1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
|
// -*- C++ -*-
// Copyright (C) 2007-2015 Free Software Foundation, Inc.
//
// This file is part of the GNU ISO C++ Library. This library is free
// software; you can redistribute it and/or modify it under the terms
// of the GNU General Public License as published by the Free Software
// Foundation; either version 3, or (at your option) any later
// version.
// This library is distributed in the hope that it will be useful, but
// WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
// General Public License for more details.
// Under Section 7 of GPL version 3, you are granted additional
// permissions described in the GCC Runtime Library Exception, version
// 3.1, as published by the Free Software Foundation.
// You should have received a copy of the GNU General Public License and
// a copy of the GCC Runtime Library Exception along with this program;
// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
// <http://www.gnu.org/licenses/>.
/** @file parallel/multiseq_selection.h
* @brief Functions to find elements of a certain global __rank in
* multiple sorted sequences. Also serves for splitting such
* sequence sets.
*
* The algorithm description can be found in
*
* P. J. Varman, S. D. Scheufler, B. R. Iyer, and G. R. Ricard.
* Merging Multiple Lists on Hierarchical-Memory Multiprocessors.
* Journal of Parallel and Distributed Computing, 12(2):171–177, 1991.
*
* This file is a GNU parallel extension to the Standard C++ Library.
*/
// Written by Johannes Singler.
#ifndef _GLIBCXX_PARALLEL_MULTISEQ_SELECTION_H
#define _GLIBCXX_PARALLEL_MULTISEQ_SELECTION_H 1
#include <vector>
#include <queue>
#include <bits/stl_algo.h>
namespace __gnu_parallel
{
/** @brief Compare __a pair of types lexicographically, ascending. */
template<typename _T1, typename _T2, typename _Compare>
class _Lexicographic
: public std::binary_function<std::pair<_T1, _T2>,
std::pair<_T1, _T2>, bool>
{
private:
_Compare& _M_comp;
public:
_Lexicographic(_Compare& __comp) : _M_comp(__comp) { }
bool
operator()(const std::pair<_T1, _T2>& __p1,
const std::pair<_T1, _T2>& __p2) const
{
if (_M_comp(__p1.first, __p2.first))
return true;
if (_M_comp(__p2.first, __p1.first))
return false;
// Firsts are equal.
return __p1.second < __p2.second;
}
};
/** @brief Compare __a pair of types lexicographically, descending. */
template<typename _T1, typename _T2, typename _Compare>
class _LexicographicReverse : public std::binary_function<_T1, _T2, bool>
{
private:
_Compare& _M_comp;
public:
_LexicographicReverse(_Compare& __comp) : _M_comp(__comp) { }
bool
operator()(const std::pair<_T1, _T2>& __p1,
const std::pair<_T1, _T2>& __p2) const
{
if (_M_comp(__p2.first, __p1.first))
return true;
if (_M_comp(__p1.first, __p2.first))
return false;
// Firsts are equal.
return __p2.second < __p1.second;
}
};
/**
* @brief Splits several sorted sequences at a certain global __rank,
* resulting in a splitting point for each sequence.
* The sequences are passed via a sequence of random-access
* iterator pairs, none of the sequences may be empty. If there
* are several equal elements across the split, the ones on the
* __left side will be chosen from sequences with smaller number.
* @param __begin_seqs Begin of the sequence of iterator pairs.
* @param __end_seqs End of the sequence of iterator pairs.
* @param __rank The global rank to partition at.
* @param __begin_offsets A random-access __sequence __begin where the
* __result will be stored in. Each element of the sequence is an
* iterator that points to the first element on the greater part of
* the respective __sequence.
* @param __comp The ordering functor, defaults to std::less<_Tp>.
*/
template<typename _RanSeqs, typename _RankType, typename _RankIterator,
typename _Compare>
void
multiseq_partition(_RanSeqs __begin_seqs, _RanSeqs __end_seqs,
_RankType __rank,
_RankIterator __begin_offsets,
_Compare __comp = std::less<
typename std::iterator_traits<typename
std::iterator_traits<_RanSeqs>::value_type::
first_type>::value_type>()) // std::less<_Tp>
{
_GLIBCXX_CALL(__end_seqs - __begin_seqs)
typedef typename std::iterator_traits<_RanSeqs>::value_type::first_type
_It;
typedef typename std::iterator_traits<_RanSeqs>::difference_type
_SeqNumber;
typedef typename std::iterator_traits<_It>::difference_type
_DifferenceType;
typedef typename std::iterator_traits<_It>::value_type _ValueType;
_Lexicographic<_ValueType, _SeqNumber, _Compare> __lcomp(__comp);
_LexicographicReverse<_ValueType, _SeqNumber, _Compare> __lrcomp(__comp);
// Number of sequences, number of elements in total (possibly
// including padding).
_DifferenceType __m = std::distance(__begin_seqs, __end_seqs), __nn = 0,
__nmax, __n, __r;
for (_SeqNumber __i = 0; __i < __m; __i++)
{
__nn += std::distance(__begin_seqs[__i].first,
__begin_seqs[__i].second);
_GLIBCXX_PARALLEL_ASSERT(
std::distance(__begin_seqs[__i].first,
__begin_seqs[__i].second) > 0);
}
if (__rank == __nn)
{
for (_SeqNumber __i = 0; __i < __m; __i++)
__begin_offsets[__i] = __begin_seqs[__i].second; // Very end.
// Return __m - 1;
return;
}
_GLIBCXX_PARALLEL_ASSERT(__m != 0);
_GLIBCXX_PARALLEL_ASSERT(__nn != 0);
_GLIBCXX_PARALLEL_ASSERT(__rank >= 0);
_GLIBCXX_PARALLEL_ASSERT(__rank < __nn);
_DifferenceType* __ns = new _DifferenceType[__m];
_DifferenceType* __a = new _DifferenceType[__m];
_DifferenceType* __b = new _DifferenceType[__m];
_DifferenceType __l;
__ns[0] = std::distance(__begin_seqs[0].first, __begin_seqs[0].second);
__nmax = __ns[0];
for (_SeqNumber __i = 0; __i < __m; __i++)
{
__ns[__i] = std::distance(__begin_seqs[__i].first,
__begin_seqs[__i].second);
__nmax = std::max(__nmax, __ns[__i]);
}
__r = __rd_log2(__nmax) + 1;
// Pad all lists to this length, at least as long as any ns[__i],
// equality iff __nmax = 2^__k - 1.
__l = (1ULL << __r) - 1;
for (_SeqNumber __i = 0; __i < __m; __i++)
{
__a[__i] = 0;
__b[__i] = __l;
}
__n = __l / 2;
// Invariants:
// 0 <= __a[__i] <= __ns[__i], 0 <= __b[__i] <= __l
#define __S(__i) (__begin_seqs[__i].first)
// Initial partition.
std::vector<std::pair<_ValueType, _SeqNumber> > __sample;
for (_SeqNumber __i = 0; __i < __m; __i++)
if (__n < __ns[__i]) //__sequence long enough
__sample.push_back(std::make_pair(__S(__i)[__n], __i));
__gnu_sequential::sort(__sample.begin(), __sample.end(), __lcomp);
for (_SeqNumber __i = 0; __i < __m; __i++) //conceptual infinity
if (__n >= __ns[__i]) //__sequence too short, conceptual infinity
__sample.push_back(
std::make_pair(__S(__i)[0] /*__dummy element*/, __i));
_DifferenceType __localrank = __rank / __l;
_SeqNumber __j;
for (__j = 0;
__j < __localrank && ((__n + 1) <= __ns[__sample[__j].second]);
++__j)
__a[__sample[__j].second] += __n + 1;
for (; __j < __m; __j++)
__b[__sample[__j].second] -= __n + 1;
// Further refinement.
while (__n > 0)
{
__n /= 2;
_SeqNumber __lmax_seq = -1; // to avoid warning
const _ValueType* __lmax = 0; // impossible to avoid the warning?
for (_SeqNumber __i = 0; __i < __m; __i++)
{
if (__a[__i] > 0)
{
if (!__lmax)
{
__lmax = &(__S(__i)[__a[__i] - 1]);
__lmax_seq = __i;
}
else
{
// Max, favor rear sequences.
if (!__comp(__S(__i)[__a[__i] - 1], *__lmax))
{
__lmax = &(__S(__i)[__a[__i] - 1]);
__lmax_seq = __i;
}
}
}
}
_SeqNumber __i;
for (__i = 0; __i < __m; __i++)
{
_DifferenceType __middle = (__b[__i] + __a[__i]) / 2;
if (__lmax && __middle < __ns[__i] &&
__lcomp(std::make_pair(__S(__i)[__middle], __i),
std::make_pair(*__lmax, __lmax_seq)))
__a[__i] = std::min(__a[__i] + __n + 1, __ns[__i]);
else
__b[__i] -= __n + 1;
}
_DifferenceType __leftsize = 0;
for (_SeqNumber __i = 0; __i < __m; __i++)
__leftsize += __a[__i] / (__n + 1);
_DifferenceType __skew = __rank / (__n + 1) - __leftsize;
if (__skew > 0)
{
// Move to the left, find smallest.
std::priority_queue<std::pair<_ValueType, _SeqNumber>,
std::vector<std::pair<_ValueType, _SeqNumber> >,
_LexicographicReverse<_ValueType, _SeqNumber, _Compare> >
__pq(__lrcomp);
for (_SeqNumber __i = 0; __i < __m; __i++)
if (__b[__i] < __ns[__i])
__pq.push(std::make_pair(__S(__i)[__b[__i]], __i));
for (; __skew != 0 && !__pq.empty(); --__skew)
{
_SeqNumber __source = __pq.top().second;
__pq.pop();
__a[__source]
= std::min(__a[__source] + __n + 1, __ns[__source]);
__b[__source] += __n + 1;
if (__b[__source] < __ns[__source])
__pq.push(
std::make_pair(__S(__source)[__b[__source]], __source));
}
}
else if (__skew < 0)
{
// Move to the right, find greatest.
std::priority_queue<std::pair<_ValueType, _SeqNumber>,
std::vector<std::pair<_ValueType, _SeqNumber> >,
_Lexicographic<_ValueType, _SeqNumber, _Compare> >
__pq(__lcomp);
for (_SeqNumber __i = 0; __i < __m; __i++)
if (__a[__i] > 0)
__pq.push(std::make_pair(__S(__i)[__a[__i] - 1], __i));
for (; __skew != 0; ++__skew)
{
_SeqNumber __source = __pq.top().second;
__pq.pop();
__a[__source] -= __n + 1;
__b[__source] -= __n + 1;
if (__a[__source] > 0)
__pq.push(std::make_pair(
__S(__source)[__a[__source] - 1], __source));
}
}
}
// Postconditions:
// __a[__i] == __b[__i] in most cases, except when __a[__i] has been
// clamped because of having reached the boundary
// Now return the result, calculate the offset.
// Compare the keys on both edges of the border.
// Maximum of left edge, minimum of right edge.
_ValueType* __maxleft = 0;
_ValueType* __minright = 0;
for (_SeqNumber __i = 0; __i < __m; __i++)
{
if (__a[__i] > 0)
{
if (!__maxleft)
__maxleft = &(__S(__i)[__a[__i] - 1]);
else
{
// Max, favor rear sequences.
if (!__comp(__S(__i)[__a[__i] - 1], *__maxleft))
__maxleft = &(__S(__i)[__a[__i] - 1]);
}
}
if (__b[__i] < __ns[__i])
{
if (!__minright)
__minright = &(__S(__i)[__b[__i]]);
else
{
// Min, favor fore sequences.
if (__comp(__S(__i)[__b[__i]], *__minright))
__minright = &(__S(__i)[__b[__i]]);
}
}
}
_SeqNumber __seq = 0;
for (_SeqNumber __i = 0; __i < __m; __i++)
__begin_offsets[__i] = __S(__i) + __a[__i];
delete[] __ns;
delete[] __a;
delete[] __b;
}
/**
* @brief Selects the element at a certain global __rank from several
* sorted sequences.
*
* The sequences are passed via a sequence of random-access
* iterator pairs, none of the sequences may be empty.
* @param __begin_seqs Begin of the sequence of iterator pairs.
* @param __end_seqs End of the sequence of iterator pairs.
* @param __rank The global rank to partition at.
* @param __offset The rank of the selected element in the global
* subsequence of elements equal to the selected element. If the
* selected element is unique, this number is 0.
* @param __comp The ordering functor, defaults to std::less.
*/
template<typename _Tp, typename _RanSeqs, typename _RankType,
typename _Compare>
_Tp
multiseq_selection(_RanSeqs __begin_seqs, _RanSeqs __end_seqs,
_RankType __rank,
_RankType& __offset, _Compare __comp = std::less<_Tp>())
{
_GLIBCXX_CALL(__end_seqs - __begin_seqs)
typedef typename std::iterator_traits<_RanSeqs>::value_type::first_type
_It;
typedef typename std::iterator_traits<_RanSeqs>::difference_type
_SeqNumber;
typedef typename std::iterator_traits<_It>::difference_type
_DifferenceType;
_Lexicographic<_Tp, _SeqNumber, _Compare> __lcomp(__comp);
_LexicographicReverse<_Tp, _SeqNumber, _Compare> __lrcomp(__comp);
// Number of sequences, number of elements in total (possibly
// including padding).
_DifferenceType __m = std::distance(__begin_seqs, __end_seqs);
_DifferenceType __nn = 0;
_DifferenceType __nmax, __n, __r;
for (_SeqNumber __i = 0; __i < __m; __i++)
__nn += std::distance(__begin_seqs[__i].first,
__begin_seqs[__i].second);
if (__m == 0 || __nn == 0 || __rank < 0 || __rank >= __nn)
{
// result undefined if there is no data or __rank is outside bounds
throw std::exception();
}
_DifferenceType* __ns = new _DifferenceType[__m];
_DifferenceType* __a = new _DifferenceType[__m];
_DifferenceType* __b = new _DifferenceType[__m];
_DifferenceType __l;
__ns[0] = std::distance(__begin_seqs[0].first, __begin_seqs[0].second);
__nmax = __ns[0];
for (_SeqNumber __i = 0; __i < __m; ++__i)
{
__ns[__i] = std::distance(__begin_seqs[__i].first,
__begin_seqs[__i].second);
__nmax = std::max(__nmax, __ns[__i]);
}
__r = __rd_log2(__nmax) + 1;
// Pad all lists to this length, at least as long as any ns[__i],
// equality iff __nmax = 2^__k - 1
__l = __round_up_to_pow2(__r) - 1;
for (_SeqNumber __i = 0; __i < __m; ++__i)
{
__a[__i] = 0;
__b[__i] = __l;
}
__n = __l / 2;
// Invariants:
// 0 <= __a[__i] <= __ns[__i], 0 <= __b[__i] <= __l
#define __S(__i) (__begin_seqs[__i].first)
// Initial partition.
std::vector<std::pair<_Tp, _SeqNumber> > __sample;
for (_SeqNumber __i = 0; __i < __m; __i++)
if (__n < __ns[__i])
__sample.push_back(std::make_pair(__S(__i)[__n], __i));
__gnu_sequential::sort(__sample.begin(), __sample.end(),
__lcomp, sequential_tag());
// Conceptual infinity.
for (_SeqNumber __i = 0; __i < __m; __i++)
if (__n >= __ns[__i])
__sample.push_back(
std::make_pair(__S(__i)[0] /*__dummy element*/, __i));
_DifferenceType __localrank = __rank / __l;
_SeqNumber __j;
for (__j = 0;
__j < __localrank && ((__n + 1) <= __ns[__sample[__j].second]);
++__j)
__a[__sample[__j].second] += __n + 1;
for (; __j < __m; ++__j)
__b[__sample[__j].second] -= __n + 1;
// Further refinement.
while (__n > 0)
{
__n /= 2;
const _Tp* __lmax = 0;
for (_SeqNumber __i = 0; __i < __m; ++__i)
{
if (__a[__i] > 0)
{
if (!__lmax)
__lmax = &(__S(__i)[__a[__i] - 1]);
else
{
if (__comp(*__lmax, __S(__i)[__a[__i] - 1])) //max
__lmax = &(__S(__i)[__a[__i] - 1]);
}
}
}
_SeqNumber __i;
for (__i = 0; __i < __m; __i++)
{
_DifferenceType __middle = (__b[__i] + __a[__i]) / 2;
if (__lmax && __middle < __ns[__i]
&& __comp(__S(__i)[__middle], *__lmax))
__a[__i] = std::min(__a[__i] + __n + 1, __ns[__i]);
else
__b[__i] -= __n + 1;
}
_DifferenceType __leftsize = 0;
for (_SeqNumber __i = 0; __i < __m; ++__i)
__leftsize += __a[__i] / (__n + 1);
_DifferenceType __skew = __rank / (__n + 1) - __leftsize;
if (__skew > 0)
{
// Move to the left, find smallest.
std::priority_queue<std::pair<_Tp, _SeqNumber>,
std::vector<std::pair<_Tp, _SeqNumber> >,
_LexicographicReverse<_Tp, _SeqNumber, _Compare> >
__pq(__lrcomp);
for (_SeqNumber __i = 0; __i < __m; ++__i)
if (__b[__i] < __ns[__i])
__pq.push(std::make_pair(__S(__i)[__b[__i]], __i));
for (; __skew != 0 && !__pq.empty(); --__skew)
{
_SeqNumber __source = __pq.top().second;
__pq.pop();
__a[__source]
= std::min(__a[__source] + __n + 1, __ns[__source]);
__b[__source] += __n + 1;
if (__b[__source] < __ns[__source])
__pq.push(
std::make_pair(__S(__source)[__b[__source]], __source));
}
}
else if (__skew < 0)
{
// Move to the right, find greatest.
std::priority_queue<std::pair<_Tp, _SeqNumber>,
std::vector<std::pair<_Tp, _SeqNumber> >,
_Lexicographic<_Tp, _SeqNumber, _Compare> > __pq(__lcomp);
for (_SeqNumber __i = 0; __i < __m; ++__i)
if (__a[__i] > 0)
__pq.push(std::make_pair(__S(__i)[__a[__i] - 1], __i));
for (; __skew != 0; ++__skew)
{
_SeqNumber __source = __pq.top().second;
__pq.pop();
__a[__source] -= __n + 1;
__b[__source] -= __n + 1;
if (__a[__source] > 0)
__pq.push(std::make_pair(
__S(__source)[__a[__source] - 1], __source));
}
}
}
// Postconditions:
// __a[__i] == __b[__i] in most cases, except when __a[__i] has been
// clamped because of having reached the boundary
// Now return the result, calculate the offset.
// Compare the keys on both edges of the border.
// Maximum of left edge, minimum of right edge.
bool __maxleftset = false, __minrightset = false;
// Impossible to avoid the warning?
_Tp __maxleft, __minright;
for (_SeqNumber __i = 0; __i < __m; ++__i)
{
if (__a[__i] > 0)
{
if (!__maxleftset)
{
__maxleft = __S(__i)[__a[__i] - 1];
__maxleftset = true;
}
else
{
// Max.
if (__comp(__maxleft, __S(__i)[__a[__i] - 1]))
__maxleft = __S(__i)[__a[__i] - 1];
}
}
if (__b[__i] < __ns[__i])
{
if (!__minrightset)
{
__minright = __S(__i)[__b[__i]];
__minrightset = true;
}
else
{
// Min.
if (__comp(__S(__i)[__b[__i]], __minright))
__minright = __S(__i)[__b[__i]];
}
}
}
// Minright is the __splitter, in any case.
if (!__maxleftset || __comp(__minright, __maxleft))
{
// Good luck, everything is split unambiguously.
__offset = 0;
}
else
{
// We have to calculate an offset.
__offset = 0;
for (_SeqNumber __i = 0; __i < __m; ++__i)
{
_DifferenceType lb
= std::lower_bound(__S(__i), __S(__i) + __ns[__i],
__minright,
__comp) - __S(__i);
__offset += __a[__i] - lb;
}
}
delete[] __ns;
delete[] __a;
delete[] __b;
return __minright;
}
}
#undef __S
#endif /* _GLIBCXX_PARALLEL_MULTISEQ_SELECTION_H */
|