src/calf/benchmark.h

   1 /* Calf DSP Library
   2  * Reusable performance measurement classes.
   3  *
   4  * Copyright (C) 2007 Krzysztof Foltman
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General
  17  * Public License along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
  19  * Boston, MA  02110-1301  USA
  20  */
  21 #ifndef __CALF_BENCHMARK_H
  22 #define __CALF_BENCHMAR_H
  23
  24 #include <time.h>
  25 #include <sys/time.h>
  26 #include <sys/resource.h>
  27 #include "primitives.h"
  28 #include <algorithm>
  29 #include <typeinfo>
  30
  31 namespace dsp {
  32 #if 0
  33 }; to keep editor happy
  34 #endif
  35
  36
  37 class median_stat
  38 {
  39 public:
  40     double *data;
  41     unsigned int pos, count;
  42     bool sorted;
  43
  44     median_stat() {
  45         data = NULL;
  46         pos = 0;
  47         count = 0;
  48         sorted = false;
  49     }
  50     void start(int items) {
  51         if (data)
  52             delete []data;
  53         data = new double[items];
  54         pos = 0;
  55         count = items;
  56         sorted = false;
  57     }
  58     void add(double value)
  59     {
  60         assert(pos < count);
  61         data[pos++] = value;
  62     }
  63     void end()
  64     {
  65         std::sort(&data[0], &data[count]);
  66         sorted = true;
  67     }
  68     float get()
  69     {
  70         assert(sorted);
  71         return data[count >> 1];
  72     }
  73 };
  74
  75 // USE_RDTSC is for testing on my own machine, a crappy 1.6GHz Pentium 4 - it gives less headaches than clock() based measurements
  76 #define USE_RDTSC 0
  77 #define CLOCK_SPEED (1.6 * 1000.0 * 1000.0 * 1000.0)
  78
  79 #if USE_RDTSC
  80 inline uint64_t rdtsc()
  81 {
  82     unsigned long long int x;
  83     __asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
  84     return x;
  85 }
  86 #endif
  87
  88 struct benchmark_globals
  89 {
  90     static bool warned;
  91 };
  92
  93 template<typename Target, class Stat>
  94 class simple_benchmark: public benchmark_globals
  95 {
  96 public:
  97     Target target;
  98     Stat &stat;
  99
 100     simple_benchmark(const Target &_target, Stat &_stat)
 101     : target(_target)
 102     , stat(_stat)
 103     {
 104     }
 105
 106     void measure(int runs, int repeats)
 107     {
 108         int priority = getpriority(PRIO_PROCESS, getpid());
 109         stat.start(runs);
 110         if (setpriority(PRIO_PROCESS, getpid(), -20) < 0) {
 111             if (!warned) {
 112                 fprintf(stderr, "Warning: could not set process priority, measurements can be worthless\n");
 113                 warned = true;
 114             }
 115         }
 116         for (int i = 0; i < runs; i++) {
 117             target.prepare();
 118             // XXXKF measuring CPU time with clock() sucks,
 119 #if USE_RDTSC
 120             uint64_t start = rdtsc();
 121 #else
 122             clock_t start = ::clock();
 123 #endif
 124             for (int j = 0; j < repeats; j++) {
 125                 target.run();
 126             }
 127 #if USE_RDTSC
 128             uint64_t end = rdtsc();
 129             double elapsed = double(end - start) / (CLOCK_SPEED * repeats * target.scaler());
 130 #else
 131             clock_t end = ::clock();
 132             double elapsed = double(end - start) / (double(CLOCKS_PER_SEC) * repeats * target.scaler());
 133 #endif
 134             stat.add(elapsed);
 135             target.cleanup();
 136             // printf("elapsed = %f start = %d end = %d diff = %d\n", elapsed, start, end, end - start);
 137         }
 138         setpriority(PRIO_PROCESS, getpid(), priority);
 139         stat.end();
 140     }
 141
 142     float get_stat()
 143     {
 144         return stat.get();
 145     }
 146 };
 147
 148 template<class T>
 149 void do_simple_benchmark(int runs = 5, int repeats = 50000)
 150 {
 151     dsp::median_stat stat;
 152     dsp::simple_benchmark<T, dsp::median_stat> benchmark(T(), stat);
 153
 154     benchmark.measure(runs, repeats);
 155
 156     printf("%-30s: %f/sec, %f/CDsr, value = %f\n", typeid(T).name(), 1.0 / stat.get(), 1.0 / (44100 * stat.get()), benchmark.target.result);
 157 }
 158
 159
 160 #if 0
 161 { to keep editor happy
 162 #endif
 163 }
 164
 165 #endif