ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/jsr166/jsr166/src/jsr166y/ForkJoinPool.java

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.58 by dl, Fri Jul 23 13:07:43 2010 UTC vs.
Revision 1.64 by dl, Tue Aug 17 18:30:32 2010 UTC

#	Line 52 | Line 52 | import java.util.concurrent.CountDownLat
52		* convenient form for informal monitoring.
53		*
54		* <p> As is the case with other ExecutorServices, there are three
55	<	* main task execution methods summarized in the follwoing
55	>	* main task execution methods summarized in the following
56		* table. These are designed to be used by clients not already engaged
57		* in fork/join computations in the current pool.  The main forms of
58		* these methods accept instances of {@code ForkJoinTask}, but
#	Line 60 | Line 60 | import java.util.concurrent.CountDownLat
60		* Runnable}- or {@code Callable}- based activities as well.  However,
61		* tasks that are already executing in a pool should normally
62		* <em>NOT</em> use these pool execution methods, but instead use the
63	<	* within-computation forms listed in the table.
63	>	* within-computation forms listed in the table.
64		*
65		* <table BORDER CELLPADDING=3 CELLSPACING=1>
66		*  <tr>
#	Line 84 | Line 84 | import java.util.concurrent.CountDownLat
84		*    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
85		*  </tr>
86		* </table>
87	<	*
87	>	*
88		* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
89		* used for all parallel task execution in a program or subsystem.
90		* Otherwise, use would not usually outweigh the construction and
#	Line 110 | Line 110 | import java.util.concurrent.CountDownLat
110		*
111		* <p>This implementation rejects submitted tasks (that is, by throwing
112		* {@link RejectedExecutionException}) only when the pool is shut down
113	<	* or internal resources have been exhuasted.
113	>	* or internal resources have been exhausted.
114		*
115		* @since 1.7
116		* @author Doug Lea
#	Line 138 | Line 138 | public class ForkJoinPool extends Abstra
138		* cache pollution effects.)
139		*
140		* Beyond work-stealing support and essential bookkeeping, the
141	<	* main responsibility of this framework is to arrange tactics for
142	<	* when one worker is waiting to join a task stolen (or always
143	<	* held by) another.  Becauae we are multiplexing many tasks on to
144	<	* a pool of workers, we can't just let them block (as in
145	<	* Thread.join).  We also cannot just reassign the joiner's
146	<	* run-time stack with another and replace it later, which would
147	<	* be a form of "continuation", that even if possible is not
148	<	* necessarily a good idea. Given that the creation costs of most
149	<	* threads on most systems mainly surrounds setting up runtime
150	<	* stacks, thread creation and switching is usually not much more
151	<	* expensive than stack creation and switching, and is more
152	<	* flexible). Instead we combine two tactics:
141	>	* main responsibility of this framework is to take actions when
142	>	* one worker is waiting to join a task stolen (or always held by)
143	>	* another.  Becauae we are multiplexing many tasks on to a pool
144	>	* of workers, we can't just let them block (as in Thread.join).
145	>	* We also cannot just reassign the joiner's run-time stack with
146	>	* another and replace it later, which would be a form of
147	>	* "continuation", that even if possible is not necessarily a good
148	>	* idea. Given that the creation costs of most threads on most
149	>	* systems mainly surrounds setting up runtime stacks, thread
150	>	* creation and switching is usually not much more expensive than
151	>	* stack creation and switching, and is more flexible). Instead we
152	>	* combine two tactics:
153		*
154	<	*   1. Arranging for the joiner to execute some task that it
154	>	*   Helping: Arranging for the joiner to execute some task that it
155		*      would be running if the steal had not occurred.  Method
156		*      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
157		*      links to try to find such a task.
158		*
159	<	*   2. Unless there are already enough live threads, creating or
160	<	*      or re-activating a spare thread to compensate for the
161	<	*      (blocked) joiner until it unblocks.  Spares then suspend
162	<	*      at their next opportunity or eventually die if unused for
163	<	*      too long.  See below and the internal documentation
164	<	*      for tryAwaitJoin for more details about compensation
165	<	*      rules.
159	>	*   Compensating: Unless there are already enough live threads,
160	>	*      method helpMaintainParallelism() may create or or
161	>	*      re-activate a spare thread to compensate for blocked
162	>	*      joiners until they unblock.
163		*
164		* Because the determining existence of conservatively safe
165		* helping targets, the availability of already-created spares,
166		* and the apparent need to create new spares are all racy and
167	<	* require heuristic guidance, joins (in
168	<	* ForkJoinWorkerThread.joinTask) interleave these options until
169	<	* successful.  Creating a new spare always succeeds, but also
170	<	* increases application footprint, so we try to avoid it, within
171	<	* reason.
167	>	* require heuristic guidance, we rely on multiple retries of
168	>	* each. Further, because it is impossible to keep exactly the
169	>	* target (parallelism) number of threads running at any given
170	>	* time, we allow compensation during joins to fail, and enlist
171	>	* all other threads to help out whenever they are not otherwise
172	>	* occupied (i.e., mainly in method preStep).
173		*
174	<	* The ManagedBlocker extension API can't use option (1) so uses a
175	<	* special version of (2) in method awaitBlocker.
174	>	* The ManagedBlocker extension API can't use helping so relies
175	>	* only on compensation in method awaitBlocker.
176		*
177		* The main throughput advantages of work-stealing stem from
178		* decentralized control -- workers mostly steal tasks from each
#	Line 207 | Line 205 | public class ForkJoinPool extends Abstra
205		* blocked workers. However, all other support code is set up to
206		* work with other policies.
207		*
208	+	* To ensure that we do not hold on to worker references that
209	+	* would prevent GC, ALL accesses to workers are via indices into
210	+	* the workers array (which is one source of some of the unusual
211	+	* code constructions here). In essence, the workers array serves
212	+	* as a WeakReference mechanism. Thus for example the event queue
213	+	* stores worker indices, not worker references. Access to the
214	+	* workers in associated methods (for example releaseEventWaiters)
215	+	* must both index-check and null-check the IDs. All such accesses
216	+	* ignore bad IDs by returning out early from what they are doing,
217	+	* since this can only be associated with shutdown, in which case
218	+	* it is OK to give up. On termination, we just clobber these
219	+	* data structures without trying to use them.
220	+	*
221		* 2. Bookkeeping for dynamically adding and removing workers. We
222		* aim to approximately maintain the given level of parallelism.
223		* When some workers are known to be blocked (on joins or via
#	Line 248 | Line 259 | public class ForkJoinPool extends Abstra
259		* workers that previously could not find a task to now find one:
260		* Submission of a new task to the pool, or another worker pushing
261		* a task onto a previously empty queue.  (We also use this
262	<	* mechanism for termination and reconfiguration actions that
262	>	* mechanism for configuration and termination actions that
263		* require wakeups of idle workers).  Each worker maintains its
264		* last known event count, and blocks when a scan for work did not
265		* find a task AND its lastEventCount matches the current
#	Line 259 | Line 270 | public class ForkJoinPool extends Abstra
270		* a record (field nextEventWaiter) for the next waiting worker.
271		* In addition to allowing simpler decisions about need for
272		* wakeup, the event count bits in eventWaiters serve the role of
273	<	* tags to avoid ABA errors in Treiber stacks.  To reduce delays
274	<	* in task diffusion, workers not otherwise occupied may invoke
275	<	* method releaseWaiters, that removes and signals (unparks)
276	<	* workers not waiting on current count. To minimize task
277	<	* production stalls associate with signalling, any worker pushing
278	<	* a task on an empty queue invokes the weaker method signalWork,
279	<	* that only releases idle workers until it detects interference
269	<	* by other threads trying to release, and lets them take
270	<	* over. The net effect is a tree-like diffusion of signals, where
271	<	* released threads (and possibly others) help with unparks.  To
272	<	* further reduce contention effects a bit, failed CASes to
273	<	* increment field eventCount are tolerated without retries.
273	>	* tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
274	>	* released threads also try to release others (but give up upon
275	>	* contention to reduce useless flailing).  The net effect is a
276	>	* tree-like diffusion of signals, where released threads (and
277	>	* possibly others) help with unparks.  To further reduce
278	>	* contention effects a bit, failed CASes to increment field
279	>	* eventCount are tolerated without retries in signalWork.
280		* Conceptually they are merged into the same event, which is OK
281		* when their only purpose is to enable workers to scan for work.
282		*
#	Line 285 | Line 291 | public class ForkJoinPool extends Abstra
291		* spare threads from normal "core" threads: On each call to
292		* preStep (the only point at which we can do this) a worker
293		* checks to see if there are now too many running workers, and if
294	<	* so, suspends itself.  Methods tryAwaitJoin and awaitBlocker
295	<	* look for suspended threads to resume before considering
296	<	* creating a new replacement. We don't need a special data
297	<	* structure to maintain spares; simply scanning the workers array
298	<	* looking for worker.isSuspended() is fine because the calling
299	<	* thread is otherwise not doing anything useful anyway; we are at
300	<	* least as happy if after locating a spare, the caller doesn't
301	<	* actually block because the join is ready before we try to
302	<	* adjust and compensate.  Note that this is intrinsically racy.
303	<	* One thread may become a spare at about the same time as another
304	<	* is needlessly being created. We counteract this and related
305	<	* slop in part by requiring resumed spares to immediately recheck
306	<	* (in preStep) to see whether they they should re-suspend. The
307	<	* only effective difference between "extra" and "core" threads is
308	<	* that we allow the "extra" ones to time out and die if they are
309	<	* not resumed within a keep-alive interval of a few seconds. This
310	<	* is implemented mainly within ForkJoinWorkerThread, but requires
311	<	* some coordination (isTrimmed() -- meaning killed while
312	<	* suspended) to correctly maintain pool counts.
313	<	*
314	<	* 6. Deciding when to create new workers. The main dynamic
315	<	* control in this class is deciding when to create extra threads,
316	<	* in methods awaitJoin and awaitBlocker. We always need to create
317	<	* one when the number of running threads would become zero and
318	<	* all workers are busy. However, this is not easy to detect
319	<	* reliably in the presence of transients so we use retries and
320	<	* allow slack (in tryAwaitJoin) to reduce false alarms.  These
321	<	* effectively reduce churn at the price of systematically
322	<	* undershooting target parallelism when many threads are blocked.
323	<	* However, biasing toward undeshooting partially compensates for
324	<	* the above mechanics to suspend extra threads, that normally
325	<	* lead to overshoot because we can only suspend workers
326	<	* in-between top-level actions. It also better copes with the
327	<	* fact that some of the methods in this class tend to never
328	<	* become compiled (but are interpreted), so some components of
329	<	* the entire set of controls might execute many times faster than
330	<	* others. And similarly for cases where the apparent lack of work
331	<	* is just due to GC stalls and other transient system activity.
294	>	* so, suspends itself.  Method helpMaintainParallelism looks for
295	>	* suspended threads to resume before considering creating a new
296	>	* replacement. The spares themselves are encoded on another
297	>	* variant of a Treiber Stack, headed at field "spareWaiters".
298	>	* Note that the use of spares is intrinsically racy.  One thread
299	>	* may become a spare at about the same time as another is
300	>	* needlessly being created. We counteract this and related slop
301	>	* in part by requiring resumed spares to immediately recheck (in
302	>	* preStep) to see whether they they should re-suspend.
303	>	*
304	>	* 6. Killing off unneeded workers. The Spare and Event queues use
305	>	* similar mechanisms to shed unused workers: The oldest (first)
306	>	* waiter uses a timed rather than hard wait. When this wait times
307	>	* out without a normal wakeup, it tries to shutdown any one (for
308	>	* convenience the newest) other waiter via tryShutdownSpare or
309	>	* tryShutdownWaiter, respectively. The wakeup rates for spares
310	>	* are much shorter than for waiters. Together, they will
311	>	* eventually reduce the number of worker threads to a minimum of
312	>	* one after a long enough period without use.
313	>	*
314	>	* 7. Deciding when to create new workers. The main dynamic
315	>	* control in this class is deciding when to create extra threads
316	>	* in method helpMaintainParallelism. We would like to keep
317	>	* exactly #parallelism threads running, which is an impossble
318	>	* task. We always need to create one when the number of running
319	>	* threads would become zero and all workers are busy. Beyond
320	>	* this, we must rely on heuristics that work well in the the
321	>	* presence of transients phenomena such as GC stalls, dynamic
322	>	* compilation, and wake-up lags. These transients are extremely
323	>	* common -- we are normally trying to fully saturate the CPUs on
324	>	* a machine, so almost any activity other than running tasks
325	>	* impedes accuracy. Our main defense is to allow some slack in
326	>	* creation thresholds, using rules that reflect the fact that the
327	>	* more threads we have running, the more likely that we are
328	>	* underestimating the number running threads. (We also include
329	>	* some heuristic use of Thread.yield when all workers appear to
330	>	* be busy, to improve likelihood of counts settling.) The rules
331	>	* also better cope with the fact that some of the methods in this
332	>	* class tend to never become compiled (but are interpreted), so
333	>	* some components of the entire set of controls might execute 100
334	>	* times faster than others. And similarly for cases where the
335	>	* apparent lack of work is just due to GC stalls and other
336	>	* transient system activity.
337		*
338		* Beware that there is a lot of representation-level coupling
339		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 335 | Line 346 | public class ForkJoinPool extends Abstra
346		*
347		* Style notes: There are lots of inline assignments (of form
348		* "while ((local = field) != 0)") which are usually the simplest
349	<	* way to ensure read orderings. Also several occurrences of the
350	<	* unusual "do {} while(!cas...)" which is the simplest way to
351	<	* force an update of a CAS'ed variable. There are also other
352	<	* coding oddities that help some methods perform reasonably even
353	<	* when interpreted (not compiled), at the expense of messiness.
349	>	* way to ensure the required read orderings (which are sometimes
350	>	* critical). Also several occurrences of the unusual "do {}
351	>	* while(!cas...)" which is the simplest way to force an update of
352	>	* a CAS'ed variable. There are also other coding oddities that
353	>	* help some methods perform reasonably even when interpreted (not
354	>	* compiled), at the expense of some messy constructions that
355	>	* reduce byte code counts.
356		*
357		* The order of declarations in this file is: (1) statics (2)
358		* fields (along with constants used when unpacking some of them)
#	Line 407 | Line 420 | public class ForkJoinPool extends Abstra
420		new AtomicInteger();
421
422		/**
423	<	* Absolute bound for parallelism level. Twice this number must
424	<	* fit into a 16bit field to enable word-packing for some counts.
423	>	* The wakeup interval (in nanoseconds) for the oldest worker
424	>	* worker waiting for an event invokes tryShutdownWaiter to shrink
425	>	* the number of workers.  The exact value does not matter too
426	>	* much, but should be long enough to slowly release resources
427	>	* during long periods without use without disrupting normal use.
428	>	*/
429	>	private static final long SHRINK_RATE_NANOS =
430	>	60L * 1000L * 1000L * 1000L; // one minute
431	>
432	>	/**
433	>	* Absolute bound for parallelism level. Twice this number plus
434	>	* one (i.e., 0xfff) must fit into a 16bit field to enable
435	>	* word-packing for some counts and indices.
436		*/
437	<	private static final int MAX_THREADS = 0x7fff;
437	>	private static final int MAX_WORKERS   = 0x7fff;
438
439		/**
440		* Array holding all worker threads in the pool.  Array size must
#	Line 452 | Line 476 | public class ForkJoinPool extends Abstra
476		/**
477		* Encoded record of top of treiber stack of threads waiting for
478		* events. The top 32 bits contain the count being waited for. The
479	<	* bottom word contains one plus the pool index of waiting worker
480	<	* thread.
479	>	* bottom 16 bits contains one plus the pool index of waiting
480	>	* worker thread. (Bits 16-31 are unused.)
481		*/
482		private volatile long eventWaiters;
483
484		private static final int  EVENT_COUNT_SHIFT = 32;
485	<	private static final long WAITER_ID_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
485	>	private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
486
487		/**
488		* A counter for events that may wake up worker threads:
489		*   - Submission of a new task to the pool
490		*   - A worker pushing a task on an empty queue
491	<	*   - termination and reconfiguration
491	>	*   - termination
492		*/
493		private volatile int eventCount;
494
495		/**
496	+	* Encoded record of top of treiber stack of spare threads waiting
497	+	* for resumption. The top 16 bits contain an arbitrary count to
498	+	* avoid ABA effects. The bottom 16bits contains one plus the pool
499	+	* index of waiting worker thread.
500	+	*/
501	+	private volatile int spareWaiters;
502	+
503	+	private static final int SPARE_COUNT_SHIFT = 16;
504	+	private static final int SPARE_ID_MASK     = (1 << 16) - 1;
505	+
506	+	/**
507		* Lifecycle control. The low word contains the number of workers
508		* that are (probably) executing tasks. This value is atomically
509		* incremented before a worker gets a task to run, and decremented
#	Line 479 | Line 514 | public class ForkJoinPool extends Abstra
514		* These are bundled together to ensure consistent read for
515		* termination checks (i.e., that runLevel is at least SHUTDOWN
516		* and active threads is zero).
517	+	*
518	+	* Notes: Most direct CASes are dependent on these bitfield
519	+	* positions.  Also, this field is non-private to enable direct
520	+	* performance-sensitive CASes in ForkJoinWorkerThread.
521		*/
522	<	private volatile int runState;
522	>	volatile int runState;
523
524		// Note: The order among run level values matters.
525		private static final int RUNLEVEL_SHIFT     = 16;
#	Line 488 | Line 527 | public class ForkJoinPool extends Abstra
527		private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
528		private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
529		private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
491	–	private static final int ONE_ACTIVE         = 1; // active update delta
530
531		/**
532		* Holds number of total (i.e., created and not yet terminated)
#	Line 497 | Line 535 | public class ForkJoinPool extends Abstra
535		* making decisions about creating and suspending spare
536		* threads. Updated only by CAS. Note that adding a new worker
537		* requires incrementing both counts, since workers start off in
538	<	* running state.  This field is also used for memory-fencing
501	<	* configuration parameters.
538	>	* running state.
539		*/
540		private volatile int workerCounts;
541
#	Line 530 | Line 567 | public class ForkJoinPool extends Abstra
567		*/
568		private final int poolNumber;
569
570	<	// Utilities for CASing fields. Note that several of these
571	<	// are manually inlined by callers
570	>
571	>	// Utilities for CASing fields. Note that most of these
572	>	// are usually manually inlined by callers
573
574		/**
575	<	* Increments running count.  Also used by ForkJoinTask.
575	>	* Increments running count part of workerCounts
576		*/
577		final void incrementRunningCount() {
578		int c;
579		do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
580	<	c = workerCounts,
580	>	c = workerCounts,
581		c + ONE_RUNNING));
582		}
583
#	Line 555 | Line 593 | public class ForkJoinPool extends Abstra
593		}
594
595		/**
596	<	* Tries to increment running count
596	>	* Forces decrement of encoded workerCounts, awaiting nonzero if
597	>	* (rarely) necessary when other count updates lag.
598	>	*
599	>	* @param dr -- either zero or ONE_RUNNING
600	>	* @param dt == either zero or ONE_TOTAL
601		*/
602	<	final boolean tryIncrementRunningCount() {
603	<	int wc;
604	<	return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
605	<	wc = workerCounts, wc + ONE_RUNNING);
602	>	private void decrementWorkerCounts(int dr, int dt) {
603	>	for (;;) {
604	>	int wc = workerCounts;
605	>	if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
606	>	(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
607	>	if ((runState & TERMINATED) != 0)
608	>	return; // lagging termination on a backout
609	>	Thread.yield();
610	>	}
611	>	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
612	>	wc, wc - (dr + dt)))
613	>	return;
614	>	}
615	>	}
616	>
617	>	/**
618	>	* Increments event count
619	>	*/
620	>	private void advanceEventCount() {
621	>	int c;
622	>	do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset,
623	>	c = eventCount, c+1));
624		}
625
626		/**
#	Line 572 | Line 632 | public class ForkJoinPool extends Abstra
632		final boolean tryIncrementActiveCount() {
633		int c;
634		return UNSAFE.compareAndSwapInt(this, runStateOffset,
635	<	c = runState, c + ONE_ACTIVE);
635	>	c = runState, c + 1);
636		}
637
638		/**
#	Line 582 | Line 642 | public class ForkJoinPool extends Abstra
642		final boolean tryDecrementActiveCount() {
643		int c;
644		return UNSAFE.compareAndSwapInt(this, runStateOffset,
645	<	c = runState, c - ONE_ACTIVE);
645	>	c = runState, c - 1);
646		}
647
648		/**
#	Line 611 | Line 671 | public class ForkJoinPool extends Abstra
671		lock.lock();
672		try {
673		ForkJoinWorkerThread[] ws = workers;
674	<	int nws = ws.length;
675	<	if (k < 0 || k >= nws || ws[k] != null) {
676	<	for (k = 0; k < nws && ws[k] != null; ++k)
674	>	int n = ws.length;
675	>	if (k < 0 || k >= n || ws[k] != null) {
676	>	for (k = 0; k < n && ws[k] != null; ++k)
677		;
678	<	if (k == nws)
679	<	ws = Arrays.copyOf(ws, nws << 1);
678	>	if (k == n)
679	>	ws = Arrays.copyOf(ws, n << 1);
680		}
681		ws[k] = w;
682		workers = ws; // volatile array write ensures slot visibility
#	Line 650 | Line 710 | public class ForkJoinPool extends Abstra
710		* are already updated to accommodate the worker, so adjusts on
711		* failure.
712		*
713	<	* @return new worker or null if creation failed
713	>	* @return the worker, or null on failure
714		*/
715		private ForkJoinWorkerThread addWorker() {
716		ForkJoinWorkerThread w = null;
#	Line 658 | Line 718 | public class ForkJoinPool extends Abstra
718		w = factory.newThread(this);
719		} finally { // Adjust on either null or exceptional factory return
720		if (w == null) {
721	<	onWorkerCreationFailure();
722	<	return null;
721	>	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
722	>	tryTerminate(false); // in case of failure during shutdown
723		}
724		}
725	<	w.start(recordWorker(w), ueh);
725	>	if (w != null) {
726	>	w.start(recordWorker(w), ueh);
727	>	advanceEventCount();
728	>	}
729		return w;
730		}
731
732		/**
733	<	* Adjusts counts upon failure to create worker
733	>	* Final callback from terminating worker.  Removes record of
734	>	* worker from array, and adjusts counts. If pool is shutting
735	>	* down, tries to complete terminatation.
736	>	*
737	>	* @param w the worker
738		*/
739	<	private void onWorkerCreationFailure() {
740	<	for (;;) {
741	<	int wc = workerCounts;
742	<	if ((wc >>> TOTAL_COUNT_SHIFT) == 0)
743	<	Thread.yield(); // wait for other counts to settle
744	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
745	<	wc - (ONE_RUNNING|ONE_TOTAL)))
739	>	final void workerTerminated(ForkJoinWorkerThread w) {
740	>	forgetWorker(w);
741	>	decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
742	>	while (w.stealCount != 0) // collect final count
743	>	tryAccumulateStealCount(w);
744	>	tryTerminate(false);
745	>	}
746	>
747	>	// Waiting for and signalling events
748	>
749	>	/**
750	>	* Releases workers blocked on a count not equal to current count.
751	>	* Normally called after precheck that eventWaiters isn't zero to
752	>	* avoid wasted array checks. Gives up upon a change in count or
753	>	* contention, letting other workers take over.
754	>	*/
755	>	private void releaseEventWaiters() {
756	>	ForkJoinWorkerThread[] ws = workers;
757	>	int n = ws.length;
758	>	long h = eventWaiters;
759	>	int ec = eventCount;
760	>	ForkJoinWorkerThread w; int id;
761	>	while ((int)(h >>> EVENT_COUNT_SHIFT) != ec &&
762	>	(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
763	>	id < n && (w = ws[id]) != null &&
764	>	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
765	>	h,  h = w.nextWaiter)) {
766	>	LockSupport.unpark(w);
767	>	if (eventWaiters != h || eventCount != ec)
768		break;
769		}
681	–	tryTerminate(false); // in case of failure during shutdown
770		}
771
772		/**
773	<	* Creates and/or resumes enough workers to establish target
774	<	* parallelism, giving up if terminating or addWorker fails
773	>	* Tries to advance eventCount and releases waiters. Called only
774	>	* from workers.
775	>	*/
776	>	final void signalWork() {
777	>	int c; // try to increment event count -- CAS failure OK
778	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
779	>	if (eventWaiters != 0L)
780	>	releaseEventWaiters();
781	>	}
782	>
783	>	/**
784	>	* Adds the given worker to event queue and blocks until
785	>	* terminating or event count advances from the workers
786	>	* lastEventCount value
787		*
788	<	* TODO: recast this to support lazier creation and automated
689	<	* parallelism maintenance
788	>	* @param w the calling worker thread
789		*/
790	<	private void ensureEnoughWorkers() {
791	<	for (;;) {
792	<	int pc = parallelism;
793	<	int wc = workerCounts;
794	<	int rc = wc & RUNNING_COUNT_MASK;
795	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
796	<	if (tc < pc) {
797	<	if (runState == TERMINATING ||
798	<	(UNSAFE.compareAndSwapInt
799	<	(this, workerCountsOffset,
800	<	wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
702	<	addWorker() == null))
703	<	break;
704	<	}
705	<	else if (tc > pc && rc < pc &&
706	<	tc > (runState & ACTIVE_COUNT_MASK)) {
707	<	ForkJoinWorkerThread spare = null;
708	<	ForkJoinWorkerThread[] ws = workers;
709	<	int nws = ws.length;
710	<	for (int i = 0; i < nws; ++i) {
711	<	ForkJoinWorkerThread w = ws[i];
712	<	if (w != null && w.isSuspended()) {
713	<	if ((workerCounts & RUNNING_COUNT_MASK) > pc ||
714	<	runState == TERMINATING)
715	<	return;
716	<	if (w.tryResumeSpare())
717	<	incrementRunningCount();
718	<	break;
719	<	}
720	<	}
721	<	}
722	<	else
790	>	private void eventSync(ForkJoinWorkerThread w) {
791	>	int ec = w.lastEventCount;
792	>	long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
793	>	long h;
794	>	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
795	>	(((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
796	>	(int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
797	>	eventCount == ec) {
798	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
799	>	w.nextWaiter = h, nh)) {
800	>	awaitEvent(w, ec);
801		break;
802	+	}
803		}
804		}
805
806		/**
807	<	* Final callback from terminating worker.  Removes record of
808	<	* worker from array, and adjusts counts. If pool is shutting
809	<	* down, tries to complete terminatation, else possibly replaces
810	<	* the worker.
807	>	* Blocks the given worker (that has already been entered as an
808	>	* event waiter) until terminating or event count advances from
809	>	* the given value. The oldest (first) waiter uses a timed wait to
810	>	* occasionally one-by-one shrink the number of workers (to a
811	>	* minumum of one) if the pool has not been used for extended
812	>	* periods.
813		*
814	<	* @param w the worker
814	>	* @param w the calling worker thread
815	>	* @param ec the count
816		*/
817	<	final void workerTerminated(ForkJoinWorkerThread w) {
818	<	if (w.active) { // force inactive
819	<	w.active = false;
820	<	do {} while (!tryDecrementActiveCount());
817	>	private void awaitEvent(ForkJoinWorkerThread w, int ec) {
818	>	while (eventCount == ec) {
819	>	if (tryAccumulateStealCount(w)) { // transfer while idle
820	>	boolean untimed = (w.nextWaiter != 0L ||
821	>	(workerCounts & RUNNING_COUNT_MASK) <= 1);
822	>	long startTime = untimed? 0 : System.nanoTime();
823	>	Thread.interrupted();         // clear/ignore interrupt
824	>	if (eventCount != ec || !w.isRunning() ||
825	>	runState >= TERMINATING)  // recheck after clear
826	>	break;
827	>	if (untimed)
828	>	LockSupport.park(w);
829	>	else {
830	>	LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
831	>	if (eventCount != ec || !w.isRunning() ||
832	>	runState >= TERMINATING)
833	>	break;
834	>	if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
835	>	tryShutdownWaiter(ec);
836	>	}
837	>	}
838		}
839	<	forgetWorker(w);
839	>	}
840
841	<	// Decrement total count, and if was running, running count
842	<	// Spin (waiting for other updates) if either would be negative
843	<	int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
844	<	int unit = ONE_TOTAL + nr;
845	<	for (;;) {
846	<	int wc = workerCounts;
847	<	int rc = wc & RUNNING_COUNT_MASK;
848	<	if (rc - nr < 0 || (wc >>> TOTAL_COUNT_SHIFT) == 0)
849	<	Thread.yield(); // back off if waiting for other updates
850	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
851	<	wc, wc - unit))
852	<	break;
841	>	/**
842	>	* Callback from the oldest waiter in awaitEvent waking up after a
843	>	* period of non-use. Tries (once) to shutdown an event waiter (or
844	>	* a spare, if one exists). Note that we don't need CAS or locks
845	>	* here because the method is called only from one thread
846	>	* occasionally waking (and even misfires are OK). Note that
847	>	* until the shutdown worker fully terminates, workerCounts
848	>	* will overestimate total count, which is tolerable.
849	>	*
850	>	* @param ec the event count waited on by caller (to abort
851	>	* attempt if count has since changed).
852	>	*/
853	>	private void tryShutdownWaiter(int ec) {
854	>	if (spareWaiters != 0) { // prefer killing spares
855	>	tryShutdownSpare();
856	>	return;
857		}
858	<
859	<	accumulateStealCount(w); // collect final count
860	<	if (!tryTerminate(false))
861	<	ensureEnoughWorkers();
858	>	ForkJoinWorkerThread[] ws = workers;
859	>	int n = ws.length;
860	>	long h = eventWaiters;
861	>	ForkJoinWorkerThread w; int id; long nh;
862	>	if (runState == 0 &&
863	>	submissionQueue.isEmpty() &&
864	>	eventCount == ec &&
865	>	(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
866	>	id < n && (w = ws[id]) != null &&
867	>	(nh = w.nextWaiter) != 0L && // keep at least one worker
868	>	UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) {
869	>	w.shutdown();
870	>	LockSupport.unpark(w);
871	>	}
872	>	releaseEventWaiters();
873		}
874
875	<	// Waiting for and signalling events
875	>	// Maintaining spares
876
877		/**
878	<	* Releases workers blocked on a count not equal to current count.
765	<	* @return true if any released
878	>	* Pushes worker onto the spare stack
879		*/
880	<	private void releaseWaiters() {
881	<	long top;
882	<	while ((top = eventWaiters) != 0L) {
883	<	ForkJoinWorkerThread[] ws = workers;
771	<	int n = ws.length;
772	<	for (;;) {
773	<	int i = ((int)(top & WAITER_ID_MASK)) - 1;
774	<	if (i < 0 || (int)(top >>> EVENT_COUNT_SHIFT) == eventCount)
775	<	return;
776	<	ForkJoinWorkerThread w;
777	<	if (i < n && (w = ws[i]) != null &&
778	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
779	<	top, w.nextWaiter)) {
780	<	LockSupport.unpark(w);
781	<	top = eventWaiters;
782	<	}
783	<	else
784	<	break;      // possibly stale; reread
785	<	}
786	<	}
880	>	final void pushSpare(ForkJoinWorkerThread w) {
881	>	int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
882	>	do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
883	>	w.nextSpare = spareWaiters,ns));
884		}
885
886		/**
887	<	* Ensures eventCount on exit is different (mod 2^32) than on
888	<	* entry and wakes up all waiters
887	>	* Callback from oldest spare occasionally waking up.  Tries
888	>	* (once) to shutdown a spare. Same idea as tryShutdownWaiter.
889		*/
890	<	private void signalEvent() {
891	<	int c;
892	<	do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
893	<	c = eventCount, c+1));
894	<	releaseWaiters();
890	>	final void tryShutdownSpare() {
891	>	int sw, id;
892	>	ForkJoinWorkerThread w;
893	>	ForkJoinWorkerThread[] ws;
894	>	if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
895	>	id < (ws = workers).length && (w = ws[id]) != null &&
896	>	(workerCounts & RUNNING_COUNT_MASK) >= parallelism &&
897	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
898	>	sw, w.nextSpare)) {
899	>	w.shutdown();
900	>	LockSupport.unpark(w);
901	>	advanceEventCount();
902	>	}
903		}
904
905		/**
906	<	* Advances eventCount and releases waiters until interference by
907	<	* other releasing threads is detected.
906	>	* Tries (once) to resume a spare if worker counts match
907	>	* the given count.
908	>	*
909	>	* @param wc workerCounts value on invocation of this method
910		*/
911	<	final void signalWork() {
912	<	int c;
913	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, c=eventCount, c+1);
914	<	long top;
915	<	while ((top = eventWaiters) != 0L) {
916	<	int ec = eventCount;
917	<	ForkJoinWorkerThread[] ws = workers;
918	<	int n = ws.length;
919	<	for (;;) {
920	<	int i = ((int)(top & WAITER_ID_MASK)) - 1;
921	<	if (i < 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec)
922	<	return;
923	<	ForkJoinWorkerThread w;
924	<	if (i < n && (w = ws[i]) != null &&
925	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
926	<	top, top = w.nextWaiter)) {
911	>	private void tryResumeSpare(int wc) {
912	>	ForkJoinWorkerThread[] ws = workers;
913	>	int n = ws.length;
914	>	int sw, id, rs;  ForkJoinWorkerThread w;
915	>	if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
916	>	id < n && (w = ws[id]) != null &&
917	>	(rs = runState) < TERMINATING &&
918	>	eventWaiters == 0L && workerCounts == wc) {
919	>	// In case all workers busy, heuristically back off to let settle
920	>	Thread.yield();
921	>	if (eventWaiters == 0L && runState == rs && // recheck
922	>	workerCounts == wc && spareWaiters == sw &&
923	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
924	>	sw, w.nextSpare)) {
925	>	int c;              // increment running count before resume
926	>	do {} while(!UNSAFE.compareAndSwapInt
927	>	(this, workerCountsOffset,
928	>	c = workerCounts, c + ONE_RUNNING));
929	>	if (w.tryUnsuspend())
930		LockSupport.unpark(w);
931	<	if (top != eventWaiters) // let someone else take over
932	<	return;
823	<	}
824	<	else
825	<	break;      // possibly stale; reread
931	>	else               // back out if w was shutdown
932	>	decrementWorkerCounts(ONE_RUNNING, 0);
933		}
934		}
935		}
936
937	+	// adding workers on demand
938	+
939		/**
940	<	* If worker is inactive, blocks until terminating or event count
941	<	* advances from last value held by worker; in any case helps
833	<	* release others.
834	<	*
835	<	* @param w the calling worker thread
836	<	* @param retries the number of scans by caller failing to find work
837	<	* @return false if now too many threads running
940	>	* Adds one or more workers if needed to establish target parallelism.
941	>	* Retries upon contention.
942		*/
943	<	private boolean eventSync(ForkJoinWorkerThread w, int retries) {
944	<	int wec = w.lastEventCount;
945	<	if (retries > 1) { // can only block after 2nd miss
946	<	long nextTop = (((long)wec << EVENT_COUNT_SHIFT) |
947	<	((long)(w.poolIndex + 1)));
948	<	long top;
949	<	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
950	<	(((int)(top = eventWaiters) & WAITER_ID_MASK) == 0 ||
847	<	(int)(top >>> EVENT_COUNT_SHIFT) == wec) &&
848	<	eventCount == wec) {
849	<	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
850	<	w.nextWaiter = top, nextTop)) {
851	<	accumulateStealCount(w); // transfer steals while idle
852	<	Thread.interrupted();    // clear/ignore interrupt
853	<	while (eventCount == wec)
854	<	w.doPark();
943	>	private void addWorkerIfBelowTarget() {
944	>	int pc = parallelism;
945	>	int wc;
946	>	while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < pc &&
947	>	runState < TERMINATING) {
948	>	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
949	>	wc + (ONE_RUNNING|ONE_TOTAL))) {
950	>	if (addWorker() == null)
951		break;
856	–	}
952		}
858	–	wec = eventCount;
953		}
954	<	releaseWaiters();
955	<	int wc = workerCounts;
956	<	if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
957	<	w.lastEventCount = wec;
958	<	return true;
954	>	}
955	>
956	>	/**
957	>	* Tries (once) to add a new worker if all existing workers are
958	>	* busy, and there are either no running workers or the deficit is
959	>	* at least twice the surplus.
960	>	*
961	>	* @param wc workerCounts value on invocation of this method
962	>	*/
963	>	private void tryAddWorkerIfBusy(int wc) {
964	>	int tc, rc, rs;
965	>	int pc = parallelism;
966	>	if ((tc = wc >>> TOTAL_COUNT_SHIFT) < MAX_WORKERS &&
967	>	((rc = wc & RUNNING_COUNT_MASK) == 0 ||
968	>	rc < pc - ((tc - pc) << 1)) &&
969	>	(rs = runState) < TERMINATING &&
970	>	(rs & ACTIVE_COUNT_MASK) == tc) {
971	>	// Since all workers busy, heuristically back off to let settle
972	>	Thread.yield();
973	>	if (eventWaiters == 0L && spareWaiters == 0 && // recheck
974	>	runState == rs && workerCounts == wc &&
975	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
976	>	wc + (ONE_RUNNING|ONE_TOTAL)))
977	>	addWorker();
978	>	}
979	>	}
980	>
981	>	/**
982	>	* Does at most one of:
983	>	*
984	>	* 1. Help wake up existing workers waiting for work via
985	>	*    releaseEventWaiters. (If any exist, then it doesn't
986	>	*    matter right now if under target parallelism level.)
987	>	*
988	>	* 2. If a spare exists, try (once) to resume it via tryResumeSpare.
989	>	*
990	>	* 3. If there are not enough total workers, add some
991	>	*    via addWorkerIfBelowTarget;
992	>	*
993	>	* 4. Try (once) to add a new worker if all existing workers
994	>	*     are busy, via tryAddWorkerIfBusy
995	>	*/
996	>	private void helpMaintainParallelism() {
997	>	long h; int pc, wc;
998	>	if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0) {
999	>	if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1000	>	releaseEventWaiters(); // avoid useless call
1001	>	}
1002	>	else if ((pc = parallelism) >
1003	>	((wc = workerCounts) & RUNNING_COUNT_MASK)) {
1004	>	if (spareWaiters != 0)
1005	>	tryResumeSpare(wc);
1006	>	else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
1007	>	addWorkerIfBelowTarget();
1008	>	else
1009	>	tryAddWorkerIfBusy(wc);
1010		}
866	–	if (wec != w.lastEventCount) // back up if may re-wait
867	–	w.lastEventCount = wec - (wc >>> TOTAL_COUNT_SHIFT);
868	–	return false;
1011		}
1012
1013		/**
1014		* Callback from workers invoked upon each top-level action (i.e.,
1015	<	* stealing a task or taking a submission and running
1016	<	* it). Performs one or both of the following:
1015	>	* stealing a task or taking a submission and running it).
1016	>	* Performs one or more of the following:
1017		*
1018	<	* * If the worker cannot find work, updates its active status to
1019	<	* inactive and updates activeCount unless there is contention, in
1020	<	* which case it may try again (either in this or a subsequent
1021	<	* call).  Additionally, awaits the next task event and/or helps
1022	<	* wake up other releasable waiters.
1023	<	*
1024	<	* * If there are too many running threads, suspends this worker
1025	<	* (first forcing inactivation if necessary).  If it is not
1026	<	* resumed before a keepAlive elapses, the worker may be "trimmed"
1027	<	* -- killed while suspended within suspendAsSpare. Otherwise,
1028	<	* upon resume it rechecks to make sure that it is still needed.
1018	>	* 1. If the worker is active, try to set its active status to
1019	>	*    inactive and update activeCount. On contention, we may try
1020	>	*    again on this or subsequent call.
1021	>	*
1022	>	* 2. Release any existing event waiters that are now relesable
1023	>	*
1024	>	* 3. If there are too many running threads, suspend this worker
1025	>	*    (first forcing inactive if necessary).  If it is not
1026	>	*    needed, it may be killed while suspended via
1027	>	*    tryShutdownSpare. Otherwise, upon resume it rechecks to make
1028	>	*    sure that it is still needed.
1029	>	*
1030	>	* 4. If more than 1 miss, await the next task event via
1031	>	*    eventSync (first forcing inactivation if necessary), upon
1032	>	*    which worker may also be killed, via tryShutdownWaiter.
1033	>	*
1034	>	* 5. Help reactivate other workers via helpMaintainParallelism
1035		*
1036		* @param w the worker
1037	<	* @param retries the number of scans by caller failing to find work
1038	<	* find any (in which case it may block waiting for work).
1037	>	* @param misses the number of scans by caller failing to find work
1038	>	* (saturating at 2 to avoid wraparound)
1039		*/
1040	<	final void preStep(ForkJoinWorkerThread w, int retries) {
1040	>	final void preStep(ForkJoinWorkerThread w, int misses) {
1041		boolean active = w.active;
1042	<	boolean inactivate = active && retries != 0;
1042	>	int pc = parallelism;
1043		for (;;) {
1044	<	int rs, wc;
1045	<	if (inactivate &&
1046	<	UNSAFE.compareAndSwapInt(this, runStateOffset,
1047	<	rs = runState, rs - ONE_ACTIVE))
1048	<	inactivate = active = w.active = false;
1049	<	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= parallelism) {
1050	<	if (active || eventSync(w, retries))
1051	<	break;
1044	>	int rs, wc, rc, ec; long h;
1045	>	if (active && UNSAFE.compareAndSwapInt(this, runStateOffset,
1046	>	rs = runState, rs - 1))
1047	>	active = w.active = false;
1048	>	if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0 &&
1049	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) {
1050	>	releaseEventWaiters();
1051	>	if (misses > 1)
1052	>	continue;                  // clear before sync below
1053		}
1054	<	else if (!(inactivate |= active) &&  // must inactivate to suspend
1055	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1056	<	wc, wc - ONE_RUNNING) &&
1057	<	!w.suspendAsSpare())             // false if trimmed
1058	<	break;
1054	>	if ((rc = ((wc = workerCounts) & RUNNING_COUNT_MASK)) > pc) {
1055	>	if (!active &&                 // must inactivate to suspend
1056	>	workerCounts == wc &&      // try to suspend as spare
1057	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1058	>	wc, wc - ONE_RUNNING)) {
1059	>	w.suspendAsSpare();
1060	>	if (!w.isRunning())
1061	>	break;                 // was killed while spare
1062	>	}
1063	>	continue;
1064	>	}
1065	>	if (misses > 0) {
1066	>	if ((ec = eventCount) == w.lastEventCount && misses > 1) {
1067	>	if (!active) {             // must inactivate to sync
1068	>	eventSync(w);
1069	>	if (w.isRunning())
1070	>	misses = 1;        // don't re-sync
1071	>	else
1072	>	break;             // was killed while waiting
1073	>	}
1074	>	continue;
1075	>	}
1076	>	w.lastEventCount = ec;
1077	>	}
1078	>	if (rc < pc)
1079	>	helpMaintainParallelism();
1080	>	break;
1081		}
1082		}
1083
1084		/**
1085	<	* Awaits join of the given task if enough threads, or can resume
1086	<	* or create a spare. Fails (in which case the given task might
1087	<	* not be done) upon contention or lack of decision about
1088	<	* blocking. Returns void because caller must check
1089	<	* task status on return anyway.
919	<	*
920	<	* We allow blocking if:
921	<	*
922	<	* 1. There would still be at least as many running threads as
923	<	*    parallelism level if this thread blocks.
924	<	*
925	<	* 2. A spare is resumed to replace this worker. We tolerate
926	<	*    slop in the decision to replace if a spare is found without
927	<	*    first decrementing run count.  This may release too many,
928	<	*    but if so, the superfluous ones will re-suspend via
929	<	*    preStep().
930	<	*
931	<	* 3. After #spares repeated checks, there are no fewer than #spare
932	<	*    threads not running. We allow this slack to avoid hysteresis
933	<	*    and as a hedge against lag/uncertainty of running count
934	<	*    estimates when signalling or unblocking stalls.
935	<	*
936	<	* 4. All existing workers are busy (as rechecked via repeated
937	<	*    retries by caller) and a new spare is created.
938	<	*
939	<	* If none of the above hold, we try to escape out by
940	<	* re-incrementing count and returning to caller, which can retry
941	<	* later.
1085	>	* Helps and/or blocks awaiting join of the given task.
1086	>	* Alternates between helpJoinTask() and helpMaintainParallelism()
1087	>	* as many times as there is a deficit in running count (or longer
1088	>	* if running count would become zero), then blocks if task still
1089	>	* not done.
1090		*
1091		* @param joinMe the task to join
944	–	* @param retries if negative, then serve only as a precheck
945	–	*   that the thread can be replaced by a spare. Otherwise,
946	–	*   the number of repeated calls to this method returning busy
947	–	* @return true if the call must be retried because there
948	–	*   none of the blocking checks hold
1092		*/
1093	<	final boolean tryAwaitJoin(ForkJoinTask<?> joinMe, int retries) {
1094	<	if (joinMe.status < 0) // precheck to prime loop
1095	<	return false;
1096	<	int pc = parallelism;
1097	<	boolean running = true; // false when running count decremented
1098	<	outer:for (;;) {
1099	<	int wc = workerCounts;
1100	<	int rc = wc & RUNNING_COUNT_MASK;
1101	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
1102	<	if (running) { // replace with spare or decrement count
1103	<	if (rc <= pc && tc > pc &&
1104	<	(retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1105	<	ForkJoinWorkerThread[] ws = workers;
963	<	int nws = ws.length;
964	<	for (int i = 0; i < nws; ++i) { // search for spare
965	<	ForkJoinWorkerThread w = ws[i];
966	<	if (w != null) {
967	<	if (joinMe.status < 0)
968	<	return false;
969	<	if (w.isSuspended()) {
970	<	if ((workerCounts & RUNNING_COUNT_MASK)>=pc &&
971	<	w.tryResumeSpare()) {
972	<	running = false;
973	<	break outer;
974	<	}
975	<	continue outer; // rescan
976	<	}
977	<	}
978	<	}
979	<	}
980	<	if (retries < 0 || // < 0 means replacement check only
981	<	rc == 0 || joinMe.status < 0 || workerCounts != wc ||
982	<	!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
983	<	wc, wc - ONE_RUNNING))
984	<	return false; // done or inconsistent or contended
985	<	running = false;
986	<	if (rc > pc)
987	<	break;
1093	>	final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1094	>	int threshold = parallelism;         // descend blocking thresholds
1095	>	while (joinMe.status >= 0) {
1096	>	boolean block; int wc;
1097	>	worker.helpJoinTask(joinMe);
1098	>	if (joinMe.status < 0)
1099	>	break;
1100	>	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1101	>	if (threshold > 0)
1102	>	--threshold;
1103	>	else
1104	>	advanceEventCount(); // force release
1105	>	block = false;
1106		}
1107	<	else { // allow blocking if enough threads
1108	<	if (rc >= pc || joinMe.status < 0)
1109	<	break;
1110	<	int sc = tc - pc + 1; // = spare threads, plus the one to add
1111	<	if (retries > sc) {
1112	<	if (rc > 0 && rc >= pc - sc) // allow slack
1113	<	break;
1114	<	if (tc < MAX_THREADS &&
1115	<	tc == (runState & ACTIVE_COUNT_MASK) &&
1116	<	workerCounts == wc &&
1117	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1000	<	wc+(ONE_RUNNING|ONE_TOTAL))) {
1001	<	addWorker();
1002	<	break;
1003	<	}
1004	<	}
1005	<	if (workerCounts == wc &&        // back out to allow rescan
1006	<	UNSAFE.compareAndSwapInt (this, workerCountsOffset,
1007	<	wc, wc + ONE_RUNNING)) {
1008	<	releaseWaiters();            // help others progress
1009	<	return true;                 // let caller retry
1010	<	}
1107	>	else
1108	>	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1109	>	wc, wc - ONE_RUNNING);
1110	>	helpMaintainParallelism();
1111	>	if (block) {
1112	>	int c;
1113	>	joinMe.internalAwaitDone();
1114	>	do {} while (!UNSAFE.compareAndSwapInt
1115	>	(this, workerCountsOffset,
1116	>	c = workerCounts, c + ONE_RUNNING));
1117	>	break;
1118		}
1119		}
1013	–	// arrive here if can block
1014	–	joinMe.internalAwaitDone();
1015	–	int c;                      // to inline incrementRunningCount
1016	–	do {} while (!UNSAFE.compareAndSwapInt
1017	–	(this, workerCountsOffset,
1018	–	c = workerCounts, c + ONE_RUNNING));
1019	–	return false;
1120		}
1121
1122		/**
1123	<	* Same idea as (and shares many code snippets with) tryAwaitJoin,
1024	<	* but self-contained because there are no caller retries.
1025	<	* TODO: Rework to use simpler API.
1123	>	* Same idea as awaitJoin, but no helping
1124		*/
1125		final void awaitBlocker(ManagedBlocker blocker)
1126		throws InterruptedException {
1127	<	boolean done;
1128	<	if (done = blocker.isReleasable())
1129	<	return;
1130	<	int pc = parallelism;
1131	<	int retries = 0;
1132	<	boolean running = true; // false when running count decremented
1133	<	outer:for (;;) {
1134	<	int wc = workerCounts;
1135	<	int rc = wc & RUNNING_COUNT_MASK;
1038	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
1039	<	if (running) {
1040	<	if (rc <= pc && tc > pc &&
1041	<	(retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1042	<	ForkJoinWorkerThread[] ws = workers;
1043	<	int nws = ws.length;
1044	<	for (int i = 0; i < nws; ++i) {
1045	<	ForkJoinWorkerThread w = ws[i];
1046	<	if (w != null) {
1047	<	if (done = blocker.isReleasable())
1048	<	return;
1049	<	if (w.isSuspended()) {
1050	<	if ((workerCounts & RUNNING_COUNT_MASK)>=pc &&
1051	<	w.tryResumeSpare()) {
1052	<	running = false;
1053	<	break outer;
1054	<	}
1055	<	continue outer; // rescan
1056	<	}
1057	<	}
1058	<	}
1059	<	}
1060	<	if (done = blocker.isReleasable())
1061	<	return;
1062	<	if (rc == 0 || workerCounts != wc ||
1063	<	!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1064	<	wc, wc - ONE_RUNNING))
1065	<	continue;
1066	<	running = false;
1067	<	if (rc > pc)
1068	<	break;
1127	>	int threshold = parallelism;
1128	>	while (!blocker.isReleasable()) {
1129	>	boolean block; int wc;
1130	>	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1131	>	if (threshold > 0)
1132	>	--threshold;
1133	>	else
1134	>	advanceEventCount();
1135	>	block = false;
1136		}
1137	<	else {
1138	<	if (rc >= pc || (done = blocker.isReleasable()))
1139	<	break;
1140	<	int sc = tc - pc + 1;
1141	<	if (retries++ > sc) {
1142	<	if (rc > 0 && rc >= pc - sc)
1143	<	break;
1144	<	if (tc < MAX_THREADS &&
1145	<	tc == (runState & ACTIVE_COUNT_MASK) &&
1146	<	workerCounts == wc &&
1147	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1148	<	wc+(ONE_RUNNING|ONE_TOTAL))) {
1082	<	addWorker();
1083	<	break;
1084	<	}
1137	>	else
1138	>	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1139	>	wc, wc - ONE_RUNNING);
1140	>	helpMaintainParallelism();
1141	>	if (block) {
1142	>	try {
1143	>	do {} while (!blocker.isReleasable() && !blocker.block());
1144	>	} finally {
1145	>	int c;
1146	>	do {} while (!UNSAFE.compareAndSwapInt
1147	>	(this, workerCountsOffset,
1148	>	c = workerCounts, c + ONE_RUNNING));
1149		}
1150	<	Thread.yield();
1087	<	}
1088	<	}
1089	<
1090	<	try {
1091	<	if (!done)
1092	<	do {} while (!blocker.isReleasable() && !blocker.block());
1093	<	} finally {
1094	<	if (!running) {
1095	<	int c;
1096	<	do {} while (!UNSAFE.compareAndSwapInt
1097	<	(this, workerCountsOffset,
1098	<	c = workerCounts, c + ONE_RUNNING));
1150	>	break;
1151		}
1152		}
1153	<	}
1153	>	}
1154
1155		/**
1156		* Possibly initiates and/or completes termination.
#	Line 1128 | Line 1180 | public class ForkJoinPool extends Abstra
1180
1181		/**
1182		* Actions on transition to TERMINATING
1183	+	*
1184	+	* Runs up to four passes through workers: (0) shutting down each
1185	+	* (without waking up if parked) to quickly spread notifications
1186	+	* without unnecessary bouncing around event queues etc (1) wake
1187	+	* up and help cancel tasks (2) interrupt (3) mop up races with
1188	+	* interrupted workers
1189		*/
1190		private void startTerminating() {
1191	<	for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1192	<	cancelSubmissions();
1193	<	shutdownWorkers();
1194	<	cancelWorkerTasks();
1195	<	signalEvent();
1196	<	interruptWorkers();
1191	>	cancelSubmissions();
1192	>	for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1193	>	advanceEventCount();
1194	>	eventWaiters = 0L; // clobber lists
1195	>	spareWaiters = 0;
1196	>	ForkJoinWorkerThread[] ws = workers;
1197	>	int n = ws.length;
1198	>	for (int i = 0; i < n; ++i) {
1199	>	ForkJoinWorkerThread w = ws[i];
1200	>	if (w != null) {
1201	>	w.shutdown();
1202	>	if (passes > 0 && !w.isTerminated()) {
1203	>	w.cancelTasks();
1204	>	LockSupport.unpark(w);
1205	>	if (passes > 1) {
1206	>	try {
1207	>	w.interrupt();
1208	>	} catch (SecurityException ignore) {
1209	>	}
1210	>	}
1211	>	}
1212	>	}
1213	>	}
1214		}
1215		}
1216
#	Line 1152 | Line 1227 | public class ForkJoinPool extends Abstra
1227		}
1228		}
1229
1155	–	/**
1156	–	* Sets all worker run states to at least shutdown,
1157	–	* also resuming suspended workers
1158	–	*/
1159	–	private void shutdownWorkers() {
1160	–	ForkJoinWorkerThread[] ws = workers;
1161	–	int nws = ws.length;
1162	–	for (int i = 0; i < nws; ++i) {
1163	–	ForkJoinWorkerThread w = ws[i];
1164	–	if (w != null)
1165	–	w.shutdown();
1166	–	}
1167	–	}
1168	–
1169	–	/**
1170	–	* Clears out and cancels all locally queued tasks
1171	–	*/
1172	–	private void cancelWorkerTasks() {
1173	–	ForkJoinWorkerThread[] ws = workers;
1174	–	int nws = ws.length;
1175	–	for (int i = 0; i < nws; ++i) {
1176	–	ForkJoinWorkerThread w = ws[i];
1177	–	if (w != null)
1178	–	w.cancelTasks();
1179	–	}
1180	–	}
1181	–
1182	–	/**
1183	–	* Unsticks all workers blocked on joins etc
1184	–	*/
1185	–	private void interruptWorkers() {
1186	–	ForkJoinWorkerThread[] ws = workers;
1187	–	int nws = ws.length;
1188	–	for (int i = 0; i < nws; ++i) {
1189	–	ForkJoinWorkerThread w = ws[i];
1190	–	if (w != null && !w.isTerminated()) {
1191	–	try {
1192	–	w.interrupt();
1193	–	} catch (SecurityException ignore) {
1194	–	}
1195	–	}
1196	–	}
1197	–	}
1198	–
1230		// misc support for ForkJoinWorkerThread
1231
1232		/**
#	Line 1206 | Line 1237 | public class ForkJoinPool extends Abstra
1237		}
1238
1239		/**
1240	<	* Accumulates steal count from a worker, clearing
1241	<	* the worker's value
1240	>	* Tries to accumulates steal count from a worker, clearing
1241	>	* the worker's value.
1242	>	*
1243	>	* @return true if worker steal count now zero
1244		*/
1245	<	final void accumulateStealCount(ForkJoinWorkerThread w) {
1245	>	final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1246		int sc = w.stealCount;
1247	<	if (sc != 0) {
1248	<	long c;
1249	<	w.stealCount = 0;
1250	<	do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1251	<	c = stealCount, c + sc));
1247	>	long c = stealCount;
1248	>	// CAS even if zero, for fence effects
1249	>	if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1250	>	if (sc != 0)
1251	>	w.stealCount = 0;
1252	>	return true;
1253		}
1254	+	return sc == 0;
1255		}
1256
1257		/**
#	Line 1275 | Line 1310 | public class ForkJoinPool extends Abstra
1310		* use {@link java.lang.Runtime#availableProcessors}.
1311		* @param factory the factory for creating new threads. For default value,
1312		* use {@link #defaultForkJoinWorkerThreadFactory}.
1313	<	* @param handler the handler for internal worker threads that
1314	<	* terminate due to unrecoverable errors encountered while executing
1313	>	* @param handler the handler for internal worker threads that
1314	>	* terminate due to unrecoverable errors encountered while executing
1315		* tasks. For default value, use <code>null</code>.
1316	<	* @param asyncMode if true,
1316	>	* @param asyncMode if true,
1317		* establishes local first-in-first-out scheduling mode for forked
1318		* tasks that are never joined. This mode may be more appropriate
1319		* than default locally stack-based mode in applications in which
#	Line 1292 | Line 1327 | public class ForkJoinPool extends Abstra
1327		*         because it does not hold {@link
1328		*         java.lang.RuntimePermission}{@code ("modifyThread")}
1329		*/
1330	<	public ForkJoinPool(int parallelism,
1330	>	public ForkJoinPool(int parallelism,
1331		ForkJoinWorkerThreadFactory factory,
1332		Thread.UncaughtExceptionHandler handler,
1333		boolean asyncMode) {
1334		checkPermission();
1335		if (factory == null)
1336		throw new NullPointerException();
1337	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
1337	>	if (parallelism <= 0 || parallelism > MAX_WORKERS)
1338		throw new IllegalArgumentException();
1339		this.parallelism = parallelism;
1340		this.factory = factory;
#	Line 1318 | Line 1353 | public class ForkJoinPool extends Abstra
1353		* @param pc the initial parallelism level
1354		*/
1355		private static int initialArraySizeFor(int pc) {
1356	<	// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1357	<	int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1356	>	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1357	>	int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1358		size |= size >>> 1;
1359		size |= size >>> 2;
1360		size |= size >>> 4;
#	Line 1338 | Line 1373 | public class ForkJoinPool extends Abstra
1373		if (runState >= SHUTDOWN)
1374		throw new RejectedExecutionException();
1375		submissionQueue.offer(task);
1376	<	signalEvent();
1377	<	ensureEnoughWorkers();
1376	>	advanceEventCount();
1377	>	if (eventWaiters != 0L)
1378	>	releaseEventWaiters();
1379	>	if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
1380	>	addWorkerIfBelowTarget();
1381		}
1382
1383		/**
1384		* Performs the given task, returning its result upon completion.
1347	–	* If the caller is already engaged in a fork/join computation in
1348	–	* the current pool, this method is equivalent in effect to
1349	–	* {@link ForkJoinTask#invoke}.
1385		*
1386		* @param task the task
1387		* @return the task's result
#	Line 1361 | Line 1396 | public class ForkJoinPool extends Abstra
1396
1397		/**
1398		* Arranges for (asynchronous) execution of the given task.
1364	–	* If the caller is already engaged in a fork/join computation in
1365	–	* the current pool, this method is equivalent in effect to
1366	–	* {@link ForkJoinTask#fork}.
1399		*
1400		* @param task the task
1401		* @throws NullPointerException if the task is null
#	Line 1392 | Line 1424 | public class ForkJoinPool extends Abstra
1424
1425		/**
1426		* Submits a ForkJoinTask for execution.
1395	–	* If the caller is already engaged in a fork/join computation in
1396	–	* the current pool, this method is equivalent in effect to
1397	–	* {@link ForkJoinTask#fork}.
1427		*
1428		* @param task the task to submit
1429		* @return the task
#	Line 1586 | Line 1615 | public class ForkJoinPool extends Abstra
1615		public long getQueuedTaskCount() {
1616		long count = 0;
1617		ForkJoinWorkerThread[] ws = workers;
1618	<	int nws = ws.length;
1619	<	for (int i = 0; i < nws; ++i) {
1618	>	int n = ws.length;
1619	>	for (int i = 0; i < n; ++i) {
1620		ForkJoinWorkerThread w = ws[i];
1621		if (w != null)
1622		count += w.getQueueSize();
#	Line 1645 | Line 1674 | public class ForkJoinPool extends Abstra
1674		* @return the number of elements transferred
1675		*/
1676		protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1677	<	int n = submissionQueue.drainTo(c);
1649	<	ForkJoinWorkerThread[] ws = workers;
1650	<	int nws = ws.length;
1651	<	for (int i = 0; i < nws; ++i) {
1652	<	ForkJoinWorkerThread w = ws[i];
1653	<	if (w != null)
1654	<	n += w.drainTasksTo(c);
1655	<	}
1656	<	return n;
1657	<	}
1658	<
1659	<	/**
1660	<	* Returns count of total parks by existing workers.
1661	<	* Used during development only since not meaningful to users.
1662	<	*/
1663	<	private int collectParkCount() {
1664	<	int count = 0;
1677	>	int count = submissionQueue.drainTo(c);
1678		ForkJoinWorkerThread[] ws = workers;
1679	<	int nws = ws.length;
1680	<	for (int i = 0; i < nws; ++i) {
1679	>	int n = ws.length;
1680	>	for (int i = 0; i < n; ++i) {
1681		ForkJoinWorkerThread w = ws[i];
1682		if (w != null)
1683	<	count += w.parkCount;
1683	>	count += w.drainTasksTo(c);
1684		}
1685		return count;
1686		}
#	Line 1689 | Line 1702 | public class ForkJoinPool extends Abstra
1702		int pc = parallelism;
1703		int rs = runState;
1704		int ac = rs & ACTIVE_COUNT_MASK;
1692	–	//        int pk = collectParkCount();
1705		return super.toString() +
1706		"[" + runLevelToString(rs) +
1707		", parallelism = " + pc +
#	Line 1699 | Line 1711 | public class ForkJoinPool extends Abstra
1711		", steals = " + st +
1712		", tasks = " + qt +
1713		", submissions = " + qs +
1702	–	//            ", parks = " + pk +
1714		"]";
1715		}
1716
#	Line 1806 | Line 1817 | public class ForkJoinPool extends Abstra
1817		* Interface for extending managed parallelism for tasks running
1818		* in {@link ForkJoinPool}s.
1819		*
1820	<	* <p>A {@code ManagedBlocker} provides two methods.
1821	<	* Method {@code isReleasable} must return {@code true} if
1822	<	* blocking is not necessary. Method {@code block} blocks the
1823	<	* current thread if necessary (perhaps internally invoking
1824	<	* {@code isReleasable} before actually blocking).
1820	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
1821	>	* {@code isReleasable} must return {@code true} if blocking is
1822	>	* not necessary. Method {@code block} blocks the current thread
1823	>	* if necessary (perhaps internally invoking {@code isReleasable}
1824	>	* before actually blocking). The unusual methods in this API
1825	>	* accommodate synchronizers that may, but don't usually, block
1826	>	* for long periods. Similarly, they allow more efficient internal
1827	>	* handling of cases in which additional workers may be, but
1828	>	* usually are not, needed to ensure sufficient parallelism.
1829	>	* Toward this end, implementations of method {@code isReleasable}
1830	>	* must be amenable to repeated invocation.
1831		*
1832		* <p>For example, here is a ManagedBlocker based on a
1833		* ReentrantLock:
#	Line 1828 | Line 1845 | public class ForkJoinPool extends Abstra
1845		*     return hasLock || (hasLock = lock.tryLock());
1846		*   }
1847		* }}</pre>
1848	+	*
1849	+	* <p>Here is a class that possibly blocks waiting for an
1850	+	* item on a given queue:
1851	+	*  <pre> {@code
1852	+	* class QueueTaker<E> implements ManagedBlocker {
1853	+	*   final BlockingQueue<E> queue;
1854	+	*   volatile E item = null;
1855	+	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1856	+	*   public boolean block() throws InterruptedException {
1857	+	*     if (item == null)
1858	+	*       item = queue.take
1859	+	*     return true;
1860	+	*   }
1861	+	*   public boolean isReleasable() {
1862	+	*     return item != null || (item = queue.poll) != null;
1863	+	*   }
1864	+	*   public E getItem() { // call after pool.managedBlock completes
1865	+	*     return item;
1866	+	*   }
1867	+	* }}</pre>
1868		*/
1869		public static interface ManagedBlocker {
1870		/**
#	Line 1870 | Line 1907 | public class ForkJoinPool extends Abstra
1907		public static void managedBlock(ManagedBlocker blocker)
1908		throws InterruptedException {
1909		Thread t = Thread.currentThread();
1910	<	if (t instanceof ForkJoinWorkerThread)
1911	<	((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
1910	>	if (t instanceof ForkJoinWorkerThread) {
1911	>	ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1912	>	w.pool.awaitBlocker(blocker);
1913	>	}
1914		else {
1915		do {} while (!blocker.isReleasable() && !blocker.block());
1916		}
#	Line 1902 | Line 1941 | public class ForkJoinPool extends Abstra
1941		objectFieldOffset("eventWaiters",ForkJoinPool.class);
1942		private static final long stealCountOffset =
1943		objectFieldOffset("stealCount",ForkJoinPool.class);
1944	+	private static final long spareWaitersOffset =
1945	+	objectFieldOffset("spareWaiters",ForkJoinPool.class);
1946
1947		private static long objectFieldOffset(String field, Class<?> klazz) {
1948		try {

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines