1 /* 2 * Copyright (c) 2001, 2012, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef SHARE_VM_UTILITIES_TASKQUEUE_HPP 26 #define SHARE_VM_UTILITIES_TASKQUEUE_HPP 27 28 #include "memory/allocation.hpp" 29 #include "memory/allocation.inline.hpp" 30 #include "runtime/mutex.hpp" 31 #include "utilities/stack.hpp" 32 #ifdef TARGET_OS_ARCH_linux_x86 33 # include "orderAccess_linux_x86.inline.hpp" 34 #endif 35 #ifdef TARGET_OS_ARCH_linux_sparc 36 # include "orderAccess_linux_sparc.inline.hpp" 37 #endif 38 #ifdef TARGET_OS_ARCH_linux_zero 39 # include "orderAccess_linux_zero.inline.hpp" 40 #endif 41 #ifdef TARGET_OS_ARCH_solaris_x86 42 # include "orderAccess_solaris_x86.inline.hpp" 43 #endif 44 #ifdef TARGET_OS_ARCH_solaris_sparc 45 # include "orderAccess_solaris_sparc.inline.hpp" 46 #endif 47 #ifdef TARGET_OS_ARCH_windows_x86 48 # include "orderAccess_windows_x86.inline.hpp" 49 #endif 50 #ifdef TARGET_OS_ARCH_linux_arm 51 # include "orderAccess_linux_arm.inline.hpp" 52 #endif 53 #ifdef TARGET_OS_ARCH_linux_ppc 54 # include "orderAccess_linux_ppc.inline.hpp" 55 #endif 56 #ifdef TARGET_OS_ARCH_bsd_x86 57 # include "orderAccess_bsd_x86.inline.hpp" 58 #endif 59 #ifdef TARGET_OS_ARCH_bsd_zero 60 # include "orderAccess_bsd_zero.inline.hpp" 61 #endif 62 63 // Simple TaskQueue stats that are collected by default in debug builds. 64 65 #if !defined(TASKQUEUE_STATS) && defined(ASSERT) 66 #define TASKQUEUE_STATS 1 67 #elif !defined(TASKQUEUE_STATS) 68 #define TASKQUEUE_STATS 0 69 #endif 70 71 #if TASKQUEUE_STATS 72 #define TASKQUEUE_STATS_ONLY(code) code 73 #else 74 #define TASKQUEUE_STATS_ONLY(code) 75 #endif // TASKQUEUE_STATS 76 77 #if TASKQUEUE_STATS 78 class TaskQueueStats { 79 public: 80 enum StatId { 81 push, // number of taskqueue pushes 82 pop, // number of taskqueue pops 83 pop_slow, // subset of taskqueue pops that were done slow-path 84 steal_attempt, // number of taskqueue steal attempts 85 steal, // number of taskqueue steals 86 overflow, // number of overflow pushes 87 overflow_max_len, // max length of overflow stack 88 last_stat_id 89 }; 90 91 public: 92 inline TaskQueueStats() { reset(); } 93 94 inline void record_push() { ++_stats[push]; } 95 inline void record_pop() { ++_stats[pop]; } 96 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; } 97 inline void record_steal(bool success); 98 inline void record_overflow(size_t new_length); 99 100 TaskQueueStats & operator +=(const TaskQueueStats & addend); 101 102 inline size_t get(StatId id) const { return _stats[id]; } 103 inline const size_t* get() const { return _stats; } 104 105 inline void reset(); 106 107 // Print the specified line of the header (does not include a line separator). 108 static void print_header(unsigned int line, outputStream* const stream = tty, 109 unsigned int width = 10); 110 // Print the statistics (does not include a line separator). 111 void print(outputStream* const stream = tty, unsigned int width = 10) const; 112 113 DEBUG_ONLY(void verify() const;) 114 115 private: 116 size_t _stats[last_stat_id]; 117 static const char * const _names[last_stat_id]; 118 }; 119 120 void TaskQueueStats::record_steal(bool success) { 121 ++_stats[steal_attempt]; 122 if (success) ++_stats[steal]; 123 } 124 125 void TaskQueueStats::record_overflow(size_t new_len) { 126 ++_stats[overflow]; 127 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len; 128 } 129 130 void TaskQueueStats::reset() { 131 memset(_stats, 0, sizeof(_stats)); 132 } 133 #endif // TASKQUEUE_STATS 134 135 template <unsigned int N, MEMFLAGS F> 136 class TaskQueueSuper: public CHeapObj<F> { 137 protected: 138 // Internal type for indexing the queue; also used for the tag. 139 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t; 140 141 // The first free element after the last one pushed (mod N). 142 volatile uint _bottom; 143 144 enum { MOD_N_MASK = N - 1 }; 145 146 class Age { 147 public: 148 Age(size_t data = 0) { _data = data; } 149 Age(const Age& age) { _data = age._data; } 150 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; } 151 152 Age get() const volatile { return _data; } 153 void set(Age age) volatile { _data = age._data; } 154 155 idx_t top() const volatile { return _fields._top; } 156 idx_t tag() const volatile { return _fields._tag; } 157 158 // Increment top; if it wraps, increment tag also. 159 void increment() { 160 _fields._top = increment_index(_fields._top); 161 if (_fields._top == 0) ++_fields._tag; 162 } 163 164 Age cmpxchg(const Age new_age, const Age old_age) volatile { 165 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data, 166 (volatile intptr_t *)&_data, 167 (intptr_t)old_age._data); 168 } 169 170 bool operator ==(const Age& other) const { return _data == other._data; } 171 172 private: 173 struct fields { 174 idx_t _top; 175 idx_t _tag; 176 }; 177 union { 178 size_t _data; 179 fields _fields; 180 }; 181 }; 182 183 volatile Age _age; 184 185 // These both operate mod N. 186 static uint increment_index(uint ind) { 187 return (ind + 1) & MOD_N_MASK; 188 } 189 static uint decrement_index(uint ind) { 190 return (ind - 1) & MOD_N_MASK; 191 } 192 193 // Returns a number in the range [0..N). If the result is "N-1", it should be 194 // interpreted as 0. 195 uint dirty_size(uint bot, uint top) const { 196 return (bot - top) & MOD_N_MASK; 197 } 198 199 // Returns the size corresponding to the given "bot" and "top". 200 uint size(uint bot, uint top) const { 201 uint sz = dirty_size(bot, top); 202 // Has the queue "wrapped", so that bottom is less than top? There's a 203 // complicated special case here. A pair of threads could perform pop_local 204 // and pop_global operations concurrently, starting from a state in which 205 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom, 206 // and the pop_global in incrementing _top (in which case the pop_global 207 // will be awarded the contested queue element.) The resulting state must 208 // be interpreted as an empty queue. (We only need to worry about one such 209 // event: only the queue owner performs pop_local's, and several concurrent 210 // threads attempting to perform the pop_global will all perform the same 211 // CAS, and only one can succeed.) Any stealing thread that reads after 212 // either the increment or decrement will see an empty queue, and will not 213 // join the competitors. The "sz == -1 || sz == N-1" state will not be 214 // modified by concurrent queues, so the owner thread can reset the state to 215 // _bottom == top so subsequent pushes will be performed normally. 216 return (sz == N - 1) ? 0 : sz; 217 } 218 219 public: 220 TaskQueueSuper() : _bottom(0), _age() {} 221 222 // Return true if the TaskQueue contains/does not contain any tasks. 223 bool peek() const { return _bottom != _age.top(); } 224 bool is_empty() const { return size() == 0; } 225 226 // Return an estimate of the number of elements in the queue. 227 // The "careful" version admits the possibility of pop_local/pop_global 228 // races. 229 uint size() const { 230 return size(_bottom, _age.top()); 231 } 232 233 uint dirty_size() const { 234 return dirty_size(_bottom, _age.top()); 235 } 236 237 void set_empty() { 238 _bottom = 0; 239 _age.set(0); 240 } 241 242 // Maximum number of elements allowed in the queue. This is two less 243 // than the actual queue size, for somewhat complicated reasons. 244 uint max_elems() const { return N - 2; } 245 246 // Total size of queue. 247 static const uint total_size() { return N; } 248 249 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;) 250 }; 251 252 253 254 template <class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> 255 class GenericTaskQueue: public TaskQueueSuper<N, F> { 256 protected: 257 typedef typename TaskQueueSuper<N, F>::Age Age; 258 typedef typename TaskQueueSuper<N, F>::idx_t idx_t; 259 260 using TaskQueueSuper<N, F>::_bottom; 261 using TaskQueueSuper<N, F>::_age; 262 using TaskQueueSuper<N, F>::increment_index; 263 using TaskQueueSuper<N, F>::decrement_index; 264 using TaskQueueSuper<N, F>::dirty_size; 265 266 public: 267 using TaskQueueSuper<N, F>::max_elems; 268 using TaskQueueSuper<N, F>::size; 269 270 #if TASKQUEUE_STATS 271 using TaskQueueSuper<N, F>::stats; 272 #endif 273 274 private: 275 // Slow paths for push, pop_local. (pop_global has no fast path.) 276 bool push_slow(E t, uint dirty_n_elems); 277 bool pop_local_slow(uint localBot, Age oldAge); 278 279 public: 280 typedef E element_type; 281 282 // Initializes the queue to empty. 283 GenericTaskQueue(); 284 285 void initialize(); 286 287 // Push the task "t" on the queue. Returns "false" iff the queue is full. 288 inline bool push(E t); 289 290 // Attempts to claim a task from the "local" end of the queue (the most 291 // recently pushed). If successful, returns true and sets t to the task; 292 // otherwise, returns false (the queue is empty). 293 inline bool pop_local(E& t); 294 295 // Like pop_local(), but uses the "global" end of the queue (the least 296 // recently pushed). 297 bool pop_global(E& t); 298 299 // Delete any resource associated with the queue. 300 ~GenericTaskQueue(); 301 302 // apply the closure to all elements in the task queue 303 void oops_do(OopClosure* f); 304 305 private: 306 // Element array. 307 volatile E* _elems; 308 }; 309 310 template<class E, MEMFLAGS F, unsigned int N> 311 GenericTaskQueue<E, F, N>::GenericTaskQueue() { 312 assert(sizeof(Age) == sizeof(size_t), "Depends on this."); 313 } 314 315 template<class E, MEMFLAGS F, unsigned int N> 316 void GenericTaskQueue<E, F, N>::initialize() { 317 _elems = NEW_C_HEAP_ARRAY(E, N, F); 318 } 319 320 template<class E, MEMFLAGS F, unsigned int N> 321 void GenericTaskQueue<E, F, N>::oops_do(OopClosure* f) { 322 // tty->print_cr("START OopTaskQueue::oops_do"); 323 uint iters = size(); 324 uint index = _bottom; 325 for (uint i = 0; i < iters; ++i) { 326 index = decrement_index(index); 327 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, 328 // index, &_elems[index], _elems[index]); 329 E* t = (E*)&_elems[index]; // cast away volatility 330 oop* p = (oop*)t; 331 assert((*t)->is_oop_or_null(), "Not an oop or null"); 332 f->do_oop(p); 333 } 334 // tty->print_cr("END OopTaskQueue::oops_do"); 335 } 336 337 template<class E, MEMFLAGS F, unsigned int N> 338 bool GenericTaskQueue<E, F, N>::push_slow(E t, uint dirty_n_elems) { 339 if (dirty_n_elems == N - 1) { 340 // Actually means 0, so do the push. 341 uint localBot = _bottom; 342 // g++ complains if the volatile result of the assignment is unused. 343 const_cast<E&>(_elems[localBot] = t); 344 OrderAccess::release_store(&_bottom, increment_index(localBot)); 345 TASKQUEUE_STATS_ONLY(stats.record_push()); 346 return true; 347 } 348 return false; 349 } 350 351 // pop_local_slow() is done by the owning thread and is trying to 352 // get the last task in the queue. It will compete with pop_global() 353 // that will be used by other threads. The tag age is incremented 354 // whenever the queue goes empty which it will do here if this thread 355 // gets the last task or in pop_global() if the queue wraps (top == 0 356 // and pop_global() succeeds, see pop_global()). 357 template<class E, MEMFLAGS F, unsigned int N> 358 bool GenericTaskQueue<E, F, N>::pop_local_slow(uint localBot, Age oldAge) { 359 // This queue was observed to contain exactly one element; either this 360 // thread will claim it, or a competing "pop_global". In either case, 361 // the queue will be logically empty afterwards. Create a new Age value 362 // that represents the empty queue for the given value of "_bottom". (We 363 // must also increment "tag" because of the case where "bottom == 1", 364 // "top == 0". A pop_global could read the queue element in that case, 365 // then have the owner thread do a pop followed by another push. Without 366 // the incrementing of "tag", the pop_global's CAS could succeed, 367 // allowing it to believe it has claimed the stale element.) 368 Age newAge((idx_t)localBot, oldAge.tag() + 1); 369 // Perhaps a competing pop_global has already incremented "top", in which 370 // case it wins the element. 371 if (localBot == oldAge.top()) { 372 // No competing pop_global has yet incremented "top"; we'll try to 373 // install new_age, thus claiming the element. 374 Age tempAge = _age.cmpxchg(newAge, oldAge); 375 if (tempAge == oldAge) { 376 // We win. 377 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); 378 TASKQUEUE_STATS_ONLY(stats.record_pop_slow()); 379 return true; 380 } 381 } 382 // We lose; a completing pop_global gets the element. But the queue is empty 383 // and top is greater than bottom. Fix this representation of the empty queue 384 // to become the canonical one. 385 _age.set(newAge); 386 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); 387 return false; 388 } 389 390 template<class E, MEMFLAGS F, unsigned int N> 391 bool GenericTaskQueue<E, F, N>::pop_global(E& t) { 392 Age oldAge = _age.get(); 393 // Architectures with weak memory model require a fence here. The 394 // fence has a cumulative effect on getting age and getting bottom. 395 // This way it is guaranteed that bottom is not older than age, 396 // what is crucial for the correctness of the algorithm. 397 #if (defined ARM || defined IA64 || defined PPC64) 398 OrderAccess::fence(); 399 #endif 400 uint localBot = OrderAccess::load_acquire((volatile juint*)&_bottom); 401 uint n_elems = size(localBot, oldAge.top()); 402 if (n_elems == 0) { 403 return false; 404 } 405 406 const_cast<E&>(t = _elems[oldAge.top()]); 407 Age newAge(oldAge); 408 newAge.increment(); 409 Age resAge = _age.cmpxchg(newAge, oldAge); 410 411 // Note that using "_bottom" here might fail, since a pop_local might 412 // have decremented it. 413 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity"); 414 return resAge == oldAge; 415 } 416 417 template<class E, MEMFLAGS F, unsigned int N> 418 GenericTaskQueue<E, F, N>::~GenericTaskQueue() { 419 FREE_C_HEAP_ARRAY(E, _elems, F); 420 } 421 422 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for 423 // elements that do not fit in the TaskQueue. 424 // 425 // This class hides two methods from super classes: 426 // 427 // push() - push onto the task queue or, if that fails, onto the overflow stack 428 // is_empty() - return true if both the TaskQueue and overflow stack are empty 429 // 430 // Note that size() is not hidden--it returns the number of elements in the 431 // TaskQueue, and does not include the size of the overflow stack. This 432 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues. 433 template<class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> 434 class OverflowTaskQueue: public GenericTaskQueue<E, F, N> 435 { 436 public: 437 typedef Stack<E, F> overflow_t; 438 typedef GenericTaskQueue<E, F, N> taskqueue_t; 439 440 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;) 441 442 // Push task t onto the queue or onto the overflow stack. Return true. 443 inline bool push(E t); 444 445 // Attempt to pop from the overflow stack; return true if anything was popped. 446 inline bool pop_overflow(E& t); 447 448 inline overflow_t* overflow_stack() { return &_overflow_stack; } 449 450 inline bool taskqueue_empty() const { return taskqueue_t::is_empty(); } 451 inline bool overflow_empty() const { return _overflow_stack.is_empty(); } 452 inline bool is_empty() const { 453 return taskqueue_empty() && overflow_empty(); 454 } 455 456 private: 457 overflow_t _overflow_stack; 458 }; 459 460 template <class E, MEMFLAGS F, unsigned int N> 461 bool OverflowTaskQueue<E, F, N>::push(E t) 462 { 463 if (!taskqueue_t::push(t)) { 464 overflow_stack()->push(t); 465 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size())); 466 } 467 return true; 468 } 469 470 template <class E, MEMFLAGS F, unsigned int N> 471 bool OverflowTaskQueue<E, F, N>::pop_overflow(E& t) 472 { 473 if (overflow_empty()) return false; 474 t = overflow_stack()->pop(); 475 return true; 476 } 477 478 class TaskQueueSetSuper { 479 protected: 480 static int randomParkAndMiller(int* seed0); 481 public: 482 // Returns "true" if some TaskQueue in the set contains a task. 483 virtual bool peek() = 0; 484 }; 485 486 template <MEMFLAGS F> class TaskQueueSetSuperImpl: public CHeapObj<F>, public TaskQueueSetSuper { 487 }; 488 489 template<class T, MEMFLAGS F> 490 class GenericTaskQueueSet: public TaskQueueSetSuperImpl<F> { 491 private: 492 uint _n; 493 T** _queues; 494 495 public: 496 typedef typename T::element_type E; 497 498 GenericTaskQueueSet(int n) : _n(n) { 499 typedef T* GenericTaskQueuePtr; 500 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n, F); 501 for (int i = 0; i < n; i++) { 502 _queues[i] = NULL; 503 } 504 } 505 506 bool steal_best_of_2(uint queue_num, int* seed, E& t); 507 508 void register_queue(uint i, T* q); 509 510 T* queue(uint n); 511 512 // The thread with queue number "queue_num" (and whose random number seed is 513 // at "seed") is trying to steal a task from some other queue. (It may try 514 // several queues, according to some configuration parameter.) If some steal 515 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns 516 // false. 517 bool steal(uint queue_num, int* seed, E& t); 518 519 bool peek(); 520 }; 521 522 template<class T, MEMFLAGS F> void 523 GenericTaskQueueSet<T, F>::register_queue(uint i, T* q) { 524 assert(i < _n, "index out of range."); 525 _queues[i] = q; 526 } 527 528 template<class T, MEMFLAGS F> T* 529 GenericTaskQueueSet<T, F>::queue(uint i) { 530 return _queues[i]; 531 } 532 533 template<class T, MEMFLAGS F> bool 534 GenericTaskQueueSet<T, F>::steal(uint queue_num, int* seed, E& t) { 535 for (uint i = 0; i < 2 * _n; i++) { 536 if (steal_best_of_2(queue_num, seed, t)) { 537 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true)); 538 return true; 539 } 540 } 541 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false)); 542 return false; 543 } 544 545 template<class T, MEMFLAGS F> bool 546 GenericTaskQueueSet<T, F>::steal_best_of_2(uint queue_num, int* seed, E& t) { 547 if (_n > 2) { 548 uint k1 = queue_num; 549 while (k1 == queue_num) k1 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n; 550 uint k2 = queue_num; 551 while (k2 == queue_num || k2 == k1) k2 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n; 552 // Sample both and try the larger. 553 uint sz1 = _queues[k1]->size(); 554 uint sz2 = _queues[k2]->size(); 555 if (sz2 > sz1) return _queues[k2]->pop_global(t); 556 else return _queues[k1]->pop_global(t); 557 } else if (_n == 2) { 558 // Just try the other one. 559 uint k = (queue_num + 1) % 2; 560 return _queues[k]->pop_global(t); 561 } else { 562 assert(_n == 1, "can't be zero."); 563 return false; 564 } 565 } 566 567 template<class T, MEMFLAGS F> 568 bool GenericTaskQueueSet<T, F>::peek() { 569 // Try all the queues. 570 for (uint j = 0; j < _n; j++) { 571 if (_queues[j]->peek()) 572 return true; 573 } 574 return false; 575 } 576 577 // When to terminate from the termination protocol. 578 class TerminatorTerminator: public CHeapObj<mtInternal> { 579 public: 580 virtual bool should_exit_termination() = 0; 581 }; 582 583 // A class to aid in the termination of a set of parallel tasks using 584 // TaskQueueSet's for work stealing. 585 586 #undef TRACESPINNING 587 588 class ParallelTaskTerminator: public StackObj { 589 private: 590 int _n_threads; 591 TaskQueueSetSuper* _queue_set; 592 int _offered_termination; 593 594 #ifdef TRACESPINNING 595 static uint _total_yields; 596 static uint _total_spins; 597 static uint _total_peeks; 598 #endif 599 600 bool peek_in_queue_set(); 601 protected: 602 virtual void yield(); 603 void sleep(uint millis); 604 605 public: 606 607 // "n_threads" is the number of threads to be terminated. "queue_set" is a 608 // queue sets of work queues of other threads. 609 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set); 610 611 // The current thread has no work, and is ready to terminate if everyone 612 // else is. If returns "true", all threads are terminated. If returns 613 // "false", available work has been observed in one of the task queues, 614 // so the global task is not complete. 615 bool offer_termination() { 616 return offer_termination(NULL); 617 } 618 619 // As above, but it also terminates if the should_exit_termination() 620 // method of the terminator parameter returns true. If terminator is 621 // NULL, then it is ignored. 622 bool offer_termination(TerminatorTerminator* terminator); 623 624 // Reset the terminator, so that it may be reused again. 625 // The caller is responsible for ensuring that this is done 626 // in an MT-safe manner, once the previous round of use of 627 // the terminator is finished. 628 void reset_for_reuse(); 629 // Same as above but the number of parallel threads is set to the 630 // given number. 631 void reset_for_reuse(int n_threads); 632 633 #ifdef TRACESPINNING 634 static uint total_yields() { return _total_yields; } 635 static uint total_spins() { return _total_spins; } 636 static uint total_peeks() { return _total_peeks; } 637 static void print_termination_counts(); 638 #endif 639 }; 640 641 template<class E, MEMFLAGS F, unsigned int N> inline bool 642 GenericTaskQueue<E, F, N>::push(E t) { 643 uint localBot = _bottom; 644 assert(localBot < N, "_bottom out of range."); 645 idx_t top = _age.top(); 646 uint dirty_n_elems = dirty_size(localBot, top); 647 assert(dirty_n_elems < N, "n_elems out of range."); 648 if (dirty_n_elems < max_elems()) { 649 // g++ complains if the volatile result of the assignment is unused. 650 const_cast<E&>(_elems[localBot] = t); 651 OrderAccess::release_store(&_bottom, increment_index(localBot)); 652 TASKQUEUE_STATS_ONLY(stats.record_push()); 653 return true; 654 } else { 655 return push_slow(t, dirty_n_elems); 656 } 657 } 658 659 template<class E, MEMFLAGS F, unsigned int N> inline bool 660 GenericTaskQueue<E, F, N>::pop_local(E& t) { 661 uint localBot = _bottom; 662 // This value cannot be N-1. That can only occur as a result of 663 // the assignment to bottom in this method. If it does, this method 664 // resets the size to 0 before the next call (which is sequential, 665 // since this is pop_local.) 666 uint dirty_n_elems = dirty_size(localBot, _age.top()); 667 assert(dirty_n_elems != N - 1, "Shouldn't be possible..."); 668 if (dirty_n_elems == 0) return false; 669 localBot = decrement_index(localBot); 670 _bottom = localBot; 671 // This is necessary to prevent any read below from being reordered 672 // before the store just above. 673 OrderAccess::fence(); 674 const_cast<E&>(t = _elems[localBot]); 675 // This is a second read of "age"; the "size()" above is the first. 676 // If there's still at least one element in the queue, based on the 677 // "_bottom" and "age" we've read, then there can be no interference with 678 // a "pop_global" operation, and we're done. 679 idx_t tp = _age.top(); // XXX 680 if (size(localBot, tp) > 0) { 681 assert(dirty_size(localBot, tp) != N - 1, "sanity"); 682 TASKQUEUE_STATS_ONLY(stats.record_pop()); 683 return true; 684 } else { 685 // Otherwise, the queue contained exactly one element; we take the slow 686 // path. 687 return pop_local_slow(localBot, _age.get()); 688 } 689 } 690 691 typedef GenericTaskQueue<oop, mtGC> OopTaskQueue; 692 typedef GenericTaskQueueSet<OopTaskQueue, mtGC> OopTaskQueueSet; 693 694 #ifdef _MSC_VER 695 #pragma warning(push) 696 // warning C4522: multiple assignment operators specified 697 #pragma warning(disable:4522) 698 #endif 699 700 // This is a container class for either an oop* or a narrowOop*. 701 // Both are pushed onto a task queue and the consumer will test is_narrow() 702 // to determine which should be processed. 703 class StarTask { 704 void* _holder; // either union oop* or narrowOop* 705 706 enum { COMPRESSED_OOP_MASK = 1 }; 707 708 public: 709 StarTask(narrowOop* p) { 710 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!"); 711 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK); 712 } 713 StarTask(oop* p) { 714 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!"); 715 _holder = (void*)p; 716 } 717 StarTask() { _holder = NULL; } 718 operator oop*() { return (oop*)_holder; } 719 operator narrowOop*() { 720 return (narrowOop*)((uintptr_t)_holder & ~COMPRESSED_OOP_MASK); 721 } 722 723 StarTask& operator=(const StarTask& t) { 724 _holder = t._holder; 725 return *this; 726 } 727 volatile StarTask& operator=(const volatile StarTask& t) volatile { 728 _holder = t._holder; 729 return *this; 730 } 731 732 bool is_narrow() const { 733 return (((uintptr_t)_holder & COMPRESSED_OOP_MASK) != 0); 734 } 735 }; 736 737 class ObjArrayTask 738 { 739 public: 740 ObjArrayTask(oop o = NULL, int idx = 0): _obj(o), _index(idx) { } 741 ObjArrayTask(oop o, size_t idx): _obj(o), _index(int(idx)) { 742 assert(idx <= size_t(max_jint), "too big"); 743 } 744 ObjArrayTask(const ObjArrayTask& t): _obj(t._obj), _index(t._index) { } 745 746 ObjArrayTask& operator =(const ObjArrayTask& t) { 747 _obj = t._obj; 748 _index = t._index; 749 return *this; 750 } 751 volatile ObjArrayTask& 752 operator =(const volatile ObjArrayTask& t) volatile { 753 _obj = t._obj; 754 _index = t._index; 755 return *this; 756 } 757 758 inline oop obj() const { return _obj; } 759 inline int index() const { return _index; } 760 761 DEBUG_ONLY(bool is_valid() const); // Tasks to be pushed/popped must be valid. 762 763 private: 764 oop _obj; 765 int _index; 766 }; 767 768 #ifdef _MSC_VER 769 #pragma warning(pop) 770 #endif 771 772 typedef OverflowTaskQueue<StarTask, mtClass> OopStarTaskQueue; 773 typedef GenericTaskQueueSet<OopStarTaskQueue, mtClass> OopStarTaskQueueSet; 774 775 typedef OverflowTaskQueue<size_t, mtInternal> RegionTaskQueue; 776 typedef GenericTaskQueueSet<RegionTaskQueue, mtClass> RegionTaskQueueSet; 777 778 779 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP