[136] | 1 | //=====================================================
|
---|
| 2 | // File : bench.hh
|
---|
| 3 | // Author : L. Plagne <laurent.plagne@edf.fr)>
|
---|
| 4 | // Copyright (C) EDF R&D, lun sep 30 14:23:16 CEST 2002
|
---|
| 5 | //=====================================================
|
---|
| 6 | //
|
---|
| 7 | // This program is free software; you can redistribute it and/or
|
---|
| 8 | // modify it under the terms of the GNU General Public License
|
---|
| 9 | // as published by the Free Software Foundation; either version 2
|
---|
| 10 | // of the License, or (at your option) any later version.
|
---|
| 11 | //
|
---|
| 12 | // This program is distributed in the hope that it will be useful,
|
---|
| 13 | // but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 14 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 15 | // GNU General Public License for more details.
|
---|
| 16 | // You should have received a copy of the GNU General Public License
|
---|
| 17 | // along with this program; if not, write to the Free Software
|
---|
| 18 | // Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
|
---|
| 19 | //
|
---|
| 20 | #ifndef BENCH_HH
|
---|
| 21 | #define BENCH_HH
|
---|
| 22 |
|
---|
| 23 | #include "btl.hh"
|
---|
| 24 | #include "bench_parameter.hh"
|
---|
| 25 | #include <iostream>
|
---|
| 26 | #include "utilities.h"
|
---|
| 27 | #include "size_lin_log.hh"
|
---|
| 28 | #include "xy_file.hh"
|
---|
| 29 | #include <vector>
|
---|
| 30 | #include <string>
|
---|
| 31 | #include "timers/portable_perf_analyzer.hh"
|
---|
| 32 | // #include "timers/mixed_perf_analyzer.hh"
|
---|
| 33 | // #include "timers/x86_perf_analyzer.hh"
|
---|
| 34 | // #include "timers/STL_perf_analyzer.hh"
|
---|
| 35 | #ifdef HAVE_MKL
|
---|
| 36 | extern "C" void cblas_saxpy(const int, const float, const float*, const int, float *, const int);
|
---|
| 37 | #endif
|
---|
| 38 | using namespace std;
|
---|
| 39 |
|
---|
| 40 | template <template<class> class Perf_Analyzer, class Action>
|
---|
| 41 | BTL_DONT_INLINE void bench( int size_min, int size_max, int nb_point )
|
---|
| 42 | {
|
---|
| 43 | if (BtlConfig::skipAction(Action::name()))
|
---|
| 44 | return;
|
---|
| 45 |
|
---|
| 46 | string filename="bench_"+Action::name()+".dat";
|
---|
| 47 |
|
---|
| 48 | INFOS("starting " <<filename);
|
---|
| 49 |
|
---|
| 50 | // utilities
|
---|
| 51 |
|
---|
| 52 | std::vector<double> tab_mflops(nb_point);
|
---|
| 53 | std::vector<int> tab_sizes(nb_point);
|
---|
| 54 |
|
---|
| 55 | // matrices and vector size calculations
|
---|
| 56 | size_lin_log(nb_point,size_min,size_max,tab_sizes);
|
---|
| 57 |
|
---|
| 58 | std::vector<int> oldSizes;
|
---|
| 59 | std::vector<double> oldFlops;
|
---|
| 60 | bool hasOldResults = read_xy_file(filename, oldSizes, oldFlops, true);
|
---|
| 61 | int oldi = oldSizes.size() - 1;
|
---|
| 62 |
|
---|
| 63 | // loop on matrix size
|
---|
| 64 | Perf_Analyzer<Action> perf_action;
|
---|
| 65 | for (int i=nb_point-1;i>=0;i--)
|
---|
| 66 | {
|
---|
| 67 | //INFOS("size=" <<tab_sizes[i]<<" ("<<nb_point-i<<"/"<<nb_point<<")");
|
---|
| 68 | std::cout << " " << "size = " << tab_sizes[i] << " " << std::flush;
|
---|
| 69 |
|
---|
| 70 | BTL_DISABLE_SSE_EXCEPTIONS();
|
---|
| 71 | #ifdef HAVE_MKL
|
---|
| 72 | {
|
---|
| 73 | float dummy;
|
---|
| 74 | cblas_saxpy(1,0,&dummy,1,&dummy,1);
|
---|
| 75 | }
|
---|
| 76 | #endif
|
---|
| 77 |
|
---|
| 78 | tab_mflops[i] = perf_action.eval_mflops(tab_sizes[i]);
|
---|
| 79 | std::cout << tab_mflops[i];
|
---|
| 80 |
|
---|
| 81 | if (hasOldResults)
|
---|
| 82 | {
|
---|
| 83 | while (oldi>=0 && oldSizes[oldi]>tab_sizes[i])
|
---|
| 84 | --oldi;
|
---|
| 85 | if (oldi>=0 && oldSizes[oldi]==tab_sizes[i])
|
---|
| 86 | {
|
---|
| 87 | if (oldFlops[oldi]<tab_mflops[i])
|
---|
| 88 | std::cout << "\t > ";
|
---|
| 89 | else
|
---|
| 90 | std::cout << "\t < ";
|
---|
| 91 | std::cout << oldFlops[oldi];
|
---|
| 92 | }
|
---|
| 93 | --oldi;
|
---|
| 94 | }
|
---|
| 95 | std::cout << " MFlops (" << nb_point-i << "/" << nb_point << ")" << std::endl;
|
---|
| 96 | }
|
---|
| 97 |
|
---|
| 98 | if (!BtlConfig::Instance.overwriteResults)
|
---|
| 99 | {
|
---|
| 100 | if (hasOldResults)
|
---|
| 101 | {
|
---|
| 102 | // merge the two data
|
---|
| 103 | std::vector<int> newSizes;
|
---|
| 104 | std::vector<double> newFlops;
|
---|
| 105 | int i=0;
|
---|
| 106 | int j=0;
|
---|
| 107 | while (i<tab_sizes.size() && j<oldSizes.size())
|
---|
| 108 | {
|
---|
| 109 | if (tab_sizes[i] == oldSizes[j])
|
---|
| 110 | {
|
---|
| 111 | newSizes.push_back(tab_sizes[i]);
|
---|
| 112 | newFlops.push_back(std::max(tab_mflops[i], oldFlops[j]));
|
---|
| 113 | ++i;
|
---|
| 114 | ++j;
|
---|
| 115 | }
|
---|
| 116 | else if (tab_sizes[i] < oldSizes[j])
|
---|
| 117 | {
|
---|
| 118 | newSizes.push_back(tab_sizes[i]);
|
---|
| 119 | newFlops.push_back(tab_mflops[i]);
|
---|
| 120 | ++i;
|
---|
| 121 | }
|
---|
| 122 | else
|
---|
| 123 | {
|
---|
| 124 | newSizes.push_back(oldSizes[j]);
|
---|
| 125 | newFlops.push_back(oldFlops[j]);
|
---|
| 126 | ++j;
|
---|
| 127 | }
|
---|
| 128 | }
|
---|
| 129 | while (i<tab_sizes.size())
|
---|
| 130 | {
|
---|
| 131 | newSizes.push_back(tab_sizes[i]);
|
---|
| 132 | newFlops.push_back(tab_mflops[i]);
|
---|
| 133 | ++i;
|
---|
| 134 | }
|
---|
| 135 | while (j<oldSizes.size())
|
---|
| 136 | {
|
---|
| 137 | newSizes.push_back(oldSizes[j]);
|
---|
| 138 | newFlops.push_back(oldFlops[j]);
|
---|
| 139 | ++j;
|
---|
| 140 | }
|
---|
| 141 | tab_mflops = newFlops;
|
---|
| 142 | tab_sizes = newSizes;
|
---|
| 143 | }
|
---|
| 144 | }
|
---|
| 145 |
|
---|
| 146 | // dump the result in a file :
|
---|
| 147 | dump_xy_file(tab_sizes,tab_mflops,filename);
|
---|
| 148 |
|
---|
| 149 | }
|
---|
| 150 |
|
---|
| 151 | // default Perf Analyzer
|
---|
| 152 |
|
---|
| 153 | template <class Action>
|
---|
| 154 | BTL_DONT_INLINE void bench( int size_min, int size_max, int nb_point ){
|
---|
| 155 |
|
---|
| 156 | // if the rdtsc is not available :
|
---|
| 157 | bench<Portable_Perf_Analyzer,Action>(size_min,size_max,nb_point);
|
---|
| 158 | // if the rdtsc is available :
|
---|
| 159 | // bench<Mixed_Perf_Analyzer,Action>(size_min,size_max,nb_point);
|
---|
| 160 |
|
---|
| 161 |
|
---|
| 162 | // Only for small problem size. Otherwize it will be too long
|
---|
| 163 | // bench<X86_Perf_Analyzer,Action>(size_min,size_max,nb_point);
|
---|
| 164 | // bench<STL_Perf_Analyzer,Action>(size_min,size_max,nb_point);
|
---|
| 165 |
|
---|
| 166 | }
|
---|
| 167 |
|
---|
| 168 | #endif
|
---|