1 /*
2 * Copyright (c) 2001, 2010, 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 #ifdef TARGET_OS_ARCH_linux_x86
32 # include "orderAccess_linux_x86.inline.hpp"
33 #endif
34 #ifdef TARGET_OS_ARCH_linux_sparc
35 # include "orderAccess_linux_sparc.inline.hpp"
36 #endif
37 #ifdef TARGET_OS_ARCH_linux_zero
38 # include "orderAccess_linux_zero.inline.hpp"
39 #endif
40 #ifdef TARGET_OS_ARCH_solaris_x86
41 # include "orderAccess_solaris_x86.inline.hpp"
42 #endif
43 #ifdef TARGET_OS_ARCH_solaris_sparc
44 # include "orderAccess_solaris_sparc.inline.hpp"
45 #endif
46 #ifdef TARGET_OS_ARCH_windows_x86
47 # include "orderAccess_windows_x86.inline.hpp"
48 #endif
49
50 // Simple TaskQueue stats that are collected by default in debug builds.
51
52 #if !defined(TASKQUEUE_STATS) && defined(ASSERT)
53 #define TASKQUEUE_STATS 1
54 #elif !defined(TASKQUEUE_STATS)
55 #define TASKQUEUE_STATS 0
56 #endif
57
58 #if TASKQUEUE_STATS
59 #define TASKQUEUE_STATS_ONLY(code) code
60 #else
61 #define TASKQUEUE_STATS_ONLY(code)
62 #endif // TASKQUEUE_STATS
63
64 #if TASKQUEUE_STATS
65 class TaskQueueStats {
66 public:
67 enum StatId {
68 push, // number of taskqueue pushes
69 pop, // number of taskqueue pops
70 pop_slow, // subset of taskqueue pops that were done slow-path
71 steal_attempt, // number of taskqueue steal attempts
72 steal, // number of taskqueue steals
73 overflow, // number of overflow pushes
74 overflow_max_len, // max length of overflow stack
75 last_stat_id
76 };
77
78 public:
79 inline TaskQueueStats() { reset(); }
80
81 inline void record_push() { ++_stats[push]; }
82 inline void record_pop() { ++_stats[pop]; }
83 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; }
84 inline void record_steal(bool success);
85 inline void record_overflow(size_t new_length);
86
87 TaskQueueStats & operator +=(const TaskQueueStats & addend);
88
89 inline size_t get(StatId id) const { return _stats[id]; }
90 inline const size_t* get() const { return _stats; }
91
92 inline void reset();
93
94 // Print the specified line of the header (does not include a line separator).
95 static void print_header(unsigned int line, outputStream* const stream = tty,
96 unsigned int width = 10);
97 // Print the statistics (does not include a line separator).
98 void print(outputStream* const stream = tty, unsigned int width = 10) const;
99
100 DEBUG_ONLY(void verify() const;)
101
102 private:
103 size_t _stats[last_stat_id];
104 static const char * const _names[last_stat_id];
105 };
106
107 void TaskQueueStats::record_steal(bool success) {
108 ++_stats[steal_attempt];
109 if (success) ++_stats[steal];
110 }
111
112 void TaskQueueStats::record_overflow(size_t new_len) {
113 ++_stats[overflow];
114 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len;
115 }
116
117 void TaskQueueStats::reset() {
118 memset(_stats, 0, sizeof(_stats));
119 }
120 #endif // TASKQUEUE_STATS
121
122 template <unsigned int N>
123 class TaskQueueSuper: public CHeapObj {
124 protected:
125 // Internal type for indexing the queue; also used for the tag.
126 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t;
127
128 // The first free element after the last one pushed (mod N).
129 volatile uint _bottom;
130
131 enum { MOD_N_MASK = N - 1 };
132
133 class Age {
134 public:
135 Age(size_t data = 0) { _data = data; }
136 Age(const Age& age) { _data = age._data; }
137 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; }
138
139 Age get() const volatile { return _data; }
140 void set(Age age) volatile { _data = age._data; }
141
142 idx_t top() const volatile { return _fields._top; }
143 idx_t tag() const volatile { return _fields._tag; }
144
145 // Increment top; if it wraps, increment tag also.
146 void increment() {
147 _fields._top = increment_index(_fields._top);
148 if (_fields._top == 0) ++_fields._tag;
149 }
150
151 Age cmpxchg(const Age new_age, const Age old_age) volatile {
152 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data,
153 (volatile intptr_t *)&_data,
154 (intptr_t)old_age._data);
155 }
156
157 bool operator ==(const Age& other) const { return _data == other._data; }
158
159 private:
160 struct fields {
161 idx_t _top;
162 idx_t _tag;
163 };
164 union {
165 size_t _data;
166 fields _fields;
167 };
168 };
169
170 volatile Age _age;
171
172 // These both operate mod N.
173 static uint increment_index(uint ind) {
174 return (ind + 1) & MOD_N_MASK;
175 }
176 static uint decrement_index(uint ind) {
177 return (ind - 1) & MOD_N_MASK;
178 }
179
180 // Returns a number in the range [0..N). If the result is "N-1", it should be
181 // interpreted as 0.
182 uint dirty_size(uint bot, uint top) const {
183 return (bot - top) & MOD_N_MASK;
184 }
185
186 // Returns the size corresponding to the given "bot" and "top".
187 uint size(uint bot, uint top) const {
188 uint sz = dirty_size(bot, top);
189 // Has the queue "wrapped", so that bottom is less than top? There's a
190 // complicated special case here. A pair of threads could perform pop_local
191 // and pop_global operations concurrently, starting from a state in which
192 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom,
193 // and the pop_global in incrementing _top (in which case the pop_global
194 // will be awarded the contested queue element.) The resulting state must
195 // be interpreted as an empty queue. (We only need to worry about one such
196 // event: only the queue owner performs pop_local's, and several concurrent
197 // threads attempting to perform the pop_global will all perform the same
198 // CAS, and only one can succeed.) Any stealing thread that reads after
199 // either the increment or decrement will see an empty queue, and will not
200 // join the competitors. The "sz == -1 || sz == N-1" state will not be
201 // modified by concurrent queues, so the owner thread can reset the state to
202 // _bottom == top so subsequent pushes will be performed normally.
203 return (sz == N - 1) ? 0 : sz;
204 }
205
206 public:
207 TaskQueueSuper() : _bottom(0), _age() {}
208
209 // Return true if the TaskQueue contains/does not contain any tasks.
210 bool peek() const { return _bottom != _age.top(); }
211 bool is_empty() const { return size() == 0; }
212
213 // Return an estimate of the number of elements in the queue.
214 // The "careful" version admits the possibility of pop_local/pop_global
215 // races.
216 uint size() const {
217 return size(_bottom, _age.top());
218 }
219
220 uint dirty_size() const {
221 return dirty_size(_bottom, _age.top());
222 }
223
224 void set_empty() {
225 _bottom = 0;
226 _age.set(0);
227 }
228
229 // Maximum number of elements allowed in the queue. This is two less
230 // than the actual queue size, for somewhat complicated reasons.
231 uint max_elems() const { return N - 2; }
232
233 // Total size of queue.
234 static const uint total_size() { return N; }
235
236 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;)
237 };
238
239 template<class E, unsigned int N = TASKQUEUE_SIZE>
240 class GenericTaskQueue: public TaskQueueSuper<N> {
241 protected:
242 typedef typename TaskQueueSuper<N>::Age Age;
243 typedef typename TaskQueueSuper<N>::idx_t idx_t;
244
245 using TaskQueueSuper<N>::_bottom;
246 using TaskQueueSuper<N>::_age;
247 using TaskQueueSuper<N>::increment_index;
248 using TaskQueueSuper<N>::decrement_index;
249 using TaskQueueSuper<N>::dirty_size;
250
251 public:
252 using TaskQueueSuper<N>::max_elems;
253 using TaskQueueSuper<N>::size;
254 TASKQUEUE_STATS_ONLY(using TaskQueueSuper<N>::stats;)
255
256 private:
257 // Slow paths for push, pop_local. (pop_global has no fast path.)
258 bool push_slow(E t, uint dirty_n_elems);
259 bool pop_local_slow(uint localBot, Age oldAge);
260
261 public:
262 typedef E element_type;
263
264 // Initializes the queue to empty.
265 GenericTaskQueue();
266
267 void initialize();
268
269 // Push the task "t" on the queue. Returns "false" iff the queue is full.
270 inline bool push(E t);
271
272 // Attempts to claim a task from the "local" end of the queue (the most
273 // recently pushed). If successful, returns true and sets t to the task;
274 // otherwise, returns false (the queue is empty).
275 inline bool pop_local(E& t);
276
277 // Like pop_local(), but uses the "global" end of the queue (the least
278 // recently pushed).
279 bool pop_global(E& t);
280
281 // Delete any resource associated with the queue.
282 ~GenericTaskQueue();
283
284 // apply the closure to all elements in the task queue
285 void oops_do(OopClosure* f);
286
287 private:
288 // Element array.
289 volatile E* _elems;
290 };
291
292 template<class E, unsigned int N>
293 GenericTaskQueue<E, N>::GenericTaskQueue() {
294 assert(sizeof(Age) == sizeof(size_t), "Depends on this.");
295 }
296
297 template<class E, unsigned int N>
298 void GenericTaskQueue<E, N>::initialize() {
299 _elems = NEW_C_HEAP_ARRAY(E, N);
300 }
301
302 template<class E, unsigned int N>
303 void GenericTaskQueue<E, N>::oops_do(OopClosure* f) {
304 // tty->print_cr("START OopTaskQueue::oops_do");
305 uint iters = size();
306 uint index = _bottom;
307 for (uint i = 0; i < iters; ++i) {
308 index = decrement_index(index);
309 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T,
310 // index, &_elems[index], _elems[index]);
311 E* t = (E*)&_elems[index]; // cast away volatility
312 oop* p = (oop*)t;
313 assert((*t)->is_oop_or_null(), "Not an oop or null");
314 f->do_oop(p);
315 }
316 // tty->print_cr("END OopTaskQueue::oops_do");
317 }
318
319 template<class E, unsigned int N>
320 bool GenericTaskQueue<E, N>::push_slow(E t, uint dirty_n_elems) {
321 if (dirty_n_elems == N - 1) {
322 // Actually means 0, so do the push.
323 uint localBot = _bottom;
324 // g++ complains if the volatile result of the assignment is unused.
325 const_cast<E&>(_elems[localBot] = t);
326 OrderAccess::release_store(&_bottom, increment_index(localBot));
327 TASKQUEUE_STATS_ONLY(stats.record_push());
328 return true;
329 }
330 return false;
331 }
332
333 template<class E, unsigned int N>
334 bool GenericTaskQueue<E, N>::pop_local_slow(uint localBot, Age oldAge) {
335 // This queue was observed to contain exactly one element; either this
336 // thread will claim it, or a competing "pop_global". In either case,
337 // the queue will be logically empty afterwards. Create a new Age value
338 // that represents the empty queue for the given value of "_bottom". (We
339 // must also increment "tag" because of the case where "bottom == 1",
340 // "top == 0". A pop_global could read the queue element in that case,
341 // then have the owner thread do a pop followed by another push. Without
342 // the incrementing of "tag", the pop_global's CAS could succeed,
343 // allowing it to believe it has claimed the stale element.)
344 Age newAge((idx_t)localBot, oldAge.tag() + 1);
345 // Perhaps a competing pop_global has already incremented "top", in which
346 // case it wins the element.
347 if (localBot == oldAge.top()) {
348 // No competing pop_global has yet incremented "top"; we'll try to
349 // install new_age, thus claiming the element.
350 Age tempAge = _age.cmpxchg(newAge, oldAge);
351 if (tempAge == oldAge) {
352 // We win.
353 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity");
354 TASKQUEUE_STATS_ONLY(stats.record_pop_slow());
355 return true;
356 }
357 }
358 // We lose; a completing pop_global gets the element. But the queue is empty
359 // and top is greater than bottom. Fix this representation of the empty queue
360 // to become the canonical one.
361 _age.set(newAge);
362 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity");
363 return false;
364 }
365
366 template<class E, unsigned int N>
367 bool GenericTaskQueue<E, N>::pop_global(E& t) {
368 Age oldAge = _age.get();
369 uint localBot = _bottom;
370 uint n_elems = size(localBot, oldAge.top());
371 if (n_elems == 0) {
372 return false;
373 }
374
375 const_cast<E&>(t = _elems[oldAge.top()]);
376 Age newAge(oldAge);
377 newAge.increment();
378 Age resAge = _age.cmpxchg(newAge, oldAge);
379
380 // Note that using "_bottom" here might fail, since a pop_local might
381 // have decremented it.
382 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity");
383 return resAge == oldAge;
384 }
385
386 template<class E, unsigned int N>
387 GenericTaskQueue<E, N>::~GenericTaskQueue() {
388 FREE_C_HEAP_ARRAY(E, _elems);
389 }
390
391 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for
392 // elements that do not fit in the TaskQueue.
393 //
394 // Three methods from super classes are overridden:
395 //
396 // initialize() - initialize the super classes and create the overflow stack
397 // push() - push onto the task queue or, if that fails, onto the overflow stack
398 // is_empty() - return true if both the TaskQueue and overflow stack are empty
399 //
400 // Note that size() is not overridden--it returns the number of elements in the
401 // TaskQueue, and does not include the size of the overflow stack. This
402 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues.
403 template<class E, unsigned int N = TASKQUEUE_SIZE>
404 class OverflowTaskQueue: public GenericTaskQueue<E, N>
405 {
406 public:
407 typedef GrowableArray<E> overflow_t;
408 typedef GenericTaskQueue<E, N> taskqueue_t;
409
410 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;)
411
412 OverflowTaskQueue();
413 ~OverflowTaskQueue();
414 void initialize();
415
416 inline overflow_t* overflow_stack() const { return _overflow_stack; }
417
418 // Push task t onto the queue or onto the overflow stack. Return true.
419 inline bool push(E t);
420
421 // Attempt to pop from the overflow stack; return true if anything was popped.
422 inline bool pop_overflow(E& t);
423
424 inline bool taskqueue_empty() const { return taskqueue_t::is_empty(); }
425 inline bool overflow_empty() const { return overflow_stack()->is_empty(); }
426 inline bool is_empty() const {
427 return taskqueue_empty() && overflow_empty();
428 }
429
430 private:
431 overflow_t* _overflow_stack;
432 };
433
434 template <class E, unsigned int N>
435 OverflowTaskQueue<E, N>::OverflowTaskQueue()
436 {
437 _overflow_stack = NULL;
438 }
439
440 template <class E, unsigned int N>
441 OverflowTaskQueue<E, N>::~OverflowTaskQueue()
442 {
443 if (_overflow_stack != NULL) {
444 delete _overflow_stack;
445 _overflow_stack = NULL;
446 }
447 }
448
449 template <class E, unsigned int N>
450 void OverflowTaskQueue<E, N>::initialize()
451 {
452 taskqueue_t::initialize();
453 assert(_overflow_stack == NULL, "memory leak");
454 _overflow_stack = new (ResourceObj::C_HEAP) GrowableArray<E>(10, true);
455 }
456
457 template <class E, unsigned int N>
458 bool OverflowTaskQueue<E, N>::push(E t)
459 {
460 if (!taskqueue_t::push(t)) {
461 overflow_stack()->push(t);
462 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->length()));
463 }
464 return true;
465 }
466
467 template <class E, unsigned int N>
468 bool OverflowTaskQueue<E, N>::pop_overflow(E& t)
469 {
470 if (overflow_empty()) return false;
471 t = overflow_stack()->pop();
472 return true;
473 }
474
475 class TaskQueueSetSuper: public CHeapObj {
476 protected:
477 static int randomParkAndMiller(int* seed0);
478 public:
479 // Returns "true" if some TaskQueue in the set contains a task.
480 virtual bool peek() = 0;
481 };
482
483 template<class T>
484 class GenericTaskQueueSet: public TaskQueueSetSuper {
485 private:
486 uint _n;
487 T** _queues;
488
489 public:
490 typedef typename T::element_type E;
491
492 GenericTaskQueueSet(int n) : _n(n) {
493 typedef T* GenericTaskQueuePtr;
494 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n);
495 for (int i = 0; i < n; i++) {
496 _queues[i] = NULL;
497 }
498 }
499
500 bool steal_1_random(uint queue_num, int* seed, E& t);
501 bool steal_best_of_2(uint queue_num, int* seed, E& t);
502 bool steal_best_of_all(uint queue_num, int* seed, E& t);
503
504 void register_queue(uint i, T* q);
505
506 T* queue(uint n);
507
508 // The thread with queue number "queue_num" (and whose random number seed is
509 // at "seed") is trying to steal a task from some other queue. (It may try
510 // several queues, according to some configuration parameter.) If some steal
511 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns
512 // false.
513 bool steal(uint queue_num, int* seed, E& t);
514
515 bool peek();
516 };
517
518 template<class T> void
519 GenericTaskQueueSet<T>::register_queue(uint i, T* q) {
520 assert(i < _n, "index out of range.");
521 _queues[i] = q;
522 }
523
524 template<class T> T*
525 GenericTaskQueueSet<T>::queue(uint i) {
526 return _queues[i];
527 }
528
529 template<class T> bool
530 GenericTaskQueueSet<T>::steal(uint queue_num, int* seed, E& t) {
531 for (uint i = 0; i < 2 * _n; i++) {
532 if (steal_best_of_2(queue_num, seed, t)) {
533 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true));
534 return true;
535 }
536 }
537 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false));
538 return false;
539 }
540
541 template<class T> bool
542 GenericTaskQueueSet<T>::steal_best_of_all(uint queue_num, int* seed, E& t) {
543 if (_n > 2) {
544 int best_k;
545 uint best_sz = 0;
546 for (uint k = 0; k < _n; k++) {
547 if (k == queue_num) continue;
548 uint sz = _queues[k]->size();
549 if (sz > best_sz) {
550 best_sz = sz;
551 best_k = k;
552 }
553 }
554 return best_sz > 0 && _queues[best_k]->pop_global(t);
555 } else if (_n == 2) {
556 // Just try the other one.
557 int k = (queue_num + 1) % 2;
558 return _queues[k]->pop_global(t);
559 } else {
560 assert(_n == 1, "can't be zero.");
561 return false;
562 }
563 }
564
565 template<class T> bool
566 GenericTaskQueueSet<T>::steal_1_random(uint queue_num, int* seed, E& t) {
567 if (_n > 2) {
568 uint k = queue_num;
569 while (k == queue_num) k = randomParkAndMiller(seed) % _n;
570 return _queues[2]->pop_global(t);
571 } else if (_n == 2) {
572 // Just try the other one.
573 int k = (queue_num + 1) % 2;
574 return _queues[k]->pop_global(t);
575 } else {
576 assert(_n == 1, "can't be zero.");
577 return false;
578 }
579 }
580
581 template<class T> bool
582 GenericTaskQueueSet<T>::steal_best_of_2(uint queue_num, int* seed, E& t) {
583 if (_n > 2) {
584 uint k1 = queue_num;
585 while (k1 == queue_num) k1 = randomParkAndMiller(seed) % _n;
586 uint k2 = queue_num;
587 while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n;
588 // Sample both and try the larger.
589 uint sz1 = _queues[k1]->size();
590 uint sz2 = _queues[k2]->size();
591 if (sz2 > sz1) return _queues[k2]->pop_global(t);
592 else return _queues[k1]->pop_global(t);
593 } else if (_n == 2) {
594 // Just try the other one.
595 uint k = (queue_num + 1) % 2;
596 return _queues[k]->pop_global(t);
597 } else {
598 assert(_n == 1, "can't be zero.");
599 return false;
600 }
601 }
602
603 template<class T>
604 bool GenericTaskQueueSet<T>::peek() {
605 // Try all the queues.
606 for (uint j = 0; j < _n; j++) {
607 if (_queues[j]->peek())
608 return true;
609 }
610 return false;
611 }
612
613 // When to terminate from the termination protocol.
614 class TerminatorTerminator: public CHeapObj {
615 public:
616 virtual bool should_exit_termination() = 0;
617 };
618
619 // A class to aid in the termination of a set of parallel tasks using
620 // TaskQueueSet's for work stealing.
621
622 #undef TRACESPINNING
623
624 class ParallelTaskTerminator: public StackObj {
625 private:
626 int _n_threads;
627 TaskQueueSetSuper* _queue_set;
628 int _offered_termination;
629
630 #ifdef TRACESPINNING
631 static uint _total_yields;
632 static uint _total_spins;
633 static uint _total_peeks;
634 #endif
635
636 bool peek_in_queue_set();
637 protected:
638 virtual void yield();
639 void sleep(uint millis);
640
641 public:
642
643 // "n_threads" is the number of threads to be terminated. "queue_set" is a
644 // queue sets of work queues of other threads.
645 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set);
646
647 // The current thread has no work, and is ready to terminate if everyone
648 // else is. If returns "true", all threads are terminated. If returns
649 // "false", available work has been observed in one of the task queues,
650 // so the global task is not complete.
651 bool offer_termination() {
652 return offer_termination(NULL);
653 }
654
655 // As above, but it also terminates if the should_exit_termination()
656 // method of the terminator parameter returns true. If terminator is
657 // NULL, then it is ignored.
658 bool offer_termination(TerminatorTerminator* terminator);
659
660 // Reset the terminator, so that it may be reused again.
661 // The caller is responsible for ensuring that this is done
662 // in an MT-safe manner, once the previous round of use of
663 // the terminator is finished.
664 void reset_for_reuse();
665
666 #ifdef TRACESPINNING
667 static uint total_yields() { return _total_yields; }
668 static uint total_spins() { return _total_spins; }
669 static uint total_peeks() { return _total_peeks; }
670 static void print_termination_counts();
671 #endif
672 };
673
674 template<class E, unsigned int N> inline bool
675 GenericTaskQueue<E, N>::push(E t) {
676 uint localBot = _bottom;
677 assert((localBot >= 0) && (localBot < N), "_bottom out of range.");
678 idx_t top = _age.top();
679 uint dirty_n_elems = dirty_size(localBot, top);
680 assert(dirty_n_elems < N, "n_elems out of range.");
681 if (dirty_n_elems < max_elems()) {
682 // g++ complains if the volatile result of the assignment is unused.
683 const_cast<E&>(_elems[localBot] = t);
684 OrderAccess::release_store(&_bottom, increment_index(localBot));
685 TASKQUEUE_STATS_ONLY(stats.record_push());
686 return true;
687 } else {
688 return push_slow(t, dirty_n_elems);
689 }
690 }
691
692 template<class E, unsigned int N> inline bool
693 GenericTaskQueue<E, N>::pop_local(E& t) {
694 uint localBot = _bottom;
695 // This value cannot be N-1. That can only occur as a result of
696 // the assignment to bottom in this method. If it does, this method
697 // resets the size to 0 before the next call (which is sequential,
698 // since this is pop_local.)
699 uint dirty_n_elems = dirty_size(localBot, _age.top());
700 assert(dirty_n_elems != N - 1, "Shouldn't be possible...");
701 if (dirty_n_elems == 0) return false;
702 localBot = decrement_index(localBot);
703 _bottom = localBot;
704 // This is necessary to prevent any read below from being reordered
705 // before the store just above.
706 OrderAccess::fence();
707 const_cast<E&>(t = _elems[localBot]);
708 // This is a second read of "age"; the "size()" above is the first.
709 // If there's still at least one element in the queue, based on the
710 // "_bottom" and "age" we've read, then there can be no interference with
711 // a "pop_global" operation, and we're done.
712 idx_t tp = _age.top(); // XXX
713 if (size(localBot, tp) > 0) {
714 assert(dirty_size(localBot, tp) != N - 1, "sanity");
715 TASKQUEUE_STATS_ONLY(stats.record_pop());
716 return true;
717 } else {
718 // Otherwise, the queue contained exactly one element; we take the slow
719 // path.
720 return pop_local_slow(localBot, _age.get());
721 }
722 }
723
724 typedef GenericTaskQueue<oop> OopTaskQueue;
725 typedef GenericTaskQueueSet<OopTaskQueue> OopTaskQueueSet;
726
727 #ifdef _MSC_VER
728 #pragma warning(push)
729 // warning C4522: multiple assignment operators specified
730 #pragma warning(disable:4522)
731 #endif
732
733 // This is a container class for either an oop* or a narrowOop*.
734 // Both are pushed onto a task queue and the consumer will test is_narrow()
735 // to determine which should be processed.
736 class StarTask {
737 void* _holder; // either union oop* or narrowOop*
738
739 enum { COMPRESSED_OOP_MASK = 1 };
740
741 public:
742 StarTask(narrowOop* p) {
743 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
744 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK);
745 }
746 StarTask(oop* p) {
747 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
748 _holder = (void*)p;
749 }
750 StarTask() { _holder = NULL; }
751 operator oop*() { return (oop*)_holder; }
752 operator narrowOop*() {
753 return (narrowOop*)((uintptr_t)_holder & ~COMPRESSED_OOP_MASK);
754 }
755
756 StarTask& operator=(const StarTask& t) {
757 _holder = t._holder;
758 return *this;
759 }
760 volatile StarTask& operator=(const volatile StarTask& t) volatile {
761 _holder = t._holder;
762 return *this;
763 }
764
765 bool is_narrow() const {
766 return (((uintptr_t)_holder & COMPRESSED_OOP_MASK) != 0);
767 }
768 };
769
770 class ObjArrayTask
771 {
772 public:
773 ObjArrayTask(oop o = NULL, int idx = 0): _obj(o), _index(idx) { }
774 ObjArrayTask(oop o, size_t idx): _obj(o), _index(int(idx)) {
775 assert(idx <= size_t(max_jint), "too big");
776 }
777 ObjArrayTask(const ObjArrayTask& t): _obj(t._obj), _index(t._index) { }
778
779 ObjArrayTask& operator =(const ObjArrayTask& t) {
780 _obj = t._obj;
781 _index = t._index;
782 return *this;
783 }
784 volatile ObjArrayTask&
785 operator =(const volatile ObjArrayTask& t) volatile {
786 _obj = t._obj;
787 _index = t._index;
788 return *this;
789 }
790
791 inline oop obj() const { return _obj; }
792 inline int index() const { return _index; }
793
794 DEBUG_ONLY(bool is_valid() const); // Tasks to be pushed/popped must be valid.
795
796 private:
797 oop _obj;
798 int _index;
799 };
800
801 #ifdef _MSC_VER
802 #pragma warning(pop)
803 #endif
804
805 typedef OverflowTaskQueue<StarTask> OopStarTaskQueue;
806 typedef GenericTaskQueueSet<OopStarTaskQueue> OopStarTaskQueueSet;
807
808 typedef OverflowTaskQueue<size_t> RegionTaskQueue;
809 typedef GenericTaskQueueSet<RegionTaskQueue> RegionTaskQueueSet;
810
811 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP