Shuto Naruse
/
Eurobot2012_Secondary
Eurobot2012_Secondary
Fork of Eurobot_2012_Secondary by
tvmet/loop/Gemv.h@1:cc2a9eb0bd55, 2012-10-17 (annotated)
- Committer:
- narshu
- Date:
- Wed Oct 17 22:25:31 2012 +0000
- Revision:
- 1:cc2a9eb0bd55
Commit before publishing
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
narshu | 1:cc2a9eb0bd55 | 1 | /* |
narshu | 1:cc2a9eb0bd55 | 2 | * Tiny Vector Matrix Library |
narshu | 1:cc2a9eb0bd55 | 3 | * Dense Vector Matrix Libary of Tiny size using Expression Templates |
narshu | 1:cc2a9eb0bd55 | 4 | * |
narshu | 1:cc2a9eb0bd55 | 5 | * Copyright (C) 2001 - 2007 Olaf Petzold <opetzold@users.sourceforge.net> |
narshu | 1:cc2a9eb0bd55 | 6 | * |
narshu | 1:cc2a9eb0bd55 | 7 | * This library is free software; you can redistribute it and/or |
narshu | 1:cc2a9eb0bd55 | 8 | * modify it under the terms of the GNU Lesser General Public |
narshu | 1:cc2a9eb0bd55 | 9 | * License as published by the Free Software Foundation; either |
narshu | 1:cc2a9eb0bd55 | 10 | * version 2.1 of the License, or (at your option) any later version. |
narshu | 1:cc2a9eb0bd55 | 11 | * |
narshu | 1:cc2a9eb0bd55 | 12 | * This library is distributed in the hope that it will be useful, |
narshu | 1:cc2a9eb0bd55 | 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
narshu | 1:cc2a9eb0bd55 | 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
narshu | 1:cc2a9eb0bd55 | 15 | * Lesser General Public License for more details. |
narshu | 1:cc2a9eb0bd55 | 16 | * |
narshu | 1:cc2a9eb0bd55 | 17 | * You should have received a copy of the GNU Lesser General Public |
narshu | 1:cc2a9eb0bd55 | 18 | * License along with this library; if not, write to the Free Software |
narshu | 1:cc2a9eb0bd55 | 19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
narshu | 1:cc2a9eb0bd55 | 20 | * |
narshu | 1:cc2a9eb0bd55 | 21 | * $Id: Gemv.h,v 1.7 2007-06-23 15:58:59 opetzold Exp $ |
narshu | 1:cc2a9eb0bd55 | 22 | */ |
narshu | 1:cc2a9eb0bd55 | 23 | |
narshu | 1:cc2a9eb0bd55 | 24 | #ifndef TVMET_LOOP_GEMV_H |
narshu | 1:cc2a9eb0bd55 | 25 | #define TVMET_LOOP_GEMV_H |
narshu | 1:cc2a9eb0bd55 | 26 | |
narshu | 1:cc2a9eb0bd55 | 27 | namespace tvmet { |
narshu | 1:cc2a9eb0bd55 | 28 | |
narshu | 1:cc2a9eb0bd55 | 29 | namespace loop { |
narshu | 1:cc2a9eb0bd55 | 30 | |
narshu | 1:cc2a9eb0bd55 | 31 | |
narshu | 1:cc2a9eb0bd55 | 32 | /** |
narshu | 1:cc2a9eb0bd55 | 33 | * \class gemv Gemv.h "tvmet/loop/Gemv.h" |
narshu | 1:cc2a9eb0bd55 | 34 | * \brief class for matrix-vector product using loop unrolling. |
narshu | 1:cc2a9eb0bd55 | 35 | * using formula |
narshu | 1:cc2a9eb0bd55 | 36 | * \f[ |
narshu | 1:cc2a9eb0bd55 | 37 | * M\,v |
narshu | 1:cc2a9eb0bd55 | 38 | * \f] |
narshu | 1:cc2a9eb0bd55 | 39 | * \par Example: |
narshu | 1:cc2a9eb0bd55 | 40 | * \code |
narshu | 1:cc2a9eb0bd55 | 41 | * template<class T, std::size_t Rows, std::size_t Cols> |
narshu | 1:cc2a9eb0bd55 | 42 | * inline |
narshu | 1:cc2a9eb0bd55 | 43 | * void |
narshu | 1:cc2a9eb0bd55 | 44 | * prod(const Matrix<T, Rows, Cols>& lhs, const Vector<T, Cols>& rhs, |
narshu | 1:cc2a9eb0bd55 | 45 | * Vector<T, Rows>& dest) |
narshu | 1:cc2a9eb0bd55 | 46 | * { |
narshu | 1:cc2a9eb0bd55 | 47 | * for (std::size_t i = 0; i != Rows; ++i) { |
narshu | 1:cc2a9eb0bd55 | 48 | * dest(i) = tvmet::loop::gemv<Rows, Cols>().prod(lhs, rhs, i); |
narshu | 1:cc2a9eb0bd55 | 49 | * } |
narshu | 1:cc2a9eb0bd55 | 50 | * } |
narshu | 1:cc2a9eb0bd55 | 51 | * \endcode |
narshu | 1:cc2a9eb0bd55 | 52 | */ |
narshu | 1:cc2a9eb0bd55 | 53 | template<std::size_t Rows, std::size_t Cols> |
narshu | 1:cc2a9eb0bd55 | 54 | class gemv |
narshu | 1:cc2a9eb0bd55 | 55 | { |
narshu | 1:cc2a9eb0bd55 | 56 | gemv(const gemv&); |
narshu | 1:cc2a9eb0bd55 | 57 | gemv& operator=(const gemv&); |
narshu | 1:cc2a9eb0bd55 | 58 | |
narshu | 1:cc2a9eb0bd55 | 59 | private: |
narshu | 1:cc2a9eb0bd55 | 60 | enum { |
narshu | 1:cc2a9eb0bd55 | 61 | count = Cols, |
narshu | 1:cc2a9eb0bd55 | 62 | N = (count+7)/8 |
narshu | 1:cc2a9eb0bd55 | 63 | }; |
narshu | 1:cc2a9eb0bd55 | 64 | |
narshu | 1:cc2a9eb0bd55 | 65 | public: |
narshu | 1:cc2a9eb0bd55 | 66 | gemv() { } |
narshu | 1:cc2a9eb0bd55 | 67 | |
narshu | 1:cc2a9eb0bd55 | 68 | public: |
narshu | 1:cc2a9eb0bd55 | 69 | template<class E1, class E2> |
narshu | 1:cc2a9eb0bd55 | 70 | static inline |
narshu | 1:cc2a9eb0bd55 | 71 | typename PromoteTraits< |
narshu | 1:cc2a9eb0bd55 | 72 | typename E1::value_type, |
narshu | 1:cc2a9eb0bd55 | 73 | typename E2::value_type |
narshu | 1:cc2a9eb0bd55 | 74 | >::value_type |
narshu | 1:cc2a9eb0bd55 | 75 | prod(const E1& lhs, const E2& rhs, std::size_t i) { |
narshu | 1:cc2a9eb0bd55 | 76 | typename PromoteTraits< |
narshu | 1:cc2a9eb0bd55 | 77 | typename E1::value_type, |
narshu | 1:cc2a9eb0bd55 | 78 | typename E2::value_type |
narshu | 1:cc2a9eb0bd55 | 79 | >::value_type sum(0); |
narshu | 1:cc2a9eb0bd55 | 80 | std::size_t j(0); |
narshu | 1:cc2a9eb0bd55 | 81 | std::size_t n(N); |
narshu | 1:cc2a9eb0bd55 | 82 | |
narshu | 1:cc2a9eb0bd55 | 83 | // Duff's device |
narshu | 1:cc2a9eb0bd55 | 84 | switch(count % 8) { |
narshu | 1:cc2a9eb0bd55 | 85 | case 0: do { sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 86 | case 7: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 87 | case 6: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 88 | case 5: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 89 | case 4: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 90 | case 3: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 91 | case 2: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 92 | case 1: sum += lhs(i, j) * rhs(j); ++j; |
narshu | 1:cc2a9eb0bd55 | 93 | } while(--n != 0); |
narshu | 1:cc2a9eb0bd55 | 94 | } |
narshu | 1:cc2a9eb0bd55 | 95 | |
narshu | 1:cc2a9eb0bd55 | 96 | return sum; |
narshu | 1:cc2a9eb0bd55 | 97 | } |
narshu | 1:cc2a9eb0bd55 | 98 | }; |
narshu | 1:cc2a9eb0bd55 | 99 | |
narshu | 1:cc2a9eb0bd55 | 100 | |
narshu | 1:cc2a9eb0bd55 | 101 | } // namespace loop |
narshu | 1:cc2a9eb0bd55 | 102 | |
narshu | 1:cc2a9eb0bd55 | 103 | } // namespace tvmet |
narshu | 1:cc2a9eb0bd55 | 104 | |
narshu | 1:cc2a9eb0bd55 | 105 | #endif /* TVMET_LOOP_GEMV_H */ |
narshu | 1:cc2a9eb0bd55 | 106 | |
narshu | 1:cc2a9eb0bd55 | 107 | // Local Variables: |
narshu | 1:cc2a9eb0bd55 | 108 | // mode:C++ |
narshu | 1:cc2a9eb0bd55 | 109 | // tab-width:8 |
narshu | 1:cc2a9eb0bd55 | 110 | // End: |