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 1.19 by jsr166, Fri Jul 24 18:57:56 2009 UTC vs.
Revision 1.141 by jsr166, Wed Nov 14 18:45:53 2012 UTC

#	Line 1 | Line 1
1		/*
2		* Written by Doug Lea with assistance from members of JCP JSR-166
3		* Expert Group and released to the public domain, as explained at
4	<	* http://creativecommons.org/licenses/publicdomain
4	>	* http://creativecommons.org/publicdomain/zero/1.0/
5		*/
6
7		package jsr166y;
8	<	import java.util.*;
9	<	import java.util.concurrent.*;
10	<	import java.util.concurrent.locks.*;
11	<	import java.util.concurrent.atomic.*;
12	<	import sun.misc.Unsafe;
13	<	import java.lang.reflect.*;
8	>
9	>	import java.util.ArrayList;
10	>	import java.util.Arrays;
11	>	import java.util.Collection;
12	>	import java.util.Collections;
13	>	import java.util.List;
14	>	import java.util.concurrent.AbstractExecutorService;
15	>	import java.util.concurrent.Callable;
16	>	import java.util.concurrent.ExecutorService;
17	>	import java.util.concurrent.Future;
18	>	import java.util.concurrent.RejectedExecutionException;
19	>	import java.util.concurrent.RunnableFuture;
20	>	import java.util.concurrent.TimeUnit;
21
22		/**
23	<	* An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
24	<	* ForkJoinPool provides the entry point for submissions from
25	<	* non-ForkJoinTasks, as well as management and monitoring operations.
26	<	* Normally a single ForkJoinPool is used for a large number of
27	<	* submitted tasks. Otherwise, use would not usually outweigh the
28	<	* construction and bookkeeping overhead of creating a large set of
29	<	* threads.
23	>	* An {@link ExecutorService} for running {@link ForkJoinTask}s.
24	>	* A {@code ForkJoinPool} provides the entry point for submissions
25	>	* from non-{@code ForkJoinTask} clients, as well as management and
26	>	* monitoring operations.
27	>	*
28	>	* <p>A {@code ForkJoinPool} differs from other kinds of {@link
29	>	* ExecutorService} mainly by virtue of employing
30	>	* <em>work-stealing</em>: all threads in the pool attempt to find and
31	>	* execute tasks submitted to the pool and/or created by other active
32	>	* tasks (eventually blocking waiting for work if none exist). This
33	>	* enables efficient processing when most tasks spawn other subtasks
34	>	* (as do most {@code ForkJoinTask}s), as well as when many small
35	>	* tasks are submitted to the pool from external clients.  Especially
36	>	* when setting <em>asyncMode</em> to true in constructors, {@code
37	>	* ForkJoinPool}s may also be appropriate for use with event-style
38	>	* tasks that are never joined.
39		*
40	<	* <p>ForkJoinPools differ from other kinds of Executors mainly in
41	<	* that they provide <em>work-stealing</em>: all threads in the pool
42	<	* attempt to find and execute subtasks created by other active tasks
43	<	* (eventually blocking if none exist). This makes them efficient when
44	<	* most tasks spawn other subtasks (as do most ForkJoinTasks), as well
45	<	* as the mixed execution of some plain Runnable- or Callable- based
30	<	* activities along with ForkJoinTasks. When setting
31	<	* {@code setAsyncMode}, a ForkJoinPools may also be appropriate for
32	<	* use with fine-grained tasks that are never joined. Otherwise, other
33	<	* ExecutorService implementations are typically more appropriate
34	<	* choices.
40	>	* <p>A static {@link #commonPool} is available and appropriate for
41	>	* most applications. The common pool is used by any ForkJoinTask that
42	>	* is not explicitly submitted to a specified pool. Using the common
43	>	* pool normally reduces resource usage (its threads are slowly
44	>	* reclaimed during periods of non-use, and reinstated upon subsequent
45	>	* use).
46		*
47	<	* <p>A ForkJoinPool may be constructed with a given parallelism level
48	<	* (target pool size), which it attempts to maintain by dynamically
49	<	* adding, suspending, or resuming threads, even if some tasks are
50	<	* waiting to join others. However, no such adjustments are performed
51	<	* in the face of blocked IO or other unmanaged synchronization. The
52	<	* nested {@code ManagedBlocker} interface enables extension of
53	<	* the kinds of synchronization accommodated.  The target parallelism
54	<	* level may also be changed dynamically ({@code setParallelism})
55	<	* and thread construction can be limited using methods
56	<	* {@code setMaximumPoolSize} and/or
46	<	* {@code setMaintainsParallelism}.
47	>	* <p>For applications that require separate or custom pools, a {@code
48	>	* ForkJoinPool} may be constructed with a given target parallelism
49	>	* level; by default, equal to the number of available processors. The
50	>	* pool attempts to maintain enough active (or available) threads by
51	>	* dynamically adding, suspending, or resuming internal worker
52	>	* threads, even if some tasks are stalled waiting to join
53	>	* others. However, no such adjustments are guaranteed in the face of
54	>	* blocked IO or other unmanaged synchronization. The nested {@link
55	>	* ManagedBlocker} interface enables extension of the kinds of
56	>	* synchronization accommodated.
57		*
58		* <p>In addition to execution and lifecycle control methods, this
59		* class provides status check methods (for example
60	<	* {@code getStealCount}) that are intended to aid in developing,
60	>	* {@link #getStealCount}) that are intended to aid in developing,
61		* tuning, and monitoring fork/join applications. Also, method
62	<	* {@code toString} returns indications of pool state in a
62	>	* {@link #toString} returns indications of pool state in a
63		* convenient form for informal monitoring.
64		*
65	+	* <p> As is the case with other ExecutorServices, there are three
66	+	* main task execution methods summarized in the following table.
67	+	* These are designed to be used primarily by clients not already
68	+	* engaged in fork/join computations in the current pool.  The main
69	+	* forms of these methods accept instances of {@code ForkJoinTask},
70	+	* but overloaded forms also allow mixed execution of plain {@code
71	+	* Runnable}- or {@code Callable}- based activities as well.  However,
72	+	* tasks that are already executing in a pool should normally instead
73	+	* use the within-computation forms listed in the table unless using
74	+	* async event-style tasks that are not usually joined, in which case
75	+	* there is little difference among choice of methods.
76	+	*
77	+	* <table BORDER CELLPADDING=3 CELLSPACING=1>
78	+	*  <tr>
79	+	*    <td></td>
80	+	*    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
81	+	*    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
82	+	*  </tr>
83	+	*  <tr>
84	+	*    <td> <b>Arrange async execution</td>
85	+	*    <td> {@link #execute(ForkJoinTask)}</td>
86	+	*    <td> {@link ForkJoinTask#fork}</td>
87	+	*  </tr>
88	+	*  <tr>
89	+	*    <td> <b>Await and obtain result</td>
90	+	*    <td> {@link #invoke(ForkJoinTask)}</td>
91	+	*    <td> {@link ForkJoinTask#invoke}</td>
92	+	*  </tr>
93	+	*  <tr>
94	+	*    <td> <b>Arrange exec and obtain Future</td>
95	+	*    <td> {@link #submit(ForkJoinTask)}</td>
96	+	*    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
97	+	*  </tr>
98	+	* </table>
99	+	*
100	+	* <p>The common pool is by default constructed with default
101	+	* parameters, but these may be controlled by setting three {@link
102	+	* System#getProperty properties} with prefix {@code
103	+	* java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
104	+	* an integer greater than zero, {@code threadFactory} -- the class
105	+	* name of a {@link ForkJoinWorkerThreadFactory}, and {@code
106	+	* exceptionHandler} -- the class name of a {@link
107	+	* Thread.UncaughtExceptionHandler}. Upon any error in establishing
108	+	* these settings, default parameters are used.
109	+	*
110		* <p><b>Implementation notes</b>: This implementation restricts the
111		* maximum number of running threads to 32767. Attempts to create
112	<	* pools with greater than the maximum result in
113	<	* IllegalArgumentExceptions.
112	>	* pools with greater than the maximum number result in
113	>	* {@code IllegalArgumentException}.
114	>	*
115	>	* <p>This implementation rejects submitted tasks (that is, by throwing
116	>	* {@link RejectedExecutionException}) only when the pool is shut down
117	>	* or internal resources have been exhausted.
118		*
119		* @since 1.7
120		* @author Doug Lea
#	Line 63 | Line 122 | import java.lang.reflect.*;
122		public class ForkJoinPool extends AbstractExecutorService {
123
124		/*
125	<	* See the extended comments interspersed below for design,
126	<	* rationale, and walkthroughs.
125	>	* Implementation Overview
126	>	*
127	>	* This class and its nested classes provide the main
128	>	* functionality and control for a set of worker threads:
129	>	* Submissions from non-FJ threads enter into submission queues.
130	>	* Workers take these tasks and typically split them into subtasks
131	>	* that may be stolen by other workers.  Preference rules give
132	>	* first priority to processing tasks from their own queues (LIFO
133	>	* or FIFO, depending on mode), then to randomized FIFO steals of
134	>	* tasks in other queues.
135	>	*
136	>	* WorkQueues
137	>	* ==========
138	>	*
139	>	* Most operations occur within work-stealing queues (in nested
140	>	* class WorkQueue).  These are special forms of Deques that
141	>	* support only three of the four possible end-operations -- push,
142	>	* pop, and poll (aka steal), under the further constraints that
143	>	* push and pop are called only from the owning thread (or, as
144	>	* extended here, under a lock), while poll may be called from
145	>	* other threads.  (If you are unfamiliar with them, you probably
146	>	* want to read Herlihy and Shavit's book "The Art of
147	>	* Multiprocessor programming", chapter 16 describing these in
148	>	* more detail before proceeding.)  The main work-stealing queue
149	>	* design is roughly similar to those in the papers "Dynamic
150	>	* Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
151	>	* (http://research.sun.com/scalable/pubs/index.html) and
152	>	* "Idempotent work stealing" by Michael, Saraswat, and Vechev,
153	>	* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
154	>	* The main differences ultimately stem from GC requirements that
155	>	* we null out taken slots as soon as we can, to maintain as small
156	>	* a footprint as possible even in programs generating huge
157	>	* numbers of tasks. To accomplish this, we shift the CAS
158	>	* arbitrating pop vs poll (steal) from being on the indices
159	>	* ("base" and "top") to the slots themselves.  So, both a
160	>	* successful pop and poll mainly entail a CAS of a slot from
161	>	* non-null to null.  Because we rely on CASes of references, we
162	>	* do not need tag bits on base or top.  They are simple ints as
163	>	* used in any circular array-based queue (see for example
164	>	* ArrayDeque).  Updates to the indices must still be ordered in a
165	>	* way that guarantees that top == base means the queue is empty,
166	>	* but otherwise may err on the side of possibly making the queue
167	>	* appear nonempty when a push, pop, or poll have not fully
168	>	* committed. Note that this means that the poll operation,
169	>	* considered individually, is not wait-free. One thief cannot
170	>	* successfully continue until another in-progress one (or, if
171	>	* previously empty, a push) completes.  However, in the
172	>	* aggregate, we ensure at least probabilistic non-blockingness.
173	>	* If an attempted steal fails, a thief always chooses a different
174	>	* random victim target to try next. So, in order for one thief to
175	>	* progress, it suffices for any in-progress poll or new push on
176	>	* any empty queue to complete. (This is why we normally use
177	>	* method pollAt and its variants that try once at the apparent
178	>	* base index, else consider alternative actions, rather than
179	>	* method poll.)
180	>	*
181	>	* This approach also enables support of a user mode in which local
182	>	* task processing is in FIFO, not LIFO order, simply by using
183	>	* poll rather than pop.  This can be useful in message-passing
184	>	* frameworks in which tasks are never joined.  However neither
185	>	* mode considers affinities, loads, cache localities, etc, so
186	>	* rarely provide the best possible performance on a given
187	>	* machine, but portably provide good throughput by averaging over
188	>	* these factors.  (Further, even if we did try to use such
189	>	* information, we do not usually have a basis for exploiting it.
190	>	* For example, some sets of tasks profit from cache affinities,
191	>	* but others are harmed by cache pollution effects.)
192	>	*
193	>	* WorkQueues are also used in a similar way for tasks submitted
194	>	* to the pool. We cannot mix these tasks in the same queues used
195	>	* for work-stealing (this would contaminate lifo/fifo
196	>	* processing). Instead, we randomly associate submission queues
197	>	* with submitting threads, using a form of hashing.  The
198	>	* ThreadLocal Submitter class contains a value initially used as
199	>	* a hash code for choosing existing queues, but may be randomly
200	>	* repositioned upon contention with other submitters.  In
201	>	* essence, submitters act like workers except that they are
202	>	* restricted to executing local tasks that they submitted (or in
203	>	* the case of CountedCompleters, others with the same root task).
204	>	* However, because most shared/external queue operations are more
205	>	* expensive than internal, and because, at steady state, external
206	>	* submitters will compete for CPU with workers, ForkJoinTask.join
207	>	* and related methods disable them from repeatedly helping to
208	>	* process tasks if all workers are active.  Insertion of tasks in
209	>	* shared mode requires a lock (mainly to protect in the case of
210	>	* resizing) but we use only a simple spinlock (using bits in
211	>	* field qlock), because submitters encountering a busy queue move
212	>	* on to try or create other queues -- they block only when
213	>	* creating and registering new queues.
214	>	*
215	>	* Management
216	>	* ==========
217	>	*
218	>	* The main throughput advantages of work-stealing stem from
219	>	* decentralized control -- workers mostly take tasks from
220	>	* themselves or each other. We cannot negate this in the
221	>	* implementation of other management responsibilities. The main
222	>	* tactic for avoiding bottlenecks is packing nearly all
223	>	* essentially atomic control state into two volatile variables
224	>	* that are by far most often read (not written) as status and
225	>	* consistency checks.
226	>	*
227	>	* Field "ctl" contains 64 bits holding all the information needed
228	>	* to atomically decide to add, inactivate, enqueue (on an event
229	>	* queue), dequeue, and/or re-activate workers.  To enable this
230	>	* packing, we restrict maximum parallelism to (1<<15)-1 (which is
231	>	* far in excess of normal operating range) to allow ids, counts,
232	>	* and their negations (used for thresholding) to fit into 16bit
233	>	* fields.
234	>	*
235	>	* Field "plock" is a form of sequence lock with a saturating
236	>	* shutdown bit (similarly for per-queue "qlocks"), mainly
237	>	* protecting updates to the workQueues array, as well as to
238	>	* enable shutdown.  When used as a lock, it is normally only very
239	>	* briefly held, so is nearly always available after at most a
240	>	* brief spin, but we use a monitor-based backup strategy to
241	>	* blocking when needed.
242	>	*
243	>	* Recording WorkQueues.  WorkQueues are recorded in the
244	>	* "workQueues" array that is created upon first use and expanded
245	>	* if necessary.  Updates to the array while recording new workers
246	>	* and unrecording terminated ones are protected from each other
247	>	* by a lock but the array is otherwise concurrently readable, and
248	>	* accessed directly.  To simplify index-based operations, the
249	>	* array size is always a power of two, and all readers must
250	>	* tolerate null slots. Worker queues are at odd indices Shared
251	>	* (submission) queues are at even indices, up to a maximum of 64
252	>	* slots, to limit growth even if array needs to expand to add
253	>	* more workers. Grouping them together in this way simplifies and
254	>	* speeds up task scanning.
255	>	*
256	>	* All worker thread creation is on-demand, triggered by task
257	>	* submissions, replacement of terminated workers, and/or
258	>	* compensation for blocked workers. However, all other support
259	>	* code is set up to work with other policies.  To ensure that we
260	>	* do not hold on to worker references that would prevent GC, ALL
261	>	* accesses to workQueues are via indices into the workQueues
262	>	* array (which is one source of some of the messy code
263	>	* constructions here). In essence, the workQueues array serves as
264	>	* a weak reference mechanism. Thus for example the wait queue
265	>	* field of ctl stores indices, not references.  Access to the
266	>	* workQueues in associated methods (for example signalWork) must
267	>	* both index-check and null-check the IDs. All such accesses
268	>	* ignore bad IDs by returning out early from what they are doing,
269	>	* since this can only be associated with termination, in which
270	>	* case it is OK to give up.  All uses of the workQueues array
271	>	* also check that it is non-null (even if previously
272	>	* non-null). This allows nulling during termination, which is
273	>	* currently not necessary, but remains an option for
274	>	* resource-revocation-based shutdown schemes. It also helps
275	>	* reduce JIT issuance of uncommon-trap code, which tends to
276	>	* unnecessarily complicate control flow in some methods.
277	>	*
278	>	* Event Queuing. Unlike HPC work-stealing frameworks, we cannot
279	>	* let workers spin indefinitely scanning for tasks when none can
280	>	* be found immediately, and we cannot start/resume workers unless
281	>	* there appear to be tasks available.  On the other hand, we must
282	>	* quickly prod them into action when new tasks are submitted or
283	>	* generated. In many usages, ramp-up time to activate workers is
284	>	* the main limiting factor in overall performance (this is
285	>	* compounded at program start-up by JIT compilation and
286	>	* allocation). So we try to streamline this as much as possible.
287	>	* We park/unpark workers after placing in an event wait queue
288	>	* when they cannot find work. This "queue" is actually a simple
289	>	* Treiber stack, headed by the "id" field of ctl, plus a 15bit
290	>	* counter value (that reflects the number of times a worker has
291	>	* been inactivated) to avoid ABA effects (we need only as many
292	>	* version numbers as worker threads). Successors are held in
293	>	* field WorkQueue.nextWait.  Queuing deals with several intrinsic
294	>	* races, mainly that a task-producing thread can miss seeing (and
295	>	* signalling) another thread that gave up looking for work but
296	>	* has not yet entered the wait queue. We solve this by requiring
297	>	* a full sweep of all workers (via repeated calls to method
298	>	* scan()) both before and after a newly waiting worker is added
299	>	* to the wait queue. During a rescan, the worker might release
300	>	* some other queued worker rather than itself, which has the same
301	>	* net effect. Because enqueued workers may actually be rescanning
302	>	* rather than waiting, we set and clear the "parker" field of
303	>	* WorkQueues to reduce unnecessary calls to unpark.  (This
304	>	* requires a secondary recheck to avoid missed signals.)  Note
305	>	* the unusual conventions about Thread.interrupts surrounding
306	>	* parking and other blocking: Because interrupts are used solely
307	>	* to alert threads to check termination, which is checked anyway
308	>	* upon blocking, we clear status (using Thread.interrupted)
309	>	* before any call to park, so that park does not immediately
310	>	* return due to status being set via some other unrelated call to
311	>	* interrupt in user code.
312	>	*
313	>	* Signalling.  We create or wake up workers only when there
314	>	* appears to be at least one task they might be able to find and
315	>	* execute. However, many other threads may notice the same task
316	>	* and each signal to wake up a thread that might take it. So in
317	>	* general, pools will be over-signalled.  When a submission is
318	>	* added or another worker adds a task to a queue that is
319	>	* apparently empty, they signal waiting workers (or trigger
320	>	* creation of new ones if fewer than the given parallelism level
321	>	* -- see signalWork).  These primary signals are buttressed by
322	>	* signals whenever other threads scan for work or do not have a
323	>	* task to process. On most platforms, signalling (unpark)
324	>	* overhead time is noticeably long, and the time between
325	>	* signalling a thread and it actually making progress can be very
326	>	* noticeably long, so it is worth offloading these delays from
327	>	* critical paths as much as possible.
328	>	*
329	>	* Trimming workers. To release resources after periods of lack of
330	>	* use, a worker starting to wait when the pool is quiescent will
331	>	* time out and terminate if the pool has remained quiescent for a
332	>	* given period -- a short period if there are more threads than
333	>	* parallelism, longer as the number of threads decreases. This
334	>	* will slowly propagate, eventually terminating all workers after
335	>	* periods of non-use.
336	>	*
337	>	* Shutdown and Termination. A call to shutdownNow atomically sets
338	>	* a plock bit and then (non-atomically) sets each worker's
339	>	* qlock status, cancels all unprocessed tasks, and wakes up
340	>	* all waiting workers.  Detecting whether termination should
341	>	* commence after a non-abrupt shutdown() call requires more work
342	>	* and bookkeeping. We need consensus about quiescence (i.e., that
343	>	* there is no more work). The active count provides a primary
344	>	* indication but non-abrupt shutdown still requires a rechecking
345	>	* scan for any workers that are inactive but not queued.
346	>	*
347	>	* Joining Tasks
348	>	* =============
349	>	*
350	>	* Any of several actions may be taken when one worker is waiting
351	>	* to join a task stolen (or always held) by another.  Because we
352	>	* are multiplexing many tasks on to a pool of workers, we can't
353	>	* just let them block (as in Thread.join).  We also cannot just
354	>	* reassign the joiner's run-time stack with another and replace
355	>	* it later, which would be a form of "continuation", that even if
356	>	* possible is not necessarily a good idea since we sometimes need
357	>	* both an unblocked task and its continuation to progress.
358	>	* Instead we combine two tactics:
359	>	*
360	>	*   Helping: Arranging for the joiner to execute some task that it
361	>	*      would be running if the steal had not occurred.
362	>	*
363	>	*   Compensating: Unless there are already enough live threads,
364	>	*      method tryCompensate() may create or re-activate a spare
365	>	*      thread to compensate for blocked joiners until they unblock.
366	>	*
367	>	* A third form (implemented in tryRemoveAndExec) amounts to
368	>	* helping a hypothetical compensator: If we can readily tell that
369	>	* a possible action of a compensator is to steal and execute the
370	>	* task being joined, the joining thread can do so directly,
371	>	* without the need for a compensation thread (although at the
372	>	* expense of larger run-time stacks, but the tradeoff is
373	>	* typically worthwhile).
374	>	*
375	>	* The ManagedBlocker extension API can't use helping so relies
376	>	* only on compensation in method awaitBlocker.
377	>	*
378	>	* The algorithm in tryHelpStealer entails a form of "linear"
379	>	* helping: Each worker records (in field currentSteal) the most
380	>	* recent task it stole from some other worker. Plus, it records
381	>	* (in field currentJoin) the task it is currently actively
382	>	* joining. Method tryHelpStealer uses these markers to try to
383	>	* find a worker to help (i.e., steal back a task from and execute
384	>	* it) that could hasten completion of the actively joined task.
385	>	* In essence, the joiner executes a task that would be on its own
386	>	* local deque had the to-be-joined task not been stolen. This may
387	>	* be seen as a conservative variant of the approach in Wagner &
388	>	* Calder "Leapfrogging: a portable technique for implementing
389	>	* efficient futures" SIGPLAN Notices, 1993
390	>	* (http://portal.acm.org/citation.cfm?id=155354). It differs in
391	>	* that: (1) We only maintain dependency links across workers upon
392	>	* steals, rather than use per-task bookkeeping.  This sometimes
393	>	* requires a linear scan of workQueues array to locate stealers,
394	>	* but often doesn't because stealers leave hints (that may become
395	>	* stale/wrong) of where to locate them.  A stealHint is only a
396	>	* hint because a worker might have had multiple steals and the
397	>	* hint records only one of them (usually the most current).
398	>	* Hinting isolates cost to when it is needed, rather than adding
399	>	* to per-task overhead.  (2) It is "shallow", ignoring nesting
400	>	* and potentially cyclic mutual steals.  (3) It is intentionally
401	>	* racy: field currentJoin is updated only while actively joining,
402	>	* which means that we miss links in the chain during long-lived
403	>	* tasks, GC stalls etc (which is OK since blocking in such cases
404	>	* is usually a good idea).  (4) We bound the number of attempts
405	>	* to find work (see MAX_HELP) and fall back to suspending the
406	>	* worker and if necessary replacing it with another.
407	>	*
408	>	* Helping actions for CountedCompleters are much simpler: Method
409	>	* helpComplete can take and execute any task with the same root
410	>	* as the task being waited on. However, this still entails some
411	>	* traversal of completer chains, so is less efficient than using
412	>	* CountedCompleters without explicit joins.
413	>	*
414	>	* It is impossible to keep exactly the target parallelism number
415	>	* of threads running at any given time.  Determining the
416	>	* existence of conservatively safe helping targets, the
417	>	* availability of already-created spares, and the apparent need
418	>	* to create new spares are all racy, so we rely on multiple
419	>	* retries of each.  Compensation in the apparent absence of
420	>	* helping opportunities is challenging to control on JVMs, where
421	>	* GC and other activities can stall progress of tasks that in
422	>	* turn stall out many other dependent tasks, without us being
423	>	* able to determine whether they will ever require compensation.
424	>	* Even though work-stealing otherwise encounters little
425	>	* degradation in the presence of more threads than cores,
426	>	* aggressively adding new threads in such cases entails risk of
427	>	* unwanted positive feedback control loops in which more threads
428	>	* cause more dependent stalls (as well as delayed progress of
429	>	* unblocked threads to the point that we know they are available)
430	>	* leading to more situations requiring more threads, and so
431	>	* on. This aspect of control can be seen as an (analytically
432	>	* intractable) game with an opponent that may choose the worst
433	>	* (for us) active thread to stall at any time.  We take several
434	>	* precautions to bound losses (and thus bound gains), mainly in
435	>	* methods tryCompensate and awaitJoin.
436	>	*
437	>	* Common Pool
438	>	* ===========
439	>	*
440	>	* The static commonPool always exists after static
441	>	* initialization.  Since it (or any other created pool) need
442	>	* never be used, we minimize initial construction overhead and
443	>	* footprint to the setup of about a dozen fields, with no nested
444	>	* allocation. Most bootstrapping occurs within method
445	>	* fullExternalPush during the first submission to the pool.
446	>	*
447	>	* When external threads submit to the common pool, they can
448	>	* perform some subtask processing (see externalHelpJoin and
449	>	* related methods).  We do not need to record whether these
450	>	* submissions are to the common pool -- if not, externalHelpJoin
451	>	* returns quicky (at the most helping to signal some common pool
452	>	* workers). These submitters would otherwise be blocked waiting
453	>	* for completion, so the extra effort (with liberally sprinkled
454	>	* task status checks) in inapplicable cases amounts to an odd
455	>	* form of limited spin-wait before blocking in ForkJoinTask.join.
456	>	*
457	>	* Style notes
458	>	* ===========
459	>	*
460	>	* There is a lot of representation-level coupling among classes
461	>	* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask.  The
462	>	* fields of WorkQueue maintain data structures managed by
463	>	* ForkJoinPool, so are directly accessed.  There is little point
464	>	* trying to reduce this, since any associated future changes in
465	>	* representations will need to be accompanied by algorithmic
466	>	* changes anyway. Several methods intrinsically sprawl because
467	>	* they must accumulate sets of consistent reads of volatiles held
468	>	* in local variables.  Methods signalWork() and scan() are the
469	>	* main bottlenecks, so are especially heavily
470	>	* micro-optimized/mangled.  There are lots of inline assignments
471	>	* (of form "while ((local = field) != 0)") which are usually the
472	>	* simplest way to ensure the required read orderings (which are
473	>	* sometimes critical). This leads to a "C"-like style of listing
474	>	* declarations of these locals at the heads of methods or blocks.
475	>	* There are several occurrences of the unusual "do {} while
476	>	* (!cas...)"  which is the simplest way to force an update of a
477	>	* CAS'ed variable. There are also other coding oddities (including
478	>	* several unnecessary-looking hoisted null checks) that help
479	>	* some methods perform reasonably even when interpreted (not
480	>	* compiled).
481	>	*
482	>	* The order of declarations in this file is:
483	>	* (1) Static utility functions
484	>	* (2) Nested (static) classes
485	>	* (3) Static fields
486	>	* (4) Fields, along with constants used when unpacking some of them
487	>	* (5) Internal control methods
488	>	* (6) Callbacks and other support for ForkJoinTask methods
489	>	* (7) Exported methods
490	>	* (8) Static block initializing statics in minimally dependent order
491		*/
492
493	<	/** Mask for packing and unpacking shorts */
494	<	private static final int  shortMask = 0xffff;
493	>	// Static utilities
494	>
495	>	/**
496	>	* If there is a security manager, makes sure caller has
497	>	* permission to modify threads.
498	>	*/
499	>	private static void checkPermission() {
500	>	SecurityManager security = System.getSecurityManager();
501	>	if (security != null)
502	>	security.checkPermission(modifyThreadPermission);
503	>	}
504
505	<	/** Max pool size -- must be a power of two minus 1 */
74	<	private static final int MAX_THREADS =  0x7FFF;
505	>	// Nested classes
506
507		/**
508	<	* Factory for creating new ForkJoinWorkerThreads.  A
509	<	* ForkJoinWorkerThreadFactory must be defined and used for
510	<	* ForkJoinWorkerThread subclasses that extend base functionality
511	<	* or initialize threads with different contexts.
508	>	* Factory for creating new {@link ForkJoinWorkerThread}s.
509	>	* A {@code ForkJoinWorkerThreadFactory} must be defined and used
510	>	* for {@code ForkJoinWorkerThread} subclasses that extend base
511	>	* functionality or initialize threads with different contexts.
512		*/
513		public static interface ForkJoinWorkerThreadFactory {
514		/**
515		* Returns a new worker thread operating in the given pool.
516		*
517		* @param pool the pool this thread works in
518	<	* @throws NullPointerException if pool is null
518	>	* @throws NullPointerException if the pool is null
519		*/
520		public ForkJoinWorkerThread newThread(ForkJoinPool pool);
521		}
#	Line 93 | Line 524 | public class ForkJoinPool extends Abstra
524		* Default ForkJoinWorkerThreadFactory implementation; creates a
525		* new ForkJoinWorkerThread.
526		*/
527	<	static class  DefaultForkJoinWorkerThreadFactory
527	>	static class DefaultForkJoinWorkerThreadFactory
528		implements ForkJoinWorkerThreadFactory {
529		public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
530	+	return new ForkJoinWorkerThread(pool);
531	+	}
532	+	}
533	+
534	+	/**
535	+	* Class for artificial tasks that are used to replace the target
536	+	* of local joins if they are removed from an interior queue slot
537	+	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
538	+	* actually do anything beyond having a unique identity.
539	+	*/
540	+	static final class EmptyTask extends ForkJoinTask<Void> {
541	+	private static final long serialVersionUID = -7721805057305804111L;
542	+	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
543	+	public final Void getRawResult() { return null; }
544	+	public final void setRawResult(Void x) {}
545	+	public final boolean exec() { return true; }
546	+	}
547	+
548	+	/**
549	+	* Queues supporting work-stealing as well as external task
550	+	* submission. See above for main rationale and algorithms.
551	+	* Implementation relies heavily on "Unsafe" intrinsics
552	+	* and selective use of "volatile":
553	+	*
554	+	* Field "base" is the index (mod array.length) of the least valid
555	+	* queue slot, which is always the next position to steal (poll)
556	+	* from if nonempty. Reads and writes require volatile orderings
557	+	* but not CAS, because updates are only performed after slot
558	+	* CASes.
559	+	*
560	+	* Field "top" is the index (mod array.length) of the next queue
561	+	* slot to push to or pop from. It is written only by owner thread
562	+	* for push, or under lock for external/shared push, and accessed
563	+	* by other threads only after reading (volatile) base.  Both top
564	+	* and base are allowed to wrap around on overflow, but (top -
565	+	* base) (or more commonly -(base - top) to force volatile read of
566	+	* base before top) still estimates size. The lock ("qlock") is
567	+	* forced to -1 on termination, causing all further lock attempts
568	+	* to fail. (Note: we don't need CAS for termination state because
569	+	* upon pool shutdown, all shared-queues will stop being used
570	+	* anyway.)  Nearly all lock bodies are set up so that exceptions
571	+	* within lock bodies are "impossible" (modulo JVM errors that
572	+	* would cause failure anyway.)
573	+	*
574	+	* The array slots are read and written using the emulation of
575	+	* volatiles/atomics provided by Unsafe. Insertions must in
576	+	* general use putOrderedObject as a form of releasing store to
577	+	* ensure that all writes to the task object are ordered before
578	+	* its publication in the queue.  All removals entail a CAS to
579	+	* null.  The array is always a power of two. To ensure safety of
580	+	* Unsafe array operations, all accesses perform explicit null
581	+	* checks and implicit bounds checks via power-of-two masking.
582	+	*
583	+	* In addition to basic queuing support, this class contains
584	+	* fields described elsewhere to control execution. It turns out
585	+	* to work better memory-layout-wise to include them in this class
586	+	* rather than a separate class.
587	+	*
588	+	* Performance on most platforms is very sensitive to placement of
589	+	* instances of both WorkQueues and their arrays -- we absolutely
590	+	* do not want multiple WorkQueue instances or multiple queue
591	+	* arrays sharing cache lines. (It would be best for queue objects
592	+	* and their arrays to share, but there is nothing available to
593	+	* help arrange that).  Unfortunately, because they are recorded
594	+	* in a common array, WorkQueue instances are often moved to be
595	+	* adjacent by garbage collectors. To reduce impact, we use field
596	+	* padding that works OK on common platforms; this effectively
597	+	* trades off slightly slower average field access for the sake of
598	+	* avoiding really bad worst-case access. (Until better JVM
599	+	* support is in place, this padding is dependent on transient
600	+	* properties of JVM field layout rules.)  We also take care in
601	+	* allocating, sizing and resizing the array. Non-shared queue
602	+	* arrays are initialized by workers before use. Others are
603	+	* allocated on first use.
604	+	*/
605	+	static final class WorkQueue {
606	+	/**
607	+	* Capacity of work-stealing queue array upon initialization.
608	+	* Must be a power of two; at least 4, but should be larger to
609	+	* reduce or eliminate cacheline sharing among queues.
610	+	* Currently, it is much larger, as a partial workaround for
611	+	* the fact that JVMs often place arrays in locations that
612	+	* share GC bookkeeping (especially cardmarks) such that
613	+	* per-write accesses encounter serious memory contention.
614	+	*/
615	+	static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
616	+
617	+	/**
618	+	* Maximum size for queue arrays. Must be a power of two less
619	+	* than or equal to 1 << (31 - width of array entry) to ensure
620	+	* lack of wraparound of index calculations, but defined to a
621	+	* value a bit less than this to help users trap runaway
622	+	* programs before saturating systems.
623	+	*/
624	+	static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
625	+
626	+	int seed;                  // for random scanning; initialize nonzero
627	+	volatile int eventCount;   // encoded inactivation count; < 0 if inactive
628	+	int nextWait;              // encoded record of next event waiter
629	+	final int mode;            // lifo, fifo, or shared
630	+	int nsteals;               // cumulative number of steals
631	+	int poolIndex;             // index of this queue in pool (or 0)
632	+	int stealHint;             // index of most recent known stealer
633	+	volatile int qlock;        // 1: locked, -1: terminate; else 0
634	+	volatile int base;         // index of next slot for poll
635	+	int top;                   // index of next slot for push
636	+	ForkJoinTask<?>[] array;   // the elements (initially unallocated)
637	+	final ForkJoinPool pool;   // the containing pool (may be null)
638	+	final ForkJoinWorkerThread owner; // owning thread or null if shared
639	+	volatile Thread parker;    // == owner during call to park; else null
640	+	volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
641	+	ForkJoinTask<?> currentSteal; // current non-local task being executed
642	+	// Heuristic padding to ameliorate unfortunate memory placements
643	+	Object p00, p01, p02, p03, p04, p05, p06, p07;
644	+	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
645	+
646	+	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
647	+	this.mode = mode;
648	+	this.pool = pool;
649	+	this.owner = owner;
650	+	// Place indices in the center of array (that is not yet allocated)
651	+	base = top = INITIAL_QUEUE_CAPACITY >>> 1;
652	+	}
653	+
654	+	/**
655	+	* Pushes a task. Call only by owner in unshared queues.
656	+	* Cases needing resizing or rejection are relyaed to fullPush
657	+	* (that also handles shared queues).
658	+	*
659	+	* @param task the task. Caller must ensure non-null.
660	+	* @throw RejectedExecutionException if array cannot be resized
661	+	*/
662	+	final void push(ForkJoinTask<?> task) {
663	+	ForkJoinPool p; ForkJoinTask<?>[] a;
664	+	int s = top, n;
665	+	if ((a = array) != null && a.length > (n = s + 1 - base)) {
666	+	U.putOrderedObject
667	+	(a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
668	+	top = s + 1;
669	+	if (n <= 1 && (p = pool) != null)
670	+	p.signalWork(this, 1);
671	+	}
672	+	else
673	+	fullPush(task, true);
674	+	}
675	+
676	+	/**
677	+	* Pushes a task if lock is free and array is either big
678	+	* enough or can be resized to be big enough. Note: a
679	+	* specialization of a common fast path of this method is in
680	+	* ForkJoinPool.externalPush. When called from a FJWT queue,
681	+	* this can fail only if the pool has been shut down or
682	+	* an out of memory error.
683	+	*
684	+	* @param task the task. Caller must ensure non-null.
685	+	* @param owned if true, throw RJE on failure
686	+	*/
687	+	final boolean fullPush(ForkJoinTask<?> task, boolean owned) {
688	+	ForkJoinPool p; ForkJoinTask<?>[] a;
689	+	if (owned) {
690	+	if (qlock < 0) // must be shutting down
691	+	throw new RejectedExecutionException();
692	+	}
693	+	else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
694	+	return false;
695		try {
696	<	return new ForkJoinWorkerThread(pool);
697	<	} catch (OutOfMemoryError oom)  {
696	>	int s = top, oldLen, len;
697	>	if ((a = array) == null)
698	>	a = array = new ForkJoinTask<?>[len=INITIAL_QUEUE_CAPACITY];
699	>	else if ((oldLen = a.length) > s + 1 - base)
700	>	len = oldLen;
701	>	else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
702	>	throw new RejectedExecutionException("Capacity exceeded");
703	>	else {
704	>	int oldMask, b;
705	>	ForkJoinTask<?>[] oldA = a;
706	>	a = array = new ForkJoinTask<?>[len];
707	>	if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
708	>	int mask = len - 1;
709	>	do {
710	>	ForkJoinTask<?> x;
711	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
712	>	int j    = ((b &    mask) << ASHIFT) + ABASE;
713	>	x = (ForkJoinTask<?>)
714	>	U.getObjectVolatile(oldA, oldj);
715	>	if (x != null &&
716	>	U.compareAndSwapObject(oldA, oldj, x, null))
717	>	U.putObjectVolatile(a, j, x);
718	>	} while (++b != s);
719	>	}
720	>	}
721	>	U.putOrderedObject
722	>	(a, (((len - 1) & s) << ASHIFT) + ABASE, task);
723	>	top = s + 1;
724	>	} finally {
725	>	if (!owned)
726	>	qlock = 0;
727	>	}
728	>	if ((p = pool) != null)
729	>	p.signalWork(this, 1);
730	>	return true;
731	>	}
732	>
733	>	/**
734	>	* Takes next task, if one exists, in LIFO order.  Call only
735	>	* by owner in unshared queues.
736	>	*/
737	>	final ForkJoinTask<?> pop() {
738	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
739	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
740	>	for (int s; (s = top - 1) - base >= 0;) {
741	>	long j = ((m & s) << ASHIFT) + ABASE;
742	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
743	>	break;
744	>	if (U.compareAndSwapObject(a, j, t, null)) {
745	>	top = s;
746	>	return t;
747	>	}
748	>	}
749	>	}
750	>	return null;
751	>	}
752	>
753	>	/**
754	>	* Takes a task in FIFO order if b is base of queue and a task
755	>	* can be claimed without contention. Specialized versions
756	>	* appear in ForkJoinPool methods scan and tryHelpStealer.
757	>	*/
758	>	final ForkJoinTask<?> pollAt(int b) {
759	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
760	>	if ((a = array) != null) {
761	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
762	>	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
763	>	base == b &&
764	>	U.compareAndSwapObject(a, j, t, null)) {
765	>	base = b + 1;
766	>	return t;
767	>	}
768	>	}
769	>	return null;
770	>	}
771	>
772	>	/**
773	>	* Takes next task, if one exists, in FIFO order.
774	>	*/
775	>	final ForkJoinTask<?> poll() {
776	>	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
777	>	while ((b = base) - top < 0 && (a = array) != null) {
778	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
779	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
780	>	if (t != null) {
781	>	if (base == b &&
782	>	U.compareAndSwapObject(a, j, t, null)) {
783	>	base = b + 1;
784	>	return t;
785	>	}
786	>	}
787	>	else if (base == b) {
788	>	if (b + 1 == top)
789	>	break;
790	>	Thread.yield(); // wait for lagging update (very rare)
791	>	}
792	>	}
793	>	return null;
794	>	}
795	>
796	>	/**
797	>	* Takes next task, if one exists, in order specified by mode.
798	>	*/
799	>	final ForkJoinTask<?> nextLocalTask() {
800	>	return mode == 0 ? pop() : poll();
801	>	}
802	>
803	>	/**
804	>	* Returns next task, if one exists, in order specified by mode.
805	>	*/
806	>	final ForkJoinTask<?> peek() {
807	>	ForkJoinTask<?>[] a = array; int m;
808	>	if (a == null || (m = a.length - 1) < 0)
809		return null;
810	+	int i = mode == 0 ? top - 1 : base;
811	+	int j = ((i & m) << ASHIFT) + ABASE;
812	+	return (ForkJoinTask<?>)U.getObjectVolatile(a, j);
813	+	}
814	+
815	+	/**
816	+	* Pops the given task only if it is at the current top.
817	+	* (A shared version is available only via FJP.tryExternalUnpush)
818	+	*/
819	+	final boolean tryUnpush(ForkJoinTask<?> t) {
820	+	ForkJoinTask<?>[] a; int s;
821	+	if ((a = array) != null && (s = top) != base &&
822	+	U.compareAndSwapObject
823	+	(a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
824	+	top = s;
825	+	return true;
826	+	}
827	+	return false;
828	+	}
829	+
830	+	/**
831	+	* Removes and cancels all known tasks, ignoring any exceptions.
832	+	*/
833	+	final void cancelAll() {
834	+	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
835	+	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
836	+	for (ForkJoinTask<?> t; (t = poll()) != null; )
837	+	ForkJoinTask.cancelIgnoringExceptions(t);
838	+	}
839	+
840	+	/**
841	+	* Computes next value for random probes.  Scans don't require
842	+	* a very high quality generator, but also not a crummy one.
843	+	* Marsaglia xor-shift is cheap and works well enough.  Note:
844	+	* This is manually inlined in its usages in ForkJoinPool to
845	+	* avoid writes inside busy scan loops.
846	+	*/
847	+	final int nextSeed() {
848	+	int r = seed;
849	+	r ^= r << 13;
850	+	r ^= r >>> 17;
851	+	return seed = r ^= r << 5;
852	+	}
853	+
854	+	/**
855	+	* Provides a more accurate estimate of size than (top - base)
856	+	* by ordering reads and checking whether a near-empty queue
857	+	* has at least one unclaimed task.
858	+	*/
859	+	final int queueSize() {
860	+	ForkJoinTask<?>[] a; int k, s, n;
861	+	return ((n = base - (s = top)) < 0 &&
862	+	(n != -1 ||
863	+	((a = array) != null && (k = a.length) > 0 &&
864	+	U.getObject
865	+	(a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
866	+	-n : 0;
867	+	}
868	+
869	+	// Specialized execution methods
870	+
871	+	/**
872	+	* Pops and runs tasks until empty.
873	+	*/
874	+	private void popAndExecAll() {
875	+	// A bit faster than repeated pop calls
876	+	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
877	+	while ((a = array) != null && (m = a.length - 1) >= 0 &&
878	+	(s = top - 1) - base >= 0 &&
879	+	(t = ((ForkJoinTask<?>)
880	+	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
881	+	!= null) {
882	+	if (U.compareAndSwapObject(a, j, t, null)) {
883	+	top = s;
884	+	t.doExec();
885	+	}
886		}
887		}
888	+
889	+	/**
890	+	* Polls and runs tasks until empty.
891	+	*/
892	+	private void pollAndExecAll() {
893	+	for (ForkJoinTask<?> t; (t = poll()) != null;)
894	+	t.doExec();
895	+	}
896	+
897	+	/**
898	+	* If present, removes from queue and executes the given task,
899	+	* or any other cancelled task. Returns (true) on any CAS
900	+	* or consistency check failure so caller can retry.
901	+	*
902	+	* @return false if no progress can be made, else true;
903	+	*/
904	+	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
905	+	boolean stat = true, removed = false, empty = true;
906	+	ForkJoinTask<?>[] a; int m, s, b, n;
907	+	if ((a = array) != null && (m = a.length - 1) >= 0 &&
908	+	(n = (s = top) - (b = base)) > 0) {
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	+	stat = false;
936	+	break;
937	+	}
938	+	}
939	+	}
940	+	if (removed)
941	+	task.doExec();
942	+	return stat;
943	+	}
944	+
945	+	/**
946	+	* Polls for and executes the given task or any other task in
947	+	* its CountedCompleter computation
948	+	*/
949	+	final boolean pollAndExecCC(ForkJoinTask<?> root) {
950	+	ForkJoinTask<?>[] a; int b; Object o;
951	+	outer: while ((b = base) - top < 0 && (a = array) != null) {
952	+	long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
953	+	if ((o = U.getObject(a, j)) == null ||
954	+	!(o instanceof CountedCompleter))
955	+	break;
956	+	for (CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;;) {
957	+	if (r == root) {
958	+	if (base == b &&
959	+	U.compareAndSwapObject(a, j, t, null)) {
960	+	base = b + 1;
961	+	t.doExec();
962	+	return true;
963	+	}
964	+	else
965	+	break; // restart
966	+	}
967	+	if ((r = r.completer) == null)
968	+	break outer; // not part of root computation
969	+	}
970	+	}
971	+	return false;
972	+	}
973	+
974	+	/**
975	+	* Executes a top-level task and any local tasks remaining
976	+	* after execution.
977	+	*/
978	+	final void runTask(ForkJoinTask<?> t) {
979	+	if (t != null) {
980	+	(currentSteal = t).doExec();
981	+	currentSteal = null;
982	+	if (++nsteals < 0) {     // spill on overflow
983	+	ForkJoinPool p;
984	+	if ((p = pool) != null)
985	+	p.collectStealCount(this);
986	+	}
987	+	if (top != base) {       // process remaining local tasks
988	+	if (mode == 0)
989	+	popAndExecAll();
990	+	else
991	+	pollAndExecAll();
992	+	}
993	+	}
994	+	}
995	+
996	+	/**
997	+	* Executes a non-top-level (stolen) task.
998	+	*/
999	+	final void runSubtask(ForkJoinTask<?> t) {
1000	+	if (t != null) {
1001	+	ForkJoinTask<?> ps = currentSteal;
1002	+	(currentSteal = t).doExec();
1003	+	currentSteal = ps;
1004	+	}
1005	+	}
1006	+
1007	+	/**
1008	+	* Returns true if owned and not known to be blocked.
1009	+	*/
1010	+	final boolean isApparentlyUnblocked() {
1011	+	Thread wt; Thread.State s;
1012	+	return (eventCount >= 0 &&
1013	+	(wt = owner) != null &&
1014	+	(s = wt.getState()) != Thread.State.BLOCKED &&
1015	+	s != Thread.State.WAITING &&
1016	+	s != Thread.State.TIMED_WAITING);
1017	+	}
1018	+
1019	+	/**
1020	+	* If this owned and is not already interrupted, try to
1021	+	* interrupt and/or unpark, ignoring exceptions.
1022	+	*/
1023	+	final void interruptOwner() {
1024	+	Thread wt, p;
1025	+	if ((wt = owner) != null && !wt.isInterrupted()) {
1026	+	try {
1027	+	wt.interrupt();
1028	+	} catch (SecurityException ignore) {
1029	+	}
1030	+	}
1031	+	if ((p = parker) != null)
1032	+	U.unpark(p);
1033	+	}
1034	+
1035	+	// Unsafe mechanics
1036	+	private static final sun.misc.Unsafe U;
1037	+	private static final long QLOCK;
1038	+	private static final int ABASE;
1039	+	private static final int ASHIFT;
1040	+	static {
1041	+	int s;
1042	+	try {
1043	+	U = getUnsafe();
1044	+	Class<?> k = WorkQueue.class;
1045	+	Class<?> ak = ForkJoinTask[].class;
1046	+	QLOCK = U.objectFieldOffset
1047	+	(k.getDeclaredField("qlock"));
1048	+	ABASE = U.arrayBaseOffset(ak);
1049	+	s = U.arrayIndexScale(ak);
1050	+	} catch (Exception e) {
1051	+	throw new Error(e);
1052	+	}
1053	+	if ((s & (s-1)) != 0)
1054	+	throw new Error("data type scale not a power of two");
1055	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1056	+	}
1057		}
1058
1059		/**
1060	+	* Per-thread records for threads that submit to pools. Currently
1061	+	* holds only pseudo-random seed / index that is used to choose
1062	+	* submission queues in method externalPush. In the future, this may
1063	+	* also incorporate a means to implement different task rejection
1064	+	* and resubmission policies.
1065	+	*
1066	+	* Seeds for submitters and workers/workQueues work in basically
1067	+	* the same way but are initialized and updated using slightly
1068	+	* different mechanics. Both are initialized using the same
1069	+	* approach as in class ThreadLocal, where successive values are
1070	+	* unlikely to collide with previous values. Seeds are then
1071	+	* randomly modified upon collisions using xorshifts, which
1072	+	* requires a non-zero seed.
1073	+	*/
1074	+	static final class Submitter {
1075	+	int seed;
1076	+	Submitter(int s) { seed = s; }
1077	+	}
1078	+
1079	+	/** Property prefix for constructing common pool */
1080	+	private static final String propPrefix =
1081	+	"java.util.concurrent.ForkJoinPool.common.";
1082	+
1083	+	// static fields (initialized in static initializer below)
1084	+
1085	+	/**
1086		* Creates a new ForkJoinWorkerThread. This factory is used unless
1087		* overridden in ForkJoinPool constructors.
1088		*/
1089		public static final ForkJoinWorkerThreadFactory
1090	<	defaultForkJoinWorkerThreadFactory =
1091	<	new DefaultForkJoinWorkerThreadFactory();
1090	>	defaultForkJoinWorkerThreadFactory;
1091	>
1092	>	/**
1093	>	* Common (static) pool. Non-null for public use unless a static
1094	>	* construction exception, but internal usages null-check on use
1095	>	* to paranoically avoid potential initialization circularities
1096	>	* as well as to simplify generated code.
1097	>	*/
1098	>	static final ForkJoinPool commonPool;
1099
1100		/**
1101		* Permission required for callers of methods that may start or
1102		* kill threads.
1103		*/
1104	<	private static final RuntimePermission modifyThreadPermission =
120	<	new RuntimePermission("modifyThread");
1104	>	private static final RuntimePermission modifyThreadPermission;
1105
1106		/**
1107	<	* If there is a security manager, makes sure caller has
1108	<	* permission to modify threads.
1107	>	* Per-thread submission bookkeeping. Shared across all pools
1108	>	* to reduce ThreadLocal pollution and because random motion
1109	>	* to avoid contention in one pool is likely to hold for others.
1110	>	* Lazily initialized on first submission (but null-checked
1111	>	* in other contexts to avoid unnecessary initialization).
1112		*/
1113	<	private static void checkPermission() {
127	<	SecurityManager security = System.getSecurityManager();
128	<	if (security != null)
129	<	security.checkPermission(modifyThreadPermission);
130	<	}
1113	>	static final ThreadLocal<Submitter> submitters;
1114
1115		/**
1116	<	* Generator for assigning sequence numbers as pool names.
1116	>	* Common pool parallelism. Must equal commonPool.parallelism.
1117		*/
1118	<	private static final AtomicInteger poolNumberGenerator =
136	<	new AtomicInteger();
1118	>	static final int commonPoolParallelism;
1119
1120		/**
1121	<	* Array holding all worker threads in the pool. Initialized upon
140	<	* first use. Array size must be a power of two.  Updates and
141	<	* replacements are protected by workerLock, but it is always kept
142	<	* in a consistent enough state to be randomly accessed without
143	<	* locking by workers performing work-stealing.
1121	>	* Sequence number for creating workerNamePrefix.
1122		*/
1123	<	volatile ForkJoinWorkerThread[] workers;
1123	>	private static int poolNumberSequence;
1124
1125		/**
1126	<	* Lock protecting access to workers.
1126	>	* Return the next sequence number. We don't expect this to
1127	>	* ever contend so use simple builtin sync.
1128		*/
1129	<	private final ReentrantLock workerLock;
1129	>	private static final synchronized int nextPoolId() {
1130	>	return ++poolNumberSequence;
1131	>	}
1132	>
1133	>	// static constants
1134
1135		/**
1136	<	* Condition for awaitTermination.
1136	>	* Initial timeout value (in nanoseconds) for the thread
1137	>	* triggering quiescence to park waiting for new work. On timeout,
1138	>	* the thread will instead try to shrink the number of
1139	>	* workers. The value should be large enough to avoid overly
1140	>	* aggressive shrinkage during most transient stalls (long GCs
1141	>	* etc).
1142		*/
1143	<	private final Condition termination;
1143	>	private static final long IDLE_TIMEOUT      = 2000L * 1000L * 1000L; // 2sec
1144
1145		/**
1146	<	* The uncaught exception handler used when any worker
159	<	* abruptly terminates
1146	>	* Timeout value when there are more threads than parallelism level
1147		*/
1148	<	private Thread.UncaughtExceptionHandler ueh;
1148	>	private static final long FAST_IDLE_TIMEOUT =  200L * 1000L * 1000L;
1149
1150		/**
1151	<	* Creation factory for worker threads.
1151	>	* The maximum stolen->joining link depth allowed in method
1152	>	* tryHelpStealer.  Must be a power of two.  Depths for legitimate
1153	>	* chains are unbounded, but we use a fixed constant to avoid
1154	>	* (otherwise unchecked) cycles and to bound staleness of
1155	>	* traversal parameters at the expense of sometimes blocking when
1156	>	* we could be helping.
1157		*/
1158	<	private final ForkJoinWorkerThreadFactory factory;
1158	>	private static final int MAX_HELP = 64;
1159
1160		/**
1161	<	* Head of stack of threads that were created to maintain
1162	<	* parallelism when other threads blocked, but have since
171	<	* suspended when the parallelism level rose.
1161	>	* Increment for seed generators. See class ThreadLocal for
1162	>	* explanation.
1163		*/
1164	<	private volatile WaitQueueNode spareStack;
1164	>	private static final int SEED_INCREMENT = 0x61c88647;
1165
1166		/**
1167	<	* Sum of per-thread steal counts, updated only when threads are
1168	<	* idle or terminating.
1167	>	* Bits and masks for control variables
1168	>	*
1169	>	* Field ctl is a long packed with:
1170	>	* AC: Number of active running workers minus target parallelism (16 bits)
1171	>	* TC: Number of total workers minus target parallelism (16 bits)
1172	>	* ST: true if pool is terminating (1 bit)
1173	>	* EC: the wait count of top waiting thread (15 bits)
1174	>	* ID: poolIndex of top of Treiber stack of waiters (16 bits)
1175	>	*
1176	>	* When convenient, we can extract the upper 32 bits of counts and
1177	>	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
1178	>	* (int)ctl.  The ec field is never accessed alone, but always
1179	>	* together with id and st. The offsets of counts by the target
1180	>	* parallelism and the positionings of fields makes it possible to
1181	>	* perform the most common checks via sign tests of fields: When
1182	>	* ac is negative, there are not enough active workers, when tc is
1183	>	* negative, there are not enough total workers, and when e is
1184	>	* negative, the pool is terminating.  To deal with these possibly
1185	>	* negative fields, we use casts in and out of "short" and/or
1186	>	* signed shifts to maintain signedness.
1187	>	*
1188	>	* When a thread is queued (inactivated), its eventCount field is
1189	>	* set negative, which is the only way to tell if a worker is
1190	>	* prevented from executing tasks, even though it must continue to
1191	>	* scan for them to avoid queuing races. Note however that
1192	>	* eventCount updates lag releases so usage requires care.
1193	>	*
1194	>	* Field plock is an int packed with:
1195	>	* SHUTDOWN: true if shutdown is enabled (1 bit)
1196	>	* SEQ:  a sequence lock, with PL_LOCK bit set if locked (30 bits)
1197	>	* SIGNAL: set when threads may be waiting on the lock (1 bit)
1198	>	*
1199	>	* The sequence number enables simple consistency checks:
1200	>	* Staleness of read-only operations on the workQueues array can
1201	>	* be checked by comparing plock before vs after the reads.
1202		*/
179	–	private final AtomicLong stealCount;
1203
1204	<	/**
1205	<	* Queue for external submissions.
1204	>	// bit positions/shifts for fields
1205	>	private static final int  AC_SHIFT   = 48;
1206	>	private static final int  TC_SHIFT   = 32;
1207	>	private static final int  ST_SHIFT   = 31;
1208	>	private static final int  EC_SHIFT   = 16;
1209	>
1210	>	// bounds
1211	>	private static final int  SMASK      = 0xffff;  // short bits
1212	>	private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
1213	>	private static final int  EVENMASK   = 0xfffe;  // even short bits
1214	>	private static final int  SQMASK     = 0x007e;  // max 64 (even) slots
1215	>	private static final int  SHORT_SIGN = 1 << 15;
1216	>	private static final int  INT_SIGN   = 1 << 31;
1217	>
1218	>	// masks
1219	>	private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
1220	>	private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
1221	>	private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
1222	>
1223	>	// units for incrementing and decrementing
1224	>	private static final long TC_UNIT    = 1L << TC_SHIFT;
1225	>	private static final long AC_UNIT    = 1L << AC_SHIFT;
1226	>
1227	>	// masks and units for dealing with u = (int)(ctl >>> 32)
1228	>	private static final int  UAC_SHIFT  = AC_SHIFT - 32;
1229	>	private static final int  UTC_SHIFT  = TC_SHIFT - 32;
1230	>	private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
1231	>	private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
1232	>	private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
1233	>	private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
1234	>
1235	>	// masks and units for dealing with e = (int)ctl
1236	>	private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
1237	>	private static final int E_SEQ       = 1 << EC_SHIFT;
1238	>
1239	>	// plock bits
1240	>	private static final int SHUTDOWN    = 1 << 31;
1241	>	private static final int PL_LOCK     = 2;
1242	>	private static final int PL_SIGNAL   = 1;
1243	>	private static final int PL_SPINS    = 1 << 8;
1244	>
1245	>	// access mode for WorkQueue
1246	>	static final int LIFO_QUEUE          =  0;
1247	>	static final int FIFO_QUEUE          =  1;
1248	>	static final int SHARED_QUEUE        = -1;
1249	>
1250	>	// Instance fields
1251	>
1252	>	/*
1253	>	* Field layout order in this class tends to matter more than one
1254	>	* would like. Runtime layout order is only loosely related to
1255	>	* declaration order and may differ across JVMs, but the following
1256	>	* empirically works OK on current JVMs.
1257	>	*/
1258	>	volatile long stealCount;                  // collects worker counts
1259	>	volatile long ctl;                         // main pool control
1260	>	final int parallelism;                     // parallelism level
1261	>	final int localMode;                       // per-worker scheduling mode
1262	>	volatile int indexSeed;                    // worker/submitter index seed
1263	>	volatile int plock;                        // shutdown status and seqLock
1264	>	WorkQueue[] workQueues;                    // main registry
1265	>	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1266	>	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1267	>	final String workerNamePrefix;             // to create worker name string
1268	>
1269	>	/*
1270	>	* Acquires the plock lock to protect worker array and related
1271	>	* updates. This method is called only if an initial CAS on plock
1272	>	* fails. This acts as a spinLock for normal cases, but falls back
1273	>	* to builtin monitor to block when (rarely) needed. This would be
1274	>	* a terrible idea for a highly contended lock, but works fine as
1275	>	* a more conservative alternative to a pure spinlock.  See
1276	>	* internal ConcurrentHashMap documentation for further
1277	>	* explanation of nearly the same construction.
1278		*/
1279	<	private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
1279	>	private int acquirePlock() {
1280	>	int spins = PL_SPINS, r = 0, ps, nps;
1281	>	for (;;) {
1282	>	if (((ps = plock) & PL_LOCK) == 0 &&
1283	>	U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1284	>	return nps;
1285	>	else if (r == 0)
1286	>	r = ThreadLocalRandom.current().nextInt(); // randomize spins
1287	>	else if (spins >= 0) {
1288	>	r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1289	>	if (r >= 0)
1290	>	--spins;
1291	>	}
1292	>	else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
1293	>	synchronized (this) {
1294	>	if ((plock & PL_SIGNAL) != 0) {
1295	>	try {
1296	>	wait();
1297	>	} catch (InterruptedException ie) {
1298	>	try {
1299	>	Thread.currentThread().interrupt();
1300	>	} catch (SecurityException ignore) {
1301	>	}
1302	>	}
1303	>	}
1304	>	else
1305	>	notifyAll();
1306	>	}
1307	>	}
1308	>	}
1309	>	}
1310
1311		/**
1312	<	* Head of Treiber stack for barrier sync. See below for explanation.
1312	>	* Unlocks and signals any thread waiting for plock. Called only
1313	>	* when CAS of seq value for unlock fails.
1314		*/
1315	<	private volatile WaitQueueNode syncStack;
1315	>	private void releasePlock(int ps) {
1316	>	plock = ps;
1317	>	synchronized (this) { notifyAll(); }
1318	>	}
1319	>
1320	>	//  Registering and deregistering workers
1321	>
1322	>	/**
1323	>	* Callback from ForkJoinWorkerThread constructor to establish its
1324	>	* poolIndex and record its WorkQueue. To avoid scanning bias due
1325	>	* to packing entries in front of the workQueues array, we treat
1326	>	* the array as a simple power-of-two hash table using per-thread
1327	>	* seed as hash, expanding as needed.
1328	>	*
1329	>	* @param w the worker's queue
1330	>	*/
1331	>	final void registerWorker(WorkQueue w) {
1332	>	int s, ps; // generate a rarely colliding candidate index seed
1333	>	do {} while (!U.compareAndSwapInt(this, INDEXSEED,
1334	>	s = indexSeed, s += SEED_INCREMENT) ||
1335	>	s == 0); // skip 0
1336	>	if (((ps = plock) & PL_LOCK) != 0 ||
1337	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1338	>	ps = acquirePlock();
1339	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1340	>	try {
1341	>	WorkQueue[] ws;
1342	>	if (w != null && (ws = workQueues) != null) {
1343	>	w.seed = s;
1344	>	int n = ws.length, m = n - 1;
1345	>	int r = (s << 1) | 1;               // use odd-numbered indices
1346	>	if (ws[r &= m] != null) {           // collision
1347	>	int probes = 0;                 // step by approx half size
1348	>	int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1349	>	while (ws[r = (r + step) & m] != null) {
1350	>	if (++probes >= n) {
1351	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1352	>	m = n - 1;
1353	>	probes = 0;
1354	>	}
1355	>	}
1356	>	}
1357	>	w.eventCount = w.poolIndex = r;     // establish before recording
1358	>	ws[r] = w;
1359	>	}
1360	>	} finally {
1361	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1362	>	releasePlock(nps);
1363	>	}
1364	>	}
1365
1366		/**
1367	<	* The count for event barrier
1368	<	*/
1369	<	private volatile long eventCount;
1367	>	* Final callback from terminating worker, as well as upon failure
1368	>	* to construct or start a worker.  Removes record of worker from
1369	>	* array, and adjusts counts. If pool is shutting down, tries to
1370	>	* complete termination.
1371	>	*
1372	>	* @param wt the worker thread or null if construction failed
1373	>	* @param ex the exception causing failure, or null if none
1374	>	*/
1375	>	final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1376	>	WorkQueue w = null;
1377	>	if (wt != null && (w = wt.workQueue) != null) {
1378	>	int ps;
1379	>	collectStealCount(w);
1380	>	w.qlock = -1;                // ensure set
1381	>	if (((ps = plock) & PL_LOCK) != 0 ||
1382	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1383	>	ps = acquirePlock();
1384	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1385	>	try {
1386	>	int idx = w.poolIndex;
1387	>	WorkQueue[] ws = workQueues;
1388	>	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1389	>	ws[idx] = null;
1390	>	} finally {
1391	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1392	>	releasePlock(nps);
1393	>	}
1394	>	}
1395	>
1396	>	long c;                             // adjust ctl counts
1397	>	do {} while (!U.compareAndSwapLong
1398	>	(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1399	>	((c - TC_UNIT) & TC_MASK) |
1400	>	(c & ~(AC_MASK|TC_MASK)))));
1401	>
1402	>	if (!tryTerminate(false, false) && w != null) {
1403	>	w.cancelAll();                  // cancel remaining tasks
1404	>	if (w.array != null)            // suppress signal if never ran
1405	>	signalWork(null, 1);        // wake up or create replacement
1406	>	if (ex == null)                 // help clean refs on way out
1407	>	ForkJoinTask.helpExpungeStaleExceptions();
1408	>	}
1409	>
1410	>	if (ex != null)                     // rethrow
1411	>	ForkJoinTask.rethrow(ex);
1412	>	}
1413
1414		/**
1415	<	* Pool number, just for assigning useful names to worker threads
1415	>	* Collect worker steal count into total. Called on termination
1416	>	* and upon int overflow of local count. (There is a possible race
1417	>	* in the latter case vs any caller of getStealCount, which can
1418	>	* make its results less accurate than usual.)
1419		*/
1420	<	private final int poolNumber;
1420	>	final void collectStealCount(WorkQueue w) {
1421	>	if (w != null) {
1422	>	long sc;
1423	>	int ns = w.nsteals;
1424	>	w.nsteals = 0; // handle overflow
1425	>	long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
1426	>	do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1427	>	sc = stealCount, sc + steals));
1428	>	}
1429	>	}
1430	>
1431	>	// Submissions
1432	>
1433	>	/**
1434	>	* Unless shutting down, adds the given task to a submission queue
1435	>	* at submitter's current queue index (modulo submission
1436	>	* range). Only the most common path is directly handled in this
1437	>	* method. All others are relayed to fullExternalPush.
1438	>	*
1439	>	* @param task the task. Caller must ensure non-null.
1440	>	*/
1441	>	final void externalPush(ForkJoinTask<?> task) {
1442	>	WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask<?>[] a;
1443	>	if ((z = submitters.get()) != null && plock > 0 &&
1444	>	(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1445	>	(q = ws[m & z.seed & SQMASK]) != null &&
1446	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1447	>	int s = q.top, n;
1448	>	if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
1449	>	U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
1450	>	task);
1451	>	q.top = s + 1;                     // push on to deque
1452	>	q.qlock = 0;
1453	>	if (n <= 1)
1454	>	signalWork(q, 1);
1455	>	return;
1456	>	}
1457	>	q.qlock = 0;
1458	>	}
1459	>	fullExternalPush(task);
1460	>	}
1461	>
1462	>	/**
1463	>	* Full version of externalPush. This method is called, among
1464	>	* other times, upon the first submission of the first task to the
1465	>	* pool, so must perform secondary initialization: creating
1466	>	* workQueue array and setting plock to a valid value. It also
1467	>	* detects first submission by an external thread by looking up
1468	>	* its ThreadLocal, and creates a new shared queue if the one at
1469	>	* index if empty or contended. The lock bodies must be
1470	>	* exception-free (so no try/finally) so we optimistically
1471	>	* allocate new queues/arrays outside the locks and throw them
1472	>	* away if (very rarely) not needed. Note that the plock seq value
1473	>	* can eventually wrap around zero, but if so harmlessly fails to
1474	>	* reinitialize.
1475	>	*/
1476	>	private void fullExternalPush(ForkJoinTask<?> task) {
1477	>	for (Submitter z = null;;) {
1478	>	WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
1479	>	if ((ps = plock) < 0)
1480	>	throw new RejectedExecutionException();
1481	>	else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) {
1482	>	int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2;
1483	>	n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
1484	>	WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
1485	>	if ((ps & PL_LOCK) != 0 ||
1486	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1487	>	ps = acquirePlock();
1488	>	if ((ws = workQueues) == null)
1489	>	workQueues = nws;
1490	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1491	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1492	>	releasePlock(nps);
1493	>	}
1494	>	else if (z == null && (z = submitters.get()) == null) {
1495	>	if (U.compareAndSwapInt(this, INDEXSEED,
1496	>	s = indexSeed, s += SEED_INCREMENT) &&
1497	>	s != 0) // skip 0
1498	>	submitters.set(z = new Submitter(s));
1499	>	}
1500	>	else {
1501	>	int k = (r = z.seed) & m & SQMASK;
1502	>	if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
1503	>	(q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
1504	>	if (((ps = plock) & PL_LOCK) != 0 ||
1505	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1506	>	ps = acquirePlock();
1507	>	WorkQueue w = null;
1508	>	if ((ws = workQueues) != null && k < ws.length &&
1509	>	(w = ws[k]) == null)
1510	>	ws[k] = q;
1511	>	else
1512	>	q = w;
1513	>	int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1514	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1515	>	releasePlock(nps);
1516	>	}
1517	>	if (q != null && q.qlock == 0 && q.fullPush(task, false))
1518	>	return;
1519	>	r ^= r << 13;                // same xorshift as WorkQueues
1520	>	r ^= r >>> 17;
1521	>	z.seed = r ^= r << 5;        // move to a different index
1522	>	}
1523	>	}
1524	>	}
1525	>
1526	>	// Maintaining ctl counts
1527
1528		/**
1529	<	* The maximum allowed pool size
1529	>	* Increments active count; mainly called upon return from blocking.
1530		*/
1531	<	private volatile int maxPoolSize;
1531	>	final void incrementActiveCount() {
1532	>	long c;
1533	>	do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1534	>	}
1535
1536		/**
1537	<	* The desired parallelism level, updated only under workerLock.
1538	<	*/
1539	<	private volatile int parallelism;
1537	>	* Tries to create (at most one) or activate (possibly several)
1538	>	* workers if too few are active. On contention failure, continues
1539	>	* until at least one worker is signalled or the given queue is
1540	>	* empty or all workers are active.
1541	>	*
1542	>	* @param q if non-null, the queue holding tasks to be signalled
1543	>	* @param signals the target number of signals.
1544	>	*/
1545	>	final void signalWork(WorkQueue q, int signals) {
1546	>	long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
1547	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {
1548	>	if ((e = (int)c) > 0) {
1549	>	if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
1550	>	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1551	>	long nc = (((long)(w.nextWait & E_MASK)) |
1552	>	((long)(u + UAC_UNIT) << 32));
1553	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1554	>	w.eventCount = (e + E_SEQ) & E_MASK;
1555	>	if ((p = w.parker) != null)
1556	>	U.unpark(p);
1557	>	if (--signals <= 0)
1558	>	break;
1559	>	}
1560	>	else
1561	>	signals = 1;
1562	>	if ((q != null && q.queueSize() == 0))
1563	>	break;
1564	>	}
1565	>	else
1566	>	break;
1567	>	}
1568	>	else if (e == 0 && (u & SHORT_SIGN) != 0) {
1569	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1570	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1571	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1572	>	ForkJoinWorkerThread wt = null;
1573	>	Throwable ex = null;
1574	>	boolean started = false;
1575	>	try {
1576	>	ForkJoinWorkerThreadFactory fac;
1577	>	if ((fac = factory) != null &&
1578	>	(wt = fac.newThread(this)) != null) {
1579	>	wt.start();
1580	>	started = true;
1581	>	}
1582	>	} catch (Throwable rex) {
1583	>	ex = rex;
1584	>	}
1585	>	if (!started)
1586	>	deregisterWorker(wt, ex); // adjust counts on failure
1587	>	break;
1588	>	}
1589	>	}
1590	>	else
1591	>	break;
1592	>	}
1593	>	}
1594	>
1595	>	// Scanning for tasks
1596
1597		/**
1598	<	* True if use local fifo, not default lifo, for local polling
1598	>	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1599		*/
1600	<	private volatile boolean locallyFifo;
1600	>	final void runWorker(WorkQueue w) {
1601	>	// initialize queue array in this thread
1602	>	w.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY];
1603	>	do { w.runTask(scan(w)); } while (w.qlock >= 0);
1604	>	}
1605
1606		/**
1607	<	* Holds number of total (i.e., created and not yet terminated)
1608	<	* and running (i.e., not blocked on joins or other managed sync)
1609	<	* threads, packed into one int to ensure consistent snapshot when
1610	<	* making decisions about creating and suspending spare
1611	<	* threads. Updated only by CAS.  Note: CASes in
1612	<	* updateRunningCount and preJoin assume that running active count
1613	<	* is in low word, so need to be modified if this changes.
1607	>	* Scans for and, if found, returns one task, else possibly
1608	>	* inactivates the worker. This method operates on single reads of
1609	>	* volatile state and is designed to be re-invoked continuously,
1610	>	* in part because it returns upon detecting inconsistencies,
1611	>	* contention, or state changes that indicate possible success on
1612	>	* re-invocation.
1613	>	*
1614	>	* The scan searches for tasks across a random permutation of
1615	>	* queues (starting at a random index and stepping by a random
1616	>	* relative prime, checking each at least once).  The scan
1617	>	* terminates upon either finding a non-empty queue, or completing
1618	>	* the sweep. If the worker is not inactivated, it takes and
1619	>	* returns a task from this queue. Otherwise, if not activated, it
1620	>	* signals workers (that may include itself) and returns so caller
1621	>	* can retry. Also returns for trtry if the worker array may have
1622	>	* changed during an empty scan.  On failure to find a task, we
1623	>	* take one of the following actions, after which the caller will
1624	>	* retry calling this method unless terminated.
1625	>	*
1626	>	* * If pool is terminating, terminate the worker.
1627	>	*
1628	>	* * If not already enqueued, try to inactivate and enqueue the
1629	>	* worker on wait queue. Or, if inactivating has caused the pool
1630	>	* to be quiescent, relay to idleAwaitWork to check for
1631	>	* termination and possibly shrink pool.
1632	>	*
1633	>	* * If already enqueued and none of the above apply, possibly
1634	>	* (with 1/2 probablility) park awaiting signal, else lingering to
1635	>	* help scan and signal.
1636	>	*
1637	>	* @param w the worker (via its WorkQueue)
1638	>	* @return a task or null if none found
1639	>	*/
1640	>	private final ForkJoinTask<?> scan(WorkQueue w) {
1641	>	WorkQueue[] ws; WorkQueue q;           // first update random seed
1642	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1643	>	int ps = plock, m;                     // volatile read order matters
1644	>	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1645	>	int ec = w.eventCount;             // ec is negative if inactive
1646	>	int step = (r >>> 16) | 1;         // relatively prime
1647	>	for (int j = (m + 1) << 2;  ; --j, r += step) {
1648	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b, n;
1649	>	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1650	>	(a = q.array) != null) {   // probably nonempty
1651	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1652	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1653	>	if (q.base == b && ec >= 0 && t != null &&
1654	>	U.compareAndSwapObject(a, i, t, null)) {
1655	>	if ((n = q.top - (q.base = b + 1)) > 0)
1656	>	signalWork(q, n);
1657	>	return t;              // taken
1658	>	}
1659	>	if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) {
1660	>	if ((n = q.queueSize() - 1) > 0)
1661	>	signalWork(q, n);
1662	>	break;                 // let caller retry after signal
1663	>	}
1664	>	}
1665	>	else if (j < 0) {              // end of scan
1666	>	long c = ctl; int e;
1667	>	if (plock != ps)           // incomplete sweep
1668	>	break;
1669	>	if ((e = (int)c) < 0)      // pool is terminating
1670	>	w.qlock = -1;
1671	>	else if (ec >= 0) {        // try to enqueue/inactivate
1672	>	long nc = ((long)ec |
1673	>	((c - AC_UNIT) & (AC_MASK|TC_MASK)));
1674	>	w.nextWait = e;
1675	>	w.eventCount = ec | INT_SIGN; // mark as inactive
1676	>	if (ctl != c ||
1677	>	!U.compareAndSwapLong(this, CTL, c, nc))
1678	>	w.eventCount = ec; // unmark on CAS failure
1679	>	else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
1680	>	idleAwaitWork(w, nc, c);  // quiescent
1681	>	}
1682	>	else if (w.seed >= 0 && w.eventCount < 0) {
1683	>	Thread wt = Thread.currentThread();
1684	>	Thread.interrupted();  // clear status
1685	>	U.putObject(wt, PARKBLOCKER, this);
1686	>	w.parker = wt;         // emulate LockSupport.park
1687	>	if (w.eventCount < 0)  // recheck
1688	>	U.park(false, 0L);
1689	>	w.parker = null;
1690	>	U.putObject(wt, PARKBLOCKER, null);
1691	>	}
1692	>	break;
1693	>	}
1694	>	}
1695	>	}
1696	>	return null;
1697	>	}
1698	>
1699	>	/**
1700	>	* If inactivating worker w has caused the pool to become
1701	>	* quiescent, checks for pool termination, and, so long as this is
1702	>	* not the only worker, waits for event for up to a given
1703	>	* duration.  On timeout, if ctl has not changed, terminates the
1704	>	* worker, which will in turn wake up another worker to possibly
1705	>	* repeat this process.
1706	>	*
1707	>	* @param w the calling worker
1708	>	* @param currentCtl the ctl value triggering possible quiescence
1709	>	* @param prevCtl the ctl value to restore if thread is terminated
1710	>	*/
1711	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1712	>	if (w.eventCount < 0 &&
1713	>	(this == commonPool || !tryTerminate(false, false)) &&
1714	>	(int)prevCtl != 0) {
1715	>	int dc = -(short)(currentCtl >>> TC_SHIFT);
1716	>	long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1717	>	long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1718	>	Thread wt = Thread.currentThread();
1719	>	while (ctl == currentCtl) {
1720	>	Thread.interrupted();  // timed variant of version in scan()
1721	>	U.putObject(wt, PARKBLOCKER, this);
1722	>	w.parker = wt;
1723	>	if (ctl == currentCtl)
1724	>	U.park(false, parkTime);
1725	>	w.parker = null;
1726	>	U.putObject(wt, PARKBLOCKER, null);
1727	>	if (ctl != currentCtl)
1728	>	break;
1729	>	if (deadline - System.nanoTime() <= 0L &&
1730	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1731	>	w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1732	>	w.qlock = -1;   // shrink
1733	>	break;
1734	>	}
1735	>	}
1736	>	}
1737	>	}
1738	>
1739	>	/**
1740	>	* Scans through queues looking for work while joining a task;
1741	>	* if any are present, signals.
1742	>	*
1743	>	* @param task to return early if done
1744	>	* @param origin an index to start scan
1745		*/
1746	<	private volatile int workerCounts;
1746	>	final int helpSignal(ForkJoinTask<?> task, int origin) {
1747	>	WorkQueue[] ws; WorkQueue q; int m, n, s;
1748	>	if (task != null && (ws = workQueues) != null &&
1749	>	(m = ws.length - 1) >= 0) {
1750	>	for (int i = 0; i <= m; ++i) {
1751	>	if ((s = task.status) < 0)
1752	>	return s;
1753	>	if ((q = ws[(i + origin) & m]) != null &&
1754	>	(n = q.queueSize()) > 0) {
1755	>	signalWork(q, n);
1756	>	if ((int)(ctl >> AC_SHIFT) >= 0)
1757	>	break;
1758	>	}
1759	>	}
1760	>	}
1761	>	return 0;
1762	>	}
1763
1764	<	private static int totalCountOf(int s)           { return s >>> 16;  }
1765	<	private static int runningCountOf(int s)         { return s & shortMask; }
1766	<	private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
1764	>	/**
1765	>	* Tries to locate and execute tasks for a stealer of the given
1766	>	* task, or in turn one of its stealers, Traces currentSteal ->
1767	>	* currentJoin links looking for a thread working on a descendant
1768	>	* of the given task and with a non-empty queue to steal back and
1769	>	* execute tasks from. The first call to this method upon a
1770	>	* waiting join will often entail scanning/search, (which is OK
1771	>	* because the joiner has nothing better to do), but this method
1772	>	* leaves hints in workers to speed up subsequent calls. The
1773	>	* implementation is very branchy to cope with potential
1774	>	* inconsistencies or loops encountering chains that are stale,
1775	>	* unknown, or so long that they are likely cyclic.
1776	>	*
1777	>	* @param joiner the joining worker
1778	>	* @param task the task to join
1779	>	* @return 0 if no progress can be made, negative if task
1780	>	* known complete, else positive
1781	>	*/
1782	>	private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1783	>	int stat = 0, steps = 0;                    // bound to avoid cycles
1784	>	if (joiner != null && task != null) {       // hoist null checks
1785	>	restart: for (;;) {
1786	>	ForkJoinTask<?> subtask = task;     // current target
1787	>	for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
1788	>	WorkQueue[] ws; int m, s, h;
1789	>	if ((s = task.status) < 0) {
1790	>	stat = s;
1791	>	break restart;
1792	>	}
1793	>	if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1794	>	break restart;              // shutting down
1795	>	if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
1796	>	v.currentSteal != subtask) {
1797	>	for (int origin = h;;) {    // find stealer
1798	>	if (((h = (h + 2) & m) & 15) == 1 &&
1799	>	(subtask.status < 0 || j.currentJoin != subtask))
1800	>	continue restart;   // occasional staleness check
1801	>	if ((v = ws[h]) != null &&
1802	>	v.currentSteal == subtask) {
1803	>	j.stealHint = h;    // save hint
1804	>	break;
1805	>	}
1806	>	if (h == origin)
1807	>	break restart;      // cannot find stealer
1808	>	}
1809	>	}
1810	>	for (;;) { // help stealer or descend to its stealer
1811	>	ForkJoinTask[] a;  int b;
1812	>	if (subtask.status < 0)     // surround probes with
1813	>	continue restart;       //   consistency checks
1814	>	if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1815	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1816	>	ForkJoinTask<?> t =
1817	>	(ForkJoinTask<?>)U.getObjectVolatile(a, i);
1818	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1819	>	v.currentSteal != subtask)
1820	>	continue restart;   // stale
1821	>	stat = 1;               // apparent progress
1822	>	if (t != null && v.base == b &&
1823	>	U.compareAndSwapObject(a, i, t, null)) {
1824	>	v.base = b + 1;     // help stealer
1825	>	joiner.runSubtask(t);
1826	>	}
1827	>	else if (v.base == b && ++steps == MAX_HELP)
1828	>	break restart;      // v apparently stalled
1829	>	}
1830	>	else {                      // empty -- try to descend
1831	>	ForkJoinTask<?> next = v.currentJoin;
1832	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1833	>	v.currentSteal != subtask)
1834	>	continue restart;   // stale
1835	>	else if (next == null || ++steps == MAX_HELP)
1836	>	break restart;      // dead-end or maybe cyclic
1837	>	else {
1838	>	subtask = next;
1839	>	j = v;
1840	>	break;
1841	>	}
1842	>	}
1843	>	}
1844	>	}
1845	>	}
1846	>	}
1847	>	return stat;
1848	>	}
1849
1850		/**
1851	<	* Adds delta (which may be negative) to running count.  This must
1852	<	* be called before (with negative arg) and after (with positive)
1853	<	* any managed synchronization (i.e., mainly, joins).
1851	>	* Analog of tryHelpStealer for CountedCompleters. Tries to steal
1852	>	* and run tasks within the target's computation
1853	>	*
1854	>	* @param task the task to join
1855	>	* @param mode if shared, exit upon completing any task
1856	>	* if all workers are active
1857		*
236	–	* @param delta the number to add
1858		*/
1859	<	final void updateRunningCount(int delta) {
1860	<	int s;
1861	<	do {} while (!casWorkerCounts(s = workerCounts, s + delta));
1859	>	private int helpComplete(ForkJoinTask<?> task, int mode) {
1860	>	WorkQueue[] ws; WorkQueue q; int m, n, s;
1861	>	if (task != null && (ws = workQueues) != null &&
1862	>	(m = ws.length - 1) >= 0) {
1863	>	for (int j = 1, origin = j;;) {
1864	>	if ((s = task.status) < 0)
1865	>	return s;
1866	>	if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1867	>	origin = j;
1868	>	if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
1869	>	break;
1870	>	}
1871	>	else if ((j = (j + 2) & m) == origin)
1872	>	break;
1873	>	}
1874	>	}
1875	>	return 0;
1876		}
1877
1878		/**
1879	<	* Adds delta (which may be negative) to both total and running
1880	<	* count.  This must be called upon creation and termination of
1881	<	* worker threads.
1879	>	* Tries to decrement active count (sometimes implicitly) and
1880	>	* possibly release or create a compensating worker in preparation
1881	>	* for blocking. Fails on contention or termination. Otherwise,
1882	>	* adds a new thread if no idle workers are available and pool
1883	>	* may become starved.
1884	>	*/
1885	>	final boolean tryCompensate() {
1886	>	int pc = parallelism, e, u, i, tc; long c;
1887	>	WorkQueue[] ws; WorkQueue w; Thread p;
1888	>	if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
1889	>	if (e != 0 && (i = e & SMASK) < ws.length &&
1890	>	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1891	>	long nc = ((long)(w.nextWait & E_MASK) |
1892	>	(c & (AC_MASK|TC_MASK)));
1893	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1894	>	w.eventCount = (e + E_SEQ) & E_MASK;
1895	>	if ((p = w.parker) != null)
1896	>	U.unpark(p);
1897	>	return true;   // replace with idle worker
1898	>	}
1899	>	}
1900	>	else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
1901	>	(u >> UAC_SHIFT) + pc > 1) {
1902	>	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1903	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1904	>	return true;    // no compensation
1905	>	}
1906	>	else if ((tc = u + pc) < MAX_CAP) {
1907	>	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1908	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1909	>	Throwable ex = null;
1910	>	ForkJoinWorkerThread wt = null;
1911	>	try {
1912	>	ForkJoinWorkerThreadFactory fac;
1913	>	if ((fac = factory) != null &&
1914	>	(wt = fac.newThread(this)) != null) {
1915	>	wt.start();
1916	>	return true;
1917	>	}
1918	>	} catch (Throwable rex) {
1919	>	ex = rex;
1920	>	}
1921	>	deregisterWorker(wt, ex); // adjust counts etc
1922	>	}
1923	>	}
1924	>	}
1925	>	return false;
1926	>	}
1927	>
1928	>	/**
1929	>	* Helps and/or blocks until the given task is done.
1930		*
1931	<	* @param delta the number to add
1931	>	* @param joiner the joining worker
1932	>	* @param task the task
1933	>	* @return task status on exit
1934		*/
1935	<	private void updateWorkerCount(int delta) {
1936	<	int d = delta + (delta << 16); // add to both lo and hi parts
1937	<	int s;
1938	<	do {} while (!casWorkerCounts(s = workerCounts, s + d));
1935	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1936	>	int s = 0;
1937	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1938	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1939	>	joiner.currentJoin = task;
1940	>	do {} while ((s = task.status) >= 0 &&
1941	>	joiner.queueSize() > 0 &&
1942	>	joiner.tryRemoveAndExec(task)); // process local tasks
1943	>	if (s >= 0 && (s = task.status) >= 0 &&
1944	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1945	>	(task instanceof CountedCompleter))
1946	>	s = helpComplete(task, LIFO_QUEUE);
1947	>	while (s >= 0 && (s = task.status) >= 0) {
1948	>	if ((joiner.queueSize() > 0 ||           // try helping
1949	>	(s = tryHelpStealer(joiner, task)) == 0) &&
1950	>	(s = task.status) >= 0 && tryCompensate()) {
1951	>	if (task.trySetSignal() && (s = task.status) >= 0) {
1952	>	synchronized (task) {
1953	>	if (task.status >= 0) {
1954	>	try {                // see ForkJoinTask
1955	>	task.wait();     //  for explanation
1956	>	} catch (InterruptedException ie) {
1957	>	}
1958	>	}
1959	>	else
1960	>	task.notifyAll();
1961	>	}
1962	>	}
1963	>	long c;                          // re-activate
1964	>	do {} while (!U.compareAndSwapLong
1965	>	(this, CTL, c = ctl, c + AC_UNIT));
1966	>	}
1967	>	}
1968	>	joiner.currentJoin = prevJoin;
1969	>	}
1970	>	return s;
1971		}
1972
1973		/**
1974	<	* Lifecycle control. High word contains runState, low word
1975	<	* contains the number of workers that are (probably) executing
1976	<	* tasks. This value is atomically incremented before a worker
1977	<	* gets a task to run, and decremented when worker has no tasks
1978	<	* and cannot find any. These two fields are bundled together to
1979	<	* support correct termination triggering.  Note: activeCount
263	<	* CAS'es cheat by assuming active count is in low word, so need
264	<	* to be modified if this changes
1974	>	* Stripped-down variant of awaitJoin used by timed joins. Tries
1975	>	* to help join only while there is continuous progress. (Caller
1976	>	* will then enter a timed wait.)
1977	>	*
1978	>	* @param joiner the joining worker
1979	>	* @param task the task
1980		*/
1981	<	private volatile int runControl;
1981	>	final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
1982	>	int s;
1983	>	if (joiner != null && task != null && (s = task.status) >= 0) {
1984	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1985	>	joiner.currentJoin = task;
1986	>	do {} while ((s = task.status) >= 0 &&
1987	>	joiner.queueSize() > 0 &&
1988	>	joiner.tryRemoveAndExec(task));
1989	>	if (s >= 0 && (s = task.status) >= 0 &&
1990	>	(s = helpSignal(task, joiner.poolIndex)) >= 0 &&
1991	>	(task instanceof CountedCompleter))
1992	>	s = helpComplete(task, LIFO_QUEUE);
1993	>	if (s >= 0 && joiner.queueSize() == 0) {
1994	>	do {} while (task.status >= 0 &&
1995	>	tryHelpStealer(joiner, task) > 0);
1996	>	}
1997	>	joiner.currentJoin = prevJoin;
1998	>	}
1999	>	}
2000
2001	<	// RunState values. Order among values matters
2002	<	private static final int RUNNING     = 0;
2003	<	private static final int SHUTDOWN    = 1;
2004	<	private static final int TERMINATING = 2;
2005	<	private static final int TERMINATED  = 3;
2001	>	/**
2002	>	* Returns a (probably) non-empty steal queue, if one is found
2003	>	* during a random, then cyclic scan, else null.  This method must
2004	>	* be retried by caller if, by the time it tries to use the queue,
2005	>	* it is empty.
2006	>	* @param r a (random) seed for scanning
2007	>	*/
2008	>	private WorkQueue findNonEmptyStealQueue(int r) {
2009	>	int step = (r >>> 16) | 1;
2010	>	for (WorkQueue[] ws;;) {
2011	>	int ps = plock, m;
2012	>	if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2013	>	return null;
2014	>	for (int j = (m + 1) << 2; ; r += step) {
2015	>	WorkQueue q = ws[((r << 1) | 1) & m];
2016	>	if (q != null && q.queueSize() > 0)
2017	>	return q;
2018	>	else if (--j < 0) {
2019	>	if (plock == ps)
2020	>	return null;
2021	>	break;
2022	>	}
2023	>	}
2024	>	}
2025	>	}
2026
2027	<	private static int runStateOf(int c)             { return c >>> 16; }
2028	<	private static int activeCountOf(int c)          { return c & shortMask; }
2029	<	private static int runControlFor(int r, int a)   { return (r << 16) + a; }
2027	>	/**
2028	>	* Runs tasks until {@code isQuiescent()}. We piggyback on
2029	>	* active count ctl maintenance, but rather than blocking
2030	>	* when tasks cannot be found, we rescan until all others cannot
2031	>	* find tasks either.
2032	>	*/
2033	>	final void helpQuiescePool(WorkQueue w) {
2034	>	for (boolean active = true;;) {
2035	>	ForkJoinTask<?> localTask; // exhaust local queue
2036	>	while ((localTask = w.nextLocalTask()) != null)
2037	>	localTask.doExec();
2038	>	// Similar to loop in scan(), but ignoring submissions
2039	>	WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2040	>	if (q != null) {
2041	>	ForkJoinTask<?> t; int b;
2042	>	if (!active) {      // re-establish active count
2043	>	long c;
2044	>	active = true;
2045	>	do {} while (!U.compareAndSwapLong
2046	>	(this, CTL, c = ctl, c + AC_UNIT));
2047	>	}
2048	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2049	>	w.runSubtask(t);
2050	>	}
2051	>	else {
2052	>	long c;
2053	>	if (active) {       // decrement active count without queuing
2054	>	active = false;
2055	>	do {} while (!U.compareAndSwapLong
2056	>	(this, CTL, c = ctl, c -= AC_UNIT));
2057	>	}
2058	>	else
2059	>	c = ctl;        // re-increment on exit
2060	>	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2061	>	do {} while (!U.compareAndSwapLong
2062	>	(this, CTL, c = ctl, c + AC_UNIT));
2063	>	break;
2064	>	}
2065	>	}
2066	>	}
2067	>	}
2068
2069		/**
2070	<	* Tries incrementing active count; fails on contention.
280	<	* Called by workers before/during executing tasks.
2070	>	* Gets and removes a local or stolen task for the given worker.
2071		*
2072	<	* @return true on success
2072	>	* @return a task, if available
2073		*/
2074	<	final boolean tryIncrementActiveCount() {
2075	<	int c = runControl;
2076	<	return casRunControl(c, c+1);
2074	>	final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
2075	>	for (ForkJoinTask<?> t;;) {
2076	>	WorkQueue q; int b;
2077	>	if ((t = w.nextLocalTask()) != null)
2078	>	return t;
2079	>	if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2080	>	return null;
2081	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2082	>	return t;
2083	>	}
2084		}
2085
2086		/**
2087	<	* Tries decrementing active count; fails on contention.
2088	<	* Possibly triggers termination on success.
2089	<	* Called by workers when they can't find tasks.
2090	<	*
2091	<	* @return true on success
2087	>	* Returns a cheap heuristic guide for task partitioning when
2088	>	* programmers, frameworks, tools, or languages have little or no
2089	>	* idea about task granularity.  In essence by offering this
2090	>	* method, we ask users only about tradeoffs in overhead vs
2091	>	* expected throughput and its variance, rather than how finely to
2092	>	* partition tasks.
2093	>	*
2094	>	* In a steady state strict (tree-structured) computation, each
2095	>	* thread makes available for stealing enough tasks for other
2096	>	* threads to remain active. Inductively, if all threads play by
2097	>	* the same rules, each thread should make available only a
2098	>	* constant number of tasks.
2099	>	*
2100	>	* The minimum useful constant is just 1. But using a value of 1
2101	>	* would require immediate replenishment upon each steal to
2102	>	* maintain enough tasks, which is infeasible.  Further,
2103	>	* partitionings/granularities of offered tasks should minimize
2104	>	* steal rates, which in general means that threads nearer the top
2105	>	* of computation tree should generate more than those nearer the
2106	>	* bottom. In perfect steady state, each thread is at
2107	>	* approximately the same level of computation tree. However,
2108	>	* producing extra tasks amortizes the uncertainty of progress and
2109	>	* diffusion assumptions.
2110	>	*
2111	>	* So, users will want to use values larger, but not much larger
2112	>	* than 1 to both smooth over transient shortages and hedge
2113	>	* against uneven progress; as traded off against the cost of
2114	>	* extra task overhead. We leave the user to pick a threshold
2115	>	* value to compare with the results of this call to guide
2116	>	* decisions, but recommend values such as 3.
2117	>	*
2118	>	* When all threads are active, it is on average OK to estimate
2119	>	* surplus strictly locally. In steady-state, if one thread is
2120	>	* maintaining say 2 surplus tasks, then so are others. So we can
2121	>	* just use estimated queue length.  However, this strategy alone
2122	>	* leads to serious mis-estimates in some non-steady-state
2123	>	* conditions (ramp-up, ramp-down, other stalls). We can detect
2124	>	* many of these by further considering the number of "idle"
2125	>	* threads, that are known to have zero queued tasks, so
2126	>	* compensate by a factor of (#idle/#active) threads.
2127	>	*
2128	>	* Note: The approximation of #busy workers as #active workers is
2129	>	* not very good under current signalling scheme, and should be
2130	>	* improved.
2131	>	*/
2132	>	static int getSurplusQueuedTaskCount() {
2133	>	Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2134	>	if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2135	>	int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
2136	>	int p = (pool = wt.pool).parallelism;
2137	>	int a = (int)(pool.ctl >> AC_SHIFT) + p;
2138	>	return q.top - b - (a > (p >>>= 1) ? 0 :
2139	>	a > (p >>>= 1) ? 1 :
2140	>	a > (p >>>= 1) ? 2 :
2141	>	a > (p >>>= 1) ? 4 :
2142	>	8);
2143	>	}
2144	>	return 0;
2145	>	}
2146	>
2147	>	//  Termination
2148	>
2149	>	/**
2150	>	* Possibly initiates and/or completes termination.  The caller
2151	>	* triggering termination runs three passes through workQueues:
2152	>	* (0) Setting termination status, followed by wakeups of queued
2153	>	* workers; (1) cancelling all tasks; (2) interrupting lagging
2154	>	* threads (likely in external tasks, but possibly also blocked in
2155	>	* joins).  Each pass repeats previous steps because of potential
2156	>	* lagging thread creation.
2157	>	*
2158	>	* @param now if true, unconditionally terminate, else only
2159	>	* if no work and no active workers
2160	>	* @param enable if true, enable shutdown when next possible
2161	>	* @return true if now terminating or terminated
2162		*/
2163	<	final boolean tryDecrementActiveCount() {
2164	<	int c = runControl;
298	<	int nextc = c - 1;
299	<	if (!casRunControl(c, nextc))
2163	>	private boolean tryTerminate(boolean now, boolean enable) {
2164	>	if (this == commonPool)                     // cannot shut down
2165		return false;
2166	<	if (canTerminateOnShutdown(nextc))
2167	<	terminateOnShutdown();
2168	<	return true;
2166	>	for (long c;;) {
2167	>	if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2168	>	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2169	>	synchronized (this) {
2170	>	notifyAll();                // signal when 0 workers
2171	>	}
2172	>	}
2173	>	return true;
2174	>	}
2175	>	if (plock >= 0) {                       // not yet enabled
2176	>	int ps;
2177	>	if (!enable)
2178	>	return false;
2179	>	if (((ps = plock) & PL_LOCK) != 0 ||
2180	>	!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2181	>	ps = acquirePlock();
2182	>	int nps = SHUTDOWN;
2183	>	if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2184	>	releasePlock(nps);
2185	>	}
2186	>	if (!now) {                             // check if idle & no tasks
2187	>	if ((int)(c >> AC_SHIFT) != -parallelism ||
2188	>	hasQueuedSubmissions())
2189	>	return false;
2190	>	// Check for unqueued inactive workers. One pass suffices.
2191	>	WorkQueue[] ws = workQueues; WorkQueue w;
2192	>	if (ws != null) {
2193	>	for (int i = 1; i < ws.length; i += 2) {
2194	>	if ((w = ws[i]) != null && w.eventCount >= 0)
2195	>	return false;
2196	>	}
2197	>	}
2198	>	}
2199	>	if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2200	>	for (int pass = 0; pass < 3; ++pass) {
2201	>	WorkQueue[] ws = workQueues;
2202	>	if (ws != null) {
2203	>	WorkQueue w;
2204	>	int n = ws.length;
2205	>	for (int i = 0; i < n; ++i) {
2206	>	if ((w = ws[i]) != null) {
2207	>	w.qlock = -1;
2208	>	if (pass > 0) {
2209	>	w.cancelAll();
2210	>	if (pass > 1)
2211	>	w.interruptOwner();
2212	>	}
2213	>	}
2214	>	}
2215	>	// Wake up workers parked on event queue
2216	>	int i, e; long cc; Thread p;
2217	>	while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2218	>	(i = e & SMASK) < n &&
2219	>	(w = ws[i]) != null) {
2220	>	long nc = ((long)(w.nextWait & E_MASK) |
2221	>	((cc + AC_UNIT) & AC_MASK) |
2222	>	(cc & (TC_MASK|STOP_BIT)));
2223	>	if (w.eventCount == (e | INT_SIGN) &&
2224	>	U.compareAndSwapLong(this, CTL, cc, nc)) {
2225	>	w.eventCount = (e + E_SEQ) & E_MASK;
2226	>	w.qlock = -1;
2227	>	if ((p = w.parker) != null)
2228	>	U.unpark(p);
2229	>	}
2230	>	}
2231	>	}
2232	>	}
2233	>	}
2234	>	}
2235		}
2236
2237	+	// external operations on common pool
2238	+
2239		/**
2240	<	* Returns true if argument represents zero active count and
2241	<	* nonzero runstate, which is the triggering condition for
309	<	* terminating on shutdown.
2240	>	* Returns common pool queue for a thread that has submitted at
2241	>	* least one task.
2242		*/
2243	<	private static boolean canTerminateOnShutdown(int c) {
2244	<	// i.e. least bit is nonzero runState bit
2245	<	return ((c & -c) >>> 16) != 0;
2243	>	static WorkQueue commonSubmitterQueue() {
2244	>	ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2245	>	return ((z = submitters.get()) != null &&
2246	>	(p = commonPool) != null &&
2247	>	(ws = p.workQueues) != null &&
2248	>	(m = ws.length - 1) >= 0) ?
2249	>	ws[m & z.seed & SQMASK] : null;
2250		}
2251
2252		/**
2253	<	* Transition run state to at least the given state. Return true
2254	<	* if not already at least given state.
2255	<	*/
2256	<	private boolean transitionRunStateTo(int state) {
2257	<	for (;;) {
2258	<	int c = runControl;
2259	<	if (runStateOf(c) >= state)
2260	<	return false;
2261	<	if (casRunControl(c, runControlFor(state, activeCountOf(c))))
2253	>	* Tries to pop the given task from submitter's queue in common pool.
2254	>	*/
2255	>	static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2256	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2257	>	ForkJoinTask<?>[] a;  int m, s; long j;
2258	>	if ((z = submitters.get()) != null &&
2259	>	(p = commonPool) != null &&
2260	>	(ws = p.workQueues) != null &&
2261	>	(m = ws.length - 1) >= 0 &&
2262	>	(q = ws[m & z.seed & SQMASK]) != null &&
2263	>	(s = q.top) != q.base &&
2264	>	(a = q.array) != null &&
2265	>	U.getObjectVolatile
2266	>	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2267	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2268	>	if (q.array == a && q.top == s && // recheck
2269	>	U.compareAndSwapObject(a, j, t, null)) {
2270	>	q.top = s - 1;
2271	>	q.qlock = 0;
2272		return true;
2273	+	}
2274	+	q.qlock = 0;
2275	+	}
2276	+	return false;
2277	+	}
2278	+
2279	+	/**
2280	+	* Tries to pop and run local tasks within the same computation
2281	+	* as the given root. On failure, tries to help complete from
2282	+	* other queues via helpComplete.
2283	+	*/
2284	+	private void externalHelpComplete(WorkQueue q, ForkJoinTask<?> root) {
2285	+	ForkJoinTask<?>[] a; int m;
2286	+	if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2287	+	root != null && root.status >= 0) {
2288	+	for (;;) {
2289	+	int s; Object o; CountedCompleter<?> task = null;
2290	+	if ((s = q.top) - q.base > 0) {
2291	+	long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2292	+	if ((o = U.getObject(a, j)) != null &&
2293	+	(o instanceof CountedCompleter)) {
2294	+	CountedCompleter<?> t = (CountedCompleter<?>)o, r = t;
2295	+	do {
2296	+	if (r == root) {
2297	+	if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2298	+	if (q.array == a && q.top == s &&
2299	+	U.compareAndSwapObject(a, j, t, null)) {
2300	+	q.top = s - 1;
2301	+	task = t;
2302	+	}
2303	+	q.qlock = 0;
2304	+	}
2305	+	break;
2306	+	}
2307	+	} while ((r = r.completer) != null);
2308	+	}
2309	+	}
2310	+	if (task != null)
2311	+	task.doExec();
2312	+	if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0)
2313	+	break;
2314	+	if (task == null) {
2315	+	if (helpSignal(root, q.poolIndex) >= 0)
2316	+	helpComplete(root, SHARED_QUEUE);
2317	+	break;
2318	+	}
2319	+	}
2320		}
2321		}
2322
2323		/**
2324	<	* Controls whether to add spares to maintain parallelism
2324	>	* Tries to help execute or signal availability of the given task
2325	>	* from submitter's queue in common pool.
2326		*/
2327	<	private volatile boolean maintainsParallelism;
2327	>	static void externalHelpJoin(ForkJoinTask<?> t) {
2328	>	// Some hard-to-avoid overlap with tryExternalUnpush
2329	>	ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2330	>	ForkJoinTask<?>[] a;  int m, s, n; long j;
2331	>	if (t != null && t.status >= 0 &&
2332	>	(z = submitters.get()) != null &&
2333	>	(p = commonPool) != null &&
2334	>	(ws = p.workQueues) != null &&
2335	>	(m = ws.length - 1) >= 0 &&
2336	>	(q = ws[m & z.seed & SQMASK]) != null &&
2337	>	(a = q.array) != null) {
2338	>	if ((s = q.top) != q.base &&
2339	>	U.getObjectVolatile
2340	>	(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2341	>	U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2342	>	if (q.array == a && q.top == s &&
2343	>	U.compareAndSwapObject(a, j, t, null)) {
2344	>	q.top = s - 1;
2345	>	q.qlock = 0;
2346	>	t.doExec();
2347	>	}
2348	>	else
2349	>	q.qlock = 0;
2350	>	}
2351	>	if (t.status >= 0) {
2352	>	if (t instanceof CountedCompleter)
2353	>	p.externalHelpComplete(q, t);
2354	>	else
2355	>	p.helpSignal(t, q.poolIndex);
2356	>	}
2357	>	}
2358	>	}
2359	>
2360	>	/**
2361	>	* Restricted version of helpQuiescePool for external callers
2362	>	*/
2363	>	static void externalHelpQuiescePool() {
2364	>	ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2365	>	int r = ThreadLocalRandom.current().nextInt();
2366	>	if ((p = commonPool) != null &&
2367	>	(q = p.findNonEmptyStealQueue(r)) != null &&
2368	>	(b = q.base) - q.top < 0 &&
2369	>	(t = q.pollAt(b)) != null)
2370	>	t.doExec();
2371	>	}
2372	>
2373	>	// Exported methods
2374
2375		// Constructors
2376
2377		/**
2378	<	* Creates a ForkJoinPool with a pool size equal to the number of
2379	<	* processors available on the system, using the default
2380	<	* ForkJoinWorkerThreadFactory.
2378	>	* Creates a {@code ForkJoinPool} with parallelism equal to {@link
2379	>	* java.lang.Runtime#availableProcessors}, using the {@linkplain
2380	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
2381	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
2382		*
2383		* @throws SecurityException if a security manager exists and
2384		*         the caller is not permitted to modify threads
#	Line 346 | Line 2387 | public class ForkJoinPool extends Abstra
2387		*/
2388		public ForkJoinPool() {
2389		this(Runtime.getRuntime().availableProcessors(),
2390	<	defaultForkJoinWorkerThreadFactory);
2390	>	defaultForkJoinWorkerThreadFactory, null, false);
2391		}
2392
2393		/**
2394	<	* Creates a ForkJoinPool with the indicated parallelism level
2395	<	* threads and using the default ForkJoinWorkerThreadFactory.
2394	>	* Creates a {@code ForkJoinPool} with the indicated parallelism
2395	>	* level, the {@linkplain
2396	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
2397	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
2398		*
2399	<	* @param parallelism the number of worker threads
2399	>	* @param parallelism the parallelism level
2400		* @throws IllegalArgumentException if parallelism less than or
2401	<	* equal to zero
2401	>	*         equal to zero, or greater than implementation limit
2402		* @throws SecurityException if a security manager exists and
2403		*         the caller is not permitted to modify threads
2404		*         because it does not hold {@link
2405		*         java.lang.RuntimePermission}{@code ("modifyThread")}
2406		*/
2407		public ForkJoinPool(int parallelism) {
2408	<	this(parallelism, defaultForkJoinWorkerThreadFactory);
2408	>	this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
2409		}
2410
2411		/**
2412	<	* Creates a ForkJoinPool with parallelism equal to the number of
370	<	* processors available on the system and using the given
371	<	* ForkJoinWorkerThreadFactory.
2412	>	* Creates a {@code ForkJoinPool} with the given parameters.
2413		*
2414	<	* @param factory the factory for creating new threads
2415	<	* @throws NullPointerException if factory is null
2416	<	* @throws SecurityException if a security manager exists and
2417	<	*         the caller is not permitted to modify threads
2418	<	*         because it does not hold {@link
2419	<	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2420	<	*/
2421	<	public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
2422	<	this(Runtime.getRuntime().availableProcessors(), factory);
2423	<	}
2424	<
2425	<	/**
2426	<	* Creates a ForkJoinPool with the given parallelism and factory.
386	<	*
387	<	* @param parallelism the targeted number of worker threads
388	<	* @param factory the factory for creating new threads
2414	>	* @param parallelism the parallelism level. For default value,
2415	>	* use {@link java.lang.Runtime#availableProcessors}.
2416	>	* @param factory the factory for creating new threads. For default value,
2417	>	* use {@link #defaultForkJoinWorkerThreadFactory}.
2418	>	* @param handler the handler for internal worker threads that
2419	>	* terminate due to unrecoverable errors encountered while executing
2420	>	* tasks. For default value, use {@code null}.
2421	>	* @param asyncMode if true,
2422	>	* establishes local first-in-first-out scheduling mode for forked
2423	>	* tasks that are never joined. This mode may be more appropriate
2424	>	* than default locally stack-based mode in applications in which
2425	>	* worker threads only process event-style asynchronous tasks.
2426	>	* For default value, use {@code false}.
2427		* @throws IllegalArgumentException if parallelism less than or
2428	<	* equal to zero, or greater than implementation limit
2429	<	* @throws NullPointerException if factory is null
2428	>	*         equal to zero, or greater than implementation limit
2429	>	* @throws NullPointerException if the factory is null
2430		* @throws SecurityException if a security manager exists and
2431		*         the caller is not permitted to modify threads
2432		*         because it does not hold {@link
2433		*         java.lang.RuntimePermission}{@code ("modifyThread")}
2434		*/
2435	<	public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
2436	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
2437	<	throw new IllegalArgumentException();
2435	>	public ForkJoinPool(int parallelism,
2436	>	ForkJoinWorkerThreadFactory factory,
2437	>	Thread.UncaughtExceptionHandler handler,
2438	>	boolean asyncMode) {
2439	>	checkPermission();
2440		if (factory == null)
2441		throw new NullPointerException();
2442	<	checkPermission();
2443	<	this.factory = factory;
2442	>	if (parallelism <= 0 || parallelism > MAX_CAP)
2443	>	throw new IllegalArgumentException();
2444		this.parallelism = parallelism;
2445	<	this.maxPoolSize = MAX_THREADS;
2446	<	this.maintainsParallelism = true;
2447	<	this.poolNumber = poolNumberGenerator.incrementAndGet();
2448	<	this.workerLock = new ReentrantLock();
2449	<	this.termination = workerLock.newCondition();
2450	<	this.stealCount = new AtomicLong();
2451	<	this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
2452	<	// worker array and workers are lazily constructed
2453	<	}
2454	<
415	<	/**
416	<	* Creates a new worker thread using factory.
417	<	*
418	<	* @param index the index to assign worker
419	<	* @return new worker, or null of factory failed
420	<	*/
421	<	private ForkJoinWorkerThread createWorker(int index) {
422	<	Thread.UncaughtExceptionHandler h = ueh;
423	<	ForkJoinWorkerThread w = factory.newThread(this);
424	<	if (w != null) {
425	<	w.poolIndex = index;
426	<	w.setDaemon(true);
427	<	w.setAsyncMode(locallyFifo);
428	<	w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
429	<	if (h != null)
430	<	w.setUncaughtExceptionHandler(h);
431	<	}
432	<	return w;
2445	>	this.factory = factory;
2446	>	this.ueh = handler;
2447	>	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2448	>	long np = (long)(-parallelism); // offset ctl counts
2449	>	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2450	>	int pn = nextPoolId();
2451	>	StringBuilder sb = new StringBuilder("ForkJoinPool-");
2452	>	sb.append(Integer.toString(pn));
2453	>	sb.append("-worker-");
2454	>	this.workerNamePrefix = sb.toString();
2455		}
2456
2457		/**
2458	<	* Returns a good size for worker array given pool size.
2459	<	* Currently requires size to be a power of two.
2460	<	*/
2461	<	private static int arraySizeFor(int poolSize) {
2462	<	return (poolSize <= 1) ? 1 :
2463	<	(1 << (32 - Integer.numberOfLeadingZeros(poolSize-1)));
2458	>	* Constructor for common pool, suitable only for static initialization.
2459	>	* Basically the same as above, but uses smallest possible initial footprint.
2460	>	*/
2461	>	ForkJoinPool(int parallelism, long ctl,
2462	>	ForkJoinWorkerThreadFactory factory,
2463	>	Thread.UncaughtExceptionHandler handler) {
2464	>	this.parallelism = parallelism;
2465	>	this.ctl = ctl;
2466	>	this.factory = factory;
2467	>	this.ueh = handler;
2468	>	this.localMode = LIFO_QUEUE;
2469	>	this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2470		}
2471
2472		/**
2473	<	* Creates or resizes array if necessary to hold newLength.
446	<	* Call only under exclusion.
2473	>	* Returns the common pool instance.
2474		*
2475	<	* @return the array
449	<	*/
450	<	private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
451	<	ForkJoinWorkerThread[] ws = workers;
452	<	if (ws == null)
453	<	return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
454	<	else if (newLength > ws.length)
455	<	return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
456	<	else
457	<	return ws;
458	<	}
459	<
460	<	/**
461	<	* Tries to shrink workers into smaller array after one or more terminate.
462	<	*/
463	<	private void tryShrinkWorkerArray() {
464	<	ForkJoinWorkerThread[] ws = workers;
465	<	if (ws != null) {
466	<	int len = ws.length;
467	<	int last = len - 1;
468	<	while (last >= 0 && ws[last] == null)
469	<	--last;
470	<	int newLength = arraySizeFor(last+1);
471	<	if (newLength < len)
472	<	workers = Arrays.copyOf(ws, newLength);
473	<	}
474	<	}
475	<
476	<	/**
477	<	* Initializes workers if necessary.
478	<	*/
479	<	final void ensureWorkerInitialization() {
480	<	ForkJoinWorkerThread[] ws = workers;
481	<	if (ws == null) {
482	<	final ReentrantLock lock = this.workerLock;
483	<	lock.lock();
484	<	try {
485	<	ws = workers;
486	<	if (ws == null) {
487	<	int ps = parallelism;
488	<	ws = ensureWorkerArrayCapacity(ps);
489	<	for (int i = 0; i < ps; ++i) {
490	<	ForkJoinWorkerThread w = createWorker(i);
491	<	if (w != null) {
492	<	ws[i] = w;
493	<	w.start();
494	<	updateWorkerCount(1);
495	<	}
496	<	}
497	<	}
498	<	} finally {
499	<	lock.unlock();
500	<	}
501	<	}
502	<	}
503	<
504	<	/**
505	<	* Worker creation and startup for threads added via setParallelism.
2475	>	* @return the common pool instance
2476		*/
2477	<	private void createAndStartAddedWorkers() {
2478	<	resumeAllSpares();  // Allow spares to convert to nonspare
509	<	int ps = parallelism;
510	<	ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
511	<	int len = ws.length;
512	<	// Sweep through slots, to keep lowest indices most populated
513	<	int k = 0;
514	<	while (k < len) {
515	<	if (ws[k] != null) {
516	<	++k;
517	<	continue;
518	<	}
519	<	int s = workerCounts;
520	<	int tc = totalCountOf(s);
521	<	int rc = runningCountOf(s);
522	<	if (rc >= ps || tc >= ps)
523	<	break;
524	<	if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
525	<	ForkJoinWorkerThread w = createWorker(k);
526	<	if (w != null) {
527	<	ws[k++] = w;
528	<	w.start();
529	<	}
530	<	else {
531	<	updateWorkerCount(-1); // back out on failed creation
532	<	break;
533	<	}
534	<	}
535	<	}
2477	>	public static ForkJoinPool commonPool() {
2478	>	return commonPool; // cannot be null (if so, a static init error)
2479		}
2480
2481		// Execution methods
2482
2483		/**
541	–	* Common code for execute, invoke and submit
542	–	*/
543	–	private <T> void doSubmit(ForkJoinTask<T> task) {
544	–	if (isShutdown())
545	–	throw new RejectedExecutionException();
546	–	if (workers == null)
547	–	ensureWorkerInitialization();
548	–	submissionQueue.offer(task);
549	–	signalIdleWorkers();
550	–	}
551	–
552	–	/**
2484		* Performs the given task, returning its result upon completion.
2485	+	* If the computation encounters an unchecked Exception or Error,
2486	+	* it is rethrown as the outcome of this invocation.  Rethrown
2487	+	* exceptions behave in the same way as regular exceptions, but,
2488	+	* when possible, contain stack traces (as displayed for example
2489	+	* using {@code ex.printStackTrace()}) of both the current thread
2490	+	* as well as the thread actually encountering the exception;
2491	+	* minimally only the latter.
2492		*
2493		* @param task the task
2494		* @return the task's result
2495	<	* @throws NullPointerException if task is null
2496	<	* @throws RejectedExecutionException if pool is shut down
2495	>	* @throws NullPointerException if the task is null
2496	>	* @throws RejectedExecutionException if the task cannot be
2497	>	*         scheduled for execution
2498		*/
2499		public <T> T invoke(ForkJoinTask<T> task) {
2500	<	doSubmit(task);
2500	>	if (task == null)
2501	>	throw new NullPointerException();
2502	>	externalPush(task);
2503		return task.join();
2504		}
2505
#	Line 566 | Line 2507 | public class ForkJoinPool extends Abstra
2507		* Arranges for (asynchronous) execution of the given task.
2508		*
2509		* @param task the task
2510	<	* @throws NullPointerException if task is null
2511	<	* @throws RejectedExecutionException if pool is shut down
2510	>	* @throws NullPointerException if the task is null
2511	>	* @throws RejectedExecutionException if the task cannot be
2512	>	*         scheduled for execution
2513		*/
2514	<	public <T> void execute(ForkJoinTask<T> task) {
2515	<	doSubmit(task);
2514	>	public void execute(ForkJoinTask<?> task) {
2515	>	if (task == null)
2516	>	throw new NullPointerException();
2517	>	externalPush(task);
2518		}
2519
2520		// AbstractExecutorService methods
2521
2522	+	/**
2523	+	* @throws NullPointerException if the task is null
2524	+	* @throws RejectedExecutionException if the task cannot be
2525	+	*         scheduled for execution
2526	+	*/
2527		public void execute(Runnable task) {
2528	<	doSubmit(new AdaptedRunnable<Void>(task, null));
2528	>	if (task == null)
2529	>	throw new NullPointerException();
2530	>	ForkJoinTask<?> job;
2531	>	if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2532	>	job = (ForkJoinTask<?>) task;
2533	>	else
2534	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2535	>	externalPush(job);
2536		}
2537
2538	<	public <T> ForkJoinTask<T> submit(Callable<T> task) {
2539	<	ForkJoinTask<T> job = new AdaptedCallable<T>(task);
2540	<	doSubmit(job);
2541	<	return job;
2538	>	/**
2539	>	* Submits a ForkJoinTask for execution.
2540	>	*
2541	>	* @param task the task to submit
2542	>	* @return the task
2543	>	* @throws NullPointerException if the task is null
2544	>	* @throws RejectedExecutionException if the task cannot be
2545	>	*         scheduled for execution
2546	>	*/
2547	>	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2548	>	if (task == null)
2549	>	throw new NullPointerException();
2550	>	externalPush(task);
2551	>	return task;
2552		}
2553
2554	<	public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2555	<	ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
2556	<	doSubmit(job);
2554	>	/**
2555	>	* @throws NullPointerException if the task is null
2556	>	* @throws RejectedExecutionException if the task cannot be
2557	>	*         scheduled for execution
2558	>	*/
2559	>	public <T> ForkJoinTask<T> submit(Callable<T> task) {
2560	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2561	>	externalPush(job);
2562		return job;
2563		}
2564
2565	<	public ForkJoinTask<?> submit(Runnable task) {
2566	<	ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
2567	<	doSubmit(job);
2565	>	/**
2566	>	* @throws NullPointerException if the task is null
2567	>	* @throws RejectedExecutionException if the task cannot be
2568	>	*         scheduled for execution
2569	>	*/
2570	>	public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2571	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2572	>	externalPush(job);
2573		return job;
2574		}
2575
2576		/**
2577	<	* Adaptor for Runnables. This implements RunnableFuture
2578	<	* to be compliant with AbstractExecutorService constraints.
2577	>	* @throws NullPointerException if the task is null
2578	>	* @throws RejectedExecutionException if the task cannot be
2579	>	*         scheduled for execution
2580		*/
2581	<	static final class AdaptedRunnable<T> extends ForkJoinTask<T>
2582	<	implements RunnableFuture<T> {
2583	<	final Runnable runnable;
2584	<	final T resultOnCompletion;
2585	<	T result;
2586	<	AdaptedRunnable(Runnable runnable, T result) {
2587	<	if (runnable == null) throw new NullPointerException();
2588	<	this.runnable = runnable;
2589	<	this.resultOnCompletion = result;
2590	<	}
614	<	public T getRawResult() { return result; }
615	<	public void setRawResult(T v) { result = v; }
616	<	public boolean exec() {
617	<	runnable.run();
618	<	result = resultOnCompletion;
619	<	return true;
620	<	}
621	<	public void run() { invoke(); }
622	<	private static final long serialVersionUID = 5232453952276885070L;
2581	>	public ForkJoinTask<?> submit(Runnable task) {
2582	>	if (task == null)
2583	>	throw new NullPointerException();
2584	>	ForkJoinTask<?> job;
2585	>	if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2586	>	job = (ForkJoinTask<?>) task;
2587	>	else
2588	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2589	>	externalPush(job);
2590	>	return job;
2591		}
2592
2593		/**
2594	<	* Adaptor for Callables
2594	>	* @throws NullPointerException       {@inheritDoc}
2595	>	* @throws RejectedExecutionException {@inheritDoc}
2596		*/
628	–	static final class AdaptedCallable<T> extends ForkJoinTask<T>
629	–	implements RunnableFuture<T> {
630	–	final Callable<T> callable;
631	–	T result;
632	–	AdaptedCallable(Callable<T> callable) {
633	–	if (callable == null) throw new NullPointerException();
634	–	this.callable = callable;
635	–	}
636	–	public T getRawResult() { return result; }
637	–	public void setRawResult(T v) { result = v; }
638	–	public boolean exec() {
639	–	try {
640	–	result = callable.call();
641	–	return true;
642	–	} catch (Error err) {
643	–	throw err;
644	–	} catch (RuntimeException rex) {
645	–	throw rex;
646	–	} catch (Exception ex) {
647	–	throw new RuntimeException(ex);
648	–	}
649	–	}
650	–	public void run() { invoke(); }
651	–	private static final long serialVersionUID = 2838392045355241008L;
652	–	}
653	–
2597		public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2598	<	ArrayList<ForkJoinTask<T>> ts =
2599	<	new ArrayList<ForkJoinTask<T>>(tasks.size());
2600	<	for (Callable<T> c : tasks)
2601	<	ts.add(new AdaptedCallable<T>(c));
2602	<	invoke(new InvokeAll<T>(ts));
2603	<	return (List<Future<T>>) (List) ts;
2604	<	}
2598	>	// In previous versions of this class, this method constructed
2599	>	// a task to run ForkJoinTask.invokeAll, but now external
2600	>	// invocation of multiple tasks is at least as efficient.
2601	>	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2602	>	// Workaround needed because method wasn't declared with
2603	>	// wildcards in return type but should have been.
2604	>	@SuppressWarnings({"unchecked", "rawtypes"})
2605	>	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2606
2607	<	static final class InvokeAll<T> extends RecursiveAction {
2608	<	final ArrayList<ForkJoinTask<T>> tasks;
2609	<	InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
2610	<	public void compute() {
2611	<	try { invokeAll(tasks); }
2612	<	catch (Exception ignore) {}
2607	>	boolean done = false;
2608	>	try {
2609	>	for (Callable<T> t : tasks) {
2610	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2611	>	externalPush(f);
2612	>	fs.add(f);
2613	>	}
2614	>	for (ForkJoinTask<T> f : fs)
2615	>	f.quietlyJoin();
2616	>	done = true;
2617	>	return futures;
2618	>	} finally {
2619	>	if (!done)
2620	>	for (ForkJoinTask<T> f : fs)
2621	>	f.cancel(false);
2622		}
670	–	private static final long serialVersionUID = -7914297376763021607L;
2623		}
2624
673	–	// Configuration and status settings and queries
674	–
2625		/**
2626		* Returns the factory used for constructing new workers.
2627		*
#	Line 685 | Line 2635 | public class ForkJoinPool extends Abstra
2635		* Returns the handler for internal worker threads that terminate
2636		* due to unrecoverable errors encountered while executing tasks.
2637		*
2638	<	* @return the handler, or null if none
2638	>	* @return the handler, or {@code null} if none
2639		*/
2640		public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
2641	<	Thread.UncaughtExceptionHandler h;
692	<	final ReentrantLock lock = this.workerLock;
693	<	lock.lock();
694	<	try {
695	<	h = ueh;
696	<	} finally {
697	<	lock.unlock();
698	<	}
699	<	return h;
700	<	}
701	<
702	<	/**
703	<	* Sets the handler for internal worker threads that terminate due
704	<	* to unrecoverable errors encountered while executing tasks.
705	<	* Unless set, the current default or ThreadGroup handler is used
706	<	* as handler.
707	<	*
708	<	* @param h the new handler
709	<	* @return the old handler, or null if none
710	<	* @throws SecurityException if a security manager exists and
711	<	*         the caller is not permitted to modify threads
712	<	*         because it does not hold {@link
713	<	*         java.lang.RuntimePermission}{@code ("modifyThread")}
714	<	*/
715	<	public Thread.UncaughtExceptionHandler
716	<	setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
717	<	checkPermission();
718	<	Thread.UncaughtExceptionHandler old = null;
719	<	final ReentrantLock lock = this.workerLock;
720	<	lock.lock();
721	<	try {
722	<	old = ueh;
723	<	ueh = h;
724	<	ForkJoinWorkerThread[] ws = workers;
725	<	if (ws != null) {
726	<	for (int i = 0; i < ws.length; ++i) {
727	<	ForkJoinWorkerThread w = ws[i];
728	<	if (w != null)
729	<	w.setUncaughtExceptionHandler(h);
730	<	}
731	<	}
732	<	} finally {
733	<	lock.unlock();
734	<	}
735	<	return old;
2641	>	return ueh;
2642		}
2643
738	–
2644		/**
2645	<	* Sets the target parallelism level of this pool.
2645	>	* Returns the targeted parallelism level of this pool.
2646		*
2647	<	* @param parallelism the target parallelism
743	<	* @throws IllegalArgumentException if parallelism less than or
744	<	* equal to zero or greater than maximum size bounds
745	<	* @throws SecurityException if a security manager exists and
746	<	*         the caller is not permitted to modify threads
747	<	*         because it does not hold {@link
748	<	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2647	>	* @return the targeted parallelism level of this pool
2648		*/
2649	<	public void setParallelism(int parallelism) {
2650	<	checkPermission();
752	<	if (parallelism <= 0 || parallelism > maxPoolSize)
753	<	throw new IllegalArgumentException();
754	<	final ReentrantLock lock = this.workerLock;
755	<	lock.lock();
756	<	try {
757	<	if (!isTerminating()) {
758	<	int p = this.parallelism;
759	<	this.parallelism = parallelism;
760	<	if (parallelism > p)
761	<	createAndStartAddedWorkers();
762	<	else
763	<	trimSpares();
764	<	}
765	<	} finally {
766	<	lock.unlock();
767	<	}
768	<	signalIdleWorkers();
2649	>	public int getParallelism() {
2650	>	return parallelism;
2651		}
2652
2653		/**
2654	<	* Returns the targeted number of worker threads in this pool.
2654	>	* Returns the targeted parallelism level of the common pool.
2655		*
2656	<	* @return the targeted number of worker threads in this pool
2656	>	* @return the targeted parallelism level of the common pool
2657		*/
2658	<	public int getParallelism() {
2659	<	return parallelism;
2658	>	public static int getCommonPoolParallelism() {
2659	>	return commonPoolParallelism;
2660		}
2661
2662		/**
2663		* Returns the number of worker threads that have started but not
2664	<	* yet terminated.  This result returned by this method may differ
2665	<	* from {@code getParallelism} when threads are created to
2664	>	* yet terminated.  The result returned by this method may differ
2665	>	* from {@link #getParallelism} when threads are created to
2666		* maintain parallelism when others are cooperatively blocked.
2667		*
2668		* @return the number of worker threads
2669		*/
2670		public int getPoolSize() {
2671	<	return totalCountOf(workerCounts);
790	<	}
791	<
792	<	/**
793	<	* Returns the maximum number of threads allowed to exist in the
794	<	* pool, even if there are insufficient unblocked running threads.
795	<	*
796	<	* @return the maximum
797	<	*/
798	<	public int getMaximumPoolSize() {
799	<	return maxPoolSize;
800	<	}
801	<
802	<	/**
803	<	* Sets the maximum number of threads allowed to exist in the
804	<	* pool, even if there are insufficient unblocked running threads.
805	<	* Setting this value has no effect on current pool size. It
806	<	* controls construction of new threads.
807	<	*
808	<	* @throws IllegalArgumentException if negative or greater then
809	<	* internal implementation limit
810	<	*/
811	<	public void setMaximumPoolSize(int newMax) {
812	<	if (newMax < 0 || newMax > MAX_THREADS)
813	<	throw new IllegalArgumentException();
814	<	maxPoolSize = newMax;
815	<	}
816	<
817	<
818	<	/**
819	<	* Returns true if this pool dynamically maintains its target
820	<	* parallelism level. If false, new threads are added only to
821	<	* avoid possible starvation.
822	<	* This setting is by default true.
823	<	*
824	<	* @return true if maintains parallelism
825	<	*/
826	<	public boolean getMaintainsParallelism() {
827	<	return maintainsParallelism;
828	<	}
829	<
830	<	/**
831	<	* Sets whether this pool dynamically maintains its target
832	<	* parallelism level. If false, new threads are added only to
833	<	* avoid possible starvation.
834	<	*
835	<	* @param enable true to maintains parallelism
836	<	*/
837	<	public void setMaintainsParallelism(boolean enable) {
838	<	maintainsParallelism = enable;
2671	>	return parallelism + (short)(ctl >>> TC_SHIFT);
2672		}
2673
2674		/**
2675	<	* Establishes local first-in-first-out scheduling mode for forked
843	<	* tasks that are never joined. This mode may be more appropriate
844	<	* than default locally stack-based mode in applications in which
845	<	* worker threads only process asynchronous tasks.  This method is
846	<	* designed to be invoked only when pool is quiescent, and
847	<	* typically only before any tasks are submitted. The effects of
848	<	* invocations at other times may be unpredictable.
849	<	*
850	<	* @param async if true, use locally FIFO scheduling
851	<	* @return the previous mode
852	<	*/
853	<	public boolean setAsyncMode(boolean async) {
854	<	boolean oldMode = locallyFifo;
855	<	locallyFifo = async;
856	<	ForkJoinWorkerThread[] ws = workers;
857	<	if (ws != null) {
858	<	for (int i = 0; i < ws.length; ++i) {
859	<	ForkJoinWorkerThread t = ws[i];
860	<	if (t != null)
861	<	t.setAsyncMode(async);
862	<	}
863	<	}
864	<	return oldMode;
865	<	}
866	<
867	<	/**
868	<	* Returns true if this pool uses local first-in-first-out
2675	>	* Returns {@code true} if this pool uses local first-in-first-out
2676		* scheduling mode for forked tasks that are never joined.
2677		*
2678	<	* @return true if this pool uses async mode
2678	>	* @return {@code true} if this pool uses async mode
2679		*/
2680		public boolean getAsyncMode() {
2681	<	return locallyFifo;
2681	>	return localMode != 0;
2682		}
2683
2684		/**
2685		* Returns an estimate of the number of worker threads that are
2686		* not blocked waiting to join tasks or for other managed
2687	<	* synchronization.
2687	>	* synchronization. This method may overestimate the
2688	>	* number of running threads.
2689		*
2690		* @return the number of worker threads
2691		*/
2692		public int getRunningThreadCount() {
2693	<	return runningCountOf(workerCounts);
2693	>	int rc = 0;
2694	>	WorkQueue[] ws; WorkQueue w;
2695	>	if ((ws = workQueues) != null) {
2696	>	for (int i = 1; i < ws.length; i += 2) {
2697	>	if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2698	>	++rc;
2699	>	}
2700	>	}
2701	>	return rc;
2702		}
2703
2704		/**
#	Line 893 | Line 2709 | public class ForkJoinPool extends Abstra
2709		* @return the number of active threads
2710		*/
2711		public int getActiveThreadCount() {
2712	<	return activeCountOf(runControl);
2712	>	int r = parallelism + (int)(ctl >> AC_SHIFT);
2713	>	return (r <= 0) ? 0 : r; // suppress momentarily negative values
2714		}
2715
2716		/**
2717	<	* Returns an estimate of the number of threads that are currently
2718	<	* idle waiting for tasks. This method may underestimate the
2719	<	* number of idle threads.
2717	>	* Returns {@code true} if all worker threads are currently idle.
2718	>	* An idle worker is one that cannot obtain a task to execute
2719	>	* because none are available to steal from other threads, and
2720	>	* there are no pending submissions to the pool. This method is
2721	>	* conservative; it might not return {@code true} immediately upon
2722	>	* idleness of all threads, but will eventually become true if
2723	>	* threads remain inactive.
2724		*
2725	<	* @return the number of idle threads
905	<	*/
906	<	final int getIdleThreadCount() {
907	<	int c = runningCountOf(workerCounts) - activeCountOf(runControl);
908	<	return (c <= 0) ? 0 : c;
909	<	}
910	<
911	<	/**
912	<	* Returns true if all worker threads are currently idle. An idle
913	<	* worker is one that cannot obtain a task to execute because none
914	<	* are available to steal from other threads, and there are no
915	<	* pending submissions to the pool. This method is conservative;
916	<	* it might not return true immediately upon idleness of all
917	<	* threads, but will eventually become true if threads remain
918	<	* inactive.
919	<	*
920	<	* @return true if all threads are currently idle
2725	>	* @return {@code true} if all threads are currently idle
2726		*/
2727		public boolean isQuiescent() {
2728	<	return activeCountOf(runControl) == 0;
2728	>	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2729		}
2730
2731		/**
#	Line 935 | Line 2740 | public class ForkJoinPool extends Abstra
2740		* @return the number of steals
2741		*/
2742		public long getStealCount() {
2743	<	return stealCount.get();
2744	<	}
2745	<
2746	<	/**
2747	<	* Accumulates steal count from a worker.
2748	<	* Call only when worker known to be idle.
2749	<	*/
2750	<	private void updateStealCount(ForkJoinWorkerThread w) {
2751	<	int sc = w.getAndClearStealCount();
947	<	if (sc != 0)
948	<	stealCount.addAndGet(sc);
2743	>	long count = stealCount;
2744	>	WorkQueue[] ws; WorkQueue w;
2745	>	if ((ws = workQueues) != null) {
2746	>	for (int i = 1; i < ws.length; i += 2) {
2747	>	if ((w = ws[i]) != null)
2748	>	count += w.nsteals;
2749	>	}
2750	>	}
2751	>	return count;
2752		}
2753
2754		/**
#	Line 960 | Line 2763 | public class ForkJoinPool extends Abstra
2763		*/
2764		public long getQueuedTaskCount() {
2765		long count = 0;
2766	<	ForkJoinWorkerThread[] ws = workers;
2767	<	if (ws != null) {
2768	<	for (int i = 0; i < ws.length; ++i) {
2769	<	ForkJoinWorkerThread t = ws[i];
2770	<	if (t != null)
968	<	count += t.getQueueSize();
2766	>	WorkQueue[] ws; WorkQueue w;
2767	>	if ((ws = workQueues) != null) {
2768	>	for (int i = 1; i < ws.length; i += 2) {
2769	>	if ((w = ws[i]) != null)
2770	>	count += w.queueSize();
2771		}
2772		}
2773		return count;
2774		}
2775
2776		/**
2777	<	* Returns an estimate of the number tasks submitted to this pool
2778	<	* that have not yet begun executing. This method takes time
2779	<	* proportional to the number of submissions.
2777	>	* Returns an estimate of the number of tasks submitted to this
2778	>	* pool that have not yet begun executing.  This method may take
2779	>	* time proportional to the number of submissions.
2780		*
2781		* @return the number of queued submissions
2782		*/
2783		public int getQueuedSubmissionCount() {
2784	<	return submissionQueue.size();
2784	>	int count = 0;
2785	>	WorkQueue[] ws; WorkQueue w;
2786	>	if ((ws = workQueues) != null) {
2787	>	for (int i = 0; i < ws.length; i += 2) {
2788	>	if ((w = ws[i]) != null)
2789	>	count += w.queueSize();
2790	>	}
2791	>	}
2792	>	return count;
2793		}
2794
2795		/**
2796	<	* Returns true if there are any tasks submitted to this pool
2797	<	* that have not yet begun executing.
2796	>	* Returns {@code true} if there are any tasks submitted to this
2797	>	* pool that have not yet begun executing.
2798		*
2799		* @return {@code true} if there are any queued submissions
2800		*/
2801		public boolean hasQueuedSubmissions() {
2802	<	return !submissionQueue.isEmpty();
2802	>	WorkQueue[] ws; WorkQueue w;
2803	>	if ((ws = workQueues) != null) {
2804	>	for (int i = 0; i < ws.length; i += 2) {
2805	>	if ((w = ws[i]) != null && w.queueSize() != 0)
2806	>	return true;
2807	>	}
2808	>	}
2809	>	return false;
2810		}
2811
2812		/**
#	Line 997 | Line 2814 | public class ForkJoinPool extends Abstra
2814		* available.  This method may be useful in extensions to this
2815		* class that re-assign work in systems with multiple pools.
2816		*
2817	<	* @return the next submission, or null if none
2817	>	* @return the next submission, or {@code null} if none
2818		*/
2819		protected ForkJoinTask<?> pollSubmission() {
2820	<	return submissionQueue.poll();
2820	>	WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2821	>	if ((ws = workQueues) != null) {
2822	>	for (int i = 0; i < ws.length; i += 2) {
2823	>	if ((w = ws[i]) != null && (t = w.poll()) != null)
2824	>	return t;
2825	>	}
2826	>	}
2827	>	return null;
2828		}
2829
2830		/**
2831		* Removes all available unexecuted submitted and forked tasks
2832		* from scheduling queues and adds them to the given collection,
2833		* without altering their execution status. These may include
2834	<	* artificially generated or wrapped tasks. This method is designed
2835	<	* to be invoked only when the pool is known to be
2834	>	* artificially generated or wrapped tasks. This method is
2835	>	* designed to be invoked only when the pool is known to be
2836		* quiescent. Invocations at other times may not remove all
2837		* tasks. A failure encountered while attempting to add elements
2838		* to collection {@code c} may result in elements being in
#	Line 1020 | Line 2844 | public class ForkJoinPool extends Abstra
2844		* @param c the collection to transfer elements into
2845		* @return the number of elements transferred
2846		*/
2847	<	protected int drainTasksTo(Collection<ForkJoinTask<?>> c) {
2848	<	int n = submissionQueue.drainTo(c);
2849	<	ForkJoinWorkerThread[] ws = workers;
2850	<	if (ws != null) {
2847	>	protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
2848	>	int count = 0;
2849	>	WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2850	>	if ((ws = workQueues) != null) {
2851		for (int i = 0; i < ws.length; ++i) {
2852	<	ForkJoinWorkerThread w = ws[i];
2853	<	if (w != null)
2854	<	n += w.drainTasksTo(c);
2852	>	if ((w = ws[i]) != null) {
2853	>	while ((t = w.poll()) != null) {
2854	>	c.add(t);
2855	>	++count;
2856	>	}
2857	>	}
2858		}
2859		}
2860	<	return n;
2860	>	return count;
2861		}
2862
2863		/**
#	Line 1041 | Line 2868 | public class ForkJoinPool extends Abstra
2868		* @return a string identifying this pool, as well as its state
2869		*/
2870		public String toString() {
2871	<	int ps = parallelism;
2872	<	int wc = workerCounts;
2873	<	int rc = runControl;
2874	<	long st = getStealCount();
2875	<	long qt = getQueuedTaskCount();
2876	<	long qs = getQueuedSubmissionCount();
2871	>	// Use a single pass through workQueues to collect counts
2872	>	long qt = 0L, qs = 0L; int rc = 0;
2873	>	long st = stealCount;
2874	>	long c = ctl;
2875	>	WorkQueue[] ws; WorkQueue w;
2876	>	if ((ws = workQueues) != null) {
2877	>	for (int i = 0; i < ws.length; ++i) {
2878	>	if ((w = ws[i]) != null) {
2879	>	int size = w.queueSize();
2880	>	if ((i & 1) == 0)
2881	>	qs += size;
2882	>	else {
2883	>	qt += size;
2884	>	st += w.nsteals;
2885	>	if (w.isApparentlyUnblocked())
2886	>	++rc;
2887	>	}
2888	>	}
2889	>	}
2890	>	}
2891	>	int pc = parallelism;
2892	>	int tc = pc + (short)(c >>> TC_SHIFT);
2893	>	int ac = pc + (int)(c >> AC_SHIFT);
2894	>	if (ac < 0) // ignore transient negative
2895	>	ac = 0;
2896	>	String level;
2897	>	if ((c & STOP_BIT) != 0)
2898	>	level = (tc == 0) ? "Terminated" : "Terminating";
2899	>	else
2900	>	level = plock < 0 ? "Shutting down" : "Running";
2901		return super.toString() +
2902	<	"[" + runStateToString(runStateOf(rc)) +
2903	<	", parallelism = " + ps +
2904	<	", size = " + totalCountOf(wc) +
2905	<	", active = " + activeCountOf(rc) +
2906	<	", running = " + runningCountOf(wc) +
2902	>	"[" + level +
2903	>	", parallelism = " + pc +
2904	>	", size = " + tc +
2905	>	", active = " + ac +
2906	>	", running = " + rc +
2907		", steals = " + st +
2908		", tasks = " + qt +
2909		", submissions = " + qs +
2910		"]";
2911		}
2912
1062	–	private static String runStateToString(int rs) {
1063	–	switch(rs) {
1064	–	case RUNNING: return "Running";
1065	–	case SHUTDOWN: return "Shutting down";
1066	–	case TERMINATING: return "Terminating";
1067	–	case TERMINATED: return "Terminated";
1068	–	default: throw new Error("Unknown run state");
1069	–	}
1070	–	}
1071	–
1072	–	// lifecycle control
1073	–
2913		/**
2914	<	* Initiates an orderly shutdown in which previously submitted
2915	<	* tasks are executed, but no new tasks will be accepted.
2916	<	* Invocation has no additional effect if already shut down.
2917	<	* Tasks that are in the process of being submitted concurrently
2918	<	* during the course of this method may or may not be rejected.
2914	>	* Possibly initiates an orderly shutdown in which previously
2915	>	* submitted tasks are executed, but no new tasks will be
2916	>	* accepted. Invocation has no effect on execution state if this
2917	>	* is the {@link #commonPool}, and no additional effect if
2918	>	* already shut down.  Tasks that are in the process of being
2919	>	* submitted concurrently during the course of this method may or
2920	>	* may not be rejected.
2921		*
2922		* @throws SecurityException if a security manager exists and
2923		*         the caller is not permitted to modify threads
#	Line 1085 | Line 2926 | public class ForkJoinPool extends Abstra
2926		*/
2927		public void shutdown() {
2928		checkPermission();
2929	<	transitionRunStateTo(SHUTDOWN);
1089	<	if (canTerminateOnShutdown(runControl))
1090	<	terminateOnShutdown();
2929	>	tryTerminate(false, true);
2930		}
2931
2932		/**
2933	<	* Attempts to stop all actively executing tasks, and cancels all
2934	<	* waiting tasks.  Tasks that are in the process of being
2935	<	* submitted or executed concurrently during the course of this
2936	<	* method may or may not be rejected. Unlike some other executors,
2937	<	* this method cancels rather than collects non-executed tasks
2938	<	* upon termination, so always returns an empty list. However, you
2939	<	* can use method {@code drainTasksTo} before invoking this
2940	<	* method to transfer unexecuted tasks to another collection.
2933	>	* Possibly attempts to cancel and/or stop all tasks, and reject
2934	>	* all subsequently submitted tasks.  Invocation has no effect on
2935	>	* execution state if this is the {@link #commonPool}, and no
2936	>	* additional effect if already shut down. Otherwise, tasks that
2937	>	* are in the process of being submitted or executed concurrently
2938	>	* during the course of this method may or may not be
2939	>	* rejected. This method cancels both existing and unexecuted
2940	>	* tasks, in order to permit termination in the presence of task
2941	>	* dependencies. So the method always returns an empty list
2942	>	* (unlike the case for some other Executors).
2943		*
2944		* @return an empty list
2945		* @throws SecurityException if a security manager exists and
#	Line 1108 | Line 2949 | public class ForkJoinPool extends Abstra
2949		*/
2950		public List<Runnable> shutdownNow() {
2951		checkPermission();
2952	<	terminate();
2952	>	tryTerminate(true, true);
2953		return Collections.emptyList();
2954		}
2955
#	Line 1118 | Line 2959 | public class ForkJoinPool extends Abstra
2959		* @return {@code true} if all tasks have completed following shut down
2960		*/
2961		public boolean isTerminated() {
2962	<	return runStateOf(runControl) == TERMINATED;
2962	>	long c = ctl;
2963	>	return ((c & STOP_BIT) != 0L &&
2964	>	(short)(c >>> TC_SHIFT) == -parallelism);
2965		}
2966
2967		/**
2968		* Returns {@code true} if the process of termination has
2969	<	* commenced but possibly not yet completed.
2969	>	* commenced but not yet completed.  This method may be useful for
2970	>	* debugging. A return of {@code true} reported a sufficient
2971	>	* period after shutdown may indicate that submitted tasks have
2972	>	* ignored or suppressed interruption, or are waiting for IO,
2973	>	* causing this executor not to properly terminate. (See the
2974	>	* advisory notes for class {@link ForkJoinTask} stating that
2975	>	* tasks should not normally entail blocking operations.  But if
2976	>	* they do, they must abort them on interrupt.)
2977		*
2978	<	* @return {@code true} if terminating
2978	>	* @return {@code true} if terminating but not yet terminated
2979		*/
2980		public boolean isTerminating() {
2981	<	return runStateOf(runControl) >= TERMINATING;
2981	>	long c = ctl;
2982	>	return ((c & STOP_BIT) != 0L &&
2983	>	(short)(c >>> TC_SHIFT) != -parallelism);
2984		}
2985
2986		/**
#	Line 1137 | Line 2989 | public class ForkJoinPool extends Abstra
2989		* @return {@code true} if this pool has been shut down
2990		*/
2991		public boolean isShutdown() {
2992	<	return runStateOf(runControl) >= SHUTDOWN;
2992	>	return plock < 0;
2993		}
2994
2995		/**
2996	<	* Blocks until all tasks have completed execution after a shutdown
2997	<	* request, or the timeout occurs, or the current thread is
2998	<	* interrupted, whichever happens first.
2996	>	* Blocks until all tasks have completed execution after a
2997	>	* shutdown request, or the timeout occurs, or the current thread
2998	>	* is interrupted, whichever happens first. Note that the {@link
2999	>	* #commonPool()} never terminates until program shutdown so
3000	>	* this method will always time out.
3001		*
3002		* @param timeout the maximum time to wait
3003		* @param unit the time unit of the timeout argument
#	Line 1154 | Line 3008 | public class ForkJoinPool extends Abstra
3008		public boolean awaitTermination(long timeout, TimeUnit unit)
3009		throws InterruptedException {
3010		long nanos = unit.toNanos(timeout);
3011	<	final ReentrantLock lock = this.workerLock;
1158	<	lock.lock();
1159	<	try {
1160	<	for (;;) {
1161	<	if (isTerminated())
1162	<	return true;
1163	<	if (nanos <= 0)
1164	<	return false;
1165	<	nanos = termination.awaitNanos(nanos);
1166	<	}
1167	<	} finally {
1168	<	lock.unlock();
1169	<	}
1170	<	}
1171	<
1172	<	// Shutdown and termination support
1173	<
1174	<	/**
1175	<	* Callback from terminating worker. Nulls out the corresponding
1176	<	* workers slot, and if terminating, tries to terminate; else
1177	<	* tries to shrink workers array.
1178	<	*
1179	<	* @param w the worker
1180	<	*/
1181	<	final void workerTerminated(ForkJoinWorkerThread w) {
1182	<	updateStealCount(w);
1183	<	updateWorkerCount(-1);
1184	<	final ReentrantLock lock = this.workerLock;
1185	<	lock.lock();
1186	<	try {
1187	<	ForkJoinWorkerThread[] ws = workers;
1188	<	if (ws != null) {
1189	<	int idx = w.poolIndex;
1190	<	if (idx >= 0 && idx < ws.length && ws[idx] == w)
1191	<	ws[idx] = null;
1192	<	if (totalCountOf(workerCounts) == 0) {
1193	<	terminate(); // no-op if already terminating
1194	<	transitionRunStateTo(TERMINATED);
1195	<	termination.signalAll();
1196	<	}
1197	<	else if (!isTerminating()) {
1198	<	tryShrinkWorkerArray();
1199	<	tryResumeSpare(true); // allow replacement
1200	<	}
1201	<	}
1202	<	} finally {
1203	<	lock.unlock();
1204	<	}
1205	<	signalIdleWorkers();
1206	<	}
1207	<
1208	<	/**
1209	<	* Initiates termination.
1210	<	*/
1211	<	private void terminate() {
1212	<	if (transitionRunStateTo(TERMINATING)) {
1213	<	stopAllWorkers();
1214	<	resumeAllSpares();
1215	<	signalIdleWorkers();
1216	<	cancelQueuedSubmissions();
1217	<	cancelQueuedWorkerTasks();
1218	<	interruptUnterminatedWorkers();
1219	<	signalIdleWorkers(); // resignal after interrupt
1220	<	}
1221	<	}
1222	<
1223	<	/**
1224	<	* Possibly terminates when on shutdown state.
1225	<	*/
1226	<	private void terminateOnShutdown() {
1227	<	if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1228	<	terminate();
1229	<	}
1230	<
1231	<	/**
1232	<	* Clears out and cancels submissions.
1233	<	*/
1234	<	private void cancelQueuedSubmissions() {
1235	<	ForkJoinTask<?> task;
1236	<	while ((task = pollSubmission()) != null)
1237	<	task.cancel(false);
1238	<	}
1239	<
1240	<	/**
1241	<	* Cleans out worker queues.
1242	<	*/
1243	<	private void cancelQueuedWorkerTasks() {
1244	<	final ReentrantLock lock = this.workerLock;
1245	<	lock.lock();
1246	<	try {
1247	<	ForkJoinWorkerThread[] ws = workers;
1248	<	if (ws != null) {
1249	<	for (int i = 0; i < ws.length; ++i) {
1250	<	ForkJoinWorkerThread t = ws[i];
1251	<	if (t != null)
1252	<	t.cancelTasks();
1253	<	}
1254	<	}
1255	<	} finally {
1256	<	lock.unlock();
1257	<	}
1258	<	}
1259	<
1260	<	/**
1261	<	* Sets each worker's status to terminating. Requires lock to avoid
1262	<	* conflicts with add/remove.
1263	<	*/
1264	<	private void stopAllWorkers() {
1265	<	final ReentrantLock lock = this.workerLock;
1266	<	lock.lock();
1267	<	try {
1268	<	ForkJoinWorkerThread[] ws = workers;
1269	<	if (ws != null) {
1270	<	for (int i = 0; i < ws.length; ++i) {
1271	<	ForkJoinWorkerThread t = ws[i];
1272	<	if (t != null)
1273	<	t.shutdownNow();
1274	<	}
1275	<	}
1276	<	} finally {
1277	<	lock.unlock();
1278	<	}
1279	<	}
1280	<
1281	<	/**
1282	<	* Interrupts all unterminated workers.  This is not required for
1283	<	* sake of internal control, but may help unstick user code during
1284	<	* shutdown.
1285	<	*/
1286	<	private void interruptUnterminatedWorkers() {
1287	<	final ReentrantLock lock = this.workerLock;
1288	<	lock.lock();
1289	<	try {
1290	<	ForkJoinWorkerThread[] ws = workers;
1291	<	if (ws != null) {
1292	<	for (int i = 0; i < ws.length; ++i) {
1293	<	ForkJoinWorkerThread t = ws[i];
1294	<	if (t != null && !t.isTerminated()) {
1295	<	try {
1296	<	t.interrupt();
1297	<	} catch (SecurityException ignore) {
1298	<	}
1299	<	}
1300	<	}
1301	<	}
1302	<	} finally {
1303	<	lock.unlock();
1304	<	}
1305	<	}
1306	<
1307	<
1308	<	/*
1309	<	* Nodes for event barrier to manage idle threads.  Queue nodes
1310	<	* are basic Treiber stack nodes, also used for spare stack.
1311	<	*
1312	<	* The event barrier has an event count and a wait queue (actually
1313	<	* a Treiber stack).  Workers are enabled to look for work when
1314	<	* the eventCount is incremented. If they fail to find work, they
1315	<	* may wait for next count. Upon release, threads help others wake
1316	<	* up.
1317	<	*
1318	<	* Synchronization events occur only in enough contexts to
1319	<	* maintain overall liveness:
1320	<	*
1321	<	*   - Submission of a new task to the pool
1322	<	*   - Resizes or other changes to the workers array
1323	<	*   - pool termination
1324	<	*   - A worker pushing a task on an empty queue
1325	<	*
1326	<	* The case of pushing a task occurs often enough, and is heavy
1327	<	* enough compared to simple stack pushes, to require special
1328	<	* handling: Method signalWork returns without advancing count if
1329	<	* the queue appears to be empty.  This would ordinarily result in
1330	<	* races causing some queued waiters not to be woken up. To avoid
1331	<	* this, the first worker enqueued in method sync (see
1332	<	* syncIsReleasable) rescans for tasks after being enqueued, and
1333	<	* helps signal if any are found. This works well because the
1334	<	* worker has nothing better to do, and so might as well help
1335	<	* alleviate the overhead and contention on the threads actually
1336	<	* doing work.  Also, since event counts increments on task
1337	<	* availability exist to maintain liveness (rather than to force
1338	<	* refreshes etc), it is OK for callers to exit early if
1339	<	* contending with another signaller.
1340	<	*/
1341	<	static final class WaitQueueNode {
1342	<	WaitQueueNode next; // only written before enqueued
1343	<	volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1344	<	final long count; // unused for spare stack
1345	<
1346	<	WaitQueueNode(long c, ForkJoinWorkerThread w) {
1347	<	count = c;
1348	<	thread = w;
1349	<	}
1350	<
1351	<	/**
1352	<	* Wakes up waiter, returning false if known to already
1353	<	*/
1354	<	boolean signal() {
1355	<	ForkJoinWorkerThread t = thread;
1356	<	if (t == null)
1357	<	return false;
1358	<	thread = null;
1359	<	LockSupport.unpark(t);
3011	>	if (isTerminated())
3012		return true;
3013	<	}
3014	<
3015	<	/**
3016	<	* Awaits release on sync.
3017	<	*/
3018	<	void awaitSyncRelease(ForkJoinPool p) {
3019	<	while (thread != null && !p.syncIsReleasable(this))
1368	<	LockSupport.park(this);
1369	<	}
1370	<
1371	<	/**
1372	<	* Awaits resumption as spare.
1373	<	*/
1374	<	void awaitSpareRelease() {
1375	<	while (thread != null) {
1376	<	if (!Thread.interrupted())
1377	<	LockSupport.park(this);
1378	<	}
1379	<	}
1380	<	}
1381	<
1382	<	/**
1383	<	* Ensures that no thread is waiting for count to advance from the
1384	<	* current value of eventCount read on entry to this method, by
1385	<	* releasing waiting threads if necessary.
1386	<	*
1387	<	* @return the count
1388	<	*/
1389	<	final long ensureSync() {
1390	<	long c = eventCount;
1391	<	WaitQueueNode q;
1392	<	while ((q = syncStack) != null && q.count < c) {
1393	<	if (casBarrierStack(q, null)) {
1394	<	do {
1395	<	q.signal();
1396	<	} while ((q = q.next) != null);
1397	<	break;
1398	<	}
1399	<	}
1400	<	return c;
1401	<	}
1402	<
1403	<	/**
1404	<	* Increments event count and releases waiting threads.
1405	<	*/
1406	<	private void signalIdleWorkers() {
1407	<	long c;
1408	<	do {} while (!casEventCount(c = eventCount, c+1));
1409	<	ensureSync();
1410	<	}
1411	<
1412	<	/**
1413	<	* Signals threads waiting to poll a task. Because method sync
1414	<	* rechecks availability, it is OK to only proceed if queue
1415	<	* appears to be non-empty, and OK to skip under contention to
1416	<	* increment count (since some other thread succeeded).
1417	<	*/
1418	<	final void signalWork() {
1419	<	long c;
1420	<	WaitQueueNode q;
1421	<	if (syncStack != null &&
1422	<	casEventCount(c = eventCount, c+1) &&
1423	<	(((q = syncStack) != null && q.count <= c) &&
1424	<	(!casBarrierStack(q, q.next) || !q.signal())))
1425	<	ensureSync();
1426	<	}
1427	<
1428	<	/**
1429	<	* Waits until event count advances from last value held by
1430	<	* caller, or if excess threads, caller is resumed as spare, or
1431	<	* caller or pool is terminating. Updates caller's event on exit.
1432	<	*
1433	<	* @param w the calling worker thread
1434	<	*/
1435	<	final void sync(ForkJoinWorkerThread w) {
1436	<	updateStealCount(w); // Transfer w's count while it is idle
1437	<
1438	<	while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1439	<	long prev = w.lastEventCount;
1440	<	WaitQueueNode node = null;
1441	<	WaitQueueNode h;
1442	<	while (eventCount == prev &&
1443	<	((h = syncStack) == null || h.count == prev)) {
1444	<	if (node == null)
1445	<	node = new WaitQueueNode(prev, w);
1446	<	if (casBarrierStack(node.next = h, node)) {
1447	<	node.awaitSyncRelease(this);
1448	<	break;
1449	<	}
1450	<	}
1451	<	long ec = ensureSync();
1452	<	if (ec != prev) {
1453	<	w.lastEventCount = ec;
1454	<	break;
1455	<	}
1456	<	}
1457	<	}
1458	<
1459	<	/**
1460	<	* Returns true if worker waiting on sync can proceed:
1461	<	*  - on signal (thread == null)
1462	<	*  - on event count advance (winning race to notify vs signaller)
1463	<	*  - on interrupt
1464	<	*  - if the first queued node, we find work available
1465	<	* If node was not signalled and event count not advanced on exit,
1466	<	* then we also help advance event count.
1467	<	*
1468	<	* @return true if node can be released
1469	<	*/
1470	<	final boolean syncIsReleasable(WaitQueueNode node) {
1471	<	long prev = node.count;
1472	<	if (!Thread.interrupted() && node.thread != null &&
1473	<	(node.next != null ||
1474	<	!ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1475	<	eventCount == prev)
1476	<	return false;
1477	<	if (node.thread != null) {
1478	<	node.thread = null;
1479	<	long ec = eventCount;
1480	<	if (prev <= ec) // help signal
1481	<	casEventCount(ec, ec+1);
1482	<	}
1483	<	return true;
1484	<	}
1485	<
1486	<	/**
1487	<	* Returns true if a new sync event occurred since last call to
1488	<	* sync or this method, if so, updating caller's count.
1489	<	*/
1490	<	final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1491	<	long lc = w.lastEventCount;
1492	<	long ec = ensureSync();
1493	<	if (ec == lc)
1494	<	return false;
1495	<	w.lastEventCount = ec;
1496	<	return true;
1497	<	}
1498	<
1499	<	//  Parallelism maintenance
1500	<
1501	<	/**
1502	<	* Decrements running count; if too low, adds spare.
1503	<	*
1504	<	* Conceptually, all we need to do here is add or resume a
1505	<	* spare thread when one is about to block (and remove or
1506	<	* suspend it later when unblocked -- see suspendIfSpare).
1507	<	* However, implementing this idea requires coping with
1508	<	* several problems: we have imperfect information about the
1509	<	* states of threads. Some count updates can and usually do
1510	<	* lag run state changes, despite arrangements to keep them
1511	<	* accurate (for example, when possible, updating counts
1512	<	* before signalling or resuming), especially when running on
1513	<	* dynamic JVMs that don't optimize the infrequent paths that
1514	<	* update counts. Generating too many threads can make these
1515	<	* problems become worse, because excess threads are more
1516	<	* likely to be context-switched with others, slowing them all
1517	<	* down, especially if there is no work available, so all are
1518	<	* busy scanning or idling.  Also, excess spare threads can
1519	<	* only be suspended or removed when they are idle, not
1520	<	* immediately when they aren't needed. So adding threads will
1521	<	* raise parallelism level for longer than necessary.  Also,
1522	<	* FJ applications often encounter highly transient peaks when
1523	<	* many threads are blocked joining, but for less time than it
1524	<	* takes to create or resume spares.
1525	<	*
1526	<	* @param joinMe if non-null, return early if done
1527	<	* @param maintainParallelism if true, try to stay within
1528	<	* target counts, else create only to avoid starvation
1529	<	* @return true if joinMe known to be done
1530	<	*/
1531	<	final boolean preJoin(ForkJoinTask<?> joinMe,
1532	<	boolean maintainParallelism) {
1533	<	maintainParallelism &= maintainsParallelism; // overrride
1534	<	boolean dec = false;  // true when running count decremented
1535	<	while (spareStack == null || !tryResumeSpare(dec)) {
1536	<	int counts = workerCounts;
1537	<	if (dec || (dec = casWorkerCounts(counts, --counts))) {
1538	<	// CAS cheat
1539	<	if (!needSpare(counts, maintainParallelism))
1540	<	break;
1541	<	if (joinMe.status < 0)
1542	<	return true;
1543	<	if (tryAddSpare(counts))
3013	>	long startTime = System.nanoTime();
3014	>	boolean terminated = false;
3015	>	synchronized (this) {
3016	>	for (long waitTime = nanos, millis = 0L;;) {
3017	>	if (terminated = isTerminated() ||
3018	>	waitTime <= 0L ||
3019	>	(millis = unit.toMillis(waitTime)) <= 0L)
3020		break;
3021	+	wait(millis);
3022	+	waitTime = nanos - (System.nanoTime() - startTime);
3023		}
3024		}
3025	<	return false;
1548	<	}
1549	<
1550	<	/**
1551	<	* Same idea as preJoin
1552	<	*/
1553	<	final boolean preBlock(ManagedBlocker blocker,
1554	<	boolean maintainParallelism) {
1555	<	maintainParallelism &= maintainsParallelism;
1556	<	boolean dec = false;
1557	<	while (spareStack == null || !tryResumeSpare(dec)) {
1558	<	int counts = workerCounts;
1559	<	if (dec || (dec = casWorkerCounts(counts, --counts))) {
1560	<	if (!needSpare(counts, maintainParallelism))
1561	<	break;
1562	<	if (blocker.isReleasable())
1563	<	return true;
1564	<	if (tryAddSpare(counts))
1565	<	break;
1566	<	}
1567	<	}
1568	<	return false;
1569	<	}
1570	<
1571	<	/**
1572	<	* Returns true if a spare thread appears to be needed.  If
1573	<	* maintaining parallelism, returns true when the deficit in
1574	<	* running threads is more than the surplus of total threads, and
1575	<	* there is apparently some work to do.  This self-limiting rule
1576	<	* means that the more threads that have already been added, the
1577	<	* less parallelism we will tolerate before adding another.
1578	<	*
1579	<	* @param counts current worker counts
1580	<	* @param maintainParallelism try to maintain parallelism
1581	<	*/
1582	<	private boolean needSpare(int counts, boolean maintainParallelism) {
1583	<	int ps = parallelism;
1584	<	int rc = runningCountOf(counts);
1585	<	int tc = totalCountOf(counts);
1586	<	int runningDeficit = ps - rc;
1587	<	int totalSurplus = tc - ps;
1588	<	return (tc < maxPoolSize &&
1589	<	(rc == 0 || totalSurplus < 0 ||
1590	<	(maintainParallelism &&
1591	<	runningDeficit > totalSurplus &&
1592	<	ForkJoinWorkerThread.hasQueuedTasks(workers))));
1593	<	}
1594	<
1595	<	/**
1596	<	* Adds a spare worker if lock available and no more than the
1597	<	* expected numbers of threads exist.
1598	<	*
1599	<	* @return true if successful
1600	<	*/
1601	<	private boolean tryAddSpare(int expectedCounts) {
1602	<	final ReentrantLock lock = this.workerLock;
1603	<	int expectedRunning = runningCountOf(expectedCounts);
1604	<	int expectedTotal = totalCountOf(expectedCounts);
1605	<	boolean success = false;
1606	<	boolean locked = false;
1607	<	// confirm counts while locking; CAS after obtaining lock
1608	<	try {
1609	<	for (;;) {
1610	<	int s = workerCounts;
1611	<	int tc = totalCountOf(s);
1612	<	int rc = runningCountOf(s);
1613	<	if (rc > expectedRunning || tc > expectedTotal)
1614	<	break;
1615	<	if (!locked && !(locked = lock.tryLock()))
1616	<	break;
1617	<	if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1618	<	createAndStartSpare(tc);
1619	<	success = true;
1620	<	break;
1621	<	}
1622	<	}
1623	<	} finally {
1624	<	if (locked)
1625	<	lock.unlock();
1626	<	}
1627	<	return success;
1628	<	}
1629	<
1630	<	/**
1631	<	* Adds the kth spare worker. On entry, pool counts are already
1632	<	* adjusted to reflect addition.
1633	<	*/
1634	<	private void createAndStartSpare(int k) {
1635	<	ForkJoinWorkerThread w = null;
1636	<	ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1637	<	int len = ws.length;
1638	<	// Probably, we can place at slot k. If not, find empty slot
1639	<	if (k < len && ws[k] != null) {
1640	<	for (k = 0; k < len && ws[k] != null; ++k)
1641	<	;
1642	<	}
1643	<	if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1644	<	ws[k] = w;
1645	<	w.start();
1646	<	}
1647	<	else
1648	<	updateWorkerCount(-1); // adjust on failure
1649	<	signalIdleWorkers();
1650	<	}
1651	<
1652	<	/**
1653	<	* Suspends calling thread w if there are excess threads.  Called
1654	<	* only from sync.  Spares are enqueued in a Treiber stack using
1655	<	* the same WaitQueueNodes as barriers.  They are resumed mainly
1656	<	* in preJoin, but are also woken on pool events that require all
1657	<	* threads to check run state.
1658	<	*
1659	<	* @param w the caller
1660	<	*/
1661	<	private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1662	<	WaitQueueNode node = null;
1663	<	int s;
1664	<	while (parallelism < runningCountOf(s = workerCounts)) {
1665	<	if (node == null)
1666	<	node = new WaitQueueNode(0, w);
1667	<	if (casWorkerCounts(s, s-1)) { // representation-dependent
1668	<	// push onto stack
1669	<	do {} while (!casSpareStack(node.next = spareStack, node));
1670	<	// block until released by resumeSpare
1671	<	node.awaitSpareRelease();
1672	<	return true;
1673	<	}
1674	<	}
1675	<	return false;
1676	<	}
1677	<
1678	<	/**
1679	<	* Tries to pop and resume a spare thread.
1680	<	*
1681	<	* @param updateCount if true, increment running count on success
1682	<	* @return true if successful
1683	<	*/
1684	<	private boolean tryResumeSpare(boolean updateCount) {
1685	<	WaitQueueNode q;
1686	<	while ((q = spareStack) != null) {
1687	<	if (casSpareStack(q, q.next)) {
1688	<	if (updateCount)
1689	<	updateRunningCount(1);
1690	<	q.signal();
1691	<	return true;
1692	<	}
1693	<	}
1694	<	return false;
1695	<	}
1696	<
1697	<	/**
1698	<	* Pops and resumes all spare threads. Same idea as ensureSync.
1699	<	*
1700	<	* @return true if any spares released
1701	<	*/
1702	<	private boolean resumeAllSpares() {
1703	<	WaitQueueNode q;
1704	<	while ( (q = spareStack) != null) {
1705	<	if (casSpareStack(q, null)) {
1706	<	do {
1707	<	updateRunningCount(1);
1708	<	q.signal();
1709	<	} while ((q = q.next) != null);
1710	<	return true;
1711	<	}
1712	<	}
1713	<	return false;
1714	<	}
1715	<
1716	<	/**
1717	<	* Pops and shuts down excessive spare threads. Call only while
1718	<	* holding lock. This is not guaranteed to eliminate all excess
1719	<	* threads, only those suspended as spares, which are the ones
1720	<	* unlikely to be needed in the future.
1721	<	*/
1722	<	private void trimSpares() {
1723	<	int surplus = totalCountOf(workerCounts) - parallelism;
1724	<	WaitQueueNode q;
1725	<	while (surplus > 0 && (q = spareStack) != null) {
1726	<	if (casSpareStack(q, null)) {
1727	<	do {
1728	<	updateRunningCount(1);
1729	<	ForkJoinWorkerThread w = q.thread;
1730	<	if (w != null && surplus > 0 &&
1731	<	runningCountOf(workerCounts) > 0 && w.shutdown())
1732	<	--surplus;
1733	<	q.signal();
1734	<	} while ((q = q.next) != null);
1735	<	}
1736	<	}
3025	>	return terminated;
3026		}
3027
3028		/**
3029		* Interface for extending managed parallelism for tasks running
3030	<	* in ForkJoinPools. A ManagedBlocker provides two methods.
3031	<	* Method {@code isReleasable} must return true if blocking is not
3032	<	* necessary. Method {@code block} blocks the current thread if
3033	<	* necessary (perhaps internally invoking {@code isReleasable}
3034	<	* before actually blocking.).
3030	>	* in {@link ForkJoinPool}s.
3031	>	*
3032	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
3033	>	* {@code isReleasable} must return {@code true} if blocking is
3034	>	* not necessary. Method {@code block} blocks the current thread
3035	>	* if necessary (perhaps internally invoking {@code isReleasable}
3036	>	* before actually blocking). These actions are performed by any
3037	>	* thread invoking {@link ForkJoinPool#managedBlock}.  The
3038	>	* unusual methods in this API accommodate synchronizers that may,
3039	>	* but don't usually, block for long periods. Similarly, they
3040	>	* allow more efficient internal handling of cases in which
3041	>	* additional workers may be, but usually are not, needed to
3042	>	* ensure sufficient parallelism.  Toward this end,
3043	>	* implementations of method {@code isReleasable} must be amenable
3044	>	* to repeated invocation.
3045		*
3046		* <p>For example, here is a ManagedBlocker based on a
3047		* ReentrantLock:
#	Line 1760 | Line 3059 | public class ForkJoinPool extends Abstra
3059		*     return hasLock || (hasLock = lock.tryLock());
3060		*   }
3061		* }}</pre>
3062	+	*
3063	+	* <p>Here is a class that possibly blocks waiting for an
3064	+	* item on a given queue:
3065	+	*  <pre> {@code
3066	+	* class QueueTaker<E> implements ManagedBlocker {
3067	+	*   final BlockingQueue<E> queue;
3068	+	*   volatile E item = null;
3069	+	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
3070	+	*   public boolean block() throws InterruptedException {
3071	+	*     if (item == null)
3072	+	*       item = queue.take();
3073	+	*     return true;
3074	+	*   }
3075	+	*   public boolean isReleasable() {
3076	+	*     return item != null || (item = queue.poll()) != null;
3077	+	*   }
3078	+	*   public E getItem() { // call after pool.managedBlock completes
3079	+	*     return item;
3080	+	*   }
3081	+	* }}</pre>
3082		*/
3083		public static interface ManagedBlocker {
3084		/**
3085		* Possibly blocks the current thread, for example waiting for
3086		* a lock or condition.
3087		*
3088	<	* @return true if no additional blocking is necessary (i.e.,
3089	<	* if isReleasable would return true)
3088	>	* @return {@code true} if no additional blocking is necessary
3089	>	* (i.e., if isReleasable would return true)
3090		* @throws InterruptedException if interrupted while waiting
3091		* (the method is not required to do so, but is allowed to)
3092		*/
3093		boolean block() throws InterruptedException;
3094
3095		/**
3096	<	* Returns true if blocking is unnecessary.
3096	>	* Returns {@code true} if blocking is unnecessary.
3097		*/
3098		boolean isReleasable();
3099		}
3100
3101		/**
3102		* Blocks in accord with the given blocker.  If the current thread
3103	<	* is a ForkJoinWorkerThread, this method possibly arranges for a
3104	<	* spare thread to be activated if necessary to ensure parallelism
3105	<	* while the current thread is blocked.  If
1787	<	* {@code maintainParallelism} is true and the pool supports
1788	<	* it ({@link #getMaintainsParallelism}), this method attempts to
1789	<	* maintain the pool's nominal parallelism. Otherwise it activates
1790	<	* a thread only if necessary to avoid complete starvation. This
1791	<	* option may be preferable when blockages use timeouts, or are
1792	<	* almost always brief.
3103	>	* is a {@link ForkJoinWorkerThread}, this method possibly
3104	>	* arranges for a spare thread to be activated if necessary to
3105	>	* ensure sufficient parallelism while the current thread is blocked.
3106		*
3107	<	* <p> If the caller is not a ForkJoinTask, this method is behaviorally
3108	<	* equivalent to
3107	>	* <p>If the caller is not a {@link ForkJoinTask}, this method is
3108	>	* behaviorally equivalent to
3109		*  <pre> {@code
3110		* while (!blocker.isReleasable())
3111		*   if (blocker.block())
3112		*     return;
3113		* }</pre>
3114	<	* If the caller is a ForkJoinTask, then the pool may first
3115	<	* be expanded to ensure parallelism, and later adjusted.
3114	>	*
3115	>	* If the caller is a {@code ForkJoinTask}, then the pool may
3116	>	* first be expanded to ensure parallelism, and later adjusted.
3117		*
3118		* @param blocker the blocker
1805	–	* @param maintainParallelism if true and supported by this pool,
1806	–	* attempt to maintain the pool's nominal parallelism; otherwise
1807	–	* activate a thread only if necessary to avoid complete
1808	–	* starvation.
3119		* @throws InterruptedException if blocker.block did so
3120		*/
3121	<	public static void managedBlock(ManagedBlocker blocker,
1812	<	boolean maintainParallelism)
3121	>	public static void managedBlock(ManagedBlocker blocker)
3122		throws InterruptedException {
3123		Thread t = Thread.currentThread();
3124	<	ForkJoinPool pool = ((t instanceof ForkJoinWorkerThread) ?
3125	<	((ForkJoinWorkerThread) t).pool : null);
3126	<	if (!blocker.isReleasable()) {
3127	<	try {
3128	<	if (pool == null ||
3129	<	!pool.preBlock(blocker, maintainParallelism))
3130	<	awaitBlocker(blocker);
3131	<	} finally {
3132	<	if (pool != null)
3133	<	pool.updateRunningCount(1);
3124	>	if (t instanceof ForkJoinWorkerThread) {
3125	>	ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3126	>	while (!blocker.isReleasable()) { // variant of helpSignal
3127	>	WorkQueue[] ws; WorkQueue q; int m, n;
3128	>	if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3129	>	for (int i = 0; i <= m; ++i) {
3130	>	if (blocker.isReleasable())
3131	>	return;
3132	>	if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
3133	>	p.signalWork(q, n);
3134	>	if ((int)(p.ctl >> AC_SHIFT) >= 0)
3135	>	break;
3136	>	}
3137	>	}
3138	>	}
3139	>	if (p.tryCompensate()) {
3140	>	try {
3141	>	do {} while (!blocker.isReleasable() &&
3142	>	!blocker.block());
3143	>	} finally {
3144	>	p.incrementActiveCount();
3145	>	}
3146	>	break;
3147	>	}
3148		}
3149		}
3150	+	else {
3151	+	do {} while (!blocker.isReleasable() &&
3152	+	!blocker.block());
3153	+	}
3154		}
3155
3156	<	private static void awaitBlocker(ManagedBlocker blocker)
3157	<	throws InterruptedException {
3158	<	do {} while (!blocker.isReleasable() && !blocker.block());
1832	<	}
1833	<
1834	<	// AbstractExecutorService overrides
3156	>	// AbstractExecutorService overrides.  These rely on undocumented
3157	>	// fact that ForkJoinTask.adapt returns ForkJoinTasks that also
3158	>	// implement RunnableFuture.
3159
3160		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3161	<	return new AdaptedRunnable<T>(runnable, value);
3161	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3162		}
3163
3164		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3165	<	return new AdaptedCallable<T>(callable);
3165	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
3166		}
3167
3168	+	// Unsafe mechanics
3169	+	private static final sun.misc.Unsafe U;
3170	+	private static final long CTL;
3171	+	private static final long PARKBLOCKER;
3172	+	private static final int ABASE;
3173	+	private static final int ASHIFT;
3174	+	private static final long STEALCOUNT;
3175	+	private static final long PLOCK;
3176	+	private static final long INDEXSEED;
3177	+	private static final long QLOCK;
3178	+
3179	+	static {
3180	+	// Establish common pool parameters
3181	+	// TBD: limit or report ignored exceptions?
3182	+
3183	+	int par = 0;
3184	+	ForkJoinWorkerThreadFactory fac = null;
3185	+	Thread.UncaughtExceptionHandler handler = null;
3186	+	try {
3187	+	String pp = System.getProperty(propPrefix + "parallelism");
3188	+	String hp = System.getProperty(propPrefix + "exceptionHandler");
3189	+	String fp = System.getProperty(propPrefix + "threadFactory");
3190	+	if (fp != null)
3191	+	fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3192	+	getSystemClassLoader().loadClass(fp).newInstance());
3193	+	if (hp != null)
3194	+	handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3195	+	getSystemClassLoader().loadClass(hp).newInstance());
3196	+	if (pp != null)
3197	+	par = Integer.parseInt(pp);
3198	+	} catch (Exception ignore) {
3199	+	}
3200
3201	<	// Temporary Unsafe mechanics for preliminary release
1846	<	private static Unsafe getUnsafe() throws Throwable {
3201	>	int s; // initialize field offsets for CAS etc
3202		try {
3203	<	return Unsafe.getUnsafe();
3203	>	U = getUnsafe();
3204	>	Class<?> k = ForkJoinPool.class;
3205	>	CTL = U.objectFieldOffset
3206	>	(k.getDeclaredField("ctl"));
3207	>	STEALCOUNT = U.objectFieldOffset
3208	>	(k.getDeclaredField("stealCount"));
3209	>	PLOCK = U.objectFieldOffset
3210	>	(k.getDeclaredField("plock"));
3211	>	INDEXSEED = U.objectFieldOffset
3212	>	(k.getDeclaredField("indexSeed"));
3213	>	Class<?> tk = Thread.class;
3214	>	PARKBLOCKER = U.objectFieldOffset
3215	>	(tk.getDeclaredField("parkBlocker"));
3216	>	Class<?> wk = WorkQueue.class;
3217	>	QLOCK = U.objectFieldOffset
3218	>	(wk.getDeclaredField("qlock"));
3219	>	Class<?> ak = ForkJoinTask[].class;
3220	>	ABASE = U.arrayBaseOffset(ak);
3221	>	s = U.arrayIndexScale(ak);
3222	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3223	>	} catch (Exception e) {
3224	>	throw new Error(e);
3225	>	}
3226	>	if ((s & (s-1)) != 0)
3227	>	throw new Error("data type scale not a power of two");
3228	>
3229	>	/*
3230	>	* For extra caution, computations to set up pool state are
3231	>	* here; the constructor just assigns these values to fields.
3232	>	*/
3233	>	ForkJoinWorkerThreadFactory defaultFac =
3234	>	defaultForkJoinWorkerThreadFactory =
3235	>	new DefaultForkJoinWorkerThreadFactory();
3236	>	if (fac == null)
3237	>	fac = defaultFac;
3238	>	if (par <= 0)
3239	>	par = Runtime.getRuntime().availableProcessors();
3240	>	if (par > MAX_CAP)
3241	>	par = MAX_CAP;
3242	>	long np = (long)(-par); // precompute initial ctl value
3243	>	long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
3244	>
3245	>	commonPoolParallelism = par;
3246	>	commonPool = new ForkJoinPool(par, ct, fac, handler);
3247	>	modifyThreadPermission = new RuntimePermission("modifyThread");
3248	>	submitters = new ThreadLocal<Submitter>();
3249	>	}
3250	>
3251	>	/**
3252	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
3253	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
3254	>	* into a jdk.
3255	>	*
3256	>	* @return a sun.misc.Unsafe
3257	>	*/
3258	>	private static sun.misc.Unsafe getUnsafe() {
3259	>	try {
3260	>	return sun.misc.Unsafe.getUnsafe();
3261		} catch (SecurityException se) {
3262		try {
3263		return java.security.AccessController.doPrivileged
3264	<	(new java.security.PrivilegedExceptionAction<Unsafe>() {
3265	<	public Unsafe run() throws Exception {
3266	<	return getUnsafePrivileged();
3264	>	(new java.security
3265	>	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3266	>	public sun.misc.Unsafe run() throws Exception {
3267	>	java.lang.reflect.Field f = sun.misc
3268	>	.Unsafe.class.getDeclaredField("theUnsafe");
3269	>	f.setAccessible(true);
3270	>	return (sun.misc.Unsafe) f.get(null);
3271		}});
3272		} catch (java.security.PrivilegedActionException e) {
3273	<	throw e.getCause();
3273	>	throw new RuntimeException("Could not initialize intrinsics",
3274	>	e.getCause());
3275		}
3276		}
3277		}
3278
1862	–	private static Unsafe getUnsafePrivileged()
1863	–	throws NoSuchFieldException, IllegalAccessException {
1864	–	Field f = Unsafe.class.getDeclaredField("theUnsafe");
1865	–	f.setAccessible(true);
1866	–	return (Unsafe) f.get(null);
1867	–	}
1868	–
1869	–	private static long fieldOffset(String fieldName)
1870	–	throws NoSuchFieldException {
1871	–	return UNSAFE.objectFieldOffset
1872	–	(ForkJoinPool.class.getDeclaredField(fieldName));
1873	–	}
1874	–
1875	–	static final Unsafe UNSAFE;
1876	–	static final long eventCountOffset;
1877	–	static final long workerCountsOffset;
1878	–	static final long runControlOffset;
1879	–	static final long syncStackOffset;
1880	–	static final long spareStackOffset;
1881	–
1882	–	static {
1883	–	try {
1884	–	UNSAFE = getUnsafe();
1885	–	eventCountOffset = fieldOffset("eventCount");
1886	–	workerCountsOffset = fieldOffset("workerCounts");
1887	–	runControlOffset = fieldOffset("runControl");
1888	–	syncStackOffset = fieldOffset("syncStack");
1889	–	spareStackOffset = fieldOffset("spareStack");
1890	–	} catch (Throwable e) {
1891	–	throw new RuntimeException("Could not initialize intrinsics", e);
1892	–	}
1893	–	}
1894	–
1895	–	private boolean casEventCount(long cmp, long val) {
1896	–	return UNSAFE.compareAndSwapLong(this, eventCountOffset, cmp, val);
1897	–	}
1898	–	private boolean casWorkerCounts(int cmp, int val) {
1899	–	return UNSAFE.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1900	–	}
1901	–	private boolean casRunControl(int cmp, int val) {
1902	–	return UNSAFE.compareAndSwapInt(this, runControlOffset, cmp, val);
1903	–	}
1904	–	private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1905	–	return UNSAFE.compareAndSwapObject(this, spareStackOffset, cmp, val);
1906	–	}
1907	–	private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1908	–	return UNSAFE.compareAndSwapObject(this, syncStackOffset, cmp, val);
1909	–	}
3279		}

Diff Legend

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