Home | History | Annotate | Download | only in parallel
      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/omp_loop_static.h
     26  *  @brief Parallelization of embarrassingly parallel execution by
     27  *  means of an OpenMP for loop with static scheduling.
     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_OMP_LOOP_STATIC_H
     34 #define _GLIBCXX_PARALLEL_OMP_LOOP_STATIC_H 1
     35 
     36 #include <omp.h>
     37 
     38 #include <parallel/settings.h>
     39 #include <parallel/basic_iterator.h>
     40 
     41 namespace __gnu_parallel
     42 {
     43 
     44   /** @brief Embarrassingly parallel algorithm for random access
     45    * iterators, using an OpenMP for loop with static scheduling.
     46    *
     47    *  @param begin Begin iterator of element sequence.
     48    *  @param end End iterator of element sequence.
     49    *  @param o User-supplied functor (comparator, predicate, adding
     50    *  functor, ...).
     51    *  @param f Functor to "process" an element with op (depends on
     52    *  desired functionality, e. g. for std::for_each(), ...).
     53    *  @param r Functor to "add" a single result to the already processed
     54    *  elements (depends on functionality).
     55    *  @param base Base value for reduction.
     56    *  @param output Pointer to position where final result is written to
     57    *  @param bound Maximum number of elements processed (e. g. for
     58    *  std::count_n()).
     59    *  @return User-supplied functor (that may contain a part of the result).
     60    */
     61 template<typename RandomAccessIterator,
     62 	 typename Op,
     63 	 typename Fu,
     64 	 typename Red,
     65 	 typename Result>
     66   Op
     67   for_each_template_random_access_omp_loop_static(RandomAccessIterator begin,
     68 						  RandomAccessIterator end,
     69 						  Op o, Fu& f, Red r,
     70 						  Result base, Result& output,
     71 						  typename std::iterator_traits
     72 						  <RandomAccessIterator>::
     73 						  difference_type bound)
     74   {
     75     typedef typename
     76       std::iterator_traits<RandomAccessIterator>::difference_type
     77       difference_type;
     78 
     79     difference_type length = end - begin;
     80     thread_index_t num_threads =
     81       std::min<difference_type>(get_max_threads(), length);
     82 
     83     Result *thread_results;
     84 
     85 #   pragma omp parallel num_threads(num_threads)
     86       {
     87 #       pragma omp single
     88           {
     89             num_threads = omp_get_num_threads();
     90             thread_results = new Result[num_threads];
     91 
     92             for (thread_index_t i = 0; i < num_threads; ++i)
     93               thread_results[i] = Result();
     94           }
     95 
     96         thread_index_t iam = omp_get_thread_num();
     97 
     98 #       pragma omp for schedule(static, _Settings::get().workstealing_chunk_size)
     99         for (difference_type pos = 0; pos < length; ++pos)
    100           thread_results[iam] = r(thread_results[iam], f(o, begin+pos));
    101       } //parallel
    102 
    103     for (thread_index_t i = 0; i < num_threads; ++i)
    104       output = r(output, thread_results[i]);
    105 
    106     delete [] thread_results;
    107 
    108     // Points to last element processed (needed as return value for
    109     // some algorithms like transform).
    110     f.finish_iterator = begin + length;
    111 
    112     return o;
    113   }
    114 
    115 } // end namespace
    116 
    117 #endif /* _GLIBCXX_PARALLEL_OMP_LOOP_STATIC_H */
    118