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

   1 // -*- C++ -*-
   2
   3 // Copyright (C) 2007, 2008, 2009 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 3, 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 // Under Section 7 of GPL version 3, you are granted additional
  17 // permissions described in the GCC Runtime Library Exception, version
  18 // 3.1, as published by the Free Software Foundation.
  19
  20 // You should have received a copy of the GNU General Public License and
  21 // a copy of the GCC Runtime Library Exception along with this program;
  22 // see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  23 // <http://www.gnu.org/licenses/>.
  24
  25 /** @file parallel/par_loop.h
  26  *  @brief Parallelization of embarrassingly parallel execution by
  27  *  means of equal splitting.
  28  *  This file is a GNU parallel extension to the Standard C++ Library.
  29  */
  30
  31 // Written by Felix Putze.
  32
  33 #ifndef _GLIBCXX_PARALLEL_PAR_LOOP_H
  34 #define _GLIBCXX_PARALLEL_PAR_LOOP_H 1
  35
  36 #include <omp.h>
  37 #include <parallel/settings.h>
  38 #include <parallel/base.h>
  39 #include <parallel/equally_split.h>
  40
  41 namespace __gnu_parallel
  42 {
  43
  44 /** @brief Embarrassingly parallel algorithm for random access
  45   * iterators, using hand-crafted parallelization by equal splitting
  46   * the work.
  47   *
  48   *  @param __begin Begin iterator of element __sequence.
  49   *  @param __end End iterator of element __sequence.
  50   *  @param __o User-supplied functor (comparator, predicate, adding
  51   *  functor, ...)
  52   *  @param __f Functor to "process" an element with __op (depends on
  53   *  desired functionality, e. g. for std::for_each(), ...).
  54   *  @param __r Functor to "add" a single __result to the already
  55   *  processed __elements (depends on functionality).
  56   *  @param __base Base value for reduction.
  57   *  @param __output Pointer to position where final result is written to
  58   *  @param __bound Maximum number of elements processed (e. g. for
  59   *  std::count_n()).
  60   *  @return User-supplied functor (that may contain a part of the result).
  61   */
  62 template<typename _RAIter,
  63          typename _Op,
  64          typename _Fu,
  65          typename _Red,
  66          typename _Result>
  67   _Op
  68   for_each_template_random_access_ed(
  69     _RAIter __begin, _RAIter __end, _Op __o, _Fu& __f, _Red __r,
  70     _Result __base, _Result& __output,
  71     typename std::iterator_traits<_RAIter>::difference_type __bound)
  72   {
  73     typedef std::iterator_traits<_RAIter> _TraitsType;
  74     typedef typename _TraitsType::difference_type _DifferenceType;
  75     const _DifferenceType __length = __end - __begin;
  76     _Result *__thread_results;
  77     bool* __constructed;
  78
  79     _ThreadIndex __num_threads =
  80       __gnu_parallel::min<_DifferenceType>(__get_max_threads(), __length);
  81
  82 #   pragma omp parallel num_threads(__num_threads)
  83       {
  84 #       pragma omp single
  85           {
  86             __num_threads = omp_get_num_threads();
  87             __thread_results =
  88               static_cast<_Result*>(
  89                             ::operator new(__num_threads * sizeof(_Result)));
  90             __constructed = new bool[__num_threads];
  91           }
  92
  93         _ThreadIndex __iam = omp_get_thread_num();
  94
  95         // Neutral element.
  96         _Result* __reduct =
  97                    static_cast<_Result*>(::operator new(sizeof(_Result)));
  98
  99         _DifferenceType
 100             __start = equally_split_point(__length, __num_threads, __iam),
 101             __stop = equally_split_point(__length, __num_threads, __iam + 1);
 102
 103         if (__start < __stop)
 104           {
 105             new(__reduct) _Result(__f(__o, __begin + __start));
 106             ++__start;
 107             __constructed[__iam] = true;
 108           }
 109         else
 110           __constructed[__iam] = false;
 111
 112         for (; __start < __stop; ++__start)
 113           *__reduct = __r(*__reduct, __f(__o, __begin + __start));
 114
 115         __thread_results[__iam] = *__reduct;
 116       } //parallel
 117
 118     for (_ThreadIndex __i = 0; __i < __num_threads; ++__i)
 119         if (__constructed[__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._M_finish_iterator = __begin + __length;
 125
 126     delete[] __thread_results;
 127     delete[] __constructed;
 128
 129     return __o;
 130   }
 131
 132 } // end namespace
 133
 134 #endif /* _GLIBCXX_PARALLEL_PAR_LOOP_H */