OSDN Git Service

da04270b6e29f8c75c1471a5fad54fcc2c7775d8
[pf3gnuchains/gcc-fork.git] / libstdc++-v3 / 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 _IIter,
55          typename _OutputIterator,
56          typename _BinaryOperation>
57   _OutputIterator
58   __parallel_partial_sum_basecase(_IIter __begin, _IIter __end,
59                                 _OutputIterator __result, _BinaryOperation __bin_op,
60                                 typename std::iterator_traits
61                                 <_IIter>::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 _IIter,
87          typename _OutputIterator,
88          typename _BinaryOperation>
89   _OutputIterator
90   __parallel_partial_sum_linear(_IIter __begin, _IIter __end,
91                               _OutputIterator __result, _BinaryOperation __bin_op,
92                               typename std::iterator_traits
93                               <_IIter>::difference_type __n)
94   {
95     typedef std::iterator_traits<_IIter> _TraitsType;
96     typedef typename _TraitsType::value_type _ValueType;
97     typedef typename _TraitsType::difference_type _DifferenceType;
98
99     if (__begin == __end)
100       return __result;
101
102     _ThreadIndex __num_threads =
103         std::min<_DifferenceType>(__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     _DifferenceType* __borders;
113     _ValueType* __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 _DifferenceType[__num_threads + 2];
124
125             if (__s.partial_sum_dilation == 1.0f)
126               equally_split(__n, __num_threads + 1, __borders);
127             else
128               {
129                 _DifferenceType __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<_ValueType*>(::operator new(sizeof(_ValueType)
143                                                            * __num_threads));
144             _OutputIterator __target_end;
145           } //single
146
147         _ThreadIndex __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])) _ValueType(*(__result + __borders[1] - 1));
154           }
155         else
156           {
157             ::new(&(__sums[__iam]))
158               _ValueType(std::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 _IIter,
193          typename _OutputIterator,
194          typename _BinaryOperation>
195   _OutputIterator
196   __parallel_partial_sum(_IIter __begin, _IIter __end,
197                        _OutputIterator __result, _BinaryOperation __bin_op)
198   {
199     _GLIBCXX_CALL(__begin - __end)
200
201     typedef std::iterator_traits<_IIter> _TraitsType;
202     typedef typename _TraitsType::value_type _ValueType;
203     typedef typename _TraitsType::difference_type _DifferenceType;
204
205     _DifferenceType __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 */