libstdc++
partial_sum.h
Go to the documentation of this file.
1 // -*- C++ -*-
2 
3 // Copyright (C) 2007-2019 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,
60  _BinaryOperation __bin_op,
61  typename std::iterator_traits <_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  * @return End iterator of output sequence.
84  */
85  template<typename _IIter,
86  typename _OutputIterator,
87  typename _BinaryOperation>
88  _OutputIterator
89  __parallel_partial_sum_linear(_IIter __begin, _IIter __end,
90  _OutputIterator __result,
91  _BinaryOperation __bin_op,
92  typename std::iterator_traits<_IIter>::difference_type __n)
93  {
94  typedef std::iterator_traits<_IIter> _TraitsType;
95  typedef typename _TraitsType::value_type _ValueType;
96  typedef typename _TraitsType::difference_type _DifferenceType;
97 
98  if (__begin == __end)
99  return __result;
100 
101  _ThreadIndex __num_threads =
102  std::min<_DifferenceType>(__get_max_threads(), __n - 1);
103 
104  if (__num_threads < 2)
105  {
106  *__result = *__begin;
107  return __parallel_partial_sum_basecase(__begin + 1, __end,
108  __result + 1, __bin_op,
109  *__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 __first_part_length =
130  std::max<_DifferenceType>(1,
131  __n / (1.0f + __s.partial_sum_dilation * __num_threads));
132  _DifferenceType __chunk_length =
133  (__n - __first_part_length) / __num_threads;
134  _DifferenceType __borderstart =
135  __n - __num_threads * __chunk_length;
136  __borders[0] = 0;
137  for (_ThreadIndex __i = 1; __i < (__num_threads + 1); ++__i)
138  {
139  __borders[__i] = __borderstart;
140  __borderstart += __chunk_length;
141  }
142  __borders[__num_threads + 1] = __n;
143  }
144 
145  __sums = static_cast<_ValueType*>(::operator new(sizeof(_ValueType)
146  * __num_threads));
147  _OutputIterator __target_end;
148  } //single
149 
150  _ThreadIndex __iam = omp_get_thread_num();
151  if (__iam == 0)
152  {
153  *__result = *__begin;
155  __begin + __borders[1],
156  __result + 1,
157  __bin_op, *__begin);
158  ::new(&(__sums[__iam])) _ValueType(*(__result + __borders[1] - 1));
159  }
160  else
161  {
162  ::new(&(__sums[__iam]))
163  _ValueType(__gnu_parallel::accumulate(
164  __begin + __borders[__iam] + 1,
165  __begin + __borders[__iam + 1],
166  *(__begin + __borders[__iam]),
167  __bin_op,
169  }
170 
171 # pragma omp barrier
172 
173 # pragma omp single
174  __parallel_partial_sum_basecase(__sums + 1, __sums + __num_threads,
175  __sums + 1, __bin_op, __sums[0]);
176 
177 # pragma omp barrier
178 
179  // Still same team.
180  __parallel_partial_sum_basecase(__begin + __borders[__iam + 1],
181  __begin + __borders[__iam + 2],
182  __result + __borders[__iam + 1],
183  __bin_op, __sums[__iam]);
184  } //parallel
185 
186  for (_ThreadIndex __i = 0; __i < __num_threads; ++__i)
187  __sums[__i].~_ValueType();
188  ::operator delete(__sums);
189 
190  delete[] __borders;
191 
192  return __result + __n;
193  }
194 
195  /** @brief Parallel partial sum front-__end.
196  * @param __begin Begin iterator of input sequence.
197  * @param __end End iterator of input sequence.
198  * @param __result Begin iterator of output sequence.
199  * @param __bin_op Associative binary function.
200  * @return End iterator of output sequence. */
201  template<typename _IIter,
202  typename _OutputIterator,
203  typename _BinaryOperation>
204  _OutputIterator
205  __parallel_partial_sum(_IIter __begin, _IIter __end,
206  _OutputIterator __result, _BinaryOperation __bin_op)
207  {
208  _GLIBCXX_CALL(__begin - __end)
209 
210  typedef std::iterator_traits<_IIter> _TraitsType;
211  typedef typename _TraitsType::value_type _ValueType;
212  typedef typename _TraitsType::difference_type _DifferenceType;
213 
214  _DifferenceType __n = __end - __begin;
215 
216  switch (_Settings::get().partial_sum_algorithm)
217  {
218  case LINEAR:
219  // Need an initial offset.
220  return __parallel_partial_sum_linear(__begin, __end, __result,
221  __bin_op, __n);
222  default:
223  // Partial_sum algorithm not implemented.
224  _GLIBCXX_PARALLEL_ASSERT(0);
225  return __result + __n;
226  }
227  }
228 }
229 
230 #endif /* _GLIBCXX_PARALLEL_PARTIAL_SUM_H */
class _Settings Run-time settings for the parallel mode including all tunable parameters.
Definition: settings.h:123
_OutputIterator __parallel_partial_sum_linear(_IIter __begin, _IIter __end, _OutputIterator __result, _BinaryOperation __bin_op, typename std::iterator_traits< _IIter >::difference_type __n)
Parallel partial sum implementation, two-phase approach, no recursion.
Definition: partial_sum.h:89
_OutputIterator __parallel_partial_sum_basecase(_IIter __begin, _IIter __end, _OutputIterator __result, _BinaryOperation __bin_op, typename std::iterator_traits< _IIter >::value_type __value)
Base case prefix sum routine.
Definition: partial_sum.h:58
_Tp accumulate(_InputIterator __first, _InputIterator __last, _Tp __init)
Accumulate values in a range.
Definition: stl_numeric.h:128
#define _GLIBCXX_CALL(__n)
Macro to produce log message when entering a function.
_OutputIterator __parallel_partial_sum(_IIter __begin, _IIter __end, _OutputIterator __result, _BinaryOperation __bin_op)
Parallel partial sum front-__end.
Definition: partial_sum.h:205
Forces sequential execution at compile time.
Definition: tags.h:42
End-user include file. Provides advanced settings and tuning options. This file is a GNU parallel ext...
uint16_t _ThreadIndex
Unsigned integer to index a thread number. The maximum thread number (for each processor) must fit in...
Definition: types.h:123
float partial_sum_dilation
Ratio for partial_sum. Assume "sum and write result" to be this factor slower than just "sum"...
Definition: settings.h:207
GNU parallel code for public use.
_OutputIterator __equally_split(_DifferenceType __n, _ThreadIndex __num_threads, _OutputIterator __s)
function to split a sequence into parts of almost equal size.
Definition: equally_split.h:48
static const _Settings & get()
Get the global settings.