ViewVC Help

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

root/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 1.160 by dl, Sun Dec 16 19:57:04 2012 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 commonPool always exists after static
444	>	* initialization.  Since it (or any other created pool) need
445	>	* never be used, we minimize initial construction overhead and
446	>	* footprint to the setup of about a dozen fields, with no nested
447	>	* allocation. Most bootstrapping occurs within method
448	>	* fullExternalPush during the first submission to the pool.
449	>	*
450	>	* When external threads submit to the common pool, they can
451	>	* perform some subtask processing (see externalHelpJoin and
452	>	* related methods).  We do not need to record whether these
453	>	* submissions are to the common pool -- if not, externalHelpJoin
454	>	* returns quickly (at the most helping to signal some common pool
455	>	* workers). These submitters would otherwise be blocked waiting
456	>	* for completion, so the extra effort (with liberally sprinkled
457	>	* task status checks) in inapplicable cases amounts to an odd
458	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
459	>	*
460	>	* Style notes
461	>	* ===========
462	>	*
463	>	* There is a lot of representation-level coupling among classes
464	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask.  The
465	>	* fields of WorkQueue maintain data structures managed by
466	>	* ForkJoinPool, so are directly accessed.  There is little point
467	>	* trying to reduce this, since any associated future changes in
468	>	* representations will need to be accompanied by algorithmic
469	>	* changes anyway. Several methods intrinsically sprawl because
470	>	* they must accumulate sets of consistent reads of volatiles held
471	>	* in local variables.  Methods signalWork() and scan() are the
472	>	* main bottlenecks, so are especially heavily
473		* micro-optimized/mangled.  There are lots of inline assignments
474		* (of form "while ((local = field) != 0)") which are usually the
475		* simplest way to ensure the required read orderings (which are
#	Line 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.
648	–	* @param p if non-null, pool to signal if necessary
711		* @throw 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 {
#	Line 986 | Line 1031 | public class ForkJoinPool extends Abstra
1031		U = getUnsafe();
1032		Class<?> k = WorkQueue.class;
1033		Class<?> ak = ForkJoinTask[].class;
1034	<	RUNSTATE = U.objectFieldOffset
1035	<	(k.getDeclaredField("runState"));
1034	>	QLOCK = U.objectFieldOffset
1035	>	(k.getDeclaredField("qlock"));
1036		ABASE = U.arrayBaseOffset(ak);
1037		s = U.arrayIndexScale(ak);
1038		} catch (Exception e) {
#	Line 999 | Line 1044 | public class ForkJoinPool extends Abstra
1044		}
1045		}
1046
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	–
1047		// static fields (initialized in static initializer below)
1048
1049		/**
#	Line 1026 | Line 1054 | public class ForkJoinPool extends Abstra
1054		defaultForkJoinWorkerThreadFactory;
1055
1056		/**
1057	<	* Generator for assigning sequence numbers as pool names.
1057	>	* Per-thread submission bookkeeping. Shared across all pools
1058	>	* to reduce ThreadLocal pollution and because random motion
1059	>	* to avoid contention in one pool is likely to hold for others.
1060	>	* Lazily initialized on first submission (but null-checked
1061	>	* in other contexts to avoid unnecessary initialization).
1062		*/
1063	<	private static final AtomicInteger poolNumberGenerator;
1063	>	static final ThreadLocal<Submitter> submitters;
1064
1065		/**
1066		* Permission required for callers of methods that may start or
#	Line 1037 | Line 1069 | public class ForkJoinPool extends Abstra
1069		private static final RuntimePermission modifyThreadPermission;
1070
1071		/**
1072	<	* Per-thread submission bookeeping. Shared across all pools
1073	<	* to reduce ThreadLocal pollution and because random motion
1074	<	* to avoid contention in one pool is likely to hold for others.
1072	>	* Common (static) pool. Non-null for public use unless a static
1073	>	* construction exception, but internal usages null-check on use
1074	>	* to paranoically avoid potential initialization circularities
1075	>	* as well as to simplify generated code.
1076	>	*/
1077	>	static final ForkJoinPool commonPool;
1078	>
1079	>	/**
1080	>	* Common pool parallelism. Must equal commonPool.parallelism.
1081		*/
1082	<	private static final ThreadSubmitter submitters;
1082	>	static final int commonPoolParallelism;
1083	>
1084	>	/**
1085	>	* Sequence number for creating workerNamePrefix.
1086	>	*/
1087	>	private static int poolNumberSequence;
1088	>
1089	>	/**
1090	>	* Return the next sequence number. We don't expect this to
1091	>	* ever contend so use simple builtin sync.
1092	>	*/
1093	>	private static final synchronized int nextPoolId() {
1094	>	return ++poolNumberSequence;
1095	>	}
1096
1097		// static constants
1098
1099		/**
1100	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
1101	<	* task when the pool is quiescent to instead try to shrink the
1102	<	* number of workers.  The exact value does not matter too
1103	<	* much. It must be short enough to release resources during
1104	<	* sustained periods of idleness, but not so short that threads
1105	<	* are continually re-created.
1100	>	* Initial timeout value (in nanoseconds) for the thread
1101	>	* triggering quiescence to park waiting for new work. On timeout,
1102	>	* the thread will instead try to shrink the number of
1103	>	* workers. The value should be large enough to avoid overly
1104	>	* aggressive shrinkage during most transient stalls (long GCs
1105	>	* etc).
1106	>	*/
1107	>	private static final long IDLE_TIMEOUT      = 2000L * 1000L * 1000L; // 2sec
1108	>
1109	>	/**
1110	>	* Timeout value when there are more threads than parallelism level
1111		*/
1112	<	private static final long SHRINK_RATE =
1057	<	4L * 1000L * 1000L * 1000L; // 4 seconds
1112	>	private static final long FAST_IDLE_TIMEOUT =  200L * 1000L * 1000L;
1113
1114		/**
1115	<	* The timeout value for attempted shrinkage, includes
1061	<	* some slop to cope with system timer imprecision.
1115	>	* Tolerance for idle timeouts, to cope with timer undershoots
1116		*/
1117	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1117	>	private static final long TIMEOUT_SLOP = 2000000L; // 20ms
1118
1119		/**
1120	<	* The maximum stolen->joining link depth allowed in tryHelpStealer.
1121	<	* Depths for legitimate chains are unbounded, but we use a fixed
1122	<	* constant to avoid (otherwise unchecked) cycles and to bound
1123	<	* staleness of traversal parameters at the expense of sometimes
1124	<	* blocking when we could be helping.
1120	>	* The maximum stolen->joining link depth allowed in method
1121	>	* tryHelpStealer.  Must be a power of two.  Depths for legitimate
1122	>	* chains are unbounded, but we use a fixed constant to avoid
1123	>	* (otherwise unchecked) cycles and to bound staleness of
1124	>	* traversal parameters at the expense of sometimes blocking when
1125	>	* we could be helping.
1126		*/
1127	<	private static final int MAX_HELP_DEPTH = 16;
1127	>	private static final int MAX_HELP = 64;
1128	>
1129	>	/**
1130	>	* Increment for seed generators. See class ThreadLocal for
1131	>	* explanation.
1132	>	*/
1133	>	private static final int SEED_INCREMENT = 0x61c88647;
1134
1135		/**
1136		* Bits and masks for control variables
#	Line 1099 | Line 1160 | public class ForkJoinPool extends Abstra
1160		* scan for them to avoid queuing races. Note however that
1161		* eventCount updates lag releases so usage requires care.
1162		*
1163	<	* Field runState is an int packed with:
1163	>	* Field plock is an int packed with:
1164		* SHUTDOWN: true if shutdown is enabled (1 bit)
1165	<	* SEQ:  a sequence number updated upon (de)registering workers (15 bits)
1166	<	* MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
1165	>	* SEQ:  a sequence lock, with PL_LOCK bit set if locked (30 bits)
1166	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1167		*
1168	<	* The combination of mask and sequence number enables simple
1169	<	* consistency checks: Staleness of read-only operations on the
1170	<	* 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.
1168	>	* The sequence number enables simple consistency checks:
1169	>	* Staleness of read-only operations on the workQueues array can
1170	>	* be checked by comparing plock before vs after the reads.
1171		*/
1172
1173		// bit positions/shifts for fields
#	Line 1119 | Line 1177 | public class ForkJoinPool extends Abstra
1177		private static final int  EC_SHIFT   = 16;
1178
1179		// bounds
1122	–	private static final int  POOL_MAX   = 0x7fff;  // max #workers - 1
1180		private static final int  SMASK      = 0xffff;  // short bits
1181	<	private static final int  SQMASK     = 0xfffe;  // even short bits
1181	>	private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
1182	>	private static final int  EVENMASK   = 0xfffe;  // even short bits
1183	>	private static final int  SQMASK     = 0x007e;  // max 64 (even) slots
1184		private static final int  SHORT_SIGN = 1 << 15;
1185		private static final int  INT_SIGN   = 1 << 31;
1186
#	Line 1146 | Line 1205 | public class ForkJoinPool extends Abstra
1205		private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
1206		private static final int E_SEQ       = 1 << EC_SHIFT;
1207
1208	<	// runState bits
1208	>	// plock bits
1209		private static final int SHUTDOWN    = 1 << 31;
1210	<	private static final int RS_SEQ      = 1 << 16;
1211	<	private static final int RS_SEQ_MASK = 0x7fff0000;
1210	>	private static final int PL_LOCK     = 2;
1211	>	private static final int PL_SIGNAL   = 1;
1212	>	private static final int PL_SPINS    = 1 << 8;
1213
1214		// access mode for WorkQueue
1215		static final int LIFO_QUEUE          =  0;
1216		static final int FIFO_QUEUE          =  1;
1217		static final int SHARED_QUEUE        = -1;
1218
1219	+	// bounds for #steps in scan loop -- must be power 2 minus 1
1220	+	private static final int MIN_SCAN    = 0x1ff;   // cover estimation slop
1221	+	private static final int MAX_SCAN    = 0x1ffff; // 4 * max workers
1222	+
1223		// Instance fields
1224
1225		/*
1226	<	* Field layout order in this class tends to matter more than one
1227	<	* would like. Runtime layout order is only loosely related to
1226	>	* Field layout of this class tends to matter more than one would
1227	>	* like. Runtime layout order is only loosely related to
1228		* declaration order and may differ across JVMs, but the following
1229		* empirically works OK on current JVMs.
1230		*/
1231
1232	+	// Heuristic padding to ameliorate unfortunate memory placements
1233	+	volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
1234	+
1235	+	volatile long stealCount;                  // collects worker counts
1236		volatile long ctl;                         // main pool control
1237	<	final int parallelism;                     // parallelism level
1238	<	final int localMode;                       // per-worker scheduling mode
1239	<	int growHints;                             // for expanding indices/ranges
1172	<	volatile int runState;                     // shutdown status, seq, and mask
1237	>	volatile int plock;                        // shutdown status and seqLock
1238	>	volatile int indexSeed;                    // worker/submitter index seed
1239	>	final int config;                          // mode and parallelism level
1240		WorkQueue[] workQueues;                    // main registry
1241	<	final Mutex lock;                          // for registration
1175	<	final Condition termination;               // for awaitTermination
1176	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1241	>	final ForkJoinWorkerThreadFactory factory;
1242		final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1178	–	final AtomicLong stealCount;               // collect counts when terminated
1179	–	final AtomicInteger nextWorkerNumber;      // to create worker name string
1243		final String workerNamePrefix;             // to create worker name string
1244
1245	<	//  Creating, registering, deregistering and running workers
1245	>	volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
1246	>	volatile Object pad18, pad19, pad1a, pad1b;
1247
1248	<	/**
1249	<	* Tries to create and start a worker
1250	<	*/
1251	<	private void addWorker() {
1252	<	Throwable ex = null;
1253	<	ForkJoinWorkerThread wt = null;
1254	<	try {
1255	<	if ((wt = factory.newThread(this)) != null) {
1256	<	wt.start();
1257	<	return;
1248	>	/*
1249	>	* Acquires the plock lock to protect worker array and related
1250	>	* updates. This method is called only if an initial CAS on plock
1251	>	* fails. This acts as a spinLock for normal cases, but falls back
1252	>	* to builtin monitor to block when (rarely) needed. This would be
1253	>	* a terrible idea for a highly contended lock, but works fine as
1254	>	* a more conservative alternative to a pure spinlock.
1255	>	*/
1256	>	private int acquirePlock() {
1257	>	int spins = PL_SPINS, r = 0, ps, nps;
1258	>	for (;;) {
1259	>	if (((ps = plock) & PL_LOCK) == 0 &&
1260	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1261	>	return nps;
1262	>	else if (r == 0) { // randomize spins if possible
1263	>	Thread t = Thread.currentThread(); WorkQueue w; Submitter z;
1264	>	if ((t instanceof ForkJoinWorkerThread) &&
1265	>	(w = ((ForkJoinWorkerThread)t).workQueue) != null)
1266	>	r = w.seed;
1267	>	else if ((z = submitters.get()) != null)
1268	>	r = z.seed;
1269	>	else
1270	>	r = 1;
1271	>	}
1272	>	else if (spins >= 0) {
1273	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1274	>	if (r >= 0)
1275	>	--spins;
1276	>	}
1277	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
1278	>	synchronized (this) {
1279	>	if ((plock & PL_SIGNAL) != 0) {
1280	>	try {
1281	>	wait();
1282	>	} catch (InterruptedException ie) {
1283	>	try {
1284	>	Thread.currentThread().interrupt();
1285	>	} catch (SecurityException ignore) {
1286	>	}
1287	>	}
1288	>	}
1289	>	else
1290	>	notifyAll();
1291	>	}
1292		}
1195	–	} catch (Throwable e) {
1196	–	ex = e;
1293		}
1198	–	deregisterWorker(wt, ex); // adjust counts etc on failure
1294		}
1295
1296		/**
1297	<	* Callback from ForkJoinWorkerThread constructor to assign a
1298	<	* public name. This must be separate from registerWorker because
1204	<	* it is called during the "super" constructor call in
1205	<	* ForkJoinWorkerThread.
1297	>	* Unlocks and signals any thread waiting for plock. Called only
1298	>	* when CAS of seq value for unlock fails.
1299		*/
1300	<	final String nextWorkerName() {
1301	<	return workerNamePrefix.concat
1302	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1300	>	private void releasePlock(int ps) {
1301	>	plock = ps;
1302	>	synchronized (this) { notifyAll(); }
1303		}
1304
1305		/**
1306	<	* Callback from ForkJoinWorkerThread constructor to establish and
1307	<	* record its WorkQueue.
1306	>	* Performs secondary initialization, called when plock is zero.
1307	>	* Creates workQueue array and sets plock to a valid value.  The
1308	>	* lock body must be exception-free (so no try/finally) so we
1309	>	* optimistically allocate new array outside the lock and throw
1310	>	* away if (very rarely) not needed. (A similar tactic is used in
1311	>	* fullExternalPush.)  Because the plock seq value can eventually
1312	>	* wrap around zero, this method harmlessly fails to reinitialize
1313	>	* if workQueues exists, while still advancing plock.
1314	>	*
1315	>	* Additionally tries to create the first worker.
1316	>	*/
1317	>	private void initWorkers() {
1318	>	WorkQueue[] ws, nws; int ps;
1319	>	int p = config & SMASK;        // find power of two table size
1320	>	int n = (p > 1) ? p - 1 : 1;   // ensure at least 2 slots
1321	>	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
1322	>	n = (n + 1) << 1;
1323	>	if ((ws = workQueues) == null || ws.length == 0)
1324	>	nws = new WorkQueue[n];
1325	>	else
1326	>	nws = null;
1327	>	if (((ps = plock) & PL_LOCK) != 0 ||
1328	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1329	>	ps = acquirePlock();
1330	>	if (((ws = workQueues) == null || ws.length == 0) && nws != null)
1331	>	workQueues = nws;
1332	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1333	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1334	>	releasePlock(nps);
1335	>	tryAddWorker();
1336	>	}
1337	>
1338	>	/**
1339	>	* Tries to create and start one worker if fewer than target
1340	>	* parallelism level exist. Adjusts counts etc on failure.
1341	>	*/
1342	>	private void tryAddWorker() {
1343	>	long c; int u;
1344	>	while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1345	>	(u & SHORT_SIGN) != 0 && (int)c == 0) {
1346	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1347	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1348	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1349	>	ForkJoinWorkerThreadFactory fac;
1350	>	Throwable ex = null;
1351	>	ForkJoinWorkerThread wt = null;
1352	>	try {
1353	>	if ((fac = factory) != null &&
1354	>	(wt = fac.newThread(this)) != null) {
1355	>	wt.start();
1356	>	break;
1357	>	}
1358	>	} catch (Throwable e) {
1359	>	ex = e;
1360	>	}
1361	>	deregisterWorker(wt, ex);
1362	>	break;
1363	>	}
1364	>	}
1365	>	}
1366	>
1367	>	//  Registering and deregistering workers
1368	>
1369	>	/**
1370	>	* Callback from ForkJoinWorkerThread to establish and record its
1371	>	* WorkQueue. To avoid scanning bias due to packing entries in
1372	>	* front of the workQueues array, we treat the array as a simple
1373	>	* power-of-two hash table using per-thread seed as hash,
1374	>	* expanding as needed.
1375		*
1376		* @param wt the worker thread
1377	+	* @return the worker's queue
1378		*/
1379	<	final void registerWorker(ForkJoinWorkerThread wt) {
1380	<	WorkQueue w = wt.workQueue;
1381	<	Mutex lock = this.lock;
1382	<	lock.lock();
1379	>	final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1380	>	Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1381	>	wt.setDaemon(true);
1382	>	if ((handler = ueh) != null)
1383	>	wt.setUncaughtExceptionHandler(handler);
1384	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1385	>	s += SEED_INCREMENT) ||
1386	>	s == 0); // skip 0
1387	>	WorkQueue w = new WorkQueue(this, wt, config >>> 16, s);
1388	>	if (((ps = plock) & PL_LOCK) != 0 ||
1389	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1390	>	ps = acquirePlock();
1391	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1392		try {
1393	<	int g = growHints, k = g & SMASK;
1394	<	WorkQueue[] ws = workQueues;
1395	<	if (ws != null) {                       // ignore on shutdown
1396	<	int n = ws.length;
1397	<	if ((k & 1) == 0 || k >= n || ws[k] != null) {
1398	<	for (k = 1; k < n && ws[k] != null; k += 2)
1399	<	;                           // workers are at odd indices
1400	<	if (k >= n)                     // resize
1401	<	workQueues = ws = Arrays.copyOf(ws, n << 1);
1402	<	}
1403	<	w.eventCount = w.poolIndex = k;     // establish before recording
1404	<	ws[k] = w;
1405	<	growHints = (g & ~SMASK) | ((k + 2) & SMASK);
1406	<	int rs = runState;
1407	<	int m = rs & SMASK;                 // recalculate runState mask
1408	<	if (k > m)
1239	<	m = (m << 1) + 1;
1240	<	runState = (rs & SHUTDOWN) | ((rs + RS_SEQ) & RS_SEQ_MASK) | m;
1393	>	if ((ws = workQueues) != null) {    // skip if shutting down
1394	>	int n = ws.length, m = n - 1;
1395	>	int r = (s << 1) | 1;           // use odd-numbered indices
1396	>	if (ws[r &= m] != null) {       // collision
1397	>	int probes = 0;             // step by approx half size
1398	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1399	>	while (ws[r = (r + step) & m] != null) {
1400	>	if (++probes >= n) {
1401	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1402	>	m = n - 1;
1403	>	probes = 0;
1404	>	}
1405	>	}
1406	>	}
1407	>	w.eventCount = w.poolIndex = r; // volatile write orders
1408	>	ws[r] = w;
1409		}
1410		} finally {
1411	<	lock.unlock();
1411	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1412	>	releasePlock(nps);
1413		}
1414	+	wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex)));
1415	+	return w;
1416		}
1417
1418		/**
1419		* Final callback from terminating worker, as well as upon failure
1420	<	* to construct or start a worker in addWorker.  Removes record of
1421	<	* worker from array, and adjusts counts. If pool is shutting
1422	<	* down, tries to complete termination.
1420	>	* to construct or start a worker.  Removes record of worker from
1421	>	* array, and adjusts counts. If pool is shutting down, tries to
1422	>	* complete termination.
1423		*
1424	<	* @param wt the worker thread or null if addWorker failed
1424	>	* @param wt the worker thread or null if construction failed
1425		* @param ex the exception causing failure, or null if none
1426		*/
1427		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1428		WorkQueue w = null;
1429		if (wt != null && (w = wt.workQueue) != null) {
1430	<	w.runState = -1;                // ensure runState is set
1431	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1432	<	int idx = w.poolIndex;
1433	<	Mutex lock = this.lock;
1434	<	lock.lock();
1435	<	try {                           // remove record from array
1430	>	int ps;
1431	>	w.qlock = -1;                // ensure set
1432	>	long ns = w.nsteals, sc;     // collect steal count
1433	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1434	>	sc = stealCount, sc + ns));
1435	>	if (((ps = plock) & PL_LOCK) != 0 ||
1436	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1437	>	ps = acquirePlock();
1438	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1439	>	try {
1440	>	int idx = w.poolIndex;
1441		WorkQueue[] ws = workQueues;
1442	<	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w) {
1442	>	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1443		ws[idx] = null;
1268	–	growHints = (growHints & ~SMASK) | idx;
1269	–	}
1444		} finally {
1445	<	lock.unlock();
1445	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1446	>	releasePlock(nps);
1447		}
1448		}
1449
1450	<	long c;                             // adjust ctl counts
1450	>	long c;                          // adjust ctl counts
1451		do {} while (!U.compareAndSwapLong
1452		(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1453		((c - TC_UNIT) & TC_MASK) |
1454		(c & ~(AC_MASK|TC_MASK)))));
1455
1456	<	if (!tryTerminate(false, false) && w != null) {
1457	<	w.cancelAll();                  // cancel remaining tasks
1458	<	if (w.array != null)            // suppress signal if never ran
1459	<	signalWork();               // wake up or create replacement
1460	<	if (ex == null)                 // help clean refs on way out
1461	<	ForkJoinTask.helpExpungeStaleExceptions();
1456	>	if (!tryTerminate(false, false) && w != null && w.array != null) {
1457	>	w.cancelAll();               // cancel remaining tasks
1458	>	WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e;
1459	>	while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) {
1460	>	if (e > 0) {             // activate or create replacement
1461	>	if ((ws = workQueues) == null ||
1462	>	(i = e & SMASK) >= ws.length ||
1463	>	(v = ws[i]) != null)
1464	>	break;
1465	>	long nc = (((long)(v.nextWait & E_MASK)) |
1466	>	((long)(u + UAC_UNIT) << 32));
1467	>	if (v.eventCount != (e | INT_SIGN))
1468	>	break;
1469	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1470	>	v.eventCount = (e + E_SEQ) & E_MASK;
1471	>	if ((p = v.parker) != null)
1472	>	U.unpark(p);
1473	>	break;
1474	>	}
1475	>	}
1476	>	else {
1477	>	if ((short)u < 0)
1478	>	tryAddWorker();
1479	>	break;
1480	>	}
1481	>	}
1482		}
1483	<
1484	<	if (ex != null)                     // rethrow
1485	<	U.throwException(ex);
1486	<	}
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)));
1483	>	if (ex == null)                     // help clean refs on way out
1484	>	ForkJoinTask.helpExpungeStaleExceptions();
1485	>	else                                // rethrow
1486	>	ForkJoinTask.rethrow(ex);
1487		}
1488
1489		// Submissions
#	Line 1307 | Line 1491 | public class ForkJoinPool extends Abstra
1491		/**
1492		* Unless shutting down, adds the given task to a submission queue
1493		* at submitter's current queue index (modulo submission
1494	<	* range). If no queue exists at the index, one is created unless
1495	<	* pool lock is busy.  If the queue and/or lock are busy, another
1496	<	* index is randomly chosen. The mask in growHints controls the
1497	<	* 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.
1494	>	* range). Only the most common path is directly handled in this
1495	>	* method. All others are relayed to fullExternalPush.
1496	>	*
1497	>	* @param task the task. Caller must ensure non-null.
1498		*/
1499	<	private void doSubmit(ForkJoinTask<?> task) {
1500	<	if (task == null)
1501	<	throw new NullPointerException();
1502	<	Submitter s = submitters.get();
1503	<	for (int r = s.seed, m = growHints >>> 16;;) {
1504	<	WorkQueue[] ws; WorkQueue q; Mutex lk;
1505	<	int k = r & m & SQMASK;          // use only even indices
1506	<	if (runState < 0 || (ws = workQueues) == null || ws.length <= k)
1507	<	throw new RejectedExecutionException(); // shutting down
1508	<	if ((q = ws[k]) == null && (lk = lock).tryAcquire(0)) {
1509	<	try {                        // try to create new queue
1510	<	if (ws == workQueues && (q = ws[k]) == null) {
1511	<	int rs;              // update runState seq
1512	<	ws[k] = q = new WorkQueue(null, SHARED_QUEUE);
1513	<	runState = (((rs = runState) & SHUTDOWN) |
1514	<	((rs + RS_SEQ) & ~SHUTDOWN));
1499	>	final void externalPush(ForkJoinTask<?> task) {
1500	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1501	>	if ((z = submitters.get()) != null && plock > 0 &&
1502	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1503	>	(q = ws[m & z.seed & SQMASK]) != null &&
1504	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1505	>	int b = q.base, s = q.top, n, an;
1506	>	if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) {
1507	>	int j = (((an - 1) & s) << ASHIFT) + ABASE;
1508	>	U.putOrderedObject(a, j, task);
1509	>	q.top = s + 1;                     // push on to deque
1510	>	q.qlock = 0;
1511	>	if (n <= 2)
1512	>	signalWork(q);
1513	>	return;
1514	>	}
1515	>	q.qlock = 0;
1516	>	}
1517	>	fullExternalPush(task);
1518	>	}
1519	>
1520	>	/**
1521	>	* Full version of externalPush. This method is called, among
1522	>	* other times, upon the first submission of the first task to the
1523	>	* pool, so must perform secondary initialization (via
1524	>	* initWorkers). It also detects first submission by an external
1525	>	* thread by looking up its ThreadLocal, and creates a new shared
1526	>	* queue if the one at index if empty or contended. The plock lock
1527	>	* body must be exception-free (so no try/finally) so we
1528	>	* optimistically allocate new queues outside the lock and throw
1529	>	* them away if (very rarely) not needed.
1530	>	*/
1531	>	private void fullExternalPush(ForkJoinTask<?> task) {
1532	>	int r = 0; // random index seed
1533	>	for (Submitter z = submitters.get();;) {
1534	>	WorkQueue[] ws; WorkQueue q; int ps, m, k;
1535	>	if (z == null) {
1536	>	if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1537	>	r += SEED_INCREMENT) && r != 0)
1538	>	submitters.set(z = new Submitter(r));
1539	>	}
1540	>	else if (r == 0) {               // move to a different index
1541	>	r = z.seed;
1542	>	r ^= r << 13;                // same xorshift as WorkQueues
1543	>	r ^= r >>> 17;
1544	>	z.seed = r ^ (r << 5);
1545	>	}
1546	>	else if ((ps = plock) < 0)
1547	>	throw new RejectedExecutionException();
1548	>	else if (ps == 0 || (ws = workQueues) == null ||
1549	>	(m = ws.length - 1) < 0)
1550	>	initWorkers();
1551	>	else if ((q = ws[k = r & m & SQMASK]) != null) {
1552	>	if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
1553	>	ForkJoinTask<?>[] a = q.array;
1554	>	int s = q.top;
1555	>	boolean submitted = false;
1556	>	try {                      // locked version of push
1557	>	if ((a != null && a.length > s + 1 - q.base) ||
1558	>	(a = q.growArray()) != null) {   // must presize
1559	>	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
1560	>	U.putOrderedObject(a, j, task);
1561	>	q.top = s + 1;
1562	>	submitted = true;
1563	>	}
1564	>	} finally {
1565	>	q.qlock = 0;  // unlock
1566	>	}
1567	>	if (submitted) {
1568	>	signalWork(q);
1569	>	return;
1570		}
1335	–	} finally {
1336	–	lk.unlock();
1571		}
1572	+	r = 0; // move on failure
1573		}
1574	<	if (q != null) {
1575	<	if (q.trySharedPush(task)) {
1576	<	signalWork();
1577	<	return;
1578	<	}
1579	<	else if (m < parallelism - 1 && m < (runState & SMASK)) {
1580	<	Mutex lock = this.lock;
1581	<	lock.lock();             // block until lock free
1582	<	int g = growHints;
1583	<	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;
1574	>	else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1575	>	q = new WorkQueue(this, null, SHARED_QUEUE, r);
1576	>	if (((ps = plock) & PL_LOCK) != 0 ||
1577	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1578	>	ps = acquirePlock();
1579	>	if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1580	>	ws[k] = q;
1581	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1582	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1583	>	releasePlock(nps);
1584		}
1585	+	else
1586	+	r = 0; // try elsewhere while lock held
1587		}
1588		}
1589
#	Line 1371 | Line 1598 | public class ForkJoinPool extends Abstra
1598		}
1599
1600		/**
1601	<	* Tries to activate or create a worker if too few are active.
1601	>	* Tries to create or activate a worker if too few are active.
1602	>	*
1603	>	* @param q the (non-null) queue holding tasks to be signalled
1604		*/
1605	<	final void signalWork() {
1606	<	long c; int u;
1607	<	while ((u = (int)((c = ctl) >>> 32)) < 0) {     // too few active
1608	<	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1609	<	if ((e = (int)c) > 0) {                     // at least one waiting
1610	<	if (ws != null && (i = e & SMASK) < ws.length &&
1605	>	final void signalWork(WorkQueue q) {
1606	>	int hint = q.poolIndex;
1607	>	long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p;
1608	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1609	>	if ((e = (int)c) > 0) {
1610	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1611		(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1612		long nc = (((long)(w.nextWait & E_MASK)) |
1613		((long)(u + UAC_UNIT) << 32));
1614		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1615	+	w.hint = hint;
1616		w.eventCount = (e + E_SEQ) & E_MASK;
1617		if ((p = w.parker) != null)
1618	<	U.unpark(p);                // activate and release
1618	>	U.unpark(p);
1619		break;
1620		}
1621	+	if (q.top - q.base <= 0)
1622	+	break;
1623		}
1624		else
1625		break;
1626		}
1627	<	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1628	<	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1629	<	((u + UAC_UNIT) & UAC_MASK)) << 32;
1398	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1399	<	addWorker();
1400	<	break;
1401	<	}
1402	<	}
1403	<	else
1627	>	else {
1628	>	if ((short)u < 0)
1629	>	tryAddWorker();
1630		break;
1631	+	}
1632		}
1633		}
1634
1635	+	// Scanning for tasks
1636	+
1637		/**
1638	<	* 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
1638	>	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1639		*/
1640	<	final boolean tryCompensate() {
1641	<	WorkQueue w; Thread p;
1642	<	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;
1640	>	final void runWorker(WorkQueue w) {
1641	>	w.growArray(); // allocate queue
1642	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1643		}
1644
1449	–	// Scanning for tasks
1450	–
1645		/**
1646		* Scans for and, if found, returns one task, else possibly
1647		* inactivates the worker. This method operates on single reads of
1648	<	* volatile state and is designed to be re-invoked continuously in
1649	<	* part because it returns upon detecting inconsistencies,
1648	>	* volatile state and is designed to be re-invoked continuously,
1649	>	* in part because it returns upon detecting inconsistencies,
1650		* contention, or state changes that indicate possible success on
1651		* re-invocation.
1652		*
1653	<	* The scan searches for tasks across queues, randomly selecting
1654	<	* the first #queues probes, favoring steals over submissions
1655	<	* (by exploiting even/odd indexing), and then performing a
1656	<	* circular sweep of all queues.  The scan terminates upon either
1657	<	* finding a non-empty queue, or completing a full sweep. If the
1658	<	* worker is not inactivated, it takes and returns a task from
1659	<	* this queue.  On failure to find a task, we take one of the
1660	<	* following actions, after which the caller will retry calling
1661	<	* this method unless terminated.
1653	>	* The scan searches for tasks across queues (starting at a random
1654	>	* index, and relying on registerWorker to irregularly scatter
1655	>	* them within array to avoid bias), checking each at least twice.
1656	>	* The scan terminates upon either finding a non-empty queue, or
1657	>	* completing the sweep. If the worker is not inactivated, it
1658	>	* takes and returns a task from this queue. Otherwise, if not
1659	>	* activated, it signals workers (that may include itself) and
1660	>	* returns so caller can retry. Also returns for true if the
1661	>	* worker array may have changed during an empty scan.  On failure
1662	>	* to find a task, we take one of the following actions, after
1663	>	* which the caller will retry calling this method unless
1664	>	* terminated.
1665		*
1666		* * If pool is terminating, terminate the worker.
1667		*
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	–	*
1668		* * If not already enqueued, try to inactivate and enqueue the
1669	<	* worker on wait queue.
1669	>	* worker on wait queue. Or, if inactivating has caused the pool
1670	>	* to be quiescent, relay to idleAwaitWork to possibly shrink
1671	>	* pool.
1672		*
1673	<	* * If already enqueued and none of the above apply, either park
1674	<	* awaiting signal, or if this is the most recent waiter and pool
1675	<	* is quiescent, relay to idleAwaitWork to check for termination
1676	<	* and possibly shrink pool.
1673	>	* * If already enqueued and none of the above apply, possibly
1674	>	* park awaiting signal, else lingering to help scan and signal.
1675	>	*
1676	>	* * If a non-empty queue discovered or left as a hint,
1677	>	* help wake up other workers before return
1678		*
1679		* @param w the worker (via its WorkQueue)
1680	<	* @return a task or null of none found
1680	>	* @return a task or null if none found
1681		*/
1682		private final ForkJoinTask<?> scan(WorkQueue w) {
1683	<	boolean swept = false;               // true after full empty scan
1684	<	WorkQueue[] ws;                      // volatile read order matters
1685	<	int r = w.seed, ec = w.eventCount;   // ec is negative if inactive
1686	<	int rs = runState, m = rs & SMASK;
1687	<	if ((ws = workQueues) != null && ws.length > m) { // consistency check
1688	<	for (int k = 0, j = -1 - m; ; ++j) {
1689	<	WorkQueue q; int b;
1690	<	if (j < 0) {                 // random probes while j negative
1691	<	r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) | (j & 1);
1692	<	}                            // worker (not submit) for odd j
1693	<	else                         // cyclic scan when j >= 0
1694	<	k += 7;                  // step 7 reduces array packing bias
1695	<	if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
1696	<	ForkJoinTask<?> t = (ec >= 0) ? q.pollAt(b) : null;
1697	<	w.seed = r;              // save seed for next scan
1698	<	if (t != null)
1699	<	return t;
1700	<	break;
1701	<	}
1702	<	else if (j - m > m) {
1703	<	if (rs == runState)      // staleness check
1704	<	swept = true;
1705	<	break;
1683	>	WorkQueue[] ws; int m;
1684	>	int ps = plock;                          // read plock before ws
1685	>	if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1686	>	int ec = w.eventCount;               // ec is negative if inactive
1687	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1688	>	w.hint = -1;                         // update seed and clear hint
1689	>	int j = ((m + m + 1) | MIN_SCAN) & MAX_SCAN;
1690	>	do {
1691	>	WorkQueue q; ForkJoinTask<?>[] a; int b;
1692	>	if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 &&
1693	>	(a = q.array) != null) {     // probably nonempty
1694	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1695	>	ForkJoinTask<?> t = (ForkJoinTask<?>)
1696	>	U.getObjectVolatile(a, i);
1697	>	if (q.base == b && ec >= 0 && t != null &&
1698	>	U.compareAndSwapObject(a, i, t, null)) {
1699	>	if ((q.base = b + 1) - q.top < 0)
1700	>	signalWork(q);
1701	>	return t;                // taken
1702	>	}
1703	>	else if ((ec < 0 || j < m) && (int)(ctl >> AC_SHIFT) <= 0) {
1704	>	w.hint = (r + j) & m;    // help signal below
1705	>	break;                   // cannot take
1706	>	}
1707		}
1708	<	}
1708	>	} while (--j >= 0);
1709
1710	<	// Decode ctl on empty scan
1711	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1712	<	if (e < 0)                       // pool is terminating
1713	<	w.runState = -1;
1714	<	else if (!swept) {               // try to release a waiter
1715	<	WorkQueue v; Thread p;
1716	<	if (e > 0 && a < 0 && (v = ws[e & m]) != null &&
1717	<	v.eventCount == (e | INT_SIGN)) {
1718	<	long nc = ((long)(v.nextWait & E_MASK) |
1719	<	((c + AC_UNIT) & (AC_MASK|TC_MASK)));
1720	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1710	>	int h, e, ns; long c, sc; WorkQueue q;
1711	>	if ((ns = w.nsteals) != 0) {
1712	>	if (U.compareAndSwapLong(this, STEALCOUNT,
1713	>	sc = stealCount, sc + ns))
1714	>	w.nsteals = 0;               // collect steals and rescan
1715	>	}
1716	>	else if (plock != ps)                // consistency check
1717	>	;                                // skip
1718	>	else if ((e = (int)(c = ctl)) < 0)
1719	>	w.qlock = -1;                    // pool is terminating
1720	>	else {
1721	>	if ((h = w.hint) < 0) {
1722	>	if (ec >= 0) {               // try to enqueue/inactivate
1723	>	long nc = (((long)ec |
1724	>	((c - AC_UNIT) & (AC_MASK|TC_MASK))));
1725	>	w.nextWait = e;          // link and mark inactive
1726	>	w.eventCount = ec | INT_SIGN;
1727	>	if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc))
1728	>	w.eventCount = ec;   // unmark on CAS failure
1729	>	else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK))
1730	>	idleAwaitWork(w, nc, c);
1731	>	}
1732	>	else if (w.eventCount < 0 && !tryTerminate(false, false) &&
1733	>	ctl == c) {         // block
1734	>	Thread wt = Thread.currentThread();
1735	>	Thread.interrupted();    // clear status
1736	>	U.putObject(wt, PARKBLOCKER, this);
1737	>	w.parker = wt;           // emulate LockSupport.park
1738	>	if (w.eventCount < 0)    // recheck
1739	>	U.park(false, 0L);
1740	>	w.parker = null;
1741	>	U.putObject(wt, PARKBLOCKER, null);
1742	>	}
1743	>	}
1744	>	if ((h >= 0 || (h = w.hint) >= 0) &&
1745	>	(ws = workQueues) != null && h < ws.length &&
1746	>	(q = ws[h]) != null) {      // signal others before retry
1747	>	WorkQueue v; Thread p; int u, i, s;
1748	>	for (int n = (config & SMASK) >>> 1;;) {
1749	>	int idleCount = (w.eventCount < 0) ? 0 : -1;
1750	>	if (((s = idleCount - q.base + q.top) <= n &&
1751	>	(n = s) <= 0) ||
1752	>	(u = (int)((c = ctl) >>> 32)) >= 0 ||
1753	>	(e = (int)c) <= 0 || m < (i = e & SMASK) ||
1754	>	(v = ws[i]) == null)
1755	>	break;
1756	>	long nc = (((long)(v.nextWait & E_MASK)) |
1757	>	((long)(u + UAC_UNIT) << 32));
1758	>	if (v.eventCount != (e | INT_SIGN) ||
1759	>	!U.compareAndSwapLong(this, CTL, c, nc))
1760	>	break;
1761	>	v.hint = h;
1762		v.eventCount = (e + E_SEQ) & E_MASK;
1763		if ((p = v.parker) != null)
1764		U.unpark(p);
1765	+	if (--n <= 0)
1766	+	break;
1767		}
1768		}
1769		}
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	–	}
1770		}
1771		return null;
1772		}
1773
1774		/**
1775	<	* If inactivating worker w has caused pool to become quiescent,
1776	<	* checks for pool termination, and, so long as this is not the
1777	<	* only worker, waits for event for up to SHRINK_RATE nanosecs.
1778	<	* On timeout, if ctl has not changed, terminates the worker,
1779	<	* which will in turn wake up another worker to possibly repeat
1780	<	* this process.
1775	>	* If inactivating worker w has caused the pool to become
1776	>	* quiescent, checks for pool termination, and, so long as this is
1777	>	* not the only worker, waits for event for up to a given
1778	>	* duration.  On timeout, if ctl has not changed, terminates the
1779	>	* worker, which will in turn wake up another worker to possibly
1780	>	* repeat this process.
1781		*
1782		* @param w the calling worker
1783	+	* @param currentCtl the ctl value triggering possible quiescence
1784	+	* @param prevCtl the ctl value to restore if thread is terminated
1785		*/
1786	<	private void idleAwaitWork(WorkQueue w) {
1787	<	long c; int nw, ec;
1788	<	if (!tryTerminate(false, false) &&
1789	<	(int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
1790	<	(ec = w.eventCount) == ((int)c | INT_SIGN) &&
1791	<	(nw = w.nextWait) != 0) {
1792	<	long nc = ((long)(nw & E_MASK) | // ctl to restore on timeout
1793	<	((c + AC_UNIT) & AC_MASK) | (c & TC_MASK));
1592	<	ForkJoinWorkerThread wt = w.owner;
1593	<	while (ctl == c) {
1594	<	long startTime = System.nanoTime();
1786	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1787	>	if (w != null && w.eventCount < 0 &&
1788	>	!tryTerminate(false, false) && (int)prevCtl != 0) {
1789	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1790	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1791	>	long deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP;
1792	>	Thread wt = Thread.currentThread();
1793	>	while (ctl == currentCtl) {
1794		Thread.interrupted();  // timed variant of version in scan()
1795		U.putObject(wt, PARKBLOCKER, this);
1796		w.parker = wt;
1797	<	if (ctl == c)
1798	<	U.park(false, SHRINK_RATE);
1797	>	if (ctl == currentCtl)
1798	>	U.park(false, parkTime);
1799		w.parker = null;
1800		U.putObject(wt, PARKBLOCKER, null);
1801	<	if (ctl != c)
1801	>	if (ctl != currentCtl)
1802		break;
1803	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1804	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1805	<	w.eventCount = (ec + E_SEQ) | E_MASK;
1806	<	w.runState = -1;          // shrink
1803	>	if (deadline - System.nanoTime() <= 0L &&
1804	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1805	>	w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1806	>	w.qlock = -1;   // shrink
1807		break;
1808		}
1809		}
#	Line 1612 | Line 1811 | public class ForkJoinPool extends Abstra
1811		}
1812
1813		/**
1814	+	* Scans through queues looking for work while joining a task; if
1815	+	* any present, signals. May return early if more signalling is
1816	+	* detectably unneeded.
1817	+	*
1818	+	* @param task return early if done
1819	+	* @param origin an index to start scan
1820	+	*/
1821	+	private void helpSignal(ForkJoinTask<?> task, int origin) {
1822	+	WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s;
1823	+	if (task != null && task.status >= 0 &&
1824	+	(u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 &&
1825	+	(ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1826	+	outer: for (int k = origin, j = m; j >= 0; --j) {
1827	+	WorkQueue q = ws[k++ & m];
1828	+	for (int n = m;;) { // limit to at most m signals
1829	+	if (task.status < 0)
1830	+	break outer;
1831	+	if (q == null ||
1832	+	((s = -q.base + q.top) <= n && (n = s) <= 0))
1833	+	break;
1834	+	if ((u = (int)((c = ctl) >>> 32)) >= 0 ||
1835	+	(e = (int)c) <= 0 || m < (i = e & SMASK) ||
1836	+	(w = ws[i]) == null)
1837	+	break outer;
1838	+	long nc = (((long)(w.nextWait & E_MASK)) |
1839	+	((long)(u + UAC_UNIT) << 32));
1840	+	if (w.eventCount != (e | INT_SIGN))
1841	+	break outer;
1842	+	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1843	+	w.eventCount = (e + E_SEQ) & E_MASK;
1844	+	if ((p = w.parker) != null)
1845	+	U.unpark(p);
1846	+	if (--n <= 0)
1847	+	break;
1848	+	}
1849	+	}
1850	+	}
1851	+	}
1852	+	}
1853	+
1854	+	/**
1855		* Tries to locate and execute tasks for a stealer of the given
1856		* task, or in turn one of its stealers, Traces currentSteal ->
1857		* currentJoin links looking for a thread working on a descendant
#	Line 1622 | Line 1862 | public class ForkJoinPool extends Abstra
1862		* leaves hints in workers to speed up subsequent calls. The
1863		* implementation is very branchy to cope with potential
1864		* inconsistencies or loops encountering chains that are stale,
1865	<	* unknown, or of length greater than MAX_HELP_DEPTH links.  All
1626	<	* of these cases are dealt with by just retrying by caller.
1865	>	* unknown, or so long that they are likely cyclic.
1866		*
1867		* @param joiner the joining worker
1868		* @param task the task to join
1869	<	* @return true if found or ran a task (and so is immediately retryable)
1869	>	* @return 0 if no progress can be made, negative if task
1870	>	* known complete, else positive
1871		*/
1872	<	final boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1873	<	ForkJoinTask<?> subtask;    // current target
1874	<	boolean progress = false;
1875	<	int depth = 0;              // current chain depth
1876	<	int m = runState & SMASK;
1877	<	WorkQueue[] ws = workQueues;
1878	<
1879	<	if (ws != null && ws.length > m && (subtask = task).status >= 0) {
1880	<	outer:for (WorkQueue j = joiner;;) {
1881	<	// Try to find the stealer of subtask, by first using hint
1882	<	WorkQueue stealer = null;
1883	<	WorkQueue v = ws[j.stealHint & m];
1884	<	if (v != null && v.currentSteal == subtask)
1885	<	stealer = v;
1886	<	else {
1887	<	for (int i = 1; i <= m; i += 2) {
1888	<	if ((v = ws[i]) != null && v.currentSteal == subtask) {
1889	<	stealer = v;
1890	<	j.stealHint = i; // save hint
1891	<	break;
1872	>	private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1873	>	int stat = 0, steps = 0;                    // bound to avoid cycles
1874	>	if (joiner != null && task != null) {       // hoist null checks
1875	>	restart: for (;;) {
1876	>	ForkJoinTask<?> subtask = task;     // current target
1877	>	for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
1878	>	WorkQueue[] ws; int m, s, h;
1879	>	if ((s = task.status) < 0) {
1880	>	stat = s;
1881	>	break restart;
1882	>	}
1883	>	if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1884	>	break restart;              // shutting down
1885	>	if ((v = ws[h = (j.hint | 1) & m]) == null ||
1886	>	v.currentSteal != subtask) {
1887	>	for (int origin = h;;) {    // find stealer
1888	>	if (((h = (h + 2) & m) & 15) == 1 &&
1889	>	(subtask.status < 0 || j.currentJoin != subtask))
1890	>	continue restart;   // occasional staleness check
1891	>	if ((v = ws[h]) != null &&
1892	>	v.currentSteal == subtask) {
1893	>	j.hint = h;        // save hint
1894	>	break;
1895	>	}
1896	>	if (h == origin)
1897	>	break restart;      // cannot find stealer
1898	>	}
1899	>	}
1900	>	for (;;) { // help stealer or descend to its stealer
1901	>	ForkJoinTask[] a;  int b;
1902	>	if (subtask.status < 0)     // surround probes with
1903	>	continue restart;       //   consistency checks
1904	>	if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1905	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1906	>	ForkJoinTask<?> t =
1907	>	(ForkJoinTask<?>)U.getObjectVolatile(a, i);
1908	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1909	>	v.currentSteal != subtask)
1910	>	continue restart;   // stale
1911	>	stat = 1;               // apparent progress
1912	>	if (t != null && v.base == b &&
1913	>	U.compareAndSwapObject(a, i, t, null)) {
1914	>	v.base = b + 1;     // help stealer
1915	>	joiner.runSubtask(t);
1916	>	}
1917	>	else if (v.base == b && ++steps == MAX_HELP)
1918	>	break restart;      // v apparently stalled
1919	>	}
1920	>	else {                      // empty -- try to descend
1921	>	ForkJoinTask<?> next = v.currentJoin;
1922	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1923	>	v.currentSteal != subtask)
1924	>	continue restart;   // stale
1925	>	else if (next == null || ++steps == MAX_HELP)
1926	>	break restart;      // dead-end or maybe cyclic
1927	>	else {
1928	>	subtask = next;
1929	>	j = v;
1930	>	break;
1931	>	}
1932		}
1933		}
1934	<	if (stealer == null)
1934	>	}
1935	>	}
1936	>	}
1937	>	return stat;
1938	>	}
1939	>
1940	>	/**
1941	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1942	>	* and run tasks within the target's computation.
1943	>	*
1944	>	* @param task the task to join
1945	>	* @param mode if shared, exit upon completing any task
1946	>	* if all workers are active
1947	>	*
1948	>	*/
1949	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1950	>	WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1951	>	if (task != null && (ws = workQueues) != null &&
1952	>	(m = ws.length - 1) >= 0) {
1953	>	for (int j = 1, origin = j;;) {
1954	>	if ((s = task.status) < 0)
1955	>	return s;
1956	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1957	>	origin = j;
1958	>	if (mode == SHARED_QUEUE &&
1959	>	((u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0))
1960		break;
1961		}
1962	+	else if ((j = (j + 2) & m) == origin)
1963	+	break;
1964	+	}
1965	+	}
1966	+	return 0;
1967	+	}
1968
1969	<	for (WorkQueue q = stealer;;) { // Try to help stealer
1970	<	ForkJoinTask<?> t; int b;
1971	<	if (task.status < 0)
1972	<	break outer;
1973	<	if ((b = q.base) - q.top < 0) {
1974	<	progress = true;
1975	<	if (subtask.status < 0)
1976	<	break outer;               // stale
1977	<	if ((t = q.pollAt(b)) != null) {
1978	<	stealer.stealHint = joiner.poolIndex;
1979	<	joiner.runSubtask(t);
1969	>	/**
1970	>	* Tries to decrement active count (sometimes implicitly) and
1971	>	* possibly release or create a compensating worker in preparation
1972	>	* for blocking. Fails on contention or termination. Otherwise,
1973	>	* adds a new thread if no idle workers are available and pool
1974	>	* may become starved.
1975	>	*/
1976	>	final boolean tryCompensate() {
1977	>	int pc = config & SMASK, e, i, tc; long c;
1978	>	WorkQueue[] ws; WorkQueue w; Thread p;
1979	>	if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) {
1980	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1981	>	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1982	>	long nc = ((long)(w.nextWait & E_MASK) |
1983	>	(c & (AC_MASK|TC_MASK)));
1984	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1985	>	w.eventCount = (e + E_SEQ) & E_MASK;
1986	>	if ((p = w.parker) != null)
1987	>	U.unpark(p);
1988	>	return true;   // replace with idle worker
1989	>	}
1990	>	}
1991	>	else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1992	>	(int)(c >> AC_SHIFT) + pc > 1) {
1993	>	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1994	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1995	>	return true;   // no compensation
1996	>	}
1997	>	else if (tc + pc < MAX_CAP) {
1998	>	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1999	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
2000	>	ForkJoinWorkerThreadFactory fac;
2001	>	Throwable ex = null;
2002	>	ForkJoinWorkerThread wt = null;
2003	>	try {
2004	>	if ((fac = factory) != null &&
2005	>	(wt = fac.newThread(this)) != null) {
2006	>	wt.start();
2007	>	return true;
2008		}
2009	+	} catch (Throwable rex) {
2010	+	ex = rex;
2011		}
2012	<	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	<	}
2012	>	deregisterWorker(wt, ex); // clean up and return false
2013		}
2014		}
2015		}
2016	<	return progress;
2016	>	return false;
2017		}
2018
2019		/**
2020	<	* If task is at base of some steal queue, steals and executes it.
2020	>	* Helps and/or blocks until the given task is done.
2021		*
2022		* @param joiner the joining worker
2023		* @param task the task
2024	+	* @return task status on exit
2025		*/
2026	<	final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
2027	<	WorkQueue[] ws;
2028	<	int m = runState & SMASK;
2029	<	if ((ws = workQueues) != null && ws.length > m) {
2030	<	for (int j = 1; j <= m && task.status >= 0; j += 2) {
2031	<	WorkQueue q = ws[j];
2032	<	if (q != null && q.pollFor(task)) {
2033	<	joiner.runSubtask(task);
2034	<	break;
2026	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
2027	>	int s = 0;
2028	>	if (joiner != null && task != null && (s = task.status) >= 0) {
2029	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2030	>	joiner.currentJoin = task;
2031	>	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2032	>	joiner.tryRemoveAndExec(task)); // process local tasks
2033	>	if (s >= 0 && (s = task.status) >= 0) {
2034	>	helpSignal(task, joiner.poolIndex);
2035	>	if ((s = task.status) >= 0 &&
2036	>	(task instanceof CountedCompleter))
2037	>	s = helpComplete(task, LIFO_QUEUE);
2038	>	}
2039	>	while (s >= 0 && (s = task.status) >= 0) {
2040	>	if ((!joiner.isEmpty() ||           // try helping
2041	>	(s = tryHelpStealer(joiner, task)) == 0) &&
2042	>	(s = task.status) >= 0) {
2043	>	helpSignal(task, joiner.poolIndex);
2044	>	if ((s = task.status) >= 0 && tryCompensate()) {
2045	>	if (task.trySetSignal() && (s = task.status) >= 0) {
2046	>	synchronized (task) {
2047	>	if (task.status >= 0) {
2048	>	try {                // see ForkJoinTask
2049	>	task.wait();     //  for explanation
2050	>	} catch (InterruptedException ie) {
2051	>	}
2052	>	}
2053	>	else
2054	>	task.notifyAll();
2055	>	}
2056	>	}
2057	>	long c;                          // re-activate
2058	>	do {} while (!U.compareAndSwapLong
2059	>	(this, CTL, c = ctl, c + AC_UNIT));
2060	>	}
2061		}
2062		}
2063	+	joiner.currentJoin = prevJoin;
2064	+	}
2065	+	return s;
2066	+	}
2067	+
2068	+	/**
2069	+	* Stripped-down variant of awaitJoin used by timed joins. Tries
2070	+	* to help join only while there is continuous progress. (Caller
2071	+	* will then enter a timed wait.)
2072	+	*
2073	+	* @param joiner the joining worker
2074	+	* @param task the task
2075	+	*/
2076	+	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2077	+	int s;
2078	+	if (joiner != null && task != null && (s = task.status) >= 0) {
2079	+	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2080	+	joiner.currentJoin = task;
2081	+	do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2082	+	joiner.tryRemoveAndExec(task));
2083	+	if (s >= 0 && (s = task.status) >= 0) {
2084	+	helpSignal(task, joiner.poolIndex);
2085	+	if ((s = task.status) >= 0 &&
2086	+	(task instanceof CountedCompleter))
2087	+	s = helpComplete(task, LIFO_QUEUE);
2088	+	}
2089	+	if (s >= 0 && joiner.isEmpty()) {
2090	+	do {} while (task.status >= 0 &&
2091	+	tryHelpStealer(joiner, task) > 0);
2092	+	}
2093	+	joiner.currentJoin = prevJoin;
2094		}
2095		}
2096
2097		/**
2098	<	* Returns a non-empty steal queue, if one is found during a random,
2099	<	* then cyclic scan, else null.  This method must be retried by
2100	<	* caller if, by the time it tries to use the queue, it is empty.
2098	>	* Returns a (probably) non-empty steal queue, if one is found
2099	>	* during a random, then cyclic scan, else null.  This method must
2100	>	* be retried by caller if, by the time it tries to use the queue,
2101	>	* it is empty.
2102	>	* @param r a (random) seed for scanning
2103		*/
2104	<	private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
1712	<	int r = w.seed;    // Same idea as scan(), but ignoring submissions
2104	>	private WorkQueue findNonEmptyStealQueue(int r) {
2105		for (WorkQueue[] ws;;) {
2106	<	int m = runState & SMASK;
2107	<	if ((ws = workQueues) == null)
2106	>	int ps = plock, m, n;
2107	>	if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2108		return null;
2109	<	if (ws.length > m) {
2110	<	WorkQueue q;
2111	<	for (int k = 0, j = -1 - m;; ++j) {
2112	<	if (j < 0) {
2113	<	r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5;
2114	<	}
2115	<	else
2116	<	k += 7;
2117	<	if ((q = ws[(k | 1) & m]) != null && q.base - q.top < 0) {
1726	<	w.seed = r;
1727	<	return q;
1728	<	}
1729	<	else if (j - m > m)
2109	>	for (int j = (m + 1) << 2; ;) {
2110	>	WorkQueue q = ws[(((r + j) << 1) | 1) & m];
2111	>	if (q != null && (n = q.base - q.top) < 0) {
2112	>	if (n < -1)
2113	>	signalWork(q);
2114	>	return q;
2115	>	}
2116	>	else if (--j < 0) {
2117	>	if (plock == ps)
2118		return null;
2119	+	break;
2120		}
2121		}
2122		}
#	Line 1741 | Line 2130 | public class ForkJoinPool extends Abstra
2130		*/
2131		final void helpQuiescePool(WorkQueue w) {
2132		for (boolean active = true;;) {
2133	<	w.runLocalTasks();      // exhaust local queue
2134	<	WorkQueue q = findNonEmptyStealQueue(w);
2133	>	ForkJoinTask<?> localTask; // exhaust local queue
2134	>	while ((localTask = w.nextLocalTask()) != null)
2135	>	localTask.doExec();
2136	>	// Similar to loop in scan(), but ignoring submissions
2137	>	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2138		if (q != null) {
2139	<	ForkJoinTask<?> t;
2139	>	ForkJoinTask<?> t; int b;
2140		if (!active) {      // re-establish active count
2141		long c;
2142		active = true;
2143		do {} while (!U.compareAndSwapLong
2144		(this, CTL, c = ctl, c + AC_UNIT));
2145		}
2146	<	if ((t = q.poll()) != null)
2146	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2147		w.runSubtask(t);
2148		}
2149		else {
#	Line 1763 | Line 2155 | public class ForkJoinPool extends Abstra
2155		}
2156		else
2157		c = ctl;        // re-increment on exit
2158	<	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2158	>	if ((int)(c >> AC_SHIFT) + (config & SMASK) == 0) {
2159		do {} while (!U.compareAndSwapLong
2160		(this, CTL, c = ctl, c + AC_UNIT));
2161		break;
#	Line 1779 | Line 2171 | public class ForkJoinPool extends Abstra
2171		*/
2172		final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
2173		for (ForkJoinTask<?> t;;) {
2174	<	WorkQueue q;
2174	>	WorkQueue q; int b;
2175		if ((t = w.nextLocalTask()) != null)
2176		return t;
2177	<	if ((q = findNonEmptyStealQueue(w)) == null)
2177	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2178		return null;
2179	<	if ((t = q.poll()) != null)
2179	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2180		return t;
2181		}
2182		}
2183
2184		/**
2185	<	* Returns the approximate (non-atomic) number of idle threads per
2186	<	* active thread to offset steal queue size for method
2187	<	* ForkJoinTask.getSurplusQueuedTaskCount().
2188	<	*/
2189	<	final int idlePerActive() {
2190	<	// Approximate at powers of two for small values, saturate past 4
2191	<	int p = parallelism;
2192	<	int a = p + (int)(ctl >> AC_SHIFT);
2193	<	return (a > (p >>>= 1) ? 0 :
2194	<	a > (p >>>= 1) ? 1 :
2195	<	a > (p >>>= 1) ? 2 :
2196	<	a > (p >>>= 1) ? 4 :
2197	<	8);
2185	>	* Returns a cheap heuristic guide for task partitioning when
2186	>	* programmers, frameworks, tools, or languages have little or no
2187	>	* idea about task granularity.  In essence by offering this
2188	>	* method, we ask users only about tradeoffs in overhead vs
2189	>	* expected throughput and its variance, rather than how finely to
2190	>	* partition tasks.
2191	>	*
2192	>	* In a steady state strict (tree-structured) computation, each
2193	>	* thread makes available for stealing enough tasks for other
2194	>	* threads to remain active. Inductively, if all threads play by
2195	>	* the same rules, each thread should make available only a
2196	>	* constant number of tasks.
2197	>	*
2198	>	* The minimum useful constant is just 1. But using a value of 1
2199	>	* would require immediate replenishment upon each steal to
2200	>	* maintain enough tasks, which is infeasible.  Further,
2201	>	* partitionings/granularities of offered tasks should minimize
2202	>	* steal rates, which in general means that threads nearer the top
2203	>	* of computation tree should generate more than those nearer the
2204	>	* bottom. In perfect steady state, each thread is at
2205	>	* approximately the same level of computation tree. However,
2206	>	* producing extra tasks amortizes the uncertainty of progress and
2207	>	* diffusion assumptions.
2208	>	*
2209	>	* So, users will want to use values larger, but not much larger
2210	>	* than 1 to both smooth over transient shortages and hedge
2211	>	* against uneven progress; as traded off against the cost of
2212	>	* extra task overhead. We leave the user to pick a threshold
2213	>	* value to compare with the results of this call to guide
2214	>	* decisions, but recommend values such as 3.
2215	>	*
2216	>	* When all threads are active, it is on average OK to estimate
2217	>	* surplus strictly locally. In steady-state, if one thread is
2218	>	* maintaining say 2 surplus tasks, then so are others. So we can
2219	>	* just use estimated queue length.  However, this strategy alone
2220	>	* leads to serious mis-estimates in some non-steady-state
2221	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2222	>	* many of these by further considering the number of "idle"
2223	>	* threads, that are known to have zero queued tasks, so
2224	>	* compensate by a factor of (#idle/#active) threads.
2225	>	*
2226	>	* Note: The approximation of #busy workers as #active workers is
2227	>	* not very good under current signalling scheme, and should be
2228	>	* improved.
2229	>	*/
2230	>	static int getSurplusQueuedTaskCount() {
2231	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2232	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2233	>	int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK;
2234	>	int n = (q = wt.workQueue).top - q.base;
2235	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2236	>	return n - (a > (p >>>= 1) ? 0 :
2237	>	a > (p >>>= 1) ? 1 :
2238	>	a > (p >>>= 1) ? 2 :
2239	>	a > (p >>>= 1) ? 4 :
2240	>	8);
2241	>	}
2242	>	return 0;
2243		}
2244
2245		//  Termination
#	Line 1822 | Line 2259 | public class ForkJoinPool extends Abstra
2259		* @return true if now terminating or terminated
2260		*/
2261		private boolean tryTerminate(boolean now, boolean enable) {
2262	<	Mutex lock = this.lock;
2262	>	if (this == commonPool)                     // cannot shut down
2263	>	return false;
2264		for (long c;;) {
2265		if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2266	<	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2267	<	lock.lock();                    // don't need try/finally
2268	<	termination.signalAll();        // signal when 0 workers
2269	<	lock.unlock();
2266	>	if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) {
2267	>	synchronized (this) {
2268	>	notifyAll();                // signal when 0 workers
2269	>	}
2270		}
2271		return true;
2272		}
2273	<	if (runState >= 0) {                    // not yet enabled
2273	>	if (plock >= 0) {                       // not yet enabled
2274	>	int ps;
2275		if (!enable)
2276		return false;
2277	<	lock.lock();
2278	<	runState |= SHUTDOWN;
2279	<	lock.unlock();
2277	>	if (((ps = plock) & PL_LOCK) != 0 ||
2278	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2279	>	ps = acquirePlock();
2280	>	if (!U.compareAndSwapInt(this, PLOCK, ps, SHUTDOWN))
2281	>	releasePlock(SHUTDOWN);
2282		}
2283		if (!now) {                             // check if idle & no tasks
2284	<	if ((int)(c >> AC_SHIFT) != -parallelism ||
2284	>	if ((int)(c >> AC_SHIFT) != -(config & SMASK) ||
2285		hasQueuedSubmissions())
2286		return false;
2287		// Check for unqueued inactive workers. One pass suffices.
#	Line 1856 | Line 2297 | public class ForkJoinPool extends Abstra
2297		for (int pass = 0; pass < 3; ++pass) {
2298		WorkQueue[] ws = workQueues;
2299		if (ws != null) {
2300	<	WorkQueue w;
2300	>	WorkQueue w; Thread wt;
2301		int n = ws.length;
2302		for (int i = 0; i < n; ++i) {
2303		if ((w = ws[i]) != null) {
2304	<	w.runState = -1;
2304	>	w.qlock = -1;
2305		if (pass > 0) {
2306		w.cancelAll();
2307	<	if (pass > 1)
2308	<	w.interruptOwner();
2307	>	if (pass > 1 && (wt = w.owner) != null) {
2308	>	if (!wt.isInterrupted()) {
2309	>	try {
2310	>	wt.interrupt();
2311	>	} catch (SecurityException ignore) {
2312	>	}
2313	>	}
2314	>	U.unpark(wt);
2315	>	}
2316		}
2317		}
2318		}
#	Line 1879 | Line 2327 | public class ForkJoinPool extends Abstra
2327		if (w.eventCount == (e | INT_SIGN) &&
2328		U.compareAndSwapLong(this, CTL, cc, nc)) {
2329		w.eventCount = (e + E_SEQ) & E_MASK;
2330	<	w.runState = -1;
2330	>	w.qlock = -1;
2331		if ((p = w.parker) != null)
2332		U.unpark(p);
2333		}
#	Line 1890 | Line 2338 | public class ForkJoinPool extends Abstra
2338		}
2339		}
2340
2341	+	// external operations on common pool
2342	+
2343	+	/**
2344	+	* Returns common pool queue for a thread that has submitted at
2345	+	* least one task.
2346	+	*/
2347	+	static WorkQueue commonSubmitterQueue() {
2348	+	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2349	+	return ((z = submitters.get()) != null &&
2350	+	(p = commonPool) != null &&
2351	+	(ws = p.workQueues) != null &&
2352	+	(m = ws.length - 1) >= 0) ?
2353	+	ws[m & z.seed & SQMASK] : null;
2354	+	}
2355	+
2356	+	/**
2357	+	* Tries to pop the given task from submitter's queue in common pool.
2358	+	*/
2359	+	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2360	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2361	+	ForkJoinTask<?>[] a;  int m, s;
2362	+	if (t != null &&
2363	+	(z = submitters.get()) != null &&
2364	+	(p = commonPool) != null &&
2365	+	(ws = p.workQueues) != null &&
2366	+	(m = ws.length - 1) >= 0 &&
2367	+	(q = ws[m & z.seed & SQMASK]) != null &&
2368	+	(s = q.top) != q.base &&
2369	+	(a = q.array) != null) {
2370	+	long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2371	+	if (U.getObject(a, j) == t &&
2372	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2373	+	if (q.array == a && q.top == s && // recheck
2374	+	U.compareAndSwapObject(a, j, t, null)) {
2375	+	q.top = s - 1;
2376	+	q.qlock = 0;
2377	+	return true;
2378	+	}
2379	+	q.qlock = 0;
2380	+	}
2381	+	}
2382	+	return false;
2383	+	}
2384	+
2385	+	/**
2386	+	* Tries to pop and run local tasks within the same computation
2387	+	* as the given root. On failure, tries to help complete from
2388	+	* other queues via helpComplete.
2389	+	*/
2390	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2391	+	ForkJoinTask<?>[] a; int m;
2392	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2393	+	root != null && root.status >= 0) {
2394	+	for (;;) {
2395	+	int s, u; Object o; CountedCompleter<?> task = null;
2396	+	if ((s = q.top) - q.base > 0) {
2397	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2398	+	if ((o = U.getObject(a, j)) != null &&
2399	+	(o instanceof CountedCompleter)) {
2400	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2401	+	do {
2402	+	if (r == root) {
2403	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2404	+	if (q.array == a && q.top == s &&
2405	+	U.compareAndSwapObject(a, j, t, null)) {
2406	+	q.top = s - 1;
2407	+	task = t;
2408	+	}
2409	+	q.qlock = 0;
2410	+	}
2411	+	break;
2412	+	}
2413	+	} while ((r = r.completer) != null);
2414	+	}
2415	+	}
2416	+	if (task != null)
2417	+	task.doExec();
2418	+	if (root.status < 0 ||
2419	+	(u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0)
2420	+	break;
2421	+	if (task == null) {
2422	+	helpSignal(root, q.poolIndex);
2423	+	if (root.status >= 0)
2424	+	helpComplete(root, SHARED_QUEUE);
2425	+	break;
2426	+	}
2427	+	}
2428	+	}
2429	+	}
2430	+
2431	+	/**
2432	+	* Tries to help execute or signal availability of the given task
2433	+	* from submitter's queue in common pool.
2434	+	*/
2435	+	static void externalHelpJoin(ForkJoinTask<?> t) {
2436	+	// Some hard-to-avoid overlap with tryExternalUnpush
2437	+	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2438	+	ForkJoinTask<?>[] a;  int m, s, n;
2439	+	if (t != null &&
2440	+	(z = submitters.get()) != null &&
2441	+	(p = commonPool) != null &&
2442	+	(ws = p.workQueues) != null &&
2443	+	(m = ws.length - 1) >= 0 &&
2444	+	(q = ws[m & z.seed & SQMASK]) != null &&
2445	+	(a = q.array) != null) {
2446	+	int am = a.length - 1;
2447	+	if ((s = q.top) != q.base) {
2448	+	long j = ((am & (s - 1)) << ASHIFT) + ABASE;
2449	+	if (U.getObject(a, j) == t &&
2450	+	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2451	+	if (q.array == a && q.top == s &&
2452	+	U.compareAndSwapObject(a, j, t, null)) {
2453	+	q.top = s - 1;
2454	+	q.qlock = 0;
2455	+	t.doExec();
2456	+	}
2457	+	else
2458	+	q.qlock = 0;
2459	+	}
2460	+	}
2461	+	if (t.status >= 0) {
2462	+	if (t instanceof CountedCompleter)
2463	+	p.externalHelpComplete(q, t);
2464	+	else
2465	+	p.helpSignal(t, q.poolIndex);
2466	+	}
2467	+	}
2468	+	}
2469	+
2470	+	/**
2471	+	* Restricted version of helpQuiescePool for external callers
2472	+	*/
2473	+	static void externalHelpQuiescePool() {
2474	+	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2475	+	if ((p = commonPool) != null &&
2476	+	(q = p.findNonEmptyStealQueue(1)) != null &&
2477	+	(b = q.base) - q.top < 0 &&
2478	+	(t = q.pollAt(b)) != null)
2479	+	t.doExec();
2480	+	}
2481	+
2482		// Exported methods
2483
2484		// Constructors
#	Line 1959 | Line 2548 | public class ForkJoinPool extends Abstra
2548		checkPermission();
2549		if (factory == null)
2550		throw new NullPointerException();
2551	<	if (parallelism <= 0 || parallelism > POOL_MAX)
2551	>	if (parallelism <= 0 || parallelism > MAX_CAP)
2552		throw new IllegalArgumentException();
1964	–	this.parallelism = parallelism;
2553		this.factory = factory;
2554		this.ueh = handler;
2555	<	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
1968	<	this.growHints = 1;
2555	>	this.config = parallelism | (asyncMode ? (FIFO_QUEUE << 16) : 0);
2556		long np = (long)(-parallelism); // offset ctl counts
2557		this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2558	<	// 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();
2558	>	int pn = nextPoolId();
2559		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2560	<	sb.append(poolNumberGenerator.incrementAndGet());
2560	>	sb.append(Integer.toString(pn));
2561		sb.append("-worker-");
2562		this.workerNamePrefix = sb.toString();
2563		}
2564
2565	+	/**
2566	+	* Constructor for common pool, suitable only for static initialization.
2567	+	* Basically the same as above, but uses smallest possible initial footprint.
2568	+	*/
2569	+	ForkJoinPool(int parallelism, long ctl,
2570	+	ForkJoinWorkerThreadFactory factory,
2571	+	Thread.UncaughtExceptionHandler handler) {
2572	+	this.config = parallelism;
2573	+	this.ctl = ctl;
2574	+	this.factory = factory;
2575	+	this.ueh = handler;
2576	+	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2577	+	}
2578	+
2579	+	/**
2580	+	* Returns the common pool instance.
2581	+	*
2582	+	* @return the common pool instance
2583	+	*/
2584	+	public static ForkJoinPool commonPool() {
2585	+	// assert commonPool != null : "static init error";
2586	+	return commonPool;
2587	+	}
2588	+
2589		// Execution methods
2590
2591		/**
#	Line 2004 | Line 2605 | public class ForkJoinPool extends Abstra
2605		*         scheduled for execution
2606		*/
2607		public <T> T invoke(ForkJoinTask<T> task) {
2608	<	doSubmit(task);
2608	>	if (task == null)
2609	>	throw new NullPointerException();
2610	>	externalPush(task);
2611		return task.join();
2612		}
2613
#	Line 2017 | Line 2620 | public class ForkJoinPool extends Abstra
2620		*         scheduled for execution
2621		*/
2622		public void execute(ForkJoinTask<?> task) {
2623	<	doSubmit(task);
2623	>	if (task == null)
2624	>	throw new NullPointerException();
2625	>	externalPush(task);
2626		}
2627
2628		// AbstractExecutorService methods
#	Line 2034 | Line 2639 | public class ForkJoinPool extends Abstra
2639		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2640		job = (ForkJoinTask<?>) task;
2641		else
2642	<	job = ForkJoinTask.adapt(task, null);
2643	<	doSubmit(job);
2642	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2643	>	externalPush(job);
2644		}
2645
2646		/**
#	Line 2048 | Line 2653 | public class ForkJoinPool extends Abstra
2653		*         scheduled for execution
2654		*/
2655		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2656	<	doSubmit(task);
2656	>	if (task == null)
2657	>	throw new NullPointerException();
2658	>	externalPush(task);
2659		return task;
2660		}
2661
#	Line 2058 | Line 2665 | public class ForkJoinPool extends Abstra
2665		*         scheduled for execution
2666		*/
2667		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2668	<	if (task == null)
2669	<	throw new NullPointerException();
2063	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task);
2064	<	doSubmit(job);
2668	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2669	>	externalPush(job);
2670		return job;
2671		}
2672
#	Line 2071 | Line 2676 | public class ForkJoinPool extends Abstra
2676		*         scheduled for execution
2677		*/
2678		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2679	<	if (task == null)
2680	<	throw new NullPointerException();
2076	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
2077	<	doSubmit(job);
2679	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2680	>	externalPush(job);
2681		return job;
2682		}
2683
#	Line 2090 | Line 2693 | public class ForkJoinPool extends Abstra
2693		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2694		job = (ForkJoinTask<?>) task;
2695		else
2696	<	job = ForkJoinTask.adapt(task, null);
2697	<	doSubmit(job);
2696	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2697	>	externalPush(job);
2698		return job;
2699		}
2700
#	Line 2112 | Line 2715 | public class ForkJoinPool extends Abstra
2715		boolean done = false;
2716		try {
2717		for (Callable<T> t : tasks) {
2718	<	ForkJoinTask<T> f = ForkJoinTask.adapt(t);
2719	<	doSubmit(f);
2718	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2719	>	externalPush(f);
2720		fs.add(f);
2721		}
2722		for (ForkJoinTask<T> f : fs)
#	Line 2152 | Line 2755 | public class ForkJoinPool extends Abstra
2755		* @return the targeted parallelism level of this pool
2756		*/
2757		public int getParallelism() {
2758	<	return parallelism;
2758	>	return config & SMASK;
2759	>	}
2760	>
2761	>	/**
2762	>	* Returns the targeted parallelism level of the common pool.
2763	>	*
2764	>	* @return the targeted parallelism level of the common pool
2765	>	*/
2766	>	public static int getCommonPoolParallelism() {
2767	>	return commonPoolParallelism;
2768		}
2769
2770		/**
#	Line 2164 | Line 2776 | public class ForkJoinPool extends Abstra
2776		* @return the number of worker threads
2777		*/
2778		public int getPoolSize() {
2779	<	return parallelism + (short)(ctl >>> TC_SHIFT);
2779	>	return (config & SMASK) + (short)(ctl >>> TC_SHIFT);
2780		}
2781
2782		/**
#	Line 2174 | Line 2786 | public class ForkJoinPool extends Abstra
2786		* @return {@code true} if this pool uses async mode
2787		*/
2788		public boolean getAsyncMode() {
2789	<	return localMode != 0;
2789	>	return (config >>> 16) == FIFO_QUEUE;
2790		}
2791
2792		/**
#	Line 2205 | Line 2817 | public class ForkJoinPool extends Abstra
2817		* @return the number of active threads
2818		*/
2819		public int getActiveThreadCount() {
2820	<	int r = parallelism + (int)(ctl >> AC_SHIFT);
2820	>	int r = (config & SMASK) + (int)(ctl >> AC_SHIFT);
2821		return (r <= 0) ? 0 : r; // suppress momentarily negative values
2822		}
2823
#	Line 2221 | Line 2833 | public class ForkJoinPool extends Abstra
2833		* @return {@code true} if all threads are currently idle
2834		*/
2835		public boolean isQuiescent() {
2836	<	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2836	>	return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0;
2837		}
2838
2839		/**
#	Line 2236 | Line 2848 | public class ForkJoinPool extends Abstra
2848		* @return the number of steals
2849		*/
2850		public long getStealCount() {
2851	<	long count = stealCount.get();
2851	>	long count = stealCount;
2852		WorkQueue[] ws; WorkQueue w;
2853		if ((ws = workQueues) != null) {
2854		for (int i = 1; i < ws.length; i += 2) {
2855		if ((w = ws[i]) != null)
2856	<	count += w.totalSteals;
2856	>	count += w.nsteals;
2857		}
2858		}
2859		return count;
#	Line 2298 | Line 2910 | public class ForkJoinPool extends Abstra
2910		WorkQueue[] ws; WorkQueue w;
2911		if ((ws = workQueues) != null) {
2912		for (int i = 0; i < ws.length; i += 2) {
2913	<	if ((w = ws[i]) != null && w.queueSize() != 0)
2913	>	if ((w = ws[i]) != null && !w.isEmpty())
2914		return true;
2915		}
2916		}
#	Line 2366 | Line 2978 | public class ForkJoinPool extends Abstra
2978		public String toString() {
2979		// Use a single pass through workQueues to collect counts
2980		long qt = 0L, qs = 0L; int rc = 0;
2981	<	long st = stealCount.get();
2981	>	long st = stealCount;
2982		long c = ctl;
2983		WorkQueue[] ws; WorkQueue w;
2984		if ((ws = workQueues) != null) {
#	Line 2377 | Line 2989 | public class ForkJoinPool extends Abstra
2989		qs += size;
2990		else {
2991		qt += size;
2992	<	st += w.totalSteals;
2992	>	st += w.nsteals;
2993		if (w.isApparentlyUnblocked())
2994		++rc;
2995		}
2996		}
2997		}
2998		}
2999	<	int pc = parallelism;
2999	>	int pc = (config & SMASK);
3000		int tc = pc + (short)(c >>> TC_SHIFT);
3001		int ac = pc + (int)(c >> AC_SHIFT);
3002		if (ac < 0) // ignore transient negative
#	Line 2393 | Line 3005 | public class ForkJoinPool extends Abstra
3005		if ((c & STOP_BIT) != 0)
3006		level = (tc == 0) ? "Terminated" : "Terminating";
3007		else
3008	<	level = runState < 0 ? "Shutting down" : "Running";
3008	>	level = plock < 0 ? "Shutting down" : "Running";
3009		return super.toString() +
3010		"[" + level +
3011		", parallelism = " + pc +
#	Line 2407 | Line 3019 | public class ForkJoinPool extends Abstra
3019		}
3020
3021		/**
3022	<	* Initiates an orderly shutdown in which previously submitted
3023	<	* tasks are executed, but no new tasks will be accepted.
3024	<	* Invocation has no additional effect if already shut down.
3025	<	* Tasks that are in the process of being submitted concurrently
3026	<	* during the course of this method may or may not be rejected.
3022	>	* Possibly initiates an orderly shutdown in which previously
3023	>	* submitted tasks are executed, but no new tasks will be
3024	>	* accepted. Invocation has no effect on execution state if this
3025	>	* is the {@link #commonPool}, and no additional effect if
3026	>	* already shut down.  Tasks that are in the process of being
3027	>	* submitted concurrently during the course of this method may or
3028	>	* may not be rejected.
3029		*
3030		* @throws SecurityException if a security manager exists and
3031		*         the caller is not permitted to modify threads
#	Line 2424 | Line 3038 | public class ForkJoinPool extends Abstra
3038		}
3039
3040		/**
3041	<	* Attempts to cancel and/or stop all tasks, and reject all
3042	<	* subsequently submitted tasks.  Tasks that are in the process of
3043	<	* being submitted or executed concurrently during the course of
3044	<	* this method may or may not be rejected. This method cancels
3045	<	* both existing and unexecuted tasks, in order to permit
3046	<	* termination in the presence of task dependencies. So the method
3047	<	* always returns an empty list (unlike the case for some other
3048	<	* Executors).
3041	>	* Possibly attempts to cancel and/or stop all tasks, and reject
3042	>	* all subsequently submitted tasks.  Invocation has no effect on
3043	>	* execution state if this is the {@link #commonPool}, and no
3044	>	* additional effect if already shut down. Otherwise, tasks that
3045	>	* are in the process of being submitted or executed concurrently
3046	>	* during the course of this method may or may not be
3047	>	* rejected. This method cancels both existing and unexecuted
3048	>	* tasks, in order to permit termination in the presence of task
3049	>	* dependencies. So the method always returns an empty list
3050	>	* (unlike the case for some other Executors).
3051		*
3052		* @return an empty list
3053		* @throws SecurityException if a security manager exists and
#	Line 2453 | Line 3069 | public class ForkJoinPool extends Abstra
3069		public boolean isTerminated() {
3070		long c = ctl;
3071		return ((c & STOP_BIT) != 0L &&
3072	<	(short)(c >>> TC_SHIFT) == -parallelism);
3072	>	(short)(c >>> TC_SHIFT) == -(config & SMASK));
3073		}
3074
3075		/**
#	Line 2461 | Line 3077 | public class ForkJoinPool extends Abstra
3077		* commenced but not yet completed.  This method may be useful for
3078		* debugging. A return of {@code true} reported a sufficient
3079		* period after shutdown may indicate that submitted tasks have
3080	<	* ignored or suppressed interruption, or are waiting for IO,
3080	>	* ignored or suppressed interruption, or are waiting for I/O,
3081		* causing this executor not to properly terminate. (See the
3082		* advisory notes for class {@link ForkJoinTask} stating that
3083		* tasks should not normally entail blocking operations.  But if
#	Line 2472 | Line 3088 | public class ForkJoinPool extends Abstra
3088		public boolean isTerminating() {
3089		long c = ctl;
3090		return ((c & STOP_BIT) != 0L &&
3091	<	(short)(c >>> TC_SHIFT) != -parallelism);
3091	>	(short)(c >>> TC_SHIFT) != -(config & SMASK));
3092		}
3093
3094		/**
#	Line 2481 | Line 3097 | public class ForkJoinPool extends Abstra
3097		* @return {@code true} if this pool has been shut down
3098		*/
3099		public boolean isShutdown() {
3100	<	return runState < 0;
3100	>	return plock < 0;
3101		}
3102
3103		/**
3104	<	* Blocks until all tasks have completed execution after a shutdown
3105	<	* request, or the timeout occurs, or the current thread is
3106	<	* interrupted, whichever happens first.
3104	>	* Blocks until all tasks have completed execution after a
3105	>	* shutdown request, or the timeout occurs, or the current thread
3106	>	* is interrupted, whichever happens first. Note that the {@link
3107	>	* #commonPool()} never terminates until program shutdown so
3108	>	* this method will always time out.
3109		*
3110		* @param timeout the maximum time to wait
3111		* @param unit the time unit of the timeout argument
#	Line 2498 | Line 3116 | public class ForkJoinPool extends Abstra
3116		public boolean awaitTermination(long timeout, TimeUnit unit)
3117		throws InterruptedException {
3118		long nanos = unit.toNanos(timeout);
3119	<	final Mutex lock = this.lock;
3120	<	lock.lock();
3121	<	try {
3122	<	for (;;) {
3123	<	if (isTerminated())
3124	<	return true;
3125	<	if (nanos <= 0)
3126	<	return false;
3127	<	nanos = termination.awaitNanos(nanos);
3119	>	if (isTerminated())
3120	>	return true;
3121	>	long startTime = System.nanoTime();
3122	>	boolean terminated = false;
3123	>	synchronized (this) {
3124	>	for (long waitTime = nanos, millis = 0L;;) {
3125	>	if (terminated = isTerminated() ||
3126	>	waitTime <= 0L ||
3127	>	(millis = unit.toMillis(waitTime)) <= 0L)
3128	>	break;
3129	>	wait(millis);
3130	>	waitTime = nanos - (System.nanoTime() - startTime);
3131		}
2511	–	} finally {
2512	–	lock.unlock();
3132		}
3133	+	return terminated;
3134		}
3135
3136		/**
#	Line 2609 | Line 3229 | public class ForkJoinPool extends Abstra
3229		public static void managedBlock(ManagedBlocker blocker)
3230		throws InterruptedException {
3231		Thread t = Thread.currentThread();
3232	<	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3233	<	((ForkJoinWorkerThread)t).pool : null);
3234	<	while (!blocker.isReleasable()) {
3235	<	if (p == null || p.tryCompensate()) {
3236	<	try {
3237	<	do {} while (!blocker.isReleasable() && !blocker.block());
3238	<	} finally {
3239	<	if (p != null)
3232	>	if (t instanceof ForkJoinWorkerThread) {
3233	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3234	>	while (!blocker.isReleasable()) { // variant of helpSignal
3235	>	WorkQueue[] ws; WorkQueue q; int m, u;
3236	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3237	>	for (int i = 0; i <= m; ++i) {
3238	>	if (blocker.isReleasable())
3239	>	return;
3240	>	if ((q = ws[i]) != null && q.base - q.top < 0) {
3241	>	p.signalWork(q);
3242	>	if ((u = (int)(p.ctl >>> 32)) >= 0 ||
3243	>	(u >> UAC_SHIFT) >= 0)
3244	>	break;
3245	>	}
3246	>	}
3247	>	}
3248	>	if (p.tryCompensate()) {
3249	>	try {
3250	>	do {} while (!blocker.isReleasable() &&
3251	>	!blocker.block());
3252	>	} finally {
3253		p.incrementActiveCount();
3254	+	}
3255	+	break;
3256		}
2622	–	break;
3257		}
3258		}
3259	+	else {
3260	+	do {} while (!blocker.isReleasable() &&
3261	+	!blocker.block());
3262	+	}
3263		}
3264
3265		// AbstractExecutorService overrides.  These rely on undocumented
#	Line 2629 | Line 3267 | public class ForkJoinPool extends Abstra
3267		// implement RunnableFuture.
3268
3269		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3270	<	return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
3270	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3271		}
3272
3273		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3274	<	return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
3274	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
3275		}
3276
3277		// Unsafe mechanics
3278		private static final sun.misc.Unsafe U;
3279		private static final long CTL;
3280		private static final long PARKBLOCKER;
3281	+	private static final int ABASE;
3282	+	private static final int ASHIFT;
3283	+	private static final long STEALCOUNT;
3284	+	private static final long PLOCK;
3285	+	private static final long INDEXSEED;
3286	+	private static final long QLOCK;
3287
3288		static {
3289	<	poolNumberGenerator = new AtomicInteger();
2646	<	modifyThreadPermission = new RuntimePermission("modifyThread");
2647	<	defaultForkJoinWorkerThreadFactory =
2648	<	new DefaultForkJoinWorkerThreadFactory();
2649	<	submitters = new ThreadSubmitter();
3289	>	int s; // initialize field offsets for CAS etc
3290		try {
3291		U = getUnsafe();
3292		Class<?> k = ForkJoinPool.class;
3293		CTL = U.objectFieldOffset
3294		(k.getDeclaredField("ctl"));
3295	+	STEALCOUNT = U.objectFieldOffset
3296	+	(k.getDeclaredField("stealCount"));
3297	+	PLOCK = U.objectFieldOffset
3298	+	(k.getDeclaredField("plock"));
3299	+	INDEXSEED = U.objectFieldOffset
3300	+	(k.getDeclaredField("indexSeed"));
3301		Class<?> tk = Thread.class;
3302		PARKBLOCKER = U.objectFieldOffset
3303		(tk.getDeclaredField("parkBlocker"));
3304	+	Class<?> wk = WorkQueue.class;
3305	+	QLOCK = U.objectFieldOffset
3306	+	(wk.getDeclaredField("qlock"));
3307	+	Class<?> ak = ForkJoinTask[].class;
3308	+	ABASE = U.arrayBaseOffset(ak);
3309	+	s = U.arrayIndexScale(ak);
3310	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3311		} catch (Exception e) {
3312		throw new Error(e);
3313		}
3314	+	if ((s & (s-1)) != 0)
3315	+	throw new Error("data type scale not a power of two");
3316	+
3317	+	submitters = new ThreadLocal<Submitter>();
3318	+	ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory =
3319	+	new DefaultForkJoinWorkerThreadFactory();
3320	+	modifyThreadPermission = new RuntimePermission("modifyThread");
3321	+
3322	+	/*
3323	+	* Establish common pool parameters.  For extra caution,
3324	+	* computations to set up common pool state are here; the
3325	+	* constructor just assigns these values to fields.
3326	+	*/
3327	+
3328	+	int par = 0;
3329	+	Thread.UncaughtExceptionHandler handler = null;
3330	+	try {  // TBD: limit or report ignored exceptions?
3331	+	String pp = System.getProperty
3332	+	("java.util.concurrent.ForkJoinPool.common.parallelism");
3333	+	String hp = System.getProperty
3334	+	("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3335	+	String fp = System.getProperty
3336	+	("java.util.concurrent.ForkJoinPool.common.threadFactory");
3337	+	if (fp != null)
3338	+	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3339	+	getSystemClassLoader().loadClass(fp).newInstance());
3340	+	if (hp != null)
3341	+	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3342	+	getSystemClassLoader().loadClass(hp).newInstance());
3343	+	if (pp != null)
3344	+	par = Integer.parseInt(pp);
3345	+	} catch (Exception ignore) {
3346	+	}
3347	+
3348	+	if (par <= 0)
3349	+	par = Runtime.getRuntime().availableProcessors();
3350	+	if (par > MAX_CAP)
3351	+	par = MAX_CAP;
3352	+	commonPoolParallelism = par;
3353	+	long np = (long)(-par); // precompute initial ctl value
3354	+	long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
3355	+
3356	+	commonPool = new ForkJoinPool(par, ct, fac, handler);
3357		}
3358
3359		/**

Diff Legend

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