Oskar Weigl / tvmet

This is the Tiny Vector Matrix Expression Templates library found at http://tvmet.sourceforge.net. It is the fastest and most compact matrix lib out there (for < 10x10 matricies). I have done some minor tweaks to make it compile for mbed. For examples and hints on how to use, see: http://tvmet.sourceforge.net/usage.html

Dependents:   Eurobot_2012_Secondary

xpr/MMProductTransposed.h@0:feb4117d16d8, 2012-03-28 (annotated)

Committer:
madcowswe
Date:
Wed Mar 28 15:53:45 2012 +0000
Revision:
0:feb4117d16d8

        

        
Who changed what in which revision?

        

            
                
UserRevisionLine numberNew contents of line

            
            


madcowswe

0:feb4117d16d8


     1

/*





madcowswe

0:feb4117d16d8


     2

 * Tiny Vector Matrix Library





madcowswe

0:feb4117d16d8


     3

 * Dense Vector Matrix Libary of Tiny size using Expression Templates





madcowswe

0:feb4117d16d8


     4

 *





madcowswe

0:feb4117d16d8


     5

 * Copyright (C) 2001 - 2007 Olaf Petzold <opetzold@users.sourceforge.net>





madcowswe

0:feb4117d16d8


     6

 *





madcowswe

0:feb4117d16d8


     7

 * This library is free software; you can redistribute it and/or





madcowswe

0:feb4117d16d8


     8

 * modify it under the terms of the GNU Lesser General Public





madcowswe

0:feb4117d16d8


     9

 * License as published by the Free Software Foundation; either





madcowswe

0:feb4117d16d8


    10

 * version 2.1 of the License, or (at your option) any later version.





madcowswe

0:feb4117d16d8


    11

 *





madcowswe

0:feb4117d16d8


    12

 * This library is distributed in the hope that it will be useful,





madcowswe

0:feb4117d16d8


    13

 * but WITHOUT ANY WARRANTY; without even the implied warranty of





madcowswe

0:feb4117d16d8


    14

 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU





madcowswe

0:feb4117d16d8


    15

 * Lesser General Public License for more details.





madcowswe

0:feb4117d16d8


    16

 *





madcowswe

0:feb4117d16d8


    17

 * You should have received a copy of the GNU Lesser General Public





madcowswe

0:feb4117d16d8


    18

 * License along with this library; if not, write to the Free Software





madcowswe

0:feb4117d16d8


    19

 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA





madcowswe

0:feb4117d16d8


    20

 *





madcowswe

0:feb4117d16d8


    21

 * $Id: MMProductTransposed.h,v 1.20 2007-06-23 15:58:59 opetzold Exp $





madcowswe

0:feb4117d16d8


    22

 */





madcowswe

0:feb4117d16d8


    23







madcowswe

0:feb4117d16d8


    24

#ifndef TVMET_XPR_MMPRODUCT_TRANSPOSED_H





madcowswe

0:feb4117d16d8


    25

#define TVMET_XPR_MMPRODUCT_TRANSPOSED_H





madcowswe

0:feb4117d16d8


    26







madcowswe

0:feb4117d16d8


    27

#include <tvmet/meta/Gemm.h>





madcowswe

0:feb4117d16d8


    28

#include <tvmet/loop/Gemm.h>





madcowswe

0:feb4117d16d8


    29







madcowswe

0:feb4117d16d8


    30

namespace tvmet {





madcowswe

0:feb4117d16d8


    31







madcowswe

0:feb4117d16d8


    32







madcowswe

0:feb4117d16d8


    33

/**





madcowswe

0:feb4117d16d8


    34

 * \class XprMMProductTransposed MMProductTransposed.h "tvmet/xpr/MMProductTransposed.h"





madcowswe

0:feb4117d16d8


    35

 * \brief Expression for transpose(matrix-matrix product).





madcowswe

0:feb4117d16d8


    36

 *        Using formula:





madcowswe

0:feb4117d16d8


    37

 *        \f[





madcowswe

0:feb4117d16d8


    38

 *        (M_1\,M_2)^T





madcowswe

0:feb4117d16d8


    39

 *        \f]





madcowswe

0:feb4117d16d8


    40

 * \note The Rows2 has to be  equal to Cols1.





madcowswe

0:feb4117d16d8


    41

 *       The result is a (Cols2 x Rows1) matrix.





madcowswe

0:feb4117d16d8


    42

 */





madcowswe

0:feb4117d16d8


    43

template<class E1, std::size_t Rows1, std::size_t Cols1,





madcowswe

0:feb4117d16d8


    44

	 class E2, std::size_t Cols2>





madcowswe

0:feb4117d16d8


    45

class XprMMProductTransposed





madcowswe

0:feb4117d16d8


    46

  : public TvmetBase< XprMMProductTransposed<E1, Rows1, Cols1, E2, Cols2> >





madcowswe

0:feb4117d16d8


    47

{





madcowswe

0:feb4117d16d8


    48

private:





madcowswe

0:feb4117d16d8


    49

  XprMMProductTransposed();





madcowswe

0:feb4117d16d8


    50

  XprMMProductTransposed& operator=(const XprMMProductTransposed&);





madcowswe

0:feb4117d16d8


    51







madcowswe

0:feb4117d16d8


    52

public:





madcowswe

0:feb4117d16d8


    53

  typedef typename PromoteTraits<





madcowswe

0:feb4117d16d8


    54

    typename E1::value_type,





madcowswe

0:feb4117d16d8


    55

    typename E2::value_type





madcowswe

0:feb4117d16d8


    56

  >::value_type							value_type;





madcowswe

0:feb4117d16d8


    57







madcowswe

0:feb4117d16d8


    58

public:





madcowswe

0:feb4117d16d8


    59

  /** Complexity counter. */





madcowswe

0:feb4117d16d8


    60

  enum {





madcowswe

0:feb4117d16d8


    61

    ops_lhs   = E1::ops,





madcowswe

0:feb4117d16d8


    62

    ops_rhs   = E2::ops,





madcowswe

0:feb4117d16d8


    63

    M = Rows1 * Cols1 * Cols2,





madcowswe

0:feb4117d16d8


    64

    N = Rows1 * (Cols1-1) * Cols2,





madcowswe

0:feb4117d16d8


    65

    ops_plus  = M * NumericTraits<value_type>::ops_plus,





madcowswe

0:feb4117d16d8


    66

    ops_muls  = N * NumericTraits<value_type>::ops_muls,





madcowswe

0:feb4117d16d8


    67

    ops       = ops_plus + ops_muls,





madcowswe

0:feb4117d16d8


    68

    use_meta  = Cols2*Rows1 < TVMET_COMPLEXITY_MM_TRIGGER ? true : false





madcowswe

0:feb4117d16d8


    69

  };





madcowswe

0:feb4117d16d8


    70







madcowswe

0:feb4117d16d8


    71

public:





madcowswe

0:feb4117d16d8


    72

  /** Constructor. */





madcowswe

0:feb4117d16d8


    73

  explicit XprMMProductTransposed(const E1& lhs, const E2& rhs)





madcowswe

0:feb4117d16d8


    74

    : m_lhs(lhs), m_rhs(rhs) { }





madcowswe

0:feb4117d16d8


    75







madcowswe

0:feb4117d16d8


    76

  /** Copy Constructor. Not explicit! */





madcowswe

0:feb4117d16d8


    77

#if defined(TVMET_OPTIMIZE_XPR_MANUAL_CCTOR)





madcowswe

0:feb4117d16d8


    78

  XprMMProductTransposed(const XprMMProductTransposed& e)





madcowswe

0:feb4117d16d8


    79

    : m_lhs(e.m_lhs), m_rhs(e.m_rhs)





madcowswe

0:feb4117d16d8


    80

  { }





madcowswe

0:feb4117d16d8


    81

#endif





madcowswe

0:feb4117d16d8


    82







madcowswe

0:feb4117d16d8


    83

private:





madcowswe

0:feb4117d16d8


    84

  /** Wrapper for meta gemm. */





madcowswe

0:feb4117d16d8


    85

  static inline





madcowswe

0:feb4117d16d8


    86

  value_type do_gemm(dispatch<true>, const E1& lhs, const E2& rhs, std::size_t i, std::size_t j) {





madcowswe

0:feb4117d16d8


    87

    return meta::gemm<Rows1, Cols1,





madcowswe

0:feb4117d16d8


    88

                      Cols2,





madcowswe

0:feb4117d16d8


    89

                      0>::prod(lhs, rhs, i, j);





madcowswe

0:feb4117d16d8


    90

  }





madcowswe

0:feb4117d16d8


    91







madcowswe

0:feb4117d16d8


    92

  /** Wrapper for loop gemm. */





madcowswe

0:feb4117d16d8


    93

  static inline





madcowswe

0:feb4117d16d8


    94

  value_type do_gemm(dispatch<false>, const E1& lhs, const E2& rhs, std::size_t i, std::size_t j) {





madcowswe

0:feb4117d16d8


    95

    return loop::gemm<Rows1, Cols1, Cols2>::prod(lhs, rhs, i, j);





madcowswe

0:feb4117d16d8


    96

  }





madcowswe

0:feb4117d16d8


    97







madcowswe

0:feb4117d16d8


    98

public:





madcowswe

0:feb4117d16d8


    99

  /** index operator for arrays/matrices */





madcowswe

0:feb4117d16d8


   100

  value_type operator()(std::size_t i, std::size_t j) const {





madcowswe

0:feb4117d16d8


   101

    TVMET_RT_CONDITION((i < Cols2) && (j < Rows1), "XprMMProductTransposed Bounce Violation")





madcowswe

0:feb4117d16d8


   102

    return do_gemm(dispatch<use_meta>(), m_lhs, m_rhs, j, i);





madcowswe

0:feb4117d16d8


   103

  }





madcowswe

0:feb4117d16d8


   104







madcowswe

0:feb4117d16d8


   105

public: // debugging Xpr parse tree





madcowswe

0:feb4117d16d8


   106

  void print_xpr(std::ostream& os, std::size_t l=0) const {





madcowswe

0:feb4117d16d8


   107

    os << IndentLevel(l++)





madcowswe

0:feb4117d16d8


   108

       << "XprMMProductTransposed["





madcowswe

0:feb4117d16d8


   109

       << (use_meta ? "M" :  "L") << ", O=" << ops





madcowswe

0:feb4117d16d8


   110

       << ", (O1=" << ops_lhs << ", O2=" << ops_rhs << ")]<"





madcowswe

0:feb4117d16d8


   111

       << std::endl;





madcowswe

0:feb4117d16d8


   112

    m_lhs.print_xpr(os, l);





madcowswe

0:feb4117d16d8


   113

    os << IndentLevel(l)





madcowswe

0:feb4117d16d8


   114

       << "R1=" << Rows1 << ", C1=" << Cols1 << ",\n";





madcowswe

0:feb4117d16d8


   115

    m_rhs.print_xpr(os, l);





madcowswe

0:feb4117d16d8


   116

    os << IndentLevel(l)





madcowswe

0:feb4117d16d8


   117

       << "C2=" << Cols2 << ",\n"





madcowswe

0:feb4117d16d8


   118

       << IndentLevel(l)





madcowswe

0:feb4117d16d8


   119

       << "\n"





madcowswe

0:feb4117d16d8


   120

       << IndentLevel(--l)





madcowswe

0:feb4117d16d8


   121

       << ">," << std::endl;





madcowswe

0:feb4117d16d8


   122

  }





madcowswe

0:feb4117d16d8


   123







madcowswe

0:feb4117d16d8


   124

private:





madcowswe

0:feb4117d16d8


   125

  const E1		 				m_lhs;





madcowswe

0:feb4117d16d8


   126

  const E2						m_rhs;





madcowswe

0:feb4117d16d8


   127

};





madcowswe

0:feb4117d16d8


   128







madcowswe

0:feb4117d16d8


   129







madcowswe

0:feb4117d16d8


   130

} // namespace tvmet





madcowswe

0:feb4117d16d8


   131







madcowswe

0:feb4117d16d8


   132

#endif // TVMET_XPR_MMPRODUCT_TRANSPOSED_H





madcowswe

0:feb4117d16d8


   133







madcowswe

0:feb4117d16d8


   134

// Local Variables:





madcowswe

0:feb4117d16d8


   135

// mode:C++





madcowswe

0:feb4117d16d8


   136

// tab-width:8





madcowswe

0:feb4117d16d8


   137

// End: