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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC vs.
Revision 1.181 by jsr166, Tue Feb 5 17:25:09 2013 UTC

#	Line 11 \| Line 11 \| import java.util.Arrays;
11		import java.util.Collection;
12		import java.util.Collections;
13		import java.util.List;
14	–	import java.util.Random;
14		import java.util.concurrent.AbstractExecutorService;
15		import java.util.concurrent.Callable;
16		import java.util.concurrent.ExecutorService;
#	Line 19 \| 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;
22	–	import java.util.concurrent.atomic.AtomicInteger;
23	–	import java.util.concurrent.atomic.AtomicLong;
24	–	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
25	–	import java.util.concurrent.locks.Condition;
21
22		/**
23		* An {@link ExecutorService} for running {@link ForkJoinTask}s.
#	Line 42 \| 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 I/O 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 59 \| Line 62 \| import java.util.concurrent.locks.Condit
62		* {@link #toString} returns indications of pool state in a
63		* convenient form for informal monitoring.
64		*
65	<	* <p> As is the case with other ExecutorServices, there are three
65	>	* <p>As is the case with other ExecutorServices, there are three
66		* main task execution methods summarized in the following table.
67		* These are designed to be used primarily by clients not already
68		* engaged in fork/join computations in the current pool. The main
#	Line 94 \| Line 97 \| import java.util.concurrent.locks.Condit
97		* </tr>
98		* </table>
99		*
100	<	* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
101	<	* used for all parallel task execution in a program or subsystem.
102	<	* Otherwise, use would not usually outweigh the construction and
103	<	* bookkeeping overhead of creating a large set of threads. For
104	<	* example, a common pool could be used for the {@code SortTasks}
105	<	* illustrated in {@link RecursiveAction}. Because {@code
106	<	* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
107	<	* daemon} mode, there is typically no need to explicitly {@link
108	<	* #shutdown} such a pool upon program exit.
109	<	*
107	<	* <pre> {@code
108	<	* static final ForkJoinPool mainPool = new ForkJoinPool();
109	<	* ...
110	<	* public void sort(long[] array) {
111	<	* mainPool.invoke(new SortTask(array, 0, array.length));
112	<	* }}</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 system properties} with prefix {@code
103	>	* java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
104	>	* an integer greater than zero, {@code threadFactory} -- the class
105	>	* name of a {@link ForkJoinWorkerThreadFactory}, and {@code
106	>	* exceptionHandler} -- the class name of a {@link
107	>	* java.lang.Thread.UncaughtExceptionHandler
108	>	* Thread.UncaughtExceptionHandler}. Upon any error in establishing
109	>	* these settings, default parameters are used.
110		*
111		* <p><b>Implementation notes</b>: This implementation restricts the
112		* maximum number of running threads to 32767. Attempts to create
#	Line 177 \| Line 174 \| public class ForkJoinPool extends Abstra
174		* If an attempted steal fails, a thief always chooses a different
175		* random victim target to try next. So, in order for one thief to
176		* progress, it suffices for any in-progress poll or new push on
177	<	* any empty queue to complete.
177	>	* any empty queue to complete. (This is why we normally use
178	>	* method pollAt and its variants that try once at the apparent
179	>	* base index, else consider alternative actions, rather than
180	>	* method poll.)
181		*
182		* This approach also enables support of a user mode in which local
183		* task processing is in FIFO, not LIFO order, simply by using
#	Line 194 \| Line 194 \| public class ForkJoinPool extends Abstra
194		* WorkQueues are also used in a similar way for tasks submitted
195		* to the pool. We cannot mix these tasks in the same queues used
196		* for work-stealing (this would contaminate lifo/fifo
197	<	* processing). Instead, we loosely associate submission queues
197	>	* processing). Instead, we randomly associate submission queues
198		* with submitting threads, using a form of hashing. The
199		* ThreadLocal Submitter class contains a value initially used as
200		* a hash code for choosing existing queues, but may be randomly
201		* repositioned upon contention with other submitters. In
202	<	* essence, submitters act like workers except that they never
203	<	* take tasks, and they are multiplexed on to a finite number of
204	<	* shared work queues. However, classes are set up so that future
205	<	* extensions could allow submitters to optionally help perform
206	<	* tasks as well. Insertion of tasks in shared mode requires a
207	<	* lock (mainly to protect in the case of resizing) but we use
208	<	* only a simple spinlock (using bits in field runState), because
209	<	* submitters encountering a busy queue move on to try or create
210	<	* other queues, so never block.
202	>	* essence, submitters act like workers except that they are
203	>	* restricted to executing local tasks that they submitted (or in
204	>	* the case of CountedCompleters, others with the same root task).
205	>	* However, because most shared/external queue operations are more
206	>	* expensive than internal, and because, at steady state, external
207	>	* submitters will compete for CPU with workers, ForkJoinTask.join
208	>	* and related methods disable them from repeatedly helping to
209	>	* process tasks if all workers are active. Insertion of tasks in
210	>	* shared mode requires a lock (mainly to protect in the case of
211	>	* resizing) but we use only a simple spinlock (using bits in
212	>	* field qlock), because submitters encountering a busy queue move
213	>	* on to try or create other queues -- they block only when
214	>	* creating and registering new queues.
215		*
216		* Management
217		* ==========
#	Line 229 \| Line 233 \| public class ForkJoinPool extends Abstra
233		* and their negations (used for thresholding) to fit into 16bit
234		* fields.
235		*
236	<	* Field "runState" contains 32 bits needed to register and
237	<	* deregister WorkQueues, as well as to enable shutdown. It is
238	<	* only modified under a lock (normally briefly held, but
239	<	* occasionally protecting allocations and resizings) but even
240	<	* when locked remains available to check consistency. An
241	<	* auxiliary field "growHints", also only modified under lock,
242	<	* contains a candidate index for the next WorkQueue and
239	<	* a mask for submission queue indices.
236	>	* Field "plock" is a form of sequence lock with a saturating
237	>	* shutdown bit (similarly for per-queue "qlocks"), mainly
238	>	* protecting updates to the workQueues array, as well as to
239	>	* enable shutdown. When used as a lock, it is normally only very
240	>	* briefly held, so is nearly always available after at most a
241	>	* brief spin, but we use a monitor-based backup strategy to
242	>	* block when needed.
243		*
244		* Recording WorkQueues. WorkQueues are recorded in the
245	<	* "workQueues" array that is created upon pool construction and
246	<	* expanded if necessary. Updates to the array while recording
247	<	* new workers and unrecording terminated ones are protected from
248	<	* each other by a lock but the array is otherwise concurrently
249	<	* readable, and accessed directly. To simplify index-based
250	<	* operations, the array size is always a power of two, and all
251	<	* readers must tolerate null slots. Shared (submission) queues
252	<	* are at even indices, worker queues at odd indices. Grouping
253	<	* them together in this way simplifies and speeds up task
254	<	* scanning. To avoid flailing during start-up, the array is
255	<	* presized to hold twice #parallelism workers (which is unlikely
253	<	* to need further resizing during execution). But to avoid
254	<	* dealing with so many null slots, variable runState includes a
255	<	* mask for the nearest power of two that contains all currently
256	<	* used indices.
245	>	* "workQueues" array that is created upon first use and expanded
246	>	* if necessary. Updates to the array while recording new workers
247	>	* and unrecording terminated ones are protected from each other
248	>	* by a lock but the array is otherwise concurrently readable, and
249	>	* accessed directly. To simplify index-based operations, the
250	>	* array size is always a power of two, and all readers must
251	>	* tolerate null slots. Worker queues are at odd indices. Shared
252	>	* (submission) queues are at even indices, up to a maximum of 64
253	>	* slots, to limit growth even if array needs to expand to add
254	>	* more workers. Grouping them together in this way simplifies and
255	>	* speeds up task scanning.
256		*
257		* All worker thread creation is on-demand, triggered by task
258		* submissions, replacement of terminated workers, and/or
#	Line 314 \| Line 313 \| public class ForkJoinPool extends Abstra
313		*
314		* Signalling. We create or wake up workers only when there
315		* appears to be at least one task they might be able to find and
316	<	* execute. When a submission is added or another worker adds a
317	<	* task to a queue that previously had fewer than two tasks, they
318	<	* signal waiting workers (or trigger creation of new ones if
319	<	* fewer than the given parallelism level -- see signalWork).
320	<	* These primary signals are buttressed by signals during rescans;
321	<	* together these cover the signals needed in cases when more
322	<	* tasks are pushed but untaken, and improve performance compared
323	<	* to having one thread wake up all workers.
316	>	* execute. However, many other threads may notice the same task
317	>	* and each signal to wake up a thread that might take it. So in
318	>	* general, pools will be over-signalled. When a submission is
319	>	* added or another worker adds a task to a queue that has fewer
320	>	* than two tasks, they signal waiting workers (or trigger
321	>	* creation of new ones if fewer than the given parallelism level
322	>	* -- signalWork), and may leave a hint to the unparked worker to
323	>	* help signal others upon wakeup). These primary signals are
324	>	* buttressed by others (see method helpSignal) whenever other
325	>	* threads scan for work or do not have a task to process. On
326	>	* most platforms, signalling (unpark) overhead time is noticeably
327	>	* long, and the time between signalling a thread and it actually
328	>	* making progress can be very noticeably long, so it is worth
329	>	* offloading these delays from critical paths as much as
330	>	* possible.
331		*
332		* Trimming workers. To release resources after periods of lack of
333		* use, a worker starting to wait when the pool is quiescent will
334	<	* time out and terminate if the pool has remained quiescent for
335	<	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
336	<	* terminating all workers after long periods of non-use.
334	>	* time out and terminate if the pool has remained quiescent for a
335	>	* given period -- a short period if there are more threads than
336	>	* parallelism, longer as the number of threads decreases. This
337	>	* will slowly propagate, eventually terminating all workers after
338	>	* periods of non-use.
339		*
340		* Shutdown and Termination. A call to shutdownNow atomically sets
341	<	* a runState bit and then (non-atomically) sets each worker's
342	<	* runState status, cancels all unprocessed tasks, and wakes up
341	>	* a plock bit and then (non-atomically) sets each worker's
342	>	* qlock status, cancels all unprocessed tasks, and wakes up
343		* all waiting workers. Detecting whether termination should
344		* commence after a non-abrupt shutdown() call requires more work
345		* and bookkeeping. We need consensus about quiescence (i.e., that
#	Line 359 \| Line 367 \| public class ForkJoinPool extends Abstra
367		* method tryCompensate() may create or re-activate a spare
368		* thread to compensate for blocked joiners until they unblock.
369		*
370	<	* A third form (implemented in tryRemoveAndExec and
371	<	* tryPollForAndExec) amounts to helping a hypothetical
372	<	* compensator: If we can readily tell that a possible action of a
373	<	* compensator is to steal and execute the task being joined, the
374	<	* joining thread can do so directly, without the need for a
375	<	* compensation thread (although at the expense of larger run-time
376	<	* stacks, but the tradeoff is typically worthwhile).
370	>	* A third form (implemented in tryRemoveAndExec) amounts to
371	>	* helping a hypothetical compensator: If we can readily tell that
372	>	* a possible action of a compensator is to steal and execute the
373	>	* task being joined, the joining thread can do so directly,
374	>	* without the need for a compensation thread (although at the
375	>	* expense of larger run-time stacks, but the tradeoff is
376	>	* typically worthwhile).
377		*
378		* The ManagedBlocker extension API can't use helping so relies
379		* only on compensation in method awaitBlocker.
#	Line 385 \| Line 393 \| public class ForkJoinPool extends Abstra
393		* (http://portal.acm.org/citation.cfm?id=155354). It differs in
394		* that: (1) We only maintain dependency links across workers upon
395		* steals, rather than use per-task bookkeeping. This sometimes
396	<	* requires a linear scan of workQueues array to locate stealers, but
397	<	* often doesn't because stealers leave hints (that may become
398	<	* stale/wrong) of where to locate them. A stealHint is only a
399	<	* hint because a worker might have had multiple steals and the
400	<	* hint records only one of them (usually the most current).
401	<	* Hinting isolates cost to when it is needed, rather than adding
402	<	* to per-task overhead. (2) It is "shallow", ignoring nesting
403	<	* and potentially cyclic mutual steals. (3) It is intentionally
396	>	* requires a linear scan of workQueues array to locate stealers,
397	>	* but often doesn't because stealers leave hints (that may become
398	>	* stale/wrong) of where to locate them. It is only a hint
399	>	* because a worker might have had multiple steals and the hint
400	>	* records only one of them (usually the most current). Hinting
401	>	* isolates cost to when it is needed, rather than adding to
402	>	* per-task overhead. (2) It is "shallow", ignoring nesting and
403	>	* potentially cyclic mutual steals. (3) It is intentionally
404		* racy: field currentJoin is updated only while actively joining,
405		* which means that we miss links in the chain during long-lived
406		* tasks, GC stalls etc (which is OK since blocking in such cases
407		* is usually a good idea). (4) We bound the number of attempts
408	<	* to find work (see MAX_HELP_DEPTH) and fall back to suspending
409	<	* the worker and if necessary replacing it with another.
408	>	* to find work (see MAX_HELP) and fall back to suspending the
409	>	* worker and if necessary replacing it with another.
410	>	*
411	>	* Helping actions for CountedCompleters are much simpler: Method
412	>	* helpComplete can take and execute any task with the same root
413	>	* as the task being waited on. However, this still entails some
414	>	* traversal of completer chains, so is less efficient than using
415	>	* CountedCompleters without explicit joins.
416		*
417		* It is impossible to keep exactly the target parallelism number
418		* of threads running at any given time. Determining the
419		* existence of conservatively safe helping targets, the
420		* availability of already-created spares, and the apparent need
421		* to create new spares are all racy, so we rely on multiple
422	<	* retries of each. Currently, in keeping with on-demand
423	<	* signalling policy, we compensate only if blocking would leave
424	<	* less than one active (non-waiting, non-blocked) worker.
425	<	* Additionally, to avoid some false alarms due to GC, lagging
426	<	* counters, system activity, etc, compensated blocking for joins
427	<	* is only attempted after rechecks stabilize in
428	<	* ForkJoinTask.awaitJoin. (Retries are interspersed with
429	<	* Thread.yield, for good citizenship.)
430	<	*
431	<	* Style notes: There is a lot of representation-level coupling
432	<	* among classes ForkJoinPool, ForkJoinWorkerThread, and
433	<	* ForkJoinTask. The fields of WorkQueue maintain data structures
434	<	* managed by ForkJoinPool, so are directly accessed. There is
435	<	* little point trying to reduce this, since any associated future
436	<	* changes in representations will need to be accompanied by
437	<	* algorithmic changes anyway. Several methods intrinsically
438	<	* sprawl because they must accumulate sets of consistent reads of
439	<	* volatiles held in local variables. Methods signalWork() and
440	<	* scan() are the main bottlenecks, so are especially heavily
422	>	* retries of each. Compensation in the apparent absence of
423	>	* helping opportunities is challenging to control on JVMs, where
424	>	* GC and other activities can stall progress of tasks that in
425	>	* turn stall out many other dependent tasks, without us being
426	>	* able to determine whether they will ever require compensation.
427	>	* Even though work-stealing otherwise encounters little
428	>	* degradation in the presence of more threads than cores,
429	>	* aggressively adding new threads in such cases entails risk of
430	>	* unwanted positive feedback control loops in which more threads
431	>	* cause more dependent stalls (as well as delayed progress of
432	>	* unblocked threads to the point that we know they are available)
433	>	* leading to more situations requiring more threads, and so
434	>	* on. This aspect of control can be seen as an (analytically
435	>	* intractable) game with an opponent that may choose the worst
436	>	* (for us) active thread to stall at any time. We take several
437	>	* precautions to bound losses (and thus bound gains), mainly in
438	>	* methods tryCompensate and awaitJoin.
439	>	*
440	>	* Common Pool
441	>	* ===========
442	>	*
443	>	* The static common Pool always exists after static
444	>	* initialization. Since it (or any other created pool) need
445	>	* never be used, we minimize initial construction overhead and
446	>	* footprint to the setup of about a dozen fields, with no nested
447	>	* allocation. Most bootstrapping occurs within method
448	>	* fullExternalPush during the first submission to the pool.
449	>	*
450	>	* When external threads submit to the common pool, they can
451	>	* perform some subtask processing (see externalHelpJoin and
452	>	* related methods). We do not need to record whether these
453	>	* submissions are to the common pool -- if not, externalHelpJoin
454	>	* returns quickly (at the most helping to signal some common pool
455	>	* workers). These submitters would otherwise be blocked waiting
456	>	* for completion, so the extra effort (with liberally sprinkled
457	>	* task status checks) in inapplicable cases amounts to an odd
458	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
459	>	*
460	>	* Style notes
461	>	* ===========
462	>	*
463	>	* There is a lot of representation-level coupling among classes
464	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
465	>	* fields of WorkQueue maintain data structures managed by
466	>	* ForkJoinPool, so are directly accessed. There is little point
467	>	* trying to reduce this, since any associated future changes in
468	>	* representations will need to be accompanied by algorithmic
469	>	* changes anyway. Several methods intrinsically sprawl because
470	>	* they must accumulate sets of consistent reads of volatiles held
471	>	* in local variables. Methods signalWork() and scan() are the
472	>	* main bottlenecks, so are especially heavily
473		* micro-optimized/mangled. There are lots of inline assignments
474		* (of form "while ((local = field) != 0)") which are usually the
475		* simplest way to ensure the required read orderings (which are
#	Line 431 \| Line 477 \| public class ForkJoinPool extends Abstra
477		* declarations of these locals at the heads of methods or blocks.
478		* There are several occurrences of the unusual "do {} while
479		* (!cas...)" which is the simplest way to force an update of a
480	<	* CAS'ed variable. There are also other coding oddities that help
480	>	* CAS'ed variable. There are also other coding oddities (including
481	>	* several unnecessary-looking hoisted null checks) that help
482		* some methods perform reasonably even when interpreted (not
483		* compiled).
484		*
#	Line 449 \| Line 496 \| public class ForkJoinPool extends Abstra
496		// Static utilities
497
498		/**
452	–	* Computes an initial hash code (also serving as a non-zero
453	–	* random seed) for a thread id. This method is expected to
454	–	* provide higher-quality hash codes than using method hashCode().
455	–	*/
456	–	static final int hashId(long id) {
457	–	int h = (int)id ^ (int)(id >>> 32); // Use MurmurHash of thread id
458	–	h ^= h >>> 16; h *= 0x85ebca6b;
459	–	h ^= h >>> 13; h *= 0xc2b2ae35;
460	–	h ^= h >>> 16;
461	–	return (h == 0)? 1 : h; // ensure nonzero
462	–	}
463	–
464	–	/**
499		* If there is a security manager, makes sure caller has
500		* permission to modify threads.
501		*/
#	Line 493 \| Line 527 \| public class ForkJoinPool extends Abstra
527		* Default ForkJoinWorkerThreadFactory implementation; creates a
528		* new ForkJoinWorkerThread.
529		*/
530	<	static class DefaultForkJoinWorkerThreadFactory
530	>	static final class DefaultForkJoinWorkerThreadFactory
531		implements ForkJoinWorkerThreadFactory {
532	<	public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
532	>	public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
533		return new ForkJoinWorkerThread(pool);
534		}
535		}
536
537		/**
538	<	* A simple non-reentrant lock used for exclusion when managing
539	<	* queues and workers. We use a custom lock so that we can readily
540	<	* probe lock state in constructions that check among alternative
541	<	* actions. The lock is normally only very briefly held, and
542	<	* sometimes treated as a spinlock, but other usages block to
543	<	* reduce overall contention in those cases where locked code
544	<	* bodies perform allocation/resizing.
545	<	*/
546	<	static final class Mutex extends AbstractQueuedSynchronizer {
547	<	public final boolean tryAcquire(int ignore) {
548	<	return compareAndSetState(0, 1);
549	<	}
550	<	public final boolean tryRelease(int ignore) {
551	<	setState(0);
552	<	return true;
553	<	}
554	<	public final void lock() { acquire(0); }
521	<	public final void unlock() { release(0); }
522	<	public final boolean isHeldExclusively() { return getState() == 1; }
523	<	public final Condition newCondition() { return new ConditionObject(); }
538	>	* Per-thread records for threads that submit to pools. Currently
539	>	* holds only pseudo-random seed / index that is used to choose
540	>	* submission queues in method externalPush. In the future, this may
541	>	* also incorporate a means to implement different task rejection
542	>	* and resubmission policies.
543	>	*
544	>	* Seeds for submitters and workers/workQueues work in basically
545	>	* the same way but are initialized and updated using slightly
546	>	* different mechanics. Both are initialized using the same
547	>	* approach as in class ThreadLocal, where successive values are
548	>	* unlikely to collide with previous values. Seeds are then
549	>	* randomly modified upon collisions using xorshifts, which
550	>	* requires a non-zero seed.
551	>	*/
552	>	static final class Submitter {
553	>	int seed;
554	>	Submitter(int s) { seed = s; }
555		}
556
557		/**
#	Line 530 \| Line 561 \| public class ForkJoinPool extends Abstra
561		* actually do anything beyond having a unique identity.
562		*/
563		static final class EmptyTask extends ForkJoinTask<Void> {
564	+	private static final long serialVersionUID = -7721805057305804111L;
565		EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
566		public final Void getRawResult() { return null; }
567		public final void setRawResult(Void x) {}
#	Line 550 \| Line 582 \| public class ForkJoinPool extends Abstra
582		*
583		* Field "top" is the index (mod array.length) of the next queue
584		* slot to push to or pop from. It is written only by owner thread
585	<	* for push, or under lock for trySharedPush, and accessed by
586	<	* other threads only after reading (volatile) base. Both top and
587	<	* base are allowed to wrap around on overflow, but (top - base)
588	<	* (or more commonly -(base - top) to force volatile read of base
589	<	* before top) still estimates size.
585	>	* for push, or under lock for external/shared push, and accessed
586	>	* by other threads only after reading (volatile) base. Both top
587	>	* and base are allowed to wrap around on overflow, but (top -
588	>	* base) (or more commonly -(base - top) to force volatile read of
589	>	* base before top) still estimates size. The lock ("qlock") is
590	>	* forced to -1 on termination, causing all further lock attempts
591	>	* to fail. (Note: we don't need CAS for termination state because
592	>	* upon pool shutdown, all shared-queues will stop being used
593	>	* anyway.) Nearly all lock bodies are set up so that exceptions
594	>	* within lock bodies are "impossible" (modulo JVM errors that
595	>	* would cause failure anyway.)
596		*
597		* The array slots are read and written using the emulation of
598		* volatiles/atomics provided by Unsafe. Insertions must in
599		* general use putOrderedObject as a form of releasing store to
600		* ensure that all writes to the task object are ordered before
601	<	* its publication in the queue. (Although we can avoid one case
602	<	* of this when locked in trySharedPush.) All removals entail a
603	<	* CAS to null. The array is always a power of two. To ensure
604	<	* safety of Unsafe array operations, all accesses perform
567	<	* explicit null checks and implicit bounds checks via
568	<	* power-of-two masking.
601	>	* its publication in the queue. All removals entail a CAS to
602	>	* null. The array is always a power of two. To ensure safety of
603	>	* Unsafe array operations, all accesses perform explicit null
604	>	* checks and implicit bounds checks via power-of-two masking.
605		*
606		* In addition to basic queuing support, this class contains
607		* fields described elsewhere to control execution. It turns out
608	<	* to work better memory-layout-wise to include them in this
609	<	* class rather than a separate class.
608	>	* to work better memory-layout-wise to include them in this class
609	>	* rather than a separate class.
610		*
611		* Performance on most platforms is very sensitive to placement of
612		* instances of both WorkQueues and their arrays -- we absolutely
#	Line 584 \| Line 620 \| public class ForkJoinPool extends Abstra
620		* trades off slightly slower average field access for the sake of
621		* avoiding really bad worst-case access. (Until better JVM
622		* support is in place, this padding is dependent on transient
623	<	* properties of JVM field layout rules.) We also take care in
623	>	* properties of JVM field layout rules.) We also take care in
624		* allocating, sizing and resizing the array. Non-shared queue
625	<	* arrays are initialized (via method growArray) by workers before
626	<	* use. Others are allocated on first use.
625	>	* arrays are initialized by workers before use. Others are
626	>	* allocated on first use.
627		*/
628		static final class WorkQueue {
629		/**
630		* Capacity of work-stealing queue array upon initialization.
631	<	* Must be a power of two; at least 4, but set larger to
632	<	* reduce cacheline sharing among queues.
631	>	* Must be a power of two; at least 4, but should be larger to
632	>	* reduce or eliminate cacheline sharing among queues.
633	>	* Currently, it is much larger, as a partial workaround for
634	>	* the fact that JVMs often place arrays in locations that
635	>	* share GC bookkeeping (especially cardmarks) such that
636	>	* per-write accesses encounter serious memory contention.
637		*/
638	<	static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
638	>	static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
639
640		/**
641		* Maximum size for queue arrays. Must be a power of two less
#	Line 606 \| Line 646 \| public class ForkJoinPool extends Abstra
646		*/
647		static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
648
649	<	volatile long totalSteals; // cumulative number of steals
649	>	// Heuristic padding to ameliorate unfortunate memory placements
650	>	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
651	>
652		int seed; // for random scanning; initialize nonzero
653		volatile int eventCount; // encoded inactivation count; < 0 if inactive
654		int nextWait; // encoded record of next event waiter
655	<	int rescans; // remaining scans until block
614	<	int nsteals; // top-level task executions since last idle
615	<	final int mode; // lifo, fifo, or shared
655	>	int hint; // steal or signal hint (index)
656		int poolIndex; // index of this queue in pool (or 0)
657	<	int stealHint; // index of most recent known stealer
658	<	volatile int runState; // 1: locked, -1: terminate; else 0
657	>	final int mode; // 0: lifo, > 0: fifo, < 0: shared
658	>	int nsteals; // number of steals
659	>	volatile int qlock; // 1: locked, -1: terminate; else 0
660		volatile int base; // index of next slot for poll
661		int top; // index of next slot for push
662		ForkJoinTask<?>[] array; // the elements (initially unallocated)
663	+	final ForkJoinPool pool; // the containing pool (may be null)
664		final ForkJoinWorkerThread owner; // owning thread or null if shared
665		volatile Thread parker; // == owner during call to park; else null
666	<	ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
666	>	volatile ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
667		ForkJoinTask<?> currentSteal; // current non-local task being executed
626	–	// Heuristic padding to ameliorate unfortunate memory placements
627	–	Object p00, p01, p02, p03, p04, p05, p06, p07, p08, p09, p0a;
668
669	<	WorkQueue(ForkJoinWorkerThread owner, int mode) {
669	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
670	>	volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d;
671	>
672	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
673	>	int seed) {
674	>	this.pool = pool;
675		this.owner = owner;
676		this.mode = mode;
677	+	this.seed = seed;
678		// Place indices in the center of array (that is not yet allocated)
679		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
680		}
681
682		/**
683	<	* Returns number of tasks in the queue.
683	>	* Returns the approximate number of tasks in the queue.
684		*/
685		final int queueSize() {
686	<	int n = base - top; // non-owner callers must read base first
687	<	return (n >= 0) ? 0 : -n;
686	>	int n = base - top; // non-owner callers must read base first
687	>	return (n >= 0) ? 0 : -n; // ignore transient negative
688	>	}
689	>
690	>	/**
691	>	* Provides a more accurate estimate of whether this queue has
692	>	* any tasks than does queueSize, by checking whether a
693	>	* near-empty queue has at least one unclaimed task.
694	>	*/
695	>	final boolean isEmpty() {
696	>	ForkJoinTask<?>[] a; int m, s;
697	>	int n = base - (s = top);
698	>	return (n >= 0 \|\|
699	>	(n == -1 &&
700	>	((a = array) == null \|\|
701	>	(m = a.length - 1) < 0 \|\|
702	>	U.getObject
703	>	(a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null)));
704		}
705
706		/**
707	<	* Pushes a task. Call only by owner in unshared queues.
707	>	* Pushes a task. Call only by owner in unshared queues. (The
708	>	* shared-queue version is embedded in method externalPush.)
709		*
710		* @param task the task. Caller must ensure non-null.
711	<	* @param p if non-null, pool to signal if necessary
649	<	* @throw RejectedExecutionException if array cannot be resized
711	>	* @throws RejectedExecutionException if array cannot be resized
712		*/
713	<	final void push(ForkJoinTask<?> task, ForkJoinPool p) {
714	<	ForkJoinTask<?>[] a;
713	>	final void push(ForkJoinTask<?> task) {
714	>	ForkJoinTask<?>[] a; ForkJoinPool p;
715		int s = top, m, n;
716		if ((a = array) != null) { // ignore if queue removed
717	<	U.putOrderedObject
718	<	(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
717	>	int j = (((m = a.length - 1) & s) << ASHIFT) + ABASE;
718	>	U.putOrderedObject(a, j, task);
719		if ((n = (top = s + 1) - base) <= 2) {
720	<	if (p != null)
721	<	p.signalWork();
720	>	if ((p = pool) != null)
721	>	p.signalWork(this);
722		}
723		else if (n >= m)
724	<	growArray(true);
724	>	growArray();
725	>	}
726	>	}
727	>
728	>	/**
729	>	* Initializes or doubles the capacity of array. Call either
730	>	* by owner or with lock held -- it is OK for base, but not
731	>	* top, to move while resizings are in progress.
732	>	*/
733	>	final ForkJoinTask<?>[] growArray() {
734	>	ForkJoinTask<?>[] oldA = array;
735	>	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
736	>	if (size > MAXIMUM_QUEUE_CAPACITY)
737	>	throw new RejectedExecutionException("Queue capacity exceeded");
738	>	int oldMask, t, b;
739	>	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
740	>	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
741	>	(t = top) - (b = base) > 0) {
742	>	int mask = size - 1;
743	>	do {
744	>	ForkJoinTask<?> x;
745	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
746	>	int j = ((b & mask) << ASHIFT) + ABASE;
747	>	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
748	>	if (x != null &&
749	>	U.compareAndSwapObject(oldA, oldj, x, null))
750	>	U.putObjectVolatile(a, j, x);
751	>	} while (++b != t);
752		}
753	+	return a;
754		}
755
756		/**
757	<	* Pushes a task if lock is free and array is either big
758	<	* enough or can be resized to be big enough.
669	<	*
670	<	* @param task the task. Caller must ensure non-null.
671	<	* @return true if submitted
757	>	* Takes next task, if one exists, in LIFO order. Call only
758	>	* by owner in unshared queues.
759		*/
760	<	final boolean trySharedPush(ForkJoinTask<?> task) {
761	<	boolean submitted = false;
762	<	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
763	<	ForkJoinTask<?>[] a = array;
764	<	int s = top;
765	<	try {
766	<	if ((a != null && a.length > s + 1 - base) \|\|
767	<	(a = growArray(false)) != null) { // must presize
768	<	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
769	<	U.putObject(a, (long)j, task); // don't need "ordered"
683	<	top = s + 1;
684	<	submitted = true;
760	>	final ForkJoinTask<?> pop() {
761	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
762	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
763	>	for (int s; (s = top - 1) - base >= 0;) {
764	>	long j = ((m & s) << ASHIFT) + ABASE;
765	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
766	>	break;
767	>	if (U.compareAndSwapObject(a, j, t, null)) {
768	>	top = s;
769	>	return t;
770		}
686	–	} finally {
687	–	runState = 0; // unlock
771		}
772		}
773	<	return submitted;
773	>	return null;
774		}
775
776		/**
777	<	* Takes next task, if one exists, in FIFO order.
777	>	* Takes a task in FIFO order if b is base of queue and a task
778	>	* can be claimed without contention. Specialized versions
779	>	* appear in ForkJoinPool methods scan and tryHelpStealer.
780		*/
781	<	final ForkJoinTask<?> poll() {
782	<	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
783	<	while ((b = base) - top < 0 && (a = array) != null) {
781	>	final ForkJoinTask<?> pollAt(int b) {
782	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
783	>	if ((a = array) != null) {
784		int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
785		if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
786		base == b &&
#	Line 708 \| Line 793 \| public class ForkJoinPool extends Abstra
793		}
794
795		/**
796	<	* Takes next task, if one exists, in LIFO order. Call only
712	<	* by owner in unshared queues. (We do not have a shared
713	<	* version of this method because it is never needed.)
796	>	* Takes next task, if one exists, in FIFO order.
797		*/
798	<	final ForkJoinTask<?> pop() {
799	<	ForkJoinTask<?> t; int m;
800	<	ForkJoinTask<?>[] a = array;
801	<	if (a != null && (m = a.length - 1) >= 0) {
802	<	for (int s; (s = top - 1) - base >= 0;) {
803	<	int j = ((m & s) << ASHIFT) + ABASE;
804	<	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) == null)
805	<	break;
806	<	if (U.compareAndSwapObject(a, j, t, null)) {
724	<	top = s;
798	>	final ForkJoinTask<?> poll() {
799	>	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
800	>	while ((b = base) - top < 0 && (a = array) != null) {
801	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
802	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
803	>	if (t != null) {
804	>	if (base == b &&
805	>	U.compareAndSwapObject(a, j, t, null)) {
806	>	base = b + 1;
807		return t;
808		}
809		}
810	+	else if (base == b) {
811	+	if (b + 1 == top)
812	+	break;
813	+	Thread.yield(); // wait for lagging update (very rare)
814	+	}
815		}
816		return null;
817		}
#	Line 749 \| Line 836 \| public class ForkJoinPool extends Abstra
836		}
837
838		/**
752	–	* Returns task at index b if b is current base of queue.
753	–	*/
754	–	final ForkJoinTask<?> pollAt(int b) {
755	–	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
756	–	if ((a = array) != null) {
757	–	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
758	–	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
759	–	base == b &&
760	–	U.compareAndSwapObject(a, j, t, null)) {
761	–	base = b + 1;
762	–	return t;
763	–	}
764	–	}
765	–	return null;
766	–	}
767	–
768	–	/**
839		* Pops the given task only if it is at the current top.
840	+	* (A shared version is available only via FJP.tryExternalUnpush)
841		*/
842		final boolean tryUnpush(ForkJoinTask<?> t) {
843		ForkJoinTask<?>[] a; int s;
#	Line 780 \| Line 851 \| public class ForkJoinPool extends Abstra
851		}
852
853		/**
854	<	* Polls the given task only if it is at the current base.
854	>	* Removes and cancels all known tasks, ignoring any exceptions.
855		*/
856	<	final boolean pollFor(ForkJoinTask<?> task) {
857	<	ForkJoinTask<?>[] a; int b;
858	<	if ((b = base) - top < 0 && (a = array) != null) {
859	<	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
860	<	if (U.getObjectVolatile(a, j) == task && base == b &&
861	<	U.compareAndSwapObject(a, j, task, null)) {
862	<	base = b + 1;
863	<	return true;
856	>	final void cancelAll() {
857	>	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
858	>	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
859	>	for (ForkJoinTask<?> t; (t = poll()) != null; )
860	>	ForkJoinTask.cancelIgnoringExceptions(t);
861	>	}
862	>
863	>	/**
864	>	* Computes next value for random probes. Scans don't require
865	>	* a very high quality generator, but also not a crummy one.
866	>	* Marsaglia xor-shift is cheap and works well enough. Note:
867	>	* This is manually inlined in its usages in ForkJoinPool to
868	>	* avoid writes inside busy scan loops.
869	>	*/
870	>	final int nextSeed() {
871	>	int r = seed;
872	>	r ^= r << 13;
873	>	r ^= r >>> 17;
874	>	return seed = r ^= r << 5;
875	>	}
876	>
877	>	// Specialized execution methods
878	>
879	>	/**
880	>	* Pops and runs tasks until empty.
881	>	*/
882	>	private void popAndExecAll() {
883	>	// A bit faster than repeated pop calls
884	>	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
885	>	while ((a = array) != null && (m = a.length - 1) >= 0 &&
886	>	(s = top - 1) - base >= 0 &&
887	>	(t = ((ForkJoinTask<?>)
888	>	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
889	>	!= null) {
890	>	if (U.compareAndSwapObject(a, j, t, null)) {
891	>	top = s;
892	>	t.doExec();
893		}
894		}
795	–	return false;
895		}
896
897		/**
898	<	* If present, removes from queue and executes the given task, or
899	<	* any other cancelled task. Returns (true) immediately on any CAS
898	>	* Polls and runs tasks until empty.
899	>	*/
900	>	private void pollAndExecAll() {
901	>	for (ForkJoinTask<?> t; (t = poll()) != null;)
902	>	t.doExec();
903	>	}
904	>
905	>	/**
906	>	* If present, removes from queue and executes the given task,
907	>	* or any other cancelled task. Returns (true) on any CAS
908		* or consistency check failure so caller can retry.
909		*
910	<	* @return false if no progress can be made
910	>	* @return false if no progress can be made, else true;
911		*/
912		final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
913	<	boolean removed = false, empty = true, progress = true;
913	>	boolean stat = true, removed = false, empty = true;
914		ForkJoinTask<?>[] a; int m, s, b, n;
915		if ((a = array) != null && (m = a.length - 1) >= 0 &&
916		(n = (s = top) - (b = base)) > 0) {
#	Line 833 \| Line 940 \| public class ForkJoinPool extends Abstra
940		}
941		if (--n == 0) {
942		if (!empty && base == b)
943	<	progress = false;
943	>	stat = false;
944		break;
945		}
946		}
947		}
948		if (removed)
949		task.doExec();
950	<	return progress;
950	>	return stat;
951		}
952
953		/**
954	<	* Initializes or doubles the capacity of array. Call either
955	<	* by owner or with lock held -- it is OK for base, but not
849	<	* top, to move while resizings are in progress.
850	<	*
851	<	* @param rejectOnFailure if true, throw exception if capacity
852	<	* exceeded (relayed ultimately to user); else return null.
954	>	* Polls for and executes the given task or any other task in
955	>	* its CountedCompleter computation.
956		*/
957	<	final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
958	<	ForkJoinTask<?>[] oldA = array;
959	<	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
960	<	if (size <= MAXIMUM_QUEUE_CAPACITY) {
961	<	int oldMask, t, b;
962	<	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
963	<	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
964	<	(t = top) - (b = base) > 0) {
965	<	int mask = size - 1;
966	<	do {
967	<	ForkJoinTask<?> x;
968	<	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
969	<	int j = ((b & mask) << ASHIFT) + ABASE;
970	<	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
971	<	if (x != null &&
972	<	U.compareAndSwapObject(oldA, oldj, x, null))
973	<	U.putObjectVolatile(a, j, x);
974	<	} while (++b != t);
957	>	final boolean pollAndExecCC(ForkJoinTask<?> root) {
958	>	ForkJoinTask<?>[] a; int b; Object o;
959	>	outer: while ((b = base) - top < 0 && (a = array) != null) {
960	>	long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
961	>	if ((o = U.getObject(a, j)) == null \|\|
962	>	!(o instanceof CountedCompleter))
963	>	break;
964	>	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
965	>	if (r == root) {
966	>	if (base == b &&
967	>	U.compareAndSwapObject(a, j, t, null)) {
968	>	base = b + 1;
969	>	t.doExec();
970	>	return true;
971	>	}
972	>	else
973	>	break; // restart
974	>	}
975	>	if ((r = r.completer) == null)
976	>	break outer; // not part of root computation
977		}
873	–	return a;
874	–	}
875	–	else if (!rejectOnFailure)
876	–	return null;
877	–	else
878	–	throw new RejectedExecutionException("Queue capacity exceeded");
879	–	}
880	–
881	–	/**
882	–	* Removes and cancels all known tasks, ignoring any exceptions.
883	–	*/
884	–	final void cancelAll() {
885	–	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
886	–	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
887	–	for (ForkJoinTask<?> t; (t = poll()) != null; )
888	–	ForkJoinTask.cancelIgnoringExceptions(t);
889	–	}
890	–
891	–	/**
892	–	* Computes next value for random probes. Scans don't require
893	–	* a very high quality generator, but also not a crummy one.
894	–	* Marsaglia xor-shift is cheap and works well enough. Note:
895	–	* This is manually inlined in several usages in ForkJoinPool
896	–	* to avoid writes inside busy scan loops.
897	–	*/
898	–	final int nextSeed() {
899	–	int r = seed;
900	–	r ^= r << 13;
901	–	r ^= r >>> 17;
902	–	return seed = r ^= r << 5;
903	–	}
904	–
905	–	// Execution methods
906	–
907	–	/**
908	–	* Removes and runs tasks until empty, using local mode
909	–	* ordering.
910	–	*/
911	–	final void runLocalTasks() {
912	–	if (base - top < 0) {
913	–	for (ForkJoinTask<?> t; (t = nextLocalTask()) != null; )
914	–	t.doExec();
978		}
979	+	return false;
980		}
981
982		/**
983		* Executes a top-level task and any local tasks remaining
984		* after execution.
921	–	*
922	–	* @return true unless terminating
985		*/
986	<	final boolean runTask(ForkJoinTask<?> t) {
925	<	boolean alive = true;
986	>	final void runTask(ForkJoinTask<?> t) {
987		if (t != null) {
988	<	currentSteal = t;
928	<	t.doExec();
929	<	runLocalTasks();
930	<	++nsteals;
988	>	(currentSteal = t).doExec();
989		currentSteal = null;
990	+	++nsteals;
991	+	if (base - top < 0) { // process remaining local tasks
992	+	if (mode == 0)
993	+	popAndExecAll();
994	+	else
995	+	pollAndExecAll();
996	+	}
997		}
933	–	else if (runState < 0) // terminating
934	–	alive = false;
935	–	return alive;
998		}
999
1000		/**
#	Line 941 \| Line 1003 \| public class ForkJoinPool extends Abstra
1003		final void runSubtask(ForkJoinTask<?> t) {
1004		if (t != null) {
1005		ForkJoinTask<?> ps = currentSteal;
1006	<	currentSteal = t;
945	<	t.doExec();
1006	>	(currentSteal = t).doExec();
1007		currentSteal = ps;
1008		}
1009		}
#	Line 959 \| Line 1020 \| public class ForkJoinPool extends Abstra
1020		s != Thread.State.TIMED_WAITING);
1021		}
1022
962	–	/**
963	–	* If this owned and is not already interrupted, try to
964	–	* interrupt and/or unpark, ignoring exceptions.
965	–	*/
966	–	final void interruptOwner() {
967	–	Thread wt, p;
968	–	if ((wt = owner) != null && !wt.isInterrupted()) {
969	–	try {
970	–	wt.interrupt();
971	–	} catch (SecurityException ignore) {
972	–	}
973	–	}
974	–	if ((p = parker) != null)
975	–	U.unpark(p);
976	–	}
977	–
1023		// Unsafe mechanics
1024		private static final sun.misc.Unsafe U;
1025	<	private static final long RUNSTATE;
1025	>	private static final long QLOCK;
1026		private static final int ABASE;
1027		private static final int ASHIFT;
1028		static {
984	–	int s;
1029		try {
1030		U = getUnsafe();
1031		Class<?> k = WorkQueue.class;
1032		Class<?> ak = ForkJoinTask[].class;
1033	<	RUNSTATE = U.objectFieldOffset
1034	<	(k.getDeclaredField("runState"));
1033	>	QLOCK = U.objectFieldOffset
1034	>	(k.getDeclaredField("qlock"));
1035		ABASE = U.arrayBaseOffset(ak);
1036	<	s = U.arrayIndexScale(ak);
1036	>	int scale = U.arrayIndexScale(ak);
1037	>	if ((scale & (scale - 1)) != 0)
1038	>	throw new Error("data type scale not a power of two");
1039	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
1040		} catch (Exception e) {
1041		throw new Error(e);
1042		}
996	–	if ((s & (s-1)) != 0)
997	–	throw new Error("data type scale not a power of two");
998	–	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1043		}
1044		}
1045
1002	–	/**
1003	–	* Per-thread records for threads that submit to pools. Currently
1004	–	* holds only pseudo-random seed / index that is used to choose
1005	–	* submission queues in method doSubmit. In the future, this may
1006	–	* also incorporate a means to implement different task rejection
1007	–	* and resubmission policies.
1008	–	*/
1009	–	static final class Submitter {
1010	–	int seed;
1011	–	Submitter() { seed = hashId(Thread.currentThread().getId()); }
1012	–	}
1013	–
1014	–	/** ThreadLocal class for Submitters */
1015	–	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1016	–	public Submitter initialValue() { return new Submitter(); }
1017	–	}
1018	–
1046		// static fields (initialized in static initializer below)
1047
1048		/**
#	Line 1026 \| Line 1053 \| public class ForkJoinPool extends Abstra
1053		defaultForkJoinWorkerThreadFactory;
1054
1055		/**
1056	<	* Generator for assigning sequence numbers as pool names.
1056	>	* Per-thread submission bookkeeping. Shared across all pools
1057	>	* to reduce ThreadLocal pollution and because random motion
1058	>	* to avoid contention in one pool is likely to hold for others.
1059	>	* Lazily initialized on first submission (but null-checked
1060	>	* in other contexts to avoid unnecessary initialization).
1061		*/
1062	<	private static final AtomicInteger poolNumberGenerator;
1062	>	static final ThreadLocal<Submitter> submitters;
1063
1064		/**
1065		* Permission required for callers of methods that may start or
#	Line 1037 \| Line 1068 \| public class ForkJoinPool extends Abstra
1068		private static final RuntimePermission modifyThreadPermission;
1069
1070		/**
1071	<	* Per-thread submission bookeeping. Shared across all pools
1072	<	* to reduce ThreadLocal pollution and because random motion
1073	<	* to avoid contention in one pool is likely to hold for others.
1071	>	* Common (static) pool. Non-null for public use unless a static
1072	>	* construction exception, but internal usages null-check on use
1073	>	* to paranoically avoid potential initialization circularities
1074	>	* as well as to simplify generated code.
1075		*/
1076	<	private static final ThreadSubmitter submitters;
1076	>	static final ForkJoinPool common;
1077	>
1078	>	/**
1079	>	* Common pool parallelism. Must equal common.parallelism.
1080	>	*/
1081	>	static final int commonParallelism;
1082	>
1083	>	/**
1084	>	* Sequence number for creating workerNamePrefix.
1085	>	*/
1086	>	private static int poolNumberSequence;
1087	>
1088	>	/**
1089	>	* Returns the next sequence number. We don't expect this to
1090	>	* ever contend, so use simple builtin sync.
1091	>	*/
1092	>	private static final synchronized int nextPoolId() {
1093	>	return ++poolNumberSequence;
1094	>	}
1095
1096		// static constants
1097
1098		/**
1099	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
1100	<	* task when the pool is quiescent to instead try to shrink the
1101	<	* number of workers. The exact value does not matter too
1102	<	* much. It must be short enough to release resources during
1103	<	* sustained periods of idleness, but not so short that threads
1104	<	* are continually re-created.
1099	>	* Initial timeout value (in nanoseconds) for the thread
1100	>	* triggering quiescence to park waiting for new work. On timeout,
1101	>	* the thread will instead try to shrink the number of
1102	>	* workers. The value should be large enough to avoid overly
1103	>	* aggressive shrinkage during most transient stalls (long GCs
1104	>	* etc).
1105	>	*/
1106	>	private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
1107	>
1108	>	/**
1109	>	* Timeout value when there are more threads than parallelism level
1110		*/
1111	<	private static final long SHRINK_RATE =
1057	<	4L * 1000L * 1000L * 1000L; // 4 seconds
1111	>	private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
1112
1113		/**
1114	<	* The timeout value for attempted shrinkage, includes
1061	<	* some slop to cope with system timer imprecision.
1114	>	* Tolerance for idle timeouts, to cope with timer undershoots
1115		*/
1116	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1116	>	private static final long TIMEOUT_SLOP = 2000000L;
1117
1118		/**
1119	<	* The maximum stolen->joining link depth allowed in tryHelpStealer.
1120	<	* Depths for legitimate chains are unbounded, but we use a fixed
1121	<	* constant to avoid (otherwise unchecked) cycles and to bound
1122	<	* staleness of traversal parameters at the expense of sometimes
1123	<	* blocking when we could be helping.
1119	>	* The maximum stolen->joining link depth allowed in method
1120	>	* tryHelpStealer. Must be a power of two. Depths for legitimate
1121	>	* chains are unbounded, but we use a fixed constant to avoid
1122	>	* (otherwise unchecked) cycles and to bound staleness of
1123	>	* traversal parameters at the expense of sometimes blocking when
1124	>	* we could be helping.
1125		*/
1126	<	private static final int MAX_HELP_DEPTH = 16;
1126	>	private static final int MAX_HELP = 64;
1127	>
1128	>	/**
1129	>	* Increment for seed generators. See class ThreadLocal for
1130	>	* explanation.
1131	>	*/
1132	>	private static final int SEED_INCREMENT = 0x61c88647;
1133
1134		/**
1135		* Bits and masks for control variables
#	Line 1099 \| Line 1159 \| public class ForkJoinPool extends Abstra
1159		* scan for them to avoid queuing races. Note however that
1160		* eventCount updates lag releases so usage requires care.
1161		*
1162	<	* Field runState is an int packed with:
1162	>	* Field plock is an int packed with:
1163		* SHUTDOWN: true if shutdown is enabled (1 bit)
1164	<	* SEQ: a sequence number updated upon (de)registering workers (15 bits)
1165	<	* MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
1164	>	* SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
1165	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1166		*
1167	<	* The combination of mask and sequence number enables simple
1168	<	* consistency checks: Staleness of read-only operations on the
1169	<	* workQueues array can be checked by comparing runState before vs
1110	<	* after the reads. The low 16 bits (i.e, anding with SMASK) hold
1111	<	* the smallest power of two covering all indices, minus
1112	<	* one.
1167	>	* The sequence number enables simple consistency checks:
1168	>	* Staleness of read-only operations on the workQueues array can
1169	>	* be checked by comparing plock before vs after the reads.
1170		*/
1171
1172		// bit positions/shifts for fields
#	Line 1119 \| Line 1176 \| public class ForkJoinPool extends Abstra
1176		private static final int EC_SHIFT = 16;
1177
1178		// bounds
1122	–	private static final int POOL_MAX = 0x7fff; // max #workers - 1
1179		private static final int SMASK = 0xffff; // short bits
1180	<	private static final int SQMASK = 0xfffe; // even short bits
1180	>	private static final int MAX_CAP = 0x7fff; // max #workers - 1
1181	>	private static final int EVENMASK = 0xfffe; // even short bits
1182	>	private static final int SQMASK = 0x007e; // max 64 (even) slots
1183		private static final int SHORT_SIGN = 1 << 15;
1184		private static final int INT_SIGN = 1 << 31;
1185
#	Line 1146 \| Line 1204 \| public class ForkJoinPool extends Abstra
1204		private static final int E_MASK = 0x7fffffff; // no STOP_BIT
1205		private static final int E_SEQ = 1 << EC_SHIFT;
1206
1207	<	// runState bits
1207	>	// plock bits
1208		private static final int SHUTDOWN = 1 << 31;
1209	<	private static final int RS_SEQ = 1 << 16;
1210	<	private static final int RS_SEQ_MASK = 0x7fff0000;
1209	>	private static final int PL_LOCK = 2;
1210	>	private static final int PL_SIGNAL = 1;
1211	>	private static final int PL_SPINS = 1 << 8;
1212
1213		// access mode for WorkQueue
1214		static final int LIFO_QUEUE = 0;
1215		static final int FIFO_QUEUE = 1;
1216		static final int SHARED_QUEUE = -1;
1217
1218	+	// bounds for #steps in scan loop -- must be power 2 minus 1
1219	+	private static final int MIN_SCAN = 0x1ff; // cover estimation slop
1220	+	private static final int MAX_SCAN = 0x1ffff; // 4 * max workers
1221	+
1222		// Instance fields
1223
1224		/*
1225	<	* Field layout order in this class tends to matter more than one
1226	<	* would like. Runtime layout order is only loosely related to
1225	>	* Field layout of this class tends to matter more than one would
1226	>	* like. Runtime layout order is only loosely related to
1227		* declaration order and may differ across JVMs, but the following
1228		* empirically works OK on current JVMs.
1229		*/
1230
1231	+	// Heuristic padding to ameliorate unfortunate memory placements
1232	+	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
1233	+
1234	+	volatile long stealCount; // collects worker counts
1235		volatile long ctl; // main pool control
1236	<	final int parallelism; // parallelism level
1237	<	final int localMode; // per-worker scheduling mode
1238	<	int growHints; // for expanding indices/ranges
1172	<	volatile int runState; // shutdown status, seq, and mask
1236	>	volatile int plock; // shutdown status and seqLock
1237	>	volatile int indexSeed; // worker/submitter index seed
1238	>	final int config; // mode and parallelism level
1239		WorkQueue[] workQueues; // main registry
1240	<	final Mutex lock; // for registration
1175	<	final Condition termination; // for awaitTermination
1176	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1240	>	final ForkJoinWorkerThreadFactory factory;
1241		final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1178	–	final AtomicLong stealCount; // collect counts when terminated
1179	–	final AtomicInteger nextWorkerNumber; // to create worker name string
1242		final String workerNamePrefix; // to create worker name string
1243
1244	<	// Creating, registering, deregistering and running workers
1244	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
1245	>	volatile Object pad18, pad19, pad1a, pad1b;
1246
1247		/**
1248	<	* Tries to create and start a worker
1249	<	*/
1250	<	private void addWorker() {
1251	<	Throwable ex = null;
1252	<	ForkJoinWorkerThread wt = null;
1253	<	try {
1254	<	if ((wt = factory.newThread(this)) != null) {
1255	<	wt.start();
1256	<	return;
1248	>	* Acquires the plock lock to protect worker array and related
1249	>	* updates. This method is called only if an initial CAS on plock
1250	>	* fails. This acts as a spinlock for normal cases, but falls back
1251	>	* to builtin monitor to block when (rarely) needed. This would be
1252	>	* a terrible idea for a highly contended lock, but works fine as
1253	>	* a more conservative alternative to a pure spinlock.
1254	>	*/
1255	>	private int acquirePlock() {
1256	>	int spins = PL_SPINS, r = 0, ps, nps;
1257	>	for (;;) {
1258	>	if (((ps = plock) & PL_LOCK) == 0 &&
1259	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1260	>	return nps;
1261	>	else if (r == 0) { // randomize spins if possible
1262	>	Thread t = Thread.currentThread(); WorkQueue w; Submitter z;
1263	>	if ((t instanceof ForkJoinWorkerThread) &&
1264	>	(w = ((ForkJoinWorkerThread)t).workQueue) != null)
1265	>	r = w.seed;
1266	>	else if ((z = submitters.get()) != null)
1267	>	r = z.seed;
1268	>	else
1269	>	r = 1;
1270	>	}
1271	>	else if (spins >= 0) {
1272	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1273	>	if (r >= 0)
1274	>	--spins;
1275	>	}
1276	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps \| PL_SIGNAL)) {
1277	>	synchronized (this) {
1278	>	if ((plock & PL_SIGNAL) != 0) {
1279	>	try {
1280	>	wait();
1281	>	} catch (InterruptedException ie) {
1282	>	try {
1283	>	Thread.currentThread().interrupt();
1284	>	} catch (SecurityException ignore) {
1285	>	}
1286	>	}
1287	>	}
1288	>	else
1289	>	notifyAll();
1290	>	}
1291		}
1195	–	} catch (Throwable e) {
1196	–	ex = e;
1292		}
1198	–	deregisterWorker(wt, ex); // adjust counts etc on failure
1293		}
1294
1295		/**
1296	<	* Callback from ForkJoinWorkerThread constructor to assign a
1297	<	* public name. This must be separate from registerWorker because
1298	<	* it is called during the "super" constructor call in
1299	<	* ForkJoinWorkerThread.
1296	>	* Unlocks and signals any thread waiting for plock. Called only
1297	>	* when CAS of seq value for unlock fails.
1298	>	*/
1299	>	private void releasePlock(int ps) {
1300	>	plock = ps;
1301	>	synchronized (this) { notifyAll(); }
1302	>	}
1303	>
1304	>	/**
1305	>	* Tries to create and start one worker if fewer than target
1306	>	* parallelism level exist. Adjusts counts etc on failure.
1307		*/
1308	<	final String nextWorkerName() {
1309	<	return workerNamePrefix.concat
1310	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1308	>	private void tryAddWorker() {
1309	>	long c; int u;
1310	>	while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1311	>	(u & SHORT_SIGN) != 0 && (int)c == 0) {
1312	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1313	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1314	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1315	>	ForkJoinWorkerThreadFactory fac;
1316	>	Throwable ex = null;
1317	>	ForkJoinWorkerThread wt = null;
1318	>	try {
1319	>	if ((fac = factory) != null &&
1320	>	(wt = fac.newThread(this)) != null) {
1321	>	wt.start();
1322	>	break;
1323	>	}
1324	>	} catch (Throwable e) {
1325	>	ex = e;
1326	>	}
1327	>	deregisterWorker(wt, ex);
1328	>	break;
1329	>	}
1330	>	}
1331		}
1332
1333	+	// Registering and deregistering workers
1334	+
1335		/**
1336	<	* Callback from ForkJoinWorkerThread constructor to establish and
1337	<	* record its WorkQueue.
1336	>	* Callback from ForkJoinWorkerThread to establish and record its
1337	>	* WorkQueue. To avoid scanning bias due to packing entries in
1338	>	* front of the workQueues array, we treat the array as a simple
1339	>	* power-of-two hash table using per-thread seed as hash,
1340	>	* expanding as needed.
1341		*
1342		* @param wt the worker thread
1343	+	* @return the worker's queue
1344		*/
1345	<	final void registerWorker(ForkJoinWorkerThread wt) {
1346	<	WorkQueue w = wt.workQueue;
1347	<	Mutex lock = this.lock;
1348	<	lock.lock();
1345	>	final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1346	>	Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1347	>	wt.setDaemon(true);
1348	>	if ((handler = ueh) != null)
1349	>	wt.setUncaughtExceptionHandler(handler);
1350	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1351	>	s += SEED_INCREMENT) \|\|
1352	>	s == 0); // skip 0
1353	>	WorkQueue w = new WorkQueue(this, wt, config >>> 16, s);
1354	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1355	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1356	>	ps = acquirePlock();
1357	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1358		try {
1359	<	int g = growHints, k = g & SMASK;
1360	<	WorkQueue[] ws = workQueues;
1361	<	if (ws != null) { // ignore on shutdown
1362	<	int n = ws.length;
1363	<	if ((k & 1) == 0 \|\| k >= n \|\| ws[k] != null) {
1364	<	for (k = 1; k < n && ws[k] != null; k += 2)
1365	<	; // workers are at odd indices
1366	<	if (k >= n) // resize
1367	<	workQueues = ws = Arrays.copyOf(ws, n << 1);
1368	<	}
1369	<	w.eventCount = w.poolIndex = k; // establish before recording
1370	<	ws[k] = w;
1371	<	growHints = (g & ~SMASK) \| ((k + 2) & SMASK);
1372	<	int rs = runState;
1373	<	int m = rs & SMASK; // recalculate runState mask
1374	<	if (k > m)
1239	<	m = (m << 1) + 1;
1240	<	runState = (rs & SHUTDOWN) \| ((rs + RS_SEQ) & RS_SEQ_MASK) \| m;
1359	>	if ((ws = workQueues) != null) { // skip if shutting down
1360	>	int n = ws.length, m = n - 1;
1361	>	int r = (s << 1) \| 1; // use odd-numbered indices
1362	>	if (ws[r &= m] != null) { // collision
1363	>	int probes = 0; // step by approx half size
1364	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1365	>	while (ws[r = (r + step) & m] != null) {
1366	>	if (++probes >= n) {
1367	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1368	>	m = n - 1;
1369	>	probes = 0;
1370	>	}
1371	>	}
1372	>	}
1373	>	w.eventCount = w.poolIndex = r; // volatile write orders
1374	>	ws[r] = w;
1375		}
1376		} finally {
1377	<	lock.unlock();
1377	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1378	>	releasePlock(nps);
1379		}
1380	+	wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex)));
1381	+	return w;
1382		}
1383
1384		/**
1385		* Final callback from terminating worker, as well as upon failure
1386	<	* to construct or start a worker in addWorker. Removes record of
1387	<	* worker from array, and adjusts counts. If pool is shutting
1388	<	* down, tries to complete termination.
1386	>	* to construct or start a worker. Removes record of worker from
1387	>	* array, and adjusts counts. If pool is shutting down, tries to
1388	>	* complete termination.
1389		*
1390	<	* @param wt the worker thread or null if addWorker failed
1390	>	* @param wt the worker thread or null if construction failed
1391		* @param ex the exception causing failure, or null if none
1392		*/
1393		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1394		WorkQueue w = null;
1395		if (wt != null && (w = wt.workQueue) != null) {
1396	<	w.runState = -1; // ensure runState is set
1397	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1398	<	int idx = w.poolIndex;
1399	<	Mutex lock = this.lock;
1400	<	lock.lock();
1401	<	try { // remove record from array
1396	>	int ps;
1397	>	w.qlock = -1; // ensure set
1398	>	long ns = w.nsteals, sc; // collect steal count
1399	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1400	>	sc = stealCount, sc + ns));
1401	>	if (((ps = plock) & PL_LOCK) != 0 \|\|
1402	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1403	>	ps = acquirePlock();
1404	>	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1405	>	try {
1406	>	int idx = w.poolIndex;
1407		WorkQueue[] ws = workQueues;
1408	<	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w) {
1408	>	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1409		ws[idx] = null;
1268	–	growHints = (growHints & ~SMASK) \| idx;
1269	–	}
1410		} finally {
1411	<	lock.unlock();
1411	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1412	>	releasePlock(nps);
1413		}
1414		}
1415
1416	<	long c; // adjust ctl counts
1416	>	long c; // adjust ctl counts
1417		do {} while (!U.compareAndSwapLong
1418		(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) \|
1419		((c - TC_UNIT) & TC_MASK) \|
1420		(c & ~(AC_MASK\|TC_MASK)))));
1421
1422	<	if (!tryTerminate(false, false) && w != null) {
1423	<	w.cancelAll(); // cancel remaining tasks
1424	<	if (w.array != null) // suppress signal if never ran
1425	<	signalWork(); // wake up or create replacement
1426	<	if (ex == null) // help clean refs on way out
1427	<	ForkJoinTask.helpExpungeStaleExceptions();
1422	>	if (!tryTerminate(false, false) && w != null && w.array != null) {
1423	>	w.cancelAll(); // cancel remaining tasks
1424	>	WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e;
1425	>	while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) {
1426	>	if (e > 0) { // activate or create replacement
1427	>	if ((ws = workQueues) == null \|\|
1428	>	(i = e & SMASK) >= ws.length \|\|
1429	>	(v = ws[i]) == null)
1430	>	break;
1431	>	long nc = (((long)(v.nextWait & E_MASK)) \|
1432	>	((long)(u + UAC_UNIT) << 32));
1433	>	if (v.eventCount != (e \| INT_SIGN))
1434	>	break;
1435	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1436	>	v.eventCount = (e + E_SEQ) & E_MASK;
1437	>	if ((p = v.parker) != null)
1438	>	U.unpark(p);
1439	>	break;
1440	>	}
1441	>	}
1442	>	else {
1443	>	if ((short)u < 0)
1444	>	tryAddWorker();
1445	>	break;
1446	>	}
1447	>	}
1448		}
1449	<
1450	<	if (ex != null) // rethrow
1451	<	U.throwException(ex);
1452	<	}
1292	<
1293	<	/**
1294	<	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1295	<	*/
1296	<	final void runWorker(ForkJoinWorkerThread wt) {
1297	<	// Initialize queue array and seed in this thread
1298	<	WorkQueue w = wt.workQueue;
1299	<	w.growArray(false);
1300	<	w.seed = hashId(Thread.currentThread().getId());
1301	<
1302	<	do {} while (w.runTask(scan(w)));
1449	>	if (ex == null) // help clean refs on way out
1450	>	ForkJoinTask.helpExpungeStaleExceptions();
1451	>	else // rethrow
1452	>	ForkJoinTask.rethrow(ex);
1453		}
1454
1455		// Submissions
#	Line 1307 \| Line 1457 \| public class ForkJoinPool extends Abstra
1457		/**
1458		* Unless shutting down, adds the given task to a submission queue
1459		* at submitter's current queue index (modulo submission
1460	<	* range). If no queue exists at the index, one is created unless
1461	<	* pool lock is busy. If the queue and/or lock are busy, another
1462	<	* index is randomly chosen. The mask in growHints controls the
1463	<	* effective index range of queues considered. The mask is
1314	<	* expanded, up to the current workerQueue mask, upon any detected
1315	<	* contention but otherwise remains small to avoid needlessly
1316	<	* creating queues when there is no contention.
1460	>	* range). Only the most common path is directly handled in this
1461	>	* method. All others are relayed to fullExternalPush.
1462	>	*
1463	>	* @param task the task. Caller must ensure non-null.
1464		*/
1465	<	private void doSubmit(ForkJoinTask<?> task) {
1466	<	if (task == null)
1467	<	throw new NullPointerException();
1468	<	Submitter s = submitters.get();
1469	<	for (int r = s.seed, m = growHints >>> 16;;) {
1470	<	WorkQueue[] ws; WorkQueue q; Mutex lk;
1471	<	int k = r & m & SQMASK; // use only even indices
1472	<	if (runState < 0 \|\| (ws = workQueues) == null \|\| ws.length <= k)
1473	<	throw new RejectedExecutionException(); // shutting down
1474	<	if ((q = ws[k]) == null && (lk = lock).tryAcquire(0)) {
1475	<	try { // try to create new queue
1476	<	if (ws == workQueues && (q = ws[k]) == null) {
1477	<	int rs; // update runState seq
1478	<	ws[k] = q = new WorkQueue(null, SHARED_QUEUE);
1479	<	runState = (((rs = runState) & SHUTDOWN) \|
1333	<	((rs + RS_SEQ) & ~SHUTDOWN));
1334	<	}
1335	<	} finally {
1336	<	lk.unlock();
1337	<	}
1338	<	}
1339	<	if (q != null) {
1340	<	if (q.trySharedPush(task)) {
1341	<	signalWork();
1342	<	return;
1343	<	}
1344	<	else if (m < parallelism - 1 && m < (runState & SMASK)) {
1345	<	Mutex lock = this.lock;
1346	<	lock.lock(); // block until lock free
1347	<	int g = growHints;
1348	<	if (g >>> 16 == m) // expand range
1349	<	growHints = (((m << 1) + 1) << 16) \| (g & SMASK);
1350	<	lock.unlock(); // no need for try/finally
1351	<	}
1352	<	else if ((r & m) == 0)
1353	<	Thread.yield(); // occasionally yield if busy
1354	<	}
1355	<	if (m == (m = growHints >>> 16)) {
1356	<	r ^= r << 13; // update seed unless new range
1357	<	r ^= r >>> 17; // same xorshift as WorkQueues
1358	<	s.seed = r ^= r << 5;
1465	>	final void externalPush(ForkJoinTask<?> task) {
1466	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1467	>	if ((z = submitters.get()) != null && plock > 0 &&
1468	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1469	>	(q = ws[m & z.seed & SQMASK]) != null &&
1470	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1471	>	int b = q.base, s = q.top, n, an;
1472	>	if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) {
1473	>	int j = (((an - 1) & s) << ASHIFT) + ABASE;
1474	>	U.putOrderedObject(a, j, task);
1475	>	q.top = s + 1; // push on to deque
1476	>	q.qlock = 0;
1477	>	if (n <= 2)
1478	>	signalWork(q);
1479	>	return;
1480		}
1481	+	q.qlock = 0;
1482	+	}
1483	+	fullExternalPush(task);
1484	+	}
1485	+
1486	+	/**
1487	+	* Full version of externalPush. This method is called, among
1488	+	* other times, upon the first submission of the first task to the
1489	+	* pool, so must perform secondary initialization. It also
1490	+	* detects first submission by an external thread by looking up
1491	+	* its ThreadLocal, and creates a new shared queue if the one at
1492	+	* index if empty or contended. The plock lock body must be
1493	+	* exception-free (so no try/finally) so we optimistically
1494	+	* allocate new queues outside the lock and throw them away if
1495	+	* (very rarely) not needed.
1496	+	*
1497	+	* Secondary initialization occurs when plock is zero, to create
1498	+	* workQueue array and set plock to a valid value. This lock body
1499	+	* must also be exception-free. Because the plock seq value can
1500	+	* eventually wrap around zero, this method harmlessly fails to
1501	+	* reinitialize if workQueues exists, while still advancing plock.
1502	+	*/
1503	+	private void fullExternalPush(ForkJoinTask<?> task) {
1504	+	int r = 0; // random index seed
1505	+	for (Submitter z = submitters.get();;) {
1506	+	WorkQueue[] ws; WorkQueue q; int ps, m, k;
1507	+	if (z == null) {
1508	+	if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1509	+	r += SEED_INCREMENT) && r != 0)
1510	+	submitters.set(z = new Submitter(r));
1511	+	}
1512	+	else if (r == 0) { // move to a different index
1513	+	r = z.seed;
1514	+	r ^= r << 13; // same xorshift as WorkQueues
1515	+	r ^= r >>> 17;
1516	+	z.seed = r ^ (r << 5);
1517	+	}
1518	+	else if ((ps = plock) < 0)
1519	+	throw new RejectedExecutionException();
1520	+	else if (ps == 0 \|\| (ws = workQueues) == null \|\|
1521	+	(m = ws.length - 1) < 0) { // initialize workQueues
1522	+	int p = config & SMASK; // find power of two table size
1523	+	int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots
1524	+	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4;
1525	+	n \|= n >>> 8; n \|= n >>> 16; n = (n + 1) << 1;
1526	+	WorkQueue[] nws = ((ws = workQueues) == null \|\| ws.length == 0 ?
1527	+	new WorkQueue[n] : null);
1528	+	if (((ps = plock) & PL_LOCK) != 0 \|\|
1529	+	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1530	+	ps = acquirePlock();
1531	+	if (((ws = workQueues) == null \|\| ws.length == 0) && nws != null)
1532	+	workQueues = nws;
1533	+	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1534	+	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1535	+	releasePlock(nps);
1536	+	}
1537	+	else if ((q = ws[k = r & m & SQMASK]) != null) {
1538	+	if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
1539	+	ForkJoinTask<?>[] a = q.array;
1540	+	int s = q.top;
1541	+	boolean submitted = false;
1542	+	try { // locked version of push
1543	+	if ((a != null && a.length > s + 1 - q.base) \|\|
1544	+	(a = q.growArray()) != null) { // must presize
1545	+	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
1546	+	U.putOrderedObject(a, j, task);
1547	+	q.top = s + 1;
1548	+	submitted = true;
1549	+	}
1550	+	} finally {
1551	+	q.qlock = 0; // unlock
1552	+	}
1553	+	if (submitted) {
1554	+	signalWork(q);
1555	+	return;
1556	+	}
1557	+	}
1558	+	r = 0; // move on failure
1559	+	}
1560	+	else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1561	+	q = new WorkQueue(this, null, SHARED_QUEUE, r);
1562	+	if (((ps = plock) & PL_LOCK) != 0 \|\|
1563	+	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1564	+	ps = acquirePlock();
1565	+	if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1566	+	ws[k] = q;
1567	+	int nps = (ps & SHUTDOWN) \| ((ps + PL_LOCK) & ~SHUTDOWN);
1568	+	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1569	+	releasePlock(nps);
1570	+	}
1571	+	else
1572	+	r = 0; // try elsewhere while lock held
1573		}
1574		}
1575
#	Line 1371 \| Line 1584 \| public class ForkJoinPool extends Abstra
1584		}
1585
1586		/**
1587	<	* Tries to activate or create a worker if too few are active.
1587	>	* Tries to create or activate a worker if too few are active.
1588	>	*
1589	>	* @param q the (non-null) queue holding tasks to be signalled
1590		*/
1591	<	final void signalWork() {
1592	<	long c; int u;
1593	<	while ((u = (int)((c = ctl) >>> 32)) < 0) { // too few active
1594	<	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1595	<	if ((e = (int)c) > 0) { // at least one waiting
1596	<	if (ws != null && (i = e & SMASK) < ws.length &&
1591	>	final void signalWork(WorkQueue q) {
1592	>	int hint = q.poolIndex;
1593	>	long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p;
1594	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1595	>	if ((e = (int)c) > 0) {
1596	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1597		(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1598		long nc = (((long)(w.nextWait & E_MASK)) \|
1599		((long)(u + UAC_UNIT) << 32));
1600		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1601	+	w.hint = hint;
1602		w.eventCount = (e + E_SEQ) & E_MASK;
1603		if ((p = w.parker) != null)
1604	<	U.unpark(p); // activate and release
1604	>	U.unpark(p);
1605		break;
1606		}
1607	+	if (q.top - q.base <= 0)
1608	+	break;
1609		}
1610		else
1611		break;
1612		}
1613	<	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1614	<	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1615	<	((u + UAC_UNIT) & UAC_MASK)) << 32;
1398	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1399	<	addWorker();
1400	<	break;
1401	<	}
1402	<	}
1403	<	else
1613	>	else {
1614	>	if ((short)u < 0)
1615	>	tryAddWorker();
1616		break;
1617	+	}
1618		}
1619		}
1620
1621	+	// Scanning for tasks
1622	+
1623		/**
1624	<	* Tries to decrement active count (sometimes implicitly) and
1410	<	* possibly release or create a compensating worker in preparation
1411	<	* for blocking. Fails on contention or termination.
1412	<	*
1413	<	* @return true if the caller can block, else should recheck and retry
1624	>	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1625		*/
1626	<	final boolean tryCompensate() {
1627	<	WorkQueue w; Thread p;
1628	<	int pc = parallelism, e, u, ac, tc, i;
1418	<	long c = ctl;
1419	<	WorkQueue[] ws = workQueues;
1420	<	if ((e = (int)c) >= 0) {
1421	<	if ((ac = ((u = (int)(c >>> 32)) >> UAC_SHIFT)) <= 0 &&
1422	<	e != 0 && ws != null && (i = e & SMASK) < ws.length &&
1423	<	(w = ws[i]) != null) {
1424	<	long nc = (long)(w.nextWait & E_MASK) \| (c & (AC_MASK\|TC_MASK));
1425	<	if (w.eventCount == (e \| INT_SIGN) &&
1426	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1427	<	w.eventCount = (e + E_SEQ) & E_MASK;
1428	<	if ((p = w.parker) != null)
1429	<	U.unpark(p);
1430	<	return true; // release an idle worker
1431	<	}
1432	<	}
1433	<	else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
1434	<	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1435	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1436	<	return true; // no compensation needed
1437	<	}
1438	<	else if (tc + pc < POOL_MAX) {
1439	<	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1440	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1441	<	addWorker();
1442	<	return true; // create replacement
1443	<	}
1444	<	}
1445	<	}
1446	<	return false;
1626	>	final void runWorker(WorkQueue w) {
1627	>	w.growArray(); // allocate queue
1628	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1629		}
1630
1449	–	// Scanning for tasks
1450	–
1631		/**
1632		* Scans for and, if found, returns one task, else possibly
1633		* inactivates the worker. This method operates on single reads of
1634	<	* volatile state and is designed to be re-invoked continuously in
1635	<	* part because it returns upon detecting inconsistencies,
1634	>	* volatile state and is designed to be re-invoked continuously,
1635	>	* in part because it returns upon detecting inconsistencies,
1636		* contention, or state changes that indicate possible success on
1637		* re-invocation.
1638		*
1639	<	* The scan searches for tasks across queues, randomly selecting
1640	<	* the first #queues probes, favoring steals over submissions
1641	<	* (by exploiting even/odd indexing), and then performing a
1642	<	* circular sweep of all queues. The scan terminates upon either
1643	<	* finding a non-empty queue, or completing a full sweep. If the
1644	<	* worker is not inactivated, it takes and returns a task from
1645	<	* this queue. On failure to find a task, we take one of the
1646	<	* following actions, after which the caller will retry calling
1647	<	* this method unless terminated.
1639	>	* The scan searches for tasks across queues (starting at a random
1640	>	* index, and relying on registerWorker to irregularly scatter
1641	>	* them within array to avoid bias), checking each at least twice.
1642	>	* The scan terminates upon either finding a non-empty queue, or
1643	>	* completing the sweep. If the worker is not inactivated, it
1644	>	* takes and returns a task from this queue. Otherwise, if not
1645	>	* activated, it signals workers (that may include itself) and
1646	>	* returns so caller can retry. Also returns for true if the
1647	>	* worker array may have changed during an empty scan. On failure
1648	>	* to find a task, we take one of the following actions, after
1649	>	* which the caller will retry calling this method unless
1650	>	* terminated.
1651		*
1652		* * If pool is terminating, terminate the worker.
1653		*
1471	–	* * If not a complete sweep, try to release a waiting worker. If
1472	–	* the scan terminated because the worker is inactivated, then the
1473	–	* released worker will often be the calling worker, and it can
1474	–	* succeed obtaining a task on the next call. Or maybe it is
1475	–	* another worker, but with same net effect. Releasing in other
1476	–	* cases as well ensures that we have enough workers running.
1477	–	*
1478	–	* * If the caller has run a task since the last empty scan,
1479	–	* return (to allow rescan) if other workers are not also yet
1480	–	* enqueued. Field WorkQueue.rescans counts down on each scan to
1481	–	* ensure eventual inactivation and blocking.
1482	–	*
1654		* * If not already enqueued, try to inactivate and enqueue the
1655	<	* worker on wait queue.
1655	>	* worker on wait queue. Or, if inactivating has caused the pool
1656	>	* to be quiescent, relay to idleAwaitWork to possibly shrink
1657	>	* pool.
1658	>	*
1659	>	* * If already enqueued and none of the above apply, possibly
1660	>	* park awaiting signal, else lingering to help scan and signal.
1661		*
1662	<	* * If already enqueued and none of the above apply, either park
1663	<	* awaiting signal, or if this is the most recent waiter and pool
1488	<	* is quiescent, relay to idleAwaitWork to check for termination
1489	<	* and possibly shrink pool.
1662	>	* * If a non-empty queue discovered or left as a hint,
1663	>	* help wake up other workers before return.
1664		*
1665		* @param w the worker (via its WorkQueue)
1666	<	* @return a task or null of none found
1666	>	* @return a task or null if none found
1667		*/
1668		private final ForkJoinTask<?> scan(WorkQueue w) {
1669	<	boolean swept = false; // true after full empty scan
1670	<	WorkQueue[] ws; // volatile read order matters
1671	<	int r = w.seed, ec = w.eventCount; // ec is negative if inactive
1672	<	int rs = runState, m = rs & SMASK;
1673	<	if ((ws = workQueues) != null && ws.length > m) { // consistency check
1674	<	for (int k = 0, j = -1 - m; ; ++j) {
1675	<	WorkQueue q; int b;
1676	<	if (j < 0) { // random probes while j negative
1677	<	r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) \| (j & 1);
1678	<	} // worker (not submit) for odd j
1679	<	else // cyclic scan when j >= 0
1680	<	k += 7; // step 7 reduces array packing bias
1681	<	if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
1682	<	ForkJoinTask<?> t = (ec >= 0) ? q.pollAt(b) : null;
1683	<	w.seed = r; // save seed for next scan
1684	<	if (t != null)
1685	<	return t;
1686	<	break;
1687	<	}
1688	<	else if (j - m > m) {
1689	<	if (rs == runState) // staleness check
1690	<	swept = true;
1691	<	break;
1669	>	WorkQueue[] ws; int m;
1670	>	int ps = plock; // read plock before ws
1671	>	if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1672	>	int ec = w.eventCount; // ec is negative if inactive
1673	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1674	>	w.hint = -1; // update seed and clear hint
1675	>	int j = ((m + m + 1) \| MIN_SCAN) & MAX_SCAN;
1676	>	do {
1677	>	WorkQueue q; ForkJoinTask<?>[] a; int b;
1678	>	if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 &&
1679	>	(a = q.array) != null) { // probably nonempty
1680	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1681	>	ForkJoinTask<?> t = (ForkJoinTask<?>)
1682	>	U.getObjectVolatile(a, i);
1683	>	if (q.base == b && ec >= 0 && t != null &&
1684	>	U.compareAndSwapObject(a, i, t, null)) {
1685	>	if ((q.base = b + 1) - q.top < 0)
1686	>	signalWork(q);
1687	>	return t; // taken
1688	>	}
1689	>	else if ((ec < 0 \|\| j < m) && (int)(ctl >> AC_SHIFT) <= 0) {
1690	>	w.hint = (r + j) & m; // help signal below
1691	>	break; // cannot take
1692	>	}
1693		}
1694	<	}
1694	>	} while (--j >= 0);
1695
1696	<	// Decode ctl on empty scan
1697	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1698	<	if (e < 0) // pool is terminating
1699	<	w.runState = -1;
1700	<	else if (!swept) { // try to release a waiter
1701	<	WorkQueue v; Thread p;
1702	<	if (e > 0 && a < 0 && (v = ws[e & m]) != null &&
1703	<	v.eventCount == (e \| INT_SIGN)) {
1704	<	long nc = ((long)(v.nextWait & E_MASK) \|
1705	<	((c + AC_UNIT) & (AC_MASK\|TC_MASK)));
1706	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1696	>	int h, e, ns; long c, sc; WorkQueue q;
1697	>	if ((ns = w.nsteals) != 0) {
1698	>	if (U.compareAndSwapLong(this, STEALCOUNT,
1699	>	sc = stealCount, sc + ns))
1700	>	w.nsteals = 0; // collect steals and rescan
1701	>	}
1702	>	else if (plock != ps) // consistency check
1703	>	; // skip
1704	>	else if ((e = (int)(c = ctl)) < 0)
1705	>	w.qlock = -1; // pool is terminating
1706	>	else {
1707	>	if ((h = w.hint) < 0) {
1708	>	if (ec >= 0) { // try to enqueue/inactivate
1709	>	long nc = (((long)ec \|
1710	>	((c - AC_UNIT) & (AC_MASK\|TC_MASK))));
1711	>	w.nextWait = e; // link and mark inactive
1712	>	w.eventCount = ec \| INT_SIGN;
1713	>	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1714	>	w.eventCount = ec; // unmark on CAS failure
1715	>	else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK))
1716	>	idleAwaitWork(w, nc, c);
1717	>	}
1718	>	else if (w.eventCount < 0 && ctl == c) {
1719	>	Thread wt = Thread.currentThread();
1720	>	Thread.interrupted(); // clear status
1721	>	U.putObject(wt, PARKBLOCKER, this);
1722	>	w.parker = wt; // emulate LockSupport.park
1723	>	if (w.eventCount < 0) // recheck
1724	>	U.park(false, 0L); // block
1725	>	w.parker = null;
1726	>	U.putObject(wt, PARKBLOCKER, null);
1727	>	}
1728	>	}
1729	>	if ((h >= 0 \|\| (h = w.hint) >= 0) &&
1730	>	(ws = workQueues) != null && h < ws.length &&
1731	>	(q = ws[h]) != null) { // signal others before retry
1732	>	WorkQueue v; Thread p; int u, i, s;
1733	>	for (int n = (config & SMASK) - 1;;) {
1734	>	int idleCount = (w.eventCount < 0) ? 0 : -1;
1735	>	if (((s = idleCount - q.base + q.top) <= n &&
1736	>	(n = s) <= 0) \|\|
1737	>	(u = (int)((c = ctl) >>> 32)) >= 0 \|\|
1738	>	(e = (int)c) <= 0 \|\| m < (i = e & SMASK) \|\|
1739	>	(v = ws[i]) == null)
1740	>	break;
1741	>	long nc = (((long)(v.nextWait & E_MASK)) \|
1742	>	((long)(u + UAC_UNIT) << 32));
1743	>	if (v.eventCount != (e \| INT_SIGN) \|\|
1744	>	!U.compareAndSwapLong(this, CTL, c, nc))
1745	>	break;
1746	>	v.hint = h;
1747		v.eventCount = (e + E_SEQ) & E_MASK;
1748		if ((p = v.parker) != null)
1749		U.unpark(p);
1750	+	if (--n <= 0)
1751	+	break;
1752		}
1753		}
1754		}
1538	–	else if ((nr = w.rescans) > 0) { // continue rescanning
1539	–	int ac = a + parallelism;
1540	–	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0 &&
1541	–	w.eventCount == ec)
1542	–	Thread.yield(); // occasionally yield
1543	–	}
1544	–	else if (ec >= 0) { // try to enqueue
1545	–	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1546	–	w.nextWait = e;
1547	–	w.eventCount = ec \| INT_SIGN;// mark as inactive
1548	–	if (!U.compareAndSwapLong(this, CTL, c, nc))
1549	–	w.eventCount = ec; // unmark on CAS failure
1550	–	else if ((ns = w.nsteals) != 0) {
1551	–	w.nsteals = 0; // set rescans if ran task
1552	–	w.rescans = a + parallelism;
1553	–	w.totalSteals += ns;
1554	–	}
1555	–	}
1556	–	else{ // already queued
1557	–	if (parallelism == -a)
1558	–	idleAwaitWork(w); // quiescent
1559	–	if (w.eventCount == ec) {
1560	–	Thread.interrupted(); // clear status
1561	–	ForkJoinWorkerThread wt = w.owner;
1562	–	U.putObject(wt, PARKBLOCKER, this);
1563	–	w.parker = wt; // emulate LockSupport.park
1564	–	if (w.eventCount == ec) // recheck
1565	–	U.park(false, 0L); // block
1566	–	w.parker = null;
1567	–	U.putObject(wt, PARKBLOCKER, null);
1568	–	}
1569	–	}
1755		}
1756		return null;
1757		}
1758
1759		/**
1760	<	* If inactivating worker w has caused pool to become quiescent,
1761	<	* checks for pool termination, and, so long as this is not the
1762	<	* only worker, waits for event for up to SHRINK_RATE nanosecs.
1763	<	* On timeout, if ctl has not changed, terminates the worker,
1764	<	* which will in turn wake up another worker to possibly repeat
1765	<	* this process.
1760	>	* If inactivating worker w has caused the pool to become
1761	>	* quiescent, checks for pool termination, and, so long as this is
1762	>	* not the only worker, waits for event for up to a given
1763	>	* duration. On timeout, if ctl has not changed, terminates the
1764	>	* worker, which will in turn wake up another worker to possibly
1765	>	* repeat this process.
1766		*
1767		* @param w the calling worker
1768	+	* @param currentCtl the ctl value triggering possible quiescence
1769	+	* @param prevCtl the ctl value to restore if thread is terminated
1770		*/
1771	<	private void idleAwaitWork(WorkQueue w) {
1772	<	long c; int nw, ec;
1773	<	if (!tryTerminate(false, false) &&
1774	<	(int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
1775	<	(ec = w.eventCount) == ((int)c \| INT_SIGN) &&
1776	<	(nw = w.nextWait) != 0) {
1777	<	long nc = ((long)(nw & E_MASK) \| // ctl to restore on timeout
1778	<	((c + AC_UNIT) & AC_MASK) \| (c & TC_MASK));
1779	<	ForkJoinWorkerThread wt = w.owner;
1593	<	while (ctl == c) {
1594	<	long startTime = System.nanoTime();
1771	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1772	>	if (w != null && w.eventCount < 0 &&
1773	>	!tryTerminate(false, false) && (int)prevCtl != 0 &&
1774	>	ctl == currentCtl) {
1775	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1776	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1777	>	long deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP;
1778	>	Thread wt = Thread.currentThread();
1779	>	while (ctl == currentCtl) {
1780		Thread.interrupted(); // timed variant of version in scan()
1781		U.putObject(wt, PARKBLOCKER, this);
1782		w.parker = wt;
1783	<	if (ctl == c)
1784	<	U.park(false, SHRINK_RATE);
1783	>	if (ctl == currentCtl)
1784	>	U.park(false, parkTime);
1785		w.parker = null;
1786		U.putObject(wt, PARKBLOCKER, null);
1787	<	if (ctl != c)
1787	>	if (ctl != currentCtl)
1788		break;
1789	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1790	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1791	<	w.eventCount = (ec + E_SEQ) \| E_MASK;
1792	<	w.runState = -1; // shrink
1789	>	if (deadline - System.nanoTime() <= 0L &&
1790	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1791	>	w.eventCount = (w.eventCount + E_SEQ) \| E_MASK;
1792	>	w.hint = -1;
1793	>	w.qlock = -1; // shrink
1794		break;
1795		}
1796		}
#	Line 1612 \| Line 1798 \| public class ForkJoinPool extends Abstra
1798		}
1799
1800		/**
1801	+	* Scans through queues looking for work while joining a task; if
1802	+	* any present, signals. May return early if more signalling is
1803	+	* detectably unneeded.
1804	+	*
1805	+	* @param task return early if done
1806	+	* @param origin an index to start scan
1807	+	*/
1808	+	private void helpSignal(ForkJoinTask<?> task, int origin) {
1809	+	WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s;
1810	+	if (task != null && task.status >= 0 &&
1811	+	(u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 &&
1812	+	(ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1813	+	outer: for (int k = origin, j = m; j >= 0; --j) {
1814	+	WorkQueue q = ws[k++ & m];
1815	+	for (int n = m;;) { // limit to at most m signals
1816	+	if (task.status < 0)
1817	+	break outer;
1818	+	if (q == null \|\|
1819	+	((s = -q.base + q.top) <= n && (n = s) <= 0))
1820	+	break;
1821	+	if ((u = (int)((c = ctl) >>> 32)) >= 0 \|\|
1822	+	(e = (int)c) <= 0 \|\| m < (i = e & SMASK) \|\|
1823	+	(w = ws[i]) == null)
1824	+	break outer;
1825	+	long nc = (((long)(w.nextWait & E_MASK)) \|
1826	+	((long)(u + UAC_UNIT) << 32));
1827	+	if (w.eventCount != (e \| INT_SIGN))
1828	+	break outer;
1829	+	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1830	+	w.eventCount = (e + E_SEQ) & E_MASK;
1831	+	if ((p = w.parker) != null)
1832	+	U.unpark(p);
1833	+	if (--n <= 0)
1834	+	break;
1835	+	}
1836	+	}
1837	+	}
1838	+	}
1839	+	}
1840	+
1841	+	/**
1842		* Tries to locate and execute tasks for a stealer of the given
1843		* task, or in turn one of its stealers, Traces currentSteal ->
1844		* currentJoin links looking for a thread working on a descendant
#	Line 1622 \| Line 1849 \| public class ForkJoinPool extends Abstra
1849		* leaves hints in workers to speed up subsequent calls. The
1850		* implementation is very branchy to cope with potential
1851		* inconsistencies or loops encountering chains that are stale,
1852	<	* unknown, or of length greater than MAX_HELP_DEPTH links. All
1626	<	* of these cases are dealt with by just retrying by caller.
1852	>	* unknown, or so long that they are likely cyclic.
1853		*
1854		* @param joiner the joining worker
1855		* @param task the task to join
1856	<	* @return true if found or ran a task (and so is immediately retryable)
1856	>	* @return 0 if no progress can be made, negative if task
1857	>	* known complete, else positive
1858		*/
1859	<	final boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1860	<	ForkJoinTask<?> subtask; // current target
1861	<	boolean progress = false;
1862	<	int depth = 0; // current chain depth
1863	<	int m = runState & SMASK;
1864	<	WorkQueue[] ws = workQueues;
1865	<
1866	<	if (ws != null && ws.length > m && (subtask = task).status >= 0) {
1867	<	outer:for (WorkQueue j = joiner;;) {
1868	<	// Try to find the stealer of subtask, by first using hint
1869	<	WorkQueue stealer = null;
1870	<	WorkQueue v = ws[j.stealHint & m];
1871	<	if (v != null && v.currentSteal == subtask)
1872	<	stealer = v;
1873	<	else {
1874	<	for (int i = 1; i <= m; i += 2) {
1875	<	if ((v = ws[i]) != null && v.currentSteal == subtask) {
1876	<	stealer = v;
1877	<	j.stealHint = i; // save hint
1878	<	break;
1859	>	private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1860	>	int stat = 0, steps = 0; // bound to avoid cycles
1861	>	if (joiner != null && task != null) { // hoist null checks
1862	>	restart: for (;;) {
1863	>	ForkJoinTask<?> subtask = task; // current target
1864	>	for (WorkQueue j = joiner, v;;) { // v is stealer of subtask
1865	>	WorkQueue[] ws; int m, s, h;
1866	>	if ((s = task.status) < 0) {
1867	>	stat = s;
1868	>	break restart;
1869	>	}
1870	>	if ((ws = workQueues) == null \|\| (m = ws.length - 1) <= 0)
1871	>	break restart; // shutting down
1872	>	if ((v = ws[h = (j.hint \| 1) & m]) == null \|\|
1873	>	v.currentSteal != subtask) {
1874	>	for (int origin = h;;) { // find stealer
1875	>	if (((h = (h + 2) & m) & 15) == 1 &&
1876	>	(subtask.status < 0 \|\| j.currentJoin != subtask))
1877	>	continue restart; // occasional staleness check
1878	>	if ((v = ws[h]) != null &&
1879	>	v.currentSteal == subtask) {
1880	>	j.hint = h; // save hint
1881	>	break;
1882	>	}
1883	>	if (h == origin)
1884	>	break restart; // cannot find stealer
1885		}
1886		}
1887	<	if (stealer == null)
1887	>	for (;;) { // help stealer or descend to its stealer
1888	>	ForkJoinTask[] a; int b;
1889	>	if (subtask.status < 0) // surround probes with
1890	>	continue restart; // consistency checks
1891	>	if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1892	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1893	>	ForkJoinTask<?> t =
1894	>	(ForkJoinTask<?>)U.getObjectVolatile(a, i);
1895	>	if (subtask.status < 0 \|\| j.currentJoin != subtask \|\|
1896	>	v.currentSteal != subtask)
1897	>	continue restart; // stale
1898	>	stat = 1; // apparent progress
1899	>	if (t != null && v.base == b &&
1900	>	U.compareAndSwapObject(a, i, t, null)) {
1901	>	v.base = b + 1; // help stealer
1902	>	joiner.runSubtask(t);
1903	>	}
1904	>	else if (v.base == b && ++steps == MAX_HELP)
1905	>	break restart; // v apparently stalled
1906	>	}
1907	>	else { // empty -- try to descend
1908	>	ForkJoinTask<?> next = v.currentJoin;
1909	>	if (subtask.status < 0 \|\| j.currentJoin != subtask \|\|
1910	>	v.currentSteal != subtask)
1911	>	continue restart; // stale
1912	>	else if (next == null \|\| ++steps == MAX_HELP)
1913	>	break restart; // dead-end or maybe cyclic
1914	>	else {
1915	>	subtask = next;
1916	>	j = v;
1917	>	break;
1918	>	}
1919	>	}
1920	>	}
1921	>	}
1922	>	}
1923	>	}
1924	>	return stat;
1925	>	}
1926	>
1927	>	/**
1928	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1929	>	* and run tasks within the target's computation.
1930	>	*
1931	>	* @param task the task to join
1932	>	* @param mode if shared, exit upon completing any task
1933	>	* if all workers are active
1934	>	*/
1935	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1936	>	WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1937	>	if (task != null && (ws = workQueues) != null &&
1938	>	(m = ws.length - 1) >= 0) {
1939	>	for (int j = 1, origin = j;;) {
1940	>	if ((s = task.status) < 0)
1941	>	return s;
1942	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1943	>	origin = j;
1944	>	if (mode == SHARED_QUEUE &&
1945	>	((u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0))
1946		break;
1947		}
1948	+	else if ((j = (j + 2) & m) == origin)
1949	+	break;
1950	+	}
1951	+	}
1952	+	return 0;
1953	+	}
1954
1955	<	for (WorkQueue q = stealer;;) { // Try to help stealer
1956	<	ForkJoinTask<?> t; int b;
1957	<	if (task.status < 0)
1958	<	break outer;
1959	<	if ((b = q.base) - q.top < 0) {
1960	<	progress = true;
1961	<	if (subtask.status < 0)
1962	<	break outer; // stale
1963	<	if ((t = q.pollAt(b)) != null) {
1964	<	stealer.stealHint = joiner.poolIndex;
1965	<	joiner.runSubtask(t);
1955	>	/**
1956	>	* Tries to decrement active count (sometimes implicitly) and
1957	>	* possibly release or create a compensating worker in preparation
1958	>	* for blocking. Fails on contention or termination. Otherwise,
1959	>	* adds a new thread if no idle workers are available and pool
1960	>	* may become starved.
1961	>	*/
1962	>	final boolean tryCompensate() {
1963	>	int pc = config & SMASK, e, i, tc; long c;
1964	>	WorkQueue[] ws; WorkQueue w; Thread p;
1965	>	if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) {
1966	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1967	>	(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1968	>	long nc = ((long)(w.nextWait & E_MASK) \|
1969	>	(c & (AC_MASK\|TC_MASK)));
1970	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1971	>	w.eventCount = (e + E_SEQ) & E_MASK;
1972	>	if ((p = w.parker) != null)
1973	>	U.unpark(p);
1974	>	return true; // replace with idle worker
1975	>	}
1976	>	}
1977	>	else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1978	>	(int)(c >> AC_SHIFT) + pc > 1) {
1979	>	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1980	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1981	>	return true; // no compensation
1982	>	}
1983	>	else if (tc + pc < MAX_CAP) {
1984	>	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1985	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1986	>	ForkJoinWorkerThreadFactory fac;
1987	>	Throwable ex = null;
1988	>	ForkJoinWorkerThread wt = null;
1989	>	try {
1990	>	if ((fac = factory) != null &&
1991	>	(wt = fac.newThread(this)) != null) {
1992	>	wt.start();
1993	>	return true;
1994		}
1995	+	} catch (Throwable rex) {
1996	+	ex = rex;
1997		}
1998	<	else { // empty - try to descend to find stealer's stealer
1672	<	ForkJoinTask<?> next = stealer.currentJoin;
1673	<	if (++depth == MAX_HELP_DEPTH \|\| subtask.status < 0 \|\|
1674	<	next == null \|\| next == subtask)
1675	<	break outer; // max depth, stale, dead-end, cyclic
1676	<	subtask = next;
1677	<	j = stealer;
1678	<	break;
1679	<	}
1998	>	deregisterWorker(wt, ex); // clean up and return false
1999		}
2000		}
2001		}
2002	<	return progress;
2002	>	return false;
2003		}
2004
2005		/**
2006	<	* If task is at base of some steal queue, steals and executes it.
2006	>	* Helps and/or blocks until the given task is done.
2007		*
2008		* @param joiner the joining worker
2009		* @param task the task
2010	+	* @return task status on exit
2011		*/
2012	<	final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
2013	<	WorkQueue[] ws;
2014	<	int m = runState & SMASK;
2015	<	if ((ws = workQueues) != null && ws.length > m) {
2016	<	for (int j = 1; j <= m && task.status >= 0; j += 2) {
2017	<	WorkQueue q = ws[j];
2018	<	if (q != null && q.pollFor(task)) {
2019	<	joiner.runSubtask(task);
2020	<	break;
2012	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
2013	>	int s = 0;
2014	>	if (joiner != null && task != null && (s = task.status) >= 0) {
2015	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2016	>	joiner.currentJoin = task;
2017	>	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2018	>	joiner.tryRemoveAndExec(task)); // process local tasks
2019	>	if (s >= 0 && (s = task.status) >= 0) {
2020	>	helpSignal(task, joiner.poolIndex);
2021	>	if ((s = task.status) >= 0 &&
2022	>	(task instanceof CountedCompleter))
2023	>	s = helpComplete(task, LIFO_QUEUE);
2024	>	}
2025	>	while (s >= 0 && (s = task.status) >= 0) {
2026	>	if ((!joiner.isEmpty() \|\| // try helping
2027	>	(s = tryHelpStealer(joiner, task)) == 0) &&
2028	>	(s = task.status) >= 0) {
2029	>	helpSignal(task, joiner.poolIndex);
2030	>	if ((s = task.status) >= 0 && tryCompensate()) {
2031	>	if (task.trySetSignal() && (s = task.status) >= 0) {
2032	>	synchronized (task) {
2033	>	if (task.status >= 0) {
2034	>	try { // see ForkJoinTask
2035	>	task.wait(); // for explanation
2036	>	} catch (InterruptedException ie) {
2037	>	}
2038	>	}
2039	>	else
2040	>	task.notifyAll();
2041	>	}
2042	>	}
2043	>	long c; // re-activate
2044	>	do {} while (!U.compareAndSwapLong
2045	>	(this, CTL, c = ctl, c + AC_UNIT));
2046	>	}
2047		}
2048		}
2049	+	joiner.currentJoin = prevJoin;
2050		}
2051	+	return s;
2052		}
2053
2054		/**
2055	<	* Returns a non-empty steal queue, if one is found during a random,
2056	<	* then cyclic scan, else null. This method must be retried by
2055	>	* Stripped-down variant of awaitJoin used by timed joins. Tries
2056	>	* to help join only while there is continuous progress. (Caller
2057	>	* will then enter a timed wait.)
2058	>	*
2059	>	* @param joiner the joining worker
2060	>	* @param task the task
2061	>	*/
2062	>	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2063	>	int s;
2064	>	if (joiner != null && task != null && (s = task.status) >= 0) {
2065	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2066	>	joiner.currentJoin = task;
2067	>	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2068	>	joiner.tryRemoveAndExec(task));
2069	>	if (s >= 0 && (s = task.status) >= 0) {
2070	>	helpSignal(task, joiner.poolIndex);
2071	>	if ((s = task.status) >= 0 &&
2072	>	(task instanceof CountedCompleter))
2073	>	s = helpComplete(task, LIFO_QUEUE);
2074	>	}
2075	>	if (s >= 0 && joiner.isEmpty()) {
2076	>	do {} while (task.status >= 0 &&
2077	>	tryHelpStealer(joiner, task) > 0);
2078	>	}
2079	>	joiner.currentJoin = prevJoin;
2080	>	}
2081	>	}
2082	>
2083	>	/**
2084	>	* Returns a (probably) non-empty steal queue, if one is found
2085	>	* during a scan, else null. This method must be retried by
2086		* caller if, by the time it tries to use the queue, it is empty.
2087	+	* @param r a (random) seed for scanning
2088		*/
2089	<	private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
2090	<	int r = w.seed; // Same idea as scan(), but ignoring submissions
2091	<	for (WorkQueue[] ws;;) {
2092	<	int m = runState & SMASK;
2093	<	if ((ws = workQueues) == null)
2094	<	return null;
2095	<	if (ws.length > m) {
1718	<	WorkQueue q;
1719	<	for (int k = 0, j = -1 - m;; ++j) {
1720	<	if (j < 0) {
1721	<	r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5;
1722	<	}
1723	<	else
1724	<	k += 7;
1725	<	if ((q = ws[(k \| 1) & m]) != null && q.base - q.top < 0) {
1726	<	w.seed = r;
2089	>	private WorkQueue findNonEmptyStealQueue(int r) {
2090	>	for (;;) {
2091	>	int ps = plock, m; WorkQueue[] ws; WorkQueue q;
2092	>	if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) {
2093	>	for (int j = (m + 1) << 2; j >= 0; --j) {
2094	>	if ((q = ws[(((r + j) << 1) \| 1) & m]) != null &&
2095	>	q.base - q.top < 0)
2096		return q;
1728	–	}
1729	–	else if (j - m > m)
1730	–	return null;
2097		}
2098		}
2099	+	if (plock == ps)
2100	+	return null;
2101		}
2102		}
2103
#	Line 1741 \| Line 2109 \| public class ForkJoinPool extends Abstra
2109		*/
2110		final void helpQuiescePool(WorkQueue w) {
2111		for (boolean active = true;;) {
2112	<	w.runLocalTasks(); // exhaust local queue
2113	<	WorkQueue q = findNonEmptyStealQueue(w);
2114	<	if (q != null) {
2115	<	ForkJoinTask<?> t;
2112	>	long c; WorkQueue q; ForkJoinTask<?> t; int b;
2113	>	while ((t = w.nextLocalTask()) != null) {
2114	>	if (w.base - w.top < 0)
2115	>	signalWork(w);
2116	>	t.doExec();
2117	>	}
2118	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) != null) {
2119		if (!active) { // re-establish active count
1749	–	long c;
2120		active = true;
2121		do {} while (!U.compareAndSwapLong
2122		(this, CTL, c = ctl, c + AC_UNIT));
2123		}
2124	<	if ((t = q.poll()) != null)
2124	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2125	>	if (q.base - q.top < 0)
2126	>	signalWork(q);
2127		w.runSubtask(t);
2128	+	}
2129		}
2130	<	else {
2131	<	long c;
2132	<	if (active) { // decrement active count without queuing
2130	>	else if (active) { // decrement active count without queuing
2131	>	long nc = (c = ctl) - AC_UNIT;
2132	>	if ((int)(nc >> AC_SHIFT) + (config & SMASK) == 0)
2133	>	return; // bypass decrement-then-increment
2134	>	if (U.compareAndSwapLong(this, CTL, c, nc))
2135		active = false;
1761	–	do {} while (!U.compareAndSwapLong
1762	–	(this, CTL, c = ctl, c -= AC_UNIT));
1763	–	}
1764	–	else
1765	–	c = ctl; // re-increment on exit
1766	–	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
1767	–	do {} while (!U.compareAndSwapLong
1768	–	(this, CTL, c = ctl, c + AC_UNIT));
1769	–	break;
1770	–	}
2136		}
2137	+	else if ((int)((c = ctl) >> AC_SHIFT) + (config & SMASK) == 0 &&
2138	+	U.compareAndSwapLong(this, CTL, c, c + AC_UNIT))
2139	+	return;
2140		}
2141		}
2142
#	Line 1779 \| Line 2147 \| public class ForkJoinPool extends Abstra
2147		*/
2148		final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
2149		for (ForkJoinTask<?> t;;) {
2150	<	WorkQueue q;
2150	>	WorkQueue q; int b;
2151		if ((t = w.nextLocalTask()) != null)
2152		return t;
2153	<	if ((q = findNonEmptyStealQueue(w)) == null)
2153	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2154		return null;
2155	<	if ((t = q.poll()) != null)
2155	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2156	>	if (q.base - q.top < 0)
2157	>	signalWork(q);
2158		return t;
2159	+	}
2160		}
2161		}
2162
2163		/**
2164	<	* Returns the approximate (non-atomic) number of idle threads per
2165	<	* active thread to offset steal queue size for method
2166	<	* ForkJoinTask.getSurplusQueuedTaskCount().
2167	<	*/
2168	<	final int idlePerActive() {
2169	<	// Approximate at powers of two for small values, saturate past 4
2170	<	int p = parallelism;
2171	<	int a = p + (int)(ctl >> AC_SHIFT);
2172	<	return (a > (p >>>= 1) ? 0 :
2173	<	a > (p >>>= 1) ? 1 :
2174	<	a > (p >>>= 1) ? 2 :
2175	<	a > (p >>>= 1) ? 4 :
2176	<	8);
2164	>	* Returns a cheap heuristic guide for task partitioning when
2165	>	* programmers, frameworks, tools, or languages have little or no
2166	>	* idea about task granularity. In essence by offering this
2167	>	* method, we ask users only about tradeoffs in overhead vs
2168	>	* expected throughput and its variance, rather than how finely to
2169	>	* partition tasks.
2170	>	*
2171	>	* In a steady state strict (tree-structured) computation, each
2172	>	* thread makes available for stealing enough tasks for other
2173	>	* threads to remain active. Inductively, if all threads play by
2174	>	* the same rules, each thread should make available only a
2175	>	* constant number of tasks.
2176	>	*
2177	>	* The minimum useful constant is just 1. But using a value of 1
2178	>	* would require immediate replenishment upon each steal to
2179	>	* maintain enough tasks, which is infeasible. Further,
2180	>	* partitionings/granularities of offered tasks should minimize
2181	>	* steal rates, which in general means that threads nearer the top
2182	>	* of computation tree should generate more than those nearer the
2183	>	* bottom. In perfect steady state, each thread is at
2184	>	* approximately the same level of computation tree. However,
2185	>	* producing extra tasks amortizes the uncertainty of progress and
2186	>	* diffusion assumptions.
2187	>	*
2188	>	* So, users will want to use values larger, but not much larger
2189	>	* than 1 to both smooth over transient shortages and hedge
2190	>	* against uneven progress; as traded off against the cost of
2191	>	* extra task overhead. We leave the user to pick a threshold
2192	>	* value to compare with the results of this call to guide
2193	>	* decisions, but recommend values such as 3.
2194	>	*
2195	>	* When all threads are active, it is on average OK to estimate
2196	>	* surplus strictly locally. In steady-state, if one thread is
2197	>	* maintaining say 2 surplus tasks, then so are others. So we can
2198	>	* just use estimated queue length. However, this strategy alone
2199	>	* leads to serious mis-estimates in some non-steady-state
2200	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2201	>	* many of these by further considering the number of "idle"
2202	>	* threads, that are known to have zero queued tasks, so
2203	>	* compensate by a factor of (#idle/#active) threads.
2204	>	*
2205	>	* Note: The approximation of #busy workers as #active workers is
2206	>	* not very good under current signalling scheme, and should be
2207	>	* improved.
2208	>	*/
2209	>	static int getSurplusQueuedTaskCount() {
2210	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2211	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2212	>	int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK;
2213	>	int n = (q = wt.workQueue).top - q.base;
2214	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2215	>	return n - (a > (p >>>= 1) ? 0 :
2216	>	a > (p >>>= 1) ? 1 :
2217	>	a > (p >>>= 1) ? 2 :
2218	>	a > (p >>>= 1) ? 4 :
2219	>	8);
2220	>	}
2221	>	return 0;
2222		}
2223
2224		// Termination
#	Line 1822 \| Line 2238 \| public class ForkJoinPool extends Abstra
2238		* @return true if now terminating or terminated
2239		*/
2240		private boolean tryTerminate(boolean now, boolean enable) {
2241	<	Mutex lock = this.lock;
2241	>	int ps;
2242	>	if (this == common) // cannot shut down
2243	>	return false;
2244	>	if ((ps = plock) >= 0) { // enable by setting plock
2245	>	if (!enable)
2246	>	return false;
2247	>	if ((ps & PL_LOCK) != 0 \|\|
2248	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2249	>	ps = acquirePlock();
2250	>	int nps = ((ps + PL_LOCK) & ~SHUTDOWN) \| SHUTDOWN;
2251	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2252	>	releasePlock(nps);
2253	>	}
2254		for (long c;;) {
2255	<	if (((c = ctl) & STOP_BIT) != 0) { // already terminating
2256	<	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2257	<	lock.lock(); // don't need try/finally
2258	<	termination.signalAll(); // signal when 0 workers
2259	<	lock.unlock();
2255	>	if (((c = ctl) & STOP_BIT) != 0) { // already terminating
2256	>	if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) {
2257	>	synchronized (this) {
2258	>	notifyAll(); // signal when 0 workers
2259	>	}
2260		}
2261		return true;
2262		}
2263	<	if (runState >= 0) { // not yet enabled
2264	<	if (!enable)
2263	>	if (!now) { // check if idle & no tasks
2264	>	WorkQueue[] ws; WorkQueue w;
2265	>	if ((int)(c >> AC_SHIFT) != -(config & SMASK))
2266		return false;
2267	<	lock.lock();
2268	<	runState \|= SHUTDOWN;
2269	<	lock.unlock();
2270	<	}
2271	<	if (!now) { // check if idle & no tasks
2272	<	if ((int)(c >> AC_SHIFT) != -parallelism \|\|
2273	<	hasQueuedSubmissions())
2274	<	return false;
2275	<	// Check for unqueued inactive workers. One pass suffices.
2276	<	WorkQueue[] ws = workQueues; WorkQueue w;
1848	<	if (ws != null) {
1849	<	for (int i = 1; i < ws.length; i += 2) {
1850	<	if ((w = ws[i]) != null && w.eventCount >= 0)
1851	<	return false;
2267	>	if ((ws = workQueues) != null) {
2268	>	for (int i = 0; i < ws.length; ++i) {
2269	>	if ((w = ws[i]) != null) {
2270	>	if (!w.isEmpty()) { // signal unprocessed tasks
2271	>	signalWork(w);
2272	>	return false;
2273	>	}
2274	>	if ((i & 1) != 0 && w.eventCount >= 0)
2275	>	return false; // unqueued inactive worker
2276	>	}
2277		}
2278		}
2279		}
2280		if (U.compareAndSwapLong(this, CTL, c, c \| STOP_BIT)) {
2281		for (int pass = 0; pass < 3; ++pass) {
2282	<	WorkQueue[] ws = workQueues;
2283	<	if (ws != null) {
1859	<	WorkQueue w;
2282	>	WorkQueue[] ws; WorkQueue w; Thread wt;
2283	>	if ((ws = workQueues) != null) {
2284		int n = ws.length;
2285		for (int i = 0; i < n; ++i) {
2286		if ((w = ws[i]) != null) {
2287	<	w.runState = -1;
2287	>	w.qlock = -1;
2288		if (pass > 0) {
2289		w.cancelAll();
2290	<	if (pass > 1)
2291	<	w.interruptOwner();
2290	>	if (pass > 1 && (wt = w.owner) != null) {
2291	>	if (!wt.isInterrupted()) {
2292	>	try {
2293	>	wt.interrupt();
2294	>	} catch (Throwable ignore) {
2295	>	}
2296	>	}
2297	>	U.unpark(wt);
2298	>	}
2299		}
2300		}
2301		}
2302		// Wake up workers parked on event queue
2303		int i, e; long cc; Thread p;
2304		while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2305	<	(i = e & SMASK) < n &&
2305	>	(i = e & SMASK) < n && i >= 0 &&
2306		(w = ws[i]) != null) {
2307		long nc = ((long)(w.nextWait & E_MASK) \|
2308		((cc + AC_UNIT) & AC_MASK) \|
#	Line 1879 \| Line 2310 \| public class ForkJoinPool extends Abstra
2310		if (w.eventCount == (e \| INT_SIGN) &&
2311		U.compareAndSwapLong(this, CTL, cc, nc)) {
2312		w.eventCount = (e + E_SEQ) & E_MASK;
2313	<	w.runState = -1;
2313	>	w.qlock = -1;
2314		if ((p = w.parker) != null)
2315		U.unpark(p);
2316		}
#	Line 1890 \| Line 2321 \| public class ForkJoinPool extends Abstra
2321		}
2322		}
2323
2324	+	// external operations on common pool
2325	+
2326	+	/**
2327	+	* Returns common pool queue for a thread that has submitted at
2328	+	* least one task.
2329	+	*/
2330	+	static WorkQueue commonSubmitterQueue() {
2331	+	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2332	+	return ((z = submitters.get()) != null &&
2333	+	(p = common) != null &&
2334	+	(ws = p.workQueues) != null &&
2335	+	(m = ws.length - 1) >= 0) ?
2336	+	ws[m & z.seed & SQMASK] : null;
2337	+	}
2338	+
2339	+	/**
2340	+	* Tries to pop the given task from submitter's queue in common pool.
2341	+	*/
2342	+	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2343	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2344	+	ForkJoinTask<?>[] a; int m, s;
2345	+	if (t != null &&
2346	+	(z = submitters.get()) != null &&
2347	+	(p = common) != null &&
2348	+	(ws = p.workQueues) != null &&
2349	+	(m = ws.length - 1) >= 0 &&
2350	+	(q = ws[m & z.seed & SQMASK]) != null &&
2351	+	(s = q.top) != q.base &&
2352	+	(a = q.array) != null) {
2353	+	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2354	+	if (U.getObject(a, j) == t &&
2355	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2356	+	if (q.array == a && q.top == s && // recheck
2357	+	U.compareAndSwapObject(a, j, t, null)) {
2358	+	q.top = s - 1;
2359	+	q.qlock = 0;
2360	+	return true;
2361	+	}
2362	+	q.qlock = 0;
2363	+	}
2364	+	}
2365	+	return false;
2366	+	}
2367	+
2368	+	/**
2369	+	* Tries to pop and run local tasks within the same computation
2370	+	* as the given root. On failure, tries to help complete from
2371	+	* other queues via helpComplete.
2372	+	*/
2373	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2374	+	ForkJoinTask<?>[] a; int m;
2375	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2376	+	root != null && root.status >= 0) {
2377	+	for (;;) {
2378	+	int s, u; Object o; CountedCompleter<?> task = null;
2379	+	if ((s = q.top) - q.base > 0) {
2380	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2381	+	if ((o = U.getObject(a, j)) != null &&
2382	+	(o instanceof CountedCompleter)) {
2383	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2384	+	do {
2385	+	if (r == root) {
2386	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2387	+	if (q.array == a && q.top == s &&
2388	+	U.compareAndSwapObject(a, j, t, null)) {
2389	+	q.top = s - 1;
2390	+	task = t;
2391	+	}
2392	+	q.qlock = 0;
2393	+	}
2394	+	break;
2395	+	}
2396	+	} while ((r = r.completer) != null);
2397	+	}
2398	+	}
2399	+	if (task != null)
2400	+	task.doExec();
2401	+	if (root.status < 0 \|\|
2402	+	(u = (int)(ctl >>> 32)) >= 0 \|\| (u >> UAC_SHIFT) >= 0)
2403	+	break;
2404	+	if (task == null) {
2405	+	helpSignal(root, q.poolIndex);
2406	+	if (root.status >= 0)
2407	+	helpComplete(root, SHARED_QUEUE);
2408	+	break;
2409	+	}
2410	+	}
2411	+	}
2412	+	}
2413	+
2414	+	/**
2415	+	* Tries to help execute or signal availability of the given task
2416	+	* from submitter's queue in common pool.
2417	+	*/
2418	+	static void externalHelpJoin(ForkJoinTask<?> t) {
2419	+	// Some hard-to-avoid overlap with tryExternalUnpush
2420	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2421	+	ForkJoinTask<?>[] a; int m, s, n;
2422	+	if (t != null &&
2423	+	(z = submitters.get()) != null &&
2424	+	(p = common) != null &&
2425	+	(ws = p.workQueues) != null &&
2426	+	(m = ws.length - 1) >= 0 &&
2427	+	(q = ws[m & z.seed & SQMASK]) != null &&
2428	+	(a = q.array) != null) {
2429	+	int am = a.length - 1;
2430	+	if ((s = q.top) != q.base) {
2431	+	long j = ((am & (s - 1)) << ASHIFT) + ABASE;
2432	+	if (U.getObject(a, j) == t &&
2433	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2434	+	if (q.array == a && q.top == s &&
2435	+	U.compareAndSwapObject(a, j, t, null)) {
2436	+	q.top = s - 1;
2437	+	q.qlock = 0;
2438	+	t.doExec();
2439	+	}
2440	+	else
2441	+	q.qlock = 0;
2442	+	}
2443	+	}
2444	+	if (t.status >= 0) {
2445	+	if (t instanceof CountedCompleter)
2446	+	p.externalHelpComplete(q, t);
2447	+	else
2448	+	p.helpSignal(t, q.poolIndex);
2449	+	}
2450	+	}
2451	+	}
2452	+
2453		// Exported methods
2454
2455		// Constructors
#	Line 1906 \| Line 2466 \| public class ForkJoinPool extends Abstra
2466		* java.lang.RuntimePermission}{@code ("modifyThread")}
2467		*/
2468		public ForkJoinPool() {
2469	<	this(Runtime.getRuntime().availableProcessors(),
2469	>	this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
2470		defaultForkJoinWorkerThreadFactory, null, false);
2471		}
2472
#	Line 1959 \| Line 2519 \| public class ForkJoinPool extends Abstra
2519		checkPermission();
2520		if (factory == null)
2521		throw new NullPointerException();
2522	<	if (parallelism <= 0 \|\| parallelism > POOL_MAX)
2522	>	if (parallelism <= 0 \|\| parallelism > MAX_CAP)
2523		throw new IllegalArgumentException();
1964	–	this.parallelism = parallelism;
2524		this.factory = factory;
2525		this.ueh = handler;
2526	<	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
1968	<	this.growHints = 1;
2526	>	this.config = parallelism \| (asyncMode ? (FIFO_QUEUE << 16) : 0);
2527		long np = (long)(-parallelism); // offset ctl counts
2528		this.ctl = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
2529	<	// initialize workQueues array with room for 2*parallelism if possible
1972	<	int n = parallelism << 1;
1973	<	if (n >= POOL_MAX)
1974	<	n = POOL_MAX;
1975	<	else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
1976	<	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8;
1977	<	}
1978	<	this.workQueues = new WorkQueue[(n + 1) << 1]; // #slots = 2 * #workers
1979	<	this.termination = (this.lock = new Mutex()).newCondition();
1980	<	this.stealCount = new AtomicLong();
1981	<	this.nextWorkerNumber = new AtomicInteger();
2529	>	int pn = nextPoolId();
2530		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2531	<	sb.append(poolNumberGenerator.incrementAndGet());
2531	>	sb.append(Integer.toString(pn));
2532		sb.append("-worker-");
2533		this.workerNamePrefix = sb.toString();
2534		}
2535
2536	+	/**
2537	+	* Constructor for common pool, suitable only for static initialization.
2538	+	* Basically the same as above, but uses smallest possible initial footprint.
2539	+	*/
2540	+	ForkJoinPool(int parallelism, long ctl,
2541	+	ForkJoinWorkerThreadFactory factory,
2542	+	Thread.UncaughtExceptionHandler handler) {
2543	+	this.config = parallelism;
2544	+	this.ctl = ctl;
2545	+	this.factory = factory;
2546	+	this.ueh = handler;
2547	+	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2548	+	}
2549	+
2550	+	/**
2551	+	* Returns the common pool instance. This pool is statically
2552	+	* constructed; its run state is unaffected by attempts to {@link
2553	+	* #shutdown} or {@link #shutdownNow}. However this pool and any
2554	+	* ongoing processing are automatically terminated upon program
2555	+	* {@link System#exit}. Any program that relies on asynchronous
2556	+	* task processing to complete before program termination should
2557	+	* invoke {@code commonPool().}{@link #awaitQuiescence}, before
2558	+	* exit.
2559	+	*
2560	+	* @return the common pool instance
2561	+	* @since 1.8
2562	+	*/
2563	+	public static ForkJoinPool commonPool() {
2564	+	// assert common != null : "static init error";
2565	+	return common;
2566	+	}
2567	+
2568		// Execution methods
2569
2570		/**
#	Line 2004 \| Line 2584 \| public class ForkJoinPool extends Abstra
2584		* scheduled for execution
2585		*/
2586		public <T> T invoke(ForkJoinTask<T> task) {
2587	<	doSubmit(task);
2587	>	if (task == null)
2588	>	throw new NullPointerException();
2589	>	externalPush(task);
2590		return task.join();
2591		}
2592
#	Line 2017 \| Line 2599 \| public class ForkJoinPool extends Abstra
2599		* scheduled for execution
2600		*/
2601		public void execute(ForkJoinTask<?> task) {
2602	<	doSubmit(task);
2602	>	if (task == null)
2603	>	throw new NullPointerException();
2604	>	externalPush(task);
2605		}
2606
2607		// AbstractExecutorService methods
#	Line 2034 \| Line 2618 \| public class ForkJoinPool extends Abstra
2618		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2619		job = (ForkJoinTask<?>) task;
2620		else
2621	<	job = ForkJoinTask.adapt(task, null);
2622	<	doSubmit(job);
2621	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2622	>	externalPush(job);
2623		}
2624
2625		/**
#	Line 2048 \| Line 2632 \| public class ForkJoinPool extends Abstra
2632		* scheduled for execution
2633		*/
2634		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2635	<	doSubmit(task);
2635	>	if (task == null)
2636	>	throw new NullPointerException();
2637	>	externalPush(task);
2638		return task;
2639		}
2640
#	Line 2058 \| Line 2644 \| public class ForkJoinPool extends Abstra
2644		* scheduled for execution
2645		*/
2646		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2647	<	if (task == null)
2648	<	throw new NullPointerException();
2063	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task);
2064	<	doSubmit(job);
2647	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2648	>	externalPush(job);
2649		return job;
2650		}
2651
#	Line 2071 \| Line 2655 \| public class ForkJoinPool extends Abstra
2655		* scheduled for execution
2656		*/
2657		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2658	<	if (task == null)
2659	<	throw new NullPointerException();
2076	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
2077	<	doSubmit(job);
2658	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2659	>	externalPush(job);
2660		return job;
2661		}
2662
#	Line 2090 \| Line 2672 \| public class ForkJoinPool extends Abstra
2672		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2673		job = (ForkJoinTask<?>) task;
2674		else
2675	<	job = ForkJoinTask.adapt(task, null);
2676	<	doSubmit(job);
2675	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2676	>	externalPush(job);
2677		return job;
2678		}
2679
#	Line 2103 \| Line 2685 \| public class ForkJoinPool extends Abstra
2685		// In previous versions of this class, this method constructed
2686		// a task to run ForkJoinTask.invokeAll, but now external
2687		// invocation of multiple tasks is at least as efficient.
2688	<	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2107	<	// Workaround needed because method wasn't declared with
2108	<	// wildcards in return type but should have been.
2109	<	@SuppressWarnings({"unchecked", "rawtypes"})
2110	<	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2688	>	ArrayList<Future<T>> futures = new ArrayList<Future<T>>(tasks.size());
2689
2690		boolean done = false;
2691		try {
2692		for (Callable<T> t : tasks) {
2693	<	ForkJoinTask<T> f = ForkJoinTask.adapt(t);
2694	<	doSubmit(f);
2695	<	fs.add(f);
2693	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2694	>	futures.add(f);
2695	>	externalPush(f);
2696		}
2697	<	for (ForkJoinTask<T> f : fs)
2698	<	f.quietlyJoin();
2697	>	for (int i = 0, size = futures.size(); i < size; i++)
2698	>	((ForkJoinTask<?>)futures.get(i)).quietlyJoin();
2699		done = true;
2700		return futures;
2701		} finally {
2702		if (!done)
2703	<	for (ForkJoinTask<T> f : fs)
2704	<	f.cancel(false);
2703	>	for (int i = 0, size = futures.size(); i < size; i++)
2704	>	futures.get(i).cancel(false);
2705		}
2706		}
2707
#	Line 2152 \| Line 2730 \| public class ForkJoinPool extends Abstra
2730		* @return the targeted parallelism level of this pool
2731		*/
2732		public int getParallelism() {
2733	<	return parallelism;
2733	>	return config & SMASK;
2734	>	}
2735	>
2736	>	/**
2737	>	* Returns the targeted parallelism level of the common pool.
2738	>	*
2739	>	* @return the targeted parallelism level of the common pool
2740	>	* @since 1.8
2741	>	*/
2742	>	public static int getCommonPoolParallelism() {
2743	>	return commonParallelism;
2744		}
2745
2746		/**
#	Line 2164 \| Line 2752 \| public class ForkJoinPool extends Abstra
2752		* @return the number of worker threads
2753		*/
2754		public int getPoolSize() {
2755	<	return parallelism + (short)(ctl >>> TC_SHIFT);
2755	>	return (config & SMASK) + (short)(ctl >>> TC_SHIFT);
2756		}
2757
2758		/**
#	Line 2174 \| Line 2762 \| public class ForkJoinPool extends Abstra
2762		* @return {@code true} if this pool uses async mode
2763		*/
2764		public boolean getAsyncMode() {
2765	<	return localMode != 0;
2765	>	return (config >>> 16) == FIFO_QUEUE;
2766		}
2767
2768		/**
#	Line 2205 \| Line 2793 \| public class ForkJoinPool extends Abstra
2793		* @return the number of active threads
2794		*/
2795		public int getActiveThreadCount() {
2796	<	int r = parallelism + (int)(ctl >> AC_SHIFT);
2796	>	int r = (config & SMASK) + (int)(ctl >> AC_SHIFT);
2797		return (r <= 0) ? 0 : r; // suppress momentarily negative values
2798		}
2799
#	Line 2221 \| Line 2809 \| public class ForkJoinPool extends Abstra
2809		* @return {@code true} if all threads are currently idle
2810		*/
2811		public boolean isQuiescent() {
2812	<	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2812	>	return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0;
2813		}
2814
2815		/**
#	Line 2236 \| Line 2824 \| public class ForkJoinPool extends Abstra
2824		* @return the number of steals
2825		*/
2826		public long getStealCount() {
2827	<	long count = stealCount.get();
2827	>	long count = stealCount;
2828		WorkQueue[] ws; WorkQueue w;
2829		if ((ws = workQueues) != null) {
2830		for (int i = 1; i < ws.length; i += 2) {
2831		if ((w = ws[i]) != null)
2832	<	count += w.totalSteals;
2832	>	count += w.nsteals;
2833		}
2834		}
2835		return count;
#	Line 2298 \| Line 2886 \| public class ForkJoinPool extends Abstra
2886		WorkQueue[] ws; WorkQueue w;
2887		if ((ws = workQueues) != null) {
2888		for (int i = 0; i < ws.length; i += 2) {
2889	<	if ((w = ws[i]) != null && w.queueSize() != 0)
2889	>	if ((w = ws[i]) != null && !w.isEmpty())
2890		return true;
2891		}
2892		}
#	Line 2366 \| Line 2954 \| public class ForkJoinPool extends Abstra
2954		public String toString() {
2955		// Use a single pass through workQueues to collect counts
2956		long qt = 0L, qs = 0L; int rc = 0;
2957	<	long st = stealCount.get();
2957	>	long st = stealCount;
2958		long c = ctl;
2959		WorkQueue[] ws; WorkQueue w;
2960		if ((ws = workQueues) != null) {
#	Line 2377 \| Line 2965 \| public class ForkJoinPool extends Abstra
2965		qs += size;
2966		else {
2967		qt += size;
2968	<	st += w.totalSteals;
2968	>	st += w.nsteals;
2969		if (w.isApparentlyUnblocked())
2970		++rc;
2971		}
2972		}
2973		}
2974		}
2975	<	int pc = parallelism;
2975	>	int pc = (config & SMASK);
2976		int tc = pc + (short)(c >>> TC_SHIFT);
2977		int ac = pc + (int)(c >> AC_SHIFT);
2978		if (ac < 0) // ignore transient negative
#	Line 2393 \| Line 2981 \| public class ForkJoinPool extends Abstra
2981		if ((c & STOP_BIT) != 0)
2982		level = (tc == 0) ? "Terminated" : "Terminating";
2983		else
2984	<	level = runState < 0 ? "Shutting down" : "Running";
2984	>	level = plock < 0 ? "Shutting down" : "Running";
2985		return super.toString() +
2986		"[" + level +
2987		", parallelism = " + pc +
#	Line 2407 \| Line 2995 \| public class ForkJoinPool extends Abstra
2995		}
2996
2997		/**
2998	<	* Initiates an orderly shutdown in which previously submitted
2999	<	* tasks are executed, but no new tasks will be accepted.
3000	<	* Invocation has no additional effect if already shut down.
3001	<	* Tasks that are in the process of being submitted concurrently
3002	<	* during the course of this method may or may not be rejected.
2998	>	* Possibly initiates an orderly shutdown in which previously
2999	>	* submitted tasks are executed, but no new tasks will be
3000	>	* accepted. Invocation has no effect on execution state if this
3001	>	* is the {@link #commonPool()}, and no additional effect if
3002	>	* already shut down. Tasks that are in the process of being
3003	>	* submitted concurrently during the course of this method may or
3004	>	* may not be rejected.
3005		*
3006		* @throws SecurityException if a security manager exists and
3007		* the caller is not permitted to modify threads
#	Line 2424 \| Line 3014 \| public class ForkJoinPool extends Abstra
3014		}
3015
3016		/**
3017	<	* Attempts to cancel and/or stop all tasks, and reject all
3018	<	* subsequently submitted tasks. Tasks that are in the process of
3019	<	* being submitted or executed concurrently during the course of
3020	<	* this method may or may not be rejected. This method cancels
3021	<	* both existing and unexecuted tasks, in order to permit
3022	<	* termination in the presence of task dependencies. So the method
3023	<	* always returns an empty list (unlike the case for some other
3024	<	* Executors).
3017	>	* Possibly attempts to cancel and/or stop all tasks, and reject
3018	>	* all subsequently submitted tasks. Invocation has no effect on
3019	>	* execution state if this is the {@link #commonPool()}, and no
3020	>	* additional effect if already shut down. Otherwise, tasks that
3021	>	* are in the process of being submitted or executed concurrently
3022	>	* during the course of this method may or may not be
3023	>	* rejected. This method cancels both existing and unexecuted
3024	>	* tasks, in order to permit termination in the presence of task
3025	>	* dependencies. So the method always returns an empty list
3026	>	* (unlike the case for some other Executors).
3027		*
3028		* @return an empty list
3029		* @throws SecurityException if a security manager exists and
#	Line 2453 \| Line 3045 \| public class ForkJoinPool extends Abstra
3045		public boolean isTerminated() {
3046		long c = ctl;
3047		return ((c & STOP_BIT) != 0L &&
3048	<	(short)(c >>> TC_SHIFT) == -parallelism);
3048	>	(short)(c >>> TC_SHIFT) == -(config & SMASK));
3049		}
3050
3051		/**
#	Line 2461 \| Line 3053 \| public class ForkJoinPool extends Abstra
3053		* commenced but not yet completed. This method may be useful for
3054		* debugging. A return of {@code true} reported a sufficient
3055		* period after shutdown may indicate that submitted tasks have
3056	<	* ignored or suppressed interruption, or are waiting for IO,
3056	>	* ignored or suppressed interruption, or are waiting for I/O,
3057		* causing this executor not to properly terminate. (See the
3058		* advisory notes for class {@link ForkJoinTask} stating that
3059		* tasks should not normally entail blocking operations. But if
#	Line 2472 \| Line 3064 \| public class ForkJoinPool extends Abstra
3064		public boolean isTerminating() {
3065		long c = ctl;
3066		return ((c & STOP_BIT) != 0L &&
3067	<	(short)(c >>> TC_SHIFT) != -parallelism);
3067	>	(short)(c >>> TC_SHIFT) != -(config & SMASK));
3068		}
3069
3070		/**
#	Line 2481 \| Line 3073 \| public class ForkJoinPool extends Abstra
3073		* @return {@code true} if this pool has been shut down
3074		*/
3075		public boolean isShutdown() {
3076	<	return runState < 0;
3076	>	return plock < 0;
3077		}
3078
3079		/**
3080	<	* Blocks until all tasks have completed execution after a shutdown
3081	<	* request, or the timeout occurs, or the current thread is
3082	<	* interrupted, whichever happens first.
3080	>	* Blocks until all tasks have completed execution after a
3081	>	* shutdown request, or the timeout occurs, or the current thread
3082	>	* is interrupted, whichever happens first. Because the {@link
3083	>	* #commonPool()} never terminates until program shutdown, when
3084	>	* applied to the common pool, this method is equivalent to {@link
3085	>	* #awaitQuiescence} but always returns {@code false}.
3086		*
3087		* @param timeout the maximum time to wait
3088		* @param unit the time unit of the timeout argument
#	Line 2497 \| Line 3092 \| public class ForkJoinPool extends Abstra
3092		*/
3093		public boolean awaitTermination(long timeout, TimeUnit unit)
3094		throws InterruptedException {
3095	+	if (Thread.interrupted())
3096	+	throw new InterruptedException();
3097	+	if (this == common) {
3098	+	awaitQuiescence(timeout, unit);
3099	+	return false;
3100	+	}
3101		long nanos = unit.toNanos(timeout);
3102	<	final Mutex lock = this.lock;
3103	<	lock.lock();
3104	<	try {
3105	<	for (;;) {
3106	<	if (isTerminated())
3107	<	return true;
3108	<	if (nanos <= 0)
3102	>	if (isTerminated())
3103	>	return true;
3104	>	long startTime = System.nanoTime();
3105	>	boolean terminated = false;
3106	>	synchronized (this) {
3107	>	for (long waitTime = nanos, millis = 0L;;) {
3108	>	if (terminated = isTerminated() \|\|
3109	>	waitTime <= 0L \|\|
3110	>	(millis = unit.toMillis(waitTime)) <= 0L)
3111	>	break;
3112	>	wait(millis);
3113	>	waitTime = nanos - (System.nanoTime() - startTime);
3114	>	}
3115	>	}
3116	>	return terminated;
3117	>	}
3118	>
3119	>	/**
3120	>	* If called by a ForkJoinTask operating in this pool, equivalent
3121	>	* in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise,
3122	>	* waits and/or attempts to assist performing tasks until this
3123	>	* pool {@link #isQuiescent} or the indicated timeout elapses.
3124	>	*
3125	>	* @param timeout the maximum time to wait
3126	>	* @param unit the time unit of the timeout argument
3127	>	* @return {@code true} if quiescent; {@code false} if the
3128	>	* timeout elapsed.
3129	>	*/
3130	>	public boolean awaitQuiescence(long timeout, TimeUnit unit) {
3131	>	long nanos = unit.toNanos(timeout);
3132	>	ForkJoinWorkerThread wt;
3133	>	Thread thread = Thread.currentThread();
3134	>	if ((thread instanceof ForkJoinWorkerThread) &&
3135	>	(wt = (ForkJoinWorkerThread)thread).pool == this) {
3136	>	helpQuiescePool(wt.workQueue);
3137	>	return true;
3138	>	}
3139	>	long startTime = System.nanoTime();
3140	>	WorkQueue[] ws;
3141	>	int r = 0, m;
3142	>	boolean found = true;
3143	>	while (!isQuiescent() && (ws = workQueues) != null &&
3144	>	(m = ws.length - 1) >= 0) {
3145	>	if (!found) {
3146	>	if ((System.nanoTime() - startTime) > nanos)
3147		return false;
3148	<	nanos = termination.awaitNanos(nanos);
3148	>	Thread.yield(); // cannot block
3149	>	}
3150	>	found = false;
3151	>	for (int j = (m + 1) << 2; j >= 0; --j) {
3152	>	ForkJoinTask<?> t; WorkQueue q; int b;
3153	>	if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) {
3154	>	found = true;
3155	>	if ((t = q.pollAt(b)) != null) {
3156	>	if (q.base - q.top < 0)
3157	>	signalWork(q);
3158	>	t.doExec();
3159	>	}
3160	>	break;
3161	>	}
3162		}
2511	–	} finally {
2512	–	lock.unlock();
3163		}
3164	+	return true;
3165	+	}
3166	+
3167	+	/**
3168	+	* Waits and/or attempts to assist performing tasks indefinitely
3169	+	* until the {@link #commonPool()} {@link #isQuiescent}
3170	+	*/
3171	+	static void quiesceCommonPool() {
3172	+	common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
3173		}
3174
3175		/**
#	Line 2609 \| Line 3268 \| public class ForkJoinPool extends Abstra
3268		public static void managedBlock(ManagedBlocker blocker)
3269		throws InterruptedException {
3270		Thread t = Thread.currentThread();
3271	<	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3272	<	((ForkJoinWorkerThread)t).pool : null);
3273	<	while (!blocker.isReleasable()) {
3274	<	if (p == null \|\| p.tryCompensate()) {
3275	<	try {
3276	<	do {} while (!blocker.isReleasable() && !blocker.block());
3277	<	} finally {
3278	<	if (p != null)
3271	>	if (t instanceof ForkJoinWorkerThread) {
3272	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3273	>	while (!blocker.isReleasable()) { // variant of helpSignal
3274	>	WorkQueue[] ws; WorkQueue q; int m, u;
3275	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3276	>	for (int i = 0; i <= m; ++i) {
3277	>	if (blocker.isReleasable())
3278	>	return;
3279	>	if ((q = ws[i]) != null && q.base - q.top < 0) {
3280	>	p.signalWork(q);
3281	>	if ((u = (int)(p.ctl >>> 32)) >= 0 \|\|
3282	>	(u >> UAC_SHIFT) >= 0)
3283	>	break;
3284	>	}
3285	>	}
3286	>	}
3287	>	if (p.tryCompensate()) {
3288	>	try {
3289	>	do {} while (!blocker.isReleasable() &&
3290	>	!blocker.block());
3291	>	} finally {
3292		p.incrementActiveCount();
3293	+	}
3294	+	break;
3295		}
2622	–	break;
3296		}
3297		}
3298	+	else {
3299	+	do {} while (!blocker.isReleasable() &&
3300	+	!blocker.block());
3301	+	}
3302		}
3303
3304		// AbstractExecutorService overrides. These rely on undocumented
#	Line 2629 \| Line 3306 \| public class ForkJoinPool extends Abstra
3306		// implement RunnableFuture.
3307
3308		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3309	<	return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
3309	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3310		}
3311
3312		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3313	<	return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
3313	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
3314		}
3315
3316		// Unsafe mechanics
3317		private static final sun.misc.Unsafe U;
3318		private static final long CTL;
3319		private static final long PARKBLOCKER;
3320	+	private static final int ABASE;
3321	+	private static final int ASHIFT;
3322	+	private static final long STEALCOUNT;
3323	+	private static final long PLOCK;
3324	+	private static final long INDEXSEED;
3325	+	private static final long QLOCK;
3326
3327		static {
3328	<	poolNumberGenerator = new AtomicInteger();
2646	<	modifyThreadPermission = new RuntimePermission("modifyThread");
2647	<	defaultForkJoinWorkerThreadFactory =
2648	<	new DefaultForkJoinWorkerThreadFactory();
2649	<	submitters = new ThreadSubmitter();
3328	>	// initialize field offsets for CAS etc
3329		try {
3330		U = getUnsafe();
3331		Class<?> k = ForkJoinPool.class;
3332		CTL = U.objectFieldOffset
3333		(k.getDeclaredField("ctl"));
3334	+	STEALCOUNT = U.objectFieldOffset
3335	+	(k.getDeclaredField("stealCount"));
3336	+	PLOCK = U.objectFieldOffset
3337	+	(k.getDeclaredField("plock"));
3338	+	INDEXSEED = U.objectFieldOffset
3339	+	(k.getDeclaredField("indexSeed"));
3340		Class<?> tk = Thread.class;
3341		PARKBLOCKER = U.objectFieldOffset
3342		(tk.getDeclaredField("parkBlocker"));
3343	+	Class<?> wk = WorkQueue.class;
3344	+	QLOCK = U.objectFieldOffset
3345	+	(wk.getDeclaredField("qlock"));
3346	+	Class<?> ak = ForkJoinTask[].class;
3347	+	ABASE = U.arrayBaseOffset(ak);
3348	+	int scale = U.arrayIndexScale(ak);
3349	+	if ((scale & (scale - 1)) != 0)
3350	+	throw new Error("data type scale not a power of two");
3351	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
3352		} catch (Exception e) {
3353		throw new Error(e);
3354		}
3355	+
3356	+	submitters = new ThreadLocal<Submitter>();
3357	+	ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory =
3358	+	new DefaultForkJoinWorkerThreadFactory();
3359	+	modifyThreadPermission = new RuntimePermission("modifyThread");
3360	+
3361	+	/*
3362	+	* Establish common pool parameters. For extra caution,
3363	+	* computations to set up common pool state are here; the
3364	+	* constructor just assigns these values to fields.
3365	+	*/
3366	+
3367	+	int par = 0;
3368	+	Thread.UncaughtExceptionHandler handler = null;
3369	+	try { // TBD: limit or report ignored exceptions?
3370	+	String pp = System.getProperty
3371	+	("java.util.concurrent.ForkJoinPool.common.parallelism");
3372	+	String hp = System.getProperty
3373	+	("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3374	+	String fp = System.getProperty
3375	+	("java.util.concurrent.ForkJoinPool.common.threadFactory");
3376	+	if (fp != null)
3377	+	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3378	+	getSystemClassLoader().loadClass(fp).newInstance());
3379	+	if (hp != null)
3380	+	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3381	+	getSystemClassLoader().loadClass(hp).newInstance());
3382	+	if (pp != null)
3383	+	par = Integer.parseInt(pp);
3384	+	} catch (Exception ignore) {
3385	+	}
3386	+
3387	+	if (par <= 0)
3388	+	par = Runtime.getRuntime().availableProcessors();
3389	+	if (par > MAX_CAP)
3390	+	par = MAX_CAP;
3391	+	commonParallelism = par;
3392	+	long np = (long)(-par); // precompute initial ctl value
3393	+	long ct = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
3394	+
3395	+	common = new ForkJoinPool(par, ct, fac, handler);
3396		}
3397
3398		/**
#	Line 2670 \| Line 3405 \| public class ForkJoinPool extends Abstra
3405		private static sun.misc.Unsafe getUnsafe() {
3406		try {
3407		return sun.misc.Unsafe.getUnsafe();
3408	<	} catch (SecurityException se) {
3409	<	try {
3410	<	return java.security.AccessController.doPrivileged
3411	<	(new java.security
3412	<	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3413	<	public sun.misc.Unsafe run() throws Exception {
3414	<	java.lang.reflect.Field f = sun.misc
3415	<	.Unsafe.class.getDeclaredField("theUnsafe");
3416	<	f.setAccessible(true);
3417	<	return (sun.misc.Unsafe) f.get(null);
3418	<	}});
3419	<	} catch (java.security.PrivilegedActionException e) {
3420	<	throw new RuntimeException("Could not initialize intrinsics",
3421	<	e.getCause());
3422	<	}
3408	>	} catch (SecurityException tryReflectionInstead) {}
3409	>	try {
3410	>	return java.security.AccessController.doPrivileged
3411	>	(new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3412	>	public sun.misc.Unsafe run() throws Exception {
3413	>	Class<sun.misc.Unsafe> k = sun.misc.Unsafe.class;
3414	>	for (java.lang.reflect.Field f : k.getDeclaredFields()) {
3415	>	f.setAccessible(true);
3416	>	Object x = f.get(null);
3417	>	if (k.isInstance(x))
3418	>	return k.cast(x);
3419	>	}
3420	>	throw new NoSuchFieldError("the Unsafe");
3421	>	}});
3422	>	} catch (java.security.PrivilegedActionException e) {
3423	>	throw new RuntimeException("Could not initialize intrinsics",
3424	>	e.getCause());
3425		}
3426		}
2690	–
3427		}

Diff Legend

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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents): Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC vs. Revision 1.181 by jsr166, Tue Feb 5 17:25:09 2013 UTC

Diff Legend

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC vs.
Revision 1.181 by jsr166, Tue Feb 5 17:25:09 2013 UTC