ViewVC Help

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

root/jsr166/jsr166/src/jsr166e/ForkJoinPool.java

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.15 by jsr166, Sun Nov 18 06:31:13 2012 UTC vs.
Revision 1.62 by jsr166, Fri Jul 19 19:34:43 2013 UTC

#	Line 6 | Line 6
6
7		package jsr166e;
8
9	+	import java.lang.Thread.UncaughtExceptionHandler;
10		import java.util.ArrayList;
11		import java.util.Arrays;
12		import java.util.Collection;
#	Line 17 | Line 18 | import java.util.concurrent.ExecutorServ
18		import java.util.concurrent.Future;
19		import java.util.concurrent.RejectedExecutionException;
20		import java.util.concurrent.RunnableFuture;
20	–	import java.util.concurrent.ThreadLocalRandom;
21		import java.util.concurrent.TimeUnit;
22
23		/**
#	Line 38 | Line 38 | import java.util.concurrent.TimeUnit;
38		* ForkJoinPool}s may also be appropriate for use with event-style
39		* tasks that are never joined.
40		*
41	<	* <p>A static {@link #commonPool} is available and appropriate for
41	>	* <p>A static {@link #commonPool()} is available and appropriate for
42		* most applications. The common pool is used by any ForkJoinTask that
43		* is not explicitly submitted to a specified pool. Using the common
44		* pool normally reduces resource usage (its threads are slowly
#	Line 50 | Line 50 | import java.util.concurrent.TimeUnit;
50		* level; by default, equal to the number of available processors. The
51		* pool attempts to maintain enough active (or available) threads by
52		* dynamically adding, suspending, or resuming internal worker
53	<	* threads, even if some tasks are stalled waiting to join
54	<	* others. However, no such adjustments are guaranteed in the face of
55	<	* blocked IO or other unmanaged synchronization. The nested {@link
53	>	* threads, even if some tasks are stalled waiting to join others.
54	>	* However, no such adjustments are guaranteed in the face of blocked
55	>	* I/O or other unmanaged synchronization. The nested {@link
56		* ManagedBlocker} interface enables extension of the kinds of
57		* synchronization accommodated.
58		*
#	Line 63 | Line 63 | import java.util.concurrent.TimeUnit;
63		* {@link #toString} returns indications of pool state in a
64		* convenient form for informal monitoring.
65		*
66	<	* <p> As is the case with other ExecutorServices, there are three
66	>	* <p>As is the case with other ExecutorServices, there are three
67		* main task execution methods summarized in the following table.
68		* These are designed to be used primarily by clients not already
69		* engaged in fork/join computations in the current pool.  The main
#	Line 76 | Line 76 | import java.util.concurrent.TimeUnit;
76		* there is little difference among choice of methods.
77		*
78		* <table BORDER CELLPADDING=3 CELLSPACING=1>
79	+	* <caption>Summary of task execution methods</caption>
80		*  <tr>
81		*    <td></td>
82		*    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
83		*    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
84		*  </tr>
85		*  <tr>
86	<	*    <td> <b>Arrange async execution</td>
86	>	*    <td> <b>Arrange async execution</b></td>
87		*    <td> {@link #execute(ForkJoinTask)}</td>
88		*    <td> {@link ForkJoinTask#fork}</td>
89		*  </tr>
90		*  <tr>
91	<	*    <td> <b>Await and obtain result</td>
91	>	*    <td> <b>Await and obtain result</b></td>
92		*    <td> {@link #invoke(ForkJoinTask)}</td>
93		*    <td> {@link ForkJoinTask#invoke}</td>
94		*  </tr>
95		*  <tr>
96	<	*    <td> <b>Arrange exec and obtain Future</td>
96	>	*    <td> <b>Arrange exec and obtain Future</b></td>
97		*    <td> {@link #submit(ForkJoinTask)}</td>
98		*    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
99		*  </tr>
100		* </table>
101		*
102		* <p>The common pool is by default constructed with default
103	<	* parameters, but these may be controlled by setting three {@link
104	<	* System#getProperty properties} with prefix {@code
105	<	* java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
106	<	* an integer greater than zero, {@code threadFactory} -- the class
107	<	* name of a {@link ForkJoinWorkerThreadFactory}, and {@code
108	<	* exceptionHandler} -- the class name of a {@link
109	<	* java.lang.Thread.UncaughtExceptionHandler
110	<	* Thread.UncaughtExceptionHandler}. Upon any error in establishing
111	<	* these settings, default parameters are used.
103	>	* parameters, but these may be controlled by setting three
104	>	* {@linkplain System#getProperty system properties}:
105	>	* <ul>
106	>	* <li>{@code java.util.concurrent.ForkJoinPool.common.parallelism}
107	>	* - the parallelism level, a non-negative integer
108	>	* <li>{@code java.util.concurrent.ForkJoinPool.common.threadFactory}
109	>	* - the class name of a {@link ForkJoinWorkerThreadFactory}
110	>	* <li>{@code java.util.concurrent.ForkJoinPool.common.exceptionHandler}
111	>	* - the class name of a {@link UncaughtExceptionHandler}
112	>	* </ul>
113	>	* The system class loader is used to load these classes.
114	>	* Upon any error in establishing these settings, default parameters
115	>	* are used. It is possible to disable or limit the use of threads in
116	>	* the common pool by setting the parallelism property to zero, and/or
117	>	* using a factory that may return {@code null}.
118		*
119		* <p><b>Implementation notes</b>: This implementation restricts the
120		* maximum number of running threads to 32767. Attempts to create
#	Line 153 | Line 160 | public class ForkJoinPool extends Abstra
160		* (http://research.sun.com/scalable/pubs/index.html) and
161		* "Idempotent work stealing" by Michael, Saraswat, and Vechev,
162		* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
163	<	* The main differences ultimately stem from GC requirements that
164	<	* we null out taken slots as soon as we can, to maintain as small
165	<	* a footprint as possible even in programs generating huge
166	<	* numbers of tasks. To accomplish this, we shift the CAS
167	<	* arbitrating pop vs poll (steal) from being on the indices
168	<	* ("base" and "top") to the slots themselves.  So, both a
169	<	* successful pop and poll mainly entail a CAS of a slot from
170	<	* non-null to null.  Because we rely on CASes of references, we
171	<	* do not need tag bits on base or top.  They are simple ints as
172	<	* used in any circular array-based queue (see for example
173	<	* ArrayDeque).  Updates to the indices must still be ordered in a
174	<	* way that guarantees that top == base means the queue is empty,
175	<	* but otherwise may err on the side of possibly making the queue
176	<	* appear nonempty when a push, pop, or poll have not fully
177	<	* committed. Note that this means that the poll operation,
178	<	* considered individually, is not wait-free. One thief cannot
179	<	* successfully continue until another in-progress one (or, if
180	<	* previously empty, a push) completes.  However, in the
181	<	* aggregate, we ensure at least probabilistic non-blockingness.
182	<	* If an attempted steal fails, a thief always chooses a different
183	<	* random victim target to try next. So, in order for one thief to
184	<	* progress, it suffices for any in-progress poll or new push on
185	<	* any empty queue to complete. (This is why we normally use
186	<	* method pollAt and its variants that try once at the apparent
187	<	* base index, else consider alternative actions, rather than
188	<	* method poll.)
163	>	* See also "Correct and Efficient Work-Stealing for Weak Memory
164	>	* Models" by Le, Pop, Cohen, and Nardelli, PPoPP 2013
165	>	* (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an
166	>	* analysis of memory ordering (atomic, volatile etc) issues.  The
167	>	* main differences ultimately stem from GC requirements that we
168	>	* null out taken slots as soon as we can, to maintain as small a
169	>	* footprint as possible even in programs generating huge numbers
170	>	* of tasks. To accomplish this, we shift the CAS arbitrating pop
171	>	* vs poll (steal) from being on the indices ("base" and "top") to
172	>	* the slots themselves.  So, both a successful pop and poll
173	>	* mainly entail a CAS of a slot from non-null to null.  Because
174	>	* we rely on CASes of references, we do not need tag bits on base
175	>	* or top.  They are simple ints as used in any circular
176	>	* array-based queue (see for example ArrayDeque).  Updates to the
177	>	* indices must still be ordered in a way that guarantees that top
178	>	* == base means the queue is empty, but otherwise may err on the
179	>	* side of possibly making the queue appear nonempty when a push,
180	>	* pop, or poll have not fully committed. Note that this means
181	>	* that the poll operation, considered individually, is not
182	>	* wait-free. One thief cannot successfully continue until another
183	>	* in-progress one (or, if previously empty, a push) completes.
184	>	* However, in the aggregate, we ensure at least probabilistic
185	>	* non-blockingness.  If an attempted steal fails, a thief always
186	>	* chooses a different random victim target to try next. So, in
187	>	* order for one thief to progress, it suffices for any
188	>	* in-progress poll or new push on any empty queue to
189	>	* complete. (This is why we normally use method pollAt and its
190	>	* variants that try once at the apparent base index, else
191	>	* consider alternative actions, rather than method poll.)
192		*
193		* This approach also enables support of a user mode in which local
194		* task processing is in FIFO, not LIFO order, simply by using
#	Line 197 | Line 207 | public class ForkJoinPool extends Abstra
207		* for work-stealing (this would contaminate lifo/fifo
208		* processing). Instead, we randomly associate submission queues
209		* with submitting threads, using a form of hashing.  The
210	<	* ThreadLocal Submitter class contains a value initially used as
211	<	* a hash code for choosing existing queues, but may be randomly
212	<	* repositioned upon contention with other submitters.  In
213	<	* essence, submitters act like workers except that they are
214	<	* restricted to executing local tasks that they submitted (or in
215	<	* the case of CountedCompleters, others with the same root task).
216	<	* However, because most shared/external queue operations are more
217	<	* expensive than internal, and because, at steady state, external
218	<	* submitters will compete for CPU with workers, ForkJoinTask.join
219	<	* and related methods disable them from repeatedly helping to
220	<	* process tasks if all workers are active.  Insertion of tasks in
221	<	* shared mode requires a lock (mainly to protect in the case of
210	>	* Submitter probe value serves as a hash code for
211	>	* choosing existing queues, and may be randomly repositioned upon
212	>	* contention with other submitters.  In essence, submitters act
213	>	* like workers except that they are restricted to executing local
214	>	* tasks that they submitted (or in the case of CountedCompleters,
215	>	* others with the same root task).  However, because most
216	>	* shared/external queue operations are more expensive than
217	>	* internal, and because, at steady state, external submitters
218	>	* will compete for CPU with workers, ForkJoinTask.join and
219	>	* related methods disable them from repeatedly helping to process
220	>	* tasks if all workers are active.  Insertion of tasks in shared
221	>	* mode requires a lock (mainly to protect in the case of
222		* resizing) but we use only a simple spinlock (using bits in
223		* field qlock), because submitters encountering a busy queue move
224		* on to try or create other queues -- they block only when
#	Line 240 | Line 250 | public class ForkJoinPool extends Abstra
250		* enable shutdown.  When used as a lock, it is normally only very
251		* briefly held, so is nearly always available after at most a
252		* brief spin, but we use a monitor-based backup strategy to
253	<	* blocking when needed.
253	>	* block when needed.
254		*
255		* Recording WorkQueues.  WorkQueues are recorded in the
256		* "workQueues" array that is created upon first use and expanded
#	Line 249 | Line 259 | public class ForkJoinPool extends Abstra
259		* by a lock but the array is otherwise concurrently readable, and
260		* accessed directly.  To simplify index-based operations, the
261		* array size is always a power of two, and all readers must
262	<	* tolerate null slots. Worker queues are at odd indices Shared
262	>	* tolerate null slots. Worker queues are at odd indices. Shared
263		* (submission) queues are at even indices, up to a maximum of 64
264		* slots, to limit growth even if array needs to expand to add
265		* more workers. Grouping them together in this way simplifies and
#	Line 298 | Line 308 | public class ForkJoinPool extends Abstra
308		* has not yet entered the wait queue. We solve this by requiring
309		* a full sweep of all workers (via repeated calls to method
310		* scan()) both before and after a newly waiting worker is added
311	<	* to the wait queue. During a rescan, the worker might release
312	<	* some other queued worker rather than itself, which has the same
313	<	* net effect. Because enqueued workers may actually be rescanning
314	<	* rather than waiting, we set and clear the "parker" field of
315	<	* WorkQueues to reduce unnecessary calls to unpark.  (This
316	<	* requires a secondary recheck to avoid missed signals.)  Note
317	<	* the unusual conventions about Thread.interrupts surrounding
318	<	* parking and other blocking: Because interrupts are used solely
319	<	* to alert threads to check termination, which is checked anyway
320	<	* upon blocking, we clear status (using Thread.interrupted)
321	<	* before any call to park, so that park does not immediately
312	<	* return due to status being set via some other unrelated call to
313	<	* interrupt in user code.
311	>	* to the wait queue.  Because enqueued workers may actually be
312	>	* rescanning rather than waiting, we set and clear the "parker"
313	>	* field of WorkQueues to reduce unnecessary calls to unpark.
314	>	* (This requires a secondary recheck to avoid missed signals.)
315	>	* Note the unusual conventions about Thread.interrupts
316	>	* surrounding parking and other blocking: Because interrupts are
317	>	* used solely to alert threads to check termination, which is
318	>	* checked anyway upon blocking, we clear status (using
319	>	* Thread.interrupted) before any call to park, so that park does
320	>	* not immediately return due to status being set via some other
321	>	* unrelated call to interrupt in user code.
322		*
323		* Signalling.  We create or wake up workers only when there
324		* appears to be at least one task they might be able to find and
325	<	* execute. However, many other threads may notice the same task
326	<	* and each signal to wake up a thread that might take it. So in
327	<	* general, pools will be over-signalled.  When a submission is
328	<	* added or another worker adds a task to a queue that is
329	<	* apparently empty, they signal waiting workers (or trigger
330	<	* creation of new ones if fewer than the given parallelism level
331	<	* -- see signalWork).  These primary signals are buttressed by
332	<	* signals whenever other threads scan for work or do not have a
333	<	* task to process. On most platforms, signalling (unpark)
334	<	* overhead time is noticeably long, and the time between
335	<	* signalling a thread and it actually making progress can be very
336	<	* noticeably long, so it is worth offloading these delays from
337	<	* critical paths as much as possible.
325	>	* execute.  When a submission is added or another worker adds a
326	>	* task to a queue that has fewer than two tasks, they signal
327	>	* waiting workers (or trigger creation of new ones if fewer than
328	>	* the given parallelism level -- signalWork).  These primary
329	>	* signals are buttressed by others whenever other threads remove
330	>	* a task from a queue and notice that there are other tasks there
331	>	* as well.  So in general, pools will be over-signalled. On most
332	>	* platforms, signalling (unpark) overhead time is noticeably
333	>	* long, and the time between signalling a thread and it actually
334	>	* making progress can be very noticeably long, so it is worth
335	>	* offloading these delays from critical paths as much as
336	>	* possible. Additionally, workers spin-down gradually, by staying
337	>	* alive so long as they see the ctl state changing.  Similar
338	>	* stability-sensing techniques are also used before blocking in
339	>	* awaitJoin and helpComplete.
340		*
341		* Trimming workers. To release resources after periods of lack of
342		* use, a worker starting to wait when the pool is quiescent will
#	Line 394 | Line 404 | public class ForkJoinPool extends Abstra
404		* steals, rather than use per-task bookkeeping.  This sometimes
405		* requires a linear scan of workQueues array to locate stealers,
406		* but often doesn't because stealers leave hints (that may become
407	<	* stale/wrong) of where to locate them.  A stealHint is only a
408	<	* hint because a worker might have had multiple steals and the
409	<	* hint records only one of them (usually the most current).
410	<	* Hinting isolates cost to when it is needed, rather than adding
411	<	* to per-task overhead.  (2) It is "shallow", ignoring nesting
412	<	* and potentially cyclic mutual steals.  (3) It is intentionally
407	>	* stale/wrong) of where to locate them.  It is only a hint
408	>	* because a worker might have had multiple steals and the hint
409	>	* records only one of them (usually the most current).  Hinting
410	>	* isolates cost to when it is needed, rather than adding to
411	>	* per-task overhead.  (2) It is "shallow", ignoring nesting and
412	>	* potentially cyclic mutual steals.  (3) It is intentionally
413		* racy: field currentJoin is updated only while actively joining,
414		* which means that we miss links in the chain during long-lived
415		* tasks, GC stalls etc (which is OK since blocking in such cases
#	Line 439 | Line 449 | public class ForkJoinPool extends Abstra
449		* Common Pool
450		* ===========
451		*
452	<	* The static commonPool always exists after static
452	>	* The static common pool always exists after static
453		* initialization.  Since it (or any other created pool) need
454		* never be used, we minimize initial construction overhead and
455		* footprint to the setup of about a dozen fields, with no nested
#	Line 447 | Line 457 | public class ForkJoinPool extends Abstra
457		* fullExternalPush during the first submission to the pool.
458		*
459		* When external threads submit to the common pool, they can
460	<	* perform some subtask processing (see externalHelpJoin and
461	<	* related methods).  We do not need to record whether these
460	>	* perform subtask processing (see externalHelpJoin and related
461	>	* methods).  This caller-helps policy makes it sensible to set
462	>	* common pool parallelism level to one (or more) less than the
463	>	* total number of available cores, or even zero for pure
464	>	* caller-runs.  We do not need to record whether external
465		* submissions are to the common pool -- if not, externalHelpJoin
466		* returns quickly (at the most helping to signal some common pool
467		* workers). These submitters would otherwise be blocked waiting
#	Line 517 | Line 530 | public class ForkJoinPool extends Abstra
530		* Returns a new worker thread operating in the given pool.
531		*
532		* @param pool the pool this thread works in
533	+	* @return the new worker thread
534		* @throws NullPointerException if the pool is null
535		*/
536		public ForkJoinWorkerThread newThread(ForkJoinPool pool);
#	Line 526 | Line 540 | public class ForkJoinPool extends Abstra
540		* Default ForkJoinWorkerThreadFactory implementation; creates a
541		* new ForkJoinWorkerThread.
542		*/
543	<	static class DefaultForkJoinWorkerThreadFactory
543	>	static final class DefaultForkJoinWorkerThreadFactory
544		implements ForkJoinWorkerThreadFactory {
545	<	public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
545	>	public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
546		return new ForkJoinWorkerThread(pool);
547		}
548		}
#	Line 592 | Line 606 | public class ForkJoinPool extends Abstra
606		* do not want multiple WorkQueue instances or multiple queue
607		* arrays sharing cache lines. (It would be best for queue objects
608		* and their arrays to share, but there is nothing available to
609	<	* help arrange that).  Unfortunately, because they are recorded
610	<	* in a common array, WorkQueue instances are often moved to be
597	<	* adjacent by garbage collectors. To reduce impact, we use field
598	<	* padding that works OK on common platforms; this effectively
599	<	* trades off slightly slower average field access for the sake of
600	<	* avoiding really bad worst-case access. (Until better JVM
601	<	* support is in place, this padding is dependent on transient
602	<	* properties of JVM field layout rules.)  We also take care in
603	<	* allocating, sizing and resizing the array. Non-shared queue
604	<	* arrays are initialized by workers before use. Others are
605	<	* allocated on first use.
609	>	* help arrange that). The @Contended annotation alerts JVMs to
610	>	* try to keep instances apart.
611		*/
612		static final class WorkQueue {
613		/**
#	Line 625 | Line 630 | public class ForkJoinPool extends Abstra
630		*/
631		static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
632
633	<	int seed;                  // for random scanning; initialize nonzero
633	>	// Heuristic padding to ameliorate unfortunate memory placements
634	>	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
635	>
636		volatile int eventCount;   // encoded inactivation count; < 0 if inactive
637		int nextWait;              // encoded record of next event waiter
638	<	final int mode;            // lifo, fifo, or shared
639	<	int nsteals;               // cumulative number of steals
640	<	int poolIndex;             // index of this queue in pool (or 0)
641	<	int stealHint;             // index of most recent known stealer
638	>	int nsteals;               // number of steals
639	>	int hint;                  // steal index hint
640	>	short poolIndex;           // index of this queue in pool
641	>	final short mode;          // 0: lifo, > 0: fifo, < 0: shared
642		volatile int qlock;        // 1: locked, -1: terminate; else 0
643		volatile int base;         // index of next slot for poll
644		int top;                   // index of next slot for push
#	Line 641 | Line 648 | public class ForkJoinPool extends Abstra
648		volatile Thread parker;    // == owner during call to park; else null
649		volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
650		ForkJoinTask<?> currentSteal; // current non-local task being executed
644	–	// Heuristic padding to ameliorate unfortunate memory placements
645	–	Object p00, p01, p02, p03, p04, p05, p06, p07;
646	–	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
651
652	<	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
653	<	this.mode = mode;
652	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
653	>	volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d;
654	>
655	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
656	>	int seed) {
657		this.pool = pool;
658		this.owner = owner;
659	+	this.mode = (short)mode;
660	+	this.hint = seed; // store initial seed for runWorker
661		// Place indices in the center of array (that is not yet allocated)
662		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
663		}
664
665		/**
666	<	* Pushes a task. Call only by owner in unshared queues.
667	<	* Cases needing resizing or rejection are relayed to fullPush
668	<	* (that also handles shared queues).
666	>	* Returns the approximate number of tasks in the queue.
667	>	*/
668	>	final int queueSize() {
669	>	int n = base - top;       // non-owner callers must read base first
670	>	return (n >= 0) ? 0 : -n; // ignore transient negative
671	>	}
672	>
673	>	/**
674	>	* Provides a more accurate estimate of whether this queue has
675	>	* any tasks than does queueSize, by checking whether a
676	>	* near-empty queue has at least one unclaimed task.
677	>	*/
678	>	final boolean isEmpty() {
679	>	ForkJoinTask<?>[] a; int m, s;
680	>	int n = base - (s = top);
681	>	return (n >= 0 ||
682	>	(n == -1 &&
683	>	((a = array) == null ||
684	>	(m = a.length - 1) < 0 ||
685	>	U.getObject
686	>	(a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null)));
687	>	}
688	>
689	>	/**
690	>	* Pushes a task. Call only by owner in unshared queues.  (The
691	>	* shared-queue version is embedded in method externalPush.)
692		*
693		* @param task the task. Caller must ensure non-null.
694	<	* @throw RejectedExecutionException if array cannot be resized
694	>	* @throws RejectedExecutionException if array cannot be resized
695		*/
696		final void push(ForkJoinTask<?> task) {
697	<	ForkJoinPool p; ForkJoinTask<?>[] a;
697	>	ForkJoinTask<?>[] a; ForkJoinPool p;
698		int s = top, n;
699	<	if ((a = array) != null && a.length > (n = s + 1 - base)) {
700	<	U.putOrderedObject
701	<	(a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
702	<	top = s + 1;
703	<	if (n <= 1 && (p = pool) != null)
704	<	p.signalWork(this, 1);
699	>	if ((a = array) != null) {    // ignore if queue removed
700	>	int m = a.length - 1;
701	>	U.putOrderedObject(a, ((m & s) << ASHIFT) + ABASE, task);
702	>	if ((n = (top = s + 1) - base) <= 2)
703	>	(p = pool).signalWork(p.workQueues, this);
704	>	else if (n >= m)
705	>	growArray();
706		}
674	–	else
675	–	fullPush(task, true);
707		}
708
709		/**
710	<	* Pushes a task if lock is free and array is either big
711	<	* enough or can be resized to be big enough. Note: a
712	<	* specialization of a common fast path of this method is in
713	<	* ForkJoinPool.externalPush. When called from a FJWT queue,
714	<	* this can fail only if the pool has been shut down or
715	<	* an out of memory error.
716	<	*
717	<	* @param task the task. Caller must ensure non-null.
718	<	* @param owned if true, throw RJE on failure
719	<	*/
720	<	final boolean fullPush(ForkJoinTask<?> task, boolean owned) {
721	<	ForkJoinPool p; ForkJoinTask<?>[] a;
722	<	if (owned) {
723	<	if (qlock < 0) // must be shutting down
724	<	throw new RejectedExecutionException();
710	>	* Initializes or doubles the capacity of array. Call either
711	>	* by owner or with lock held -- it is OK for base, but not
712	>	* top, to move while resizings are in progress.
713	>	*/
714	>	final ForkJoinTask<?>[] growArray() {
715	>	ForkJoinTask<?>[] oldA = array;
716	>	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
717	>	if (size > MAXIMUM_QUEUE_CAPACITY)
718	>	throw new RejectedExecutionException("Queue capacity exceeded");
719	>	int oldMask, t, b;
720	>	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
721	>	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
722	>	(t = top) - (b = base) > 0) {
723	>	int mask = size - 1;
724	>	do {
725	>	ForkJoinTask<?> x;
726	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
727	>	int j    = ((b &    mask) << ASHIFT) + ABASE;
728	>	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
729	>	if (x != null &&
730	>	U.compareAndSwapObject(oldA, oldj, x, null))
731	>	U.putObjectVolatile(a, j, x);
732	>	} while (++b != t);
733		}
734	<	else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
696	<	return false;
697	<	try {
698	<	int s = top, oldLen, len;
699	<	if ((a = array) == null)
700	<	a = array = new ForkJoinTask<?>[len=INITIAL_QUEUE_CAPACITY];
701	<	else if ((oldLen = a.length) > s + 1 - base)
702	<	len = oldLen;
703	<	else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
704	<	throw new RejectedExecutionException("Capacity exceeded");
705	<	else {
706	<	int oldMask, b;
707	<	ForkJoinTask<?>[] oldA = a;
708	<	a = array = new ForkJoinTask<?>[len];
709	<	if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
710	<	int mask = len - 1;
711	<	do {
712	<	ForkJoinTask<?> x;
713	<	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
714	<	int j    = ((b &    mask) << ASHIFT) + ABASE;
715	<	x = (ForkJoinTask<?>)
716	<	U.getObjectVolatile(oldA, oldj);
717	<	if (x != null &&
718	<	U.compareAndSwapObject(oldA, oldj, x, null))
719	<	U.putObjectVolatile(a, j, x);
720	<	} while (++b != s);
721	<	}
722	<	}
723	<	U.putOrderedObject
724	<	(a, (((len - 1) & s) << ASHIFT) + ABASE, task);
725	<	top = s + 1;
726	<	} finally {
727	<	if (!owned)
728	<	qlock = 0;
729	<	}
730	<	if ((p = pool) != null)
731	<	p.signalWork(this, 1);
732	<	return true;
734	>	return a;
735		}
736
737		/**
#	Line 762 | Line 764 | public class ForkJoinPool extends Abstra
764		if ((a = array) != null) {
765		int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
766		if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
767	<	base == b &&
768	<	U.compareAndSwapObject(a, j, t, null)) {
767	<	base = b + 1;
767	>	base == b && U.compareAndSwapObject(a, j, t, null)) {
768	>	U.putOrderedInt(this, QBASE, b + 1);
769		return t;
770		}
771		}
#	Line 780 | Line 781 | public class ForkJoinPool extends Abstra
781		int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
782		t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
783		if (t != null) {
784	<	if (base == b &&
785	<	U.compareAndSwapObject(a, j, t, null)) {
785	<	base = b + 1;
784	>	if (U.compareAndSwapObject(a, j, t, null)) {
785	>	U.putOrderedInt(this, QBASE, b + 1);
786		return t;
787		}
788		}
#	Line 839 | Line 839 | public class ForkJoinPool extends Abstra
839		ForkJoinTask.cancelIgnoringExceptions(t);
840		}
841
842	<	/**
843	<	* Computes next value for random probes.  Scans don't require
844	<	* a very high quality generator, but also not a crummy one.
845	<	* Marsaglia xor-shift is cheap and works well enough.  Note:
846	<	* This is manually inlined in its usages in ForkJoinPool to
847	<	* avoid writes inside busy scan loops.
848	<	*/
849	<	final int nextSeed() {
850	<	int r = seed;
851	<	r ^= r << 13;
852	<	r ^= r >>> 17;
853	<	return seed = r ^= r << 5;
854	<	}
842	>	// Specialized execution methods
843
844		/**
845	<	* Provides a more accurate estimate of size than (top - base)
858	<	* by ordering reads and checking whether a near-empty queue
859	<	* has at least one unclaimed task.
845	>	* Polls and runs tasks until empty.
846		*/
847	<	final int queueSize() {
848	<	ForkJoinTask<?>[] a; int k, s, n;
849	<	return ((n = base - (s = top)) < 0 &&
864	<	(n != -1 ||
865	<	((a = array) != null && (k = a.length) > 0 &&
866	<	U.getObject
867	<	(a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
868	<	-n : 0;
847	>	final void pollAndExecAll() {
848	>	for (ForkJoinTask<?> t; (t = poll()) != null;)
849	>	t.doExec();
850		}
851
871	–	// Specialized execution methods
872	–
852		/**
853	<	* Pops and runs tasks until empty.
853	>	* Executes a top-level task and any local tasks remaining
854	>	* after execution.
855		*/
856	<	private void popAndExecAll() {
857	<	// A bit faster than repeated pop calls
858	<	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
859	<	while ((a = array) != null && (m = a.length - 1) >= 0 &&
860	<	(s = top - 1) - base >= 0 &&
861	<	(t = ((ForkJoinTask<?>)
862	<	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
863	<	!= null) {
864	<	if (U.compareAndSwapObject(a, j, t, null)) {
865	<	top = s;
866	<	t.doExec();
856	>	final void runTask(ForkJoinTask<?> task) {
857	>	if ((currentSteal = task) != null) {
858	>	task.doExec();
859	>	ForkJoinTask<?>[] a = array;
860	>	int md = mode;
861	>	++nsteals;
862	>	currentSteal = null;
863	>	if (md != 0)
864	>	pollAndExecAll();
865	>	else if (a != null) {
866	>	int s, m = a.length - 1;
867	>	while ((s = top - 1) - base >= 0) {
868	>	long i = ((m & s) << ASHIFT) + ABASE;
869	>	ForkJoinTask<?> t = (ForkJoinTask<?>)U.getObject(a, i);
870	>	if (t == null)
871	>	break;
872	>	if (U.compareAndSwapObject(a, i, t, null)) {
873	>	top = s;
874	>	t.doExec();
875	>	}
876	>	}
877		}
878		}
879		}
880
881		/**
892	–	* Polls and runs tasks until empty.
893	–	*/
894	–	private void pollAndExecAll() {
895	–	for (ForkJoinTask<?> t; (t = poll()) != null;)
896	–	t.doExec();
897	–	}
898	–
899	–	/**
882		* If present, removes from queue and executes the given task,
883		* or any other cancelled task. Returns (true) on any CAS
884		* or consistency check failure so caller can retry.
885		*
886	<	* @return false if no progress can be made, else true;
886	>	* @return false if no progress can be made, else true
887		*/
888		final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
889	<	boolean stat = true, removed = false, empty = true;
889	>	boolean stat;
890		ForkJoinTask<?>[] a; int m, s, b, n;
891	<	if ((a = array) != null && (m = a.length - 1) >= 0 &&
891	>	if (task != null && (a = array) != null && (m = a.length - 1) >= 0 &&
892		(n = (s = top) - (b = base)) > 0) {
893	+	boolean removed = false, empty = true;
894	+	stat = true;
895		for (ForkJoinTask<?> t;;) {           // traverse from s to b
896	<	int j = ((--s & m) << ASHIFT) + ABASE;
897	<	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
896	>	long j = ((--s & m) << ASHIFT) + ABASE;
897	>	t = (ForkJoinTask<?>)U.getObject(a, j);
898		if (t == null)                    // inconsistent length
899		break;
900		else if (t == task) {
#	Line 938 | Line 922 | public class ForkJoinPool extends Abstra
922		break;
923		}
924		}
925	+	if (removed)
926	+	task.doExec();
927		}
928	<	if (removed)
929	<	task.doExec();
928	>	else
929	>	stat = false;
930		return stat;
931		}
932
933		/**
934	<	* Polls for and executes the given task or any other task in
935	<	* its CountedCompleter computation
934	>	* Tries to poll for and execute the given task or any other
935	>	* task in its CountedCompleter computation.
936		*/
937	<	final boolean pollAndExecCC(ForkJoinTask<?> root) {
938	<	ForkJoinTask<?>[] a; int b; Object o;
939	<	outer: while ((b = base) - top < 0 && (a = array) != null) {
937	>	final boolean pollAndExecCC(CountedCompleter<?> root) {
938	>	ForkJoinTask<?>[] a; int b; Object o; CountedCompleter<?> t, r;
939	>	if ((b = base) - top < 0 && (a = array) != null) {
940		long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
941	<	if ((o = U.getObject(a, j)) == null ||
942	<	!(o instanceof CountedCompleter))
943	<	break;
944	<	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
945	<	if (r == root) {
946	<	if (base == b &&
947	<	U.compareAndSwapObject(a, j, t, null)) {
948	<	base = b + 1;
949	<	t.doExec();
941	>	if ((o = U.getObjectVolatile(a, j)) == null)
942	>	return true; // retry
943	>	if (o instanceof CountedCompleter) {
944	>	for (t = (CountedCompleter<?>)o, r = t;;) {
945	>	if (r == root) {
946	>	if (base == b &&
947	>	U.compareAndSwapObject(a, j, t, null)) {
948	>	U.putOrderedInt(this, QBASE, b + 1);
949	>	t.doExec();
950	>	}
951		return true;
952		}
953	<	else
954	<	break; // restart
953	>	else if ((r = r.completer) == null)
954	>	break; // not part of root computation
955		}
969	–	if ((r = r.completer) == null)
970	–	break outer; // not part of root computation
956		}
957		}
958		return false;
959		}
960
961		/**
962	<	* Executes a top-level task and any local tasks remaining
963	<	* after execution.
962	>	* Tries to pop and execute the given task or any other task
963	>	* in its CountedCompleter computation.
964		*/
965	<	final void runTask(ForkJoinTask<?> t) {
966	<	if (t != null) {
967	<	(currentSteal = t).doExec();
968	<	currentSteal = null;
969	<	if (++nsteals < 0) {     // spill on overflow
970	<	ForkJoinPool p;
971	<	if ((p = pool) != null)
972	<	p.collectStealCount(this);
973	<	}
974	<	if (top != base) {       // process remaining local tasks
975	<	if (mode == 0)
976	<	popAndExecAll();
977	<	else
978	<	pollAndExecAll();
965	>	final boolean externalPopAndExecCC(CountedCompleter<?> root) {
966	>	ForkJoinTask<?>[] a; int s; Object o; CountedCompleter<?> t, r;
967	>	if (base - (s = top) < 0 && (a = array) != null) {
968	>	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
969	>	if ((o = U.getObject(a, j)) instanceof CountedCompleter) {
970	>	for (t = (CountedCompleter<?>)o, r = t;;) {
971	>	if (r == root) {
972	>	if (U.compareAndSwapInt(this, QLOCK, 0, 1)) {
973	>	if (top == s && array == a &&
974	>	U.compareAndSwapObject(a, j, t, null)) {
975	>	top = s - 1;
976	>	qlock = 0;
977	>	t.doExec();
978	>	}
979	>	else
980	>	qlock = 0;
981	>	}
982	>	return true;
983	>	}
984	>	else if ((r = r.completer) == null)
985	>	break;
986	>	}
987		}
988		}
989	+	return false;
990		}
991
992		/**
993	<	* Executes a non-top-level (stolen) task.
993	>	* Internal version
994		*/
995	<	final void runSubtask(ForkJoinTask<?> t) {
996	<	if (t != null) {
997	<	ForkJoinTask<?> ps = currentSteal;
998	<	(currentSteal = t).doExec();
999	<	currentSteal = ps;
995	>	final boolean internalPopAndExecCC(CountedCompleter<?> root) {
996	>	ForkJoinTask<?>[] a; int s; Object o; CountedCompleter<?> t, r;
997	>	if (base - (s = top) < 0 && (a = array) != null) {
998	>	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
999	>	if ((o = U.getObject(a, j)) instanceof CountedCompleter) {
1000	>	for (t = (CountedCompleter<?>)o, r = t;;) {
1001	>	if (r == root) {
1002	>	if (U.compareAndSwapObject(a, j, t, null)) {
1003	>	top = s - 1;
1004	>	t.doExec();
1005	>	}
1006	>	return true;
1007	>	}
1008	>	else if ((r = r.completer) == null)
1009	>	break;
1010	>	}
1011	>	}
1012		}
1013	+	return false;
1014		}
1015
1016		/**
#	Line 1018 | Line 1025 | public class ForkJoinPool extends Abstra
1025		s != Thread.State.TIMED_WAITING);
1026		}
1027
1021	–	/**
1022	–	* If this owned and is not already interrupted, try to
1023	–	* interrupt and/or unpark, ignoring exceptions.
1024	–	*/
1025	–	final void interruptOwner() {
1026	–	Thread wt, p;
1027	–	if ((wt = owner) != null && !wt.isInterrupted()) {
1028	–	try {
1029	–	wt.interrupt();
1030	–	} catch (SecurityException ignore) {
1031	–	}
1032	–	}
1033	–	if ((p = parker) != null)
1034	–	U.unpark(p);
1035	–	}
1036	–
1028		// Unsafe mechanics
1029		private static final sun.misc.Unsafe U;
1030	+	private static final long QBASE;
1031		private static final long QLOCK;
1032		private static final int ABASE;
1033		private static final int ASHIFT;
1034		static {
1043	–	int s;
1035		try {
1036		U = getUnsafe();
1037		Class<?> k = WorkQueue.class;
1038		Class<?> ak = ForkJoinTask[].class;
1039	+	QBASE = U.objectFieldOffset
1040	+	(k.getDeclaredField("base"));
1041		QLOCK = U.objectFieldOffset
1042		(k.getDeclaredField("qlock"));
1043		ABASE = U.arrayBaseOffset(ak);
1044	<	s = U.arrayIndexScale(ak);
1044	>	int scale = U.arrayIndexScale(ak);
1045	>	if ((scale & (scale - 1)) != 0)
1046	>	throw new Error("data type scale not a power of two");
1047	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
1048		} catch (Exception e) {
1049		throw new Error(e);
1050		}
1055	–	if ((s & (s-1)) != 0)
1056	–	throw new Error("data type scale not a power of two");
1057	–	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1051		}
1052		}
1053
1054	+	// static fields (initialized in static initializer below)
1055	+
1056		/**
1057	<	* Per-thread records for threads that submit to pools. Currently
1058	<	* holds only pseudo-random seed / index that is used to choose
1059	<	* submission queues in method externalPush. In the future, this may
1060	<	* also incorporate a means to implement different task rejection
1061	<	* and resubmission policies.
1067	<	*
1068	<	* Seeds for submitters and workers/workQueues work in basically
1069	<	* the same way but are initialized and updated using slightly
1070	<	* different mechanics. Both are initialized using the same
1071	<	* approach as in class ThreadLocal, where successive values are
1072	<	* unlikely to collide with previous values. Seeds are then
1073	<	* randomly modified upon collisions using xorshifts, which
1074	<	* requires a non-zero seed.
1057	>	* Per-thread submission bookkeeping. Shared across all pools
1058	>	* to reduce ThreadLocal pollution and because random motion
1059	>	* to avoid contention in one pool is likely to hold for others.
1060	>	* Lazily initialized on first submission (but null-checked
1061	>	* in other contexts to avoid unnecessary initialization).
1062		*/
1063	<	static final class Submitter {
1077	<	int seed;
1078	<	Submitter(int s) { seed = s; }
1079	<	}
1080	<
1081	<	/** Property prefix for constructing common pool */
1082	<	private static final String propPrefix =
1083	<	"java.util.concurrent.ForkJoinPool.common.";
1084	<
1085	<	// static fields (initialized in static initializer below)
1063	>	static final ThreadLocal<Submitter> submitters;
1064
1065		/**
1066		* Creates a new ForkJoinWorkerThread. This factory is used unless
#	Line 1092 | Line 1070 | public class ForkJoinPool extends Abstra
1070		defaultForkJoinWorkerThreadFactory;
1071
1072		/**
1095	–	* Common (static) pool. Non-null for public use unless a static
1096	–	* construction exception, but internal usages null-check on use
1097	–	* to paranoically avoid potential initialization circularities
1098	–	* as well as to simplify generated code.
1099	–	*/
1100	–	static final ForkJoinPool commonPool;
1101	–
1102	–	/**
1073		* Permission required for callers of methods that may start or
1074		* kill threads.
1075		*/
1076		private static final RuntimePermission modifyThreadPermission;
1077
1078		/**
1079	<	* Per-thread submission bookkeeping. Shared across all pools
1080	<	* to reduce ThreadLocal pollution and because random motion
1081	<	* to avoid contention in one pool is likely to hold for others.
1082	<	* Lazily initialized on first submission (but null-checked
1113	<	* in other contexts to avoid unnecessary initialization).
1079	>	* Common (static) pool. Non-null for public use unless a static
1080	>	* construction exception, but internal usages null-check on use
1081	>	* to paranoically avoid potential initialization circularities
1082	>	* as well as to simplify generated code.
1083		*/
1084	<	static final ThreadLocal<Submitter> submitters;
1084	>	static final ForkJoinPool common;
1085
1086		/**
1087	<	* Common pool parallelism. Must equal commonPool.parallelism.
1087	>	* Common pool parallelism. To allow simpler use and management
1088	>	* when common pool threads are disabled, we allow the underlying
1089	>	* common.parallelism field to be zero, but in that case still report
1090	>	* parallelism as 1 to reflect resulting caller-runs mechanics.
1091		*/
1092	<	static final int commonPoolParallelism;
1092	>	static final int commonParallelism;
1093
1094		/**
1095		* Sequence number for creating workerNamePrefix.
#	Line 1125 | Line 1097 | public class ForkJoinPool extends Abstra
1097		private static int poolNumberSequence;
1098
1099		/**
1100	<	* Return the next sequence number. We don't expect this to
1101	<	* ever contend so use simple builtin sync.
1100	>	* Returns the next sequence number. We don't expect this to
1101	>	* ever contend, so use simple builtin sync.
1102		*/
1103		private static final synchronized int nextPoolId() {
1104		return ++poolNumberSequence;
#	Line 1150 | Line 1122 | public class ForkJoinPool extends Abstra
1122		private static final long FAST_IDLE_TIMEOUT =  200L * 1000L * 1000L;
1123
1124		/**
1125	+	* Tolerance for idle timeouts, to cope with timer undershoots
1126	+	*/
1127	+	private static final long TIMEOUT_SLOP = 2000000L;
1128	+
1129	+	/**
1130		* The maximum stolen->joining link depth allowed in method
1131		* tryHelpStealer.  Must be a power of two.  Depths for legitimate
1132		* chains are unbounded, but we use a fixed constant to avoid
#	Line 1165 | Line 1142 | public class ForkJoinPool extends Abstra
1142		*/
1143		private static final int SEED_INCREMENT = 0x61c88647;
1144
1145	<	/**
1145	>	/*
1146		* Bits and masks for control variables
1147		*
1148		* Field ctl is a long packed with:
#	Line 1249 | Line 1226 | public class ForkJoinPool extends Abstra
1226		static final int FIFO_QUEUE          =  1;
1227		static final int SHARED_QUEUE        = -1;
1228
1229	<	// Instance fields
1229	>	// Heuristic padding to ameliorate unfortunate memory placements
1230	>	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
1231
1232	<	/*
1255	<	* Field layout order in this class tends to matter more than one
1256	<	* would like. Runtime layout order is only loosely related to
1257	<	* declaration order and may differ across JVMs, but the following
1258	<	* empirically works OK on current JVMs.
1259	<	*/
1232	>	// Instance fields
1233		volatile long stealCount;                  // collects worker counts
1234		volatile long ctl;                         // main pool control
1262	–	final int parallelism;                     // parallelism level
1263	–	final int localMode;                       // per-worker scheduling mode
1264	–	volatile int indexSeed;                    // worker/submitter index seed
1235		volatile int plock;                        // shutdown status and seqLock
1236	+	volatile int indexSeed;                    // worker/submitter index seed
1237	+	final short parallelism;                   // parallelism level
1238	+	final short mode;                          // LIFO/FIFO
1239		WorkQueue[] workQueues;                    // main registry
1240	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1241	<	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1240	>	final ForkJoinWorkerThreadFactory factory;
1241	>	final UncaughtExceptionHandler ueh;        // per-worker UEH
1242		final String workerNamePrefix;             // to create worker name string
1243
1244	<	/*
1244	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
1245	>	volatile Object pad18, pad19, pad1a, pad1b;
1246	>
1247	>	/**
1248		* Acquires the plock lock to protect worker array and related
1249		* updates. This method is called only if an initial CAS on plock
1250	<	* fails. This acts as a spinLock for normal cases, but falls back
1250	>	* fails. This acts as a spinlock for normal cases, but falls back
1251		* to builtin monitor to block when (rarely) needed. This would be
1252		* a terrible idea for a highly contended lock, but works fine as
1253	<	* a more conservative alternative to a pure spinlock.  See
1278	<	* internal ConcurrentHashMap documentation for further
1279	<	* explanation of nearly the same construction.
1253	>	* a more conservative alternative to a pure spinlock.
1254		*/
1255		private int acquirePlock() {
1256	<	int spins = PL_SPINS, r = 0, ps, nps;
1256	>	int spins = PL_SPINS, ps, nps;
1257		for (;;) {
1258		if (((ps = plock) & PL_LOCK) == 0 &&
1259		U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1260		return nps;
1287	–	else if (r == 0)
1288	–	r = ThreadLocalRandom.current().nextInt(); // randomize spins
1261		else if (spins >= 0) {
1262	<	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1291	<	if (r >= 0)
1262	>	if (ThreadLocalRandom.current().nextInt() >= 0)
1263		--spins;
1264		}
1265		else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
#	Line 1319 | Line 1290 | public class ForkJoinPool extends Abstra
1290		synchronized (this) { notifyAll(); }
1291		}
1292
1293	+	/**
1294	+	* Tries to create and start one worker if fewer than target
1295	+	* parallelism level exist. Adjusts counts etc on failure.
1296	+	*/
1297	+	private void tryAddWorker() {
1298	+	long c; int u, e;
1299	+	while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1300	+	(u & SHORT_SIGN) != 0 && (e = (int)c) >= 0) {
1301	+	long nc = ((long)(((u + UTC_UNIT) & UTC_MASK) |
1302	+	((u + UAC_UNIT) & UAC_MASK)) << 32) | (long)e;
1303	+	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1304	+	ForkJoinWorkerThreadFactory fac;
1305	+	Throwable ex = null;
1306	+	ForkJoinWorkerThread wt = null;
1307	+	try {
1308	+	if ((fac = factory) != null &&
1309	+	(wt = fac.newThread(this)) != null) {
1310	+	wt.start();
1311	+	break;
1312	+	}
1313	+	} catch (Throwable rex) {
1314	+	ex = rex;
1315	+	}
1316	+	deregisterWorker(wt, ex);
1317	+	break;
1318	+	}
1319	+	}
1320	+	}
1321	+
1322		//  Registering and deregistering workers
1323
1324		/**
1325	<	* Callback from ForkJoinWorkerThread constructor to establish its
1326	<	* poolIndex and record its WorkQueue. To avoid scanning bias due
1327	<	* to packing entries in front of the workQueues array, we treat
1328	<	* the array as a simple power-of-two hash table using per-thread
1329	<	* seed as hash, expanding as needed.
1330	<	*
1331	<	* @param w the worker's queue
1332	<	*/
1333	<	final void registerWorker(WorkQueue w) {
1334	<	int s, ps; // generate a rarely colliding candidate index seed
1335	<	do {} while (!U.compareAndSwapInt(this, INDEXSEED,
1336	<	s = indexSeed, s += SEED_INCREMENT) ||
1325	>	* Callback from ForkJoinWorkerThread to establish and record its
1326	>	* WorkQueue. To avoid scanning bias due to packing entries in
1327	>	* front of the workQueues array, we treat the array as a simple
1328	>	* power-of-two hash table using per-thread seed as hash,
1329	>	* expanding as needed.
1330	>	*
1331	>	* @param wt the worker thread
1332	>	* @return the worker's queue
1333	>	*/
1334	>	final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1335	>	UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1336	>	wt.setDaemon(true);
1337	>	if ((handler = ueh) != null)
1338	>	wt.setUncaughtExceptionHandler(handler);
1339	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1340	>	s += SEED_INCREMENT) ||
1341		s == 0); // skip 0
1342	+	WorkQueue w = new WorkQueue(this, wt, mode, s);
1343		if (((ps = plock) & PL_LOCK) != 0 ||
1344		!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1345		ps = acquirePlock();
1346		int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1347		try {
1348	<	WorkQueue[] ws;
1344	<	if (w != null && (ws = workQueues) != null) {
1345	<	w.seed = s;
1348	>	if ((ws = workQueues) != null) {    // skip if shutting down
1349		int n = ws.length, m = n - 1;
1350	<	int r = (s << 1) | 1;               // use odd-numbered indices
1351	<	if (ws[r &= m] != null) {           // collision
1352	<	int probes = 0;                 // step by approx half size
1350	>	int r = (s << 1) | 1;           // use odd-numbered indices
1351	>	if (ws[r &= m] != null) {       // collision
1352	>	int probes = 0;             // step by approx half size
1353		int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1354		while (ws[r = (r + step) & m] != null) {
1355		if (++probes >= n) {
#	Line 1356 | Line 1359 | public class ForkJoinPool extends Abstra
1359		}
1360		}
1361		}
1362	<	w.eventCount = w.poolIndex = r;     // establish before recording
1362	>	w.poolIndex = (short)r;
1363	>	w.eventCount = r; // volatile write orders
1364		ws[r] = w;
1365		}
1366		} finally {
1367		if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1368		releasePlock(nps);
1369		}
1370	+	wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex >>> 1)));
1371	+	return w;
1372		}
1373
1374		/**
#	Line 1371 | Line 1377 | public class ForkJoinPool extends Abstra
1377		* array, and adjusts counts. If pool is shutting down, tries to
1378		* complete termination.
1379		*
1380	<	* @param wt the worker thread or null if construction failed
1380	>	* @param wt the worker thread, or null if construction failed
1381		* @param ex the exception causing failure, or null if none
1382		*/
1383		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1384		WorkQueue w = null;
1385		if (wt != null && (w = wt.workQueue) != null) {
1386	<	int ps;
1381	<	collectStealCount(w);
1386	>	int ps; long sc;
1387		w.qlock = -1;                // ensure set
1388	+	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1389	+	sc = stealCount,
1390	+	sc + w.nsteals));
1391		if (((ps = plock) & PL_LOCK) != 0 ||
1392		!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1393		ps = acquirePlock();
#	Line 1395 | Line 1403 | public class ForkJoinPool extends Abstra
1403		}
1404		}
1405
1406	<	long c;                             // adjust ctl counts
1406	>	long c;                          // adjust ctl counts
1407		do {} while (!U.compareAndSwapLong
1408		(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1409		((c - TC_UNIT) & TC_MASK) |
1410		(c & ~(AC_MASK|TC_MASK)))));
1411
1412	<	if (!tryTerminate(false, false) && w != null) {
1413	<	w.cancelAll();                  // cancel remaining tasks
1414	<	if (w.array != null)            // suppress signal if never ran
1415	<	signalWork(null, 1);        // wake up or create replacement
1416	<	if (ex == null)                 // help clean refs on way out
1417	<	ForkJoinTask.helpExpungeStaleExceptions();
1412	>	if (!tryTerminate(false, false) && w != null && w.array != null) {
1413	>	w.cancelAll();               // cancel remaining tasks
1414	>	WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e;
1415	>	while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) {
1416	>	if (e > 0) {             // activate or create replacement
1417	>	if ((ws = workQueues) == null ||
1418	>	(i = e & SMASK) >= ws.length ||
1419	>	(v = ws[i]) == null)
1420	>	break;
1421	>	long nc = (((long)(v.nextWait & E_MASK)) |
1422	>	((long)(u + UAC_UNIT) << 32));
1423	>	if (v.eventCount != (e | INT_SIGN))
1424	>	break;
1425	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1426	>	v.eventCount = (e + E_SEQ) & E_MASK;
1427	>	if ((p = v.parker) != null)
1428	>	U.unpark(p);
1429	>	break;
1430	>	}
1431	>	}
1432	>	else {
1433	>	if ((short)u < 0)
1434	>	tryAddWorker();
1435	>	break;
1436	>	}
1437	>	}
1438		}
1439	<
1440	<	if (ex != null)                     // rethrow
1439	>	if (ex == null)                     // help clean refs on way out
1440	>	ForkJoinTask.helpExpungeStaleExceptions();
1441	>	else                                // rethrow
1442		ForkJoinTask.rethrow(ex);
1443		}
1444
1445	+	// Submissions
1446	+
1447		/**
1448	<	* Collect worker steal count into total. Called on termination
1449	<	* and upon int overflow of local count. (There is a possible race
1450	<	* in the latter case vs any caller of getStealCount, which can
1451	<	* make its results less accurate than usual.)
1452	<	*/
1453	<	final void collectStealCount(WorkQueue w) {
1454	<	if (w != null) {
1455	<	long sc;
1456	<	int ns = w.nsteals;
1457	<	w.nsteals = 0; // handle overflow
1458	<	long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
1459	<	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1460	<	sc = stealCount, sc + steals));
1461	<	}
1448	>	* Per-thread records for threads that submit to pools. Currently
1449	>	* holds only pseudo-random seed / index that is used to choose
1450	>	* submission queues in method externalPush. In the future, this may
1451	>	* also incorporate a means to implement different task rejection
1452	>	* and resubmission policies.
1453	>	*
1454	>	* Seeds for submitters and workers/workQueues work in basically
1455	>	* the same way but are initialized and updated using slightly
1456	>	* different mechanics. Both are initialized using the same
1457	>	* approach as in class ThreadLocal, where successive values are
1458	>	* unlikely to collide with previous values. Seeds are then
1459	>	* randomly modified upon collisions using xorshifts, which
1460	>	* requires a non-zero seed.
1461	>	*/
1462	>	static final class Submitter {
1463	>	int seed;
1464	>	Submitter(int s) { seed = s; }
1465		}
1466
1433	–	// Submissions
1434	–
1467		/**
1468		* Unless shutting down, adds the given task to a submission queue
1469		* at submitter's current queue index (modulo submission
#	Line 1441 | Line 1473 | public class ForkJoinPool extends Abstra
1473		* @param task the task. Caller must ensure non-null.
1474		*/
1475		final void externalPush(ForkJoinTask<?> task) {
1476	<	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1477	<	if ((z = submitters.get()) != null && plock > 0 &&
1478	<	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1479	<	(q = ws[m & z.seed & SQMASK]) != null &&
1476	>	Submitter z = submitters.get();
1477	>	WorkQueue q; int r, m, s, n, am; ForkJoinTask<?>[] a;
1478	>	int ps = plock;
1479	>	WorkQueue[] ws = workQueues;
1480	>	if (z != null && ps > 0 && ws != null && (m = (ws.length - 1)) >= 0 &&
1481	>	(q = ws[m & (r = z.seed) & SQMASK]) != null && r != 0 &&
1482		U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1483	<	int s = q.top, n;
1484	<	if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
1485	<	U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
1486	<	task);
1483	>	if ((a = q.array) != null &&
1484	>	(am = a.length - 1) > (n = (s = q.top) - q.base)) {
1485	>	int j = ((am & s) << ASHIFT) + ABASE;
1486	>	U.putOrderedObject(a, j, task);
1487		q.top = s + 1;                     // push on to deque
1488		q.qlock = 0;
1489		if (n <= 1)
1490	<	signalWork(q, 1);
1490	>	signalWork(ws, q);
1491		return;
1492		}
1493		q.qlock = 0;
#	Line 1464 | Line 1498 | public class ForkJoinPool extends Abstra
1498		/**
1499		* Full version of externalPush. This method is called, among
1500		* other times, upon the first submission of the first task to the
1501	<	* pool, so must perform secondary initialization: creating
1468	<	* workQueue array and setting plock to a valid value. It also
1501	>	* pool, so must perform secondary initialization.  It also
1502		* detects first submission by an external thread by looking up
1503		* its ThreadLocal, and creates a new shared queue if the one at
1504	<	* index if empty or contended. The lock bodies must be
1504	>	* index if empty or contended. The plock lock body must be
1505		* exception-free (so no try/finally) so we optimistically
1506	<	* allocate new queues/arrays outside the locks and throw them
1507	<	* away if (very rarely) not needed. Note that the plock seq value
1508	<	* can eventually wrap around zero, but if so harmlessly fails to
1509	<	* reinitialize.
1506	>	* allocate new queues outside the lock and throw them away if
1507	>	* (very rarely) not needed.
1508	>	*
1509	>	* Secondary initialization occurs when plock is zero, to create
1510	>	* workQueue array and set plock to a valid value.  This lock body
1511	>	* must also be exception-free. Because the plock seq value can
1512	>	* eventually wrap around zero, this method harmlessly fails to
1513	>	* reinitialize if workQueues exists, while still advancing plock.
1514		*/
1515		private void fullExternalPush(ForkJoinTask<?> task) {
1516	<	for (Submitter z = null;;) {
1517	<	WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
1516	>	int r = 0; // random index seed
1517	>	for (Submitter z = submitters.get();;) {
1518	>	WorkQueue[] ws; WorkQueue q; int ps, m, k;
1519	>	if (z == null) {
1520	>	if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1521	>	r += SEED_INCREMENT) && r != 0)
1522	>	submitters.set(z = new Submitter(r));
1523	>	}
1524	>	else if (r == 0) {                  // move to a different index
1525	>	r = z.seed;
1526	>	r ^= r << 13;                   // same xorshift as WorkQueues
1527	>	r ^= r >>> 17;
1528	>	z.seed = r ^= (r << 5);
1529	>	}
1530		if ((ps = plock) < 0)
1531		throw new RejectedExecutionException();
1532	<	else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) {
1533	<	int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2;
1534	<	n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
1535	<	WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
1536	<	if ((ps & PL_LOCK) != 0 ||
1532	>	else if (ps == 0 || (ws = workQueues) == null ||
1533	>	(m = ws.length - 1) < 0) { // initialize workQueues
1534	>	int p = parallelism;            // find power of two table size
1535	>	int n = (p > 1) ? p - 1 : 1;    // ensure at least 2 slots
1536	>	n |= n >>> 1; n |= n >>> 2;  n |= n >>> 4;
1537	>	n |= n >>> 8; n |= n >>> 16; n = (n + 1) << 1;
1538	>	WorkQueue[] nws = ((ws = workQueues) == null || ws.length == 0 ?
1539	>	new WorkQueue[n] : null);
1540	>	if (((ps = plock) & PL_LOCK) != 0 ||
1541		!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1542		ps = acquirePlock();
1543	<	if ((ws = workQueues) == null)
1543	>	if (((ws = workQueues) == null || ws.length == 0) && nws != null)
1544		workQueues = nws;
1545		int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1546		if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1547		releasePlock(nps);
1548		}
1549	<	else if (z == null && (z = submitters.get()) == null) {
1550	<	if (U.compareAndSwapInt(this, INDEXSEED,
1551	<	s = indexSeed, s += SEED_INCREMENT) &&
1552	<	s != 0) // skip 0
1553	<	submitters.set(z = new Submitter(s));
1549	>	else if ((q = ws[k = r & m & SQMASK]) != null) {
1550	>	if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
1551	>	ForkJoinTask<?>[] a = q.array;
1552	>	int s = q.top;
1553	>	boolean submitted = false;
1554	>	try {                      // locked version of push
1555	>	if ((a != null && a.length > s + 1 - q.base) ||
1556	>	(a = q.growArray()) != null) {   // must presize
1557	>	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
1558	>	U.putOrderedObject(a, j, task);
1559	>	q.top = s + 1;
1560	>	submitted = true;
1561	>	}
1562	>	} finally {
1563	>	q.qlock = 0;  // unlock
1564	>	}
1565	>	if (submitted) {
1566	>	signalWork(ws, q);
1567	>	return;
1568	>	}
1569	>	}
1570	>	r = 0; // move on failure
1571		}
1572	<	else {
1573	<	int k = (r = z.seed) & m & SQMASK;
1574	<	if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
1575	<	(q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
1576	<	if (((ps = plock) & PL_LOCK) != 0 ||
1577	<	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1578	<	ps = acquirePlock();
1579	<	WorkQueue w = null;
1580	<	if ((ws = workQueues) != null && k < ws.length &&
1581	<	(w = ws[k]) == null)
1582	<	ws[k] = q;
1513	<	else
1514	<	q = w;
1515	<	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1516	<	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1517	<	releasePlock(nps);
1518	<	}
1519	<	if (q != null && q.qlock == 0 && q.fullPush(task, false))
1520	<	return;
1521	<	r ^= r << 13;                // same xorshift as WorkQueues
1522	<	r ^= r >>> 17;
1523	<	z.seed = r ^= r << 5;        // move to a different index
1572	>	else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1573	>	q = new WorkQueue(this, null, SHARED_QUEUE, r);
1574	>	q.poolIndex = (short)k;
1575	>	if (((ps = plock) & PL_LOCK) != 0 ||
1576	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1577	>	ps = acquirePlock();
1578	>	if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1579	>	ws[k] = q;
1580	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1581	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1582	>	releasePlock(nps);
1583		}
1584	+	else
1585	+	r = 0;
1586		}
1587		}
1588
#	Line 1532 | Line 1593 | public class ForkJoinPool extends Abstra
1593		*/
1594		final void incrementActiveCount() {
1595		long c;
1596	<	do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1596	>	do {} while (!U.compareAndSwapLong
1597	>	(this, CTL, c = ctl, ((c & ~AC_MASK) |
1598	>	((c & AC_MASK) + AC_UNIT))));
1599		}
1600
1601		/**
1602	<	* Tries to create (at most one) or activate (possibly several)
1540	<	* workers if too few are active. On contention failure, continues
1541	<	* until at least one worker is signalled or the given queue is
1542	<	* empty or all workers are active.
1602	>	* Tries to create or activate a worker if too few are active.
1603		*
1604	<	* @param q if non-null, the queue holding tasks to be signalled
1605	<	* @param signals the target number of signals.
1604	>	* @param ws the worker array to use to find signallees
1605	>	* @param q if non-null, the queue holding tasks to be processed
1606		*/
1607	<	final void signalWork(WorkQueue q, int signals) {
1608	<	long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
1609	<	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1610	<	if ((e = (int)c) > 0) {
1611	<	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1612	<	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1613	<	long nc = (((long)(w.nextWait & E_MASK)) |
1614	<	((long)(u + UAC_UNIT) << 32));
1615	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1556	<	w.eventCount = (e + E_SEQ) & E_MASK;
1557	<	if ((p = w.parker) != null)
1558	<	U.unpark(p);
1559	<	if (--signals <= 0)
1560	<	break;
1561	<	}
1562	<	else
1563	<	signals = 1;
1564	<	if ((q != null && q.queueSize() == 0))
1565	<	break;
1566	<	}
1567	<	else
1568	<	break;
1607	>	final void signalWork(WorkQueue[] ws, WorkQueue q) {
1608	>	for (;;) {
1609	>	long c; int e, u, i; WorkQueue w; Thread p;
1610	>	if ((u = (int)((c = ctl) >>> 32)) >= 0)
1611	>	break;
1612	>	if ((e = (int)c) <= 0) {
1613	>	if ((short)u < 0)
1614	>	tryAddWorker();
1615	>	break;
1616		}
1617	<	else if (e == 0 && (u & SHORT_SIGN) != 0) {
1618	<	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1619	<	((u + UAC_UNIT) & UAC_MASK)) << 32;
1620	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1621	<	ForkJoinWorkerThread wt = null;
1622	<	Throwable ex = null;
1623	<	boolean started = false;
1624	<	try {
1625	<	ForkJoinWorkerThreadFactory fac;
1626	<	if ((fac = factory) != null &&
1627	<	(wt = fac.newThread(this)) != null) {
1628	<	wt.start();
1582	<	started = true;
1583	<	}
1584	<	} catch (Throwable rex) {
1585	<	ex = rex;
1586	<	}
1587	<	if (!started)
1588	<	deregisterWorker(wt, ex); // adjust counts on failure
1589	<	break;
1590	<	}
1617	>	if (ws == null || ws.length <= (i = e & SMASK) ||
1618	>	(w = ws[i]) == null)
1619	>	break;
1620	>	long nc = (((long)(w.nextWait & E_MASK)) |
1621	>	((long)(u + UAC_UNIT)) << 32);
1622	>	int ne = (e + E_SEQ) & E_MASK;
1623	>	if (w.eventCount == (e | INT_SIGN) &&
1624	>	U.compareAndSwapLong(this, CTL, c, nc)) {
1625	>	w.eventCount = ne;
1626	>	if ((p = w.parker) != null)
1627	>	U.unpark(p);
1628	>	break;
1629		}
1630	<	else
1630	>	if (q != null && q.base >= q.top)
1631		break;
1632		}
1633		}
#	Line 1600 | Line 1638 | public class ForkJoinPool extends Abstra
1638		* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1639		*/
1640		final void runWorker(WorkQueue w) {
1641	<	// initialize queue array in this thread
1642	<	w.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY];
1643	<	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1641	>	w.growArray(); // allocate queue
1642	>	for (int r = w.hint; scan(w, r) == 0; ) {
1643	>	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
1644	>	}
1645		}
1646
1647		/**
1648	<	* Scans for and, if found, returns one task, else possibly
1648	>	* Scans for and, if found, runs one task, else possibly
1649		* inactivates the worker. This method operates on single reads of
1650		* volatile state and is designed to be re-invoked continuously,
1651		* in part because it returns upon detecting inconsistencies,
1652		* contention, or state changes that indicate possible success on
1653		* re-invocation.
1654		*
1655	<	* The scan searches for tasks across a random permutation of
1656	<	* queues (starting at a random index and stepping by a random
1657	<	* relative prime, checking each at least once).  The scan
1658	<	* terminates upon either finding a non-empty queue, or completing
1659	<	* the sweep. If the worker is not inactivated, it takes and
1660	<	* returns a task from this queue. Otherwise, if not activated, it
1661	<	* signals workers (that may include itself) and returns so caller
1662	<	* can retry. Also returns for trtry if the worker array may have
1663	<	* changed during an empty scan.  On failure to find a task, we
1625	<	* take one of the following actions, after which the caller will
1626	<	* retry calling this method unless terminated.
1627	<	*
1628	<	* * If pool is terminating, terminate the worker.
1629	<	*
1630	<	* * If not already enqueued, try to inactivate and enqueue the
1631	<	* worker on wait queue. Or, if inactivating has caused the pool
1632	<	* to be quiescent, relay to idleAwaitWork to check for
1633	<	* termination and possibly shrink pool.
1634	<	*
1635	<	* * If already enqueued and none of the above apply, possibly
1636	<	* (with 1/2 probability) park awaiting signal, else lingering to
1637	<	* help scan and signal.
1655	>	* The scan searches for tasks across queues starting at a random
1656	>	* index, checking each at least twice.  The scan terminates upon
1657	>	* either finding a non-empty queue, or completing the sweep. If
1658	>	* the worker is not inactivated, it takes and runs a task from
1659	>	* this queue. Otherwise, if not activated, it tries to activate
1660	>	* itself or some other worker by signalling. On failure to find a
1661	>	* task, returns (for retry) if pool state may have changed during
1662	>	* an empty scan, or tries to inactivate if active, else possibly
1663	>	* blocks or terminates via method awaitWork.
1664		*
1665		* @param w the worker (via its WorkQueue)
1666	<	* @return a task or null if none found
1666	>	* @param r a random seed
1667	>	* @return worker qlock status if would have waited, else 0
1668		*/
1669	<	private final ForkJoinTask<?> scan(WorkQueue w) {
1670	<	WorkQueue[] ws; WorkQueue q;           // first update random seed
1671	<	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1672	<	int ps = plock, m;                     // volatile read order matters
1673	<	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1674	<	int ec = w.eventCount;             // ec is negative if inactive
1675	<	int step = (r >>> 16) | 1;         // relatively prime
1676	<	for (int j = (m + 1) << 2;  ; --j, r += step) {
1677	<	ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b, n;
1678	<	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1679	<	(a = q.array) != null) {   // probably nonempty
1680	<	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1681	<	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1682	<	if (q.base == b && ec >= 0 && t != null &&
1683	<	U.compareAndSwapObject(a, i, t, null)) {
1684	<	if ((n = q.top - (q.base = b + 1)) > 0)
1685	<	signalWork(q, n);
1686	<	return t;              // taken
1687	<	}
1688	<	if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) {
1689	<	if ((n = q.queueSize() - 1) > 0)
1690	<	signalWork(q, n);
1691	<	break;                 // let caller retry after signal
1692	<	}
1693	<	}
1694	<	else if (j < 0) {              // end of scan
1695	<	long c = ctl; int e;
1696	<	if (plock != ps)           // incomplete sweep
1670	<	break;
1671	<	if ((e = (int)c) < 0)      // pool is terminating
1672	<	w.qlock = -1;
1673	<	else if (ec >= 0) {        // try to enqueue/inactivate
1674	<	long nc = ((long)ec |
1675	<	((c - AC_UNIT) & (AC_MASK|TC_MASK)));
1669	>	private final int scan(WorkQueue w, int r) {
1670	>	WorkQueue[] ws; int m;
1671	>	long c = ctl;                            // for consistency check
1672	>	if ((ws = workQueues) != null && (m = ws.length - 1) >= 0 && w != null) {
1673	>	for (int j = m + m + 1, ec = w.eventCount;;) {
1674	>	WorkQueue q; int b, e; ForkJoinTask<?>[] a; ForkJoinTask<?> t;
1675	>	if ((q = ws[(r - j) & m]) != null &&
1676	>	(b = q.base) - q.top < 0 && (a = q.array) != null) {
1677	>	long i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1678	>	if ((t = ((ForkJoinTask<?>)
1679	>	U.getObjectVolatile(a, i))) != null) {
1680	>	if (ec < 0)
1681	>	helpRelease(c, ws, w, q, b);
1682	>	else if (q.base == b &&
1683	>	U.compareAndSwapObject(a, i, t, null)) {
1684	>	U.putOrderedInt(q, QBASE, b + 1);
1685	>	if ((b + 1) - q.top < 0)
1686	>	signalWork(ws, q);
1687	>	w.runTask(t);
1688	>	}
1689	>	}
1690	>	break;
1691	>	}
1692	>	else if (--j < 0) {
1693	>	if ((ec | (e = (int)c)) < 0) // inactive or terminating
1694	>	return awaitWork(w, c, ec);
1695	>	else if (ctl == c) {         // try to inactivate and enqueue
1696	>	long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
1697		w.nextWait = e;
1698	<	w.eventCount = ec | INT_SIGN; // mark as inactive
1699	<	if (ctl != c ||
1700	<	!U.compareAndSwapLong(this, CTL, c, nc))
1680	<	w.eventCount = ec; // unmark on CAS failure
1681	<	else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
1682	<	idleAwaitWork(w, nc, c);  // quiescent
1683	<	}
1684	<	else if (w.seed >= 0 && w.eventCount < 0) {
1685	<	Thread wt = Thread.currentThread();
1686	<	Thread.interrupted();  // clear status
1687	<	U.putObject(wt, PARKBLOCKER, this);
1688	<	w.parker = wt;         // emulate LockSupport.park
1689	<	if (w.eventCount < 0)  // recheck
1690	<	U.park(false, 0L);
1691	<	w.parker = null;
1692	<	U.putObject(wt, PARKBLOCKER, null);
1698	>	w.eventCount = ec | INT_SIGN;
1699	>	if (!U.compareAndSwapLong(this, CTL, c, nc))
1700	>	w.eventCount = ec;   // back out
1701		}
1702		break;
1703		}
1704		}
1705		}
1706	<	return null;
1706	>	return 0;
1707		}
1708
1709		/**
1710	<	* If inactivating worker w has caused the pool to become
1711	<	* quiescent, checks for pool termination, and, so long as this is
1712	<	* not the only worker, waits for event for up to a given
1713	<	* duration.  On timeout, if ctl has not changed, terminates the
1714	<	* worker, which will in turn wake up another worker to possibly
1715	<	* repeat this process.
1710	>	* A continuation of scan(), possibly blocking or terminating
1711	>	* worker w. Returns without blocking if pool state has apparently
1712	>	* changed since last invocation.  Also, if inactivating w has
1713	>	* caused the pool to become quiescent, checks for pool
1714	>	* termination, and, so long as this is not the only worker, waits
1715	>	* for event for up to a given duration.  On timeout, if ctl has
1716	>	* not changed, terminates the worker, which will in turn wake up
1717	>	* another worker to possibly repeat this process.
1718		*
1719		* @param w the calling worker
1720	<	* @param currentCtl the ctl value triggering possible quiescence
1721	<	* @param prevCtl the ctl value to restore if thread is terminated
1720	>	* @param c the ctl value on entry to scan
1721	>	* @param ec the worker's eventCount on entry to scan
1722		*/
1723	<	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1724	<	if (w.eventCount < 0 &&
1725	<	(this == commonPool || !tryTerminate(false, false)) &&
1726	<	(int)prevCtl != 0) {
1727	<	int dc = -(short)(currentCtl >>> TC_SHIFT);
1728	<	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1729	<	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1730	<	Thread wt = Thread.currentThread();
1731	<	while (ctl == currentCtl) {
1732	<	Thread.interrupted();  // timed variant of version in scan()
1733	<	U.putObject(wt, PARKBLOCKER, this);
1734	<	w.parker = wt;
1735	<	if (ctl == currentCtl)
1736	<	U.park(false, parkTime);
1737	<	w.parker = null;
1738	<	U.putObject(wt, PARKBLOCKER, null);
1739	<	if (ctl != currentCtl)
1740	<	break;
1741	<	if (deadline - System.nanoTime() <= 0L &&
1742	<	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1743	<	w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1744	<	w.qlock = -1;   // shrink
1745	<	break;
1723	>	private final int awaitWork(WorkQueue w, long c, int ec) {
1724	>	int stat, ns; long parkTime, deadline;
1725	>	if ((stat = w.qlock) >= 0 && w.eventCount == ec && ctl == c &&
1726	>	!Thread.interrupted()) {
1727	>	int e = (int)c;
1728	>	int u = (int)(c >>> 32);
1729	>	int d = (u >> UAC_SHIFT) + parallelism; // active count
1730	>
1731	>	if (e < 0 || (d <= 0 && tryTerminate(false, false)))
1732	>	stat = w.qlock = -1;          // pool is terminating
1733	>	else if ((ns = w.nsteals) != 0) { // collect steals and retry
1734	>	long sc;
1735	>	w.nsteals = 0;
1736	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1737	>	sc = stealCount, sc + ns));
1738	>	}
1739	>	else {
1740	>	long pc = ((d > 0 || ec != (e | INT_SIGN)) ? 0L :
1741	>	((long)(w.nextWait & E_MASK)) | // ctl to restore
1742	>	((long)(u + UAC_UNIT)) << 32);
1743	>	if (pc != 0L) {               // timed wait if last waiter
1744	>	int dc = -(short)(c >>> TC_SHIFT);
1745	>	parkTime = (dc < 0 ? FAST_IDLE_TIMEOUT:
1746	>	(dc + 1) * IDLE_TIMEOUT);
1747	>	deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP;
1748	>	}
1749	>	else
1750	>	parkTime = deadline = 0L;
1751	>	if (w.eventCount == ec && ctl == c) {
1752	>	Thread wt = Thread.currentThread();
1753	>	U.putObject(wt, PARKBLOCKER, this);
1754	>	w.parker = wt;            // emulate LockSupport.park
1755	>	if (w.eventCount == ec && ctl == c)
1756	>	U.park(false, parkTime);  // must recheck before park
1757	>	w.parker = null;
1758	>	U.putObject(wt, PARKBLOCKER, null);
1759	>	if (parkTime != 0L && ctl == c &&
1760	>	deadline - System.nanoTime() <= 0L &&
1761	>	U.compareAndSwapLong(this, CTL, c, pc))
1762	>	stat = w.qlock = -1;  // shrink pool
1763		}
1764		}
1765		}
1766	+	return stat;
1767		}
1768
1769		/**
1770	<	* Scans through queues looking for work while joining a task;
1771	<	* if any are present, signals.
1772	<	*
1773	<	* @param task to return early if done
1774	<	* @param origin an index to start scan
1775	<	*/
1776	<	final int helpSignal(ForkJoinTask<?> task, int origin) {
1777	<	WorkQueue[] ws; WorkQueue q; int m, n, s;
1778	<	if (task != null && (ws = workQueues) != null &&
1779	<	(m = ws.length - 1) >= 0) {
1780	<	for (int i = 0; i <= m; ++i) {
1781	<	if ((s = task.status) < 0)
1782	<	return s;
1783	<	if ((q = ws[(i + origin) & m]) != null &&
1784	<	(n = q.queueSize()) > 0) {
1785	<	signalWork(q, n);
1786	<	if ((int)(ctl >> AC_SHIFT) >= 0)
1787	<	break;
1788	<	}
1770	>	* Possibly releases (signals) a worker. Called only from scan()
1771	>	* when a worker with apparently inactive status finds a non-empty
1772	>	* queue. This requires revalidating all of the associated state
1773	>	* from caller.
1774	>	*/
1775	>	private final void helpRelease(long c, WorkQueue[] ws, WorkQueue w,
1776	>	WorkQueue q, int b) {
1777	>	WorkQueue v; int e, i; Thread p;
1778	>	if (w != null && w.eventCount < 0 && (e = (int)c) > 0 &&
1779	>	ws != null && ws.length > (i = e & SMASK) &&
1780	>	(v = ws[i]) != null && ctl == c) {
1781	>	long nc = (((long)(v.nextWait & E_MASK)) |
1782	>	((long)((int)(c >>> 32) + UAC_UNIT)) << 32);
1783	>	int ne = (e + E_SEQ) & E_MASK;
1784	>	if (q != null && q.base == b && w.eventCount < 0 &&
1785	>	v.eventCount == (e | INT_SIGN) &&
1786	>	U.compareAndSwapLong(this, CTL, c, nc)) {
1787	>	v.eventCount = ne;
1788	>	if ((p = v.parker) != null)
1789	>	U.unpark(p);
1790		}
1791		}
1763	–	return 0;
1792		}
1793
1794		/**
#	Line 1783 | Line 1811 | public class ForkJoinPool extends Abstra
1811		*/
1812		private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1813		int stat = 0, steps = 0;                    // bound to avoid cycles
1814	<	if (joiner != null && task != null) {       // hoist null checks
1814	>	if (task != null && joiner != null &&
1815	>	joiner.base - joiner.top >= 0) {        // hoist checks
1816		restart: for (;;) {
1817		ForkJoinTask<?> subtask = task;     // current target
1818		for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
#	Line 1794 | Line 1823 | public class ForkJoinPool extends Abstra
1823		}
1824		if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1825		break restart;              // shutting down
1826	<	if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
1826	>	if ((v = ws[h = (j.hint | 1) & m]) == null ||
1827		v.currentSteal != subtask) {
1828		for (int origin = h;;) {    // find stealer
1829		if (((h = (h + 2) & m) & 15) == 1 &&
#	Line 1802 | Line 1831 | public class ForkJoinPool extends Abstra
1831		continue restart;   // occasional staleness check
1832		if ((v = ws[h]) != null &&
1833		v.currentSteal == subtask) {
1834	<	j.stealHint = h;    // save hint
1834	>	j.hint = h;        // save hint
1835		break;
1836		}
1837		if (h == origin)
#	Line 1810 | Line 1839 | public class ForkJoinPool extends Abstra
1839		}
1840		}
1841		for (;;) { // help stealer or descend to its stealer
1842	<	ForkJoinTask[] a;  int b;
1842	>	ForkJoinTask[] a; int b;
1843		if (subtask.status < 0)     // surround probes with
1844		continue restart;       //   consistency checks
1845		if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
#	Line 1821 | Line 1850 | public class ForkJoinPool extends Abstra
1850		v.currentSteal != subtask)
1851		continue restart;   // stale
1852		stat = 1;               // apparent progress
1853	<	if (t != null && v.base == b &&
1854	<	U.compareAndSwapObject(a, i, t, null)) {
1855	<	v.base = b + 1;     // help stealer
1856	<	joiner.runSubtask(t);
1853	>	if (v.base == b) {
1854	>	if (t == null)
1855	>	break restart;
1856	>	if (U.compareAndSwapObject(a, i, t, null)) {
1857	>	U.putOrderedInt(v, QBASE, b + 1);
1858	>	ForkJoinTask<?> ps = joiner.currentSteal;
1859	>	int jt = joiner.top;
1860	>	do {
1861	>	joiner.currentSteal = t;
1862	>	t.doExec(); // clear local tasks too
1863	>	} while (task.status >= 0 &&
1864	>	joiner.top != jt &&
1865	>	(t = joiner.pop()) != null);
1866	>	joiner.currentSteal = ps;
1867	>	break restart;
1868	>	}
1869		}
1829	–	else if (v.base == b && ++steps == MAX_HELP)
1830	–	break restart;      // v apparently stalled
1870		}
1871		else {                      // empty -- try to descend
1872		ForkJoinTask<?> next = v.currentJoin;
#	Line 1851 | Line 1890 | public class ForkJoinPool extends Abstra
1890
1891		/**
1892		* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1893	<	* and run tasks within the target's computation
1893	>	* and run tasks within the target's computation.
1894		*
1895		* @param task the task to join
1857	–	* @param mode if shared, exit upon completing any task
1858	–	* if all workers are active
1859	–	*
1896		*/
1897	<	private int helpComplete(ForkJoinTask<?> task, int mode) {
1898	<	WorkQueue[] ws; WorkQueue q; int m, n, s;
1899	<	if (task != null && (ws = workQueues) != null &&
1900	<	(m = ws.length - 1) >= 0) {
1901	<	for (int j = 1, origin = j;;) {
1897	>	private int helpComplete(WorkQueue joiner, CountedCompleter<?> task) {
1898	>	WorkQueue[] ws; int m;
1899	>	int s = 0;
1900	>	if ((ws = workQueues) != null && (m = ws.length - 1) >= 0 &&
1901	>	joiner != null && task != null) {
1902	>	int j = joiner.poolIndex;
1903	>	int scans = m + m + 1;
1904	>	long c = 0L;              // for stability check
1905	>	for (int k = scans; ; j += 2) {
1906	>	WorkQueue q;
1907		if ((s = task.status) < 0)
1908	<	return s;
1909	<	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1910	<	origin = j;
1911	<	if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
1908	>	break;
1909	>	else if (joiner.internalPopAndExecCC(task))
1910	>	k = scans;
1911	>	else if ((s = task.status) < 0)
1912	>	break;
1913	>	else if ((q = ws[j & m]) != null && q.pollAndExecCC(task))
1914	>	k = scans;
1915	>	else if (--k < 0) {
1916	>	if (c == (c = ctl))
1917		break;
1918	+	k = scans;
1919		}
1873	–	else if ((j = (j + 2) & m) == origin)
1874	–	break;
1920		}
1921		}
1922	<	return 0;
1922	>	return s;
1923		}
1924
1925		/**
#	Line 1883 | Line 1928 | public class ForkJoinPool extends Abstra
1928		* for blocking. Fails on contention or termination. Otherwise,
1929		* adds a new thread if no idle workers are available and pool
1930		* may become starved.
1931	+	*
1932	+	* @param c the assumed ctl value
1933		*/
1934	<	final boolean tryCompensate() {
1935	<	int pc = parallelism, e, u, i, tc; long c;
1936	<	WorkQueue[] ws; WorkQueue w; Thread p;
1937	<	if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
1938	<	if (e != 0 && (i = e & SMASK) < ws.length &&
1939	<	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1934	>	final boolean tryCompensate(long c) {
1935	>	WorkQueue[] ws = workQueues;
1936	>	int pc = parallelism, e = (int)c, m, tc;
1937	>	if (ws != null && (m = ws.length - 1) >= 0 && e >= 0 && ctl == c) {
1938	>	WorkQueue w = ws[e & m];
1939	>	if (e != 0 && w != null) {
1940	>	Thread p;
1941		long nc = ((long)(w.nextWait & E_MASK) |
1942		(c & (AC_MASK|TC_MASK)));
1943	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1944	<	w.eventCount = (e + E_SEQ) & E_MASK;
1943	>	int ne = (e + E_SEQ) & E_MASK;
1944	>	if (w.eventCount == (e | INT_SIGN) &&
1945	>	U.compareAndSwapLong(this, CTL, c, nc)) {
1946	>	w.eventCount = ne;
1947		if ((p = w.parker) != null)
1948		U.unpark(p);
1949		return true;   // replace with idle worker
1950		}
1951		}
1952	<	else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
1953	<	(u >> UAC_SHIFT) + pc > 1) {
1952	>	else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1953	>	(int)(c >> AC_SHIFT) + pc > 1) {
1954		long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1955		if (U.compareAndSwapLong(this, CTL, c, nc))
1956	<	return true;    // no compensation
1956	>	return true;   // no compensation
1957		}
1958	<	else if ((tc = u + pc) < MAX_CAP) {
1958	>	else if (tc + pc < MAX_CAP) {
1959		long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1960		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1961	+	ForkJoinWorkerThreadFactory fac;
1962		Throwable ex = null;
1963		ForkJoinWorkerThread wt = null;
1964		try {
1914	–	ForkJoinWorkerThreadFactory fac;
1965		if ((fac = factory) != null &&
1966		(wt = fac.newThread(this)) != null) {
1967		wt.start();
#	Line 1920 | Line 1970 | public class ForkJoinPool extends Abstra
1970		} catch (Throwable rex) {
1971		ex = rex;
1972		}
1973	<	deregisterWorker(wt, ex); // adjust counts etc
1973	>	deregisterWorker(wt, ex); // clean up and return false
1974		}
1975		}
1976		}
#	Line 1936 | Line 1986 | public class ForkJoinPool extends Abstra
1986		*/
1987		final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1988		int s = 0;
1989	<	if (joiner != null && task != null && (s = task.status) >= 0) {
1989	>	if (task != null && (s = task.status) >= 0 && joiner != null) {
1990		ForkJoinTask<?> prevJoin = joiner.currentJoin;
1991		joiner.currentJoin = task;
1992	<	do {} while ((s = task.status) >= 0 &&
1993	<	joiner.queueSize() > 0 &&
1994	<	joiner.tryRemoveAndExec(task)); // process local tasks
1995	<	if (s >= 0 && (s = task.status) >= 0 &&
1996	<	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1947	<	(task instanceof CountedCompleter))
1948	<	s = helpComplete(task, LIFO_QUEUE);
1992	>	do {} while (joiner.tryRemoveAndExec(task) && // process local tasks
1993	>	(s = task.status) >= 0);
1994	>	if (s >= 0 && (task instanceof CountedCompleter))
1995	>	s = helpComplete(joiner, (CountedCompleter<?>)task);
1996	>	long cc = 0;        // for stability checks
1997		while (s >= 0 && (s = task.status) >= 0) {
1998	<	if ((joiner.queueSize() > 0 ||           // try helping
1999	<	(s = tryHelpStealer(joiner, task)) == 0) &&
2000	<	(s = task.status) >= 0 && tryCompensate()) {
2001	<	if (task.trySetSignal() && (s = task.status) >= 0) {
2002	<	synchronized (task) {
2003	<	if (task.status >= 0) {
2004	<	try {                // see ForkJoinTask
2005	<	task.wait();     //  for explanation
2006	<	} catch (InterruptedException ie) {
1998	>	if ((s = tryHelpStealer(joiner, task)) == 0 &&
1999	>	(s = task.status) >= 0) {
2000	>	if (!tryCompensate(cc))
2001	>	cc = ctl;
2002	>	else {
2003	>	if (task.trySetSignal() && (s = task.status) >= 0) {
2004	>	synchronized (task) {
2005	>	if (task.status >= 0) {
2006	>	try {                // see ForkJoinTask
2007	>	task.wait();     //  for explanation
2008	>	} catch (InterruptedException ie) {
2009	>	}
2010		}
2011	+	else
2012	+	task.notifyAll();
2013		}
1961	–	else
1962	–	task.notifyAll();
2014		}
2015	+	long c; // reactivate
2016	+	do {} while (!U.compareAndSwapLong
2017	+	(this, CTL, c = ctl,
2018	+	((c & ~AC_MASK) |
2019	+	((c & AC_MASK) + AC_UNIT))));
2020		}
1965	–	long c;                          // re-activate
1966	–	do {} while (!U.compareAndSwapLong
1967	–	(this, CTL, c = ctl, c + AC_UNIT));
2021		}
2022		}
2023		joiner.currentJoin = prevJoin;
#	Line 1985 | Line 2038 | public class ForkJoinPool extends Abstra
2038		if (joiner != null && task != null && (s = task.status) >= 0) {
2039		ForkJoinTask<?> prevJoin = joiner.currentJoin;
2040		joiner.currentJoin = task;
2041	<	do {} while ((s = task.status) >= 0 &&
2042	<	joiner.queueSize() > 0 &&
2043	<	joiner.tryRemoveAndExec(task));
2044	<	if (s >= 0 && (s = task.status) >= 0 &&
2045	<	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1993	<	(task instanceof CountedCompleter))
1994	<	s = helpComplete(task, LIFO_QUEUE);
1995	<	if (s >= 0 && joiner.queueSize() == 0) {
2041	>	do {} while (joiner.tryRemoveAndExec(task) && // process local tasks
2042	>	(s = task.status) >= 0);
2043	>	if (s >= 0) {
2044	>	if (task instanceof CountedCompleter)
2045	>	helpComplete(joiner, (CountedCompleter<?>)task);
2046		do {} while (task.status >= 0 &&
2047		tryHelpStealer(joiner, task) > 0);
2048		}
#	Line 2002 | Line 2052 | public class ForkJoinPool extends Abstra
2052
2053		/**
2054		* Returns a (probably) non-empty steal queue, if one is found
2055	<	* during a random, then cyclic scan, else null.  This method must
2056	<	* be retried by caller if, by the time it tries to use the queue,
2057	<	* it is empty.
2058	<	* @param r a (random) seed for scanning
2059	<	*/
2060	<	private WorkQueue findNonEmptyStealQueue(int r) {
2061	<	int step = (r >>> 16) | 1;
2062	<	for (WorkQueue[] ws;;) {
2063	<	int ps = plock, m;
2064	<	if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2065	<	return null;
2066	<	for (int j = (m + 1) << 2; ; r += step) {
2017	<	WorkQueue q = ws[((r << 1) | 1) & m];
2018	<	if (q != null && q.queueSize() > 0)
2019	<	return q;
2020	<	else if (--j < 0) {
2021	<	if (plock == ps)
2022	<	return null;
2023	<	break;
2055	>	* during a scan, else null.  This method must be retried by
2056	>	* caller if, by the time it tries to use the queue, it is empty.
2057	>	*/
2058	>	private WorkQueue findNonEmptyStealQueue() {
2059	>	int r = ThreadLocalRandom.current().nextInt();
2060	>	for (;;) {
2061	>	int ps = plock, m; WorkQueue[] ws; WorkQueue q;
2062	>	if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) {
2063	>	for (int j = (m + 1) << 2; j >= 0; --j) {
2064	>	if ((q = ws[(((r - j) << 1) | 1) & m]) != null &&
2065	>	q.base - q.top < 0)
2066	>	return q;
2067		}
2068		}
2069	+	if (plock == ps)
2070	+	return null;
2071		}
2072		}
2073
#	Line 2033 | Line 2078 | public class ForkJoinPool extends Abstra
2078		* find tasks either.
2079		*/
2080		final void helpQuiescePool(WorkQueue w) {
2081	+	ForkJoinTask<?> ps = w.currentSteal;
2082		for (boolean active = true;;) {
2083	<	ForkJoinTask<?> localTask; // exhaust local queue
2084	<	while ((localTask = w.nextLocalTask()) != null)
2085	<	localTask.doExec();
2086	<	// Similar to loop in scan(), but ignoring submissions
2041	<	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2042	<	if (q != null) {
2043	<	ForkJoinTask<?> t; int b;
2083	>	long c; WorkQueue q; ForkJoinTask<?> t; int b;
2084	>	while ((t = w.nextLocalTask()) != null)
2085	>	t.doExec();
2086	>	if ((q = findNonEmptyStealQueue()) != null) {
2087		if (!active) {      // re-establish active count
2045	–	long c;
2088		active = true;
2089		do {} while (!U.compareAndSwapLong
2090	<	(this, CTL, c = ctl, c + AC_UNIT));
2090	>	(this, CTL, c = ctl,
2091	>	((c & ~AC_MASK) |
2092	>	((c & AC_MASK) + AC_UNIT))));
2093	>	}
2094	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2095	>	(w.currentSteal = t).doExec();
2096	>	w.currentSteal = ps;
2097		}
2050	–	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2051	–	w.runSubtask(t);
2098		}
2099	<	else {
2100	<	long c;
2101	<	if (active) {       // decrement active count without queuing
2099	>	else if (active) {       // decrement active count without queuing
2100	>	long nc = ((c = ctl) & ~AC_MASK) | ((c & AC_MASK) - AC_UNIT);
2101	>	if ((int)(nc >> AC_SHIFT) + parallelism == 0)
2102	>	break;          // bypass decrement-then-increment
2103	>	if (U.compareAndSwapLong(this, CTL, c, nc))
2104		active = false;
2057	–	do {} while (!U.compareAndSwapLong
2058	–	(this, CTL, c = ctl, c -= AC_UNIT));
2059	–	}
2060	–	else
2061	–	c = ctl;        // re-increment on exit
2062	–	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2063	–	do {} while (!U.compareAndSwapLong
2064	–	(this, CTL, c = ctl, c + AC_UNIT));
2065	–	break;
2066	–	}
2105		}
2106	+	else if ((int)((c = ctl) >> AC_SHIFT) + parallelism <= 0 &&
2107	+	U.compareAndSwapLong
2108	+	(this, CTL, c, ((c & ~AC_MASK) |
2109	+	((c & AC_MASK) + AC_UNIT))))
2110	+	break;
2111		}
2112		}
2113
#	Line 2078 | Line 2121 | public class ForkJoinPool extends Abstra
2121		WorkQueue q; int b;
2122		if ((t = w.nextLocalTask()) != null)
2123		return t;
2124	<	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2124	>	if ((q = findNonEmptyStealQueue()) == null)
2125		return null;
2126		if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2127		return t;
#	Line 2110 | Line 2153 | public class ForkJoinPool extends Abstra
2153		* producing extra tasks amortizes the uncertainty of progress and
2154		* diffusion assumptions.
2155		*
2156	<	* So, users will want to use values larger, but not much larger
2156	>	* So, users will want to use values larger (but not much larger)
2157		* than 1 to both smooth over transient shortages and hedge
2158		* against uneven progress; as traded off against the cost of
2159		* extra task overhead. We leave the user to pick a threshold
#	Line 2134 | Line 2177 | public class ForkJoinPool extends Abstra
2177		static int getSurplusQueuedTaskCount() {
2178		Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2179		if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2180	<	int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
2181	<	int p = (pool = wt.pool).parallelism;
2180	>	int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).parallelism;
2181	>	int n = (q = wt.workQueue).top - q.base;
2182		int a = (int)(pool.ctl >> AC_SHIFT) + p;
2183	<	return q.top - b - (a > (p >>>= 1) ? 0 :
2184	<	a > (p >>>= 1) ? 1 :
2185	<	a > (p >>>= 1) ? 2 :
2186	<	a > (p >>>= 1) ? 4 :
2187	<	8);
2183	>	return n - (a > (p >>>= 1) ? 0 :
2184	>	a > (p >>>= 1) ? 1 :
2185	>	a > (p >>>= 1) ? 2 :
2186	>	a > (p >>>= 1) ? 4 :
2187	>	8);
2188		}
2189		return 0;
2190		}
#	Line 2163 | Line 2206 | public class ForkJoinPool extends Abstra
2206		* @return true if now terminating or terminated
2207		*/
2208		private boolean tryTerminate(boolean now, boolean enable) {
2209	<	if (this == commonPool)                     // cannot shut down
2209	>	int ps;
2210	>	if (this == common)                        // cannot shut down
2211		return false;
2212	+	if ((ps = plock) >= 0) {                   // enable by setting plock
2213	+	if (!enable)
2214	+	return false;
2215	+	if ((ps & PL_LOCK) != 0 ||
2216	+	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2217	+	ps = acquirePlock();
2218	+	int nps = ((ps + PL_LOCK) & ~SHUTDOWN) | SHUTDOWN;
2219	+	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2220	+	releasePlock(nps);
2221	+	}
2222		for (long c;;) {
2223	<	if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2224	<	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2223	>	if (((c = ctl) & STOP_BIT) != 0) {     // already terminating
2224	>	if ((short)(c >>> TC_SHIFT) + parallelism <= 0) {
2225		synchronized (this) {
2226	<	notifyAll();                // signal when 0 workers
2226	>	notifyAll();               // signal when 0 workers
2227		}
2228		}
2229		return true;
2230		}
2231	<	if (plock >= 0) {                       // not yet enabled
2232	<	int ps;
2233	<	if (!enable)
2180	<	return false;
2181	<	if (((ps = plock) & PL_LOCK) != 0 ||
2182	<	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2183	<	ps = acquirePlock();
2184	<	int nps = SHUTDOWN;
2185	<	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2186	<	releasePlock(nps);
2187	<	}
2188	<	if (!now) {                             // check if idle & no tasks
2189	<	if ((int)(c >> AC_SHIFT) != -parallelism ||
2190	<	hasQueuedSubmissions())
2231	>	if (!now) {                            // check if idle & no tasks
2232	>	WorkQueue[] ws; WorkQueue w;
2233	>	if ((int)(c >> AC_SHIFT) + parallelism > 0)
2234		return false;
2235	<	// Check for unqueued inactive workers. One pass suffices.
2236	<	WorkQueue[] ws = workQueues; WorkQueue w;
2237	<	if (ws != null) {
2238	<	for (int i = 1; i < ws.length; i += 2) {
2239	<	if ((w = ws[i]) != null && w.eventCount >= 0)
2235	>	if ((ws = workQueues) != null) {
2236	>	for (int i = 0; i < ws.length; ++i) {
2237	>	if ((w = ws[i]) != null &&
2238	>	(!w.isEmpty() ||
2239	>	((i & 1) != 0 && w.eventCount >= 0))) {
2240	>	signalWork(ws, w);
2241		return false;
2242	+	}
2243		}
2244		}
2245		}
2246		if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2247		for (int pass = 0; pass < 3; ++pass) {
2248	<	WorkQueue[] ws = workQueues;
2249	<	if (ws != null) {
2205	<	WorkQueue w;
2248	>	WorkQueue[] ws; WorkQueue w; Thread wt;
2249	>	if ((ws = workQueues) != null) {
2250		int n = ws.length;
2251		for (int i = 0; i < n; ++i) {
2252		if ((w = ws[i]) != null) {
2253		w.qlock = -1;
2254		if (pass > 0) {
2255		w.cancelAll();
2256	<	if (pass > 1)
2257	<	w.interruptOwner();
2256	>	if (pass > 1 && (wt = w.owner) != null) {
2257	>	if (!wt.isInterrupted()) {
2258	>	try {
2259	>	wt.interrupt();
2260	>	} catch (Throwable ignore) {
2261	>	}
2262	>	}
2263	>	U.unpark(wt);
2264	>	}
2265		}
2266		}
2267		}
2268		// Wake up workers parked on event queue
2269		int i, e; long cc; Thread p;
2270		while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2271	<	(i = e & SMASK) < n &&
2271	>	(i = e & SMASK) < n && i >= 0 &&
2272		(w = ws[i]) != null) {
2273		long nc = ((long)(w.nextWait & E_MASK) |
2274		((cc + AC_UNIT) & AC_MASK) |
#	Line 2243 | Line 2294 | public class ForkJoinPool extends Abstra
2294		* least one task.
2295		*/
2296		static WorkQueue commonSubmitterQueue() {
2297	<	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2297	>	Submitter z; ForkJoinPool p; WorkQueue[] ws; int m, r;
2298		return ((z = submitters.get()) != null &&
2299	<	(p = commonPool) != null &&
2299	>	(p = common) != null &&
2300		(ws = p.workQueues) != null &&
2301		(m = ws.length - 1) >= 0) ?
2302		ws[m & z.seed & SQMASK] : null;
#	Line 2254 | Line 2305 | public class ForkJoinPool extends Abstra
2305		/**
2306		* Tries to pop the given task from submitter's queue in common pool.
2307		*/
2308	<	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2309	<	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2310	<	ForkJoinTask<?>[] a;  int m, s; long j;
2311	<	if ((z = submitters.get()) != null &&
2312	<	(p = commonPool) != null &&
2313	<	(ws = p.workQueues) != null &&
2314	<	(m = ws.length - 1) >= 0 &&
2315	<	(q = ws[m & z.seed & SQMASK]) != null &&
2316	<	(s = q.top) != q.base &&
2317	<	(a = q.array) != null &&
2318	<	U.getObjectVolatile
2319	<	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2320	<	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2321	<	if (q.array == a && q.top == s && // recheck
2322	<	U.compareAndSwapObject(a, j, t, null)) {
2323	<	q.top = s - 1;
2324	<	q.qlock = 0;
2325	<	return true;
2326	<	}
2327	<	q.qlock = 0;
2328	<	}
2329	<	return false;
2330	<	}
2331	<
2332	<	/**
2333	<	* Tries to pop and run local tasks within the same computation
2334	<	* as the given root. On failure, tries to help complete from
2335	<	* other queues via helpComplete.
2336	<	*/
2337	<	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2338	<	ForkJoinTask<?>[] a; int m;
2339	<	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2340	<	root != null && root.status >= 0) {
2341	<	for (;;) {
2342	<	int s; Object o; CountedCompleter<?> task = null;
2343	<	if ((s = q.top) - q.base > 0) {
2293	<	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2294	<	if ((o = U.getObject(a, j)) != null &&
2295	<	(o instanceof CountedCompleter)) {
2296	<	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2297	<	do {
2298	<	if (r == root) {
2299	<	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2300	<	if (q.array == a && q.top == s &&
2301	<	U.compareAndSwapObject(a, j, t, null)) {
2302	<	q.top = s - 1;
2303	<	task = t;
2304	<	}
2305	<	q.qlock = 0;
2306	<	}
2307	<	break;
2308	<	}
2309	<	} while ((r = r.completer) != null);
2310	<	}
2311	<	}
2312	<	if (task != null)
2313	<	task.doExec();
2314	<	if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0)
2308	>	final boolean tryExternalUnpush(ForkJoinTask<?> task) {
2309	>	WorkQueue joiner; ForkJoinTask<?>[] a; int m, s;
2310	>	Submitter z = submitters.get();
2311	>	WorkQueue[] ws = workQueues;
2312	>	boolean popped = false;
2313	>	if (z != null && ws != null && (m = ws.length - 1) >= 0 &&
2314	>	(joiner = ws[z.seed & m & SQMASK]) != null &&
2315	>	joiner.base != (s = joiner.top) &&
2316	>	(a = joiner.array) != null) {
2317	>	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2318	>	if (U.getObject(a, j) == task &&
2319	>	U.compareAndSwapInt(joiner, QLOCK, 0, 1)) {
2320	>	if (joiner.top == s && joiner.array == a &&
2321	>	U.compareAndSwapObject(a, j, task, null)) {
2322	>	joiner.top = s - 1;
2323	>	popped = true;
2324	>	}
2325	>	joiner.qlock = 0;
2326	>	}
2327	>	}
2328	>	return popped;
2329	>	}
2330	>
2331	>	final int externalHelpComplete(CountedCompleter<?> task) {
2332	>	WorkQueue joiner; int m, j;
2333	>	Submitter z = submitters.get();
2334	>	WorkQueue[] ws = workQueues;
2335	>	int s = 0;
2336	>	if (z != null && ws != null && (m = ws.length - 1) >= 0 &&
2337	>	(joiner = ws[(j = z.seed) & m & SQMASK]) != null && task != null) {
2338	>	int scans = m + m + 1;
2339	>	long c = 0L;             // for stability check
2340	>	j |= 1;                  // poll odd queues
2341	>	for (int k = scans; ; j += 2) {
2342	>	WorkQueue q;
2343	>	if ((s = task.status) < 0)
2344		break;
2345	<	if (task == null) {
2346	<	if (helpSignal(root, q.poolIndex) >= 0)
2347	<	helpComplete(root, SHARED_QUEUE);
2345	>	else if (joiner.externalPopAndExecCC(task))
2346	>	k = scans;
2347	>	else if ((s = task.status) < 0)
2348		break;
2349	+	else if ((q = ws[j & m]) != null && q.pollAndExecCC(task))
2350	+	k = scans;
2351	+	else if (--k < 0) {
2352	+	if (c == (c = ctl))
2353	+	break;
2354	+	k = scans;
2355		}
2356		}
2357		}
2358	<	}
2324	<
2325	<	/**
2326	<	* Tries to help execute or signal availability of the given task
2327	<	* from submitter's queue in common pool.
2328	<	*/
2329	<	static void externalHelpJoin(ForkJoinTask<?> t) {
2330	<	// Some hard-to-avoid overlap with tryExternalUnpush
2331	<	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2332	<	ForkJoinTask<?>[] a;  int m, s, n; long j;
2333	<	if (t != null && t.status >= 0 &&
2334	<	(z = submitters.get()) != null &&
2335	<	(p = commonPool) != null &&
2336	<	(ws = p.workQueues) != null &&
2337	<	(m = ws.length - 1) >= 0 &&
2338	<	(q = ws[m & z.seed & SQMASK]) != null &&
2339	<	(a = q.array) != null) {
2340	<	if ((s = q.top) != q.base &&
2341	<	U.getObjectVolatile
2342	<	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2343	<	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2344	<	if (q.array == a && q.top == s &&
2345	<	U.compareAndSwapObject(a, j, t, null)) {
2346	<	q.top = s - 1;
2347	<	q.qlock = 0;
2348	<	t.doExec();
2349	<	}
2350	<	else
2351	<	q.qlock = 0;
2352	<	}
2353	<	if (t.status >= 0) {
2354	<	if (t instanceof CountedCompleter)
2355	<	p.externalHelpComplete(q, t);
2356	<	else
2357	<	p.helpSignal(t, q.poolIndex);
2358	<	}
2359	<	}
2360	<	}
2361	<
2362	<	/**
2363	<	* Restricted version of helpQuiescePool for external callers
2364	<	*/
2365	<	static void externalHelpQuiescePool() {
2366	<	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2367	<	int r = ThreadLocalRandom.current().nextInt();
2368	<	if ((p = commonPool) != null &&
2369	<	(q = p.findNonEmptyStealQueue(r)) != null &&
2370	<	(b = q.base) - q.top < 0 &&
2371	<	(t = q.pollAt(b)) != null)
2372	<	t.doExec();
2358	>	return s;
2359		}
2360
2361		// Exported methods
#	Line 2388 | Line 2374 | public class ForkJoinPool extends Abstra
2374		*         java.lang.RuntimePermission}{@code ("modifyThread")}
2375		*/
2376		public ForkJoinPool() {
2377	<	this(Runtime.getRuntime().availableProcessors(),
2377	>	this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
2378		defaultForkJoinWorkerThreadFactory, null, false);
2379		}
2380
#	Line 2436 | Line 2422 | public class ForkJoinPool extends Abstra
2422		*/
2423		public ForkJoinPool(int parallelism,
2424		ForkJoinWorkerThreadFactory factory,
2425	<	Thread.UncaughtExceptionHandler handler,
2425	>	UncaughtExceptionHandler handler,
2426		boolean asyncMode) {
2427	+	this(checkParallelism(parallelism),
2428	+	checkFactory(factory),
2429	+	handler,
2430	+	(asyncMode ? FIFO_QUEUE : LIFO_QUEUE),
2431	+	"ForkJoinPool-" + nextPoolId() + "-worker-");
2432		checkPermission();
2433	<	if (factory == null)
2434	<	throw new NullPointerException();
2433	>	}
2434	>
2435	>	private static int checkParallelism(int parallelism) {
2436		if (parallelism <= 0 || parallelism > MAX_CAP)
2437		throw new IllegalArgumentException();
2438	<	this.parallelism = parallelism;
2439	<	this.factory = factory;
2440	<	this.ueh = handler;
2441	<	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2442	<	long np = (long)(-parallelism); // offset ctl counts
2443	<	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2444	<	int pn = nextPoolId();
2445	<	StringBuilder sb = new StringBuilder("ForkJoinPool-");
2454	<	sb.append(Integer.toString(pn));
2455	<	sb.append("-worker-");
2456	<	this.workerNamePrefix = sb.toString();
2438	>	return parallelism;
2439	>	}
2440	>
2441	>	private static ForkJoinWorkerThreadFactory checkFactory
2442	>	(ForkJoinWorkerThreadFactory factory) {
2443	>	if (factory == null)
2444	>	throw new NullPointerException();
2445	>	return factory;
2446		}
2447
2448		/**
2449	<	* Constructor for common pool, suitable only for static initialization.
2450	<	* Basically the same as above, but uses smallest possible initial footprint.
2451	<	*/
2452	<	ForkJoinPool(int parallelism, long ctl,
2453	<	ForkJoinWorkerThreadFactory factory,
2454	<	Thread.UncaughtExceptionHandler handler) {
2455	<	this.parallelism = parallelism;
2456	<	this.ctl = ctl;
2449	>	* Creates a {@code ForkJoinPool} with the given parameters, without
2450	>	* any security checks or parameter validation.  Invoked directly by
2451	>	* makeCommonPool.
2452	>	*/
2453	>	private ForkJoinPool(int parallelism,
2454	>	ForkJoinWorkerThreadFactory factory,
2455	>	UncaughtExceptionHandler handler,
2456	>	int mode,
2457	>	String workerNamePrefix) {
2458	>	this.workerNamePrefix = workerNamePrefix;
2459		this.factory = factory;
2460		this.ueh = handler;
2461	<	this.localMode = LIFO_QUEUE;
2462	<	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2461	>	this.mode = (short)mode;
2462	>	this.parallelism = (short)parallelism;
2463	>	long np = (long)(-parallelism); // offset ctl counts
2464	>	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2465		}
2466
2467		/**
2468	<	* Returns the common pool instance.
2468	>	* Returns the common pool instance. This pool is statically
2469	>	* constructed; its run state is unaffected by attempts to {@link
2470	>	* #shutdown} or {@link #shutdownNow}. However this pool and any
2471	>	* ongoing processing are automatically terminated upon program
2472	>	* {@link System#exit}.  Any program that relies on asynchronous
2473	>	* task processing to complete before program termination should
2474	>	* invoke {@code commonPool().}{@link #awaitQuiescence awaitQuiescence},
2475	>	* before exit.
2476		*
2477		* @return the common pool instance
2478	+	* @since 1.8
2479		*/
2480		public static ForkJoinPool commonPool() {
2481	<	return commonPool; // cannot be null (if so, a static init error)
2481	>	// assert common != null : "static init error";
2482	>	return common;
2483		}
2484
2485		// Execution methods
#	Line 2533 | Line 2535 | public class ForkJoinPool extends Abstra
2535		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2536		job = (ForkJoinTask<?>) task;
2537		else
2538	<	job = new ForkJoinTask.AdaptedRunnableAction(task);
2538	>	job = new ForkJoinTask.RunnableExecuteAction(task);
2539		externalPush(job);
2540		}
2541
#	Line 2600 | Line 2602 | public class ForkJoinPool extends Abstra
2602		// In previous versions of this class, this method constructed
2603		// a task to run ForkJoinTask.invokeAll, but now external
2604		// invocation of multiple tasks is at least as efficient.
2605	<	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2604	<	// Workaround needed because method wasn't declared with
2605	<	// wildcards in return type but should have been.
2606	<	@SuppressWarnings({"unchecked", "rawtypes"})
2607	<	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2605	>	ArrayList<Future<T>> futures = new ArrayList<Future<T>>(tasks.size());
2606
2607		boolean done = false;
2608		try {
2609		for (Callable<T> t : tasks) {
2610		ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2611	+	futures.add(f);
2612		externalPush(f);
2614	–	fs.add(f);
2613		}
2614	<	for (ForkJoinTask<T> f : fs)
2615	<	f.quietlyJoin();
2614	>	for (int i = 0, size = futures.size(); i < size; i++)
2615	>	((ForkJoinTask<?>)futures.get(i)).quietlyJoin();
2616		done = true;
2617		return futures;
2618		} finally {
2619		if (!done)
2620	<	for (ForkJoinTask<T> f : fs)
2621	<	f.cancel(false);
2620	>	for (int i = 0, size = futures.size(); i < size; i++)
2621	>	futures.get(i).cancel(false);
2622		}
2623		}
2624
#	Line 2639 | Line 2637 | public class ForkJoinPool extends Abstra
2637		*
2638		* @return the handler, or {@code null} if none
2639		*/
2640	<	public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
2640	>	public UncaughtExceptionHandler getUncaughtExceptionHandler() {
2641		return ueh;
2642		}
2643
#	Line 2649 | Line 2647 | public class ForkJoinPool extends Abstra
2647		* @return the targeted parallelism level of this pool
2648		*/
2649		public int getParallelism() {
2650	<	return parallelism;
2650	>	int par;
2651	>	return ((par = parallelism) > 0) ? par : 1;
2652		}
2653
2654		/**
2655		* Returns the targeted parallelism level of the common pool.
2656		*
2657		* @return the targeted parallelism level of the common pool
2658	+	* @since 1.8
2659		*/
2660		public static int getCommonPoolParallelism() {
2661	<	return commonPoolParallelism;
2661	>	return commonParallelism;
2662		}
2663
2664		/**
#	Line 2680 | Line 2680 | public class ForkJoinPool extends Abstra
2680		* @return {@code true} if this pool uses async mode
2681		*/
2682		public boolean getAsyncMode() {
2683	<	return localMode != 0;
2683	>	return mode == FIFO_QUEUE;
2684		}
2685
2686		/**
#	Line 2727 | Line 2727 | public class ForkJoinPool extends Abstra
2727		* @return {@code true} if all threads are currently idle
2728		*/
2729		public boolean isQuiescent() {
2730	<	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2730	>	return parallelism + (int)(ctl >> AC_SHIFT) <= 0;
2731		}
2732
2733		/**
#	Line 2804 | Line 2804 | public class ForkJoinPool extends Abstra
2804		WorkQueue[] ws; WorkQueue w;
2805		if ((ws = workQueues) != null) {
2806		for (int i = 0; i < ws.length; i += 2) {
2807	<	if ((w = ws[i]) != null && w.queueSize() != 0)
2807	>	if ((w = ws[i]) != null && !w.isEmpty())
2808		return true;
2809		}
2810		}
#	Line 2916 | Line 2916 | public class ForkJoinPool extends Abstra
2916		* Possibly initiates an orderly shutdown in which previously
2917		* submitted tasks are executed, but no new tasks will be
2918		* accepted. Invocation has no effect on execution state if this
2919	<	* is the {@link #commonPool}, and no additional effect if
2919	>	* is the {@link #commonPool()}, and no additional effect if
2920		* already shut down.  Tasks that are in the process of being
2921		* submitted concurrently during the course of this method may or
2922		* may not be rejected.
#	Line 2934 | Line 2934 | public class ForkJoinPool extends Abstra
2934		/**
2935		* Possibly attempts to cancel and/or stop all tasks, and reject
2936		* all subsequently submitted tasks.  Invocation has no effect on
2937	<	* execution state if this is the {@link #commonPool}, and no
2937	>	* execution state if this is the {@link #commonPool()}, and no
2938		* additional effect if already shut down. Otherwise, tasks that
2939		* are in the process of being submitted or executed concurrently
2940		* during the course of this method may or may not be
#	Line 2963 | Line 2963 | public class ForkJoinPool extends Abstra
2963		public boolean isTerminated() {
2964		long c = ctl;
2965		return ((c & STOP_BIT) != 0L &&
2966	<	(short)(c >>> TC_SHIFT) == -parallelism);
2966	>	(short)(c >>> TC_SHIFT) + parallelism <= 0);
2967		}
2968
2969		/**
#	Line 2971 | Line 2971 | public class ForkJoinPool extends Abstra
2971		* commenced but not yet completed.  This method may be useful for
2972		* debugging. A return of {@code true} reported a sufficient
2973		* period after shutdown may indicate that submitted tasks have
2974	<	* ignored or suppressed interruption, or are waiting for IO,
2974	>	* ignored or suppressed interruption, or are waiting for I/O,
2975		* causing this executor not to properly terminate. (See the
2976		* advisory notes for class {@link ForkJoinTask} stating that
2977		* tasks should not normally entail blocking operations.  But if
#	Line 2982 | Line 2982 | public class ForkJoinPool extends Abstra
2982		public boolean isTerminating() {
2983		long c = ctl;
2984		return ((c & STOP_BIT) != 0L &&
2985	<	(short)(c >>> TC_SHIFT) != -parallelism);
2985	>	(short)(c >>> TC_SHIFT) + parallelism > 0);
2986		}
2987
2988		/**
#	Line 2997 | Line 2997 | public class ForkJoinPool extends Abstra
2997		/**
2998		* Blocks until all tasks have completed execution after a
2999		* shutdown request, or the timeout occurs, or the current thread
3000	<	* is interrupted, whichever happens first. Note that the {@link
3001	<	* #commonPool()} never terminates until program shutdown so
3002	<	* this method will always time out.
3000	>	* is interrupted, whichever happens first. Because the {@link
3001	>	* #commonPool()} never terminates until program shutdown, when
3002	>	* applied to the common pool, this method is equivalent to {@link
3003	>	* #awaitQuiescence(long, TimeUnit)} but always returns {@code false}.
3004		*
3005		* @param timeout the maximum time to wait
3006		* @param unit the time unit of the timeout argument
#	Line 3009 | Line 3010 | public class ForkJoinPool extends Abstra
3010		*/
3011		public boolean awaitTermination(long timeout, TimeUnit unit)
3012		throws InterruptedException {
3013	+	if (Thread.interrupted())
3014	+	throw new InterruptedException();
3015	+	if (this == common) {
3016	+	awaitQuiescence(timeout, unit);
3017	+	return false;
3018	+	}
3019		long nanos = unit.toNanos(timeout);
3020		if (isTerminated())
3021		return true;
3022	<	long startTime = System.nanoTime();
3023	<	boolean terminated = false;
3022	>	if (nanos <= 0L)
3023	>	return false;
3024	>	long deadline = System.nanoTime() + nanos;
3025		synchronized (this) {
3026	<	for (long waitTime = nanos, millis = 0L;;) {
3027	<	if (terminated = isTerminated() ||
3028	<	waitTime <= 0L ||
3029	<	(millis = unit.toMillis(waitTime)) <= 0L)
3026	>	for (;;) {
3027	>	if (isTerminated())
3028	>	return true;
3029	>	if (nanos <= 0L)
3030	>	return false;
3031	>	long millis = TimeUnit.NANOSECONDS.toMillis(nanos);
3032	>	wait(millis > 0L ? millis : 1L);
3033	>	nanos = deadline - System.nanoTime();
3034	>	}
3035	>	}
3036	>	}
3037	>
3038	>	/**
3039	>	* If called by a ForkJoinTask operating in this pool, equivalent
3040	>	* in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise,
3041	>	* waits and/or attempts to assist performing tasks until this
3042	>	* pool {@link #isQuiescent} or the indicated timeout elapses.
3043	>	*
3044	>	* @param timeout the maximum time to wait
3045	>	* @param unit the time unit of the timeout argument
3046	>	* @return {@code true} if quiescent; {@code false} if the
3047	>	* timeout elapsed.
3048	>	*/
3049	>	public boolean awaitQuiescence(long timeout, TimeUnit unit) {
3050	>	long nanos = unit.toNanos(timeout);
3051	>	ForkJoinWorkerThread wt;
3052	>	Thread thread = Thread.currentThread();
3053	>	if ((thread instanceof ForkJoinWorkerThread) &&
3054	>	(wt = (ForkJoinWorkerThread)thread).pool == this) {
3055	>	helpQuiescePool(wt.workQueue);
3056	>	return true;
3057	>	}
3058	>	long startTime = System.nanoTime();
3059	>	WorkQueue[] ws;
3060	>	int r = 0, m;
3061	>	boolean found = true;
3062	>	while (!isQuiescent() && (ws = workQueues) != null &&
3063	>	(m = ws.length - 1) >= 0) {
3064	>	if (!found) {
3065	>	if ((System.nanoTime() - startTime) > nanos)
3066	>	return false;
3067	>	Thread.yield(); // cannot block
3068	>	}
3069	>	found = false;
3070	>	for (int j = (m + 1) << 2; j >= 0; --j) {
3071	>	ForkJoinTask<?> t; WorkQueue q; int b;
3072	>	if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) {
3073	>	found = true;
3074	>	if ((t = q.pollAt(b)) != null)
3075	>	t.doExec();
3076		break;
3077	<	wait(millis);
3024	<	waitTime = nanos - (System.nanoTime() - startTime);
3077	>	}
3078		}
3079		}
3080	<	return terminated;
3080	>	return true;
3081	>	}
3082	>
3083	>	/**
3084	>	* Waits and/or attempts to assist performing tasks indefinitely
3085	>	* until the {@link #commonPool()} {@link #isQuiescent}.
3086	>	*/
3087	>	static void quiesceCommonPool() {
3088	>	common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
3089		}
3090
3091		/**
#	Line 3036 | Line 3097 | public class ForkJoinPool extends Abstra
3097		* not necessary. Method {@code block} blocks the current thread
3098		* if necessary (perhaps internally invoking {@code isReleasable}
3099		* before actually blocking). These actions are performed by any
3100	<	* thread invoking {@link ForkJoinPool#managedBlock}.  The
3101	<	* unusual methods in this API accommodate synchronizers that may,
3102	<	* but don't usually, block for long periods. Similarly, they
3100	>	* thread invoking {@link ForkJoinPool#managedBlock(ManagedBlocker)}.
3101	>	* The unusual methods in this API accommodate synchronizers that
3102	>	* may, but don't usually, block for long periods. Similarly, they
3103		* allow more efficient internal handling of cases in which
3104		* additional workers may be, but usually are not, needed to
3105		* ensure sufficient parallelism.  Toward this end,
#	Line 3096 | Line 3157 | public class ForkJoinPool extends Abstra
3157
3158		/**
3159		* Returns {@code true} if blocking is unnecessary.
3160	+	* @return {@code true} if blocking is unnecessary
3161		*/
3162		boolean isReleasable();
3163		}
#	Line 3125 | Line 3187 | public class ForkJoinPool extends Abstra
3187		Thread t = Thread.currentThread();
3188		if (t instanceof ForkJoinWorkerThread) {
3189		ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3190	<	while (!blocker.isReleasable()) { // variant of helpSignal
3191	<	WorkQueue[] ws; WorkQueue q; int m, n;
3130	<	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3131	<	for (int i = 0; i <= m; ++i) {
3132	<	if (blocker.isReleasable())
3133	<	return;
3134	<	if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
3135	<	p.signalWork(q, n);
3136	<	if ((int)(p.ctl >> AC_SHIFT) >= 0)
3137	<	break;
3138	<	}
3139	<	}
3140	<	}
3141	<	if (p.tryCompensate()) {
3190	>	while (!blocker.isReleasable()) {
3191	>	if (p.tryCompensate(p.ctl)) {
3192		try {
3193		do {} while (!blocker.isReleasable() &&
3194		!blocker.block());
#	Line 3176 | Line 3226 | public class ForkJoinPool extends Abstra
3226		private static final long STEALCOUNT;
3227		private static final long PLOCK;
3228		private static final long INDEXSEED;
3229	+	private static final long QBASE;
3230		private static final long QLOCK;
3231
3232		static {
3233	<	// Establish common pool parameters
3183	<	// TBD: limit or report ignored exceptions?
3184	<
3185	<	int par = 0;
3186	<	ForkJoinWorkerThreadFactory fac = null;
3187	<	Thread.UncaughtExceptionHandler handler = null;
3188	<	try {
3189	<	String pp = System.getProperty(propPrefix + "parallelism");
3190	<	String hp = System.getProperty(propPrefix + "exceptionHandler");
3191	<	String fp = System.getProperty(propPrefix + "threadFactory");
3192	<	if (fp != null)
3193	<	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3194	<	getSystemClassLoader().loadClass(fp).newInstance());
3195	<	if (hp != null)
3196	<	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3197	<	getSystemClassLoader().loadClass(hp).newInstance());
3198	<	if (pp != null)
3199	<	par = Integer.parseInt(pp);
3200	<	} catch (Exception ignore) {
3201	<	}
3202	<
3203	<	int s; // initialize field offsets for CAS etc
3233	>	// initialize field offsets for CAS etc
3234		try {
3235		U = getUnsafe();
3236		Class<?> k = ForkJoinPool.class;
#	Line 3216 | Line 3246 | public class ForkJoinPool extends Abstra
3246		PARKBLOCKER = U.objectFieldOffset
3247		(tk.getDeclaredField("parkBlocker"));
3248		Class<?> wk = WorkQueue.class;
3249	+	QBASE = U.objectFieldOffset
3250	+	(wk.getDeclaredField("base"));
3251		QLOCK = U.objectFieldOffset
3252		(wk.getDeclaredField("qlock"));
3253		Class<?> ak = ForkJoinTask[].class;
3254		ABASE = U.arrayBaseOffset(ak);
3255	<	s = U.arrayIndexScale(ak);
3256	<	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3255	>	int scale = U.arrayIndexScale(ak);
3256	>	if ((scale & (scale - 1)) != 0)
3257	>	throw new Error("data type scale not a power of two");
3258	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
3259		} catch (Exception e) {
3260		throw new Error(e);
3261		}
3228	–	if ((s & (s-1)) != 0)
3229	–	throw new Error("data type scale not a power of two");
3262
3263	<	/*
3264	<	* For extra caution, computations to set up pool state are
3233	<	* here; the constructor just assigns these values to fields.
3234	<	*/
3235	<	ForkJoinWorkerThreadFactory defaultFac =
3236	<	defaultForkJoinWorkerThreadFactory =
3263	>	submitters = new ThreadLocal<Submitter>();
3264	>	defaultForkJoinWorkerThreadFactory =
3265		new DefaultForkJoinWorkerThreadFactory();
3238	–	if (fac == null)
3239	–	fac = defaultFac;
3240	–	if (par <= 0)
3241	–	par = Runtime.getRuntime().availableProcessors();
3242	–	if (par > MAX_CAP)
3243	–	par = MAX_CAP;
3244	–	long np = (long)(-par); // precompute initial ctl value
3245	–	long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
3246	–
3247	–	commonPoolParallelism = par;
3248	–	commonPool = new ForkJoinPool(par, ct, fac, handler);
3266		modifyThreadPermission = new RuntimePermission("modifyThread");
3267	<	submitters = new ThreadLocal<Submitter>();
3267	>
3268	>	common = java.security.AccessController.doPrivileged
3269	>	(new java.security.PrivilegedAction<ForkJoinPool>() {
3270	>	public ForkJoinPool run() { return makeCommonPool(); }});
3271	>	int par = common.parallelism; // report 1 even if threads disabled
3272	>	commonParallelism = par > 0 ? par : 1;
3273	>	}
3274	>
3275	>	/**
3276	>	* Creates and returns the common pool, respecting user settings
3277	>	* specified via system properties.
3278	>	*/
3279	>	private static ForkJoinPool makeCommonPool() {
3280	>	int parallelism = -1;
3281	>	ForkJoinWorkerThreadFactory factory
3282	>	= defaultForkJoinWorkerThreadFactory;
3283	>	UncaughtExceptionHandler handler = null;
3284	>	try {  // ignore exceptions in accessing/parsing properties
3285	>	String pp = System.getProperty
3286	>	("java.util.concurrent.ForkJoinPool.common.parallelism");
3287	>	String fp = System.getProperty
3288	>	("java.util.concurrent.ForkJoinPool.common.threadFactory");
3289	>	String hp = System.getProperty
3290	>	("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3291	>	if (pp != null)
3292	>	parallelism = Integer.parseInt(pp);
3293	>	if (fp != null)
3294	>	factory = ((ForkJoinWorkerThreadFactory)ClassLoader.
3295	>	getSystemClassLoader().loadClass(fp).newInstance());
3296	>	if (hp != null)
3297	>	handler = ((UncaughtExceptionHandler)ClassLoader.
3298	>	getSystemClassLoader().loadClass(hp).newInstance());
3299	>	} catch (Exception ignore) {
3300	>	}
3301	>
3302	>	if (parallelism < 0 && // default 1 less than #cores
3303	>	(parallelism = Runtime.getRuntime().availableProcessors() - 1) < 0)
3304	>	parallelism = 0;
3305	>	if (parallelism > MAX_CAP)
3306	>	parallelism = MAX_CAP;
3307	>	return new ForkJoinPool(parallelism, factory, handler, LIFO_QUEUE,
3308	>	"ForkJoinPool.commonPool-worker-");
3309		}
3310
3311		/**
#	Line 3260 | Line 3318 | public class ForkJoinPool extends Abstra
3318		private static sun.misc.Unsafe getUnsafe() {
3319		try {
3320		return sun.misc.Unsafe.getUnsafe();
3321	<	} catch (SecurityException se) {
3322	<	try {
3323	<	return java.security.AccessController.doPrivileged
3324	<	(new java.security
3325	<	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3326	<	public sun.misc.Unsafe run() throws Exception {
3327	<	java.lang.reflect.Field f = sun.misc
3328	<	.Unsafe.class.getDeclaredField("theUnsafe");
3329	<	f.setAccessible(true);
3330	<	return (sun.misc.Unsafe) f.get(null);
3331	<	}});
3332	<	} catch (java.security.PrivilegedActionException e) {
3333	<	throw new RuntimeException("Could not initialize intrinsics",
3334	<	e.getCause());
3335	<	}
3321	>	} catch (SecurityException tryReflectionInstead) {}
3322	>	try {
3323	>	return java.security.AccessController.doPrivileged
3324	>	(new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3325	>	public sun.misc.Unsafe run() throws Exception {
3326	>	Class<sun.misc.Unsafe> k = sun.misc.Unsafe.class;
3327	>	for (java.lang.reflect.Field f : k.getDeclaredFields()) {
3328	>	f.setAccessible(true);
3329	>	Object x = f.get(null);
3330	>	if (k.isInstance(x))
3331	>	return k.cast(x);
3332	>	}
3333	>	throw new NoSuchFieldError("the Unsafe");
3334	>	}});
3335	>	} catch (java.security.PrivilegedActionException e) {
3336	>	throw new RuntimeException("Could not initialize intrinsics",
3337	>	e.getCause());
3338		}
3339		}
3280	–
3340		}

Diff Legend

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