[ViewVC] Diff of: jsr166/jsr166/src/jsr166y/ForkJoinPool.java

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.124 by jsr166, Mon Feb 20 23:32:24 2012 UTC vs.
Revision 1.142 by jsr166, Wed Nov 14 19:05:03 2012 UTC

#	Line 5 \| Line 5
5		*/
6
7		package jsr166y;
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 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.
105	<	*
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	>	* Thread.UncaughtExceptionHandler}. Upon any error in establishing
108	>	* these settings, default parameters are used.
109		*
110		* <p><b>Implementation notes</b>: This implementation restricts the
111		* maximum number of running threads to 32767. Attempts to create
#	Line 196 \| Line 193 \| public class ForkJoinPool extends Abstra
193		* WorkQueues are also used in a similar way for tasks submitted
194		* to the pool. We cannot mix these tasks in the same queues used
195		* for work-stealing (this would contaminate lifo/fifo
196	<	* processing). Instead, we loosely associate submission queues
196	>	* processing). Instead, we randomly associate submission queues
197		* with submitting threads, using a form of hashing. The
198		* ThreadLocal Submitter class contains a value initially used as
199		* a hash code for choosing existing queues, but may be randomly
200		* repositioned upon contention with other submitters. In
201	<	* essence, submitters act like workers except that they never
202	<	* take tasks, and they are multiplexed on to a finite number of
203	<	* shared work queues. However, classes are set up so that future
204	<	* extensions could allow submitters to optionally help perform
205	<	* tasks as well. Insertion of tasks in shared mode requires a
206	<	* lock (mainly to protect in the case of resizing) but we use
207	<	* only a simple spinlock (using bits in field runState), because
208	<	* submitters encountering a busy queue move on to try or create
209	<	* other queues -- they block only when creating and registering
210	<	* new queues.
201	>	* essence, submitters act like workers except that they are
202	>	* restricted to executing local tasks that they submitted (or in
203	>	* the case of CountedCompleters, others with the same root task).
204	>	* However, because most shared/external queue operations are more
205	>	* expensive than internal, and because, at steady state, external
206	>	* submitters will compete for CPU with workers, ForkJoinTask.join
207	>	* and related methods disable them from repeatedly helping to
208	>	* process tasks if all workers are active. Insertion of tasks in
209	>	* shared mode requires a lock (mainly to protect in the case of
210	>	* resizing) but we use only a simple spinlock (using bits in
211	>	* field qlock), because submitters encountering a busy queue move
212	>	* on to try or create other queues -- they block only when
213	>	* creating and registering new queues.
214		*
215		* Management
216		* ==========
#	Line 232 \| Line 232 \| public class ForkJoinPool extends Abstra
232		* and their negations (used for thresholding) to fit into 16bit
233		* fields.
234		*
235	<	* Field "runState" contains 32 bits needed to register and
236	<	* deregister WorkQueues, as well as to enable shutdown. It is
237	<	* only modified under a lock (normally briefly held, but
238	<	* occasionally protecting allocations and resizings) but even
239	<	* when locked remains available to check consistency.
235	>	* Field "plock" is a form of sequence lock with a saturating
236	>	* shutdown bit (similarly for per-queue "qlocks"), mainly
237	>	* protecting updates to the workQueues array, as well as to
238	>	* enable shutdown. When used as a lock, it is normally only very
239	>	* briefly held, so is nearly always available after at most a
240	>	* brief spin, but we use a monitor-based backup strategy to
241	>	* blocking when needed.
242		*
243		* Recording WorkQueues. WorkQueues are recorded in the
244	<	* "workQueues" array that is created upon pool construction and
245	<	* expanded if necessary. Updates to the array while recording
246	<	* new workers and unrecording terminated ones are protected from
247	<	* each other by a lock but the array is otherwise concurrently
248	<	* readable, and accessed directly. To simplify index-based
249	<	* operations, the array size is always a power of two, and all
250	<	* readers must tolerate null slots. Shared (submission) queues
251	<	* are at even indices, worker queues at odd indices. Grouping
252	<	* them together in this way simplifies and speeds up task
253	<	* scanning.
244	>	* "workQueues" array that is created upon first use and expanded
245	>	* if necessary. Updates to the array while recording new workers
246	>	* and unrecording terminated ones are protected from each other
247	>	* by a lock but the array is otherwise concurrently readable, and
248	>	* accessed directly. To simplify index-based operations, the
249	>	* array size is always a power of two, and all readers must
250	>	* tolerate null slots. Worker queues are at odd indices Shared
251	>	* (submission) queues are at even indices, up to a maximum of 64
252	>	* slots, to limit growth even if array needs to expand to add
253	>	* more workers. Grouping them together in this way simplifies and
254	>	* speeds up task scanning.
255		*
256		* All worker thread creation is on-demand, triggered by task
257		* submissions, replacement of terminated workers, and/or
#	Line 309 \| Line 312 \| public class ForkJoinPool extends Abstra
312		*
313		* Signalling. We create or wake up workers only when there
314		* appears to be at least one task they might be able to find and
315	<	* execute. When a submission is added or another worker adds a
316	<	* task to a queue that previously had fewer than two tasks, they
317	<	* signal waiting workers (or trigger creation of new ones if
318	<	* fewer than the given parallelism level -- see signalWork).
319	<	* These primary signals are buttressed by signals during rescans;
320	<	* together these cover the signals needed in cases when more
321	<	* tasks are pushed but untaken, and improve performance compared
322	<	* to having one thread wake up all workers.
315	>	* execute. However, many other threads may notice the same task
316	>	* and each signal to wake up a thread that might take it. So in
317	>	* general, pools will be over-signalled. When a submission is
318	>	* added or another worker adds a task to a queue that is
319	>	* apparently empty, they signal waiting workers (or trigger
320	>	* creation of new ones if fewer than the given parallelism level
321	>	* -- see signalWork). These primary signals are buttressed by
322	>	* signals whenever other threads scan for work or do not have a
323	>	* task to process. On most platforms, signalling (unpark)
324	>	* overhead time is noticeably long, and the time between
325	>	* signalling a thread and it actually making progress can be very
326	>	* noticeably long, so it is worth offloading these delays from
327	>	* critical paths as much as possible.
328		*
329		* Trimming workers. To release resources after periods of lack of
330		* use, a worker starting to wait when the pool is quiescent will
331	<	* time out and terminate if the pool has remained quiescent for
332	<	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
333	<	* terminating all workers after long periods of non-use.
331	>	* time out and terminate if the pool has remained quiescent for a
332	>	* given period -- a short period if there are more threads than
333	>	* parallelism, longer as the number of threads decreases. This
334	>	* will slowly propagate, eventually terminating all workers after
335	>	* periods of non-use.
336		*
337		* Shutdown and Termination. A call to shutdownNow atomically sets
338	<	* a runState bit and then (non-atomically) sets each worker's
339	<	* runState status, cancels all unprocessed tasks, and wakes up
338	>	* a plock bit and then (non-atomically) sets each worker's
339	>	* qlock status, cancels all unprocessed tasks, and wakes up
340		* all waiting workers. Detecting whether termination should
341		* commence after a non-abrupt shutdown() call requires more work
342		* and bookkeeping. We need consensus about quiescence (i.e., that
#	Line 354 \| Line 364 \| public class ForkJoinPool extends Abstra
364		* method tryCompensate() may create or re-activate a spare
365		* thread to compensate for blocked joiners until they unblock.
366		*
367	<	* A third form (implemented in tryRemoveAndExec and
368	<	* tryPollForAndExec) amounts to helping a hypothetical
369	<	* compensator: If we can readily tell that a possible action of a
370	<	* compensator is to steal and execute the task being joined, the
371	<	* joining thread can do so directly, without the need for a
372	<	* compensation thread (although at the expense of larger run-time
373	<	* stacks, but the tradeoff is typically worthwhile).
367	>	* A third form (implemented in tryRemoveAndExec) amounts to
368	>	* helping a hypothetical compensator: If we can readily tell that
369	>	* a possible action of a compensator is to steal and execute the
370	>	* task being joined, the joining thread can do so directly,
371	>	* without the need for a compensation thread (although at the
372	>	* expense of larger run-time stacks, but the tradeoff is
373	>	* typically worthwhile).
374		*
375		* The ManagedBlocker extension API can't use helping so relies
376		* only on compensation in method awaitBlocker.
#	Line 395 \| Line 405 \| public class ForkJoinPool extends Abstra
405		* to find work (see MAX_HELP) and fall back to suspending the
406		* worker and if necessary replacing it with another.
407		*
408	+	* Helping actions for CountedCompleters are much simpler: Method
409	+	* helpComplete can take and execute any task with the same root
410	+	* as the task being waited on. However, this still entails some
411	+	* traversal of completer chains, so is less efficient than using
412	+	* CountedCompleters without explicit joins.
413	+	*
414		* It is impossible to keep exactly the target parallelism number
415		* of threads running at any given time. Determining the
416		* existence of conservatively safe helping targets, the
#	Line 413 \| Line 429 \| public class ForkJoinPool extends Abstra
429		* unblocked threads to the point that we know they are available)
430		* leading to more situations requiring more threads, and so
431		* on. This aspect of control can be seen as an (analytically
432	<	* intractible) game with an opponent that may choose the worst
432	>	* intractable) game with an opponent that may choose the worst
433		* (for us) active thread to stall at any time. We take several
434		* precautions to bound losses (and thus bound gains), mainly in
435	<	* methods tryCompensate and awaitJoin: (1) We only try
436	<	* compensation after attempting enough helping steps (measured
437	<	* via counting and timing) that we have already consumed the
438	<	* estimated cost of creating and activating a new thread. (2) We
439	<	* allow up to 50% of threads to be blocked before initially
440	<	* adding any others, and unless completely saturated, check that
441	<	* some work is available for a new worker before adding. Also, we
442	<	* create up to only 50% more threads until entering a mode that
443	<	* only adds a thread if all others are possibly blocked. All
444	<	* together, this means that we might be half as fast to react,
445	<	* and create half as many threads as possible in the ideal case,
446	<	* but present vastly fewer anomalies in all other cases compared
447	<	* to both more aggressive and more conservative alternatives.
448	<	*
449	<	* Style notes: There is a lot of representation-level coupling
450	<	* among classes ForkJoinPool, ForkJoinWorkerThread, and
451	<	* ForkJoinTask. The fields of WorkQueue maintain data structures
452	<	* managed by ForkJoinPool, so are directly accessed. There is
453	<	* little point trying to reduce this, since any associated future
454	<	* changes in representations will need to be accompanied by
455	<	* algorithmic changes anyway. Several methods intrinsically
456	<	* sprawl because they must accumulate sets of consistent reads of
457	<	* volatiles held in local variables. Methods signalWork() and
458	<	* scan() are the main bottlenecks, so are especially heavily
435	>	* methods tryCompensate and awaitJoin.
436	>	*
437	>	* Common Pool
438	>	* ===========
439	>	*
440	>	* The static commonPool always exists after static
441	>	* initialization. Since it (or any other created pool) need
442	>	* never be used, we minimize initial construction overhead and
443	>	* footprint to the setup of about a dozen fields, with no nested
444	>	* allocation. Most bootstrapping occurs within method
445	>	* fullExternalPush during the first submission to the pool.
446	>	*
447	>	* When external threads submit to the common pool, they can
448	>	* perform some subtask processing (see externalHelpJoin and
449	>	* related methods). We do not need to record whether these
450	>	* submissions are to the common pool -- if not, externalHelpJoin
451	>	* returns quickly (at the most helping to signal some common pool
452	>	* workers). These submitters would otherwise be blocked waiting
453	>	* for completion, so the extra effort (with liberally sprinkled
454	>	* task status checks) in inapplicable cases amounts to an odd
455	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
456	>	*
457	>	* Style notes
458	>	* ===========
459	>	*
460	>	* There is a lot of representation-level coupling among classes
461	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
462	>	* fields of WorkQueue maintain data structures managed by
463	>	* ForkJoinPool, so are directly accessed. There is little point
464	>	* trying to reduce this, since any associated future changes in
465	>	* representations will need to be accompanied by algorithmic
466	>	* changes anyway. Several methods intrinsically sprawl because
467	>	* they must accumulate sets of consistent reads of volatiles held
468	>	* in local variables. Methods signalWork() and scan() are the
469	>	* main bottlenecks, so are especially heavily
470		* micro-optimized/mangled. There are lots of inline assignments
471		* (of form "while ((local = field) != 0)") which are usually the
472		* simplest way to ensure the required read orderings (which are
#	Line 447 \| Line 474 \| public class ForkJoinPool extends Abstra
474		* declarations of these locals at the heads of methods or blocks.
475		* There are several occurrences of the unusual "do {} while
476		* (!cas...)" which is the simplest way to force an update of a
477	<	* CAS'ed variable. There are also other coding oddities that help
477	>	* CAS'ed variable. There are also other coding oddities (including
478	>	* several unnecessary-looking hoisted null checks) that help
479		* some methods perform reasonably even when interpreted (not
480		* compiled).
481		*
#	Line 504 \| Line 532 \| public class ForkJoinPool extends Abstra
532		}
533
534		/**
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	–	/**
535		* Class for artificial tasks that are used to replace the target
536		* of local joins if they are removed from an interior queue slot
537		* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
538		* actually do anything beyond having a unique identity.
539		*/
540		static final class EmptyTask extends ForkJoinTask<Void> {
541	+	private static final long serialVersionUID = -7721805057305804111L;
542		EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
543		public final Void getRawResult() { return null; }
544		public final void setRawResult(Void x) {}
#	Line 553 \| Line 559 \| public class ForkJoinPool extends Abstra
559		*
560		* Field "top" is the index (mod array.length) of the next queue
561		* slot to push to or pop from. It is written only by owner thread
562	<	* for push, or under lock for trySharedPush, and accessed by
563	<	* other threads only after reading (volatile) base. Both top and
564	<	* base are allowed to wrap around on overflow, but (top - base)
565	<	* (or more commonly -(base - top) to force volatile read of base
566	<	* before top) still estimates size.
562	>	* for push, or under lock for external/shared push, and accessed
563	>	* by other threads only after reading (volatile) base. Both top
564	>	* and base are allowed to wrap around on overflow, but (top -
565	>	* base) (or more commonly -(base - top) to force volatile read of
566	>	* base before top) still estimates size. The lock ("qlock") is
567	>	* forced to -1 on termination, causing all further lock attempts
568	>	* to fail. (Note: we don't need CAS for termination state because
569	>	* upon pool shutdown, all shared-queues will stop being used
570	>	* anyway.) Nearly all lock bodies are set up so that exceptions
571	>	* within lock bodies are "impossible" (modulo JVM errors that
572	>	* would cause failure anyway.)
573		*
574		* The array slots are read and written using the emulation of
575		* volatiles/atomics provided by Unsafe. Insertions must in
576		* general use putOrderedObject as a form of releasing store to
577		* ensure that all writes to the task object are ordered before
578	<	* its publication in the queue. (Although we can avoid one case
579	<	* of this when locked in trySharedPush.) All removals entail a
580	<	* CAS to null. The array is always a power of two. To ensure
581	<	* safety of Unsafe array operations, all accesses perform
570	<	* explicit null checks and implicit bounds checks via
571	<	* power-of-two masking.
578	>	* its publication in the queue. All removals entail a CAS to
579	>	* null. The array is always a power of two. To ensure safety of
580	>	* Unsafe array operations, all accesses perform explicit null
581	>	* checks and implicit bounds checks via power-of-two masking.
582		*
583		* In addition to basic queuing support, this class contains
584		* fields described elsewhere to control execution. It turns out
585	<	* to work better memory-layout-wise to include them in this
586	<	* class rather than a separate class.
585	>	* to work better memory-layout-wise to include them in this class
586	>	* rather than a separate class.
587		*
588		* Performance on most platforms is very sensitive to placement of
589		* instances of both WorkQueues and their arrays -- we absolutely
#	Line 589 \| Line 599 \| public class ForkJoinPool extends Abstra
599		* support is in place, this padding is dependent on transient
600		* properties of JVM field layout rules.) We also take care in
601		* allocating, sizing and resizing the array. Non-shared queue
602	<	* arrays are initialized (via method growArray) by workers before
603	<	* use. Others are allocated on first use.
602	>	* arrays are initialized by workers before use. Others are
603	>	* allocated on first use.
604		*/
605		static final class WorkQueue {
606		/**
#	Line 613 \| Line 623 \| public class ForkJoinPool extends Abstra
623		*/
624		static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
625
616	–	volatile long totalSteals; // cumulative number of steals
626		int seed; // for random scanning; initialize nonzero
627		volatile int eventCount; // encoded inactivation count; < 0 if inactive
628		int nextWait; // encoded record of next event waiter
620	–	int rescans; // remaining scans until block
621	–	int nsteals; // top-level task executions since last idle
629		final int mode; // lifo, fifo, or shared
630	+	int nsteals; // cumulative number of steals
631		int poolIndex; // index of this queue in pool (or 0)
632		int stealHint; // index of most recent known stealer
633	<	volatile int runState; // 1: locked, -1: terminate; else 0
633	>	volatile int qlock; // 1: locked, -1: terminate; else 0
634		volatile int base; // index of next slot for poll
635		int top; // index of next slot for push
636		ForkJoinTask<?>[] array; // the elements (initially unallocated)
637		final ForkJoinPool pool; // the containing pool (may be null)
638		final ForkJoinWorkerThread owner; // owning thread or null if shared
639		volatile Thread parker; // == owner during call to park; else null
640	<	ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
640	>	volatile ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
641		ForkJoinTask<?> currentSteal; // current non-local task being executed
642		// Heuristic padding to ameliorate unfortunate memory placements
643		Object p00, p01, p02, p03, p04, p05, p06, p07;
#	Line 644 \| Line 652 \| public class ForkJoinPool extends Abstra
652		}
653
654		/**
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	–	/**
655		* Pushes a task. Call only by owner in unshared queues.
656	+	* Cases needing resizing or rejection are relayed to fullPush
657	+	* (that also handles shared queues).
658		*
659		* @param task the task. Caller must ensure non-null.
660		* @throw RejectedExecutionException if array cannot be resized
661		*/
662		final void push(ForkJoinTask<?> task) {
663	<	ForkJoinTask<?>[] a; ForkJoinPool p;
664	<	int s = top, m, n;
665	<	if ((a = array) != null) { // ignore if queue removed
663	>	ForkJoinPool p; ForkJoinTask<?>[] a;
664	>	int s = top, n;
665	>	if ((a = array) != null && a.length > (n = s + 1 - base)) {
666		U.putOrderedObject
667	<	(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
668	<	if ((n = (top = s + 1) - base) <= 2) {
669	<	if ((p = pool) != null)
670	<	p.signalWork();
685	<	}
686	<	else if (n >= m)
687	<	growArray(true);
667	>	(a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
668	>	top = s + 1;
669	>	if (n <= 1 && (p = pool) != null)
670	>	p.signalWork(this, 1);
671		}
672	+	else
673	+	fullPush(task, true);
674		}
675
676		/**
677		* Pushes a task if lock is free and array is either big
678	<	* enough or can be resized to be big enough.
678	>	* enough or can be resized to be big enough. Note: a
679	>	* specialization of a common fast path of this method is in
680	>	* ForkJoinPool.externalPush. When called from a FJWT queue,
681	>	* this can fail only if the pool has been shut down or
682	>	* an out of memory error.
683		*
684		* @param task the task. Caller must ensure non-null.
685	<	* @return true if submitted
685	>	* @param owned if true, throw RJE on failure
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;
691	<	int s = top;
692	<	try {
693	<	if ((a != null && a.length > s + 1 - base) \|\|
694	<	(a = growArray(false)) != null) { // must presize
695	<	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
696	<	U.putObject(a, (long)j, task); // don't need "ordered"
697	<	top = s + 1;
698	<	submitted = true;
687	>	final boolean fullPush(ForkJoinTask<?> task, boolean owned) {
688	>	ForkJoinPool p; ForkJoinTask<?>[] a;
689	>	if (owned) {
690	>	if (qlock < 0) // must be shutting down
691	>	throw new RejectedExecutionException();
692	>	}
693	>	else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
694	>	return false;
695	>	try {
696	>	int s = top, oldLen, len;
697	>	if ((a = array) == null)
698	>	a = array = new ForkJoinTask<?>[len=INITIAL_QUEUE_CAPACITY];
699	>	else if ((oldLen = a.length) > s + 1 - base)
700	>	len = oldLen;
701	>	else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
702	>	throw new RejectedExecutionException("Capacity exceeded");
703	>	else {
704	>	int oldMask, b;
705	>	ForkJoinTask<?>[] oldA = a;
706	>	a = array = new ForkJoinTask<?>[len];
707	>	if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
708	>	int mask = len - 1;
709	>	do {
710	>	ForkJoinTask<?> x;
711	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
712	>	int j = ((b & mask) << ASHIFT) + ABASE;
713	>	x = (ForkJoinTask<?>)
714	>	U.getObjectVolatile(oldA, oldj);
715	>	if (x != null &&
716	>	U.compareAndSwapObject(oldA, oldj, x, null))
717	>	U.putObjectVolatile(a, j, x);
718	>	} while (++b != s);
719		}
711	–	} finally {
712	–	runState = 0; // unlock
720		}
721	+	U.putOrderedObject
722	+	(a, (((len - 1) & s) << ASHIFT) + ABASE, task);
723	+	top = s + 1;
724	+	} finally {
725	+	if (!owned)
726	+	qlock = 0;
727		}
728	<	return submitted;
728	>	if ((p = pool) != null)
729	>	p.signalWork(this, 1);
730	>	return true;
731		}
732
733		/**
734		* Takes next task, if one exists, in LIFO order. Call only
735	<	* by owner in unshared queues. (We do not have a shared
721	<	* version of this method because it is never needed.)
735	>	* by owner in unshared queues.
736		*/
737		final ForkJoinTask<?> pop() {
738	<	ForkJoinTask<?> t; int m;
739	<	ForkJoinTask<?>[] a = array;
726	<	if (a != null && (m = a.length - 1) >= 0) {
738	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
739	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
740		for (int s; (s = top - 1) - base >= 0;) {
741	<	int j = ((m & s) << ASHIFT) + ABASE;
742	<	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) == null)
741	>	long j = ((m & s) << ASHIFT) + ABASE;
742	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
743		break;
744		if (U.compareAndSwapObject(a, j, t, null)) {
745		top = s;
#	Line 774 \| Line 787 \| public class ForkJoinPool extends Abstra
787		else if (base == b) {
788		if (b + 1 == top)
789		break;
790	<	Thread.yield(); // wait for lagging update
790	>	Thread.yield(); // wait for lagging update (very rare)
791		}
792		}
793		return null;
#	Line 801 \| Line 814 \| public class ForkJoinPool extends Abstra
814
815		/**
816		* Pops the given task only if it is at the current top.
817	+	* (A shared version is available only via FJP.tryExternalUnpush)
818		*/
819		final boolean tryUnpush(ForkJoinTask<?> t) {
820		ForkJoinTask<?>[] a; int s;
#	Line 814 \| Line 828 \| public class ForkJoinPool extends Abstra
828		}
829
830		/**
831	<	* Polls the given task only if it is at the current base.
831	>	* Removes and cancels all known tasks, ignoring any exceptions.
832		*/
833	<	final boolean pollFor(ForkJoinTask<?> task) {
834	<	ForkJoinTask<?>[] a; int b;
835	<	if ((b = base) - top < 0 && (a = array) != null) {
836	<	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
837	<	if (U.getObjectVolatile(a, j) == task && base == b &&
838	<	U.compareAndSwapObject(a, j, task, null)) {
839	<	base = b + 1;
840	<	return true;
833	>	final void cancelAll() {
834	>	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
835	>	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
836	>	for (ForkJoinTask<?> t; (t = poll()) != null; )
837	>	ForkJoinTask.cancelIgnoringExceptions(t);
838	>	}
839	>
840	>	/**
841	>	* Computes next value for random probes. Scans don't require
842	>	* a very high quality generator, but also not a crummy one.
843	>	* Marsaglia xor-shift is cheap and works well enough. Note:
844	>	* This is manually inlined in its usages in ForkJoinPool to
845	>	* avoid writes inside busy scan loops.
846	>	*/
847	>	final int nextSeed() {
848	>	int r = seed;
849	>	r ^= r << 13;
850	>	r ^= r >>> 17;
851	>	return seed = r ^= r << 5;
852	>	}
853	>
854	>	/**
855	>	* Provides a more accurate estimate of size than (top - base)
856	>	* by ordering reads and checking whether a near-empty queue
857	>	* has at least one unclaimed task.
858	>	*/
859	>	final int queueSize() {
860	>	ForkJoinTask<?>[] a; int k, s, n;
861	>	return ((n = base - (s = top)) < 0 &&
862	>	(n != -1 \|\|
863	>	((a = array) != null && (k = a.length) > 0 &&
864	>	U.getObject
865	>	(a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
866	>	-n : 0;
867	>	}
868	>
869	>	// Specialized execution methods
870	>
871	>	/**
872	>	* Pops and runs tasks until empty.
873	>	*/
874	>	private void popAndExecAll() {
875	>	// A bit faster than repeated pop calls
876	>	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
877	>	while ((a = array) != null && (m = a.length - 1) >= 0 &&
878	>	(s = top - 1) - base >= 0 &&
879	>	(t = ((ForkJoinTask<?>)
880	>	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
881	>	!= null) {
882	>	if (U.compareAndSwapObject(a, j, t, null)) {
883	>	top = s;
884	>	t.doExec();
885		}
886		}
829	–	return false;
887		}
888
889		/**
890	<	* If present, removes from queue and executes the given task, or
891	<	* any other cancelled task. Returns (true) immediately on any CAS
890	>	* Polls and runs tasks until empty.
891	>	*/
892	>	private void pollAndExecAll() {
893	>	for (ForkJoinTask<?> t; (t = poll()) != null;)
894	>	t.doExec();
895	>	}
896	>
897	>	/**
898	>	* If present, removes from queue and executes the given task,
899	>	* or any other cancelled task. Returns (true) on any CAS
900		* or consistency check failure so caller can retry.
901		*
902	<	* @return false if no progress can be made
902	>	* @return false if no progress can be made, else true;
903		*/
904		final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
905	<	boolean removed = false, empty = true, progress = true;
905	>	boolean stat = true, removed = false, empty = true;
906		ForkJoinTask<?>[] a; int m, s, b, n;
907		if ((a = array) != null && (m = a.length - 1) >= 0 &&
908		(n = (s = top) - (b = base)) > 0) {
#	Line 867 \| Line 932 \| public class ForkJoinPool extends Abstra
932		}
933		if (--n == 0) {
934		if (!empty && base == b)
935	<	progress = false;
935	>	stat = false;
936		break;
937		}
938		}
939		}
940		if (removed)
941		task.doExec();
942	<	return progress;
878	<	}
879	<
880	<	/**
881	<	* Initializes or doubles the capacity of array. Call either
882	<	* by owner or with lock held -- it is OK for base, but not
883	<	* top, to move while resizings are in progress.
884	<	*
885	<	* @param rejectOnFailure if true, throw exception if capacity
886	<	* exceeded (relayed ultimately to user); else return null.
887	<	*/
888	<	final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
889	<	ForkJoinTask<?>[] oldA = array;
890	<	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
891	<	if (size <= MAXIMUM_QUEUE_CAPACITY) {
892	<	int oldMask, t, b;
893	<	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
894	<	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
895	<	(t = top) - (b = base) > 0) {
896	<	int mask = size - 1;
897	<	do {
898	<	ForkJoinTask<?> x;
899	<	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
900	<	int j = ((b & mask) << ASHIFT) + ABASE;
901	<	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
902	<	if (x != null &&
903	<	U.compareAndSwapObject(oldA, oldj, x, null))
904	<	U.putObjectVolatile(a, j, x);
905	<	} while (++b != t);
906	<	}
907	<	return a;
908	<	}
909	<	else if (!rejectOnFailure)
910	<	return null;
911	<	else
912	<	throw new RejectedExecutionException("Queue capacity exceeded");
913	<	}
914	<
915	<	/**
916	<	* Removes and cancels all known tasks, ignoring any exceptions.
917	<	*/
918	<	final void cancelAll() {
919	<	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
920	<	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
921	<	for (ForkJoinTask<?> t; (t = poll()) != null; )
922	<	ForkJoinTask.cancelIgnoringExceptions(t);
942	>	return stat;
943		}
944
945		/**
946	<	* Computes next value for random probes. Scans don't require
947	<	* a very high quality generator, but also not a crummy one.
928	<	* Marsaglia xor-shift is cheap and works well enough. Note:
929	<	* This is manually inlined in its usages in ForkJoinPool to
930	<	* avoid writes inside busy scan loops.
931	<	*/
932	<	final int nextSeed() {
933	<	int r = seed;
934	<	r ^= r << 13;
935	<	r ^= r >>> 17;
936	<	return seed = r ^= r << 5;
937	<	}
938	<
939	<	// Execution methods
940	<
941	<	/**
942	<	* Removes and runs tasks until empty, using local mode
943	<	* ordering. Normally called only after checking for apparent
944	<	* non-emptiness.
946	>	* Polls for and executes the given task or any other task in
947	>	* its CountedCompleter computation
948		*/
949	<	final void runLocalTasks() {
950	<	// hoist checks from repeated pop/poll
951	<	ForkJoinTask<?>[] a; int m;
952	<	if ((a = array) != null && (m = a.length - 1) >= 0) {
953	<	if (mode == 0) {
954	<	for (int s; (s = top - 1) - base >= 0;) {
955	<	int j = ((m & s) << ASHIFT) + ABASE;
956	<	ForkJoinTask<?> t =
957	<	(ForkJoinTask<?>)U.getObjectVolatile(a, j);
958	<	if (t != null) {
959	<	if (U.compareAndSwapObject(a, j, t, null)) {
960	<	top = s;
961	<	t.doExec();
962	<	}
949	>	final boolean pollAndExecCC(ForkJoinTask<?> root) {
950	>	ForkJoinTask<?>[] a; int b; Object o;
951	>	outer: while ((b = base) - top < 0 && (a = array) != null) {
952	>	long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
953	>	if ((o = U.getObject(a, j)) == null \|\|
954	>	!(o instanceof CountedCompleter))
955	>	break;
956	>	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
957	>	if (r == root) {
958	>	if (base == b &&
959	>	U.compareAndSwapObject(a, j, t, null)) {
960	>	base = b + 1;
961	>	t.doExec();
962	>	return true;
963		}
964		else
965	<	break;
963	<	}
964	<	}
965	<	else {
966	<	for (int b; (b = base) - top < 0;) {
967	<	int j = ((m & b) << ASHIFT) + ABASE;
968	<	ForkJoinTask<?> t =
969	<	(ForkJoinTask<?>)U.getObjectVolatile(a, j);
970	<	if (t != null) {
971	<	if (base == b &&
972	<	U.compareAndSwapObject(a, j, t, null)) {
973	<	base = b + 1;
974	<	t.doExec();
975	<	}
976	<	} else if (base == b) {
977	<	if (b + 1 == top)
978	<	break;
979	<	Thread.yield(); // wait for lagging update
980	<	}
965	>	break; // restart
966		}
967	+	if ((r = r.completer) == null)
968	+	break outer; // not part of root computation
969		}
970		}
971	+	return false;
972		}
973
974		/**
975		* Executes a top-level task and any local tasks remaining
976		* after execution.
989	–	*
990	–	* @return true unless terminating
977		*/
978	<	final boolean runTask(ForkJoinTask<?> t) {
993	<	boolean alive = true;
978	>	final void runTask(ForkJoinTask<?> t) {
979		if (t != null) {
980	<	currentSteal = t;
996	<	t.doExec();
997	<	if (top != base) // conservative guard
998	<	runLocalTasks();
999	<	++nsteals;
980	>	(currentSteal = t).doExec();
981		currentSteal = null;
982	+	if (++nsteals < 0) { // spill on overflow
983	+	ForkJoinPool p;
984	+	if ((p = pool) != null)
985	+	p.collectStealCount(this);
986	+	}
987	+	if (top != base) { // process remaining local tasks
988	+	if (mode == 0)
989	+	popAndExecAll();
990	+	else
991	+	pollAndExecAll();
992	+	}
993		}
1002	–	else if (runState < 0) // terminating
1003	–	alive = false;
1004	–	return alive;
994		}
995
996		/**
#	Line 1010 \| Line 999 \| public class ForkJoinPool extends Abstra
999		final void runSubtask(ForkJoinTask<?> t) {
1000		if (t != null) {
1001		ForkJoinTask<?> ps = currentSteal;
1002	<	currentSteal = t;
1014	<	t.doExec();
1002	>	(currentSteal = t).doExec();
1003		currentSteal = ps;
1004		}
1005		}
#	Line 1046 \| 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 1055 \| 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 1071 \| Line 1059 \| public class ForkJoinPool extends Abstra
1059		/**
1060		* Per-thread records for threads that submit to pools. Currently
1061		* holds only pseudo-random seed / index that is used to choose
1062	<	* submission queues in method doSubmit. In the future, this may
1062	>	* submission queues in method externalPush. In the future, this may
1063		* also incorporate a means to implement different task rejection
1064		* and resubmission policies.
1065		*
#	Line 1079 \| Line 1067 \| public class ForkJoinPool extends Abstra
1067		* the same way but are initialized and updated using slightly
1068		* different mechanics. Both are initialized using the same
1069		* approach as in class ThreadLocal, where successive values are
1070	<	* unlikely to collide with previous values. This is done during
1071	<	* registration for workers, but requires a separate AtomicInteger
1072	<	* for submitters. Seeds are then randomly modified upon
1085	<	* collisions using xorshifts, which requires a non-zero seed.
1070	>	* unlikely to collide with previous values. Seeds are then
1071	>	* randomly modified upon collisions using xorshifts, which
1072	>	* requires a non-zero seed.
1073		*/
1074		static final class Submitter {
1075		int seed;
1076	<	Submitter() {
1090	<	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1091	<	seed = (s == 0) ? 1 : s; // ensure non-zero
1092	<	}
1076	>	Submitter(int s) { seed = s; }
1077		}
1078
1079	<	/** ThreadLocal class for Submitters */
1080	<	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1081	<	public Submitter initialValue() { return new Submitter(); }
1098	<	}
1079	>	/** Property prefix for constructing common pool */
1080	>	private static final String propPrefix =
1081	>	"java.util.concurrent.ForkJoinPool.common.";
1082
1083		// static fields (initialized in static initializer below)
1084
#	Line 1107 \| Line 1090 \| public class ForkJoinPool extends Abstra
1090		defaultForkJoinWorkerThreadFactory;
1091
1092		/**
1093	<	* Generator for assigning sequence numbers as pool names.
1093	>	* Common (static) pool. Non-null for public use unless a static
1094	>	* construction exception, but internal usages null-check on use
1095	>	* to paranoically avoid potential initialization circularities
1096	>	* as well as to simplify generated code.
1097		*/
1098	<	private static final AtomicInteger poolNumberGenerator;
1113	<
1114	<	/**
1115	<	* Generator for initial hashes/seeds for submitters. Accessed by
1116	<	* Submitter class constructor.
1117	<	*/
1118	<	static final AtomicInteger nextSubmitterSeed;
1098	>	static final ForkJoinPool commonPool;
1099
1100		/**
1101		* Permission required for callers of methods that may start or
#	Line 1124 \| Line 1104 \| public class ForkJoinPool extends Abstra
1104		private static final RuntimePermission modifyThreadPermission;
1105
1106		/**
1107	<	* Per-thread submission bookeeping. Shared across all pools
1107	>	* Per-thread submission bookkeeping. Shared across all pools
1108		* to reduce ThreadLocal pollution and because random motion
1109		* to avoid contention in one pool is likely to hold for others.
1110	+	* Lazily initialized on first submission (but null-checked
1111	+	* in other contexts to avoid unnecessary initialization).
1112	+	*/
1113	+	static final ThreadLocal<Submitter> submitters;
1114	+
1115	+	/**
1116	+	* Common pool parallelism. Must equal commonPool.parallelism.
1117		*/
1118	<	private static final ThreadSubmitter submitters;
1118	>	static final int commonPoolParallelism;
1119	>
1120	>	/**
1121	>	* Sequence number for creating workerNamePrefix.
1122	>	*/
1123	>	private static int poolNumberSequence;
1124	>
1125	>	/**
1126	>	* Return the next sequence number. We don't expect this to
1127	>	* ever contend so use simple builtin sync.
1128	>	*/
1129	>	private static final synchronized int nextPoolId() {
1130	>	return ++poolNumberSequence;
1131	>	}
1132
1133		// static constants
1134
1135		/**
1136	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
1137	<	* task when the pool is quiescent to instead try to shrink the
1138	<	* number of workers. The exact value does not matter too
1139	<	* much. It must be short enough to release resources during
1140	<	* sustained periods of idleness, but not so short that threads
1141	<	* are continually re-created.
1136	>	* Initial timeout value (in nanoseconds) for the thread
1137	>	* triggering quiescence to park waiting for new work. On timeout,
1138	>	* the thread will instead try to shrink the number of
1139	>	* workers. The value should be large enough to avoid overly
1140	>	* aggressive shrinkage during most transient stalls (long GCs
1141	>	* etc).
1142		*/
1143	<	private static final long SHRINK_RATE =
1144	<	4L * 1000L * 1000L * 1000L; // 4 seconds
1143	>	private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
1144
1145		/**
1146	<	* The timeout value for attempted shrinkage, includes
1148	<	* some slop to cope with system timer imprecision.
1146	>	* Timeout value when there are more threads than parallelism level
1147		*/
1148	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1148	>	private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
1149
1150		/**
1151		* The maximum stolen->joining link depth allowed in method
1152	<	* tryHelpStealer. Must be a power of two. This value also
1155	<	* controls the maximum number of times to try to help join a task
1156	<	* without any apparent progress or change in pool state before
1157	<	* giving up and blocking (see awaitJoin). Depths for legitimate
1152	>	* tryHelpStealer. Must be a power of two. Depths for legitimate
1153		* chains are unbounded, but we use a fixed constant to avoid
1154		* (otherwise unchecked) cycles and to bound staleness of
1155		* traversal parameters at the expense of sometimes blocking when
1156		* we could be helping.
1157		*/
1158	<	private static final int MAX_HELP = 32;
1164	<
1165	<	/**
1166	<	* Secondary time-based bound (in nanosecs) for helping attempts
1167	<	* before trying compensated blocking in awaitJoin. Used in
1168	<	* conjunction with MAX_HELP to reduce variance due to different
1169	<	* polling rates associated with different helping options. The
1170	<	* value should roughly approximate the time required to create
1171	<	* and/or activate a worker thread.
1172	<	*/
1173	<	private static final long COMPENSATION_DELAY = 100L * 1000L; // 0.1 millisec
1158	>	private static final int MAX_HELP = 64;
1159
1160		/**
1161		* Increment for seed generators. See class ThreadLocal for
#	Line 1206 \| Line 1191 \| public class ForkJoinPool extends Abstra
1191		* scan for them to avoid queuing races. Note however that
1192		* eventCount updates lag releases so usage requires care.
1193		*
1194	<	* Field runState is an int packed with:
1194	>	* Field plock is an int packed with:
1195		* SHUTDOWN: true if shutdown is enabled (1 bit)
1196	<	* SEQ: a sequence number updated upon (de)registering workers (30 bits)
1197	<	* INIT: set true after workQueues array construction (1 bit)
1196	>	* SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
1197	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1198		*
1199		* The sequence number enables simple consistency checks:
1200		* Staleness of read-only operations on the workQueues array can
1201	<	* be checked by comparing runState before vs after the reads.
1201	>	* be checked by comparing plock before vs after the reads.
1202		*/
1203
1204		// bit positions/shifts for fields
#	Line 1225 \| Line 1210 \| public class ForkJoinPool extends Abstra
1210		// bounds
1211		private static final int SMASK = 0xffff; // short bits
1212		private static final int MAX_CAP = 0x7fff; // max #workers - 1
1213	<	private static final int SQMASK = 0xfffe; // even short bits
1213	>	private static final int EVENMASK = 0xfffe; // even short bits
1214	>	private static final int SQMASK = 0x007e; // max 64 (even) slots
1215		private static final int SHORT_SIGN = 1 << 15;
1216		private static final int INT_SIGN = 1 << 31;
1217
#	Line 1250 \| Line 1236 \| public class ForkJoinPool extends Abstra
1236		private static final int E_MASK = 0x7fffffff; // no STOP_BIT
1237		private static final int E_SEQ = 1 << EC_SHIFT;
1238
1239	<	// runState bits
1239	>	// plock bits
1240		private static final int SHUTDOWN = 1 << 31;
1241	+	private static final int PL_LOCK = 2;
1242	+	private static final int PL_SIGNAL = 1;
1243	+	private static final int PL_SPINS = 1 << 8;
1244
1245		// access mode for WorkQueue
1246		static final int LIFO_QUEUE = 0;
#	Line 1266 \| Line 1255 \| public class ForkJoinPool extends Abstra
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
1263	<	int nextSeed; // for initializing worker seeds
1275	<	volatile int runState; // shutdown status and seq
1262	>	volatile int indexSeed; // worker/submitter index seed
1263	>	volatile int plock; // shutdown status and seqLock
1264		WorkQueue[] workQueues; // main registry
1277	–	final Mutex lock; // for registration
1278	–	final Condition termination; // for awaitTermination
1265		final ForkJoinWorkerThreadFactory factory; // factory for new workers
1266		final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1281	–	final AtomicLong stealCount; // collect counts when terminated
1282	–	final AtomicInteger nextWorkerNumber; // to create worker name string
1267		final String workerNamePrefix; // to create worker name string
1268
1269	<	// Creating, registering, and deregistering workers
1270	<
1271	<	/**
1272	<	* Tries to create and start a worker
1273	<	*/
1274	<	private void addWorker() {
1275	<	Throwable ex = null;
1276	<	ForkJoinWorkerThread wt = null;
1277	<	try {
1278	<	if ((wt = factory.newThread(this)) != null) {
1279	<	wt.start();
1280	<	return;
1269	>	/*
1270	>	* Acquires the plock lock to protect worker array and related
1271	>	* updates. This method is called only if an initial CAS on plock
1272	>	* fails. This acts as a spinLock for normal cases, but falls back
1273	>	* to builtin monitor to block when (rarely) needed. This would be
1274	>	* a terrible idea for a highly contended lock, but works fine as
1275	>	* a more conservative alternative to a pure spinlock. See
1276	>	* internal ConcurrentHashMap documentation for further
1277	>	* explanation of nearly the same construction.
1278	>	*/
1279	>	private int acquirePlock() {
1280	>	int spins = PL_SPINS, r = 0, ps, nps;
1281	>	for (;;) {
1282	>	if (((ps = plock) & PL_LOCK) == 0 &&
1283	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1284	>	return nps;
1285	>	else if (r == 0)
1286	>	r = ThreadLocalRandom.current().nextInt(); // randomize spins
1287	>	else if (spins >= 0) {
1288	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1289	>	if (r >= 0)
1290	>	--spins;
1291	>	}
1292	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps \| PL_SIGNAL)) {
1293	>	synchronized (this) {
1294	>	if ((plock & PL_SIGNAL) != 0) {
1295	>	try {
1296	>	wait();
1297	>	} catch (InterruptedException ie) {
1298	>	try {
1299	>	Thread.currentThread().interrupt();
1300	>	} catch (SecurityException ignore) {
1301	>	}
1302	>	}
1303	>	}
1304	>	else
1305	>	notifyAll();
1306	>	}
1307		}
1298	–	} catch (Throwable e) {
1299	–	ex = e;
1308		}
1301	–	deregisterWorker(wt, ex); // adjust counts etc on failure
1309		}
1310
1311		/**
1312	<	* Callback from ForkJoinWorkerThread constructor to assign a
1313	<	* public name. This must be separate from registerWorker because
1307	<	* it is called during the "super" constructor call in
1308	<	* ForkJoinWorkerThread.
1312	>	* Unlocks and signals any thread waiting for plock. Called only
1313	>	* when CAS of seq value for unlock fails.
1314		*/
1315	<	final String nextWorkerName() {
1316	<	return workerNamePrefix.concat
1317	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1315	>	private void releasePlock(int ps) {
1316	>	plock = ps;
1317	>	synchronized (this) { notifyAll(); }
1318		}
1319
1320	+	// Registering and deregistering workers
1321	+
1322		/**
1323		* Callback from ForkJoinWorkerThread constructor to establish its
1324		* poolIndex and record its WorkQueue. To avoid scanning bias due
#	Line 1322 \| Line 1329 \| public class ForkJoinPool extends Abstra
1329		* @param w the worker's queue
1330		*/
1331		final void registerWorker(WorkQueue w) {
1332	<	Mutex lock = this.lock;
1333	<	lock.lock();
1332	>	int s, ps; // generate a rarely colliding candidate index seed
1333	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED,
1334	>	s = indexSeed, s += SEED_INCREMENT) \|\|
1335	>	s == 0); // skip 0
1336	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1337	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1338	>	ps = acquirePlock();
1339	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1340		try {
1341	<	WorkQueue[] ws = workQueues;
1342	<	if (w != null && ws != null) { // skip on shutdown/failure
1343	<	int rs, n;
1344	<	while ((n = ws.length) < // ensure can hold total
1332	<	(parallelism + (short)(ctl >>> TC_SHIFT) << 1))
1333	<	workQueues = ws = Arrays.copyOf(ws, n << 1);
1334	<	int m = n - 1;
1335	<	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1336	<	w.seed = (s == 0) ? 1 : s; // ensure non-zero seed
1341	>	WorkQueue[] ws;
1342	>	if (w != null && (ws = workQueues) != null) {
1343	>	w.seed = s;
1344	>	int n = ws.length, m = n - 1;
1345		int r = (s << 1) \| 1; // use odd-numbered indices
1346	<	while (ws[r &= m] != null) // step by approx half size
1347	<	r += ((n >>> 1) & SQMASK) + 2;
1346	>	if (ws[r &= m] != null) { // collision
1347	>	int probes = 0; // step by approx half size
1348	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1349	>	while (ws[r = (r + step) & m] != null) {
1350	>	if (++probes >= n) {
1351	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1352	>	m = n - 1;
1353	>	probes = 0;
1354	>	}
1355	>	}
1356	>	}
1357		w.eventCount = w.poolIndex = r; // establish before recording
1358	<	ws[r] = w; // also update seq
1342	<	runState = ((rs = runState) & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN);
1358	>	ws[r] = w;
1359		}
1360		} finally {
1361	<	lock.unlock();
1361	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1362	>	releasePlock(nps);
1363		}
1364		}
1365
1366		/**
1367		* Final callback from terminating worker, as well as upon failure
1368	<	* to construct or start a worker in addWorker. Removes record of
1369	<	* worker from array, and adjusts counts. If pool is shutting
1370	<	* down, tries to complete termination.
1368	>	* to construct or start a worker. Removes record of worker from
1369	>	* array, and adjusts counts. If pool is shutting down, tries to
1370	>	* complete termination.
1371		*
1372	<	* @param wt the worker thread or null if addWorker failed
1372	>	* @param wt the worker thread or null if construction failed
1373		* @param ex the exception causing failure, or null if none
1374		*/
1375		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1359	–	Mutex lock = this.lock;
1376		WorkQueue w = null;
1377		if (wt != null && (w = wt.workQueue) != null) {
1378	<	w.runState = -1; // ensure runState is set
1379	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1380	<	int idx = w.poolIndex;
1381	<	lock.lock();
1382	<	try { // remove record from array
1378	>	int ps;
1379	>	collectStealCount(w);
1380	>	w.qlock = -1; // ensure set
1381	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1382	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1383	>	ps = acquirePlock();
1384	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1385	>	try {
1386	>	int idx = w.poolIndex;
1387		WorkQueue[] ws = workQueues;
1388		if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1389		ws[idx] = null;
1390		} finally {
1391	<	lock.unlock();
1391	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1392	>	releasePlock(nps);
1393		}
1394		}
1395
#	Line 1381 \| Line 1402 \| public class ForkJoinPool extends Abstra
1402		if (!tryTerminate(false, false) && w != null) {
1403		w.cancelAll(); // cancel remaining tasks
1404		if (w.array != null) // suppress signal if never ran
1405	<	signalWork(); // wake up or create replacement
1405	>	signalWork(null, 1); // wake up or create replacement
1406		if (ex == null) // help clean refs on way out
1407		ForkJoinTask.helpExpungeStaleExceptions();
1408		}
1409
1410		if (ex != null) // rethrow
1411	<	U.throwException(ex);
1411	>	ForkJoinTask.rethrow(ex);
1412		}
1413
1414	+	/**
1415	+	* Collect worker steal count into total. Called on termination
1416	+	* and upon int overflow of local count. (There is a possible race
1417	+	* in the latter case vs any caller of getStealCount, which can
1418	+	* make its results less accurate than usual.)
1419	+	*/
1420	+	final void collectStealCount(WorkQueue w) {
1421	+	if (w != null) {
1422	+	long sc;
1423	+	int ns = w.nsteals;
1424	+	w.nsteals = 0; // handle overflow
1425	+	long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
1426	+	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1427	+	sc = stealCount, sc + steals));
1428	+	}
1429	+	}
1430
1431		// Submissions
1432
1433		/**
1434		* Unless shutting down, adds the given task to a submission queue
1435		* at submitter's current queue index (modulo submission
1436	<	* range). If no queue exists at the index, one is created. If
1437	<	* the queue is busy, another index is randomly chosen. The
1401	<	* submitMask bounds the effective number of queues to the
1402	<	* (nearest poswer of two for) parallelism level.
1436	>	* range). Only the most common path is directly handled in this
1437	>	* method. All others are relayed to fullExternalPush.
1438		*
1439		* @param task the task. Caller must ensure non-null.
1440		*/
1441	<	private void doSubmit(ForkJoinTask<?> task) {
1442	<	Submitter s = submitters.get();
1443	<	for (int r = s.seed, m = submitMask;;) {
1444	<	WorkQueue[] ws; WorkQueue q;
1445	<	int k = r & m & SQMASK; // use only even indices
1446	<	if (runState < 0 \|\| (ws = workQueues) == null \|\| ws.length <= k)
1447	<	throw new RejectedExecutionException(); // shutting down
1448	<	else if ((q = ws[k]) == null) { // create new queue
1449	<	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1450	<	Mutex lock = this.lock; // construct outside lock
1451	<	lock.lock();
1452	<	try { // recheck under lock
1453	<	int rs = runState; // to update seq
1454	<	if (ws == workQueues && ws[k] == null) {
1420	<	ws[k] = nq;
1421	<	runState = ((rs & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN));
1422	<	}
1423	<	} finally {
1424	<	lock.unlock();
1425	<	}
1426	<	}
1427	<	else if (q.trySharedPush(task)) {
1428	<	signalWork();
1441	>	final void externalPush(ForkJoinTask<?> task) {
1442	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1443	>	if ((z = submitters.get()) != null && plock > 0 &&
1444	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1445	>	(q = ws[m & z.seed & SQMASK]) != null &&
1446	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1447	>	int s = q.top, n;
1448	>	if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
1449	>	U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
1450	>	task);
1451	>	q.top = s + 1; // push on to deque
1452	>	q.qlock = 0;
1453	>	if (n <= 1)
1454	>	signalWork(q, 1);
1455		return;
1456		}
1457	<	else if (m > 1) { // move to a different index
1457	>	q.qlock = 0;
1458	>	}
1459	>	fullExternalPush(task);
1460	>	}
1461	>
1462	>	/**
1463	>	* Full version of externalPush. This method is called, among
1464	>	* other times, upon the first submission of the first task to the
1465	>	* pool, so must perform secondary initialization: creating
1466	>	* workQueue array and setting plock to a valid value. It also
1467	>	* detects first submission by an external thread by looking up
1468	>	* its ThreadLocal, and creates a new shared queue if the one at
1469	>	* index if empty or contended. The lock bodies must be
1470	>	* exception-free (so no try/finally) so we optimistically
1471	>	* allocate new queues/arrays outside the locks and throw them
1472	>	* away if (very rarely) not needed. Note that the plock seq value
1473	>	* can eventually wrap around zero, but if so harmlessly fails to
1474	>	* reinitialize.
1475	>	*/
1476	>	private void fullExternalPush(ForkJoinTask<?> task) {
1477	>	for (Submitter z = null;;) {
1478	>	WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
1479	>	if ((ps = plock) < 0)
1480	>	throw new RejectedExecutionException();
1481	>	else if ((ws = workQueues) == null \|\| (m = ws.length - 1) < 0) {
1482	>	int n = parallelism - 1; n \|= n >>> 1; n \|= n >>> 2;
1483	>	n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
1484	>	WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
1485	>	if ((ps & PL_LOCK) != 0 \|\|
1486	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1487	>	ps = acquirePlock();
1488	>	if ((ws = workQueues) == null)
1489	>	workQueues = nws;
1490	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1491	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1492	>	releasePlock(nps);
1493	>	}
1494	>	else if (z == null && (z = submitters.get()) == null) {
1495	>	if (U.compareAndSwapInt(this, INDEXSEED,
1496	>	s = indexSeed, s += SEED_INCREMENT) &&
1497	>	s != 0) // skip 0
1498	>	submitters.set(z = new Submitter(s));
1499	>	}
1500	>	else {
1501	>	int k = (r = z.seed) & m & SQMASK;
1502	>	if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
1503	>	(q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
1504	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1505	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1506	>	ps = acquirePlock();
1507	>	WorkQueue w = null;
1508	>	if ((ws = workQueues) != null && k < ws.length &&
1509	>	(w = ws[k]) == null)
1510	>	ws[k] = q;
1511	>	else
1512	>	q = w;
1513	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1514	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1515	>	releasePlock(nps);
1516	>	}
1517	>	if (q != null && q.qlock == 0 && q.fullPush(task, false))
1518	>	return;
1519		r ^= r << 13; // same xorshift as WorkQueues
1520		r ^= r >>> 17;
1521	<	s.seed = r ^= r << 5;
1521	>	z.seed = r ^= r << 5; // move to a different index
1522		}
1436	–	else
1437	–	Thread.yield(); // yield if no alternatives
1523		}
1524		}
1525
#	Line 1449 \| Line 1534 \| public class ForkJoinPool extends Abstra
1534		}
1535
1536		/**
1537	<	* Tries to activate or create a worker if too few are active.
1538	<	*/
1539	<	final void signalWork() {
1540	<	long c; int u;
1541	<	while ((u = (int)((c = ctl) >>> 32)) < 0) { // too few active
1542	<	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1543	<	if ((e = (int)c) > 0) { // at least one waiting
1544	<	if (ws != null && (i = e & SMASK) < ws.length &&
1537	>	* Tries to create (at most one) or activate (possibly several)
1538	>	* workers if too few are active. On contention failure, continues
1539	>	* until at least one worker is signalled or the given queue is
1540	>	* empty or all workers are active.
1541	>	*
1542	>	* @param q if non-null, the queue holding tasks to be signalled
1543	>	* @param signals the target number of signals.
1544	>	*/
1545	>	final void signalWork(WorkQueue q, int signals) {
1546	>	long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
1547	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1548	>	if ((e = (int)c) > 0) {
1549	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1550		(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1551		long nc = (((long)(w.nextWait & E_MASK)) \|
1552		((long)(u + UAC_UNIT) << 32));
1553		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1554		w.eventCount = (e + E_SEQ) & E_MASK;
1555		if ((p = w.parker) != null)
1556	<	U.unpark(p); // activate and release
1557	<	break;
1556	>	U.unpark(p);
1557	>	if (--signals <= 0)
1558	>	break;
1559		}
1560	+	else
1561	+	signals = 1;
1562	+	if ((q != null && q.queueSize() == 0))
1563	+	break;
1564		}
1565		else
1566		break;
1567		}
1568	<	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1568	>	else if (e == 0 && (u & SHORT_SIGN) != 0) {
1569		long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1570		((u + UAC_UNIT) & UAC_MASK)) << 32;
1571		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1572	<	addWorker();
1572	>	ForkJoinWorkerThread wt = null;
1573	>	Throwable ex = null;
1574	>	boolean started = false;
1575	>	try {
1576	>	ForkJoinWorkerThreadFactory fac;
1577	>	if ((fac = factory) != null &&
1578	>	(wt = fac.newThread(this)) != null) {
1579	>	wt.start();
1580	>	started = true;
1581	>	}
1582	>	} catch (Throwable rex) {
1583	>	ex = rex;
1584	>	}
1585	>	if (!started)
1586	>	deregisterWorker(wt, ex); // adjust counts on failure
1587		break;
1588		}
1589		}
#	Line 1483 \| Line 1592 \| public class ForkJoinPool extends Abstra
1592		}
1593		}
1594
1486	–
1595		// Scanning for tasks
1596
1597		/**
1598		* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1599		*/
1600		final void runWorker(WorkQueue w) {
1601	<	w.growArray(false); // initialize queue array in this thread
1602	<	do {} while (w.runTask(scan(w)));
1601	>	// initialize queue array in this thread
1602	>	w.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY];
1603	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1604		}
1605
1606		/**
#	Line 1507 \| Line 1616 \| public class ForkJoinPool extends Abstra
1616		* relative prime, checking each at least once). The scan
1617		* terminates upon either finding a non-empty queue, or completing
1618		* the sweep. If the worker is not inactivated, it takes and
1619	<	* returns a task from this queue. On failure to find a task, we
1619	>	* returns a task from this queue. Otherwise, if not activated, it
1620	>	* signals workers (that may include itself) and returns so caller
1621	>	* can retry. Also returns for trtry if the worker array may have
1622	>	* changed during an empty scan. On failure to find a task, we
1623		* take one of the following actions, after which the caller will
1624		* retry calling this method unless terminated.
1625		*
1626		* * If pool is terminating, terminate the worker.
1627		*
1516	–	* * If not a complete sweep, try to release a waiting worker. If
1517	–	* the scan terminated because the worker is inactivated, then the
1518	–	* released worker will often be the calling worker, and it can
1519	–	* succeed obtaining a task on the next call. Or maybe it is
1520	–	* another worker, but with same net effect. Releasing in other
1521	–	* cases as well ensures that we have enough workers running.
1522	–	*
1628		* * If not already enqueued, try to inactivate and enqueue the
1629		* worker on wait queue. Or, if inactivating has caused the pool
1630		* to be quiescent, relay to idleAwaitWork to check for
1631		* termination and possibly shrink pool.
1632		*
1633	<	* * If already inactive, and the caller has run a task since the
1634	<	* last empty scan, return (to allow rescan) unless others are
1635	<	* also inactivated. Field WorkQueue.rescans counts down on each
1531	<	* scan to ensure eventual inactivation and blocking.
1532	<	*
1533	<	* * If already enqueued and none of the above apply, park
1534	<	* awaiting signal,
1633	>	* * If already enqueued and none of the above apply, possibly
1634	>	* (with 1/2 probability) park awaiting signal, else lingering to
1635	>	* help scan and signal.
1636		*
1637		* @param w the worker (via its WorkQueue)
1638	<	* @return a task or null of none found
1638	>	* @return a task or null if none found
1639		*/
1640		private final ForkJoinTask<?> scan(WorkQueue w) {
1641	<	WorkQueue[] ws; // first update random seed
1641	>	WorkQueue[] ws; WorkQueue q; // first update random seed
1642		int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1643	<	int rs = runState, m; // volatile read order matters
1643	>	int ps = plock, m; // volatile read order matters
1644		if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1645	<	int ec = w.eventCount; // ec is negative if inactive
1646	<	int step = (r >>> 16) \| 1; // relative prime
1647	<	for (int j = (m + 1) << 2; ; r += step) {
1648	<	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1645	>	int ec = w.eventCount; // ec is negative if inactive
1646	>	int step = (r >>> 16) \| 1; // relatively prime
1647	>	for (int j = (m + 1) << 2; ; --j, r += step) {
1648	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b, n;
1649		if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1650	<	(a = q.array) != null) { // probably nonempty
1650	>	(a = q.array) != null) { // probably nonempty
1651		int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1652		t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1653		if (q.base == b && ec >= 0 && t != null &&
1654		U.compareAndSwapObject(a, i, t, null)) {
1655	<	q.base = b + 1; // specialization of pollAt
1656	<	return t;
1657	<	}
1658	<	else if ((t != null \|\| b + 1 != q.top) &&
1659	<	(ec < 0 \|\| j <= m)) {
1660	<	rs = 0; // mark scan as imcomplete
1661	<	break; // caller can retry after release
1655	>	if ((n = q.top - (q.base = b + 1)) > 0)
1656	>	signalWork(q, n);
1657	>	return t; // taken
1658	>	}
1659	>	if (j < m \|\| (ec < 0 && (ec = w.eventCount) < 0)) {
1660	>	if ((n = q.queueSize() - 1) > 0)
1661	>	signalWork(q, n);
1662	>	break; // let caller retry after signal
1663		}
1664		}
1665	<	if (--j < 0)
1666	<	break;
1667	<	}
1668	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1669	<	if (e < 0) // decode ctl on empty scan
1670	<	w.runState = -1; // pool is terminating
1671	<	else if (rs == 0 \|\| rs != runState) { // incomplete scan
1672	<	WorkQueue v; Thread p; // try to release a waiter
1673	<	if (e > 0 && a < 0 && w.eventCount == ec &&
1674	<	(v = ws[e & m]) != null && v.eventCount == (e \| INT_SIGN)) {
1675	<	long nc = ((long)(v.nextWait & E_MASK) \|
1676	<	((c + AC_UNIT) & (AC_MASK\|TC_MASK)));
1677	<	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1678	<	v.eventCount = (e + E_SEQ) & E_MASK;
1679	<	if ((p = v.parker) != null)
1680	<	U.unpark(p);
1665	>	else if (j < 0) { // end of scan
1666	>	long c = ctl; int e;
1667	>	if (plock != ps) // incomplete sweep
1668	>	break;
1669	>	if ((e = (int)c) < 0) // pool is terminating
1670	>	w.qlock = -1;
1671	>	else if (ec >= 0) { // try to enqueue/inactivate
1672	>	long nc = ((long)ec \|
1673	>	((c - AC_UNIT) & (AC_MASK\|TC_MASK)));
1674	>	w.nextWait = e;
1675	>	w.eventCount = ec \| INT_SIGN; // mark as inactive
1676	>	if (ctl != c \|\|
1677	>	!U.compareAndSwapLong(this, CTL, c, nc))
1678	>	w.eventCount = ec; // unmark on CAS failure
1679	>	else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
1680	>	idleAwaitWork(w, nc, c); // quiescent
1681	>	}
1682	>	else if (w.seed >= 0 && w.eventCount < 0) {
1683	>	Thread wt = Thread.currentThread();
1684	>	Thread.interrupted(); // clear status
1685	>	U.putObject(wt, PARKBLOCKER, this);
1686	>	w.parker = wt; // emulate LockSupport.park
1687	>	if (w.eventCount < 0) // recheck
1688	>	U.park(false, 0L);
1689	>	w.parker = null;
1690	>	U.putObject(wt, PARKBLOCKER, null);
1691		}
1692	<	}
1581	<	}
1582	<	else if (ec >= 0) { // try to enqueue/inactivate
1583	<	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1584	<	w.nextWait = e;
1585	<	w.eventCount = ec \| INT_SIGN; // mark as inactive
1586	<	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1587	<	w.eventCount = ec; // unmark on CAS failure
1588	<	else {
1589	<	if ((ns = w.nsteals) != 0) {
1590	<	w.nsteals = 0; // set rescans if ran task
1591	<	w.rescans = (a > 0) ? 0 : a + parallelism;
1592	<	w.totalSteals += ns;
1593	<	}
1594	<	if (a == 1 - parallelism) // quiescent
1595	<	idleAwaitWork(w, nc, c);
1596	<	}
1597	<	}
1598	<	else if (w.eventCount < 0) { // already queued
1599	<	if ((nr = w.rescans) > 0) { // continue rescanning
1600	<	int ac = a + parallelism;
1601	<	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0)
1602	<	Thread.yield(); // yield before block
1603	<	}
1604	<	else {
1605	<	Thread.interrupted(); // clear status
1606	<	Thread wt = Thread.currentThread();
1607	<	U.putObject(wt, PARKBLOCKER, this);
1608	<	w.parker = wt; // emulate LockSupport.park
1609	<	if (w.eventCount < 0) // recheck
1610	<	U.park(false, 0L);
1611	<	w.parker = null;
1612	<	U.putObject(wt, PARKBLOCKER, null);
1692	>	break;
1693		}
1694		}
1695		}
#	Line 1619 \| Line 1699 \| public class ForkJoinPool extends Abstra
1699		/**
1700		* If inactivating worker w has caused the pool to become
1701		* quiescent, checks for pool termination, and, so long as this is
1702	<	* not the only worker, waits for event for up to SHRINK_RATE
1703	<	* nanosecs. On timeout, if ctl has not changed, terminates the
1702	>	* not the only worker, waits for event for up to a given
1703	>	* duration. On timeout, if ctl has not changed, terminates the
1704		* worker, which will in turn wake up another worker to possibly
1705		* repeat this process.
1706		*
#	Line 1629 \| Line 1709 \| public class ForkJoinPool extends Abstra
1709		* @param prevCtl the ctl value to restore if thread is terminated
1710		*/
1711		private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1712	<	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1713	<	(int)prevCtl != 0 && ctl == currentCtl) {
1712	>	if (w.eventCount < 0 &&
1713	>	(this == commonPool \|\| !tryTerminate(false, false)) &&
1714	>	(int)prevCtl != 0) {
1715	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1716	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1717	>	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1718		Thread wt = Thread.currentThread();
1635	–	Thread.yield(); // yield before block
1719		while (ctl == currentCtl) {
1637	–	long startTime = System.nanoTime();
1720		Thread.interrupted(); // timed variant of version in scan()
1721		U.putObject(wt, PARKBLOCKER, this);
1722		w.parker = wt;
1723		if (ctl == currentCtl)
1724	<	U.park(false, SHRINK_RATE);
1724	>	U.park(false, parkTime);
1725		w.parker = null;
1726		U.putObject(wt, PARKBLOCKER, null);
1727		if (ctl != currentCtl)
1728		break;
1729	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1729	>	if (deadline - System.nanoTime() <= 0L &&
1730		U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1731		w.eventCount = (w.eventCount + E_SEQ) \| E_MASK;
1732	<	w.runState = -1; // shrink
1732	>	w.qlock = -1; // shrink
1733		break;
1734		}
1735		}
#	Line 1655 \| Line 1737 \| public class ForkJoinPool extends Abstra
1737		}
1738
1739		/**
1740	+	* Scans through queues looking for work while joining a task;
1741	+	* if any are present, signals.
1742	+	*
1743	+	* @param task to return early if done
1744	+	* @param origin an index to start scan
1745	+	*/
1746	+	final int helpSignal(ForkJoinTask<?> task, int origin) {
1747	+	WorkQueue[] ws; WorkQueue q; int m, n, s;
1748	+	if (task != null && (ws = workQueues) != null &&
1749	+	(m = ws.length - 1) >= 0) {
1750	+	for (int i = 0; i <= m; ++i) {
1751	+	if ((s = task.status) < 0)
1752	+	return s;
1753	+	if ((q = ws[(i + origin) & m]) != null &&
1754	+	(n = q.queueSize()) > 0) {
1755	+	signalWork(q, n);
1756	+	if ((int)(ctl >> AC_SHIFT) >= 0)
1757	+	break;
1758	+	}
1759	+	}
1760	+	}
1761	+	return 0;
1762	+	}
1763	+
1764	+	/**
1765		* Tries to locate and execute tasks for a stealer of the given
1766		* task, or in turn one of its stealers, Traces currentSteal ->
1767		* currentJoin links looking for a thread working on a descendant
#	Line 1665 \| Line 1772 \| public class ForkJoinPool extends Abstra
1772		* leaves hints in workers to speed up subsequent calls. The
1773		* implementation is very branchy to cope with potential
1774		* inconsistencies or loops encountering chains that are stale,
1775	<	* unknown, or so long that they are likely cyclic. All of these
1669	<	* cases are dealt with by just retrying by caller.
1775	>	* unknown, or so long that they are likely cyclic.
1776		*
1777		* @param joiner the joining worker
1778		* @param task the task to join
1779	<	* @return true if found or ran a task (and so is immediately retryable)
1779	>	* @return 0 if no progress can be made, negative if task
1780	>	* known complete, else positive
1781		*/
1782	<	private boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1783	<	WorkQueue[] ws;
1784	<	int m, depth = MAX_HELP; // remaining chain depth
1785	<	boolean progress = false;
1786	<	if ((ws = workQueues) != null && (m = ws.length - 1) > 0 &&
1787	<	task.status >= 0) {
1788	<	ForkJoinTask<?> subtask = task; // current target
1789	<	outer: for (WorkQueue j = joiner;;) {
1790	<	WorkQueue stealer = null; // find stealer of subtask
1791	<	WorkQueue v = ws[j.stealHint & m]; // try hint
1792	<	if (v != null && v.currentSteal == subtask)
1793	<	stealer = v;
1794	<	else { // scan
1795	<	for (int i = 1; i <= m; i += 2) {
1796	<	if ((v = ws[i]) != null && v.currentSteal == subtask &&
1797	<	v != joiner) {
1798	<	stealer = v;
1799	<	j.stealHint = i; // save hint
1800	<	break;
1782	>	private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1783	>	int stat = 0, steps = 0; // bound to avoid cycles
1784	>	if (joiner != null && task != null) { // hoist null checks
1785	>	restart: for (;;) {
1786	>	ForkJoinTask<?> subtask = task; // current target
1787	>	for (WorkQueue j = joiner, v;;) { // v is stealer of subtask
1788	>	WorkQueue[] ws; int m, s, h;
1789	>	if ((s = task.status) < 0) {
1790	>	stat = s;
1791	>	break restart;
1792	>	}
1793	>	if ((ws = workQueues) == null \|\| (m = ws.length - 1) <= 0)
1794	>	break restart; // shutting down
1795	>	if ((v = ws[h = (j.stealHint \| 1) & m]) == null \|\|
1796	>	v.currentSteal != subtask) {
1797	>	for (int origin = h;;) { // find stealer
1798	>	if (((h = (h + 2) & m) & 15) == 1 &&
1799	>	(subtask.status < 0 \|\| j.currentJoin != subtask))
1800	>	continue restart; // occasional staleness check
1801	>	if ((v = ws[h]) != null &&
1802	>	v.currentSteal == subtask) {
1803	>	j.stealHint = h; // save hint
1804	>	break;
1805	>	}
1806	>	if (h == origin)
1807	>	break restart; // cannot find stealer
1808		}
1809		}
1810	<	if (stealer == null)
1811	<	break;
1812	<	}
1813	<
1814	<	for (WorkQueue q = stealer;;) { // try to help stealer
1815	<	ForkJoinTask[] a; ForkJoinTask<?> t; int b;
1816	<	if (task.status < 0)
1817	<	break outer;
1818	<	if ((b = q.base) - q.top < 0 && (a = q.array) != null) {
1819	<	progress = true;
1820	<	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1821	<	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1822	<	if (subtask.status < 0) // must recheck before taking
1823	<	break outer;
1824	<	if (t != null &&
1825	<	q.base == b &&
1826	<	U.compareAndSwapObject(a, i, t, null)) {
1827	<	q.base = b + 1;
1828	<	joiner.runSubtask(t);
1810	>	for (;;) { // help stealer or descend to its stealer
1811	>	ForkJoinTask[] a; int b;
1812	>	if (subtask.status < 0) // surround probes with
1813	>	continue restart; // consistency checks
1814	>	if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1815	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1816	>	ForkJoinTask<?> t =
1817	>	(ForkJoinTask<?>)U.getObjectVolatile(a, i);
1818	>	if (subtask.status < 0 \|\| j.currentJoin != subtask \|\|
1819	>	v.currentSteal != subtask)
1820	>	continue restart; // stale
1821	>	stat = 1; // apparent progress
1822	>	if (t != null && v.base == b &&
1823	>	U.compareAndSwapObject(a, i, t, null)) {
1824	>	v.base = b + 1; // help stealer
1825	>	joiner.runSubtask(t);
1826	>	}
1827	>	else if (v.base == b && ++steps == MAX_HELP)
1828	>	break restart; // v apparently stalled
1829	>	}
1830	>	else { // empty -- try to descend
1831	>	ForkJoinTask<?> next = v.currentJoin;
1832	>	if (subtask.status < 0 \|\| j.currentJoin != subtask \|\|
1833	>	v.currentSteal != subtask)
1834	>	continue restart; // stale
1835	>	else if (next == null \|\| ++steps == MAX_HELP)
1836	>	break restart; // dead-end or maybe cyclic
1837	>	else {
1838	>	subtask = next;
1839	>	j = v;
1840	>	break;
1841	>	}
1842		}
1716	–	else if (q.base == b)
1717	–	break outer; // possibly stalled
1718	–	}
1719	–	else { // descend
1720	–	ForkJoinTask<?> next = stealer.currentJoin;
1721	–	if (--depth <= 0 \|\| subtask.status < 0 \|\|
1722	–	next == null \|\| next == subtask)
1723	–	break outer; // stale, dead-end, or cyclic
1724	–	subtask = next;
1725	–	j = stealer;
1726	–	break;
1843		}
1844		}
1845		}
1846		}
1847	<	return progress;
1847	>	return stat;
1848		}
1849
1850		/**
1851	<	* If task is at base of some steal queue, steals and executes it.
1851	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1852	>	* and run tasks within the target's computation
1853	>	*
1854	>	* @param task the task to join
1855	>	* @param mode if shared, exit upon completing any task
1856	>	* if all workers are active
1857		*
1737	–	* @param joiner the joining worker
1738	–	* @param task the task
1858		*/
1859	<	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1860	<	WorkQueue[] ws;
1861	<	if ((ws = workQueues) != null) {
1862	<	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1863	<	WorkQueue q = ws[j];
1864	<	if (q != null && q.pollFor(task)) {
1865	<	joiner.runSubtask(task);
1866	<	break;
1859	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1860	>	WorkQueue[] ws; WorkQueue q; int m, n, s;
1861	>	if (task != null && (ws = workQueues) != null &&
1862	>	(m = ws.length - 1) >= 0) {
1863	>	for (int j = 1, origin = j;;) {
1864	>	if ((s = task.status) < 0)
1865	>	return s;
1866	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1867	>	origin = j;
1868	>	if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
1869	>	break;
1870		}
1871	+	else if ((j = (j + 2) & m) == origin)
1872	+	break;
1873		}
1874		}
1875	+	return 0;
1876		}
1877
1878		/**
1879		* Tries to decrement active count (sometimes implicitly) and
1880		* possibly release or create a compensating worker in preparation
1881		* for blocking. Fails on contention or termination. Otherwise,
1882	<	* adds a new thread if no idle workers are available and either
1883	<	* pool would become completely starved or: (at least half
1759	<	* starved, and fewer than 50% spares exist, and there is at least
1760	<	* one task apparently available). Even though the availablity
1761	<	* check requires a full scan, it is worthwhile in reducing false
1762	<	* alarms.
1763	<	*
1764	<	* @param task if nonnull, a task being waited for
1765	<	* @param blocker if nonnull, a blocker being waited for
1766	<	* @return true if the caller can block, else should recheck and retry
1882	>	* adds a new thread if no idle workers are available and pool
1883	>	* may become starved.
1884		*/
1885	<	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1886	<	int pc = parallelism, e;
1887	<	long c = ctl;
1888	<	WorkQueue[] ws = workQueues;
1889	<	if ((e = (int)c) >= 0 && ws != null) {
1890	<	int u, a, ac, hc;
1891	<	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1892	<	boolean replace = false;
1893	<	if ((a = u >> UAC_SHIFT) <= 0) {
1894	<	if ((ac = a + pc) <= 1)
1895	<	replace = true;
1896	<	else if ((e > 0 \|\| (task != null &&
1897	<	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1781	<	WorkQueue w;
1782	<	for (int j = 0; j < ws.length; ++j) {
1783	<	if ((w = ws[j]) != null && !w.isEmpty()) {
1784	<	replace = true;
1785	<	break; // in compensation range and tasks available
1786	<	}
1787	<	}
1885	>	final boolean tryCompensate() {
1886	>	int pc = parallelism, e, u, i, tc; long c;
1887	>	WorkQueue[] ws; WorkQueue w; Thread p;
1888	>	if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
1889	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1890	>	(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1891	>	long nc = ((long)(w.nextWait & E_MASK) \|
1892	>	(c & (AC_MASK\|TC_MASK)));
1893	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1894	>	w.eventCount = (e + E_SEQ) & E_MASK;
1895	>	if ((p = w.parker) != null)
1896	>	U.unpark(p);
1897	>	return true; // replace with idle worker
1898		}
1899		}
1900	<	if ((task == null \|\| task.status >= 0) && // recheck need to block
1901	<	(blocker == null \|\| !blocker.isReleasable()) && ctl == c) {
1902	<	if (!replace) { // no compensation
1903	<	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1904	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1905	<	return true;
1906	<	}
1907	<	else if (e != 0) { // release an idle worker
1908	<	WorkQueue w; Thread p; int i;
1909	<	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
1910	<	long nc = ((long)(w.nextWait & E_MASK) \|
1911	<	(c & (AC_MASK\|TC_MASK)));
1912	<	if (w.eventCount == (e \| INT_SIGN) &&
1913	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1914	<	w.eventCount = (e + E_SEQ) & E_MASK;
1915	<	if ((p = w.parker) != null)
1806	<	U.unpark(p);
1900	>	else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
1901	>	(u >> UAC_SHIFT) + pc > 1) {
1902	>	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1903	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1904	>	return true; // no compensation
1905	>	}
1906	>	else if ((tc = u + pc) < MAX_CAP) {
1907	>	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1908	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1909	>	Throwable ex = null;
1910	>	ForkJoinWorkerThread wt = null;
1911	>	try {
1912	>	ForkJoinWorkerThreadFactory fac;
1913	>	if ((fac = factory) != null &&
1914	>	(wt = fac.newThread(this)) != null) {
1915	>	wt.start();
1916		return true;
1917		}
1918	+	} catch (Throwable rex) {
1919	+	ex = rex;
1920		}
1921	<	}
1811	<	else if (tc < MAX_CAP) { // create replacement
1812	<	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1813	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1814	<	addWorker();
1815	<	return true;
1816	<	}
1921	>	deregisterWorker(wt, ex); // adjust counts etc
1922		}
1923		}
1924		}
#	Line 1821 \| Line 1926 \| public class ForkJoinPool extends Abstra
1926		}
1927
1928		/**
1929	<	* Helps and/or blocks until the given task is done
1929	>	* Helps and/or blocks until the given task is done.
1930		*
1931		* @param joiner the joining worker
1932		* @param task the task
1933		* @return task status on exit
1934		*/
1935		final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1936	<	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1937	<	joiner.currentJoin = task;
1938	<	long startTime = 0L;
1939	<	for (int k = 0, s; ; ++k) {
1940	<	if ((joiner.isEmpty() ? // try to help
1941	<	!tryHelpStealer(joiner, task) :
1942	<	!joiner.tryRemoveAndExec(task))) {
1943	<	if (k == 0) {
1944	<	startTime = System.nanoTime();
1945	<	tryPollForAndExec(joiner, task); // check uncommon case
1946	<	}
1947	<	else if ((k & (MAX_HELP - 1)) == 0 &&
1948	<	System.nanoTime() - startTime >= COMPENSATION_DELAY &&
1949	<	tryCompensate(task, null)) {
1950	<	if (task.trySetSignal() && task.status >= 0) {
1936	>	int s = 0;
1937	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1938	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1939	>	joiner.currentJoin = task;
1940	>	do {} while ((s = task.status) >= 0 &&
1941	>	joiner.queueSize() > 0 &&
1942	>	joiner.tryRemoveAndExec(task)); // process local tasks
1943	>	if (s >= 0 && (s = task.status) >= 0 &&
1944	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1945	>	(task instanceof CountedCompleter))
1946	>	s = helpComplete(task, LIFO_QUEUE);
1947	>	while (s >= 0 && (s = task.status) >= 0) {
1948	>	if ((joiner.queueSize() > 0 \|\| // try helping
1949	>	(s = tryHelpStealer(joiner, task)) == 0) &&
1950	>	(s = task.status) >= 0 && tryCompensate()) {
1951	>	if (task.trySetSignal() && (s = task.status) >= 0) {
1952		synchronized (task) {
1953		if (task.status >= 0) {
1954		try { // see ForkJoinTask
#	Line 1859 \| Line 1965 \| public class ForkJoinPool extends Abstra
1965		(this, CTL, c = ctl, c + AC_UNIT));
1966		}
1967		}
1968	<	if ((s = task.status) < 0) {
1863	<	joiner.currentJoin = prevJoin;
1864	<	return s;
1865	<	}
1866	<	else if ((k & (MAX_HELP - 1)) == MAX_HELP >>> 1)
1867	<	Thread.yield(); // for politeness
1968	>	joiner.currentJoin = prevJoin;
1969		}
1970	+	return s;
1971		}
1972
1973		/**
#	Line 1875 \| Line 1977 \| public class ForkJoinPool extends Abstra
1977		*
1978		* @param joiner the joining worker
1979		* @param task the task
1878	–	* @return task status on exit
1980		*/
1981	<	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
1981	>	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
1982		int s;
1983	<	while ((s = task.status) >= 0 &&
1984	<	(joiner.isEmpty() ?
1985	<	tryHelpStealer(joiner, task) :
1986	<	joiner.tryRemoveAndExec(task)))
1987	<	;
1988	<	return s;
1983	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1984	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1985	>	joiner.currentJoin = task;
1986	>	do {} while ((s = task.status) >= 0 &&
1987	>	joiner.queueSize() > 0 &&
1988	>	joiner.tryRemoveAndExec(task));
1989	>	if (s >= 0 && (s = task.status) >= 0 &&
1990	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1991	>	(task instanceof CountedCompleter))
1992	>	s = helpComplete(task, LIFO_QUEUE);
1993	>	if (s >= 0 && joiner.queueSize() == 0) {
1994	>	do {} while (task.status >= 0 &&
1995	>	tryHelpStealer(joiner, task) > 0);
1996	>	}
1997	>	joiner.currentJoin = prevJoin;
1998	>	}
1999		}
2000
2001		/**
#	Line 1892 \| Line 2003 \| public class ForkJoinPool extends Abstra
2003		* during a random, then cyclic scan, else null. This method must
2004		* be retried by caller if, by the time it tries to use the queue,
2005		* it is empty.
2006	+	* @param r a (random) seed for scanning
2007		*/
2008	<	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;
2008	>	private WorkQueue findNonEmptyStealQueue(int r) {
2009		int step = (r >>> 16) \| 1;
2010		for (WorkQueue[] ws;;) {
2011	<	int rs = runState, m;
2011	>	int ps = plock, m;
2012		if ((ws = workQueues) == null \|\| (m = ws.length - 1) < 1)
2013		return null;
2014		for (int j = (m + 1) << 2; ; r += step) {
2015		WorkQueue q = ws[((r << 1) \| 1) & m];
2016	<	if (q != null && !q.isEmpty())
2016	>	if (q != null && q.queueSize() > 0)
2017		return q;
2018		else if (--j < 0) {
2019	<	if (runState == rs)
2019	>	if (plock == ps)
2020		return null;
2021		break;
2022		}
#	Line 1922 \| Line 2032 \| public class ForkJoinPool extends Abstra
2032		*/
2033		final void helpQuiescePool(WorkQueue w) {
2034		for (boolean active = true;;) {
2035	<	if (w.base - w.top < 0)
2036	<	w.runLocalTasks(); // exhaust local queue
2037	<	WorkQueue q = findNonEmptyStealQueue(w);
2035	>	ForkJoinTask<?> localTask; // exhaust local queue
2036	>	while ((localTask = w.nextLocalTask()) != null)
2037	>	localTask.doExec();
2038	>	// Similar to loop in scan(), but ignoring submissions
2039	>	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2040		if (q != null) {
2041		ForkJoinTask<?> t; int b;
2042		if (!active) { // re-establish active count
#	Line 1964 \| Line 2076 \| public class ForkJoinPool extends Abstra
2076		WorkQueue q; int b;
2077		if ((t = w.nextLocalTask()) != null)
2078		return t;
2079	<	if ((q = findNonEmptyStealQueue(w)) == null)
2079	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2080		return null;
2081		if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2082		return t;
#	Line 1972 \| Line 2084 \| public class ForkJoinPool extends Abstra
2084		}
2085
2086		/**
2087	<	* Returns the approximate (non-atomic) number of idle threads per
2088	<	* active thread to offset steal queue size for method
2089	<	* ForkJoinTask.getSurplusQueuedTaskCount().
2090	<	*/
2091	<	final int idlePerActive() {
2092	<	// Approximate at powers of two for small values, saturate past 4
2093	<	int p = parallelism;
2094	<	int a = p + (int)(ctl >> AC_SHIFT);
2095	<	return (a > (p >>>= 1) ? 0 :
2096	<	a > (p >>>= 1) ? 1 :
2097	<	a > (p >>>= 1) ? 2 :
2098	<	a > (p >>>= 1) ? 4 :
2099	<	8);
2087	>	* Returns a cheap heuristic guide for task partitioning when
2088	>	* programmers, frameworks, tools, or languages have little or no
2089	>	* idea about task granularity. In essence by offering this
2090	>	* method, we ask users only about tradeoffs in overhead vs
2091	>	* expected throughput and its variance, rather than how finely to
2092	>	* partition tasks.
2093	>	*
2094	>	* In a steady state strict (tree-structured) computation, each
2095	>	* thread makes available for stealing enough tasks for other
2096	>	* threads to remain active. Inductively, if all threads play by
2097	>	* the same rules, each thread should make available only a
2098	>	* constant number of tasks.
2099	>	*
2100	>	* The minimum useful constant is just 1. But using a value of 1
2101	>	* would require immediate replenishment upon each steal to
2102	>	* maintain enough tasks, which is infeasible. Further,
2103	>	* partitionings/granularities of offered tasks should minimize
2104	>	* steal rates, which in general means that threads nearer the top
2105	>	* of computation tree should generate more than those nearer the
2106	>	* bottom. In perfect steady state, each thread is at
2107	>	* approximately the same level of computation tree. However,
2108	>	* producing extra tasks amortizes the uncertainty of progress and
2109	>	* diffusion assumptions.
2110	>	*
2111	>	* So, users will want to use values larger, but not much larger
2112	>	* than 1 to both smooth over transient shortages and hedge
2113	>	* against uneven progress; as traded off against the cost of
2114	>	* extra task overhead. We leave the user to pick a threshold
2115	>	* value to compare with the results of this call to guide
2116	>	* decisions, but recommend values such as 3.
2117	>	*
2118	>	* When all threads are active, it is on average OK to estimate
2119	>	* surplus strictly locally. In steady-state, if one thread is
2120	>	* maintaining say 2 surplus tasks, then so are others. So we can
2121	>	* just use estimated queue length. However, this strategy alone
2122	>	* leads to serious mis-estimates in some non-steady-state
2123	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2124	>	* many of these by further considering the number of "idle"
2125	>	* threads, that are known to have zero queued tasks, so
2126	>	* compensate by a factor of (#idle/#active) threads.
2127	>	*
2128	>	* Note: The approximation of #busy workers as #active workers is
2129	>	* not very good under current signalling scheme, and should be
2130	>	* improved.
2131	>	*/
2132	>	static int getSurplusQueuedTaskCount() {
2133	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2134	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2135	>	int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
2136	>	int p = (pool = wt.pool).parallelism;
2137	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2138	>	return q.top - b - (a > (p >>>= 1) ? 0 :
2139	>	a > (p >>>= 1) ? 1 :
2140	>	a > (p >>>= 1) ? 2 :
2141	>	a > (p >>>= 1) ? 4 :
2142	>	8);
2143	>	}
2144	>	return 0;
2145		}
2146
2147		// Termination
#	Line 2004 \| Line 2161 \| public class ForkJoinPool extends Abstra
2161		* @return true if now terminating or terminated
2162		*/
2163		private boolean tryTerminate(boolean now, boolean enable) {
2164	<	Mutex lock = this.lock;
2164	>	if (this == commonPool) // cannot shut down
2165	>	return false;
2166		for (long c;;) {
2167		if (((c = ctl) & STOP_BIT) != 0) { // already terminating
2168		if ((short)(c >>> TC_SHIFT) == -parallelism) {
2169	<	lock.lock(); // don't need try/finally
2170	<	termination.signalAll(); // signal when 0 workers
2171	<	lock.unlock();
2169	>	synchronized (this) {
2170	>	notifyAll(); // signal when 0 workers
2171	>	}
2172		}
2173		return true;
2174		}
2175	<	if (runState >= 0) { // not yet enabled
2175	>	if (plock >= 0) { // not yet enabled
2176	>	int ps;
2177		if (!enable)
2178		return false;
2179	<	lock.lock();
2180	<	runState \|= SHUTDOWN;
2181	<	lock.unlock();
2179	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
2180	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2181	>	ps = acquirePlock();
2182	>	int nps = SHUTDOWN;
2183	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2184	>	releasePlock(nps);
2185		}
2186		if (!now) { // check if idle & no tasks
2187		if ((int)(c >> AC_SHIFT) != -parallelism \|\|
#	Line 2042 \| Line 2204 \| public class ForkJoinPool extends Abstra
2204		int n = ws.length;
2205		for (int i = 0; i < n; ++i) {
2206		if ((w = ws[i]) != null) {
2207	<	w.runState = -1;
2207	>	w.qlock = -1;
2208		if (pass > 0) {
2209		w.cancelAll();
2210		if (pass > 1)
#	Line 2061 \| Line 2223 \| public class ForkJoinPool extends Abstra
2223		if (w.eventCount == (e \| INT_SIGN) &&
2224		U.compareAndSwapLong(this, CTL, cc, nc)) {
2225		w.eventCount = (e + E_SEQ) & E_MASK;
2226	<	w.runState = -1;
2226	>	w.qlock = -1;
2227		if ((p = w.parker) != null)
2228		U.unpark(p);
2229		}
#	Line 2072 \| Line 2234 \| public class ForkJoinPool extends Abstra
2234		}
2235		}
2236
2237	+	// external operations on common pool
2238	+
2239	+	/**
2240	+	* Returns common pool queue for a thread that has submitted at
2241	+	* least one task.
2242	+	*/
2243	+	static WorkQueue commonSubmitterQueue() {
2244	+	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2245	+	return ((z = submitters.get()) != null &&
2246	+	(p = commonPool) != null &&
2247	+	(ws = p.workQueues) != null &&
2248	+	(m = ws.length - 1) >= 0) ?
2249	+	ws[m & z.seed & SQMASK] : null;
2250	+	}
2251	+
2252	+	/**
2253	+	* Tries to pop the given task from submitter's queue in common pool.
2254	+	*/
2255	+	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2256	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2257	+	ForkJoinTask<?>[] a; int m, s; long j;
2258	+	if ((z = submitters.get()) != null &&
2259	+	(p = commonPool) != null &&
2260	+	(ws = p.workQueues) != null &&
2261	+	(m = ws.length - 1) >= 0 &&
2262	+	(q = ws[m & z.seed & SQMASK]) != null &&
2263	+	(s = q.top) != q.base &&
2264	+	(a = q.array) != null &&
2265	+	U.getObjectVolatile
2266	+	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2267	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2268	+	if (q.array == a && q.top == s && // recheck
2269	+	U.compareAndSwapObject(a, j, t, null)) {
2270	+	q.top = s - 1;
2271	+	q.qlock = 0;
2272	+	return true;
2273	+	}
2274	+	q.qlock = 0;
2275	+	}
2276	+	return false;
2277	+	}
2278	+
2279	+	/**
2280	+	* Tries to pop and run local tasks within the same computation
2281	+	* as the given root. On failure, tries to help complete from
2282	+	* other queues via helpComplete.
2283	+	*/
2284	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2285	+	ForkJoinTask<?>[] a; int m;
2286	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2287	+	root != null && root.status >= 0) {
2288	+	for (;;) {
2289	+	int s; Object o; CountedCompleter<?> task = null;
2290	+	if ((s = q.top) - q.base > 0) {
2291	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2292	+	if ((o = U.getObject(a, j)) != null &&
2293	+	(o instanceof CountedCompleter)) {
2294	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2295	+	do {
2296	+	if (r == root) {
2297	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2298	+	if (q.array == a && q.top == s &&
2299	+	U.compareAndSwapObject(a, j, t, null)) {
2300	+	q.top = s - 1;
2301	+	task = t;
2302	+	}
2303	+	q.qlock = 0;
2304	+	}
2305	+	break;
2306	+	}
2307	+	} while ((r = r.completer) != null);
2308	+	}
2309	+	}
2310	+	if (task != null)
2311	+	task.doExec();
2312	+	if (root.status < 0 \|\| (int)(ctl >> AC_SHIFT) >= 0)
2313	+	break;
2314	+	if (task == null) {
2315	+	if (helpSignal(root, q.poolIndex) >= 0)
2316	+	helpComplete(root, SHARED_QUEUE);
2317	+	break;
2318	+	}
2319	+	}
2320	+	}
2321	+	}
2322	+
2323	+	/**
2324	+	* Tries to help execute or signal availability of the given task
2325	+	* from submitter's queue in common pool.
2326	+	*/
2327	+	static void externalHelpJoin(ForkJoinTask<?> t) {
2328	+	// Some hard-to-avoid overlap with tryExternalUnpush
2329	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2330	+	ForkJoinTask<?>[] a; int m, s, n; long j;
2331	+	if (t != null && t.status >= 0 &&
2332	+	(z = submitters.get()) != null &&
2333	+	(p = commonPool) != null &&
2334	+	(ws = p.workQueues) != null &&
2335	+	(m = ws.length - 1) >= 0 &&
2336	+	(q = ws[m & z.seed & SQMASK]) != null &&
2337	+	(a = q.array) != null) {
2338	+	if ((s = q.top) != q.base &&
2339	+	U.getObjectVolatile
2340	+	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2341	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2342	+	if (q.array == a && q.top == s &&
2343	+	U.compareAndSwapObject(a, j, t, null)) {
2344	+	q.top = s - 1;
2345	+	q.qlock = 0;
2346	+	t.doExec();
2347	+	}
2348	+	else
2349	+	q.qlock = 0;
2350	+	}
2351	+	if (t.status >= 0) {
2352	+	if (t instanceof CountedCompleter)
2353	+	p.externalHelpComplete(q, t);
2354	+	else
2355	+	p.helpSignal(t, q.poolIndex);
2356	+	}
2357	+	}
2358	+	}
2359	+
2360	+	/**
2361	+	* Restricted version of helpQuiescePool for external callers
2362	+	*/
2363	+	static void externalHelpQuiescePool() {
2364	+	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2365	+	int r = ThreadLocalRandom.current().nextInt();
2366	+	if ((p = commonPool) != null &&
2367	+	(q = p.findNonEmptyStealQueue(r)) != null &&
2368	+	(b = q.base) - q.top < 0 &&
2369	+	(t = q.pollAt(b)) != null)
2370	+	t.doExec();
2371	+	}
2372	+
2373		// Exported methods
2374
2375		// Constructors
#	Line 2149 \| Line 2447 \| public class ForkJoinPool extends Abstra
2447		this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2448		long np = (long)(-parallelism); // offset ctl counts
2449		this.ctl = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
2450	<	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2153	<	int n = parallelism - 1;
2154	<	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
2155	<	int size = (n + 1) << 1; // #slots = 2*#workers
2156	<	this.submitMask = size - 1; // room for max # of submit queues
2157	<	this.workQueues = new WorkQueue[size];
2158	<	this.termination = (this.lock = new Mutex()).newCondition();
2159	<	this.stealCount = new AtomicLong();
2160	<	this.nextWorkerNumber = new AtomicInteger();
2161	<	int pn = poolNumberGenerator.incrementAndGet();
2450	>	int pn = nextPoolId();
2451		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2452		sb.append(Integer.toString(pn));
2453		sb.append("-worker-");
2454		this.workerNamePrefix = sb.toString();
2455	<	lock.lock();
2456	<	this.runState = 1; // set init flag
2457	<	lock.unlock();
2455	>	}
2456	>
2457	>	/**
2458	>	* Constructor for common pool, suitable only for static initialization.
2459	>	* Basically the same as above, but uses smallest possible initial footprint.
2460	>	*/
2461	>	ForkJoinPool(int parallelism, long ctl,
2462	>	ForkJoinWorkerThreadFactory factory,
2463	>	Thread.UncaughtExceptionHandler handler) {
2464	>	this.parallelism = parallelism;
2465	>	this.ctl = ctl;
2466	>	this.factory = factory;
2467	>	this.ueh = handler;
2468	>	this.localMode = LIFO_QUEUE;
2469	>	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2470	>	}
2471	>
2472	>	/**
2473	>	* Returns the common pool instance.
2474	>	*
2475	>	* @return the common pool instance
2476	>	*/
2477	>	public static ForkJoinPool commonPool() {
2478	>	return commonPool; // cannot be null (if so, a static init error)
2479		}
2480
2481		// Execution methods
#	Line 2189 \| Line 2499 \| public class ForkJoinPool extends Abstra
2499		public <T> T invoke(ForkJoinTask<T> task) {
2500		if (task == null)
2501		throw new NullPointerException();
2502	<	doSubmit(task);
2502	>	externalPush(task);
2503		return task.join();
2504		}
2505
#	Line 2204 \| Line 2514 \| public class ForkJoinPool extends Abstra
2514		public void execute(ForkJoinTask<?> task) {
2515		if (task == null)
2516		throw new NullPointerException();
2517	<	doSubmit(task);
2517	>	externalPush(task);
2518		}
2519
2520		// AbstractExecutorService methods
#	Line 2222 \| Line 2532 \| public class ForkJoinPool extends Abstra
2532		job = (ForkJoinTask<?>) task;
2533		else
2534		job = new ForkJoinTask.AdaptedRunnableAction(task);
2535	<	doSubmit(job);
2535	>	externalPush(job);
2536		}
2537
2538		/**
#	Line 2237 \| Line 2547 \| public class ForkJoinPool extends Abstra
2547		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2548		if (task == null)
2549		throw new NullPointerException();
2550	<	doSubmit(task);
2550	>	externalPush(task);
2551		return task;
2552		}
2553
#	Line 2248 \| Line 2558 \| public class ForkJoinPool extends Abstra
2558		*/
2559		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2560		ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2561	<	doSubmit(job);
2561	>	externalPush(job);
2562		return job;
2563		}
2564
#	Line 2259 \| Line 2569 \| public class ForkJoinPool extends Abstra
2569		*/
2570		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2571		ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2572	<	doSubmit(job);
2572	>	externalPush(job);
2573		return job;
2574		}
2575
#	Line 2276 \| Line 2586 \| public class ForkJoinPool extends Abstra
2586		job = (ForkJoinTask<?>) task;
2587		else
2588		job = new ForkJoinTask.AdaptedRunnableAction(task);
2589	<	doSubmit(job);
2589	>	externalPush(job);
2590		return job;
2591		}
2592
#	Line 2298 \| Line 2608 \| public class ForkJoinPool extends Abstra
2608		try {
2609		for (Callable<T> t : tasks) {
2610		ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2611	<	doSubmit(f);
2611	>	externalPush(f);
2612		fs.add(f);
2613		}
2614		for (ForkJoinTask<T> f : fs)
#	Line 2341 \| Line 2651 \| public class ForkJoinPool extends Abstra
2651		}
2652
2653		/**
2654	+	* Returns the targeted parallelism level of the common pool.
2655	+	*
2656	+	* @return the targeted parallelism level of the common pool
2657	+	*/
2658	+	public static int getCommonPoolParallelism() {
2659	+	return commonPoolParallelism;
2660	+	}
2661	+
2662	+	/**
2663		* Returns the number of worker threads that have started but not
2664		* yet terminated. The result returned by this method may differ
2665		* from {@link #getParallelism} when threads are created to
#	Line 2421 \| Line 2740 \| public class ForkJoinPool extends Abstra
2740		* @return the number of steals
2741		*/
2742		public long getStealCount() {
2743	<	long count = stealCount.get();
2743	>	long count = stealCount;
2744		WorkQueue[] ws; WorkQueue w;
2745		if ((ws = workQueues) != null) {
2746		for (int i = 1; i < ws.length; i += 2) {
2747		if ((w = ws[i]) != null)
2748	<	count += w.totalSteals;
2748	>	count += w.nsteals;
2749		}
2750		}
2751		return count;
#	Line 2483 \| Line 2802 \| public class ForkJoinPool extends Abstra
2802		WorkQueue[] ws; WorkQueue w;
2803		if ((ws = workQueues) != null) {
2804		for (int i = 0; i < ws.length; i += 2) {
2805	<	if ((w = ws[i]) != null && !w.isEmpty())
2805	>	if ((w = ws[i]) != null && w.queueSize() != 0)
2806		return true;
2807		}
2808		}
#	Line 2551 \| Line 2870 \| public class ForkJoinPool extends Abstra
2870		public String toString() {
2871		// Use a single pass through workQueues to collect counts
2872		long qt = 0L, qs = 0L; int rc = 0;
2873	<	long st = stealCount.get();
2873	>	long st = stealCount;
2874		long c = ctl;
2875		WorkQueue[] ws; WorkQueue w;
2876		if ((ws = workQueues) != null) {
#	Line 2562 \| Line 2881 \| public class ForkJoinPool extends Abstra
2881		qs += size;
2882		else {
2883		qt += size;
2884	<	st += w.totalSteals;
2884	>	st += w.nsteals;
2885		if (w.isApparentlyUnblocked())
2886		++rc;
2887		}
#	Line 2578 \| Line 2897 \| public class ForkJoinPool extends Abstra
2897		if ((c & STOP_BIT) != 0)
2898		level = (tc == 0) ? "Terminated" : "Terminating";
2899		else
2900	<	level = runState < 0 ? "Shutting down" : "Running";
2900	>	level = plock < 0 ? "Shutting down" : "Running";
2901		return super.toString() +
2902		"[" + level +
2903		", parallelism = " + pc +
#	Line 2592 \| Line 2911 \| public class ForkJoinPool extends Abstra
2911		}
2912
2913		/**
2914	<	* Initiates an orderly shutdown in which previously submitted
2915	<	* tasks are executed, but no new tasks will be accepted.
2916	<	* Invocation has no additional effect if already shut down.
2917	<	* Tasks that are in the process of being submitted concurrently
2918	<	* during the course of this method may or may not be rejected.
2914	>	* Possibly initiates an orderly shutdown in which previously
2915	>	* submitted tasks are executed, but no new tasks will be
2916	>	* accepted. Invocation has no effect on execution state if this
2917	>	* is the {@link #commonPool}, and no additional effect if
2918	>	* already shut down. Tasks that are in the process of being
2919	>	* submitted concurrently during the course of this method may or
2920	>	* may not be rejected.
2921		*
2922		* @throws SecurityException if a security manager exists and
2923		* the caller is not permitted to modify threads
#	Line 2609 \| Line 2930 \| public class ForkJoinPool extends Abstra
2930		}
2931
2932		/**
2933	<	* Attempts to cancel and/or stop all tasks, and reject all
2934	<	* subsequently submitted tasks. Tasks that are in the process of
2935	<	* being submitted or executed concurrently during the course of
2936	<	* this method may or may not be rejected. This method cancels
2937	<	* both existing and unexecuted tasks, in order to permit
2938	<	* termination in the presence of task dependencies. So the method
2939	<	* always returns an empty list (unlike the case for some other
2940	<	* Executors).
2933	>	* Possibly attempts to cancel and/or stop all tasks, and reject
2934	>	* all subsequently submitted tasks. Invocation has no effect on
2935	>	* execution state if this is the {@link #commonPool}, and no
2936	>	* additional effect if already shut down. Otherwise, tasks that
2937	>	* are in the process of being submitted or executed concurrently
2938	>	* during the course of this method may or may not be
2939	>	* rejected. This method cancels both existing and unexecuted
2940	>	* tasks, in order to permit termination in the presence of task
2941	>	* dependencies. So the method always returns an empty list
2942	>	* (unlike the case for some other Executors).
2943		*
2944		* @return an empty list
2945		* @throws SecurityException if a security manager exists and
#	Line 2666 \| Line 2989 \| public class ForkJoinPool extends Abstra
2989		* @return {@code true} if this pool has been shut down
2990		*/
2991		public boolean isShutdown() {
2992	<	return runState < 0;
2992	>	return plock < 0;
2993		}
2994
2995		/**
2996	<	* Blocks until all tasks have completed execution after a shutdown
2997	<	* request, or the timeout occurs, or the current thread is
2998	<	* interrupted, whichever happens first.
2996	>	* Blocks until all tasks have completed execution after a
2997	>	* shutdown request, or the timeout occurs, or the current thread
2998	>	* is interrupted, whichever happens first. Note that the {@link
2999	>	* #commonPool()} never terminates until program shutdown so
3000	>	* this method will always time out.
3001		*
3002		* @param timeout the maximum time to wait
3003		* @param unit the time unit of the timeout argument
#	Line 2683 \| Line 3008 \| public class ForkJoinPool extends Abstra
3008		public boolean awaitTermination(long timeout, TimeUnit unit)
3009		throws InterruptedException {
3010		long nanos = unit.toNanos(timeout);
3011	<	final Mutex lock = this.lock;
3012	<	lock.lock();
3013	<	try {
3014	<	for (;;) {
3015	<	if (isTerminated())
3016	<	return true;
3017	<	if (nanos <= 0)
3018	<	return false;
3019	<	nanos = termination.awaitNanos(nanos);
3011	>	if (isTerminated())
3012	>	return true;
3013	>	long startTime = System.nanoTime();
3014	>	boolean terminated = false;
3015	>	synchronized (this) {
3016	>	for (long waitTime = nanos, millis = 0L;;) {
3017	>	if (terminated = isTerminated() \|\|
3018	>	waitTime <= 0L \|\|
3019	>	(millis = unit.toMillis(waitTime)) <= 0L)
3020	>	break;
3021	>	wait(millis);
3022	>	waitTime = nanos - (System.nanoTime() - startTime);
3023		}
2696	–	} finally {
2697	–	lock.unlock();
3024		}
3025	+	return terminated;
3026		}
3027
3028		/**
#	Line 2794 \| Line 3121 \| public class ForkJoinPool extends Abstra
3121		public static void managedBlock(ManagedBlocker blocker)
3122		throws InterruptedException {
3123		Thread t = Thread.currentThread();
3124	<	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3125	<	((ForkJoinWorkerThread)t).pool : null);
3126	<	while (!blocker.isReleasable()) {
3127	<	if (p == null \|\| p.tryCompensate(null, blocker)) {
3128	<	try {
3129	<	do {} while (!blocker.isReleasable() && !blocker.block());
3130	<	} finally {
3131	<	if (p != null)
3124	>	if (t instanceof ForkJoinWorkerThread) {
3125	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3126	>	while (!blocker.isReleasable()) { // variant of helpSignal
3127	>	WorkQueue[] ws; WorkQueue q; int m, n;
3128	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3129	>	for (int i = 0; i <= m; ++i) {
3130	>	if (blocker.isReleasable())
3131	>	return;
3132	>	if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
3133	>	p.signalWork(q, n);
3134	>	if ((int)(p.ctl >> AC_SHIFT) >= 0)
3135	>	break;
3136	>	}
3137	>	}
3138	>	}
3139	>	if (p.tryCompensate()) {
3140	>	try {
3141	>	do {} while (!blocker.isReleasable() &&
3142	>	!blocker.block());
3143	>	} finally {
3144		p.incrementActiveCount();
3145	+	}
3146	+	break;
3147		}
2807	–	break;
3148		}
3149		}
3150	+	else {
3151	+	do {} while (!blocker.isReleasable() &&
3152	+	!blocker.block());
3153	+	}
3154		}
3155
3156		// AbstractExecutorService overrides. These rely on undocumented
#	Line 2827 \| Line 3171 \| public class ForkJoinPool extends Abstra
3171		private static final long PARKBLOCKER;
3172		private static final int ABASE;
3173		private static final int ASHIFT;
3174	+	private static final long STEALCOUNT;
3175	+	private static final long PLOCK;
3176	+	private static final long INDEXSEED;
3177	+	private static final long QLOCK;
3178
3179		static {
3180	<	poolNumberGenerator = new AtomicInteger();
3181	<	nextSubmitterSeed = new AtomicInteger(0x55555555);
3182	<	modifyThreadPermission = new RuntimePermission("modifyThread");
3183	<	defaultForkJoinWorkerThreadFactory =
3184	<	new DefaultForkJoinWorkerThreadFactory();
3185	<	submitters = new ThreadSubmitter();
3186	<	int s;
3180	>	// Establish common pool parameters
3181	>	// TBD: limit or report ignored exceptions?
3182	>
3183	>	int par = 0;
3184	>	ForkJoinWorkerThreadFactory fac = null;
3185	>	Thread.UncaughtExceptionHandler handler = null;
3186	>	try {
3187	>	String pp = System.getProperty(propPrefix + "parallelism");
3188	>	String hp = System.getProperty(propPrefix + "exceptionHandler");
3189	>	String fp = System.getProperty(propPrefix + "threadFactory");
3190	>	if (fp != null)
3191	>	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3192	>	getSystemClassLoader().loadClass(fp).newInstance());
3193	>	if (hp != null)
3194	>	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3195	>	getSystemClassLoader().loadClass(hp).newInstance());
3196	>	if (pp != null)
3197	>	par = Integer.parseInt(pp);
3198	>	} catch (Exception ignore) {
3199	>	}
3200	>
3201	>	int s; // initialize field offsets for CAS etc
3202		try {
3203		U = getUnsafe();
3204		Class<?> k = ForkJoinPool.class;
2842	–	Class<?> ak = ForkJoinTask[].class;
3205		CTL = U.objectFieldOffset
3206		(k.getDeclaredField("ctl"));
3207	+	STEALCOUNT = U.objectFieldOffset
3208	+	(k.getDeclaredField("stealCount"));
3209	+	PLOCK = U.objectFieldOffset
3210	+	(k.getDeclaredField("plock"));
3211	+	INDEXSEED = U.objectFieldOffset
3212	+	(k.getDeclaredField("indexSeed"));
3213		Class<?> tk = Thread.class;
3214		PARKBLOCKER = U.objectFieldOffset
3215		(tk.getDeclaredField("parkBlocker"));
3216	+	Class<?> wk = WorkQueue.class;
3217	+	QLOCK = U.objectFieldOffset
3218	+	(wk.getDeclaredField("qlock"));
3219	+	Class<?> ak = ForkJoinTask[].class;
3220		ABASE = U.arrayBaseOffset(ak);
3221		s = U.arrayIndexScale(ak);
3222	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3223		} catch (Exception e) {
3224		throw new Error(e);
3225		}
3226		if ((s & (s-1)) != 0)
3227		throw new Error("data type scale not a power of two");
3228	<	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3228	>
3229	>	/*
3230	>	* For extra caution, computations to set up pool state are
3231	>	* here; the constructor just assigns these values to fields.
3232	>	*/
3233	>	ForkJoinWorkerThreadFactory defaultFac =
3234	>	defaultForkJoinWorkerThreadFactory =
3235	>	new DefaultForkJoinWorkerThreadFactory();
3236	>	if (fac == null)
3237	>	fac = defaultFac;
3238	>	if (par <= 0)
3239	>	par = Runtime.getRuntime().availableProcessors();
3240	>	if (par > MAX_CAP)
3241	>	par = MAX_CAP;
3242	>	long np = (long)(-par); // precompute initial ctl value
3243	>	long ct = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
3244	>
3245	>	commonPoolParallelism = par;
3246	>	commonPool = new ForkJoinPool(par, ct, fac, handler);
3247	>	modifyThreadPermission = new RuntimePermission("modifyThread");
3248	>	submitters = new ThreadLocal<Submitter>();
3249		}
3250
3251		/**

Diff Legend

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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents): Revision 1.124 by jsr166, Mon Feb 20 23:32:24 2012 UTC vs. Revision 1.142 by jsr166, Wed Nov 14 19:05:03 2012 UTC

Diff Legend

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.124 by jsr166, Mon Feb 20 23:32:24 2012 UTC vs.
Revision 1.142 by jsr166, Wed Nov 14 19:05:03 2012 UTC