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 1.60 by dl, Sat Jul 24 20:28:18 2010 UTC vs.
Revision 1.78 by dl, Tue Sep 7 14:52:24 2010 UTC

#	Line 6 | Line 6
6
7		package jsr166y;
8
9	–	import java.util.concurrent.*;
10	–
9		import java.util.ArrayList;
10		import java.util.Arrays;
11		import java.util.Collection;
12		import java.util.Collections;
13		import java.util.List;
14	+	import java.util.concurrent.AbstractExecutorService;
15	+	import java.util.concurrent.Callable;
16	+	import java.util.concurrent.CountDownLatch;
17	+	import java.util.concurrent.ExecutorService;
18	+	import java.util.concurrent.Future;
19	+	import java.util.concurrent.RejectedExecutionException;
20	+	import java.util.concurrent.RunnableFuture;
21	+	import java.util.concurrent.TimeUnit;
22	+	import java.util.concurrent.TimeoutException;
23	+	import java.util.concurrent.atomic.AtomicInteger;
24		import java.util.concurrent.locks.LockSupport;
25		import java.util.concurrent.locks.ReentrantLock;
18	–	import java.util.concurrent.atomic.AtomicInteger;
19	–	import java.util.concurrent.CountDownLatch;
26
27		/**
28		* An {@link ExecutorService} for running {@link ForkJoinTask}s.
#	Line 69 | Line 75 | import java.util.concurrent.CountDownLat
75		*    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
76		*  </tr>
77		*  <tr>
78	<	*    <td> <b>Arange async execution</td>
78	>	*    <td> <b>Arrange async execution</td>
79		*    <td> {@link #execute(ForkJoinTask)}</td>
80		*    <td> {@link ForkJoinTask#fork}</td>
81		*  </tr>
#	Line 110 | Line 116 | import java.util.concurrent.CountDownLat
116		*
117		* <p>This implementation rejects submitted tasks (that is, by throwing
118		* {@link RejectedExecutionException}) only when the pool is shut down
119	<	* or internal resources have been exhuasted.
119	>	* or internal resources have been exhausted.
120		*
121		* @since 1.7
122		* @author Doug Lea
#	Line 140 | Line 146 | public class ForkJoinPool extends Abstra
146		* Beyond work-stealing support and essential bookkeeping, the
147		* main responsibility of this framework is to take actions when
148		* one worker is waiting to join a task stolen (or always held by)
149	<	* another.  Becauae we are multiplexing many tasks on to a pool
149	>	* another.  Because we are multiplexing many tasks on to a pool
150		* of workers, we can't just let them block (as in Thread.join).
151		* We also cannot just reassign the joiner's run-time stack with
152		* another and replace it later, which would be a form of
#	Line 156 | Line 162 | public class ForkJoinPool extends Abstra
162		*      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
163		*      links to try to find such a task.
164		*
165	<	*   Compensating: Unless there are already enough live threads,
166	<	*      creating or or re-activating a spare thread to compensate
167	<	*      for the (blocked) joiner until it unblocks.  Spares then
168	<	*      suspend at their next opportunity or eventually die if
169	<	*      unused for too long.  See below and the internal
170	<	*      documentation for tryAwaitJoin for more details about
171	<	*      compensation rules.
172	<	*
173	<	* Because the determining existence of conservatively safe
174	<	* helping targets, the availability of already-created spares,
175	<	* and the apparent need to create new spares are all racy and
176	<	* require heuristic guidance, joins (in
177	<	* ForkJoinWorkerThread.joinTask) interleave these options until
178	<	* successful.  Creating a new spare always succeeds, but also
179	<	* increases application footprint, so we try to avoid it, within
180	<	* reason.
165	>	*   Compensating: Unless there are already enough live threads,
166	>	*      method helpMaintainParallelism() may create or
167	>	*      re-activate a spare thread to compensate for blocked
168	>	*      joiners until they unblock.
169	>	*
170	>	* It is impossible to keep exactly the target (parallelism)
171	>	* number of threads running at any given time.  Determining
172	>	* existence of conservatively safe helping targets, the
173	>	* availability of already-created spares, and the apparent need
174	>	* to create new spares are all racy and require heuristic
175	>	* guidance, so we rely on multiple retries of each.  Compensation
176	>	* occurs in slow-motion. It is triggered only upon timeouts of
177	>	* Object.wait used for joins. This reduces poor decisions that
178	>	* would otherwise be made when threads are waiting for others
179	>	* that are stalled because of unrelated activities such as
180	>	* garbage collection.
181		*
182	<	* The ManagedBlocker extension API can't use helping so uses a
183	<	* special version of compensation in method awaitBlocker.
182	>	* The ManagedBlocker extension API can't use helping so relies
183	>	* only on compensation in method awaitBlocker.
184		*
185		* The main throughput advantages of work-stealing stem from
186		* decentralized control -- workers mostly steal tasks from each
#	Line 207 | Line 213 | public class ForkJoinPool extends Abstra
213		* blocked workers. However, all other support code is set up to
214		* work with other policies.
215		*
216	+	* To ensure that we do not hold on to worker references that
217	+	* would prevent GC, ALL accesses to workers are via indices into
218	+	* the workers array (which is one source of some of the unusual
219	+	* code constructions here). In essence, the workers array serves
220	+	* as a WeakReference mechanism. Thus for example the event queue
221	+	* stores worker indices, not worker references. Access to the
222	+	* workers in associated methods (for example releaseEventWaiters)
223	+	* must both index-check and null-check the IDs. All such accesses
224	+	* ignore bad IDs by returning out early from what they are doing,
225	+	* since this can only be associated with shutdown, in which case
226	+	* it is OK to give up. On termination, we just clobber these
227	+	* data structures without trying to use them.
228	+	*
229		* 2. Bookkeeping for dynamically adding and removing workers. We
230		* aim to approximately maintain the given level of parallelism.
231		* When some workers are known to be blocked (on joins or via
232		* ManagedBlocker), we may create or resume others to take their
233		* place until they unblock (see below). Implementing this
234		* requires counts of the number of "running" threads (i.e., those
235	<	* that are neither blocked nor artifically suspended) as well as
235	>	* that are neither blocked nor artificially suspended) as well as
236		* the total number.  These two values are packed into one field,
237		* "workerCounts" because we need accurate snapshots when deciding
238		* to create, resume or suspend.  Note however that the
239	<	* correspondance of these counts to reality is not guaranteed. In
239	>	* correspondence of these counts to reality is not guaranteed. In
240		* particular updates for unblocked threads may lag until they
241		* actually wake up.
242		*
#	Line 248 | Line 267 | public class ForkJoinPool extends Abstra
267		* workers that previously could not find a task to now find one:
268		* Submission of a new task to the pool, or another worker pushing
269		* a task onto a previously empty queue.  (We also use this
270	<	* mechanism for termination and reconfiguration actions that
270	>	* mechanism for configuration and termination actions that
271		* require wakeups of idle workers).  Each worker maintains its
272		* last known event count, and blocks when a scan for work did not
273		* find a task AND its lastEventCount matches the current
#	Line 259 | Line 278 | public class ForkJoinPool extends Abstra
278		* a record (field nextEventWaiter) for the next waiting worker.
279		* In addition to allowing simpler decisions about need for
280		* wakeup, the event count bits in eventWaiters serve the role of
281	<	* tags to avoid ABA errors in Treiber stacks.  To reduce delays
282	<	* in task diffusion, workers not otherwise occupied may invoke
283	<	* method releaseWaiters, that removes and signals (unparks)
284	<	* workers not waiting on current count. To minimize task
285	<	* production stalls associate with signalling, any worker pushing
286	<	* a task on an empty queue invokes the weaker method signalWork,
268	<	* 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.
281	>	* tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
282	>	* released threads also try to release at most two others.  The
283	>	* net effect is a tree-like diffusion of signals, where released
284	>	* threads (and possibly others) help with unparks.  To further
285	>	* reduce contention effects a bit, failed CASes to increment
286	>	* field eventCount are tolerated without retries in signalWork.
287		* Conceptually they are merged into the same event, which is OK
288		* when their only purpose is to enable workers to scan for work.
289		*
290	<	* 5. Managing suspension of extra workers. When a worker is about
291	<	* to block waiting for a join (or via ManagedBlockers), we may
292	<	* create a new thread to maintain parallelism level, or at least
293	<	* avoid starvation. Usually, extra threads are needed for only
294	<	* very short periods, yet join dependencies are such that we
295	<	* sometimes need them in bursts. Rather than create new threads
296	<	* each time this happens, we suspend no-longer-needed extra ones
297	<	* as "spares". For most purposes, we don't distinguish "extra"
298	<	* spare threads from normal "core" threads: On each call to
299	<	* preStep (the only point at which we can do this) a worker
300	<	* checks to see if there are now too many running workers, and if
301	<	* so, suspends itself.  Methods tryAwaitJoin and awaitBlocker
302	<	* look for suspended threads to resume before considering
303	<	* creating a new replacement. We don't need a special data
304	<	* structure to maintain spares; simply scanning the workers array
305	<	* looking for worker.isSuspended() is fine because the calling
306	<	* thread is otherwise not doing anything useful anyway; we are at
307	<	* least as happy if after locating a spare, the caller doesn't
308	<	* actually block because the join is ready before we try to
309	<	* adjust and compensate.  Note that this is intrinsically racy.
310	<	* One thread may become a spare at about the same time as another
311	<	* is needlessly being created. We counteract this and related
312	<	* slop in part by requiring resumed spares to immediately recheck
313	<	* (in preStep) to see whether they they should re-suspend. The
314	<	* only effective difference between "extra" and "core" threads is
315	<	* that we allow the "extra" ones to time out and die if they are
316	<	* not resumed within a keep-alive interval of a few seconds. This
317	<	* is implemented mainly within ForkJoinWorkerThread, but requires
318	<	* some coordination (isTrimmed() -- meaning killed while
319	<	* suspended) to correctly maintain pool counts.
320	<	*
321	<	* 6. Deciding when to create new workers. The main dynamic
322	<	* control in this class is deciding when to create extra threads,
323	<	* in methods awaitJoin and awaitBlocker. We always need to create
324	<	* one when the number of running threads would become zero and
325	<	* all workers are busy. However, this is not easy to detect
326	<	* reliably in the presence of transients so we use retries and
327	<	* allow slack (in tryAwaitJoin) to reduce false alarms.  These
328	<	* effectively reduce churn at the price of systematically
329	<	* undershooting target parallelism when many threads are blocked.
330	<	* However, biasing toward undeshooting partially compensates for
331	<	* the above mechanics to suspend extra threads, that normally
332	<	* lead to overshoot because we can only suspend workers
333	<	* in-between top-level actions. It also better copes with the
334	<	* fact that some of the methods in this class tend to never
335	<	* become compiled (but are interpreted), so some components of
336	<	* the entire set of controls might execute many times faster than
337	<	* others. And similarly for cases where the apparent lack of work
338	<	* is just due to GC stalls and other transient system activity.
290	>	* 5. Managing suspension of extra workers. When a worker notices
291	>	* (usually upon timeout of a wait()) that there are too few
292	>	* running threads, we may create a new thread to maintain
293	>	* parallelism level, or at least avoid starvation. Usually, extra
294	>	* threads are needed for only very short periods, yet join
295	>	* dependencies are such that we sometimes need them in
296	>	* bursts. Rather than create new threads each time this happens,
297	>	* we suspend no-longer-needed extra ones as "spares". For most
298	>	* purposes, we don't distinguish "extra" spare threads from
299	>	* normal "core" threads: On each call to preStep (the only point
300	>	* at which we can do this) a worker checks to see if there are
301	>	* now too many running workers, and if so, suspends itself.
302	>	* Method helpMaintainParallelism looks for suspended threads to
303	>	* resume before considering creating a new replacement. The
304	>	* spares themselves are encoded on another variant of a Treiber
305	>	* Stack, headed at field "spareWaiters".  Note that the use of
306	>	* spares is intrinsically racy.  One thread may become a spare at
307	>	* about the same time as another is needlessly being created. We
308	>	* counteract this and related slop in part by requiring resumed
309	>	* spares to immediately recheck (in preStep) to see whether they
310	>	* should re-suspend.
311	>	*
312	>	* 6. Killing off unneeded workers. A timeout mechanism is used to
313	>	* shed unused workers: The oldest (first) event queue waiter uses
314	>	* a timed rather than hard wait. When this wait times out without
315	>	* a normal wakeup, it tries to shutdown any one (for convenience
316	>	* the newest) other spare or event waiter via
317	>	* tryShutdownUnusedWorker. This eventually reduces the number of
318	>	* worker threads to a minimum of one after a long enough period
319	>	* without use.
320	>	*
321	>	* 7. Deciding when to create new workers. The main dynamic
322	>	* control in this class is deciding when to create extra threads
323	>	* in method helpMaintainParallelism. We would like to keep
324	>	* exactly #parallelism threads running, which is an impossible
325	>	* task. We always need to create one when the number of running
326	>	* threads would become zero and all workers are busy. Beyond
327	>	* this, we must rely on heuristics that work well in the
328	>	* presence of transient phenomena such as GC stalls, dynamic
329	>	* compilation, and wake-up lags. These transients are extremely
330	>	* common -- we are normally trying to fully saturate the CPUs on
331	>	* a machine, so almost any activity other than running tasks
332	>	* impedes accuracy. Our main defense is to allow parallelism to
333	>	* lapse for a while during joins, and use a timeout to see if,
334	>	* after the resulting settling, there is still a need for
335	>	* additional workers.  This also better copes with the fact that
336	>	* some of the methods in this class tend to never become compiled
337	>	* (but are interpreted), so some components of the entire set of
338	>	* controls might execute 100 times faster than others. And
339	>	* similarly for cases where the apparent lack of work is just due
340	>	* to GC stalls and other transient system activity.
341		*
342		* Beware that there is a lot of representation-level coupling
343		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 335 | Line 350 | public class ForkJoinPool extends Abstra
350		*
351		* Style notes: There are lots of inline assignments (of form
352		* "while ((local = field) != 0)") which are usually the simplest
353	<	* way to ensure read orderings. Also several occurrences of the
354	<	* unusual "do {} while(!cas...)" which is the simplest way to
355	<	* force an update of a CAS'ed variable. There are also other
356	<	* coding oddities that help some methods perform reasonably even
357	<	* when interpreted (not compiled), at the expense of messiness.
353	>	* way to ensure the required read orderings (which are sometimes
354	>	* critical). Also several occurrences of the unusual "do {}
355	>	* while (!cas...)" which is the simplest way to force an update of
356	>	* a CAS'ed variable. There are also other coding oddities that
357	>	* help some methods perform reasonably even when interpreted (not
358	>	* compiled), at the expense of some messy constructions that
359	>	* reduce byte code counts.
360		*
361		* The order of declarations in this file is: (1) statics (2)
362		* fields (along with constants used when unpacking some of them)
#	Line 407 | Line 424 | public class ForkJoinPool extends Abstra
424		new AtomicInteger();
425
426		/**
427	<	* Absolute bound for parallelism level. Twice this number must
428	<	* fit into a 16bit field to enable word-packing for some counts.
427	>	* The time to block in a join (see awaitJoin) before checking if
428	>	* a new worker should be (re)started to maintain parallelism
429	>	* level. The value should be short enough to maintain global
430	>	* responsiveness and progress but long enough to avoid
431	>	* counterproductive firings during GC stalls or unrelated system
432	>	* activity, and to not bog down systems with continual re-firings
433	>	* on GCs or legitimately long waits.
434	>	*/
435	>	private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
436	>
437	>	/**
438	>	* The wakeup interval (in nanoseconds) for the oldest worker
439	>	* waiting for an event to invoke tryShutdownUnusedWorker to
440	>	* shrink the number of workers.  The exact value does not matter
441	>	* too much. It must be short enough to release resources during
442	>	* sustained periods of idleness, but not so short that threads
443	>	* are continually re-created.
444		*/
445	<	private static final int MAX_THREADS = 0x7fff;
445	>	private static final long SHRINK_RATE_NANOS =
446	>	30L * 1000L * 1000L * 1000L; // 2 per minute
447	>
448	>	/**
449	>	* Absolute bound for parallelism level. Twice this number plus
450	>	* one (i.e., 0xfff) must fit into a 16bit field to enable
451	>	* word-packing for some counts and indices.
452	>	*/
453	>	private static final int MAX_WORKERS   = 0x7fff;
454
455		/**
456		* Array holding all worker threads in the pool.  Array size must
#	Line 450 | Line 490 | public class ForkJoinPool extends Abstra
490		private volatile long stealCount;
491
492		/**
493	<	* Encoded record of top of treiber stack of threads waiting for
493	>	* Encoded record of top of Treiber stack of threads waiting for
494		* events. The top 32 bits contain the count being waited for. The
495	<	* bottom word contains one plus the pool index of waiting worker
496	<	* thread.
495	>	* bottom 16 bits contains one plus the pool index of waiting
496	>	* worker thread. (Bits 16-31 are unused.)
497		*/
498		private volatile long eventWaiters;
499
500		private static final int  EVENT_COUNT_SHIFT = 32;
501	<	private static final long WAITER_ID_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
501	>	private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
502
503		/**
504		* A counter for events that may wake up worker threads:
505		*   - Submission of a new task to the pool
506		*   - A worker pushing a task on an empty queue
507	<	*   - termination and reconfiguration
507	>	*   - termination
508		*/
509		private volatile int eventCount;
510
511		/**
512	+	* Encoded record of top of Treiber stack of spare threads waiting
513	+	* for resumption. The top 16 bits contain an arbitrary count to
514	+	* avoid ABA effects. The bottom 16bits contains one plus the pool
515	+	* index of waiting worker thread.
516	+	*/
517	+	private volatile int spareWaiters;
518	+
519	+	private static final int SPARE_COUNT_SHIFT = 16;
520	+	private static final int SPARE_ID_MASK     = (1 << 16) - 1;
521	+
522	+	/**
523		* Lifecycle control. The low word contains the number of workers
524		* that are (probably) executing tasks. This value is atomically
525		* incremented before a worker gets a task to run, and decremented
#	Line 479 | Line 530 | public class ForkJoinPool extends Abstra
530		* These are bundled together to ensure consistent read for
531		* termination checks (i.e., that runLevel is at least SHUTDOWN
532		* and active threads is zero).
533	+	*
534	+	* Notes: Most direct CASes are dependent on these bitfield
535	+	* positions.  Also, this field is non-private to enable direct
536	+	* performance-sensitive CASes in ForkJoinWorkerThread.
537		*/
538	<	private volatile int runState;
538	>	volatile int runState;
539
540		// Note: The order among run level values matters.
541		private static final int RUNLEVEL_SHIFT     = 16;
#	Line 488 | Line 543 | public class ForkJoinPool extends Abstra
543		private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
544		private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
545		private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
491	–	private static final int ONE_ACTIVE         = 1; // active update delta
546
547		/**
548		* Holds number of total (i.e., created and not yet terminated)
#	Line 529 | Line 583 | public class ForkJoinPool extends Abstra
583		*/
584		private final int poolNumber;
585
586	<	// Utilities for CASing fields. Note that several of these
587	<	// are manually inlined by callers
586	>	// Utilities for CASing fields. Note that most of these
587	>	// are usually manually inlined by callers
588
589		/**
590	<	* Increments running count.  Also used by ForkJoinTask.
590	>	* Increments running count part of workerCounts
591		*/
592		final void incrementRunningCount() {
593		int c;
#	Line 554 | Line 608 | public class ForkJoinPool extends Abstra
608		}
609
610		/**
611	<	* Tries to increment running count
612	<	*/
559	<	final boolean tryIncrementRunningCount() {
560	<	int wc;
561	<	return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
562	<	wc = workerCounts, wc + ONE_RUNNING);
563	<	}
564	<
565	<	/**
566	<	* Tries incrementing active count; fails on contention.
567	<	* Called by workers before executing tasks.
611	>	* Forces decrement of encoded workerCounts, awaiting nonzero if
612	>	* (rarely) necessary when other count updates lag.
613		*
614	<	* @return true on success
614	>	* @param dr -- either zero or ONE_RUNNING
615	>	* @param dt -- either zero or ONE_TOTAL
616		*/
617	<	final boolean tryIncrementActiveCount() {
618	<	int c;
619	<	return UNSAFE.compareAndSwapInt(this, runStateOffset,
620	<	c = runState, c + ONE_ACTIVE);
617	>	private void decrementWorkerCounts(int dr, int dt) {
618	>	for (;;) {
619	>	int wc = workerCounts;
620	>	if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
621	>	(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
622	>	if ((runState & TERMINATED) != 0)
623	>	return; // lagging termination on a backout
624	>	Thread.yield();
625	>	}
626	>	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
627	>	wc, wc - (dr + dt)))
628	>	return;
629	>	}
630		}
631
632		/**
#	Line 581 | Line 636 | public class ForkJoinPool extends Abstra
636		final boolean tryDecrementActiveCount() {
637		int c;
638		return UNSAFE.compareAndSwapInt(this, runStateOffset,
639	<	c = runState, c - ONE_ACTIVE);
639	>	c = runState, c - 1);
640		}
641
642		/**
#	Line 610 | Line 665 | public class ForkJoinPool extends Abstra
665		lock.lock();
666		try {
667		ForkJoinWorkerThread[] ws = workers;
668	<	int nws = ws.length;
669	<	if (k < 0 || k >= nws || ws[k] != null) {
670	<	for (k = 0; k < nws && ws[k] != null; ++k)
668	>	int n = ws.length;
669	>	if (k < 0 || k >= n || ws[k] != null) {
670	>	for (k = 0; k < n && ws[k] != null; ++k)
671		;
672	<	if (k == nws)
673	<	ws = Arrays.copyOf(ws, nws << 1);
672	>	if (k == n)
673	>	ws = Arrays.copyOf(ws, n << 1);
674		}
675		ws[k] = w;
676		workers = ws; // volatile array write ensures slot visibility
#	Line 626 | Line 681 | public class ForkJoinPool extends Abstra
681		}
682
683		/**
684	<	* Nulls out record of worker in workers array
684	>	* Nulls out record of worker in workers array.
685		*/
686		private void forgetWorker(ForkJoinWorkerThread w) {
687		int idx = w.poolIndex;
688	<	// Locking helps method recordWorker avoid unecessary expansion
688	>	// Locking helps method recordWorker avoid unnecessary expansion
689		final ReentrantLock lock = this.workerLock;
690		lock.lock();
691		try {
#	Line 642 | Line 697 | public class ForkJoinPool extends Abstra
697		}
698		}
699
645	–	// adding and removing workers
646	–
700		/**
701	<	* Tries to create and add new worker. Assumes that worker counts
702	<	* are already updated to accommodate the worker, so adjusts on
703	<	* failure.
701	>	* Final callback from terminating worker.  Removes record of
702	>	* worker from array, and adjusts counts. If pool is shutting
703	>	* down, tries to complete termination.
704		*
705	<	* @return new worker or null if creation failed
705	>	* @param w the worker
706		*/
707	<	private ForkJoinWorkerThread addWorker() {
708	<	ForkJoinWorkerThread w = null;
709	<	try {
710	<	w = factory.newThread(this);
711	<	} finally { // Adjust on either null or exceptional factory return
712	<	if (w == null)
660	<	onWorkerCreationFailure();
661	<	}
662	<	if (w != null)
663	<	w.start(recordWorker(w), ueh);
664	<	return w;
707	>	final void workerTerminated(ForkJoinWorkerThread w) {
708	>	forgetWorker(w);
709	>	decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
710	>	while (w.stealCount != 0) // collect final count
711	>	tryAccumulateStealCount(w);
712	>	tryTerminate(false);
713		}
714
715	+	// Waiting for and signalling events
716	+
717		/**
718	<	* Adjusts counts upon failure to create worker
718	>	* Releases workers blocked on a count not equal to current count.
719	>	* Normally called after precheck that eventWaiters isn't zero to
720	>	* avoid wasted array checks. Gives up upon a change in count or
721	>	* upon releasing two workers, letting others take over.
722		*/
723	<	private void onWorkerCreationFailure() {
724	<	for (;;) {
725	<	int wc = workerCounts;
726	<	int rc = wc & RUNNING_COUNT_MASK;
727	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
728	<	if (rc == 0 || wc == 0)
729	<	Thread.yield(); // must wait for other counts to settle
730	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
731	<	wc - (ONE_RUNNING|ONE_TOTAL)))
723	>	private void releaseEventWaiters() {
724	>	ForkJoinWorkerThread[] ws = workers;
725	>	int n = ws.length;
726	>	long h = eventWaiters;
727	>	int ec = eventCount;
728	>	boolean releasedOne = false;
729	>	ForkJoinWorkerThread w; int id;
730	>	while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
731	>	(int)(h >>> EVENT_COUNT_SHIFT) != ec &&
732	>	id < n && (w = ws[id]) != null) {
733	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
734	>	h,  w.nextWaiter)) {
735	>	LockSupport.unpark(w);
736	>	if (releasedOne) // exit on second release
737	>	break;
738	>	releasedOne = true;
739	>	}
740	>	if (eventCount != ec)
741		break;
742	+	h = eventWaiters;
743		}
681	–	tryTerminate(false); // in case of failure during shutdown
744		}
745
746		/**
747	<	* Creates enough total workers to establish target parallelism,
748	<	* giving up if terminating or addWorker fails
747	>	* Tries to advance eventCount and releases waiters. Called only
748	>	* from workers.
749		*/
750	<	private void ensureEnoughTotalWorkers() {
751	<	int wc;
752	<	while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
753	<	runState < TERMINATING) {
754	<	if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
693	<	wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
694	<	addWorker() == null))
695	<	break;
696	<	}
750	>	final void signalWork() {
751	>	int c; // try to increment event count -- CAS failure OK
752	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
753	>	if (eventWaiters != 0L)
754	>	releaseEventWaiters();
755		}
756
757		/**
758	<	* Final callback from terminating worker.  Removes record of
759	<	* worker from array, and adjusts counts. If pool is shutting
702	<	* down, tries to complete terminatation, else possibly replaces
703	<	* the worker.
758	>	* Adds the given worker to event queue and blocks until
759	>	* terminating or event count advances from the given value
760		*
761	<	* @param w the worker
761	>	* @param w the calling worker thread
762	>	* @param ec the count
763		*/
764	<	final void workerTerminated(ForkJoinWorkerThread w) {
765	<	if (w.active) { // force inactive
766	<	w.active = false;
767	<	do {} while (!tryDecrementActiveCount());
768	<	}
769	<	forgetWorker(w);
770	<
771	<	// Decrement total count, and if was running, running count
772	<	// Spin (waiting for other updates) if either would be negative
773	<	int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
717	<	int unit = ONE_TOTAL + nr;
718	<	for (;;) {
719	<	int wc = workerCounts;
720	<	int rc = wc & RUNNING_COUNT_MASK;
721	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
722	<	if (rc - nr < 0 || tc == 0)
723	<	Thread.yield(); // back off if waiting for other updates
724	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
725	<	wc, wc - unit))
764	>	private void eventSync(ForkJoinWorkerThread w, int ec) {
765	>	long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
766	>	long h;
767	>	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
768	>	(((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
769	>	(int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
770	>	eventCount == ec) {
771	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
772	>	w.nextWaiter = h, nh)) {
773	>	awaitEvent(w, ec);
774		break;
775	+	}
776		}
728	–
729	–	accumulateStealCount(w); // collect final count
730	–	if (!tryTerminate(false))
731	–	ensureEnoughTotalWorkers();
777		}
778
734	–	// Waiting for and signalling events
735	–
779		/**
780	<	* Releases workers blocked on a count not equal to current count.
781	<	* @return true if any released
780	>	* Blocks the given worker (that has already been entered as an
781	>	* event waiter) until terminating or event count advances from
782	>	* the given value. The oldest (first) waiter uses a timed wait to
783	>	* occasionally one-by-one shrink the number of workers (to a
784	>	* minimum of one) if the pool has not been used for extended
785	>	* periods.
786	>	*
787	>	* @param w the calling worker thread
788	>	* @param ec the count
789		*/
790	<	private void releaseWaiters() {
791	<	long top;
792	<	while ((top = eventWaiters) != 0L) {
793	<	ForkJoinWorkerThread[] ws = workers;
794	<	int n = ws.length;
795	<	for (;;) {
796	<	int i = ((int)(top & WAITER_ID_MASK)) - 1;
797	<	int e = (int)(top >>> EVENT_COUNT_SHIFT);
798	<	if (i < 0 || e == eventCount)
799	<	return;
800	<	ForkJoinWorkerThread w;
801	<	if (i < n && (w = ws[i]) != null &&
802	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
803	<	top, w.nextWaiter)) {
804	<	LockSupport.unpark(w);
805	<	top = eventWaiters;
790	>	private void awaitEvent(ForkJoinWorkerThread w, int ec) {
791	>	while (eventCount == ec) {
792	>	if (tryAccumulateStealCount(w)) { // transfer while idle
793	>	boolean untimed = (w.nextWaiter != 0L ||
794	>	(workerCounts & RUNNING_COUNT_MASK) <= 1);
795	>	long startTime = untimed? 0 : System.nanoTime();
796	>	Thread.interrupted();         // clear/ignore interrupt
797	>	if (eventCount != ec || w.runState != 0 ||
798	>	runState >= TERMINATING)  // recheck after clear
799	>	break;
800	>	if (untimed)
801	>	LockSupport.park(w);
802	>	else {
803	>	LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
804	>	if (eventCount != ec || w.runState != 0 ||
805	>	runState >= TERMINATING)
806	>	break;
807	>	if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
808	>	tryShutdownUnusedWorker(ec);
809		}
757	–	else
758	–	break;      // possibly stale; reread
810		}
811		}
812		}
813
814	+	// Maintaining parallelism
815	+
816		/**
817	<	* Ensures eventCount on exit is different (mod 2^32) than on
765	<	* entry and wakes up all waiters
817	>	* Pushes worker onto the spare stack.
818		*/
819	<	private void signalEvent() {
820	<	int c;
821	<	do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
822	<	c = eventCount, c+1));
771	<	releaseWaiters();
819	>	final void pushSpare(ForkJoinWorkerThread w) {
820	>	int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
821	>	do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
822	>	w.nextSpare = spareWaiters,ns));
823		}
824
825		/**
826	<	* Advances eventCount and releases waiters until interference by
827	<	* other releasing threads is detected.
826	>	* Tries (once) to resume a spare if the number of running
827	>	* threads is less than target.
828		*/
829	<	final void signalWork() {
830	<	int c;
831	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, c=eventCount, c+1);
832	<	long top;
833	<	while ((top = eventWaiters) != 0L) {
834	<	int ec = eventCount;
835	<	ForkJoinWorkerThread[] ws = workers;
836	<	int n = ws.length;
837	<	for (;;) {
838	<	int i = ((int)(top & WAITER_ID_MASK)) - 1;
839	<	int e = (int)(top >>> EVENT_COUNT_SHIFT);
840	<	if (i < 0 || e == ec)
841	<	return;
842	<	ForkJoinWorkerThread w;
843	<	if (i < n && (w = ws[i]) != null &&
844	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
845	<	top, top = w.nextWaiter)) {
846	<	LockSupport.unpark(w);
847	<	if (top != eventWaiters) // let someone else take over
848	<	return;
829	>	private void tryResumeSpare() {
830	>	int sw, id;
831	>	ForkJoinWorkerThread[] ws = workers;
832	>	int n = ws.length;
833	>	ForkJoinWorkerThread w;
834	>	if ((sw = spareWaiters) != 0 &&
835	>	(id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
836	>	id < n && (w = ws[id]) != null &&
837	>	(workerCounts & RUNNING_COUNT_MASK) < parallelism &&
838	>	spareWaiters == sw &&
839	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
840	>	sw, w.nextSpare)) {
841	>	int c; // increment running count before resume
842	>	do {} while (!UNSAFE.compareAndSwapInt
843	>	(this, workerCountsOffset,
844	>	c = workerCounts, c + ONE_RUNNING));
845	>	if (w.tryUnsuspend())
846	>	LockSupport.unpark(w);
847	>	else   // back out if w was shutdown
848	>	decrementWorkerCounts(ONE_RUNNING, 0);
849	>	}
850	>	}
851	>
852	>	/**
853	>	* Tries to increase the number of running workers if below target
854	>	* parallelism: If a spare exists tries to resume it via
855	>	* tryResumeSpare.  Otherwise, if not enough total workers or all
856	>	* existing workers are busy, adds a new worker. In all cases also
857	>	* helps wake up releasable workers waiting for work.
858	>	*/
859	>	private void helpMaintainParallelism() {
860	>	int pc = parallelism;
861	>	int wc, rs, tc;
862	>	while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
863	>	(rs = runState) < TERMINATING) {
864	>	if (spareWaiters != 0)
865	>	tryResumeSpare();
866	>	else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
867	>	(tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
868	>	break;   // enough total
869	>	else if (runState == rs && workerCounts == wc &&
870	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
871	>	wc + (ONE_RUNNING|ONE_TOTAL))) {
872	>	ForkJoinWorkerThread w = null;
873	>	try {
874	>	w = factory.newThread(this);
875	>	} finally { // adjust on null or exceptional factory return
876	>	if (w == null) {
877	>	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
878	>	tryTerminate(false); // handle failure during shutdown
879	>	}
880	>	}
881	>	if (w == null)
882	>	break;
883	>	w.start(recordWorker(w), ueh);
884	>	if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
885	>	int c; // advance event count
886	>	UNSAFE.compareAndSwapInt(this, eventCountOffset,
887	>	c = eventCount, c+1);
888	>	break; // add at most one unless total below target
889		}
799	–	else
800	–	break;      // possibly stale; reread
890		}
891		}
892	+	if (eventWaiters != 0L)
893	+	releaseEventWaiters();
894		}
895
896		/**
897	<	* Blockss worker until terminating or event count
898	<	* advances from last value held by worker
897	>	* Callback from the oldest waiter in awaitEvent waking up after a
898	>	* period of non-use. If all workers are idle, tries (once) to
899	>	* shutdown an event waiter or a spare, if one exists. Note that
900	>	* we don't need CAS or locks here because the method is called
901	>	* only from one thread occasionally waking (and even misfires are
902	>	* OK). Note that until the shutdown worker fully terminates,
903	>	* workerCounts will overestimate total count, which is tolerable.
904		*
905	<	* @param w the calling worker thread
905	>	* @param ec the event count waited on by caller (to abort
906	>	* attempt if count has since changed).
907		*/
908	<	private void eventSync(ForkJoinWorkerThread w) {
909	<	int wec = w.lastEventCount;
910	<	long nextTop = (((long)wec << EVENT_COUNT_SHIFT) |
911	<	((long)(w.poolIndex + 1)));
912	<	long top;
913	<	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
914	<	(((int)(top = eventWaiters) & WAITER_ID_MASK) == 0 ||
915	<	(int)(top >>> EVENT_COUNT_SHIFT) == wec) &&
916	<	eventCount == wec) {
917	<	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
918	<	w.nextWaiter = top, nextTop)) {
919	<	accumulateStealCount(w); // transfer steals while idle
920	<	Thread.interrupted();    // clear/ignore interrupt
921	<	while (eventCount == wec)
922	<	w.doPark();
923	<	break;
908	>	private void tryShutdownUnusedWorker(int ec) {
909	>	if (runState == 0 && eventCount == ec) { // only trigger if all idle
910	>	ForkJoinWorkerThread[] ws = workers;
911	>	int n = ws.length;
912	>	ForkJoinWorkerThread w = null;
913	>	boolean shutdown = false;
914	>	int sw;
915	>	long h;
916	>	if ((sw = spareWaiters) != 0) { // prefer killing spares
917	>	int id = (sw & SPARE_ID_MASK) - 1;
918	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
919	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
920	>	sw, w.nextSpare))
921	>	shutdown = true;
922	>	}
923	>	else if ((h = eventWaiters) != 0L) {
924	>	long nh;
925	>	int id = ((int)(h & WAITER_ID_MASK)) - 1;
926	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
927	>	(nh = w.nextWaiter) != 0L && // keep at least one worker
928	>	UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
929	>	shutdown = true;
930	>	}
931	>	if (w != null && shutdown) {
932	>	w.shutdown();
933	>	LockSupport.unpark(w);
934		}
935		}
936	<	w.lastEventCount = eventCount;
936	>	releaseEventWaiters(); // in case of interference
937		}
938
939		/**
940		* Callback from workers invoked upon each top-level action (i.e.,
941	<	* stealing a task or taking a submission and running
942	<	* it). Performs one or both of the following:
941	>	* stealing a task or taking a submission and running it).
942	>	* Performs one or more of the following:
943		*
944	<	* * If the worker cannot find work, updates its active status to
945	<	* inactive and updates activeCount unless there is contention, in
946	<	* which case it may try again (either in this or a subsequent
947	<	* call).  Additionally, awaits the next task event and/or helps
948	<	* wake up other releasable waiters.
949	<	*
950	<	* * If there are too many running threads, suspends this worker
951	<	* (first forcing inactivation if necessary).  If it is not
952	<	* resumed before a keepAlive elapses, the worker may be "trimmed"
953	<	* -- killed while suspended within suspendAsSpare. Otherwise,
954	<	* upon resume it rechecks to make sure that it is still needed.
944	>	* 1. If the worker is active and either did not run a task
945	>	*    or there are too many workers, try to set its active status
946	>	*    to inactive and update activeCount. On contention, we may
947	>	*    try again in this or a subsequent call.
948	>	*
949	>	* 2. If not enough total workers, help create some.
950	>	*
951	>	* 3. If there are too many running workers, suspend this worker
952	>	*    (first forcing inactive if necessary).  If it is not needed,
953	>	*    it may be shutdown while suspended (via
954	>	*    tryShutdownUnusedWorker).  Otherwise, upon resume it
955	>	*    rechecks running thread count and need for event sync.
956	>	*
957	>	* 4. If worker did not run a task, await the next task event via
958	>	*    eventSync if necessary (first forcing inactivation), upon
959	>	*    which the worker may be shutdown via
960	>	*    tryShutdownUnusedWorker.  Otherwise, help release any
961	>	*    existing event waiters that are now releasable,
962		*
963		* @param w the worker
964	<	* @param retries the number of scans by caller failing to find work
851	<	* find any (in which case it may block waiting for work).
964	>	* @param ran true if worker ran a task since last call to this method
965		*/
966	<	final void preStep(ForkJoinWorkerThread w, int retries) {
966	>	final void preStep(ForkJoinWorkerThread w, boolean ran) {
967	>	int wec = w.lastEventCount;
968		boolean active = w.active;
969	<	boolean inactivate = active && retries > 0;
970	<	for (;;) {
971	<	int rs, wc;
972	<	if (inactivate &&
973	<	UNSAFE.compareAndSwapInt(this, runStateOffset,
974	<	rs = runState, rs - ONE_ACTIVE))
969	>	boolean inactivate = false;
970	>	int pc = parallelism;
971	>	int rs;
972	>	while (w.runState == 0 && (rs = runState) < TERMINATING) {
973	>	if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
974	>	UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
975		inactivate = active = w.active = false;
976	<	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= parallelism) {
977	<	if (retries > 0) {
978	<	if (retries > 1 && !active)
979	<	eventSync(w);
980	<	releaseWaiters();
976	>	int wc = workerCounts;
977	>	if ((wc & RUNNING_COUNT_MASK) > pc) {
978	>	if (!(inactivate |= active) && // must inactivate to suspend
979	>	workerCounts == wc &&      // try to suspend as spare
980	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
981	>	wc, wc - ONE_RUNNING))
982	>	w.suspendAsSpare();
983	>	}
984	>	else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
985	>	helpMaintainParallelism();     // not enough workers
986	>	else if (!ran) {
987	>	long h = eventWaiters;
988	>	int ec = eventCount;
989	>	if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
990	>	releaseEventWaiters();     // release others before waiting
991	>	else if (ec != wec) {
992	>	w.lastEventCount = ec;     // no need to wait
993	>	break;
994		}
995	<	break;
995	>	else if (!(inactivate |= active))
996	>	eventSync(w, wec);         // must inactivate before sync
997		}
998	<	if (!(inactivate |= active) &&  // must inactivate to suspend
871	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
872	<	wc, wc - ONE_RUNNING) &&
873	<	!w.suspendAsSpare())             // false if trimmed
998	>	else
999		break;
1000		}
1001		}
1002
1003		/**
1004	<	* Awaits join of the given task if enough threads, or can resume
1005	<	* or create a spare. Fails (in which case the given task might
881	<	* not be done) upon contention or lack of decision about
882	<	* blocking.
883	<	*
884	<	* We allow blocking if:
885	<	*
886	<	* 1. There would still be at least as many running threads as
887	<	*    parallelism level if this thread blocks.
888	<	*
889	<	* 2. A spare is resumed to replace this worker. We tolerate
890	<	*    races in the decision to replace when a spare is found.
891	<	*    This may release too many, but if so, the superfluous ones
892	<	*    will re-suspend via preStep().
893	<	*
894	<	* 3. After #spares repeated retries, there are fewer than #spare
895	<	*    threads not running. We allow this slack to avoid hysteresis
896	<	*    and as a hedge against lag/uncertainty of running count
897	<	*    estimates when signalling or unblocking stalls.
898	<	*
899	<	* 4. All existing workers are busy (as rechecked via #spares
900	<	*    repeated retries by caller) and a new spare is created.
901	<	*
902	<	* If none of the above hold, we escape out by re-incrementing
903	<	* count and returning to caller, which can retry later.
1004	>	* Helps and/or blocks awaiting join of the given task.
1005	>	* See above for explanation.
1006		*
1007		* @param joinMe the task to join
1008	<	* @param retries the number of calls to this method for this join
1008	>	* @param worker the current worker thread
1009		*/
1010	<	final void tryAwaitJoin(ForkJoinTask<?> joinMe, int retries) {
1011	<	int pc = parallelism;
1012	<	boolean running = true; // false when running count decremented
1013	<	outer:while (joinMe.status >= 0) {
1014	<	int wc = workerCounts;
1015	<	int rc = wc & RUNNING_COUNT_MASK;
1016	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
1017	<	if (running) { // replace with spare or decrement count
1018	<	if (rc <= pc && tc > pc &&
1019	<	(retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1020	<	ForkJoinWorkerThread[] ws = workers; // search for spare
1021	<	int nws = ws.length;
1022	<	for (int i = 0; i < nws; ++i) {
1023	<	ForkJoinWorkerThread w = ws[i];
1024	<	if (w != null && w.isSuspended()) {
1025	<	if ((workerCounts & RUNNING_COUNT_MASK) > pc)
1026	<	continue outer;
1027	<	if (joinMe.status < 0)
1028	<	break outer;
1029	<	if (w.tryResumeSpare()) {
1030	<	running = false;
1031	<	break outer;
1032	<	}
1033	<	continue outer; // rescan on failure to resume
1034	<	}
1035	<	}
1036	<	}
935	<	if ((rc <= pc && (rc == 0 || --retries < 0)) || // no retry
936	<	joinMe.status < 0)
937	<	break;
938	<	if (workerCounts == wc &&
939	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
940	<	wc, wc - ONE_RUNNING))
941	<	running = false;
942	<	}
943	<	else { // allow blocking if enough threads
944	<	int sc = tc - pc + 1;          // = spares, plus the one to add
945	<	if (sc > 0 && rc > 0 && rc >= pc - sc && rc > pc - retries)
946	<	break;
947	<	if (--retries > sc && tc < MAX_THREADS &&
948	<	tc == (runState & ACTIVE_COUNT_MASK) &&
949	<	workerCounts == wc &&
950	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
951	<	wc + (ONE_RUNNING|ONE_TOTAL))) {
952	<	addWorker();
953	<	break;
954	<	}
955	<	if (workerCounts == wc &&
956	<	UNSAFE.compareAndSwapInt (this, workerCountsOffset,
957	<	wc, wc + ONE_RUNNING)) {
958	<	running = true;            // back out; allow retry
959	<	break;
960	<	}
1010	>	final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1011	>	int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1012	>	while (joinMe.status >= 0) {
1013	>	int wc;
1014	>	worker.helpJoinTask(joinMe);
1015	>	if (joinMe.status < 0)
1016	>	break;
1017	>	else if (retries > 0)
1018	>	--retries;
1019	>	else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1020	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1021	>	wc, wc - ONE_RUNNING)) {
1022	>	int stat, c; long h;
1023	>	while ((stat = joinMe.status) >= 0 &&
1024	>	(h = eventWaiters) != 0L && // help release others
1025	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1026	>	releaseEventWaiters();
1027	>	if (stat >= 0 &&
1028	>	((workerCounts & RUNNING_COUNT_MASK) == 0 ||
1029	>	(stat =
1030	>	joinMe.internalAwaitDone(JOIN_TIMEOUT_MILLIS)) >= 0))
1031	>	helpMaintainParallelism(); // timeout or no running workers
1032	>	do {} while (!UNSAFE.compareAndSwapInt
1033	>	(this, workerCountsOffset,
1034	>	c = workerCounts, c + ONE_RUNNING));
1035	>	if (stat < 0)
1036	>	break;   // else restart
1037		}
1038		}
963	–	if (!running) { // can block
964	–	int c;                      // to inline incrementRunningCount
965	–	joinMe.internalAwaitDone();
966	–	do {} while (!UNSAFE.compareAndSwapInt
967	–	(this, workerCountsOffset,
968	–	c = workerCounts, c + ONE_RUNNING));
969	–	}
1039		}
1040
1041		/**
1042	<	* Same idea as (and shares many code snippets with) tryAwaitJoin,
974	<	* but self-contained because there are no caller retries.
975	<	* TODO: Rework to use simpler API.
1042	>	* Same idea as awaitJoin, but no helping, retries, or timeouts.
1043		*/
1044		final void awaitBlocker(ManagedBlocker blocker)
1045		throws InterruptedException {
1046	<	int pc = parallelism;
980	<	boolean running = true;
981	<	int retries = 0;
982	<	boolean done;
983	<	outer:while (!(done = blocker.isReleasable())) {
1046	>	while (!blocker.isReleasable()) {
1047		int wc = workerCounts;
1048	<	int rc = wc & RUNNING_COUNT_MASK;
1049	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
1050	<	if (running) {
1051	<	if (rc <= pc && tc > pc &&
1052	<	(retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1053	<	ForkJoinWorkerThread[] ws = workers;
1054	<	int nws = ws.length;
1055	<	for (int i = 0; i < nws; ++i) {
1056	<	ForkJoinWorkerThread w = ws[i];
1057	<	if (w != null && w.isSuspended()) {
1058	<	if ((workerCounts & RUNNING_COUNT_MASK) > pc)
1059	<	continue outer;
1060	<	if (done = blocker.isReleasable())
1061	<	break outer;
999	<	if (w.tryResumeSpare()) {
1000	<	running = false;
1001	<	break outer;
1002	<	}
1003	<	continue outer;
1004	<	}
1048	>	if ((wc & RUNNING_COUNT_MASK) != 0 &&
1049	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1050	>	wc, wc - ONE_RUNNING)) {
1051	>	try {
1052	>	while (!blocker.isReleasable()) {
1053	>	long h = eventWaiters;
1054	>	if (h != 0L &&
1055	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1056	>	releaseEventWaiters();
1057	>	else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1058	>	runState < TERMINATING)
1059	>	helpMaintainParallelism();
1060	>	else if (blocker.block())
1061	>	break;
1062		}
1063	<	if (done = blocker.isReleasable())
1064	<	break;
1063	>	} finally {
1064	>	int c;
1065	>	do {} while (!UNSAFE.compareAndSwapInt
1066	>	(this, workerCountsOffset,
1067	>	c = workerCounts, c + ONE_RUNNING));
1068		}
1009	–	if (rc > 0 && workerCounts == wc &&
1010	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1011	–	wc, wc - ONE_RUNNING)) {
1012	–	running = false;
1013	–	if (rc > pc)
1014	–	break;
1015	–	}
1016	–	}
1017	–	else if (rc >= pc)
1069		break;
1019	–	else if (tc < MAX_THREADS &&
1020	–	tc == (runState & ACTIVE_COUNT_MASK) &&
1021	–	workerCounts == wc &&
1022	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1023	–	wc + (ONE_RUNNING|ONE_TOTAL))) {
1024	–	addWorker();
1025	–	break;
1026	–	}
1027	–	else if (workerCounts == wc &&
1028	–	UNSAFE.compareAndSwapInt (this, workerCountsOffset,
1029	–	wc, wc + ONE_RUNNING)) {
1030	–	Thread.yield();
1031	–	++retries;
1032	–	running = true;            // allow rescan
1033	–	}
1034	–	}
1035	–
1036	–	try {
1037	–	if (!done)
1038	–	do {} while (!blocker.isReleasable() && !blocker.block());
1039	–	} finally {
1040	–	if (!running) {
1041	–	int c;
1042	–	do {} while (!UNSAFE.compareAndSwapInt
1043	–	(this, workerCountsOffset,
1044	–	c = workerCounts, c + ONE_RUNNING));
1070		}
1071		}
1072		}
#	Line 1074 | Line 1099 | public class ForkJoinPool extends Abstra
1099
1100		/**
1101		* Actions on transition to TERMINATING
1102	+	*
1103	+	* Runs up to four passes through workers: (0) shutting down each
1104	+	* (without waking up if parked) to quickly spread notifications
1105	+	* without unnecessary bouncing around event queues etc (1) wake
1106	+	* up and help cancel tasks (2) interrupt (3) mop up races with
1107	+	* interrupted workers
1108		*/
1109		private void startTerminating() {
1110	<	for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1111	<	cancelSubmissions();
1112	<	shutdownWorkers();
1113	<	cancelWorkerTasks();
1114	<	signalEvent();
1115	<	interruptWorkers();
1110	>	cancelSubmissions();
1111	>	for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1112	>	int c; // advance event count
1113	>	UNSAFE.compareAndSwapInt(this, eventCountOffset,
1114	>	c = eventCount, c+1);
1115	>	eventWaiters = 0L; // clobber lists
1116	>	spareWaiters = 0;
1117	>	for (ForkJoinWorkerThread w : workers) {
1118	>	if (w != null) {
1119	>	w.shutdown();
1120	>	if (passes > 0 && !w.isTerminated()) {
1121	>	w.cancelTasks();
1122	>	LockSupport.unpark(w);
1123	>	if (passes > 1) {
1124	>	try {
1125	>	w.interrupt();
1126	>	} catch (SecurityException ignore) {
1127	>	}
1128	>	}
1129	>	}
1130	>	}
1131	>	}
1132		}
1133		}
1134
1135		/**
1136	<	* Clear out and cancel submissions, ignoring exceptions
1136	>	* Clears out and cancels submissions, ignoring exceptions.
1137		*/
1138		private void cancelSubmissions() {
1139		ForkJoinTask<?> task;
#	Line 1098 | Line 1145 | public class ForkJoinPool extends Abstra
1145		}
1146		}
1147
1101	–	/**
1102	–	* Sets all worker run states to at least shutdown,
1103	–	* also resuming suspended workers
1104	–	*/
1105	–	private void shutdownWorkers() {
1106	–	ForkJoinWorkerThread[] ws = workers;
1107	–	int nws = ws.length;
1108	–	for (int i = 0; i < nws; ++i) {
1109	–	ForkJoinWorkerThread w = ws[i];
1110	–	if (w != null)
1111	–	w.shutdown();
1112	–	}
1113	–	}
1114	–
1115	–	/**
1116	–	* Clears out and cancels all locally queued tasks
1117	–	*/
1118	–	private void cancelWorkerTasks() {
1119	–	ForkJoinWorkerThread[] ws = workers;
1120	–	int nws = ws.length;
1121	–	for (int i = 0; i < nws; ++i) {
1122	–	ForkJoinWorkerThread w = ws[i];
1123	–	if (w != null)
1124	–	w.cancelTasks();
1125	–	}
1126	–	}
1127	–
1128	–	/**
1129	–	* Unsticks all workers blocked on joins etc
1130	–	*/
1131	–	private void interruptWorkers() {
1132	–	ForkJoinWorkerThread[] ws = workers;
1133	–	int nws = ws.length;
1134	–	for (int i = 0; i < nws; ++i) {
1135	–	ForkJoinWorkerThread w = ws[i];
1136	–	if (w != null && !w.isTerminated()) {
1137	–	try {
1138	–	w.interrupt();
1139	–	} catch (SecurityException ignore) {
1140	–	}
1141	–	}
1142	–	}
1143	–	}
1144	–
1148		// misc support for ForkJoinWorkerThread
1149
1150		/**
1151	<	* Returns pool number
1151	>	* Returns pool number.
1152		*/
1153		final int getPoolNumber() {
1154		return poolNumber;
1155		}
1156
1157		/**
1158	<	* Accumulates steal count from a worker, clearing
1159	<	* the worker's value
1158	>	* Tries to accumulate steal count from a worker, clearing
1159	>	* the worker's value if successful.
1160	>	*
1161	>	* @return true if worker steal count now zero
1162		*/
1163	<	final void accumulateStealCount(ForkJoinWorkerThread w) {
1163	>	final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1164		int sc = w.stealCount;
1165	<	if (sc != 0) {
1166	<	long c;
1167	<	w.stealCount = 0;
1168	<	do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1169	<	c = stealCount, c + sc));
1165	>	long c = stealCount;
1166	>	// CAS even if zero, for fence effects
1167	>	if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1168	>	if (sc != 0)
1169	>	w.stealCount = 0;
1170	>	return true;
1171		}
1172	+	return sc == 0;
1173		}
1174
1175		/**
#	Line 1171 | Line 1178 | public class ForkJoinPool extends Abstra
1178		*/
1179		final int idlePerActive() {
1180		int pc = parallelism; // use parallelism, not rc
1181	<	int ac = runState;    // no mask -- artifically boosts during shutdown
1181	>	int ac = runState;    // no mask -- artificially boosts during shutdown
1182		// Use exact results for small values, saturate past 4
1183	<	return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1183	>	return ((pc <= ac) ? 0 :
1184	>	(pc >>> 1 <= ac) ? 1 :
1185	>	(pc >>> 2 <= ac) ? 3 :
1186	>	pc >>> 3);
1187		}
1188
1189		// Public and protected methods
#	Line 1223 | Line 1233 | public class ForkJoinPool extends Abstra
1233		* use {@link #defaultForkJoinWorkerThreadFactory}.
1234		* @param handler the handler for internal worker threads that
1235		* terminate due to unrecoverable errors encountered while executing
1236	<	* tasks. For default value, use <code>null</code>.
1236	>	* tasks. For default value, use {@code null}.
1237		* @param asyncMode if true,
1238		* establishes local first-in-first-out scheduling mode for forked
1239		* tasks that are never joined. This mode may be more appropriate
1240		* than default locally stack-based mode in applications in which
1241		* worker threads only process event-style asynchronous tasks.
1242	<	* For default value, use <code>false</code>.
1242	>	* For default value, use {@code false}.
1243		* @throws IllegalArgumentException if parallelism less than or
1244		*         equal to zero, or greater than implementation limit
1245		* @throws NullPointerException if the factory is null
#	Line 1245 | Line 1255 | public class ForkJoinPool extends Abstra
1255		checkPermission();
1256		if (factory == null)
1257		throw new NullPointerException();
1258	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
1258	>	if (parallelism <= 0 || parallelism > MAX_WORKERS)
1259		throw new IllegalArgumentException();
1260		this.parallelism = parallelism;
1261		this.factory = factory;
#	Line 1264 | Line 1274 | public class ForkJoinPool extends Abstra
1274		* @param pc the initial parallelism level
1275		*/
1276		private static int initialArraySizeFor(int pc) {
1277	<	// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1278	<	int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1277	>	// If possible, initially allocate enough space for one spare
1278	>	int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1279	>	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1280		size |= size >>> 1;
1281		size |= size >>> 2;
1282		size |= size >>> 4;
#	Line 1284 | Line 1295 | public class ForkJoinPool extends Abstra
1295		if (runState >= SHUTDOWN)
1296		throw new RejectedExecutionException();
1297		submissionQueue.offer(task);
1298	<	signalEvent();
1299	<	ensureEnoughTotalWorkers();
1298	>	int c; // try to increment event count -- CAS failure OK
1299	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1300	>	helpMaintainParallelism(); // create, start, or resume some workers
1301		}
1302
1303		/**
1304		* Performs the given task, returning its result upon completion.
1293	–	* If the caller is already engaged in a fork/join computation in
1294	–	* the current pool, this method is equivalent in effect to
1295	–	* {@link ForkJoinTask#invoke}.
1305		*
1306		* @param task the task
1307		* @return the task's result
#	Line 1307 | Line 1316 | public class ForkJoinPool extends Abstra
1316
1317		/**
1318		* Arranges for (asynchronous) execution of the given task.
1310	–	* If the caller is already engaged in a fork/join computation in
1311	–	* the current pool, this method is equivalent in effect to
1312	–	* {@link ForkJoinTask#fork}.
1319		*
1320		* @param task the task
1321		* @throws NullPointerException if the task is null
#	Line 1338 | Line 1344 | public class ForkJoinPool extends Abstra
1344
1345		/**
1346		* Submits a ForkJoinTask for execution.
1341	–	* If the caller is already engaged in a fork/join computation in
1342	–	* the current pool, this method is equivalent in effect to
1343	–	* {@link ForkJoinTask#fork}.
1347		*
1348		* @param task the task to submit
1349		* @return the task
#	Line 1446 | Line 1449 | public class ForkJoinPool extends Abstra
1449
1450		/**
1451		* Returns the number of worker threads that have started but not
1452	<	* yet terminated.  This result returned by this method may differ
1452	>	* yet terminated.  The result returned by this method may differ
1453		* from {@link #getParallelism} when threads are created to
1454		* maintain parallelism when others are cooperatively blocked.
1455		*
#	Line 1531 | Line 1534 | public class ForkJoinPool extends Abstra
1534		*/
1535		public long getQueuedTaskCount() {
1536		long count = 0;
1537	<	ForkJoinWorkerThread[] ws = workers;
1535	<	int nws = ws.length;
1536	<	for (int i = 0; i < nws; ++i) {
1537	<	ForkJoinWorkerThread w = ws[i];
1537	>	for (ForkJoinWorkerThread w : workers)
1538		if (w != null)
1539		count += w.getQueueSize();
1540	–	}
1540		return count;
1541		}
1542
#	Line 1591 | Line 1590 | public class ForkJoinPool extends Abstra
1590		* @return the number of elements transferred
1591		*/
1592		protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1593	<	int n = submissionQueue.drainTo(c);
1594	<	ForkJoinWorkerThread[] ws = workers;
1596	<	int nws = ws.length;
1597	<	for (int i = 0; i < nws; ++i) {
1598	<	ForkJoinWorkerThread w = ws[i];
1599	<	if (w != null)
1600	<	n += w.drainTasksTo(c);
1601	<	}
1602	<	return n;
1603	<	}
1604	<
1605	<	/**
1606	<	* Returns count of total parks by existing workers.
1607	<	* Used during development only since not meaningful to users.
1608	<	*/
1609	<	private int collectParkCount() {
1610	<	int count = 0;
1611	<	ForkJoinWorkerThread[] ws = workers;
1612	<	int nws = ws.length;
1613	<	for (int i = 0; i < nws; ++i) {
1614	<	ForkJoinWorkerThread w = ws[i];
1593	>	int count = submissionQueue.drainTo(c);
1594	>	for (ForkJoinWorkerThread w : workers)
1595		if (w != null)
1596	<	count += w.parkCount;
1617	<	}
1596	>	count += w.drainTasksTo(c);
1597		return count;
1598		}
1599
#	Line 1635 | Line 1614 | public class ForkJoinPool extends Abstra
1614		int pc = parallelism;
1615		int rs = runState;
1616		int ac = rs & ACTIVE_COUNT_MASK;
1638	–	//        int pk = collectParkCount();
1617		return super.toString() +
1618		"[" + runLevelToString(rs) +
1619		", parallelism = " + pc +
#	Line 1645 | Line 1623 | public class ForkJoinPool extends Abstra
1623		", steals = " + st +
1624		", tasks = " + qt +
1625		", submissions = " + qs +
1648	–	//            ", parks = " + pk +
1626		"]";
1627		}
1628
#	Line 1743 | Line 1720 | public class ForkJoinPool extends Abstra
1720		throws InterruptedException {
1721		try {
1722		return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1723	<	} catch(TimeoutException ex) {
1723	>	} catch (TimeoutException ex) {
1724		return false;
1725		}
1726		}
#	Line 1752 | Line 1729 | public class ForkJoinPool extends Abstra
1729		* Interface for extending managed parallelism for tasks running
1730		* in {@link ForkJoinPool}s.
1731		*
1732	<	* <p>A {@code ManagedBlocker} provides two methods.
1733	<	* Method {@code isReleasable} must return {@code true} if
1734	<	* blocking is not necessary. Method {@code block} blocks the
1735	<	* current thread if necessary (perhaps internally invoking
1736	<	* {@code isReleasable} before actually blocking).
1732	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
1733	>	* {@code isReleasable} must return {@code true} if blocking is
1734	>	* not necessary. Method {@code block} blocks the current thread
1735	>	* if necessary (perhaps internally invoking {@code isReleasable}
1736	>	* before actually blocking). The unusual methods in this API
1737	>	* accommodate synchronizers that may, but don't usually, block
1738	>	* for long periods. Similarly, they allow more efficient internal
1739	>	* handling of cases in which additional workers may be, but
1740	>	* usually are not, needed to ensure sufficient parallelism.
1741	>	* Toward this end, implementations of method {@code isReleasable}
1742	>	* must be amenable to repeated invocation.
1743		*
1744		* <p>For example, here is a ManagedBlocker based on a
1745		* ReentrantLock:
#	Line 1774 | Line 1757 | public class ForkJoinPool extends Abstra
1757		*     return hasLock || (hasLock = lock.tryLock());
1758		*   }
1759		* }}</pre>
1760	+	*
1761	+	* <p>Here is a class that possibly blocks waiting for an
1762	+	* item on a given queue:
1763	+	*  <pre> {@code
1764	+	* class QueueTaker<E> implements ManagedBlocker {
1765	+	*   final BlockingQueue<E> queue;
1766	+	*   volatile E item = null;
1767	+	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1768	+	*   public boolean block() throws InterruptedException {
1769	+	*     if (item == null)
1770	+	*       item = queue.take();
1771	+	*     return true;
1772	+	*   }
1773	+	*   public boolean isReleasable() {
1774	+	*     return item != null || (item = queue.poll()) != null;
1775	+	*   }
1776	+	*   public E getItem() { // call after pool.managedBlock completes
1777	+	*     return item;
1778	+	*   }
1779	+	* }}</pre>
1780		*/
1781		public static interface ManagedBlocker {
1782		/**
#	Line 1816 | Line 1819 | public class ForkJoinPool extends Abstra
1819		public static void managedBlock(ManagedBlocker blocker)
1820		throws InterruptedException {
1821		Thread t = Thread.currentThread();
1822	<	if (t instanceof ForkJoinWorkerThread)
1823	<	((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
1822	>	if (t instanceof ForkJoinWorkerThread) {
1823	>	ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1824	>	w.pool.awaitBlocker(blocker);
1825	>	}
1826		else {
1827		do {} while (!blocker.isReleasable() && !blocker.block());
1828		}
#	Line 1845 | Line 1850 | public class ForkJoinPool extends Abstra
1850		private static final long eventCountOffset =
1851		objectFieldOffset("eventCount", ForkJoinPool.class);
1852		private static final long eventWaitersOffset =
1853	<	objectFieldOffset("eventWaiters",ForkJoinPool.class);
1853	>	objectFieldOffset("eventWaiters", ForkJoinPool.class);
1854		private static final long stealCountOffset =
1855	<	objectFieldOffset("stealCount",ForkJoinPool.class);
1855	>	objectFieldOffset("stealCount", ForkJoinPool.class);
1856	>	private static final long spareWaitersOffset =
1857	>	objectFieldOffset("spareWaiters", ForkJoinPool.class);
1858
1859		private static long objectFieldOffset(String field, Class<?> klazz) {
1860		try {

Diff Legend

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