@smake/eigen 1.1.0 → 1.1.1

package/eigen/Eigen/src/LU/FullPivLU.h

@@ -10,487 +10,469 @@
 #ifndef EIGEN_LU_H
 #define EIGEN_LU_H
 
+// IWYU pragma: private
+#include "./InternalHeaderCheck.h"
+
 namespace Eigen {
 
 namespace internal {
-template<typename _MatrixType> struct traits<FullPivLU<_MatrixType> >
- : traits<_MatrixType>
-{
+template <typename MatrixType_, typename PermutationIndex_>
+struct traits<FullPivLU<MatrixType_, PermutationIndex_> > : traits<MatrixType_> {
   typedef MatrixXpr XprKind;
   typedef SolverStorage StorageKind;
-  typedef int StorageIndex;
+  typedef PermutationIndex_ StorageIndex;
   enum { Flags = 0 };
 };
 
-} // end namespace internal
+}  // end namespace internal
 
 /** \ingroup LU_Module
-  *
-  * \class FullPivLU
-  *
-  * \brief LU decomposition of a matrix with complete pivoting, and related features
-  *
-  * \tparam _MatrixType the type of the matrix of which we are computing the LU decomposition
-  *
-  * This class represents a LU decomposition of any matrix, with complete pivoting: the matrix A is
-  * decomposed as \f$ A = P^{-1} L U Q^{-1} \f$ where L is unit-lower-triangular, U is
-  * upper-triangular, and P and Q are permutation matrices. This is a rank-revealing LU
-  * decomposition. The eigenvalues (diagonal coefficients) of U are sorted in such a way that any
-  * zeros are at the end.
-  *
-  * This decomposition provides the generic approach to solving systems of linear equations, computing
-  * the rank, invertibility, inverse, kernel, and determinant.
-  *
-  * This LU decomposition is very stable and well tested with large matrices. However there are use cases where the SVD
-  * decomposition is inherently more stable and/or flexible. For example, when computing the kernel of a matrix,
-  * working with the SVD allows to select the smallest singular values of the matrix, something that
-  * the LU decomposition doesn't see.
-  *
-  * The data of the LU decomposition can be directly accessed through the methods matrixLU(),
-  * permutationP(), permutationQ().
-  *
-  * As an example, here is how the original matrix can be retrieved:
-  * \include class_FullPivLU.cpp
-  * Output: \verbinclude class_FullPivLU.out
-  *
-  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
-  *
-  * \sa MatrixBase::fullPivLu(), MatrixBase::determinant(), MatrixBase::inverse()
-  */
-template<typename _MatrixType> class FullPivLU
-  : public SolverBase<FullPivLU<_MatrixType> >
-{
-  public:
-    typedef _MatrixType MatrixType;
-    typedef SolverBase<FullPivLU> Base;
-    friend class SolverBase<FullPivLU>;
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(FullPivLU)
-    enum {
-      MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
-    };
-    typedef typename internal::plain_row_type<MatrixType, StorageIndex>::type IntRowVectorType;
-    typedef typename internal::plain_col_type<MatrixType, StorageIndex>::type IntColVectorType;
-    typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationQType;
-    typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationPType;
-    typedef typename MatrixType::PlainObject PlainObject;
-
-    /**
-      * \brief Default Constructor.
-      *
-      * The default constructor is useful in cases in which the user intends to
-      * perform decompositions via LU::compute(const MatrixType&).
-      */
-    FullPivLU();
-
-    /** \brief Default Constructor with memory preallocation
-      *
-      * Like the default constructor but with preallocation of the internal data
-      * according to the specified problem \a size.
-      * \sa FullPivLU()
-      */
-    FullPivLU(Index rows, Index cols);
-
-    /** Constructor.
-      *
-      * \param matrix the matrix of which to compute the LU decomposition.
-      *               It is required to be nonzero.
-      */
-    template<typename InputType>
-    explicit FullPivLU(const EigenBase<InputType>& matrix);
-
-    /** \brief Constructs a LU factorization from a given matrix
-      *
-      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
-      *
-      * \sa FullPivLU(const EigenBase&)
-      */
-    template<typename InputType>
-    explicit FullPivLU(EigenBase<InputType>& matrix);
-
-    /** Computes the LU decomposition of the given matrix.
-      *
-      * \param matrix the matrix of which to compute the LU decomposition.
-      *               It is required to be nonzero.
-      *
-      * \returns a reference to *this
-      */
-    template<typename InputType>
-    FullPivLU& compute(const EigenBase<InputType>& matrix) {
-      m_lu = matrix.derived();
-      computeInPlace();
-      return *this;
-    }
+ *
+ * \class FullPivLU
+ *
+ * \brief LU decomposition of a matrix with complete pivoting, and related features
+ *
+ * \tparam MatrixType_ the type of the matrix of which we are computing the LU decomposition
+ *
+ * This class represents a LU decomposition of any matrix, with complete pivoting: the matrix A is
+ * decomposed as \f$ A = P^{-1} L U Q^{-1} \f$ where L is unit-lower-triangular, U is
+ * upper-triangular, and P and Q are permutation matrices. This is a rank-revealing LU
+ * decomposition. The eigenvalues (diagonal coefficients) of U are sorted in such a way that any
+ * zeros are at the end.
+ *
+ * This decomposition provides the generic approach to solving systems of linear equations, computing
+ * the rank, invertibility, inverse, kernel, and determinant.
+ *
+ * This LU decomposition is very stable and well tested with large matrices. However there are use cases where the SVD
+ * decomposition is inherently more stable and/or flexible. For example, when computing the kernel of a matrix,
+ * working with the SVD allows to select the smallest singular values of the matrix, something that
+ * the LU decomposition doesn't see.
+ *
+ * The data of the LU decomposition can be directly accessed through the methods matrixLU(),
+ * permutationP(), permutationQ().
+ *
+ * As an example, here is how the original matrix can be retrieved:
+ * \include class_FullPivLU.cpp
+ * Output: \verbinclude class_FullPivLU.out
+ *
+ * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+ *
+ * \sa MatrixBase::fullPivLu(), MatrixBase::determinant(), MatrixBase::inverse()
+ */
+template <typename MatrixType_, typename PermutationIndex_>
+class FullPivLU : public SolverBase<FullPivLU<MatrixType_, PermutationIndex_> > {
+ public:
+  typedef MatrixType_ MatrixType;
+  typedef SolverBase<FullPivLU> Base;
+  friend class SolverBase<FullPivLU>;
+
+  EIGEN_GENERIC_PUBLIC_INTERFACE(FullPivLU)
+  enum {
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+  };
+  using PermutationIndex = PermutationIndex_;
+  typedef typename internal::plain_row_type<MatrixType, PermutationIndex>::type IntRowVectorType;
+  typedef typename internal::plain_col_type<MatrixType, PermutationIndex>::type IntColVectorType;
+  typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime, PermutationIndex> PermutationQType;
+  typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime, PermutationIndex> PermutationPType;
+  typedef typename MatrixType::PlainObject PlainObject;
+
+  /** \brief Reports whether the LU factorization was successful.
+   *
+   * \note This function always returns \c Success. It is provided for compatibility
+   * with other factorization routines.
+   * \returns \c Success
+   */
+  ComputationInfo info() const {
+    eigen_assert(m_isInitialized && "FullPivLU is not initialized.");
+    return Success;
+  }
 
-    /** \returns the LU decomposition matrix: the upper-triangular part is U, the
-      * unit-lower-triangular part is L (at least for square matrices; in the non-square
-      * case, special care is needed, see the documentation of class FullPivLU).
-      *
-      * \sa matrixL(), matrixU()
-      */
-    inline const MatrixType& matrixLU() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return m_lu;
-    }
+  /**
+   * \brief Default Constructor.
+   *
+   * The default constructor is useful in cases in which the user intends to
+   * perform decompositions via LU::compute(const MatrixType&).
+   */
+  FullPivLU();
 
-    /** \returns the number of nonzero pivots in the LU decomposition.
-      * Here nonzero is meant in the exact sense, not in a fuzzy sense.
-      * So that notion isn't really intrinsically interesting, but it is
-      * still useful when implementing algorithms.
-      *
-      * \sa rank()
-      */
-    inline Index nonzeroPivots() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return m_nonzero_pivots;
-    }
+  /** \brief Default Constructor with memory preallocation
+   *
+   * Like the default constructor but with preallocation of the internal data
+   * according to the specified problem \a size.
+   * \sa FullPivLU()
+   */
+  FullPivLU(Index rows, Index cols);
 
-    /** \returns the absolute value of the biggest pivot, i.e. the biggest
-      *          diagonal coefficient of U.
-      */
-    RealScalar maxPivot() const { return m_maxpivot; }
-
-    /** \returns the permutation matrix P
-      *
-      * \sa permutationQ()
-      */
-    EIGEN_DEVICE_FUNC inline const PermutationPType& permutationP() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return m_p;
-    }
+  /** Constructor.
+   *
+   * \param matrix the matrix of which to compute the LU decomposition.
+   *               It is required to be nonzero.
+   */
+  template <typename InputType>
+  explicit FullPivLU(const EigenBase<InputType>& matrix);
+
+  /** \brief Constructs a LU factorization from a given matrix
+   *
+   * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c
+   * MatrixType is a Eigen::Ref.
+   *
+   * \sa FullPivLU(const EigenBase&)
+   */
+  template <typename InputType>
+  explicit FullPivLU(EigenBase<InputType>& matrix);
+
+  /** Computes the LU decomposition of the given matrix.
+   *
+   * \param matrix the matrix of which to compute the LU decomposition.
+   *               It is required to be nonzero.
+   *
+   * \returns a reference to *this
+   */
+  template <typename InputType>
+  FullPivLU& compute(const EigenBase<InputType>& matrix) {
+    m_lu = matrix.derived();
+    computeInPlace();
+    return *this;
+  }
 
-    /** \returns the permutation matrix Q
-      *
-      * \sa permutationP()
-      */
-    inline const PermutationQType& permutationQ() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return m_q;
-    }
+  /** \returns the LU decomposition matrix: the upper-triangular part is U, the
+   * unit-lower-triangular part is L (at least for square matrices; in the non-square
+   * case, special care is needed, see the documentation of class FullPivLU).
+   *
+   * \sa matrixL(), matrixU()
+   */
+  inline const MatrixType& matrixLU() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return m_lu;
+  }
 
-    /** \returns the kernel of the matrix, also called its null-space. The columns of the returned matrix
-      * will form a basis of the kernel.
-      *
-      * \note If the kernel has dimension zero, then the returned matrix is a column-vector filled with zeros.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      *
-      * Example: \include FullPivLU_kernel.cpp
-      * Output: \verbinclude FullPivLU_kernel.out
-      *
-      * \sa image()
-      */
-    inline const internal::kernel_retval<FullPivLU> kernel() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return internal::kernel_retval<FullPivLU>(*this);
-    }
+  /** \returns the number of nonzero pivots in the LU decomposition.
+   * Here nonzero is meant in the exact sense, not in a fuzzy sense.
+   * So that notion isn't really intrinsically interesting, but it is
+   * still useful when implementing algorithms.
+   *
+   * \sa rank()
+   */
+  inline Index nonzeroPivots() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return m_nonzero_pivots;
+  }
 
-    /** \returns the image of the matrix, also called its column-space. The columns of the returned matrix
-      * will form a basis of the image (column-space).
-      *
-      * \param originalMatrix the original matrix, of which *this is the LU decomposition.
-      *                       The reason why it is needed to pass it here, is that this allows
-      *                       a large optimization, as otherwise this method would need to reconstruct it
-      *                       from the LU decomposition.
-      *
-      * \note If the image has dimension zero, then the returned matrix is a column-vector filled with zeros.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      *
-      * Example: \include FullPivLU_image.cpp
-      * Output: \verbinclude FullPivLU_image.out
-      *
-      * \sa kernel()
-      */
-    inline const internal::image_retval<FullPivLU>
-      image(const MatrixType& originalMatrix) const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return internal::image_retval<FullPivLU>(*this, originalMatrix);
-    }
+  /** \returns the absolute value of the biggest pivot, i.e. the biggest
+   *          diagonal coefficient of U.
+   */
+  RealScalar maxPivot() const { return m_maxpivot; }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    /** \return a solution x to the equation Ax=b, where A is the matrix of which
-      * *this is the LU decomposition.
-      *
-      * \param b the right-hand-side of the equation to solve. Can be a vector or a matrix,
-      *          the only requirement in order for the equation to make sense is that
-      *          b.rows()==A.rows(), where A is the matrix of which *this is the LU decomposition.
-      *
-      * \returns a solution.
-      *
-      * \note_about_checking_solutions
-      *
-      * \note_about_arbitrary_choice_of_solution
-      * \note_about_using_kernel_to_study_multiple_solutions
-      *
-      * Example: \include FullPivLU_solve.cpp
-      * Output: \verbinclude FullPivLU_solve.out
-      *
-      * \sa TriangularView::solve(), kernel(), inverse()
-      */
-    template<typename Rhs>
-    inline const Solve<FullPivLU, Rhs>
-    solve(const MatrixBase<Rhs>& b) const;
-    #endif
-
-    /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
-        the LU decomposition.
-      */
-    inline RealScalar rcond() const
-    {
-      eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
-      return internal::rcond_estimate_helper(m_l1_norm, *this);
-    }
+  /** \returns the permutation matrix P
+   *
+   * \sa permutationQ()
+   */
+  EIGEN_DEVICE_FUNC inline const PermutationPType& permutationP() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return m_p;
+  }
 
-    /** \returns the determinant of the matrix of which
-      * *this is the LU decomposition. It has only linear complexity
-      * (that is, O(n) where n is the dimension of the square matrix)
-      * as the LU decomposition has already been computed.
-      *
-      * \note This is only for square matrices.
-      *
-      * \note For fixed-size matrices of size up to 4, MatrixBase::determinant() offers
-      *       optimized paths.
-      *
-      * \warning a determinant can be very big or small, so for matrices
-      * of large enough dimension, there is a risk of overflow/underflow.
-      *
-      * \sa MatrixBase::determinant()
-      */
-    typename internal::traits<MatrixType>::Scalar determinant() const;
-
-    /** Allows to prescribe a threshold to be used by certain methods, such as rank(),
-      * who need to determine when pivots are to be considered nonzero. This is not used for the
-      * LU decomposition itself.
-      *
-      * When it needs to get the threshold value, Eigen calls threshold(). By default, this
-      * uses a formula to automatically determine a reasonable threshold.
-      * Once you have called the present method setThreshold(const RealScalar&),
-      * your value is used instead.
-      *
-      * \param threshold The new value to use as the threshold.
-      *
-      * A pivot will be considered nonzero if its absolute value is strictly greater than
-      *  \f$ \vert pivot \vert \leqslant threshold \times \vert maxpivot \vert \f$
-      * where maxpivot is the biggest pivot.
-      *
-      * If you want to come back to the default behavior, call setThreshold(Default_t)
-      */
-    FullPivLU& setThreshold(const RealScalar& threshold)
-    {
-      m_usePrescribedThreshold = true;
-      m_prescribedThreshold = threshold;
-      return *this;
-    }
+  /** \returns the permutation matrix Q
+   *
+   * \sa permutationP()
+   */
+  inline const PermutationQType& permutationQ() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return m_q;
+  }
 
-    /** Allows to come back to the default behavior, letting Eigen use its default formula for
-      * determining the threshold.
-      *
-      * You should pass the special object Eigen::Default as parameter here.
-      * \code lu.setThreshold(Eigen::Default); \endcode
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    FullPivLU& setThreshold(Default_t)
-    {
-      m_usePrescribedThreshold = false;
-      return *this;
-    }
+  /** \returns the kernel of the matrix, also called its null-space. The columns of the returned matrix
+   * will form a basis of the kernel.
+   *
+   * \note If the kernel has dimension zero, then the returned matrix is a column-vector filled with zeros.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   *
+   * Example: \include FullPivLU_kernel.cpp
+   * Output: \verbinclude FullPivLU_kernel.out
+   *
+   * \sa image()
+   */
+  inline const internal::kernel_retval<FullPivLU> kernel() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return internal::kernel_retval<FullPivLU>(*this);
+  }
 
-    /** Returns the threshold that will be used by certain methods such as rank().
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    RealScalar threshold() const
-    {
-      eigen_assert(m_isInitialized || m_usePrescribedThreshold);
-      return m_usePrescribedThreshold ? m_prescribedThreshold
-      // this formula comes from experimenting (see "LU precision tuning" thread on the list)
-      // and turns out to be identical to Higham's formula used already in LDLt.
-          : NumTraits<Scalar>::epsilon() * RealScalar(m_lu.diagonalSize());
-    }
+  /** \returns the image of the matrix, also called its column-space. The columns of the returned matrix
+   * will form a basis of the image (column-space).
+   *
+   * \param originalMatrix the original matrix, of which *this is the LU decomposition.
+   *                       The reason why it is needed to pass it here, is that this allows
+   *                       a large optimization, as otherwise this method would need to reconstruct it
+   *                       from the LU decomposition.
+   *
+   * \note If the image has dimension zero, then the returned matrix is a column-vector filled with zeros.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   *
+   * Example: \include FullPivLU_image.cpp
+   * Output: \verbinclude FullPivLU_image.out
+   *
+   * \sa kernel()
+   */
+  inline const internal::image_retval<FullPivLU> image(const MatrixType& originalMatrix) const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return internal::image_retval<FullPivLU>(*this, originalMatrix);
+  }
 
-    /** \returns the rank of the matrix of which *this is the LU decomposition.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline Index rank() const
-    {
-      using std::abs;
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      RealScalar premultiplied_threshold = abs(m_maxpivot) * threshold();
-      Index result = 0;
-      for(Index i = 0; i < m_nonzero_pivots; ++i)
-        result += (abs(m_lu.coeff(i,i)) > premultiplied_threshold);
-      return result;
-    }
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  /** \return a solution x to the equation Ax=b, where A is the matrix of which
+   * *this is the LU decomposition.
+   *
+   * \param b the right-hand-side of the equation to solve. Can be a vector or a matrix,
+   *          the only requirement in order for the equation to make sense is that
+   *          b.rows()==A.rows(), where A is the matrix of which *this is the LU decomposition.
+   *
+   * \returns a solution.
+   *
+   * \note_about_checking_solutions
+   *
+   * \note_about_arbitrary_choice_of_solution
+   * \note_about_using_kernel_to_study_multiple_solutions
+   *
+   * Example: \include FullPivLU_solve.cpp
+   * Output: \verbinclude FullPivLU_solve.out
+   *
+   * \sa TriangularView::solve(), kernel(), inverse()
+   */
+  template <typename Rhs>
+  inline const Solve<FullPivLU, Rhs> solve(const MatrixBase<Rhs>& b) const;
+#endif
 
-    /** \returns the dimension of the kernel of the matrix of which *this is the LU decomposition.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline Index dimensionOfKernel() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return cols() - rank();
+  /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
+      the LU decomposition.
+    */
+  inline RealScalar rcond() const {
+    eigen_assert(m_isInitialized && "FullPivLU is not initialized.");
+    if (!isInvertible()) {
+      return RealScalar(0);
     }
+    return internal::rcond_estimate_helper(m_l1_norm, *this);
+  }
 
-    /** \returns true if the matrix of which *this is the LU decomposition represents an injective
-      *          linear map, i.e. has trivial kernel; false otherwise.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline bool isInjective() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return rank() == cols();
-    }
+  /** \returns the determinant of the matrix of which
+   * *this is the LU decomposition. It has only linear complexity
+   * (that is, O(n) where n is the dimension of the square matrix)
+   * as the LU decomposition has already been computed.
+   *
+   * \note This is only for square matrices.
+   *
+   * \note For fixed-size matrices of size up to 4, MatrixBase::determinant() offers
+   *       optimized paths.
+   *
+   * \warning a determinant can be very big or small, so for matrices
+   * of large enough dimension, there is a risk of overflow/underflow.
+   *
+   * \sa MatrixBase::determinant()
+   */
+  typename internal::traits<MatrixType>::Scalar determinant() const;
+
+  /** Allows to prescribe a threshold to be used by certain methods, such as rank(),
+   * who need to determine when pivots are to be considered nonzero. This is not used for the
+   * LU decomposition itself.
+   *
+   * When it needs to get the threshold value, Eigen calls threshold(). By default, this
+   * uses a formula to automatically determine a reasonable threshold.
+   * Once you have called the present method setThreshold(const RealScalar&),
+   * your value is used instead.
+   *
+   * \param threshold The new value to use as the threshold.
+   *
+   * A pivot will be considered nonzero if its absolute value is strictly greater than
+   *  \f$ \vert pivot \vert \leqslant threshold \times \vert maxpivot \vert \f$
+   * where maxpivot is the biggest pivot.
+   *
+   * If you want to come back to the default behavior, call setThreshold(Default_t)
+   */
+  FullPivLU& setThreshold(const RealScalar& threshold) {
+    m_usePrescribedThreshold = true;
+    m_prescribedThreshold = threshold;
+    return *this;
+  }
 
-    /** \returns true if the matrix of which *this is the LU decomposition represents a surjective
-      *          linear map; false otherwise.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline bool isSurjective() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return rank() == rows();
-    }
+  /** Allows to come back to the default behavior, letting Eigen use its default formula for
+   * determining the threshold.
+   *
+   * You should pass the special object Eigen::Default as parameter here.
+   * \code lu.setThreshold(Eigen::Default); \endcode
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  FullPivLU& setThreshold(Default_t) {
+    m_usePrescribedThreshold = false;
+    return *this;
+  }
 
-    /** \returns true if the matrix of which *this is the LU decomposition is invertible.
-      *
-      * \note This method has to determine which pivots should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline bool isInvertible() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      return isInjective() && (m_lu.rows() == m_lu.cols());
-    }
+  /** Returns the threshold that will be used by certain methods such as rank().
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  RealScalar threshold() const {
+    eigen_assert(m_isInitialized || m_usePrescribedThreshold);
+    return m_usePrescribedThreshold ? m_prescribedThreshold
+                                    // this formula comes from experimenting (see "LU precision tuning" thread on the
+                                    // list) and turns out to be identical to Higham's formula used already in LDLt.
+                                    : NumTraits<Scalar>::epsilon() * RealScalar(m_lu.diagonalSize());
+  }
 
-    /** \returns the inverse of the matrix of which *this is the LU decomposition.
-      *
-      * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
-      *       Use isInvertible() to first determine whether this matrix is invertible.
-      *
-      * \sa MatrixBase::inverse()
-      */
-    inline const Inverse<FullPivLU> inverse() const
-    {
-      eigen_assert(m_isInitialized && "LU is not initialized.");
-      eigen_assert(m_lu.rows() == m_lu.cols() && "You can't take the inverse of a non-square matrix!");
-      return Inverse<FullPivLU>(*this);
-    }
+  /** \returns the rank of the matrix of which *this is the LU decomposition.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   */
+  inline Index rank() const {
+    using std::abs;
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    RealScalar premultiplied_threshold = abs(m_maxpivot) * threshold();
+    Index result = 0;
+    for (Index i = 0; i < m_nonzero_pivots; ++i) result += (abs(m_lu.coeff(i, i)) > premultiplied_threshold);
+    return result;
+  }
 
-    MatrixType reconstructedMatrix() const;
+  /** \returns the dimension of the kernel of the matrix of which *this is the LU decomposition.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   */
+  inline Index dimensionOfKernel() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return cols() - rank();
+  }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index rows() const EIGEN_NOEXCEPT { return m_lu.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_lu.cols(); }
+  /** \returns true if the matrix of which *this is the LU decomposition represents an injective
+   *          linear map, i.e. has trivial kernel; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   */
+  inline bool isInjective() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return rank() == cols();
+  }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename RhsType, typename DstType>
-    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+  /** \returns true if the matrix of which *this is the LU decomposition represents a surjective
+   *          linear map; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   */
+  inline bool isSurjective() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return rank() == rows();
+  }
 
-    template<bool Conjugate, typename RhsType, typename DstType>
-    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
-    #endif
+  /** \returns true if the matrix of which *this is the LU decomposition is invertible.
+   *
+   * \note This method has to determine which pivots should be considered nonzero.
+   *       For that, it uses the threshold value that you can control by calling
+   *       setThreshold(const RealScalar&).
+   */
+  inline bool isInvertible() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    return isInjective() && (m_lu.rows() == m_lu.cols());
+  }
 
-  protected:
+  /** \returns the inverse of the matrix of which *this is the LU decomposition.
+   *
+   * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
+   *       Use isInvertible() to first determine whether this matrix is invertible.
+   *
+   * \sa MatrixBase::inverse()
+   */
+  inline const Inverse<FullPivLU> inverse() const {
+    eigen_assert(m_isInitialized && "LU is not initialized.");
+    eigen_assert(m_lu.rows() == m_lu.cols() && "You can't take the inverse of a non-square matrix!");
+    return Inverse<FullPivLU>(*this);
+  }
 
-    static void check_template_parameters()
-    {
-      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
-    }
+  MatrixType reconstructedMatrix() const;
 
-    void computeInPlace();
-
-    MatrixType m_lu;
-    PermutationPType m_p;
-    PermutationQType m_q;
-    IntColVectorType m_rowsTranspositions;
-    IntRowVectorType m_colsTranspositions;
-    Index m_nonzero_pivots;
-    RealScalar m_l1_norm;
-    RealScalar m_maxpivot, m_prescribedThreshold;
-    signed char m_det_pq;
-    bool m_isInitialized, m_usePrescribedThreshold;
-};
+  EIGEN_DEVICE_FUNC constexpr Index rows() const noexcept { return m_lu.rows(); }
+  EIGEN_DEVICE_FUNC constexpr Index cols() const noexcept { return m_lu.cols(); }
 
-template<typename MatrixType>
-FullPivLU<MatrixType>::FullPivLU()
-  : m_isInitialized(false), m_usePrescribedThreshold(false)
-{
-}
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  template <typename RhsType, typename DstType>
+  void _solve_impl(const RhsType& rhs, DstType& dst) const;
 
-template<typename MatrixType>
-FullPivLU<MatrixType>::FullPivLU(Index rows, Index cols)
-  : m_lu(rows, cols),
-    m_p(rows),
-    m_q(cols),
-    m_rowsTranspositions(rows),
-    m_colsTranspositions(cols),
-    m_isInitialized(false),
-    m_usePrescribedThreshold(false)
-{
-}
+  template <bool Conjugate, typename RhsType, typename DstType>
+  void _solve_impl_transposed(const RhsType& rhs, DstType& dst) const;
+#endif
 
-template<typename MatrixType>
-template<typename InputType>
-FullPivLU<MatrixType>::FullPivLU(const EigenBase<InputType>& matrix)
-  : m_lu(matrix.rows(), matrix.cols()),
-    m_p(matrix.rows()),
-    m_q(matrix.cols()),
-    m_rowsTranspositions(matrix.rows()),
-    m_colsTranspositions(matrix.cols()),
-    m_isInitialized(false),
-    m_usePrescribedThreshold(false)
-{
+ protected:
+  EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar)
+
+  void computeInPlace();
+
+  MatrixType m_lu;
+  PermutationPType m_p;
+  PermutationQType m_q;
+  IntColVectorType m_rowsTranspositions;
+  IntRowVectorType m_colsTranspositions;
+  Index m_nonzero_pivots;
+  RealScalar m_l1_norm;
+  RealScalar m_maxpivot, m_prescribedThreshold;
+  signed char m_det_pq;
+  bool m_isInitialized, m_usePrescribedThreshold;
+};
+
+template <typename MatrixType, typename PermutationIndex>
+FullPivLU<MatrixType, PermutationIndex>::FullPivLU() : m_isInitialized(false), m_usePrescribedThreshold(false) {}
+
+template <typename MatrixType, typename PermutationIndex>
+FullPivLU<MatrixType, PermutationIndex>::FullPivLU(Index rows, Index cols)
+    : m_lu(rows, cols),
+      m_p(rows),
+      m_q(cols),
+      m_rowsTranspositions(rows),
+      m_colsTranspositions(cols),
+      m_isInitialized(false),
+      m_usePrescribedThreshold(false) {}
+
+template <typename MatrixType, typename PermutationIndex>
+template <typename InputType>
+FullPivLU<MatrixType, PermutationIndex>::FullPivLU(const EigenBase<InputType>& matrix)
+    : m_lu(matrix.rows(), matrix.cols()),
+      m_p(matrix.rows()),
+      m_q(matrix.cols()),
+      m_rowsTranspositions(matrix.rows()),
+      m_colsTranspositions(matrix.cols()),
+      m_isInitialized(false),
+      m_usePrescribedThreshold(false) {
   compute(matrix.derived());
 }
 
-template<typename MatrixType>
-template<typename InputType>
-FullPivLU<MatrixType>::FullPivLU(EigenBase<InputType>& matrix)
-  : m_lu(matrix.derived()),
-    m_p(matrix.rows()),
-    m_q(matrix.cols()),
-    m_rowsTranspositions(matrix.rows()),
-    m_colsTranspositions(matrix.cols()),
-    m_isInitialized(false),
-    m_usePrescribedThreshold(false)
-{
+template <typename MatrixType, typename PermutationIndex>
+template <typename InputType>
+FullPivLU<MatrixType, PermutationIndex>::FullPivLU(EigenBase<InputType>& matrix)
+    : m_lu(matrix.derived()),
+      m_p(matrix.rows()),
+      m_q(matrix.cols()),
+      m_rowsTranspositions(matrix.rows()),
+      m_colsTranspositions(matrix.cols()),
+      m_isInitialized(false),
+      m_usePrescribedThreshold(false) {
   computeInPlace();
 }
 
-template<typename MatrixType>
-void FullPivLU<MatrixType>::computeInPlace()
-{
-  check_template_parameters();
-
-  // the permutations are stored as int indices, so just to be sure:
-  eigen_assert(m_lu.rows()<=NumTraits<int>::highest() && m_lu.cols()<=NumTraits<int>::highest());
+template <typename MatrixType, typename PermutationIndex>
+void FullPivLU<MatrixType, PermutationIndex>::computeInPlace() {
+  eigen_assert(m_lu.rows() <= NumTraits<PermutationIndex>::highest() &&
+               m_lu.cols() <= NumTraits<PermutationIndex>::highest());
 
   m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
 
@@ -502,13 +484,12 @@ void FullPivLU<MatrixType>::computeInPlace()
   // can't accumulate on-the-fly because that will be done in reverse order for the rows.
   m_rowsTranspositions.resize(m_lu.rows());
   m_colsTranspositions.resize(m_lu.cols());
-  Index number_of_transpositions = 0; // number of NONTRIVIAL transpositions, i.e. m_rowsTranspositions[i]!=i
+  Index number_of_transpositions = 0;  // number of NONTRIVIAL transpositions, i.e. m_rowsTranspositions[i]!=i
 
-  m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
+  m_nonzero_pivots = size;  // the generic case is that in which all pivots are nonzero (invertible case)
   m_maxpivot = RealScalar(0);
 
-  for(Index k = 0; k < size; ++k)
-  {
+  for (Index k = 0; k < size; ++k) {
     // First, we need to find the pivot.
 
     // biggest coefficient in the remaining bottom-right corner (starting at row k, col k)
@@ -516,38 +497,37 @@ void FullPivLU<MatrixType>::computeInPlace()
     typedef internal::scalar_score_coeff_op<Scalar> Scoring;
     typedef typename Scoring::result_type Score;
     Score biggest_in_corner;
-    biggest_in_corner = m_lu.bottomRightCorner(rows-k, cols-k)
-                        .unaryExpr(Scoring())
-                        .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
-    row_of_biggest_in_corner += k; // correct the values! since they were computed in the corner,
-    col_of_biggest_in_corner += k; // need to add k to them.
-
-    if(biggest_in_corner==Score(0))
-    {
+    biggest_in_corner = m_lu.bottomRightCorner(rows - k, cols - k)
+                            .unaryExpr(Scoring())
+                            .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
+    row_of_biggest_in_corner += k;  // correct the values! since they were computed in the corner,
+    col_of_biggest_in_corner += k;  // need to add k to them.
+
+    if (numext::is_exactly_zero(biggest_in_corner)) {
       // before exiting, make sure to initialize the still uninitialized transpositions
       // in a sane state without destroying what we already have.
       m_nonzero_pivots = k;
-      for(Index i = k; i < size; ++i)
-      {
+      for (Index i = k; i < size; ++i) {
         m_rowsTranspositions.coeffRef(i) = internal::convert_index<StorageIndex>(i);
         m_colsTranspositions.coeffRef(i) = internal::convert_index<StorageIndex>(i);
       }
       break;
     }
 
-    RealScalar abs_pivot = internal::abs_knowing_score<Scalar>()(m_lu(row_of_biggest_in_corner, col_of_biggest_in_corner), biggest_in_corner);
-    if(abs_pivot > m_maxpivot) m_maxpivot = abs_pivot;
+    RealScalar abs_pivot = internal::abs_knowing_score<Scalar>()(
+        m_lu(row_of_biggest_in_corner, col_of_biggest_in_corner), biggest_in_corner);
+    if (abs_pivot > m_maxpivot) m_maxpivot = abs_pivot;
 
     // Now that we've found the pivot, we need to apply the row/col swaps to
     // bring it to the location (k,k).
 
     m_rowsTranspositions.coeffRef(k) = internal::convert_index<StorageIndex>(row_of_biggest_in_corner);
     m_colsTranspositions.coeffRef(k) = internal::convert_index<StorageIndex>(col_of_biggest_in_corner);
-    if(k != row_of_biggest_in_corner) {
+    if (k != row_of_biggest_in_corner) {
       m_lu.row(k).swap(m_lu.row(row_of_biggest_in_corner));
       ++number_of_transpositions;
     }
-    if(k != col_of_biggest_in_corner) {
+    if (k != col_of_biggest_in_corner) {
       m_lu.col(k).swap(m_lu.col(col_of_biggest_in_corner));
       ++number_of_transpositions;
     }
@@ -555,31 +535,28 @@ void FullPivLU<MatrixType>::computeInPlace()
     // Now that the pivot is at the right location, we update the remaining
     // bottom-right corner by Gaussian elimination.
 
-    if(k<rows-1)
-      m_lu.col(k).tail(rows-k-1) /= m_lu.coeff(k,k);
-    if(k<size-1)
-      m_lu.block(k+1,k+1,rows-k-1,cols-k-1).noalias() -= m_lu.col(k).tail(rows-k-1) * m_lu.row(k).tail(cols-k-1);
+    if (k < rows - 1) m_lu.col(k).tail(rows - k - 1) /= m_lu.coeff(k, k);
+    if (k < size - 1)
+      m_lu.block(k + 1, k + 1, rows - k - 1, cols - k - 1).noalias() -=
+          m_lu.col(k).tail(rows - k - 1) * m_lu.row(k).tail(cols - k - 1);
   }
 
   // the main loop is over, we still have to accumulate the transpositions to find the
   // permutations P and Q
 
   m_p.setIdentity(rows);
-  for(Index k = size-1; k >= 0; --k)
-    m_p.applyTranspositionOnTheRight(k, m_rowsTranspositions.coeff(k));
+  for (Index k = size - 1; k >= 0; --k) m_p.applyTranspositionOnTheRight(k, m_rowsTranspositions.coeff(k));
 
   m_q.setIdentity(cols);
-  for(Index k = 0; k < size; ++k)
-    m_q.applyTranspositionOnTheRight(k, m_colsTranspositions.coeff(k));
+  for (Index k = 0; k < size; ++k) m_q.applyTranspositionOnTheRight(k, m_colsTranspositions.coeff(k));
 
-  m_det_pq = (number_of_transpositions%2) ? -1 : 1;
+  m_det_pq = (number_of_transpositions % 2) ? -1 : 1;
 
   m_isInitialized = true;
 }
 
-template<typename MatrixType>
-typename internal::traits<MatrixType>::Scalar FullPivLU<MatrixType>::determinant() const
-{
+template <typename MatrixType, typename PermutationIndex>
+typename internal::traits<MatrixType>::Scalar FullPivLU<MatrixType, PermutationIndex>::determinant() const {
   eigen_assert(m_isInitialized && "LU is not initialized.");
   eigen_assert(m_lu.rows() == m_lu.cols() && "You can't take the determinant of a non-square matrix!");
   return Scalar(m_det_pq) * Scalar(m_lu.diagonal().prod());
@@ -588,18 +565,15 @@ typename internal::traits<MatrixType>::Scalar FullPivLU<MatrixType>::determinant
 /** \returns the matrix represented by the decomposition,
  * i.e., it returns the product: \f$ P^{-1} L U Q^{-1} \f$.
  * This function is provided for debug purposes. */
-template<typename MatrixType>
-MatrixType FullPivLU<MatrixType>::reconstructedMatrix() const
-{
+template <typename MatrixType, typename PermutationIndex>
+MatrixType FullPivLU<MatrixType, PermutationIndex>::reconstructedMatrix() const {
   eigen_assert(m_isInitialized && "LU is not initialized.");
   const Index smalldim = (std::min)(m_lu.rows(), m_lu.cols());
   // LU
-  MatrixType res(m_lu.rows(),m_lu.cols());
+  MatrixType res(m_lu.rows(), m_lu.cols());
   // FIXME the .toDenseMatrix() should not be needed...
-  res = m_lu.leftCols(smalldim)
-            .template triangularView<UnitLower>().toDenseMatrix()
-      * m_lu.topRows(smalldim)
-            .template triangularView<Upper>().toDenseMatrix();
+  res = m_lu.leftCols(smalldim).template triangularView<UnitLower>().toDenseMatrix() *
+        m_lu.topRows(smalldim).template triangularView<Upper>().toDenseMatrix();
 
   // P^{-1}(LU)
   res = m_p.inverse() * res;
@@ -613,23 +587,21 @@ MatrixType FullPivLU<MatrixType>::reconstructedMatrix() const
 /********* Implementation of kernel() **************************************************/
 
 namespace internal {
-template<typename _MatrixType>
-struct kernel_retval<FullPivLU<_MatrixType> >
-  : kernel_retval_base<FullPivLU<_MatrixType> >
-{
-  EIGEN_MAKE_KERNEL_HELPERS(FullPivLU<_MatrixType>)
-
-  enum { MaxSmallDimAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(
-            MatrixType::MaxColsAtCompileTime,
-            MatrixType::MaxRowsAtCompileTime)
+template <typename MatrixType_, typename PermutationIndex_>
+struct kernel_retval<FullPivLU<MatrixType_, PermutationIndex_> >
+    : kernel_retval_base<FullPivLU<MatrixType_, PermutationIndex_> > {
+  using DecompositionType = FullPivLU<MatrixType_, PermutationIndex_>;
+  EIGEN_MAKE_KERNEL_HELPERS(DecompositionType)
+
+  enum {
+    MaxSmallDimAtCompileTime = min_size_prefer_fixed(MatrixType::MaxColsAtCompileTime, MatrixType::MaxRowsAtCompileTime)
   };
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
+  template <typename Dest>
+  void evalTo(Dest& dst) const {
     using std::abs;
     const Index cols = dec().matrixLU().cols(), dimker = cols - rank();
-    if(dimker == 0)
-    {
+    if (dimker == 0) {
       // The Kernel is just {0}, so it doesn't have a basis properly speaking, but let's
       // avoid crashing/asserting as that depends on floating point calculations. Let's
       // just return a single column vector filled with zeros.
@@ -638,83 +610,74 @@ struct kernel_retval<FullPivLU<_MatrixType> >
     }
 
     /* Let us use the following lemma:
-      *
-      * Lemma: If the matrix A has the LU decomposition PAQ = LU,
-      * then Ker A = Q(Ker U).
-      *
-      * Proof: trivial: just keep in mind that P, Q, L are invertible.
-      */
+     *
+     * Lemma: If the matrix A has the LU decomposition PAQ = LU,
+     * then Ker A = Q(Ker U).
+     *
+     * Proof: trivial: just keep in mind that P, Q, L are invertible.
+     */
 
     /* Thus, all we need to do is to compute Ker U, and then apply Q.
-      *
-      * U is upper triangular, with eigenvalues sorted so that any zeros appear at the end.
-      * Thus, the diagonal of U ends with exactly
-      * dimKer zero's. Let us use that to construct dimKer linearly
-      * independent vectors in Ker U.
-      */
+     *
+     * U is upper triangular, with eigenvalues sorted so that any zeros appear at the end.
+     * Thus, the diagonal of U ends with exactly
+     * dimKer zero's. Let us use that to construct dimKer linearly
+     * independent vectors in Ker U.
+     */
 
     Matrix<Index, Dynamic, 1, 0, MaxSmallDimAtCompileTime, 1> pivots(rank());
     RealScalar premultiplied_threshold = dec().maxPivot() * dec().threshold();
     Index p = 0;
-    for(Index i = 0; i < dec().nonzeroPivots(); ++i)
-      if(abs(dec().matrixLU().coeff(i,i)) > premultiplied_threshold)
-        pivots.coeffRef(p++) = i;
+    for (Index i = 0; i < dec().nonzeroPivots(); ++i)
+      if (abs(dec().matrixLU().coeff(i, i)) > premultiplied_threshold) pivots.coeffRef(p++) = i;
     eigen_internal_assert(p == rank());
 
     // we construct a temporaty trapezoid matrix m, by taking the U matrix and
     // permuting the rows and cols to bring the nonnegligible pivots to the top of
     // the main diagonal. We need that to be able to apply our triangular solvers.
     // FIXME when we get triangularView-for-rectangular-matrices, this can be simplified
-    Matrix<typename MatrixType::Scalar, Dynamic, Dynamic, MatrixType::Options,
-           MaxSmallDimAtCompileTime, MatrixType::MaxColsAtCompileTime>
-      m(dec().matrixLU().block(0, 0, rank(), cols));
-    for(Index i = 0; i < rank(); ++i)
-    {
-      if(i) m.row(i).head(i).setZero();
-      m.row(i).tail(cols-i) = dec().matrixLU().row(pivots.coeff(i)).tail(cols-i);
+    Matrix<typename MatrixType::Scalar, Dynamic, Dynamic, traits<MatrixType>::Options, MaxSmallDimAtCompileTime,
+           MatrixType::MaxColsAtCompileTime>
+        m(dec().matrixLU().block(0, 0, rank(), cols));
+    for (Index i = 0; i < rank(); ++i) {
+      if (i) m.row(i).head(i).setZero();
+      m.row(i).tail(cols - i) = dec().matrixLU().row(pivots.coeff(i)).tail(cols - i);
     }
     m.block(0, 0, rank(), rank());
     m.block(0, 0, rank(), rank()).template triangularView<StrictlyLower>().setZero();
-    for(Index i = 0; i < rank(); ++i)
-      m.col(i).swap(m.col(pivots.coeff(i)));
+    for (Index i = 0; i < rank(); ++i) m.col(i).swap(m.col(pivots.coeff(i)));
 
     // ok, we have our trapezoid matrix, we can apply the triangular solver.
     // notice that the math behind this suggests that we should apply this to the
     // negative of the RHS, but for performance we just put the negative sign elsewhere, see below.
-    m.topLeftCorner(rank(), rank())
-     .template triangularView<Upper>().solveInPlace(
-        m.topRightCorner(rank(), dimker)
-      );
+    m.topLeftCorner(rank(), rank()).template triangularView<Upper>().solveInPlace(m.topRightCorner(rank(), dimker));
 
     // now we must undo the column permutation that we had applied!
-    for(Index i = rank()-1; i >= 0; --i)
-      m.col(i).swap(m.col(pivots.coeff(i)));
+    for (Index i = rank() - 1; i >= 0; --i) m.col(i).swap(m.col(pivots.coeff(i)));
 
     // see the negative sign in the next line, that's what we were talking about above.
-    for(Index i = 0; i < rank(); ++i) dst.row(dec().permutationQ().indices().coeff(i)) = -m.row(i).tail(dimker);
-    for(Index i = rank(); i < cols; ++i) dst.row(dec().permutationQ().indices().coeff(i)).setZero();
-    for(Index k = 0; k < dimker; ++k) dst.coeffRef(dec().permutationQ().indices().coeff(rank()+k), k) = Scalar(1);
+    for (Index i = 0; i < rank(); ++i) dst.row(dec().permutationQ().indices().coeff(i)) = -m.row(i).tail(dimker);
+    for (Index i = rank(); i < cols; ++i) dst.row(dec().permutationQ().indices().coeff(i)).setZero();
+    for (Index k = 0; k < dimker; ++k) dst.coeffRef(dec().permutationQ().indices().coeff(rank() + k), k) = Scalar(1);
   }
 };
 
 /***** Implementation of image() *****************************************************/
 
-template<typename _MatrixType>
-struct image_retval<FullPivLU<_MatrixType> >
-  : image_retval_base<FullPivLU<_MatrixType> >
-{
-  EIGEN_MAKE_IMAGE_HELPERS(FullPivLU<_MatrixType>)
+template <typename MatrixType_, typename PermutationIndex_>
+struct image_retval<FullPivLU<MatrixType_, PermutationIndex_> >
+    : image_retval_base<FullPivLU<MatrixType_, PermutationIndex_> > {
+  using DecompositionType = FullPivLU<MatrixType_, PermutationIndex_>;
+  EIGEN_MAKE_IMAGE_HELPERS(DecompositionType)
 
-  enum { MaxSmallDimAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(
-            MatrixType::MaxColsAtCompileTime,
-            MatrixType::MaxRowsAtCompileTime)
+  enum {
+    MaxSmallDimAtCompileTime = min_size_prefer_fixed(MatrixType::MaxColsAtCompileTime, MatrixType::MaxRowsAtCompileTime)
   };
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
+  template <typename Dest>
+  void evalTo(Dest& dst) const {
     using std::abs;
-    if(rank() == 0)
-    {
+    if (rank() == 0) {
       // The Image is just {0}, so it doesn't have a basis properly speaking, but let's
       // avoid crashing/asserting as that depends on floating point calculations. Let's
       // just return a single column vector filled with zeros.
@@ -725,40 +688,35 @@ struct image_retval<FullPivLU<_MatrixType> >
     Matrix<Index, Dynamic, 1, 0, MaxSmallDimAtCompileTime, 1> pivots(rank());
     RealScalar premultiplied_threshold = dec().maxPivot() * dec().threshold();
     Index p = 0;
-    for(Index i = 0; i < dec().nonzeroPivots(); ++i)
-      if(abs(dec().matrixLU().coeff(i,i)) > premultiplied_threshold)
-        pivots.coeffRef(p++) = i;
+    for (Index i = 0; i < dec().nonzeroPivots(); ++i)
+      if (abs(dec().matrixLU().coeff(i, i)) > premultiplied_threshold) pivots.coeffRef(p++) = i;
     eigen_internal_assert(p == rank());
 
-    for(Index i = 0; i < rank(); ++i)
+    for (Index i = 0; i < rank(); ++i)
       dst.col(i) = originalMatrix().col(dec().permutationQ().indices().coeff(pivots.coeff(i)));
   }
 };
 
 /***** Implementation of solve() *****************************************************/
 
-} // end namespace internal
+}  // end namespace internal
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-template<typename _MatrixType>
-template<typename RhsType, typename DstType>
-void FullPivLU<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
-{
+template <typename MatrixType_, typename PermutationIndex_>
+template <typename RhsType, typename DstType>
+void FullPivLU<MatrixType_, PermutationIndex_>::_solve_impl(const RhsType& rhs, DstType& dst) const {
   /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1}.
-  * So we proceed as follows:
-  * Step 1: compute c = P * rhs.
-  * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
-  * Step 3: replace c by the solution x to Ux = c. May or may not exist.
-  * Step 4: result = Q * c;
-  */
-
-  const Index rows = this->rows(),
-              cols = this->cols(),
-              nonzero_pivots = this->rank();
+   * So we proceed as follows:
+   * Step 1: compute c = P * rhs.
+   * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
+   * Step 3: replace c by the solution x to Ux = c. May or may not exist.
+   * Step 4: result = Q * c;
+   */
+
+  const Index rows = this->rows(), cols = this->cols(), nonzero_pivots = this->rank();
   const Index smalldim = (std::min)(rows, cols);
 
-  if(nonzero_pivots == 0)
-  {
+  if (nonzero_pivots == 0) {
     dst.setZero();
     return;
   }
@@ -769,11 +727,8 @@ void FullPivLU<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
   c = permutationP() * rhs;
 
   // Step 2
-  m_lu.topLeftCorner(smalldim,smalldim)
-      .template triangularView<UnitLower>()
-      .solveInPlace(c.topRows(smalldim));
-  if(rows>cols)
-    c.bottomRows(rows-cols) -= m_lu.bottomRows(rows-cols) * c.topRows(cols);
+  m_lu.topLeftCorner(smalldim, smalldim).template triangularView<UnitLower>().solveInPlace(c.topRows(smalldim));
+  if (rows > cols) c.bottomRows(rows - cols).noalias() -= m_lu.bottomRows(rows - cols) * c.topRows(cols);
 
   // Step 3
   m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
@@ -781,16 +736,13 @@ void FullPivLU<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
       .solveInPlace(c.topRows(nonzero_pivots));
 
   // Step 4
-  for(Index i = 0; i < nonzero_pivots; ++i)
-    dst.row(permutationQ().indices().coeff(i)) = c.row(i);
-  for(Index i = nonzero_pivots; i < m_lu.cols(); ++i)
-    dst.row(permutationQ().indices().coeff(i)).setZero();
+  for (Index i = 0; i < nonzero_pivots; ++i) dst.row(permutationQ().indices().coeff(i)) = c.row(i);
+  for (Index i = nonzero_pivots; i < m_lu.cols(); ++i) dst.row(permutationQ().indices().coeff(i)).setZero();
 }
 
-template<typename _MatrixType>
-template<bool Conjugate, typename RhsType, typename DstType>
-void FullPivLU<_MatrixType>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
-{
+template <typename MatrixType_, typename PermutationIndex_>
+template <bool Conjugate, typename RhsType, typename DstType>
+void FullPivLU<MatrixType_, PermutationIndex_>::_solve_impl_transposed(const RhsType& rhs, DstType& dst) const {
   /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1},
    * and since permutations are real and unitary, we can write this
    * as   A^T = Q U^T L^T P,
@@ -802,12 +754,10 @@ void FullPivLU<_MatrixType>::_solve_impl_transposed(const RhsType &rhs, DstType
    * If Conjugate is true, replace "^T" by "^*" above.
    */
 
-  const Index rows = this->rows(), cols = this->cols(),
-    nonzero_pivots = this->rank();
+  const Index rows = this->rows(), cols = this->cols(), nonzero_pivots = this->rank();
   const Index smalldim = (std::min)(rows, cols);
 
-  if(nonzero_pivots == 0)
-  {
+  if (nonzero_pivots == 0) {
     dst.setZero();
     return;
   }
@@ -833,45 +783,44 @@ void FullPivLU<_MatrixType>::_solve_impl_transposed(const RhsType &rhs, DstType
 
   // Step 4
   PermutationPType invp = permutationP().inverse().eval();
-  for(Index i = 0; i < smalldim; ++i)
-    dst.row(invp.indices().coeff(i)) = c.row(i);
-  for(Index i = smalldim; i < rows; ++i)
-    dst.row(invp.indices().coeff(i)).setZero();
+  for (Index i = 0; i < smalldim; ++i) dst.row(invp.indices().coeff(i)) = c.row(i);
+  for (Index i = smalldim; i < rows; ++i) dst.row(invp.indices().coeff(i)).setZero();
 }
 
 #endif
 
 namespace internal {
 
-
 /***** Implementation of inverse() *****************************************************/
-template<typename DstXprType, typename MatrixType>
-struct Assignment<DstXprType, Inverse<FullPivLU<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename FullPivLU<MatrixType>::Scalar>, Dense2Dense>
-{
-  typedef FullPivLU<MatrixType> LuType;
+template <typename DstXprType, typename MatrixType, typename PermutationIndex>
+struct Assignment<
+    DstXprType, Inverse<FullPivLU<MatrixType, PermutationIndex> >,
+    internal::assign_op<typename DstXprType::Scalar, typename FullPivLU<MatrixType, PermutationIndex>::Scalar>,
+    Dense2Dense> {
+  typedef FullPivLU<MatrixType, PermutationIndex> LuType;
   typedef Inverse<LuType> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename MatrixType::Scalar> &)
-  {
+  static void run(DstXprType& dst, const SrcXprType& src,
+                  const internal::assign_op<typename DstXprType::Scalar, typename MatrixType::Scalar>&) {
     dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
-} // end namespace internal
+}  // end namespace internal
 
 /******* MatrixBase methods *****************************************************************/
 
 /** \lu_module
-  *
-  * \return the full-pivoting LU decomposition of \c *this.
-  *
-  * \sa class FullPivLU
-  */
-template<typename Derived>
-inline const FullPivLU<typename MatrixBase<Derived>::PlainObject>
-MatrixBase<Derived>::fullPivLu() const
-{
-  return FullPivLU<PlainObject>(eval());
+ *
+ * \return the full-pivoting LU decomposition of \c *this.
+ *
+ * \sa class FullPivLU
+ */
+template <typename Derived>
+template <typename PermutationIndex>
+inline const FullPivLU<typename MatrixBase<Derived>::PlainObject, PermutationIndex> MatrixBase<Derived>::fullPivLu()
+    const {
+  return FullPivLU<PlainObject, PermutationIndex>(eval());
 }
 
-} // end namespace Eigen
+}  // end namespace Eigen
 
-#endif // EIGEN_LU_H
+#endif  // EIGEN_LU_H