Oskar Weigl
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
loop/Gemtm.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?
| User | Revision | Line number | New 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: Gemtm.h,v 1.9 2007-06-23 15:58:59 opetzold Exp $ |
| madcowswe | 0:feb4117d16d8 | 22 | */ |
| madcowswe | 0:feb4117d16d8 | 23 | |
| madcowswe | 0:feb4117d16d8 | 24 | #ifndef TVMET_LOOP_GEMTM_H |
| madcowswe | 0:feb4117d16d8 | 25 | #define TVMET_LOOP_GEMTM_H |
| madcowswe | 0:feb4117d16d8 | 26 | |
| madcowswe | 0:feb4117d16d8 | 27 | namespace tvmet { |
| madcowswe | 0:feb4117d16d8 | 28 | |
| madcowswe | 0:feb4117d16d8 | 29 | namespace loop { |
| madcowswe | 0:feb4117d16d8 | 30 | |
| madcowswe | 0:feb4117d16d8 | 31 | |
| madcowswe | 0:feb4117d16d8 | 32 | /** |
| madcowswe | 0:feb4117d16d8 | 33 | * \class gemtm Gemtm.h "tvmet/loop/Gemtm.h" |
| madcowswe | 0:feb4117d16d8 | 34 | * \brief class for matrix-matrix product using loop unrolling. |
| madcowswe | 0:feb4117d16d8 | 35 | * using formula |
| madcowswe | 0:feb4117d16d8 | 36 | * \f[ |
| madcowswe | 0:feb4117d16d8 | 37 | * M_1^{T}\,M_2 |
| madcowswe | 0:feb4117d16d8 | 38 | * \f] |
| madcowswe | 0:feb4117d16d8 | 39 | * \par Example: |
| madcowswe | 0:feb4117d16d8 | 40 | * \code |
| madcowswe | 0:feb4117d16d8 | 41 | * template<class T, std::size_t Rows1, std::size_t Cols1, std::size_t Cols2> |
| madcowswe | 0:feb4117d16d8 | 42 | * inline |
| madcowswe | 0:feb4117d16d8 | 43 | * void |
| madcowswe | 0:feb4117d16d8 | 44 | * prod(const Matrix<T, Rows1, Cols1>& lhs, const Matrix<T, Rows1, Cols2>& rhs, |
| madcowswe | 0:feb4117d16d8 | 45 | * Matrix<T, Cols2, Cols1>& dest) |
| madcowswe | 0:feb4117d16d8 | 46 | * { |
| madcowswe | 0:feb4117d16d8 | 47 | * for (std::size_t i = 0; i != Cols1; ++i) { |
| madcowswe | 0:feb4117d16d8 | 48 | * for (std::size_t j = 0; j != Cols2; ++j) { |
| madcowswe | 0:feb4117d16d8 | 49 | * dest(i, j) = tvmet::loop::gemtm<Rows1, Cols1, Cols2>::prod(lhs, rhs, i, j); |
| madcowswe | 0:feb4117d16d8 | 50 | * } |
| madcowswe | 0:feb4117d16d8 | 51 | * } |
| madcowswe | 0:feb4117d16d8 | 52 | * } |
| madcowswe | 0:feb4117d16d8 | 53 | * \endcode |
| madcowswe | 0:feb4117d16d8 | 54 | * \note The number of rows of rhs matrix have to be equal rows of rhs matrix, |
| madcowswe | 0:feb4117d16d8 | 55 | * since lhs matrix 1 is transposed. |
| madcowswe | 0:feb4117d16d8 | 56 | * The result is a (Cols1 x Cols2) matrix. |
| madcowswe | 0:feb4117d16d8 | 57 | */ |
| madcowswe | 0:feb4117d16d8 | 58 | template<std::size_t Rows1, std::size_t Cols1, |
| madcowswe | 0:feb4117d16d8 | 59 | std::size_t Cols2> |
| madcowswe | 0:feb4117d16d8 | 60 | class gemtm |
| madcowswe | 0:feb4117d16d8 | 61 | { |
| madcowswe | 0:feb4117d16d8 | 62 | gemtm(const gemtm&); |
| madcowswe | 0:feb4117d16d8 | 63 | gemtm& operator=(const gemtm&); |
| madcowswe | 0:feb4117d16d8 | 64 | |
| madcowswe | 0:feb4117d16d8 | 65 | private: |
| madcowswe | 0:feb4117d16d8 | 66 | enum { |
| madcowswe | 0:feb4117d16d8 | 67 | count = Cols1, |
| madcowswe | 0:feb4117d16d8 | 68 | N = (count+7)/8 |
| madcowswe | 0:feb4117d16d8 | 69 | }; |
| madcowswe | 0:feb4117d16d8 | 70 | |
| madcowswe | 0:feb4117d16d8 | 71 | public: |
| madcowswe | 0:feb4117d16d8 | 72 | gemtm() { } |
| madcowswe | 0:feb4117d16d8 | 73 | |
| madcowswe | 0:feb4117d16d8 | 74 | public: |
| madcowswe | 0:feb4117d16d8 | 75 | template<class E1, class E2> |
| madcowswe | 0:feb4117d16d8 | 76 | static inline |
| madcowswe | 0:feb4117d16d8 | 77 | typename PromoteTraits< |
| madcowswe | 0:feb4117d16d8 | 78 | typename E1::value_type, |
| madcowswe | 0:feb4117d16d8 | 79 | typename E2::value_type |
| madcowswe | 0:feb4117d16d8 | 80 | >::value_type |
| madcowswe | 0:feb4117d16d8 | 81 | prod(const E1& lhs, const E2& rhs, std::size_t i, std::size_t j) { |
| madcowswe | 0:feb4117d16d8 | 82 | typename PromoteTraits< |
| madcowswe | 0:feb4117d16d8 | 83 | typename E1::value_type, |
| madcowswe | 0:feb4117d16d8 | 84 | typename E2::value_type |
| madcowswe | 0:feb4117d16d8 | 85 | >::value_type sum(0); |
| madcowswe | 0:feb4117d16d8 | 86 | std::size_t k(0); |
| madcowswe | 0:feb4117d16d8 | 87 | std::size_t n(N); |
| madcowswe | 0:feb4117d16d8 | 88 | |
| madcowswe | 0:feb4117d16d8 | 89 | // Duff's device |
| madcowswe | 0:feb4117d16d8 | 90 | switch(count % 8) { |
| madcowswe | 0:feb4117d16d8 | 91 | case 0: do { sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 92 | case 7: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 93 | case 6: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 94 | case 5: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 95 | case 4: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 96 | case 3: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 97 | case 2: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 98 | case 1: sum += lhs(k, i) * rhs(k, j); ++k; |
| madcowswe | 0:feb4117d16d8 | 99 | } while(--n != 0); |
| madcowswe | 0:feb4117d16d8 | 100 | } |
| madcowswe | 0:feb4117d16d8 | 101 | |
| madcowswe | 0:feb4117d16d8 | 102 | return sum; |
| madcowswe | 0:feb4117d16d8 | 103 | } |
| madcowswe | 0:feb4117d16d8 | 104 | }; |
| madcowswe | 0:feb4117d16d8 | 105 | |
| madcowswe | 0:feb4117d16d8 | 106 | |
| madcowswe | 0:feb4117d16d8 | 107 | } // namespace loop |
| madcowswe | 0:feb4117d16d8 | 108 | |
| madcowswe | 0:feb4117d16d8 | 109 | } // namespace tvmet |
| madcowswe | 0:feb4117d16d8 | 110 | |
| madcowswe | 0:feb4117d16d8 | 111 | #endif /* TVMET_LOOP_GEMTM_H */ |
| madcowswe | 0:feb4117d16d8 | 112 | |
| madcowswe | 0:feb4117d16d8 | 113 | // Local Variables: |
| madcowswe | 0:feb4117d16d8 | 114 | // mode:C++ |
| madcowswe | 0:feb4117d16d8 | 115 | // tab-width:8 |
| madcowswe | 0:feb4117d16d8 | 116 | // End: |