libstdc++-v3/include/parallel/par_loop.h

   1 // -*- C++ -*-
   2
   3 // Copyright (C) 2007 Free Software Foundation, Inc.
   4 //
   5 // This file is part of the GNU ISO C++ Library.  This library is free
   6 // software; you can redistribute it and/or modify it under the terms
   7 // of the GNU General Public License as published by the Free Software
   8 // Foundation; either version 2, or (at your option) any later
   9 // version.
  10
  11 // This library is distributed in the hope that it will be useful, but
  12 // WITHOUT ANY WARRANTY; without even the implied warranty of
  13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 // General Public License for more details.
  15
  16 // You should have received a copy of the GNU General Public License
  17 // along with this library; see the file COPYING.  If not, write to
  18 // the Free Software Foundation, 59 Temple Place - Suite 330, Boston,
  19 // MA 02111-1307, USA.
  20
  21 // As a special exception, you may use this file as part of a free
  22 // software library without restriction.  Specifically, if other files
  23 // instantiate templates or use macros or inline functions from this
  24 // file, or you compile this file and link it with other files to
  25 // produce an executable, this file does not by itself cause the
  26 // resulting executable to be covered by the GNU General Public
  27 // License.  This exception does not however invalidate any other
  28 // reasons why the executable file might be covered by the GNU General
  29 // Public License.
  30
  31 /** @file parallel/par_loop.h
  32  *  @brief Parallelization of embarrassingly parallel execution by
  33  *  means of equal splitting.
  34  *  This file is a GNU parallel extension to the Standard C++ Library.
  35  */
  36
  37 // Written by Felix Putze.
  38
  39 #ifndef _GLIBCXX_PARALLEL_PAR_LOOP_H
  40 #define _GLIBCXX_PARALLEL_PAR_LOOP_H 1
  41
  42 #include <omp.h>
  43 #include <parallel/settings.h>
  44
  45 namespace __gnu_parallel
  46 {
  47
  48   /** @brief Embarrassingly parallel algorithm for random access
  49    * iterators, using hand-crafted parallelization by equal splitting
  50    * the work.
  51    *
  52    *  @param begin Begin iterator of element sequence.
  53    *  @param end End iterator of element sequence.
  54    *  @param o User-supplied functor (comparator, predicate, adding
  55    *  functor, ...)
  56    *  @param f Functor to "process" an element with op (depends on
  57    *  desired functionality, e. g. for std::for_each(), ...).
  58    *  @param r Functor to "add" a single result to the already
  59    *  processed elements (depends on functionality).
  60    *  @param base Base value for reduction.
  61    *  @param output Pointer to position where final result is written to
  62    *  @param bound Maximum number of elements processed (e. g. for
  63    *  std::count_n()).
  64    *  @return User-supplied functor (that may contain a part of the result).
  65    */
  66   template<typename RandomAccessIterator, typename Op, typename Fu, typename Red, typename Result>
  67   Op
  68   for_each_template_random_access_ed(RandomAccessIterator begin,
  69                                      RandomAccessIterator end, Op o, Fu& f,
  70                                      Red r, Result base, Result& output,
  71                                      typename std::iterator_traits<RandomAccessIterator>::difference_type bound)
  72   {
  73     typedef std::iterator_traits<RandomAccessIterator> traits_type;
  74     typedef typename traits_type::difference_type difference_type;
  75
  76     const difference_type length = end - begin;
  77     const difference_type settings_threads = static_cast<difference_type>(get_max_threads());
  78     const difference_type dmin = settings_threads < length ? settings_threads : length;
  79     const difference_type dmax = dmin > 1 ? dmin : 1;
  80
  81     thread_index_t num_threads = static_cast<thread_index_t>(dmax);
  82
  83
  84     Result *thread_results = new Result[num_threads];
  85
  86 #pragma omp parallel num_threads(num_threads)
  87     {
  88       // Neutral element.
  89       Result reduct = Result();
  90
  91       thread_index_t p = num_threads;
  92       thread_index_t iam = omp_get_thread_num();
  93       difference_type start = iam * length / p;
  94       difference_type limit = (iam == p - 1) ? length : (iam + 1) * length / p;
  95
  96       if (start < limit)
  97         {
  98           reduct = f(o, begin + start);
  99           start++;
 100         }
 101
 102       for (; start < limit; start++)
 103         reduct = r(reduct, f(o, begin + start));
 104
 105       thread_results[iam] = reduct;
 106     }
 107
 108     for (thread_index_t i = 0; i < num_threads; i++)
 109       output = r(output, thread_results[i]);
 110
 111     // Points to last element processed (needed as return value for
 112     // some algorithms like transform).
 113     f.finish_iterator = begin + length;
 114
 115     return o;
 116   }
 117
 118 } // end namespace
 119
 120 #endif