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 1.112 by dl, Thu Jan 26 18:15:12 2012 UTC vs.
Revision 1.139 by dl, Wed Oct 31 12:49:24 2012 UTC

#	Line 21 | Line 21 | import java.util.concurrent.RunnableFutu
21		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.ReentrantLock;
24	>	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
25		import java.util.concurrent.locks.Condition;
26
27		/**
#	Line 42 | Line 42 | import java.util.concurrent.locks.Condit
42		* ForkJoinPool}s may also be appropriate for use with event-style
43		* tasks that are never joined.
44		*
45	<	* <p>A {@code ForkJoinPool} is constructed with a given target
46	<	* parallelism level; by default, equal to the number of available
47	<	* processors. The pool attempts to maintain enough active (or
48	<	* available) threads by dynamically adding, suspending, or resuming
49	<	* internal worker threads, even if some tasks are stalled waiting to
50	<	* join others. However, no such adjustments are guaranteed in the
51	<	* face of blocked IO or other unmanaged synchronization. The nested
52	<	* {@link ManagedBlocker} interface enables extension of the kinds of
45	>	* <p>A static {@link #commonPool} is available and appropriate for
46	>	* most applications. The common pool is used by any ForkJoinTask that
47	>	* is not explicitly submitted to a specified pool. Using the common
48	>	* pool normally reduces resource usage (its threads are slowly
49	>	* reclaimed during periods of non-use, and reinstated upon subsequent
50	>	* use).  The common pool is by default constructed with default
51	>	* parameters, but these may be controlled by setting any or all of
52	>	* the three properties {@code
53	>	* java.util.concurrent.ForkJoinPool.common.{parallelism,
54	>	* threadFactory, exceptionHandler}}.
55	>	*
56	>	* <p>For applications that require separate or custom pools, a {@code
57	>	* ForkJoinPool} may be constructed with a given target parallelism
58	>	* level; by default, equal to the number of available processors. The
59	>	* pool attempts to maintain enough active (or available) threads by
60	>	* dynamically adding, suspending, or resuming internal worker
61	>	* threads, even if some tasks are stalled waiting to join
62	>	* others. However, no such adjustments are guaranteed in the face of
63	>	* blocked IO or other unmanaged synchronization. The nested {@link
64	>	* ManagedBlocker} interface enables extension of the kinds of
65		* synchronization accommodated.
66		*
67		* <p>In addition to execution and lifecycle control methods, this
#	Line 60 | Line 72 | import java.util.concurrent.locks.Condit
72		* convenient form for informal monitoring.
73		*
74		* <p> As is the case with other ExecutorServices, there are three
75	<	* main task execution methods summarized in the following
76	<	* table. These are designed to be used primarily by clients not
77	<	* already engaged in fork/join computations in the current pool.  The
78	<	* main forms of these methods accept instances of {@code
79	<	* ForkJoinTask}, but overloaded forms also allow mixed execution of
80	<	* plain {@code Runnable}- or {@code Callable}- based activities as
81	<	* well.  However, tasks that are already executing in a pool should
82	<	* normally instead use the within-computation forms listed in the
83	<	* table unless using async event-style tasks that are not usually
84	<	* joined, in which case there is little difference among choice of
73	<	* methods.
75	>	* main task execution methods summarized in the following table.
76	>	* These are designed to be used primarily by clients not already
77	>	* engaged in fork/join computations in the current pool.  The main
78	>	* forms of these methods accept instances of {@code ForkJoinTask},
79	>	* but overloaded forms also allow mixed execution of plain {@code
80	>	* Runnable}- or {@code Callable}- based activities as well.  However,
81	>	* tasks that are already executing in a pool should normally instead
82	>	* use the within-computation forms listed in the table unless using
83	>	* async event-style tasks that are not usually joined, in which case
84	>	* there is little difference among choice of methods.
85		*
86		* <table BORDER CELLPADDING=3 CELLSPACING=1>
87		*  <tr>
#	Line 95 | Line 106 | import java.util.concurrent.locks.Condit
106		*  </tr>
107		* </table>
108		*
98	–	* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
99	–	* used for all parallel task execution in a program or subsystem.
100	–	* Otherwise, use would not usually outweigh the construction and
101	–	* bookkeeping overhead of creating a large set of threads. For
102	–	* example, a common pool could be used for the {@code SortTasks}
103	–	* illustrated in {@link RecursiveAction}. Because {@code
104	–	* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
105	–	* daemon} mode, there is typically no need to explicitly {@link
106	–	* #shutdown} such a pool upon program exit.
107	–	*
108	–	*  <pre> {@code
109	–	* static final ForkJoinPool mainPool = new ForkJoinPool();
110	–	* ...
111	–	* public void sort(long[] array) {
112	–	*   mainPool.invoke(new SortTask(array, 0, array.length));
113	–	* }}</pre>
114	–	*
109		* <p><b>Implementation notes</b>: This implementation restricts the
110		* maximum number of running threads to 32767. Attempts to create
111		* pools with greater than the maximum number result in
#	Line 131 | Line 125 | public class ForkJoinPool extends Abstra
125		*
126		* This class and its nested classes provide the main
127		* functionality and control for a set of worker threads:
128	<	* Submissions from non-FJ threads enter into submission
129	<	* queues. Workers take these tasks and typically split them into
130	<	* subtasks that may be stolen by other workers.  Preference rules
131	<	* give first priority to processing tasks from their own queues
132	<	* (LIFO or FIFO, depending on mode), then to randomized FIFO
133	<	* steals of tasks in other queues.
128	>	* Submissions from non-FJ threads enter into submission queues.
129	>	* Workers take these tasks and typically split them into subtasks
130	>	* that may be stolen by other workers.  Preference rules give
131	>	* first priority to processing tasks from their own queues (LIFO
132	>	* or FIFO, depending on mode), then to randomized FIFO steals of
133	>	* tasks in other queues.
134		*
135	<	* WorkQueues.
135	>	* WorkQueues
136		* ==========
137		*
138		* Most operations occur within work-stealing queues (in nested
#	Line 156 | Line 150 | public class ForkJoinPool extends Abstra
150		* (http://research.sun.com/scalable/pubs/index.html) and
151		* "Idempotent work stealing" by Michael, Saraswat, and Vechev,
152		* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
153	<	* The main differences ultimately stem from gc requirements that
153	>	* The main differences ultimately stem from GC requirements that
154		* we null out taken slots as soon as we can, to maintain as small
155		* a footprint as possible even in programs generating huge
156		* numbers of tasks. To accomplish this, we shift the CAS
#	Line 178 | Line 172 | public class ForkJoinPool extends Abstra
172		* If an attempted steal fails, a thief always chooses a different
173		* random victim target to try next. So, in order for one thief to
174		* progress, it suffices for any in-progress poll or new push on
175	<	* any empty queue to complete.
175	>	* any empty queue to complete. (This is why we normally use
176	>	* method pollAt and its variants that try once at the apparent
177	>	* base index, else consider alternative actions, rather than
178	>	* method poll.)
179		*
180		* This approach also enables support of a user mode in which local
181		* task processing is in FIFO, not LIFO order, simply by using
#	Line 188 | Line 185 | public class ForkJoinPool extends Abstra
185		* rarely provide the best possible performance on a given
186		* machine, but portably provide good throughput by averaging over
187		* these factors.  (Further, even if we did try to use such
188	<	* information, we do not usually have a basis for exploiting
189	<	* it. For example, some sets of tasks profit from cache
190	<	* affinities, but others are harmed by cache pollution effects.)
188	>	* information, we do not usually have a basis for exploiting it.
189	>	* For example, some sets of tasks profit from cache affinities,
190	>	* but others are harmed by cache pollution effects.)
191		*
192		* WorkQueues are also used in a similar way for tasks submitted
193		* to the pool. We cannot mix these tasks in the same queues used
194		* for work-stealing (this would contaminate lifo/fifo
195	<	* processing). Instead, we loosely associate (via hashing)
196	<	* submission queues with submitting threads, and randomly scan
197	<	* these queues as well when looking for work. In essence,
198	<	* submitters act like workers except that they never take tasks,
199	<	* and they are multiplexed on to a finite number of shared work
200	<	* queues. However, classes are set up so that future extensions
201	<	* could allow submitters to optionally help perform tasks as
202	<	* well. Pool submissions from internal workers are also allowed,
203	<	* but use randomized rather than thread-hashed queue indices to
204	<	* avoid imbalance.  Insertion of tasks in shared mode requires a
195	>	* processing). Instead, we loosely associate submission queues
196	>	* with submitting threads, using a form of hashing.  The
197	>	* ThreadLocal Submitter class contains a value initially used as
198	>	* a hash code for choosing existing queues, but may be randomly
199	>	* repositioned upon contention with other submitters.  In
200	>	* essence, submitters act like workers except that they never
201	>	* take tasks, and they are multiplexed on to a finite number of
202	>	* shared work queues. However, classes are set up so that future
203	>	* extensions could allow submitters to optionally help perform
204	>	* tasks as well. Insertion of tasks in shared mode requires a
205		* lock (mainly to protect in the case of resizing) but we use
206		* only a simple spinlock (using bits in field runState), because
207	<	* submitters encountering a busy queue try or create others so
208	<	* never block.
207	>	* submitters encountering a busy queue move on to try or create
208	>	* other queues -- they block only when creating and registering
209	>	* new queues.
210		*
211	<	* Management.
211	>	* Management
212		* ==========
213		*
214		* The main throughput advantages of work-stealing stem from
#	Line 220 | Line 218 | public class ForkJoinPool extends Abstra
218		* tactic for avoiding bottlenecks is packing nearly all
219		* essentially atomic control state into two volatile variables
220		* that are by far most often read (not written) as status and
221	<	* consistency checks
221	>	* consistency checks.
222		*
223		* Field "ctl" contains 64 bits holding all the information needed
224		* to atomically decide to add, inactivate, enqueue (on an event
#	Line 237 | Line 235 | public class ForkJoinPool extends Abstra
235		* when locked remains available to check consistency.
236		*
237		* Recording WorkQueues.  WorkQueues are recorded in the
238	<	* "workQueues" array that is created upon pool construction and
239	<	* expanded if necessary.  Updates to the array while recording
240	<	* new workers and unrecording terminated ones are protected from
241	<	* each other by a lock but the array is otherwise concurrently
242	<	* readable, and accessed directly.  To simplify index-based
243	<	* operations, the array size is always a power of two, and all
244	<	* readers must tolerate null slots. Shared (submission) queues
245	<	* are at even indices, worker queues at odd indices. Grouping
246	<	* them together in this way simplifies and speeds up task
247	<	* scanning. To avoid flailing during start-up, the array is
248	<	* presized to hold twice #parallelism workers (which is unlikely
249	<	* to need further resizing during execution). But to avoid
252	<	* dealing with so many null slots, variable runState includes a
253	<	* mask for the nearest power of two that contains all current
254	<	* workers.  All worker thread creation is on-demand, triggered by
255	<	* task submissions, replacement of terminated workers, and/or
238	>	* "workQueues" array that is created upon first use and expanded
239	>	* if necessary.  Updates to the array while recording new workers
240	>	* and unrecording terminated ones are protected from each other
241	>	* by a lock but the array is otherwise concurrently readable, and
242	>	* accessed directly.  To simplify index-based operations, the
243	>	* array size is always a power of two, and all readers must
244	>	* tolerate null slots. Shared (submission) queues are at even
245	>	* indices, worker queues at odd indices. Grouping them together
246	>	* in this way simplifies and speeds up task scanning.
247	>	*
248	>	* All worker thread creation is on-demand, triggered by task
249	>	* submissions, replacement of terminated workers, and/or
250		* compensation for blocked workers. However, all other support
251		* code is set up to work with other policies.  To ensure that we
252		* do not hold on to worker references that would prevent GC, ALL
#	Line 265 | Line 259 | public class ForkJoinPool extends Abstra
259		* both index-check and null-check the IDs. All such accesses
260		* ignore bad IDs by returning out early from what they are doing,
261		* since this can only be associated with termination, in which
262	<	* case it is OK to give up.
263	<	*
264	<	* All uses of the workQueues array check that it is non-null
265	<	* (even if previously non-null). This allows nulling during
266	<	* termination, which is currently not necessary, but remains an
267	<	* option for resource-revocation-based shutdown schemes. It also
274	<	* helps reduce JIT issuance of uncommon-trap code, which tends to
262	>	* case it is OK to give up.  All uses of the workQueues array
263	>	* also check that it is non-null (even if previously
264	>	* non-null). This allows nulling during termination, which is
265	>	* currently not necessary, but remains an option for
266	>	* resource-revocation-based shutdown schemes. It also helps
267	>	* reduce JIT issuance of uncommon-trap code, which tends to
268		* unnecessarily complicate control flow in some methods.
269		*
270		* Event Queuing. Unlike HPC work-stealing frameworks, we cannot
#	Line 299 | Line 292 | public class ForkJoinPool extends Abstra
292		* some other queued worker rather than itself, which has the same
293		* net effect. Because enqueued workers may actually be rescanning
294		* rather than waiting, we set and clear the "parker" field of
295	<	* Workqueues to reduce unnecessary calls to unpark.  (This
295	>	* WorkQueues to reduce unnecessary calls to unpark.  (This
296		* requires a secondary recheck to avoid missed signals.)  Note
297		* the unusual conventions about Thread.interrupts surrounding
298		* parking and other blocking: Because interrupts are used solely
#	Line 322 | Line 315 | public class ForkJoinPool extends Abstra
315		*
316		* Trimming workers. To release resources after periods of lack of
317		* use, a worker starting to wait when the pool is quiescent will
318	<	* time out and terminate if the pool has remained quiescent for
319	<	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
320	<	* terminating all workers after long periods of non-use.
318	>	* time out and terminate if the pool has remained quiescent for a
319	>	* given period -- a short period if there are more threads than
320	>	* parallelism, longer as the number of threads decreases. This
321	>	* will slowly propagate, eventually terminating all workers after
322	>	* periods of non-use.
323		*
324		* Shutdown and Termination. A call to shutdownNow atomically sets
325	<	* a runState bit and then (non-atomically) sets each workers
325	>	* a runState bit and then (non-atomically) sets each worker's
326		* runState status, cancels all unprocessed tasks, and wakes up
327		* all waiting workers.  Detecting whether termination should
328		* commence after a non-abrupt shutdown() call requires more work
#	Line 336 | Line 331 | public class ForkJoinPool extends Abstra
331		* indication but non-abrupt shutdown still requires a rechecking
332		* scan for any workers that are inactive but not queued.
333		*
334	<	* Joining Tasks.
335	<	* ==============
334	>	* Joining Tasks
335	>	* =============
336		*
337		* Any of several actions may be taken when one worker is waiting
338	<	* to join a task stolen (or always held by) another.  Because we
338	>	* to join a task stolen (or always held) by another.  Because we
339		* are multiplexing many tasks on to a pool of workers, we can't
340		* just let them block (as in Thread.join).  We also cannot just
341		* reassign the joiner's run-time stack with another and replace
342		* it later, which would be a form of "continuation", that even if
343		* possible is not necessarily a good idea since we sometimes need
344	<	* both an unblocked task and its continuation to
345	<	* progress. Instead we combine two tactics:
344	>	* both an unblocked task and its continuation to progress.
345	>	* Instead we combine two tactics:
346		*
347		*   Helping: Arranging for the joiner to execute some task that it
348		*      would be running if the steal had not occurred.
#	Line 382 | Line 377 | public class ForkJoinPool extends Abstra
377		* (http://portal.acm.org/citation.cfm?id=155354). It differs in
378		* that: (1) We only maintain dependency links across workers upon
379		* steals, rather than use per-task bookkeeping.  This sometimes
380	<	* requires a linear scan of workers array to locate stealers, but
381	<	* often doesn't because stealers leave hints (that may become
380	>	* requires a linear scan of workQueues array to locate stealers,
381	>	* but often doesn't because stealers leave hints (that may become
382		* stale/wrong) of where to locate them.  A stealHint is only a
383		* hint because a worker might have had multiple steals and the
384		* hint records only one of them (usually the most current).
#	Line 394 | Line 389 | public class ForkJoinPool extends Abstra
389		* which means that we miss links in the chain during long-lived
390		* tasks, GC stalls etc (which is OK since blocking in such cases
391		* is usually a good idea).  (4) We bound the number of attempts
392	<	* to find work (see MAX_HELP_DEPTH) and fall back to suspending
393	<	* the worker and if necessary replacing it with another.
392	>	* to find work (see MAX_HELP) and fall back to suspending the
393	>	* worker and if necessary replacing it with another.
394		*
395		* It is impossible to keep exactly the target parallelism number
396		* of threads running at any given time.  Determining the
397		* existence of conservatively safe helping targets, the
398		* availability of already-created spares, and the apparent need
399		* to create new spares are all racy, so we rely on multiple
400	<	* retries of each.  Currently, in keeping with on-demand
401	<	* signalling policy, we compensate only if blocking would leave
402	<	* less than one active (non-waiting, non-blocked) worker.
403	<	* Additionally, to avoid some false alarms due to GC, lagging
404	<	* counters, system activity, etc, compensated blocking for joins
405	<	* is only attempted after rechecks stabilize in
406	<	* ForkJoinTask.awaitJoin. (Retries are interspersed with
407	<	* Thread.yield, for good citizenship.)
400	>	* retries of each.  Compensation in the apparent absence of
401	>	* helping opportunities is challenging to control on JVMs, where
402	>	* GC and other activities can stall progress of tasks that in
403	>	* turn stall out many other dependent tasks, without us being
404	>	* able to determine whether they will ever require compensation.
405	>	* Even though work-stealing otherwise encounters little
406	>	* degradation in the presence of more threads than cores,
407	>	* aggressively adding new threads in such cases entails risk of
408	>	* unwanted positive feedback control loops in which more threads
409	>	* cause more dependent stalls (as well as delayed progress of
410	>	* unblocked threads to the point that we know they are available)
411	>	* leading to more situations requiring more threads, and so
412	>	* on. This aspect of control can be seen as an (analytically
413	>	* intractable) game with an opponent that may choose the worst
414	>	* (for us) active thread to stall at any time.  We take several
415	>	* precautions to bound losses (and thus bound gains), mainly in
416	>	* methods tryCompensate and awaitJoin: (1) We only try
417	>	* compensation after attempting enough helping steps (measured
418	>	* via counting and timing) that we have already consumed the
419	>	* estimated cost of creating and activating a new thread.  (2) We
420	>	* allow up to 50% of threads to be blocked before initially
421	>	* adding any others, and unless completely saturated, check that
422	>	* some work is available for a new worker before adding. Also, we
423	>	* create up to only 50% more threads until entering a mode that
424	>	* only adds a thread if all others are possibly blocked.  All
425	>	* together, this means that we might be half as fast to react,
426	>	* and create half as many threads as possible in the ideal case,
427	>	* but present vastly fewer anomalies in all other cases compared
428	>	* to both more aggressive and more conservative alternatives.
429		*
430		* Style notes: There is a lot of representation-level coupling
431		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 417 | Line 433 | public class ForkJoinPool extends Abstra
433		* managed by ForkJoinPool, so are directly accessed.  There is
434		* little point trying to reduce this, since any associated future
435		* changes in representations will need to be accompanied by
436	<	* algorithmic changes anyway. All together, these low-level
437	<	* implementation choices produce as much as a factor of 4
438	<	* performance improvement compared to naive implementations, and
439	<	* enable the processing of billions of tasks per second, at the
440	<	* expense of some ugliness.
441	<	*
442	<	* Methods signalWork() and scan() are the main bottlenecks so are
443	<	* especially heavily micro-optimized/mangled.  There are lots of
444	<	* inline assignments (of form "while ((local = field) != 0)")
445	<	* which are usually the simplest way to ensure the required read
446	<	* orderings (which are sometimes critical). This leads to a
447	<	* "C"-like style of listing declarations of these locals at the
448	<	* heads of methods or blocks.  There are several occurrences of
449	<	* the unusual "do {} while (!cas...)"  which is the simplest way
450	<	* to force an update of a CAS'ed variable. There are also other
451	<	* coding oddities that help some methods perform reasonably even
452	<	* when interpreted (not compiled).
453	<	*
454	<	* The order of declarations in this file is: (1) declarations of
455	<	* statics (2) fields (along with constants used when unpacking
456	<	* some of them), listed in an order that tends to reduce
457	<	* contention among them a bit under most JVMs; (3) nested
458	<	* classes; (4) internal control methods; (5) callbacks and other
459	<	* support for ForkJoinTask methods; (6) exported methods (plus a
460	<	* few little helpers); (7) static block initializing all statics
461	<	* in a minimally dependent order.
436	>	* algorithmic changes anyway. Several methods intrinsically
437	>	* sprawl because they must accumulate sets of consistent reads of
438	>	* volatiles held in local variables.  Methods signalWork() and
439	>	* scan() are the main bottlenecks, so are especially heavily
440	>	* micro-optimized/mangled.  There are lots of inline assignments
441	>	* (of form "while ((local = field) != 0)") which are usually the
442	>	* simplest way to ensure the required read orderings (which are
443	>	* sometimes critical). This leads to a "C"-like style of listing
444	>	* declarations of these locals at the heads of methods or blocks.
445	>	* There are several occurrences of the unusual "do {} while
446	>	* (!cas...)"  which is the simplest way to force an update of a
447	>	* CAS'ed variable. There are also other coding oddities that help
448	>	* some methods perform reasonably even when interpreted (not
449	>	* compiled).
450	>	*
451	>	* The order of declarations in this file is:
452	>	* (1) Static utility functions
453	>	* (2) Nested (static) classes
454	>	* (3) Static fields
455	>	* (4) Fields, along with constants used when unpacking some of them
456	>	* (5) Internal control methods
457	>	* (6) Callbacks and other support for ForkJoinTask methods
458	>	* (7) Exported methods
459	>	* (8) Static block initializing statics in minimally dependent order
460	>	*/
461	>
462	>	// Static utilities
463	>
464	>	/**
465	>	* If there is a security manager, makes sure caller has
466	>	* permission to modify threads.
467		*/
468	+	private static void checkPermission() {
469	+	SecurityManager security = System.getSecurityManager();
470	+	if (security != null)
471	+	security.checkPermission(modifyThreadPermission);
472	+	}
473	+
474	+	// Nested classes
475
476		/**
477		* Factory for creating new {@link ForkJoinWorkerThread}s.
#	Line 473 | Line 501 | public class ForkJoinPool extends Abstra
501		}
502
503		/**
504	<	* Creates a new ForkJoinWorkerThread. This factory is used unless
505	<	* overridden in ForkJoinPool constructors.
506	<	*/
507	<	public static final ForkJoinWorkerThreadFactory
480	<	defaultForkJoinWorkerThreadFactory;
481	<
482	<	/**
483	<	* Permission required for callers of methods that may start or
484	<	* kill threads.
485	<	*/
486	<	private static final RuntimePermission modifyThreadPermission;
487	<
488	<	/**
489	<	* If there is a security manager, makes sure caller has
490	<	* permission to modify threads.
504	>	* Class for artificial tasks that are used to replace the target
505	>	* of local joins if they are removed from an interior queue slot
506	>	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
507	>	* actually do anything beyond having a unique identity.
508		*/
509	<	private static void checkPermission() {
510	<	SecurityManager security = System.getSecurityManager();
511	<	if (security != null)
512	<	security.checkPermission(modifyThreadPermission);
509	>	static final class EmptyTask extends ForkJoinTask<Void> {
510	>	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
511	>	public final Void getRawResult() { return null; }
512	>	public final void setRawResult(Void x) {}
513	>	public final boolean exec() { return true; }
514		}
515
516		/**
499	–	* Generator for assigning sequence numbers as pool names.
500	–	*/
501	–	private static final AtomicInteger poolNumberGenerator;
502	–
503	–	/**
504	–	* Bits and masks for control variables
505	–	*
506	–	* Field ctl is a long packed with:
507	–	* AC: Number of active running workers minus target parallelism (16 bits)
508	–	* TC: Number of total workers minus target parallelism (16 bits)
509	–	* ST: true if pool is terminating (1 bit)
510	–	* EC: the wait count of top waiting thread (15 bits)
511	–	* ID: ~(poolIndex >>> 1) of top of Treiber stack of waiters (16 bits)
512	–	*
513	–	* When convenient, we can extract the upper 32 bits of counts and
514	–	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
515	–	* (int)ctl.  The ec field is never accessed alone, but always
516	–	* together with id and st. The offsets of counts by the target
517	–	* parallelism and the positionings of fields makes it possible to
518	–	* perform the most common checks via sign tests of fields: When
519	–	* ac is negative, there are not enough active workers, when tc is
520	–	* negative, there are not enough total workers, when id is
521	–	* negative, there is at least one waiting worker, and when e is
522	–	* negative, the pool is terminating.  To deal with these possibly
523	–	* negative fields, we use casts in and out of "short" and/or
524	–	* signed shifts to maintain signedness.
525	–	*
526	–	* When a thread is queued (inactivated), its eventCount field is
527	–	* negative, which is the only way to tell if a worker is
528	–	* prevented from executing tasks, even though it must continue to
529	–	* scan for them to avoid queuing races.
530	–	*
531	–	* Field runState is an int packed with:
532	–	* SHUTDOWN: true if shutdown is enabled (1 bit)
533	–	* SEQ:  a sequence number updated upon (de)registering workers (15 bits)
534	–	* MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
535	–	*
536	–	* The combination of mask and sequence number enables simple
537	–	* consistency checks: Staleness of read-only operations on the
538	–	* workers and queues arrays can be checked by comparing runState
539	–	* before vs after the reads. The low 16 bits (i.e, anding with
540	–	* SMASK) hold (the smallest power of two covering all worker
541	–	* indices, minus one.  The mask for queues (vs workers) is twice
542	–	* this value plus 1.
543	–	*/
544	–
545	–	// bit positions/shifts for fields
546	–	private static final int  AC_SHIFT   = 48;
547	–	private static final int  TC_SHIFT   = 32;
548	–	private static final int  ST_SHIFT   = 31;
549	–	private static final int  EC_SHIFT   = 16;
550	–
551	–	// bounds
552	–	private static final int  MAX_ID     = 0x7fff;  // max poolIndex
553	–	private static final int  SMASK      = 0xffff;  // mask short bits
554	–	private static final int  SHORT_SIGN = 1 << 15;
555	–	private static final int  INT_SIGN   = 1 << 31;
556	–
557	–	// masks
558	–	private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
559	–	private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
560	–	private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
561	–
562	–	// units for incrementing and decrementing
563	–	private static final long TC_UNIT    = 1L << TC_SHIFT;
564	–	private static final long AC_UNIT    = 1L << AC_SHIFT;
565	–
566	–	// masks and units for dealing with u = (int)(ctl >>> 32)
567	–	private static final int  UAC_SHIFT  = AC_SHIFT - 32;
568	–	private static final int  UTC_SHIFT  = TC_SHIFT - 32;
569	–	private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
570	–	private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
571	–	private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
572	–	private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
573	–
574	–	// masks and units for dealing with e = (int)ctl
575	–	private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
576	–	private static final int E_SEQ       = 1 << EC_SHIFT;
577	–
578	–	// runState bits
579	–	private static final int SHUTDOWN    = 1 << 31;
580	–	private static final int RS_SEQ      = 1 << 16;
581	–	private static final int RS_SEQ_MASK = 0x7fff0000;
582	–
583	–	// access mode for WorkQueue
584	–	static final int LIFO_QUEUE          =  0;
585	–	static final int FIFO_QUEUE          =  1;
586	–	static final int SHARED_QUEUE        = -1;
587	–
588	–	/**
589	–	* The wakeup interval (in nanoseconds) for a worker waiting for a
590	–	* task when the pool is quiescent to instead try to shrink the
591	–	* number of workers.  The exact value does not matter too
592	–	* much. It must be short enough to release resources during
593	–	* sustained periods of idleness, but not so short that threads
594	–	* are continually re-created.
595	–	*/
596	–	private static final long SHRINK_RATE =
597	–	4L * 1000L * 1000L * 1000L; // 4 seconds
598	–
599	–	/**
600	–	* The timeout value for attempted shrinkage, includes
601	–	* some slop to cope with system timer imprecision.
602	–	*/
603	–	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
604	–
605	–	/**
606	–	* The maximum stolen->joining link depth allowed in tryHelpStealer.
607	–	* Depths for legitimate chains are unbounded, but we use a fixed
608	–	* constant to avoid (otherwise unchecked) cycles and to bound
609	–	* staleness of traversal parameters at the expense of sometimes
610	–	* blocking when we could be helping.
611	–	*/
612	–	private static final int MAX_HELP_DEPTH = 16;
613	–
614	–	/*
615	–	* Field layout order in this class tends to matter more than one
616	–	* would like. Runtime layout order is only loosely related to
617	–	* declaration order and may differ across JVMs, but the following
618	–	* empirically works OK on current JVMs.
619	–	*/
620	–
621	–	volatile long ctl;                       // main pool control
622	–	final int parallelism;                   // parallelism level
623	–	final int localMode;                     // per-worker scheduling mode
624	–	int nextPoolIndex;                       // hint used in registerWorker
625	–	volatile int runState;                   // shutdown status, seq, and mask
626	–	WorkQueue[] workQueues;                  // main registry
627	–	final ReentrantLock lock;                // for registration
628	–	final Condition termination;             // for awaitTermination
629	–	final ForkJoinWorkerThreadFactory factory; // factory for new workers
630	–	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
631	–	final AtomicLong stealCount;             // collect counts when terminated
632	–	final AtomicInteger nextWorkerNumber;    // to create worker name string
633	–	final String workerNamePrefix;           // Prefix for assigning worker names
634	–
635	–	/**
517		* Queues supporting work-stealing as well as external task
518		* submission. See above for main rationale and algorithms.
519		* Implementation relies heavily on "Unsafe" intrinsics
#	Line 649 | Line 530 | public class ForkJoinPool extends Abstra
530		* for push, or under lock for trySharedPush, and accessed by
531		* other threads only after reading (volatile) base.  Both top and
532		* base are allowed to wrap around on overflow, but (top - base)
533	<	* (or more comonly -(base - top) to force volatile read of base
533	>	* (or more commonly -(base - top) to force volatile read of base
534		* before top) still estimates size.
535		*
536		* The array slots are read and written using the emulation of
#	Line 681 | Line 562 | public class ForkJoinPool extends Abstra
562		* avoiding really bad worst-case access. (Until better JVM
563		* support is in place, this padding is dependent on transient
564		* properties of JVM field layout rules.)  We also take care in
565	<	* allocating and sizing and resizing the array. Non-shared queue
565	>	* allocating, sizing and resizing the array. Non-shared queue
566		* arrays are initialized (via method growArray) by workers before
567		* use. Others are allocated on first use.
568		*/
569		static final class WorkQueue {
570		/**
571		* Capacity of work-stealing queue array upon initialization.
572	<	* Must be a power of two; at least 4, but set larger to
573	<	* reduce cacheline sharing among queues.
572	>	* Must be a power of two; at least 4, but should be larger to
573	>	* reduce or eliminate cacheline sharing among queues.
574	>	* Currently, it is much larger, as a partial workaround for
575	>	* the fact that JVMs often place arrays in locations that
576	>	* share GC bookkeeping (especially cardmarks) such that
577	>	* per-write accesses encounter serious memory contention.
578		*/
579	<	static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
579	>	static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
580
581		/**
582		* Maximum size for queue arrays. Must be a power of two less
#	Line 715 | Line 600 | public class ForkJoinPool extends Abstra
600		volatile int base;         // index of next slot for poll
601		int top;                   // index of next slot for push
602		ForkJoinTask<?>[] array;   // the elements (initially unallocated)
603	+	final ForkJoinPool pool;   // the containing pool (may be null)
604		final ForkJoinWorkerThread owner; // owning thread or null if shared
605		volatile Thread parker;    // == owner during call to park; else null
606	<	ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
606	>	volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
607		ForkJoinTask<?> currentSteal; // current non-local task being executed
608		// Heuristic padding to ameliorate unfortunate memory placements
609	<	Object p00, p01, p02, p03, p04, p05, p06, p07, p08, p09, p0a;
609	>	Object p00, p01, p02, p03, p04, p05, p06, p07;
610	>	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
611
612	<	WorkQueue(ForkJoinWorkerThread owner, int mode) {
726	<	this.owner = owner;
612	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
613		this.mode = mode;
614	+	this.pool = pool;
615	+	this.owner = owner;
616		// Place indices in the center of array (that is not yet allocated)
617		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
618		}
619
620		/**
621	<	* Returns number of tasks in the queue
621	>	* Returns the approximate number of tasks in the queue.
622		*/
623		final int queueSize() {
624	<	int n = base - top; // non-owner callers must read base first
625	<	return (n >= 0) ? 0 : -n;
624	>	int n = base - top;       // non-owner callers must read base first
625	>	return (n >= 0) ? 0 : -n; // ignore transient negative
626	>	}
627	>
628	>	/**
629	>	* Provides a more accurate estimate of whether this queue has
630	>	* any tasks than does queueSize, by checking whether a
631	>	* near-empty queue has at least one unclaimed task.
632	>	*/
633	>	final boolean isEmpty() {
634	>	ForkJoinTask<?>[] a; int m, s;
635	>	int n = base - (s = top);
636	>	return (n >= 0 ||
637	>	(n == -1 &&
638	>	((a = array) == null ||
639	>	(m = a.length - 1) < 0 ||
640	>	U.getObjectVolatile
641	>	(a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
642		}
643
644		/**
645		* Pushes a task. Call only by owner in unshared queues.
646		*
647		* @param task the task. Caller must ensure non-null.
648	<	* @param p, if non-null, pool to signal if necessary
745	<	* @throw RejectedExecutionException if array cannot
746	<	* be resized
648	>	* @throw RejectedExecutionException if array cannot be resized
649		*/
650	<	final void push(ForkJoinTask<?> task, ForkJoinPool p) {
651	<	ForkJoinTask<?>[] a;
650	>	final void push(ForkJoinTask<?> task) {
651	>	ForkJoinTask<?>[] a; ForkJoinPool p;
652		int s = top, m, n;
653		if ((a = array) != null) {    // ignore if queue removed
654		U.putOrderedObject
655		(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
656		if ((n = (top = s + 1) - base) <= 2) {
657	<	if (p != null)
657	>	if ((p = pool) != null)
658		p.signalWork();
659		}
660		else if (n >= m)
#	Line 771 | Line 673 | public class ForkJoinPool extends Abstra
673		boolean submitted = false;
674		if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
675		ForkJoinTask<?>[] a = array;
676	<	int s = top, n = s - base;
676	>	int s = top;
677		try {
678	<	if ((a != null && n < a.length - 1) ||
678	>	if ((a != null && a.length > s + 1 - base) ||
679		(a = growArray(false)) != null) { // must presize
680		int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
681		U.putObject(a, (long)j, task);    // don't need "ordered"
#	Line 788 | Line 690 | public class ForkJoinPool extends Abstra
690		}
691
692		/**
693	<	* Takes next task, if one exists, in FIFO order.
693	>	* Takes next task, if one exists, in LIFO order.  Call only
694	>	* by owner in unshared queues.
695		*/
696	<	final ForkJoinTask<?> poll() {
697	<	ForkJoinTask<?>[] a; int b, i;
698	<	while ((b = base) - top < 0 && (a = array) != null &&
699	<	(i = (a.length - 1) & b) >= 0) {
700	<	int j = (i << ASHIFT) + ABASE;
701	<	ForkJoinTask<?> t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
702	<	if (t != null && base == b &&
696	>	final ForkJoinTask<?> pop() {
697	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
698	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
699	>	for (int s; (s = top - 1) - base >= 0;) {
700	>	long j = ((m & s) << ASHIFT) + ABASE;
701	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
702	>	break;
703	>	if (U.compareAndSwapObject(a, j, t, null)) {
704	>	top = s;
705	>	return t;
706	>	}
707	>	}
708	>	}
709	>	return null;
710	>	}
711	>
712	>	final ForkJoinTask<?> sharedPop() {
713	>	ForkJoinTask<?> task = null;
714	>	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
715	>	try {
716	>	ForkJoinTask<?>[] a; int m;
717	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
718	>	for (int s; (s = top - 1) - base >= 0;) {
719	>	long j = ((m & s) << ASHIFT) + ABASE;
720	>	ForkJoinTask<?> t =
721	>	(ForkJoinTask<?>)U.getObject(a, j);
722	>	if (t == null)
723	>	break;
724	>	if (U.compareAndSwapObject(a, j, t, null)) {
725	>	top = s;
726	>	task = t;
727	>	break;
728	>	}
729	>	}
730	>	}
731	>	} finally {
732	>	runState = 0;
733	>	}
734	>	}
735	>	return task;
736	>	}
737	>
738	>
739	>	/**
740	>	* Takes a task in FIFO order if b is base of queue and a task
741	>	* can be claimed without contention. Specialized versions
742	>	* appear in ForkJoinPool methods scan and tryHelpStealer.
743	>	*/
744	>	final ForkJoinTask<?> pollAt(int b) {
745	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
746	>	if ((a = array) != null) {
747	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
748	>	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
749	>	base == b &&
750		U.compareAndSwapObject(a, j, t, null)) {
751		base = b + 1;
752		return t;
#	Line 806 | Line 756 | public class ForkJoinPool extends Abstra
756		}
757
758		/**
759	<	* Takes next task, if one exists, in LIFO order.
810	<	* Call only by owner in unshared queues.
759	>	* Takes next task, if one exists, in FIFO order.
760		*/
761	<	final ForkJoinTask<?> pop() {
762	<	ForkJoinTask<?> t; int m;
763	<	ForkJoinTask<?>[] a = array;
764	<	if (a != null && (m = a.length - 1) >= 0) {
765	<	for (int s; (s = top - 1) - base >= 0;) {
766	<	int j = ((m & s) << ASHIFT) + ABASE;
767	<	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) == null)
768	<	break;
769	<	if (U.compareAndSwapObject(a, j, t, null)) {
821	<	top = s;
761	>	final ForkJoinTask<?> poll() {
762	>	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
763	>	while ((b = base) - top < 0 && (a = array) != null) {
764	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
765	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
766	>	if (t != null) {
767	>	if (base == b &&
768	>	U.compareAndSwapObject(a, j, t, null)) {
769	>	base = b + 1;
770		return t;
771		}
772		}
773	+	else if (base == b) {
774	+	if (b + 1 == top)
775	+	break;
776	+	Thread.yield(); // wait for lagging update
777	+	}
778		}
779		return null;
780		}
#	Line 846 | Line 799 | public class ForkJoinPool extends Abstra
799		}
800
801		/**
849	–	* Returns task at index b if b is current base of queue.
850	–	*/
851	–	final ForkJoinTask<?> pollAt(int b) {
852	–	ForkJoinTask<?>[] a; int i;
853	–	ForkJoinTask<?> task = null;
854	–	if ((a = array) != null && (i = ((a.length - 1) & b)) >= 0) {
855	–	int j = (i << ASHIFT) + ABASE;
856	–	ForkJoinTask<?> t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
857	–	if (t != null && base == b &&
858	–	U.compareAndSwapObject(a, j, t, null)) {
859	–	base = b + 1;
860	–	task = t;
861	–	}
862	–	}
863	–	return task;
864	–	}
865	–
866	–	/**
802		* Pops the given task only if it is at the current top.
803		*/
804		final boolean tryUnpush(ForkJoinTask<?> t) {
#	Line 878 | Line 813 | public class ForkJoinPool extends Abstra
813		}
814
815		/**
816	+	* Version of tryUnpush for shared queues; called by non-FJ
817	+	* submitters after prechecking that task probably exists.
818	+	*/
819	+	final boolean trySharedUnpush(ForkJoinTask<?> t) {
820	+	boolean success = false;
821	+	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
822	+	try {
823	+	ForkJoinTask<?>[] a; int s;
824	+	if ((a = array) != null && (s = top) != base &&
825	+	U.compareAndSwapObject
826	+	(a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
827	+	top = s;
828	+	success = true;
829	+	}
830	+	} finally {
831	+	runState = 0;                         // unlock
832	+	}
833	+	}
834	+	return success;
835	+	}
836	+
837	+	/**
838		* Polls the given task only if it is at the current base.
839		*/
840		final boolean pollFor(ForkJoinTask<?> task) {
841	<	ForkJoinTask<?>[] a; int b, i;
842	<	if ((b = base) - top < 0 && (a = array) != null &&
843	<	(i = (a.length - 1) & b) >= 0) {
887	<	int j = (i << ASHIFT) + ABASE;
841	>	ForkJoinTask<?>[] a; int b;
842	>	if ((b = base) - top < 0 && (a = array) != null) {
843	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
844		if (U.getObjectVolatile(a, j) == task && base == b &&
845		U.compareAndSwapObject(a, j, task, null)) {
846		base = b + 1;
#	Line 895 | Line 851 | public class ForkJoinPool extends Abstra
851		}
852
853		/**
898	–	* If present, removes from queue and executes the given task, or
899	–	* any other cancelled task. Returns (true) immediately on any CAS
900	–	* or consistency check failure so caller can retry.
901	–	*
902	–	* @return false if no progress can be made
903	–	*/
904	–	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
905	–	boolean removed = false, empty = true, progress = true;
906	–	ForkJoinTask<?>[] a; int m, s, b, n;
907	–	if ((a = array) != null && (m = a.length - 1) >= 0 &&
908	–	(n = (s = top) - (b = base)) > 0) {
909	–	for (ForkJoinTask<?> t;;) {           // traverse from s to b
910	–	int j = ((--s & m) << ASHIFT) + ABASE;
911	–	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
912	–	if (t == null)                    // inconsistent length
913	–	break;
914	–	else if (t == task) {
915	–	if (s + 1 == top) {           // pop
916	–	if (!U.compareAndSwapObject(a, j, task, null))
917	–	break;
918	–	top = s;
919	–	removed = true;
920	–	}
921	–	else if (base == b)           // replace with proxy
922	–	removed = U.compareAndSwapObject(a, j, task,
923	–	new EmptyTask());
924	–	break;
925	–	}
926	–	else if (t.status >= 0)
927	–	empty = false;
928	–	else if (s + 1 == top) {          // pop and throw away
929	–	if (U.compareAndSwapObject(a, j, t, null))
930	–	top = s;
931	–	break;
932	–	}
933	–	if (--n == 0) {
934	–	if (!empty && base == b)
935	–	progress = false;
936	–	break;
937	–	}
938	–	}
939	–	}
940	–	if (removed)
941	–	task.doExec();
942	–	return progress;
943	–	}
944	–
945	–	/**
854		* Initializes or doubles the capacity of array. Call either
855		* by owner or with lock held -- it is OK for base, but not
856		* top, to move while resizings are in progress.
#	Line 978 | Line 886 | public class ForkJoinPool extends Abstra
886		}
887
888		/**
889	<	* Removes and cancels all known tasks, ignoring any exceptions
889	>	* Removes and cancels all known tasks, ignoring any exceptions.
890		*/
891		final void cancelAll() {
892		ForkJoinTask.cancelIgnoringExceptions(currentJoin);
#	Line 987 | Line 895 | public class ForkJoinPool extends Abstra
895		ForkJoinTask.cancelIgnoringExceptions(t);
896		}
897
898	<	// Execution methods
898	>	/**
899	>	* Computes next value for random probes.  Scans don't require
900	>	* a very high quality generator, but also not a crummy one.
901	>	* Marsaglia xor-shift is cheap and works well enough.  Note:
902	>	* This is manually inlined in its usages in ForkJoinPool to
903	>	* avoid writes inside busy scan loops.
904	>	*/
905	>	final int nextSeed() {
906	>	int r = seed;
907	>	r ^= r << 13;
908	>	r ^= r >>> 17;
909	>	return seed = r ^= r << 5;
910	>	}
911	>
912	>	// Specialized execution methods
913
914		/**
915	<	* Removes and runs tasks until empty, using local mode
994	<	* ordering.
915	>	* Pops and runs tasks until empty.
916		*/
917	<	final void runLocalTasks() {
918	<	if (base - top < 0) {
919	<	for (ForkJoinTask<?> t; (t = nextLocalTask()) != null; )
917	>	private void popAndExecAll() {
918	>	// A bit faster than repeated pop calls
919	>	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
920	>	while ((a = array) != null && (m = a.length - 1) >= 0 &&
921	>	(s = top - 1) - base >= 0 &&
922	>	(t = ((ForkJoinTask<?>)
923	>	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
924	>	!= null) {
925	>	if (U.compareAndSwapObject(a, j, t, null)) {
926	>	top = s;
927		t.doExec();
928	+	}
929		}
930		}
931
932		/**
933	+	* Polls and runs tasks until empty.
934	+	*/
935	+	private void pollAndExecAll() {
936	+	for (ForkJoinTask<?> t; (t = poll()) != null;)
937	+	t.doExec();
938	+	}
939	+
940	+	/**
941	+	* If present, removes from queue and executes the given task, or
942	+	* any other cancelled task. Returns (true) immediately on any CAS
943	+	* or consistency check failure so caller can retry.
944	+	*
945	+	* @return 0 if no progress can be made, else positive
946	+	* (this unusual convention simplifies use with tryHelpStealer.)
947	+	*/
948	+	final int tryRemoveAndExec(ForkJoinTask<?> task) {
949	+	int stat = 1;
950	+	boolean removed = false, empty = true;
951	+	ForkJoinTask<?>[] a; int m, s, b, n;
952	+	if ((a = array) != null && (m = a.length - 1) >= 0 &&
953	+	(n = (s = top) - (b = base)) > 0) {
954	+	for (ForkJoinTask<?> t;;) {           // traverse from s to b
955	+	int j = ((--s & m) << ASHIFT) + ABASE;
956	+	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
957	+	if (t == null)                    // inconsistent length
958	+	break;
959	+	else if (t == task) {
960	+	if (s + 1 == top) {           // pop
961	+	if (!U.compareAndSwapObject(a, j, task, null))
962	+	break;
963	+	top = s;
964	+	removed = true;
965	+	}
966	+	else if (base == b)           // replace with proxy
967	+	removed = U.compareAndSwapObject(a, j, task,
968	+	new EmptyTask());
969	+	break;
970	+	}
971	+	else if (t.status >= 0)
972	+	empty = false;
973	+	else if (s + 1 == top) {          // pop and throw away
974	+	if (U.compareAndSwapObject(a, j, t, null))
975	+	top = s;
976	+	break;
977	+	}
978	+	if (--n == 0) {
979	+	if (!empty && base == b)
980	+	stat = 0;
981	+	break;
982	+	}
983	+	}
984	+	}
985	+	if (removed)
986	+	task.doExec();
987	+	return stat;
988	+	}
989	+
990	+	/**
991	+	* Version of shared pop that takes top element only if it
992	+	* its root is the given CountedCompleter.
993	+	*/
994	+	final CountedCompleter<?> sharedPopCC(CountedCompleter<?> root) {
995	+	CountedCompleter<?> task = null;
996	+	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
997	+	try {
998	+	ForkJoinTask<?>[] a; int m;
999	+	if ((a = array) != null && (m = a.length - 1) >= 0) {
1000	+	outer:for (int s; (s = top - 1) - base >= 0;) {
1001	+	long j = ((m & s) << ASHIFT) + ABASE;
1002	+	ForkJoinTask<?> t =
1003	+	(ForkJoinTask<?>)U.getObject(a, j);
1004	+	if (t == null || !(t instanceof CountedCompleter))
1005	+	break;
1006	+	CountedCompleter<?> cc = (CountedCompleter<?>)t;
1007	+	for (CountedCompleter<?> q = cc, p;;) {
1008	+	if (q == root) {
1009	+	if (U.compareAndSwapObject(a, j, cc, null)) {
1010	+	top = s;
1011	+	task = cc;
1012	+	break outer;
1013	+	}
1014	+	break;
1015	+	}
1016	+	if ((p = q.completer) == null)
1017	+	break outer;
1018	+	q = p;
1019	+	}
1020	+	}
1021	+	}
1022	+	} finally {
1023	+	runState = 0;
1024	+	}
1025	+	}
1026	+	return task;
1027	+	}
1028	+
1029	+	/**
1030		* Executes a top-level task and any local tasks remaining
1031		* after execution.
1006	–	*
1007	–	* @return true unless terminating
1032		*/
1033	<	final boolean runTask(ForkJoinTask<?> t) {
1010	<	boolean alive = true;
1033	>	final void runTask(ForkJoinTask<?> t) {
1034		if (t != null) {
1035		currentSteal = t;
1036		t.doExec();
1037	<	runLocalTasks();
1037	>	if (top != base) {       // process remaining local tasks
1038	>	if (mode == 0)
1039	>	popAndExecAll();
1040	>	else
1041	>	pollAndExecAll();
1042	>	}
1043		++nsteals;
1044		currentSteal = null;
1045		}
1018	–	else if (runState < 0)            // terminating
1019	–	alive = false;
1020	–	return alive;
1046		}
1047
1048		/**
1049	<	* Executes a non-top-level (stolen) task
1049	>	* Executes a non-top-level (stolen) task.
1050		*/
1051		final void runSubtask(ForkJoinTask<?> t) {
1052		if (t != null) {
#	Line 1033 | Line 1058 | public class ForkJoinPool extends Abstra
1058		}
1059
1060		/**
1061	<	* Computes next value for random probes.  Scans don't require
1037	<	* a very high quality generator, but also not a crummy one.
1038	<	* Marsaglia xor-shift is cheap and works well enough.  Note:
1039	<	* This is manually inlined in several usages in ForkJoinPool
1040	<	* to avoid writes inside busy scan loops.
1061	>	* Returns true if owned and not known to be blocked.
1062		*/
1063	<	final int nextSeed() {
1064	<	int r = seed;
1065	<	r ^= r << 13;
1066	<	r ^= r >>> 17;
1067	<	r ^= r << 5;
1068	<	return seed = r;
1063	>	final boolean isApparentlyUnblocked() {
1064	>	Thread wt; Thread.State s;
1065	>	return (eventCount >= 0 &&
1066	>	(wt = owner) != null &&
1067	>	(s = wt.getState()) != Thread.State.BLOCKED &&
1068	>	s != Thread.State.WAITING &&
1069	>	s != Thread.State.TIMED_WAITING);
1070	>	}
1071	>
1072	>	/**
1073	>	* If this owned and is not already interrupted, try to
1074	>	* interrupt and/or unpark, ignoring exceptions.
1075	>	*/
1076	>	final void interruptOwner() {
1077	>	Thread wt, p;
1078	>	if ((wt = owner) != null && !wt.isInterrupted()) {
1079	>	try {
1080	>	wt.interrupt();
1081	>	} catch (SecurityException ignore) {
1082	>	}
1083	>	}
1084	>	if ((p = parker) != null)
1085	>	U.unpark(p);
1086		}
1087
1088		// Unsafe mechanics
#	Line 1072 | Line 1110 | public class ForkJoinPool extends Abstra
1110		}
1111
1112		/**
1113	<	* Class for artificial tasks that are used to replace the target
1114	<	* of local joins if they are removed from an interior queue slot
1115	<	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
1116	<	* actually do anything beyond having a unique identity.
1117	<	*/
1118	<	static final class EmptyTask extends ForkJoinTask<Void> {
1119	<	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
1120	<	public Void getRawResult() { return null; }
1121	<	public void setRawResult(Void x) {}
1122	<	public boolean exec() { return true; }
1113	>	* Per-thread records for threads that submit to pools. Currently
1114	>	* holds only pseudo-random seed / index that is used to choose
1115	>	* submission queues in method doSubmit. In the future, this may
1116	>	* also incorporate a means to implement different task rejection
1117	>	* and resubmission policies.
1118	>	*
1119	>	* Seeds for submitters and workers/workQueues work in basically
1120	>	* the same way but are initialized and updated using slightly
1121	>	* different mechanics. Both are initialized using the same
1122	>	* approach as in class ThreadLocal, where successive values are
1123	>	* unlikely to collide with previous values. This is done during
1124	>	* registration for workers, but requires a separate AtomicInteger
1125	>	* for submitters. Seeds are then randomly modified upon
1126	>	* collisions using xorshifts, which requires a non-zero seed.
1127	>	*/
1128	>	static final class Submitter {
1129	>	int seed;
1130	>	Submitter() {
1131	>	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1132	>	seed = (s == 0) ? 1 : s; // ensure non-zero
1133	>	}
1134		}
1135
1136	+	/** ThreadLocal class for Submitters */
1137	+	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1138	+	public Submitter initialValue() { return new Submitter(); }
1139	+	}
1140	+
1141	+	// static fields (initialized in static initializer below)
1142	+
1143		/**
1144	<	* Computes a hash code for the given thread. This method is
1145	<	* expected to provide higher-quality hash codes than those using
1090	<	* method hashCode().
1144	>	* Creates a new ForkJoinWorkerThread. This factory is used unless
1145	>	* overridden in ForkJoinPool constructors.
1146		*/
1147	<	static final int hashThread(Thread t) {
1148	<	long id = (t == null)? 0L : t.getId(); // Use MurmurHash of thread id
1149	<	int h = (int)id ^ (int)(id >>> 32);
1150	<	h ^= h >>> 16;
1151	<	h *= 0x85ebca6b;
1152	<	h ^= h >>> 13;
1153	<	h *= 0xc2b2ae35;
1154	<	return h ^ (h >>> 16);
1155	<	}
1147	>	public static final ForkJoinWorkerThreadFactory
1148	>	defaultForkJoinWorkerThreadFactory;
1149	>
1150	>	/** Property prefix for constructing common pool */
1151	>	private static final String propPrefix =
1152	>	"java.util.concurrent.ForkJoinPool.common.";
1153	>
1154	>	/**
1155	>	* Common (static) pool. Non-null for public use unless a static
1156	>	* construction exception, but internal usages must null-check on
1157	>	* use.
1158	>	*/
1159	>	static final ForkJoinPool commonPool;
1160	>
1161	>	/**
1162	>	* Common pool parallelism. Must equal commonPool.parallelism.
1163	>	*/
1164	>	static final int commonPoolParallelism;
1165	>
1166	>	/**
1167	>	* Generator for assigning sequence numbers as pool names.
1168	>	*/
1169	>	private static final AtomicInteger poolNumberGenerator;
1170	>
1171	>	/**
1172	>	* Generator for initial hashes/seeds for submitters. Accessed by
1173	>	* Submitter class constructor.
1174	>	*/
1175	>	static final AtomicInteger nextSubmitterSeed;
1176	>
1177	>	/**
1178	>	* Permission required for callers of methods that may start or
1179	>	* kill threads.
1180	>	*/
1181	>	private static final RuntimePermission modifyThreadPermission;
1182	>
1183	>	/**
1184	>	* Per-thread submission bookkeeping. Shared across all pools
1185	>	* to reduce ThreadLocal pollution and because random motion
1186	>	* to avoid contention in one pool is likely to hold for others.
1187	>	*/
1188	>	private static final ThreadSubmitter submitters;
1189	>
1190	>	// static constants
1191	>
1192	>	/**
1193	>	* Initial timeout value (in nanoseconds) for the thread triggering
1194	>	* quiescence to park waiting for new work. On timeout, the thread
1195	>	* will instead try to shrink the number of workers.
1196	>	*/
1197	>	private static final long IDLE_TIMEOUT      = 1000L * 1000L * 1000L; // 1sec
1198	>
1199	>	/**
1200	>	* Timeout value when there are more threads than parallelism level
1201	>	*/
1202	>	private static final long FAST_IDLE_TIMEOUT =  100L * 1000L * 1000L;
1203	>
1204	>	/**
1205	>	* The maximum stolen->joining link depth allowed in method
1206	>	* tryHelpStealer.  Must be a power of two. This value also
1207	>	* controls the maximum number of times to try to help join a task
1208	>	* without any apparent progress or change in pool state before
1209	>	* giving up and blocking (see awaitJoin).  Depths for legitimate
1210	>	* chains are unbounded, but we use a fixed constant to avoid
1211	>	* (otherwise unchecked) cycles and to bound staleness of
1212	>	* traversal parameters at the expense of sometimes blocking when
1213	>	* we could be helping.
1214	>	*/
1215	>	private static final int MAX_HELP = 64;
1216	>
1217	>	/**
1218	>	* Secondary time-based bound (in nanosecs) for helping attempts
1219	>	* before trying compensated blocking in awaitJoin. Used in
1220	>	* conjunction with MAX_HELP to reduce variance due to different
1221	>	* polling rates associated with different helping options. The
1222	>	* value should roughly approximate the time required to create
1223	>	* and/or activate a worker thread.
1224	>	*/
1225	>	private static final long COMPENSATION_DELAY = 1L << 18; // ~0.25 millisec
1226	>
1227	>	/**
1228	>	* Increment for seed generators. See class ThreadLocal for
1229	>	* explanation.
1230	>	*/
1231	>	private static final int SEED_INCREMENT = 0x61c88647;
1232
1233		/**
1234	<	* Top-level runloop for workers
1234	>	* Bits and masks for control variables
1235	>	*
1236	>	* Field ctl is a long packed with:
1237	>	* AC: Number of active running workers minus target parallelism (16 bits)
1238	>	* TC: Number of total workers minus target parallelism (16 bits)
1239	>	* ST: true if pool is terminating (1 bit)
1240	>	* EC: the wait count of top waiting thread (15 bits)
1241	>	* ID: poolIndex of top of Treiber stack of waiters (16 bits)
1242	>	*
1243	>	* When convenient, we can extract the upper 32 bits of counts and
1244	>	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
1245	>	* (int)ctl.  The ec field is never accessed alone, but always
1246	>	* together with id and st. The offsets of counts by the target
1247	>	* parallelism and the positionings of fields makes it possible to
1248	>	* perform the most common checks via sign tests of fields: When
1249	>	* ac is negative, there are not enough active workers, when tc is
1250	>	* negative, there are not enough total workers, and when e is
1251	>	* negative, the pool is terminating.  To deal with these possibly
1252	>	* negative fields, we use casts in and out of "short" and/or
1253	>	* signed shifts to maintain signedness.
1254	>	*
1255	>	* When a thread is queued (inactivated), its eventCount field is
1256	>	* set negative, which is the only way to tell if a worker is
1257	>	* prevented from executing tasks, even though it must continue to
1258	>	* scan for them to avoid queuing races. Note however that
1259	>	* eventCount updates lag releases so usage requires care.
1260	>	*
1261	>	* Field runState is an int packed with:
1262	>	* SHUTDOWN: true if shutdown is enabled (1 bit)
1263	>	* SEQ:  a sequence number updated upon (de)registering workers (30 bits)
1264	>	* INIT: set true after workQueues array construction (1 bit)
1265	>	*
1266	>	* The sequence number enables simple consistency checks:
1267	>	* Staleness of read-only operations on the workQueues array can
1268	>	* be checked by comparing runState before vs after the reads.
1269	>	*/
1270	>
1271	>	// bit positions/shifts for fields
1272	>	private static final int  AC_SHIFT   = 48;
1273	>	private static final int  TC_SHIFT   = 32;
1274	>	private static final int  ST_SHIFT   = 31;
1275	>	private static final int  EC_SHIFT   = 16;
1276	>
1277	>	// bounds
1278	>	private static final int  SMASK      = 0xffff;  // short bits
1279	>	private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
1280	>	private static final int  SQMASK     = 0xfffe;  // even short bits
1281	>	private static final int  SHORT_SIGN = 1 << 15;
1282	>	private static final int  INT_SIGN   = 1 << 31;
1283	>
1284	>	// masks
1285	>	private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
1286	>	private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
1287	>	private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
1288	>
1289	>	// units for incrementing and decrementing
1290	>	private static final long TC_UNIT    = 1L << TC_SHIFT;
1291	>	private static final long AC_UNIT    = 1L << AC_SHIFT;
1292	>
1293	>	// masks and units for dealing with u = (int)(ctl >>> 32)
1294	>	private static final int  UAC_SHIFT  = AC_SHIFT - 32;
1295	>	private static final int  UTC_SHIFT  = TC_SHIFT - 32;
1296	>	private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
1297	>	private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
1298	>	private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
1299	>	private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
1300	>
1301	>	// masks and units for dealing with e = (int)ctl
1302	>	private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
1303	>	private static final int E_SEQ       = 1 << EC_SHIFT;
1304	>
1305	>	// runState bits
1306	>	private static final int SHUTDOWN    = 1 << 31;
1307	>
1308	>	// access mode for WorkQueue
1309	>	static final int LIFO_QUEUE          =  0;
1310	>	static final int FIFO_QUEUE          =  1;
1311	>	static final int SHARED_QUEUE        = -1;
1312	>
1313	>	// Instance fields
1314	>
1315	>	/*
1316	>	* Field layout order in this class tends to matter more than one
1317	>	* would like. Runtime layout order is only loosely related to
1318	>	* declaration order and may differ across JVMs, but the following
1319	>	* empirically works OK on current JVMs.
1320		*/
1105	–	final void runWorker(ForkJoinWorkerThread wt) {
1106	–	WorkQueue w = wt.workQueue;
1107	–	w.growArray(false);     // Initialize queue array and seed in this thread
1108	–	w.seed = hashThread(Thread.currentThread()) | (1 << 31); // force < 0
1321
1322	<	do {} while (w.runTask(scan(w)));
1322	>	volatile long stealCount;                  // collects worker counts
1323	>	volatile long ctl;                         // main pool control
1324	>	final int parallelism;                     // parallelism level
1325	>	final int localMode;                       // per-worker scheduling mode
1326	>	volatile int nextWorkerNumber;             // to create worker name string
1327	>	final int submitMask;                      // submit queue index bound
1328	>	int nextSeed;                              // for initializing worker seeds
1329	>	volatile int mainLock;                     // spinlock for array updates
1330	>	volatile int runState;                     // shutdown status and seq
1331	>	WorkQueue[] workQueues;                    // main registry
1332	>	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1333	>	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1334	>	final String workerNamePrefix;             // to create worker name string
1335	>
1336	>	/*
1337	>	* Mechanics for main lock protecting worker array updates.  Uses
1338	>	* the same strategy as ConcurrentHashMap bins -- a spinLock for
1339	>	* normal cases, but falling back to builtin lock when (rarely)
1340	>	* needed.  See internal ConcurrentHashMap documentation for
1341	>	* explanation.
1342	>	*/
1343	>
1344	>	static final int LOCK_WAITING = 2; // bit to indicate need for signal
1345	>	static final int MAX_LOCK_SPINS = 1 << 8;
1346	>
1347	>	private void tryAwaitMainLock() {
1348	>	int spins = MAX_LOCK_SPINS, r = 0, h;
1349	>	while (((h = mainLock) & 1) != 0) {
1350	>	if (r == 0)
1351	>	r = ThreadLocalRandom.current().nextInt(); // randomize spins
1352	>	else if (spins >= 0) {
1353	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1354	>	if (r >= 0)
1355	>	--spins;
1356	>	}
1357	>	else if (U.compareAndSwapInt(this, MAINLOCK, h, h | LOCK_WAITING)) {
1358	>	synchronized (this) {
1359	>	if ((mainLock & LOCK_WAITING) != 0) {
1360	>	try {
1361	>	wait();
1362	>	} catch (InterruptedException ie) {
1363	>	try {
1364	>	Thread.currentThread().interrupt();
1365	>	} catch (SecurityException ignore) {
1366	>	}
1367	>	}
1368	>	}
1369	>	else
1370	>	notifyAll(); // possibly won race vs signaller
1371	>	}
1372	>	break;
1373	>	}
1374	>	}
1375		}
1376
1377	<	// Creating, registering and deregistering workers
1377	>	//  Creating, registering, and deregistering workers
1378
1379		/**
1380		* Tries to create and start a worker
1381		*/
1382		private void addWorker() {
1383		Throwable ex = null;
1384	<	ForkJoinWorkerThread w = null;
1384	>	ForkJoinWorkerThread wt = null;
1385		try {
1386	<	if ((w = factory.newThread(this)) != null) {
1387	<	w.start();
1386	>	if ((wt = factory.newThread(this)) != null) {
1387	>	wt.start();
1388		return;
1389		}
1390		} catch (Throwable e) {
1391		ex = e;
1392		}
1393	<	deregisterWorker(w, ex);
1393	>	deregisterWorker(wt, ex); // adjust counts etc on failure
1394		}
1395
1396		/**
#	Line 1136 | Line 1400 | public class ForkJoinPool extends Abstra
1400		* ForkJoinWorkerThread.
1401		*/
1402		final String nextWorkerName() {
1403	<	return workerNamePrefix.concat
1404	<	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1403	>	int n;
1404	>	do {} while (!U.compareAndSwapInt(this, NEXTWORKERNUMBER,
1405	>	n = nextWorkerNumber, ++n));
1406	>	return workerNamePrefix.concat(Integer.toString(n));
1407		}
1408
1409		/**
1410	<	* Callback from ForkJoinWorkerThread constructor to establish and
1411	<	* record its WorkQueue
1410	>	* Callback from ForkJoinWorkerThread constructor to establish its
1411	>	* poolIndex and record its WorkQueue. To avoid scanning bias due
1412	>	* to packing entries in front of the workQueues array, we treat
1413	>	* the array as a simple power-of-two hash table using per-thread
1414	>	* seed as hash, expanding as needed.
1415		*
1416	<	* @param wt the worker thread
1416	>	* @param w the worker's queue
1417		*/
1418	<	final void registerWorker(ForkJoinWorkerThread wt) {
1419	<	WorkQueue w = wt.workQueue;
1420	<	ReentrantLock lock = this.lock;
1152	<	lock.lock();
1418	>	final void registerWorker(WorkQueue w) {
1419	>	while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1420	>	tryAwaitMainLock();
1421		try {
1422	<	int k = nextPoolIndex;
1423	<	WorkQueue[] ws = workQueues;
1424	<	if (ws != null) {                       // ignore on shutdown
1425	<	int n = ws.length;
1426	<	if (k < 0 || (k & 1) == 0 || k >= n || ws[k] != null) {
1427	<	for (k = 1; k < n && ws[k] != null; k += 2)
1428	<	;                           // workers are at odd indices
1429	<	if (k >= n)                     // resize
1430	<	workQueues = ws = Arrays.copyOf(ws, n << 1);
1431	<	}
1432	<	w.poolIndex = k;
1433	<	w.eventCount = ~(k >>> 1) & SMASK;  // Set up wait count
1434	<	ws[k] = w;                          // record worker
1435	<	nextPoolIndex = k + 2;
1436	<	int rs = runState;
1437	<	int m = rs & SMASK;                 // recalculate runState mask
1438	<	if (k > m)
1439	<	m = (m << 1) + 1;
1440	<	runState = (rs & SHUTDOWN) | ((rs + RS_SEQ) & RS_SEQ_MASK) | m;
1422	>	WorkQueue[] ws;
1423	>	if ((ws = workQueues) == null)
1424	>	ws = workQueues = new WorkQueue[submitMask + 1];
1425	>	if (w != null) {
1426	>	int rs, n =  ws.length, m = n - 1;
1427	>	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1428	>	w.seed = (s == 0) ? 1 : s;          // ensure non-zero seed
1429	>	int r = (s << 1) | 1;               // use odd-numbered indices
1430	>	if (ws[r &= m] != null) {           // collision
1431	>	int probes = 0;                 // step by approx half size
1432	>	int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1433	>	while (ws[r = (r + step) & m] != null) {
1434	>	if (++probes >= n) {
1435	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1436	>	m = n - 1;
1437	>	probes = 0;
1438	>	}
1439	>	}
1440	>	}
1441	>	w.eventCount = w.poolIndex = r;     // establish before recording
1442	>	ws[r] = w;                          // also update seq
1443	>	runState = ((rs = runState) & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN);
1444		}
1445		} finally {
1446	<	lock.unlock();
1446	>	if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1447	>	mainLock = 0;
1448	>	synchronized (this) { notifyAll(); };
1449	>	}
1450		}
1451		}
1452
1453		/**
1454	<	* Final callback from terminating worker, as well as failure to
1455	<	* construct or start a worker in addWorker.  Removes record of
1454	>	* Final callback from terminating worker, as well as upon failure
1455	>	* to construct or start a worker in addWorker.  Removes record of
1456		* worker from array, and adjusts counts. If pool is shutting
1457		* down, tries to complete termination.
1458		*
#	Line 1189 | Line 1463 | public class ForkJoinPool extends Abstra
1463		WorkQueue w = null;
1464		if (wt != null && (w = wt.workQueue) != null) {
1465		w.runState = -1;                // ensure runState is set
1466	<	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1466	>	long steals = w.totalSteals + w.nsteals, sc;
1467	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1468	>	sc = stealCount, sc + steals));
1469		int idx = w.poolIndex;
1470	<	ReentrantLock lock = this.lock;
1471	<	lock.lock();
1472	<	try {                           // remove record from array
1470	>	while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1471	>	tryAwaitMainLock();
1472	>	try {
1473		WorkQueue[] ws = workQueues;
1474		if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1475	<	ws[nextPoolIndex = idx] = null;
1475	>	ws[idx] = null;
1476		} finally {
1477	<	lock.unlock();
1477	>	if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1478	>	mainLock = 0;
1479	>	synchronized (this) { notifyAll(); };
1480	>	}
1481		}
1482		}
1483
#	Line 1208 | Line 1487 | public class ForkJoinPool extends Abstra
1487		((c - TC_UNIT) & TC_MASK) |
1488		(c & ~(AC_MASK|TC_MASK)))));
1489
1490	<	if (!tryTerminate(false) && w != null) {
1490	>	if (!tryTerminate(false, false) && w != null) {
1491		w.cancelAll();                  // cancel remaining tasks
1492		if (w.array != null)            // suppress signal if never ran
1493		signalWork();               // wake up or create replacement
1494	+	if (ex == null)                 // help clean refs on way out
1495	+	ForkJoinTask.helpExpungeStaleExceptions();
1496		}
1497
1498		if (ex != null)                     // rethrow
1499	<	U.throwException(ex);
1499	>	ForkJoinTask.rethrow(ex);
1500		}
1501
1502	<
1222	<	// Maintaining ctl counts
1223	<
1224	<	/**
1225	<	* Increments active count; mainly called upon return from blocking
1226	<	*/
1227	<	final void incrementActiveCount() {
1228	<	long c;
1229	<	do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1230	<	}
1502	>	// Submissions
1503
1504		/**
1505	<	* Activates or creates a worker
1505	>	* Unless shutting down, adds the given task to a submission queue
1506	>	* at submitter's current queue index (modulo submission
1507	>	* range). If no queue exists at the index, one is created.  If
1508	>	* the queue is busy, another index is randomly chosen. The
1509	>	* submitMask bounds the effective number of queues to the
1510	>	* (nearest power of two for) parallelism level.
1511	>	*
1512	>	* @param task the task. Caller must ensure non-null.
1513		*/
1514	<	final void signalWork() {
1515	<	/*
1516	<	* The while condition is true if: (there is are too few total
1517	<	* workers OR there is at least one waiter) AND (there are too
1518	<	* few active workers OR the pool is terminating).  The value
1519	<	* of e distinguishes the remaining cases: zero (no waiters)
1520	<	* for create, negative if terminating (in which case do
1521	<	* nothing), else release a waiter. The secondary checks for
1522	<	* release (non-null array etc) can fail if the pool begins
1523	<	* terminating after the test, and don't impose any added cost
1524	<	* because JVMs must perform null and bounds checks anyway.
1525	<	*/
1526	<	long c; int e, u;
1527	<	while ((((e = (int)(c = ctl)) | (u = (int)(c >>> 32))) &
1528	<	(INT_SIGN|SHORT_SIGN)) == (INT_SIGN|SHORT_SIGN)) {
1529	<	WorkQueue[] ws = workQueues; int i; WorkQueue w; Thread p;
1530	<	if (e == 0) {                    // add a new worker
1531	<	if (U.compareAndSwapLong
1253	<	(this, CTL, c, (long)(((u + UTC_UNIT) & UTC_MASK) |
1254	<	((u + UAC_UNIT) & UAC_MASK)) << 32)) {
1255	<	addWorker();
1256	<	break;
1514	>	private void doSubmit(ForkJoinTask<?> task) {
1515	>	Submitter s = submitters.get();
1516	>	for (int r = s.seed, m = submitMask;;) {
1517	>	WorkQueue[] ws; WorkQueue q;
1518	>	int k = r & m & SQMASK;          // use only even indices
1519	>	if (runState < 0)
1520	>	throw new RejectedExecutionException(); // shutting down
1521	>	else if ((ws = workQueues) == null || ws.length <= k) {
1522	>	while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1523	>	tryAwaitMainLock();
1524	>	try {
1525	>	if (workQueues == null)
1526	>	workQueues = new WorkQueue[submitMask + 1];
1527	>	} finally {
1528	>	if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1529	>	mainLock = 0;
1530	>	synchronized (this) { notifyAll(); };
1531	>	}
1532		}
1533		}
1534	<	else if (e > 0 && ws != null &&
1535	<	(i = ((~e << 1) | 1) & SMASK) < ws.length &&
1536	<	(w = ws[i]) != null &&
1537	<	w.eventCount == (e | INT_SIGN)) {
1538	<	if (U.compareAndSwapLong
1539	<	(this, CTL, c, (((long)(w.nextWait & E_MASK)) |
1540	<	((long)(u + UAC_UNIT) << 32)))) {
1541	<	w.eventCount = (e + E_SEQ) & E_MASK;
1542	<	if ((p = w.parker) != null)
1543	<	U.unpark(p);         // release a waiting worker
1544	<	break;
1534	>	else if ((q = ws[k]) == null) {  // create new queue
1535	>	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1536	>	while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1537	>	tryAwaitMainLock();
1538	>	try {
1539	>	int rs = runState;       // to update seq
1540	>	if (ws == workQueues && ws[k] == null) {
1541	>	ws[k] = nq;
1542	>	runState = ((rs & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN));
1543	>	}
1544	>	} finally {
1545	>	if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1546	>	mainLock = 0;
1547	>	synchronized (this) { notifyAll(); };
1548	>	}
1549		}
1550		}
1551	+	else if (q.trySharedPush(task)) {
1552	+	signalWork();
1553	+	return;
1554	+	}
1555	+	else if (m > 1) {                // move to a different index
1556	+	r ^= r << 13;                // same xorshift as WorkQueues
1557	+	r ^= r >>> 17;
1558	+	s.seed = r ^= r << 5;
1559	+	}
1560		else
1561	<	break;
1561	>	Thread.yield();              // yield if no alternatives
1562		}
1563		}
1564
1565		/**
1566	<	* Tries to decrement active count (sometimes implicitly) and
1279	<	* possibly release or create a compensating worker in preparation
1280	<	* for blocking. Fails on contention or termination.
1281	<	*
1282	<	* @return true if the caller can block, else should recheck and retry
1566	>	* Submits the given (non-null) task to the common pool, if possible.
1567		*/
1568	<	final boolean tryCompensate() {
1569	<	WorkQueue[] ws; WorkQueue w; Thread p;
1570	<	int pc = parallelism, e, u, ac, tc, i;
1571	<	long c = ctl;
1568	>	static void submitToCommonPool(ForkJoinTask<?> task) {
1569	>	ForkJoinPool p;
1570	>	if ((p = commonPool) == null)
1571	>	throw new RejectedExecutionException("Common Pool Unavailable");
1572	>	p.doSubmit(task);
1573	>	}
1574
1575	<	if ((e = (int)c) >= 0) {
1576	<	if ((ac = ((u = (int)(c >>> 32)) >> UAC_SHIFT)) <= 0 &&
1577	<	e != 0 && (ws = workQueues) != null &&
1578	<	(i = ((~e << 1) | 1) & SMASK) < ws.length &&
1579	<	(w = ws[i]) != null) {
1580	<	if (w.eventCount == (e | INT_SIGN) &&
1581	<	U.compareAndSwapLong
1582	<	(this, CTL, c, ((long)(w.nextWait & E_MASK) |
1583	<	(c & (AC_MASK|TC_MASK))))) {
1584	<	w.eventCount = (e + E_SEQ) & E_MASK;
1585	<	if ((p = w.parker) != null)
1586	<	U.unpark(p);
1587	<	return true;             // release an idle worker
1588	<	}
1589	<	}
1590	<	else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
1591	<	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1592	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1593	<	return true;             // no compensation needed
1594	<	}
1595	<	else if (tc + pc < MAX_ID) {
1596	<	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1597	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1598	<	addWorker();
1599	<	return true;             // create replacement
1575	>	/**
1576	>	* Returns true if the given task was submitted to common pool
1577	>	* and has not yet commenced execution, and is available for
1578	>	* removal according to execution policies; if so removing the
1579	>	* submission from the pool.
1580	>	*
1581	>	* @param task the task
1582	>	* @return true if successful
1583	>	*/
1584	>	static boolean tryUnsubmitFromCommonPool(ForkJoinTask<?> task) {
1585	>	// If not oversaturating platform, peek, looking for task and
1586	>	// eligibility before using trySharedUnpush to actually take
1587	>	// it under lock
1588	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue w, q;
1589	>	ForkJoinTask<?>[] a; int ac, s, m;
1590	>	if ((p = commonPool) != null && (ws = p.workQueues) != null) {
1591	>	int k = submitters.get().seed & p.submitMask & SQMASK;
1592	>	if ((m = ws.length - 1) >= k && (q = ws[k]) != null &&
1593	>	(ac = (int)(p.ctl >> AC_SHIFT)) <= 0) {
1594	>	if (ac == 0) { // double check if all workers active
1595	>	for (int i = 1; i <= m; i += 2) {
1596	>	if ((w = ws[i]) != null && w.parker != null) {
1597	>	ac = -1;
1598	>	break;
1599	>	}
1600	>	}
1601		}
1602	+	return (ac < 0 && (a = q.array) != null &&
1603	+	(s = q.top - 1) - q.base >= 0 &&
1604	+	s >= 0 && s < a.length &&
1605	+	a[s] == task &&
1606	+	q.trySharedUnpush(task));
1607		}
1608		}
1609		return false;
1610		}
1611
1320	–	// Submissions
1321	–
1612		/**
1613	<	* Unless shutting down, adds the given task to some submission
1324	<	* queue; using a randomly chosen queue index if the caller is a
1325	<	* ForkJoinWorkerThread, else one based on caller thread's hash
1326	<	* code. If no queue exists at the index, one is created.  If the
1327	<	* queue is busy, another is chosen by sweeping through the queues
1328	<	* array.
1613	>	* Tries to pop and run a task within same computation from common pool
1614		*/
1615	<	private void doSubmit(ForkJoinTask<?> task) {
1616	<	if (task == null)
1617	<	throw new NullPointerException();
1618	<	Thread t = Thread.currentThread();
1619	<	int r = ((t instanceof ForkJoinWorkerThread) ?
1620	<	((ForkJoinWorkerThread)t).workQueue.nextSeed() : hashThread(t));
1621	<	for (;;) {
1622	<	int rs = runState, m = rs & SMASK;
1623	<	int j = r &= (m & ~1);                      // even numbered queues
1624	<	WorkQueue[] ws = workQueues;
1625	<	if (rs < 0 || ws == null)
1626	<	throw new RejectedExecutionException(); // shutting down
1627	<	if (ws.length > m) {                        // consistency check
1628	<	for (WorkQueue q;;) {                   // circular sweep
1629	<	if (((q = ws[j]) != null ||
1345	<	(q = tryAddSharedQueue(j)) != null) &&
1346	<	q.trySharedPush(task)) {
1347	<	signalWork();
1348	<	return;
1349	<	}
1350	<	if ((j = (j + 2) & m) == r) {
1351	<	Thread.yield();                 // all queues busy
1352	<	break;
1615	>	static void popAndExecCCFromCommonPool(CountedCompleter<?> cc) {
1616	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; int m, ac;
1617	>	CountedCompleter<?> par, task;
1618	>	if ((p = commonPool) != null && (ws = p.workQueues) != null) {
1619	>	while ((par = cc.completer) != null) // find root
1620	>	cc = par;
1621	>	int k = submitters.get().seed & p.submitMask & SQMASK;
1622	>	if ((m = ws.length - 1) >= k && (q = ws[k]) != null &&
1623	>	(ac = (int)(p.ctl >> AC_SHIFT)) <= 0) {
1624	>	if (ac == 0) {
1625	>	for (int i = 1; i <= m; i += 2) {
1626	>	if ((w = ws[i]) != null && w.parker != null) {
1627	>	ac = -1;
1628	>	break;
1629	>	}
1630		}
1631		}
1632	+	if (ac < 0 && q.top - q.base > 0 &&
1633	+	(task = q.sharedPopCC(cc)) != null)
1634	+	task.exec();
1635		}
1636		}
1637		}
1638
1639	+	// Maintaining ctl counts
1640	+
1641		/**
1642	<	* Tries to add and register a new queue at the given index.
1643	<	*
1644	<	* @param idx the workQueues array index to register the queue
1645	<	* @return the queue, or null if could not add because could
1646	<	* not acquire lock or idx is unusable
1647	<	*/
1648	<	private WorkQueue tryAddSharedQueue(int idx) {
1649	<	WorkQueue q = null;
1650	<	ReentrantLock lock = this.lock;
1651	<	if (idx >= 0 && (idx & 1) == 0 && !lock.isLocked()) {
1652	<	// create queue outside of lock but only if apparently free
1653	<	WorkQueue nq = new WorkQueue(null, SHARED_QUEUE);
1654	<	if (lock.tryLock()) {
1655	<	try {
1656	<	WorkQueue[] ws = workQueues;
1657	<	if (ws != null && idx < ws.length) {
1658	<	if ((q = ws[idx]) == null) {
1659	<	int rs;         // update runState seq
1660	<	ws[idx] = q = nq;
1661	<	runState = (((rs = runState) & SHUTDOWN) |
1662	<	((rs + RS_SEQ) & ~SHUTDOWN));
1663	<	}
1642	>	* Increments active count; mainly called upon return from blocking.
1643	>	*/
1644	>	final void incrementActiveCount() {
1645	>	long c;
1646	>	do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1647	>	}
1648	>
1649	>	/**
1650	>	* Tries to create one or activate one or more workers if too few are active.
1651	>	*/
1652	>	final void signalWork() {
1653	>	long c; int u;
1654	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {     // too few active
1655	>	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1656	>	if ((e = (int)c) > 0) {                     // at least one waiting
1657	>	if (ws != null && (i = e & SMASK) < ws.length &&
1658	>	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1659	>	long nc = (((long)(w.nextWait & E_MASK)) |
1660	>	((long)(u + UAC_UNIT) << 32));
1661	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1662	>	w.eventCount = (e + E_SEQ) & E_MASK;
1663	>	if ((p = w.parker) != null)
1664	>	U.unpark(p);                // activate and release
1665	>	break;
1666		}
1667	<	} finally {
1668	<	lock.unlock();
1667	>	}
1668	>	else
1669	>	break;
1670	>	}
1671	>	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1672	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1673	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1674	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1675	>	addWorker();
1676	>	break;
1677		}
1678		}
1679	+	else
1680	+	break;
1681		}
1388	–	return q;
1682		}
1683
1684		// Scanning for tasks
1685
1686		/**
1687	+	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1688	+	*/
1689	+	final void runWorker(WorkQueue w) {
1690	+	w.growArray(false);         // initialize queue array in this thread
1691	+	do { w.runTask(scan(w)); } while (w.runState >= 0);
1692	+	}
1693	+
1694	+	/**
1695		* Scans for and, if found, returns one task, else possibly
1696		* inactivates the worker. This method operates on single reads of
1697	<	* volatile state and is designed to be re-invoked continuously in
1698	<	* part because it returns upon detecting inconsistencies,
1697	>	* volatile state and is designed to be re-invoked continuously,
1698	>	* in part because it returns upon detecting inconsistencies,
1699		* contention, or state changes that indicate possible success on
1700		* re-invocation.
1701		*
1702	<	* The scan searches for tasks across queues, randomly selecting
1703	<	* the first #queues probes, favoring steals 2:1 over submissions
1704	<	* (by exploiting even/odd indexing), and then performing a
1705	<	* circular sweep of all queues.  The scan terminates upon either
1706	<	* finding a non-empty queue, or completing a full sweep. If the
1707	<	* worker is not inactivated, it takes and returns a task from
1708	<	* this queue.  On failure to find a task, we take one of the
1709	<	* following actions, after which the caller will retry calling
1710	<	* this method unless terminated.
1702	>	* The scan searches for tasks across a random permutation of
1703	>	* queues (starting at a random index and stepping by a random
1704	>	* relative prime, checking each at least once).  The scan
1705	>	* terminates upon either finding a non-empty queue, or completing
1706	>	* the sweep. If the worker is not inactivated, it takes and
1707	>	* returns a task from this queue.  On failure to find a task, we
1708	>	* take one of the following actions, after which the caller will
1709	>	* retry calling this method unless terminated.
1710	>	*
1711	>	* * If pool is terminating, terminate the worker.
1712		*
1713		* * If not a complete sweep, try to release a waiting worker.  If
1714		* the scan terminated because the worker is inactivated, then the
#	Line 1415 | Line 1717 | public class ForkJoinPool extends Abstra
1717		* another worker, but with same net effect. Releasing in other
1718		* cases as well ensures that we have enough workers running.
1719		*
1418	–	* * If the caller has run a task since the the last empty scan,
1419	–	* return (to allow rescan) if other workers are not also yet
1420	–	* enqueued.  Field WorkQueue.rescans counts down on each scan to
1421	–	* ensure eventual inactivation, and occasional calls to
1422	–	* Thread.yield to help avoid interference with more useful
1423	–	* activities on the system.
1424	–	*
1425	–	* * If pool is terminating, terminate the worker
1426	–	*
1720		* * If not already enqueued, try to inactivate and enqueue the
1721	<	* worker on wait queue.
1721	>	* worker on wait queue. Or, if inactivating has caused the pool
1722	>	* to be quiescent, relay to idleAwaitWork to check for
1723	>	* termination and possibly shrink pool.
1724	>	*
1725	>	* * If already inactive, and the caller has run a task since the
1726	>	* last empty scan, return (to allow rescan) unless others are
1727	>	* also inactivated.  Field WorkQueue.rescans counts down on each
1728	>	* scan to ensure eventual inactivation and blocking.
1729		*
1730	<	* * If already enqueued and none of the above apply, either park
1731	<	* awaiting signal, or if this is the most recent waiter and pool
1432	<	* is quiescent, relay to idleAwaitWork to check for termination
1433	<	* and possibly shrink pool.
1730	>	* * If already enqueued and none of the above apply, park
1731	>	* awaiting signal,
1732		*
1733		* @param w the worker (via its WorkQueue)
1734	<	* @return a task or null of none found
1734	>	* @return a task or null if none found
1735		*/
1736		private final ForkJoinTask<?> scan(WorkQueue w) {
1737	<	boolean swept = false;                 // true after full empty scan
1738	<	WorkQueue[] ws;                        // volatile read order matters
1739	<	int r = w.seed, ec = w.eventCount;     // ec is negative if inactive
1740	<	int rs = runState, m = rs & SMASK;
1741	<	if ((ws = workQueues) != null && ws.length > m) {
1742	<	ForkJoinTask<?> task = null;
1743	<	for (int k = 0, j = -2 - m; ; ++j) {
1744	<	WorkQueue q; int b;
1745	<	if (j < 0) {                    // random probes while j negative
1746	<	r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) | (j & 1);
1747	<	}                               // worker (not submit) for odd j
1748	<	else                            // cyclic scan when j >= 0
1749	<	k += (m >>> 1) | 1;         // step by half to reduce bias
1750	<
1751	<	if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
1752	<	if (ec >= 0)
1753	<	task = q.pollAt(b);     // steal
1754	<	break;
1737	>	WorkQueue[] ws;                       // first update random seed
1738	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1739	>	int rs = runState, m;                 // volatile read order matters
1740	>	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1741	>	int ec = w.eventCount;            // ec is negative if inactive
1742	>	int step = (r >>> 16) | 1;        // relative prime
1743	>	for (int j = (m + 1) << 2; ; r += step) {
1744	>	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1745	>	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1746	>	(a = q.array) != null) {  // probably nonempty
1747	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1748	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1749	>	if (q.base == b && ec >= 0 && t != null &&
1750	>	U.compareAndSwapObject(a, i, t, null)) {
1751	>	if (q.top - (q.base = b + 1) > 0)
1752	>	signalWork();    // help pushes signal
1753	>	return t;
1754	>	}
1755	>	else if (ec < 0 || j <= m) {
1756	>	rs = 0;               // mark scan as imcomplete
1757	>	break;                // caller can retry after release
1758	>	}
1759		}
1760	<	else if (j > m) {
1459	<	if (rs == runState)        // staleness check
1460	<	swept = true;
1760	>	if (--j < 0)
1761		break;
1762	+	}
1763	+
1764	+	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1765	+	if (e < 0)                        // decode ctl on empty scan
1766	+	w.runState = -1;              // pool is terminating
1767	+	else if (rs == 0 || rs != runState) { // incomplete scan
1768	+	WorkQueue v; Thread p;        // try to release a waiter
1769	+	if (e > 0 && a < 0 && w.eventCount == ec &&
1770	+	(v = ws[e & m]) != null && v.eventCount == (e | INT_SIGN)) {
1771	+	long nc = ((long)(v.nextWait & E_MASK) |
1772	+	((c + AC_UNIT) & (AC_MASK|TC_MASK)));
1773	+	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1774	+	v.eventCount = (e + E_SEQ) & E_MASK;
1775	+	if ((p = v.parker) != null)
1776	+	U.unpark(p);
1777	+	}
1778		}
1779		}
1780	<	w.seed = r;                        // save seed for next scan
1781	<	if (task != null)
1782	<	return task;
1783	<	}
1784	<
1785	<	// Decode ctl on empty scan
1786	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1787	<	if (!swept) {                          // try to release a waiter
1788	<	WorkQueue v; Thread p;
1789	<	if (e > 0 && a < 0 && ws != null &&
1790	<	(v = ws[((~e << 1) | 1) & m]) != null &&
1791	<	v.eventCount == (e | INT_SIGN) && U.compareAndSwapLong
1792	<	(this, CTL, c, ((long)(v.nextWait & E_MASK) |
1793	<	((c + AC_UNIT) & (AC_MASK|TC_MASK))))) {
1794	<	v.eventCount = (e + E_SEQ) & E_MASK;
1795	<	if ((p = v.parker) != null)
1796	<	U.unpark(p);
1797	<	}
1798	<	}
1799	<	else if ((nr = w.rescans) > 0) {       // continue rescanning
1800	<	int ac = a + parallelism;
1801	<	if ((w.rescans = (ac < nr) ? ac : nr - 1) > 0 && w.seed < 0 &&
1802	<	w.eventCount == ec)
1803	<	Thread.yield();                // 1 bit randomness for yield call
1804	<	}
1805	<	else if (e < 0)                        // pool is terminating
1806	<	w.runState = -1;
1807	<	else if (ec >= 0) {                    // try to enqueue
1808	<	long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
1809	<	w.nextWait = e;
1494	<	w.eventCount = ec | INT_SIGN;      // mark as inactive
1495	<	if (!U.compareAndSwapLong(this, CTL, c, nc))
1496	<	w.eventCount = ec;             // back out on CAS failure
1497	<	else if ((ns = w.nsteals) != 0) {  // set rescans if ran task
1498	<	if (a <= 0)                    // ... unless too many active
1499	<	w.rescans = a + parallelism;
1500	<	w.nsteals = 0;
1501	<	w.totalSteals += ns;
1502	<	}
1503	<	}
1504	<	else{                                  // already queued
1505	<	if (parallelism == -a)
1506	<	idleAwaitWork(w);              // quiescent
1507	<	if (w.eventCount == ec) {
1508	<	Thread.interrupted();          // clear status
1509	<	ForkJoinWorkerThread wt = w.owner;
1510	<	U.putObject(wt, PARKBLOCKER, this);
1511	<	w.parker = wt;                 // emulate LockSupport.park
1512	<	if (w.eventCount == ec)        // recheck
1513	<	U.park(false, 0L);         // block
1514	<	w.parker = null;
1515	<	U.putObject(wt, PARKBLOCKER, null);
1780	>	else if (ec >= 0) {               // try to enqueue/inactivate
1781	>	long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
1782	>	w.nextWait = e;
1783	>	w.eventCount = ec | INT_SIGN; // mark as inactive
1784	>	if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc))
1785	>	w.eventCount = ec;        // unmark on CAS failure
1786	>	else {
1787	>	if ((ns = w.nsteals) != 0) {
1788	>	w.nsteals = 0;        // set rescans if ran task
1789	>	w.rescans = (a > 0) ? 0 : a + parallelism;
1790	>	w.totalSteals += ns;
1791	>	}
1792	>	if (a == 1 - parallelism) // quiescent
1793	>	idleAwaitWork(w, nc, c);
1794	>	}
1795	>	}
1796	>	else if (w.eventCount < 0) {      // already queued
1797	>	int ac = a + parallelism;
1798	>	if ((nr = w.rescans) > 0)     // continue rescanning
1799	>	w.rescans = (ac < nr) ? ac : nr - 1;
1800	>	else if (((w.seed >>> 16) & ac) == 0) { // randomize park
1801	>	Thread.interrupted();     // clear status
1802	>	Thread wt = Thread.currentThread();
1803	>	U.putObject(wt, PARKBLOCKER, this);
1804	>	w.parker = wt;            // emulate LockSupport.park
1805	>	if (w.eventCount < 0)     // recheck
1806	>	U.park(false, 0L);
1807	>	w.parker = null;
1808	>	U.putObject(wt, PARKBLOCKER, null);
1809	>	}
1810		}
1811		}
1812		return null;
1813		}
1814
1815		/**
1816	<	* If inactivating worker w has caused pool to become quiescent,
1817	<	* check for pool termination, and, so long as this is not the
1818	<	* only worker, wait for event for up to SHRINK_RATE nanosecs On
1819	<	* timeout, if ctl has not changed, terminate the worker, which
1820	<	* will in turn wake up another worker to possibly repeat this
1821	<	* process.
1816	>	* If inactivating worker w has caused the pool to become
1817	>	* quiescent, checks for pool termination, and, so long as this is
1818	>	* not the only worker, waits for event for up to a given
1819	>	* duration.  On timeout, if ctl has not changed, terminates the
1820	>	* worker, which will in turn wake up another worker to possibly
1821	>	* repeat this process.
1822		*
1823		* @param w the calling worker
1824	+	* @param currentCtl the ctl value triggering possible quiescence
1825	+	* @param prevCtl the ctl value to restore if thread is terminated
1826		*/
1827	<	private void idleAwaitWork(WorkQueue w) {
1828	<	long c; int nw, ec;
1829	<	if (!tryTerminate(false) &&
1830	<	(int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
1831	<	(ec = w.eventCount) == ((int)c | INT_SIGN) &&
1832	<	(nw = w.nextWait) != 0) {
1833	<	long nc = ((long)(nw & E_MASK) | // ctl to restore on timeout
1834	<	((c + AC_UNIT) & AC_MASK) | (c & TC_MASK));
1539	<	ForkJoinTask.helpExpungeStaleExceptions(); // help clean
1540	<	ForkJoinWorkerThread wt = w.owner;
1541	<	while (ctl == c) {
1542	<	long startTime = System.nanoTime();
1827	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1828	>	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1829	>	(int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1830	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1831	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1832	>	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1833	>	Thread wt = Thread.currentThread();
1834	>	while (ctl == currentCtl) {
1835		Thread.interrupted();  // timed variant of version in scan()
1836		U.putObject(wt, PARKBLOCKER, this);
1837		w.parker = wt;
1838	<	if (ctl == c)
1839	<	U.park(false, SHRINK_RATE);
1838	>	if (ctl == currentCtl)
1839	>	U.park(false, parkTime);
1840		w.parker = null;
1841		U.putObject(wt, PARKBLOCKER, null);
1842	<	if (ctl != c)
1842	>	if (ctl != currentCtl)
1843		break;
1844	<	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1845	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1846	<	w.runState = -1;          // shrink
1847	<	w.eventCount = (ec + E_SEQ) | E_MASK;
1844	>	if (deadline - System.nanoTime() <= 0L &&
1845	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1846	>	w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1847	>	w.runState = -1;   // shrink
1848		break;
1849		}
1850		}
#	Line 1570 | 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
1574	<	* 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
1590	<	WorkQueue stealer = null;
1591	<	WorkQueue v = ws[j.stealHint & m];
1592	<	if (v != null && v.currentSteal == subtask)
1593	<	stealer = v;
1594	<	else {
1595	<	for (int i = 1; i <= m; i += 2) {
1596	<	if ((v = ws[i]) != null && v.currentSteal == subtask) {
1597	<	stealer = v;
1598	<	j.stealHint = i; // save hint
1599	<	break;
1600	<	}
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 (stealer == null)
1884	<	break;
1885	<	}
1886	<
1887	<	for (WorkQueue q = stealer;;) { // Try to help stealer
1888	<	ForkJoinTask<?> t; int b;
1889	<	if (task.status < 0)
1890	<	break outer;
1891	<	if ((b = q.base) - q.top < 0) {
1892	<	progress = true;
1893	<	if (subtask.status < 0)
1894	<	break outer;               // stale
1895	<	if ((t = q.pollAt(b)) != null) {
1896	<	stealer.stealHint = joiner.poolIndex;
1897	<	joiner.runSubtask(t);
1883	>	if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1884	>	break restart;              // shutting down
1885	>	if ((v = ws[h = (j.stealHint | 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.stealHint = h;    // save hint
1894	>	break;
1895	>	}
1896	>	if (h == origin)
1897	>	break restart;      // cannot find stealer
1898		}
1899		}
1900	<	else { // empty - try to descend to find stealer's stealer
1901	<	ForkJoinTask<?> next = stealer.currentJoin;
1902	<	if (++depth == MAX_HELP_DEPTH || subtask.status < 0 ||
1903	<	next == null || next == subtask)
1904	<	break outer;  // max depth, stale, dead-end, cyclic
1905	<	subtask = next;
1906	<	j = stealer;
1907	<	break;
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		}
1935		}
1936		}
1937	<	return progress;
1937	>	return stat;
1938		}
1939
1940		/**
#	Line 1637 | Line 1943 | public class ForkJoinPool extends Abstra
1943		* @param joiner the joining worker
1944		* @param task the task
1945		*/
1946	<	final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1946	>	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1947		WorkQueue[] ws;
1948	<	int m = runState & SMASK;
1949	<	if ((ws = workQueues) != null && ws.length > m) {
1644	<	for (int j = 1; j <= m && task.status >= 0; j += 2) {
1948	>	if ((ws = workQueues) != null) {
1949	>	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1950		WorkQueue q = ws[j];
1951		if (q != null && q.pollFor(task)) {
1952		joiner.runSubtask(task);
#	Line 1652 | Line 1957 | public class ForkJoinPool extends Abstra
1957		}
1958
1959		/**
1960	<	* Returns a non-empty steal queue, if one is found during a random,
1961	<	* then cyclic scan, else null.  This method must be retried by
1962	<	* caller if, by the time it tries to use the queue, it is empty.
1960	>	* Tries to decrement active count (sometimes implicitly) and
1961	>	* possibly release or create a compensating worker in preparation
1962	>	* for blocking. Fails on contention or termination. Otherwise,
1963	>	* adds a new thread if no idle workers are available and either
1964	>	* pool would become completely starved or: (at least half
1965	>	* starved, and fewer than 50% spares exist, and there is at least
1966	>	* one task apparently available). Even though the availability
1967	>	* check requires a full scan, it is worthwhile in reducing false
1968	>	* alarms.
1969	>	*
1970	>	* @param task if non-null, a task being waited for
1971	>	* @param blocker if non-null, a blocker being waited for
1972	>	* @return true if the caller can block, else should recheck and retry
1973	>	*/
1974	>	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1975	>	int pc = parallelism, e;
1976	>	long c = ctl;
1977	>	WorkQueue[] ws = workQueues;
1978	>	if ((e = (int)c) >= 0 && ws != null) {
1979	>	int u, a, ac, hc;
1980	>	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1981	>	boolean replace = false;
1982	>	if ((a = u >> UAC_SHIFT) <= 0) {
1983	>	if ((ac = a + pc) <= 1)
1984	>	replace = true;
1985	>	else if ((e > 0 || (task != null &&
1986	>	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1987	>	WorkQueue w;
1988	>	for (int j = 0; j < ws.length; ++j) {
1989	>	if ((w = ws[j]) != null && !w.isEmpty()) {
1990	>	replace = true;
1991	>	break;   // in compensation range and tasks available
1992	>	}
1993	>	}
1994	>	}
1995	>	}
1996	>	if ((task == null || task.status >= 0) && // recheck need to block
1997	>	(blocker == null || !blocker.isReleasable()) && ctl == c) {
1998	>	if (!replace) {          // no compensation
1999	>	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
2000	>	if (U.compareAndSwapLong(this, CTL, c, nc))
2001	>	return true;
2002	>	}
2003	>	else if (e != 0) {       // release an idle worker
2004	>	WorkQueue w; Thread p; int i;
2005	>	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
2006	>	long nc = ((long)(w.nextWait & E_MASK) |
2007	>	(c & (AC_MASK|TC_MASK)));
2008	>	if (w.eventCount == (e | INT_SIGN) &&
2009	>	U.compareAndSwapLong(this, CTL, c, nc)) {
2010	>	w.eventCount = (e + E_SEQ) & E_MASK;
2011	>	if ((p = w.parker) != null)
2012	>	U.unpark(p);
2013	>	return true;
2014	>	}
2015	>	}
2016	>	}
2017	>	else if (tc < MAX_CAP) { // create replacement
2018	>	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
2019	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
2020	>	addWorker();
2021	>	return true;
2022	>	}
2023	>	}
2024	>	}
2025	>	}
2026	>	return false;
2027	>	}
2028	>
2029	>	/**
2030	>	* Helps and/or blocks until the given task is done.
2031	>	*
2032	>	* @param joiner the joining worker
2033	>	* @param task the task
2034	>	* @return task status on exit
2035	>	*/
2036	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
2037	>	int s;
2038	>	if ((s = task.status) >= 0) {
2039	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
2040	>	joiner.currentJoin = task;
2041	>	long startTime = 0L;
2042	>	for (int k = 0;;) {
2043	>	if ((s = (joiner.isEmpty() ?           // try to help
2044	>	tryHelpStealer(joiner, task) :
2045	>	joiner.tryRemoveAndExec(task))) == 0 &&
2046	>	(s = task.status) >= 0) {
2047	>	if (k == 0) {
2048	>	startTime = System.nanoTime();
2049	>	tryPollForAndExec(joiner, task); // check uncommon case
2050	>	}
2051	>	else if ((k & (MAX_HELP - 1)) == 0 &&
2052	>	System.nanoTime() - startTime >=
2053	>	COMPENSATION_DELAY &&
2054	>	tryCompensate(task, null)) {
2055	>	if (task.trySetSignal()) {
2056	>	synchronized (task) {
2057	>	if (task.status >= 0) {
2058	>	try {                // see ForkJoinTask
2059	>	task.wait();     //  for explanation
2060	>	} catch (InterruptedException ie) {
2061	>	}
2062	>	}
2063	>	else
2064	>	task.notifyAll();
2065	>	}
2066	>	}
2067	>	long c;                          // re-activate
2068	>	do {} while (!U.compareAndSwapLong
2069	>	(this, CTL, c = ctl, c + AC_UNIT));
2070	>	}
2071	>	}
2072	>	if (s < 0 || (s = task.status) < 0) {
2073	>	joiner.currentJoin = prevJoin;
2074	>	break;
2075	>	}
2076	>	else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
2077	>	Thread.yield();                     // for politeness
2078	>	}
2079	>	}
2080	>	return s;
2081	>	}
2082	>
2083	>	/**
2084	>	* Stripped-down variant of awaitJoin used by timed joins. Tries
2085	>	* to help join only while there is continuous progress. (Caller
2086	>	* will then enter a timed wait.)
2087	>	*
2088	>	* @param joiner the joining worker
2089	>	* @param task the task
2090	>	* @return task status on exit
2091	>	*/
2092	>	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2093	>	int s;
2094	>	while ((s = task.status) >= 0 &&
2095	>	(joiner.isEmpty() ?
2096	>	tryHelpStealer(joiner, task) :
2097	>	joiner.tryRemoveAndExec(task)) != 0)
2098	>	;
2099	>	return s;
2100	>	}
2101	>
2102	>	/**
2103	>	* Returns a (probably) non-empty steal queue, if one is found
2104	>	* during a random, then cyclic scan, else null.  This method must
2105	>	* be retried by caller if, by the time it tries to use the queue,
2106	>	* it is empty.
2107		*/
2108		private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
2109	<	int r = w.seed;    // Same idea as scan(), but ignoring submissions
2109	>	// Similar to loop in scan(), but ignoring submissions
2110	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
2111	>	int step = (r >>> 16) | 1;
2112		for (WorkQueue[] ws;;) {
2113	<	int m = runState & SMASK;
2114	<	if ((ws = workQueues) == null)
2113	>	int rs = runState, m;
2114	>	if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2115		return null;
2116	<	if (ws.length > m) {
2117	<	WorkQueue q;
2118	<	for (int n = m << 2, k = r, j = -n;;) {
2119	<	r ^= r << 13; r ^= r >>> 17; r ^= r << 5;
2120	<	if ((q = ws[(k | 1) & m]) != null && q.base - q.top < 0) {
2121	<	w.seed = r;
1671	<	return q;
1672	<	}
1673	<	else if (j > n)
2116	>	for (int j = (m + 1) << 2; ; r += step) {
2117	>	WorkQueue q = ws[((r << 1) | 1) & m];
2118	>	if (q != null && !q.isEmpty())
2119	>	return q;
2120	>	else if (--j < 0) {
2121	>	if (runState == rs)
2122		return null;
2123	<	else
1676	<	k = (j++ < 0) ? r : k + ((m >>> 1) | 1);
1677	<
2123	>	break;
2124		}
2125		}
2126		}
#	Line 1688 | Line 2134 | public class ForkJoinPool extends Abstra
2134		*/
2135		final void helpQuiescePool(WorkQueue w) {
2136		for (boolean active = true;;) {
2137	<	w.runLocalTasks();      // exhaust local queue
2137	>	ForkJoinTask<?> localTask; // exhaust local queue
2138	>	while ((localTask = w.nextLocalTask()) != null)
2139	>	localTask.doExec();
2140		WorkQueue q = findNonEmptyStealQueue(w);
2141		if (q != null) {
2142	<	ForkJoinTask<?> t;
2142	>	ForkJoinTask<?> t; int b;
2143		if (!active) {      // re-establish active count
2144		long c;
2145		active = true;
2146		do {} while (!U.compareAndSwapLong
2147		(this, CTL, c = ctl, c + AC_UNIT));
2148		}
2149	<	if ((t = q.poll()) != null)
2149	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2150		w.runSubtask(t);
2151		}
2152		else {
#	Line 1720 | Line 2168 | public class ForkJoinPool extends Abstra
2168		}
2169
2170		/**
2171	<	* Gets and removes a local or stolen task for the given worker
2171	>	* Restricted version of helpQuiescePool for non-FJ callers
2172	>	*/
2173	>	static void externalHelpQuiescePool() {
2174	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, sq;
2175	>	ForkJoinTask<?>[] a; int b;
2176	>	ForkJoinTask<?> t = null;
2177	>	int k = submitters.get().seed & SQMASK;
2178	>	if ((p = commonPool) != null &&
2179	>	(ws = p.workQueues) != null &&
2180	>	ws.length > (k &= p.submitMask) &&
2181	>	(q = ws[k]) != null) {
2182	>	while (q.top - q.base > 0) {
2183	>	if ((t = q.sharedPop()) != null)
2184	>	break;
2185	>	}
2186	>	if (t == null && (sq = p.findNonEmptyStealQueue(q)) != null &&
2187	>	(b = sq.base) - sq.top < 0)
2188	>	t = sq.pollAt(b);
2189	>	if (t != null)
2190	>	t.doExec();
2191	>	}
2192	>	}
2193	>
2194	>	/**
2195	>	* Gets and removes a local or stolen task for the given worker.
2196		*
2197		* @return a task, if available
2198		*/
2199		final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
2200		for (ForkJoinTask<?> t;;) {
2201	<	WorkQueue q;
2201	>	WorkQueue q; int b;
2202		if ((t = w.nextLocalTask()) != null)
2203		return t;
2204		if ((q = findNonEmptyStealQueue(w)) == null)
2205		return null;
2206	<	if ((t = q.poll()) != null)
2206	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2207		return t;
2208		}
2209		}
#	Line 1752 | Line 2224 | public class ForkJoinPool extends Abstra
2224		8);
2225		}
2226
1755	–	// Termination
1756	–
2227		/**
2228	<	* Sets SHUTDOWN bit of runState under lock
2228	>	* Returns approximate submission queue length for the given caller
2229		*/
2230	<	private void enableShutdown() {
2231	<	ReentrantLock lock = this.lock;
2232	<	if (runState >= 0) {
2233	<	lock.lock();                       // don't need try/finally
2234	<	runState |= SHUTDOWN;
2235	<	lock.unlock();
2236	<	}
2230	>	static int getEstimatedSubmitterQueueLength() {
2231	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue q;
2232	>	int k = submitters.get().seed & SQMASK;
2233	>	return ((p = commonPool) != null && (ws = p.workQueues) != null &&
2234	>	ws.length > (k &= p.submitMask) &&
2235	>	(q = ws[k]) != null) ?
2236	>	q.queueSize() : 0;
2237		}
2238
2239	+	//  Termination
2240	+
2241		/**
2242	<	* Possibly initiates and/or completes termination.  Upon
2243	<	* termination, cancels all queued tasks and then
2242	>	* Possibly initiates and/or completes termination.  The caller
2243	>	* triggering termination runs three passes through workQueues:
2244	>	* (0) Setting termination status, followed by wakeups of queued
2245	>	* workers; (1) cancelling all tasks; (2) interrupting lagging
2246	>	* threads (likely in external tasks, but possibly also blocked in
2247	>	* joins).  Each pass repeats previous steps because of potential
2248	>	* lagging thread creation.
2249		*
2250		* @param now if true, unconditionally terminate, else only
2251		* if no work and no active workers
2252	+	* @param enable if true, enable shutdown when next possible
2253		* @return true if now terminating or terminated
2254		*/
2255	<	private boolean tryTerminate(boolean now) {
2255	>	private boolean tryTerminate(boolean now, boolean enable) {
2256		for (long c;;) {
2257		if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2258		if ((short)(c >>> TC_SHIFT) == -parallelism) {
2259	<	ReentrantLock lock = this.lock; // signal when no workers
2260	<	lock.lock();                    // don't need try/finally
2261	<	termination.signalAll();        // signal when 0 workers
1784	<	lock.unlock();
2259	>	synchronized (this) {
2260	>	notifyAll();                // signal when 0 workers
2261	>	}
2262		}
2263		return true;
2264		}
2265	<	if (!now) {
2266	<	if ((int)(c >> AC_SHIFT) != -parallelism || runState >= 0 ||
2265	>	if (runState >= 0) {                    // not yet enabled
2266	>	if (!enable)
2267	>	return false;
2268	>	while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
2269	>	tryAwaitMainLock();
2270	>	try {
2271	>	runState |= SHUTDOWN;
2272	>	} finally {
2273	>	if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
2274	>	mainLock = 0;
2275	>	synchronized (this) { notifyAll(); };
2276	>	}
2277	>	}
2278	>	}
2279	>	if (!now) {                             // check if idle & no tasks
2280	>	if ((int)(c >> AC_SHIFT) != -parallelism ||
2281		hasQueuedSubmissions())
2282		return false;
2283		// Check for unqueued inactive workers. One pass suffices.
2284		WorkQueue[] ws = workQueues; WorkQueue w;
2285		if (ws != null) {
2286	<	int n = ws.length;
1796	<	for (int i = 1; i < n; i += 2) {
2286	>	for (int i = 1; i < ws.length; i += 2) {
2287		if ((w = ws[i]) != null && w.eventCount >= 0)
2288		return false;
2289		}
2290		}
2291		}
2292	<	if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT))
2293	<	startTerminating();
2294	<	}
2295	<	}
2296	<
2297	<	/**
2298	<	* Initiates termination: Runs three passes through workQueues:
2299	<	* (0) Setting termination status, followed by wakeups of queued
2300	<	* workers; (1) cancelling all tasks; (2) interrupting lagging
2301	<	* threads (likely in external tasks, but possibly also blocked in
2302	<	* joins).  Each pass repeats previous steps because of potential
2303	<	* lagging thread creation.
2304	<	*/
1815	<	private void startTerminating() {
1816	<	for (int pass = 0; pass < 3; ++pass) {
1817	<	WorkQueue[] ws = workQueues;
1818	<	if (ws != null) {
1819	<	WorkQueue w; Thread wt;
1820	<	int n = ws.length;
1821	<	for (int j = 0; j < n; ++j) {
1822	<	if ((w = ws[j]) != null) {
1823	<	w.runState = -1;
1824	<	if (pass > 0) {
1825	<	w.cancelAll();
1826	<	if (pass > 1 && (wt = w.owner) != null &&
1827	<	!wt.isInterrupted()) {
1828	<	try {
1829	<	wt.interrupt();
1830	<	} catch (SecurityException ignore) {
2292	>	if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2293	>	for (int pass = 0; pass < 3; ++pass) {
2294	>	WorkQueue[] ws = workQueues;
2295	>	if (ws != null) {
2296	>	WorkQueue w;
2297	>	int n = ws.length;
2298	>	for (int i = 0; i < n; ++i) {
2299	>	if ((w = ws[i]) != null) {
2300	>	w.runState = -1;
2301	>	if (pass > 0) {
2302	>	w.cancelAll();
2303	>	if (pass > 1)
2304	>	w.interruptOwner();
2305		}
2306		}
2307		}
2308	<	}
2309	<	}
2310	<	// Wake up workers parked on event queue
2311	<	int i, e; long c; Thread p;
2312	<	while ((i = ((~(e = (int)(c = ctl)) << 1) | 1) & SMASK) < n &&
2313	<	(w = ws[i]) != null &&
2314	<	w.eventCount == (e | INT_SIGN)) {
2315	<	long nc = ((long)(w.nextWait & E_MASK) |
2316	<	((c + AC_UNIT) & AC_MASK) |
2317	<	(c & (TC_MASK|STOP_BIT)));
2318	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
2319	<	w.eventCount = (e + E_SEQ) & E_MASK;
2320	<	if ((p = w.parker) != null)
2321	<	U.unpark(p);
2308	>	// Wake up workers parked on event queue
2309	>	int i, e; long cc; Thread p;
2310	>	while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2311	>	(i = e & SMASK) < n &&
2312	>	(w = ws[i]) != null) {
2313	>	long nc = ((long)(w.nextWait & E_MASK) |
2314	>	((cc + AC_UNIT) & AC_MASK) |
2315	>	(cc & (TC_MASK|STOP_BIT)));
2316	>	if (w.eventCount == (e | INT_SIGN) &&
2317	>	U.compareAndSwapLong(this, CTL, cc, nc)) {
2318	>	w.eventCount = (e + E_SEQ) & E_MASK;
2319	>	w.runState = -1;
2320	>	if ((p = w.parker) != null)
2321	>	U.unpark(p);
2322	>	}
2323	>	}
2324		}
2325		}
2326		}
#	Line 1920 | Line 2396 | public class ForkJoinPool extends Abstra
2396		checkPermission();
2397		if (factory == null)
2398		throw new NullPointerException();
2399	<	if (parallelism <= 0 || parallelism > MAX_ID)
2399	>	if (parallelism <= 0 || parallelism > MAX_CAP)
2400		throw new IllegalArgumentException();
2401		this.parallelism = parallelism;
2402		this.factory = factory;
2403		this.ueh = handler;
2404		this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
1929	–	this.nextPoolIndex = 1;
2405		long np = (long)(-parallelism); // offset ctl counts
2406		this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2407	<	// initialize workQueues array with room for 2*parallelism if possible
2408	<	int n = parallelism << 1;
2409	<	if (n >= MAX_ID)
2410	<	n = MAX_ID;
2411	<	else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
1937	<	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8;
1938	<	}
1939	<	this.workQueues = new WorkQueue[(n + 1) << 1];
1940	<	ReentrantLock lck = this.lock = new ReentrantLock();
1941	<	this.termination = lck.newCondition();
1942	<	this.stealCount = new AtomicLong();
1943	<	this.nextWorkerNumber = new AtomicInteger();
2407	>	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2408	>	int n = parallelism - 1;
2409	>	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
2410	>	this.submitMask = ((n + 1) << 1) - 1;
2411	>	int pn = poolNumberGenerator.incrementAndGet();
2412		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2413	<	sb.append(poolNumberGenerator.incrementAndGet());
2413	>	sb.append(Integer.toString(pn));
2414		sb.append("-worker-");
2415		this.workerNamePrefix = sb.toString();
2416	<	// Create initial submission queue
2417	<	WorkQueue sq = tryAddSharedQueue(0);
2418	<	if (sq != null)
2419	<	sq.growArray(false);
2416	>	this.runState = 1;              // set init flag
2417	>	}
2418	>
2419	>	/**
2420	>	* Constructor for common pool, suitable only for static initialization.
2421	>	* Basically the same as above, but uses smallest possible initial footprint.
2422	>	*/
2423	>	ForkJoinPool(int parallelism, int submitMask,
2424	>	ForkJoinWorkerThreadFactory factory,
2425	>	Thread.UncaughtExceptionHandler handler) {
2426	>	this.factory = factory;
2427	>	this.ueh = handler;
2428	>	this.submitMask = submitMask;
2429	>	this.parallelism = parallelism;
2430	>	long np = (long)(-parallelism);
2431	>	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2432	>	this.localMode = LIFO_QUEUE;
2433	>	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2434	>	this.runState = 1;
2435	>	}
2436	>
2437	>	/**
2438	>	* Returns the common pool instance.
2439	>	*
2440	>	* @return the common pool instance
2441	>	*/
2442	>	public static ForkJoinPool commonPool() {
2443	>	ForkJoinPool p;
2444	>	if ((p = commonPool) == null)
2445	>	throw new Error("Common Pool Unavailable");
2446	>	return p;
2447		}
2448
2449		// Execution methods
#	Line 1970 | Line 2465 | public class ForkJoinPool extends Abstra
2465		*         scheduled for execution
2466		*/
2467		public <T> T invoke(ForkJoinTask<T> task) {
2468	+	if (task == null)
2469	+	throw new NullPointerException();
2470		doSubmit(task);
2471		return task.join();
2472		}
#	Line 1983 | Line 2480 | public class ForkJoinPool extends Abstra
2480		*         scheduled for execution
2481		*/
2482		public void execute(ForkJoinTask<?> task) {
2483	+	if (task == null)
2484	+	throw new NullPointerException();
2485		doSubmit(task);
2486		}
2487
#	Line 2000 | Line 2499 | public class ForkJoinPool extends Abstra
2499		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2500		job = (ForkJoinTask<?>) task;
2501		else
2502	<	job = ForkJoinTask.adapt(task, null);
2502	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2503		doSubmit(job);
2504		}
2505
#	Line 2014 | Line 2513 | public class ForkJoinPool extends Abstra
2513		*         scheduled for execution
2514		*/
2515		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2516	+	if (task == null)
2517	+	throw new NullPointerException();
2518		doSubmit(task);
2519		return task;
2520		}
#	Line 2024 | Line 2525 | public class ForkJoinPool extends Abstra
2525		*         scheduled for execution
2526		*/
2527		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2528	<	if (task == null)
2028	<	throw new NullPointerException();
2029	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task);
2528	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2529		doSubmit(job);
2530		return job;
2531		}
#	Line 2037 | Line 2536 | public class ForkJoinPool extends Abstra
2536		*         scheduled for execution
2537		*/
2538		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2539	<	if (task == null)
2041	<	throw new NullPointerException();
2042	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
2539	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2540		doSubmit(job);
2541		return job;
2542		}
#	Line 2056 | Line 2553 | public class ForkJoinPool extends Abstra
2553		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2554		job = (ForkJoinTask<?>) task;
2555		else
2556	<	job = ForkJoinTask.adapt(task, null);
2556	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2557		doSubmit(job);
2558		return job;
2559		}
#	Line 2066 | Line 2563 | public class ForkJoinPool extends Abstra
2563		* @throws RejectedExecutionException {@inheritDoc}
2564		*/
2565		public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2566	<	ArrayList<ForkJoinTask<T>> forkJoinTasks =
2567	<	new ArrayList<ForkJoinTask<T>>(tasks.size());
2568	<	for (Callable<T> task : tasks)
2569	<	forkJoinTasks.add(ForkJoinTask.adapt(task));
2570	<	invoke(new InvokeAll<T>(forkJoinTasks));
2571	<
2566	>	// In previous versions of this class, this method constructed
2567	>	// a task to run ForkJoinTask.invokeAll, but now external
2568	>	// invocation of multiple tasks is at least as efficient.
2569	>	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2570	>	// Workaround needed because method wasn't declared with
2571	>	// wildcards in return type but should have been.
2572		@SuppressWarnings({"unchecked", "rawtypes"})
2573	<	List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
2077	<	return futures;
2078	<	}
2573	>	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2574
2575	<	static final class InvokeAll<T> extends RecursiveAction {
2576	<	final ArrayList<ForkJoinTask<T>> tasks;
2577	<	InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
2578	<	public void compute() {
2579	<	try { invokeAll(tasks); }
2580	<	catch (Exception ignore) {}
2575	>	boolean done = false;
2576	>	try {
2577	>	for (Callable<T> t : tasks) {
2578	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2579	>	doSubmit(f);
2580	>	fs.add(f);
2581	>	}
2582	>	for (ForkJoinTask<T> f : fs)
2583	>	f.quietlyJoin();
2584	>	done = true;
2585	>	return futures;
2586	>	} finally {
2587	>	if (!done)
2588	>	for (ForkJoinTask<T> f : fs)
2589	>	f.cancel(false);
2590		}
2087	–	private static final long serialVersionUID = -7914297376763021607L;
2591		}
2592
2593		/**
#	Line 2116 | Line 2619 | public class ForkJoinPool extends Abstra
2619		}
2620
2621		/**
2622	+	* Returns the targeted parallelism level of the common pool.
2623	+	*
2624	+	* @return the targeted parallelism level of the common pool
2625	+	*/
2626	+	public static int getCommonPoolParallelism() {
2627	+	return commonPoolParallelism;
2628	+	}
2629	+
2630	+	/**
2631		* Returns the number of worker threads that have started but not
2632		* yet terminated.  The result returned by this method may differ
2633		* from {@link #getParallelism} when threads are created to
#	Line 2149 | Line 2661 | public class ForkJoinPool extends Abstra
2661		int rc = 0;
2662		WorkQueue[] ws; WorkQueue w;
2663		if ((ws = workQueues) != null) {
2664	<	int n = ws.length;
2665	<	for (int i = 1; i < n; i += 2) {
2154	<	Thread.State s; ForkJoinWorkerThread wt;
2155	<	if ((w = ws[i]) != null && (wt = w.owner) != null &&
2156	<	w.eventCount >= 0 &&
2157	<	(s = wt.getState()) != Thread.State.BLOCKED &&
2158	<	s != Thread.State.WAITING &&
2159	<	s != Thread.State.TIMED_WAITING)
2664	>	for (int i = 1; i < ws.length; i += 2) {
2665	>	if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2666		++rc;
2667		}
2668		}
#	Line 2202 | Line 2708 | public class ForkJoinPool extends Abstra
2708		* @return the number of steals
2709		*/
2710		public long getStealCount() {
2711	<	long count = stealCount.get();
2711	>	long count = stealCount;
2712		WorkQueue[] ws; WorkQueue w;
2713		if ((ws = workQueues) != null) {
2714	<	int n = ws.length;
2209	<	for (int i = 1; i < n; i += 2) {
2714	>	for (int i = 1; i < ws.length; i += 2) {
2715		if ((w = ws[i]) != null)
2716		count += w.totalSteals;
2717		}
#	Line 2228 | Line 2733 | public class ForkJoinPool extends Abstra
2733		long count = 0;
2734		WorkQueue[] ws; WorkQueue w;
2735		if ((ws = workQueues) != null) {
2736	<	int n = ws.length;
2232	<	for (int i = 1; i < n; i += 2) {
2736	>	for (int i = 1; i < ws.length; i += 2) {
2737		if ((w = ws[i]) != null)
2738		count += w.queueSize();
2739		}
#	Line 2248 | Line 2752 | public class ForkJoinPool extends Abstra
2752		int count = 0;
2753		WorkQueue[] ws; WorkQueue w;
2754		if ((ws = workQueues) != null) {
2755	<	int n = ws.length;
2252	<	for (int i = 0; i < n; i += 2) {
2755	>	for (int i = 0; i < ws.length; i += 2) {
2756		if ((w = ws[i]) != null)
2757		count += w.queueSize();
2758		}
#	Line 2266 | Line 2769 | public class ForkJoinPool extends Abstra
2769		public boolean hasQueuedSubmissions() {
2770		WorkQueue[] ws; WorkQueue w;
2771		if ((ws = workQueues) != null) {
2772	<	int n = ws.length;
2773	<	for (int i = 0; i < n; i += 2) {
2271	<	if ((w = ws[i]) != null && w.queueSize() != 0)
2772	>	for (int i = 0; i < ws.length; i += 2) {
2773	>	if ((w = ws[i]) != null && !w.isEmpty())
2774		return true;
2775		}
2776		}
#	Line 2285 | Line 2787 | public class ForkJoinPool extends Abstra
2787		protected ForkJoinTask<?> pollSubmission() {
2788		WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2789		if ((ws = workQueues) != null) {
2790	<	int n = ws.length;
2289	<	for (int i = 0; i < n; i += 2) {
2790	>	for (int i = 0; i < ws.length; i += 2) {
2791		if ((w = ws[i]) != null && (t = w.poll()) != null)
2792		return t;
2793		}
#	Line 2315 | Line 2816 | public class ForkJoinPool extends Abstra
2816		int count = 0;
2817		WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2818		if ((ws = workQueues) != null) {
2819	<	int n = ws.length;
2319	<	for (int i = 0; i < n; ++i) {
2819	>	for (int i = 0; i < ws.length; ++i) {
2820		if ((w = ws[i]) != null) {
2821		while ((t = w.poll()) != null) {
2822		c.add(t);
#	Line 2336 | Line 2836 | public class ForkJoinPool extends Abstra
2836		* @return a string identifying this pool, as well as its state
2837		*/
2838		public String toString() {
2839	<	long st = getStealCount();
2840	<	long qt = getQueuedTaskCount();
2841	<	long qs = getQueuedSubmissionCount();
2342	<	int rc = getRunningThreadCount();
2343	<	int pc = parallelism;
2839	>	// Use a single pass through workQueues to collect counts
2840	>	long qt = 0L, qs = 0L; int rc = 0;
2841	>	long st = stealCount;
2842		long c = ctl;
2843	+	WorkQueue[] ws; WorkQueue w;
2844	+	if ((ws = workQueues) != null) {
2845	+	for (int i = 0; i < ws.length; ++i) {
2846	+	if ((w = ws[i]) != null) {
2847	+	int size = w.queueSize();
2848	+	if ((i & 1) == 0)
2849	+	qs += size;
2850	+	else {
2851	+	qt += size;
2852	+	st += w.totalSteals;
2853	+	if (w.isApparentlyUnblocked())
2854	+	++rc;
2855	+	}
2856	+	}
2857	+	}
2858	+	}
2859	+	int pc = parallelism;
2860		int tc = pc + (short)(c >>> TC_SHIFT);
2861		int ac = pc + (int)(c >> AC_SHIFT);
2862		if (ac < 0) // ignore transient negative
#	Line 2364 | Line 2879 | public class ForkJoinPool extends Abstra
2879		}
2880
2881		/**
2882	<	* Initiates an orderly shutdown in which previously submitted
2883	<	* tasks are executed, but no new tasks will be accepted.
2884	<	* Invocation has no additional effect if already shut down.
2885	<	* Tasks that are in the process of being submitted concurrently
2886	<	* during the course of this method may or may not be rejected.
2882	>	* Possibly initiates an orderly shutdown in which previously
2883	>	* submitted tasks are executed, but no new tasks will be
2884	>	* accepted. Invocation has no effect on execution state if this
2885	>	* is the {@link #commonPool}, and no additional effect if
2886	>	* already shut down.  Tasks that are in the process of being
2887	>	* submitted concurrently during the course of this method may or
2888	>	* may not be rejected.
2889		*
2890		* @throws SecurityException if a security manager exists and
2891		*         the caller is not permitted to modify threads
#	Line 2377 | Line 2894 | public class ForkJoinPool extends Abstra
2894		*/
2895		public void shutdown() {
2896		checkPermission();
2897	<	enableShutdown();
2898	<	tryTerminate(false);
2897	>	if (this != commonPool)
2898	>	tryTerminate(false, true);
2899		}
2900
2901		/**
2902	<	* Attempts to cancel and/or stop all tasks, and reject all
2903	<	* subsequently submitted tasks.  Tasks that are in the process of
2904	<	* being submitted or executed concurrently during the course of
2905	<	* this method may or may not be rejected. This method cancels
2906	<	* both existing and unexecuted tasks, in order to permit
2907	<	* termination in the presence of task dependencies. So the method
2908	<	* always returns an empty list (unlike the case for some other
2909	<	* Executors).
2902	>	* Possibly attempts to cancel and/or stop all tasks, and reject
2903	>	* all subsequently submitted tasks.  Invocation has no effect on
2904	>	* execution state if this is the {@link #commonPool}, and no
2905	>	* additional effect if already shut down. Otherwise, tasks that
2906	>	* are in the process of being submitted or executed concurrently
2907	>	* during the course of this method may or may not be
2908	>	* rejected. This method cancels both existing and unexecuted
2909	>	* tasks, in order to permit termination in the presence of task
2910	>	* dependencies. So the method always returns an empty list
2911	>	* (unlike the case for some other Executors).
2912		*
2913		* @return an empty list
2914		* @throws SecurityException if a security manager exists and
#	Line 2399 | Line 2918 | public class ForkJoinPool extends Abstra
2918		*/
2919		public List<Runnable> shutdownNow() {
2920		checkPermission();
2921	<	enableShutdown();
2922	<	tryTerminate(true);
2921	>	if (this != commonPool)
2922	>	tryTerminate(true, true);
2923		return Collections.emptyList();
2924		}
2925
#	Line 2457 | Line 2976 | public class ForkJoinPool extends Abstra
2976		public boolean awaitTermination(long timeout, TimeUnit unit)
2977		throws InterruptedException {
2978		long nanos = unit.toNanos(timeout);
2979	<	final ReentrantLock lock = this.lock;
2980	<	lock.lock();
2981	<	try {
2982	<	for (;;) {
2983	<	if (isTerminated())
2984	<	return true;
2985	<	if (nanos <= 0)
2986	<	return false;
2987	<	nanos = termination.awaitNanos(nanos);
2979	>	if (isTerminated())
2980	>	return true;
2981	>	long startTime = System.nanoTime();
2982	>	boolean terminated = false;
2983	>	synchronized (this) {
2984	>	for (long waitTime = nanos, millis = 0L;;) {
2985	>	if (terminated = isTerminated() ||
2986	>	waitTime <= 0L ||
2987	>	(millis = unit.toMillis(waitTime)) <= 0L)
2988	>	break;
2989	>	wait(millis);
2990	>	waitTime = nanos - (System.nanoTime() - startTime);
2991		}
2470	–	} finally {
2471	–	lock.unlock();
2992		}
2993	+	return terminated;
2994		}
2995
2996		/**
#	Line 2553 | Line 3074 | public class ForkJoinPool extends Abstra
3074		*
3075		* <p>If the caller is not a {@link ForkJoinTask}, this method is
3076		* behaviorally equivalent to
3077	<	a     *  <pre> {@code
3077	>	*  <pre> {@code
3078		* while (!blocker.isReleasable())
3079		*   if (blocker.block())
3080		*     return;
#	Line 2571 | Line 3092 | a     *  <pre> {@code
3092		ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3093		((ForkJoinWorkerThread)t).pool : null);
3094		while (!blocker.isReleasable()) {
3095	<	if (p == null || p.tryCompensate()) {
3095	>	if (p == null || p.tryCompensate(null, blocker)) {
3096		try {
3097		do {} while (!blocker.isReleasable() && !blocker.block());
3098		} finally {
#	Line 2588 | Line 3109 | a     *  <pre> {@code
3109		// implement RunnableFuture.
3110
3111		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3112	<	return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
3112	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3113		}
3114
3115		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3116	<	return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
3116	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
3117		}
3118
3119		// Unsafe mechanics
3120		private static final sun.misc.Unsafe U;
3121		private static final long CTL;
2601	–	private static final long RUNSTATE;
3122		private static final long PARKBLOCKER;
3123	+	private static final int ABASE;
3124	+	private static final int ASHIFT;
3125	+	private static final long NEXTWORKERNUMBER;
3126	+	private static final long STEALCOUNT;
3127	+	private static final long MAINLOCK;
3128
3129		static {
3130		poolNumberGenerator = new AtomicInteger();
3131	+	nextSubmitterSeed = new AtomicInteger(0x55555555);
3132		modifyThreadPermission = new RuntimePermission("modifyThread");
3133		defaultForkJoinWorkerThreadFactory =
3134		new DefaultForkJoinWorkerThreadFactory();
3135	+	submitters = new ThreadSubmitter();
3136		int s;
3137		try {
3138		U = getUnsafe();
3139		Class<?> k = ForkJoinPool.class;
3140	<	Class<?> tk = Thread.class;
3140	>	Class<?> ak = ForkJoinTask[].class;
3141		CTL = U.objectFieldOffset
3142		(k.getDeclaredField("ctl"));
3143	<	RUNSTATE = U.objectFieldOffset
3144	<	(k.getDeclaredField("runState"));
3143	>	NEXTWORKERNUMBER = U.objectFieldOffset
3144	>	(k.getDeclaredField("nextWorkerNumber"));
3145	>	STEALCOUNT = U.objectFieldOffset
3146	>	(k.getDeclaredField("stealCount"));
3147	>	MAINLOCK = U.objectFieldOffset
3148	>	(k.getDeclaredField("mainLock"));
3149	>	Class<?> tk = Thread.class;
3150		PARKBLOCKER = U.objectFieldOffset
3151		(tk.getDeclaredField("parkBlocker"));
3152	+	ABASE = U.arrayBaseOffset(ak);
3153	+	s = U.arrayIndexScale(ak);
3154	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3155	+	} catch (Exception e) {
3156	+	throw new Error(e);
3157	+	}
3158	+	if ((s & (s-1)) != 0)
3159	+	throw new Error("data type scale not a power of two");
3160	+	try { // Establish common pool
3161	+	String pp = System.getProperty(propPrefix + "parallelism");
3162	+	String fp = System.getProperty(propPrefix + "threadFactory");
3163	+	String up = System.getProperty(propPrefix + "exceptionHandler");
3164	+	ForkJoinWorkerThreadFactory fac = (fp == null) ?
3165	+	defaultForkJoinWorkerThreadFactory :
3166	+	((ForkJoinWorkerThreadFactory)ClassLoader.
3167	+	getSystemClassLoader().loadClass(fp).newInstance());
3168	+	Thread.UncaughtExceptionHandler ueh = (up == null) ? null :
3169	+	((Thread.UncaughtExceptionHandler)ClassLoader.
3170	+	getSystemClassLoader().loadClass(up).newInstance());
3171	+	int par;
3172	+	if ((pp == null || (par = Integer.parseInt(pp)) <= 0))
3173	+	par = Runtime.getRuntime().availableProcessors();
3174	+	if (par > MAX_CAP)
3175	+	par = MAX_CAP;
3176	+	commonPoolParallelism = par;
3177	+	int n = par - 1; // precompute submit mask
3178	+	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4;
3179	+	n |= n >>> 8; n |= n >>> 16;
3180	+	int mask = ((n + 1) << 1) - 1;
3181	+	commonPool = new ForkJoinPool(par, mask, fac, ueh);
3182		} catch (Exception e) {
3183		throw new Error(e);
3184		}

Diff Legend

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