1 /*
2 * Copyright (c) 2001, 2011, 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
57 // Simple TaskQueue stats that are collected by default in debug builds.
58
59 #if !defined(TASKQUEUE_STATS) && defined(ASSERT)
60 #define TASKQUEUE_STATS 1
61 #elif !defined(TASKQUEUE_STATS)
62 #define TASKQUEUE_STATS 0
63 #endif
64
65 #if TASKQUEUE_STATS
66 #define TASKQUEUE_STATS_ONLY(code) code
67 #else
68 #define TASKQUEUE_STATS_ONLY(code)
69 #endif // TASKQUEUE_STATS
70
71 #if TASKQUEUE_STATS
72 class TaskQueueStats {
73 public:
74 enum StatId {
75 push, // number of taskqueue pushes
76 pop, // number of taskqueue pops
77 pop_slow, // subset of taskqueue pops that were done slow-path
78 steal_attempt, // number of taskqueue steal attempts
79 steal, // number of taskqueue steals
80 overflow, // number of overflow pushes
81 overflow_max_len, // max length of overflow stack
82 last_stat_id
83 };
84
85 public:
86 inline TaskQueueStats() { reset(); }
87
88 inline void record_push() { ++_stats[push]; }
89 inline void record_pop() { ++_stats[pop]; }
90 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; }
91 inline void record_steal(bool success);
92 inline void record_overflow(size_t new_length);
93
94 TaskQueueStats & operator +=(const TaskQueueStats & addend);
95
96 inline size_t get(StatId id) const { return _stats[id]; }
97 inline const size_t* get() const { return _stats; }
98
99 inline void reset();
100
101 // Print the specified line of the header (does not include a line separator).
102 static void print_header(unsigned int line, outputStream* const stream = tty,
103 unsigned int width = 10);
104 // Print the statistics (does not include a line separator).
105 void print(outputStream* const stream = tty, unsigned int width = 10) const;
106
107 DEBUG_ONLY(void verify() const;)
108
109 private:
110 size_t _stats[last_stat_id];
111 static const char * const _names[last_stat_id];
112 };
113
114 void TaskQueueStats::record_steal(bool success) {
115 ++_stats[steal_attempt];
116 if (success) ++_stats[steal];
117 }
118
119 void TaskQueueStats::record_overflow(size_t new_len) {
120 ++_stats[overflow];
121 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len;
122 }
123
124 void TaskQueueStats::reset() {
125 memset(_stats, 0, sizeof(_stats));
126 }
127 #endif // TASKQUEUE_STATS
128
129 template <unsigned int N>
130 class TaskQueueSuper: public CHeapObj {
131 protected:
132 // Internal type for indexing the queue; also used for the tag.
133 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t;
134
135 // The first free element after the last one pushed (mod N).
136 volatile uint _bottom;
137
138 enum { MOD_N_MASK = N - 1 };
139
140 class Age {
141 public:
142 Age(size_t data = 0) { _data = data; }
143 Age(const Age& age) { _data = age._data; }
144 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; }
145
146 Age get() const volatile { return _data; }
147 void set(Age age) volatile { _data = age._data; }
148
149 idx_t top() const volatile { return _fields._top; }
150 idx_t tag() const volatile { return _fields._tag; }
151
152 // Increment top; if it wraps, increment tag also.
153 void increment() {
154 _fields._top = increment_index(_fields._top);
155 if (_fields._top == 0) ++_fields._tag;
156 }
157
158 Age cmpxchg(const Age new_age, const Age old_age) volatile {
159 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data,
160 (volatile intptr_t *)&_data,
161 (intptr_t)old_age._data);
162 }
163
164 bool operator ==(const Age& other) const { return _data == other._data; }
165
166 private:
167 struct fields {
168 idx_t _top;
169 idx_t _tag;
170 };
171 union {
172 size_t _data;
173 fields _fields;
174 };
175 };
176
177 volatile Age _age;
178
179 // These both operate mod N.
180 static uint increment_index(uint ind) {
181 return (ind + 1) & MOD_N_MASK;
182 }
183 static uint decrement_index(uint ind) {
184 return (ind - 1) & MOD_N_MASK;
185 }
186
187 // Returns a number in the range [0..N). If the result is "N-1", it should be
188 // interpreted as 0.
189 uint dirty_size(uint bot, uint top) const {
190 return (bot - top) & MOD_N_MASK;
191 }
192
193 // Returns the size corresponding to the given "bot" and "top".
194 uint size(uint bot, uint top) const {
195 uint sz = dirty_size(bot, top);
196 // Has the queue "wrapped", so that bottom is less than top? There's a
197 // complicated special case here. A pair of threads could perform pop_local
198 // and pop_global operations concurrently, starting from a state in which
199 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom,
200 // and the pop_global in incrementing _top (in which case the pop_global
201 // will be awarded the contested queue element.) The resulting state must
202 // be interpreted as an empty queue. (We only need to worry about one such
203 // event: only the queue owner performs pop_local's, and several concurrent
204 // threads attempting to perform the pop_global will all perform the same
205 // CAS, and only one can succeed.) Any stealing thread that reads after
206 // either the increment or decrement will see an empty queue, and will not
207 // join the competitors. The "sz == -1 || sz == N-1" state will not be
208 // modified by concurrent queues, so the owner thread can reset the state to
209 // _bottom == top so subsequent pushes will be performed normally.
210 return (sz == N - 1) ? 0 : sz;
211 }
212
213 public:
214 TaskQueueSuper() : _bottom(0), _age() {}
215
216 // Return true if the TaskQueue contains/does not contain any tasks.
217 bool peek() const { return _bottom != _age.top(); }
218 bool is_empty() const { return size() == 0; }
219
220 // Return an estimate of the number of elements in the queue.
221 // The "careful" version admits the possibility of pop_local/pop_global
222 // races.
223 uint size() const {
224 return size(_bottom, _age.top());
225 }
226
227 uint dirty_size() const {
228 return dirty_size(_bottom, _age.top());
229 }
230
231 void set_empty() {
232 _bottom = 0;
233 _age.set(0);
234 }
235
236 // Maximum number of elements allowed in the queue. This is two less
237 // than the actual queue size, for somewhat complicated reasons.
238 uint max_elems() const { return N - 2; }
239
240 // Total size of queue.
241 static const uint total_size() { return N; }
242
243 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;)
244 };
245
246 template<class E, unsigned int N = TASKQUEUE_SIZE>
247 class GenericTaskQueue: public TaskQueueSuper<N> {
248 protected:
249 typedef typename TaskQueueSuper<N>::Age Age;
250 typedef typename TaskQueueSuper<N>::idx_t idx_t;
251
252 using TaskQueueSuper<N>::_bottom;
253 using TaskQueueSuper<N>::_age;
254 using TaskQueueSuper<N>::increment_index;
255 using TaskQueueSuper<N>::decrement_index;
256 using TaskQueueSuper<N>::dirty_size;
257
258 public:
259 using TaskQueueSuper<N>::max_elems;
260 using TaskQueueSuper<N>::size;
261 TASKQUEUE_STATS_ONLY(using TaskQueueSuper<N>::stats;)
262
263 private:
264 // Slow paths for push, pop_local. (pop_global has no fast path.)
265 bool push_slow(E t, uint dirty_n_elems);
266 bool pop_local_slow(uint localBot, Age oldAge);
267
268 public:
269 typedef E element_type;
270
271 // Initializes the queue to empty.
272 GenericTaskQueue();
273
274 void initialize();
275
276 // Push the task "t" on the queue. Returns "false" iff the queue is full.
277 inline bool push(E t);
278
279 // Attempts to claim a task from the "local" end of the queue (the most
280 // recently pushed). If successful, returns true and sets t to the task;
281 // otherwise, returns false (the queue is empty).
282 inline bool pop_local(E& t);
283
284 // Like pop_local(), but uses the "global" end of the queue (the least
285 // recently pushed).
286 bool pop_global(E& t);
287
288 // Delete any resource associated with the queue.
289 ~GenericTaskQueue();
290
291 // apply the closure to all elements in the task queue
292 void oops_do(OopClosure* f);
293
294 private:
295 // Element array.
296 volatile E* _elems;
297 };
298
299 template<class E, unsigned int N>
300 GenericTaskQueue<E, N>::GenericTaskQueue() {
301 assert(sizeof(Age) == sizeof(size_t), "Depends on this.");
302 }
303
304 template<class E, unsigned int N>
305 void GenericTaskQueue<E, N>::initialize() {
306 _elems = NEW_C_HEAP_ARRAY(E, N);
307 }
308
309 template<class E, unsigned int N>
310 void GenericTaskQueue<E, N>::oops_do(OopClosure* f) {
311 // tty->print_cr("START OopTaskQueue::oops_do");
312 uint iters = size();
313 uint index = _bottom;
314 for (uint i = 0; i < iters; ++i) {
315 index = decrement_index(index);
316 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T,
317 // index, &_elems[index], _elems[index]);
318 E* t = (E*)&_elems[index]; // cast away volatility
319 oop* p = (oop*)t;
320 assert((*t)->is_oop_or_null(), "Not an oop or null");
321 f->do_oop(p);
322 }
323 // tty->print_cr("END OopTaskQueue::oops_do");
324 }
325
326 template<class E, unsigned int N>
327 bool GenericTaskQueue<E, N>::push_slow(E t, uint dirty_n_elems) {
328 if (dirty_n_elems == N - 1) {
329 // Actually means 0, so do the push.
330 uint localBot = _bottom;
331 // g++ complains if the volatile result of the assignment is unused.
332 const_cast<E&>(_elems[localBot] = t);
333 OrderAccess::release_store(&_bottom, increment_index(localBot));
334 TASKQUEUE_STATS_ONLY(stats.record_push());
335 return true;
336 }
337 return false;
338 }
339
340 // pop_local_slow() is done by the owning thread and is trying to
341 // get the last task in the queue. It will compete with pop_global()
342 // that will be used by other threads. The tag age is incremented
343 // whenever the queue goes empty which it will do here if this thread
344 // gets the last task or in pop_global() if the queue wraps (top == 0
345 // and pop_global() succeeds, see pop_global()).
346 template<class E, unsigned int N>
347 bool GenericTaskQueue<E, N>::pop_local_slow(uint localBot, Age oldAge) {
348 // This queue was observed to contain exactly one element; either this
349 // thread will claim it, or a competing "pop_global". In either case,
350 // the queue will be logically empty afterwards. Create a new Age value
351 // that represents the empty queue for the given value of "_bottom". (We
352 // must also increment "tag" because of the case where "bottom == 1",
353 // "top == 0". A pop_global could read the queue element in that case,
354 // then have the owner thread do a pop followed by another push. Without
355 // the incrementing of "tag", the pop_global's CAS could succeed,
356 // allowing it to believe it has claimed the stale element.)
357 Age newAge((idx_t)localBot, oldAge.tag() + 1);
358 // Perhaps a competing pop_global has already incremented "top", in which
359 // case it wins the element.
360 if (localBot == oldAge.top()) {
361 // No competing pop_global has yet incremented "top"; we'll try to
362 // install new_age, thus claiming the element.
363 Age tempAge = _age.cmpxchg(newAge, oldAge);
364 if (tempAge == oldAge) {
365 // We win.
366 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity");
367 TASKQUEUE_STATS_ONLY(stats.record_pop_slow());
368 return true;
369 }
370 }
371 // We lose; a completing pop_global gets the element. But the queue is empty
372 // and top is greater than bottom. Fix this representation of the empty queue
373 // to become the canonical one.
374 _age.set(newAge);
375 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity");
376 return false;
377 }
378
379 template<class E, unsigned int N>
380 bool GenericTaskQueue<E, N>::pop_global(E& t) {
381 Age oldAge = _age.get();
382 uint localBot = _bottom;
383 uint n_elems = size(localBot, oldAge.top());
384 if (n_elems == 0) {
385 return false;
386 }
387
388 const_cast<E&>(t = _elems[oldAge.top()]);
389 Age newAge(oldAge);
390 newAge.increment();
391 Age resAge = _age.cmpxchg(newAge, oldAge);
392
393 // Note that using "_bottom" here might fail, since a pop_local might
394 // have decremented it.
395 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity");
396 return resAge == oldAge;
397 }
398
399 template<class E, unsigned int N>
400 GenericTaskQueue<E, N>::~GenericTaskQueue() {
401 FREE_C_HEAP_ARRAY(E, _elems);
402 }
403
404 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for
405 // elements that do not fit in the TaskQueue.
406 //
407 // This class hides two methods from super classes:
408 //
409 // push() - push onto the task queue or, if that fails, onto the overflow stack
410 // is_empty() - return true if both the TaskQueue and overflow stack are empty
411 //
412 // Note that size() is not hidden--it returns the number of elements in the
413 // TaskQueue, and does not include the size of the overflow stack. This
414 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues.
415 template<class E, unsigned int N = TASKQUEUE_SIZE>
416 class OverflowTaskQueue: public GenericTaskQueue<E, N>
417 {
418 public:
419 typedef Stack<E> overflow_t;
420 typedef GenericTaskQueue<E, N> taskqueue_t;
421
422 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;)
423
424 // Push task t onto the queue or onto the overflow stack. Return true.
425 inline bool push(E t);
426
427 // Attempt to pop from the overflow stack; return true if anything was popped.
428 inline bool pop_overflow(E& t);
429
430 inline overflow_t* overflow_stack() { return &_overflow_stack; }
431
432 inline bool taskqueue_empty() const { return taskqueue_t::is_empty(); }
433 inline bool overflow_empty() const { return _overflow_stack.is_empty(); }
434 inline bool is_empty() const {
435 return taskqueue_empty() && overflow_empty();
436 }
437
438 private:
439 overflow_t _overflow_stack;
440 };
441
442 template <class E, unsigned int N>
443 bool OverflowTaskQueue<E, N>::push(E t)
444 {
445 if (!taskqueue_t::push(t)) {
446 overflow_stack()->push(t);
447 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size()));
448 }
449 return true;
450 }
451
452 template <class E, unsigned int N>
453 bool OverflowTaskQueue<E, N>::pop_overflow(E& t)
454 {
455 if (overflow_empty()) return false;
456 t = overflow_stack()->pop();
457 return true;
458 }
459
460 class TaskQueueSetSuper: public CHeapObj {
461 protected:
462 static int randomParkAndMiller(int* seed0);
463 public:
464 // Returns "true" if some TaskQueue in the set contains a task.
465 virtual bool peek() = 0;
466 };
467
468 template<class T>
469 class GenericTaskQueueSet: public TaskQueueSetSuper {
470 private:
471 uint _n;
472 T** _queues;
473
474 public:
475 typedef typename T::element_type E;
476
477 GenericTaskQueueSet(int n) : _n(n) {
478 typedef T* GenericTaskQueuePtr;
479 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n);
480 for (int i = 0; i < n; i++) {
481 _queues[i] = NULL;
482 }
483 }
484
485 bool steal_1_random(uint queue_num, int* seed, E& t);
486 bool steal_best_of_2(uint queue_num, int* seed, E& t);
487 bool steal_best_of_all(uint queue_num, int* seed, E& t);
488
489 void register_queue(uint i, T* q);
490
491 T* queue(uint n);
492
493 // The thread with queue number "queue_num" (and whose random number seed is
494 // at "seed") is trying to steal a task from some other queue. (It may try
495 // several queues, according to some configuration parameter.) If some steal
496 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns
497 // false.
498 bool steal(uint queue_num, int* seed, E& t);
499
500 bool peek();
501 };
502
503 template<class T> void
504 GenericTaskQueueSet<T>::register_queue(uint i, T* q) {
505 assert(i < _n, "index out of range.");
506 _queues[i] = q;
507 }
508
509 template<class T> T*
510 GenericTaskQueueSet<T>::queue(uint i) {
511 return _queues[i];
512 }
513
514 template<class T> bool
515 GenericTaskQueueSet<T>::steal(uint queue_num, int* seed, E& t) {
516 for (uint i = 0; i < 2 * _n; i++) {
517 if (steal_best_of_2(queue_num, seed, t)) {
518 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true));
519 return true;
520 }
521 }
522 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false));
523 return false;
524 }
525
526 template<class T> bool
527 GenericTaskQueueSet<T>::steal_best_of_all(uint queue_num, int* seed, E& t) {
528 if (_n > 2) {
529 int best_k;
530 uint best_sz = 0;
531 for (uint k = 0; k < _n; k++) {
532 if (k == queue_num) continue;
533 uint sz = _queues[k]->size();
534 if (sz > best_sz) {
535 best_sz = sz;
536 best_k = k;
537 }
538 }
539 return best_sz > 0 && _queues[best_k]->pop_global(t);
540 } else if (_n == 2) {
541 // Just try the other one.
542 int k = (queue_num + 1) % 2;
543 return _queues[k]->pop_global(t);
544 } else {
545 assert(_n == 1, "can't be zero.");
546 return false;
547 }
548 }
549
550 template<class T> bool
551 GenericTaskQueueSet<T>::steal_1_random(uint queue_num, int* seed, E& t) {
552 if (_n > 2) {
553 uint k = queue_num;
554 while (k == queue_num) k = randomParkAndMiller(seed) % _n;
555 return _queues[2]->pop_global(t);
556 } else if (_n == 2) {
557 // Just try the other one.
558 int k = (queue_num + 1) % 2;
559 return _queues[k]->pop_global(t);
560 } else {
561 assert(_n == 1, "can't be zero.");
562 return false;
563 }
564 }
565
566 template<class T> bool
567 GenericTaskQueueSet<T>::steal_best_of_2(uint queue_num, int* seed, E& t) {
568 if (_n > 2) {
569 uint k1 = queue_num;
570 while (k1 == queue_num) k1 = randomParkAndMiller(seed) % _n;
571 uint k2 = queue_num;
572 while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n;
573 // Sample both and try the larger.
574 uint sz1 = _queues[k1]->size();
575 uint sz2 = _queues[k2]->size();
576 if (sz2 > sz1) return _queues[k2]->pop_global(t);
577 else return _queues[k1]->pop_global(t);
578 } else if (_n == 2) {
579 // Just try the other one.
580 uint k = (queue_num + 1) % 2;
581 return _queues[k]->pop_global(t);
582 } else {
583 assert(_n == 1, "can't be zero.");
584 return false;
585 }
586 }
587
588 template<class T>
589 bool GenericTaskQueueSet<T>::peek() {
590 // Try all the queues.
591 for (uint j = 0; j < _n; j++) {
592 if (_queues[j]->peek())
593 return true;
594 }
595 return false;
596 }
597
598 // When to terminate from the termination protocol.
599 class TerminatorTerminator: public CHeapObj {
600 public:
601 virtual bool should_exit_termination() = 0;
602 };
603
604 // A class to aid in the termination of a set of parallel tasks using
605 // TaskQueueSet's for work stealing.
606
607 #undef TRACESPINNING
608
609 class ParallelTaskTerminator: public StackObj {
610 private:
611 int _n_threads;
612 TaskQueueSetSuper* _queue_set;
613 int _offered_termination;
614
615 #ifdef TRACESPINNING
616 static uint _total_yields;
617 static uint _total_spins;
618 static uint _total_peeks;
619 #endif
620
621 bool peek_in_queue_set();
622 protected:
623 virtual void yield();
624 void sleep(uint millis);
625
626 public:
627
628 // "n_threads" is the number of threads to be terminated. "queue_set" is a
629 // queue sets of work queues of other threads.
630 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set);
631
632 // The current thread has no work, and is ready to terminate if everyone
633 // else is. If returns "true", all threads are terminated. If returns
634 // "false", available work has been observed in one of the task queues,
635 // so the global task is not complete.
636 bool offer_termination() {
637 return offer_termination(NULL);
638 }
639
640 // As above, but it also terminates if the should_exit_termination()
641 // method of the terminator parameter returns true. If terminator is
642 // NULL, then it is ignored.
643 bool offer_termination(TerminatorTerminator* terminator);
644
645 // Reset the terminator, so that it may be reused again.
646 // The caller is responsible for ensuring that this is done
647 // in an MT-safe manner, once the previous round of use of
648 // the terminator is finished.
649 void reset_for_reuse();
650 // Same as above but the number of parallel threads is set to the
651 // given number.
652 void reset_for_reuse(int n_threads);
653
654 #ifdef TRACESPINNING
655 static uint total_yields() { return _total_yields; }
656 static uint total_spins() { return _total_spins; }
657 static uint total_peeks() { return _total_peeks; }
658 static void print_termination_counts();
659 #endif
660 };
661
662 template<class E, unsigned int N> inline bool
663 GenericTaskQueue<E, N>::push(E t) {
664 uint localBot = _bottom;
665 assert((localBot >= 0) && (localBot < N), "_bottom out of range.");
666 idx_t top = _age.top();
667 uint dirty_n_elems = dirty_size(localBot, top);
668 assert(dirty_n_elems < N, "n_elems out of range.");
669 if (dirty_n_elems < max_elems()) {
670 // g++ complains if the volatile result of the assignment is unused.
671 const_cast<E&>(_elems[localBot] = t);
672 OrderAccess::release_store(&_bottom, increment_index(localBot));
673 TASKQUEUE_STATS_ONLY(stats.record_push());
674 return true;
675 } else {
676 return push_slow(t, dirty_n_elems);
677 }
678 }
679
680 template<class E, unsigned int N> inline bool
681 GenericTaskQueue<E, N>::pop_local(E& t) {
682 uint localBot = _bottom;
683 // This value cannot be N-1. That can only occur as a result of
684 // the assignment to bottom in this method. If it does, this method
685 // resets the size to 0 before the next call (which is sequential,
686 // since this is pop_local.)
687 uint dirty_n_elems = dirty_size(localBot, _age.top());
688 assert(dirty_n_elems != N - 1, "Shouldn't be possible...");
689 if (dirty_n_elems == 0) return false;
690 localBot = decrement_index(localBot);
691 _bottom = localBot;
692 // This is necessary to prevent any read below from being reordered
693 // before the store just above.
694 OrderAccess::fence();
695 const_cast<E&>(t = _elems[localBot]);
696 // This is a second read of "age"; the "size()" above is the first.
697 // If there's still at least one element in the queue, based on the
698 // "_bottom" and "age" we've read, then there can be no interference with
699 // a "pop_global" operation, and we're done.
700 idx_t tp = _age.top(); // XXX
701 if (size(localBot, tp) > 0) {
702 assert(dirty_size(localBot, tp) != N - 1, "sanity");
703 TASKQUEUE_STATS_ONLY(stats.record_pop());
704 return true;
705 } else {
706 // Otherwise, the queue contained exactly one element; we take the slow
707 // path.
708 return pop_local_slow(localBot, _age.get());
709 }
710 }
711
712 typedef GenericTaskQueue<oop> OopTaskQueue;
713 typedef GenericTaskQueueSet<OopTaskQueue> OopTaskQueueSet;
714
715 #ifdef _MSC_VER
716 #pragma warning(push)
717 // warning C4522: multiple assignment operators specified
718 #pragma warning(disable:4522)
719 #endif
720
721 // This is a container class for either an oop* or a narrowOop*.
722 // Both are pushed onto a task queue and the consumer will test is_narrow()
723 // to determine which should be processed.
724 class StarTask {
725 void* _holder; // either union oop* or narrowOop*
726
727 enum { COMPRESSED_OOP_MASK = 1 };
728
729 public:
730 StarTask(narrowOop* p) {
731 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
732 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK);
733 }
734 StarTask(oop* p) {
735 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
736 _holder = (void*)p;
737 }
738 StarTask() { _holder = NULL; }
739 operator oop*() { return (oop*)_holder; }
740 operator narrowOop*() {
741 return (narrowOop*)((uintptr_t)_holder & ~COMPRESSED_OOP_MASK);
742 }
743
744 StarTask& operator=(const StarTask& t) {
745 _holder = t._holder;
746 return *this;
747 }
748 volatile StarTask& operator=(const volatile StarTask& t) volatile {
749 _holder = t._holder;
750 return *this;
751 }
752
753 bool is_narrow() const {
754 return (((uintptr_t)_holder & COMPRESSED_OOP_MASK) != 0);
755 }
756 };
757
758 class ObjArrayTask
759 {
760 public:
761 ObjArrayTask(oop o = NULL, int idx = 0): _obj(o), _index(idx) { }
762 ObjArrayTask(oop o, size_t idx): _obj(o), _index(int(idx)) {
763 assert(idx <= size_t(max_jint), "too big");
764 }
765 ObjArrayTask(const ObjArrayTask& t): _obj(t._obj), _index(t._index) { }
766
767 ObjArrayTask& operator =(const ObjArrayTask& t) {
768 _obj = t._obj;
769 _index = t._index;
770 return *this;
771 }
772 volatile ObjArrayTask&
773 operator =(const volatile ObjArrayTask& t) volatile {
774 _obj = t._obj;
775 _index = t._index;
776 return *this;
777 }
778
779 inline oop obj() const { return _obj; }
780 inline int index() const { return _index; }
781
782 DEBUG_ONLY(bool is_valid() const); // Tasks to be pushed/popped must be valid.
783
784 private:
785 oop _obj;
786 int _index;
787 };
788
789 #ifdef _MSC_VER
790 #pragma warning(pop)
791 #endif
792
793 typedef OverflowTaskQueue<StarTask> OopStarTaskQueue;
794 typedef GenericTaskQueueSet<OopStarTaskQueue> OopStarTaskQueueSet;
795
796 typedef OverflowTaskQueue<size_t> RegionTaskQueue;
797 typedef GenericTaskQueueSet<RegionTaskQueue> RegionTaskQueueSet;
798
799
800 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP