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 #include <parallel/base.h>
  45
  46 namespace __gnu_parallel
  47 {
  48
  49 /** @brief Embarrassingly parallel algorithm for random access
  50   * iterators, using hand-crafted parallelization by equal splitting
  51   * the work.
  52   *
  53   *  @param begin Begin iterator of element sequence.
  54   *  @param end End iterator of element sequence.
  55   *  @param o User-supplied functor (comparator, predicate, adding
  56   *  functor, ...)
  57   *  @param f Functor to "process" an element with op (depends on
  58   *  desired functionality, e. g. for std::for_each(), ...).
  59   *  @param r Functor to "add" a single result to the already
  60   *  processed elements (depends on functionality).
  61   *  @param base Base value for reduction.
  62   *  @param output Pointer to position where final result is written to
  63   *  @param bound Maximum number of elements processed (e. g. for
  64   *  std::count_n()).
  65   *  @return User-supplied functor (that may contain a part of the result).
  66   */
  67 template<
  68     typename RandomAccessIterator,
  69     typename Op,
  70     typename Fu,
  71     typename Red,
  72     typename Result>
  73   Op
  74   for_each_template_random_access_ed(
  75               RandomAccessIterator begin,
  76               RandomAccessIterator end,
  77               Op o, Fu& f, Red r, Result base, Result& output,
  78               typename std::iterator_traits<RandomAccessIterator>::
  79                   difference_type bound)
  80   {
  81     typedef std::iterator_traits<RandomAccessIterator> traits_type;
  82     typedef typename traits_type::difference_type difference_type;
  83
  84     const difference_type length = end - begin;
  85     Result *thread_results;
  86
  87     thread_index_t num_threads =
  88         __gnu_parallel::min<difference_type>(get_max_threads(), length);
  89
  90 #   pragma omp parallel num_threads(num_threads)
  91       {
  92 #       pragma omp single
  93           {
  94             num_threads = omp_get_num_threads();
  95             thread_results = new Result[num_threads];
  96           }
  97
  98         thread_index_t iam = omp_get_thread_num();
  99
 100         // Neutral element.
 101         Result reduct = Result();
 102
 103         difference_type
 104             start = equally_split_point(length, num_threads, iam),
 105             stop = equally_split_point(length, num_threads, iam + 1);
 106
 107         if (start < stop)
 108           {
 109             reduct = f(o, begin + start);
 110             ++start;
 111           }
 112
 113         for (; start < stop; ++start)
 114           reduct = r(reduct, f(o, begin + start));
 115
 116         thread_results[iam] = reduct;
 117       } //parallel
 118
 119     for (thread_index_t i = 0; i < num_threads; i++)
 120       output = r(output, thread_results[i]);
 121
 122     // Points to last element processed (needed as return value for
 123     // some algorithms like transform).
 124     f.finish_iterator = begin + length;
 125
 126     return o;
 127   }
 128
 129 } // end namespace
 130
 131 #endif