Ryan Savitski / Mbed 2 deprecated ICRSEurobot14

2014 Eurobot fork

Dependencies:   mbed-rtos mbed QEI

tvmet/loop/Gemm.h@92:4a1225fbb146, 2013-10-15 (annotated)

Committer:
rsavitski
Date:
Tue Oct 15 12:19:32 2013 +0000
Revision:
92:4a1225fbb146
Parent:
15:9c5aaeda36dc 
touch: ripped out 2013-specific bits. Need to address "2014" comments. Rewrite AI layer and other deleted parts.

Who changed what in which revision?

UserRevisionLine numberNew contents of line
madcowswe 15:9c5aaeda36dc 1 /*
madcowswe 15:9c5aaeda36dc 2 * Tiny Vector Matrix Library
madcowswe 15:9c5aaeda36dc 3 * Dense Vector Matrix Libary of Tiny size using Expression Templates
madcowswe 15:9c5aaeda36dc 4 *
madcowswe 15:9c5aaeda36dc 5 * Copyright (C) 2001 - 2007 Olaf Petzold <opetzold@users.sourceforge.net>
madcowswe 15:9c5aaeda36dc 6 *
madcowswe 15:9c5aaeda36dc 7 * This library is free software; you can redistribute it and/or
madcowswe 15:9c5aaeda36dc 8 * modify it under the terms of the GNU Lesser General Public
madcowswe 15:9c5aaeda36dc 9 * License as published by the Free Software Foundation; either
madcowswe 15:9c5aaeda36dc 10 * version 2.1 of the License, or (at your option) any later version.
madcowswe 15:9c5aaeda36dc 11 *
madcowswe 15:9c5aaeda36dc 12 * This library is distributed in the hope that it will be useful,
madcowswe 15:9c5aaeda36dc 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
madcowswe 15:9c5aaeda36dc 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
madcowswe 15:9c5aaeda36dc 15 * Lesser General Public License for more details.
madcowswe 15:9c5aaeda36dc 16 *
madcowswe 15:9c5aaeda36dc 17 * You should have received a copy of the GNU Lesser General Public
madcowswe 15:9c5aaeda36dc 18 * License along with this library; if not, write to the Free Software
madcowswe 15:9c5aaeda36dc 19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
madcowswe 15:9c5aaeda36dc 20 *
madcowswe 15:9c5aaeda36dc 21 * $Id: Gemm.h,v 1.12 2007-06-23 15:58:59 opetzold Exp $
madcowswe 15:9c5aaeda36dc 22 */
madcowswe 15:9c5aaeda36dc 23
madcowswe 15:9c5aaeda36dc 24 #ifndef TVMET_LOOP_GEMM_H
madcowswe 15:9c5aaeda36dc 25 #define TVMET_LOOP_GEMM_H
madcowswe 15:9c5aaeda36dc 26
madcowswe 15:9c5aaeda36dc 27 namespace tvmet {
madcowswe 15:9c5aaeda36dc 28
madcowswe 15:9c5aaeda36dc 29 namespace loop {
madcowswe 15:9c5aaeda36dc 30
madcowswe 15:9c5aaeda36dc 31
madcowswe 15:9c5aaeda36dc 32 /**
madcowswe 15:9c5aaeda36dc 33 * \class gemm Gemm.h "tvmet/loop/Gemm.h"
madcowswe 15:9c5aaeda36dc 34 * \brief class for matrix-matrix product using loop unrolling.
madcowswe 15:9c5aaeda36dc 35 * using formula
madcowswe 15:9c5aaeda36dc 36 * \f[
madcowswe 15:9c5aaeda36dc 37 * M_1\,M_2
madcowswe 15:9c5aaeda36dc 38 * \f]
madcowswe 15:9c5aaeda36dc 39 * \par Example:
madcowswe 15:9c5aaeda36dc 40 * \code
madcowswe 15:9c5aaeda36dc 41 * template<class T, std::size_t Rows1, std::size_t Cols1, std::size_t Cols2>
madcowswe 15:9c5aaeda36dc 42 * inline
madcowswe 15:9c5aaeda36dc 43 * void
madcowswe 15:9c5aaeda36dc 44 * prod(const Matrix<T, Rows1, Cols1>& lhs, const Matrix<T, Cols1, Cols2>& rhs,
madcowswe 15:9c5aaeda36dc 45 * Matrix<T, Rows1, Cols2>& dest)
madcowswe 15:9c5aaeda36dc 46 * {
madcowswe 15:9c5aaeda36dc 47 * for (std::size_t i = 0; i != Rows1; ++i) {
madcowswe 15:9c5aaeda36dc 48 * for (std::size_t j = 0; j != Cols2; ++j) {
madcowswe 15:9c5aaeda36dc 49 * dest(i, j) = tvmet::loop::gemm<Rows1, Cols1, Cols2>().prod(lhs, rhs, i, j);
madcowswe 15:9c5aaeda36dc 50 * }
madcowswe 15:9c5aaeda36dc 51 * }
madcowswe 15:9c5aaeda36dc 52 * }
madcowswe 15:9c5aaeda36dc 53 * \endcode
madcowswe 15:9c5aaeda36dc 54 * \note The number of rows of rhs matrix have to be equal to cols of lhs matrix.
madcowswe 15:9c5aaeda36dc 55 * The result is a (Rows1 x Cols2) matrix.
madcowswe 15:9c5aaeda36dc 56 */
madcowswe 15:9c5aaeda36dc 57 template<std::size_t Rows1, std::size_t Cols1,
madcowswe 15:9c5aaeda36dc 58 std::size_t Cols2>
madcowswe 15:9c5aaeda36dc 59 class gemm
madcowswe 15:9c5aaeda36dc 60 {
madcowswe 15:9c5aaeda36dc 61 gemm(const gemm&);
madcowswe 15:9c5aaeda36dc 62 gemm& operator=(const gemm&);
madcowswe 15:9c5aaeda36dc 63
madcowswe 15:9c5aaeda36dc 64 private:
madcowswe 15:9c5aaeda36dc 65 enum {
madcowswe 15:9c5aaeda36dc 66 count = Cols1,
madcowswe 15:9c5aaeda36dc 67 N = (count+7)/8
madcowswe 15:9c5aaeda36dc 68 };
madcowswe 15:9c5aaeda36dc 69
madcowswe 15:9c5aaeda36dc 70 public:
madcowswe 15:9c5aaeda36dc 71 gemm() { }
madcowswe 15:9c5aaeda36dc 72
madcowswe 15:9c5aaeda36dc 73 public:
madcowswe 15:9c5aaeda36dc 74 template<class E1, class E2>
madcowswe 15:9c5aaeda36dc 75 static inline
madcowswe 15:9c5aaeda36dc 76 typename PromoteTraits<
madcowswe 15:9c5aaeda36dc 77 typename E1::value_type,
madcowswe 15:9c5aaeda36dc 78 typename E2::value_type
madcowswe 15:9c5aaeda36dc 79 >::value_type
madcowswe 15:9c5aaeda36dc 80 prod(const E1& lhs, const E2& rhs, std::size_t i, std::size_t j) {
madcowswe 15:9c5aaeda36dc 81 typename PromoteTraits<
madcowswe 15:9c5aaeda36dc 82 typename E1::value_type,
madcowswe 15:9c5aaeda36dc 83 typename E2::value_type
madcowswe 15:9c5aaeda36dc 84 >::value_type sum(0);
madcowswe 15:9c5aaeda36dc 85 std::size_t k(0);
madcowswe 15:9c5aaeda36dc 86 std::size_t n(N);
madcowswe 15:9c5aaeda36dc 87
madcowswe 15:9c5aaeda36dc 88 // Duff's device
madcowswe 15:9c5aaeda36dc 89 switch(count % 8) {
madcowswe 15:9c5aaeda36dc 90 case 0: do { sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 91 case 7: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 92 case 6: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 93 case 5: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 94 case 4: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 95 case 3: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 96 case 2: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 97 case 1: sum += lhs(i, k) * rhs(k, j); ++k;
madcowswe 15:9c5aaeda36dc 98 } while(--n != 0);
madcowswe 15:9c5aaeda36dc 99 }
madcowswe 15:9c5aaeda36dc 100
madcowswe 15:9c5aaeda36dc 101 return sum;
madcowswe 15:9c5aaeda36dc 102 }
madcowswe 15:9c5aaeda36dc 103 };
madcowswe 15:9c5aaeda36dc 104
madcowswe 15:9c5aaeda36dc 105
madcowswe 15:9c5aaeda36dc 106 } // namespace loop
madcowswe 15:9c5aaeda36dc 107
madcowswe 15:9c5aaeda36dc 108 } // namespace tvmet
madcowswe 15:9c5aaeda36dc 109
madcowswe 15:9c5aaeda36dc 110 #endif /* TVMET_LOOP_GEMM_H */
madcowswe 15:9c5aaeda36dc 111
madcowswe 15:9c5aaeda36dc 112 // Local Variables:
madcowswe 15:9c5aaeda36dc 113 // mode:C++
madcowswe 15:9c5aaeda36dc 114 // tab-width:8
madcowswe 15:9c5aaeda36dc 115 // End: