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

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.2 by jsr166, Mon Aug 13 18:25:53 2012 UTC vs.
Revision 1.20 by jsr166, Tue Nov 20 06:18:39 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 is
320	>	* apparently empty, they signal waiting workers (or trigger
321	>	* creation of new ones if fewer than the given parallelism
322	>	* level). These primary signals are buttressed by signals
323	>	* whenever other threads scan for work or do not have a task to
324	>	* process (including the case of leaving a hint to unparked
325	>	* threads to help signal others upon wakeup). On most platforms,
326	>	* signalling (unpark) overhead time is noticeably long, and the
327	>	* time between signalling a thread and it actually making
328	>	* progress can be very noticeably long, so it is worth offloading
329	>	* these delays from critical paths as much as possible.
330		*
331		* Trimming workers. To release resources after periods of lack of
332		* use, a worker starting to wait when the pool is quiescent will
333	<	* time out and terminate if the pool has remained quiescent for
334	<	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
335	<	* terminating all workers after long periods of non-use.
333	>	* time out and terminate if the pool has remained quiescent for a
334	>	* given period -- a short period if there are more threads than
335	>	* parallelism, longer as the number of threads decreases. This
336	>	* will slowly propagate, eventually terminating all workers after
337	>	* periods of non-use.
338		*
339		* Shutdown and Termination. A call to shutdownNow atomically sets
340	<	* a runState bit and then (non-atomically) sets each worker's
341	<	* runState status, cancels all unprocessed tasks, and wakes up
340	>	* a plock bit and then (non-atomically) sets each worker's
341	>	* qlock status, cancels all unprocessed tasks, and wakes up
342		* all waiting workers. Detecting whether termination should
343		* commence after a non-abrupt shutdown() call requires more work
344		* and bookkeeping. We need consensus about quiescence (i.e., that
#	Line 354 \| Line 366 \| public class ForkJoinPool extends Abstra
366		* method tryCompensate() may create or re-activate a spare
367		* thread to compensate for blocked joiners until they unblock.
368		*
369	<	* A third form (implemented in tryRemoveAndExec and
370	<	* tryPollForAndExec) amounts to helping a hypothetical
371	<	* compensator: If we can readily tell that a possible action of a
372	<	* compensator is to steal and execute the task being joined, the
373	<	* joining thread can do so directly, without the need for a
374	<	* compensation thread (although at the expense of larger run-time
375	<	* stacks, but the tradeoff is typically worthwhile).
369	>	* A third form (implemented in tryRemoveAndExec) amounts to
370	>	* helping a hypothetical compensator: If we can readily tell that
371	>	* a possible action of a compensator is to steal and execute the
372	>	* task being joined, the joining thread can do so directly,
373	>	* without the need for a compensation thread (although at the
374	>	* expense of larger run-time stacks, but the tradeoff is
375	>	* typically worthwhile).
376		*
377		* The ManagedBlocker extension API can't use helping so relies
378		* only on compensation in method awaitBlocker.
#	Line 382 \| Line 394 \| public class ForkJoinPool extends Abstra
394		* steals, rather than use per-task bookkeeping. This sometimes
395		* requires a linear scan of workQueues array to locate stealers,
396		* but often doesn't because stealers leave hints (that may become
397	<	* stale/wrong) of where to locate them. A stealHint is only a
398	<	* hint because a worker might have had multiple steals and the
399	<	* hint records only one of them (usually the most current).
400	<	* Hinting isolates cost to when it is needed, rather than adding
401	<	* to per-task overhead. (2) It is "shallow", ignoring nesting
402	<	* and potentially cyclic mutual steals. (3) It is intentionally
397	>	* stale/wrong) of where to locate them. It is only a hint
398	>	* because a worker might have had multiple steals and the hint
399	>	* records only one of them (usually the most current). Hinting
400	>	* isolates cost to when it is needed, rather than adding to
401	>	* per-task overhead. (2) It is "shallow", ignoring nesting and
402	>	* potentially cyclic mutual steals. (3) It is intentionally
403		* racy: field currentJoin is updated only while actively joining,
404		* which means that we miss links in the chain during long-lived
405		* tasks, GC stalls etc (which is OK since blocking in such cases
#	Line 395 \| Line 407 \| public class ForkJoinPool extends Abstra
407		* to find work (see MAX_HELP) and fall back to suspending the
408		* worker and if necessary replacing it with another.
409		*
410	+	* Helping actions for CountedCompleters are much simpler: Method
411	+	* helpComplete can take and execute any task with the same root
412	+	* as the task being waited on. However, this still entails some
413	+	* traversal of completer chains, so is less efficient than using
414	+	* CountedCompleters without explicit joins.
415	+	*
416		* It is impossible to keep exactly the target parallelism number
417		* of threads running at any given time. Determining the
418		* existence of conservatively safe helping targets, the
#	Line 416 \| Line 434 \| public class ForkJoinPool extends Abstra
434		* intractable) game with an opponent that may choose the worst
435		* (for us) active thread to stall at any time. We take several
436		* precautions to bound losses (and thus bound gains), mainly in
437	<	* methods tryCompensate and awaitJoin: (1) We only try
438	<	* compensation after attempting enough helping steps (measured
439	<	* via counting and timing) that we have already consumed the
440	<	* estimated cost of creating and activating a new thread. (2) We
441	<	* allow up to 50% of threads to be blocked before initially
442	<	* adding any others, and unless completely saturated, check that
443	<	* some work is available for a new worker before adding. Also, we
444	<	* create up to only 50% more threads until entering a mode that
445	<	* only adds a thread if all others are possibly blocked. All
446	<	* together, this means that we might be half as fast to react,
447	<	* and create half as many threads as possible in the ideal case,
448	<	* but present vastly fewer anomalies in all other cases compared
449	<	* to both more aggressive and more conservative alternatives.
450	<	*
451	<	* Style notes: There is a lot of representation-level coupling
452	<	* among classes ForkJoinPool, ForkJoinWorkerThread, and
453	<	* ForkJoinTask. The fields of WorkQueue maintain data structures
454	<	* managed by ForkJoinPool, so are directly accessed. There is
455	<	* little point trying to reduce this, since any associated future
456	<	* changes in representations will need to be accompanied by
457	<	* algorithmic changes anyway. Several methods intrinsically
458	<	* sprawl because they must accumulate sets of consistent reads of
459	<	* volatiles held in local variables. Methods signalWork() and
460	<	* scan() are the main bottlenecks, so are especially heavily
437	>	* methods tryCompensate and awaitJoin.
438	>	*
439	>	* Common Pool
440	>	* ===========
441	>	*
442	>	* The static commonPool always exists after static
443	>	* initialization. Since it (or any other created pool) need
444	>	* never be used, we minimize initial construction overhead and
445	>	* footprint to the setup of about a dozen fields, with no nested
446	>	* allocation. Most bootstrapping occurs within method
447	>	* fullExternalPush during the first submission to the pool.
448	>	*
449	>	* When external threads submit to the common pool, they can
450	>	* perform some subtask processing (see externalHelpJoin and
451	>	* related methods). We do not need to record whether these
452	>	* submissions are to the common pool -- if not, externalHelpJoin
453	>	* returns quickly (at the most helping to signal some common pool
454	>	* workers). These submitters would otherwise be blocked waiting
455	>	* for completion, so the extra effort (with liberally sprinkled
456	>	* task status checks) in inapplicable cases amounts to an odd
457	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
458	>	*
459	>	* Style notes
460	>	* ===========
461	>	*
462	>	* There is a lot of representation-level coupling among classes
463	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
464	>	* fields of WorkQueue maintain data structures managed by
465	>	* ForkJoinPool, so are directly accessed. There is little point
466	>	* trying to reduce this, since any associated future changes in
467	>	* representations will need to be accompanied by algorithmic
468	>	* changes anyway. Several methods intrinsically sprawl because
469	>	* they must accumulate sets of consistent reads of volatiles held
470	>	* in local variables. Methods signalWork() and scan() are the
471	>	* main bottlenecks, so are especially heavily
472		* micro-optimized/mangled. There are lots of inline assignments
473		* (of form "while ((local = field) != 0)") which are usually the
474		* simplest way to ensure the required read orderings (which are
#	Line 447 \| Line 476 \| public class ForkJoinPool extends Abstra
476		* declarations of these locals at the heads of methods or blocks.
477		* There are several occurrences of the unusual "do {} while
478		* (!cas...)" which is the simplest way to force an update of a
479	<	* CAS'ed variable. There are also other coding oddities that help
479	>	* CAS'ed variable. There are also other coding oddities (including
480	>	* several unnecessary-looking hoisted null checks) that help
481		* some methods perform reasonably even when interpreted (not
482		* compiled).
483		*
#	Line 496 \| Line 526 \| public class ForkJoinPool extends Abstra
526		* Default ForkJoinWorkerThreadFactory implementation; creates a
527		* new ForkJoinWorkerThread.
528		*/
529	<	static class DefaultForkJoinWorkerThreadFactory
529	>	static final class DefaultForkJoinWorkerThreadFactory
530		implements ForkJoinWorkerThreadFactory {
531	<	public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
531	>	public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
532		return new ForkJoinWorkerThread(pool);
533		}
534		}
535
536		/**
507	–	* A simple non-reentrant lock used for exclusion when managing
508	–	* queues and workers. We use a custom lock so that we can readily
509	–	* probe lock state in constructions that check among alternative
510	–	* actions. The lock is normally only very briefly held, and
511	–	* sometimes treated as a spinlock, but other usages block to
512	–	* reduce overall contention in those cases where locked code
513	–	* bodies perform allocation/resizing.
514	–	*/
515	–	static final class Mutex extends AbstractQueuedSynchronizer {
516	–	public final boolean tryAcquire(int ignore) {
517	–	return compareAndSetState(0, 1);
518	–	}
519	–	public final boolean tryRelease(int ignore) {
520	–	setState(0);
521	–	return true;
522	–	}
523	–	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(); }
527	–	}
528	–
529	–	/**
537		* Class for artificial tasks that are used to replace the target
538		* of local joins if they are removed from an interior queue slot
539		* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
540		* actually do anything beyond having a unique identity.
541		*/
542		static final class EmptyTask extends ForkJoinTask<Void> {
543	+	private static final long serialVersionUID = -7721805057305804111L;
544		EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
545		public final Void getRawResult() { return null; }
546		public final void setRawResult(Void x) {}
#	Line 553 \| Line 561 \| public class ForkJoinPool extends Abstra
561		*
562		* Field "top" is the index (mod array.length) of the next queue
563		* slot to push to or pop from. It is written only by owner thread
564	<	* for push, or under lock for trySharedPush, and accessed by
565	<	* other threads only after reading (volatile) base. Both top and
566	<	* base are allowed to wrap around on overflow, but (top - base)
567	<	* (or more commonly -(base - top) to force volatile read of base
568	<	* before top) still estimates size.
564	>	* for push, or under lock for external/shared push, and accessed
565	>	* by other threads only after reading (volatile) base. Both top
566	>	* and base are allowed to wrap around on overflow, but (top -
567	>	* base) (or more commonly -(base - top) to force volatile read of
568	>	* base before top) still estimates size. The lock ("qlock") is
569	>	* forced to -1 on termination, causing all further lock attempts
570	>	* to fail. (Note: we don't need CAS for termination state because
571	>	* upon pool shutdown, all shared-queues will stop being used
572	>	* anyway.) Nearly all lock bodies are set up so that exceptions
573	>	* within lock bodies are "impossible" (modulo JVM errors that
574	>	* would cause failure anyway.)
575		*
576		* The array slots are read and written using the emulation of
577		* volatiles/atomics provided by Unsafe. Insertions must in
578		* general use putOrderedObject as a form of releasing store to
579		* ensure that all writes to the task object are ordered before
580	<	* its publication in the queue. (Although we can avoid one case
581	<	* of this when locked in trySharedPush.) All removals entail a
582	<	* CAS to null. The array is always a power of two. To ensure
583	<	* safety of Unsafe array operations, all accesses perform
570	<	* explicit null checks and implicit bounds checks via
571	<	* power-of-two masking.
580	>	* its publication in the queue. All removals entail a CAS to
581	>	* null. The array is always a power of two. To ensure safety of
582	>	* Unsafe array operations, all accesses perform explicit null
583	>	* checks and implicit bounds checks via power-of-two masking.
584		*
585		* In addition to basic queuing support, this class contains
586		* fields described elsewhere to control execution. It turns out
587	<	* to work better memory-layout-wise to include them in this
588	<	* class rather than a separate class.
587	>	* to work better memory-layout-wise to include them in this class
588	>	* rather than a separate class.
589		*
590		* Performance on most platforms is very sensitive to placement of
591		* instances of both WorkQueues and their arrays -- we absolutely
#	Line 587 \| Line 599 \| public class ForkJoinPool extends Abstra
599		* trades off slightly slower average field access for the sake of
600		* avoiding really bad worst-case access. (Until better JVM
601		* support is in place, this padding is dependent on transient
602	<	* properties of JVM field layout rules.) We also take care in
591	<	* allocating, sizing and resizing the array. Non-shared queue
592	<	* arrays are initialized (via method growArray) by workers before
593	<	* use. Others are allocated on first use.
602	>	* properties of JVM field layout rules.)
603		*/
604		static final class WorkQueue {
605		/**
#	Line 613 \| Line 622 \| public class ForkJoinPool extends Abstra
622		*/
623		static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
624
616	–	volatile long totalSteals; // cumulative number of steals
625		int seed; // for random scanning; initialize nonzero
626		volatile int eventCount; // encoded inactivation count; < 0 if inactive
627		int nextWait; // encoded record of next event waiter
628	<	int rescans; // remaining scans until block
621	<	int nsteals; // top-level task executions since last idle
622	<	final int mode; // lifo, fifo, or shared
628	>	int hint; // steal or signal hint (index)
629		int poolIndex; // index of this queue in pool (or 0)
630	<	int stealHint; // index of most recent known stealer
631	<	volatile int runState; // 1: locked, -1: terminate; else 0
630	>	final int mode; // 0: lifo, > 0: fifo, < 0: shared
631	>	int nsteals; // number of steals
632	>	volatile int qlock; // 1: locked, -1: terminate; else 0
633		volatile int base; // index of next slot for poll
634		int top; // index of next slot for push
635		ForkJoinTask<?>[] array; // the elements (initially unallocated)
#	Line 631 \| Line 638 \| public class ForkJoinPool extends Abstra
638		volatile Thread parker; // == owner during call to park; else null
639		volatile ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
640		ForkJoinTask<?> currentSteal; // current non-local task being executed
641	+
642		// Heuristic padding to ameliorate unfortunate memory placements
643		Object p00, p01, p02, p03, p04, p05, p06, p07;
644	<	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
644	>	Object p08, p09, p0a, p0b, p0c;
645
646	<	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
647	<	this.mode = mode;
646	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
647	>	int seed) {
648	>	this.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY];
649		this.pool = pool;
650		this.owner = owner;
651	<	// Place indices in the center of array (that is not yet allocated)
651	>	this.mode = mode;
652	>	this.seed = seed;
653	>	// Place indices in the center of array
654		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
655		}
656
657		/**
647	–	* Returns the approximate number of tasks in the queue.
648	–	*/
649	–	final int queueSize() {
650	–	int n = base - top; // non-owner callers must read base first
651	–	return (n >= 0) ? 0 : -n; // ignore transient negative
652	–	}
653	–
654	–	/**
655	–	* Provides a more accurate estimate of whether this queue has
656	–	* any tasks than does queueSize, by checking whether a
657	–	* near-empty queue has at least one unclaimed task.
658	–	*/
659	–	final boolean isEmpty() {
660	–	ForkJoinTask<?>[] a; int m, s;
661	–	int n = base - (s = top);
662	–	return (n >= 0 \|\|
663	–	(n == -1 &&
664	–	((a = array) == null \|\|
665	–	(m = a.length - 1) < 0 \|\|
666	–	U.getObjectVolatile
667	–	(a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
668	–	}
669	–
670	–	/**
658		* Pushes a task. Call only by owner in unshared queues.
659	+	* Cases needing resizing or rejection are relayed to fullPush
660	+	* (that also handles shared queues).
661		*
662		* @param task the task. Caller must ensure non-null.
663		* @throw RejectedExecutionException if array cannot be resized
#	Line 679 \| Line 668 \| public class ForkJoinPool extends Abstra
668		if ((a = array) != null) { // ignore if queue removed
669		U.putOrderedObject
670		(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
671	<	if ((n = (top = s + 1) - base) <= 2) {
671	>	if ((n = (top = s + 1) - base) <= 1) {
672		if ((p = pool) != null)
673	<	p.signalWork();
673	>	p.signalWork(this, 0);
674		}
675		else if (n >= m)
676	<	growArray(true);
676	>	growArray();
677		}
678		}
679
#	Line 697 \| Line 686 \| public class ForkJoinPool extends Abstra
686		*/
687		final boolean trySharedPush(ForkJoinTask<?> task) {
688		boolean submitted = false;
689	<	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
690	<	ForkJoinTask<?>[] a = array;
689	>	if (qlock == 0 && U.compareAndSwapInt(this, QLOCK, 0, 1)) {
690	>	ForkJoinTask<?>[] a = array; ForkJoinPool p;
691		int s = top;
692		try {
693		if ((a != null && a.length > s + 1 - base) \|\|
694	<	(a = growArray(false)) != null) { // must presize
694	>	(a = growArray()) != null) { // must presize
695		int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
696	<	U.putObject(a, (long)j, task); // don't need "ordered"
696	>	U.putOrderedObject(a, j, task);
697		top = s + 1;
698		submitted = true;
699		}
700		} finally {
701	<	runState = 0; // unlock
701	>	qlock = 0; // unlock
702		}
703	+	if (submitted && (p = pool) != null)
704	+	p.signalWork(this, 0);
705		}
706		return submitted;
707		}
708
709	+	/**
710	+	* Initializes or doubles the capacity of array. Call either
711	+	* by owner or with lock held -- it is OK for base, but not
712	+	* top, to move while resizings are in progress.
713	+	*/
714	+	final ForkJoinTask<?>[] growArray() {
715	+	ForkJoinTask<?>[] oldA = array;
716	+	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
717	+	if (size > MAXIMUM_QUEUE_CAPACITY)
718	+	throw new RejectedExecutionException("Queue capacity exceeded");
719	+	int oldMask, t, b;
720	+	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
721	+	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
722	+	(t = top) - (b = base) > 0) {
723	+	int mask = size - 1;
724	+	do {
725	+	ForkJoinTask<?> x;
726	+	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
727	+	int j = ((b & mask) << ASHIFT) + ABASE;
728	+	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
729	+	if (x != null &&
730	+	U.compareAndSwapObject(oldA, oldj, x, null))
731	+	U.putObjectVolatile(a, j, x);
732	+	} while (++b != t);
733	+	}
734	+	return a;
735	+	}
736	+
737		/**
738		* Takes next task, if one exists, in LIFO order. Call only
739	<	* by owner in unshared queues. (We do not have a shared
721	<	* version of this method because it is never needed.)
739	>	* by owner in unshared queues.
740		*/
741		final ForkJoinTask<?> pop() {
742		ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
#	Line 773 \| Line 791 \| public class ForkJoinPool extends Abstra
791		else if (base == b) {
792		if (b + 1 == top)
793		break;
794	<	Thread.yield(); // wait for lagging update
794	>	Thread.yield(); // wait for lagging update (very rare)
795		}
796		}
797		return null;
#	Line 800 \| Line 818 \| public class ForkJoinPool extends Abstra
818
819		/**
820		* Pops the given task only if it is at the current top.
821	+	* (A shared version is available only via FJP.tryExternalUnpush)
822		*/
823		final boolean tryUnpush(ForkJoinTask<?> t) {
824		ForkJoinTask<?>[] a; int s;
#	Line 813 \| Line 832 \| public class ForkJoinPool extends Abstra
832		}
833
834		/**
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	–	/**
835		* Removes and cancels all known tasks, ignoring any exceptions.
836		*/
837		final void cancelAll() {
#	Line 887 \| Line 855 \| public class ForkJoinPool extends Abstra
855		return seed = r ^= r << 5;
856		}
857
858	<	// Execution methods
858	>	/**
859	>	* Provides a more accurate estimate of size than (top - base)
860	>	* by ordering reads and checking whether a near-empty queue
861	>	* has at least one unclaimed task.
862	>	*/
863	>	final int queueSize() {
864	>	ForkJoinTask<?>[] a; int k, s, n;
865	>	return ((n = base - (s = top)) < 0 &&
866	>	(n != -1 \|\|
867	>	((a = array) != null && (k = a.length) > 0 &&
868	>	U.getObject
869	>	(a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
870	>	-n : 0;
871	>	}
872	>
873	>	// Specialized execution methods
874
875		/**
876		* Pops and runs tasks until empty.
#	Line 916 \| Line 899 \| public class ForkJoinPool extends Abstra
899		}
900
901		/**
902	<	* If present, removes from queue and executes the given task, or
903	<	* any other cancelled task. Returns (true) immediately on any CAS
902	>	* If present, removes from queue and executes the given task,
903	>	* or any other cancelled task. Returns (true) on any CAS
904		* or consistency check failure so caller can retry.
905		*
906	<	* @return 0 if no progress can be made, else positive
924	<	* (this unusual convention simplifies use with tryHelpStealer.)
906	>	* @return false if no progress can be made, else true;
907		*/
908	<	final int tryRemoveAndExec(ForkJoinTask<?> task) {
909	<	int stat = 1;
928	<	boolean removed = false, empty = true;
908	>	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
909	>	boolean stat = true, removed = false, empty = true;
910		ForkJoinTask<?>[] a; int m, s, b, n;
911		if ((a = array) != null && (m = a.length - 1) >= 0 &&
912		(n = (s = top) - (b = base)) > 0) {
#	Line 955 \| Line 936 \| public class ForkJoinPool extends Abstra
936		}
937		if (--n == 0) {
938		if (!empty && base == b)
939	<	stat = 0;
939	>	stat = false;
940		break;
941		}
942		}
#	Line 966 \| Line 947 \| public class ForkJoinPool extends Abstra
947		}
948
949		/**
950	+	* Polls for and executes the given task or any other task in
951	+	* its CountedCompleter computation
952	+	*/
953	+	final boolean pollAndExecCC(ForkJoinTask<?> root) {
954	+	ForkJoinTask<?>[] a; int b; Object o;
955	+	outer: while ((b = base) - top < 0 && (a = array) != null) {
956	+	long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
957	+	if ((o = U.getObject(a, j)) == null \|\|
958	+	!(o instanceof CountedCompleter))
959	+	break;
960	+	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
961	+	if (r == root) {
962	+	if (base == b &&
963	+	U.compareAndSwapObject(a, j, t, null)) {
964	+	base = b + 1;
965	+	t.doExec();
966	+	return true;
967	+	}
968	+	else
969	+	break; // restart
970	+	}
971	+	if ((r = r.completer) == null)
972	+	break outer; // not part of root computation
973	+	}
974	+	}
975	+	return false;
976	+	}
977	+
978	+	/**
979		* Executes a top-level task and any local tasks remaining
980		* after execution.
981		*/
982		final void runTask(ForkJoinTask<?> t) {
983		if (t != null) {
984	<	currentSteal = t;
985	<	t.doExec();
984	>	(currentSteal = t).doExec();
985	>	currentSteal = null;
986	>	++nsteals;
987		if (top != base) { // process remaining local tasks
988		if (mode == 0)
989		popAndExecAll();
990		else
991		pollAndExecAll();
992		}
982	–	++nsteals;
983	–	currentSteal = null;
993		}
994		}
995
#	Line 990 \| Line 999 \| public class ForkJoinPool extends Abstra
999		final void runSubtask(ForkJoinTask<?> t) {
1000		if (t != null) {
1001		ForkJoinTask<?> ps = currentSteal;
1002	<	currentSteal = t;
994	<	t.doExec();
1002	>	(currentSteal = t).doExec();
1003		currentSteal = ps;
1004		}
1005		}
#	Line 1026 \| Line 1034 \| public class ForkJoinPool extends Abstra
1034
1035		// Unsafe mechanics
1036		private static final sun.misc.Unsafe U;
1037	<	private static final long RUNSTATE;
1037	>	private static final long QLOCK;
1038		private static final int ABASE;
1039		private static final int ASHIFT;
1040		static {
#	Line 1035 \| Line 1043 \| public class ForkJoinPool extends Abstra
1043		U = getUnsafe();
1044		Class<?> k = WorkQueue.class;
1045		Class<?> ak = ForkJoinTask[].class;
1046	<	RUNSTATE = U.objectFieldOffset
1047	<	(k.getDeclaredField("runState"));
1046	>	QLOCK = U.objectFieldOffset
1047	>	(k.getDeclaredField("qlock"));
1048		ABASE = U.arrayBaseOffset(ak);
1049		s = U.arrayIndexScale(ak);
1050		} catch (Exception e) {
#	Line 1047 \| Line 1055 \| public class ForkJoinPool extends Abstra
1055		ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1056		}
1057		}
1058	+
1059	+	// static fields (initialized in static initializer below)
1060	+
1061	+	/**
1062	+	* Creates a new ForkJoinWorkerThread. This factory is used unless
1063	+	* overridden in ForkJoinPool constructors.
1064	+	*/
1065	+	public static final ForkJoinWorkerThreadFactory
1066	+	defaultForkJoinWorkerThreadFactory;
1067	+
1068		/**
1069		* Per-thread records for threads that submit to pools. Currently
1070		* holds only pseudo-random seed / index that is used to choose
1071	<	* submission queues in method doSubmit. In the future, this may
1071	>	* submission queues in method externalPush. In the future, this may
1072		* also incorporate a means to implement different task rejection
1073		* and resubmission policies.
1074		*
#	Line 1058 \| Line 1076 \| public class ForkJoinPool extends Abstra
1076		* the same way but are initialized and updated using slightly
1077		* different mechanics. Both are initialized using the same
1078		* approach as in class ThreadLocal, where successive values are
1079	<	* unlikely to collide with previous values. This is done during
1080	<	* registration for workers, but requires a separate AtomicInteger
1081	<	* for submitters. Seeds are then randomly modified upon
1064	<	* collisions using xorshifts, which requires a non-zero seed.
1079	>	* unlikely to collide with previous values. Seeds are then
1080	>	* randomly modified upon collisions using xorshifts, which
1081	>	* requires a non-zero seed.
1082		*/
1083		static final class Submitter {
1084		int seed;
1085	<	Submitter() {
1069	<	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1070	<	seed = (s == 0) ? 1 : s; // ensure non-zero
1071	<	}
1072	<	}
1073	<
1074	<	/** ThreadLocal class for Submitters */
1075	<	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1076	<	public Submitter initialValue() { return new Submitter(); }
1085	>	Submitter(int s) { seed = s; }
1086		}
1087
1088	<	// static fields (initialized in static initializer below)
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.
1092	>	* Lazily initialized on first submission (but null-checked
1093	>	* in other contexts to avoid unnecessary initialization).
1094	>	*/
1095	>	static final ThreadLocal<Submitter> submitters;
1096
1097		/**
1098	<	* Creates a new ForkJoinWorkerThread. This factory is used unless
1099	<	* overridden in ForkJoinPool constructors.
1098	>	* Common (static) pool. Non-null for public use unless a static
1099	>	* construction exception, but internal usages null-check on use
1100	>	* to paranoically avoid potential initialization circularities
1101	>	* as well as to simplify generated code.
1102		*/
1103	<	public static final ForkJoinWorkerThreadFactory
1086	<	defaultForkJoinWorkerThreadFactory;
1103	>	static final ForkJoinPool commonPool;
1104
1105		/**
1106	<	* Generator for assigning sequence numbers as pool names.
1106	>	* Permission required for callers of methods that may start or
1107	>	* kill threads.
1108		*/
1109	<	private static final AtomicInteger poolNumberGenerator;
1109	>	private static final RuntimePermission modifyThreadPermission;
1110
1111		/**
1112	<	* Generator for initial hashes/seeds for submitters. Accessed by
1095	<	* Submitter class constructor.
1112	>	* Common pool parallelism. Must equal commonPool.parallelism.
1113		*/
1114	<	static final AtomicInteger nextSubmitterSeed;
1114	>	static final int commonPoolParallelism;
1115
1116		/**
1117	<	* Permission required for callers of methods that may start or
1101	<	* kill threads.
1117	>	* Sequence number for creating workerNamePrefix.
1118		*/
1119	<	private static final RuntimePermission modifyThreadPermission;
1119	>	private static int poolNumberSequence;
1120
1121		/**
1122	<	* Per-thread submission bookkeeping. Shared across all pools
1123	<	* to reduce ThreadLocal pollution and because random motion
1108	<	* to avoid contention in one pool is likely to hold for others.
1122	>	* Return the next sequence number. We don't expect this to
1123	>	* ever contend so use simple builtin sync.
1124		*/
1125	<	private static final ThreadSubmitter submitters;
1125	>	private static final synchronized int nextPoolId() {
1126	>	return ++poolNumberSequence;
1127	>	}
1128
1129		// static constants
1130
1131		/**
1132	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
1133	<	* task when the pool is quiescent to instead try to shrink the
1134	<	* number of workers. The exact value does not matter too
1135	<	* much. It must be short enough to release resources during
1136	<	* sustained periods of idleness, but not so short that threads
1137	<	* are continually re-created.
1132	>	* Initial timeout value (in nanoseconds) for the thread
1133	>	* triggering quiescence to park waiting for new work. On timeout,
1134	>	* the thread will instead try to shrink the number of
1135	>	* workers. The value should be large enough to avoid overly
1136	>	* aggressive shrinkage during most transient stalls (long GCs
1137	>	* etc).
1138		*/
1139	<	private static final long SHRINK_RATE =
1123	<	4L * 1000L * 1000L * 1000L; // 4 seconds
1139	>	private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
1140
1141		/**
1142	<	* The timeout value for attempted shrinkage, includes
1127	<	* some slop to cope with system timer imprecision.
1142	>	* Timeout value when there are more threads than parallelism level
1143		*/
1144	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1144	>	private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
1145
1146		/**
1147		* The maximum stolen->joining link depth allowed in method
1148	<	* tryHelpStealer. Must be a power of two. This value also
1134	<	* controls the maximum number of times to try to help join a task
1135	<	* without any apparent progress or change in pool state before
1136	<	* giving up and blocking (see awaitJoin). Depths for legitimate
1148	>	* tryHelpStealer. Must be a power of two. Depths for legitimate
1149		* chains are unbounded, but we use a fixed constant to avoid
1150		* (otherwise unchecked) cycles and to bound staleness of
1151		* traversal parameters at the expense of sometimes blocking when
#	Line 1142 \| Line 1154 \| public class ForkJoinPool extends Abstra
1154		private static final int MAX_HELP = 64;
1155
1156		/**
1145	–	* Secondary time-based bound (in nanosecs) for helping attempts
1146	–	* before trying compensated blocking in awaitJoin. Used in
1147	–	* conjunction with MAX_HELP to reduce variance due to different
1148	–	* polling rates associated with different helping options. The
1149	–	* value should roughly approximate the time required to create
1150	–	* and/or activate a worker thread.
1151	–	*/
1152	–	private static final long COMPENSATION_DELAY = 1L << 18; // ~0.25 millisec
1153	–
1154	–	/**
1157		* Increment for seed generators. See class ThreadLocal for
1158		* explanation.
1159		*/
#	Line 1185 \| Line 1187 \| public class ForkJoinPool extends Abstra
1187		* scan for them to avoid queuing races. Note however that
1188		* eventCount updates lag releases so usage requires care.
1189		*
1190	<	* Field runState is an int packed with:
1190	>	* Field plock is an int packed with:
1191		* SHUTDOWN: true if shutdown is enabled (1 bit)
1192	<	* SEQ: a sequence number updated upon (de)registering workers (30 bits)
1193	<	* INIT: set true after workQueues array construction (1 bit)
1192	>	* SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
1193	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1194		*
1195		* The sequence number enables simple consistency checks:
1196		* Staleness of read-only operations on the workQueues array can
1197	<	* be checked by comparing runState before vs after the reads.
1197	>	* be checked by comparing plock before vs after the reads.
1198		*/
1199
1200		// bit positions/shifts for fields
#	Line 1204 \| Line 1206 \| public class ForkJoinPool extends Abstra
1206		// bounds
1207		private static final int SMASK = 0xffff; // short bits
1208		private static final int MAX_CAP = 0x7fff; // max #workers - 1
1209	<	private static final int SQMASK = 0xfffe; // even short bits
1209	>	private static final int EVENMASK = 0xfffe; // even short bits
1210	>	private static final int SQMASK = 0x007e; // max 64 (even) slots
1211		private static final int SHORT_SIGN = 1 << 15;
1212		private static final int INT_SIGN = 1 << 31;
1213
#	Line 1229 \| Line 1232 \| public class ForkJoinPool extends Abstra
1232		private static final int E_MASK = 0x7fffffff; // no STOP_BIT
1233		private static final int E_SEQ = 1 << EC_SHIFT;
1234
1235	<	// runState bits
1235	>	// plock bits
1236		private static final int SHUTDOWN = 1 << 31;
1237	+	private static final int PL_LOCK = 2;
1238	+	private static final int PL_SIGNAL = 1;
1239	+	private static final int PL_SPINS = 1 << 8;
1240
1241		// access mode for WorkQueue
1242		static final int LIFO_QUEUE = 0;
1243		static final int FIFO_QUEUE = 1;
1244		static final int SHARED_QUEUE = -1;
1245
1246	+	// bounds for #steps in scan loop -- must be power 2 minus 1
1247	+	private static final int MIN_SCAN = 0x1ff; // cover estimation slop
1248	+	private static final int MAX_SCAN = 0x1ffff; // 4 * max workers
1249	+
1250		// Instance fields
1251
1252		/*
1253	<	* Field layout order in this class tends to matter more than one
1254	<	* would like. Runtime layout order is only loosely related to
1253	>	* Field layout of this class tends to matter more than one would
1254	>	* like. Runtime layout order is only loosely related to
1255		* declaration order and may differ across JVMs, but the following
1256		* empirically works OK on current JVMs.
1257		*/
1258	<
1258	>	volatile long stealCount; // collects worker counts
1259		volatile long ctl; // main pool control
1260	<	final int parallelism; // parallelism level
1261	<	final int localMode; // per-worker scheduling mode
1262	<	final int submitMask; // submit queue index bound
1253	<	int nextSeed; // for initializing worker seeds
1254	<	volatile int runState; // shutdown status and seq
1260	>	volatile int plock; // shutdown status and seqLock
1261	>	volatile int indexSeed; // worker/submitter index seed
1262	>	final int config; // mode and parallelism level
1263		WorkQueue[] workQueues; // main registry
1264	<	final Mutex lock; // for registration
1257	<	final Condition termination; // for awaitTermination
1258	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1264	>	final ForkJoinWorkerThreadFactory factory;
1265		final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1260	–	final AtomicLong stealCount; // collect counts when terminated
1261	–	final AtomicInteger nextWorkerNumber; // to create worker name string
1266		final String workerNamePrefix; // to create worker name string
1267
1268	<	// Creating, registering, and deregistering workers
1268	>	/*
1269	>	* Acquires the plock lock to protect worker array and related
1270	>	* updates. This method is called only if an initial CAS on plock
1271	>	* fails. This acts as a spinLock for normal cases, but falls back
1272	>	* to builtin monitor to block when (rarely) needed. This would be
1273	>	* a terrible idea for a highly contended lock, but works fine as
1274	>	* a more conservative alternative to a pure spinlock. See
1275	>	* internal ConcurrentHashMap documentation for further
1276	>	* explanation of nearly the same construction.
1277	>	*/
1278	>	private int acquirePlock() {
1279	>	int spins = PL_SPINS, r = 0, ps, nps;
1280	>	for (;;) {
1281	>	if (((ps = plock) & PL_LOCK) == 0 &&
1282	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1283	>	return nps;
1284	>	else if (r == 0) { // randomize spins if possible
1285	>	Thread t = Thread.currentThread(); WorkQueue w; Submitter z;
1286	>	if ((t instanceof ForkJoinWorkerThread) &&
1287	>	(w = ((ForkJoinWorkerThread)t).workQueue) != null)
1288	>	r = w.seed;
1289	>	else if ((z = submitters.get()) != null)
1290	>	r = z.seed;
1291	>	else
1292	>	r = 1;
1293	>	}
1294	>	else if (spins >= 0) {
1295	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1296	>	if (r >= 0)
1297	>	--spins;
1298	>	}
1299	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps \| PL_SIGNAL)) {
1300	>	synchronized (this) {
1301	>	if ((plock & PL_SIGNAL) != 0) {
1302	>	try {
1303	>	wait();
1304	>	} catch (InterruptedException ie) {
1305	>	try {
1306	>	Thread.currentThread().interrupt();
1307	>	} catch (SecurityException ignore) {
1308	>	}
1309	>	}
1310	>	}
1311	>	else
1312	>	notifyAll();
1313	>	}
1314	>	}
1315	>	}
1316	>	}
1317	>
1318	>	/**
1319	>	* Unlocks and signals any thread waiting for plock. Called only
1320	>	* when CAS of seq value for unlock fails.
1321	>	*/
1322	>	private void releasePlock(int ps) {
1323	>	plock = ps;
1324	>	synchronized (this) { notifyAll(); }
1325	>	}
1326
1327		/**
1328	<	* Tries to create and start a worker
1328	>	* Tries to create and start a worker; adjusts counts etc on failure
1329		*/
1330		private void addWorker() {
1270	–	Throwable ex = null;
1331		ForkJoinWorkerThread wt = null;
1332		try {
1333	<	if ((wt = factory.newThread(this)) != null) {
1334	<	wt.start();
1335	<	return;
1276	<	}
1277	<	} catch (Throwable e) {
1278	<	ex = e;
1333	>	(wt = factory.newThread(this)).start();
1334	>	} catch (Throwable ex) {
1335	>	deregisterWorker(wt, ex); // adjust on failure
1336		}
1280	–	deregisterWorker(wt, ex); // adjust counts etc on failure
1337		}
1338
1339		/**
1340	<	* Callback from ForkJoinWorkerThread constructor to assign a
1341	<	* public name. This must be separate from registerWorker because
1342	<	* it is called during the "super" constructor call in
1343	<	* ForkJoinWorkerThread.
1344	<	*/
1345	<	final String nextWorkerName() {
1346	<	return workerNamePrefix.concat
1347	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1340	>	* Performs secondary initialization, called when plock is zero.
1341	>	* Creates workQueue array and sets plock to a valid value. The
1342	>	* lock body must be exception-free (so no try/finally) so we
1343	>	* optimistically allocate new array outside the lock and throw
1344	>	* away if (very rarely) not needed. (A similar tactic is used in
1345	>	* fullExternalPush.) Because the plock seq value can eventually
1346	>	* wrap around zero, this method harmlessly fails to reinitialize
1347	>	* if workQueues exists, while still advancing plock.
1348	>	*/
1349	>	private void initWorkQueuesArray() {
1350	>	WorkQueue[] ws; int ps;
1351	>	int p = config & SMASK; // find power of two table size
1352	>	int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots
1353	>	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
1354	>	WorkQueue[] nws = new WorkQueue[(n + 1) << 1];
1355	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1356	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1357	>	ps = acquirePlock();
1358	>	if ((ws = workQueues) == null \|\| ws.length == 0)
1359	>	workQueues = nws;
1360	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1361	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1362	>	releasePlock(nps);
1363	>	long c; int u;
1364	>	if ((u = (int)((c = ctl) >>> 32)) < 0 && (int)c == 0) {
1365	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1366	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1367	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1368	>	addWorker();
1369	>	}
1370	>
1371		}
1372
1373	+	// Registering and deregistering workers
1374	+
1375		/**
1376	<	* Callback from ForkJoinWorkerThread constructor to establish its
1377	<	* poolIndex and record its WorkQueue. To avoid scanning bias due
1378	<	* to packing entries in front of the workQueues array, we treat
1379	<	* the array as a simple power-of-two hash table using per-thread
1380	<	* seed as hash, expanding as needed.
1376	>	* Callback from ForkJoinWorkerThread to establish and record its
1377	>	* WorkQueue. To avoid scanning bias due to packing entries in
1378	>	* front of the workQueues array, we treat the array as a simple
1379	>	* power-of-two hash table using per-thread seed as hash,
1380	>	* expanding as needed.
1381		*
1382	<	* @param w the worker's queue
1382	>	* @param wt the worker thread
1383		*/
1384	<
1385	<	final void registerWorker(WorkQueue w) {
1386	<	Mutex lock = this.lock;
1387	<	lock.lock();
1388	<	try {
1389	<	WorkQueue[] ws = workQueues;
1390	<	if (w != null && ws != null) { // skip on shutdown/failure
1391	<	int rs, n = ws.length, m = n - 1;
1392	<	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1393	<	w.seed = (s == 0) ? 1 : s; // ensure non-zero seed
1394	<	int r = (s << 1) \| 1; // use odd-numbered indices
1395	<	if (ws[r &= m] != null) { // collision
1396	<	int probes = 0; // step by approx half size
1397	<	int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1398	<	while (ws[r = (r + step) & m] != null) {
1399	<	if (++probes >= n) {
1400	<	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1401	<	m = n - 1;
1402	<	probes = 0;
1384	>	final void registerWorker(ForkJoinWorkerThread wt) {
1385	>	if (wt != null && wt.workQueue == null) {
1386	>	int s, ps; // generate a rarely colliding candidate index seed
1387	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1388	>	s += SEED_INCREMENT) \|\|
1389	>	s == 0); // skip 0
1390	>	WorkQueue w = new WorkQueue(this, wt, config >>> 16, s);
1391	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1392	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1393	>	ps = acquirePlock();
1394	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1395	>	try {
1396	>	WorkQueue[] ws;
1397	>	if ((ws = workQueues) != null && wt.workQueue == null) {
1398	>	int n = ws.length, m = n - 1;
1399	>	int r = (s << 1) \| 1; // use odd-numbered indices
1400	>	if (ws[r &= m] != null) { // collision
1401	>	int probes = 0; // step by approx half size
1402	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1403	>	while (ws[r = (r + step) & m] != null) {
1404	>	if (++probes >= n) {
1405	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1406	>	m = n - 1;
1407	>	probes = 0;
1408	>	}
1409		}
1410		}
1411	+	w.eventCount = w.poolIndex = r; // volatile write orders
1412	+	wt.workQueue = ws[r] = w;
1413		}
1414	<	w.eventCount = w.poolIndex = r; // establish before recording
1415	<	ws[r] = w; // also update seq
1416	<	runState = ((rs = runState) & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN);
1414	>	} finally {
1415	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1416	>	releasePlock(nps);
1417		}
1329	–	} finally {
1330	–	lock.unlock();
1418		}
1419		}
1420
1421		/**
1422		* Final callback from terminating worker, as well as upon failure
1423	<	* to construct or start a worker in addWorker. Removes record of
1424	<	* worker from array, and adjusts counts. If pool is shutting
1425	<	* down, tries to complete termination.
1423	>	* to construct or start a worker. Removes record of worker from
1424	>	* array, and adjusts counts. If pool is shutting down, tries to
1425	>	* complete termination.
1426		*
1427	<	* @param wt the worker thread or null if addWorker failed
1427	>	* @param wt the worker thread or null if construction failed
1428		* @param ex the exception causing failure, or null if none
1429		*/
1430		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1344	–	Mutex lock = this.lock;
1431		WorkQueue w = null;
1432		if (wt != null && (w = wt.workQueue) != null) {
1433	<	w.runState = -1; // ensure runState is set
1434	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1435	<	int idx = w.poolIndex;
1436	<	lock.lock();
1437	<	try { // remove record from array
1433	>	int ps;
1434	>	w.qlock = -1; // ensure set
1435	>	long ns = w.nsteals, sc; // collect steal count
1436	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1437	>	sc = stealCount, sc + ns));
1438	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1439	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1440	>	ps = acquirePlock();
1441	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1442	>	try {
1443	>	int idx = w.poolIndex;
1444		WorkQueue[] ws = workQueues;
1445		if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1446		ws[idx] = null;
1447		} finally {
1448	<	lock.unlock();
1448	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1449	>	releasePlock(nps);
1450		}
1451		}
1452
#	Line 1366 \| Line 1459 \| public class ForkJoinPool extends Abstra
1459		if (!tryTerminate(false, false) && w != null) {
1460		w.cancelAll(); // cancel remaining tasks
1461		if (w.array != null) // suppress signal if never ran
1462	<	signalWork(); // wake up or create replacement
1462	>	helpSignal(null, 0); // wake up or create replacement
1463		if (ex == null) // help clean refs on way out
1464		ForkJoinTask.helpExpungeStaleExceptions();
1465		}
1466
1467		if (ex != null) // rethrow
1468	<	U.throwException(ex);
1468	>	ForkJoinTask.rethrow(ex);
1469		}
1470
1378	–
1471		// Submissions
1472
1473		/**
1474		* Unless shutting down, adds the given task to a submission queue
1475		* at submitter's current queue index (modulo submission
1476	<	* range). If no queue exists at the index, one is created. If
1477	<	* the queue is busy, another index is randomly chosen. The
1386	<	* submitMask bounds the effective number of queues to the
1387	<	* (nearest power of two for) parallelism level.
1476	>	* range). Only the most common path is directly handled in this
1477	>	* method. All others are relayed to fullExternalPush.
1478		*
1479		* @param task the task. Caller must ensure non-null.
1480		*/
1481	<	private void doSubmit(ForkJoinTask<?> task) {
1482	<	Submitter s = submitters.get();
1483	<	for (int r = s.seed, m = submitMask;;) {
1484	<	WorkQueue[] ws; WorkQueue q;
1485	<	int k = r & m & SQMASK; // use only even indices
1486	<	if (runState < 0 \|\| (ws = workQueues) == null \|\| ws.length <= k)
1487	<	throw new RejectedExecutionException(); // shutting down
1488	<	else if ((q = ws[k]) == null) { // create new queue
1489	<	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1490	<	Mutex lock = this.lock; // construct outside lock
1491	<	lock.lock();
1492	<	try { // recheck under lock
1493	<	int rs = runState; // to update seq
1404	<	if (ws == workQueues && ws[k] == null) {
1405	<	ws[k] = nq;
1406	<	runState = ((rs & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN));
1407	<	}
1408	<	} finally {
1409	<	lock.unlock();
1410	<	}
1411	<	}
1412	<	else if (q.trySharedPush(task)) {
1413	<	signalWork();
1481	>	final void externalPush(ForkJoinTask<?> task) {
1482	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1483	>	if ((z = submitters.get()) != null && plock > 0 &&
1484	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1485	>	(q = ws[m & z.seed & SQMASK]) != null &&
1486	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1487	>	int b = q.base, s = q.top, n, an;
1488	>	if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) {
1489	>	U.putObject(a, (long)(((an - 1) & s) << ASHIFT) + ABASE, task);
1490	>	q.top = s + 1; // push on to deque
1491	>	q.qlock = 0;
1492	>	if (n <= 2)
1493	>	signalWork(q, 0);
1494		return;
1495		}
1496	<	else if (m > 1) { // move to a different index
1496	>	q.qlock = 0;
1497	>	}
1498	>	fullExternalPush(task);
1499	>	}
1500	>
1501	>	/**
1502	>	* Full version of externalPush. This method is called, among
1503	>	* other times, upon the first submission of the first task to the
1504	>	* pool, so must perform secondary initialization (via
1505	>	* initWorkQueuesArray). It also detects first submission by an
1506	>	* external thread by looking up its ThreadLocal, and creates a
1507	>	* new shared queue if the one at index if empty or contended. The
1508	>	* lock body must be exception-free (so no try/finally) so we
1509	>	* optimistically allocate new queues outside the lock and throw
1510	>	* them away if (very rarely) not needed.
1511	>	*/
1512	>	private void fullExternalPush(ForkJoinTask<?> task) {
1513	>	int r = 0;
1514	>	for (Submitter z = submitters.get();;) {
1515	>	WorkQueue[] ws; WorkQueue q; int ps, m, k;
1516	>	if (z == null) {
1517	>	if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1518	>	r += SEED_INCREMENT) && r != 0)
1519	>	submitters.set(z = new Submitter(r));
1520	>	}
1521	>	else if (r == 0) { // move to a different index
1522	>	r = z.seed;
1523		r ^= r << 13; // same xorshift as WorkQueues
1524		r ^= r >>> 17;
1525	<	s.seed = r ^= r << 5;
1525	>	z.seed = r ^ (r << 5);
1526	>	}
1527	>	else if ((ps = plock) < 0)
1528	>	throw new RejectedExecutionException();
1529	>	else if (ps == 0 \|\| (ws = workQueues) == null \|\|
1530	>	(m = ws.length - 1) < 0)
1531	>	initWorkQueuesArray();
1532	>	else if ((q = ws[k = r & m & SQMASK]) != null) {
1533	>	if (q.trySharedPush(task))
1534	>	return;
1535	>	else
1536	>	r = 0; // move on contention
1537	>	}
1538	>	else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1539	>	q = new WorkQueue(this, null, SHARED_QUEUE, r);
1540	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1541	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1542	>	ps = acquirePlock();
1543	>	if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1544	>	ws[k] = q;
1545	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1546	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1547	>	releasePlock(nps);
1548		}
1549		else
1550	<	Thread.yield(); // yield if no alternatives
1550	>	r = 0; // try elsewhere while lock held
1551		}
1552		}
1553
#	Line 1434 \| Line 1562 \| public class ForkJoinPool extends Abstra
1562		}
1563
1564		/**
1565	<	* Tries to activate or create a worker if too few are active.
1566	<	*/
1567	<	final void signalWork() {
1568	<	long c; int u;
1569	<	while ((u = (int)((c = ctl) >>> 32)) < 0) { // too few active
1570	<	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1571	<	if ((e = (int)c) > 0) { // at least one waiting
1572	<	if (ws != null && (i = e & SMASK) < ws.length &&
1565	>	* Tries to create (at most one) or activate (possibly several)
1566	>	* workers if too few are active. On contention failure, continues
1567	>	* until at least one worker is signalled or the given queue is
1568	>	* empty or all workers are active.
1569	>	*
1570	>	* @param q if non-null, the queue holding tasks to be signalled
1571	>	* @param signals the target number of signals (at least one --
1572	>	* if argument is zero also sets signallee hint if parked).
1573	>	*/
1574	>	final void signalWork(WorkQueue q, int signals) {
1575	>	long c; int e, u, i, s; WorkQueue[] ws; WorkQueue w; Thread p;
1576	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1577	>	if ((e = (int)c) > 0) {
1578	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1579		(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1580		long nc = (((long)(w.nextWait & E_MASK)) \|
1581		((long)(u + UAC_UNIT) << 32));
1582		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1583		w.eventCount = (e + E_SEQ) & E_MASK;
1584	<	if ((p = w.parker) != null)
1585	<	U.unpark(p); // activate and release
1586	<	break;
1584	>	if ((p = w.parker) != null) {
1585	>	if (q != null && signals == 0)
1586	>	w.hint = q.poolIndex;
1587	>	U.unpark(p);
1588	>	}
1589	>	if (--signals <= 0)
1590	>	break;
1591		}
1592	+	if (q != null && (s = q.queueSize()) <= signals &&
1593	+	(signals = s) <= 0)
1594	+	break;
1595		}
1596		else
1597		break;
1598		}
1599	<	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1599	>	else if (e == 0 && (u & SHORT_SIGN) != 0) {
1600		long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1601		((u + UAC_UNIT) & UAC_MASK)) << 32;
1602		if (U.compareAndSwapLong(this, CTL, c, nc)) {
#	Line 1474 \| Line 1615 \| public class ForkJoinPool extends Abstra
1615		* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1616		*/
1617		final void runWorker(WorkQueue w) {
1618	<	w.growArray(false); // initialize queue array in this thread
1619	<	do { w.runTask(scan(w)); } while (w.runState >= 0);
1618	>	if (w != null) // skip on initialization failure
1619	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1620		}
1621
1622		/**
#	Line 1486 \| Line 1627 \| public class ForkJoinPool extends Abstra
1627		* contention, or state changes that indicate possible success on
1628		* re-invocation.
1629		*
1630	<	* The scan searches for tasks across a random permutation of
1631	<	* queues (starting at a random index and stepping by a random
1632	<	* relative prime, checking each at least once). The scan
1633	<	* terminates upon either finding a non-empty queue, or completing
1634	<	* the sweep. If the worker is not inactivated, it takes and
1635	<	* returns a task from this queue. On failure to find a task, we
1636	<	* take one of the following actions, after which the caller will
1637	<	* retry calling this method unless terminated.
1630	>	* The scan searches for tasks across queues (starting at a random
1631	>	* index, and relying on registerWorker to irregularly scatter
1632	>	* them within array to avoid bias), checking each at least twice.
1633	>	* The scan terminates upon either finding a non-empty queue, or
1634	>	* completing the sweep. If the worker is not inactivated, it
1635	>	* takes and returns a task from this queue. Otherwise, if not
1636	>	* activated, it signals workers (that may include itself) and
1637	>	* returns so caller can retry. Also returns for true if the
1638	>	* worker array may have changed during an empty scan. On failure
1639	>	* to find a task, we take one of the following actions, after
1640	>	* which the caller will retry calling this method unless
1641	>	* terminated.
1642		*
1643		* * If pool is terminating, terminate the worker.
1644		*
1500	–	* * If not a complete sweep, try to release a waiting worker. If
1501	–	* the scan terminated because the worker is inactivated, then the
1502	–	* released worker will often be the calling worker, and it can
1503	–	* succeed obtaining a task on the next call. Or maybe it is
1504	–	* another worker, but with same net effect. Releasing in other
1505	–	* cases as well ensures that we have enough workers running.
1506	–	*
1645		* * If not already enqueued, try to inactivate and enqueue the
1646		* worker on wait queue. Or, if inactivating has caused the pool
1647		* to be quiescent, relay to idleAwaitWork to check for
1648		* termination and possibly shrink pool.
1649		*
1650	<	* * If already inactive, and the caller has run a task since the
1651	<	* last empty scan, return (to allow rescan) unless others are
1652	<	* also inactivated. Field WorkQueue.rescans counts down on each
1515	<	* scan to ensure eventual inactivation and blocking.
1516	<	*
1517	<	* * If already enqueued and none of the above apply, park
1518	<	* awaiting signal,
1650	>	* * If already enqueued and none of the above apply, possibly
1651	>	* (with 1/2 probability) park awaiting signal, else lingering to
1652	>	* help scan and signal.
1653		*
1654		* @param w the worker (via its WorkQueue)
1655	<	* @return a task or null of none found
1655	>	* @return a task or null if none found
1656		*/
1657		private final ForkJoinTask<?> scan(WorkQueue w) {
1658	<	WorkQueue[] ws; // first update random seed
1659	<	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1660	<	int rs = runState, m; // volatile read order matters
1661	<	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1662	<	int ec = w.eventCount; // ec is negative if inactive
1663	<	int step = (r >>> 16) \| 1; // relative prime
1664	<	for (int j = (m + 1) << 2; ; r += step) {
1665	<	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1666	<	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1533	<	(a = q.array) != null) { // probably nonempty
1658	>	WorkQueue[] ws; int m, hint;
1659	>	int ps = plock; // read plock before ws
1660	>	if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1661	>	int ec = w.eventCount; // ec is negative if inactive
1662	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1663	>	for (int j = ((m + m + 1) \| MIN_SCAN) & MAX_SCAN; ; --j) {
1664	>	WorkQueue q; ForkJoinTask<?>[] a; int b;
1665	>	if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 &&
1666	>	(a = q.array) != null) { // probably nonempty
1667		int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1668	<	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1668	>	ForkJoinTask<?> t = (ForkJoinTask<?>)
1669	>	U.getObjectVolatile(a, i);
1670		if (q.base == b && ec >= 0 && t != null &&
1671		U.compareAndSwapObject(a, i, t, null)) {
1672	<	if (q.top - (q.base = b + 1) > 1)
1673	<	signalWork(); // help pushes signal
1674	<	return t;
1672	>	if ((q.base = b + 1) - q.top < 0)
1673	>	signalWork(q, 0);
1674	>	return t; // taken
1675		}
1676	<	else if (ec < 0 \|\| j <= m) {
1677	<	rs = 0; // mark scan as imcomplete
1678	<	break; // caller can retry after release
1676	>	else if (ec < 0 \|\| j < m) { // cannot take or cannot rescan
1677	>	w.hint = q.poolIndex; // use hint below
1678	>	break; // let caller retry after signal
1679		}
1680		}
1681	<	if (--j < 0)
1682	<	break;
1683	<	}
1684	<
1685	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1686	<	if (e < 0) // decode ctl on empty scan
1553	<	w.runState = -1; // pool is terminating
1554	<	else if (rs == 0 \|\| rs != runState) { // incomplete scan
1555	<	WorkQueue v; Thread p; // try to release a waiter
1556	<	if (e > 0 && a < 0 && w.eventCount == ec &&
1557	<	(v = ws[e & m]) != null && v.eventCount == (e \| INT_SIGN)) {
1558	<	long nc = ((long)(v.nextWait & E_MASK) \|
1559	<	((c + AC_UNIT) & (AC_MASK\|TC_MASK)));
1560	<	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1561	<	v.eventCount = (e + E_SEQ) & E_MASK;
1562	<	if ((p = v.parker) != null)
1563	<	U.unpark(p);
1681	>	else if (j < 0) { // end of scan; in loop to simplify code
1682	>	long c, sc; int e, ns;
1683	>	if ((ns = w.nsteals) != 0) {
1684	>	if (U.compareAndSwapLong(this, STEALCOUNT,
1685	>	sc = stealCount, sc + ns))
1686	>	w.nsteals = 0; // collect steals
1687		}
1688	+	else if (plock != ps) // ws may have changed
1689	+	break;
1690	+	else if ((e = (int)(c = ctl)) < 0)
1691	+	w.qlock = -1; // pool is terminating
1692	+	else if (ec >= 0) { // try to enqueue/inactivate
1693	+	long nc = ((long)ec \|
1694	+	((c - AC_UNIT) & (AC_MASK\|TC_MASK)));
1695	+	w.nextWait = e; // link and mark inactive
1696	+	w.hint = -1; // use hint if set while parked
1697	+	w.eventCount = ec \| INT_SIGN;
1698	+	if (ctl != c \|\|
1699	+	!U.compareAndSwapLong(this, CTL, c, nc))
1700	+	w.eventCount = ec; // unmark on CAS failure
1701	+	else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK))
1702	+	idleAwaitWork(w, nc, c);
1703	+	}
1704	+	else if (w.eventCount < 0) { // block
1705	+	Thread wt = Thread.currentThread();
1706	+	Thread.interrupted(); // clear status
1707	+	U.putObject(wt, PARKBLOCKER, this);
1708	+	w.parker = wt; // emulate LockSupport.park
1709	+	if (w.eventCount < 0) // recheck
1710	+	U.park(false, 0L);
1711	+	w.parker = null;
1712	+	U.putObject(wt, PARKBLOCKER, null);
1713	+	}
1714	+	break;
1715		}
1716		}
1717	<	else if (ec >= 0) { // try to enqueue/inactivate
1718	<	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1719	<	w.nextWait = e;
1720	<	w.eventCount = ec \| INT_SIGN; // mark as inactive
1721	<	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1722	<	w.eventCount = ec; // unmark on CAS failure
1573	<	else {
1574	<	if ((ns = w.nsteals) != 0) {
1575	<	w.nsteals = 0; // set rescans if ran task
1576	<	w.rescans = (a > 0) ? 0 : a + parallelism;
1577	<	w.totalSteals += ns;
1578	<	}
1579	<	if (a == 1 - parallelism) // quiescent
1580	<	idleAwaitWork(w, nc, c);
1581	<	}
1582	<	}
1583	<	else if (w.eventCount < 0) { // already queued
1584	<	if ((nr = w.rescans) > 0) { // continue rescanning
1585	<	int ac = a + parallelism;
1586	<	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0)
1587	<	Thread.yield(); // yield before block
1588	<	}
1589	<	else {
1590	<	Thread.interrupted(); // clear status
1591	<	Thread wt = Thread.currentThread();
1592	<	U.putObject(wt, PARKBLOCKER, this);
1593	<	w.parker = wt; // emulate LockSupport.park
1594	<	if (w.eventCount < 0) // recheck
1595	<	U.park(false, 0L);
1596	<	w.parker = null;
1597	<	U.putObject(wt, PARKBLOCKER, null);
1598	<	}
1717	>	if ((hint = w.hint) >= 0) { // help signal
1718	>	WorkQueue[] vs; WorkQueue v; int k;
1719	>	w.hint = -1; // suppress resignal
1720	>	if ((vs = workQueues) != null && hint < vs.length &&
1721	>	(v = vs[hint]) != null && (k = v.base - v.top) < -1)
1722	>	signalWork(v, 1 - k);
1723		}
1724		}
1725		return null;
#	Line 1604 \| Line 1728 \| public class ForkJoinPool extends Abstra
1728		/**
1729		* If inactivating worker w has caused the pool to become
1730		* quiescent, checks for pool termination, and, so long as this is
1731	<	* not the only worker, waits for event for up to SHRINK_RATE
1732	<	* nanosecs. On timeout, if ctl has not changed, terminates the
1731	>	* not the only worker, waits for event for up to a given
1732	>	* duration. On timeout, if ctl has not changed, terminates the
1733		* worker, which will in turn wake up another worker to possibly
1734		* repeat this process.
1735		*
#	Line 1614 \| Line 1738 \| public class ForkJoinPool extends Abstra
1738		* @param prevCtl the ctl value to restore if thread is terminated
1739		*/
1740		private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1741	<	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1742	<	(int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1741	>	if (w != null && w.eventCount < 0 &&
1742	>	!tryTerminate(false, false) && (int)prevCtl != 0) {
1743	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1744	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1745	>	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1746		Thread wt = Thread.currentThread();
1620	–	Thread.yield(); // yield before block
1747		while (ctl == currentCtl) {
1622	–	long startTime = System.nanoTime();
1748		Thread.interrupted(); // timed variant of version in scan()
1749		U.putObject(wt, PARKBLOCKER, this);
1750		w.parker = wt;
1751		if (ctl == currentCtl)
1752	<	U.park(false, SHRINK_RATE);
1752	>	U.park(false, parkTime);
1753		w.parker = null;
1754		U.putObject(wt, PARKBLOCKER, null);
1755		if (ctl != currentCtl)
1756		break;
1757	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1757	>	if (deadline - System.nanoTime() <= 0L &&
1758		U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1759		w.eventCount = (w.eventCount + E_SEQ) \| E_MASK;
1760	<	w.runState = -1; // shrink
1760	>	w.qlock = -1; // shrink
1761	>	w.hint = -1; // suppress helping
1762		break;
1763		}
1764		}
#	Line 1640 \| Line 1766 \| public class ForkJoinPool extends Abstra
1766		}
1767
1768		/**
1769	+	* Scans through queues looking for work (optionally, while
1770	+	* joining a task); if any are present, signals. May return early
1771	+	* if more signalling is detectably unneeded.
1772	+	*
1773	+	* @param task if non-null, return early if done
1774	+	* @param origin an index to start scan
1775	+	*/
1776	+	final int helpSignal(ForkJoinTask<?> task, int origin) {
1777	+	WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1778	+	if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1779	+	for (int i = 0; i <= m; ++i) {
1780	+	if (task != null && (s = task.status) < 0)
1781	+	return s;
1782	+	if ((q = ws[(i + origin) & m]) != null &&
1783	+	(n = q.queueSize()) > 0) {
1784	+	signalWork(q, n);
1785	+	if ((u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0)
1786	+	break;
1787	+	}
1788	+	}
1789	+	}
1790	+	return 0;
1791	+	}
1792	+
1793	+	/**
1794		* Tries to locate and execute tasks for a stealer of the given
1795		* task, or in turn one of its stealers, Traces currentSteal ->
1796		* currentJoin links looking for a thread working on a descendant
#	Line 1670 \| Line 1821 \| public class ForkJoinPool extends Abstra
1821		}
1822		if ((ws = workQueues) == null \|\| (m = ws.length - 1) <= 0)
1823		break restart; // shutting down
1824	<	if ((v = ws[h = (j.stealHint \| 1) & m]) == null \|\|
1824	>	if ((v = ws[h = (j.hint \| 1) & m]) == null \|\|
1825		v.currentSteal != subtask) {
1826		for (int origin = h;;) { // find stealer
1827		if (((h = (h + 2) & m) & 15) == 1 &&
#	Line 1678 \| Line 1829 \| public class ForkJoinPool extends Abstra
1829		continue restart; // occasional staleness check
1830		if ((v = ws[h]) != null &&
1831		v.currentSteal == subtask) {
1832	<	j.stealHint = h; // save hint
1832	>	j.hint = h; // save hint
1833		break;
1834		}
1835		if (h == origin)
#	Line 1726 \| Line 1877 \| public class ForkJoinPool extends Abstra
1877		}
1878
1879		/**
1880	<	* If task is at base of some steal queue, steals and executes it.
1880	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1881	>	* and run tasks within the target's computation.
1882	>	*
1883	>	* @param task the task to join
1884	>	* @param mode if shared, exit upon completing any task
1885	>	* if all workers are active
1886		*
1731	–	* @param joiner the joining worker
1732	–	* @param task the task
1887		*/
1888	<	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1889	<	WorkQueue[] ws;
1890	<	if ((ws = workQueues) != null) {
1891	<	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1892	<	WorkQueue q = ws[j];
1893	<	if (q != null && q.pollFor(task)) {
1894	<	joiner.runSubtask(task);
1895	<	break;
1888	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1889	>	WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1890	>	if (task != null && (ws = workQueues) != null &&
1891	>	(m = ws.length - 1) >= 0) {
1892	>	for (int j = 1, origin = j;;) {
1893	>	if ((s = task.status) < 0)
1894	>	return s;
1895	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1896	>	origin = j;
1897	>	if (mode == SHARED_QUEUE &&
1898	>	((u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0))
1899	>	break;
1900		}
1901	+	else if ((j = (j + 2) & m) == origin)
1902	+	break;
1903		}
1904		}
1905	+	return 0;
1906		}
1907
1908		/**
1909		* Tries to decrement active count (sometimes implicitly) and
1910		* possibly release or create a compensating worker in preparation
1911		* for blocking. Fails on contention or termination. Otherwise,
1912	<	* adds a new thread if no idle workers are available and either
1913	<	* pool would become completely starved or: (at least half
1753	<	* starved, and fewer than 50% spares exist, and there is at least
1754	<	* one task apparently available). Even though the availability
1755	<	* check requires a full scan, it is worthwhile in reducing false
1756	<	* alarms.
1757	<	*
1758	<	* @param task if non-null, a task being waited for
1759	<	* @param blocker if non-null, a blocker being waited for
1760	<	* @return true if the caller can block, else should recheck and retry
1912	>	* adds a new thread if no idle workers are available and pool
1913	>	* may become starved.
1914		*/
1915	<	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1916	<	int pc = parallelism, e;
1917	<	long c = ctl;
1918	<	WorkQueue[] ws = workQueues;
1919	<	if ((e = (int)c) >= 0 && ws != null) {
1920	<	int u, a, ac, hc;
1921	<	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1922	<	boolean replace = false;
1923	<	if ((a = u >> UAC_SHIFT) <= 0) {
1924	<	if ((ac = a + pc) <= 1)
1925	<	replace = true;
1926	<	else if ((e > 0 \|\| (task != null &&
1927	<	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1775	<	WorkQueue w;
1776	<	for (int j = 0; j < ws.length; ++j) {
1777	<	if ((w = ws[j]) != null && !w.isEmpty()) {
1778	<	replace = true;
1779	<	break; // in compensation range and tasks available
1780	<	}
1781	<	}
1915	>	final boolean tryCompensate() {
1916	>	int pc = config & SMASK, e, i, tc; long c;
1917	>	WorkQueue[] ws; WorkQueue w; Thread p;
1918	>	if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) {
1919	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1920	>	(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1921	>	long nc = ((long)(w.nextWait & E_MASK) \|
1922	>	(c & (AC_MASK\|TC_MASK)));
1923	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1924	>	w.eventCount = (e + E_SEQ) & E_MASK;
1925	>	if ((p = w.parker) != null)
1926	>	U.unpark(p);
1927	>	return true; // replace with idle worker
1928		}
1929		}
1930	<	if ((task == null \|\| task.status >= 0) && // recheck need to block
1931	<	(blocker == null \|\| !blocker.isReleasable()) && ctl == c) {
1932	<	if (!replace) { // no compensation
1933	<	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1934	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1935	<	return true;
1936	<	}
1937	<	else if (e != 0) { // release an idle worker
1938	<	WorkQueue w; Thread p; int i;
1939	<	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
1940	<	long nc = ((long)(w.nextWait & E_MASK) \|
1795	<	(c & (AC_MASK\|TC_MASK)));
1796	<	if (w.eventCount == (e \| INT_SIGN) &&
1797	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1798	<	w.eventCount = (e + E_SEQ) & E_MASK;
1799	<	if ((p = w.parker) != null)
1800	<	U.unpark(p);
1801	<	return true;
1802	<	}
1803	<	}
1804	<	}
1805	<	else if (tc < MAX_CAP) { // create replacement
1806	<	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1807	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1808	<	addWorker();
1809	<	return true;
1810	<	}
1930	>	else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1931	>	(int)(c >> AC_SHIFT) + pc > 1) {
1932	>	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1933	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1934	>	return true; // no compensation
1935	>	}
1936	>	else if (tc + pc < MAX_CAP) {
1937	>	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1938	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1939	>	addWorker();
1940	>	return true;
1941		}
1942		}
1943		}
#	Line 1822 \| Line 1952 \| public class ForkJoinPool extends Abstra
1952		* @return task status on exit
1953		*/
1954		final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1955	<	int s;
1956	<	if ((s = task.status) >= 0) {
1955	>	int s = 0;
1956	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1957		ForkJoinTask<?> prevJoin = joiner.currentJoin;
1958		joiner.currentJoin = task;
1959	<	long startTime = 0L;
1960	<	for (int k = 0;;) {
1961	<	if ((s = (joiner.isEmpty() ? // try to help
1962	<	tryHelpStealer(joiner, task) :
1963	<	joiner.tryRemoveAndExec(task))) == 0 &&
1959	>	do {} while ((s = task.status) >= 0 &&
1960	>	joiner.queueSize() > 0 &&
1961	>	joiner.tryRemoveAndExec(task)); // process local tasks
1962	>	if (s >= 0 && (s = task.status) >= 0 &&
1963	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1964	>	(task instanceof CountedCompleter))
1965	>	s = helpComplete(task, LIFO_QUEUE);
1966	>	int k = 0; // to perform pre-block yield for politeness
1967	>	while (s >= 0 && (s = task.status) >= 0) {
1968	>	if ((joiner.queueSize() > 0 \|\| // try helping
1969	>	(s = tryHelpStealer(joiner, task)) == 0) &&
1970		(s = task.status) >= 0) {
1971	<	if (k == 0) {
1972	<	startTime = System.nanoTime();
1973	<	tryPollForAndExec(joiner, task); // check uncommon case
1974	<	}
1975	<	else if ((k & (MAX_HELP - 1)) == 0 &&
1976	<	System.nanoTime() - startTime >=
1977	<	COMPENSATION_DELAY &&
1978	<	tryCompensate(task, null)) {
1979	<	if (task.trySetSignal()) {
1971	>	if (k < 3) {
1972	>	if (++k < 3)
1973	>	s = helpSignal(task, joiner.poolIndex);
1974	>	else
1975	>	Thread.yield();
1976	>	}
1977	>	else if (!tryCompensate())
1978	>	k = 0;
1979	>	else {
1980	>	if (task.trySetSignal() && (s = task.status) >= 0) {
1981		synchronized (task) {
1982		if (task.status >= 0) {
1983		try { // see ForkJoinTask
#	Line 1857 \| Line 1994 \| public class ForkJoinPool extends Abstra
1994		(this, CTL, c = ctl, c + AC_UNIT));
1995		}
1996		}
1860	–	if (s < 0 \|\| (s = task.status) < 0) {
1861	–	joiner.currentJoin = prevJoin;
1862	–	break;
1863	–	}
1864	–	else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
1865	–	Thread.yield(); // for politeness
1997		}
1998	+	joiner.currentJoin = prevJoin;
1999		}
2000		return s;
2001		}
#	Line 1875 \| Line 2007 \| public class ForkJoinPool extends Abstra
2007		*
2008		* @param joiner the joining worker
2009		* @param task the task
1878	–	* @return task status on exit
2010		*/
2011	<	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2011	>	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2012		int s;
2013	<	while ((s = task.status) >= 0 &&
2014	<	(joiner.isEmpty() ?
2015	<	tryHelpStealer(joiner, task) :
2016	<	joiner.tryRemoveAndExec(task)) != 0)
2017	<	;
2018	<	return s;
2013	>	if (joiner != null && task != null && (s = task.status) >= 0) {
2014	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2015	>	joiner.currentJoin = task;
2016	>	do {} while ((s = task.status) >= 0 &&
2017	>	joiner.queueSize() > 0 &&
2018	>	joiner.tryRemoveAndExec(task));
2019	>	if (s >= 0 && (s = task.status) >= 0 &&
2020	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
2021	>	(task instanceof CountedCompleter))
2022	>	s = helpComplete(task, LIFO_QUEUE);
2023	>	if (s >= 0 && joiner.queueSize() == 0) {
2024	>	do {} while (task.status >= 0 &&
2025	>	tryHelpStealer(joiner, task) > 0);
2026	>	}
2027	>	joiner.currentJoin = prevJoin;
2028	>	}
2029		}
2030
2031		/**
#	Line 1892 \| Line 2033 \| public class ForkJoinPool extends Abstra
2033		* during a random, then cyclic scan, else null. This method must
2034		* be retried by caller if, by the time it tries to use the queue,
2035		* it is empty.
2036	+	* @param r a (random) seed for scanning
2037		*/
2038	<	private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
1897	<	// Similar to loop in scan(), but ignoring submissions
1898	<	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1899	<	int step = (r >>> 16) \| 1;
2038	>	private WorkQueue findNonEmptyStealQueue(int r) {
2039		for (WorkQueue[] ws;;) {
2040	<	int rs = runState, m;
2040	>	int ps = plock, m, n;
2041		if ((ws = workQueues) == null \|\| (m = ws.length - 1) < 1)
2042		return null;
2043	<	for (int j = (m + 1) << 2; ; r += step) {
2044	<	WorkQueue q = ws[((r << 1) \| 1) & m];
2045	<	if (q != null && !q.isEmpty())
2043	>	for (int j = (m + 1) << 2; ;) {
2044	>	WorkQueue q = ws[(((r + j) << 1) \| 1) & m];
2045	>	if (q != null && (n = q.queueSize()) > 0) {
2046	>	if (n > 1)
2047	>	signalWork(q, 0);
2048		return q;
2049	+	}
2050		else if (--j < 0) {
2051	<	if (runState == rs)
2051	>	if (plock == ps)
2052		return null;
2053		break;
2054		}
#	Line 1914 \| Line 2056 \| public class ForkJoinPool extends Abstra
2056		}
2057		}
2058
1917	–
2059		/**
2060		* Runs tasks until {@code isQuiescent()}. We piggyback on
2061		* active count ctl maintenance, but rather than blocking
#	Line 1926 \| Line 2067 \| public class ForkJoinPool extends Abstra
2067		ForkJoinTask<?> localTask; // exhaust local queue
2068		while ((localTask = w.nextLocalTask()) != null)
2069		localTask.doExec();
2070	<	WorkQueue q = findNonEmptyStealQueue(w);
2070	>	// Similar to loop in scan(), but ignoring submissions
2071	>	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2072		if (q != null) {
2073		ForkJoinTask<?> t; int b;
2074		if (!active) { // re-establish active count
#	Line 1947 \| Line 2089 \| public class ForkJoinPool extends Abstra
2089		}
2090		else
2091		c = ctl; // re-increment on exit
2092	<	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2092	>	if ((int)(c >> AC_SHIFT) + (config & SMASK) == 0) {
2093		do {} while (!U.compareAndSwapLong
2094		(this, CTL, c = ctl, c + AC_UNIT));
2095		break;
#	Line 1966 \| Line 2108 \| public class ForkJoinPool extends Abstra
2108		WorkQueue q; int b;
2109		if ((t = w.nextLocalTask()) != null)
2110		return t;
2111	<	if ((q = findNonEmptyStealQueue(w)) == null)
2111	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2112		return null;
2113		if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2114		return t;
#	Line 1974 \| Line 2116 \| public class ForkJoinPool extends Abstra
2116		}
2117
2118		/**
2119	<	* Returns the approximate (non-atomic) number of idle threads per
2120	<	* active thread to offset steal queue size for method
2121	<	* ForkJoinTask.getSurplusQueuedTaskCount().
2122	<	*/
2123	<	final int idlePerActive() {
2124	<	// Approximate at powers of two for small values, saturate past 4
2125	<	int p = parallelism;
2126	<	int a = p + (int)(ctl >> AC_SHIFT);
2127	<	return (a > (p >>>= 1) ? 0 :
2128	<	a > (p >>>= 1) ? 1 :
2129	<	a > (p >>>= 1) ? 2 :
2130	<	a > (p >>>= 1) ? 4 :
2131	<	8);
2119	>	* Returns a cheap heuristic guide for task partitioning when
2120	>	* programmers, frameworks, tools, or languages have little or no
2121	>	* idea about task granularity. In essence by offering this
2122	>	* method, we ask users only about tradeoffs in overhead vs
2123	>	* expected throughput and its variance, rather than how finely to
2124	>	* partition tasks.
2125	>	*
2126	>	* In a steady state strict (tree-structured) computation, each
2127	>	* thread makes available for stealing enough tasks for other
2128	>	* threads to remain active. Inductively, if all threads play by
2129	>	* the same rules, each thread should make available only a
2130	>	* constant number of tasks.
2131	>	*
2132	>	* The minimum useful constant is just 1. But using a value of 1
2133	>	* would require immediate replenishment upon each steal to
2134	>	* maintain enough tasks, which is infeasible. Further,
2135	>	* partitionings/granularities of offered tasks should minimize
2136	>	* steal rates, which in general means that threads nearer the top
2137	>	* of computation tree should generate more than those nearer the
2138	>	* bottom. In perfect steady state, each thread is at
2139	>	* approximately the same level of computation tree. However,
2140	>	* producing extra tasks amortizes the uncertainty of progress and
2141	>	* diffusion assumptions.
2142	>	*
2143	>	* So, users will want to use values larger, but not much larger
2144	>	* than 1 to both smooth over transient shortages and hedge
2145	>	* against uneven progress; as traded off against the cost of
2146	>	* extra task overhead. We leave the user to pick a threshold
2147	>	* value to compare with the results of this call to guide
2148	>	* decisions, but recommend values such as 3.
2149	>	*
2150	>	* When all threads are active, it is on average OK to estimate
2151	>	* surplus strictly locally. In steady-state, if one thread is
2152	>	* maintaining say 2 surplus tasks, then so are others. So we can
2153	>	* just use estimated queue length. However, this strategy alone
2154	>	* leads to serious mis-estimates in some non-steady-state
2155	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2156	>	* many of these by further considering the number of "idle"
2157	>	* threads, that are known to have zero queued tasks, so
2158	>	* compensate by a factor of (#idle/#active) threads.
2159	>	*
2160	>	* Note: The approximation of #busy workers as #active workers is
2161	>	* not very good under current signalling scheme, and should be
2162	>	* improved.
2163	>	*/
2164	>	static int getSurplusQueuedTaskCount() {
2165	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2166	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2167	>	int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK;
2168	>	int n = (q = wt.workQueue).top - q.base;
2169	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2170	>	return n - (a > (p >>>= 1) ? 0 :
2171	>	a > (p >>>= 1) ? 1 :
2172	>	a > (p >>>= 1) ? 2 :
2173	>	a > (p >>>= 1) ? 4 :
2174	>	8);
2175	>	}
2176	>	return 0;
2177		}
2178
2179		// Termination
#	Line 2006 \| Line 2193 \| public class ForkJoinPool extends Abstra
2193		* @return true if now terminating or terminated
2194		*/
2195		private boolean tryTerminate(boolean now, boolean enable) {
2196	<	Mutex lock = this.lock;
2196	>	if (this == commonPool) // cannot shut down
2197	>	return false;
2198		for (long c;;) {
2199		if (((c = ctl) & STOP_BIT) != 0) { // already terminating
2200	<	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2201	<	lock.lock(); // don't need try/finally
2202	<	termination.signalAll(); // signal when 0 workers
2203	<	lock.unlock();
2200	>	if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) {
2201	>	synchronized (this) {
2202	>	notifyAll(); // signal when 0 workers
2203	>	}
2204		}
2205		return true;
2206		}
2207	<	if (runState >= 0) { // not yet enabled
2207	>	if (plock >= 0) { // not yet enabled
2208	>	int ps;
2209		if (!enable)
2210		return false;
2211	<	lock.lock();
2212	<	runState \|= SHUTDOWN;
2213	<	lock.unlock();
2211	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
2212	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2213	>	ps = acquirePlock();
2214	>	int nps = SHUTDOWN;
2215	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2216	>	releasePlock(nps);
2217		}
2218		if (!now) { // check if idle & no tasks
2219	<	if ((int)(c >> AC_SHIFT) != -parallelism \|\|
2219	>	if ((int)(c >> AC_SHIFT) != -(config & SMASK) \|\|
2220		hasQueuedSubmissions())
2221		return false;
2222		// Check for unqueued inactive workers. One pass suffices.
#	Line 2044 \| Line 2236 \| public class ForkJoinPool extends Abstra
2236		int n = ws.length;
2237		for (int i = 0; i < n; ++i) {
2238		if ((w = ws[i]) != null) {
2239	<	w.runState = -1;
2239	>	w.qlock = -1;
2240		if (pass > 0) {
2241		w.cancelAll();
2242		if (pass > 1)
#	Line 2063 \| Line 2255 \| public class ForkJoinPool extends Abstra
2255		if (w.eventCount == (e \| INT_SIGN) &&
2256		U.compareAndSwapLong(this, CTL, cc, nc)) {
2257		w.eventCount = (e + E_SEQ) & E_MASK;
2258	<	w.runState = -1;
2258	>	w.qlock = -1;
2259		if ((p = w.parker) != null)
2260		U.unpark(p);
2261		}
#	Line 2074 \| Line 2266 \| public class ForkJoinPool extends Abstra
2266		}
2267		}
2268
2269	+	// external operations on common pool
2270	+
2271	+	/**
2272	+	* Returns common pool queue for a thread that has submitted at
2273	+	* least one task.
2274	+	*/
2275	+	static WorkQueue commonSubmitterQueue() {
2276	+	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2277	+	return ((z = submitters.get()) != null &&
2278	+	(p = commonPool) != null &&
2279	+	(ws = p.workQueues) != null &&
2280	+	(m = ws.length - 1) >= 0) ?
2281	+	ws[m & z.seed & SQMASK] : null;
2282	+	}
2283	+
2284	+	/**
2285	+	* Tries to pop the given task from submitter's queue in common pool.
2286	+	*/
2287	+	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2288	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2289	+	ForkJoinTask<?>[] a; int m, s; long j;
2290	+	if ((z = submitters.get()) != null &&
2291	+	(p = commonPool) != null &&
2292	+	(ws = p.workQueues) != null &&
2293	+	(m = ws.length - 1) >= 0 &&
2294	+	(q = ws[m & z.seed & SQMASK]) != null &&
2295	+	(s = q.top) != q.base &&
2296	+	(a = q.array) != null &&
2297	+	U.getObjectVolatile
2298	+	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2299	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2300	+	if (q.array == a && q.top == s && // recheck
2301	+	U.compareAndSwapObject(a, j, t, null)) {
2302	+	q.top = s - 1;
2303	+	q.qlock = 0;
2304	+	return true;
2305	+	}
2306	+	q.qlock = 0;
2307	+	}
2308	+	return false;
2309	+	}
2310	+
2311	+	/**
2312	+	* Tries to pop and run local tasks within the same computation
2313	+	* as the given root. On failure, tries to help complete from
2314	+	* other queues via helpComplete.
2315	+	*/
2316	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2317	+	ForkJoinTask<?>[] a; int m;
2318	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2319	+	root != null && root.status >= 0) {
2320	+	for (;;) {
2321	+	int s, u; Object o; CountedCompleter<?> task = null;
2322	+	if ((s = q.top) - q.base > 0) {
2323	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2324	+	if ((o = U.getObject(a, j)) != null &&
2325	+	(o instanceof CountedCompleter)) {
2326	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2327	+	do {
2328	+	if (r == root) {
2329	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2330	+	if (q.array == a && q.top == s &&
2331	+	U.compareAndSwapObject(a, j, t, null)) {
2332	+	q.top = s - 1;
2333	+	task = t;
2334	+	}
2335	+	q.qlock = 0;
2336	+	}
2337	+	break;
2338	+	}
2339	+	} while ((r = r.completer) != null);
2340	+	}
2341	+	}
2342	+	if (task != null)
2343	+	task.doExec();
2344	+	if (root.status < 0 \|\|
2345	+	(u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0)
2346	+	break;
2347	+	if (task == null) {
2348	+	if (helpSignal(root, q.poolIndex) >= 0)
2349	+	helpComplete(root, SHARED_QUEUE);
2350	+	break;
2351	+	}
2352	+	}
2353	+	}
2354	+	}
2355	+
2356	+	/**
2357	+	* Tries to help execute or signal availability of the given task
2358	+	* from submitter's queue in common pool.
2359	+	*/
2360	+	static void externalHelpJoin(ForkJoinTask<?> t) {
2361	+	// Some hard-to-avoid overlap with tryExternalUnpush
2362	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2363	+	ForkJoinTask<?>[] a; int m, s, n; long j;
2364	+	if (t != null &&
2365	+	(z = submitters.get()) != null &&
2366	+	(p = commonPool) != null &&
2367	+	(ws = p.workQueues) != null &&
2368	+	(m = ws.length - 1) >= 0 &&
2369	+	(q = ws[m & z.seed & SQMASK]) != null &&
2370	+	(a = q.array) != null &&
2371	+	t.status >= 0) {
2372	+	if ((s = q.top) != q.base &&
2373	+	U.getObjectVolatile
2374	+	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2375	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2376	+	if (q.array == a && q.top == s &&
2377	+	U.compareAndSwapObject(a, j, t, null)) {
2378	+	q.top = s - 1;
2379	+	q.qlock = 0;
2380	+	t.doExec();
2381	+	}
2382	+	else
2383	+	q.qlock = 0;
2384	+	}
2385	+	if (t.status >= 0) {
2386	+	if (t instanceof CountedCompleter)
2387	+	p.externalHelpComplete(q, t);
2388	+	else
2389	+	p.helpSignal(t, q.poolIndex);
2390	+	}
2391	+	}
2392	+	}
2393	+
2394	+	/**
2395	+	* Restricted version of helpQuiescePool for external callers
2396	+	*/
2397	+	static void externalHelpQuiescePool() {
2398	+	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2399	+	if ((p = commonPool) != null &&
2400	+	(q = p.findNonEmptyStealQueue(1)) != null &&
2401	+	(b = q.base) - q.top < 0 &&
2402	+	(t = q.pollAt(b)) != null)
2403	+	t.doExec();
2404	+	}
2405	+
2406		// Exported methods
2407
2408		// Constructors
#	Line 2145 \| Line 2474 \| public class ForkJoinPool extends Abstra
2474		throw new NullPointerException();
2475		if (parallelism <= 0 \|\| parallelism > MAX_CAP)
2476		throw new IllegalArgumentException();
2148	–	this.parallelism = parallelism;
2477		this.factory = factory;
2478		this.ueh = handler;
2479	<	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2479	>	this.config = parallelism \| (asyncMode ? (FIFO_QUEUE << 16) : 0);
2480		long np = (long)(-parallelism); // offset ctl counts
2481		this.ctl = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
2482	<	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2155	<	int n = parallelism - 1;
2156	<	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
2157	<	int size = (n + 1) << 1; // #slots = 2*#workers
2158	<	this.submitMask = size - 1; // room for max # of submit queues
2159	<	this.workQueues = new WorkQueue[size];
2160	<	this.termination = (this.lock = new Mutex()).newCondition();
2161	<	this.stealCount = new AtomicLong();
2162	<	this.nextWorkerNumber = new AtomicInteger();
2163	<	int pn = poolNumberGenerator.incrementAndGet();
2482	>	int pn = nextPoolId();
2483		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2484		sb.append(Integer.toString(pn));
2485		sb.append("-worker-");
2486		this.workerNamePrefix = sb.toString();
2487	<	lock.lock();
2488	<	this.runState = 1; // set init flag
2489	<	lock.unlock();
2487	>	}
2488	>
2489	>	/**
2490	>	* Constructor for common pool, suitable only for static initialization.
2491	>	* Basically the same as above, but uses smallest possible initial footprint.
2492	>	*/
2493	>	ForkJoinPool(int parallelism, long ctl,
2494	>	ForkJoinWorkerThreadFactory factory,
2495	>	Thread.UncaughtExceptionHandler handler) {
2496	>	this.config = parallelism;
2497	>	this.ctl = ctl;
2498	>	this.factory = factory;
2499	>	this.ueh = handler;
2500	>	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2501	>	}
2502	>
2503	>	/**
2504	>	* Returns the common pool instance.
2505	>	*
2506	>	* @return the common pool instance
2507	>	*/
2508	>	public static ForkJoinPool commonPool() {
2509	>	// assert commonPool != null : "static init error";
2510	>	return commonPool;
2511		}
2512
2513		// Execution methods
#	Line 2191 \| Line 2531 \| public class ForkJoinPool extends Abstra
2531		public <T> T invoke(ForkJoinTask<T> task) {
2532		if (task == null)
2533		throw new NullPointerException();
2534	<	doSubmit(task);
2534	>	externalPush(task);
2535		return task.join();
2536		}
2537
#	Line 2206 \| Line 2546 \| public class ForkJoinPool extends Abstra
2546		public void execute(ForkJoinTask<?> task) {
2547		if (task == null)
2548		throw new NullPointerException();
2549	<	doSubmit(task);
2549	>	externalPush(task);
2550		}
2551
2552		// AbstractExecutorService methods
#	Line 2224 \| Line 2564 \| public class ForkJoinPool extends Abstra
2564		job = (ForkJoinTask<?>) task;
2565		else
2566		job = new ForkJoinTask.AdaptedRunnableAction(task);
2567	<	doSubmit(job);
2567	>	externalPush(job);
2568		}
2569
2570		/**
#	Line 2239 \| Line 2579 \| public class ForkJoinPool extends Abstra
2579		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2580		if (task == null)
2581		throw new NullPointerException();
2582	<	doSubmit(task);
2582	>	externalPush(task);
2583		return task;
2584		}
2585
#	Line 2250 \| Line 2590 \| public class ForkJoinPool extends Abstra
2590		*/
2591		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2592		ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2593	<	doSubmit(job);
2593	>	externalPush(job);
2594		return job;
2595		}
2596
#	Line 2261 \| Line 2601 \| public class ForkJoinPool extends Abstra
2601		*/
2602		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2603		ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2604	<	doSubmit(job);
2604	>	externalPush(job);
2605		return job;
2606		}
2607
#	Line 2278 \| Line 2618 \| public class ForkJoinPool extends Abstra
2618		job = (ForkJoinTask<?>) task;
2619		else
2620		job = new ForkJoinTask.AdaptedRunnableAction(task);
2621	<	doSubmit(job);
2621	>	externalPush(job);
2622		return job;
2623		}
2624
#	Line 2300 \| Line 2640 \| public class ForkJoinPool extends Abstra
2640		try {
2641		for (Callable<T> t : tasks) {
2642		ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2643	<	doSubmit(f);
2643	>	externalPush(f);
2644		fs.add(f);
2645		}
2646		for (ForkJoinTask<T> f : fs)
#	Line 2339 \| Line 2679 \| public class ForkJoinPool extends Abstra
2679		* @return the targeted parallelism level of this pool
2680		*/
2681		public int getParallelism() {
2682	<	return parallelism;
2682	>	return config & SMASK;
2683	>	}
2684	>
2685	>	/**
2686	>	* Returns the targeted parallelism level of the common pool.
2687	>	*
2688	>	* @return the targeted parallelism level of the common pool
2689	>	*/
2690	>	public static int getCommonPoolParallelism() {
2691	>	return commonPoolParallelism;
2692		}
2693
2694		/**
#	Line 2351 \| Line 2700 \| public class ForkJoinPool extends Abstra
2700		* @return the number of worker threads
2701		*/
2702		public int getPoolSize() {
2703	<	return parallelism + (short)(ctl >>> TC_SHIFT);
2703	>	return (config & SMASK) + (short)(ctl >>> TC_SHIFT);
2704		}
2705
2706		/**
#	Line 2361 \| Line 2710 \| public class ForkJoinPool extends Abstra
2710		* @return {@code true} if this pool uses async mode
2711		*/
2712		public boolean getAsyncMode() {
2713	<	return localMode != 0;
2713	>	return (config >>> 16) == FIFO_QUEUE;
2714		}
2715
2716		/**
#	Line 2392 \| Line 2741 \| public class ForkJoinPool extends Abstra
2741		* @return the number of active threads
2742		*/
2743		public int getActiveThreadCount() {
2744	<	int r = parallelism + (int)(ctl >> AC_SHIFT);
2744	>	int r = (config & SMASK) + (int)(ctl >> AC_SHIFT);
2745		return (r <= 0) ? 0 : r; // suppress momentarily negative values
2746		}
2747
#	Line 2408 \| Line 2757 \| public class ForkJoinPool extends Abstra
2757		* @return {@code true} if all threads are currently idle
2758		*/
2759		public boolean isQuiescent() {
2760	<	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2760	>	return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0;
2761		}
2762
2763		/**
#	Line 2423 \| Line 2772 \| public class ForkJoinPool extends Abstra
2772		* @return the number of steals
2773		*/
2774		public long getStealCount() {
2775	<	long count = stealCount.get();
2775	>	long count = stealCount;
2776		WorkQueue[] ws; WorkQueue w;
2777		if ((ws = workQueues) != null) {
2778		for (int i = 1; i < ws.length; i += 2) {
2779		if ((w = ws[i]) != null)
2780	<	count += w.totalSteals;
2780	>	count += w.nsteals;
2781		}
2782		}
2783		return count;
#	Line 2485 \| Line 2834 \| public class ForkJoinPool extends Abstra
2834		WorkQueue[] ws; WorkQueue w;
2835		if ((ws = workQueues) != null) {
2836		for (int i = 0; i < ws.length; i += 2) {
2837	<	if ((w = ws[i]) != null && !w.isEmpty())
2837	>	if ((w = ws[i]) != null && w.queueSize() != 0)
2838		return true;
2839		}
2840		}
#	Line 2553 \| Line 2902 \| public class ForkJoinPool extends Abstra
2902		public String toString() {
2903		// Use a single pass through workQueues to collect counts
2904		long qt = 0L, qs = 0L; int rc = 0;
2905	<	long st = stealCount.get();
2905	>	long st = stealCount;
2906		long c = ctl;
2907		WorkQueue[] ws; WorkQueue w;
2908		if ((ws = workQueues) != null) {
#	Line 2564 \| Line 2913 \| public class ForkJoinPool extends Abstra
2913		qs += size;
2914		else {
2915		qt += size;
2916	<	st += w.totalSteals;
2916	>	st += w.nsteals;
2917		if (w.isApparentlyUnblocked())
2918		++rc;
2919		}
2920		}
2921		}
2922		}
2923	<	int pc = parallelism;
2923	>	int pc = (config & SMASK);
2924		int tc = pc + (short)(c >>> TC_SHIFT);
2925		int ac = pc + (int)(c >> AC_SHIFT);
2926		if (ac < 0) // ignore transient negative
#	Line 2580 \| Line 2929 \| public class ForkJoinPool extends Abstra
2929		if ((c & STOP_BIT) != 0)
2930		level = (tc == 0) ? "Terminated" : "Terminating";
2931		else
2932	<	level = runState < 0 ? "Shutting down" : "Running";
2932	>	level = plock < 0 ? "Shutting down" : "Running";
2933		return super.toString() +
2934		"[" + level +
2935		", parallelism = " + pc +
#	Line 2594 \| Line 2943 \| public class ForkJoinPool extends Abstra
2943		}
2944
2945		/**
2946	<	* Initiates an orderly shutdown in which previously submitted
2947	<	* tasks are executed, but no new tasks will be accepted.
2948	<	* Invocation has no additional effect if already shut down.
2949	<	* Tasks that are in the process of being submitted concurrently
2950	<	* during the course of this method may or may not be rejected.
2946	>	* Possibly initiates an orderly shutdown in which previously
2947	>	* submitted tasks are executed, but no new tasks will be
2948	>	* accepted. Invocation has no effect on execution state if this
2949	>	* is the {@link #commonPool}, and no additional effect if
2950	>	* already shut down. Tasks that are in the process of being
2951	>	* submitted concurrently during the course of this method may or
2952	>	* may not be rejected.
2953		*
2954		* @throws SecurityException if a security manager exists and
2955		* the caller is not permitted to modify threads
#	Line 2611 \| Line 2962 \| public class ForkJoinPool extends Abstra
2962		}
2963
2964		/**
2965	<	* Attempts to cancel and/or stop all tasks, and reject all
2966	<	* subsequently submitted tasks. Tasks that are in the process of
2967	<	* being submitted or executed concurrently during the course of
2968	<	* this method may or may not be rejected. This method cancels
2969	<	* both existing and unexecuted tasks, in order to permit
2970	<	* termination in the presence of task dependencies. So the method
2971	<	* always returns an empty list (unlike the case for some other
2972	<	* Executors).
2965	>	* Possibly attempts to cancel and/or stop all tasks, and reject
2966	>	* all subsequently submitted tasks. Invocation has no effect on
2967	>	* execution state if this is the {@link #commonPool}, and no
2968	>	* additional effect if already shut down. Otherwise, tasks that
2969	>	* are in the process of being submitted or executed concurrently
2970	>	* during the course of this method may or may not be
2971	>	* rejected. This method cancels both existing and unexecuted
2972	>	* tasks, in order to permit termination in the presence of task
2973	>	* dependencies. So the method always returns an empty list
2974	>	* (unlike the case for some other Executors).
2975		*
2976		* @return an empty list
2977		* @throws SecurityException if a security manager exists and
#	Line 2640 \| Line 2993 \| public class ForkJoinPool extends Abstra
2993		public boolean isTerminated() {
2994		long c = ctl;
2995		return ((c & STOP_BIT) != 0L &&
2996	<	(short)(c >>> TC_SHIFT) == -parallelism);
2996	>	(short)(c >>> TC_SHIFT) == -(config & SMASK));
2997		}
2998
2999		/**
#	Line 2659 \| Line 3012 \| public class ForkJoinPool extends Abstra
3012		public boolean isTerminating() {
3013		long c = ctl;
3014		return ((c & STOP_BIT) != 0L &&
3015	<	(short)(c >>> TC_SHIFT) != -parallelism);
3015	>	(short)(c >>> TC_SHIFT) != -(config & SMASK));
3016		}
3017
3018		/**
#	Line 2668 \| Line 3021 \| public class ForkJoinPool extends Abstra
3021		* @return {@code true} if this pool has been shut down
3022		*/
3023		public boolean isShutdown() {
3024	<	return runState < 0;
3024	>	return plock < 0;
3025		}
3026
3027		/**
3028	<	* Blocks until all tasks have completed execution after a shutdown
3029	<	* request, or the timeout occurs, or the current thread is
3030	<	* interrupted, whichever happens first.
3028	>	* Blocks until all tasks have completed execution after a
3029	>	* shutdown request, or the timeout occurs, or the current thread
3030	>	* is interrupted, whichever happens first. Note that the {@link
3031	>	* #commonPool()} never terminates until program shutdown so
3032	>	* this method will always time out.
3033		*
3034		* @param timeout the maximum time to wait
3035		* @param unit the time unit of the timeout argument
#	Line 2685 \| Line 3040 \| public class ForkJoinPool extends Abstra
3040		public boolean awaitTermination(long timeout, TimeUnit unit)
3041		throws InterruptedException {
3042		long nanos = unit.toNanos(timeout);
3043	<	final Mutex lock = this.lock;
3044	<	lock.lock();
3045	<	try {
3046	<	for (;;) {
3047	<	if (isTerminated())
3048	<	return true;
3049	<	if (nanos <= 0)
3050	<	return false;
3051	<	nanos = termination.awaitNanos(nanos);
3043	>	if (isTerminated())
3044	>	return true;
3045	>	long startTime = System.nanoTime();
3046	>	boolean terminated = false;
3047	>	synchronized (this) {
3048	>	for (long waitTime = nanos, millis = 0L;;) {
3049	>	if (terminated = isTerminated() \|\|
3050	>	waitTime <= 0L \|\|
3051	>	(millis = unit.toMillis(waitTime)) <= 0L)
3052	>	break;
3053	>	wait(millis);
3054	>	waitTime = nanos - (System.nanoTime() - startTime);
3055		}
2698	–	} finally {
2699	–	lock.unlock();
3056		}
3057	+	return terminated;
3058		}
3059
3060		/**
#	Line 2796 \| Line 3153 \| public class ForkJoinPool extends Abstra
3153		public static void managedBlock(ManagedBlocker blocker)
3154		throws InterruptedException {
3155		Thread t = Thread.currentThread();
3156	<	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3157	<	((ForkJoinWorkerThread)t).pool : null);
3158	<	while (!blocker.isReleasable()) {
3159	<	if (p == null \|\| p.tryCompensate(null, blocker)) {
3160	<	try {
3161	<	do {} while (!blocker.isReleasable() && !blocker.block());
3162	<	} finally {
3163	<	if (p != null)
3156	>	if (t instanceof ForkJoinWorkerThread) {
3157	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3158	>	while (!blocker.isReleasable()) { // variant of helpSignal
3159	>	WorkQueue[] ws; WorkQueue q; int m, n, u;
3160	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3161	>	for (int i = 0; i <= m; ++i) {
3162	>	if (blocker.isReleasable())
3163	>	return;
3164	>	if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
3165	>	p.signalWork(q, n);
3166	>	if ((u = (int)(p.ctl >>> 32)) >= 0 \|\|
3167	>	(u >> UAC_SHIFT) >= 0)
3168	>	break;
3169	>	}
3170	>	}
3171	>	}
3172	>	if (p.tryCompensate()) {
3173	>	try {
3174	>	do {} while (!blocker.isReleasable() &&
3175	>	!blocker.block());
3176	>	} finally {
3177		p.incrementActiveCount();
3178	+	}
3179	+	break;
3180		}
2809	–	break;
3181		}
3182		}
3183	+	else {
3184	+	do {} while (!blocker.isReleasable() &&
3185	+	!blocker.block());
3186	+	}
3187		}
3188
3189		// AbstractExecutorService overrides. These rely on undocumented
#	Line 2829 \| Line 3204 \| public class ForkJoinPool extends Abstra
3204		private static final long PARKBLOCKER;
3205		private static final int ABASE;
3206		private static final int ASHIFT;
3207	+	private static final long STEALCOUNT;
3208	+	private static final long PLOCK;
3209	+	private static final long INDEXSEED;
3210	+	private static final long QLOCK;
3211
3212		static {
3213	<	poolNumberGenerator = new AtomicInteger();
2835	<	nextSubmitterSeed = new AtomicInteger(0x55555555);
2836	<	modifyThreadPermission = new RuntimePermission("modifyThread");
2837	<	defaultForkJoinWorkerThreadFactory =
2838	<	new DefaultForkJoinWorkerThreadFactory();
2839	<	submitters = new ThreadSubmitter();
2840	<	int s;
3213	>	int s; // initialize field offsets for CAS etc
3214		try {
3215		U = getUnsafe();
3216		Class<?> k = ForkJoinPool.class;
2844	–	Class<?> ak = ForkJoinTask[].class;
3217		CTL = U.objectFieldOffset
3218		(k.getDeclaredField("ctl"));
3219	+	STEALCOUNT = U.objectFieldOffset
3220	+	(k.getDeclaredField("stealCount"));
3221	+	PLOCK = U.objectFieldOffset
3222	+	(k.getDeclaredField("plock"));
3223	+	INDEXSEED = U.objectFieldOffset
3224	+	(k.getDeclaredField("indexSeed"));
3225		Class<?> tk = Thread.class;
3226		PARKBLOCKER = U.objectFieldOffset
3227		(tk.getDeclaredField("parkBlocker"));
3228	+	Class<?> wk = WorkQueue.class;
3229	+	QLOCK = U.objectFieldOffset
3230	+	(wk.getDeclaredField("qlock"));
3231	+	Class<?> ak = ForkJoinTask[].class;
3232		ABASE = U.arrayBaseOffset(ak);
3233		s = U.arrayIndexScale(ak);
3234	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3235		} catch (Exception e) {
3236		throw new Error(e);
3237		}
3238		if ((s & (s-1)) != 0)
3239		throw new Error("data type scale not a power of two");
3240	<	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3240	>
3241	>	submitters = new ThreadLocal<Submitter>();
3242	>	ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory =
3243	>	new DefaultForkJoinWorkerThreadFactory();
3244	>	/*
3245	>	* Establish common pool parameters. For extra caution,
3246	>	* computations to set up common pool state are here; the
3247	>	* constructor just assigns these values to fields.
3248	>	*/
3249	>
3250	>	int par = 0;
3251	>	Thread.UncaughtExceptionHandler handler = null;
3252	>	try { // TBD: limit or report ignored exceptions?
3253	>	String pp = System.getProperty
3254	>	("java.util.concurrent.ForkJoinPool.common.parallelism");
3255	>	String hp = System.getProperty
3256	>	("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3257	>	String fp = System.getProperty
3258	>	("java.util.concurrent.ForkJoinPool.common.threadFactory");
3259	>	if (fp != null)
3260	>	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3261	>	getSystemClassLoader().loadClass(fp).newInstance());
3262	>	if (hp != null)
3263	>	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3264	>	getSystemClassLoader().loadClass(hp).newInstance());
3265	>	if (pp != null)
3266	>	par = Integer.parseInt(pp);
3267	>	} catch (Exception ignore) {
3268	>	}
3269	>
3270	>	if (par <= 0)
3271	>	par = Runtime.getRuntime().availableProcessors();
3272	>	if (par > MAX_CAP)
3273	>	par = MAX_CAP;
3274	>	commonPoolParallelism = par;
3275	>	long np = (long)(-par); // precompute initial ctl value
3276	>	long ct = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
3277	>
3278	>	commonPool = new ForkJoinPool(par, ct, fac, handler);
3279	>	modifyThreadPermission = new RuntimePermission("modifyThread");
3280		}
3281
3282		/**

Diff Legend

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

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents): Revision 1.2 by jsr166, Mon Aug 13 18:25:53 2012 UTC vs. Revision 1.20 by jsr166, Tue Nov 20 06:18:39 2012 UTC

Diff Legend

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.2 by jsr166, Mon Aug 13 18:25:53 2012 UTC vs.
Revision 1.20 by jsr166, Tue Nov 20 06:18:39 2012 UTC