[USUB].mem = m_Ustore.InnerIndexPtr();
m_Glu.ucol = m_Glu.expanders[UCOL].mem = m_Ustore.valuePtr();
expand<ScalarVector>(Glu.lsup, nzlumax, LUSUP, 0, 0, Glu);
expand<ScalarVector>(Glu.ucol, nzumax, UCOL, 0, 0, Glu);
expand<IndexVector>(Glu.lsub, nzlmax, LSUB, 0, 0, Glu);
expand<IndexVector>(Glu.usub, nzumax, USUB, 0, 1, Glu);
}
// LUWorkInit : Now, allocate known working storage
int isize = (2 * m_panel_size + 3 + LU_NO_MARKER) * m + n;
int dsize = m * m_panel_size + LU_NUM_TEMPV(m, m_panel_size, m_maxsuper, m_rowblk);
m_iwork = new Index(isize);
iwork = new Index(isize);
eigen_assert( (m_iwork != 0) && "Malloc fails for iwork");
m_work = new Scalar(dsize);
work = new Scalar(dsize);
eigen_assert( (m_work != 0) && "Malloc fails for dwork");
++m_Glu.num_expansions;
++Glu.num_expansions;
return 0;
} // end LuMemInit
/**
* Expand the existing storage to accomodate more fill-ins
* \param vec Valid pointer to a vector to allocate or expand
* \param [in,out]prev_len At input, length from previous call. At output, length of the newly allocated vector
* \param type Which part of the memory to expand
* \param len_to_copy Size of the memory to be copied to new store
* \param keep_prev true: use prev_len; Do not expand this vector; false: compute new_len and expand
*/
template <typename DestType >
DestType* SparseLU::expand(int& prev_len, // Length from previous call
MemType type, // Which part of the memory to expand
int len_to_copy, // Size of the memory to be copied to new store
int keep_prev) // = 1: use prev_len; Do not expand this vector
// = 0: compute new_len to expand)
template <typename VectorType >
int SparseLU::expand(VectorType& vec, int& prev_len, MemType type, int len_to_copy, bool keep_prev, GlobalLU_t& Glu)
{
float alpha = 1.5; // Ratio of the memory increase
int new_len; // New size of the allocated memory
if(m_Glu.num_expansions == 0 || keep_prev)
new_len = prev_len;
if(Glu.num_expansions == 0 || keep_prev)
new_len = prev_len; // First time allocate requested
else
new_len = alpha * prev_len;
// Allocate new space
DestType *new_mem, *old_mem;
new_mem = new DestType(new_len);
if ( m_Glu.num_expansions != 0 ) // The memory has been expanded before
// Allocate new space
// vec = new VectorType(new_len);
VectorType old_vec(vec);
if ( Glu.num_expansions != 0 ) // The memory has been expanded before
{
int tries = 0;
vec.resize(new_len); //expand the current vector
if (keep_prev)
{
if (!new_mem) return 0;
if (!vec.size()) return -1 ; // FIXME could throw an exception somehow
}
else
{
while ( !new_mem)
while (!vec.size())
{
// Reduce the size and allocate again
if ( ++tries > 10) return 0;
if ( ++tries > 10) return -1
alpha = LU_Reduce(alpha);
new_len = alpha * prev_len;
new_mem = new DestType(new_len);
vec->resize(new_len);
}
} // keep_prev
} // end allocation
//Copy the previous values to the newly allocated space
ExpHeader<DestType>* expanders = m_Glu.expanders;
std::memcpy(new_mem, expanders[type].mem, len_to_copy);
delete [] expanders[type].mem;
}
expanders[type].mem = new_mem;
expanders[type].size = new_len;
for (int i = 0; i < old_vec.size(); i++)
vec(i) = old_vec(i);
} // end expansion
// expanders[type].mem = vec;
// expanders[type].size = new_len;
prev_len = new_len;
if(m_Glu.num_expansions) ++m_Glu.num_expansions;
return expanders[type].mem;
if(Glu.num_expansions) ++Glu.num_expansions;
return 0;
}
/**
@ -224,15 +210,16 @@ DestType* SparseLU::expand(int& prev_len, // Length from previous call
*
* \return a pointer to the newly allocated space
*/
template <typename DestType>
DestType* SparseLU::LUMemXpand(int jcol, int next, MemType mem_type, int& maxlen)
template <typename VectorType>
VectorType* SparseLU::LUMemXpand(int jcol, int next, MemType mem_type, int& maxlen)
{
DestType *newmem;
VectorType *newmem;
if (memtype == USUB)
new_mem = expand<DestType>(maxlen, mem_type, next, 1);
vec = expand<VectorType>(vec, maxlen, mem_type, next, 1);
else
new_mem = expand<DestType>(maxlen, mem_type, next, 0);
eigen_assert(new_mem && "Can't expand memory"); // FIXME Should be an exception
vec = expand<VectorType>(vec, maxlen, mem_type, next, 0);
// FIXME Should be an exception instead of an assert
eigen_assert(new_mem.size() && "Can't expand memory");
return new_mem;

37
Eigen/src/SparseLU/SparseLU_Structs.h
View File

@ -23,7 +23,7 @@
// Eigen. If not, see <http://www.gnu.org/licenses/>.
/*
* NOTE: Part of this file is the modified version of files slu_[s,d,c,z]defs.h
* NOTE: This file comes from a partly modified version of files slu_[s,d,c,z]defs.h
* -- SuperLU routine (version 4.1) --
* Univ. of California Berkeley, Xerox Palo Alto Research Center,
* and Lawrence Berkeley National Lab.
@ -84,36 +84,39 @@
#define EIGEN_LU_STRUCTS
namespace Eigen {
#define NO_MEMTYPE 4 /* 0: lusup
#define LU_NBR_MEMTYPE 4 /* 0: lusup
1: ucol
2: lsub
3: usub */
typedef enum {NATURAL, MMD_ATA, MMD_AT_PLUS_A, COLAMD, MY_PREMC} colperm_t;
typedef enum {DOFACT, SamePattern, SamePattern_SameRowPerm, Factored} fact_t;
typedef enum {NATURAL, MMD_ATA, MMD_AT_PLUS_A, COLAMD, MY_PERMC} colperm_t;
typedef enum {DOFACT, SamePattern, Factored} fact_t;
typedef enum {LUSUP, UCOL, LSUB, USUB, LLVL, ULVL} MemType;
/** Headers for dynamically managed memory
\tparam BaseType can be int, real scalar or complex scalar*/
template <typename BaseType>
\tparam IndexVectorType can be int, real scalar or complex scalar*/
template <typename VectorType>
struct ExpHeader {
int size; // Length of the memory that has been used */
BaseType *mem;
VectorType *mem; // Save the current pointer of the newly allocated memory
} ExpHeader;
template <typename VectorType, typename Index>
template <typename ScalarVector, typename IndexVector>
struct {
VectorXi xsup; // supernode and column mapping
VectorXi supno; // Supernode number corresponding to this column
VectorXi lsub; // Compressed L subscripts of rectangular supernodes
VectorXi xlsub; // xlsub(j) points to the starting location of the j-th column in lsub
VectorXi xlusup;
VectorXi xusub;
VectorType lusup; // L supernodes
VectorType ucol; // U columns
IndexVector* xsup; //First supernode column ... xsup(s) points to the beginning of the s-th supernode
IndexVector* supno; // Supernode number corresponding to this column (column to supernode mapping)
ScalarVector* lusup; // nonzero values of L ordered by columns
IndexVector* lsub; // Compressed row indices of L rectangular supernodes.
IndexVector* xlusup; // pointers to the beginning of each column in lusup
IndexVector* xlsub; // pointers to the beginning of each column in lsub
Index nzlmax; // Current max size of lsub
Index nzumax; // Current max size of ucol
Index nzlumax; // Current max size of lusup
ScalarVector* ucol; // nonzero values of U ordered by columns
IndexVector* usub; // row indices of U columns in ucol
IndexVector* xusub; // Pointers to the beginning of each column of U in ucol
Index nzumax; // Current max size of ucol
Index n; // Number of columns in the matrix
int num_expansions;
ExpHeader *expanders; // Array of pointers to 4 types of memory
} GlobalLU_t;