GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-25 15:53:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

working version of sparse LU with unsymmetric supernodes and fill-reducing permutation

Browse Source
This commit is contained in:
Desire NUENTSA 2012-07-13 17:32:25 +02:00
parent e529bc9cc1
commit 773804691a
4 changed files with 35 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

75
Eigen/src/SparseLU/SparseLU.h
View File

@ -137,15 +137,16 @@ class SparseLU
EIGEN_STATIC_ASSERT((Dest::Flags&RowMajorBit)==0,
THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
X = B; /* on return, X is overwritten by the computed solution */
int nrhs = B.cols();
Index n = B.rows();
// Permute the right hand side to form Pr*B
X = m_perm_r * X;
// Permute the right hand side to form X = Pr*B
// on return, X is overwritten by the computed solution
X.resize(n,nrhs);
X = m_perm_r * B;
// Forward solve PLy = Pb;
Index n = B.rows();
Index fsupc; // First column of the current supernode
Index istart; // Pointer index to the subscript of the current column
Index nsupr; // Number of rows in the current supernode
@ -324,13 +325,9 @@ void SparseLU<MatrixType, OrderingType>::analyzePattern(const MatrixType& mat)
ord(mat,m_perm_c);
//FIXME Check the right semantic behind m_perm_c
// that is, column j of mat goes to column m_perm_c(j) of mat * m_perm_c;
//DEBUG : Set the natural ordering
for (int i = 0; i < mat.cols(); i++)
m_perm_c.indices()(i) = i;
// Apply the permutation to the column of the input matrix
m_mat = mat * m_perm_c.inverse();
m_mat = mat * m_perm_c.inverse(); //FIXME It should be less expensive here to permute only the structural pattern of the matrix
// Compute the column elimination tree of the permuted matrix
@ -352,15 +349,12 @@ void SparseLU<MatrixType, OrderingType>::analyzePattern(const MatrixType& mat)
m_etree = iwork;
// Postmultiply A*Pc by post, i.e reorder the matrix according to the postorder of the etree
PermutationType post_perm(m); //FIXME Use vector constructor
PermutationType post_perm(m); //FIXME Use directly a constructor with post
for (int i = 0; i < m; i++)
post_perm.indices()(i) = post(i);
// m_mat = m_mat * post_perm.inverse(); // FIXME This should surely be in factorize()
// Composition of the two permutations
m_perm_c = m_perm_c * post_perm;
// Combine the two permutations : postorder the permutation for future use
m_perm_c = post_perm * m_perm_c;
} // end postordering
@ -413,16 +407,11 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
m_mat = matrix * m_perm_c.inverse();
m_mat.makeCompressed();
// DEBUG ... Watch matrix permutation
const int *asub_in = matrix.innerIndexPtr();
const int *colptr_in = matrix.outerIndexPtr();
int * asub = m_mat.innerIndexPtr();
int * colptr = m_mat.outerIndexPtr();
int m = m_mat.rows();
int n = m_mat.cols();
int nnz = m_mat.nonZeros();
int maxpanel = m_panel_size * m;
// Allocate storage common to the factor routines
// Allocate working storage common to the factor routines
int lwork = 0;
int info = LUMemInit(m, n, nnz, lwork, m_fillfactor, m_panel_size, m_glu);
if (info)
@ -432,30 +421,14 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
return ;
}
// Set up pointers for integer working arrays
// int idx = 0;
// VectorBlock<IndexVector> segrep(iwork, idx, m);
// idx += m;
// VectorBlock<IndexVector> parent(iwork, idx, m);
// idx += m;
// VectorBlock<IndexVector> xplore(iwork, idx, m);
// idx += m;
// VectorBlock<IndexVector> repfnz(iwork, idx, maxpanel);
// idx += maxpanel;
// VectorBlock<IndexVector> panel_lsub(iwork, idx, maxpanel);
// idx += maxpanel;
// VectorBlock<IndexVector> xprune(iwork, idx, n);
// idx += n;
// VectorBlock<IndexVector> marker(iwork, idx, m * LU_NO_MARKER);
// Set up pointers for integer working arrays
IndexVector segrep(m);
IndexVector parent(m);
IndexVector xplore(m);
IndexVector segrep(m); segrep.setZero();
IndexVector parent(m); parent.setZero();
IndexVector xplore(m); xplore.setZero();
IndexVector repfnz(maxpanel);
IndexVector panel_lsub(maxpanel);
IndexVector xprune(n); xprune.setZero();
IndexVector marker(m*LU_NO_MARKER);
IndexVector marker(m*LU_NO_MARKER); marker.setZero();
repfnz.setConstant(-1);
panel_lsub.setConstant(-1);
@ -466,10 +439,8 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
ScalarVector tempv;
tempv.setZero(LU_NUM_TEMPV(m, m_panel_size, m_maxsuper, m_rowblk) );
// Setup Permutation vectors
// Compute the inverse of perm_c
// PermutationType iperm_c (m_perm_c.inverse() );
PermutationType iperm_c (m_perm_c);
PermutationType iperm_c(m_perm_c.inverse());
// Identify initial relaxed snodes
IndexVector relax_end(n);
@ -478,11 +449,9 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
else
LU_relax_snode<IndexVector>(n, m_etree, m_relax, marker, relax_end);
//DEBUG
// std::cout<< "relax_end " <<relax_end.transpose() << std::endl;
m_perm_r.resize(m);
m_perm_r.indices().setConstant(-1); //FIXME
m_perm_r.indices().setConstant(-1);
marker.setConstant(-1);
IndexVector& xsup = m_glu.xsup;
@ -493,7 +462,7 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
ScalarVector& lusup = m_glu.lusup;
Index& nzlumax = m_glu.nzlumax;
supno(0) = IND_EMPTY;
supno(0) = IND_EMPTY; xsup.setConstant(0);
xsup(0) = xlsub(0) = xusub(0) = xlusup(0) = Index(0);
// Work on one 'panel' at a time. A panel is one of the following :
@ -552,7 +521,6 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
// Eliminate the current column
info = LU_pivotL(icol, m_diagpivotthresh, m_perm_r.indices(), iperm_c.indices(), pivrow, m_glu);
eigen_assert(info==0 && " SINGULAR MATRIX");
if ( info )
{
m_info = NumericalIssue;
@ -626,7 +594,6 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
// Form the L-segment
info = LU_pivotL(jj, m_diagpivotthresh, m_perm_r.indices(), iperm_c.indices(), pivrow, m_glu);
eigen_assert(info==0 && " SINGULAR MATRIX");
if ( info )
{
std::cerr<< "THE MATRIX IS STRUCTURALLY SINGULAR ... ZERO COLUMN AT " << info <<std::endl;
@ -652,16 +619,14 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
// Count the number of nonzeros in factors
LU_countnz(n, m_nnzL, m_nnzU, m_glu);
// Apply permutation to the L subscripts
LU_fixupL/*<IndexVector, ScalarVector>*/(n, m_perm_r.indices(), m_glu);
LU_fixupL(n, m_perm_r.indices(), m_glu);
// Create supernode matrix L
m_Lstore.setInfos(m, n, m_glu.lusup, m_glu.xlusup, m_glu.lsub, m_glu.xlsub, m_glu.supno, m_glu.xsup);
// Create the column major upper sparse matrix U;
// it is assumed here that MatrixType = SparseMatrix<Scalar,ColumnMajor>
new (&m_Ustore) MappedSparseMatrix<Scalar> ( m, n, m_nnzU, m_glu.xusub.data(), m_glu.usub.data(), m_glu.ucol.data() );
//this.m_Ustore = m_Ustore; //FIXME Is it necessary
m_info = Success;
m_factorizationIsOk = true;

22
Eigen/src/SparseLU/SparseLU_Memory.h
View File

@ -60,12 +60,12 @@
* Expand the existing storage to accomodate more fill-ins
* \param vec Valid pointer to the vector to allocate or expand
* \param [in,out]length At input, contain the current length of the vector that is to be increased. At output, length of the newly allocated vector
* \param [in]len_to_copy Current number of elements in the factors
* \param [in]nbElts Current number of elements in the factors
* \param keep_prev 1: use length and do not expand the vector; 0: compute new_len and expand
* \param [in,out]num_expansions Number of times the memory has been expanded
*/
template <typename VectorType >
int expand(VectorType& vec, int& length, int len_to_copy, int keep_prev, int& num_expansions)
int expand(VectorType& vec, int& length, int nbElts, int keep_prev, int& num_expansions)
{
float alpha = 1.5; // Ratio of the memory increase
@ -77,8 +77,8 @@ int expand(VectorType& vec, int& length, int len_to_copy, int keep_prev, int& n
new_len = alpha * length ;
VectorType old_vec; // Temporary vector to hold the previous values
if (len_to_copy > 0 )
old_vec = vec; // old_vec should be of size len_to_copy... to be checked
if (nbElts > 0 )
old_vec = vec.segment(0,nbElts); // old_vec should be of size nbElts... to be checked
//expand the current vector //FIXME Should be in a try ... catch region
vec.resize(new_len);
@ -107,8 +107,8 @@ int expand(VectorType& vec, int& length, int len_to_copy, int keep_prev, int& n
} // end allocation
//Copy the previous values to the newly allocated space
if (len_to_copy > 0)
vec.segment(0, len_to_copy) = old_vec;
if (nbElts > 0)
vec.segment(0, nbElts) = old_vec;
} // end expansion
length = new_len;
if(num_expansions) ++num_expansions;
@ -137,7 +137,7 @@ int LUMemInit(int m, int n, int annz, int lwork, int fillratio, int panel_size,
Index& nzlmax = glu.nzlmax;
Index& nzumax = glu.nzumax;
Index& nzlumax = glu.nzlumax;
nzumax = nzlumax = fillratio * annz; // estimated number of nonzeros in U
nzumax = nzlumax = std::max(fillratio * annz, m*n); // estimated number of nonzeros in U
nzlmax = std::max(1., fillratio/4.) * annz; // estimated nnz in L factor
// Return the estimated size to the user if necessary
@ -191,18 +191,18 @@ int LUMemInit(int m, int n, int annz, int lwork, int fillratio, int panel_size,
* \brief Expand the existing storage
* \param vec vector to expand
* \param [in,out]maxlen On input, previous size of vec (Number of elements to copy ). on output, new size
* \param next current number of elements in the vector.
* \param nbElts current number of elements in the vector.
* \param glu Global data structure
* \return 0 on success, > 0 size of the memory allocated so far
*/
template <typename VectorType>
int LUMemXpand(VectorType& vec, int& maxlen, int next, LU_MemType memtype, int& num_expansions)
int LUMemXpand(VectorType& vec, int& maxlen, int nbElts, LU_MemType memtype, int& num_expansions)
{
int failed_size;
if (memtype == USUB)
failed_size = expand<VectorType>(vec, maxlen, next, 1, num_expansions);
failed_size = expand<VectorType>(vec, maxlen, nbElts, 1, num_expansions);
else
failed_size = expand<VectorType>(vec, maxlen, next, 0, num_expansions);
failed_size = expand<VectorType>(vec, maxlen, nbElts, 0, num_expansions);
if (failed_size)
return failed_size;

4
Eigen/src/SparseLU/SparseLU_column_dfs.h
View File

@ -44,7 +44,6 @@
*/
#ifndef SPARSELU_COLUMN_DFS_H
#define SPARSELU_COLUMN_DFS_H
/**
* \brief Performs a symbolic factorization on column jcol and decide the supernode boundary
*
@ -234,6 +233,9 @@ int LU_column_dfs(const int m, const int jcol, IndexVector& perm_r, int maxsuper
jptr = xlsub(jcol); // Not yet compressed
jm1ptr = xlsub(jcolm1);
// Use supernodes of type T2 : see SuperLU paper
if ( (nextl-jptr != jptr-jm1ptr-1) ) jsuper = IND_EMPTY;
// Make sure the number of columns in a supernode doesn't
// exceed threshold
if ( (jcol - fsupc) >= maxsuper) jsuper = IND_EMPTY;

2
Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
View File

@ -108,7 +108,7 @@ int LU_copy_to_ucol(const int jcol, const int nseg, SegRepType& segrep, RepfnzTy
for (i = 0; i < segsize; i++)
{
irow = lsub(isub);
usub(nextu) = perm_r(irow); // Unlike teh L part, the U part is stored in its final order
usub(nextu) = perm_r(irow); // Unlike the L part, the U part is stored in its final order
ucol(nextu) = dense(irow);
dense(irow) = Scalar(0.0);
nextu++;