[ViewVC] Diff of: jsr166/jsr166/src/jsr166e/ForkJoinPool.java

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.4 by jsr166, Fri Oct 12 16:46:37 2012 UTC vs.
Revision 1.22 by jsr166, Wed Nov 21 22:15:41 2012 UTC

#	Line 5 \| Line 5
5		*/
6
7		package jsr166e;
8	+
9		import java.util.ArrayList;
10		import java.util.Arrays;
11		import java.util.Collection;
12		import java.util.Collections;
13		import java.util.List;
13	–	import java.util.Random;
14		import java.util.concurrent.AbstractExecutorService;
15		import java.util.concurrent.Callable;
16		import java.util.concurrent.ExecutorService;
#	Line 18 \| Line 18 \| import java.util.concurrent.Future;
18		import java.util.concurrent.RejectedExecutionException;
19		import java.util.concurrent.RunnableFuture;
20		import java.util.concurrent.TimeUnit;
21	–	import java.util.concurrent.atomic.AtomicInteger;
22	–	import java.util.concurrent.atomic.AtomicLong;
23	–	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
24	–	import java.util.concurrent.locks.Condition;
21
22		/**
23		* An {@link ExecutorService} for running {@link ForkJoinTask}s.
#	Line 41 \| Line 37 \| import java.util.concurrent.locks.Condit
37		* ForkJoinPool}s may also be appropriate for use with event-style
38		* tasks that are never joined.
39		*
40	<	* <p>A {@code ForkJoinPool} is constructed with a given target
41	<	* parallelism level; by default, equal to the number of available
42	<	* processors. The pool attempts to maintain enough active (or
43	<	* available) threads by dynamically adding, suspending, or resuming
44	<	* internal worker threads, even if some tasks are stalled waiting to
45	<	* join others. However, no such adjustments are guaranteed in the
46	<	* face of blocked IO or other unmanaged synchronization. The nested
47	<	* {@link ManagedBlocker} interface enables extension of the kinds of
40	>	* <p>A static {@link #commonPool()} is available and appropriate for
41	>	* most applications. The common pool is used by any ForkJoinTask that
42	>	* is not explicitly submitted to a specified pool. Using the common
43	>	* pool normally reduces resource usage (its threads are slowly
44	>	* reclaimed during periods of non-use, and reinstated upon subsequent
45	>	* use).
46	>	*
47	>	* <p>For applications that require separate or custom pools, a {@code
48	>	* ForkJoinPool} may be constructed with a given target parallelism
49	>	* level; by default, equal to the number of available processors. The
50	>	* pool attempts to maintain enough active (or available) threads by
51	>	* dynamically adding, suspending, or resuming internal worker
52	>	* threads, even if some tasks are stalled waiting to join
53	>	* others. However, no such adjustments are guaranteed in the face of
54	>	* blocked IO or other unmanaged synchronization. The nested {@link
55	>	* ManagedBlocker} interface enables extension of the kinds of
56		* synchronization accommodated.
57		*
58		* <p>In addition to execution and lifecycle control methods, this
#	Line 58 \| Line 62 \| import java.util.concurrent.locks.Condit
62		* {@link #toString} returns indications of pool state in a
63		* convenient form for informal monitoring.
64		*
65	<	* <p> As is the case with other ExecutorServices, there are three
65	>	* <p>As is the case with other ExecutorServices, there are three
66		* main task execution methods summarized in the following table.
67		* These are designed to be used primarily by clients not already
68		* engaged in fork/join computations in the current pool. The main
#	Line 93 \| Line 97 \| import java.util.concurrent.locks.Condit
97		* </tr>
98		* </table>
99		*
100	<	* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
101	<	* used for all parallel task execution in a program or subsystem.
102	<	* Otherwise, use would not usually outweigh the construction and
103	<	* bookkeeping overhead of creating a large set of threads. For
104	<	* example, a common pool could be used for the {@code SortTasks}
105	<	* illustrated in {@link RecursiveAction}. Because {@code
106	<	* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
107	<	* daemon} mode, there is typically no need to explicitly {@link
108	<	* #shutdown} such a pool upon program exit.
109	<	*
106	<	* <pre> {@code
107	<	* static final ForkJoinPool mainPool = new ForkJoinPool();
108	<	* ...
109	<	* public void sort(long[] array) {
110	<	* mainPool.invoke(new SortTask(array, 0, array.length));
111	<	* }}</pre>
100	>	* <p>The common pool is by default constructed with default
101	>	* parameters, but these may be controlled by setting three {@link
102	>	* System#getProperty properties} with prefix {@code
103	>	* java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
104	>	* an integer greater than zero, {@code threadFactory} -- the class
105	>	* name of a {@link ForkJoinWorkerThreadFactory}, and {@code
106	>	* exceptionHandler} -- the class name of a {@link
107	>	* java.lang.Thread.UncaughtExceptionHandler
108	>	* Thread.UncaughtExceptionHandler}. Upon any error in establishing
109	>	* these settings, default parameters are used.
110		*
111		* <p><b>Implementation notes</b>: This implementation restricts the
112		* maximum number of running threads to 32767. Attempts to create
#	Line 196 \| Line 194 \| public class ForkJoinPool extends Abstra
194		* WorkQueues are also used in a similar way for tasks submitted
195		* to the pool. We cannot mix these tasks in the same queues used
196		* for work-stealing (this would contaminate lifo/fifo
197	<	* processing). Instead, we loosely associate submission queues
197	>	* processing). Instead, we randomly associate submission queues
198		* with submitting threads, using a form of hashing. The
199		* ThreadLocal Submitter class contains a value initially used as
200		* a hash code for choosing existing queues, but may be randomly
201		* repositioned upon contention with other submitters. In
202	<	* essence, submitters act like workers except that they never
203	<	* take tasks, and they are multiplexed on to a finite number of
204	<	* shared work queues. However, classes are set up so that future
205	<	* extensions could allow submitters to optionally help perform
206	<	* tasks as well. Insertion of tasks in shared mode requires a
207	<	* lock (mainly to protect in the case of resizing) but we use
208	<	* only a simple spinlock (using bits in field runState), because
209	<	* submitters encountering a busy queue move on to try or create
210	<	* other queues -- they block only when creating and registering
211	<	* new queues.
202	>	* essence, submitters act like workers except that they are
203	>	* restricted to executing local tasks that they submitted (or in
204	>	* the case of CountedCompleters, others with the same root task).
205	>	* However, because most shared/external queue operations are more
206	>	* expensive than internal, and because, at steady state, external
207	>	* submitters will compete for CPU with workers, ForkJoinTask.join
208	>	* and related methods disable them from repeatedly helping to
209	>	* process tasks if all workers are active. Insertion of tasks in
210	>	* shared mode requires a lock (mainly to protect in the case of
211	>	* resizing) but we use only a simple spinlock (using bits in
212	>	* field qlock), because submitters encountering a busy queue move
213	>	* on to try or create other queues -- they block only when
214	>	* creating and registering new queues.
215		*
216		* Management
217		* ==========
#	Line 232 \| Line 233 \| public class ForkJoinPool extends Abstra
233		* and their negations (used for thresholding) to fit into 16bit
234		* fields.
235		*
236	<	* Field "runState" contains 32 bits needed to register and
237	<	* deregister WorkQueues, as well as to enable shutdown. It is
238	<	* only modified under a lock (normally briefly held, but
239	<	* occasionally protecting allocations and resizings) but even
240	<	* when locked remains available to check consistency.
236	>	* Field "plock" is a form of sequence lock with a saturating
237	>	* shutdown bit (similarly for per-queue "qlocks"), mainly
238	>	* protecting updates to the workQueues array, as well as to
239	>	* enable shutdown. When used as a lock, it is normally only very
240	>	* briefly held, so is nearly always available after at most a
241	>	* brief spin, but we use a monitor-based backup strategy to
242	>	* block when needed.
243		*
244		* Recording WorkQueues. WorkQueues are recorded in the
245	<	* "workQueues" array that is created upon pool construction and
246	<	* expanded if necessary. Updates to the array while recording
247	<	* new workers and unrecording terminated ones are protected from
248	<	* each other by a lock but the array is otherwise concurrently
249	<	* readable, and accessed directly. To simplify index-based
250	<	* operations, the array size is always a power of two, and all
251	<	* readers must tolerate null slots. Shared (submission) queues
252	<	* are at even indices, worker queues at odd indices. Grouping
253	<	* them together in this way simplifies and speeds up task
254	<	* scanning.
245	>	* "workQueues" array that is created upon first use and expanded
246	>	* if necessary. Updates to the array while recording new workers
247	>	* and unrecording terminated ones are protected from each other
248	>	* by a lock but the array is otherwise concurrently readable, and
249	>	* accessed directly. To simplify index-based operations, the
250	>	* array size is always a power of two, and all readers must
251	>	* tolerate null slots. Worker queues are at odd indices. Shared
252	>	* (submission) queues are at even indices, up to a maximum of 64
253	>	* slots, to limit growth even if array needs to expand to add
254	>	* more workers. Grouping them together in this way simplifies and
255	>	* speeds up task scanning.
256		*
257		* All worker thread creation is on-demand, triggered by task
258		* submissions, replacement of terminated workers, and/or
#	Line 309 \| Line 313 \| public class ForkJoinPool extends Abstra
313		*
314		* Signalling. We create or wake up workers only when there
315		* appears to be at least one task they might be able to find and
316	<	* execute. When a submission is added or another worker adds a
317	<	* task to a queue that previously had fewer than two tasks, they
318	<	* signal waiting workers (or trigger creation of new ones if
319	<	* fewer than the given parallelism level -- see signalWork).
320	<	* These primary signals are buttressed by signals during rescans;
321	<	* together these cover the signals needed in cases when more
322	<	* tasks are pushed but untaken, and improve performance compared
323	<	* to having one thread wake up all workers.
316	>	* execute. However, many other threads may notice the same task
317	>	* and each signal to wake up a thread that might take it. So in
318	>	* general, pools will be over-signalled. When a submission is
319	>	* added or another worker adds a task to a queue that has fewer
320	>	* than two tasks, they signal waiting workers (or trigger
321	>	* creation of new ones if fewer than the given parallelism level
322	>	* -- signalWork), and may leave a hint to the unparked worker to
323	>	* help signal others upon wakeup). These primary signals are
324	>	* buttressed by others (see method helpSignal) whenever other
325	>	* threads scan for work or do not have a task to process. On
326	>	* most platforms, signalling (unpark) overhead time is noticeably
327	>	* long, and the time between signalling a thread and it actually
328	>	* making progress can be very noticeably long, so it is worth
329	>	* offloading these delays from critical paths as much as
330	>	* possible.
331		*
332		* Trimming workers. To release resources after periods of lack of
333		* use, a worker starting to wait when the pool is quiescent will
334	<	* time out and terminate if the pool has remained quiescent for
335	<	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
336	<	* terminating all workers after long periods of non-use.
334	>	* time out and terminate if the pool has remained quiescent for a
335	>	* given period -- a short period if there are more threads than
336	>	* parallelism, longer as the number of threads decreases. This
337	>	* will slowly propagate, eventually terminating all workers after
338	>	* periods of non-use.
339		*
340		* Shutdown and Termination. A call to shutdownNow atomically sets
341	<	* a runState bit and then (non-atomically) sets each worker's
342	<	* runState status, cancels all unprocessed tasks, and wakes up
341	>	* a plock bit and then (non-atomically) sets each worker's
342	>	* qlock status, cancels all unprocessed tasks, and wakes up
343		* all waiting workers. Detecting whether termination should
344		* commence after a non-abrupt shutdown() call requires more work
345		* and bookkeeping. We need consensus about quiescence (i.e., that
#	Line 354 \| Line 367 \| public class ForkJoinPool extends Abstra
367		* method tryCompensate() may create or re-activate a spare
368		* thread to compensate for blocked joiners until they unblock.
369		*
370	<	* A third form (implemented in tryRemoveAndExec and
371	<	* tryPollForAndExec) amounts to helping a hypothetical
372	<	* compensator: If we can readily tell that a possible action of a
373	<	* compensator is to steal and execute the task being joined, the
374	<	* joining thread can do so directly, without the need for a
375	<	* compensation thread (although at the expense of larger run-time
376	<	* stacks, but the tradeoff is typically worthwhile).
370	>	* A third form (implemented in tryRemoveAndExec) amounts to
371	>	* helping a hypothetical compensator: If we can readily tell that
372	>	* a possible action of a compensator is to steal and execute the
373	>	* task being joined, the joining thread can do so directly,
374	>	* without the need for a compensation thread (although at the
375	>	* expense of larger run-time stacks, but the tradeoff is
376	>	* typically worthwhile).
377		*
378		* The ManagedBlocker extension API can't use helping so relies
379		* only on compensation in method awaitBlocker.
#	Line 382 \| Line 395 \| public class ForkJoinPool extends Abstra
395		* steals, rather than use per-task bookkeeping. This sometimes
396		* requires a linear scan of workQueues array to locate stealers,
397		* but often doesn't because stealers leave hints (that may become
398	<	* stale/wrong) of where to locate them. A stealHint is only a
399	<	* hint because a worker might have had multiple steals and the
400	<	* hint records only one of them (usually the most current).
401	<	* Hinting isolates cost to when it is needed, rather than adding
402	<	* to per-task overhead. (2) It is "shallow", ignoring nesting
403	<	* and potentially cyclic mutual steals. (3) It is intentionally
398	>	* stale/wrong) of where to locate them. It is only a hint
399	>	* because a worker might have had multiple steals and the hint
400	>	* records only one of them (usually the most current). Hinting
401	>	* isolates cost to when it is needed, rather than adding to
402	>	* per-task overhead. (2) It is "shallow", ignoring nesting and
403	>	* potentially cyclic mutual steals. (3) It is intentionally
404		* racy: field currentJoin is updated only while actively joining,
405		* which means that we miss links in the chain during long-lived
406		* tasks, GC stalls etc (which is OK since blocking in such cases
#	Line 395 \| Line 408 \| public class ForkJoinPool extends Abstra
408		* to find work (see MAX_HELP) and fall back to suspending the
409		* worker and if necessary replacing it with another.
410		*
411	+	* Helping actions for CountedCompleters are much simpler: Method
412	+	* helpComplete can take and execute any task with the same root
413	+	* as the task being waited on. However, this still entails some
414	+	* traversal of completer chains, so is less efficient than using
415	+	* CountedCompleters without explicit joins.
416	+	*
417		* It is impossible to keep exactly the target parallelism number
418		* of threads running at any given time. Determining the
419		* existence of conservatively safe helping targets, the
#	Line 416 \| Line 435 \| public class ForkJoinPool extends Abstra
435		* intractable) game with an opponent that may choose the worst
436		* (for us) active thread to stall at any time. We take several
437		* precautions to bound losses (and thus bound gains), mainly in
438	<	* methods tryCompensate and awaitJoin: (1) We only try
439	<	* compensation after attempting enough helping steps (measured
440	<	* via counting and timing) that we have already consumed the
441	<	* estimated cost of creating and activating a new thread. (2) We
442	<	* allow up to 50% of threads to be blocked before initially
443	<	* adding any others, and unless completely saturated, check that
444	<	* some work is available for a new worker before adding. Also, we
445	<	* create up to only 50% more threads until entering a mode that
446	<	* only adds a thread if all others are possibly blocked. All
447	<	* together, this means that we might be half as fast to react,
448	<	* and create half as many threads as possible in the ideal case,
449	<	* but present vastly fewer anomalies in all other cases compared
450	<	* to both more aggressive and more conservative alternatives.
451	<	*
452	<	* Style notes: There is a lot of representation-level coupling
453	<	* among classes ForkJoinPool, ForkJoinWorkerThread, and
454	<	* ForkJoinTask. The fields of WorkQueue maintain data structures
455	<	* managed by ForkJoinPool, so are directly accessed. There is
456	<	* little point trying to reduce this, since any associated future
457	<	* changes in representations will need to be accompanied by
458	<	* algorithmic changes anyway. Several methods intrinsically
459	<	* sprawl because they must accumulate sets of consistent reads of
460	<	* volatiles held in local variables. Methods signalWork() and
461	<	* scan() are the main bottlenecks, so are especially heavily
438	>	* methods tryCompensate and awaitJoin.
439	>	*
440	>	* Common Pool
441	>	* ===========
442	>	*
443	>	* The static commonPool always exists after static
444	>	* initialization. Since it (or any other created pool) need
445	>	* never be used, we minimize initial construction overhead and
446	>	* footprint to the setup of about a dozen fields, with no nested
447	>	* allocation. Most bootstrapping occurs within method
448	>	* fullExternalPush during the first submission to the pool.
449	>	*
450	>	* When external threads submit to the common pool, they can
451	>	* perform some subtask processing (see externalHelpJoin and
452	>	* related methods). We do not need to record whether these
453	>	* submissions are to the common pool -- if not, externalHelpJoin
454	>	* returns quickly (at the most helping to signal some common pool
455	>	* workers). These submitters would otherwise be blocked waiting
456	>	* for completion, so the extra effort (with liberally sprinkled
457	>	* task status checks) in inapplicable cases amounts to an odd
458	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
459	>	*
460	>	* Style notes
461	>	* ===========
462	>	*
463	>	* There is a lot of representation-level coupling among classes
464	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
465	>	* fields of WorkQueue maintain data structures managed by
466	>	* ForkJoinPool, so are directly accessed. There is little point
467	>	* trying to reduce this, since any associated future changes in
468	>	* representations will need to be accompanied by algorithmic
469	>	* changes anyway. Several methods intrinsically sprawl because
470	>	* they must accumulate sets of consistent reads of volatiles held
471	>	* in local variables. Methods signalWork() and scan() are the
472	>	* main bottlenecks, so are especially heavily
473		* micro-optimized/mangled. There are lots of inline assignments
474		* (of form "while ((local = field) != 0)") which are usually the
475		* simplest way to ensure the required read orderings (which are
#	Line 447 \| Line 477 \| public class ForkJoinPool extends Abstra
477		* declarations of these locals at the heads of methods or blocks.
478		* There are several occurrences of the unusual "do {} while
479		* (!cas...)" which is the simplest way to force an update of a
480	<	* CAS'ed variable. There are also other coding oddities that help
480	>	* CAS'ed variable. There are also other coding oddities (including
481	>	* several unnecessary-looking hoisted null checks) that help
482		* some methods perform reasonably even when interpreted (not
483		* compiled).
484		*
#	Line 496 \| Line 527 \| public class ForkJoinPool extends Abstra
527		* Default ForkJoinWorkerThreadFactory implementation; creates a
528		* new ForkJoinWorkerThread.
529		*/
530	<	static class DefaultForkJoinWorkerThreadFactory
530	>	static final class DefaultForkJoinWorkerThreadFactory
531		implements ForkJoinWorkerThreadFactory {
532	<	public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
532	>	public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
533		return new ForkJoinWorkerThread(pool);
534		}
535		}
536
537		/**
538	<	* A simple non-reentrant lock used for exclusion when managing
539	<	* queues and workers. We use a custom lock so that we can readily
540	<	* probe lock state in constructions that check among alternative
541	<	* actions. The lock is normally only very briefly held, and
542	<	* sometimes treated as a spinlock, but other usages block to
543	<	* reduce overall contention in those cases where locked code
544	<	* bodies perform allocation/resizing.
545	<	*/
546	<	static final class Mutex extends AbstractQueuedSynchronizer {
547	<	public final boolean tryAcquire(int ignore) {
548	<	return compareAndSetState(0, 1);
549	<	}
550	<	public final boolean tryRelease(int ignore) {
551	<	setState(0);
552	<	return true;
553	<	}
554	<	public final void lock() { acquire(0); }
524	<	public final void unlock() { release(0); }
525	<	public final boolean isHeldExclusively() { return getState() == 1; }
526	<	public final Condition newCondition() { return new ConditionObject(); }
538	>	* Per-thread records for threads that submit to pools. Currently
539	>	* holds only pseudo-random seed / index that is used to choose
540	>	* submission queues in method externalPush. In the future, this may
541	>	* also incorporate a means to implement different task rejection
542	>	* and resubmission policies.
543	>	*
544	>	* Seeds for submitters and workers/workQueues work in basically
545	>	* the same way but are initialized and updated using slightly
546	>	* different mechanics. Both are initialized using the same
547	>	* approach as in class ThreadLocal, where successive values are
548	>	* unlikely to collide with previous values. Seeds are then
549	>	* randomly modified upon collisions using xorshifts, which
550	>	* requires a non-zero seed.
551	>	*/
552	>	static final class Submitter {
553	>	int seed;
554	>	Submitter(int s) { seed = s; }
555		}
556
557		/**
#	Line 533 \| Line 561 \| public class ForkJoinPool extends Abstra
561		* actually do anything beyond having a unique identity.
562		*/
563		static final class EmptyTask extends ForkJoinTask<Void> {
564	+	private static final long serialVersionUID = -7721805057305804111L;
565		EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
566		public final Void getRawResult() { return null; }
567		public final void setRawResult(Void x) {}
#	Line 553 \| Line 582 \| public class ForkJoinPool extends Abstra
582		*
583		* Field "top" is the index (mod array.length) of the next queue
584		* slot to push to or pop from. It is written only by owner thread
585	<	* for push, or under lock for trySharedPush, and accessed by
586	<	* other threads only after reading (volatile) base. Both top and
587	<	* base are allowed to wrap around on overflow, but (top - base)
588	<	* (or more commonly -(base - top) to force volatile read of base
589	<	* before top) still estimates size.
585	>	* for push, or under lock for external/shared push, and accessed
586	>	* by other threads only after reading (volatile) base. Both top
587	>	* and base are allowed to wrap around on overflow, but (top -
588	>	* base) (or more commonly -(base - top) to force volatile read of
589	>	* base before top) still estimates size. The lock ("qlock") is
590	>	* forced to -1 on termination, causing all further lock attempts
591	>	* to fail. (Note: we don't need CAS for termination state because
592	>	* upon pool shutdown, all shared-queues will stop being used
593	>	* anyway.) Nearly all lock bodies are set up so that exceptions
594	>	* within lock bodies are "impossible" (modulo JVM errors that
595	>	* would cause failure anyway.)
596		*
597		* The array slots are read and written using the emulation of
598		* volatiles/atomics provided by Unsafe. Insertions must in
599		* general use putOrderedObject as a form of releasing store to
600		* ensure that all writes to the task object are ordered before
601	<	* its publication in the queue. (Although we can avoid one case
602	<	* of this when locked in trySharedPush.) All removals entail a
603	<	* CAS to null. The array is always a power of two. To ensure
604	<	* safety of Unsafe array operations, all accesses perform
570	<	* explicit null checks and implicit bounds checks via
571	<	* power-of-two masking.
601	>	* its publication in the queue. All removals entail a CAS to
602	>	* null. The array is always a power of two. To ensure safety of
603	>	* Unsafe array operations, all accesses perform explicit null
604	>	* checks and implicit bounds checks via power-of-two masking.
605		*
606		* In addition to basic queuing support, this class contains
607		* fields described elsewhere to control execution. It turns out
608	<	* to work better memory-layout-wise to include them in this
609	<	* class rather than a separate class.
608	>	* to work better memory-layout-wise to include them in this class
609	>	* rather than a separate class.
610		*
611		* Performance on most platforms is very sensitive to placement of
612		* instances of both WorkQueues and their arrays -- we absolutely
#	Line 587 \| Line 620 \| public class ForkJoinPool extends Abstra
620		* trades off slightly slower average field access for the sake of
621		* avoiding really bad worst-case access. (Until better JVM
622		* support is in place, this padding is dependent on transient
623	<	* properties of JVM field layout rules.) We also take care in
623	>	* properties of JVM field layout rules.) We also take care in
624		* allocating, sizing and resizing the array. Non-shared queue
625	<	* arrays are initialized (via method growArray) by workers before
626	<	* use. Others are allocated on first use.
625	>	* arrays are initialized by workers before use. Others are
626	>	* allocated on first use.
627		*/
628		static final class WorkQueue {
629		/**
#	Line 613 \| Line 646 \| public class ForkJoinPool extends Abstra
646		*/
647		static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
648
649	<	volatile long totalSteals; // cumulative number of steals
649	>	// Heuristic padding to ameliorate unfortunate memory placements
650	>	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
651	>
652		int seed; // for random scanning; initialize nonzero
653		volatile int eventCount; // encoded inactivation count; < 0 if inactive
654		int nextWait; // encoded record of next event waiter
655	<	int rescans; // remaining scans until block
621	<	int nsteals; // top-level task executions since last idle
622	<	final int mode; // lifo, fifo, or shared
655	>	int hint; // steal or signal hint (index)
656		int poolIndex; // index of this queue in pool (or 0)
657	<	int stealHint; // index of most recent known stealer
658	<	volatile int runState; // 1: locked, -1: terminate; else 0
657	>	final int mode; // 0: lifo, > 0: fifo, < 0: shared
658	>	int nsteals; // number of steals
659	>	volatile int qlock; // 1: locked, -1: terminate; else 0
660		volatile int base; // index of next slot for poll
661		int top; // index of next slot for push
662		ForkJoinTask<?>[] array; // the elements (initially unallocated)
#	Line 631 \| Line 665 \| public class ForkJoinPool extends Abstra
665		volatile Thread parker; // == owner during call to park; else null
666		volatile ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
667		ForkJoinTask<?> currentSteal; // current non-local task being executed
634	–	// Heuristic padding to ameliorate unfortunate memory placements
635	–	Object p00, p01, p02, p03, p04, p05, p06, p07;
636	–	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
668
669	<	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
670	<	this.mode = mode;
669	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
670	>	volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d;
671	>
672	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
673	>	int seed) {
674		this.pool = pool;
675		this.owner = owner;
676	+	this.mode = mode;
677	+	this.seed = seed;
678		// Place indices in the center of array (that is not yet allocated)
679		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
680		}
#	Line 651 \| Line 687 \| public class ForkJoinPool extends Abstra
687		return (n >= 0) ? 0 : -n; // ignore transient negative
688		}
689
690	<	/**
690	>	/**
691		* Provides a more accurate estimate of whether this queue has
692		* any tasks than does queueSize, by checking whether a
693		* near-empty queue has at least one unclaimed task.
#	Line 663 \| Line 699 \| public class ForkJoinPool extends Abstra
699		(n == -1 &&
700		((a = array) == null \|\|
701		(m = a.length - 1) < 0 \|\|
702	<	U.getObjectVolatile
703	<	(a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
702	>	U.getObject
703	>	(a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null)));
704		}
705
706		/**
707	<	* Pushes a task. Call only by owner in unshared queues.
707	>	* Pushes a task. Call only by owner in unshared queues. (The
708	>	* shared-queue version is embedded in method externalPush.)
709		*
710		* @param task the task. Caller must ensure non-null.
711		* @throw RejectedExecutionException if array cannot be resized
#	Line 677 \| Line 714 \| public class ForkJoinPool extends Abstra
714		ForkJoinTask<?>[] a; ForkJoinPool p;
715		int s = top, m, n;
716		if ((a = array) != null) { // ignore if queue removed
717	<	U.putOrderedObject
718	<	(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
717	>	int j = (((m = a.length - 1) & s) << ASHIFT) + ABASE;
718	>	U.putOrderedObject(a, j, task);
719		if ((n = (top = s + 1) - base) <= 2) {
720		if ((p = pool) != null)
721	<	p.signalWork();
721	>	p.signalWork(this);
722		}
723		else if (n >= m)
724	<	growArray(true);
724	>	growArray();
725		}
726		}
727
728	<	/**
729	<	* Pushes a task if lock is free and array is either big
730	<	* enough or can be resized to be big enough.
731	<	*
695	<	* @param task the task. Caller must ensure non-null.
696	<	* @return true if submitted
728	>	/**
729	>	* Initializes or doubles the capacity of array. Call either
730	>	* by owner or with lock held -- it is OK for base, but not
731	>	* top, to move while resizings are in progress.
732		*/
733	<	final boolean trySharedPush(ForkJoinTask<?> task) {
734	<	boolean submitted = false;
735	<	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
736	<	ForkJoinTask<?>[] a = array;
737	<	int s = top;
738	<	try {
739	<	if ((a != null && a.length > s + 1 - base) \|\|
740	<	(a = growArray(false)) != null) { // must presize
741	<	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
742	<	U.putObject(a, (long)j, task); // don't need "ordered"
743	<	top = s + 1;
744	<	submitted = true;
745	<	}
746	<	} finally {
747	<	runState = 0; // unlock
748	<	}
733	>	final ForkJoinTask<?>[] growArray() {
734	>	ForkJoinTask<?>[] oldA = array;
735	>	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
736	>	if (size > MAXIMUM_QUEUE_CAPACITY)
737	>	throw new RejectedExecutionException("Queue capacity exceeded");
738	>	int oldMask, t, b;
739	>	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
740	>	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
741	>	(t = top) - (b = base) > 0) {
742	>	int mask = size - 1;
743	>	do {
744	>	ForkJoinTask<?> x;
745	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
746	>	int j = ((b & mask) << ASHIFT) + ABASE;
747	>	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
748	>	if (x != null &&
749	>	U.compareAndSwapObject(oldA, oldj, x, null))
750	>	U.putObjectVolatile(a, j, x);
751	>	} while (++b != t);
752		}
753	<	return submitted;
753	>	return a;
754		}
755
756		/**
757		* Takes next task, if one exists, in LIFO order. Call only
758	<	* by owner in unshared queues. (We do not have a shared
721	<	* version of this method because it is never needed.)
758	>	* by owner in unshared queues.
759		*/
760		final ForkJoinTask<?> pop() {
761		ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
#	Line 773 \| Line 810 \| public class ForkJoinPool extends Abstra
810		else if (base == b) {
811		if (b + 1 == top)
812		break;
813	<	Thread.yield(); // wait for lagging update
813	>	Thread.yield(); // wait for lagging update (very rare)
814		}
815		}
816		return null;
#	Line 800 \| Line 837 \| public class ForkJoinPool extends Abstra
837
838		/**
839		* Pops the given task only if it is at the current top.
840	+	* (A shared version is available only via FJP.tryExternalUnpush)
841		*/
842		final boolean tryUnpush(ForkJoinTask<?> t) {
843		ForkJoinTask<?>[] a; int s;
#	Line 813 \| Line 851 \| public class ForkJoinPool extends Abstra
851		}
852
853		/**
816	–	* Polls the given task only if it is at the current base.
817	–	*/
818	–	final boolean pollFor(ForkJoinTask<?> task) {
819	–	ForkJoinTask<?>[] a; int b;
820	–	if ((b = base) - top < 0 && (a = array) != null) {
821	–	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
822	–	if (U.getObjectVolatile(a, j) == task && base == b &&
823	–	U.compareAndSwapObject(a, j, task, null)) {
824	–	base = b + 1;
825	–	return true;
826	–	}
827	–	}
828	–	return false;
829	–	}
830	–
831	–	/**
832	–	* Initializes or doubles the capacity of array. Call either
833	–	* by owner or with lock held -- it is OK for base, but not
834	–	* top, to move while resizings are in progress.
835	–	*
836	–	* @param rejectOnFailure if true, throw exception if capacity
837	–	* exceeded (relayed ultimately to user); else return null.
838	–	*/
839	–	final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
840	–	ForkJoinTask<?>[] oldA = array;
841	–	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
842	–	if (size <= MAXIMUM_QUEUE_CAPACITY) {
843	–	int oldMask, t, b;
844	–	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
845	–	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
846	–	(t = top) - (b = base) > 0) {
847	–	int mask = size - 1;
848	–	do {
849	–	ForkJoinTask<?> x;
850	–	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
851	–	int j = ((b & mask) << ASHIFT) + ABASE;
852	–	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
853	–	if (x != null &&
854	–	U.compareAndSwapObject(oldA, oldj, x, null))
855	–	U.putObjectVolatile(a, j, x);
856	–	} while (++b != t);
857	–	}
858	–	return a;
859	–	}
860	–	else if (!rejectOnFailure)
861	–	return null;
862	–	else
863	–	throw new RejectedExecutionException("Queue capacity exceeded");
864	–	}
865	–
866	–	/**
854		* Removes and cancels all known tasks, ignoring any exceptions.
855		*/
856		final void cancelAll() {
#	Line 887 \| Line 874 \| public class ForkJoinPool extends Abstra
874		return seed = r ^= r << 5;
875		}
876
877	<	// Execution methods
877	>	// Specialized execution methods
878
879		/**
880		* Pops and runs tasks until empty.
#	Line 916 \| Line 903 \| public class ForkJoinPool extends Abstra
903		}
904
905		/**
906	<	* If present, removes from queue and executes the given task, or
907	<	* any other cancelled task. Returns (true) immediately on any CAS
906	>	* If present, removes from queue and executes the given task,
907	>	* or any other cancelled task. Returns (true) on any CAS
908		* or consistency check failure so caller can retry.
909		*
910	<	* @return 0 if no progress can be made, else positive
924	<	* (this unusual convention simplifies use with tryHelpStealer.)
910	>	* @return false if no progress can be made, else true;
911		*/
912	<	final int tryRemoveAndExec(ForkJoinTask<?> task) {
913	<	int stat = 1;
928	<	boolean removed = false, empty = true;
912	>	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
913	>	boolean stat = true, removed = false, empty = true;
914		ForkJoinTask<?>[] a; int m, s, b, n;
915		if ((a = array) != null && (m = a.length - 1) >= 0 &&
916		(n = (s = top) - (b = base)) > 0) {
#	Line 955 \| Line 940 \| public class ForkJoinPool extends Abstra
940		}
941		if (--n == 0) {
942		if (!empty && base == b)
943	<	stat = 0;
943	>	stat = false;
944		break;
945		}
946		}
#	Line 966 \| Line 951 \| public class ForkJoinPool extends Abstra
951		}
952
953		/**
954	+	* Polls for and executes the given task or any other task in
955	+	* its CountedCompleter computation
956	+	*/
957	+	final boolean pollAndExecCC(ForkJoinTask<?> root) {
958	+	ForkJoinTask<?>[] a; int b; Object o;
959	+	outer: while ((b = base) - top < 0 && (a = array) != null) {
960	+	long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
961	+	if ((o = U.getObject(a, j)) == null \|\|
962	+	!(o instanceof CountedCompleter))
963	+	break;
964	+	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
965	+	if (r == root) {
966	+	if (base == b &&
967	+	U.compareAndSwapObject(a, j, t, null)) {
968	+	base = b + 1;
969	+	t.doExec();
970	+	return true;
971	+	}
972	+	else
973	+	break; // restart
974	+	}
975	+	if ((r = r.completer) == null)
976	+	break outer; // not part of root computation
977	+	}
978	+	}
979	+	return false;
980	+	}
981	+
982	+	/**
983		* Executes a top-level task and any local tasks remaining
984		* after execution.
985		*/
986		final void runTask(ForkJoinTask<?> t) {
987		if (t != null) {
988	<	currentSteal = t;
989	<	t.doExec();
990	<	if (top != base) { // process remaining local tasks
988	>	(currentSteal = t).doExec();
989	>	currentSteal = null;
990	>	if (base - top < 0) { // process remaining local tasks
991		if (mode == 0)
992		popAndExecAll();
993		else
994		pollAndExecAll();
995		}
996		++nsteals;
997	<	currentSteal = null;
997	>	hint = -1;
998		}
999		}
1000
#	Line 990 \| Line 1004 \| public class ForkJoinPool extends Abstra
1004		final void runSubtask(ForkJoinTask<?> t) {
1005		if (t != null) {
1006		ForkJoinTask<?> ps = currentSteal;
1007	<	currentSteal = t;
994	<	t.doExec();
1007	>	(currentSteal = t).doExec();
1008		currentSteal = ps;
1009		}
1010		}
#	Line 1026 \| Line 1039 \| public class ForkJoinPool extends Abstra
1039
1040		// Unsafe mechanics
1041		private static final sun.misc.Unsafe U;
1042	<	private static final long RUNSTATE;
1042	>	private static final long QLOCK;
1043		private static final int ABASE;
1044		private static final int ASHIFT;
1045		static {
#	Line 1035 \| Line 1048 \| public class ForkJoinPool extends Abstra
1048		U = getUnsafe();
1049		Class<?> k = WorkQueue.class;
1050		Class<?> ak = ForkJoinTask[].class;
1051	<	RUNSTATE = U.objectFieldOffset
1052	<	(k.getDeclaredField("runState"));
1051	>	QLOCK = U.objectFieldOffset
1052	>	(k.getDeclaredField("qlock"));
1053		ABASE = U.arrayBaseOffset(ak);
1054		s = U.arrayIndexScale(ak);
1055		} catch (Exception e) {
#	Line 1048 \| Line 1061 \| public class ForkJoinPool extends Abstra
1061		}
1062		}
1063
1051	–	/**
1052	–	* Per-thread records for threads that submit to pools. Currently
1053	–	* holds only pseudo-random seed / index that is used to choose
1054	–	* submission queues in method doSubmit. In the future, this may
1055	–	* also incorporate a means to implement different task rejection
1056	–	* and resubmission policies.
1057	–	*
1058	–	* Seeds for submitters and workers/workQueues work in basically
1059	–	* the same way but are initialized and updated using slightly
1060	–	* different mechanics. Both are initialized using the same
1061	–	* approach as in class ThreadLocal, where successive values are
1062	–	* unlikely to collide with previous values. This is done during
1063	–	* registration for workers, but requires a separate AtomicInteger
1064	–	* for submitters. Seeds are then randomly modified upon
1065	–	* collisions using xorshifts, which requires a non-zero seed.
1066	–	*/
1067	–	static final class Submitter {
1068	–	int seed;
1069	–	Submitter() {
1070	–	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1071	–	seed = (s == 0) ? 1 : s; // ensure non-zero
1072	–	}
1073	–	}
1074	–
1075	–	/** ThreadLocal class for Submitters */
1076	–	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1077	–	public Submitter initialValue() { return new Submitter(); }
1078	–	}
1079	–
1064		// static fields (initialized in static initializer below)
1065
1066		/**
#	Line 1087 \| Line 1071 \| public class ForkJoinPool extends Abstra
1071		defaultForkJoinWorkerThreadFactory;
1072
1073		/**
1074	<	* Generator for assigning sequence numbers as pool names.
1075	<	*/
1076	<	private static final AtomicInteger poolNumberGenerator;
1077	<
1078	<	/**
1095	<	* Generator for initial hashes/seeds for submitters. Accessed by
1096	<	* Submitter class constructor.
1074	>	* Per-thread submission bookkeeping. Shared across all pools
1075	>	* to reduce ThreadLocal pollution and because random motion
1076	>	* to avoid contention in one pool is likely to hold for others.
1077	>	* Lazily initialized on first submission (but null-checked
1078	>	* in other contexts to avoid unnecessary initialization).
1079		*/
1080	<	static final AtomicInteger nextSubmitterSeed;
1080	>	static final ThreadLocal<Submitter> submitters;
1081
1082		/**
1083		* Permission required for callers of methods that may start or
#	Line 1104 \| Line 1086 \| public class ForkJoinPool extends Abstra
1086		private static final RuntimePermission modifyThreadPermission;
1087
1088		/**
1089	<	* Per-thread submission bookkeeping. Shared across all pools
1090	<	* to reduce ThreadLocal pollution and because random motion
1091	<	* to avoid contention in one pool is likely to hold for others.
1089	>	* Common (static) pool. Non-null for public use unless a static
1090	>	* construction exception, but internal usages null-check on use
1091	>	* to paranoically avoid potential initialization circularities
1092	>	* as well as to simplify generated code.
1093	>	*/
1094	>	static final ForkJoinPool commonPool;
1095	>
1096	>	/**
1097	>	* Common pool parallelism. Must equal commonPool.parallelism.
1098	>	*/
1099	>	static final int commonPoolParallelism;
1100	>
1101	>	/**
1102	>	* Sequence number for creating workerNamePrefix.
1103		*/
1104	<	private static final ThreadSubmitter submitters;
1104	>	private static int poolNumberSequence;
1105	>
1106	>	/**
1107	>	* Return the next sequence number. We don't expect this to
1108	>	* ever contend so use simple builtin sync.
1109	>	*/
1110	>	private static final synchronized int nextPoolId() {
1111	>	return ++poolNumberSequence;
1112	>	}
1113
1114		// static constants
1115
1116		/**
1117	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
1118	<	* task when the pool is quiescent to instead try to shrink the
1119	<	* number of workers. The exact value does not matter too
1120	<	* much. It must be short enough to release resources during
1121	<	* sustained periods of idleness, but not so short that threads
1122	<	* are continually re-created.
1117	>	* Initial timeout value (in nanoseconds) for the thread
1118	>	* triggering quiescence to park waiting for new work. On timeout,
1119	>	* the thread will instead try to shrink the number of
1120	>	* workers. The value should be large enough to avoid overly
1121	>	* aggressive shrinkage during most transient stalls (long GCs
1122	>	* etc).
1123		*/
1124	<	private static final long SHRINK_RATE =
1124	<	4L * 1000L * 1000L * 1000L; // 4 seconds
1124	>	private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
1125
1126		/**
1127	<	* The timeout value for attempted shrinkage, includes
1128	<	* some slop to cope with system timer imprecision.
1127	>	* Timeout value when there are more threads than parallelism level
1128		*/
1129	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1129	>	private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
1130
1131		/**
1132		* The maximum stolen->joining link depth allowed in method
1133	<	* tryHelpStealer. Must be a power of two. This value also
1135	<	* controls the maximum number of times to try to help join a task
1136	<	* without any apparent progress or change in pool state before
1137	<	* giving up and blocking (see awaitJoin). Depths for legitimate
1133	>	* tryHelpStealer. Must be a power of two. Depths for legitimate
1134		* chains are unbounded, but we use a fixed constant to avoid
1135		* (otherwise unchecked) cycles and to bound staleness of
1136		* traversal parameters at the expense of sometimes blocking when
#	Line 1143 \| Line 1139 \| public class ForkJoinPool extends Abstra
1139		private static final int MAX_HELP = 64;
1140
1141		/**
1146	–	* Secondary time-based bound (in nanosecs) for helping attempts
1147	–	* before trying compensated blocking in awaitJoin. Used in
1148	–	* conjunction with MAX_HELP to reduce variance due to different
1149	–	* polling rates associated with different helping options. The
1150	–	* value should roughly approximate the time required to create
1151	–	* and/or activate a worker thread.
1152	–	*/
1153	–	private static final long COMPENSATION_DELAY = 1L << 18; // ~0.25 millisec
1154	–
1155	–	/**
1142		* Increment for seed generators. See class ThreadLocal for
1143		* explanation.
1144		*/
#	Line 1186 \| Line 1172 \| public class ForkJoinPool extends Abstra
1172		* scan for them to avoid queuing races. Note however that
1173		* eventCount updates lag releases so usage requires care.
1174		*
1175	<	* Field runState is an int packed with:
1175	>	* Field plock is an int packed with:
1176		* SHUTDOWN: true if shutdown is enabled (1 bit)
1177	<	* SEQ: a sequence number updated upon (de)registering workers (30 bits)
1178	<	* INIT: set true after workQueues array construction (1 bit)
1177	>	* SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
1178	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1179		*
1180		* The sequence number enables simple consistency checks:
1181		* Staleness of read-only operations on the workQueues array can
1182	<	* be checked by comparing runState before vs after the reads.
1182	>	* be checked by comparing plock before vs after the reads.
1183		*/
1184
1185		// bit positions/shifts for fields
#	Line 1205 \| Line 1191 \| public class ForkJoinPool extends Abstra
1191		// bounds
1192		private static final int SMASK = 0xffff; // short bits
1193		private static final int MAX_CAP = 0x7fff; // max #workers - 1
1194	<	private static final int SQMASK = 0xfffe; // even short bits
1194	>	private static final int EVENMASK = 0xfffe; // even short bits
1195	>	private static final int SQMASK = 0x007e; // max 64 (even) slots
1196		private static final int SHORT_SIGN = 1 << 15;
1197		private static final int INT_SIGN = 1 << 31;
1198
#	Line 1230 \| Line 1217 \| public class ForkJoinPool extends Abstra
1217		private static final int E_MASK = 0x7fffffff; // no STOP_BIT
1218		private static final int E_SEQ = 1 << EC_SHIFT;
1219
1220	<	// runState bits
1220	>	// plock bits
1221		private static final int SHUTDOWN = 1 << 31;
1222	+	private static final int PL_LOCK = 2;
1223	+	private static final int PL_SIGNAL = 1;
1224	+	private static final int PL_SPINS = 1 << 8;
1225
1226		// access mode for WorkQueue
1227		static final int LIFO_QUEUE = 0;
1228		static final int FIFO_QUEUE = 1;
1229		static final int SHARED_QUEUE = -1;
1230
1231	+	// bounds for #steps in scan loop -- must be power 2 minus 1
1232	+	private static final int MIN_SCAN = 0x1ff; // cover estimation slop
1233	+	private static final int MAX_SCAN = 0x1ffff; // 4 * max workers
1234	+
1235		// Instance fields
1236
1237		/*
1238	<	* Field layout order in this class tends to matter more than one
1239	<	* would like. Runtime layout order is only loosely related to
1238	>	* Field layout of this class tends to matter more than one would
1239	>	* like. Runtime layout order is only loosely related to
1240		* declaration order and may differ across JVMs, but the following
1241		* empirically works OK on current JVMs.
1242		*/
1243
1244	+	// Heuristic padding to ameliorate unfortunate memory placements
1245	+	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
1246	+
1247	+	volatile long stealCount; // collects worker counts
1248		volatile long ctl; // main pool control
1249	<	final int parallelism; // parallelism level
1250	<	final int localMode; // per-worker scheduling mode
1251	<	final int submitMask; // submit queue index bound
1254	<	int nextSeed; // for initializing worker seeds
1255	<	volatile int runState; // shutdown status and seq
1249	>	volatile int plock; // shutdown status and seqLock
1250	>	volatile int indexSeed; // worker/submitter index seed
1251	>	final int config; // mode and parallelism level
1252		WorkQueue[] workQueues; // main registry
1253	<	final Mutex lock; // for registration
1258	<	final Condition termination; // for awaitTermination
1259	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1253	>	final ForkJoinWorkerThreadFactory factory;
1254		final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1261	–	final AtomicLong stealCount; // collect counts when terminated
1262	–	final AtomicInteger nextWorkerNumber; // to create worker name string
1255		final String workerNamePrefix; // to create worker name string
1256
1257	<	// Creating, registering, and deregistering workers
1257	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
1258	>	volatile Object pad18, pad19, pad1a, pad1b;
1259
1260	<	/**
1261	<	* Tries to create and start a worker
1262	<	*/
1263	<	private void addWorker() {
1264	<	Throwable ex = null;
1265	<	ForkJoinWorkerThread wt = null;
1266	<	try {
1267	<	if ((wt = factory.newThread(this)) != null) {
1268	<	wt.start();
1269	<	return;
1260	>	/*
1261	>	* Acquires the plock lock to protect worker array and related
1262	>	* updates. This method is called only if an initial CAS on plock
1263	>	* fails. This acts as a spinLock for normal cases, but falls back
1264	>	* to builtin monitor to block when (rarely) needed. This would be
1265	>	* a terrible idea for a highly contended lock, but works fine as
1266	>	* a more conservative alternative to a pure spinlock. See
1267	>	* internal ConcurrentHashMap documentation for further
1268	>	* explanation of nearly the same construction.
1269	>	*/
1270	>	private int acquirePlock() {
1271	>	int spins = PL_SPINS, r = 0, ps, nps;
1272	>	for (;;) {
1273	>	if (((ps = plock) & PL_LOCK) == 0 &&
1274	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1275	>	return nps;
1276	>	else if (r == 0) { // randomize spins if possible
1277	>	Thread t = Thread.currentThread(); WorkQueue w; Submitter z;
1278	>	if ((t instanceof ForkJoinWorkerThread) &&
1279	>	(w = ((ForkJoinWorkerThread)t).workQueue) != null)
1280	>	r = w.seed;
1281	>	else if ((z = submitters.get()) != null)
1282	>	r = z.seed;
1283	>	else
1284	>	r = 1;
1285	>	}
1286	>	else if (spins >= 0) {
1287	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1288	>	if (r >= 0)
1289	>	--spins;
1290	>	}
1291	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps \| PL_SIGNAL)) {
1292	>	synchronized (this) {
1293	>	if ((plock & PL_SIGNAL) != 0) {
1294	>	try {
1295	>	wait();
1296	>	} catch (InterruptedException ie) {
1297	>	try {
1298	>	Thread.currentThread().interrupt();
1299	>	} catch (SecurityException ignore) {
1300	>	}
1301	>	}
1302	>	}
1303	>	else
1304	>	notifyAll();
1305	>	}
1306		}
1278	–	} catch (Throwable e) {
1279	–	ex = e;
1307		}
1281	–	deregisterWorker(wt, ex); // adjust counts etc on failure
1308		}
1309
1310		/**
1311	<	* Callback from ForkJoinWorkerThread constructor to assign a
1312	<	* public name. This must be separate from registerWorker because
1287	<	* it is called during the "super" constructor call in
1288	<	* ForkJoinWorkerThread.
1311	>	* Unlocks and signals any thread waiting for plock. Called only
1312	>	* when CAS of seq value for unlock fails.
1313		*/
1314	<	final String nextWorkerName() {
1315	<	return workerNamePrefix.concat
1316	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1314	>	private void releasePlock(int ps) {
1315	>	plock = ps;
1316	>	synchronized (this) { notifyAll(); }
1317		}
1318
1319		/**
1320	<	* Callback from ForkJoinWorkerThread constructor to establish its
1321	<	* poolIndex and record its WorkQueue. To avoid scanning bias due
1322	<	* to packing entries in front of the workQueues array, we treat
1323	<	* the array as a simple power-of-two hash table using per-thread
1324	<	* seed as hash, expanding as needed.
1320	>	* Performs secondary initialization, called when plock is zero.
1321	>	* Creates workQueue array and sets plock to a valid value. The
1322	>	* lock body must be exception-free (so no try/finally) so we
1323	>	* optimistically allocate new array outside the lock and throw
1324	>	* away if (very rarely) not needed. (A similar tactic is used in
1325	>	* fullExternalPush.) Because the plock seq value can eventually
1326	>	* wrap around zero, this method harmlessly fails to reinitialize
1327	>	* if workQueues exists, while still advancing plock.
1328		*
1329	<	* @param w the worker's queue
1329	>	* Additonally tries to create the first worker.
1330	>	*/
1331	>	private void initWorkers() {
1332	>	WorkQueue[] ws, nws; int ps;
1333	>	int p = config & SMASK; // find power of two table size
1334	>	int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots
1335	>	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
1336	>	n = (n + 1) << 1;
1337	>	if ((ws = workQueues) == null \|\| ws.length == 0)
1338	>	nws = new WorkQueue[n];
1339	>	else
1340	>	nws = null;
1341	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1342	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1343	>	ps = acquirePlock();
1344	>	if (((ws = workQueues) == null \|\| ws.length == 0) && nws != null)
1345	>	workQueues = nws;
1346	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1347	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1348	>	releasePlock(nps);
1349	>	tryAddWorker();
1350	>	}
1351	>
1352	>	/**
1353	>	* Tries to create and start one worker. Adjusts counts etc on
1354	>	* failure.
1355		*/
1356	+	private void tryAddWorker() {
1357	+	long c; int u;
1358	+	while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1359	+	(u & SHORT_SIGN) != 0 && (int)c == 0) {
1360	+	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1361	+	((u + UAC_UNIT) & UAC_MASK)) << 32;
1362	+	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1363	+	ForkJoinWorkerThreadFactory fac;
1364	+	Throwable ex = null;
1365	+	ForkJoinWorkerThread wt = null;
1366	+	try {
1367	+	if ((fac = factory) != null &&
1368	+	(wt = fac.newThread(this)) != null) {
1369	+	wt.start();
1370	+	break;
1371	+	}
1372	+	} catch (Throwable e) {
1373	+	ex = e;
1374	+	}
1375	+	deregisterWorker(wt, ex);
1376	+	break;
1377	+	}
1378	+	}
1379	+	}
1380	+
1381	+	// Registering and deregistering workers
1382
1383	<	final void registerWorker(WorkQueue w) {
1384	<	Mutex lock = this.lock;
1385	<	lock.lock();
1383	>	/**
1384	>	* Callback from ForkJoinWorkerThread to establish and record its
1385	>	* WorkQueue. To avoid scanning bias due to packing entries in
1386	>	* front of the workQueues array, we treat the array as a simple
1387	>	* power-of-two hash table using per-thread seed as hash,
1388	>	* expanding as needed.
1389	>	*
1390	>	* @param wt the worker thread
1391	>	* @return the worker's queue
1392	>	*/
1393	>	final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1394	>	Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1395	>	wt.setDaemon(true);
1396	>	if ((handler = ueh) != null)
1397	>	wt.setUncaughtExceptionHandler(handler);
1398	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1399	>	s += SEED_INCREMENT) \|\|
1400	>	s == 0); // skip 0
1401	>	WorkQueue w = new WorkQueue(this, wt, config >>> 16, s);
1402	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1403	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1404	>	ps = acquirePlock();
1405	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1406		try {
1407	<	WorkQueue[] ws = workQueues;
1408	<	if (w != null && ws != null) { // skip on shutdown/failure
1409	<	int rs, n = ws.length, m = n - 1;
1410	<	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1411	<	w.seed = (s == 0) ? 1 : s; // ensure non-zero seed
1412	<	int r = (s << 1) \| 1; // use odd-numbered indices
1315	<	if (ws[r &= m] != null) { // collision
1316	<	int probes = 0; // step by approx half size
1317	<	int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1407	>	if ((ws = workQueues) != null) { // skip if shutting down
1408	>	int n = ws.length, m = n - 1;
1409	>	int r = (s << 1) \| 1; // use odd-numbered indices
1410	>	if (ws[r &= m] != null) { // collision
1411	>	int probes = 0; // step by approx half size
1412	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1413		while (ws[r = (r + step) & m] != null) {
1414		if (++probes >= n) {
1415		workQueues = ws = Arrays.copyOf(ws, n <<= 1);
#	Line 1323 \| Line 1418 \| public class ForkJoinPool extends Abstra
1418		}
1419		}
1420		}
1421	<	w.eventCount = w.poolIndex = r; // establish before recording
1422	<	ws[r] = w; // also update seq
1328	<	runState = ((rs = runState) & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN);
1421	>	w.eventCount = w.poolIndex = r; // volatile write orders
1422	>	ws[r] = w;
1423		}
1424		} finally {
1425	<	lock.unlock();
1425	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1426	>	releasePlock(nps);
1427		}
1428	+	wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex)));
1429	+	return w;
1430		}
1431
1432		/**
1433		* Final callback from terminating worker, as well as upon failure
1434	<	* to construct or start a worker in addWorker. Removes record of
1435	<	* worker from array, and adjusts counts. If pool is shutting
1436	<	* down, tries to complete termination.
1434	>	* to construct or start a worker. Removes record of worker from
1435	>	* array, and adjusts counts. If pool is shutting down, tries to
1436	>	* complete termination.
1437		*
1438	<	* @param wt the worker thread or null if addWorker failed
1438	>	* @param wt the worker thread or null if construction failed
1439		* @param ex the exception causing failure, or null if none
1440		*/
1441		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1345	–	Mutex lock = this.lock;
1442		WorkQueue w = null;
1443		if (wt != null && (w = wt.workQueue) != null) {
1444	<	w.runState = -1; // ensure runState is set
1445	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1446	<	int idx = w.poolIndex;
1447	<	lock.lock();
1448	<	try { // remove record from array
1444	>	int ps;
1445	>	w.qlock = -1; // ensure set
1446	>	long ns = w.nsteals, sc; // collect steal count
1447	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1448	>	sc = stealCount, sc + ns));
1449	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1450	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1451	>	ps = acquirePlock();
1452	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1453	>	try {
1454	>	int idx = w.poolIndex;
1455		WorkQueue[] ws = workQueues;
1456		if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1457		ws[idx] = null;
1458		} finally {
1459	<	lock.unlock();
1459	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1460	>	releasePlock(nps);
1461		}
1462		}
1463
#	Line 1367 \| Line 1470 \| public class ForkJoinPool extends Abstra
1470		if (!tryTerminate(false, false) && w != null) {
1471		w.cancelAll(); // cancel remaining tasks
1472		if (w.array != null) // suppress signal if never ran
1473	<	signalWork(); // wake up or create replacement
1473	>	tryAddWorker(); // create replacement
1474		if (ex == null) // help clean refs on way out
1475		ForkJoinTask.helpExpungeStaleExceptions();
1476		}
1477
1478		if (ex != null) // rethrow
1479	<	U.throwException(ex);
1479	>	ForkJoinTask.rethrow(ex);
1480		}
1481
1379	–
1482		// Submissions
1483
1484		/**
1485		* Unless shutting down, adds the given task to a submission queue
1486		* at submitter's current queue index (modulo submission
1487	<	* range). If no queue exists at the index, one is created. If
1488	<	* the queue is busy, another index is randomly chosen. The
1387	<	* submitMask bounds the effective number of queues to the
1388	<	* (nearest power of two for) parallelism level.
1487	>	* range). Only the most common path is directly handled in this
1488	>	* method. All others are relayed to fullExternalPush.
1489		*
1490		* @param task the task. Caller must ensure non-null.
1491		*/
1492	<	private void doSubmit(ForkJoinTask<?> task) {
1493	<	Submitter s = submitters.get();
1494	<	for (int r = s.seed, m = submitMask;;) {
1495	<	WorkQueue[] ws; WorkQueue q;
1496	<	int k = r & m & SQMASK; // use only even indices
1497	<	if (runState < 0 \|\| (ws = workQueues) == null \|\| ws.length <= k)
1498	<	throw new RejectedExecutionException(); // shutting down
1499	<	else if ((q = ws[k]) == null) { // create new queue
1500	<	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1501	<	Mutex lock = this.lock; // construct outside lock
1502	<	lock.lock();
1503	<	try { // recheck under lock
1504	<	int rs = runState; // to update seq
1505	<	if (ws == workQueues && ws[k] == null) {
1406	<	ws[k] = nq;
1407	<	runState = ((rs & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN));
1408	<	}
1409	<	} finally {
1410	<	lock.unlock();
1411	<	}
1412	<	}
1413	<	else if (q.trySharedPush(task)) {
1414	<	signalWork();
1492	>	final void externalPush(ForkJoinTask<?> task) {
1493	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1494	>	if ((z = submitters.get()) != null && plock > 0 &&
1495	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1496	>	(q = ws[m & z.seed & SQMASK]) != null &&
1497	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1498	>	int b = q.base, s = q.top, n, an;
1499	>	if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) {
1500	>	int j = (((an - 1) & s) << ASHIFT) + ABASE;
1501	>	U.putOrderedObject(a, j, task);
1502	>	q.top = s + 1; // push on to deque
1503	>	q.qlock = 0;
1504	>	if (n <= 2)
1505	>	signalWork(q);
1506		return;
1507		}
1508	<	else if (m > 1) { // move to a different index
1508	>	q.qlock = 0;
1509	>	}
1510	>	fullExternalPush(task);
1511	>	}
1512	>
1513	>	/**
1514	>	* Full version of externalPush. This method is called, among
1515	>	* other times, upon the first submission of the first task to the
1516	>	* pool, so must perform secondary initialization (via
1517	>	* initWorkers). It also detects first submission by an external
1518	>	* thread by looking up its ThreadLocal, and creates a new shared
1519	>	* queue if the one at index if empty or contended. The plock lock
1520	>	* body must be exception-free (so no try/finally) so we
1521	>	* optimistically allocate new queues outside the lock and throw
1522	>	* them away if (very rarely) not needed.
1523	>	*/
1524	>	private void fullExternalPush(ForkJoinTask<?> task) {
1525	>	int r = 0; // random index seed
1526	>	for (Submitter z = submitters.get();;) {
1527	>	WorkQueue[] ws; WorkQueue q; int ps, m, k;
1528	>	if (z == null) {
1529	>	if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1530	>	r += SEED_INCREMENT) && r != 0)
1531	>	submitters.set(z = new Submitter(r));
1532	>	}
1533	>	else if (r == 0) { // move to a different index
1534	>	r = z.seed;
1535		r ^= r << 13; // same xorshift as WorkQueues
1536		r ^= r >>> 17;
1537	<	s.seed = r ^= r << 5;
1537	>	z.seed = r ^ (r << 5);
1538	>	}
1539	>	else if ((ps = plock) < 0)
1540	>	throw new RejectedExecutionException();
1541	>	else if (ps == 0 \|\| (ws = workQueues) == null \|\|
1542	>	(m = ws.length - 1) < 0)
1543	>	initWorkers();
1544	>	else if ((q = ws[k = r & m & SQMASK]) != null) {
1545	>	if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
1546	>	ForkJoinTask<?>[] a = q.array;
1547	>	int s = q.top;
1548	>	boolean submitted = false;
1549	>	try { // locked version of push
1550	>	if ((a != null && a.length > s + 1 - q.base) \|\|
1551	>	(a = q.growArray()) != null) { // must presize
1552	>	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
1553	>	U.putOrderedObject(a, j, task);
1554	>	q.top = s + 1;
1555	>	submitted = true;
1556	>	}
1557	>	} finally {
1558	>	q.qlock = 0; // unlock
1559	>	}
1560	>	if (submitted) {
1561	>	signalWork(q);
1562	>	return;
1563	>	}
1564	>	}
1565	>	r = 0; // move on failure
1566	>	}
1567	>	else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1568	>	q = new WorkQueue(this, null, SHARED_QUEUE, r);
1569	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1570	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1571	>	ps = acquirePlock();
1572	>	if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1573	>	ws[k] = q;
1574	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1575	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1576	>	releasePlock(nps);
1577		}
1578		else
1579	<	Thread.yield(); // yield if no alternatives
1579	>	r = 0; // try elsewhere while lock held
1580		}
1581		}
1582
#	Line 1435 \| Line 1591 \| public class ForkJoinPool extends Abstra
1591		}
1592
1593		/**
1594	<	* Tries to activate or create a worker if too few are active.
1594	>	* Tries to create or activate a worker if too few are active.
1595	>	*
1596	>	* @param q the (non-null) queue holding tasks to be signalled
1597		*/
1598	<	final void signalWork() {
1599	<	long c; int u;
1600	<	while ((u = (int)((c = ctl) >>> 32)) < 0) { // too few active
1601	<	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1602	<	if ((e = (int)c) > 0) { // at least one waiting
1603	<	if (ws != null && (i = e & SMASK) < ws.length &&
1598	>	final void signalWork(WorkQueue q) {
1599	>	int hint = q.poolIndex;
1600	>	long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p;
1601	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1602	>	if ((e = (int)c) > 0) {
1603	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1604		(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1605		long nc = (((long)(w.nextWait & E_MASK)) \|
1606		((long)(u + UAC_UNIT) << 32));
1607		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1608	+	w.hint = hint;
1609		w.eventCount = (e + E_SEQ) & E_MASK;
1610		if ((p = w.parker) != null)
1611	<	U.unpark(p); // activate and release
1611	>	U.unpark(p);
1612		break;
1613		}
1614	+	if (q.top - q.base <= 0)
1615	+	break;
1616		}
1617		else
1618		break;
1619		}
1620	<	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1621	<	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1622	<	((u + UAC_UNIT) & UAC_MASK)) << 32;
1462	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1463	<	addWorker();
1464	<	break;
1465	<	}
1466	<	}
1467	<	else
1620	>	else {
1621	>	if ((short)u < 0)
1622	>	tryAddWorker();
1623		break;
1624	+	}
1625		}
1626		}
1627
#	Line 1475 \| Line 1631 \| public class ForkJoinPool extends Abstra
1631		* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1632		*/
1633		final void runWorker(WorkQueue w) {
1634	<	w.growArray(false); // initialize queue array in this thread
1635	<	do { w.runTask(scan(w)); } while (w.runState >= 0);
1634	>	w.growArray(); // allocate queue
1635	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1636		}
1637
1638		/**
#	Line 1487 \| Line 1643 \| public class ForkJoinPool extends Abstra
1643		* contention, or state changes that indicate possible success on
1644		* re-invocation.
1645		*
1646	<	* The scan searches for tasks across a random permutation of
1647	<	* queues (starting at a random index and stepping by a random
1648	<	* relative prime, checking each at least once). The scan
1649	<	* terminates upon either finding a non-empty queue, or completing
1650	<	* the sweep. If the worker is not inactivated, it takes and
1651	<	* returns a task from this queue. On failure to find a task, we
1652	<	* take one of the following actions, after which the caller will
1653	<	* retry calling this method unless terminated.
1646	>	* The scan searches for tasks across queues (starting at a random
1647	>	* index, and relying on registerWorker to irregularly scatter
1648	>	* them within array to avoid bias), checking each at least twice.
1649	>	* The scan terminates upon either finding a non-empty queue, or
1650	>	* completing the sweep. If the worker is not inactivated, it
1651	>	* takes and returns a task from this queue. Otherwise, if not
1652	>	* activated, it signals workers (that may include itself) and
1653	>	* returns so caller can retry. Also returns for true if the
1654	>	* worker array may have changed during an empty scan. On failure
1655	>	* to find a task, we take one of the following actions, after
1656	>	* which the caller will retry calling this method unless
1657	>	* terminated.
1658		*
1659		* * If pool is terminating, terminate the worker.
1660		*
1501	–	* * If not a complete sweep, try to release a waiting worker. If
1502	–	* the scan terminated because the worker is inactivated, then the
1503	–	* released worker will often be the calling worker, and it can
1504	–	* succeed obtaining a task on the next call. Or maybe it is
1505	–	* another worker, but with same net effect. Releasing in other
1506	–	* cases as well ensures that we have enough workers running.
1507	–	*
1661		* * If not already enqueued, try to inactivate and enqueue the
1662		* worker on wait queue. Or, if inactivating has caused the pool
1663		* to be quiescent, relay to idleAwaitWork to check for
1664		* termination and possibly shrink pool.
1665		*
1666	<	* * If already inactive, and the caller has run a task since the
1667	<	* last empty scan, return (to allow rescan) unless others are
1668	<	* also inactivated. Field WorkQueue.rescans counts down on each
1516	<	* scan to ensure eventual inactivation and blocking.
1517	<	*
1518	<	* * If already enqueued and none of the above apply, park
1519	<	* awaiting signal,
1666	>	* * If already enqueued and none of the above apply, possibly
1667	>	* (with 1/2 probability) park awaiting signal, else lingering to
1668	>	* help scan and signal.
1669		*
1670		* @param w the worker (via its WorkQueue)
1671	<	* @return a task or null of none found
1671	>	* @return a task or null if none found
1672		*/
1673		private final ForkJoinTask<?> scan(WorkQueue w) {
1674	<	WorkQueue[] ws; // first update random seed
1675	<	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1676	<	int rs = runState, m; // volatile read order matters
1677	<	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1678	<	int ec = w.eventCount; // ec is negative if inactive
1679	<	int step = (r >>> 16) \| 1; // relative prime
1680	<	for (int j = (m + 1) << 2; ; r += step) {
1681	<	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1682	<	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1683	<	(a = q.array) != null) { // probably nonempty
1674	>	WorkQueue[] ws; int m;
1675	>	int ps = plock; // read plock before ws
1676	>	if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1677	>	int ec = w.eventCount; // ec is negative if inactive
1678	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1679	>	int j = ((m + m + 1) \| MIN_SCAN) & MAX_SCAN;
1680	>	do {
1681	>	WorkQueue q; ForkJoinTask<?>[] a; int b;
1682	>	if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 &&
1683	>	(a = q.array) != null) { // probably nonempty
1684		int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1685	<	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1685	>	ForkJoinTask<?> t = (ForkJoinTask<?>)
1686	>	U.getObjectVolatile(a, i);
1687		if (q.base == b && ec >= 0 && t != null &&
1688		U.compareAndSwapObject(a, i, t, null)) {
1689	<	if (q.top - (q.base = b + 1) > 1)
1690	<	signalWork(); // help pushes signal
1691	<	return t;
1692	<	}
1693	<	else if (ec < 0 \|\| j <= m) {
1694	<	rs = 0; // mark scan as imcomplete
1695	<	break; // caller can retry after release
1696	<	}
1697	<	}
1698	<	if (--j < 0)
1699	<	break;
1700	<	}
1701	<
1702	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1703	<	if (e < 0) // decode ctl on empty scan
1704	<	w.runState = -1; // pool is terminating
1705	<	else if (rs == 0 \|\| rs != runState) { // incomplete scan
1706	<	WorkQueue v; Thread p; // try to release a waiter
1707	<	if (e > 0 && a < 0 && w.eventCount == ec &&
1708	<	(v = ws[e & m]) != null && v.eventCount == (e \| INT_SIGN)) {
1709	<	long nc = ((long)(v.nextWait & E_MASK) \|
1710	<	((c + AC_UNIT) & (AC_MASK\|TC_MASK)));
1711	<	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1712	<	v.eventCount = (e + E_SEQ) & E_MASK;
1713	<	if ((p = v.parker) != null)
1714	<	U.unpark(p);
1715	<	}
1716	<	}
1717	<	}
1568	<	else if (ec >= 0) { // try to enqueue/inactivate
1569	<	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1570	<	w.nextWait = e;
1571	<	w.eventCount = ec \| INT_SIGN; // mark as inactive
1572	<	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1573	<	w.eventCount = ec; // unmark on CAS failure
1574	<	else {
1575	<	if ((ns = w.nsteals) != 0) {
1576	<	w.nsteals = 0; // set rescans if ran task
1577	<	w.rescans = (a > 0) ? 0 : a + parallelism;
1578	<	w.totalSteals += ns;
1579	<	}
1580	<	if (a == 1 - parallelism) // quiescent
1689	>	if ((q.base = b + 1) - q.top < 0)
1690	>	signalWork(q);
1691	>	return t; // taken
1692	>	}
1693	>	else if ((ec < 0 \|\| j < m) && (int)(ctl >> AC_SHIFT) <= 0) {
1694	>	w.hint = (r + j) & m; // help signal below
1695	>	break; // cannot take
1696	>	}
1697	>	}
1698	>	} while (--j >= 0);
1699	>
1700	>	long c, sc; int e, ns, h;
1701	>	if ((h = w.hint) < 0) {
1702	>	if ((ns = w.nsteals) != 0) {
1703	>	if (U.compareAndSwapLong(this, STEALCOUNT,
1704	>	sc = stealCount, sc + ns))
1705	>	w.nsteals = 0; // collect steals
1706	>	}
1707	>	else if (plock != ps) // consistency check
1708	>	; // skip
1709	>	else if ((e = (int)(c = ctl)) < 0)
1710	>	w.qlock = -1; // pool is terminating
1711	>	else if (ec >= 0) { // try to enqueue/inactivate
1712	>	long nc = ((long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK)));
1713	>	w.nextWait = e; // link and mark inactive
1714	>	w.eventCount = ec \| INT_SIGN;
1715	>	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1716	>	w.eventCount = ec; // unmark on CAS failure
1717	>	else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK))
1718		idleAwaitWork(w, nc, c);
1719		}
1720	<	}
1584	<	else if (w.eventCount < 0) { // already queued
1585	<	if ((nr = w.rescans) > 0) { // continue rescanning
1586	<	int ac = a + parallelism;
1587	<	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0)
1588	<	Thread.yield(); // yield before block
1589	<	}
1590	<	else {
1591	<	Thread.interrupted(); // clear status
1720	>	else if (w.eventCount < 0) { // block
1721		Thread wt = Thread.currentThread();
1722	+	Thread.interrupted(); // clear status
1723		U.putObject(wt, PARKBLOCKER, this);
1724	<	w.parker = wt; // emulate LockSupport.park
1725	<	if (w.eventCount < 0) // recheck
1724	>	w.parker = wt; // emulate LockSupport.park
1725	>	if (w.eventCount < 0) // recheck
1726		U.park(false, 0L);
1727		w.parker = null;
1728		U.putObject(wt, PARKBLOCKER, null);
1729		}
1730		}
1731	+	if (h >= 0 \|\| (h = w.hint) >= 0) { // signal others before retry
1732	+	w.hint = -1; // reset
1733	+	helpSignal(null, h, true);
1734	+	}
1735		}
1736		return null;
1737		}
#	Line 1605 \| Line 1739 \| public class ForkJoinPool extends Abstra
1739		/**
1740		* If inactivating worker w has caused the pool to become
1741		* quiescent, checks for pool termination, and, so long as this is
1742	<	* not the only worker, waits for event for up to SHRINK_RATE
1743	<	* nanosecs. On timeout, if ctl has not changed, terminates the
1742	>	* not the only worker, waits for event for up to a given
1743	>	* duration. On timeout, if ctl has not changed, terminates the
1744		* worker, which will in turn wake up another worker to possibly
1745		* repeat this process.
1746		*
#	Line 1615 \| Line 1749 \| public class ForkJoinPool extends Abstra
1749		* @param prevCtl the ctl value to restore if thread is terminated
1750		*/
1751		private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1752	<	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1753	<	(int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1752	>	if (w != null && w.eventCount < 0 &&
1753	>	!tryTerminate(false, false) && (int)prevCtl != 0) {
1754	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1755	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1756	>	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1757		Thread wt = Thread.currentThread();
1621	–	Thread.yield(); // yield before block
1758		while (ctl == currentCtl) {
1623	–	long startTime = System.nanoTime();
1759		Thread.interrupted(); // timed variant of version in scan()
1760		U.putObject(wt, PARKBLOCKER, this);
1761		w.parker = wt;
1762		if (ctl == currentCtl)
1763	<	U.park(false, SHRINK_RATE);
1763	>	U.park(false, parkTime);
1764		w.parker = null;
1765		U.putObject(wt, PARKBLOCKER, null);
1766		if (ctl != currentCtl)
1767		break;
1768	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1768	>	if (deadline - System.nanoTime() <= 0L &&
1769		U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1770		w.eventCount = (w.eventCount + E_SEQ) \| E_MASK;
1771	<	w.runState = -1; // shrink
1771	>	w.qlock = -1; // shrink
1772		break;
1773		}
1774		}
#	Line 1641 \| Line 1776 \| public class ForkJoinPool extends Abstra
1776		}
1777
1778		/**
1779	+	* Scans through queues looking for work (optionally, while
1780	+	* joining a task); if any present, signals. May return early if
1781	+	* more signalling is detectably unneeded.
1782	+	*
1783	+	* @param task if non-null, return early if done
1784	+	* @param origin an index to start scan
1785	+	* @param once if only the origin should be checked
1786	+	*/
1787	+	private void helpSignal(ForkJoinTask<?> task, int origin, boolean once) {
1788	+	WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s;
1789	+	if ((u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 &&
1790	+	(ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1791	+	outer: for (int k = origin, j = once ? 0 : m; j >= 0; --j) {
1792	+	WorkQueue q = ws[k++ & m];
1793	+	for (int n = m;;) { // limit to at most m signals
1794	+	if (task != null && task.status < 0)
1795	+	break outer;
1796	+	if (q == null \|\|
1797	+	((s = (task == null ? -1 : 0) - q.base + q.top) <= n &&
1798	+	(n = s) <= 0))
1799	+	break;
1800	+	if ((u = (int)((c = ctl) >>> 32)) >= 0 \|\|
1801	+	(e = (int)c) <= 0 \|\| m < (i = e & SMASK) \|\|
1802	+	(w = ws[i]) == null)
1803	+	break outer;
1804	+	long nc = (((long)(w.nextWait & E_MASK)) \|
1805	+	((long)(u + UAC_UNIT) << 32));
1806	+	if (w.eventCount == (e \| INT_SIGN) &&
1807	+	U.compareAndSwapLong(this, CTL, c, nc)) {
1808	+	w.eventCount = (e + E_SEQ) & E_MASK;
1809	+	if ((p = w.parker) != null)
1810	+	U.unpark(p);
1811	+	if (--n <= 0)
1812	+	break;
1813	+	}
1814	+	}
1815	+	}
1816	+	}
1817	+	}
1818	+
1819	+	/**
1820		* Tries to locate and execute tasks for a stealer of the given
1821		* task, or in turn one of its stealers, Traces currentSteal ->
1822		* currentJoin links looking for a thread working on a descendant
#	Line 1671 \| Line 1847 \| public class ForkJoinPool extends Abstra
1847		}
1848		if ((ws = workQueues) == null \|\| (m = ws.length - 1) <= 0)
1849		break restart; // shutting down
1850	<	if ((v = ws[h = (j.stealHint \| 1) & m]) == null \|\|
1850	>	if ((v = ws[h = (j.hint \| 1) & m]) == null \|\|
1851		v.currentSteal != subtask) {
1852		for (int origin = h;;) { // find stealer
1853		if (((h = (h + 2) & m) & 15) == 1 &&
#	Line 1679 \| Line 1855 \| public class ForkJoinPool extends Abstra
1855		continue restart; // occasional staleness check
1856		if ((v = ws[h]) != null &&
1857		v.currentSteal == subtask) {
1858	<	j.stealHint = h; // save hint
1858	>	j.hint = h; // save hint
1859		break;
1860		}
1861		if (h == origin)
#	Line 1727 \| Line 1903 \| public class ForkJoinPool extends Abstra
1903		}
1904
1905		/**
1906	<	* If task is at base of some steal queue, steals and executes it.
1906	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1907	>	* and run tasks within the target's computation.
1908	>	*
1909	>	* @param task the task to join
1910	>	* @param mode if shared, exit upon completing any task
1911	>	* if all workers are active
1912		*
1732	–	* @param joiner the joining worker
1733	–	* @param task the task
1913		*/
1914	<	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1915	<	WorkQueue[] ws;
1916	<	if ((ws = workQueues) != null) {
1917	<	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1918	<	WorkQueue q = ws[j];
1919	<	if (q != null && q.pollFor(task)) {
1920	<	joiner.runSubtask(task);
1921	<	break;
1914	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1915	>	WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1916	>	if (task != null && (ws = workQueues) != null &&
1917	>	(m = ws.length - 1) >= 0) {
1918	>	for (int j = 1, origin = j;;) {
1919	>	if ((s = task.status) < 0)
1920	>	return s;
1921	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1922	>	origin = j;
1923	>	if (mode == SHARED_QUEUE &&
1924	>	((u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0))
1925	>	break;
1926		}
1927	+	else if ((j = (j + 2) & m) == origin)
1928	+	break;
1929		}
1930		}
1931	+	return 0;
1932		}
1933
1934		/**
1935		* Tries to decrement active count (sometimes implicitly) and
1936		* possibly release or create a compensating worker in preparation
1937		* for blocking. Fails on contention or termination. Otherwise,
1938	<	* adds a new thread if no idle workers are available and either
1939	<	* pool would become completely starved or: (at least half
1754	<	* starved, and fewer than 50% spares exist, and there is at least
1755	<	* one task apparently available). Even though the availability
1756	<	* check requires a full scan, it is worthwhile in reducing false
1757	<	* alarms.
1758	<	*
1759	<	* @param task if non-null, a task being waited for
1760	<	* @param blocker if non-null, a blocker being waited for
1761	<	* @return true if the caller can block, else should recheck and retry
1938	>	* adds a new thread if no idle workers are available and pool
1939	>	* may become starved.
1940		*/
1941	<	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1942	<	int pc = parallelism, e;
1943	<	long c = ctl;
1944	<	WorkQueue[] ws = workQueues;
1945	<	if ((e = (int)c) >= 0 && ws != null) {
1946	<	int u, a, ac, hc;
1947	<	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1948	<	boolean replace = false;
1949	<	if ((a = u >> UAC_SHIFT) <= 0) {
1950	<	if ((ac = a + pc) <= 1)
1951	<	replace = true;
1952	<	else if ((e > 0 \|\| (task != null &&
1953	<	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1776	<	WorkQueue w;
1777	<	for (int j = 0; j < ws.length; ++j) {
1778	<	if ((w = ws[j]) != null && !w.isEmpty()) {
1779	<	replace = true;
1780	<	break; // in compensation range and tasks available
1781	<	}
1782	<	}
1941	>	final boolean tryCompensate() {
1942	>	int pc = config & SMASK, e, i, tc; long c;
1943	>	WorkQueue[] ws; WorkQueue w; Thread p;
1944	>	if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) {
1945	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1946	>	(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1947	>	long nc = ((long)(w.nextWait & E_MASK) \|
1948	>	(c & (AC_MASK\|TC_MASK)));
1949	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1950	>	w.eventCount = (e + E_SEQ) & E_MASK;
1951	>	if ((p = w.parker) != null)
1952	>	U.unpark(p);
1953	>	return true; // replace with idle worker
1954		}
1955		}
1956	<	if ((task == null \|\| task.status >= 0) && // recheck need to block
1957	<	(blocker == null \|\| !blocker.isReleasable()) && ctl == c) {
1958	<	if (!replace) { // no compensation
1959	<	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1960	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1961	<	return true;
1962	<	}
1963	<	else if (e != 0) { // release an idle worker
1964	<	WorkQueue w; Thread p; int i;
1965	<	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
1966	<	long nc = ((long)(w.nextWait & E_MASK) \|
1967	<	(c & (AC_MASK\|TC_MASK)));
1968	<	if (w.eventCount == (e \| INT_SIGN) &&
1969	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1970	<	w.eventCount = (e + E_SEQ) & E_MASK;
1971	<	if ((p = w.parker) != null)
1801	<	U.unpark(p);
1956	>	else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1957	>	(int)(c >> AC_SHIFT) + pc > 1) {
1958	>	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1959	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1960	>	return true; // no compensation
1961	>	}
1962	>	else if (tc + pc < MAX_CAP) {
1963	>	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1964	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1965	>	ForkJoinWorkerThreadFactory fac;
1966	>	Throwable ex = null;
1967	>	ForkJoinWorkerThread wt = null;
1968	>	try {
1969	>	if ((fac = factory) != null &&
1970	>	(wt = fac.newThread(this)) != null) {
1971	>	wt.start();
1972		return true;
1973		}
1974	+	} catch (Throwable rex) {
1975	+	ex = rex;
1976		}
1977	<	}
1806	<	else if (tc < MAX_CAP) { // create replacement
1807	<	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1808	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1809	<	addWorker();
1810	<	return true;
1811	<	}
1977	>	deregisterWorker(wt, ex); // clean up and return false
1978		}
1979		}
1980		}
#	Line 1823 \| Line 1989 \| public class ForkJoinPool extends Abstra
1989		* @return task status on exit
1990		*/
1991		final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1992	<	int s;
1993	<	if ((s = task.status) >= 0) {
1992	>	int s = 0;
1993	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1994		ForkJoinTask<?> prevJoin = joiner.currentJoin;
1995		joiner.currentJoin = task;
1996	<	long startTime = 0L;
1997	<	for (int k = 0;;) {
1998	<	if ((s = (joiner.isEmpty() ? // try to help
1999	<	tryHelpStealer(joiner, task) :
2000	<	joiner.tryRemoveAndExec(task))) == 0 &&
1996	>	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
1997	>	joiner.tryRemoveAndExec(task)); // process local tasks
1998	>	if (s >= 0 && (s = task.status) >= 0) {
1999	>	helpSignal(task, joiner.poolIndex, false);
2000	>	if ((s = task.status) >= 0 &&
2001	>	(task instanceof CountedCompleter))
2002	>	s = helpComplete(task, LIFO_QUEUE);
2003	>	}
2004	>	while (s >= 0 && (s = task.status) >= 0) {
2005	>	if ((!joiner.isEmpty() \|\| // try helping
2006	>	(s = tryHelpStealer(joiner, task)) == 0) &&
2007		(s = task.status) >= 0) {
2008	<	if (k == 0) {
2009	<	startTime = System.nanoTime();
2010	<	tryPollForAndExec(joiner, task); // check uncommon case
1839	<	}
1840	<	else if ((k & (MAX_HELP - 1)) == 0 &&
1841	<	System.nanoTime() - startTime >=
1842	<	COMPENSATION_DELAY &&
1843	<	tryCompensate(task, null)) {
1844	<	if (task.trySetSignal()) {
2008	>	helpSignal(task, joiner.poolIndex, false);
2009	>	if ((s = task.status) >= 0 && tryCompensate()) {
2010	>	if (task.trySetSignal() && (s = task.status) >= 0) {
2011		synchronized (task) {
2012		if (task.status >= 0) {
2013		try { // see ForkJoinTask
#	Line 1858 \| Line 2024 \| public class ForkJoinPool extends Abstra
2024		(this, CTL, c = ctl, c + AC_UNIT));
2025		}
2026		}
1861	–	if (s < 0 \|\| (s = task.status) < 0) {
1862	–	joiner.currentJoin = prevJoin;
1863	–	break;
1864	–	}
1865	–	else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
1866	–	Thread.yield(); // for politeness
2027		}
2028	+	joiner.currentJoin = prevJoin;
2029		}
2030		return s;
2031		}
#	Line 1876 \| Line 2037 \| public class ForkJoinPool extends Abstra
2037		*
2038		* @param joiner the joining worker
2039		* @param task the task
1879	–	* @return task status on exit
2040		*/
2041	<	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2041	>	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2042		int s;
2043	<	while ((s = task.status) >= 0 &&
2044	<	(joiner.isEmpty() ?
2045	<	tryHelpStealer(joiner, task) :
2046	<	joiner.tryRemoveAndExec(task)) != 0)
2047	<	;
2048	<	return s;
2043	>	if (joiner != null && task != null && (s = task.status) >= 0) {
2044	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2045	>	joiner.currentJoin = task;
2046	>	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2047	>	joiner.tryRemoveAndExec(task));
2048	>	if (s >= 0 && (s = task.status) >= 0) {
2049	>	helpSignal(task, joiner.poolIndex, false);
2050	>	if ((s = task.status) >= 0 &&
2051	>	(task instanceof CountedCompleter))
2052	>	s = helpComplete(task, LIFO_QUEUE);
2053	>	}
2054	>	if (s >= 0 && joiner.isEmpty()) {
2055	>	do {} while (task.status >= 0 &&
2056	>	tryHelpStealer(joiner, task) > 0);
2057	>	}
2058	>	joiner.currentJoin = prevJoin;
2059	>	}
2060		}
2061
2062		/**
#	Line 1893 \| Line 2064 \| public class ForkJoinPool extends Abstra
2064		* during a random, then cyclic scan, else null. This method must
2065		* be retried by caller if, by the time it tries to use the queue,
2066		* it is empty.
2067	+	* @param r a (random) seed for scanning
2068		*/
2069	<	private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
1898	<	// Similar to loop in scan(), but ignoring submissions
1899	<	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1900	<	int step = (r >>> 16) \| 1;
2069	>	private WorkQueue findNonEmptyStealQueue(int r) {
2070		for (WorkQueue[] ws;;) {
2071	<	int rs = runState, m;
2071	>	int ps = plock, m, n;
2072		if ((ws = workQueues) == null \|\| (m = ws.length - 1) < 1)
2073		return null;
2074	<	for (int j = (m + 1) << 2; ; r += step) {
2075	<	WorkQueue q = ws[((r << 1) \| 1) & m];
2076	<	if (q != null && !q.isEmpty())
2074	>	for (int j = (m + 1) << 2; ;) {
2075	>	WorkQueue q = ws[(((r + j) << 1) \| 1) & m];
2076	>	if (q != null && (n = q.base - q.top) < 0) {
2077	>	if (n < -1)
2078	>	signalWork(q);
2079		return q;
2080	+	}
2081		else if (--j < 0) {
2082	<	if (runState == rs)
2082	>	if (plock == ps)
2083		return null;
2084		break;
2085		}
#	Line 1915 \| Line 2087 \| public class ForkJoinPool extends Abstra
2087		}
2088		}
2089
1918	–
2090		/**
2091		* Runs tasks until {@code isQuiescent()}. We piggyback on
2092		* active count ctl maintenance, but rather than blocking
#	Line 1927 \| Line 2098 \| public class ForkJoinPool extends Abstra
2098		ForkJoinTask<?> localTask; // exhaust local queue
2099		while ((localTask = w.nextLocalTask()) != null)
2100		localTask.doExec();
2101	<	WorkQueue q = findNonEmptyStealQueue(w);
2101	>	// Similar to loop in scan(), but ignoring submissions
2102	>	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2103		if (q != null) {
2104		ForkJoinTask<?> t; int b;
2105		if (!active) { // re-establish active count
#	Line 1948 \| Line 2120 \| public class ForkJoinPool extends Abstra
2120		}
2121		else
2122		c = ctl; // re-increment on exit
2123	<	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2123	>	if ((int)(c >> AC_SHIFT) + (config & SMASK) == 0) {
2124		do {} while (!U.compareAndSwapLong
2125		(this, CTL, c = ctl, c + AC_UNIT));
2126		break;
#	Line 1967 \| Line 2139 \| public class ForkJoinPool extends Abstra
2139		WorkQueue q; int b;
2140		if ((t = w.nextLocalTask()) != null)
2141		return t;
2142	<	if ((q = findNonEmptyStealQueue(w)) == null)
2142	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2143		return null;
2144		if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2145		return t;
#	Line 1975 \| Line 2147 \| public class ForkJoinPool extends Abstra
2147		}
2148
2149		/**
2150	<	* Returns the approximate (non-atomic) number of idle threads per
2151	<	* active thread to offset steal queue size for method
2152	<	* ForkJoinTask.getSurplusQueuedTaskCount().
2153	<	*/
2154	<	final int idlePerActive() {
2155	<	// Approximate at powers of two for small values, saturate past 4
2156	<	int p = parallelism;
2157	<	int a = p + (int)(ctl >> AC_SHIFT);
2158	<	return (a > (p >>>= 1) ? 0 :
2159	<	a > (p >>>= 1) ? 1 :
2160	<	a > (p >>>= 1) ? 2 :
2161	<	a > (p >>>= 1) ? 4 :
2162	<	8);
2150	>	* Returns a cheap heuristic guide for task partitioning when
2151	>	* programmers, frameworks, tools, or languages have little or no
2152	>	* idea about task granularity. In essence by offering this
2153	>	* method, we ask users only about tradeoffs in overhead vs
2154	>	* expected throughput and its variance, rather than how finely to
2155	>	* partition tasks.
2156	>	*
2157	>	* In a steady state strict (tree-structured) computation, each
2158	>	* thread makes available for stealing enough tasks for other
2159	>	* threads to remain active. Inductively, if all threads play by
2160	>	* the same rules, each thread should make available only a
2161	>	* constant number of tasks.
2162	>	*
2163	>	* The minimum useful constant is just 1. But using a value of 1
2164	>	* would require immediate replenishment upon each steal to
2165	>	* maintain enough tasks, which is infeasible. Further,
2166	>	* partitionings/granularities of offered tasks should minimize
2167	>	* steal rates, which in general means that threads nearer the top
2168	>	* of computation tree should generate more than those nearer the
2169	>	* bottom. In perfect steady state, each thread is at
2170	>	* approximately the same level of computation tree. However,
2171	>	* producing extra tasks amortizes the uncertainty of progress and
2172	>	* diffusion assumptions.
2173	>	*
2174	>	* So, users will want to use values larger, but not much larger
2175	>	* than 1 to both smooth over transient shortages and hedge
2176	>	* against uneven progress; as traded off against the cost of
2177	>	* extra task overhead. We leave the user to pick a threshold
2178	>	* value to compare with the results of this call to guide
2179	>	* decisions, but recommend values such as 3.
2180	>	*
2181	>	* When all threads are active, it is on average OK to estimate
2182	>	* surplus strictly locally. In steady-state, if one thread is
2183	>	* maintaining say 2 surplus tasks, then so are others. So we can
2184	>	* just use estimated queue length. However, this strategy alone
2185	>	* leads to serious mis-estimates in some non-steady-state
2186	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2187	>	* many of these by further considering the number of "idle"
2188	>	* threads, that are known to have zero queued tasks, so
2189	>	* compensate by a factor of (#idle/#active) threads.
2190	>	*
2191	>	* Note: The approximation of #busy workers as #active workers is
2192	>	* not very good under current signalling scheme, and should be
2193	>	* improved.
2194	>	*/
2195	>	static int getSurplusQueuedTaskCount() {
2196	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2197	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2198	>	int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK;
2199	>	int n = (q = wt.workQueue).top - q.base;
2200	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2201	>	return n - (a > (p >>>= 1) ? 0 :
2202	>	a > (p >>>= 1) ? 1 :
2203	>	a > (p >>>= 1) ? 2 :
2204	>	a > (p >>>= 1) ? 4 :
2205	>	8);
2206	>	}
2207	>	return 0;
2208		}
2209
2210		// Termination
#	Line 2007 \| Line 2224 \| public class ForkJoinPool extends Abstra
2224		* @return true if now terminating or terminated
2225		*/
2226		private boolean tryTerminate(boolean now, boolean enable) {
2227	<	Mutex lock = this.lock;
2227	>	if (this == commonPool) // cannot shut down
2228	>	return false;
2229		for (long c;;) {
2230		if (((c = ctl) & STOP_BIT) != 0) { // already terminating
2231	<	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2232	<	lock.lock(); // don't need try/finally
2233	<	termination.signalAll(); // signal when 0 workers
2234	<	lock.unlock();
2231	>	if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) {
2232	>	synchronized (this) {
2233	>	notifyAll(); // signal when 0 workers
2234	>	}
2235		}
2236		return true;
2237		}
2238	<	if (runState >= 0) { // not yet enabled
2238	>	if (plock >= 0) { // not yet enabled
2239	>	int ps;
2240		if (!enable)
2241		return false;
2242	<	lock.lock();
2243	<	runState \|= SHUTDOWN;
2244	<	lock.unlock();
2242	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
2243	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2244	>	ps = acquirePlock();
2245	>	int nps = SHUTDOWN;
2246	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2247	>	releasePlock(nps);
2248		}
2249		if (!now) { // check if idle & no tasks
2250	<	if ((int)(c >> AC_SHIFT) != -parallelism \|\|
2250	>	if ((int)(c >> AC_SHIFT) != -(config & SMASK) \|\|
2251		hasQueuedSubmissions())
2252		return false;
2253		// Check for unqueued inactive workers. One pass suffices.
#	Line 2045 \| Line 2267 \| public class ForkJoinPool extends Abstra
2267		int n = ws.length;
2268		for (int i = 0; i < n; ++i) {
2269		if ((w = ws[i]) != null) {
2270	<	w.runState = -1;
2270	>	w.qlock = -1;
2271		if (pass > 0) {
2272		w.cancelAll();
2273		if (pass > 1)
#	Line 2064 \| Line 2286 \| public class ForkJoinPool extends Abstra
2286		if (w.eventCount == (e \| INT_SIGN) &&
2287		U.compareAndSwapLong(this, CTL, cc, nc)) {
2288		w.eventCount = (e + E_SEQ) & E_MASK;
2289	<	w.runState = -1;
2289	>	w.qlock = -1;
2290		if ((p = w.parker) != null)
2291		U.unpark(p);
2292		}
#	Line 2075 \| Line 2297 \| public class ForkJoinPool extends Abstra
2297		}
2298		}
2299
2300	+	// external operations on common pool
2301	+
2302	+	/**
2303	+	* Returns common pool queue for a thread that has submitted at
2304	+	* least one task.
2305	+	*/
2306	+	static WorkQueue commonSubmitterQueue() {
2307	+	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2308	+	return ((z = submitters.get()) != null &&
2309	+	(p = commonPool) != null &&
2310	+	(ws = p.workQueues) != null &&
2311	+	(m = ws.length - 1) >= 0) ?
2312	+	ws[m & z.seed & SQMASK] : null;
2313	+	}
2314	+
2315	+	/**
2316	+	* Tries to pop the given task from submitter's queue in common pool.
2317	+	*/
2318	+	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2319	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2320	+	ForkJoinTask<?>[] a; int m, s;
2321	+	if (t != null &&
2322	+	(z = submitters.get()) != null &&
2323	+	(p = commonPool) != null &&
2324	+	(ws = p.workQueues) != null &&
2325	+	(m = ws.length - 1) >= 0 &&
2326	+	(q = ws[m & z.seed & SQMASK]) != null &&
2327	+	(s = q.top) != q.base &&
2328	+	(a = q.array) != null) {
2329	+	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2330	+	if (U.getObject(a, j) == t &&
2331	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2332	+	if (q.array == a && q.top == s && // recheck
2333	+	U.compareAndSwapObject(a, j, t, null)) {
2334	+	q.top = s - 1;
2335	+	q.qlock = 0;
2336	+	return true;
2337	+	}
2338	+	q.qlock = 0;
2339	+	}
2340	+	}
2341	+	return false;
2342	+	}
2343	+
2344	+	/**
2345	+	* Tries to pop and run local tasks within the same computation
2346	+	* as the given root. On failure, tries to help complete from
2347	+	* other queues via helpComplete.
2348	+	*/
2349	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2350	+	ForkJoinTask<?>[] a; int m;
2351	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2352	+	root != null && root.status >= 0) {
2353	+	for (;;) {
2354	+	int s, u; Object o; CountedCompleter<?> task = null;
2355	+	if ((s = q.top) - q.base > 0) {
2356	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2357	+	if ((o = U.getObject(a, j)) != null &&
2358	+	(o instanceof CountedCompleter)) {
2359	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2360	+	do {
2361	+	if (r == root) {
2362	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2363	+	if (q.array == a && q.top == s &&
2364	+	U.compareAndSwapObject(a, j, t, null)) {
2365	+	q.top = s - 1;
2366	+	task = t;
2367	+	}
2368	+	q.qlock = 0;
2369	+	}
2370	+	break;
2371	+	}
2372	+	} while ((r = r.completer) != null);
2373	+	}
2374	+	}
2375	+	if (task != null)
2376	+	task.doExec();
2377	+	if (root.status < 0 \|\|
2378	+	(u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0)
2379	+	break;
2380	+	if (task == null) {
2381	+	helpSignal(root, q.poolIndex, false);
2382	+	if (root.status >= 0)
2383	+	helpComplete(root, SHARED_QUEUE);
2384	+	break;
2385	+	}
2386	+	}
2387	+	}
2388	+	}
2389	+
2390	+	/**
2391	+	* Tries to help execute or signal availability of the given task
2392	+	* from submitter's queue in common pool.
2393	+	*/
2394	+	static void externalHelpJoin(ForkJoinTask<?> t) {
2395	+	// Some hard-to-avoid overlap with tryExternalUnpush
2396	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2397	+	ForkJoinTask<?>[] a; int m, s, n;
2398	+	if (t != null &&
2399	+	(z = submitters.get()) != null &&
2400	+	(p = commonPool) != null &&
2401	+	(ws = p.workQueues) != null &&
2402	+	(m = ws.length - 1) >= 0 &&
2403	+	(q = ws[m & z.seed & SQMASK]) != null &&
2404	+	(a = q.array) != null) {
2405	+	int am = a.length - 1;
2406	+	if ((s = q.top) != q.base) {
2407	+	long j = ((am & (s - 1)) << ASHIFT) + ABASE;
2408	+	if (U.getObject(a, j) == t &&
2409	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2410	+	if (q.array == a && q.top == s &&
2411	+	U.compareAndSwapObject(a, j, t, null)) {
2412	+	q.top = s - 1;
2413	+	q.qlock = 0;
2414	+	t.doExec();
2415	+	}
2416	+	else
2417	+	q.qlock = 0;
2418	+	}
2419	+	}
2420	+	if (t.status >= 0) {
2421	+	if (t instanceof CountedCompleter)
2422	+	p.externalHelpComplete(q, t);
2423	+	else
2424	+	p.helpSignal(t, q.poolIndex, false);
2425	+	}
2426	+	}
2427	+	}
2428	+
2429	+	/**
2430	+	* Restricted version of helpQuiescePool for external callers
2431	+	*/
2432	+	static void externalHelpQuiescePool() {
2433	+	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2434	+	if ((p = commonPool) != null &&
2435	+	(q = p.findNonEmptyStealQueue(1)) != null &&
2436	+	(b = q.base) - q.top < 0 &&
2437	+	(t = q.pollAt(b)) != null)
2438	+	t.doExec();
2439	+	}
2440	+
2441		// Exported methods
2442
2443		// Constructors
#	Line 2146 \| Line 2509 \| public class ForkJoinPool extends Abstra
2509		throw new NullPointerException();
2510		if (parallelism <= 0 \|\| parallelism > MAX_CAP)
2511		throw new IllegalArgumentException();
2149	–	this.parallelism = parallelism;
2512		this.factory = factory;
2513		this.ueh = handler;
2514	<	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2514	>	this.config = parallelism \| (asyncMode ? (FIFO_QUEUE << 16) : 0);
2515		long np = (long)(-parallelism); // offset ctl counts
2516		this.ctl = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
2517	<	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2156	<	int n = parallelism - 1;
2157	<	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
2158	<	int size = (n + 1) << 1; // #slots = 2*#workers
2159	<	this.submitMask = size - 1; // room for max # of submit queues
2160	<	this.workQueues = new WorkQueue[size];
2161	<	this.termination = (this.lock = new Mutex()).newCondition();
2162	<	this.stealCount = new AtomicLong();
2163	<	this.nextWorkerNumber = new AtomicInteger();
2164	<	int pn = poolNumberGenerator.incrementAndGet();
2517	>	int pn = nextPoolId();
2518		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2519		sb.append(Integer.toString(pn));
2520		sb.append("-worker-");
2521		this.workerNamePrefix = sb.toString();
2522	<	lock.lock();
2523	<	this.runState = 1; // set init flag
2524	<	lock.unlock();
2522	>	}
2523	>
2524	>	/**
2525	>	* Constructor for common pool, suitable only for static initialization.
2526	>	* Basically the same as above, but uses smallest possible initial footprint.
2527	>	*/
2528	>	ForkJoinPool(int parallelism, long ctl,
2529	>	ForkJoinWorkerThreadFactory factory,
2530	>	Thread.UncaughtExceptionHandler handler) {
2531	>	this.config = parallelism;
2532	>	this.ctl = ctl;
2533	>	this.factory = factory;
2534	>	this.ueh = handler;
2535	>	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2536	>	}
2537	>
2538	>	/**
2539	>	* Returns the common pool instance.
2540	>	*
2541	>	* @return the common pool instance
2542	>	*/
2543	>	public static ForkJoinPool commonPool() {
2544	>	// assert commonPool != null : "static init error";
2545	>	return commonPool;
2546		}
2547
2548		// Execution methods
#	Line 2192 \| Line 2566 \| public class ForkJoinPool extends Abstra
2566		public <T> T invoke(ForkJoinTask<T> task) {
2567		if (task == null)
2568		throw new NullPointerException();
2569	<	doSubmit(task);
2569	>	externalPush(task);
2570		return task.join();
2571		}
2572
#	Line 2207 \| Line 2581 \| public class ForkJoinPool extends Abstra
2581		public void execute(ForkJoinTask<?> task) {
2582		if (task == null)
2583		throw new NullPointerException();
2584	<	doSubmit(task);
2584	>	externalPush(task);
2585		}
2586
2587		// AbstractExecutorService methods
#	Line 2225 \| Line 2599 \| public class ForkJoinPool extends Abstra
2599		job = (ForkJoinTask<?>) task;
2600		else
2601		job = new ForkJoinTask.AdaptedRunnableAction(task);
2602	<	doSubmit(job);
2602	>	externalPush(job);
2603		}
2604
2605		/**
#	Line 2240 \| Line 2614 \| public class ForkJoinPool extends Abstra
2614		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2615		if (task == null)
2616		throw new NullPointerException();
2617	<	doSubmit(task);
2617	>	externalPush(task);
2618		return task;
2619		}
2620
#	Line 2251 \| Line 2625 \| public class ForkJoinPool extends Abstra
2625		*/
2626		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2627		ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2628	<	doSubmit(job);
2628	>	externalPush(job);
2629		return job;
2630		}
2631
#	Line 2262 \| Line 2636 \| public class ForkJoinPool extends Abstra
2636		*/
2637		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2638		ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2639	<	doSubmit(job);
2639	>	externalPush(job);
2640		return job;
2641		}
2642
#	Line 2279 \| Line 2653 \| public class ForkJoinPool extends Abstra
2653		job = (ForkJoinTask<?>) task;
2654		else
2655		job = new ForkJoinTask.AdaptedRunnableAction(task);
2656	<	doSubmit(job);
2656	>	externalPush(job);
2657		return job;
2658		}
2659
#	Line 2301 \| Line 2675 \| public class ForkJoinPool extends Abstra
2675		try {
2676		for (Callable<T> t : tasks) {
2677		ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2678	<	doSubmit(f);
2678	>	externalPush(f);
2679		fs.add(f);
2680		}
2681		for (ForkJoinTask<T> f : fs)
#	Line 2340 \| Line 2714 \| public class ForkJoinPool extends Abstra
2714		* @return the targeted parallelism level of this pool
2715		*/
2716		public int getParallelism() {
2717	<	return parallelism;
2717	>	return config & SMASK;
2718	>	}
2719	>
2720	>	/**
2721	>	* Returns the targeted parallelism level of the common pool.
2722	>	*
2723	>	* @return the targeted parallelism level of the common pool
2724	>	*/
2725	>	public static int getCommonPoolParallelism() {
2726	>	return commonPoolParallelism;
2727		}
2728
2729		/**
#	Line 2352 \| Line 2735 \| public class ForkJoinPool extends Abstra
2735		* @return the number of worker threads
2736		*/
2737		public int getPoolSize() {
2738	<	return parallelism + (short)(ctl >>> TC_SHIFT);
2738	>	return (config & SMASK) + (short)(ctl >>> TC_SHIFT);
2739		}
2740
2741		/**
#	Line 2362 \| Line 2745 \| public class ForkJoinPool extends Abstra
2745		* @return {@code true} if this pool uses async mode
2746		*/
2747		public boolean getAsyncMode() {
2748	<	return localMode != 0;
2748	>	return (config >>> 16) == FIFO_QUEUE;
2749		}
2750
2751		/**
#	Line 2393 \| Line 2776 \| public class ForkJoinPool extends Abstra
2776		* @return the number of active threads
2777		*/
2778		public int getActiveThreadCount() {
2779	<	int r = parallelism + (int)(ctl >> AC_SHIFT);
2779	>	int r = (config & SMASK) + (int)(ctl >> AC_SHIFT);
2780		return (r <= 0) ? 0 : r; // suppress momentarily negative values
2781		}
2782
#	Line 2409 \| Line 2792 \| public class ForkJoinPool extends Abstra
2792		* @return {@code true} if all threads are currently idle
2793		*/
2794		public boolean isQuiescent() {
2795	<	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2795	>	return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0;
2796		}
2797
2798		/**
#	Line 2424 \| Line 2807 \| public class ForkJoinPool extends Abstra
2807		* @return the number of steals
2808		*/
2809		public long getStealCount() {
2810	<	long count = stealCount.get();
2810	>	long count = stealCount;
2811		WorkQueue[] ws; WorkQueue w;
2812		if ((ws = workQueues) != null) {
2813		for (int i = 1; i < ws.length; i += 2) {
2814		if ((w = ws[i]) != null)
2815	<	count += w.totalSteals;
2815	>	count += w.nsteals;
2816		}
2817		}
2818		return count;
#	Line 2554 \| Line 2937 \| public class ForkJoinPool extends Abstra
2937		public String toString() {
2938		// Use a single pass through workQueues to collect counts
2939		long qt = 0L, qs = 0L; int rc = 0;
2940	<	long st = stealCount.get();
2940	>	long st = stealCount;
2941		long c = ctl;
2942		WorkQueue[] ws; WorkQueue w;
2943		if ((ws = workQueues) != null) {
#	Line 2565 \| Line 2948 \| public class ForkJoinPool extends Abstra
2948		qs += size;
2949		else {
2950		qt += size;
2951	<	st += w.totalSteals;
2951	>	st += w.nsteals;
2952		if (w.isApparentlyUnblocked())
2953		++rc;
2954		}
2955		}
2956		}
2957		}
2958	<	int pc = parallelism;
2958	>	int pc = (config & SMASK);
2959		int tc = pc + (short)(c >>> TC_SHIFT);
2960		int ac = pc + (int)(c >> AC_SHIFT);
2961		if (ac < 0) // ignore transient negative
#	Line 2581 \| Line 2964 \| public class ForkJoinPool extends Abstra
2964		if ((c & STOP_BIT) != 0)
2965		level = (tc == 0) ? "Terminated" : "Terminating";
2966		else
2967	<	level = runState < 0 ? "Shutting down" : "Running";
2967	>	level = plock < 0 ? "Shutting down" : "Running";
2968		return super.toString() +
2969		"[" + level +
2970		", parallelism = " + pc +
#	Line 2595 \| Line 2978 \| public class ForkJoinPool extends Abstra
2978		}
2979
2980		/**
2981	<	* Initiates an orderly shutdown in which previously submitted
2982	<	* tasks are executed, but no new tasks will be accepted.
2983	<	* Invocation has no additional effect if already shut down.
2984	<	* Tasks that are in the process of being submitted concurrently
2985	<	* during the course of this method may or may not be rejected.
2981	>	* Possibly initiates an orderly shutdown in which previously
2982	>	* submitted tasks are executed, but no new tasks will be
2983	>	* accepted. Invocation has no effect on execution state if this
2984	>	* is the {@link #commonPool}, and no additional effect if
2985	>	* already shut down. Tasks that are in the process of being
2986	>	* submitted concurrently during the course of this method may or
2987	>	* may not be rejected.
2988		*
2989		* @throws SecurityException if a security manager exists and
2990		* the caller is not permitted to modify threads
#	Line 2612 \| Line 2997 \| public class ForkJoinPool extends Abstra
2997		}
2998
2999		/**
3000	<	* Attempts to cancel and/or stop all tasks, and reject all
3001	<	* subsequently submitted tasks. Tasks that are in the process of
3002	<	* being submitted or executed concurrently during the course of
3003	<	* this method may or may not be rejected. This method cancels
3004	<	* both existing and unexecuted tasks, in order to permit
3005	<	* termination in the presence of task dependencies. So the method
3006	<	* always returns an empty list (unlike the case for some other
3007	<	* Executors).
3000	>	* Possibly attempts to cancel and/or stop all tasks, and reject
3001	>	* all subsequently submitted tasks. Invocation has no effect on
3002	>	* execution state if this is the {@link #commonPool}, and no
3003	>	* additional effect if already shut down. Otherwise, tasks that
3004	>	* are in the process of being submitted or executed concurrently
3005	>	* during the course of this method may or may not be
3006	>	* rejected. This method cancels both existing and unexecuted
3007	>	* tasks, in order to permit termination in the presence of task
3008	>	* dependencies. So the method always returns an empty list
3009	>	* (unlike the case for some other Executors).
3010		*
3011		* @return an empty list
3012		* @throws SecurityException if a security manager exists and
#	Line 2641 \| Line 3028 \| public class ForkJoinPool extends Abstra
3028		public boolean isTerminated() {
3029		long c = ctl;
3030		return ((c & STOP_BIT) != 0L &&
3031	<	(short)(c >>> TC_SHIFT) == -parallelism);
3031	>	(short)(c >>> TC_SHIFT) == -(config & SMASK));
3032		}
3033
3034		/**
#	Line 2660 \| Line 3047 \| public class ForkJoinPool extends Abstra
3047		public boolean isTerminating() {
3048		long c = ctl;
3049		return ((c & STOP_BIT) != 0L &&
3050	<	(short)(c >>> TC_SHIFT) != -parallelism);
3050	>	(short)(c >>> TC_SHIFT) != -(config & SMASK));
3051		}
3052
3053		/**
#	Line 2669 \| Line 3056 \| public class ForkJoinPool extends Abstra
3056		* @return {@code true} if this pool has been shut down
3057		*/
3058		public boolean isShutdown() {
3059	<	return runState < 0;
3059	>	return plock < 0;
3060		}
3061
3062		/**
3063	<	* Blocks until all tasks have completed execution after a shutdown
3064	<	* request, or the timeout occurs, or the current thread is
3065	<	* interrupted, whichever happens first.
3063	>	* Blocks until all tasks have completed execution after a
3064	>	* shutdown request, or the timeout occurs, or the current thread
3065	>	* is interrupted, whichever happens first. Note that the {@link
3066	>	* #commonPool()} never terminates until program shutdown so
3067	>	* this method will always time out.
3068		*
3069		* @param timeout the maximum time to wait
3070		* @param unit the time unit of the timeout argument
#	Line 2686 \| Line 3075 \| public class ForkJoinPool extends Abstra
3075		public boolean awaitTermination(long timeout, TimeUnit unit)
3076		throws InterruptedException {
3077		long nanos = unit.toNanos(timeout);
3078	<	final Mutex lock = this.lock;
3079	<	lock.lock();
3080	<	try {
3081	<	for (;;) {
3082	<	if (isTerminated())
3083	<	return true;
3084	<	if (nanos <= 0)
3085	<	return false;
3086	<	nanos = termination.awaitNanos(nanos);
3078	>	if (isTerminated())
3079	>	return true;
3080	>	long startTime = System.nanoTime();
3081	>	boolean terminated = false;
3082	>	synchronized (this) {
3083	>	for (long waitTime = nanos, millis = 0L;;) {
3084	>	if (terminated = isTerminated() \|\|
3085	>	waitTime <= 0L \|\|
3086	>	(millis = unit.toMillis(waitTime)) <= 0L)
3087	>	break;
3088	>	wait(millis);
3089	>	waitTime = nanos - (System.nanoTime() - startTime);
3090		}
2699	–	} finally {
2700	–	lock.unlock();
3091		}
3092	+	return terminated;
3093		}
3094
3095		/**
#	Line 2797 \| Line 3188 \| public class ForkJoinPool extends Abstra
3188		public static void managedBlock(ManagedBlocker blocker)
3189		throws InterruptedException {
3190		Thread t = Thread.currentThread();
3191	<	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3192	<	((ForkJoinWorkerThread)t).pool : null);
3193	<	while (!blocker.isReleasable()) {
3194	<	if (p == null \|\| p.tryCompensate(null, blocker)) {
3195	<	try {
3196	<	do {} while (!blocker.isReleasable() && !blocker.block());
3197	<	} finally {
3198	<	if (p != null)
3191	>	if (t instanceof ForkJoinWorkerThread) {
3192	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3193	>	while (!blocker.isReleasable()) { // variant of helpSignal
3194	>	WorkQueue[] ws; WorkQueue q; int m, u;
3195	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3196	>	for (int i = 0; i <= m; ++i) {
3197	>	if (blocker.isReleasable())
3198	>	return;
3199	>	if ((q = ws[i]) != null && q.base - q.top < 0) {
3200	>	p.signalWork(q);
3201	>	if ((u = (int)(p.ctl >>> 32)) >= 0 \|\|
3202	>	(u >> UAC_SHIFT) >= 0)
3203	>	break;
3204	>	}
3205	>	}
3206	>	}
3207	>	if (p.tryCompensate()) {
3208	>	try {
3209	>	do {} while (!blocker.isReleasable() &&
3210	>	!blocker.block());
3211	>	} finally {
3212		p.incrementActiveCount();
3213	+	}
3214	+	break;
3215		}
2810	–	break;
3216		}
3217		}
3218	+	else {
3219	+	do {} while (!blocker.isReleasable() &&
3220	+	!blocker.block());
3221	+	}
3222		}
3223
3224		// AbstractExecutorService overrides. These rely on undocumented
#	Line 2830 \| Line 3239 \| public class ForkJoinPool extends Abstra
3239		private static final long PARKBLOCKER;
3240		private static final int ABASE;
3241		private static final int ASHIFT;
3242	+	private static final long STEALCOUNT;
3243	+	private static final long PLOCK;
3244	+	private static final long INDEXSEED;
3245	+	private static final long QLOCK;
3246
3247		static {
3248	<	poolNumberGenerator = new AtomicInteger();
2836	<	nextSubmitterSeed = new AtomicInteger(0x55555555);
2837	<	modifyThreadPermission = new RuntimePermission("modifyThread");
2838	<	defaultForkJoinWorkerThreadFactory =
2839	<	new DefaultForkJoinWorkerThreadFactory();
2840	<	submitters = new ThreadSubmitter();
2841	<	int s;
3248	>	int s; // initialize field offsets for CAS etc
3249		try {
3250		U = getUnsafe();
3251		Class<?> k = ForkJoinPool.class;
2845	–	Class<?> ak = ForkJoinTask[].class;
3252		CTL = U.objectFieldOffset
3253		(k.getDeclaredField("ctl"));
3254	+	STEALCOUNT = U.objectFieldOffset
3255	+	(k.getDeclaredField("stealCount"));
3256	+	PLOCK = U.objectFieldOffset
3257	+	(k.getDeclaredField("plock"));
3258	+	INDEXSEED = U.objectFieldOffset
3259	+	(k.getDeclaredField("indexSeed"));
3260		Class<?> tk = Thread.class;
3261		PARKBLOCKER = U.objectFieldOffset
3262		(tk.getDeclaredField("parkBlocker"));
3263	+	Class<?> wk = WorkQueue.class;
3264	+	QLOCK = U.objectFieldOffset
3265	+	(wk.getDeclaredField("qlock"));
3266	+	Class<?> ak = ForkJoinTask[].class;
3267		ABASE = U.arrayBaseOffset(ak);
3268		s = U.arrayIndexScale(ak);
3269	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3270		} catch (Exception e) {
3271		throw new Error(e);
3272		}
3273		if ((s & (s-1)) != 0)
3274		throw new Error("data type scale not a power of two");
3275	<	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3275	>
3276	>	submitters = new ThreadLocal<Submitter>();
3277	>	ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory =
3278	>	new DefaultForkJoinWorkerThreadFactory();
3279	>	modifyThreadPermission = new RuntimePermission("modifyThread");
3280	>
3281	>	/*
3282	>	* Establish common pool parameters. For extra caution,
3283	>	* computations to set up common pool state are here; the
3284	>	* constructor just assigns these values to fields.
3285	>	*/
3286	>
3287	>	int par = 0;
3288	>	Thread.UncaughtExceptionHandler handler = null;
3289	>	try { // TBD: limit or report ignored exceptions?
3290	>	String pp = System.getProperty
3291	>	("java.util.concurrent.ForkJoinPool.common.parallelism");
3292	>	String hp = System.getProperty
3293	>	("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3294	>	String fp = System.getProperty
3295	>	("java.util.concurrent.ForkJoinPool.common.threadFactory");
3296	>	if (fp != null)
3297	>	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3298	>	getSystemClassLoader().loadClass(fp).newInstance());
3299	>	if (hp != null)
3300	>	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3301	>	getSystemClassLoader().loadClass(hp).newInstance());
3302	>	if (pp != null)
3303	>	par = Integer.parseInt(pp);
3304	>	} catch (Exception ignore) {
3305	>	}
3306	>
3307	>	if (par <= 0)
3308	>	par = Runtime.getRuntime().availableProcessors();
3309	>	if (par > MAX_CAP)
3310	>	par = MAX_CAP;
3311	>	commonPoolParallelism = par;
3312	>	long np = (long)(-par); // precompute initial ctl value
3313	>	long ct = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
3314	>
3315	>	commonPool = new ForkJoinPool(par, ct, fac, handler);
3316		}
3317
3318		/**

Diff Legend

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

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents): Revision 1.4 by jsr166, Fri Oct 12 16:46:37 2012 UTC vs. Revision 1.22 by jsr166, Wed Nov 21 22:15:41 2012 UTC

Diff Legend

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.4 by jsr166, Fri Oct 12 16:46:37 2012 UTC vs.
Revision 1.22 by jsr166, Wed Nov 21 22:15:41 2012 UTC