Home | History | Annotate | Download | only in parallel
      1 // -*- C++ -*-
      2 
      3 // Copyright (C) 2007-2013 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;
    154             __parallel_partial_sum_basecase(__begin + 1,
    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,
    168                                          __gnu_parallel::sequential_tag()));
    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 */
    231