l4/pkg/l4re-core/libstdc++-v3/contrib/libstdc++-v3-4.4/include/parallel/partial_sum.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/partial_sum.h
  26  *  @brief Parallel implementation of std::partial_sum(), i. e. prefix
  27  *  sums.
  28  *  This file is a GNU parallel extension to the Standard C++ Library.
  29  */
  30
  31 // Written by Johannes Singler.
  32
  33 #ifndef _GLIBCXX_PARALLEL_PARTIAL_SUM_H
  34 #define _GLIBCXX_PARALLEL_PARTIAL_SUM_H 1
  35
  36 #include <omp.h>
  37 #include <new>
  38 #include <bits/stl_algobase.h>
  39 #include <parallel/parallel.h>
  40 #include <parallel/numericfwd.h>
  41
  42 namespace __gnu_parallel
  43 {
  44   // Problem: there is no 0-element given.
  45
  46 /** @brief Base case prefix sum routine.
  47   *  @param begin Begin iterator of input sequence.
  48   *  @param end End iterator of input sequence.
  49   *  @param result Begin iterator of output sequence.
  50   *  @param bin_op Associative binary function.
  51   *  @param value Start value. Must be passed since the neutral
  52   *  element is unknown in general.
  53   *  @return End iterator of output sequence. */
  54 template<typename InputIterator,
  55          typename OutputIterator,
  56          typename BinaryOperation>
  57   OutputIterator
  58   parallel_partial_sum_basecase(InputIterator begin, InputIterator end,
  59                                 OutputIterator result, BinaryOperation bin_op,
  60                                 typename std::iterator_traits
  61                                 <InputIterator>::value_type value)
  62   {
  63     if (begin == end)
  64       return result;
  65
  66     while (begin != end)
  67       {
  68         value = bin_op(value, *begin);
  69         *result = value;
  70         ++result;
  71         ++begin;
  72       }
  73     return result;
  74   }
  75
  76 /** @brief Parallel partial sum implementation, two-phase approach,
  77     no recursion.
  78     *  @param begin Begin iterator of input sequence.
  79     *  @param end End iterator of input sequence.
  80     *  @param result Begin iterator of output sequence.
  81     *  @param bin_op Associative binary function.
  82     *  @param n Length of sequence.
  83     *  @param num_threads Number of threads to use.
  84     *  @return End iterator of output sequence.
  85     */
  86 template<typename InputIterator,
  87          typename OutputIterator,
  88          typename BinaryOperation>
  89   OutputIterator
  90   parallel_partial_sum_linear(InputIterator begin, InputIterator end,
  91                               OutputIterator result, BinaryOperation bin_op,
  92                               typename std::iterator_traits
  93                               <InputIterator>::difference_type n)
  94   {
  95     typedef std::iterator_traits<InputIterator> traits_type;
  96     typedef typename traits_type::value_type value_type;
  97     typedef typename traits_type::difference_type difference_type;
  98
  99     if (begin == end)
 100       return result;
 101
 102     thread_index_t num_threads =
 103         std::min<difference_type>(get_max_threads(), n - 1);
 104
 105     if (num_threads < 2)
 106       {
 107         *result = *begin;
 108         return parallel_partial_sum_basecase(
 109             begin + 1, end, result + 1, bin_op, *begin);
 110       }
 111
 112     difference_type* borders;
 113     value_type* sums;
 114
 115     const _Settings& __s = _Settings::get();
 116
 117 #   pragma omp parallel num_threads(num_threads)
 118       {
 119 #       pragma omp single
 120           {
 121             num_threads = omp_get_num_threads();
 122
 123             borders = new difference_type[num_threads + 2];
 124
 125             if (__s.partial_sum_dilation == 1.0f)
 126               equally_split(n, num_threads + 1, borders);
 127             else
 128               {
 129                 difference_type chunk_length =
 130                     ((double)n
 131                      / ((double)num_threads + __s.partial_sum_dilation)),
 132                   borderstart = n - num_threads * chunk_length;
 133                 borders[0] = 0;
 134                 for (int i = 1; i < (num_threads + 1); ++i)
 135                   {
 136                     borders[i] = borderstart;
 137                     borderstart += chunk_length;
 138                   }
 139                 borders[num_threads + 1] = n;
 140               }
 141
 142             sums = static_cast<value_type*>(::operator new(sizeof(value_type)
 143                                                            * num_threads));
 144             OutputIterator target_end;
 145           } //single
 146
 147         thread_index_t iam = omp_get_thread_num();
 148         if (iam == 0)
 149           {
 150             *result = *begin;
 151             parallel_partial_sum_basecase(begin + 1, begin + borders[1],
 152                                           result + 1, bin_op, *begin);
 153             ::new(&(sums[iam])) value_type(*(result + borders[1] - 1));
 154           }
 155         else
 156           {
 157             ::new(&(sums[iam]))
 158               value_type(__gnu_parallel::accumulate(begin + borders[iam] + 1,
 159                                          begin + borders[iam + 1],
 160                                          *(begin + borders[iam]),
 161                                          bin_op,
 162                                          __gnu_parallel::sequential_tag()));
 163           }
 164
 165 #       pragma omp barrier
 166
 167 #       pragma omp single
 168           parallel_partial_sum_basecase(
 169               sums + 1, sums + num_threads, sums + 1, bin_op, sums[0]);
 170
 171 #       pragma omp barrier
 172
 173         // Still same team.
 174         parallel_partial_sum_basecase(begin + borders[iam + 1],
 175                                       begin + borders[iam + 2],
 176                                       result + borders[iam + 1], bin_op,
 177                                       sums[iam]);
 178       } //parallel
 179
 180     ::operator delete(sums);
 181     delete[] borders;
 182
 183     return result + n;
 184   }
 185
 186 /** @brief Parallel partial sum front-end.
 187   *  @param begin Begin iterator of input sequence.
 188   *  @param end End iterator of input sequence.
 189   *  @param result Begin iterator of output sequence.
 190   *  @param bin_op Associative binary function.
 191   *  @return End iterator of output sequence. */
 192 template<typename InputIterator,
 193          typename OutputIterator,
 194          typename BinaryOperation>
 195   OutputIterator
 196   parallel_partial_sum(InputIterator begin, InputIterator end,
 197                        OutputIterator result, BinaryOperation bin_op)
 198   {
 199     _GLIBCXX_CALL(begin - end)
 200
 201     typedef std::iterator_traits<InputIterator> traits_type;
 202     typedef typename traits_type::value_type value_type;
 203     typedef typename traits_type::difference_type difference_type;
 204
 205     difference_type n = end - begin;
 206
 207     switch (_Settings::get().partial_sum_algorithm)
 208       {
 209       case LINEAR:
 210         // Need an initial offset.
 211         return parallel_partial_sum_linear(begin, end, result, bin_op, n);
 212       default:
 213     // Partial_sum algorithm not implemented.
 214         _GLIBCXX_PARALLEL_ASSERT(0);
 215         return result + n;
 216       }
 217   }
 218 }
 219
 220 #endif /* _GLIBCXX_PARALLEL_PARTIAL_SUM_H */