Merge -r 127928:132243 from trunk
[official-gcc.git] / libstdc++-v3 / include / parallel / partial_sum.h
blobf7ca754720d1acc9a6aab155e1a77c5bafe139e6
1 // -*- C++ -*-
3 // Copyright (C) 2007, 2008 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.
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.
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.
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.
31 /** @file parallel/partial_sum.h
32 * @brief Parallel implementation of std::partial_sum(), i. e. prefix
33 * sums.
34 * This file is a GNU parallel extension to the Standard C++ Library.
37 // Written by Johannes Singler.
39 #ifndef _GLIBCXX_PARALLEL_PARTIAL_SUM_H
40 #define _GLIBCXX_PARALLEL_PARTIAL_SUM_H 1
42 #include <omp.h>
43 #include <new>
44 #include <bits/stl_algobase.h>
45 #include <parallel/parallel.h>
46 #include <parallel/numericfwd.h>
48 namespace __gnu_parallel
50 // Problem: there is no 0-element given.
52 /** @brief Base case prefix sum routine.
53 * @param begin Begin iterator of input sequence.
54 * @param end End iterator of input sequence.
55 * @param result Begin iterator of output sequence.
56 * @param bin_op Associative binary function.
57 * @param value Start value. Must be passed since the neutral
58 * element is unknown in general.
59 * @return End iterator of output sequence. */
60 template<typename InputIterator,
61 typename OutputIterator,
62 typename BinaryOperation>
63 OutputIterator
64 parallel_partial_sum_basecase(InputIterator begin, InputIterator end,
65 OutputIterator result, BinaryOperation bin_op,
66 typename std::iterator_traits
67 <InputIterator>::value_type value)
69 if (begin == end)
70 return result;
72 while (begin != end)
74 value = bin_op(value, *begin);
75 *result = value;
76 ++result;
77 ++begin;
79 return result;
82 /** @brief Parallel partial sum implementation, two-phase approach,
83 no recursion.
84 * @param begin Begin iterator of input sequence.
85 * @param end End iterator of input sequence.
86 * @param result Begin iterator of output sequence.
87 * @param bin_op Associative binary function.
88 * @param n Length of sequence.
89 * @param num_threads Number of threads to use.
90 * @return End iterator of output sequence.
92 template<typename InputIterator,
93 typename OutputIterator,
94 typename BinaryOperation>
95 OutputIterator
96 parallel_partial_sum_linear(InputIterator begin, InputIterator end,
97 OutputIterator result, BinaryOperation bin_op,
98 typename std::iterator_traits
99 <InputIterator>::difference_type n)
101 typedef std::iterator_traits<InputIterator> traits_type;
102 typedef typename traits_type::value_type value_type;
103 typedef typename traits_type::difference_type difference_type;
105 if (begin == end)
106 return result;
108 thread_index_t num_threads =
109 std::min<difference_type>(get_max_threads(), n - 1);
111 if (num_threads < 2)
113 *result = *begin;
114 return parallel_partial_sum_basecase(
115 begin + 1, end, result + 1, bin_op, *begin);
118 difference_type* borders;
119 value_type* sums;
121 # pragma omp parallel num_threads(num_threads)
123 # pragma omp single
125 num_threads = omp_get_num_threads();
127 borders = new difference_type[num_threads + 2];
129 if (Settings::partial_sum_dilatation == 1.0f)
130 equally_split(n, num_threads + 1, borders);
131 else
133 difference_type chunk_length =
134 ((double)n
135 / ((double)num_threads
136 + Settings::partial_sum_dilatation)),
137 borderstart = n - num_threads * chunk_length;
138 borders[0] = 0;
139 for (int i = 1; i < (num_threads + 1); ++i)
141 borders[i] = borderstart;
142 borderstart += chunk_length;
144 borders[num_threads + 1] = n;
147 sums = static_cast<value_type*>(::operator new(sizeof(value_type)
148 * num_threads));
149 OutputIterator target_end;
150 } //single
152 thread_index_t iam = omp_get_thread_num();
153 if (iam == 0)
155 *result = *begin;
156 parallel_partial_sum_basecase(begin + 1, begin + borders[1],
157 result + 1, bin_op, *begin);
158 ::new(&(sums[iam])) value_type(*(result + borders[1] - 1));
160 else
162 ::new(&(sums[iam]))
163 value_type(std::accumulate(begin + borders[iam] + 1,
164 begin + borders[iam + 1],
165 *(begin + borders[iam]),
166 bin_op,
167 __gnu_parallel::sequential_tag()));
170 # pragma omp barrier
172 # pragma omp single
173 parallel_partial_sum_basecase(
174 sums + 1, sums + num_threads, sums + 1, bin_op, sums[0]);
176 # pragma omp barrier
178 // Still same team.
179 parallel_partial_sum_basecase(begin + borders[iam + 1],
180 begin + borders[iam + 2],
181 result + borders[iam + 1], bin_op,
182 sums[iam]);
183 } //parallel
185 ::operator delete(sums);
186 delete[] borders;
188 return result + n;
191 /** @brief Parallel partial sum front-end.
192 * @param begin Begin iterator of input sequence.
193 * @param end End iterator of input sequence.
194 * @param result Begin iterator of output sequence.
195 * @param bin_op Associative binary function.
196 * @return End iterator of output sequence. */
197 template<typename InputIterator,
198 typename OutputIterator,
199 typename BinaryOperation>
200 OutputIterator
201 parallel_partial_sum(InputIterator begin, InputIterator end,
202 OutputIterator result, BinaryOperation bin_op)
204 _GLIBCXX_CALL(begin - end)
206 typedef std::iterator_traits<InputIterator> traits_type;
207 typedef typename traits_type::value_type value_type;
208 typedef typename traits_type::difference_type difference_type;
210 difference_type n = end - begin;
212 switch (Settings::partial_sum_algorithm)
214 case Settings::LINEAR:
215 // Need an initial offset.
216 return parallel_partial_sum_linear(begin, end, result, bin_op, n);
217 default:
218 // Partial_sum algorithm not implemented.
219 _GLIBCXX_PARALLEL_ASSERT(0);
220 return result + n;
225 #endif