libstdc++
omp_loop.h
Go to the documentation of this file.
00001 // -*- C++ -*-
00002 
00003 // Copyright (C) 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
00004 //
00005 // This file is part of the GNU ISO C++ Library.  This library is free
00006 // software; you can redistribute it and/or modify it under the terms
00007 // of the GNU General Public License as published by the Free Software
00008 // Foundation; either version 3, or (at your option) any later
00009 // version.
00010 
00011 // This library is distributed in the hope that it will be useful, but
00012 // WITHOUT ANY WARRANTY; without even the implied warranty of
00013 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014 // General Public License for more details.
00015 
00016 // Under Section 7 of GPL version 3, you are granted additional
00017 // permissions described in the GCC Runtime Library Exception, version
00018 // 3.1, as published by the Free Software Foundation.
00019 
00020 // You should have received a copy of the GNU General Public License and
00021 // a copy of the GCC Runtime Library Exception along with this program;
00022 // see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
00023 // <http://www.gnu.org/licenses/>.
00024 
00025 /** @file parallel/omp_loop.h
00026  *  @brief Parallelization of embarrassingly parallel execution by
00027  *  means of an OpenMP for loop.
00028  *  This file is a GNU parallel extension to the Standard C++ Library.
00029  */
00030 
00031 // Written by Felix Putze.
00032 
00033 #ifndef _GLIBCXX_PARALLEL_OMP_LOOP_H
00034 #define _GLIBCXX_PARALLEL_OMP_LOOP_H 1
00035 
00036 #include <omp.h>
00037 
00038 #include <parallel/settings.h>
00039 #include <parallel/basic_iterator.h>
00040 #include <parallel/base.h>
00041 
00042 namespace __gnu_parallel
00043 {
00044   /** @brief Embarrassingly parallel algorithm for random access
00045    * iterators, using an OpenMP for loop.
00046    *
00047    *  @param __begin Begin iterator of element sequence.
00048    *  @param __end End iterator of element sequence.
00049    *  @param __o User-supplied functor (comparator, predicate, adding
00050    *  functor, etc.).
00051    *  @param __f Functor to @a process an element with __op (depends on
00052    *  desired functionality, e. g. for std::for_each(), ...).
00053    *  @param __r Functor to @a add a single __result to the already
00054    *  processed elements (depends on functionality).
00055    *  @param __base Base value for reduction.
00056    *  @param __output Pointer to position where final result is written to
00057    *  @param __bound Maximum number of elements processed (e. g. for
00058    *  std::count_n()).
00059    *  @return User-supplied functor (that may contain a part of the result).
00060    */
00061   template<typename _RAIter,
00062        typename _Op,
00063        typename _Fu,
00064        typename _Red,
00065        typename _Result>
00066     _Op
00067     __for_each_template_random_access_omp_loop(_RAIter __begin, _RAIter __end,
00068                            _Op __o, _Fu& __f, _Red __r,
00069                            _Result __base,
00070                            _Result& __output,
00071       typename std::iterator_traits<_RAIter>::difference_type __bound)
00072     {
00073       typedef typename std::iterator_traits<_RAIter>::difference_type
00074         _DifferenceType;
00075 
00076       _DifferenceType __length = __end - __begin;
00077       _ThreadIndex __num_threads = __gnu_parallel::min<_DifferenceType>
00078     (__get_max_threads(), __length);
00079 
00080       _Result *__thread_results;
00081 
00082 #     pragma omp parallel num_threads(__num_threads)
00083       {
00084 #       pragma omp single
00085     {
00086       __num_threads = omp_get_num_threads();
00087       __thread_results = new _Result[__num_threads];
00088 
00089       for (_ThreadIndex __i = 0; __i < __num_threads; ++__i)
00090         __thread_results[__i] = _Result();
00091     }
00092 
00093         _ThreadIndex __iam = omp_get_thread_num();
00094 
00095 #pragma omp for schedule(dynamic, _Settings::get().workstealing_chunk_size)
00096         for (_DifferenceType __pos = 0; __pos < __length; ++__pos)
00097           __thread_results[__iam] = __r(__thread_results[__iam],
00098                     __f(__o, __begin+__pos));
00099       } //parallel
00100 
00101       for (_ThreadIndex __i = 0; __i < __num_threads; ++__i)
00102         __output = __r(__output, __thread_results[__i]);
00103 
00104       delete [] __thread_results;
00105 
00106       // Points to last element processed (needed as return value for
00107       // some algorithms like transform).
00108       __f._M_finish_iterator = __begin + __length;
00109 
00110       return __o;
00111     }
00112 
00113 } // end namespace
00114 
00115 #endif /* _GLIBCXX_PARALLEL_OMP_LOOP_H */