/* *_________________________________________________________________________* * POEMS: PARALLELIZABLE OPEN SOURCE EFFICIENT MULTIBODY SOFTWARE * * DESCRIPTION: SEE READ-ME * * FILE NAME: mat6x6.h * * AUTHORS: See Author List * * GRANTS: See Grants List * * COPYRIGHT: (C) 2005 by Authors as listed in Author's List * * LICENSE: Please see License Agreement * * DOWNLOAD: Free at www.rpi.edu/~anderk5 * * ADMINISTRATOR: Prof. Kurt Anderson * * Computational Dynamics Lab * * Rensselaer Polytechnic Institute * * 110 8th St. Troy NY 12180 * * CONTACT: anderk5@rpi.edu * *_________________________________________________________________________*/ #ifndef MAT6X6_H #define MAT6X6_H #include "virtualmatrix.h" class Matrix; class Mat3x3; class Vect6; class Vect3; class Mat6x6 : public VirtualMatrix { double elements[6][6]; public: Mat6x6(); ~Mat6x6(); Mat6x6(const Mat6x6& A); // copy constructor Mat6x6(const VirtualMatrix& A); // copy constructor double& operator_2int(int i, int j); // array access double Get_2int(int i, int j) const; void Set_2int(int i, int j, double value); double BasicGet_2int(int i, int j) const; void BasicSet_2int(int i, int j, double value); void BasicIncrement_2int(int i, int j, double value); void Const(double value); MatrixType GetType() const; std::istream& ReadData(std::istream& c); // input std::ostream& WriteData(std::ostream& c) const; // output void AssignVM(const VirtualMatrix& A); Mat6x6& operator=(const Mat6x6& A); // assignment operator Mat6x6& operator=(const VirtualMatrix& A); // overloaded = Mat6x6& operator*=(double b); Mat6x6& Identity(); friend Mat6x6 T(const Mat6x6& A); // a wasteful transpose // fast matrix operations friend void FastAdd(Mat6x6& A, Mat6x6& B, Mat6x6& C); friend void FastSubt(Mat6x6& A, Mat6x6& B, Mat6x6& C); friend void FastMultT(Matrix& A, Matrix& B, Mat6x6& C); friend void FastMult(Mat6x6& A, Mat6x6& B, Mat6x6& C); // C = A*B friend void FastMult(Mat6x6& A, Matrix& B, Matrix& C); friend void FastMultT(Mat6x6& A, Mat6x6& B, Mat6x6& C); // C = A*B^T friend void FastTMult(Mat6x6& A, Mat6x6& B, Mat6x6& C); // C = A^T*B friend void FastMult(Mat6x6& A, Vect6& B, Vect6& C); friend void FastTMult(Mat6x6& A, Vect6& B, Vect6& C); friend void FastLDLT(Mat6x6& A, Mat6x6& C); friend void FastLDLTSubs(Mat6x6& LD, Mat6x6& B, Mat6x6& C); friend void FastLDLTSubs(Mat6x6& LD, Vect6& B, Vect6& C); friend void OnPopulateSC(Vect3& gamma, Mat3x3& C, Mat6x6& SC); friend void OnPopulateSI(Mat3x3& inertia, double mass, Mat6x6& sI); }; #endif