3 // Copyright (C) 2007, 2008 Free Software Foundation, Inc.
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
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
31 /** @file parallel/multiseq_selection.h
32 * @brief Functions to find elements of a certain global rank in
33 * multiple sorted sequences. Also serves for splitting such
36 * The algorithm description can be found in
38 * P. J. Varman, S. D. Scheufler, B. R. Iyer, and G. R. Ricard.
39 * Merging Multiple Lists on Hierarchical-Memory Multiprocessors.
40 * Journal of Parallel and Distributed Computing, 12(2):171–177, 1991.
42 * This file is a GNU parallel extension to the Standard C++ Library.
45 // Written by Johannes Singler.
47 #ifndef _GLIBCXX_PARALLEL_MULTISEQ_SELECTION_H
48 #define _GLIBCXX_PARALLEL_MULTISEQ_SELECTION_H 1
53 #include <bits/stl_algo.h>
55 #include <parallel/sort.h>
57 namespace __gnu_parallel
59 /** @brief Compare a pair of types lexicographically, ascending. */
60 template<typename T1, typename T2, typename Comparator>
62 : public std::binary_function<std::pair<T1, T2>, std::pair<T1, T2>, bool>
68 lexicographic(Comparator& _comp) : comp(_comp) { }
71 operator()(const std::pair<T1, T2>& p1,
72 const std::pair<T1, T2>& p2) const
74 if (comp(p1.first, p2.first))
77 if (comp(p2.first, p1.first))
81 return p1.second < p2.second;
85 /** @brief Compare a pair of types lexicographically, descending. */
86 template<typename T1, typename T2, typename Comparator>
87 class lexicographic_reverse : public std::binary_function<T1, T2, bool>
93 lexicographic_reverse(Comparator& _comp) : comp(_comp) { }
96 operator()(const std::pair<T1, T2>& p1,
97 const std::pair<T1, T2>& p2) const
99 if (comp(p2.first, p1.first))
102 if (comp(p1.first, p2.first))
106 return p2.second < p1.second;
111 * @brief Splits several sorted sequences at a certain global rank,
112 * resulting in a splitting point for each sequence.
113 * The sequences are passed via a sequence of random-access
114 * iterator pairs, none of the sequences may be empty. If there
115 * are several equal elements across the split, the ones on the
116 * left side will be chosen from sequences with smaller number.
117 * @param begin_seqs Begin of the sequence of iterator pairs.
118 * @param end_seqs End of the sequence of iterator pairs.
119 * @param rank The global rank to partition at.
120 * @param begin_offsets A random-access sequence begin where the
121 * result will be stored in. Each element of the sequence is an
122 * iterator that points to the first element on the greater part of
123 * the respective sequence.
124 * @param comp The ordering functor, defaults to std::less<T>.
126 template<typename RanSeqs, typename RankType, typename RankIterator,
129 multiseq_partition(RanSeqs begin_seqs, RanSeqs end_seqs,
131 RankIterator begin_offsets,
132 Comparator comp = std::less<
133 typename std::iterator_traits<typename
134 std::iterator_traits<RanSeqs>::value_type::
135 first_type>::value_type>()) // std::less<T>
137 _GLIBCXX_CALL(end_seqs - begin_seqs)
139 typedef typename std::iterator_traits<RanSeqs>::value_type::first_type
141 typedef typename std::iterator_traits<It>::difference_type
143 typedef typename std::iterator_traits<It>::value_type value_type;
145 lexicographic<value_type, int, Comparator> lcomp(comp);
146 lexicographic_reverse<value_type, int, Comparator> lrcomp(comp);
148 // Number of sequences, number of elements in total (possibly
149 // including padding).
150 difference_type m = std::distance(begin_seqs, end_seqs), N = 0,
153 for (int i = 0; i < m; i++)
155 N += std::distance(begin_seqs[i].first, begin_seqs[i].second);
156 _GLIBCXX_PARALLEL_ASSERT(
157 std::distance(begin_seqs[i].first, begin_seqs[i].second) > 0);
162 for (int i = 0; i < m; i++)
163 begin_offsets[i] = begin_seqs[i].second; // Very end.
168 _GLIBCXX_PARALLEL_ASSERT(m != 0);
169 _GLIBCXX_PARALLEL_ASSERT(N != 0);
170 _GLIBCXX_PARALLEL_ASSERT(rank >= 0);
171 _GLIBCXX_PARALLEL_ASSERT(rank < N);
173 difference_type* ns = new difference_type[m];
174 difference_type* a = new difference_type[m];
175 difference_type* b = new difference_type[m];
178 ns[0] = std::distance(begin_seqs[0].first, begin_seqs[0].second);
180 for (int i = 0; i < m; i++)
182 ns[i] = std::distance(begin_seqs[i].first, begin_seqs[i].second);
183 nmax = std::max(nmax, ns[i]);
188 // Pad all lists to this length, at least as long as any ns[i],
189 // equality iff nmax = 2^k - 1.
192 // From now on, including padding.
195 for (int i = 0; i < m; i++)
203 // 0 <= a[i] <= ns[i], 0 <= b[i] <= l
205 #define S(i) (begin_seqs[i].first)
207 // Initial partition.
208 std::vector<std::pair<value_type, int> > sample;
210 for (int i = 0; i < m; i++)
211 if (n < ns[i]) //sequence long enough
212 sample.push_back(std::make_pair(S(i)[n], i));
213 __gnu_sequential::sort(sample.begin(), sample.end(), lcomp);
215 for (int i = 0; i < m; i++) //conceptual infinity
216 if (n >= ns[i]) //sequence too short, conceptual infinity
217 sample.push_back(std::make_pair(S(i)[0] /*dummy element*/, i));
219 difference_type localrank = rank * m / N ;
222 for (j = 0; j < localrank && ((n + 1) <= ns[sample[j].second]); ++j)
223 a[sample[j].second] += n + 1;
225 b[sample[j].second] -= n + 1;
227 // Further refinement.
232 int lmax_seq = -1; // to avoid warning
233 const value_type* lmax = NULL; // impossible to avoid the warning?
234 for (int i = 0; i < m; i++)
240 lmax = &(S(i)[a[i] - 1]);
245 // Max, favor rear sequences.
246 if (!comp(S(i)[a[i] - 1], *lmax))
248 lmax = &(S(i)[a[i] - 1]);
256 for (i = 0; i < m; i++)
258 difference_type middle = (b[i] + a[i]) / 2;
259 if (lmax && middle < ns[i] &&
260 lcomp(std::make_pair(S(i)[middle], i),
261 std::make_pair(*lmax, lmax_seq)))
262 a[i] = std::min(a[i] + n + 1, ns[i]);
267 difference_type leftsize = 0, total = 0;
268 for (int i = 0; i < m; i++)
270 leftsize += a[i] / (n + 1);
271 total += l / (n + 1);
274 difference_type skew = static_cast<difference_type>
275 (static_cast<uint64>(total) * rank / N - leftsize);
279 // Move to the left, find smallest.
280 std::priority_queue<std::pair<value_type, int>,
281 std::vector<std::pair<value_type, int> >,
282 lexicographic_reverse<value_type, int, Comparator> >
285 for (int i = 0; i < m; i++)
287 pq.push(std::make_pair(S(i)[b[i]], i));
289 for (; skew != 0 && !pq.empty(); --skew)
291 int source = pq.top().second;
294 a[source] = std::min(a[source] + n + 1, ns[source]);
297 if (b[source] < ns[source])
298 pq.push(std::make_pair(S(source)[b[source]], source));
303 // Move to the right, find greatest.
304 std::priority_queue<std::pair<value_type, int>,
305 std::vector<std::pair<value_type, int> >,
306 lexicographic<value_type, int, Comparator> > pq(lcomp);
308 for (int i = 0; i < m; i++)
310 pq.push(std::make_pair(S(i)[a[i] - 1], i));
312 for (; skew != 0; ++skew)
314 int source = pq.top().second;
321 pq.push(std::make_pair(S(source)[a[source] - 1], source));
327 // a[i] == b[i] in most cases, except when a[i] has been clamped
328 // because of having reached the boundary
330 // Now return the result, calculate the offset.
332 // Compare the keys on both edges of the border.
334 // Maximum of left edge, minimum of right edge.
335 value_type* maxleft = NULL;
336 value_type* minright = NULL;
337 for (int i = 0; i < m; i++)
342 maxleft = &(S(i)[a[i] - 1]);
345 // Max, favor rear sequences.
346 if (!comp(S(i)[a[i] - 1], *maxleft))
347 maxleft = &(S(i)[a[i] - 1]);
353 minright = &(S(i)[b[i]]);
356 // Min, favor fore sequences.
357 if (comp(S(i)[b[i]], *minright))
358 minright = &(S(i)[b[i]]);
364 for (int i = 0; i < m; i++)
365 begin_offsets[i] = S(i) + a[i];
374 * @brief Selects the element at a certain global rank from several
377 * The sequences are passed via a sequence of random-access
378 * iterator pairs, none of the sequences may be empty.
379 * @param begin_seqs Begin of the sequence of iterator pairs.
380 * @param end_seqs End of the sequence of iterator pairs.
381 * @param rank The global rank to partition at.
382 * @param offset The rank of the selected element in the global
383 * subsequence of elements equal to the selected element. If the
384 * selected element is unique, this number is 0.
385 * @param comp The ordering functor, defaults to std::less.
387 template<typename T, typename RanSeqs, typename RankType,
390 multiseq_selection(RanSeqs begin_seqs, RanSeqs end_seqs, RankType rank,
391 RankType& offset, Comparator comp = std::less<T>())
393 _GLIBCXX_CALL(end_seqs - begin_seqs)
395 typedef typename std::iterator_traits<RanSeqs>::value_type::first_type
397 typedef typename std::iterator_traits<It>::difference_type
400 lexicographic<T, int, Comparator> lcomp(comp);
401 lexicographic_reverse<T, int, Comparator> lrcomp(comp);
403 // Number of sequences, number of elements in total (possibly
404 // including padding).
405 difference_type m = std::distance(begin_seqs, end_seqs);
406 difference_type N = 0;
407 difference_type nmax, n, r;
409 for (int i = 0; i < m; i++)
410 N += std::distance(begin_seqs[i].first, begin_seqs[i].second);
412 if (m == 0 || N == 0 || rank < 0 || rank >= N)
414 // Result undefined when there is no data or rank is outside bounds.
415 throw std::exception();
419 difference_type* ns = new difference_type[m];
420 difference_type* a = new difference_type[m];
421 difference_type* b = new difference_type[m];
424 ns[0] = std::distance(begin_seqs[0].first, begin_seqs[0].second);
426 for (int i = 0; i < m; ++i)
428 ns[i] = std::distance(begin_seqs[i].first, begin_seqs[i].second);
429 nmax = std::max(nmax, ns[i]);
434 // Pad all lists to this length, at least as long as any ns[i],
435 // equality iff nmax = 2^k - 1
438 // From now on, including padding.
441 for (int i = 0; i < m; ++i)
449 // 0 <= a[i] <= ns[i], 0 <= b[i] <= l
451 #define S(i) (begin_seqs[i].first)
453 // Initial partition.
454 std::vector<std::pair<T, int> > sample;
456 for (int i = 0; i < m; i++)
458 sample.push_back(std::make_pair(S(i)[n], i));
459 __gnu_sequential::sort(sample.begin(), sample.end(),
460 lcomp, sequential_tag());
462 // Conceptual infinity.
463 for (int i = 0; i < m; i++)
465 sample.push_back(std::make_pair(S(i)[0] /*dummy element*/, i));
467 difference_type localrank = rank * m / N ;
470 for (j = 0; j < localrank && ((n + 1) <= ns[sample[j].second]); ++j)
471 a[sample[j].second] += n + 1;
473 b[sample[j].second] -= n + 1;
475 // Further refinement.
480 const T* lmax = NULL;
481 for (int i = 0; i < m; ++i)
486 lmax = &(S(i)[a[i] - 1]);
489 if (comp(*lmax, S(i)[a[i] - 1])) //max
490 lmax = &(S(i)[a[i] - 1]);
496 for (i = 0; i < m; i++)
498 difference_type middle = (b[i] + a[i]) / 2;
499 if (lmax && middle < ns[i] && comp(S(i)[middle], *lmax))
500 a[i] = std::min(a[i] + n + 1, ns[i]);
505 difference_type leftsize = 0, total = 0;
506 for (int i = 0; i < m; ++i)
508 leftsize += a[i] / (n + 1);
509 total += l / (n + 1);
512 difference_type skew = ((unsigned long long)total * rank / N
517 // Move to the left, find smallest.
518 std::priority_queue<std::pair<T, int>,
519 std::vector<std::pair<T, int> >,
520 lexicographic_reverse<T, int, Comparator> > pq(lrcomp);
522 for (int i = 0; i < m; ++i)
524 pq.push(std::make_pair(S(i)[b[i]], i));
526 for (; skew != 0 && !pq.empty(); --skew)
528 int source = pq.top().second;
531 a[source] = std::min(a[source] + n + 1, ns[source]);
534 if (b[source] < ns[source])
535 pq.push(std::make_pair(S(source)[b[source]], source));
540 // Move to the right, find greatest.
541 std::priority_queue<std::pair<T, int>,
542 std::vector<std::pair<T, int> >,
543 lexicographic<T, int, Comparator> > pq(lcomp);
545 for (int i = 0; i < m; ++i)
547 pq.push(std::make_pair(S(i)[a[i] - 1], i));
549 for (; skew != 0; ++skew)
551 int source = pq.top().second;
558 pq.push(std::make_pair(S(source)[a[source] - 1], source));
564 // a[i] == b[i] in most cases, except when a[i] has been clamped
565 // because of having reached the boundary
567 // Now return the result, calculate the offset.
569 // Compare the keys on both edges of the border.
571 // Maximum of left edge, minimum of right edge.
572 bool maxleftset = false, minrightset = false;
574 // Impossible to avoid the warning?
576 for (int i = 0; i < m; ++i)
582 maxleft = S(i)[a[i] - 1];
588 if (comp(maxleft, S(i)[a[i] - 1]))
589 maxleft = S(i)[a[i] - 1];
596 minright = S(i)[b[i]];
602 if (comp(S(i)[b[i]], minright))
603 minright = S(i)[b[i]];
608 // Minright is the splitter, in any case.
610 if (!maxleftset || comp(minright, maxleft))
612 // Good luck, everything is split unambiguously.
617 // We have to calculate an offset.
620 for (int i = 0; i < m; ++i)
622 difference_type lb = std::lower_bound(S(i), S(i) + ns[i],