Oskar Weigl
/
Eurobot2013
We are going to win! wohoo
tvmet/loop/Gemmt.h@1:6799c07fe510, 2012-11-07 (annotated)
- Committer:
- sv
- Date:
- Wed Nov 07 14:37:35 2012 +0000
- Revision:
- 1:6799c07fe510
Preliminary copy of 2012 code
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
sv | 1:6799c07fe510 | 1 | /* |
sv | 1:6799c07fe510 | 2 | * Tiny Vector Matrix Library |
sv | 1:6799c07fe510 | 3 | * Dense Vector Matrix Libary of Tiny size using Expression Templates |
sv | 1:6799c07fe510 | 4 | * |
sv | 1:6799c07fe510 | 5 | * Copyright (C) 2001 - 2007 Olaf Petzold <opetzold@users.sourceforge.net> |
sv | 1:6799c07fe510 | 6 | * |
sv | 1:6799c07fe510 | 7 | * This library is free software; you can redistribute it and/or |
sv | 1:6799c07fe510 | 8 | * modify it under the terms of the GNU Lesser General Public |
sv | 1:6799c07fe510 | 9 | * License as published by the Free Software Foundation; either |
sv | 1:6799c07fe510 | 10 | * version 2.1 of the License, or (at your option) any later version. |
sv | 1:6799c07fe510 | 11 | * |
sv | 1:6799c07fe510 | 12 | * This library is distributed in the hope that it will be useful, |
sv | 1:6799c07fe510 | 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
sv | 1:6799c07fe510 | 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
sv | 1:6799c07fe510 | 15 | * Lesser General Public License for more details. |
sv | 1:6799c07fe510 | 16 | * |
sv | 1:6799c07fe510 | 17 | * You should have received a copy of the GNU Lesser General Public |
sv | 1:6799c07fe510 | 18 | * License along with this library; if not, write to the Free Software |
sv | 1:6799c07fe510 | 19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
sv | 1:6799c07fe510 | 20 | * |
sv | 1:6799c07fe510 | 21 | * $Id: Gemmt.h,v 1.9 2007-06-23 15:58:59 opetzold Exp $ |
sv | 1:6799c07fe510 | 22 | */ |
sv | 1:6799c07fe510 | 23 | |
sv | 1:6799c07fe510 | 24 | #ifndef TVMET_LOOP_GEMMT_H |
sv | 1:6799c07fe510 | 25 | #define TVMET_LOOP_GEMMT_H |
sv | 1:6799c07fe510 | 26 | |
sv | 1:6799c07fe510 | 27 | namespace tvmet { |
sv | 1:6799c07fe510 | 28 | |
sv | 1:6799c07fe510 | 29 | namespace loop { |
sv | 1:6799c07fe510 | 30 | |
sv | 1:6799c07fe510 | 31 | |
sv | 1:6799c07fe510 | 32 | /** |
sv | 1:6799c07fe510 | 33 | * \class gemmt Gemmt.h "tvmet/loop/Gemmt.h" |
sv | 1:6799c07fe510 | 34 | * \brief class for for product matrix-transpose(matrix) operations. |
sv | 1:6799c07fe510 | 35 | * using formula |
sv | 1:6799c07fe510 | 36 | * \f[ |
sv | 1:6799c07fe510 | 37 | * M_1\,M_2^{T} |
sv | 1:6799c07fe510 | 38 | * \f] |
sv | 1:6799c07fe510 | 39 | * \par Example: |
sv | 1:6799c07fe510 | 40 | * \code |
sv | 1:6799c07fe510 | 41 | * template<class T, std::size_t Rows1, std::size_t Cols1, std::size_t Cols2> |
sv | 1:6799c07fe510 | 42 | * inline |
sv | 1:6799c07fe510 | 43 | * void |
sv | 1:6799c07fe510 | 44 | * prod(const Matrix<T, Rows1, Cols1>& lhs, const Matrix<T, Rows2, Cols1>& rhs, |
sv | 1:6799c07fe510 | 45 | * Matrix<T, Rows1, Rows2>& dest) |
sv | 1:6799c07fe510 | 46 | * { |
sv | 1:6799c07fe510 | 47 | * for (std::size_t i = 0; i != Rows1; ++i) { |
sv | 1:6799c07fe510 | 48 | * for (std::size_t j = 0; j != Rows2; ++j) { |
sv | 1:6799c07fe510 | 49 | * dest(i, j) = tvmet::loop::gemmt<Rows1, Cols1, Cols1>().prod(lhs, rhs, i, j); |
sv | 1:6799c07fe510 | 50 | * } |
sv | 1:6799c07fe510 | 51 | * } |
sv | 1:6799c07fe510 | 52 | * } |
sv | 1:6799c07fe510 | 53 | * \endcode |
sv | 1:6799c07fe510 | 54 | * \note The number of cols of rhs matrix have to be equal to cols of rhs matrix. |
sv | 1:6799c07fe510 | 55 | * The result is a (Rows1 x Rows2) matrix. |
sv | 1:6799c07fe510 | 56 | */ |
sv | 1:6799c07fe510 | 57 | template<std::size_t Rows1, std::size_t Cols1, |
sv | 1:6799c07fe510 | 58 | std::size_t Cols2 /* unused */> |
sv | 1:6799c07fe510 | 59 | class gemmt |
sv | 1:6799c07fe510 | 60 | { |
sv | 1:6799c07fe510 | 61 | gemmt(const gemmt&); |
sv | 1:6799c07fe510 | 62 | gemmt& operator=(const gemmt&); |
sv | 1:6799c07fe510 | 63 | |
sv | 1:6799c07fe510 | 64 | private: |
sv | 1:6799c07fe510 | 65 | enum { |
sv | 1:6799c07fe510 | 66 | count = Cols1, |
sv | 1:6799c07fe510 | 67 | N = (count+7)/8 |
sv | 1:6799c07fe510 | 68 | }; |
sv | 1:6799c07fe510 | 69 | |
sv | 1:6799c07fe510 | 70 | public: |
sv | 1:6799c07fe510 | 71 | gemmt() { } |
sv | 1:6799c07fe510 | 72 | |
sv | 1:6799c07fe510 | 73 | public: |
sv | 1:6799c07fe510 | 74 | template<class E1, class E2> |
sv | 1:6799c07fe510 | 75 | static inline |
sv | 1:6799c07fe510 | 76 | typename PromoteTraits< |
sv | 1:6799c07fe510 | 77 | typename E1::value_type, |
sv | 1:6799c07fe510 | 78 | typename E2::value_type |
sv | 1:6799c07fe510 | 79 | >::value_type |
sv | 1:6799c07fe510 | 80 | prod(const E1& lhs, const E2& rhs, std::size_t i, std::size_t j) { |
sv | 1:6799c07fe510 | 81 | typename PromoteTraits< |
sv | 1:6799c07fe510 | 82 | typename E1::value_type, |
sv | 1:6799c07fe510 | 83 | typename E2::value_type |
sv | 1:6799c07fe510 | 84 | >::value_type sum(0); |
sv | 1:6799c07fe510 | 85 | std::size_t k(0); |
sv | 1:6799c07fe510 | 86 | std::size_t n(N); |
sv | 1:6799c07fe510 | 87 | |
sv | 1:6799c07fe510 | 88 | // Duff's device |
sv | 1:6799c07fe510 | 89 | switch(count % 8) { |
sv | 1:6799c07fe510 | 90 | case 0: do { sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 91 | case 7: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 92 | case 6: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 93 | case 5: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 94 | case 4: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 95 | case 3: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 96 | case 2: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 97 | case 1: sum += lhs(i, k) * rhs(j, k); ++k; |
sv | 1:6799c07fe510 | 98 | } while(--n != 0); |
sv | 1:6799c07fe510 | 99 | } |
sv | 1:6799c07fe510 | 100 | |
sv | 1:6799c07fe510 | 101 | return sum; |
sv | 1:6799c07fe510 | 102 | } |
sv | 1:6799c07fe510 | 103 | }; |
sv | 1:6799c07fe510 | 104 | |
sv | 1:6799c07fe510 | 105 | |
sv | 1:6799c07fe510 | 106 | } // namespace loop |
sv | 1:6799c07fe510 | 107 | |
sv | 1:6799c07fe510 | 108 | } // namespace tvmet |
sv | 1:6799c07fe510 | 109 | |
sv | 1:6799c07fe510 | 110 | #endif /* TVMET_LOOP_GEMMT_H */ |
sv | 1:6799c07fe510 | 111 | |
sv | 1:6799c07fe510 | 112 | // Local Variables: |
sv | 1:6799c07fe510 | 113 | // mode:C++ |
sv | 1:6799c07fe510 | 114 | // tab-width:8 |
sv | 1:6799c07fe510 | 115 | // End: |