1 // -*- C++ -*- 2 3 // Copyright (C) 2007, 2008, 2009, 2010 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/balanced_quicksort.h 26 * @brief Implementation of a dynamically load-balanced parallel quicksort. 27 * 28 * It works in-place and needs only logarithmic extra memory. 29 * The algorithm is similar to the one proposed in 30 * 31 * P. Tsigas and Y. Zhang. 32 * A simple, fast parallel implementation of quicksort and 33 * its performance evaluation on SUN enterprise 10000. 34 * In 11th Euromicro Conference on Parallel, Distributed and 35 * Network-Based Processing, page 372, 2003. 36 * 37 * This file is a GNU parallel extension to the Standard C++ Library. 38 */ 39 40 // Written by Johannes Singler. 41 42 #ifndef _GLIBCXX_PARALLEL_BALANCED_QUICKSORT_H 43 #define _GLIBCXX_PARALLEL_BALANCED_QUICKSORT_H 1 44 45 #include <parallel/basic_iterator.h> 46 #include <bits/stl_algo.h> 47 #include <bits/stl_function.h> 48 49 #include <parallel/settings.h> 50 #include <parallel/partition.h> 51 #include <parallel/random_number.h> 52 #include <parallel/queue.h> 53 54 #if _GLIBCXX_ASSERTIONS 55 #include <parallel/checkers.h> 56 #endif 57 58 namespace __gnu_parallel 59 { 60 /** @brief Information local to one thread in the parallel quicksort run. */ 61 template<typename _RAIter> 62 struct _QSBThreadLocal 63 { 64 typedef std::iterator_traits<_RAIter> _TraitsType; 65 typedef typename _TraitsType::difference_type _DifferenceType; 66 67 /** @brief Continuous part of the sequence, described by an 68 iterator pair. */ 69 typedef std::pair<_RAIter, _RAIter> _Piece; 70 71 /** @brief Initial piece to work on. */ 72 _Piece _M_initial; 73 74 /** @brief Work-stealing queue. */ 75 _RestrictedBoundedConcurrentQueue<_Piece> _M_leftover_parts; 76 77 /** @brief Number of threads involved in this algorithm. */ 78 _ThreadIndex _M_num_threads; 79 80 /** @brief Pointer to a counter of elements left over to sort. */ 81 volatile _DifferenceType* _M_elements_leftover; 82 83 /** @brief The complete sequence to sort. */ 84 _Piece _M_global; 85 86 /** @brief Constructor. 87 * @param __queue_size size of the work-stealing queue. */ 88 _QSBThreadLocal(int __queue_size) : _M_leftover_parts(__queue_size) { } 89 }; 90 91 /** @brief Balanced quicksort divide step. 92 * @param __begin Begin iterator of subsequence. 93 * @param __end End iterator of subsequence. 94 * @param __comp Comparator. 95 * @param __num_threads Number of threads that are allowed to work on 96 * this part. 97 * @pre @c (__end-__begin)>=1 */ 98 template<typename _RAIter, typename _Compare> 99 typename std::iterator_traits<_RAIter>::difference_type 100 __qsb_divide(_RAIter __begin, _RAIter __end, 101 _Compare __comp, _ThreadIndex __num_threads) 102 { 103 _GLIBCXX_PARALLEL_ASSERT(__num_threads > 0); 104 105 typedef std::iterator_traits<_RAIter> _TraitsType; 106 typedef typename _TraitsType::value_type _ValueType; 107 typedef typename _TraitsType::difference_type _DifferenceType; 108 109 _RAIter __pivot_pos = 110 __median_of_three_iterators(__begin, __begin + (__end - __begin) / 2, 111 __end - 1, __comp); 112 113 #if defined(_GLIBCXX_ASSERTIONS) 114 // Must be in between somewhere. 115 _DifferenceType __n = __end - __begin; 116 117 _GLIBCXX_PARALLEL_ASSERT((!__comp(*__pivot_pos, *__begin) 118 && !__comp(*(__begin + __n / 2), 119 *__pivot_pos)) 120 || (!__comp(*__pivot_pos, *__begin) 121 && !__comp(*(__end - 1), *__pivot_pos)) 122 || (!__comp(*__pivot_pos, *(__begin + __n / 2)) 123 && !__comp(*__begin, *__pivot_pos)) 124 || (!__comp(*__pivot_pos, *(__begin + __n / 2)) 125 && !__comp(*(__end - 1), *__pivot_pos)) 126 || (!__comp(*__pivot_pos, *(__end - 1)) 127 && !__comp(*__begin, *__pivot_pos)) 128 || (!__comp(*__pivot_pos, *(__end - 1)) 129 && !__comp(*(__begin + __n / 2), 130 *__pivot_pos))); 131 #endif 132 133 // Swap pivot value to end. 134 if (__pivot_pos != (__end - 1)) 135 std::iter_swap(__pivot_pos, __end - 1); 136 __pivot_pos = __end - 1; 137 138 __gnu_parallel::__binder2nd<_Compare, _ValueType, _ValueType, bool> 139 __pred(__comp, *__pivot_pos); 140 141 // Divide, returning __end - __begin - 1 in the worst case. 142 _DifferenceType __split_pos = __parallel_partition(__begin, __end - 1, 143 __pred, 144 __num_threads); 145 146 // Swap back pivot to middle. 147 std::iter_swap(__begin + __split_pos, __pivot_pos); 148 __pivot_pos = __begin + __split_pos; 149 150 #if _GLIBCXX_ASSERTIONS 151 _RAIter __r; 152 for (__r = __begin; __r != __pivot_pos; ++__r) 153 _GLIBCXX_PARALLEL_ASSERT(__comp(*__r, *__pivot_pos)); 154 for (; __r != __end; ++__r) 155 _GLIBCXX_PARALLEL_ASSERT(!__comp(*__r, *__pivot_pos)); 156 #endif 157 158 return __split_pos; 159 } 160 161 /** @brief Quicksort conquer step. 162 * @param __tls Array of thread-local storages. 163 * @param __begin Begin iterator of subsequence. 164 * @param __end End iterator of subsequence. 165 * @param __comp Comparator. 166 * @param __iam Number of the thread processing this function. 167 * @param __num_threads 168 * Number of threads that are allowed to work on this part. */ 169 template<typename _RAIter, typename _Compare> 170 void 171 __qsb_conquer(_QSBThreadLocal<_RAIter>** __tls, 172 _RAIter __begin, _RAIter __end, 173 _Compare __comp, 174 _ThreadIndex __iam, _ThreadIndex __num_threads, 175 bool __parent_wait) 176 { 177 typedef std::iterator_traits<_RAIter> _TraitsType; 178 typedef typename _TraitsType::value_type _ValueType; 179 typedef typename _TraitsType::difference_type _DifferenceType; 180 181 _DifferenceType __n = __end - __begin; 182 183 if (__num_threads <= 1 || __n <= 1) 184 { 185 __tls[__iam]->_M_initial.first = __begin; 186 __tls[__iam]->_M_initial.second = __end; 187 188 __qsb_local_sort_with_helping(__tls, __comp, __iam, __parent_wait); 189 190 return; 191 } 192 193 // Divide step. 194 _DifferenceType __split_pos = 195 __qsb_divide(__begin, __end, __comp, __num_threads); 196 197 #if _GLIBCXX_ASSERTIONS 198 _GLIBCXX_PARALLEL_ASSERT(0 <= __split_pos && 199 __split_pos < (__end - __begin)); 200 #endif 201 202 _ThreadIndex 203 __num_threads_leftside = std::max<_ThreadIndex> 204 (1, std::min<_ThreadIndex>(__num_threads - 1, __split_pos 205 * __num_threads / __n)); 206 207 # pragma omp atomic 208 *__tls[__iam]->_M_elements_leftover -= (_DifferenceType)1; 209 210 // Conquer step. 211 # pragma omp parallel num_threads(2) 212 { 213 bool __wait; 214 if(omp_get_num_threads() < 2) 215 __wait = false; 216 else 217 __wait = __parent_wait; 218 219 # pragma omp sections 220 { 221 # pragma omp section 222 { 223 __qsb_conquer(__tls, __begin, __begin + __split_pos, __comp, 224 __iam, __num_threads_leftside, __wait); 225 __wait = __parent_wait; 226 } 227 // The pivot_pos is left in place, to ensure termination. 228 # pragma omp section 229 { 230 __qsb_conquer(__tls, __begin + __split_pos + 1, __end, __comp, 231 __iam + __num_threads_leftside, 232 __num_threads - __num_threads_leftside, __wait); 233 __wait = __parent_wait; 234 } 235 } 236 } 237 } 238 239 /** 240 * @brief Quicksort step doing load-balanced local sort. 241 * @param __tls Array of thread-local storages. 242 * @param __comp Comparator. 243 * @param __iam Number of the thread processing this function. 244 */ 245 template<typename _RAIter, typename _Compare> 246 void 247 __qsb_local_sort_with_helping(_QSBThreadLocal<_RAIter>** __tls, 248 _Compare& __comp, _ThreadIndex __iam, 249 bool __wait) 250 { 251 typedef std::iterator_traits<_RAIter> _TraitsType; 252 typedef typename _TraitsType::value_type _ValueType; 253 typedef typename _TraitsType::difference_type _DifferenceType; 254 typedef std::pair<_RAIter, _RAIter> _Piece; 255 256 _QSBThreadLocal<_RAIter>& __tl = *__tls[__iam]; 257 258 _DifferenceType 259 __base_case_n = _Settings::get().sort_qsb_base_case_maximal_n; 260 if (__base_case_n < 2) 261 __base_case_n = 2; 262 _ThreadIndex __num_threads = __tl._M_num_threads; 263 264 // Every thread has its own random number generator. 265 _RandomNumber __rng(__iam + 1); 266 267 _Piece __current = __tl._M_initial; 268 269 _DifferenceType __elements_done = 0; 270 #if _GLIBCXX_ASSERTIONS 271 _DifferenceType __total_elements_done = 0; 272 #endif 273 274 for (;;) 275 { 276 // Invariant: __current must be a valid (maybe empty) range. 277 _RAIter __begin = __current.first, __end = __current.second; 278 _DifferenceType __n = __end - __begin; 279 280 if (__n > __base_case_n) 281 { 282 // Divide. 283 _RAIter __pivot_pos = __begin + __rng(__n); 284 285 // Swap __pivot_pos value to end. 286 if (__pivot_pos != (__end - 1)) 287 std::iter_swap(__pivot_pos, __end - 1); 288 __pivot_pos = __end - 1; 289 290 __gnu_parallel::__binder2nd 291 <_Compare, _ValueType, _ValueType, bool> 292 __pred(__comp, *__pivot_pos); 293 294 // Divide, leave pivot unchanged in last place. 295 _RAIter __split_pos1, __split_pos2; 296 __split_pos1 = __gnu_sequential::partition(__begin, __end - 1, 297 __pred); 298 299 // Left side: < __pivot_pos; __right side: >= __pivot_pos. 300 #if _GLIBCXX_ASSERTIONS 301 _GLIBCXX_PARALLEL_ASSERT(__begin <= __split_pos1 302 && __split_pos1 < __end); 303 #endif 304 // Swap pivot back to middle. 305 if (__split_pos1 != __pivot_pos) 306 std::iter_swap(__split_pos1, __pivot_pos); 307 __pivot_pos = __split_pos1; 308 309 // In case all elements are equal, __split_pos1 == 0. 310 if ((__split_pos1 + 1 - __begin) < (__n >> 7) 311 || (__end - __split_pos1) < (__n >> 7)) 312 { 313 // Very unequal split, one part smaller than one 128th 314 // elements not strictly larger than the pivot. 315 __gnu_parallel::__unary_negate<__gnu_parallel::__binder1st 316 <_Compare, _ValueType, _ValueType, bool>, _ValueType> 317 __pred(__gnu_parallel::__binder1st 318 <_Compare, _ValueType, _ValueType, bool> 319 (__comp, *__pivot_pos)); 320 321 // Find other end of pivot-equal range. 322 __split_pos2 = __gnu_sequential::partition(__split_pos1 + 1, 323 __end, __pred); 324 } 325 else 326 // Only skip the pivot. 327 __split_pos2 = __split_pos1 + 1; 328 329 // Elements equal to pivot are done. 330 __elements_done += (__split_pos2 - __split_pos1); 331 #if _GLIBCXX_ASSERTIONS 332 __total_elements_done += (__split_pos2 - __split_pos1); 333 #endif 334 // Always push larger part onto stack. 335 if (((__split_pos1 + 1) - __begin) < (__end - (__split_pos2))) 336 { 337 // Right side larger. 338 if ((__split_pos2) != __end) 339 __tl._M_leftover_parts.push_front 340 (std::make_pair(__split_pos2, __end)); 341 342 //__current.first = __begin; //already set anyway 343 __current.second = __split_pos1; 344 continue; 345 } 346 else 347 { 348 // Left side larger. 349 if (__begin != __split_pos1) 350 __tl._M_leftover_parts.push_front(std::make_pair 351 (__begin, __split_pos1)); 352 353 __current.first = __split_pos2; 354 //__current.second = __end; //already set anyway 355 continue; 356 } 357 } 358 else 359 { 360 __gnu_sequential::sort(__begin, __end, __comp); 361 __elements_done += __n; 362 #if _GLIBCXX_ASSERTIONS 363 __total_elements_done += __n; 364 #endif 365 366 // Prefer own stack, small pieces. 367 if (__tl._M_leftover_parts.pop_front(__current)) 368 continue; 369 370 # pragma omp atomic 371 *__tl._M_elements_leftover -= __elements_done; 372 373 __elements_done = 0; 374 375 #if _GLIBCXX_ASSERTIONS 376 double __search_start = omp_get_wtime(); 377 #endif 378 379 // Look for new work. 380 bool __successfully_stolen = false; 381 while (__wait && *__tl._M_elements_leftover > 0 382 && !__successfully_stolen 383 #if _GLIBCXX_ASSERTIONS 384 // Possible dead-lock. 385 && (omp_get_wtime() < (__search_start + 1.0)) 386 #endif 387 ) 388 { 389 _ThreadIndex __victim; 390 __victim = __rng(__num_threads); 391 392 // Large pieces. 393 __successfully_stolen = (__victim != __iam) 394 && __tls[__victim]->_M_leftover_parts.pop_back(__current); 395 if (!__successfully_stolen) 396 __yield(); 397 #if !defined(__ICC) && !defined(__ECC) 398 # pragma omp flush 399 #endif 400 } 401 402 #if _GLIBCXX_ASSERTIONS 403 if (omp_get_wtime() >= (__search_start + 1.0)) 404 { 405 sleep(1); 406 _GLIBCXX_PARALLEL_ASSERT(omp_get_wtime() 407 < (__search_start + 1.0)); 408 } 409 #endif 410 if (!__successfully_stolen) 411 { 412 #if _GLIBCXX_ASSERTIONS 413 _GLIBCXX_PARALLEL_ASSERT(*__tl._M_elements_leftover == 0); 414 #endif 415 return; 416 } 417 } 418 } 419 } 420 421 /** @brief Top-level quicksort routine. 422 * @param __begin Begin iterator of sequence. 423 * @param __end End iterator of sequence. 424 * @param __comp Comparator. 425 * @param __num_threads Number of threads that are allowed to work on 426 * this part. 427 */ 428 template<typename _RAIter, typename _Compare> 429 void 430 __parallel_sort_qsb(_RAIter __begin, _RAIter __end, 431 _Compare __comp, _ThreadIndex __num_threads) 432 { 433 _GLIBCXX_CALL(__end - __begin) 434 435 typedef std::iterator_traits<_RAIter> _TraitsType; 436 typedef typename _TraitsType::value_type _ValueType; 437 typedef typename _TraitsType::difference_type _DifferenceType; 438 typedef std::pair<_RAIter, _RAIter> _Piece; 439 440 typedef _QSBThreadLocal<_RAIter> _TLSType; 441 442 _DifferenceType __n = __end - __begin; 443 444 if (__n <= 1) 445 return; 446 447 // At least one element per processor. 448 if (__num_threads > __n) 449 __num_threads = static_cast<_ThreadIndex>(__n); 450 451 // Initialize thread local storage 452 _TLSType** __tls = new _TLSType*[__num_threads]; 453 _DifferenceType __queue_size = (__num_threads 454 * (_ThreadIndex)(__rd_log2(__n) + 1)); 455 for (_ThreadIndex __t = 0; __t < __num_threads; ++__t) 456 __tls[__t] = new _QSBThreadLocal<_RAIter>(__queue_size); 457 458 // There can never be more than ceil(__rd_log2(__n)) ranges on the 459 // stack, because 460 // 1. Only one processor pushes onto the stack 461 // 2. The largest range has at most length __n 462 // 3. Each range is larger than half of the range remaining 463 volatile _DifferenceType __elements_leftover = __n; 464 for (_ThreadIndex __i = 0; __i < __num_threads; ++__i) 465 { 466 __tls[__i]->_M_elements_leftover = &__elements_leftover; 467 __tls[__i]->_M_num_threads = __num_threads; 468 __tls[__i]->_M_global = std::make_pair(__begin, __end); 469 470 // Just in case nothing is left to assign. 471 __tls[__i]->_M_initial = std::make_pair(__end, __end); 472 } 473 474 // Main recursion call. 475 __qsb_conquer(__tls, __begin, __begin + __n, __comp, 0, 476 __num_threads, true); 477 478 #if _GLIBCXX_ASSERTIONS 479 // All stack must be empty. 480 _Piece __dummy; 481 for (_ThreadIndex __i = 1; __i < __num_threads; ++__i) 482 _GLIBCXX_PARALLEL_ASSERT( 483 !__tls[__i]->_M_leftover_parts.pop_back(__dummy)); 484 #endif 485 486 for (_ThreadIndex __i = 0; __i < __num_threads; ++__i) 487 delete __tls[__i]; 488 delete[] __tls; 489 } 490 } // namespace __gnu_parallel 491 492 #endif /* _GLIBCXX_PARALLEL_BALANCED_QUICKSORT_H */ 493