1 // -*- C++ -*- 2 3 // Copyright (C) 2007-2022 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/multiway_mergesort.h 26 * @brief Parallel multiway merge sort. 27 * This file is a GNU parallel extension to the Standard C++ Library. 28 */ 29 30 // Written by Johannes Singler. 31 32 #ifndef _GLIBCXX_PARALLEL_MULTIWAY_MERGESORT_H 33 #define _GLIBCXX_PARALLEL_MULTIWAY_MERGESORT_H 1 34 35 #include <vector> 36 37 #include <parallel/basic_iterator.h> 38 #include <bits/stl_algo.h> 39 #include <parallel/parallel.h> 40 #include <parallel/multiway_merge.h> 41 42 namespace __gnu_parallel 43 { 44 /** @brief Subsequence description. */ 45 template<typename _DifferenceTp> 46 struct _Piece 47 { 48 typedef _DifferenceTp _DifferenceType; 49 50 /** @brief Begin of subsequence. */ 51 _DifferenceType _M_begin; 52 53 /** @brief End of subsequence. */ 54 _DifferenceType _M_end; 55 }; 56 57 /** @brief Data accessed by all threads. 58 * 59 * PMWMS = parallel multiway mergesort */ 60 template<typename _RAIter> 61 struct _PMWMSSortingData 62 { 63 typedef std::iterator_traits<_RAIter> _TraitsType; 64 typedef typename _TraitsType::value_type _ValueType; 65 typedef typename _TraitsType::difference_type _DifferenceType; 66 67 /** @brief Number of threads involved. */ 68 _ThreadIndex _M_num_threads; 69 70 /** @brief Input __begin. */ 71 _RAIter _M_source; 72 73 /** @brief Start indices, per thread. */ 74 _DifferenceType* _M_starts; 75 76 /** @brief Storage in which to sort. */ 77 _ValueType** _M_temporary; 78 79 /** @brief Samples. */ 80 _ValueType* _M_samples; 81 82 /** @brief Offsets to add to the found positions. */ 83 _DifferenceType* _M_offsets; 84 85 /** @brief Pieces of data to merge @c [thread][__sequence] */ 86 std::vector<_Piece<_DifferenceType> >* _M_pieces; 87 }; 88 89 /** 90 * @brief Select _M_samples from a sequence. 91 * @param __sd Pointer to algorithm data. _Result will be placed in 92 * @c __sd->_M_samples. 93 * @param __num_samples Number of _M_samples to select. 94 */ 95 template<typename _RAIter, typename _DifferenceTp> 96 void __determine_samples(_PMWMSSortingData<_RAIter> * __sd,_DifferenceTp __num_samples)97 __determine_samples(_PMWMSSortingData<_RAIter>* __sd, 98 _DifferenceTp __num_samples) 99 { 100 typedef std::iterator_traits<_RAIter> _TraitsType; 101 typedef typename _TraitsType::value_type _ValueType; 102 typedef _DifferenceTp _DifferenceType; 103 104 _ThreadIndex __iam = omp_get_thread_num(); 105 106 _DifferenceType* __es = new _DifferenceType[__num_samples + 2]; 107 108 __equally_split(__sd->_M_starts[__iam + 1] - __sd->_M_starts[__iam], 109 __num_samples + 1, __es); 110 111 for (_DifferenceType __i = 0; __i < __num_samples; ++__i) 112 ::new(&(__sd->_M_samples[__iam * __num_samples + __i])) 113 _ValueType(__sd->_M_source[__sd->_M_starts[__iam] 114 + __es[__i + 1]]); 115 116 delete[] __es; 117 } 118 119 /** @brief Split consistently. */ 120 template<bool __exact, typename _RAIter, 121 typename _Compare, typename _SortingPlacesIterator> 122 struct _SplitConsistently 123 { }; 124 125 /** @brief Split by exact splitting. */ 126 template<typename _RAIter, typename _Compare, 127 typename _SortingPlacesIterator> 128 struct _SplitConsistently<true, _RAIter, _Compare, _SortingPlacesIterator> 129 { 130 void 131 operator()(const _ThreadIndex __iam, 132 _PMWMSSortingData<_RAIter>* __sd, 133 _Compare& __comp, 134 const typename 135 std::iterator_traits<_RAIter>::difference_type 136 __num_samples) const 137 { 138 # pragma omp barrier 139 140 std::vector<std::pair<_SortingPlacesIterator, 141 _SortingPlacesIterator> > 142 __seqs(__sd->_M_num_threads); 143 for (_ThreadIndex __s = 0; __s < __sd->_M_num_threads; __s++) 144 __seqs[__s] = std::make_pair(__sd->_M_temporary[__s], 145 __sd->_M_temporary[__s] 146 + (__sd->_M_starts[__s + 1] 147 - __sd->_M_starts[__s])); 148 149 std::vector<_SortingPlacesIterator> __offsets(__sd->_M_num_threads); 150 151 // if not last thread 152 if (__iam < __sd->_M_num_threads - 1) 153 multiseq_partition(__seqs.begin(), __seqs.end(), 154 __sd->_M_starts[__iam + 1], __offsets.begin(), 155 __comp); 156 157 for (_ThreadIndex __seq = 0; __seq < __sd->_M_num_threads; __seq++) 158 { 159 // for each sequence 160 if (__iam < (__sd->_M_num_threads - 1)) 161 __sd->_M_pieces[__iam][__seq]._M_end 162 = __offsets[__seq] - __seqs[__seq].first; 163 else 164 // very end of this sequence 165 __sd->_M_pieces[__iam][__seq]._M_end = 166 __sd->_M_starts[__seq + 1] - __sd->_M_starts[__seq]; 167 } 168 169 # pragma omp barrier 170 171 for (_ThreadIndex __seq = 0; __seq < __sd->_M_num_threads; __seq++) 172 { 173 // For each sequence. 174 if (__iam > 0) 175 __sd->_M_pieces[__iam][__seq]._M_begin = 176 __sd->_M_pieces[__iam - 1][__seq]._M_end; 177 else 178 // Absolute beginning. 179 __sd->_M_pieces[__iam][__seq]._M_begin = 0; 180 } 181 } 182 }; 183 184 /** @brief Split by sampling. */ 185 template<typename _RAIter, typename _Compare, 186 typename _SortingPlacesIterator> 187 struct _SplitConsistently<false, _RAIter, _Compare, _SortingPlacesIterator> 188 { 189 void 190 operator()(const _ThreadIndex __iam, 191 _PMWMSSortingData<_RAIter>* __sd, 192 _Compare& __comp, 193 const typename 194 std::iterator_traits<_RAIter>::difference_type 195 __num_samples) const 196 { 197 typedef std::iterator_traits<_RAIter> _TraitsType; 198 typedef typename _TraitsType::value_type _ValueType; 199 typedef typename _TraitsType::difference_type _DifferenceType; 200 201 __determine_samples(__sd, __num_samples); 202 203 # pragma omp barrier 204 205 # pragma omp single 206 __gnu_sequential::sort(__sd->_M_samples, 207 __sd->_M_samples 208 + (__num_samples * __sd->_M_num_threads), 209 __comp); 210 211 # pragma omp barrier 212 213 for (_ThreadIndex __s = 0; __s < __sd->_M_num_threads; ++__s) 214 { 215 // For each sequence. 216 if (__num_samples * __iam > 0) 217 __sd->_M_pieces[__iam][__s]._M_begin = 218 std::lower_bound(__sd->_M_temporary[__s], 219 __sd->_M_temporary[__s] 220 + (__sd->_M_starts[__s + 1] 221 - __sd->_M_starts[__s]), 222 __sd->_M_samples[__num_samples * __iam], 223 __comp) 224 - __sd->_M_temporary[__s]; 225 else 226 // Absolute beginning. 227 __sd->_M_pieces[__iam][__s]._M_begin = 0; 228 229 if ((__num_samples * (__iam + 1)) < 230 (__num_samples * __sd->_M_num_threads)) 231 __sd->_M_pieces[__iam][__s]._M_end = 232 std::lower_bound(__sd->_M_temporary[__s], 233 __sd->_M_temporary[__s] 234 + (__sd->_M_starts[__s + 1] 235 - __sd->_M_starts[__s]), 236 __sd->_M_samples[__num_samples * (__iam + 1)], 237 __comp) 238 - __sd->_M_temporary[__s]; 239 else 240 // Absolute end. 241 __sd->_M_pieces[__iam][__s]._M_end = (__sd->_M_starts[__s + 1] 242 - __sd->_M_starts[__s]); 243 } 244 } 245 }; 246 247 template<bool __stable, typename _RAIter, typename _Compare> 248 struct __possibly_stable_sort 249 { }; 250 251 template<typename _RAIter, typename _Compare> 252 struct __possibly_stable_sort<true, _RAIter, _Compare> 253 { 254 void operator()(const _RAIter& __begin, 255 const _RAIter& __end, _Compare& __comp) const 256 { __gnu_sequential::stable_sort(__begin, __end, __comp); } 257 }; 258 259 template<typename _RAIter, typename _Compare> 260 struct __possibly_stable_sort<false, _RAIter, _Compare> 261 { 262 void operator()(const _RAIter __begin, 263 const _RAIter __end, _Compare& __comp) const 264 { __gnu_sequential::sort(__begin, __end, __comp); } 265 }; 266 267 template<bool __stable, typename _Seq_RAIter, 268 typename _RAIter, typename _Compare, 269 typename _DiffType> 270 struct __possibly_stable_multiway_merge 271 { }; 272 273 template<typename _Seq_RAIter, typename _RAIter, 274 typename _Compare, typename _DiffType> 275 struct __possibly_stable_multiway_merge<true, _Seq_RAIter, 276 _RAIter, _Compare, _DiffType> 277 { 278 void operator()(const _Seq_RAIter& __seqs_begin, 279 const _Seq_RAIter& __seqs_end, 280 const _RAIter& __target, 281 _Compare& __comp, 282 _DiffType __length_am) const 283 { stable_multiway_merge(__seqs_begin, __seqs_end, __target, 284 __length_am, __comp, sequential_tag()); } 285 }; 286 287 template<typename _Seq_RAIter, typename _RAIter, 288 typename _Compare, typename _DiffType> 289 struct __possibly_stable_multiway_merge<false, _Seq_RAIter, 290 _RAIter, _Compare, _DiffType> 291 { 292 void operator()(const _Seq_RAIter& __seqs_begin, 293 const _Seq_RAIter& __seqs_end, 294 const _RAIter& __target, 295 _Compare& __comp, 296 _DiffType __length_am) const 297 { multiway_merge(__seqs_begin, __seqs_end, __target, __length_am, 298 __comp, sequential_tag()); } 299 }; 300 301 /** @brief PMWMS code executed by each thread. 302 * @param __sd Pointer to algorithm data. 303 * @param __comp Comparator. 304 */ 305 template<bool __stable, bool __exact, typename _RAIter, 306 typename _Compare> 307 void 308 parallel_sort_mwms_pu(_PMWMSSortingData<_RAIter>* __sd, 309 _Compare& __comp) 310 { 311 typedef std::iterator_traits<_RAIter> _TraitsType; 312 typedef typename _TraitsType::value_type _ValueType; 313 typedef typename _TraitsType::difference_type _DifferenceType; 314 315 _ThreadIndex __iam = omp_get_thread_num(); 316 317 // Length of this thread's chunk, before merging. 318 _DifferenceType __length_local = 319 __sd->_M_starts[__iam + 1] - __sd->_M_starts[__iam]; 320 321 // Sort in temporary storage, leave space for sentinel. 322 323 typedef _ValueType* _SortingPlacesIterator; 324 325 __sd->_M_temporary[__iam] = 326 static_cast<_ValueType*>(::operator new(sizeof(_ValueType) 327 * (__length_local + 1))); 328 329 // Copy there. 330 std::uninitialized_copy(__sd->_M_source + __sd->_M_starts[__iam], 331 __sd->_M_source + __sd->_M_starts[__iam] 332 + __length_local, 333 __sd->_M_temporary[__iam]); 334 335 __possibly_stable_sort<__stable, _SortingPlacesIterator, _Compare>() 336 (__sd->_M_temporary[__iam], 337 __sd->_M_temporary[__iam] + __length_local, 338 __comp); 339 340 // Invariant: locally sorted subsequence in sd->_M_temporary[__iam], 341 // __sd->_M_temporary[__iam] + __length_local. 342 343 // No barrier here: Synchronization is done by the splitting routine. 344 345 _DifferenceType __num_samples = 346 _Settings::get().sort_mwms_oversampling * __sd->_M_num_threads - 1; 347 _SplitConsistently<__exact, _RAIter, _Compare, _SortingPlacesIterator>() 348 (__iam, __sd, __comp, __num_samples); 349 350 // Offset from __target __begin, __length after merging. 351 _DifferenceType __offset = 0, __length_am = 0; 352 for (_ThreadIndex __s = 0; __s < __sd->_M_num_threads; __s++) 353 { 354 __length_am += (__sd->_M_pieces[__iam][__s]._M_end 355 - __sd->_M_pieces[__iam][__s]._M_begin); 356 __offset += __sd->_M_pieces[__iam][__s]._M_begin; 357 } 358 359 typedef std::vector< 360 std::pair<_SortingPlacesIterator, _SortingPlacesIterator> > 361 _SeqVector; 362 _SeqVector __seqs(__sd->_M_num_threads); 363 364 for (_ThreadIndex __s = 0; __s < __sd->_M_num_threads; ++__s) 365 { 366 __seqs[__s] = 367 std::make_pair(__sd->_M_temporary[__s] 368 + __sd->_M_pieces[__iam][__s]._M_begin, 369 __sd->_M_temporary[__s] 370 + __sd->_M_pieces[__iam][__s]._M_end); 371 } 372 373 __possibly_stable_multiway_merge< 374 __stable, typename _SeqVector::iterator, 375 _RAIter, _Compare, _DifferenceType>()(__seqs.begin(), __seqs.end(), 376 __sd->_M_source + __offset, __comp, 377 __length_am); 378 379 # pragma omp barrier 380 381 for (_DifferenceType __i = 0; __i < __length_local; ++__i) 382 __sd->_M_temporary[__iam][__i].~_ValueType(); 383 ::operator delete(__sd->_M_temporary[__iam]); 384 } 385 386 /** @brief PMWMS main call. 387 * @param __begin Begin iterator of sequence. 388 * @param __end End iterator of sequence. 389 * @param __comp Comparator. 390 * @param __num_threads Number of threads to use. 391 */ 392 template<bool __stable, bool __exact, typename _RAIter, 393 typename _Compare> 394 void 395 parallel_sort_mwms(_RAIter __begin, _RAIter __end, 396 _Compare __comp, 397 _ThreadIndex __num_threads) 398 { 399 _GLIBCXX_CALL(__end - __begin) 400 401 typedef std::iterator_traits<_RAIter> _TraitsType; 402 typedef typename _TraitsType::value_type _ValueType; 403 typedef typename _TraitsType::difference_type _DifferenceType; 404 405 _DifferenceType __n = __end - __begin; 406 407 if (__n <= 1) 408 return; 409 410 // at least one element per thread 411 if (__num_threads > __n) 412 __num_threads = static_cast<_ThreadIndex>(__n); 413 414 // shared variables 415 _PMWMSSortingData<_RAIter> __sd; 416 _DifferenceType* __starts; 417 _DifferenceType __size; 418 419 # pragma omp parallel num_threads(__num_threads) 420 { 421 __num_threads = omp_get_num_threads(); //no more threads than requested 422 423 # pragma omp single 424 { 425 __sd._M_num_threads = __num_threads; 426 __sd._M_source = __begin; 427 428 __sd._M_temporary = new _ValueType*[__num_threads]; 429 430 if (!__exact) 431 { 432 __size = 433 (_Settings::get().sort_mwms_oversampling * __num_threads - 1) 434 * __num_threads; 435 __sd._M_samples = static_cast<_ValueType*> 436 (::operator new(__size * sizeof(_ValueType))); 437 } 438 else 439 __sd._M_samples = 0; 440 441 __sd._M_offsets = new _DifferenceType[__num_threads - 1]; 442 __sd._M_pieces 443 = new std::vector<_Piece<_DifferenceType> >[__num_threads]; 444 for (_ThreadIndex __s = 0; __s < __num_threads; ++__s) 445 __sd._M_pieces[__s].resize(__num_threads); 446 __starts = __sd._M_starts = new _DifferenceType[__num_threads + 1]; 447 448 _DifferenceType __chunk_length = __n / __num_threads; 449 _DifferenceType __split = __n % __num_threads; 450 _DifferenceType __pos = 0; 451 for (_ThreadIndex __i = 0; __i < __num_threads; ++__i) 452 { 453 __starts[__i] = __pos; 454 __pos += ((__i < __split) 455 ? (__chunk_length + 1) : __chunk_length); 456 } 457 __starts[__num_threads] = __pos; 458 } //single 459 460 // Now sort in parallel. 461 parallel_sort_mwms_pu<__stable, __exact>(&__sd, __comp); 462 } //parallel 463 464 delete[] __starts; 465 delete[] __sd._M_temporary; 466 467 if (!__exact) 468 { 469 for (_DifferenceType __i = 0; __i < __size; ++__i) 470 __sd._M_samples[__i].~_ValueType(); 471 ::operator delete(__sd._M_samples); 472 } 473 474 delete[] __sd._M_offsets; 475 delete[] __sd._M_pieces; 476 } 477 478 } //namespace __gnu_parallel 479 480 #endif /* _GLIBCXX_PARALLEL_MULTIWAY_MERGESORT_H */ 481