xref: /dflybsd-src/contrib/gcc-8.0/libstdc++-v3/include/parallel/list_partition.h (revision 38fd149817dfbff97799f62fcb70be98c4e32523)
1*38fd1498Szrj // -*- C++ -*-
2*38fd1498Szrj 
3*38fd1498Szrj // Copyright (C) 2007-2018 Free Software Foundation, Inc.
4*38fd1498Szrj //
5*38fd1498Szrj // This file is part of the GNU ISO C++ Library.  This library is free
6*38fd1498Szrj // software; you can redistribute __it and/or modify __it under the terms
7*38fd1498Szrj // of the GNU General Public License as published by the Free Software
8*38fd1498Szrj // Foundation; either version 3, or (at your option) any later
9*38fd1498Szrj // version.
10*38fd1498Szrj 
11*38fd1498Szrj // This library is distributed in the hope that __it will be useful, but
12*38fd1498Szrj // WITHOUT ANY WARRANTY; without even the implied warranty of
13*38fd1498Szrj // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14*38fd1498Szrj // General Public License for more details.
15*38fd1498Szrj 
16*38fd1498Szrj // Under Section 7 of GPL version 3, you are granted additional
17*38fd1498Szrj // permissions described in the GCC Runtime Library Exception, version
18*38fd1498Szrj // 3.1, as published by the Free Software Foundation.
19*38fd1498Szrj 
20*38fd1498Szrj // You should have received a copy of the GNU General Public License and
21*38fd1498Szrj // a copy of the GCC Runtime Library Exception along with this program;
22*38fd1498Szrj // see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
23*38fd1498Szrj // <http://www.gnu.org/licenses/>.
24*38fd1498Szrj 
25*38fd1498Szrj /** @file parallel/list_partition.h
26*38fd1498Szrj  *  @brief _Functionality to split __sequence referenced by only input
27*38fd1498Szrj  *  iterators.
28*38fd1498Szrj  *  This file is a GNU parallel extension to the Standard C++ Library.
29*38fd1498Szrj  */
30*38fd1498Szrj 
31*38fd1498Szrj // Written by Leonor Frias Moya and Johannes Singler.
32*38fd1498Szrj 
33*38fd1498Szrj #ifndef _GLIBCXX_PARALLEL_LIST_PARTITION_H
34*38fd1498Szrj #define _GLIBCXX_PARALLEL_LIST_PARTITION_H 1
35*38fd1498Szrj 
36*38fd1498Szrj #include <parallel/parallel.h>
37*38fd1498Szrj #include <vector>
38*38fd1498Szrj 
39*38fd1498Szrj namespace __gnu_parallel
40*38fd1498Szrj {
41*38fd1498Szrj   /** @brief Shrinks and doubles the ranges.
42*38fd1498Szrj    *  @param __os_starts Start positions worked on (oversampled).
43*38fd1498Szrj    *  @param __count_to_two Counts up to 2.
44*38fd1498Szrj    *  @param __range_length Current length of a chunk.
45*38fd1498Szrj    *  @param __make_twice Whether the @c __os_starts is allowed to be
46*38fd1498Szrj    *  grown or not
47*38fd1498Szrj    */
48*38fd1498Szrj   template<typename _IIter>
49*38fd1498Szrj     void
__shrink_and_double(std::vector<_IIter> & __os_starts,size_t & __count_to_two,size_t & __range_length,const bool __make_twice)50*38fd1498Szrj     __shrink_and_double(std::vector<_IIter>& __os_starts,
51*38fd1498Szrj 			size_t& __count_to_two, size_t& __range_length,
52*38fd1498Szrj 			const bool __make_twice)
53*38fd1498Szrj     {
54*38fd1498Szrj       ++__count_to_two;
55*38fd1498Szrj       if (!__make_twice || __count_to_two < 2)
56*38fd1498Szrj         __shrink(__os_starts, __count_to_two, __range_length);
57*38fd1498Szrj       else
58*38fd1498Szrj         {
59*38fd1498Szrj           __os_starts.resize((__os_starts.size() - 1) * 2 + 1);
60*38fd1498Szrj           __count_to_two = 0;
61*38fd1498Szrj         }
62*38fd1498Szrj     }
63*38fd1498Szrj 
64*38fd1498Szrj   /** @brief Combines two ranges into one and thus halves the number of ranges.
65*38fd1498Szrj    *  @param __os_starts Start positions worked on (oversampled).
66*38fd1498Szrj    *  @param __count_to_two Counts up to 2.
67*38fd1498Szrj    *  @param __range_length Current length of a chunk. */
68*38fd1498Szrj   template<typename _IIter>
69*38fd1498Szrj     void
__shrink(std::vector<_IIter> & __os_starts,size_t & __count_to_two,size_t & __range_length)70*38fd1498Szrj     __shrink(std::vector<_IIter>& __os_starts, size_t& __count_to_two,
71*38fd1498Szrj 	     size_t& __range_length)
72*38fd1498Szrj     {
73*38fd1498Szrj       for (typename std::vector<_IIter>::size_type __i = 0;
74*38fd1498Szrj            __i <= (__os_starts.size() / 2); ++__i)
75*38fd1498Szrj         __os_starts[__i] = __os_starts[__i * 2];
76*38fd1498Szrj       __range_length *= 2;
77*38fd1498Szrj     }
78*38fd1498Szrj 
79*38fd1498Szrj   /** @brief Splits a sequence given by input iterators into parts of
80*38fd1498Szrj    * almost equal size
81*38fd1498Szrj    *
82*38fd1498Szrj    *  The function needs only one pass over the sequence.
83*38fd1498Szrj    *  @param __begin Begin iterator of input sequence.
84*38fd1498Szrj    *  @param __end End iterator of input sequence.
85*38fd1498Szrj    *  @param __starts Start iterators for the resulting parts, dimension
86*38fd1498Szrj    *  @c __num_parts+1. For convenience, @c __starts @c [__num_parts]
87*38fd1498Szrj    *  contains the end iterator of the sequence.
88*38fd1498Szrj    *  @param __lengths Length of the resulting parts.
89*38fd1498Szrj    *  @param __num_parts Number of parts to split the sequence into.
90*38fd1498Szrj    *  @param __f Functor to be applied to each element by traversing __it
91*38fd1498Szrj    *  @param __oversampling Oversampling factor. If 0, then the
92*38fd1498Szrj    *  partitions will differ in at most
93*38fd1498Szrj    *  \f$\sqrt{\mathrm{end} - \mathrm{begin}}\f$
94*38fd1498Szrj    *  elements. Otherwise, the ratio between the
95*38fd1498Szrj    *  longest and the shortest part is bounded by
96*38fd1498Szrj    *  \f$1/(\mathrm{oversampling} \cdot \mathrm{num\_parts})\f$
97*38fd1498Szrj    *  @return Length of the whole sequence.
98*38fd1498Szrj    */
99*38fd1498Szrj   template<typename _IIter, typename _FunctorType>
100*38fd1498Szrj     size_t
101*38fd1498Szrj     list_partition(const _IIter __begin, const _IIter __end,
102*38fd1498Szrj                    _IIter* __starts, size_t* __lengths, const int __num_parts,
103*38fd1498Szrj                    _FunctorType& __f, int __oversampling = 0)
104*38fd1498Szrj     {
105*38fd1498Szrj       bool __make_twice = false;
106*38fd1498Szrj 
107*38fd1498Szrj       // The resizing algorithm is chosen according to the oversampling factor.
108*38fd1498Szrj       if (__oversampling == 0)
109*38fd1498Szrj         {
110*38fd1498Szrj           __make_twice = true;
111*38fd1498Szrj           __oversampling = 1;
112*38fd1498Szrj         }
113*38fd1498Szrj 
114*38fd1498Szrj       std::vector<_IIter> __os_starts(2 * __oversampling * __num_parts + 1);
115*38fd1498Szrj 
116*38fd1498Szrj       __os_starts[0] = __begin;
117*38fd1498Szrj       _IIter __prev  = __begin, __it = __begin;
118*38fd1498Szrj       size_t __dist_limit = 0, __dist = 0;
119*38fd1498Szrj       size_t __cur = 1, __next = 1;
120*38fd1498Szrj       size_t __range_length = 1;
121*38fd1498Szrj       size_t __count_to_two = 0;
122*38fd1498Szrj       while (__it != __end)
123*38fd1498Szrj         {
124*38fd1498Szrj           __cur = __next;
125*38fd1498Szrj           for (; __cur < __os_starts.size() and __it != __end; ++__cur)
126*38fd1498Szrj             {
127*38fd1498Szrj               for (__dist_limit += __range_length;
128*38fd1498Szrj                    __dist < __dist_limit and __it != __end; ++__dist)
129*38fd1498Szrj                 {
130*38fd1498Szrj                   __f(__it);
131*38fd1498Szrj                   ++__it;
132*38fd1498Szrj                 }
133*38fd1498Szrj               __os_starts[__cur] = __it;
134*38fd1498Szrj             }
135*38fd1498Szrj 
136*38fd1498Szrj           // Must compare for end and not __cur < __os_starts.size() , because
137*38fd1498Szrj           // __cur could be == __os_starts.size() as well
138*38fd1498Szrj           if (__it == __end)
139*38fd1498Szrj             break;
140*38fd1498Szrj 
141*38fd1498Szrj           __shrink_and_double(__os_starts, __count_to_two, __range_length,
142*38fd1498Szrj                               __make_twice);
143*38fd1498Szrj           __next = __os_starts.size() / 2 + 1;
144*38fd1498Szrj         }
145*38fd1498Szrj 
146*38fd1498Szrj       // Calculation of the parts (one must be extracted from __current
147*38fd1498Szrj       // because the partition beginning at end, consists only of
148*38fd1498Szrj       // itself).
149*38fd1498Szrj       size_t __size_part = (__cur - 1) / __num_parts;
150*38fd1498Szrj       int __size_greater = static_cast<int>((__cur - 1) % __num_parts);
151*38fd1498Szrj       __starts[0] = __os_starts[0];
152*38fd1498Szrj 
153*38fd1498Szrj       size_t __index = 0;
154*38fd1498Szrj 
155*38fd1498Szrj       // Smallest partitions.
156*38fd1498Szrj       for (int __i = 1; __i < (__num_parts + 1 - __size_greater); ++__i)
157*38fd1498Szrj         {
158*38fd1498Szrj           __lengths[__i - 1] =  __size_part * __range_length;
159*38fd1498Szrj           __index += __size_part;
160*38fd1498Szrj           __starts[__i] = __os_starts[__index];
161*38fd1498Szrj         }
162*38fd1498Szrj 
163*38fd1498Szrj       // Biggest partitions.
164*38fd1498Szrj       for (int __i = __num_parts + 1 - __size_greater; __i <= __num_parts;
165*38fd1498Szrj            ++__i)
166*38fd1498Szrj         {
167*38fd1498Szrj           __lengths[__i - 1] =  (__size_part+1) * __range_length;
168*38fd1498Szrj           __index += (__size_part+1);
169*38fd1498Szrj           __starts[__i] = __os_starts[__index];
170*38fd1498Szrj         }
171*38fd1498Szrj 
172*38fd1498Szrj       // Correction of the end size (the end iteration has not finished).
173*38fd1498Szrj       __lengths[__num_parts - 1] -= (__dist_limit - __dist);
174*38fd1498Szrj 
175*38fd1498Szrj       return __dist;
176*38fd1498Szrj     }
177*38fd1498Szrj }
178*38fd1498Szrj 
179*38fd1498Szrj #endif /* _GLIBCXX_PARALLEL_LIST_PARTITION_H */
180