Fri, 05 Oct 2012 18:57:10 -0700
7177003: C1: LogCompilation support
Summary: add LogCompilation support in C1 - both client and tiered mode.
Reviewed-by: twisti, kvn
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 */
25 #ifndef SHARE_VM_UTILITIES_TASKQUEUE_HPP
26 #define SHARE_VM_UTILITIES_TASKQUEUE_HPP
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
63 // Simple TaskQueue stats that are collected by default in debug builds.
65 #if !defined(TASKQUEUE_STATS) && defined(ASSERT)
66 #define TASKQUEUE_STATS 1
67 #elif !defined(TASKQUEUE_STATS)
68 #define TASKQUEUE_STATS 0
69 #endif
71 #if TASKQUEUE_STATS
72 #define TASKQUEUE_STATS_ONLY(code) code
73 #else
74 #define TASKQUEUE_STATS_ONLY(code)
75 #endif // TASKQUEUE_STATS
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 };
91 public:
92 inline TaskQueueStats() { reset(); }
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);
100 TaskQueueStats & operator +=(const TaskQueueStats & addend);
102 inline size_t get(StatId id) const { return _stats[id]; }
103 inline const size_t* get() const { return _stats; }
105 inline void reset();
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;
113 DEBUG_ONLY(void verify() const;)
115 private:
116 size_t _stats[last_stat_id];
117 static const char * const _names[last_stat_id];
118 };
120 void TaskQueueStats::record_steal(bool success) {
121 ++_stats[steal_attempt];
122 if (success) ++_stats[steal];
123 }
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 }
130 void TaskQueueStats::reset() {
131 memset(_stats, 0, sizeof(_stats));
132 }
133 #endif // TASKQUEUE_STATS
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;
141 // The first free element after the last one pushed (mod N).
142 volatile uint _bottom;
144 enum { MOD_N_MASK = N - 1 };
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; }
152 Age get() const volatile { return _data; }
153 void set(Age age) volatile { _data = age._data; }
155 idx_t top() const volatile { return _fields._top; }
156 idx_t tag() const volatile { return _fields._tag; }
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 }
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 }
170 bool operator ==(const Age& other) const { return _data == other._data; }
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 };
183 volatile Age _age;
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 }
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 }
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 }
219 public:
220 TaskQueueSuper() : _bottom(0), _age() {}
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; }
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 }
233 uint dirty_size() const {
234 return dirty_size(_bottom, _age.top());
235 }
237 void set_empty() {
238 _bottom = 0;
239 _age.set(0);
240 }
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; }
246 // Total size of queue.
247 static const uint total_size() { return N; }
249 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;)
250 };
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;
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;
266 public:
267 using TaskQueueSuper<N, F>::max_elems;
268 using TaskQueueSuper<N, F>::size;
270 #if TASKQUEUE_STATS
271 using TaskQueueSuper<N, F>::stats;
272 #endif
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);
279 public:
280 typedef E element_type;
282 // Initializes the queue to empty.
283 GenericTaskQueue();
285 void initialize();
287 // Push the task "t" on the queue. Returns "false" iff the queue is full.
288 inline bool push(E t);
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);
295 // Like pop_local(), but uses the "global" end of the queue (the least
296 // recently pushed).
297 bool pop_global(E& t);
299 // Delete any resource associated with the queue.
300 ~GenericTaskQueue();
302 // apply the closure to all elements in the task queue
303 void oops_do(OopClosure* f);
305 private:
306 // Element array.
307 volatile E* _elems;
308 };
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 }
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 }
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 }
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 }
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 }
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 uint localBot = _bottom;
394 uint n_elems = size(localBot, oldAge.top());
395 if (n_elems == 0) {
396 return false;
397 }
399 const_cast<E&>(t = _elems[oldAge.top()]);
400 Age newAge(oldAge);
401 newAge.increment();
402 Age resAge = _age.cmpxchg(newAge, oldAge);
404 // Note that using "_bottom" here might fail, since a pop_local might
405 // have decremented it.
406 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity");
407 return resAge == oldAge;
408 }
410 template<class E, MEMFLAGS F, unsigned int N>
411 GenericTaskQueue<E, F, N>::~GenericTaskQueue() {
412 FREE_C_HEAP_ARRAY(E, _elems, F);
413 }
415 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for
416 // elements that do not fit in the TaskQueue.
417 //
418 // This class hides two methods from super classes:
419 //
420 // push() - push onto the task queue or, if that fails, onto the overflow stack
421 // is_empty() - return true if both the TaskQueue and overflow stack are empty
422 //
423 // Note that size() is not hidden--it returns the number of elements in the
424 // TaskQueue, and does not include the size of the overflow stack. This
425 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues.
426 template<class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE>
427 class OverflowTaskQueue: public GenericTaskQueue<E, F, N>
428 {
429 public:
430 typedef Stack<E, F> overflow_t;
431 typedef GenericTaskQueue<E, F, N> taskqueue_t;
433 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;)
435 // Push task t onto the queue or onto the overflow stack. Return true.
436 inline bool push(E t);
438 // Attempt to pop from the overflow stack; return true if anything was popped.
439 inline bool pop_overflow(E& t);
441 inline overflow_t* overflow_stack() { return &_overflow_stack; }
443 inline bool taskqueue_empty() const { return taskqueue_t::is_empty(); }
444 inline bool overflow_empty() const { return _overflow_stack.is_empty(); }
445 inline bool is_empty() const {
446 return taskqueue_empty() && overflow_empty();
447 }
449 private:
450 overflow_t _overflow_stack;
451 };
453 template <class E, MEMFLAGS F, unsigned int N>
454 bool OverflowTaskQueue<E, F, N>::push(E t)
455 {
456 if (!taskqueue_t::push(t)) {
457 overflow_stack()->push(t);
458 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size()));
459 }
460 return true;
461 }
463 template <class E, MEMFLAGS F, unsigned int N>
464 bool OverflowTaskQueue<E, F, N>::pop_overflow(E& t)
465 {
466 if (overflow_empty()) return false;
467 t = overflow_stack()->pop();
468 return true;
469 }
471 class TaskQueueSetSuper {
472 protected:
473 static int randomParkAndMiller(int* seed0);
474 public:
475 // Returns "true" if some TaskQueue in the set contains a task.
476 virtual bool peek() = 0;
477 };
479 template <MEMFLAGS F> class TaskQueueSetSuperImpl: public CHeapObj<F>, public TaskQueueSetSuper {
480 };
482 template<class T, MEMFLAGS F>
483 class GenericTaskQueueSet: public TaskQueueSetSuperImpl<F> {
484 private:
485 uint _n;
486 T** _queues;
488 public:
489 typedef typename T::element_type E;
491 GenericTaskQueueSet(int n) : _n(n) {
492 typedef T* GenericTaskQueuePtr;
493 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n, F);
494 for (int i = 0; i < n; i++) {
495 _queues[i] = NULL;
496 }
497 }
499 bool steal_1_random(uint queue_num, int* seed, E& t);
500 bool steal_best_of_2(uint queue_num, int* seed, E& t);
501 bool steal_best_of_all(uint queue_num, int* seed, E& t);
503 void register_queue(uint i, T* q);
505 T* queue(uint n);
507 // The thread with queue number "queue_num" (and whose random number seed is
508 // at "seed") is trying to steal a task from some other queue. (It may try
509 // several queues, according to some configuration parameter.) If some steal
510 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns
511 // false.
512 bool steal(uint queue_num, int* seed, E& t);
514 bool peek();
515 };
517 template<class T, MEMFLAGS F> void
518 GenericTaskQueueSet<T, F>::register_queue(uint i, T* q) {
519 assert(i < _n, "index out of range.");
520 _queues[i] = q;
521 }
523 template<class T, MEMFLAGS F> T*
524 GenericTaskQueueSet<T, F>::queue(uint i) {
525 return _queues[i];
526 }
528 template<class T, MEMFLAGS F> bool
529 GenericTaskQueueSet<T, F>::steal(uint queue_num, int* seed, E& t) {
530 for (uint i = 0; i < 2 * _n; i++) {
531 if (steal_best_of_2(queue_num, seed, t)) {
532 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true));
533 return true;
534 }
535 }
536 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false));
537 return false;
538 }
540 template<class T, MEMFLAGS F> bool
541 GenericTaskQueueSet<T, F>::steal_best_of_all(uint queue_num, int* seed, E& t) {
542 if (_n > 2) {
543 int best_k;
544 uint best_sz = 0;
545 for (uint k = 0; k < _n; k++) {
546 if (k == queue_num) continue;
547 uint sz = _queues[k]->size();
548 if (sz > best_sz) {
549 best_sz = sz;
550 best_k = k;
551 }
552 }
553 return best_sz > 0 && _queues[best_k]->pop_global(t);
554 } else if (_n == 2) {
555 // Just try the other one.
556 int k = (queue_num + 1) % 2;
557 return _queues[k]->pop_global(t);
558 } else {
559 assert(_n == 1, "can't be zero.");
560 return false;
561 }
562 }
564 template<class T, MEMFLAGS F> bool
565 GenericTaskQueueSet<T, F>::steal_1_random(uint queue_num, int* seed, E& t) {
566 if (_n > 2) {
567 uint k = queue_num;
568 while (k == queue_num) k = TaskQueueSetSuper::randomParkAndMiller(seed) % _n;
569 return _queues[2]->pop_global(t);
570 } else if (_n == 2) {
571 // Just try the other one.
572 int k = (queue_num + 1) % 2;
573 return _queues[k]->pop_global(t);
574 } else {
575 assert(_n == 1, "can't be zero.");
576 return false;
577 }
578 }
580 template<class T, MEMFLAGS F> bool
581 GenericTaskQueueSet<T, F>::steal_best_of_2(uint queue_num, int* seed, E& t) {
582 if (_n > 2) {
583 uint k1 = queue_num;
584 while (k1 == queue_num) k1 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n;
585 uint k2 = queue_num;
586 while (k2 == queue_num || k2 == k1) k2 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n;
587 // Sample both and try the larger.
588 uint sz1 = _queues[k1]->size();
589 uint sz2 = _queues[k2]->size();
590 if (sz2 > sz1) return _queues[k2]->pop_global(t);
591 else return _queues[k1]->pop_global(t);
592 } else if (_n == 2) {
593 // Just try the other one.
594 uint k = (queue_num + 1) % 2;
595 return _queues[k]->pop_global(t);
596 } else {
597 assert(_n == 1, "can't be zero.");
598 return false;
599 }
600 }
602 template<class T, MEMFLAGS F>
603 bool GenericTaskQueueSet<T, F>::peek() {
604 // Try all the queues.
605 for (uint j = 0; j < _n; j++) {
606 if (_queues[j]->peek())
607 return true;
608 }
609 return false;
610 }
612 // When to terminate from the termination protocol.
613 class TerminatorTerminator: public CHeapObj<mtInternal> {
614 public:
615 virtual bool should_exit_termination() = 0;
616 };
618 // A class to aid in the termination of a set of parallel tasks using
619 // TaskQueueSet's for work stealing.
621 #undef TRACESPINNING
623 class ParallelTaskTerminator: public StackObj {
624 private:
625 int _n_threads;
626 TaskQueueSetSuper* _queue_set;
627 int _offered_termination;
629 #ifdef TRACESPINNING
630 static uint _total_yields;
631 static uint _total_spins;
632 static uint _total_peeks;
633 #endif
635 bool peek_in_queue_set();
636 protected:
637 virtual void yield();
638 void sleep(uint millis);
640 public:
642 // "n_threads" is the number of threads to be terminated. "queue_set" is a
643 // queue sets of work queues of other threads.
644 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set);
646 // The current thread has no work, and is ready to terminate if everyone
647 // else is. If returns "true", all threads are terminated. If returns
648 // "false", available work has been observed in one of the task queues,
649 // so the global task is not complete.
650 bool offer_termination() {
651 return offer_termination(NULL);
652 }
654 // As above, but it also terminates if the should_exit_termination()
655 // method of the terminator parameter returns true. If terminator is
656 // NULL, then it is ignored.
657 bool offer_termination(TerminatorTerminator* terminator);
659 // Reset the terminator, so that it may be reused again.
660 // The caller is responsible for ensuring that this is done
661 // in an MT-safe manner, once the previous round of use of
662 // the terminator is finished.
663 void reset_for_reuse();
664 // Same as above but the number of parallel threads is set to the
665 // given number.
666 void reset_for_reuse(int n_threads);
668 #ifdef TRACESPINNING
669 static uint total_yields() { return _total_yields; }
670 static uint total_spins() { return _total_spins; }
671 static uint total_peeks() { return _total_peeks; }
672 static void print_termination_counts();
673 #endif
674 };
676 template<class E, MEMFLAGS F, unsigned int N> inline bool
677 GenericTaskQueue<E, F, N>::push(E t) {
678 uint localBot = _bottom;
679 assert((localBot >= 0) && (localBot < N), "_bottom out of range.");
680 idx_t top = _age.top();
681 uint dirty_n_elems = dirty_size(localBot, top);
682 assert(dirty_n_elems < N, "n_elems out of range.");
683 if (dirty_n_elems < max_elems()) {
684 // g++ complains if the volatile result of the assignment is unused.
685 const_cast<E&>(_elems[localBot] = t);
686 OrderAccess::release_store(&_bottom, increment_index(localBot));
687 TASKQUEUE_STATS_ONLY(stats.record_push());
688 return true;
689 } else {
690 return push_slow(t, dirty_n_elems);
691 }
692 }
694 template<class E, MEMFLAGS F, unsigned int N> inline bool
695 GenericTaskQueue<E, F, N>::pop_local(E& t) {
696 uint localBot = _bottom;
697 // This value cannot be N-1. That can only occur as a result of
698 // the assignment to bottom in this method. If it does, this method
699 // resets the size to 0 before the next call (which is sequential,
700 // since this is pop_local.)
701 uint dirty_n_elems = dirty_size(localBot, _age.top());
702 assert(dirty_n_elems != N - 1, "Shouldn't be possible...");
703 if (dirty_n_elems == 0) return false;
704 localBot = decrement_index(localBot);
705 _bottom = localBot;
706 // This is necessary to prevent any read below from being reordered
707 // before the store just above.
708 OrderAccess::fence();
709 const_cast<E&>(t = _elems[localBot]);
710 // This is a second read of "age"; the "size()" above is the first.
711 // If there's still at least one element in the queue, based on the
712 // "_bottom" and "age" we've read, then there can be no interference with
713 // a "pop_global" operation, and we're done.
714 idx_t tp = _age.top(); // XXX
715 if (size(localBot, tp) > 0) {
716 assert(dirty_size(localBot, tp) != N - 1, "sanity");
717 TASKQUEUE_STATS_ONLY(stats.record_pop());
718 return true;
719 } else {
720 // Otherwise, the queue contained exactly one element; we take the slow
721 // path.
722 return pop_local_slow(localBot, _age.get());
723 }
724 }
726 typedef GenericTaskQueue<oop, mtGC> OopTaskQueue;
727 typedef GenericTaskQueueSet<OopTaskQueue, mtGC> OopTaskQueueSet;
729 #ifdef _MSC_VER
730 #pragma warning(push)
731 // warning C4522: multiple assignment operators specified
732 #pragma warning(disable:4522)
733 #endif
735 // This is a container class for either an oop* or a narrowOop*.
736 // Both are pushed onto a task queue and the consumer will test is_narrow()
737 // to determine which should be processed.
738 class StarTask {
739 void* _holder; // either union oop* or narrowOop*
741 enum { COMPRESSED_OOP_MASK = 1 };
743 public:
744 StarTask(narrowOop* p) {
745 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
746 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK);
747 }
748 StarTask(oop* p) {
749 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
750 _holder = (void*)p;
751 }
752 StarTask() { _holder = NULL; }
753 operator oop*() { return (oop*)_holder; }
754 operator narrowOop*() {
755 return (narrowOop*)((uintptr_t)_holder & ~COMPRESSED_OOP_MASK);
756 }
758 StarTask& operator=(const StarTask& t) {
759 _holder = t._holder;
760 return *this;
761 }
762 volatile StarTask& operator=(const volatile StarTask& t) volatile {
763 _holder = t._holder;
764 return *this;
765 }
767 bool is_narrow() const {
768 return (((uintptr_t)_holder & COMPRESSED_OOP_MASK) != 0);
769 }
770 };
772 class ObjArrayTask
773 {
774 public:
775 ObjArrayTask(oop o = NULL, int idx = 0): _obj(o), _index(idx) { }
776 ObjArrayTask(oop o, size_t idx): _obj(o), _index(int(idx)) {
777 assert(idx <= size_t(max_jint), "too big");
778 }
779 ObjArrayTask(const ObjArrayTask& t): _obj(t._obj), _index(t._index) { }
781 ObjArrayTask& operator =(const ObjArrayTask& t) {
782 _obj = t._obj;
783 _index = t._index;
784 return *this;
785 }
786 volatile ObjArrayTask&
787 operator =(const volatile ObjArrayTask& t) volatile {
788 _obj = t._obj;
789 _index = t._index;
790 return *this;
791 }
793 inline oop obj() const { return _obj; }
794 inline int index() const { return _index; }
796 DEBUG_ONLY(bool is_valid() const); // Tasks to be pushed/popped must be valid.
798 private:
799 oop _obj;
800 int _index;
801 };
803 #ifdef _MSC_VER
804 #pragma warning(pop)
805 #endif
807 typedef OverflowTaskQueue<StarTask, mtClass> OopStarTaskQueue;
808 typedef GenericTaskQueueSet<OopStarTaskQueue, mtClass> OopStarTaskQueueSet;
810 typedef OverflowTaskQueue<size_t, mtInternal> RegionTaskQueue;
811 typedef GenericTaskQueueSet<RegionTaskQueue, mtClass> RegionTaskQueueSet;
814 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP