@smake/eigen 1.0.2 → 1.1.0

package/eigen/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h

@@ -75,12 +75,12 @@ class MappedSuperNodalMatrix
     /**
      * Number of rows
      */
-    Index rows() { return m_row; }
+    Index rows() const { return m_row; }
     
     /**
      * Number of columns
      */
-    Index cols() { return m_col; }
+    Index cols() const { return m_col; }
     
     /**
      * Return the array of nonzero values packed by column
@@ -156,6 +156,9 @@ class MappedSuperNodalMatrix
     class InnerIterator; 
     template<typename Dest>
     void solveInPlace( MatrixBase<Dest>&X) const;
+    template<bool Conjugate, typename Dest>
+    void solveTransposedInPlace( MatrixBase<Dest>&X) const;
+
     
       
       
@@ -294,6 +297,77 @@ void MappedSuperNodalMatrix<Scalar,Index_>::solveInPlace( MatrixBase<Dest>&X) co
     } 
 }
 
+template<typename Scalar, typename Index_>
+template<bool Conjugate, typename Dest>
+void MappedSuperNodalMatrix<Scalar,Index_>::solveTransposedInPlace( MatrixBase<Dest>&X) const
+{
+    using numext::conj;
+  Index n    = int(X.rows());
+  Index nrhs = Index(X.cols());
+  const Scalar * Lval = valuePtr();                 // Nonzero values
+  Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor> work(n, nrhs);     // working vector
+  work.setZero();
+  for (Index k = nsuper(); k >= 0; k--)
+  {
+    Index fsupc = supToCol()[k];                    // First column of the current supernode
+    Index istart = rowIndexPtr()[fsupc];            // Pointer index to the subscript of the current column
+    Index nsupr = rowIndexPtr()[fsupc+1] - istart;  // Number of rows in the current supernode
+    Index nsupc = supToCol()[k+1] - fsupc;          // Number of columns in the current supernode
+    Index nrow = nsupr - nsupc;                     // Number of rows in the non-diagonal part of the supernode
+    Index irow;                                     //Current index row
+
+    if (nsupc == 1 )
+    {
+      for (Index j = 0; j < nrhs; j++)
+      {
+        InnerIterator it(*this, fsupc);
+        ++it; // Skip the diagonal element
+        for (; it; ++it)
+        {
+          irow = it.row();
+          X(fsupc,j) -= X(irow, j) * (Conjugate?conj(it.value()):it.value());
+        }
+      }
+    }
+    else
+    {
+      // The supernode has more than one column
+      Index luptr = colIndexPtr()[fsupc];
+      Index lda = colIndexPtr()[fsupc+1] - luptr;
+
+      //Begin Gather
+      for (Index j = 0; j < nrhs; j++)
+      {
+        Index iptr = istart + nsupc;
+        for (Index i = 0; i < nrow; i++)
+        {
+          irow = rowIndex()[iptr];
+          work.topRows(nrow)(i,j)= X(irow,j); // Gather operation
+          iptr++;
+        }
+      }
+
+      // Matrix-vector product with transposed submatrix
+      Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(Lval[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) );
+      Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
+      if(Conjugate)
+        U = U - A.adjoint() * work.topRows(nrow);
+      else
+        U = U - A.transpose() * work.topRows(nrow);
+
+      // Triangular solve (of transposed diagonal block)
+      new (&A) Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > ( &(Lval[luptr]), nsupc, nsupc, OuterStride<>(lda) );
+      if(Conjugate)
+        U = A.adjoint().template triangularView<UnitUpper>().solve(U);
+      else
+        U = A.transpose().template triangularView<UnitUpper>().solve(U);
+
+    }
+
+  }
+}
+
+
 } // end namespace internal
 
 } // end namespace Eigen

package/eigen/Eigen/src/SparseLU/SparseLU_column_dfs.h

@@ -151,7 +151,7 @@ Index SparseLUImpl<Scalar,StorageIndex>::column_dfs(const Index m, const Index j
         StorageIndex ito = glu.xlsub(fsupc+1);
         glu.xlsub(jcolm1) = ito; 
         StorageIndex istop = ito + jptr - jm1ptr; 
-        xprune(jcolm1) = istop; // intialize xprune(jcol-1)
+        xprune(jcolm1) = istop; // initialize xprune(jcol-1)
         glu.xlsub(jcol) = istop; 
         
         for (StorageIndex ifrom = jm1ptr; ifrom < nextl; ++ifrom, ++ito)
@@ -166,7 +166,7 @@ Index SparseLUImpl<Scalar,StorageIndex>::column_dfs(const Index m, const Index j
   // Tidy up the pointers before exit
   glu.xsup(nsuper+1) = jcolp1; 
   glu.supno(jcolp1) = nsuper; 
-  xprune(jcol) = StorageIndex(nextl);  // Intialize upper bound for pruning
+  xprune(jcol) = StorageIndex(nextl);  // Initialize upper bound for pruning
   glu.xlsub(jcolp1) = StorageIndex(nextl); 
   
   return 0; 

package/eigen/Eigen/src/SparseLU/SparseLU_gemm_kernel.h

@@ -215,7 +215,7 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize);  }\
                    pstore(C0+i+(I)*PacketSize, c0);
         
-        // agressive vectorization and peeling
+        // aggressive vectorization and peeling
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
         {
           EIGEN_ASM_COMMENT("SPARSELU_GEMML_KERNEL2");

package/eigen/Eigen/src/SparseLU/SparseLU_panel_bmod.h

@@ -38,7 +38,7 @@ namespace internal {
  * \brief Performs numeric block updates (sup-panel) in topological order.
  * 
  * Before entering this routine, the original nonzeros in the panel
- * were already copied i nto the spa[m,w]
+ * were already copied into the spa[m,w]
  * 
  * \param m number of rows in the matrix
  * \param w Panel size

package/eigen/Eigen/src/SparseQR/SparseQR.h

@@ -41,15 +41,16 @@ namespace internal {
 /**
   * \ingroup SparseQR_Module
   * \class SparseQR
-  * \brief Sparse left-looking rank-revealing QR factorization
+  * \brief Sparse left-looking QR factorization with numerical column pivoting
   * 
-  * This class implements a left-looking rank-revealing QR decomposition 
-  * of sparse matrices. When a column has a norm less than a given tolerance
+  * This class implements a left-looking QR decomposition of sparse matrices
+  * with numerical column pivoting.
+  * When a column has a norm less than a given tolerance
   * it is implicitly permuted to the end. The QR factorization thus obtained is 
   * given by A*P = Q*R where R is upper triangular or trapezoidal. 
   * 
   * P is the column permutation which is the product of the fill-reducing and the
-  * rank-revealing permutations. Use colsPermutation() to get it.
+  * numerical permutations. Use colsPermutation() to get it.
   * 
   * Q is the orthogonal matrix represented as products of Householder reflectors. 
   * Use matrixQ() to get an expression and matrixQ().adjoint() to get the adjoint.
@@ -64,6 +65,17 @@ namespace internal {
   * 
   * \implsparsesolverconcept
   *
+  * The numerical pivoting strategy and default threshold are the same as in SuiteSparse QR, and
+  * detailed in the following paper:
+  * <i>
+  * Tim Davis, "Algorithm 915, SuiteSparseQR: Multifrontal Multithreaded Rank-Revealing
+  * Sparse QR Factorization, ACM Trans. on Math. Soft. 38(1), 2011.
+  * </i>
+  * Even though it is qualified as "rank-revealing", this strategy might fail for some 
+  * rank deficient problems. When this class is used to solve linear or least-square problems
+  * it is thus strongly recommended to check the accuracy of the computed solution. If it
+  * failed, it usually helps to increase the threshold with setPivotThreshold.
+  * 
   * \warning The input sparse matrix A must be in compressed mode (see SparseMatrix::makeCompressed()).
   * \warning For complex matrices matrixQ().transpose() will actually return the adjoint matrix.
   * 
@@ -331,7 +343,7 @@ void SparseQR<MatrixType,OrderingType>::analyzePattern(const MatrixType& mat)
   m_R.resize(m, n);
   m_Q.resize(m, diagSize);
   
-  // Allocate space for nonzero elements : rough estimation
+  // Allocate space for nonzero elements: rough estimation
   m_R.reserve(2*mat.nonZeros()); //FIXME Get a more accurate estimation through symbolic factorization with the etree
   m_Q.reserve(2*mat.nonZeros());
   m_hcoeffs.resize(diagSize);
@@ -640,7 +652,8 @@ struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived
       // Compute res = Q * other column by column
       for(Index j = 0; j < res.cols(); j++)
       {
-        for (Index k = diagSize-1; k >=0; k--)
+        Index start_k = internal::is_identity<Derived>::value ? numext::mini(j,diagSize-1) : diagSize-1;
+        for (Index k = start_k; k >=0; k--)
         {
           Scalar tau = Scalar(0);
           tau = m_qr.m_Q.col(k).dot(res.col(j));

package/eigen/Eigen/src/StlSupport/StdDeque.h

@@ -36,7 +36,7 @@ namespace std \
     deque(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : deque_base(first, last, a) {} \
     deque(const deque& c) : deque_base(c) {}  \
     explicit deque(size_type num, const value_type& val = value_type()) : deque_base(num, val) {} \
-    deque(iterator start, iterator end) : deque_base(start, end) {}  \
+    deque(iterator start_, iterator end_) : deque_base(start_, end_) {}  \
     deque& operator=(const deque& x) {  \
       deque_base::operator=(x);  \
       return *this;  \
@@ -62,7 +62,7 @@ namespace std {
     : deque_base(first, last, a) {} \
     deque(const deque& c) : deque_base(c) {}  \
     explicit deque(size_type num, const value_type& val = value_type()) : deque_base(num, val) {} \
-    deque(iterator start, iterator end) : deque_base(start, end) {}  \
+    deque(iterator start_, iterator end_) : deque_base(start_, end_) {}  \
     deque& operator=(const deque& x) {  \
       deque_base::operator=(x);  \
       return *this;  \
@@ -98,19 +98,7 @@ namespace std {
   { return deque_base::insert(position,x); }
   void insert(const_iterator position, size_type new_size, const value_type& x)
   { deque_base::insert(position, new_size, x); }
-#elif defined(_GLIBCXX_DEQUE) && EIGEN_GNUC_AT_LEAST(4,2) && !EIGEN_GNUC_AT_LEAST(10, 1)
-  // workaround GCC std::deque implementation
-  // GCC 10.1 doesn't let us access _Deque_impl _M_impl anymore and we have to
-  // fall-back to the default case
-  void resize(size_type new_size, const value_type& x)
-  {
-    if (new_size < deque_base::size())
-      deque_base::_M_erase_at_end(this->_M_impl._M_start + new_size);
-    else
-      deque_base::insert(deque_base::end(), new_size - deque_base::size(), x);
-  }
 #else
-  // either non-GCC or GCC between 4.1 and 10.1
   // default implementation which should always work.
   void resize(size_type new_size, const value_type& x)
   {

package/eigen/Eigen/src/StlSupport/StdList.h

@@ -35,7 +35,7 @@ namespace std \
     list(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : list_base(first, last, a) {} \
     list(const list& c) : list_base(c) {}  \
     explicit list(size_type num, const value_type& val = value_type()) : list_base(num, val) {} \
-    list(iterator start, iterator end) : list_base(start, end) {}  \
+    list(iterator start_, iterator end_) : list_base(start_, end_) {}  \
     list& operator=(const list& x) {  \
       list_base::operator=(x);  \
       return *this;  \
@@ -62,7 +62,7 @@ namespace std
     : list_base(first, last, a) {} \
     list(const list& c) : list_base(c) {}  \
     explicit list(size_type num, const value_type& val = value_type()) : list_base(num, val) {} \
-    list(iterator start, iterator end) : list_base(start, end) {}  \
+    list(iterator start_, iterator end_) : list_base(start_, end_) {}  \
     list& operator=(const list& x) {  \
     list_base::operator=(x);  \
     return *this; \

package/eigen/Eigen/src/StlSupport/StdVector.h

@@ -36,7 +36,7 @@ namespace std \
     vector(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : vector_base(first, last, a) {} \
     vector(const vector& c) : vector_base(c) {}  \
     explicit vector(size_type num, const value_type& val = value_type()) : vector_base(num, val) {} \
-    vector(iterator start, iterator end) : vector_base(start, end) {}  \
+    vector(iterator start_, iterator end_) : vector_base(start_, end_) {}  \
     vector& operator=(const vector& x) {  \
       vector_base::operator=(x);  \
       return *this;  \
@@ -62,7 +62,7 @@ namespace std {
     : vector_base(first, last, a) {} \
     vector(const vector& c) : vector_base(c) {}  \
     explicit vector(size_type num, const value_type& val = value_type()) : vector_base(num, val) {} \
-    vector(iterator start, iterator end) : vector_base(start, end) {}  \
+    vector(iterator start_, iterator end_) : vector_base(start_, end_) {}  \
     vector& operator=(const vector& x) {  \
       vector_base::operator=(x);  \
       return *this;  \

package/eigen/Eigen/src/SuperLUSupport/SuperLUSupport.h

@@ -217,12 +217,12 @@ struct SluMatrix : SuperMatrix
     res.setScalarType<typename MatrixType::Scalar>();
 
     // FIXME the following is not very accurate
-    if (MatrixType::Flags & Upper)
+    if (int(MatrixType::Flags) & int(Upper))
       res.Mtype = SLU_TRU;
-    if (MatrixType::Flags & Lower)
+    if (int(MatrixType::Flags) & int(Lower))
       res.Mtype = SLU_TRL;
 
-    eigen_assert(((MatrixType::Flags & SelfAdjoint)==0) && "SelfAdjoint matrix shape not supported by SuperLU");
+    eigen_assert(((int(MatrixType::Flags) & int(SelfAdjoint))==0) && "SelfAdjoint matrix shape not supported by SuperLU");
 
     return res;
   }
@@ -352,7 +352,7 @@ class SuperLUBase : public SparseSolverBase<Derived>
     
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was succesful,
+      * \returns \c Success if computation was successful,
       *          \c NumericalIssue if the matrix.appears to be negative.
       */
     ComputationInfo info() const
@@ -650,9 +650,8 @@ void SuperLU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
-  const Index size = m_matrix.rows();
   const Index rhsCols = b.cols();
-  eigen_assert(size==b.rows());
+  eigen_assert(m_matrix.rows()==b.rows());
 
   m_sluOptions.Trans = NOTRANS;
   m_sluOptions.Fact = FACTORED;
@@ -974,9 +973,8 @@ void SuperILU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
-  const int size = m_matrix.rows();
   const int rhsCols = b.cols();
-  eigen_assert(size==b.rows());
+  eigen_assert(m_matrix.rows()==b.rows());
 
   m_sluOptions.Trans = NOTRANS;
   m_sluOptions.Fact = FACTORED;