casadi 3.6.7__cp313-none-manylinux2014_aarch64.whl

casadi/include/casadi/core/generic_matrix.hpp

@@ -0,0 +1,1598 @@
+/*
+ *    This file is part of CasADi.
+ *
+ *    CasADi -- A symbolic framework for dynamic optimization.
+ *    Copyright (C) 2010-2023 Joel Andersson, Joris Gillis, Moritz Diehl,
+ *                            KU Leuven. All rights reserved.
+ *    Copyright (C) 2011-2014 Greg Horn
+ *
+ *    CasADi is free software; you can redistribute it and/or
+ *    modify it under the terms of the GNU Lesser General Public
+ *    License as published by the Free Software Foundation; either
+ *    version 3 of the License, or (at your option) any later version.
+ *
+ *    CasADi is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ *    Lesser General Public License for more details.
+ *
+ *    You should have received a copy of the GNU Lesser General Public
+ *    License along with CasADi; if not, write to the Free Software
+ *    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ */
+
+
+#ifndef CASADI_GENERIC_MATRIX_HPP
+#define CASADI_GENERIC_MATRIX_HPP
+
+#include "slice.hpp"
+#include "submatrix.hpp"
+#include "nonzeros.hpp"
+#include "sparsity.hpp"
+#include "calculus.hpp"
+#include "sparsity_interface.hpp"
+#include "generic_type.hpp"
+
+namespace casadi {
+  /** \brief Empty Base
+
+      This class is extended in SWIG.
+
+      \identifier{1al} */
+  struct CASADI_EXPORT GenericMatrixCommon {};
+
+  /** \brief Matrix base class
+
+      This is a common base class for MX and Matrix<>, introducing a uniform syntax and implementing
+      common functionality using the curiously recurring template pattern (CRTP) idiom.\n
+
+      The class is designed with the idea that "everything is a matrix",
+      that is, also scalars and vectors.\n
+      This philosophy makes it easy to use and to interface in particularly
+      with Python and Matlab/Octave.\n
+
+      The syntax tries to stay as close as possible to the ublas syntax
+      when it comes to vector/matrix operations.\n
+
+      Index starts with 0.\n
+      Index vec happens as follows: (rr, cc) -> k = rr+cc*size1()\n
+      Vectors are column vectors.\n
+
+      The storage format is Compressed Column Storage (CCS),
+      similar to that used for sparse matrices in Matlab, \n
+      but unlike this format, we do allow for elements to be structurally non-zero
+      but numerically zero.\n
+
+      The sparsity pattern, which is reference counted and cached,
+      can be accessed with Sparsity& sparsity()\n
+
+      \author Joel Andersson
+      \date 2012
+
+      \identifier{1am} */
+  template<typename MatType>
+  class GenericMatrix
+    : public GenericMatrixCommon,
+      public SWIG_IF_ELSE(SparsityInterfaceCommon, SparsityInterface<MatType>) {
+    using SparsityInterface<MatType>::self;
+  public:
+
+    /** \brief Get the number of (structural) non-zero elements
+
+        \identifier{1an} */
+    casadi_int nnz() const;
+
+    /** \brief Get the number of non-zeros in the lower triangular half
+
+        \identifier{1ao} */
+    casadi_int nnz_lower() const;
+
+    /** \brief Get the number of non-zeros in the upper triangular half
+
+        \identifier{1ap} */
+    casadi_int nnz_upper() const;
+
+    /** \brief Get get the number of non-zeros on the diagonal
+
+        \identifier{1aq} */
+    casadi_int nnz_diag() const;
+
+    /** \brief Get the number of elements
+
+        \identifier{1ar} */
+    casadi_int numel() const;
+
+    /** \brief Get the first dimension (i.e. number of rows)
+
+        \identifier{1as} */
+    casadi_int size1() const;
+
+    /** \brief Get the number of rows, Octave-style syntax
+
+        \identifier{1at} */
+    casadi_int rows() const {return size1();}
+
+    /** \brief Get the second dimension (i.e. number of columns)
+
+        \identifier{1au} */
+    casadi_int size2() const;
+
+    /** \brief Get the number of columns, Octave-style syntax
+
+        \identifier{1av} */
+    casadi_int columns() const {return size2();}
+
+    /** \brief Get string representation of dimensions.
+
+        The representation is e.g. "4x5" or  "4x5,10nz"
+
+        \identifier{1aw} */
+    std::string dim(bool with_nz=false) const;
+
+    /** \brief  Get the shape
+
+        \identifier{1ax} */
+    std::pair<casadi_int, casadi_int> size() const;
+
+    /** \brief  Get the size along a particular dimensions
+
+        \identifier{1ay} */
+    casadi_int size(casadi_int axis) const;
+
+    /** \brief Check if the sparsity is empty, i.e. if one of the dimensions is zero
+
+     * (or optionally both dimensions)
+
+        \identifier{1az} */
+    bool is_empty(bool both=false) const { return sparsity().is_empty(both);}
+
+    /** \brief  Check if the matrix expression is dense
+
+        \identifier{1b0} */
+    bool is_dense() const { return sparsity().is_dense();}
+
+    /** \brief  Check if the matrix expression is scalar
+
+        \identifier{1b1} */
+    bool is_scalar(bool scalar_and_dense=false) const;
+
+    /** \brief  Check if the matrix expression is square
+
+        \identifier{1b2} */
+    bool is_square() const { return sparsity().is_square();}
+
+    /** \brief  Check if the matrix is a row or column vector
+
+        \identifier{1b3} */
+    bool is_vector() const { return sparsity().is_vector();}
+
+    /** \brief  Check if the matrix is a row vector (i.e. size1()==1)
+
+        \identifier{1b4} */
+    bool is_row() const { return sparsity().is_row();}
+
+    /** \brief  Check if the matrix is a column vector (i.e. size2()==1)
+
+        \identifier{1b5} */
+    bool is_column() const { return sparsity().is_column();}
+
+    /** \brief Check if the matrix is upper triangular
+
+        \identifier{1b6} */
+    bool is_triu() const { return sparsity().is_triu();}
+
+    /** \brief Check if the matrix is lower triangular
+
+        \identifier{1b7} */
+    bool is_tril() const { return sparsity().is_tril();}
+
+    ///@{
+    /** \brief Get the sparsity pattern. See the Sparsity class for details.
+
+        \identifier{1b8} */
+    std::vector<casadi_int> get_row() const { return sparsity().get_row(); }
+    std::vector<casadi_int> get_colind() const { return sparsity().get_colind(); }
+#ifndef SWIG
+    const casadi_int* row() const { return sparsity().row(); }
+    const casadi_int* colind() const { return sparsity().colind(); }
+#endif
+    casadi_int row(casadi_int el) const { return sparsity().row(el); }
+    casadi_int colind(casadi_int col) const { return sparsity().colind(col); }
+    ///@}
+
+    /** \brief Get the sparsity pattern
+
+        \identifier{1b9} */
+    SWIG_CONSTREF(Sparsity) sparsity() const;
+
+#ifndef SWIG
+    /// \cond CLUTTER
+    /**  @{  */
+    /// Functions called by friend functions defined here
+    static MatType interp1d(const std::vector<double>& x, const MatType &v,
+         const std::vector<double>& xq, const std::string& mode, bool equidistant);
+    static casadi_int sprank(const MatType &x) { return Sparsity::sprank(x.sparsity());}
+    static casadi_int norm_0_mul(const MatType &x, const MatType &y) {
+      return Sparsity::norm_0_mul(x.sparsity(), y.sparsity());
+    }
+    static MatType tril(const MatType &x, bool includeDiagonal=true) {
+      return project(x, Sparsity::tril(x.sparsity(), includeDiagonal));
+    }
+    static MatType triu(const MatType &x, bool includeDiagonal=true) {
+      return project(x, Sparsity::triu(x.sparsity(), includeDiagonal));
+    }
+    static MatType sumsqr(const MatType &x) { return dot(x, x);}
+    static MatType linspace(const MatType &a, const MatType &b, casadi_int nsteps);
+    static MatType cross(const MatType &a, const MatType &b, casadi_int dim=-1);
+    static MatType skew(const MatType &a);
+    static MatType inv_skew(const MatType &a);
+    static MatType tril2symm(const MatType &x);
+    static MatType triu2symm(const MatType &x);
+    static MatType repsum(const MatType &x, casadi_int n, casadi_int m=1);
+    static MatType diff(const MatType &x, casadi_int n=1, casadi_int axis=-1);
+
+    static bool is_linear(const MatType &expr, const MatType &var);
+    static bool is_quadratic(const MatType &expr, const MatType &var);
+    static void quadratic_coeff(const MatType &expr, const MatType &var,
+        MatType& A, MatType& b, MatType& c, bool check);
+    static void linear_coeff(const MatType &expr, const MatType &var,
+        MatType& A, MatType& b, bool check);
+    /** @}  */
+    /// \endcond
+
+    /** \brief  Get vector nonzero or slice of nonzeros
+
+        \identifier{1bb} */
+    template<typename K>
+    const MatType nz(const K& k) const {
+      MatType ret;
+      self().get_nz(ret, false, k);
+      return ret;
+    }
+
+    /** \brief  Access vector nonzero or slice of nonzeros
+
+        \identifier{1bc} */
+    template<typename K>
+    NonZeros<MatType, K> nz(const K& k) {
+      return NonZeros<MatType, K>(self(), k);
+    }
+
+    /** \brief  Get vector element or slice
+
+        \identifier{1bd} */
+    template<typename RR>
+    const MatType operator()(const RR& rr) const {
+      MatType ret;
+      self().get(ret, false, rr);
+      return ret;
+    }
+
+    /** \brief  Get Matrix element or slice
+
+        \identifier{1be} */
+    template<typename RR, typename CC>
+    const MatType operator()(const RR& rr, const CC& cc) const {
+      MatType ret;
+      self().get(ret, false, rr, cc);
+      return ret;
+    }
+
+    /** \brief Access Matrix elements (one argument)
+
+        \identifier{1bf} */
+    template<typename RR>
+    SubIndex<MatType, RR> operator()(const RR& rr) {
+      return SubIndex<MatType, RR>(self(), rr);
+    }
+
+    /** \brief Access Matrix elements (two arguments)
+
+        \identifier{1bg} */
+    template<typename RR, typename CC>
+    SubMatrix<MatType, RR, CC> operator()(const RR& rr, const CC& cc) {
+      return SubMatrix<MatType, RR, CC>(self(), rr, cc);
+    }
+#endif // SWIG
+
+#if !defined(SWIG) || defined(DOXYGEN)
+/**
+\addtogroup expression_tools
+@{
+*/
+
+     /** \brief Performs 1d linear interpolation
+     *
+     * The data-points to be interpolated are given as (x[i], v[i]).
+     * xq[j] is used as interplating value
+
+         \identifier{1bh} */
+     inline friend MatType interp1d(const std::vector<double>& x, const MatType&v,
+         const std::vector<double>& xq, const std::string& mode, bool equidistant=false) {
+       return MatType::interp1d(x, v, xq, mode, equidistant);
+     }
+
+    /** \brief Matrix power x^n
+
+        \identifier{1bi} */
+    inline friend MatType mpower(const MatType& x, const MatType& n) {
+      return MatType::mpower(x, n);
+    }
+
+    /** \brief Construct second-order-convex
+     *
+     * \param[in] x vector expression of size n
+     * \param[in] y scalar expression
+     *
+     * soc(x,y) computes [y*eye(n) x; x' y]
+     *
+     *  soc(x,y) positive semi definite
+     *        <=> || x ||_2 <= y
+     *
+        \identifier{1bj} */
+    inline friend MatType soc(const MatType& x, const MatType& y) {
+      return MatType::soc(x, y);
+    }
+
+    /** \brief Compute any contraction of two dense tensors, using index/einstein notation
+
+        einstein(A, B, a, b, c) -> C
+
+        Given two tensors, A and B, computes a third tensor C such that:
+
+        C_c = A_a * B_b
+
+        With a, b, c representing einstein indices.
+        Instead of the classical index labels i,j,k,... we employ -1,-2,-3,...
+
+        A, B, C are represented as CasADi vectors, with dim_a, dim_b, dim_c
+        indictating theire tensorial dimensions.
+
+        \identifier{1bk} */
+    ///@{
+    inline friend MatType
+      einstein(const MatType &A, const MatType &B, const MatType &C,
+        const std::vector<casadi_int>& dim_a, const std::vector<casadi_int>& dim_b,
+        const std::vector<casadi_int>& dim_c,
+        const std::vector<casadi_int>& a, const std::vector<casadi_int>& b,
+        const std::vector<casadi_int>& c) {
+      return MatType::einstein(A, B, C, dim_a, dim_b, dim_c, a, b, c);
+    }
+
+    inline friend MatType
+      einstein(const MatType &A, const MatType &B,
+        const std::vector<casadi_int>& dim_a, const std::vector<casadi_int>& dim_b,
+        const std::vector<casadi_int>& dim_c,
+        const std::vector<casadi_int>& a, const std::vector<casadi_int>& b,
+        const std::vector<casadi_int>& c) {
+      return MatType::einstein(A, B, dim_a, dim_b, dim_c, a, b, c);
+    }
+    ///@}
+
+    /** \brief Matrix divide (cf. slash '/' in MATLAB)
+
+        \identifier{1bl} */
+    inline friend MatType mrdivide(const MatType& x, const MatType& n) {
+      return MatType::mrdivide(x, n);
+    }
+
+    /** \brief Matrix divide (cf. backslash '\' in MATLAB)
+
+        \identifier{1bm} */
+    inline friend MatType mldivide(const MatType& x, const MatType& n) {
+      return MatType::mldivide(x, n);
+    }
+
+    /** \brief Get all symbols contained in the supplied expression
+
+     * Get all symbols on which the supplied expression depends
+     * \see SXFunction::getFree(), MXFunction::getFree()
+
+        \identifier{1bn} */
+    inline friend std::vector<MatType> symvar(const MatType& x) {
+      return MatType::symvar(x);
+    }
+
+    ///@{
+    /** \brief Calculate bilinear/quadratic form x^T A y
+     *
+     * \param[in] y can be omitted, in which case x^T A x is calculated
+
+        \identifier{1bo} */
+    inline friend MatType bilin(const MatType &A, const MatType &x, const MatType &y) {
+      return MatType::bilin(A, x, y);
+    }
+    inline friend MatType bilin(const MatType &A, const MatType &x) {
+      return MatType::bilin(A, x, x);
+    }
+    static MatType bilin(const MatType& A, const MatType& x, const MatType& y);
+    ///@}
+
+    ///@{
+    /** \brief Make a rank-1 update to a matrix A
+
+     * Calculates A + 1/2 * alpha * x*y'
+
+        \identifier{1bp} */
+    inline friend MatType rank1(const MatType &A, const MatType &alpha,
+                                const MatType &x, const MatType &y) {
+      return MatType::rank1(A, alpha, x, y);
+    }
+    static MatType rank1(const MatType& A, const MatType& alpha,
+                         const MatType& x, const MatType& y);
+    ///@}
+
+    /** \brief Calculate sum of squares: sum_ij X_ij^2
+
+        \identifier{1bq} */
+    inline friend MatType sumsqr(const MatType &x) {
+      return MatType::sumsqr(x);
+    }
+
+    /** \brief x -> log(sum_i exp(x_i))
+     * 
+     * Can be used to achieve a smooth max.
+     * 
+     * Implementations for DM/SX/MX are hardened against overflow
+     * Implementations for DM/MX are more accurate with log1p (*)
+     * 
+     * (*) https://nhigham.com/2021/01/05/what-is-the-log-sum-exp-function/
+     *
+        \identifier{1br} */
+    inline friend MatType logsumexp(const MatType& x) {
+      return MatType::logsumexp(x);
+    }
+    /** \brief Scaled version of logsumexp
+     * 
+     * Scaled such that max(x) <= logsumexp(x, margin) <= max(x)+margin
+     *
+        \identifier{1bs} */
+    inline friend MatType logsumexp(const MatType& x, const MatType& margin) {
+      MatType alpha = log(x.size1()) / margin;
+      return MatType::logsumexp(alpha*x)/alpha;
+    }
+    static MatType logsumexp(const MatType& x);
+
+    /** \brief Matlab's \c linspace command
+
+        \identifier{1bt} */
+    inline friend MatType linspace(const MatType &a, const MatType &b, casadi_int nsteps) {
+      return MatType::linspace(a, b, nsteps);
+    }
+
+    /** \brief Matlab's \c cross command
+
+        \identifier{1bu} */
+    inline friend MatType cross(const MatType &a, const MatType &b, casadi_int dim = -1) {
+      return MatType::cross(a, b, dim);
+    }
+
+    /** \brief Generate a skew symmetric matrix from a 3-vector
+
+        \identifier{1bv} */
+    inline friend MatType skew(const MatType &a) {
+      return MatType::skew(a);
+    }
+
+    /** \brief Generate the 3-vector progenitor of a skew symmetric matrix
+
+        \identifier{1bw} */
+    inline friend MatType inv_skew(const MatType &a) {
+      return MatType::inv_skew(a);
+    }
+
+    /** \brief Matrix determinant (experimental)
+
+        \identifier{1bx} */
+    inline friend MatType det(const MatType& A) { return MatType::det(A);}
+
+    /** \brief Matrix inverse (experimental)
+
+        \identifier{1by} */
+    inline friend MatType inv_minor(const MatType& A) { return MatType::inv_minor(A);}
+
+    /** \brief Matrix inverse
+
+        \identifier{1bz} */
+    inline friend MatType inv(const MatType& A) {
+        return MatType::inv(A);
+    }
+
+    /** \brief Matrix inverse
+
+        \identifier{1c0} */
+    inline friend MatType inv(const MatType& A,
+      const std::string& lsolver,
+      const Dict& options=Dict()) {
+        return MatType::inv(A, lsolver, options);
+    }
+
+    /** \brief Matrix trace
+
+        \identifier{1c1} */
+    inline friend MatType trace(const MatType& x) { return MatType::trace(x);}
+
+    /** \brief Convert a lower triangular matrix to a symmetric one
+
+        \identifier{1c2} */
+    inline friend MatType tril2symm(const MatType &a) { return MatType::tril2symm(a);}
+
+    /** \brief Convert a upper triangular matrix to a symmetric one
+
+        \identifier{1c3} */
+    inline friend MatType triu2symm(const MatType &a) { return MatType::triu2symm(a);}
+
+    /** \brief  Frobenius norm
+
+        \identifier{1c4} */
+    inline friend MatType norm_fro(const MatType &x) { return MatType::norm_fro(x);}
+
+    /** \brief  2-norm
+
+        \identifier{1c5} */
+    inline friend MatType norm_2(const MatType &x) { return MatType::norm_2(x);}
+
+    /** \brief 1-norm
+
+        \identifier{1c6} */
+    inline friend MatType norm_1(const MatType &x) { return MatType::norm_1(x);}
+
+    /** \brief Infinity-norm
+
+        \identifier{1c7} */
+    inline friend MatType norm_inf(const MatType &x) { return MatType::norm_inf(x);}
+
+    /** \brief Returns difference (n-th order) along given axis (MATLAB convention)
+
+        \identifier{1c8} */
+    inline friend MatType diff(const MatType &x, casadi_int n=1, casadi_int axis=-1) {
+      return MatType::diff(x, n, axis);
+    }
+
+    /** \brief Returns cumulative sum along given axis (MATLAB convention)
+
+        \identifier{1c9} */
+    inline friend MatType cumsum(const MatType &x, casadi_int axis=-1) {
+      return MatType::cumsum(x, axis);
+    }
+
+    /** \brief Inner product of two matrices
+
+        with x and y matrices of the same dimension
+
+        \identifier{1ca} */
+    inline friend MatType dot(const MatType &x, const MatType &y) {
+      return MatType::dot(x, y);
+    }
+
+    /** \brief Computes the nullspace of a matrix A
+     *
+     * Finds Z m-by-(m-n) such that AZ = 0
+     * with A n-by-m with m > n
+     *
+     * Assumes A is full rank
+     *
+     * Inspired by Numerical Methods in Scientific Computing by Ake Bjorck
+
+        \identifier{1cb} */
+    inline friend MatType nullspace(const MatType& A) {
+      return MatType::nullspace(A);
+    }
+
+    /** \brief  Evaluate a polynomial with coefficients p in x
+
+        \identifier{1cc} */
+    inline friend MatType polyval(const MatType& p, const MatType& x) {
+      return MatType::polyval(p, x);
+    }
+
+    /** \brief   Get the diagonal of a matrix or construct a diagonal
+
+        When the input is square, the diagonal elements are returned.
+        If the input is vector-like, a diagonal matrix is constructed with it.
+
+        \identifier{1cd} */
+    inline friend MatType diag(const MatType &A) {
+      return MatType::diag(A);
+    }
+
+    /** \brief  Unite two matrices no overlapping sparsity
+
+        \identifier{1ce} */
+    inline friend MatType unite(const MatType& A, const MatType& B) {
+      return MatType::unite(A, B);
+    }
+
+    /** \brief  Make the matrix dense if not already
+
+        \identifier{1cf} */
+    inline friend MatType densify(const MatType& x) {
+      return MatType::densify(x);
+    }
+
+    /** \brief  Make the matrix dense and assign nonzeros to a value
+
+        \identifier{1cg} */
+    inline friend MatType densify(const MatType& x, const MatType& val) {
+      return MatType::densify(x, val);
+    }
+
+    /** \brief Create a new matrix with a given sparsity pattern but with the
+
+      * nonzeros taken from an existing matrix
+
+        \identifier{1ch} */
+    inline friend MatType project(const MatType& A, const Sparsity& sp,
+                                  bool intersect=false) {
+      return MatType::project(A, sp, intersect);
+    }
+
+    /** \brief Branching on MX nodes
+
+        Ternary operator, "cond ? if_true : if_false"
+
+        \identifier{1ci} */
+    inline friend MatType if_else(const MatType &cond, const MatType &if_true,
+                                  const MatType &if_false, bool short_circuit=false) {
+      return MatType::if_else(cond, if_true, if_false, short_circuit);
+    }
+
+    /** \brief Create a switch
+     *
+     * If the condition \param ind evaluates to the integer k, where 0<=k<f.size(),
+     * then x[k] will be returned, otherwise \param x_default will be returned.
+
+        \identifier{1cj} */
+    inline friend MatType conditional(const MatType& ind, const std::vector<MatType> &x,
+                                      const MatType &x_default, bool short_circuit=false) {
+      return MatType::conditional(ind, x, x_default, short_circuit);
+    }
+
+    /** \brief Check if expression depends on the argument
+
+        The argument must be symbolic
+
+        \identifier{1ck} */
+    inline friend bool depends_on(const MatType& f, const MatType &arg) {
+      return MatType::depends_on(f, arg);
+    }
+
+    /** \brief  Substitute variable v with expression vdef in an expression ex
+
+        \identifier{1cl} */
+    friend inline MatType substitute(const MatType& ex, const MatType& v,
+                                     const MatType& vdef) {
+      return MatType::substitute(ex, v, vdef);
+    }
+
+    /** \brief  Substitute variable var with expression expr in multiple expressions
+
+        \identifier{1cm} */
+    friend inline std::vector<MatType>
+      substitute(const std::vector<MatType>& ex, const std::vector<MatType>& v,
+                 const std::vector<MatType>& vdef) {
+      return MatType::substitute(ex, v, vdef);
+    }
+
+    /** \brief Inplace substitution with piggyback expressions
+
+     * Substitute variables v out of the expressions vdef sequentially,
+     * as well as out of a number of other expressions piggyback
+
+        \identifier{1cn} */
+    inline friend void
+      substitute_inplace(const std::vector<MatType>& v,
+                        std::vector<MatType>& inout_vdef,
+                        std::vector<MatType>& inout_ex, bool reverse=false) {
+      return MatType::substitute_inplace(v, inout_vdef, inout_ex, reverse);
+    }
+
+    /** \brief Common subexpression elimination
+
+        \identifier{1co} */
+    inline friend MatType cse(const MatType& e) {
+      return MatType::cse({e}).at(0);
+    }
+
+
+    /** \brief Common subexpression elimination
+
+        \identifier{1cp} */
+    inline friend std::vector<MatType> cse(const std::vector<MatType>& e) {
+      return MatType::cse(e);
+    }
+
+    /** \brief  Solve a system of equations: A*x = b
+
+        The solve routine works similar to Matlab's backslash when A is square and nonsingular.
+        The algorithm used is the following:
+        1. A simple forward or backward substitution if A is upper or lower triangular
+        2. If the linear system is at most 3-by-3, form the inverse via minor expansion and multiply
+        3. Permute the variables and equations as to get a (structurally) nonzero diagonal,
+        then perform a QR factorization without pivoting and solve the factorized system.
+
+        Note 1: If there are entries of the linear system known to be zero, these will be removed.
+        Elements that are very small, or will evaluate to be zero, can still cause numerical errors,
+        due to the lack of pivoting (which is not possible since cannot compare the size of entries)
+
+        Note 2: When permuting the linear system, a BLT (block lower triangular) transformation is
+        formed. Only the permutation part of this is however used. An improvement would be to solve
+        block-by-block if there are multiple BLT blocks.
+
+        \identifier{1cq} */
+    friend inline MatType solve(const MatType& A, const MatType& b) {
+      return MatType::solve(A, b);
+    }
+
+    /** \brief Solve a system of equations: A*x = b
+
+        \identifier{1cr} */
+    friend inline MatType solve(const MatType& A, const MatType& b,
+                                const std::string& lsolver,
+                                const Dict& dict = Dict()) {
+      return MatType::solve(A, b, lsolver, dict);
+    }
+
+    /** \brief Linearize an expression
+
+        \identifier{1cs} */
+    friend inline MatType linearize(const MatType& f, const MatType& x, const MatType& x0,
+        const Dict& opts=Dict()) {
+      return MatType::linearize(f, x, x0, opts);
+    }
+
+    /** \brief Computes the Moore-Penrose pseudo-inverse
+     *
+     * If the matrix A is fat (size1<size2), mul(A, pinv(A)) is unity.
+     *
+     *  pinv(A)' = (AA')^(-1) A
+     *
+     *
+     * If the matrix A is slender (size1>size2), mul(pinv(A), A) is unity.
+     *
+     *  pinv(A) = (A'A)^(-1) A'
+     *
+        \identifier{1ct} */
+    friend inline MatType pinv(const MatType& A) {
+      return MatType::pinv(A);
+    }
+
+    /** \brief Computes the Moore-Penrose pseudo-inverse
+     *
+     * If the matrix A is fat (size1>size2), mul(A, pinv(A)) is unity.
+     * If the matrix A is slender (size2<size1), mul(pinv(A), A) is unity.
+     *
+        \identifier{1cu} */
+    friend inline MatType pinv(const MatType& A, const std::string& lsolver,
+                               const Dict& dict = Dict()) {
+      return MatType::pinv(A, lsolver, dict);
+    }
+
+
+    /** \brief Calculate Matrix exponential
+     * 
+     * Computes expm(A*t) with A constant
+     * 
+     * \param A[in] Square matrix
+     * \param t[in] Scalar
+     *
+
+        \identifier{23v} */
+    friend inline MatType expm_const(const MatType& A, const MatType& t) {
+      return MatType::expm_const(A, t);
+    }
+
+    /** \brief Calculate Matrix exponential
+     *
+
+        \identifier{23w} */
+    friend inline MatType expm(const MatType& A) {
+      return MatType::expm(A);
+    }
+
+    /** \brief Calculate Jacobian
+     * 
+     * \return Sparse matrix
+
+        \identifier{1cv} */
+    inline friend MatType jacobian(const MatType &ex, const MatType &arg,
+                                   const Dict& opts = Dict()) {
+      return MatType::jacobian(ex, arg, opts);
+    }
+    /** \brief Calculate the gradient of an expression
+     * 
+     * \param ex[in] Scalar expression to take the gradient of
+     * \param arg[in] Vector expression of symbols
+     * \param opts[in] Options
+     * 
+     * \return Dense column vector
+
+        \identifier{23x} */
+    inline friend MatType gradient(const MatType &ex, const MatType &arg, const Dict& opts=Dict()) {
+      return MatType::gradient(ex, arg, opts);
+    }
+    /** \brief Calculate the tangent of an expression
+     *
+
+        \identifier{23y} */
+    inline friend MatType tangent(const MatType &ex, const MatType &arg, const Dict& opts=Dict()) {
+      return MatType::tangent(ex, arg, opts);
+    }
+
+    /** \brief Calculate the Jacobian and multiply by a vector from the right
+
+        This is equivalent to <tt>mul(jacobian(ex, arg), v)</tt> or
+        <tt>mul(jacobian(ex, arg).T, v)</tt> for
+        tr set to false and true respectively. If contrast to these
+        expressions, it will use directional derivatives which is typically (but
+        not necessarily) more efficient if the complete Jacobian is not needed and v has few rows.
+
+        \identifier{1cw} */
+    friend inline MatType jtimes(const MatType &ex, const MatType &arg,
+                                 const MatType &v, bool tr=false, const Dict& opts=Dict()) {
+      return MatType::jtimes(ex, arg, v, tr, opts);
+    }
+
+    /** \brief Forward directional derivative
+
+        \identifier{1cx} */
+    friend inline std::vector<std::vector<MatType> >
+    forward(const std::vector<MatType> &ex, const std::vector<MatType> &arg,
+            const std::vector<std::vector<MatType> > &v,
+            const Dict& opts = Dict()) {
+      return MatType::forward(ex, arg, v, opts);
+    }
+
+    /** \brief Reverse directional derivative
+
+        \identifier{1cy} */
+    friend inline std::vector<std::vector<MatType> >
+    reverse(const std::vector<MatType> &ex, const std::vector<MatType> &arg,
+            const std::vector<std::vector<MatType> > &v,
+            const Dict& opts = Dict()) {
+      return MatType::reverse(ex, arg, v, opts);
+    }
+
+    ///@{
+    /** \brief Hessian and (optionally) gradient
+
+        \identifier{23z} */
+    inline friend MatType hessian(const MatType &ex, const MatType &arg,
+        const Dict& opts = Dict()) {
+      return MatType::hessian(ex, arg, opts);
+    }
+    inline friend MatType hessian(const MatType &ex, const MatType &arg, MatType& output_g,
+        const Dict& opts = Dict()) {
+      return MatType::hessian(ex, arg, output_g, opts);
+    }
+    ///@}
+
+    /** \brief Find out which variables enter with some order
+
+        \identifier{1cz} */
+    inline friend std::vector<bool> which_depends(const MatType &expr, const MatType &var,
+        casadi_int order, bool tr) {
+      return MatType::which_depends(expr, var, order, tr);
+    }
+
+    /** \brief Get the sparsity pattern of a jacobian
+     * 
+     * Equivalent to, but cheaper to compute than, jacobian(f,x).sparsity()
+
+        \identifier{259} */
+    inline friend Sparsity jacobian_sparsity(const MatType &f, const MatType &x) {
+      return MatType::jacobian_sparsity(f, x);
+    }
+
+    /** \brief Is expr linear in var?
+    *
+    * False negatives are possible
+    * (an expression may not be recognised as linear while it really is),
+    * false positives not.
+
+        \identifier{1d0} */
+    inline friend bool is_linear(const MatType &expr, const MatType &var) {
+      return MatType::is_linear(expr, var);
+    }
+
+    /** \brief Is expr quadratic in var?
+    *
+    * False negatives are possible
+    * (an expression may not be recognised as quadratic while it really is),
+    * false positives not.
+
+        \identifier{1d1} */
+    inline friend bool is_quadratic(const MatType &expr, const MatType &var) {
+      return MatType::is_quadratic(expr, var);
+    }
+
+    /** \brief Recognizes quadratic form in scalar expression
+    *
+    * 1/2*x' A x + b' x + c
+    *
+    * e = 0.5*bilin(A,x,x)+dot(b,x)+c
+    *
+    * \param check[in] When true (default), A is checked to be independent of x.
+    *                  Provided to deal with false positive dependency checks.
+
+        \identifier{1d2} */
+    inline friend void quadratic_coeff(const MatType &expr, const MatType &var,
+        MatType& A, MatType& b, MatType& c, bool check=true) {
+      MatType::quadratic_coeff(expr, var, A, b, c, check);
+    }
+
+    /** \brief Recognizes linear form in vector expression
+    *
+    * A x + b
+    *
+    * \param check[in] When true (default)m, A is checked to be independent of x.
+    *                  Provided to deal with false positive dependency checks.
+
+        \identifier{1d3} */
+    inline friend void linear_coeff(const MatType &expr, const MatType &var,
+        MatType& A, MatType& b, bool check=true) {
+      MatType::linear_coeff(expr, var, A, b, check);
+    }
+
+    /** Count number of nodes */
+    inline friend casadi_int n_nodes(const MatType& A) {
+      return MatType::n_nodes(A);
+    }
+
+    /// Simplify an expression
+    friend inline MatType simplify(const MatType &x) {
+      return MatType::simplify(x);
+    }
+
+    /** \brief Get a string representation for a binary MatType, using custom arguments
+
+        \identifier{1d4} */
+    inline friend std::string
+      print_operator(const MatType& xb, const std::vector<std::string>& args) {
+      return MatType::print_operator(xb, args);
+    }
+
+    /** \brief Introduce intermediate variables for selected nodes in a graph
+
+        \identifier{1d5} */
+    inline friend void extract(std::vector<MatType>& ex,
+        std::vector<MatType>& v,
+        std::vector<MatType>& vdef,
+        const Dict& opts = Dict()) {
+      MatType::extract(ex, v, vdef, opts);
+    }
+
+    /** \brief Extract shared subexpressions from an set of expressions
+
+        \identifier{1d6} */
+    inline friend void shared(std::vector<MatType>& ex,
+        std::vector<MatType>& v,
+        std::vector<MatType>& vdef,
+        const std::string& v_prefix="v_",
+        const std::string& v_suffix="") {
+      return MatType::shared(ex, v, vdef, v_prefix, v_suffix);
+    }
+
+    /** \brief Given a repeated matrix, computes the sum of repeated parts
+
+        \identifier{1d7} */
+    inline friend MatType repsum(const MatType &A, casadi_int n, casadi_int m=1) {
+      return MatType::repsum(A, n, m);
+    }
+
+    ///@{
+    /** \brief Smallest element in a matrix
+
+        \identifier{1d8} */
+    friend inline MatType mmin(const MatType& x) {
+      return MatType::mmin(x);
+    }
+    ///@}
+
+    ///@{
+    /** \brief Largest element in a matrix
+
+        \identifier{1d9} */
+    friend inline MatType mmax(const MatType& x) {
+      return MatType::mmax(x);
+    }
+    ///@}
+
+    ///@{
+    /// Functions called by friend functions defined here
+    static MatType jtimes(const MatType &ex, const MatType &arg,
+                          const MatType &v, bool tr=false, const Dict& opts=Dict());
+    static MatType gradient(const MatType &ex, const MatType &arg, const Dict& opts=Dict());
+    static MatType tangent(const MatType &ex, const MatType &arg, const Dict& opts=Dict());
+    static MatType linearize(const MatType& f, const MatType& x, const MatType& x0,
+      const Dict& opts=Dict());
+    static MatType mpower(const MatType &x, const MatType &y);
+    static MatType soc(const MatType &x, const MatType &y);
+    ///@}
+
+/** @} */
+#endif // SWIG
+
+    /** @name Construct symbolic primitives
+        The "sym" function is intended to work in a similar way as "sym" used
+        in the Symbolic Toolbox for Matlab but instead creating a
+        CasADi symbolic primitive.
+    */
+    ///@{
+
+    /** \brief Create an nrow-by-ncol symbolic primitive
+
+        \identifier{1da} */
+    static MatType sym(const std::string& name, casadi_int nrow=1, casadi_int ncol=1) {
+      return sym(name, Sparsity::dense(nrow, ncol));
+    }
+
+    /** \brief  Construct a symbolic primitive with given dimensions
+
+        \identifier{1db} */
+    static MatType sym(const std::string& name, const std::pair<casadi_int, casadi_int> &rc) {
+      return sym(name, rc.first, rc.second);
+    }
+
+    /** \brief Create symbolic primitive with a given sparsity pattern
+
+        \identifier{1dc} */
+    static MatType sym(const std::string& name, const Sparsity& sp) {
+      return MatType::_sym(name, sp);
+    }
+
+    /** \brief Create a vector of length p with with matrices
+
+     * with symbolic primitives of given sparsity
+
+        \identifier{1dd} */
+    static std::vector<MatType > sym(const std::string& name, const Sparsity& sp, casadi_int p);
+
+    /** \brief Create a vector of length p with nrow-by-ncol symbolic primitives
+
+        \identifier{1de} */
+    static std::vector<MatType > sym(const std::string& name, casadi_int nrow,
+        casadi_int ncol, casadi_int p) {
+      return sym(name, Sparsity::dense(nrow, ncol), p);
+    }
+
+    /** \brief Create a vector of length r of vectors of length p with
+
+     * symbolic primitives with given sparsity
+
+        \identifier{1df} */
+    static std::vector<std::vector<MatType> >
+      sym(const std::string& name, const Sparsity& sp, casadi_int p, casadi_int r);
+
+    /** \brief Create a vector of length r of vectors of length p
+
+     * with nrow-by-ncol symbolic primitives
+
+        \identifier{1dg} */
+    static std::vector<std::vector<MatType> >
+      sym(const std::string& name, casadi_int nrow, casadi_int ncol, casadi_int p, casadi_int r) {
+      return sym(name, Sparsity::dense(nrow, ncol), p, r);
+    }
+    ///@}
+
+    ///@{
+    /** \brief Create a dense matrix or a matrix with specified sparsity with all entries zero
+
+        \identifier{1dh} */
+    static MatType zeros(casadi_int nrow=1, casadi_int ncol=1) {
+      return zeros(Sparsity::dense(nrow, ncol));
+    }
+    static MatType zeros(const Sparsity& sp) { return MatType(sp, 0, false);}
+    static MatType zeros(const std::pair<casadi_int, casadi_int>& rc) {
+      return zeros(rc.first, rc.second);
+    }
+    ///@}
+
+    ///@{
+    /** \brief Create a dense matrix or a matrix with specified sparsity with all entries one
+
+        \identifier{1di} */
+    static MatType ones(casadi_int nrow=1, casadi_int ncol=1) {
+      return ones(Sparsity::dense(nrow, ncol));
+    }
+    static MatType ones(const Sparsity& sp) { return MatType(sp, 1, false);}
+    static MatType ones(const std::pair<casadi_int, casadi_int>& rc) {
+      return ones(rc.first, rc.second);
+    }
+    ///@}
+  };
+
+  // Throw informative error message
+  #define CASADI_THROW_ERROR(FNAME, WHAT) \
+  throw CasadiException("Error in " + MatType::type_name() \
+    + "::" FNAME " at " + CASADI_WHERE + ":\n" + std::string(WHAT));
+
+#ifndef SWIG
+  // Implementations
+  template<typename MatType>
+  const Sparsity& GenericMatrix<MatType>::sparsity() const {
+    return self().sparsity();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::nnz() const {
+    return sparsity().nnz();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::nnz_lower() const {
+    return sparsity().nnz_lower();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::nnz_upper() const {
+    return sparsity().nnz_upper();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::nnz_diag() const {
+    return sparsity().nnz_diag();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::numel() const {
+    return sparsity().numel();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::size1() const {
+    return sparsity().size1();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::size2() const {
+    return sparsity().size2();
+  }
+
+  template<typename MatType>
+  std::pair<casadi_int, casadi_int> GenericMatrix<MatType>::size() const {
+    return sparsity().size();
+  }
+
+  template<typename MatType>
+  casadi_int GenericMatrix<MatType>::size(casadi_int axis) const {
+    return sparsity().size(axis);
+  }
+
+  template<typename MatType>
+  std::string GenericMatrix<MatType>::dim(bool with_nz) const {
+    return sparsity().dim(with_nz);
+  }
+
+  template<typename MatType>
+  bool GenericMatrix<MatType>::is_scalar(bool scalar_and_dense) const {
+    return sparsity().is_scalar(scalar_and_dense);
+  }
+
+#endif // SWIG
+
+  template<typename MatType>
+  std::vector<MatType> GenericMatrix<MatType>::sym(const std::string& name,
+                                                   const Sparsity& sp, casadi_int p) {
+    std::vector<MatType> ret(p);
+    std::stringstream ss;
+    for (casadi_int k=0; k<p; ++k) {
+      ss.str("");
+      ss << name << k;
+      ret[k] = sym(ss.str(), sp);
+    }
+    return ret;
+  }
+
+  template<typename MatType>
+  std::vector<std::vector<MatType> > GenericMatrix<MatType>::sym(const std::string& name,
+                                                                  const Sparsity& sp, casadi_int p,
+                                                                  casadi_int r) {
+    std::vector<std::vector<MatType> > ret(r);
+    for (casadi_int k=0; k<r; ++k) {
+      std::stringstream ss;
+      ss << name << "_" << k;
+      ret[k] = sym(ss.str(), sp, p);
+    }
+    return ret;
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::linspace(const MatType& a, const MatType& b, casadi_int nsteps) {
+    std::vector<MatType> ret(nsteps);
+    ret[0] = a;
+    MatType step = (b-a)/static_cast<MatType>(nsteps-1);
+
+    for (casadi_int i=1; i<nsteps-1; ++i)
+      ret[i] = a + i * step;
+
+    ret[nsteps-1] = b;
+    return vertcat(ret);
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::cross(const MatType& a, const MatType& b, casadi_int dim) {
+    casadi_assert(a.size1()==b.size1() && a.size2()==b.size2(),
+      "cross(a, b): Inconsistent dimensions. Dimension of a ("
+      + a.dim() + " ) must equal that of b (" + b.dim() + ").");
+
+    casadi_assert(a.size1()==3 || a.size2()==3,
+      "cross(a, b): One of the dimensions of a should have length 3, but got "
+        + a.dim() + ".");
+    casadi_assert(dim==-1 || dim==1 || dim==2,
+      "cross(a, b, dim): Dim must be 1, 2 or -1 (automatic).");
+
+    std::vector<MatType> ret(3);
+
+    bool t = a.size1()==3;
+
+    if (dim==1) t = true;
+    if (dim==2) t = false;
+
+    MatType a1 = t ? a(0, Slice()) : a(Slice(), 0);
+    MatType a2 = t ? a(1, Slice()) : a(Slice(), 1);
+    MatType a3 = t ? a(2, Slice()) : a(Slice(), 2);
+
+    MatType b1 = t ? b(0, Slice()) : b(Slice(), 0);
+    MatType b2 = t ? b(1, Slice()) : b(Slice(), 1);
+    MatType b3 = t ? b(2, Slice()) : b(Slice(), 2);
+
+    ret[0] = a2*b3-a3*b2;
+    ret[1] = a3*b1-a1*b3;
+    ret[2] = a1*b2-a2*b1;
+
+    return t ? vertcat(ret) : horzcat(ret);
+  }
+
+  double CASADI_EXPORT index_interp1d(const std::vector<double>& x, double xq,
+    bool equidistant=false);
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::interp1d(const std::vector<double>& x, const MatType& v,
+      const std::vector<double>& xq, const std::string& mode, bool equidistant) {
+
+    bool mode_floor = mode == "floor";
+    bool mode_ceil = mode == "ceil";
+
+    casadi_assert_dev(is_increasing(x));
+
+    casadi_assert(x.size()==v.size1(),
+      "interp1d(x, v, xq): dimensions mismatch. v expected to have " + str(x.size()) + " rows,"
+      " but got " + str(v.size1()) + " instead.");
+
+    // Need at least two elements
+    casadi_assert(x.size()>=2, "interp1d(x, v, xq): x must be at least length 2.");
+
+    // Vectors to compose a sparse matrix
+    std::vector<double> val;
+    std::vector<casadi_int> colind(1, 0);
+    std::vector<casadi_int> row;
+
+    // Number of nonzeros in to-be composed matrix
+    casadi_int nnz = 0;
+    for (casadi_int i=0;i<xq.size();++i) {
+      // Obtain index corresponding to xq[i]
+      double ind = index_interp1d(x, xq[i], equidistant);
+
+      if (mode_floor) ind = floor(ind);
+      if (mode_ceil) ind = ceil(ind);
+
+      // Split into integer and fractional part
+      double int_partd;
+      double frac_part = modf(ind, &int_partd);
+      casadi_int int_part = static_cast<casadi_int>(int_partd);
+
+      if (frac_part==0) {
+         // Create a single entry
+         val.push_back(1);
+         row.push_back(int_part);
+         nnz+=1;
+         colind.push_back(nnz);
+       } else {
+         // Create a double entry
+         val.push_back(1-frac_part);
+         val.push_back(frac_part);
+         row.push_back(int_part);
+         row.push_back(int_part+1);
+         nnz+=2;
+         colind.push_back(nnz);
+       }
+     }
+
+     // Construct sparsity for composed matrix
+     Sparsity sp(x.size(), xq.size() , colind, row);
+
+     return MatType::mtimes(MatType(sp, val).T(), v);
+
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::skew(const MatType& a) {
+    casadi_assert(a.is_vector() && (a.size1()==3 || a.size2()==3),
+      "skew(a): Expecting 3-vector, got " + a.dim() + ".");
+
+    MatType x = a(0);
+    MatType y = a(1);
+    MatType z = a(2);
+    return blockcat(std::vector< std::vector<MatType> >({{0, -z, y}, {z, 0, -x}, {-y, x, 0}}));
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::inv_skew(const MatType& a) {
+    casadi_assert(a.size1()==3 && a.size2()==3,
+      "inv_skew(a): Expecting 3-by-3 matrix, got " + a.dim() + ".");
+
+    return 0.5*vertcat(std::vector<MatType>({a(2, 1)-a(1, 2), a(0, 2)-a(2, 0), a(1, 0)-a(0, 1)}));
+  }
+
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::tril2symm(const MatType& x) {
+    casadi_assert(x.is_square(),
+      "Shape error in tril2symm. Expecting square shape but got " + x.dim());
+    casadi_assert(x.nnz_upper()-x.nnz_diag()==0,
+      "Sparsity error in tril2symm. Found above-diagonal entries in argument: " + x.dim());
+    return x +  x.T() - diag(diag(x));
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::repsum(const MatType& x, casadi_int n, casadi_int m) {
+    casadi_assert_dev(x.size1() % n==0);
+    casadi_assert_dev(x.size2() % m==0);
+    std::vector< std::vector< MatType> > s =
+      blocksplit(x, x.size1()/n, x.size2()/m);
+    MatType sum = 0;
+    for (casadi_int i=0;i<s.size();++i) {
+      for (casadi_int j=0;j<s[i].size();++j) {
+        sum = sum + s[i][j];
+      }
+    }
+    return sum;
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::triu2symm(const MatType& x) {
+    casadi_assert(x.is_square(),
+      "Shape error in triu2symm. Expecting square shape but got " + x.dim());
+    casadi_assert(x.nnz_lower()-x.nnz_diag()==0,
+      "Sparsity error in triu2symm. Found below-diagonal entries in argument: " + x.dim());
+    return x + x.T() - diag(diag(x));
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::bilin(const MatType& A, const MatType& x,
+                                        const MatType& y) {
+    // Check/correct x
+    casadi_assert_dev(x.is_vector());
+    if (!x.is_column()) return bilin(A, x.T(), y);
+    if (!x.is_dense()) return bilin(A, densify(x), y);
+
+    // Check/correct y
+    casadi_assert_dev(y.is_vector());
+    if (!y.is_column()) return bilin(A, x, y.T());
+    if (!y.is_dense()) return bilin(A, x, densify(y));
+
+    // Assert dimensions
+    casadi_assert(x.size1()==A.size1() && y.size1()==A.size2(),
+      "Dimension mismatch. Got x.size1() = " + str(x.size1())
+      + " and y.size1() = " + str(y.size1()) + " but A.size() = " + str(A.size()));
+
+    // Call the class specific method
+    return MatType::_bilin(A, x, y);
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::rank1(const MatType& A, const MatType& alpha,
+                                        const MatType& x, const MatType& y) {
+    // Check/correct x
+    casadi_assert_dev(x.is_vector());
+    if (!x.is_column()) return rank1(A, alpha, x.T(), y);
+    if (!x.is_dense()) return rank1(A, alpha, densify(x), y);
+
+    // Check/correct y
+    casadi_assert_dev(y.is_vector());
+    if (!y.is_column()) return rank1(A, alpha, x, y.T());
+    if (!y.is_dense()) return rank1(A, alpha, x, densify(y));
+
+    // Check alpha, quick return
+    casadi_assert_dev(alpha.is_scalar());
+    if (!alpha.is_dense()) return A;
+
+    // Assert dimensions
+    casadi_assert(x.size1()==A.size1() && y.size1()==A.size2(),
+      "Dimension mismatch. Got x.size1() = " + str(x.size1())
+        + " and y.size1() = " + str(y.size1())
+        + " but A.size() = " + str(A.size()));
+
+    // Call the class specific method
+    return MatType::_rank1(A, alpha, x, y);
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::logsumexp(const MatType& x) {
+    casadi_assert(x.is_dense(), "Argument must be dense");
+    casadi_assert(x.is_column(), "Argument must be column vector");
+    // Call the class specific method
+    return MatType::_logsumexp(x);
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::jtimes(const MatType &ex, const MatType &arg,
+                                         const MatType &v, bool tr, const Dict& opts) {
+    try {
+      // Assert consistent input dimensions
+      if (tr) {
+        casadi_assert(v.size1() == ex.size1() && v.size2() % ex.size2() == 0,
+                      "'v' has inconsistent dimensions: "
+                      " v " + v.dim(false) + ", ex " + ex.dim(false) + ".");
+      } else {
+        casadi_assert(v.size1() == arg.size1() && v.size2() % arg.size2() == 0,
+                      "'v' has inconsistent dimensions: "
+                      " v " + v.dim(false) + ", arg " + arg.dim(false) + ".");
+      }
+
+      // Quick return if no seeds
+      if (v.is_empty()) return MatType(tr ? arg.size1() : ex.size1(), 0);
+
+      // Split up the seed into its components
+      std::vector<MatType> w = horzsplit(v, tr ? ex.size2() : arg.size2());
+
+      // Seeds as a vector of vectors
+      std::vector<std::vector<MatType> > ww(w.size());
+      for (casadi_int i=0; i<w.size(); ++i) ww[i] = {w[i]};
+
+      // Calculate directional derivatives
+      if (tr) {
+        ww = reverse({ex}, {arg}, ww, opts);
+      } else {
+        ww = forward({ex}, {arg}, ww, opts);
+      }
+
+      // Get results
+      for (casadi_int i=0; i<w.size(); ++i) w[i] = ww[i][0];
+      return horzcat(w);
+    } catch (std::exception& e) {
+      CASADI_THROW_ERROR("jtimes", e.what());
+    }
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::gradient(const MatType &ex, const MatType &arg,
+      const Dict& opts) {
+    try {
+      casadi_assert(ex.is_scalar(),
+                    "'gradient' only defined for scalar outputs: Use 'jacobian' instead.");
+      return project(jtimes(ex, arg, MatType::ones(ex.sparsity()), true, opts), arg.sparsity());
+    } catch (std::exception& e) {
+      CASADI_THROW_ERROR("gradient", e.what());
+    }
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::tangent(const MatType &ex, const MatType &arg,
+      const Dict& opts) {
+    try {
+      casadi_assert(arg.is_scalar(),
+                    "'tangent' only defined for scalar inputs: Use 'jacobian' instead.");
+      return project(jtimes(ex, arg, MatType::ones(arg.sparsity()), false, opts), ex.sparsity());
+    } catch (std::exception& e) {
+      CASADI_THROW_ERROR("tangent", e.what());
+    }
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::
+  linearize(const MatType& f, const MatType& x, const MatType& x0, const Dict& opts) {
+    MatType x_lin = MatType::sym("x_lin", x.sparsity());
+    // mismatching dimensions
+    if (x0.size() != x.size()) {
+      // Scalar x0 is ok
+      if (x0.sparsity().is_scalar()) {
+        return linearize(f, x, MatType(x.sparsity(), x0));
+      }
+      casadi_error("Dimension mismatch in 'linearize'");
+    }
+    return substitute(f + jtimes(f, x, x_lin, false, opts),
+      MatType::vertcat({x_lin, x}), MatType::vertcat({x, x0}));
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::mpower(const MatType& a,
+                                            const MatType& b) {
+    if (a.is_scalar() && b.is_scalar()) return pow(a, b);
+    casadi_assert(a.is_square() && b.is_constant() && b.is_scalar(),
+      "Not Implemented");
+    double bv = static_cast<double>(b);
+    casadi_int N = static_cast<casadi_int>(bv);
+    casadi_assert(bv-static_cast<double>(N)==0, "mpower only defined for integer powers.");
+    casadi_assert(bv==N, "Not Implemented");
+    if (N<0) return inv(mpower(a, -N));
+    if (N==0) return MatType::eye(a.size1());
+    if (N==1) return a;
+    if (N % 2 == 0) {
+      MatType h = mpower(a, N/2); // NOLINT
+      return MatType::mtimes(h, h);
+    } else {
+      return MatType::mtimes(mpower(a, N-1), a);
+    }
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::soc(const MatType& x,
+                                            const MatType& y) {
+    casadi_assert(y.is_scalar(), "y needs to be scalar. Got " + y.dim() + ".");
+    casadi_assert(x.is_vector(), "x needs to be a vector. Got " + x.dim() + ".");
+
+    MatType x_col = x.is_column() ? x : x.T();
+
+    x_col = x_col.nz(Slice()); // NOLINT
+
+    casadi_int n = x_col.numel();
+    return blockcat(y*MatType::eye(n), x_col, x_col.T(), y);
+  }
+
+  template<typename MatType>
+  bool GenericMatrix<MatType>::is_linear(const MatType &expr, const MatType &var) {
+    return !any(MatType::which_depends(expr, var, 2, true));
+  }
+
+  template<typename MatType>
+  bool GenericMatrix<MatType>::is_quadratic(const MatType &expr, const MatType &var) {
+    return is_linear(gradient(expr, var), var);
+  }
+
+  template<typename MatType>
+  void GenericMatrix<MatType>::quadratic_coeff(const MatType &expr, const MatType &var,
+          MatType& A, MatType& b, MatType& c, bool check) {
+    casadi_assert(expr.is_scalar(), "'quadratic_coeff' only defined for scalar expressions.");
+    A = hessian(expr, var);
+    b = substitute(jacobian(expr, var), var, 0).T();
+    if (check)
+      casadi_assert(!depends_on(A, var), "'quadratic_coeff' called on non-quadratic expression.");
+    c = substitute(expr, var, 0);
+  }
+
+  template<typename MatType>
+  void GenericMatrix<MatType>::linear_coeff(const MatType &expr, const MatType &var,
+          MatType& A, MatType& b, bool check) {
+    casadi_assert(expr.is_vector(), "'linear_coeff' only defined for vector expressions.");
+    if (check)
+      casadi_assert(is_linear(expr, var), "'linear_coeff' called on non-linear expression.");
+    A = substitute(jacobian(expr, var), var, 0);
+    b = vec(substitute(expr, var, 0));
+  }
+
+  template<typename MatType>
+  MatType GenericMatrix<MatType>::diff(const MatType& x, casadi_int n, casadi_int axis) {
+    casadi_assert(axis==-1 || axis==0 || axis==1, "Axis argument invalid");
+    casadi_assert(n>=1, "n argument invalid");
+
+    MatType ret = x;
+    for (casadi_int i=0;i<n;++i) {
+      // Matlab's special case
+      if (axis==-1 && ret.is_scalar()) return MatType();
+
+      casadi_int local_axis = (axis==-1) ? ret.is_row() : axis;
+      if (local_axis==0) {
+        if (ret.size1()<=1) {
+          ret = MatType::zeros(0, ret.size2());
+        } else {
+          ret = ret(Slice(1, ret.size1()), Slice())-ret(Slice(0, ret.size1()-1), Slice());
+        }
+      } else {
+        if (ret.size2()<=1) {
+          ret = MatType::zeros(ret.size1(), 0);
+        } else {
+          ret = ret(Slice(), Slice(1, ret.size2()))-ret(Slice(), Slice(0, ret.size2()-1));
+        }
+      }
+    }
+    return ret;
+  }
+
+#undef CASADI_THROW_ERROR
+
+} // namespace casadi
+
+#endif // CASADI_GENERIC_MATRIX_HPP