libstdc++-v3/include/parallel/omp_loop_static.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/omp_loop_static.h
  32  *  @brief Parallelization of embarrassingly parallel execution by
  33  *  means of an OpenMP for loop with static scheduling.
  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_OMP_LOOP_STATIC_H
  40 #define _GLIBCXX_PARALLEL_OMP_LOOP_STATIC_H 1
  41
  42 #include <omp.h>
  43
  44 #include <parallel/settings.h>
  45 #include <parallel/basic_iterator.h>
  46
  47 namespace __gnu_parallel
  48 {
  49
  50   /** @brief Embarrassingly parallel algorithm for random access
  51    * iterators, using an OpenMP for loop with static scheduling.
  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 processed
  60    *  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<typename RandomAccessIterator, typename Op, typename Fu, typename Red, typename Result>
  68   Op
  69   for_each_template_random_access_omp_loop_static(RandomAccessIterator begin,
  70                                                   RandomAccessIterator end,
  71                                                   Op o, Fu& f, Red r,
  72                                                   Result base, Result& output,
  73                                                   typename std::iterator_traits<RandomAccessIterator>::difference_type bound)
  74   {
  75     typedef std::iterator_traits<RandomAccessIterator> traits_type;
  76     typedef typename traits_type::difference_type difference_type;
  77
  78     thread_index_t num_threads = (get_max_threads() < (end - begin)) ? get_max_threads() : (end - begin);
  79     Result *thread_results = new Result[num_threads];
  80     difference_type length = end - begin;
  81
  82     for (thread_index_t i = 0; i < num_threads; i++)
  83       {
  84         thread_results[i] = r(thread_results[i], f(o, begin+i));
  85       }
  86
  87 #pragma omp parallel num_threads(num_threads)
  88     {
  89 #pragma omp for schedule(static, Settings::workstealing_chunk_size)
  90       for (difference_type pos = 0; pos < length; pos++)
  91         {
  92           thread_results[omp_get_thread_num()] = r(thread_results[omp_get_thread_num()], f(o, begin+pos));
  93         }
  94     }
  95
  96     for (thread_index_t i = 0; i < num_threads; i++)
  97       {
  98         output = r(output, thread_results[i]);
  99       }
 100
 101     delete [] thread_results;
 102
 103     // Points to last element processed (needed as return value for
 104     // some algorithms like transform).
 105     f.finish_iterator = begin + length;
 106
 107     return o;
 108   }
 109 } // end namespace
 110
 111 #endif