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