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 1.10 by jsr166, Mon Jul 20 23:07:43 2009 UTC vs.
Revision 1.131 by jsr166, Tue Aug 14 06:00:55 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	>	import java.util.ArrayList;
9	>	import java.util.Arrays;
10	>	import java.util.Collection;
11	>	import java.util.Collections;
12	>	import java.util.List;
13	>	import java.util.Random;
14	>	import java.util.concurrent.AbstractExecutorService;
15	>	import java.util.concurrent.Callable;
16	>	import java.util.concurrent.ExecutorService;
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	>	import java.util.concurrent.atomic.AtomicInteger;
22	>	import java.util.concurrent.atomic.AtomicLong;
23	>	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
24	>	import java.util.concurrent.locks.Condition;
25
26		/**
27	<	* An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
28	<	* ForkJoinPool provides the entry point for submissions from
29	<	* non-ForkJoinTasks, as well as management and monitoring operations.
30	<	* Normally a single ForkJoinPool is used for a large number of
20	<	* submitted tasks. Otherwise, use would not usually outweigh the
21	<	* construction and bookkeeping overhead of creating a large set of
22	<	* threads.
27	>	* An {@link ExecutorService} for running {@link ForkJoinTask}s.
28	>	* A {@code ForkJoinPool} provides the entry point for submissions
29	>	* from non-{@code ForkJoinTask} clients, as well as management and
30	>	* monitoring operations.
31		*
32	<	* <p>ForkJoinPools differ from other kinds of Executors mainly in
33	<	* that they provide <em>work-stealing</em>: all threads in the pool
34	<	* attempt to find and execute subtasks created by other active tasks
35	<	* (eventually blocking if none exist). This makes them efficient when
36	<	* most tasks spawn other subtasks (as do most ForkJoinTasks), as well
37	<	* as the mixed execution of some plain Runnable- or Callable- based
38	<	* activities along with ForkJoinTasks. When setting
39	<	* {@code setAsyncMode}, a ForkJoinPools may also be appropriate for
40	<	* use with fine-grained tasks that are never joined. Otherwise, other
41	<	* ExecutorService implementations are typically more appropriate
42	<	* choices.
32	>	* <p>A {@code ForkJoinPool} differs from other kinds of {@link
33	>	* ExecutorService} mainly by virtue of employing
34	>	* <em>work-stealing</em>: all threads in the pool attempt to find and
35	>	* execute tasks submitted to the pool and/or created by other active
36	>	* tasks (eventually blocking waiting for work if none exist). This
37	>	* enables efficient processing when most tasks spawn other subtasks
38	>	* (as do most {@code ForkJoinTask}s), as well as when many small
39	>	* tasks are submitted to the pool from external clients.  Especially
40	>	* when setting <em>asyncMode</em> to true in constructors, {@code
41	>	* ForkJoinPool}s may also be appropriate for use with event-style
42	>	* tasks that are never joined.
43		*
44	<	* <p>A ForkJoinPool may be constructed with a given parallelism level
45	<	* (target pool size), which it attempts to maintain by dynamically
46	<	* adding, suspending, or resuming threads, even if some tasks are
47	<	* waiting to join others. However, no such adjustments are performed
48	<	* in the face of blocked IO or other unmanaged synchronization. The
49	<	* nested {@code ManagedBlocker} interface enables extension of
50	<	* the kinds of synchronization accommodated.  The target parallelism
51	<	* level may also be changed dynamically ({@code setParallelism})
52	<	* and thread construction can be limited using methods
45	<	* {@code setMaximumPoolSize} and/or
46	<	* {@code setMaintainsParallelism}.
44	>	* <p>A {@code ForkJoinPool} is constructed with a given target
45	>	* parallelism level; by default, equal to the number of available
46	>	* processors. The pool attempts to maintain enough active (or
47	>	* available) threads by dynamically adding, suspending, or resuming
48	>	* internal worker threads, even if some tasks are stalled waiting to
49	>	* join others. However, no such adjustments are guaranteed in the
50	>	* face of blocked IO or other unmanaged synchronization. The nested
51	>	* {@link ManagedBlocker} interface enables extension of the kinds of
52	>	* synchronization accommodated.
53		*
54		* <p>In addition to execution and lifecycle control methods, this
55		* class provides status check methods (for example
56	<	* {@code getStealCount}) that are intended to aid in developing,
56	>	* {@link #getStealCount}) that are intended to aid in developing,
57		* tuning, and monitoring fork/join applications. Also, method
58	<	* {@code toString} returns indications of pool state in a
58	>	* {@link #toString} returns indications of pool state in a
59		* convenient form for informal monitoring.
60		*
61	+	* <p> As is the case with other ExecutorServices, there are three
62	+	* main task execution methods summarized in the following table.
63	+	* These are designed to be used primarily by clients not already
64	+	* engaged in fork/join computations in the current pool.  The main
65	+	* forms of these methods accept instances of {@code ForkJoinTask},
66	+	* but overloaded forms also allow mixed execution of plain {@code
67	+	* Runnable}- or {@code Callable}- based activities as well.  However,
68	+	* tasks that are already executing in a pool should normally instead
69	+	* use the within-computation forms listed in the table unless using
70	+	* async event-style tasks that are not usually joined, in which case
71	+	* there is little difference among choice of methods.
72	+	*
73	+	* <table BORDER CELLPADDING=3 CELLSPACING=1>
74	+	*  <tr>
75	+	*    <td></td>
76	+	*    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
77	+	*    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
78	+	*  </tr>
79	+	*  <tr>
80	+	*    <td> <b>Arrange async execution</td>
81	+	*    <td> {@link #execute(ForkJoinTask)}</td>
82	+	*    <td> {@link ForkJoinTask#fork}</td>
83	+	*  </tr>
84	+	*  <tr>
85	+	*    <td> <b>Await and obtain result</td>
86	+	*    <td> {@link #invoke(ForkJoinTask)}</td>
87	+	*    <td> {@link ForkJoinTask#invoke}</td>
88	+	*  </tr>
89	+	*  <tr>
90	+	*    <td> <b>Arrange exec and obtain Future</td>
91	+	*    <td> {@link #submit(ForkJoinTask)}</td>
92	+	*    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
93	+	*  </tr>
94	+	* </table>
95	+	*
96	+	* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
97	+	* used for all parallel task execution in a program or subsystem.
98	+	* Otherwise, use would not usually outweigh the construction and
99	+	* bookkeeping overhead of creating a large set of threads. For
100	+	* example, a common pool could be used for the {@code SortTasks}
101	+	* illustrated in {@link RecursiveAction}. Because {@code
102	+	* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
103	+	* daemon} mode, there is typically no need to explicitly {@link
104	+	* #shutdown} such a pool upon program exit.
105	+	*
106	+	*  <pre> {@code
107	+	* static final ForkJoinPool mainPool = new ForkJoinPool();
108	+	* ...
109	+	* public void sort(long[] array) {
110	+	*   mainPool.invoke(new SortTask(array, 0, array.length));
111	+	* }}</pre>
112	+	*
113		* <p><b>Implementation notes</b>: This implementation restricts the
114		* maximum number of running threads to 32767. Attempts to create
115	<	* pools with greater than the maximum result in
116	<	* IllegalArgumentExceptions.
115	>	* pools with greater than the maximum number result in
116	>	* {@code IllegalArgumentException}.
117	>	*
118	>	* <p>This implementation rejects submitted tasks (that is, by throwing
119	>	* {@link RejectedExecutionException}) only when the pool is shut down
120	>	* or internal resources have been exhausted.
121	>	*
122	>	* @since 1.7
123	>	* @author Doug Lea
124		*/
125		public class ForkJoinPool extends AbstractExecutorService {
126
127		/*
128	<	* See the extended comments interspersed below for design,
129	<	* rationale, and walkthroughs.
128	>	* Implementation Overview
129	>	*
130	>	* This class and its nested classes provide the main
131	>	* functionality and control for a set of worker threads:
132	>	* Submissions from non-FJ threads enter into submission queues.
133	>	* Workers take these tasks and typically split them into subtasks
134	>	* that may be stolen by other workers.  Preference rules give
135	>	* first priority to processing tasks from their own queues (LIFO
136	>	* or FIFO, depending on mode), then to randomized FIFO steals of
137	>	* tasks in other queues.
138	>	*
139	>	* WorkQueues
140	>	* ==========
141	>	*
142	>	* Most operations occur within work-stealing queues (in nested
143	>	* class WorkQueue).  These are special forms of Deques that
144	>	* support only three of the four possible end-operations -- push,
145	>	* pop, and poll (aka steal), under the further constraints that
146	>	* push and pop are called only from the owning thread (or, as
147	>	* extended here, under a lock), while poll may be called from
148	>	* other threads.  (If you are unfamiliar with them, you probably
149	>	* want to read Herlihy and Shavit's book "The Art of
150	>	* Multiprocessor programming", chapter 16 describing these in
151	>	* more detail before proceeding.)  The main work-stealing queue
152	>	* design is roughly similar to those in the papers "Dynamic
153	>	* Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
154	>	* (http://research.sun.com/scalable/pubs/index.html) and
155	>	* "Idempotent work stealing" by Michael, Saraswat, and Vechev,
156	>	* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
157	>	* The main differences ultimately stem from GC requirements that
158	>	* we null out taken slots as soon as we can, to maintain as small
159	>	* a footprint as possible even in programs generating huge
160	>	* numbers of tasks. To accomplish this, we shift the CAS
161	>	* arbitrating pop vs poll (steal) from being on the indices
162	>	* ("base" and "top") to the slots themselves.  So, both a
163	>	* successful pop and poll mainly entail a CAS of a slot from
164	>	* non-null to null.  Because we rely on CASes of references, we
165	>	* do not need tag bits on base or top.  They are simple ints as
166	>	* used in any circular array-based queue (see for example
167	>	* ArrayDeque).  Updates to the indices must still be ordered in a
168	>	* way that guarantees that top == base means the queue is empty,
169	>	* but otherwise may err on the side of possibly making the queue
170	>	* appear nonempty when a push, pop, or poll have not fully
171	>	* committed. Note that this means that the poll operation,
172	>	* considered individually, is not wait-free. One thief cannot
173	>	* successfully continue until another in-progress one (or, if
174	>	* previously empty, a push) completes.  However, in the
175	>	* aggregate, we ensure at least probabilistic non-blockingness.
176	>	* If an attempted steal fails, a thief always chooses a different
177	>	* random victim target to try next. So, in order for one thief to
178	>	* progress, it suffices for any in-progress poll or new push on
179	>	* any empty queue to complete. (This is why we normally use
180	>	* method pollAt and its variants that try once at the apparent
181	>	* base index, else consider alternative actions, rather than
182	>	* method poll.)
183	>	*
184	>	* This approach also enables support of a user mode in which local
185	>	* task processing is in FIFO, not LIFO order, simply by using
186	>	* poll rather than pop.  This can be useful in message-passing
187	>	* frameworks in which tasks are never joined.  However neither
188	>	* mode considers affinities, loads, cache localities, etc, so
189	>	* rarely provide the best possible performance on a given
190	>	* machine, but portably provide good throughput by averaging over
191	>	* these factors.  (Further, even if we did try to use such
192	>	* information, we do not usually have a basis for exploiting it.
193	>	* For example, some sets of tasks profit from cache affinities,
194	>	* but others are harmed by cache pollution effects.)
195	>	*
196	>	* WorkQueues are also used in a similar way for tasks submitted
197	>	* to the pool. We cannot mix these tasks in the same queues used
198	>	* for work-stealing (this would contaminate lifo/fifo
199	>	* processing). Instead, we loosely associate submission queues
200	>	* with submitting threads, using a form of hashing.  The
201	>	* ThreadLocal Submitter class contains a value initially used as
202	>	* a hash code for choosing existing queues, but may be randomly
203	>	* repositioned upon contention with other submitters.  In
204	>	* essence, submitters act like workers except that they never
205	>	* take tasks, and they are multiplexed on to a finite number of
206	>	* shared work queues. However, classes are set up so that future
207	>	* extensions could allow submitters to optionally help perform
208	>	* tasks as well. Insertion of tasks in shared mode requires a
209	>	* lock (mainly to protect in the case of resizing) but we use
210	>	* only a simple spinlock (using bits in field runState), because
211	>	* submitters encountering a busy queue move on to try or create
212	>	* other queues -- they block only when creating and registering
213	>	* 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 "runState" contains 32 bits needed to register and
236	>	* deregister WorkQueues, as well as to enable shutdown. It is
237	>	* only modified under a lock (normally briefly held, but
238	>	* occasionally protecting allocations and resizings) but even
239	>	* when locked remains available to check consistency.
240	>	*
241	>	* Recording WorkQueues.  WorkQueues are recorded in the
242	>	* "workQueues" array that is created upon pool construction and
243	>	* expanded if necessary.  Updates to the array while recording
244	>	* new workers and unrecording terminated ones are protected from
245	>	* each other by a lock but the array is otherwise concurrently
246	>	* readable, and accessed directly.  To simplify index-based
247	>	* operations, the array size is always a power of two, and all
248	>	* readers must tolerate null slots. Shared (submission) queues
249	>	* are at even indices, worker queues at odd indices. Grouping
250	>	* them together in this way simplifies and speeds up task
251	>	* scanning.
252	>	*
253	>	* All worker thread creation is on-demand, triggered by task
254	>	* submissions, replacement of terminated workers, and/or
255	>	* compensation for blocked workers. However, all other support
256	>	* code is set up to work with other policies.  To ensure that we
257	>	* do not hold on to worker references that would prevent GC, ALL
258	>	* accesses to workQueues are via indices into the workQueues
259	>	* array (which is one source of some of the messy code
260	>	* constructions here). In essence, the workQueues array serves as
261	>	* a weak reference mechanism. Thus for example the wait queue
262	>	* field of ctl stores indices, not references.  Access to the
263	>	* workQueues in associated methods (for example signalWork) must
264	>	* both index-check and null-check the IDs. All such accesses
265	>	* ignore bad IDs by returning out early from what they are doing,
266	>	* since this can only be associated with termination, in which
267	>	* case it is OK to give up.  All uses of the workQueues array
268	>	* also check that it is non-null (even if previously
269	>	* non-null). This allows nulling during termination, which is
270	>	* currently not necessary, but remains an option for
271	>	* resource-revocation-based shutdown schemes. It also helps
272	>	* reduce JIT issuance of uncommon-trap code, which tends to
273	>	* unnecessarily complicate control flow in some methods.
274	>	*
275	>	* Event Queuing. Unlike HPC work-stealing frameworks, we cannot
276	>	* let workers spin indefinitely scanning for tasks when none can
277	>	* be found immediately, and we cannot start/resume workers unless
278	>	* there appear to be tasks available.  On the other hand, we must
279	>	* quickly prod them into action when new tasks are submitted or
280	>	* generated. In many usages, ramp-up time to activate workers is
281	>	* the main limiting factor in overall performance (this is
282	>	* compounded at program start-up by JIT compilation and
283	>	* allocation). So we try to streamline this as much as possible.
284	>	* We park/unpark workers after placing in an event wait queue
285	>	* when they cannot find work. This "queue" is actually a simple
286	>	* Treiber stack, headed by the "id" field of ctl, plus a 15bit
287	>	* counter value (that reflects the number of times a worker has
288	>	* been inactivated) to avoid ABA effects (we need only as many
289	>	* version numbers as worker threads). Successors are held in
290	>	* field WorkQueue.nextWait.  Queuing deals with several intrinsic
291	>	* races, mainly that a task-producing thread can miss seeing (and
292	>	* signalling) another thread that gave up looking for work but
293	>	* has not yet entered the wait queue. We solve this by requiring
294	>	* a full sweep of all workers (via repeated calls to method
295	>	* scan()) both before and after a newly waiting worker is added
296	>	* to the wait queue. During a rescan, the worker might release
297	>	* some other queued worker rather than itself, which has the same
298	>	* net effect. Because enqueued workers may actually be rescanning
299	>	* rather than waiting, we set and clear the "parker" field of
300	>	* WorkQueues to reduce unnecessary calls to unpark.  (This
301	>	* requires a secondary recheck to avoid missed signals.)  Note
302	>	* the unusual conventions about Thread.interrupts surrounding
303	>	* parking and other blocking: Because interrupts are used solely
304	>	* to alert threads to check termination, which is checked anyway
305	>	* upon blocking, we clear status (using Thread.interrupted)
306	>	* before any call to park, so that park does not immediately
307	>	* return due to status being set via some other unrelated call to
308	>	* interrupt in user code.
309	>	*
310	>	* Signalling.  We create or wake up workers only when there
311	>	* appears to be at least one task they might be able to find and
312	>	* execute.  When a submission is added or another worker adds a
313	>	* task to a queue that previously had fewer than two tasks, they
314	>	* signal waiting workers (or trigger creation of new ones if
315	>	* fewer than the given parallelism level -- see signalWork).
316	>	* These primary signals are buttressed by signals during rescans;
317	>	* together these cover the signals needed in cases when more
318	>	* tasks are pushed but untaken, and improve performance compared
319	>	* to having one thread wake up all workers.
320	>	*
321	>	* Trimming workers. To release resources after periods of lack of
322	>	* use, a worker starting to wait when the pool is quiescent will
323	>	* time out and terminate if the pool has remained quiescent for
324	>	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
325	>	* terminating all workers after long periods of non-use.
326	>	*
327	>	* Shutdown and Termination. A call to shutdownNow atomically sets
328	>	* a runState bit and then (non-atomically) sets each worker's
329	>	* runState status, cancels all unprocessed tasks, and wakes up
330	>	* all waiting workers.  Detecting whether termination should
331	>	* commence after a non-abrupt shutdown() call requires more work
332	>	* and bookkeeping. We need consensus about quiescence (i.e., that
333	>	* there is no more work). The active count provides a primary
334	>	* indication but non-abrupt shutdown still requires a rechecking
335	>	* scan for any workers that are inactive but not queued.
336	>	*
337	>	* Joining Tasks
338	>	* =============
339	>	*
340	>	* Any of several actions may be taken when one worker is waiting
341	>	* to join a task stolen (or always held) by another.  Because we
342	>	* are multiplexing many tasks on to a pool of workers, we can't
343	>	* just let them block (as in Thread.join).  We also cannot just
344	>	* reassign the joiner's run-time stack with another and replace
345	>	* it later, which would be a form of "continuation", that even if
346	>	* possible is not necessarily a good idea since we sometimes need
347	>	* both an unblocked task and its continuation to progress.
348	>	* Instead we combine two tactics:
349	>	*
350	>	*   Helping: Arranging for the joiner to execute some task that it
351	>	*      would be running if the steal had not occurred.
352	>	*
353	>	*   Compensating: Unless there are already enough live threads,
354	>	*      method tryCompensate() may create or re-activate a spare
355	>	*      thread to compensate for blocked joiners until they unblock.
356	>	*
357	>	* A third form (implemented in tryRemoveAndExec and
358	>	* tryPollForAndExec) amounts to helping a hypothetical
359	>	* compensator: If we can readily tell that a possible action of a
360	>	* compensator is to steal and execute the task being joined, the
361	>	* joining thread can do so directly, without the need for a
362	>	* compensation thread (although at the expense of larger run-time
363	>	* stacks, but the tradeoff is typically worthwhile).
364	>	*
365	>	* The ManagedBlocker extension API can't use helping so relies
366	>	* only on compensation in method awaitBlocker.
367	>	*
368	>	* The algorithm in tryHelpStealer entails a form of "linear"
369	>	* helping: Each worker records (in field currentSteal) the most
370	>	* recent task it stole from some other worker. Plus, it records
371	>	* (in field currentJoin) the task it is currently actively
372	>	* joining. Method tryHelpStealer uses these markers to try to
373	>	* find a worker to help (i.e., steal back a task from and execute
374	>	* it) that could hasten completion of the actively joined task.
375	>	* In essence, the joiner executes a task that would be on its own
376	>	* local deque had the to-be-joined task not been stolen. This may
377	>	* be seen as a conservative variant of the approach in Wagner &
378	>	* Calder "Leapfrogging: a portable technique for implementing
379	>	* efficient futures" SIGPLAN Notices, 1993
380	>	* (http://portal.acm.org/citation.cfm?id=155354). It differs in
381	>	* that: (1) We only maintain dependency links across workers upon
382	>	* steals, rather than use per-task bookkeeping.  This sometimes
383	>	* requires a linear scan of workQueues array to locate stealers,
384	>	* but often doesn't because stealers leave hints (that may become
385	>	* stale/wrong) of where to locate them.  A stealHint is only a
386	>	* hint because a worker might have had multiple steals and the
387	>	* hint records only one of them (usually the most current).
388	>	* Hinting isolates cost to when it is needed, rather than adding
389	>	* to per-task overhead.  (2) It is "shallow", ignoring nesting
390	>	* and potentially cyclic mutual steals.  (3) It is intentionally
391	>	* racy: field currentJoin is updated only while actively joining,
392	>	* which means that we miss links in the chain during long-lived
393	>	* tasks, GC stalls etc (which is OK since blocking in such cases
394	>	* is usually a good idea).  (4) We bound the number of attempts
395	>	* to find work (see MAX_HELP) and fall back to suspending the
396	>	* worker and if necessary replacing it with another.
397	>	*
398	>	* It is impossible to keep exactly the target parallelism number
399	>	* of threads running at any given time.  Determining the
400	>	* existence of conservatively safe helping targets, the
401	>	* availability of already-created spares, and the apparent need
402	>	* to create new spares are all racy, so we rely on multiple
403	>	* retries of each.  Compensation in the apparent absence of
404	>	* helping opportunities is challenging to control on JVMs, where
405	>	* GC and other activities can stall progress of tasks that in
406	>	* turn stall out many other dependent tasks, without us being
407	>	* able to determine whether they will ever require compensation.
408	>	* Even though work-stealing otherwise encounters little
409	>	* degradation in the presence of more threads than cores,
410	>	* aggressively adding new threads in such cases entails risk of
411	>	* unwanted positive feedback control loops in which more threads
412	>	* cause more dependent stalls (as well as delayed progress of
413	>	* unblocked threads to the point that we know they are available)
414	>	* leading to more situations requiring more threads, and so
415	>	* on. This aspect of control can be seen as an (analytically
416	>	* intractable) game with an opponent that may choose the worst
417	>	* (for us) active thread to stall at any time.  We take several
418	>	* precautions to bound losses (and thus bound gains), mainly in
419	>	* methods tryCompensate and awaitJoin: (1) We only try
420	>	* compensation after attempting enough helping steps (measured
421	>	* via counting and timing) that we have already consumed the
422	>	* estimated cost of creating and activating a new thread.  (2) We
423	>	* allow up to 50% of threads to be blocked before initially
424	>	* adding any others, and unless completely saturated, check that
425	>	* some work is available for a new worker before adding. Also, we
426	>	* create up to only 50% more threads until entering a mode that
427	>	* only adds a thread if all others are possibly blocked.  All
428	>	* together, this means that we might be half as fast to react,
429	>	* and create half as many threads as possible in the ideal case,
430	>	* but present vastly fewer anomalies in all other cases compared
431	>	* to both more aggressive and more conservative alternatives.
432	>	*
433	>	* Style notes: There is a lot of representation-level coupling
434	>	* among classes ForkJoinPool, ForkJoinWorkerThread, and
435	>	* ForkJoinTask.  The fields of WorkQueue maintain data structures
436	>	* managed by ForkJoinPool, so are directly accessed.  There is
437	>	* little point trying to reduce this, since any associated future
438	>	* changes in representations will need to be accompanied by
439	>	* algorithmic changes anyway. Several methods intrinsically
440	>	* sprawl because they must accumulate sets of consistent reads of
441	>	* volatiles held in local variables.  Methods signalWork() and
442	>	* scan() are the main bottlenecks, so are especially heavily
443	>	* micro-optimized/mangled.  There are lots of inline assignments
444	>	* (of form "while ((local = field) != 0)") which are usually the
445	>	* simplest way to ensure the required read orderings (which are
446	>	* sometimes critical). This leads to a "C"-like style of listing
447	>	* declarations of these locals at the heads of methods or blocks.
448	>	* There are several occurrences of the unusual "do {} while
449	>	* (!cas...)"  which is the simplest way to force an update of a
450	>	* CAS'ed variable. There are also other coding oddities that help
451	>	* some methods perform reasonably even when interpreted (not
452	>	* compiled).
453	>	*
454	>	* The order of declarations in this file is:
455	>	* (1) Static utility functions
456	>	* (2) Nested (static) classes
457	>	* (3) Static fields
458	>	* (4) Fields, along with constants used when unpacking some of them
459	>	* (5) Internal control methods
460	>	* (6) Callbacks and other support for ForkJoinTask methods
461	>	* (7) Exported methods
462	>	* (8) Static block initializing statics in minimally dependent order
463		*/
464
465	<	/** Mask for packing and unpacking shorts */
466	<	private static final int  shortMask = 0xffff;
465	>	// Static utilities
466	>
467	>	/**
468	>	* If there is a security manager, makes sure caller has
469	>	* permission to modify threads.
470	>	*/
471	>	private static void checkPermission() {
472	>	SecurityManager security = System.getSecurityManager();
473	>	if (security != null)
474	>	security.checkPermission(modifyThreadPermission);
475	>	}
476
477	<	/** Max pool size -- must be a power of two minus 1 */
71	<	private static final int MAX_THREADS =  0x7FFF;
477	>	// Nested classes
478
479		/**
480	<	* Factory for creating new ForkJoinWorkerThreads.  A
481	<	* ForkJoinWorkerThreadFactory must be defined and used for
482	<	* ForkJoinWorkerThread subclasses that extend base functionality
483	<	* or initialize threads with different contexts.
480	>	* Factory for creating new {@link ForkJoinWorkerThread}s.
481	>	* A {@code ForkJoinWorkerThreadFactory} must be defined and used
482	>	* for {@code ForkJoinWorkerThread} subclasses that extend base
483	>	* functionality or initialize threads with different contexts.
484		*/
485		public static interface ForkJoinWorkerThreadFactory {
486		/**
487		* Returns a new worker thread operating in the given pool.
488		*
489		* @param pool the pool this thread works in
490	<	* @throws NullPointerException if pool is null;
490	>	* @throws NullPointerException if the pool is null
491		*/
492		public ForkJoinWorkerThread newThread(ForkJoinPool pool);
493		}
494
495		/**
496	<	* Default ForkJoinWorkerThreadFactory implementation, creates a
496	>	* Default ForkJoinWorkerThreadFactory implementation; creates a
497		* new ForkJoinWorkerThread.
498		*/
499	<	static class  DefaultForkJoinWorkerThreadFactory
499	>	static class DefaultForkJoinWorkerThreadFactory
500		implements ForkJoinWorkerThreadFactory {
501		public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
502	<	try {
97	<	return new ForkJoinWorkerThread(pool);
98	<	} catch (OutOfMemoryError oom)  {
99	<	return null;
100	<	}
502	>	return new ForkJoinWorkerThread(pool);
503		}
504		}
505
506		/**
507	<	* Creates a new ForkJoinWorkerThread. This factory is used unless
508	<	* overridden in ForkJoinPool constructors.
507	>	* A simple non-reentrant lock used for exclusion when managing
508	>	* queues and workers. We use a custom lock so that we can readily
509	>	* probe lock state in constructions that check among alternative
510	>	* actions. The lock is normally only very briefly held, and
511	>	* sometimes treated as a spinlock, but other usages block to
512	>	* reduce overall contention in those cases where locked code
513	>	* bodies perform allocation/resizing.
514		*/
515	<	public static final ForkJoinWorkerThreadFactory
516	<	defaultForkJoinWorkerThreadFactory =
517	<	new DefaultForkJoinWorkerThreadFactory();
518	<
519	<	/**
520	<	* Permission required for callers of methods that may start or
521	<	* kill threads.
515	>	static final class Mutex extends AbstractQueuedSynchronizer {
516	>	public final boolean tryAcquire(int ignore) {
517	>	return compareAndSetState(0, 1);
518	>	}
519	>	public final boolean tryRelease(int ignore) {
520	>	setState(0);
521	>	return true;
522	>	}
523	>	public final void lock() { acquire(0); }
524	>	public final void unlock() { release(0); }
525	>	public final boolean isHeldExclusively() { return getState() == 1; }
526	>	public final Condition newCondition() { return new ConditionObject(); }
527	>	}
528	>
529	>	/**
530	>	* Class for artificial tasks that are used to replace the target
531	>	* of local joins if they are removed from an interior queue slot
532	>	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
533	>	* actually do anything beyond having a unique identity.
534	>	*/
535	>	static final class EmptyTask extends ForkJoinTask<Void> {
536	>	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
537	>	public final Void getRawResult() { return null; }
538	>	public final void setRawResult(Void x) {}
539	>	public final boolean exec() { return true; }
540	>	}
541	>
542	>	/**
543	>	* Queues supporting work-stealing as well as external task
544	>	* submission. See above for main rationale and algorithms.
545	>	* Implementation relies heavily on "Unsafe" intrinsics
546	>	* and selective use of "volatile":
547	>	*
548	>	* Field "base" is the index (mod array.length) of the least valid
549	>	* queue slot, which is always the next position to steal (poll)
550	>	* from if nonempty. Reads and writes require volatile orderings
551	>	* but not CAS, because updates are only performed after slot
552	>	* CASes.
553	>	*
554	>	* Field "top" is the index (mod array.length) of the next queue
555	>	* slot to push to or pop from. It is written only by owner thread
556	>	* for push, or under lock for trySharedPush, and accessed by
557	>	* other threads only after reading (volatile) base.  Both top and
558	>	* base are allowed to wrap around on overflow, but (top - base)
559	>	* (or more commonly -(base - top) to force volatile read of base
560	>	* before top) still estimates size.
561	>	*
562	>	* The array slots are read and written using the emulation of
563	>	* volatiles/atomics provided by Unsafe. Insertions must in
564	>	* general use putOrderedObject as a form of releasing store to
565	>	* ensure that all writes to the task object are ordered before
566	>	* its publication in the queue. (Although we can avoid one case
567	>	* of this when locked in trySharedPush.) All removals entail a
568	>	* CAS to null.  The array is always a power of two. To ensure
569	>	* safety of Unsafe array operations, all accesses perform
570	>	* explicit null checks and implicit bounds checks via
571	>	* power-of-two masking.
572	>	*
573	>	* In addition to basic queuing support, this class contains
574	>	* fields described elsewhere to control execution. It turns out
575	>	* to work better memory-layout-wise to include them in this
576	>	* class rather than a separate class.
577	>	*
578	>	* Performance on most platforms is very sensitive to placement of
579	>	* instances of both WorkQueues and their arrays -- we absolutely
580	>	* do not want multiple WorkQueue instances or multiple queue
581	>	* arrays sharing cache lines. (It would be best for queue objects
582	>	* and their arrays to share, but there is nothing available to
583	>	* help arrange that).  Unfortunately, because they are recorded
584	>	* in a common array, WorkQueue instances are often moved to be
585	>	* adjacent by garbage collectors. To reduce impact, we use field
586	>	* padding that works OK on common platforms; this effectively
587	>	* trades off slightly slower average field access for the sake of
588	>	* avoiding really bad worst-case access. (Until better JVM
589	>	* support is in place, this padding is dependent on transient
590	>	* properties of JVM field layout rules.)  We also take care in
591	>	* allocating, sizing and resizing the array. Non-shared queue
592	>	* arrays are initialized (via method growArray) by workers before
593	>	* use. Others are allocated on first use.
594		*/
595	<	private static final RuntimePermission modifyThreadPermission =
596	<	new RuntimePermission("modifyThread");
595	>	static final class WorkQueue {
596	>	/**
597	>	* Capacity of work-stealing queue array upon initialization.
598	>	* Must be a power of two; at least 4, but should be larger to
599	>	* reduce or eliminate cacheline sharing among queues.
600	>	* Currently, it is much larger, as a partial workaround for
601	>	* the fact that JVMs often place arrays in locations that
602	>	* share GC bookkeeping (especially cardmarks) such that
603	>	* per-write accesses encounter serious memory contention.
604	>	*/
605	>	static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
606
607	<	/**
608	<	* If there is a security manager, makes sure caller has
609	<	* permission to modify threads.
610	<	*/
611	<	private static void checkPermission() {
612	<	SecurityManager security = System.getSecurityManager();
613	<	if (security != null)
614	<	security.checkPermission(modifyThreadPermission);
607	>	/**
608	>	* Maximum size for queue arrays. Must be a power of two less
609	>	* than or equal to 1 << (31 - width of array entry) to ensure
610	>	* lack of wraparound of index calculations, but defined to a
611	>	* value a bit less than this to help users trap runaway
612	>	* programs before saturating systems.
613	>	*/
614	>	static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
615	>
616	>	volatile long totalSteals; // cumulative number of steals
617	>	int seed;                  // for random scanning; initialize nonzero
618	>	volatile int eventCount;   // encoded inactivation count; < 0 if inactive
619	>	int nextWait;              // encoded record of next event waiter
620	>	int rescans;               // remaining scans until block
621	>	int nsteals;               // top-level task executions since last idle
622	>	final int mode;            // lifo, fifo, or shared
623	>	int poolIndex;             // index of this queue in pool (or 0)
624	>	int stealHint;             // index of most recent known stealer
625	>	volatile int runState;     // 1: locked, -1: terminate; else 0
626	>	volatile int base;         // index of next slot for poll
627	>	int top;                   // index of next slot for push
628	>	ForkJoinTask<?>[] array;   // the elements (initially unallocated)
629	>	final ForkJoinPool pool;   // the containing pool (may be null)
630	>	final ForkJoinWorkerThread owner; // owning thread or null if shared
631	>	volatile Thread parker;    // == owner during call to park; else null
632	>	volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
633	>	ForkJoinTask<?> currentSteal; // current non-local task being executed
634	>	// Heuristic padding to ameliorate unfortunate memory placements
635	>	Object p00, p01, p02, p03, p04, p05, p06, p07;
636	>	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
637	>
638	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
639	>	this.mode = mode;
640	>	this.pool = pool;
641	>	this.owner = owner;
642	>	// Place indices in the center of array (that is not yet allocated)
643	>	base = top = INITIAL_QUEUE_CAPACITY >>> 1;
644	>	}
645	>
646	>	/**
647	>	* Returns the approximate number of tasks in the queue.
648	>	*/
649	>	final int queueSize() {
650	>	int n = base - top;       // non-owner callers must read base first
651	>	return (n >= 0) ? 0 : -n; // ignore transient negative
652	>	}
653	>
654	>	/**
655	>	* Provides a more accurate estimate of whether this queue has
656	>	* any tasks than does queueSize, by checking whether a
657	>	* near-empty queue has at least one unclaimed task.
658	>	*/
659	>	final boolean isEmpty() {
660	>	ForkJoinTask<?>[] a; int m, s;
661	>	int n = base - (s = top);
662	>	return (n >= 0 ||
663	>	(n == -1 &&
664	>	((a = array) == null ||
665	>	(m = a.length - 1) < 0 ||
666	>	U.getObjectVolatile
667	>	(a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
668	>	}
669	>
670	>	/**
671	>	* Pushes a task. Call only by owner in unshared queues.
672	>	*
673	>	* @param task the task. Caller must ensure non-null.
674	>	* @throw RejectedExecutionException if array cannot be resized
675	>	*/
676	>	final void push(ForkJoinTask<?> task) {
677	>	ForkJoinTask<?>[] a; ForkJoinPool p;
678	>	int s = top, m, n;
679	>	if ((a = array) != null) {    // ignore if queue removed
680	>	U.putOrderedObject
681	>	(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
682	>	if ((n = (top = s + 1) - base) <= 2) {
683	>	if ((p = pool) != null)
684	>	p.signalWork();
685	>	}
686	>	else if (n >= m)
687	>	growArray(true);
688	>	}
689	>	}
690	>
691	>	/**
692	>	* Pushes a task if lock is free and array is either big
693	>	* enough or can be resized to be big enough.
694	>	*
695	>	* @param task the task. Caller must ensure non-null.
696	>	* @return true if submitted
697	>	*/
698	>	final boolean trySharedPush(ForkJoinTask<?> task) {
699	>	boolean submitted = false;
700	>	if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
701	>	ForkJoinTask<?>[] a = array;
702	>	int s = top;
703	>	try {
704	>	if ((a != null && a.length > s + 1 - base) ||
705	>	(a = growArray(false)) != null) { // must presize
706	>	int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
707	>	U.putObject(a, (long)j, task);    // don't need "ordered"
708	>	top = s + 1;
709	>	submitted = true;
710	>	}
711	>	} finally {
712	>	runState = 0;                         // unlock
713	>	}
714	>	}
715	>	return submitted;
716	>	}
717	>
718	>	/**
719	>	* Takes next task, if one exists, in LIFO order.  Call only
720	>	* by owner in unshared queues. (We do not have a shared
721	>	* version of this method because it is never needed.)
722	>	*/
723	>	final ForkJoinTask<?> pop() {
724	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
725	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
726	>	for (int s; (s = top - 1) - base >= 0;) {
727	>	long j = ((m & s) << ASHIFT) + ABASE;
728	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
729	>	break;
730	>	if (U.compareAndSwapObject(a, j, t, null)) {
731	>	top = s;
732	>	return t;
733	>	}
734	>	}
735	>	}
736	>	return null;
737	>	}
738	>
739	>	/**
740	>	* Takes a task in FIFO order if b is base of queue and a task
741	>	* can be claimed without contention. Specialized versions
742	>	* appear in ForkJoinPool methods scan and tryHelpStealer.
743	>	*/
744	>	final ForkJoinTask<?> pollAt(int b) {
745	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
746	>	if ((a = array) != null) {
747	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
748	>	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
749	>	base == b &&
750	>	U.compareAndSwapObject(a, j, t, null)) {
751	>	base = b + 1;
752	>	return t;
753	>	}
754	>	}
755	>	return null;
756	>	}
757	>
758	>	/**
759	>	* Takes next task, if one exists, in FIFO order.
760	>	*/
761	>	final ForkJoinTask<?> poll() {
762	>	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
763	>	while ((b = base) - top < 0 && (a = array) != null) {
764	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
765	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
766	>	if (t != null) {
767	>	if (base == b &&
768	>	U.compareAndSwapObject(a, j, t, null)) {
769	>	base = b + 1;
770	>	return t;
771	>	}
772	>	}
773	>	else if (base == b) {
774	>	if (b + 1 == top)
775	>	break;
776	>	Thread.yield(); // wait for lagging update
777	>	}
778	>	}
779	>	return null;
780	>	}
781	>
782	>	/**
783	>	* Takes next task, if one exists, in order specified by mode.
784	>	*/
785	>	final ForkJoinTask<?> nextLocalTask() {
786	>	return mode == 0 ? pop() : poll();
787	>	}
788	>
789	>	/**
790	>	* Returns next task, if one exists, in order specified by mode.
791	>	*/
792	>	final ForkJoinTask<?> peek() {
793	>	ForkJoinTask<?>[] a = array; int m;
794	>	if (a == null || (m = a.length - 1) < 0)
795	>	return null;
796	>	int i = mode == 0 ? top - 1 : base;
797	>	int j = ((i & m) << ASHIFT) + ABASE;
798	>	return (ForkJoinTask<?>)U.getObjectVolatile(a, j);
799	>	}
800	>
801	>	/**
802	>	* Pops the given task only if it is at the current top.
803	>	*/
804	>	final boolean tryUnpush(ForkJoinTask<?> t) {
805	>	ForkJoinTask<?>[] a; int s;
806	>	if ((a = array) != null && (s = top) != base &&
807	>	U.compareAndSwapObject
808	>	(a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
809	>	top = s;
810	>	return true;
811	>	}
812	>	return false;
813	>	}
814	>
815	>	/**
816	>	* Polls the given task only if it is at the current base.
817	>	*/
818	>	final boolean pollFor(ForkJoinTask<?> task) {
819	>	ForkJoinTask<?>[] a; int b;
820	>	if ((b = base) - top < 0 && (a = array) != null) {
821	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
822	>	if (U.getObjectVolatile(a, j) == task && base == b &&
823	>	U.compareAndSwapObject(a, j, task, null)) {
824	>	base = b + 1;
825	>	return true;
826	>	}
827	>	}
828	>	return false;
829	>	}
830	>
831	>	/**
832	>	* Initializes or doubles the capacity of array. Call either
833	>	* by owner or with lock held -- it is OK for base, but not
834	>	* top, to move while resizings are in progress.
835	>	*
836	>	* @param rejectOnFailure if true, throw exception if capacity
837	>	* exceeded (relayed ultimately to user); else return null.
838	>	*/
839	>	final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
840	>	ForkJoinTask<?>[] oldA = array;
841	>	int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
842	>	if (size <= MAXIMUM_QUEUE_CAPACITY) {
843	>	int oldMask, t, b;
844	>	ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
845	>	if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
846	>	(t = top) - (b = base) > 0) {
847	>	int mask = size - 1;
848	>	do {
849	>	ForkJoinTask<?> x;
850	>	int oldj = ((b & oldMask) << ASHIFT) + ABASE;
851	>	int j    = ((b &    mask) << ASHIFT) + ABASE;
852	>	x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
853	>	if (x != null &&
854	>	U.compareAndSwapObject(oldA, oldj, x, null))
855	>	U.putObjectVolatile(a, j, x);
856	>	} while (++b != t);
857	>	}
858	>	return a;
859	>	}
860	>	else if (!rejectOnFailure)
861	>	return null;
862	>	else
863	>	throw new RejectedExecutionException("Queue capacity exceeded");
864	>	}
865	>
866	>	/**
867	>	* Removes and cancels all known tasks, ignoring any exceptions.
868	>	*/
869	>	final void cancelAll() {
870	>	ForkJoinTask.cancelIgnoringExceptions(currentJoin);
871	>	ForkJoinTask.cancelIgnoringExceptions(currentSteal);
872	>	for (ForkJoinTask<?> t; (t = poll()) != null; )
873	>	ForkJoinTask.cancelIgnoringExceptions(t);
874	>	}
875	>
876	>	/**
877	>	* Computes next value for random probes.  Scans don't require
878	>	* a very high quality generator, but also not a crummy one.
879	>	* Marsaglia xor-shift is cheap and works well enough.  Note:
880	>	* This is manually inlined in its usages in ForkJoinPool to
881	>	* avoid writes inside busy scan loops.
882	>	*/
883	>	final int nextSeed() {
884	>	int r = seed;
885	>	r ^= r << 13;
886	>	r ^= r >>> 17;
887	>	return seed = r ^= r << 5;
888	>	}
889	>
890	>	// Execution methods
891	>
892	>	/**
893	>	* Pops and runs tasks until empty.
894	>	*/
895	>	private void popAndExecAll() {
896	>	// A bit faster than repeated pop calls
897	>	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
898	>	while ((a = array) != null && (m = a.length - 1) >= 0 &&
899	>	(s = top - 1) - base >= 0 &&
900	>	(t = ((ForkJoinTask<?>)
901	>	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
902	>	!= null) {
903	>	if (U.compareAndSwapObject(a, j, t, null)) {
904	>	top = s;
905	>	t.doExec();
906	>	}
907	>	}
908	>	}
909	>
910	>	/**
911	>	* Polls and runs tasks until empty.
912	>	*/
913	>	private void pollAndExecAll() {
914	>	for (ForkJoinTask<?> t; (t = poll()) != null;)
915	>	t.doExec();
916	>	}
917	>
918	>	/**
919	>	* If present, removes from queue and executes the given task, or
920	>	* any other cancelled task. Returns (true) immediately on any CAS
921	>	* or consistency check failure so caller can retry.
922	>	*
923	>	* @return 0 if no progress can be made, else positive
924	>	* (this unusual convention simplifies use with tryHelpStealer.)
925	>	*/
926	>	final int tryRemoveAndExec(ForkJoinTask<?> task) {
927	>	int stat = 1;
928	>	boolean removed = false, empty = true;
929	>	ForkJoinTask<?>[] a; int m, s, b, n;
930	>	if ((a = array) != null && (m = a.length - 1) >= 0 &&
931	>	(n = (s = top) - (b = base)) > 0) {
932	>	for (ForkJoinTask<?> t;;) {           // traverse from s to b
933	>	int j = ((--s & m) << ASHIFT) + ABASE;
934	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
935	>	if (t == null)                    // inconsistent length
936	>	break;
937	>	else if (t == task) {
938	>	if (s + 1 == top) {           // pop
939	>	if (!U.compareAndSwapObject(a, j, task, null))
940	>	break;
941	>	top = s;
942	>	removed = true;
943	>	}
944	>	else if (base == b)           // replace with proxy
945	>	removed = U.compareAndSwapObject(a, j, task,
946	>	new EmptyTask());
947	>	break;
948	>	}
949	>	else if (t.status >= 0)
950	>	empty = false;
951	>	else if (s + 1 == top) {          // pop and throw away
952	>	if (U.compareAndSwapObject(a, j, t, null))
953	>	top = s;
954	>	break;
955	>	}
956	>	if (--n == 0) {
957	>	if (!empty && base == b)
958	>	stat = 0;
959	>	break;
960	>	}
961	>	}
962	>	}
963	>	if (removed)
964	>	task.doExec();
965	>	return stat;
966	>	}
967	>
968	>	/**
969	>	* Executes a top-level task and any local tasks remaining
970	>	* after execution.
971	>	*/
972	>	final void runTask(ForkJoinTask<?> t) {
973	>	if (t != null) {
974	>	currentSteal = t;
975	>	t.doExec();
976	>	if (top != base) {       // process remaining local tasks
977	>	if (mode == 0)
978	>	popAndExecAll();
979	>	else
980	>	pollAndExecAll();
981	>	}
982	>	++nsteals;
983	>	currentSteal = null;
984	>	}
985	>	}
986	>
987	>	/**
988	>	* Executes a non-top-level (stolen) task.
989	>	*/
990	>	final void runSubtask(ForkJoinTask<?> t) {
991	>	if (t != null) {
992	>	ForkJoinTask<?> ps = currentSteal;
993	>	currentSteal = t;
994	>	t.doExec();
995	>	currentSteal = ps;
996	>	}
997	>	}
998	>
999	>	/**
1000	>	* Returns true if owned and not known to be blocked.
1001	>	*/
1002	>	final boolean isApparentlyUnblocked() {
1003	>	Thread wt; Thread.State s;
1004	>	return (eventCount >= 0 &&
1005	>	(wt = owner) != null &&
1006	>	(s = wt.getState()) != Thread.State.BLOCKED &&
1007	>	s != Thread.State.WAITING &&
1008	>	s != Thread.State.TIMED_WAITING);
1009	>	}
1010	>
1011	>	/**
1012	>	* If this owned and is not already interrupted, try to
1013	>	* interrupt and/or unpark, ignoring exceptions.
1014	>	*/
1015	>	final void interruptOwner() {
1016	>	Thread wt, p;
1017	>	if ((wt = owner) != null && !wt.isInterrupted()) {
1018	>	try {
1019	>	wt.interrupt();
1020	>	} catch (SecurityException ignore) {
1021	>	}
1022	>	}
1023	>	if ((p = parker) != null)
1024	>	U.unpark(p);
1025	>	}
1026	>
1027	>	// Unsafe mechanics
1028	>	private static final sun.misc.Unsafe U;
1029	>	private static final long RUNSTATE;
1030	>	private static final int ABASE;
1031	>	private static final int ASHIFT;
1032	>	static {
1033	>	int s;
1034	>	try {
1035	>	U = getUnsafe();
1036	>	Class<?> k = WorkQueue.class;
1037	>	Class<?> ak = ForkJoinTask[].class;
1038	>	RUNSTATE = U.objectFieldOffset
1039	>	(k.getDeclaredField("runState"));
1040	>	ABASE = U.arrayBaseOffset(ak);
1041	>	s = U.arrayIndexScale(ak);
1042	>	} catch (Exception e) {
1043	>	throw new Error(e);
1044	>	}
1045	>	if ((s & (s-1)) != 0)
1046	>	throw new Error("data type scale not a power of two");
1047	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1048	>	}
1049		}
1050
1051		/**
1052	<	* Generator for assigning sequence numbers as pool names.
1052	>	* Per-thread records for threads that submit to pools. Currently
1053	>	* holds only pseudo-random seed / index that is used to choose
1054	>	* submission queues in method doSubmit. In the future, this may
1055	>	* also incorporate a means to implement different task rejection
1056	>	* and resubmission policies.
1057	>	*
1058	>	* Seeds for submitters and workers/workQueues work in basically
1059	>	* the same way but are initialized and updated using slightly
1060	>	* different mechanics. Both are initialized using the same
1061	>	* approach as in class ThreadLocal, where successive values are
1062	>	* unlikely to collide with previous values. This is done during
1063	>	* registration for workers, but requires a separate AtomicInteger
1064	>	* for submitters. Seeds are then randomly modified upon
1065	>	* collisions using xorshifts, which requires a non-zero seed.
1066		*/
1067	<	private static final AtomicInteger poolNumberGenerator =
1068	<	new AtomicInteger();
1067	>	static final class Submitter {
1068	>	int seed;
1069	>	Submitter() {
1070	>	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1071	>	seed = (s == 0) ? 1 : s; // ensure non-zero
1072	>	}
1073	>	}
1074
1075	<	/**
1076	<	* Array holding all worker threads in the pool. Initialized upon
1077	<	* first use. Array size must be a power of two.  Updates and
1078	<	* replacements are protected by workerLock, but it is always kept
139	<	* in a consistent enough state to be randomly accessed without
140	<	* locking by workers performing work-stealing.
141	<	*/
142	<	volatile ForkJoinWorkerThread[] workers;
1075	>	/** ThreadLocal class for Submitters */
1076	>	static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1077	>	public Submitter initialValue() { return new Submitter(); }
1078	>	}
1079
1080	<	/**
145	<	* Lock protecting access to workers.
146	<	*/
147	<	private final ReentrantLock workerLock;
1080	>	// static fields (initialized in static initializer below)
1081
1082		/**
1083	<	* Condition for awaitTermination.
1083	>	* Creates a new ForkJoinWorkerThread. This factory is used unless
1084	>	* overridden in ForkJoinPool constructors.
1085		*/
1086	<	private final Condition termination;
1086	>	public static final ForkJoinWorkerThreadFactory
1087	>	defaultForkJoinWorkerThreadFactory;
1088
1089		/**
1090	<	* The uncaught exception handler used when any worker
156	<	* abruptly terminates
1090	>	* Generator for assigning sequence numbers as pool names.
1091		*/
1092	<	private Thread.UncaughtExceptionHandler ueh;
1092	>	private static final AtomicInteger poolNumberGenerator;
1093
1094		/**
1095	<	* Creation factory for worker threads.
1095	>	* Generator for initial hashes/seeds for submitters. Accessed by
1096	>	* Submitter class constructor.
1097		*/
1098	<	private final ForkJoinWorkerThreadFactory factory;
1098	>	static final AtomicInteger nextSubmitterSeed;
1099
1100		/**
1101	<	* Head of stack of threads that were created to maintain
1102	<	* parallelism when other threads blocked, but have since
168	<	* suspended when the parallelism level rose.
1101	>	* Permission required for callers of methods that may start or
1102	>	* kill threads.
1103		*/
1104	<	private volatile WaitQueueNode spareStack;
1104	>	private static final RuntimePermission modifyThreadPermission;
1105
1106		/**
1107	<	* Sum of per-thread steal counts, updated only when threads are
1108	<	* idle or terminating.
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		*/
1111	<	private final AtomicLong stealCount;
1111	>	private static final ThreadSubmitter submitters;
1112
1113	<	/**
179	<	* Queue for external submissions.
180	<	*/
181	<	private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
1113	>	// static constants
1114
1115		/**
1116	<	* Head of Treiber stack for barrier sync. See below for explanation
1116	>	* The wakeup interval (in nanoseconds) for a worker waiting for a
1117	>	* task when the pool is quiescent to instead try to shrink the
1118	>	* number of workers.  The exact value does not matter too
1119	>	* much. It must be short enough to release resources during
1120	>	* sustained periods of idleness, but not so short that threads
1121	>	* are continually re-created.
1122		*/
1123	<	private volatile WaitQueueNode syncStack;
1123	>	private static final long SHRINK_RATE =
1124	>	4L * 1000L * 1000L * 1000L; // 4 seconds
1125
1126		/**
1127	<	* The count for event barrier
1127	>	* The timeout value for attempted shrinkage, includes
1128	>	* some slop to cope with system timer imprecision.
1129		*/
1130	<	private volatile long eventCount;
1130	>	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1131
1132		/**
1133	<	* Pool number, just for assigning useful names to worker threads
1133	>	* The maximum stolen->joining link depth allowed in method
1134	>	* tryHelpStealer.  Must be a power of two. This value also
1135	>	* controls the maximum number of times to try to help join a task
1136	>	* without any apparent progress or change in pool state before
1137	>	* giving up and blocking (see awaitJoin).  Depths for legitimate
1138	>	* chains are unbounded, but we use a fixed constant to avoid
1139	>	* (otherwise unchecked) cycles and to bound staleness of
1140	>	* traversal parameters at the expense of sometimes blocking when
1141	>	* we could be helping.
1142		*/
1143	<	private final int poolNumber;
1143	>	private static final int MAX_HELP = 64;
1144
1145		/**
1146	<	* The maximum allowed pool size
1146	>	* Secondary time-based bound (in nanosecs) for helping attempts
1147	>	* before trying compensated blocking in awaitJoin. Used in
1148	>	* conjunction with MAX_HELP to reduce variance due to different
1149	>	* polling rates associated with different helping options. The
1150	>	* value should roughly approximate the time required to create
1151	>	* and/or activate a worker thread.
1152		*/
1153	<	private volatile int maxPoolSize;
1153	>	private static final long COMPENSATION_DELAY = 1L << 18; // ~0.25 millisec
1154
1155		/**
1156	<	* The desired parallelism level, updated only under workerLock.
1156	>	* Increment for seed generators. See class ThreadLocal for
1157	>	* explanation.
1158		*/
1159	<	private volatile int parallelism;
1159	>	private static final int SEED_INCREMENT = 0x61c88647;
1160
1161		/**
1162	<	* True if use local fifo, not default lifo, for local polling
1162	>	* Bits and masks for control variables
1163	>	*
1164	>	* Field ctl is a long packed with:
1165	>	* AC: Number of active running workers minus target parallelism (16 bits)
1166	>	* TC: Number of total workers minus target parallelism (16 bits)
1167	>	* ST: true if pool is terminating (1 bit)
1168	>	* EC: the wait count of top waiting thread (15 bits)
1169	>	* ID: poolIndex of top of Treiber stack of waiters (16 bits)
1170	>	*
1171	>	* When convenient, we can extract the upper 32 bits of counts and
1172	>	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
1173	>	* (int)ctl.  The ec field is never accessed alone, but always
1174	>	* together with id and st. The offsets of counts by the target
1175	>	* parallelism and the positionings of fields makes it possible to
1176	>	* perform the most common checks via sign tests of fields: When
1177	>	* ac is negative, there are not enough active workers, when tc is
1178	>	* negative, there are not enough total workers, and when e is
1179	>	* negative, the pool is terminating.  To deal with these possibly
1180	>	* negative fields, we use casts in and out of "short" and/or
1181	>	* signed shifts to maintain signedness.
1182	>	*
1183	>	* When a thread is queued (inactivated), its eventCount field is
1184	>	* set negative, which is the only way to tell if a worker is
1185	>	* prevented from executing tasks, even though it must continue to
1186	>	* scan for them to avoid queuing races. Note however that
1187	>	* eventCount updates lag releases so usage requires care.
1188	>	*
1189	>	* Field runState is an int packed with:
1190	>	* SHUTDOWN: true if shutdown is enabled (1 bit)
1191	>	* SEQ:  a sequence number updated upon (de)registering workers (30 bits)
1192	>	* INIT: set true after workQueues array construction (1 bit)
1193	>	*
1194	>	* The sequence number enables simple consistency checks:
1195	>	* Staleness of read-only operations on the workQueues array can
1196	>	* be checked by comparing runState before vs after the reads.
1197		*/
211	–	private volatile boolean locallyFifo;
1198
1199	<	/**
1200	<	* Holds number of total (i.e., created and not yet terminated)
1201	<	* and running (i.e., not blocked on joins or other managed sync)
1202	<	* threads, packed into one int to ensure consistent snapshot when
1203	<	* making decisions about creating and suspending spare
1204	<	* threads. Updated only by CAS.  Note: CASes in
1205	<	* updateRunningCount and preJoin running active count is in low
1206	<	* word, so need to be modified if this changes
1207	<	*/
1208	<	private volatile int workerCounts;
1199	>	// bit positions/shifts for fields
1200	>	private static final int  AC_SHIFT   = 48;
1201	>	private static final int  TC_SHIFT   = 32;
1202	>	private static final int  ST_SHIFT   = 31;
1203	>	private static final int  EC_SHIFT   = 16;
1204	>
1205	>	// bounds
1206	>	private static final int  SMASK      = 0xffff;  // short bits
1207	>	private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
1208	>	private static final int  SQMASK     = 0xfffe;  // even short bits
1209	>	private static final int  SHORT_SIGN = 1 << 15;
1210	>	private static final int  INT_SIGN   = 1 << 31;
1211	>
1212	>	// masks
1213	>	private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
1214	>	private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
1215	>	private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
1216	>
1217	>	// units for incrementing and decrementing
1218	>	private static final long TC_UNIT    = 1L << TC_SHIFT;
1219	>	private static final long AC_UNIT    = 1L << AC_SHIFT;
1220
1221	<	private static int totalCountOf(int s)           { return s >>> 16;  }
1222	<	private static int runningCountOf(int s)         { return s & shortMask; }
1223	<	private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
1221	>	// masks and units for dealing with u = (int)(ctl >>> 32)
1222	>	private static final int  UAC_SHIFT  = AC_SHIFT - 32;
1223	>	private static final int  UTC_SHIFT  = TC_SHIFT - 32;
1224	>	private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
1225	>	private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
1226	>	private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
1227	>	private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
1228	>
1229	>	// masks and units for dealing with e = (int)ctl
1230	>	private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
1231	>	private static final int E_SEQ       = 1 << EC_SHIFT;
1232	>
1233	>	// runState bits
1234	>	private static final int SHUTDOWN    = 1 << 31;
1235	>
1236	>	// access mode for WorkQueue
1237	>	static final int LIFO_QUEUE          =  0;
1238	>	static final int FIFO_QUEUE          =  1;
1239	>	static final int SHARED_QUEUE        = -1;
1240	>
1241	>	// Instance fields
1242	>
1243	>	/*
1244	>	* Field layout order in this class tends to matter more than one
1245	>	* would like. Runtime layout order is only loosely related to
1246	>	* declaration order and may differ across JVMs, but the following
1247	>	* empirically works OK on current JVMs.
1248	>	*/
1249	>
1250	>	volatile long ctl;                         // main pool control
1251	>	final int parallelism;                     // parallelism level
1252	>	final int localMode;                       // per-worker scheduling mode
1253	>	final int submitMask;                      // submit queue index bound
1254	>	int nextSeed;                              // for initializing worker seeds
1255	>	volatile int runState;                     // shutdown status and seq
1256	>	WorkQueue[] workQueues;                    // main registry
1257	>	final Mutex lock;                          // for registration
1258	>	final Condition termination;               // for awaitTermination
1259	>	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1260	>	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1261	>	final AtomicLong stealCount;               // collect counts when terminated
1262	>	final AtomicInteger nextWorkerNumber;      // to create worker name string
1263	>	final String workerNamePrefix;             // to create worker name string
1264	>
1265	>	//  Creating, registering, and deregistering workers
1266	>
1267	>	/**
1268	>	* Tries to create and start a worker
1269	>	*/
1270	>	private void addWorker() {
1271	>	Throwable ex = null;
1272	>	ForkJoinWorkerThread wt = null;
1273	>	try {
1274	>	if ((wt = factory.newThread(this)) != null) {
1275	>	wt.start();
1276	>	return;
1277	>	}
1278	>	} catch (Throwable e) {
1279	>	ex = e;
1280	>	}
1281	>	deregisterWorker(wt, ex); // adjust counts etc on failure
1282	>	}
1283
1284		/**
1285	<	* Adds delta (which may be negative) to running count.  This must
1286	<	* be called before (with negative arg) and after (with positive)
1287	<	* any managed synchronization (i.e., mainly, joins).
1288	<	* @param delta the number to add
1285	>	* Callback from ForkJoinWorkerThread constructor to assign a
1286	>	* public name. This must be separate from registerWorker because
1287	>	* it is called during the "super" constructor call in
1288	>	* ForkJoinWorkerThread.
1289		*/
1290	<	final void updateRunningCount(int delta) {
1291	<	int s;
1292	<	do;while (!casWorkerCounts(s = workerCounts, s + delta));
1290	>	final String nextWorkerName() {
1291	>	return workerNamePrefix.concat
1292	>	(Integer.toString(nextWorkerNumber.addAndGet(1)));
1293		}
1294
1295		/**
1296	<	* Adds delta (which may be negative) to both total and running
1297	<	* count.  This must be called upon creation and termination of
1298	<	* worker threads.
1299	<	* @param delta the number to add
1296	>	* Callback from ForkJoinWorkerThread constructor to establish its
1297	>	* poolIndex and record its WorkQueue. To avoid scanning bias due
1298	>	* to packing entries in front of the workQueues array, we treat
1299	>	* the array as a simple power-of-two hash table using per-thread
1300	>	* seed as hash, expanding as needed.
1301	>	*
1302	>	* @param w the worker's queue
1303		*/
1304	<	private void updateWorkerCount(int delta) {
1305	<	int d = delta + (delta << 16); // add to both lo and hi parts
1306	<	int s;
1307	<	do;while (!casWorkerCounts(s = workerCounts, s + d));
1304	>
1305	>	final void registerWorker(WorkQueue w) {
1306	>	Mutex lock = this.lock;
1307	>	lock.lock();
1308	>	try {
1309	>	WorkQueue[] ws = workQueues;
1310	>	if (w != null && ws != null) {          // skip on shutdown/failure
1311	>	int rs, n =  ws.length, m = n - 1;
1312	>	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1313	>	w.seed = (s == 0) ? 1 : s;          // ensure non-zero seed
1314	>	int r = (s << 1) | 1;               // use odd-numbered indices
1315	>	if (ws[r &= m] != null) {           // collision
1316	>	int probes = 0;                 // step by approx half size
1317	>	int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1318	>	while (ws[r = (r + step) & m] != null) {
1319	>	if (++probes >= n) {
1320	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1321	>	m = n - 1;
1322	>	probes = 0;
1323	>	}
1324	>	}
1325	>	}
1326	>	w.eventCount = w.poolIndex = r;     // establish before recording
1327	>	ws[r] = w;                          // also update seq
1328	>	runState = ((rs = runState) & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN);
1329	>	}
1330	>	} finally {
1331	>	lock.unlock();
1332	>	}
1333		}
1334
1335		/**
1336	<	* Lifecycle control. High word contains runState, low word
1337	<	* contains the number of workers that are (probably) executing
1338	<	* tasks. This value is atomically incremented before a worker
1339	<	* gets a task to run, and decremented when worker has no tasks
1340	<	* and cannot find any. These two fields are bundled together to
1341	<	* support correct termination triggering.  Note: activeCount
1342	<	* CAS'es cheat by assuming active count is in low word, so need
1343	<	* to be modified if this changes
1344	<	*/
1345	<	private volatile int runControl;
1336	>	* Final callback from terminating worker, as well as upon failure
1337	>	* to construct or start a worker in addWorker.  Removes record of
1338	>	* worker from array, and adjusts counts. If pool is shutting
1339	>	* down, tries to complete termination.
1340	>	*
1341	>	* @param wt the worker thread or null if addWorker failed
1342	>	* @param ex the exception causing failure, or null if none
1343	>	*/
1344	>	final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1345	>	Mutex lock = this.lock;
1346	>	WorkQueue w = null;
1347	>	if (wt != null && (w = wt.workQueue) != null) {
1348	>	w.runState = -1;                // ensure runState is set
1349	>	stealCount.getAndAdd(w.totalSteals + w.nsteals);
1350	>	int idx = w.poolIndex;
1351	>	lock.lock();
1352	>	try {                           // remove record from array
1353	>	WorkQueue[] ws = workQueues;
1354	>	if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1355	>	ws[idx] = null;
1356	>	} finally {
1357	>	lock.unlock();
1358	>	}
1359	>	}
1360
1361	<	// RunState values. Order among values matters
1362	<	private static final int RUNNING     = 0;
1363	<	private static final int SHUTDOWN    = 1;
1364	<	private static final int TERMINATING = 2;
1365	<	private static final int TERMINATED  = 3;
1361	>	long c;                             // adjust ctl counts
1362	>	do {} while (!U.compareAndSwapLong
1363	>	(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1364	>	((c - TC_UNIT) & TC_MASK) |
1365	>	(c & ~(AC_MASK|TC_MASK)))));
1366	>
1367	>	if (!tryTerminate(false, false) && w != null) {
1368	>	w.cancelAll();                  // cancel remaining tasks
1369	>	if (w.array != null)            // suppress signal if never ran
1370	>	signalWork();               // wake up or create replacement
1371	>	if (ex == null)                 // help clean refs on way out
1372	>	ForkJoinTask.helpExpungeStaleExceptions();
1373	>	}
1374	>
1375	>	if (ex != null)                     // rethrow
1376	>	U.throwException(ex);
1377	>	}
1378
269	–	private static int runStateOf(int c)             { return c >>> 16; }
270	–	private static int activeCountOf(int c)          { return c & shortMask; }
271	–	private static int runControlFor(int r, int a)   { return (r << 16) + a; }
1379
1380	<	/**
1381	<	* Try incrementing active count; fail on contention. Called by
1382	<	* workers before/during executing tasks.
1383	<	* @return true on success;
1384	<	*/
1385	<	final boolean tryIncrementActiveCount() {
1386	<	int c = runControl;
1387	<	return casRunControl(c, c+1);
1380	>	// Submissions
1381	>
1382	>	/**
1383	>	* Unless shutting down, adds the given task to a submission queue
1384	>	* at submitter's current queue index (modulo submission
1385	>	* range). If no queue exists at the index, one is created.  If
1386	>	* the queue is busy, another index is randomly chosen. The
1387	>	* submitMask bounds the effective number of queues to the
1388	>	* (nearest power of two for) parallelism level.
1389	>	*
1390	>	* @param task the task. Caller must ensure non-null.
1391	>	*/
1392	>	private void doSubmit(ForkJoinTask<?> task) {
1393	>	Submitter s = submitters.get();
1394	>	for (int r = s.seed, m = submitMask;;) {
1395	>	WorkQueue[] ws; WorkQueue q;
1396	>	int k = r & m & SQMASK;          // use only even indices
1397	>	if (runState < 0 || (ws = workQueues) == null || ws.length <= k)
1398	>	throw new RejectedExecutionException(); // shutting down
1399	>	else if ((q = ws[k]) == null) {  // create new queue
1400	>	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1401	>	Mutex lock = this.lock;      // construct outside lock
1402	>	lock.lock();
1403	>	try {                        // recheck under lock
1404	>	int rs = runState;       // to update seq
1405	>	if (ws == workQueues && ws[k] == null) {
1406	>	ws[k] = nq;
1407	>	runState = ((rs & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN));
1408	>	}
1409	>	} finally {
1410	>	lock.unlock();
1411	>	}
1412	>	}
1413	>	else if (q.trySharedPush(task)) {
1414	>	signalWork();
1415	>	return;
1416	>	}
1417	>	else if (m > 1) {                // move to a different index
1418	>	r ^= r << 13;                // same xorshift as WorkQueues
1419	>	r ^= r >>> 17;
1420	>	s.seed = r ^= r << 5;
1421	>	}
1422	>	else
1423	>	Thread.yield();              // yield if no alternatives
1424	>	}
1425		}
1426
1427	+	// Maintaining ctl counts
1428	+
1429		/**
1430	<	* Tries decrementing active count; fails on contention.
285	<	* Possibly triggers termination on success.
286	<	* Called by workers when they can't find tasks.
287	<	* @return true on success
1430	>	* Increments active count; mainly called upon return from blocking.
1431		*/
1432	<	final boolean tryDecrementActiveCount() {
1433	<	int c = runControl;
1434	<	int nextc = c - 1;
292	<	if (!casRunControl(c, nextc))
293	<	return false;
294	<	if (canTerminateOnShutdown(nextc))
295	<	terminateOnShutdown();
296	<	return true;
1432	>	final void incrementActiveCount() {
1433	>	long c;
1434	>	do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1435		}
1436
1437		/**
1438	<	* Returns true if argument represents zero active count and
301	<	* nonzero runstate, which is the triggering condition for
302	<	* terminating on shutdown.
1438	>	* Tries to activate or create a worker if too few are active.
1439		*/
1440	<	private static boolean canTerminateOnShutdown(int c) {
1441	<	return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
1440	>	final void signalWork() {
1441	>	long c; int u;
1442	>	while ((u = (int)((c = ctl) >>> 32)) < 0) {     // too few active
1443	>	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1444	>	if ((e = (int)c) > 0) {                     // at least one waiting
1445	>	if (ws != null && (i = e & SMASK) < ws.length &&
1446	>	(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1447	>	long nc = (((long)(w.nextWait & E_MASK)) |
1448	>	((long)(u + UAC_UNIT) << 32));
1449	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1450	>	w.eventCount = (e + E_SEQ) & E_MASK;
1451	>	if ((p = w.parker) != null)
1452	>	U.unpark(p);                // activate and release
1453	>	break;
1454	>	}
1455	>	}
1456	>	else
1457	>	break;
1458	>	}
1459	>	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1460	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1461	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1462	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1463	>	addWorker();
1464	>	break;
1465	>	}
1466	>	}
1467	>	else
1468	>	break;
1469	>	}
1470		}
1471
1472	+	// Scanning for tasks
1473	+
1474		/**
1475	<	* Transition run state to at least the given state. Return true
310	<	* if not already at least given state.
1475	>	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1476		*/
1477	<	private boolean transitionRunStateTo(int state) {
1478	<	for (;;) {
1479	<	int c = runControl;
1480	<	if (runStateOf(c) >= state)
1481	<	return false;
1482	<	if (casRunControl(c, runControlFor(state, activeCountOf(c))))
1483	<	return true;
1477	>	final void runWorker(WorkQueue w) {
1478	>	w.growArray(false);         // initialize queue array in this thread
1479	>	do { w.runTask(scan(w)); } while (w.runState >= 0);
1480	>	}
1481	>
1482	>	/**
1483	>	* Scans for and, if found, returns one task, else possibly
1484	>	* inactivates the worker. This method operates on single reads of
1485	>	* volatile state and is designed to be re-invoked continuously,
1486	>	* in part because it returns upon detecting inconsistencies,
1487	>	* contention, or state changes that indicate possible success on
1488	>	* re-invocation.
1489	>	*
1490	>	* The scan searches for tasks across a random permutation of
1491	>	* queues (starting at a random index and stepping by a random
1492	>	* relative prime, checking each at least once).  The scan
1493	>	* terminates upon either finding a non-empty queue, or completing
1494	>	* the sweep. If the worker is not inactivated, it takes and
1495	>	* returns a task from this queue.  On failure to find a task, we
1496	>	* take one of the following actions, after which the caller will
1497	>	* retry calling this method unless terminated.
1498	>	*
1499	>	* * If pool is terminating, terminate the worker.
1500	>	*
1501	>	* * If not a complete sweep, try to release a waiting worker.  If
1502	>	* the scan terminated because the worker is inactivated, then the
1503	>	* released worker will often be the calling worker, and it can
1504	>	* succeed obtaining a task on the next call. Or maybe it is
1505	>	* another worker, but with same net effect. Releasing in other
1506	>	* cases as well ensures that we have enough workers running.
1507	>	*
1508	>	* * If not already enqueued, try to inactivate and enqueue the
1509	>	* worker on wait queue. Or, if inactivating has caused the pool
1510	>	* to be quiescent, relay to idleAwaitWork to check for
1511	>	* termination and possibly shrink pool.
1512	>	*
1513	>	* * If already inactive, and the caller has run a task since the
1514	>	* last empty scan, return (to allow rescan) unless others are
1515	>	* also inactivated.  Field WorkQueue.rescans counts down on each
1516	>	* scan to ensure eventual inactivation and blocking.
1517	>	*
1518	>	* * If already enqueued and none of the above apply, park
1519	>	* awaiting signal,
1520	>	*
1521	>	* @param w the worker (via its WorkQueue)
1522	>	* @return a task or null of none found
1523	>	*/
1524	>	private final ForkJoinTask<?> scan(WorkQueue w) {
1525	>	WorkQueue[] ws;                       // first update random seed
1526	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1527	>	int rs = runState, m;                 // volatile read order matters
1528	>	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1529	>	int ec = w.eventCount;            // ec is negative if inactive
1530	>	int step = (r >>> 16) | 1;        // relative prime
1531	>	for (int j = (m + 1) << 2; ; r += step) {
1532	>	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1533	>	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1534	>	(a = q.array) != null) {  // probably nonempty
1535	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1536	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1537	>	if (q.base == b && ec >= 0 && t != null &&
1538	>	U.compareAndSwapObject(a, i, t, null)) {
1539	>	if (q.top - (q.base = b + 1) > 1)
1540	>	signalWork();    // help pushes signal
1541	>	return t;
1542	>	}
1543	>	else if (ec < 0 || j <= m) {
1544	>	rs = 0;               // mark scan as imcomplete
1545	>	break;                // caller can retry after release
1546	>	}
1547	>	}
1548	>	if (--j < 0)
1549	>	break;
1550	>	}
1551	>
1552	>	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1553	>	if (e < 0)                        // decode ctl on empty scan
1554	>	w.runState = -1;              // pool is terminating
1555	>	else if (rs == 0 || rs != runState) { // incomplete scan
1556	>	WorkQueue v; Thread p;        // try to release a waiter
1557	>	if (e > 0 && a < 0 && w.eventCount == ec &&
1558	>	(v = ws[e & m]) != null && v.eventCount == (e | INT_SIGN)) {
1559	>	long nc = ((long)(v.nextWait & E_MASK) |
1560	>	((c + AC_UNIT) & (AC_MASK|TC_MASK)));
1561	>	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1562	>	v.eventCount = (e + E_SEQ) & E_MASK;
1563	>	if ((p = v.parker) != null)
1564	>	U.unpark(p);
1565	>	}
1566	>	}
1567	>	}
1568	>	else if (ec >= 0) {               // try to enqueue/inactivate
1569	>	long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
1570	>	w.nextWait = e;
1571	>	w.eventCount = ec | INT_SIGN; // mark as inactive
1572	>	if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc))
1573	>	w.eventCount = ec;        // unmark on CAS failure
1574	>	else {
1575	>	if ((ns = w.nsteals) != 0) {
1576	>	w.nsteals = 0;        // set rescans if ran task
1577	>	w.rescans = (a > 0) ? 0 : a + parallelism;
1578	>	w.totalSteals += ns;
1579	>	}
1580	>	if (a == 1 - parallelism) // quiescent
1581	>	idleAwaitWork(w, nc, c);
1582	>	}
1583	>	}
1584	>	else if (w.eventCount < 0) {      // already queued
1585	>	if ((nr = w.rescans) > 0) {   // continue rescanning
1586	>	int ac = a + parallelism;
1587	>	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0)
1588	>	Thread.yield();       // yield before block
1589	>	}
1590	>	else {
1591	>	Thread.interrupted();     // clear status
1592	>	Thread wt = Thread.currentThread();
1593	>	U.putObject(wt, PARKBLOCKER, this);
1594	>	w.parker = wt;            // emulate LockSupport.park
1595	>	if (w.eventCount < 0)     // recheck
1596	>	U.park(false, 0L);
1597	>	w.parker = null;
1598	>	U.putObject(wt, PARKBLOCKER, null);
1599	>	}
1600	>	}
1601		}
1602	+	return null;
1603		}
1604
1605		/**
1606	<	* Controls whether to add spares to maintain parallelism
1607	<	*/
1608	<	private volatile boolean maintainsParallelism;
1609	<
1610	<	// Constructors
1606	>	* If inactivating worker w has caused the pool to become
1607	>	* quiescent, checks for pool termination, and, so long as this is
1608	>	* not the only worker, waits for event for up to SHRINK_RATE
1609	>	* nanosecs.  On timeout, if ctl has not changed, terminates the
1610	>	* worker, which will in turn wake up another worker to possibly
1611	>	* repeat this process.
1612	>	*
1613	>	* @param w the calling worker
1614	>	* @param currentCtl the ctl value triggering possible quiescence
1615	>	* @param prevCtl the ctl value to restore if thread is terminated
1616	>	*/
1617	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1618	>	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1619	>	(int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1620	>	Thread wt = Thread.currentThread();
1621	>	Thread.yield();            // yield before block
1622	>	while (ctl == currentCtl) {
1623	>	long startTime = System.nanoTime();
1624	>	Thread.interrupted();  // timed variant of version in scan()
1625	>	U.putObject(wt, PARKBLOCKER, this);
1626	>	w.parker = wt;
1627	>	if (ctl == currentCtl)
1628	>	U.park(false, SHRINK_RATE);
1629	>	w.parker = null;
1630	>	U.putObject(wt, PARKBLOCKER, null);
1631	>	if (ctl != currentCtl)
1632	>	break;
1633	>	if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1634	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1635	>	w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1636	>	w.runState = -1;   // shrink
1637	>	break;
1638	>	}
1639	>	}
1640	>	}
1641	>	}
1642
1643		/**
1644	<	* Creates a ForkJoinPool with a pool size equal to the number of
1645	<	* processors available on the system and using the default
1646	<	* ForkJoinWorkerThreadFactory,
1647	<	* @throws SecurityException if a security manager exists and
1648	<	*         the caller is not permitted to modify threads
1649	<	*         because it does not hold {@link
1650	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
1651	<	*/
1652	<	public ForkJoinPool() {
1653	<	this(Runtime.getRuntime().availableProcessors(),
1654	<	defaultForkJoinWorkerThreadFactory);
1644	>	* Tries to locate and execute tasks for a stealer of the given
1645	>	* task, or in turn one of its stealers, Traces currentSteal ->
1646	>	* currentJoin links looking for a thread working on a descendant
1647	>	* of the given task and with a non-empty queue to steal back and
1648	>	* execute tasks from. The first call to this method upon a
1649	>	* waiting join will often entail scanning/search, (which is OK
1650	>	* because the joiner has nothing better to do), but this method
1651	>	* leaves hints in workers to speed up subsequent calls. The
1652	>	* implementation is very branchy to cope with potential
1653	>	* inconsistencies or loops encountering chains that are stale,
1654	>	* unknown, or so long that they are likely cyclic.
1655	>	*
1656	>	* @param joiner the joining worker
1657	>	* @param task the task to join
1658	>	* @return 0 if no progress can be made, negative if task
1659	>	* known complete, else positive
1660	>	*/
1661	>	private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1662	>	int stat = 0, steps = 0;                    // bound to avoid cycles
1663	>	if (joiner != null && task != null) {       // hoist null checks
1664	>	restart: for (;;) {
1665	>	ForkJoinTask<?> subtask = task;     // current target
1666	>	for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
1667	>	WorkQueue[] ws; int m, s, h;
1668	>	if ((s = task.status) < 0) {
1669	>	stat = s;
1670	>	break restart;
1671	>	}
1672	>	if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1673	>	break restart;              // shutting down
1674	>	if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
1675	>	v.currentSteal != subtask) {
1676	>	for (int origin = h;;) {    // find stealer
1677	>	if (((h = (h + 2) & m) & 15) == 1 &&
1678	>	(subtask.status < 0 || j.currentJoin != subtask))
1679	>	continue restart;   // occasional staleness check
1680	>	if ((v = ws[h]) != null &&
1681	>	v.currentSteal == subtask) {
1682	>	j.stealHint = h;    // save hint
1683	>	break;
1684	>	}
1685	>	if (h == origin)
1686	>	break restart;      // cannot find stealer
1687	>	}
1688	>	}
1689	>	for (;;) { // help stealer or descend to its stealer
1690	>	ForkJoinTask[] a;  int b;
1691	>	if (subtask.status < 0)     // surround probes with
1692	>	continue restart;       //   consistency checks
1693	>	if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1694	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1695	>	ForkJoinTask<?> t =
1696	>	(ForkJoinTask<?>)U.getObjectVolatile(a, i);
1697	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1698	>	v.currentSteal != subtask)
1699	>	continue restart;   // stale
1700	>	stat = 1;               // apparent progress
1701	>	if (t != null && v.base == b &&
1702	>	U.compareAndSwapObject(a, i, t, null)) {
1703	>	v.base = b + 1;     // help stealer
1704	>	joiner.runSubtask(t);
1705	>	}
1706	>	else if (v.base == b && ++steps == MAX_HELP)
1707	>	break restart;      // v apparently stalled
1708	>	}
1709	>	else {                      // empty -- try to descend
1710	>	ForkJoinTask<?> next = v.currentJoin;
1711	>	if (subtask.status < 0 || j.currentJoin != subtask ||
1712	>	v.currentSteal != subtask)
1713	>	continue restart;   // stale
1714	>	else if (next == null || ++steps == MAX_HELP)
1715	>	break restart;      // dead-end or maybe cyclic
1716	>	else {
1717	>	subtask = next;
1718	>	j = v;
1719	>	break;
1720	>	}
1721	>	}
1722	>	}
1723	>	}
1724	>	}
1725	>	}
1726	>	return stat;
1727		}
1728
1729		/**
1730	<	* Creates a ForkJoinPool with the indicated parallelism level
1731	<	* threads, and using the default ForkJoinWorkerThreadFactory,
1732	<	* @param parallelism the number of worker threads
1733	<	* @throws IllegalArgumentException if parallelism less than or
348	<	* equal to zero
349	<	* @throws SecurityException if a security manager exists and
350	<	*         the caller is not permitted to modify threads
351	<	*         because it does not hold {@link
352	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
1730	>	* If task is at base of some steal queue, steals and executes it.
1731	>	*
1732	>	* @param joiner the joining worker
1733	>	* @param task the task
1734		*/
1735	<	public ForkJoinPool(int parallelism) {
1736	<	this(parallelism, defaultForkJoinWorkerThreadFactory);
1735	>	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1736	>	WorkQueue[] ws;
1737	>	if ((ws = workQueues) != null) {
1738	>	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1739	>	WorkQueue q = ws[j];
1740	>	if (q != null && q.pollFor(task)) {
1741	>	joiner.runSubtask(task);
1742	>	break;
1743	>	}
1744	>	}
1745	>	}
1746		}
1747
1748		/**
1749	<	* Creates a ForkJoinPool with parallelism equal to the number of
1750	<	* processors available on the system and using the given
1751	<	* ForkJoinWorkerThreadFactory,
1752	<	* @param factory the factory for creating new threads
1753	<	* @throws NullPointerException if factory is null
1754	<	* @throws SecurityException if a security manager exists and
1755	<	*         the caller is not permitted to modify threads
1756	<	*         because it does not hold {@link
1757	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
1758	<	*/
1759	<	public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
1760	<	this(Runtime.getRuntime().availableProcessors(), factory);
1749	>	* Tries to decrement active count (sometimes implicitly) and
1750	>	* possibly release or create a compensating worker in preparation
1751	>	* for blocking. Fails on contention or termination. Otherwise,
1752	>	* adds a new thread if no idle workers are available and either
1753	>	* pool would become completely starved or: (at least half
1754	>	* starved, and fewer than 50% spares exist, and there is at least
1755	>	* one task apparently available). Even though the availability
1756	>	* check requires a full scan, it is worthwhile in reducing false
1757	>	* alarms.
1758	>	*
1759	>	* @param task if non-null, a task being waited for
1760	>	* @param blocker if non-null, a blocker being waited for
1761	>	* @return true if the caller can block, else should recheck and retry
1762	>	*/
1763	>	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1764	>	int pc = parallelism, e;
1765	>	long c = ctl;
1766	>	WorkQueue[] ws = workQueues;
1767	>	if ((e = (int)c) >= 0 && ws != null) {
1768	>	int u, a, ac, hc;
1769	>	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1770	>	boolean replace = false;
1771	>	if ((a = u >> UAC_SHIFT) <= 0) {
1772	>	if ((ac = a + pc) <= 1)
1773	>	replace = true;
1774	>	else if ((e > 0 || (task != null &&
1775	>	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1776	>	WorkQueue w;
1777	>	for (int j = 0; j < ws.length; ++j) {
1778	>	if ((w = ws[j]) != null && !w.isEmpty()) {
1779	>	replace = true;
1780	>	break;   // in compensation range and tasks available
1781	>	}
1782	>	}
1783	>	}
1784	>	}
1785	>	if ((task == null || task.status >= 0) && // recheck need to block
1786	>	(blocker == null || !blocker.isReleasable()) && ctl == c) {
1787	>	if (!replace) {          // no compensation
1788	>	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1789	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1790	>	return true;
1791	>	}
1792	>	else if (e != 0) {       // release an idle worker
1793	>	WorkQueue w; Thread p; int i;
1794	>	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
1795	>	long nc = ((long)(w.nextWait & E_MASK) |
1796	>	(c & (AC_MASK|TC_MASK)));
1797	>	if (w.eventCount == (e | INT_SIGN) &&
1798	>	U.compareAndSwapLong(this, CTL, c, nc)) {
1799	>	w.eventCount = (e + E_SEQ) & E_MASK;
1800	>	if ((p = w.parker) != null)
1801	>	U.unpark(p);
1802	>	return true;
1803	>	}
1804	>	}
1805	>	}
1806	>	else if (tc < MAX_CAP) { // create replacement
1807	>	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1808	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1809	>	addWorker();
1810	>	return true;
1811	>	}
1812	>	}
1813	>	}
1814	>	}
1815	>	return false;
1816		}
1817
1818		/**
1819	<	* Creates a ForkJoinPool with the given parallelism and factory.
1819	>	* Helps and/or blocks until the given task is done.
1820		*
1821	<	* @param parallelism the targeted number of worker threads
1822	<	* @param factory the factory for creating new threads
1823	<	* @throws IllegalArgumentException if parallelism less than or
379	<	* equal to zero, or greater than implementation limit.
380	<	* @throws NullPointerException if factory is null
381	<	* @throws SecurityException if a security manager exists and
382	<	*         the caller is not permitted to modify threads
383	<	*         because it does not hold {@link
384	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
1821	>	* @param joiner the joining worker
1822	>	* @param task the task
1823	>	* @return task status on exit
1824		*/
1825	<	public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
1826	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
1827	<	throw new IllegalArgumentException();
1828	<	if (factory == null)
1829	<	throw new NullPointerException();
1830	<	checkPermission();
1831	<	this.factory = factory;
1832	<	this.parallelism = parallelism;
1833	<	this.maxPoolSize = MAX_THREADS;
1834	<	this.maintainsParallelism = true;
1835	<	this.poolNumber = poolNumberGenerator.incrementAndGet();
1836	<	this.workerLock = new ReentrantLock();
1837	<	this.termination = workerLock.newCondition();
1838	<	this.stealCount = new AtomicLong();
1839	<	this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1840	<	// worker array and workers are lazily constructed
1825	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1826	>	int s;
1827	>	if ((s = task.status) >= 0) {
1828	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1829	>	joiner.currentJoin = task;
1830	>	long startTime = 0L;
1831	>	for (int k = 0;;) {
1832	>	if ((s = (joiner.isEmpty() ?           // try to help
1833	>	tryHelpStealer(joiner, task) :
1834	>	joiner.tryRemoveAndExec(task))) == 0 &&
1835	>	(s = task.status) >= 0) {
1836	>	if (k == 0) {
1837	>	startTime = System.nanoTime();
1838	>	tryPollForAndExec(joiner, task); // check uncommon case
1839	>	}
1840	>	else if ((k & (MAX_HELP - 1)) == 0 &&
1841	>	System.nanoTime() - startTime >=
1842	>	COMPENSATION_DELAY &&
1843	>	tryCompensate(task, null)) {
1844	>	if (task.trySetSignal()) {
1845	>	synchronized (task) {
1846	>	if (task.status >= 0) {
1847	>	try {                // see ForkJoinTask
1848	>	task.wait();     //  for explanation
1849	>	} catch (InterruptedException ie) {
1850	>	}
1851	>	}
1852	>	else
1853	>	task.notifyAll();
1854	>	}
1855	>	}
1856	>	long c;                          // re-activate
1857	>	do {} while (!U.compareAndSwapLong
1858	>	(this, CTL, c = ctl, c + AC_UNIT));
1859	>	}
1860	>	}
1861	>	if (s < 0 || (s = task.status) < 0) {
1862	>	joiner.currentJoin = prevJoin;
1863	>	break;
1864	>	}
1865	>	else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
1866	>	Thread.yield();                     // for politeness
1867	>	}
1868	>	}
1869	>	return s;
1870		}
1871
1872		/**
1873	<	* Create new worker using factory.
1874	<	* @param index the index to assign worker
1875	<	* @return new worker, or null of factory failed
1873	>	* Stripped-down variant of awaitJoin used by timed joins. Tries
1874	>	* to help join only while there is continuous progress. (Caller
1875	>	* will then enter a timed wait.)
1876	>	*
1877	>	* @param joiner the joining worker
1878	>	* @param task the task
1879	>	* @return task status on exit
1880		*/
1881	<	private ForkJoinWorkerThread createWorker(int index) {
1882	<	Thread.UncaughtExceptionHandler h = ueh;
1883	<	ForkJoinWorkerThread w = factory.newThread(this);
1884	<	if (w != null) {
1885	<	w.poolIndex = index;
1886	<	w.setDaemon(true);
1887	<	w.setAsyncMode(locallyFifo);
1888	<	w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
417	<	if (h != null)
418	<	w.setUncaughtExceptionHandler(h);
419	<	}
420	<	return w;
1881	>	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
1882	>	int s;
1883	>	while ((s = task.status) >= 0 &&
1884	>	(joiner.isEmpty() ?
1885	>	tryHelpStealer(joiner, task) :
1886	>	joiner.tryRemoveAndExec(task)) != 0)
1887	>	;
1888	>	return s;
1889		}
1890
1891		/**
1892	<	* Returns a good size for worker array given pool size.
1893	<	* Currently requires size to be a power of two.
1894	<	*/
1895	<	private static int arraySizeFor(int ps) {
1896	<	return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
1892	>	* Returns a (probably) non-empty steal queue, if one is found
1893	>	* during a random, then cyclic scan, else null.  This method must
1894	>	* be retried by caller if, by the time it tries to use the queue,
1895	>	* it is empty.
1896	>	*/
1897	>	private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
1898	>	// Similar to loop in scan(), but ignoring submissions
1899	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1900	>	int step = (r >>> 16) | 1;
1901	>	for (WorkQueue[] ws;;) {
1902	>	int rs = runState, m;
1903	>	if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
1904	>	return null;
1905	>	for (int j = (m + 1) << 2; ; r += step) {
1906	>	WorkQueue q = ws[((r << 1) | 1) & m];
1907	>	if (q != null && !q.isEmpty())
1908	>	return q;
1909	>	else if (--j < 0) {
1910	>	if (runState == rs)
1911	>	return null;
1912	>	break;
1913	>	}
1914	>	}
1915	>	}
1916		}
1917
1918	+
1919		/**
1920	<	* Creates or resizes array if necessary to hold newLength.
1921	<	* Call only under exclusion or lock.
1922	<	* @return the array
1923	<	*/
1924	<	private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
1925	<	ForkJoinWorkerThread[] ws = workers;
1926	<	if (ws == null)
1927	<	return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
1928	<	else if (newLength > ws.length)
1929	<	return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
1930	<	else
1931	<	return ws;
1920	>	* Runs tasks until {@code isQuiescent()}. We piggyback on
1921	>	* active count ctl maintenance, but rather than blocking
1922	>	* when tasks cannot be found, we rescan until all others cannot
1923	>	* find tasks either.
1924	>	*/
1925	>	final void helpQuiescePool(WorkQueue w) {
1926	>	for (boolean active = true;;) {
1927	>	ForkJoinTask<?> localTask; // exhaust local queue
1928	>	while ((localTask = w.nextLocalTask()) != null)
1929	>	localTask.doExec();
1930	>	WorkQueue q = findNonEmptyStealQueue(w);
1931	>	if (q != null) {
1932	>	ForkJoinTask<?> t; int b;
1933	>	if (!active) {      // re-establish active count
1934	>	long c;
1935	>	active = true;
1936	>	do {} while (!U.compareAndSwapLong
1937	>	(this, CTL, c = ctl, c + AC_UNIT));
1938	>	}
1939	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
1940	>	w.runSubtask(t);
1941	>	}
1942	>	else {
1943	>	long c;
1944	>	if (active) {       // decrement active count without queuing
1945	>	active = false;
1946	>	do {} while (!U.compareAndSwapLong
1947	>	(this, CTL, c = ctl, c -= AC_UNIT));
1948	>	}
1949	>	else
1950	>	c = ctl;        // re-increment on exit
1951	>	if ((int)(c >> AC_SHIFT) + parallelism == 0) {
1952	>	do {} while (!U.compareAndSwapLong
1953	>	(this, CTL, c = ctl, c + AC_UNIT));
1954	>	break;
1955	>	}
1956	>	}
1957	>	}
1958		}
1959
1960		/**
1961	<	* Try to shrink workers into smaller array after one or more terminate
1961	>	* Gets and removes a local or stolen task for the given worker.
1962	>	*
1963	>	* @return a task, if available
1964		*/
1965	<	private void tryShrinkWorkerArray() {
1966	<	ForkJoinWorkerThread[] ws = workers;
1967	<	if (ws != null) {
1968	<	int len = ws.length;
1969	<	int last = len - 1;
1970	<	while (last >= 0 && ws[last] == null)
1971	<	--last;
1972	<	int newLength = arraySizeFor(last+1);
1973	<	if (newLength < len)
458	<	workers = Arrays.copyOf(ws, newLength);
1965	>	final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
1966	>	for (ForkJoinTask<?> t;;) {
1967	>	WorkQueue q; int b;
1968	>	if ((t = w.nextLocalTask()) != null)
1969	>	return t;
1970	>	if ((q = findNonEmptyStealQueue(w)) == null)
1971	>	return null;
1972	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
1973	>	return t;
1974		}
1975		}
1976
1977		/**
1978	<	* Initialize workers if necessary
1979	<	*/
1980	<	final void ensureWorkerInitialization() {
1981	<	ForkJoinWorkerThread[] ws = workers;
1982	<	if (ws == null) {
1983	<	final ReentrantLock lock = this.workerLock;
1984	<	lock.lock();
1985	<	try {
1986	<	ws = workers;
1987	<	if (ws == null) {
1988	<	int ps = parallelism;
1989	<	ws = ensureWorkerArrayCapacity(ps);
1990	<	for (int i = 0; i < ps; ++i) {
1991	<	ForkJoinWorkerThread w = createWorker(i);
1992	<	if (w != null) {
1993	<	ws[i] = w;
1994	<	w.start();
1995	<	updateWorkerCount(1);
1978	>	* Returns the approximate (non-atomic) number of idle threads per
1979	>	* active thread to offset steal queue size for method
1980	>	* ForkJoinTask.getSurplusQueuedTaskCount().
1981	>	*/
1982	>	final int idlePerActive() {
1983	>	// Approximate at powers of two for small values, saturate past 4
1984	>	int p = parallelism;
1985	>	int a = p + (int)(ctl >> AC_SHIFT);
1986	>	return (a > (p >>>= 1) ? 0 :
1987	>	a > (p >>>= 1) ? 1 :
1988	>	a > (p >>>= 1) ? 2 :
1989	>	a > (p >>>= 1) ? 4 :
1990	>	8);
1991	>	}
1992	>
1993	>	//  Termination
1994	>
1995	>	/**
1996	>	* Possibly initiates and/or completes termination.  The caller
1997	>	* triggering termination runs three passes through workQueues:
1998	>	* (0) Setting termination status, followed by wakeups of queued
1999	>	* workers; (1) cancelling all tasks; (2) interrupting lagging
2000	>	* threads (likely in external tasks, but possibly also blocked in
2001	>	* joins).  Each pass repeats previous steps because of potential
2002	>	* lagging thread creation.
2003	>	*
2004	>	* @param now if true, unconditionally terminate, else only
2005	>	* if no work and no active workers
2006	>	* @param enable if true, enable shutdown when next possible
2007	>	* @return true if now terminating or terminated
2008	>	*/
2009	>	private boolean tryTerminate(boolean now, boolean enable) {
2010	>	Mutex lock = this.lock;
2011	>	for (long c;;) {
2012	>	if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2013	>	if ((short)(c >>> TC_SHIFT) == -parallelism) {
2014	>	lock.lock();                    // don't need try/finally
2015	>	termination.signalAll();        // signal when 0 workers
2016	>	lock.unlock();
2017	>	}
2018	>	return true;
2019	>	}
2020	>	if (runState >= 0) {                    // not yet enabled
2021	>	if (!enable)
2022	>	return false;
2023	>	lock.lock();
2024	>	runState |= SHUTDOWN;
2025	>	lock.unlock();
2026	>	}
2027	>	if (!now) {                             // check if idle & no tasks
2028	>	if ((int)(c >> AC_SHIFT) != -parallelism ||
2029	>	hasQueuedSubmissions())
2030	>	return false;
2031	>	// Check for unqueued inactive workers. One pass suffices.
2032	>	WorkQueue[] ws = workQueues; WorkQueue w;
2033	>	if (ws != null) {
2034	>	for (int i = 1; i < ws.length; i += 2) {
2035	>	if ((w = ws[i]) != null && w.eventCount >= 0)
2036	>	return false;
2037	>	}
2038	>	}
2039	>	}
2040	>	if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2041	>	for (int pass = 0; pass < 3; ++pass) {
2042	>	WorkQueue[] ws = workQueues;
2043	>	if (ws != null) {
2044	>	WorkQueue w;
2045	>	int n = ws.length;
2046	>	for (int i = 0; i < n; ++i) {
2047	>	if ((w = ws[i]) != null) {
2048	>	w.runState = -1;
2049	>	if (pass > 0) {
2050	>	w.cancelAll();
2051	>	if (pass > 1)
2052	>	w.interruptOwner();
2053	>	}
2054	>	}
2055	>	}
2056	>	// Wake up workers parked on event queue
2057	>	int i, e; long cc; Thread p;
2058	>	while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2059	>	(i = e & SMASK) < n &&
2060	>	(w = ws[i]) != null) {
2061	>	long nc = ((long)(w.nextWait & E_MASK) |
2062	>	((cc + AC_UNIT) & AC_MASK) |
2063	>	(cc & (TC_MASK|STOP_BIT)));
2064	>	if (w.eventCount == (e | INT_SIGN) &&
2065	>	U.compareAndSwapLong(this, CTL, cc, nc)) {
2066	>	w.eventCount = (e + E_SEQ) & E_MASK;
2067	>	w.runState = -1;
2068	>	if ((p = w.parker) != null)
2069	>	U.unpark(p);
2070	>	}
2071		}
2072		}
2073		}
484	–	} finally {
485	–	lock.unlock();
2074		}
2075		}
2076		}
2077
2078	+	// Exported methods
2079	+
2080	+	// Constructors
2081	+
2082		/**
2083	<	* Worker creation and startup for threads added via setParallelism.
2083	>	* Creates a {@code ForkJoinPool} with parallelism equal to {@link
2084	>	* java.lang.Runtime#availableProcessors}, using the {@linkplain
2085	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
2086	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
2087	>	*
2088	>	* @throws SecurityException if a security manager exists and
2089	>	*         the caller is not permitted to modify threads
2090	>	*         because it does not hold {@link
2091	>	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2092		*/
2093	<	private void createAndStartAddedWorkers() {
2094	<	resumeAllSpares();  // Allow spares to convert to nonspare
2095	<	int ps = parallelism;
496	<	ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
497	<	int len = ws.length;
498	<	// Sweep through slots, to keep lowest indices most populated
499	<	int k = 0;
500	<	while (k < len) {
501	<	if (ws[k] != null) {
502	<	++k;
503	<	continue;
504	<	}
505	<	int s = workerCounts;
506	<	int tc = totalCountOf(s);
507	<	int rc = runningCountOf(s);
508	<	if (rc >= ps || tc >= ps)
509	<	break;
510	<	if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
511	<	ForkJoinWorkerThread w = createWorker(k);
512	<	if (w != null) {
513	<	ws[k++] = w;
514	<	w.start();
515	<	}
516	<	else {
517	<	updateWorkerCount(-1); // back out on failed creation
518	<	break;
519	<	}
520	<	}
521	<	}
2093	>	public ForkJoinPool() {
2094	>	this(Runtime.getRuntime().availableProcessors(),
2095	>	defaultForkJoinWorkerThreadFactory, null, false);
2096		}
2097
2098	<	// Execution methods
2098	>	/**
2099	>	* Creates a {@code ForkJoinPool} with the indicated parallelism
2100	>	* level, the {@linkplain
2101	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
2102	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
2103	>	*
2104	>	* @param parallelism the parallelism level
2105	>	* @throws IllegalArgumentException if parallelism less than or
2106	>	*         equal to zero, or greater than implementation limit
2107	>	* @throws SecurityException if a security manager exists and
2108	>	*         the caller is not permitted to modify threads
2109	>	*         because it does not hold {@link
2110	>	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2111	>	*/
2112	>	public ForkJoinPool(int parallelism) {
2113	>	this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
2114	>	}
2115
2116		/**
2117	<	* Common code for execute, invoke and submit
2117	>	* Creates a {@code ForkJoinPool} with the given parameters.
2118	>	*
2119	>	* @param parallelism the parallelism level. For default value,
2120	>	* use {@link java.lang.Runtime#availableProcessors}.
2121	>	* @param factory the factory for creating new threads. For default value,
2122	>	* use {@link #defaultForkJoinWorkerThreadFactory}.
2123	>	* @param handler the handler for internal worker threads that
2124	>	* terminate due to unrecoverable errors encountered while executing
2125	>	* tasks. For default value, use {@code null}.
2126	>	* @param asyncMode if true,
2127	>	* establishes local first-in-first-out scheduling mode for forked
2128	>	* tasks that are never joined. This mode may be more appropriate
2129	>	* than default locally stack-based mode in applications in which
2130	>	* worker threads only process event-style asynchronous tasks.
2131	>	* For default value, use {@code false}.
2132	>	* @throws IllegalArgumentException if parallelism less than or
2133	>	*         equal to zero, or greater than implementation limit
2134	>	* @throws NullPointerException if the factory is null
2135	>	* @throws SecurityException if a security manager exists and
2136	>	*         the caller is not permitted to modify threads
2137	>	*         because it does not hold {@link
2138	>	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2139		*/
2140	<	private <T> void doSubmit(ForkJoinTask<T> task) {
2141	<	if (isShutdown())
2142	<	throw new RejectedExecutionException();
2143	<	if (workers == null)
2144	<	ensureWorkerInitialization();
2145	<	submissionQueue.offer(task);
2146	<	signalIdleWorkers();
2140	>	public ForkJoinPool(int parallelism,
2141	>	ForkJoinWorkerThreadFactory factory,
2142	>	Thread.UncaughtExceptionHandler handler,
2143	>	boolean asyncMode) {
2144	>	checkPermission();
2145	>	if (factory == null)
2146	>	throw new NullPointerException();
2147	>	if (parallelism <= 0 || parallelism > MAX_CAP)
2148	>	throw new IllegalArgumentException();
2149	>	this.parallelism = parallelism;
2150	>	this.factory = factory;
2151	>	this.ueh = handler;
2152	>	this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2153	>	long np = (long)(-parallelism); // offset ctl counts
2154	>	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2155	>	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2156	>	int n = parallelism - 1;
2157	>	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
2158	>	int size = (n + 1) << 1;        // #slots = 2*#workers
2159	>	this.submitMask = size - 1;     // room for max # of submit queues
2160	>	this.workQueues = new WorkQueue[size];
2161	>	this.termination = (this.lock = new Mutex()).newCondition();
2162	>	this.stealCount = new AtomicLong();
2163	>	this.nextWorkerNumber = new AtomicInteger();
2164	>	int pn = poolNumberGenerator.incrementAndGet();
2165	>	StringBuilder sb = new StringBuilder("ForkJoinPool-");
2166	>	sb.append(Integer.toString(pn));
2167	>	sb.append("-worker-");
2168	>	this.workerNamePrefix = sb.toString();
2169	>	lock.lock();
2170	>	this.runState = 1;              // set init flag
2171	>	lock.unlock();
2172		}
2173
2174	+	// Execution methods
2175	+
2176		/**
2177	<	* Performs the given task; returning its result upon completion
2177	>	* Performs the given task, returning its result upon completion.
2178	>	* If the computation encounters an unchecked Exception or Error,
2179	>	* it is rethrown as the outcome of this invocation.  Rethrown
2180	>	* exceptions behave in the same way as regular exceptions, but,
2181	>	* when possible, contain stack traces (as displayed for example
2182	>	* using {@code ex.printStackTrace()}) of both the current thread
2183	>	* as well as the thread actually encountering the exception;
2184	>	* minimally only the latter.
2185	>	*
2186		* @param task the task
2187		* @return the task's result
2188	<	* @throws NullPointerException if task is null
2189	<	* @throws RejectedExecutionException if pool is shut down
2188	>	* @throws NullPointerException if the task is null
2189	>	* @throws RejectedExecutionException if the task cannot be
2190	>	*         scheduled for execution
2191		*/
2192		public <T> T invoke(ForkJoinTask<T> task) {
2193	+	if (task == null)
2194	+	throw new NullPointerException();
2195		doSubmit(task);
2196		return task.join();
2197		}
2198
2199		/**
2200		* Arranges for (asynchronous) execution of the given task.
2201	+	*
2202		* @param task the task
2203	<	* @throws NullPointerException if task is null
2204	<	* @throws RejectedExecutionException if pool is shut down
2203	>	* @throws NullPointerException if the task is null
2204	>	* @throws RejectedExecutionException if the task cannot be
2205	>	*         scheduled for execution
2206		*/
2207	<	public <T> void execute(ForkJoinTask<T> task) {
2207	>	public void execute(ForkJoinTask<?> task) {
2208	>	if (task == null)
2209	>	throw new NullPointerException();
2210		doSubmit(task);
2211		}
2212
2213		// AbstractExecutorService methods
2214
2215	+	/**
2216	+	* @throws NullPointerException if the task is null
2217	+	* @throws RejectedExecutionException if the task cannot be
2218	+	*         scheduled for execution
2219	+	*/
2220		public void execute(Runnable task) {
2221	<	doSubmit(new AdaptedRunnable<Void>(task, null));
2221	>	if (task == null)
2222	>	throw new NullPointerException();
2223	>	ForkJoinTask<?> job;
2224	>	if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2225	>	job = (ForkJoinTask<?>) task;
2226	>	else
2227	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2228	>	doSubmit(job);
2229		}
2230
2231	+	/**
2232	+	* Submits a ForkJoinTask for execution.
2233	+	*
2234	+	* @param task the task to submit
2235	+	* @return the task
2236	+	* @throws NullPointerException if the task is null
2237	+	* @throws RejectedExecutionException if the task cannot be
2238	+	*         scheduled for execution
2239	+	*/
2240	+	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2241	+	if (task == null)
2242	+	throw new NullPointerException();
2243	+	doSubmit(task);
2244	+	return task;
2245	+	}
2246	+
2247	+	/**
2248	+	* @throws NullPointerException if the task is null
2249	+	* @throws RejectedExecutionException if the task cannot be
2250	+	*         scheduled for execution
2251	+	*/
2252		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2253	<	ForkJoinTask<T> job = new AdaptedCallable<T>(task);
2253	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2254		doSubmit(job);
2255		return job;
2256		}
2257
2258	+	/**
2259	+	* @throws NullPointerException if the task is null
2260	+	* @throws RejectedExecutionException if the task cannot be
2261	+	*         scheduled for execution
2262	+	*/
2263		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2264	<	ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
2264	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2265		doSubmit(job);
2266		return job;
2267		}
2268
2269	+	/**
2270	+	* @throws NullPointerException if the task is null
2271	+	* @throws RejectedExecutionException if the task cannot be
2272	+	*         scheduled for execution
2273	+	*/
2274		public ForkJoinTask<?> submit(Runnable task) {
2275	<	ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
2275	>	if (task == null)
2276	>	throw new NullPointerException();
2277	>	ForkJoinTask<?> job;
2278	>	if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2279	>	job = (ForkJoinTask<?>) task;
2280	>	else
2281	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2282		doSubmit(job);
2283		return job;
2284		}
2285
2286		/**
2287	<	* Adaptor for Runnables. This implements RunnableFuture
2288	<	* to be compliant with AbstractExecutorService constraints
587	<	*/
588	<	static final class AdaptedRunnable<T> extends ForkJoinTask<T>
589	<	implements RunnableFuture<T> {
590	<	final Runnable runnable;
591	<	final T resultOnCompletion;
592	<	T result;
593	<	AdaptedRunnable(Runnable runnable, T result) {
594	<	if (runnable == null) throw new NullPointerException();
595	<	this.runnable = runnable;
596	<	this.resultOnCompletion = result;
597	<	}
598	<	public T getRawResult() { return result; }
599	<	public void setRawResult(T v) { result = v; }
600	<	public boolean exec() {
601	<	runnable.run();
602	<	result = resultOnCompletion;
603	<	return true;
604	<	}
605	<	public void run() { invoke(); }
606	<	}
607	<
608	<	/**
609	<	* Adaptor for Callables
2287	>	* @throws NullPointerException       {@inheritDoc}
2288	>	* @throws RejectedExecutionException {@inheritDoc}
2289		*/
611	–	static final class AdaptedCallable<T> extends ForkJoinTask<T>
612	–	implements RunnableFuture<T> {
613	–	final Callable<T> callable;
614	–	T result;
615	–	AdaptedCallable(Callable<T> callable) {
616	–	if (callable == null) throw new NullPointerException();
617	–	this.callable = callable;
618	–	}
619	–	public T getRawResult() { return result; }
620	–	public void setRawResult(T v) { result = v; }
621	–	public boolean exec() {
622	–	try {
623	–	result = callable.call();
624	–	return true;
625	–	} catch (Error err) {
626	–	throw err;
627	–	} catch (RuntimeException rex) {
628	–	throw rex;
629	–	} catch (Exception ex) {
630	–	throw new RuntimeException(ex);
631	–	}
632	–	}
633	–	public void run() { invoke(); }
634	–	}
635	–
2290		public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2291	<	ArrayList<ForkJoinTask<T>> ts =
2292	<	new ArrayList<ForkJoinTask<T>>(tasks.size());
2293	<	for (Callable<T> c : tasks)
2294	<	ts.add(new AdaptedCallable<T>(c));
2295	<	invoke(new InvokeAll<T>(ts));
2296	<	return (List<Future<T>>)(List)ts;
2297	<	}
2291	>	// In previous versions of this class, this method constructed
2292	>	// a task to run ForkJoinTask.invokeAll, but now external
2293	>	// invocation of multiple tasks is at least as efficient.
2294	>	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2295	>	// Workaround needed because method wasn't declared with
2296	>	// wildcards in return type but should have been.
2297	>	@SuppressWarnings({"unchecked", "rawtypes"})
2298	>	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2299
2300	<	static final class InvokeAll<T> extends RecursiveAction {
2301	<	final ArrayList<ForkJoinTask<T>> tasks;
2302	<	InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
2303	<	public void compute() {
2304	<	try { invokeAll(tasks); } catch(Exception ignore) {}
2300	>	boolean done = false;
2301	>	try {
2302	>	for (Callable<T> t : tasks) {
2303	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2304	>	doSubmit(f);
2305	>	fs.add(f);
2306	>	}
2307	>	for (ForkJoinTask<T> f : fs)
2308	>	f.quietlyJoin();
2309	>	done = true;
2310	>	return futures;
2311	>	} finally {
2312	>	if (!done)
2313	>	for (ForkJoinTask<T> f : fs)
2314	>	f.cancel(false);
2315		}
2316		}
2317
653	–	// Configuration and status settings and queries
654	–
2318		/**
2319	<	* Returns the factory used for constructing new workers
2319	>	* Returns the factory used for constructing new workers.
2320		*
2321		* @return the factory used for constructing new workers
2322		*/
#	Line 664 | Line 2327 | public class ForkJoinPool extends Abstra
2327		/**
2328		* Returns the handler for internal worker threads that terminate
2329		* due to unrecoverable errors encountered while executing tasks.
667	–	* @return the handler, or null if none
668	–	*/
669	–	public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
670	–	Thread.UncaughtExceptionHandler h;
671	–	final ReentrantLock lock = this.workerLock;
672	–	lock.lock();
673	–	try {
674	–	h = ueh;
675	–	} finally {
676	–	lock.unlock();
677	–	}
678	–	return h;
679	–	}
680	–
681	–	/**
682	–	* Sets the handler for internal worker threads that terminate due
683	–	* to unrecoverable errors encountered while executing tasks.
684	–	* Unless set, the current default or ThreadGroup handler is used
685	–	* as handler.
2330		*
2331	<	* @param h the new handler
688	<	* @return the old handler, or null if none
689	<	* @throws SecurityException if a security manager exists and
690	<	*         the caller is not permitted to modify threads
691	<	*         because it does not hold {@link
692	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
2331	>	* @return the handler, or {@code null} if none
2332		*/
2333	<	public Thread.UncaughtExceptionHandler
2334	<	setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
696	<	checkPermission();
697	<	Thread.UncaughtExceptionHandler old = null;
698	<	final ReentrantLock lock = this.workerLock;
699	<	lock.lock();
700	<	try {
701	<	old = ueh;
702	<	ueh = h;
703	<	ForkJoinWorkerThread[] ws = workers;
704	<	if (ws != null) {
705	<	for (int i = 0; i < ws.length; ++i) {
706	<	ForkJoinWorkerThread w = ws[i];
707	<	if (w != null)
708	<	w.setUncaughtExceptionHandler(h);
709	<	}
710	<	}
711	<	} finally {
712	<	lock.unlock();
713	<	}
714	<	return old;
715	<	}
716	<
717	<
718	<	/**
719	<	* Sets the target parallelism level of this pool.
720	<	* @param parallelism the target parallelism
721	<	* @throws IllegalArgumentException if parallelism less than or
722	<	* equal to zero or greater than maximum size bounds.
723	<	* @throws SecurityException if a security manager exists and
724	<	*         the caller is not permitted to modify threads
725	<	*         because it does not hold {@link
726	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
727	<	*/
728	<	public void setParallelism(int parallelism) {
729	<	checkPermission();
730	<	if (parallelism <= 0 || parallelism > maxPoolSize)
731	<	throw new IllegalArgumentException();
732	<	final ReentrantLock lock = this.workerLock;
733	<	lock.lock();
734	<	try {
735	<	if (!isTerminating()) {
736	<	int p = this.parallelism;
737	<	this.parallelism = parallelism;
738	<	if (parallelism > p)
739	<	createAndStartAddedWorkers();
740	<	else
741	<	trimSpares();
742	<	}
743	<	} finally {
744	<	lock.unlock();
745	<	}
746	<	signalIdleWorkers();
2333	>	public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
2334	>	return ueh;
2335		}
2336
2337		/**
2338	<	* Returns the targeted number of worker threads in this pool.
2338	>	* Returns the targeted parallelism level of this pool.
2339		*
2340	<	* @return the targeted number of worker threads in this pool
2340	>	* @return the targeted parallelism level of this pool
2341		*/
2342		public int getParallelism() {
2343		return parallelism;
#	Line 757 | Line 2345 | public class ForkJoinPool extends Abstra
2345
2346		/**
2347		* Returns the number of worker threads that have started but not
2348	<	* yet terminated.  This result returned by this method may differ
2349	<	* from {@code getParallelism} when threads are created to
2348	>	* yet terminated.  The result returned by this method may differ
2349	>	* from {@link #getParallelism} when threads are created to
2350		* maintain parallelism when others are cooperatively blocked.
2351		*
2352		* @return the number of worker threads
2353		*/
2354		public int getPoolSize() {
2355	<	return totalCountOf(workerCounts);
768	<	}
769	<
770	<	/**
771	<	* Returns the maximum number of threads allowed to exist in the
772	<	* pool, even if there are insufficient unblocked running threads.
773	<	* @return the maximum
774	<	*/
775	<	public int getMaximumPoolSize() {
776	<	return maxPoolSize;
777	<	}
778	<
779	<	/**
780	<	* Sets the maximum number of threads allowed to exist in the
781	<	* pool, even if there are insufficient unblocked running threads.
782	<	* Setting this value has no effect on current pool size. It
783	<	* controls construction of new threads.
784	<	* @throws IllegalArgumentException if negative or greater then
785	<	* internal implementation limit.
786	<	*/
787	<	public void setMaximumPoolSize(int newMax) {
788	<	if (newMax < 0 || newMax > MAX_THREADS)
789	<	throw new IllegalArgumentException();
790	<	maxPoolSize = newMax;
2355	>	return parallelism + (short)(ctl >>> TC_SHIFT);
2356		}
2357
793	–
2358		/**
2359	<	* Returns true if this pool dynamically maintains its target
796	<	* parallelism level. If false, new threads are added only to
797	<	* avoid possible starvation.
798	<	* This setting is by default true;
799	<	* @return true if maintains parallelism
800	<	*/
801	<	public boolean getMaintainsParallelism() {
802	<	return maintainsParallelism;
803	<	}
804	<
805	<	/**
806	<	* Sets whether this pool dynamically maintains its target
807	<	* parallelism level. If false, new threads are added only to
808	<	* avoid possible starvation.
809	<	* @param enable true to maintains parallelism
810	<	*/
811	<	public void setMaintainsParallelism(boolean enable) {
812	<	maintainsParallelism = enable;
813	<	}
814	<
815	<	/**
816	<	* Establishes local first-in-first-out scheduling mode for forked
817	<	* tasks that are never joined. This mode may be more appropriate
818	<	* than default locally stack-based mode in applications in which
819	<	* worker threads only process asynchronous tasks.  This method is
820	<	* designed to be invoked only when pool is quiescent, and
821	<	* typically only before any tasks are submitted. The effects of
822	<	* invocations at other times may be unpredictable.
823	<	*
824	<	* @param async if true, use locally FIFO scheduling
825	<	* @return the previous mode.
826	<	*/
827	<	public boolean setAsyncMode(boolean async) {
828	<	boolean oldMode = locallyFifo;
829	<	locallyFifo = async;
830	<	ForkJoinWorkerThread[] ws = workers;
831	<	if (ws != null) {
832	<	for (int i = 0; i < ws.length; ++i) {
833	<	ForkJoinWorkerThread t = ws[i];
834	<	if (t != null)
835	<	t.setAsyncMode(async);
836	<	}
837	<	}
838	<	return oldMode;
839	<	}
840	<
841	<	/**
842	<	* Returns true if this pool uses local first-in-first-out
2359	>	* Returns {@code true} if this pool uses local first-in-first-out
2360		* scheduling mode for forked tasks that are never joined.
2361		*
2362	<	* @return true if this pool uses async mode.
2362	>	* @return {@code true} if this pool uses async mode
2363		*/
2364		public boolean getAsyncMode() {
2365	<	return locallyFifo;
2365	>	return localMode != 0;
2366		}
2367
2368		/**
2369		* Returns an estimate of the number of worker threads that are
2370		* not blocked waiting to join tasks or for other managed
2371	<	* synchronization.
2371	>	* synchronization. This method may overestimate the
2372	>	* number of running threads.
2373		*
2374		* @return the number of worker threads
2375		*/
2376		public int getRunningThreadCount() {
2377	<	return runningCountOf(workerCounts);
2377	>	int rc = 0;
2378	>	WorkQueue[] ws; WorkQueue w;
2379	>	if ((ws = workQueues) != null) {
2380	>	for (int i = 1; i < ws.length; i += 2) {
2381	>	if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2382	>	++rc;
2383	>	}
2384	>	}
2385	>	return rc;
2386		}
2387
2388		/**
2389		* Returns an estimate of the number of threads that are currently
2390		* stealing or executing tasks. This method may overestimate the
2391		* number of active threads.
2392	<	* @return the number of active threads.
2392	>	*
2393	>	* @return the number of active threads
2394		*/
2395		public int getActiveThreadCount() {
2396	<	return activeCountOf(runControl);
2396	>	int r = parallelism + (int)(ctl >> AC_SHIFT);
2397	>	return (r <= 0) ? 0 : r; // suppress momentarily negative values
2398		}
2399
2400		/**
2401	<	* Returns an estimate of the number of threads that are currently
2402	<	* idle waiting for tasks. This method may underestimate the
2403	<	* number of idle threads.
2404	<	* @return the number of idle threads.
2405	<	*/
2406	<	final int getIdleThreadCount() {
2407	<	int c = runningCountOf(workerCounts) - activeCountOf(runControl);
2408	<	return (c <= 0)? 0 : c;
2409	<	}
882	<
883	<	/**
884	<	* Returns true if all worker threads are currently idle. An idle
885	<	* worker is one that cannot obtain a task to execute because none
886	<	* are available to steal from other threads, and there are no
887	<	* pending submissions to the pool. This method is conservative:
888	<	* It might not return true immediately upon idleness of all
889	<	* threads, but will eventually become true if threads remain
890	<	* inactive.
891	<	* @return true if all threads are currently idle
2401	>	* Returns {@code true} if all worker threads are currently idle.
2402	>	* An idle worker is one that cannot obtain a task to execute
2403	>	* because none are available to steal from other threads, and
2404	>	* there are no pending submissions to the pool. This method is
2405	>	* conservative; it might not return {@code true} immediately upon
2406	>	* idleness of all threads, but will eventually become true if
2407	>	* threads remain inactive.
2408	>	*
2409	>	* @return {@code true} if all threads are currently idle
2410		*/
2411		public boolean isQuiescent() {
2412	<	return activeCountOf(runControl) == 0;
2412	>	return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2413		}
2414
2415		/**
#	Line 899 | Line 2417 | public class ForkJoinPool extends Abstra
2417		* one thread's work queue by another. The reported value
2418		* underestimates the actual total number of steals when the pool
2419		* is not quiescent. This value may be useful for monitoring and
2420	<	* tuning fork/join programs: In general, steal counts should be
2420	>	* tuning fork/join programs: in general, steal counts should be
2421		* high enough to keep threads busy, but low enough to avoid
2422		* overhead and contention across threads.
2423	<	* @return the number of steals.
2423	>	*
2424	>	* @return the number of steals
2425		*/
2426		public long getStealCount() {
2427	<	return stealCount.get();
2428	<	}
2429	<
2430	<	/**
2431	<	* Accumulate steal count from a worker. Call only
2432	<	* when worker known to be idle.
2433	<	*/
2434	<	private void updateStealCount(ForkJoinWorkerThread w) {
2435	<	int sc = w.getAndClearStealCount();
917	<	if (sc != 0)
918	<	stealCount.addAndGet(sc);
2427	>	long count = stealCount.get();
2428	>	WorkQueue[] ws; WorkQueue w;
2429	>	if ((ws = workQueues) != null) {
2430	>	for (int i = 1; i < ws.length; i += 2) {
2431	>	if ((w = ws[i]) != null)
2432	>	count += w.totalSteals;
2433	>	}
2434	>	}
2435	>	return count;
2436		}
2437
2438		/**
#	Line 925 | Line 2442 | public class ForkJoinPool extends Abstra
2442		* an approximation, obtained by iterating across all threads in
2443		* the pool. This method may be useful for tuning task
2444		* granularities.
2445	<	* @return the number of queued tasks.
2445	>	*
2446	>	* @return the number of queued tasks
2447		*/
2448		public long getQueuedTaskCount() {
2449		long count = 0;
2450	<	ForkJoinWorkerThread[] ws = workers;
2451	<	if (ws != null) {
2452	<	for (int i = 0; i < ws.length; ++i) {
2453	<	ForkJoinWorkerThread t = ws[i];
2454	<	if (t != null)
937	<	count += t.getQueueSize();
2450	>	WorkQueue[] ws; WorkQueue w;
2451	>	if ((ws = workQueues) != null) {
2452	>	for (int i = 1; i < ws.length; i += 2) {
2453	>	if ((w = ws[i]) != null)
2454	>	count += w.queueSize();
2455		}
2456		}
2457		return count;
2458		}
2459
2460		/**
2461	<	* Returns an estimate of the number tasks submitted to this pool
2462	<	* that have not yet begun executing. This method takes time
2463	<	* proportional to the number of submissions.
2464	<	* @return the number of queued submissions.
2461	>	* Returns an estimate of the number of tasks submitted to this
2462	>	* pool that have not yet begun executing.  This method may take
2463	>	* time proportional to the number of submissions.
2464	>	*
2465	>	* @return the number of queued submissions
2466		*/
2467		public int getQueuedSubmissionCount() {
2468	<	return submissionQueue.size();
2468	>	int count = 0;
2469	>	WorkQueue[] ws; WorkQueue w;
2470	>	if ((ws = workQueues) != null) {
2471	>	for (int i = 0; i < ws.length; i += 2) {
2472	>	if ((w = ws[i]) != null)
2473	>	count += w.queueSize();
2474	>	}
2475	>	}
2476	>	return count;
2477		}
2478
2479		/**
2480	<	* Returns true if there are any tasks submitted to this pool
2481	<	* that have not yet begun executing.
2482	<	* @return {@code true} if there are any queued submissions.
2480	>	* Returns {@code true} if there are any tasks submitted to this
2481	>	* pool that have not yet begun executing.
2482	>	*
2483	>	* @return {@code true} if there are any queued submissions
2484		*/
2485		public boolean hasQueuedSubmissions() {
2486	<	return !submissionQueue.isEmpty();
2486	>	WorkQueue[] ws; WorkQueue w;
2487	>	if ((ws = workQueues) != null) {
2488	>	for (int i = 0; i < ws.length; i += 2) {
2489	>	if ((w = ws[i]) != null && !w.isEmpty())
2490	>	return true;
2491	>	}
2492	>	}
2493	>	return false;
2494		}
2495
2496		/**
2497		* Removes and returns the next unexecuted submission if one is
2498		* available.  This method may be useful in extensions to this
2499		* class that re-assign work in systems with multiple pools.
2500	<	* @return the next submission, or null if none
2500	>	*
2501	>	* @return the next submission, or {@code null} if none
2502		*/
2503		protected ForkJoinTask<?> pollSubmission() {
2504	<	return submissionQueue.poll();
2504	>	WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2505	>	if ((ws = workQueues) != null) {
2506	>	for (int i = 0; i < ws.length; i += 2) {
2507	>	if ((w = ws[i]) != null && (t = w.poll()) != null)
2508	>	return t;
2509	>	}
2510	>	}
2511	>	return null;
2512		}
2513
2514		/**
2515		* Removes all available unexecuted submitted and forked tasks
2516		* from scheduling queues and adds them to the given collection,
2517		* without altering their execution status. These may include
2518	<	* artificially generated or wrapped tasks. This method is designed
2519	<	* to be invoked only when the pool is known to be
2518	>	* artificially generated or wrapped tasks. This method is
2519	>	* designed to be invoked only when the pool is known to be
2520		* quiescent. Invocations at other times may not remove all
2521		* tasks. A failure encountered while attempting to add elements
2522		* to collection {@code c} may result in elements being in
#	Line 982 | Line 2524 | public class ForkJoinPool extends Abstra
2524		* exception is thrown.  The behavior of this operation is
2525		* undefined if the specified collection is modified while the
2526		* operation is in progress.
2527	+	*
2528		* @param c the collection to transfer elements into
2529		* @return the number of elements transferred
2530		*/
2531	<	protected int drainTasksTo(Collection<ForkJoinTask<?>> c) {
2532	<	int n = submissionQueue.drainTo(c);
2533	<	ForkJoinWorkerThread[] ws = workers;
2534	<	if (ws != null) {
2531	>	protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
2532	>	int count = 0;
2533	>	WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2534	>	if ((ws = workQueues) != null) {
2535		for (int i = 0; i < ws.length; ++i) {
2536	<	ForkJoinWorkerThread w = ws[i];
2537	<	if (w != null)
2538	<	n += w.drainTasksTo(c);
2536	>	if ((w = ws[i]) != null) {
2537	>	while ((t = w.poll()) != null) {
2538	>	c.add(t);
2539	>	++count;
2540	>	}
2541	>	}
2542		}
2543		}
2544	<	return n;
2544	>	return count;
2545		}
2546
2547		/**
#	Line 1006 | Line 2552 | public class ForkJoinPool extends Abstra
2552		* @return a string identifying this pool, as well as its state
2553		*/
2554		public String toString() {
2555	<	int ps = parallelism;
2556	<	int wc = workerCounts;
2557	<	int rc = runControl;
2558	<	long st = getStealCount();
2559	<	long qt = getQueuedTaskCount();
2560	<	long qs = getQueuedSubmissionCount();
2555	>	// Use a single pass through workQueues to collect counts
2556	>	long qt = 0L, qs = 0L; int rc = 0;
2557	>	long st = stealCount.get();
2558	>	long c = ctl;
2559	>	WorkQueue[] ws; WorkQueue w;
2560	>	if ((ws = workQueues) != null) {
2561	>	for (int i = 0; i < ws.length; ++i) {
2562	>	if ((w = ws[i]) != null) {
2563	>	int size = w.queueSize();
2564	>	if ((i & 1) == 0)
2565	>	qs += size;
2566	>	else {
2567	>	qt += size;
2568	>	st += w.totalSteals;
2569	>	if (w.isApparentlyUnblocked())
2570	>	++rc;
2571	>	}
2572	>	}
2573	>	}
2574	>	}
2575	>	int pc = parallelism;
2576	>	int tc = pc + (short)(c >>> TC_SHIFT);
2577	>	int ac = pc + (int)(c >> AC_SHIFT);
2578	>	if (ac < 0) // ignore transient negative
2579	>	ac = 0;
2580	>	String level;
2581	>	if ((c & STOP_BIT) != 0)
2582	>	level = (tc == 0) ? "Terminated" : "Terminating";
2583	>	else
2584	>	level = runState < 0 ? "Shutting down" : "Running";
2585		return super.toString() +
2586	<	"[" + runStateToString(runStateOf(rc)) +
2587	<	", parallelism = " + ps +
2588	<	", size = " + totalCountOf(wc) +
2589	<	", active = " + activeCountOf(rc) +
2590	<	", running = " + runningCountOf(wc) +
2586	>	"[" + level +
2587	>	", parallelism = " + pc +
2588	>	", size = " + tc +
2589	>	", active = " + ac +
2590	>	", running = " + rc +
2591		", steals = " + st +
2592		", tasks = " + qt +
2593		", submissions = " + qs +
2594		"]";
2595		}
2596
1027	–	private static String runStateToString(int rs) {
1028	–	switch(rs) {
1029	–	case RUNNING: return "Running";
1030	–	case SHUTDOWN: return "Shutting down";
1031	–	case TERMINATING: return "Terminating";
1032	–	case TERMINATED: return "Terminated";
1033	–	default: throw new Error("Unknown run state");
1034	–	}
1035	–	}
1036	–
1037	–	// lifecycle control
1038	–
2597		/**
2598		* Initiates an orderly shutdown in which previously submitted
2599		* tasks are executed, but no new tasks will be accepted.
2600		* Invocation has no additional effect if already shut down.
2601		* Tasks that are in the process of being submitted concurrently
2602		* during the course of this method may or may not be rejected.
2603	+	*
2604		* @throws SecurityException if a security manager exists and
2605		*         the caller is not permitted to modify threads
2606		*         because it does not hold {@link
2607	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
2607	>	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2608		*/
2609		public void shutdown() {
2610		checkPermission();
2611	<	transitionRunStateTo(SHUTDOWN);
1053	<	if (canTerminateOnShutdown(runControl))
1054	<	terminateOnShutdown();
2611	>	tryTerminate(false, true);
2612		}
2613
2614		/**
2615	<	* Attempts to stop all actively executing tasks, and cancels all
2616	<	* waiting tasks.  Tasks that are in the process of being
2617	<	* submitted or executed concurrently during the course of this
2618	<	* method may or may not be rejected. Unlike some other executors,
2619	<	* this method cancels rather than collects non-executed tasks
2620	<	* upon termination, so always returns an empty list. However, you
2621	<	* can use method {@code drainTasksTo} before invoking this
2622	<	* method to transfer unexecuted tasks to another collection.
2615	>	* Attempts to cancel and/or stop all tasks, and reject all
2616	>	* subsequently submitted tasks.  Tasks that are in the process of
2617	>	* being submitted or executed concurrently during the course of
2618	>	* this method may or may not be rejected. This method cancels
2619	>	* both existing and unexecuted tasks, in order to permit
2620	>	* termination in the presence of task dependencies. So the method
2621	>	* always returns an empty list (unlike the case for some other
2622	>	* Executors).
2623	>	*
2624		* @return an empty list
2625		* @throws SecurityException if a security manager exists and
2626		*         the caller is not permitted to modify threads
2627		*         because it does not hold {@link
2628	<	*         java.lang.RuntimePermission}{@code ("modifyThread")},
2628	>	*         java.lang.RuntimePermission}{@code ("modifyThread")}
2629		*/
2630		public List<Runnable> shutdownNow() {
2631		checkPermission();
2632	<	terminate();
2632	>	tryTerminate(true, true);
2633		return Collections.emptyList();
2634		}
2635
#	Line 1081 | Line 2639 | public class ForkJoinPool extends Abstra
2639		* @return {@code true} if all tasks have completed following shut down
2640		*/
2641		public boolean isTerminated() {
2642	<	return runStateOf(runControl) == TERMINATED;
2642	>	long c = ctl;
2643	>	return ((c & STOP_BIT) != 0L &&
2644	>	(short)(c >>> TC_SHIFT) == -parallelism);
2645		}
2646
2647		/**
2648		* Returns {@code true} if the process of termination has
2649	<	* commenced but possibly not yet completed.
2649	>	* commenced but not yet completed.  This method may be useful for
2650	>	* debugging. A return of {@code true} reported a sufficient
2651	>	* period after shutdown may indicate that submitted tasks have
2652	>	* ignored or suppressed interruption, or are waiting for IO,
2653	>	* causing this executor not to properly terminate. (See the
2654	>	* advisory notes for class {@link ForkJoinTask} stating that
2655	>	* tasks should not normally entail blocking operations.  But if
2656	>	* they do, they must abort them on interrupt.)
2657		*
2658	<	* @return {@code true} if terminating
2658	>	* @return {@code true} if terminating but not yet terminated
2659		*/
2660		public boolean isTerminating() {
2661	<	return runStateOf(runControl) >= TERMINATING;
2661	>	long c = ctl;
2662	>	return ((c & STOP_BIT) != 0L &&
2663	>	(short)(c >>> TC_SHIFT) != -parallelism);
2664		}
2665
2666		/**
#	Line 1100 | Line 2669 | public class ForkJoinPool extends Abstra
2669		* @return {@code true} if this pool has been shut down
2670		*/
2671		public boolean isShutdown() {
2672	<	return runStateOf(runControl) >= SHUTDOWN;
2672	>	return runState < 0;
2673		}
2674
2675		/**
#	Line 1117 | Line 2686 | public class ForkJoinPool extends Abstra
2686		public boolean awaitTermination(long timeout, TimeUnit unit)
2687		throws InterruptedException {
2688		long nanos = unit.toNanos(timeout);
2689	<	final ReentrantLock lock = this.workerLock;
2689	>	final Mutex lock = this.lock;
2690		lock.lock();
2691		try {
2692		for (;;) {
#	Line 1132 | Line 2701 | public class ForkJoinPool extends Abstra
2701		}
2702		}
2703
1135	–	// Shutdown and termination support
1136	–
1137	–	/**
1138	–	* Callback from terminating worker. Null out the corresponding
1139	–	* workers slot, and if terminating, try to terminate, else try to
1140	–	* shrink workers array.
1141	–	* @param w the worker
1142	–	*/
1143	–	final void workerTerminated(ForkJoinWorkerThread w) {
1144	–	updateStealCount(w);
1145	–	updateWorkerCount(-1);
1146	–	final ReentrantLock lock = this.workerLock;
1147	–	lock.lock();
1148	–	try {
1149	–	ForkJoinWorkerThread[] ws = workers;
1150	–	if (ws != null) {
1151	–	int idx = w.poolIndex;
1152	–	if (idx >= 0 && idx < ws.length && ws[idx] == w)
1153	–	ws[idx] = null;
1154	–	if (totalCountOf(workerCounts) == 0) {
1155	–	terminate(); // no-op if already terminating
1156	–	transitionRunStateTo(TERMINATED);
1157	–	termination.signalAll();
1158	–	}
1159	–	else if (!isTerminating()) {
1160	–	tryShrinkWorkerArray();
1161	–	tryResumeSpare(true); // allow replacement
1162	–	}
1163	–	}
1164	–	} finally {
1165	–	lock.unlock();
1166	–	}
1167	–	signalIdleWorkers();
1168	–	}
1169	–
1170	–	/**
1171	–	* Initiate termination.
1172	–	*/
1173	–	private void terminate() {
1174	–	if (transitionRunStateTo(TERMINATING)) {
1175	–	stopAllWorkers();
1176	–	resumeAllSpares();
1177	–	signalIdleWorkers();
1178	–	cancelQueuedSubmissions();
1179	–	cancelQueuedWorkerTasks();
1180	–	interruptUnterminatedWorkers();
1181	–	signalIdleWorkers(); // resignal after interrupt
1182	–	}
1183	–	}
1184	–
1185	–	/**
1186	–	* Possibly terminates when on shutdown state.
1187	–	*/
1188	–	private void terminateOnShutdown() {
1189	–	if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1190	–	terminate();
1191	–	}
1192	–
1193	–	/**
1194	–	* Clears out and cancels submissions.
1195	–	*/
1196	–	private void cancelQueuedSubmissions() {
1197	–	ForkJoinTask<?> task;
1198	–	while ((task = pollSubmission()) != null)
1199	–	task.cancel(false);
1200	–	}
1201	–
1202	–	/**
1203	–	* Cleans out worker queues.
1204	–	*/
1205	–	private void cancelQueuedWorkerTasks() {
1206	–	final ReentrantLock lock = this.workerLock;
1207	–	lock.lock();
1208	–	try {
1209	–	ForkJoinWorkerThread[] ws = workers;
1210	–	if (ws != null) {
1211	–	for (int i = 0; i < ws.length; ++i) {
1212	–	ForkJoinWorkerThread t = ws[i];
1213	–	if (t != null)
1214	–	t.cancelTasks();
1215	–	}
1216	–	}
1217	–	} finally {
1218	–	lock.unlock();
1219	–	}
1220	–	}
1221	–
1222	–	/**
1223	–	* Sets each worker's status to terminating. Requires lock to avoid
1224	–	* conflicts with add/remove.
1225	–	*/
1226	–	private void stopAllWorkers() {
1227	–	final ReentrantLock lock = this.workerLock;
1228	–	lock.lock();
1229	–	try {
1230	–	ForkJoinWorkerThread[] ws = workers;
1231	–	if (ws != null) {
1232	–	for (int i = 0; i < ws.length; ++i) {
1233	–	ForkJoinWorkerThread t = ws[i];
1234	–	if (t != null)
1235	–	t.shutdownNow();
1236	–	}
1237	–	}
1238	–	} finally {
1239	–	lock.unlock();
1240	–	}
1241	–	}
1242	–
2704		/**
2705	<	* Interrupts all unterminated workers.  This is not required for
2706	<	* sake of internal control, but may help unstick user code during
1246	<	* shutdown.
1247	<	*/
1248	<	private void interruptUnterminatedWorkers() {
1249	<	final ReentrantLock lock = this.workerLock;
1250	<	lock.lock();
1251	<	try {
1252	<	ForkJoinWorkerThread[] ws = workers;
1253	<	if (ws != null) {
1254	<	for (int i = 0; i < ws.length; ++i) {
1255	<	ForkJoinWorkerThread t = ws[i];
1256	<	if (t != null && !t.isTerminated()) {
1257	<	try {
1258	<	t.interrupt();
1259	<	} catch (SecurityException ignore) {
1260	<	}
1261	<	}
1262	<	}
1263	<	}
1264	<	} finally {
1265	<	lock.unlock();
1266	<	}
1267	<	}
1268	<
1269	<
1270	<	/*
1271	<	* Nodes for event barrier to manage idle threads.  Queue nodes
1272	<	* are basic Treiber stack nodes, also used for spare stack.
2705	>	* Interface for extending managed parallelism for tasks running
2706	>	* in {@link ForkJoinPool}s.
2707		*
2708	<	* The event barrier has an event count and a wait queue (actually
2709	<	* a Treiber stack).  Workers are enabled to look for work when
2710	<	* the eventCount is incremented. If they fail to find work, they
2711	<	* may wait for next count. Upon release, threads help others wake
2712	<	* up.
2713	<	*
2714	<	* Synchronization events occur only in enough contexts to
2715	<	* maintain overall liveness:
2716	<	*
2717	<	*   - Submission of a new task to the pool
2718	<	*   - Resizes or other changes to the workers array
2719	<	*   - pool termination
2720	<	*   - A worker pushing a task on an empty queue
1287	<	*
1288	<	* The case of pushing a task occurs often enough, and is heavy
1289	<	* enough compared to simple stack pushes, to require special
1290	<	* handling: Method signalWork returns without advancing count if
1291	<	* the queue appears to be empty.  This would ordinarily result in
1292	<	* races causing some queued waiters not to be woken up. To avoid
1293	<	* this, the first worker enqueued in method sync (see
1294	<	* syncIsReleasable) rescans for tasks after being enqueued, and
1295	<	* helps signal if any are found. This works well because the
1296	<	* worker has nothing better to do, and so might as well help
1297	<	* alleviate the overhead and contention on the threads actually
1298	<	* doing work.  Also, since event counts increments on task
1299	<	* availability exist to maintain liveness (rather than to force
1300	<	* refreshes etc), it is OK for callers to exit early if
1301	<	* contending with another signaller.
1302	<	*/
1303	<	static final class WaitQueueNode {
1304	<	WaitQueueNode next; // only written before enqueued
1305	<	volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1306	<	final long count; // unused for spare stack
1307	<
1308	<	WaitQueueNode(long c, ForkJoinWorkerThread w) {
1309	<	count = c;
1310	<	thread = w;
1311	<	}
1312	<
1313	<	/**
1314	<	* Wakes up waiter, returning false if known to already
1315	<	*/
1316	<	boolean signal() {
1317	<	ForkJoinWorkerThread t = thread;
1318	<	if (t == null)
1319	<	return false;
1320	<	thread = null;
1321	<	LockSupport.unpark(t);
1322	<	return true;
1323	<	}
1324	<
1325	<	/**
1326	<	* Awaits release on sync.
1327	<	*/
1328	<	void awaitSyncRelease(ForkJoinPool p) {
1329	<	while (thread != null && !p.syncIsReleasable(this))
1330	<	LockSupport.park(this);
1331	<	}
1332	<
1333	<	/**
1334	<	* Awaits resumption as spare.
1335	<	*/
1336	<	void awaitSpareRelease() {
1337	<	while (thread != null) {
1338	<	if (!Thread.interrupted())
1339	<	LockSupport.park(this);
1340	<	}
1341	<	}
1342	<	}
1343	<
1344	<	/**
1345	<	* Ensures that no thread is waiting for count to advance from the
1346	<	* current value of eventCount read on entry to this method, by
1347	<	* releasing waiting threads if necessary.
1348	<	* @return the count
1349	<	*/
1350	<	final long ensureSync() {
1351	<	long c = eventCount;
1352	<	WaitQueueNode q;
1353	<	while ((q = syncStack) != null && q.count < c) {
1354	<	if (casBarrierStack(q, null)) {
1355	<	do {
1356	<	q.signal();
1357	<	} while ((q = q.next) != null);
1358	<	break;
1359	<	}
1360	<	}
1361	<	return c;
1362	<	}
1363	<
1364	<	/**
1365	<	* Increments event count and releases waiting threads.
1366	<	*/
1367	<	private void signalIdleWorkers() {
1368	<	long c;
1369	<	do;while (!casEventCount(c = eventCount, c+1));
1370	<	ensureSync();
1371	<	}
1372	<
1373	<	/**
1374	<	* Signals threads waiting to poll a task. Because method sync
1375	<	* rechecks availability, it is OK to only proceed if queue
1376	<	* appears to be non-empty, and OK to skip under contention to
1377	<	* increment count (since some other thread succeeded).
1378	<	*/
1379	<	final void signalWork() {
1380	<	long c;
1381	<	WaitQueueNode q;
1382	<	if (syncStack != null &&
1383	<	casEventCount(c = eventCount, c+1) &&
1384	<	(((q = syncStack) != null && q.count <= c) &&
1385	<	(!casBarrierStack(q, q.next) || !q.signal())))
1386	<	ensureSync();
1387	<	}
1388	<
1389	<	/**
1390	<	* Waits until event count advances from last value held by
1391	<	* caller, or if excess threads, caller is resumed as spare, or
1392	<	* caller or pool is terminating. Updates caller's event on exit.
1393	<	* @param w the calling worker thread
1394	<	*/
1395	<	final void sync(ForkJoinWorkerThread w) {
1396	<	updateStealCount(w); // Transfer w's count while it is idle
1397	<
1398	<	while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1399	<	long prev = w.lastEventCount;
1400	<	WaitQueueNode node = null;
1401	<	WaitQueueNode h;
1402	<	while (eventCount == prev &&
1403	<	((h = syncStack) == null || h.count == prev)) {
1404	<	if (node == null)
1405	<	node = new WaitQueueNode(prev, w);
1406	<	if (casBarrierStack(node.next = h, node)) {
1407	<	node.awaitSyncRelease(this);
1408	<	break;
1409	<	}
1410	<	}
1411	<	long ec = ensureSync();
1412	<	if (ec != prev) {
1413	<	w.lastEventCount = ec;
1414	<	break;
1415	<	}
1416	<	}
1417	<	}
1418	<
1419	<	/**
1420	<	* Returns true if worker waiting on sync can proceed:
1421	<	*  - on signal (thread == null)
1422	<	*  - on event count advance (winning race to notify vs signaller)
1423	<	*  - on Interrupt
1424	<	*  - if the first queued node, we find work available
1425	<	* If node was not signalled and event count not advanced on exit,
1426	<	* then we also help advance event count.
1427	<	* @return true if node can be released
1428	<	*/
1429	<	final boolean syncIsReleasable(WaitQueueNode node) {
1430	<	long prev = node.count;
1431	<	if (!Thread.interrupted() && node.thread != null &&
1432	<	(node.next != null ||
1433	<	!ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1434	<	eventCount == prev)
1435	<	return false;
1436	<	if (node.thread != null) {
1437	<	node.thread = null;
1438	<	long ec = eventCount;
1439	<	if (prev <= ec) // help signal
1440	<	casEventCount(ec, ec+1);
1441	<	}
1442	<	return true;
1443	<	}
1444	<
1445	<	/**
1446	<	* Returns true if a new sync event occurred since last call to
1447	<	* sync or this method, if so, updating caller's count.
1448	<	*/
1449	<	final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1450	<	long lc = w.lastEventCount;
1451	<	long ec = ensureSync();
1452	<	if (ec == lc)
1453	<	return false;
1454	<	w.lastEventCount = ec;
1455	<	return true;
1456	<	}
1457	<
1458	<	//  Parallelism maintenance
1459	<
1460	<	/**
1461	<	* Decrements running count; if too low, adds spare.
2708	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
2709	>	* {@code isReleasable} must return {@code true} if blocking is
2710	>	* not necessary. Method {@code block} blocks the current thread
2711	>	* if necessary (perhaps internally invoking {@code isReleasable}
2712	>	* before actually blocking). These actions are performed by any
2713	>	* thread invoking {@link ForkJoinPool#managedBlock}.  The
2714	>	* unusual methods in this API accommodate synchronizers that may,
2715	>	* but don't usually, block for long periods. Similarly, they
2716	>	* allow more efficient internal handling of cases in which
2717	>	* additional workers may be, but usually are not, needed to
2718	>	* ensure sufficient parallelism.  Toward this end,
2719	>	* implementations of method {@code isReleasable} must be amenable
2720	>	* to repeated invocation.
2721		*
1463	–	* Conceptually, all we need to do here is add or resume a
1464	–	* spare thread when one is about to block (and remove or
1465	–	* suspend it later when unblocked -- see suspendIfSpare).
1466	–	* However, implementing this idea requires coping with
1467	–	* several problems: We have imperfect information about the
1468	–	* states of threads. Some count updates can and usually do
1469	–	* lag run state changes, despite arrangements to keep them
1470	–	* accurate (for example, when possible, updating counts
1471	–	* before signalling or resuming), especially when running on
1472	–	* dynamic JVMs that don't optimize the infrequent paths that
1473	–	* update counts. Generating too many threads can make these
1474	–	* problems become worse, because excess threads are more
1475	–	* likely to be context-switched with others, slowing them all
1476	–	* down, especially if there is no work available, so all are
1477	–	* busy scanning or idling.  Also, excess spare threads can
1478	–	* only be suspended or removed when they are idle, not
1479	–	* immediately when they aren't needed. So adding threads will
1480	–	* raise parallelism level for longer than necessary.  Also,
1481	–	* FJ applications often encounter highly transient peaks when
1482	–	* many threads are blocked joining, but for less time than it
1483	–	* takes to create or resume spares.
1484	–	*
1485	–	* @param joinMe if non-null, return early if done
1486	–	* @param maintainParallelism if true, try to stay within
1487	–	* target counts, else create only to avoid starvation
1488	–	* @return true if joinMe known to be done
1489	–	*/
1490	–	final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
1491	–	maintainParallelism &= maintainsParallelism; // overrride
1492	–	boolean dec = false;  // true when running count decremented
1493	–	while (spareStack == null || !tryResumeSpare(dec)) {
1494	–	int counts = workerCounts;
1495	–	if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
1496	–	if (!needSpare(counts, maintainParallelism))
1497	–	break;
1498	–	if (joinMe.status < 0)
1499	–	return true;
1500	–	if (tryAddSpare(counts))
1501	–	break;
1502	–	}
1503	–	}
1504	–	return false;
1505	–	}
1506	–
1507	–	/**
1508	–	* Same idea as preJoin
1509	–	*/
1510	–	final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
1511	–	maintainParallelism &= maintainsParallelism;
1512	–	boolean dec = false;
1513	–	while (spareStack == null || !tryResumeSpare(dec)) {
1514	–	int counts = workerCounts;
1515	–	if (dec || (dec = casWorkerCounts(counts, --counts))) {
1516	–	if (!needSpare(counts, maintainParallelism))
1517	–	break;
1518	–	if (blocker.isReleasable())
1519	–	return true;
1520	–	if (tryAddSpare(counts))
1521	–	break;
1522	–	}
1523	–	}
1524	–	return false;
1525	–	}
1526	–
1527	–	/**
1528	–	* Returns true if a spare thread appears to be needed.  If
1529	–	* maintaining parallelism, returns true when the deficit in
1530	–	* running threads is more than the surplus of total threads, and
1531	–	* there is apparently some work to do.  This self-limiting rule
1532	–	* means that the more threads that have already been added, the
1533	–	* less parallelism we will tolerate before adding another.
1534	–	* @param counts current worker counts
1535	–	* @param maintainParallelism try to maintain parallelism
1536	–	*/
1537	–	private boolean needSpare(int counts, boolean maintainParallelism) {
1538	–	int ps = parallelism;
1539	–	int rc = runningCountOf(counts);
1540	–	int tc = totalCountOf(counts);
1541	–	int runningDeficit = ps - rc;
1542	–	int totalSurplus = tc - ps;
1543	–	return (tc < maxPoolSize &&
1544	–	(rc == 0 || totalSurplus < 0 ||
1545	–	(maintainParallelism &&
1546	–	runningDeficit > totalSurplus &&
1547	–	ForkJoinWorkerThread.hasQueuedTasks(workers))));
1548	–	}
1549	–
1550	–	/**
1551	–	* Adds a spare worker if lock available and no more than the
1552	–	* expected numbers of threads exist.
1553	–	* @return true if successful
1554	–	*/
1555	–	private boolean tryAddSpare(int expectedCounts) {
1556	–	final ReentrantLock lock = this.workerLock;
1557	–	int expectedRunning = runningCountOf(expectedCounts);
1558	–	int expectedTotal = totalCountOf(expectedCounts);
1559	–	boolean success = false;
1560	–	boolean locked = false;
1561	–	// confirm counts while locking; CAS after obtaining lock
1562	–	try {
1563	–	for (;;) {
1564	–	int s = workerCounts;
1565	–	int tc = totalCountOf(s);
1566	–	int rc = runningCountOf(s);
1567	–	if (rc > expectedRunning || tc > expectedTotal)
1568	–	break;
1569	–	if (!locked && !(locked = lock.tryLock()))
1570	–	break;
1571	–	if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1572	–	createAndStartSpare(tc);
1573	–	success = true;
1574	–	break;
1575	–	}
1576	–	}
1577	–	} finally {
1578	–	if (locked)
1579	–	lock.unlock();
1580	–	}
1581	–	return success;
1582	–	}
1583	–
1584	–	/**
1585	–	* Adds the kth spare worker. On entry, pool counts are already
1586	–	* adjusted to reflect addition.
1587	–	*/
1588	–	private void createAndStartSpare(int k) {
1589	–	ForkJoinWorkerThread w = null;
1590	–	ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1591	–	int len = ws.length;
1592	–	// Probably, we can place at slot k. If not, find empty slot
1593	–	if (k < len && ws[k] != null) {
1594	–	for (k = 0; k < len && ws[k] != null; ++k)
1595	–	;
1596	–	}
1597	–	if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1598	–	ws[k] = w;
1599	–	w.start();
1600	–	}
1601	–	else
1602	–	updateWorkerCount(-1); // adjust on failure
1603	–	signalIdleWorkers();
1604	–	}
1605	–
1606	–	/**
1607	–	* Suspends calling thread w if there are excess threads.  Called
1608	–	* only from sync.  Spares are enqueued in a Treiber stack using
1609	–	* the same WaitQueueNodes as barriers.  They are resumed mainly
1610	–	* in preJoin, but are also woken on pool events that require all
1611	–	* threads to check run state.
1612	–	* @param w the caller
1613	–	*/
1614	–	private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1615	–	WaitQueueNode node = null;
1616	–	int s;
1617	–	while (parallelism < runningCountOf(s = workerCounts)) {
1618	–	if (node == null)
1619	–	node = new WaitQueueNode(0, w);
1620	–	if (casWorkerCounts(s, s-1)) { // representation-dependent
1621	–	// push onto stack
1622	–	do;while (!casSpareStack(node.next = spareStack, node));
1623	–	// block until released by resumeSpare
1624	–	node.awaitSpareRelease();
1625	–	return true;
1626	–	}
1627	–	}
1628	–	return false;
1629	–	}
1630	–
1631	–	/**
1632	–	* Tries to pop and resume a spare thread.
1633	–	* @param updateCount if true, increment running count on success
1634	–	* @return true if successful
1635	–	*/
1636	–	private boolean tryResumeSpare(boolean updateCount) {
1637	–	WaitQueueNode q;
1638	–	while ((q = spareStack) != null) {
1639	–	if (casSpareStack(q, q.next)) {
1640	–	if (updateCount)
1641	–	updateRunningCount(1);
1642	–	q.signal();
1643	–	return true;
1644	–	}
1645	–	}
1646	–	return false;
1647	–	}
1648	–
1649	–	/**
1650	–	* Pops and resumes all spare threads. Same idea as ensureSync.
1651	–	* @return true if any spares released
1652	–	*/
1653	–	private boolean resumeAllSpares() {
1654	–	WaitQueueNode q;
1655	–	while ( (q = spareStack) != null) {
1656	–	if (casSpareStack(q, null)) {
1657	–	do {
1658	–	updateRunningCount(1);
1659	–	q.signal();
1660	–	} while ((q = q.next) != null);
1661	–	return true;
1662	–	}
1663	–	}
1664	–	return false;
1665	–	}
1666	–
1667	–	/**
1668	–	* Pops and shuts down excessive spare threads. Call only while
1669	–	* holding lock. This is not guaranteed to eliminate all excess
1670	–	* threads, only those suspended as spares, which are the ones
1671	–	* unlikely to be needed in the future.
1672	–	*/
1673	–	private void trimSpares() {
1674	–	int surplus = totalCountOf(workerCounts) - parallelism;
1675	–	WaitQueueNode q;
1676	–	while (surplus > 0 && (q = spareStack) != null) {
1677	–	if (casSpareStack(q, null)) {
1678	–	do {
1679	–	updateRunningCount(1);
1680	–	ForkJoinWorkerThread w = q.thread;
1681	–	if (w != null && surplus > 0 &&
1682	–	runningCountOf(workerCounts) > 0 && w.shutdown())
1683	–	--surplus;
1684	–	q.signal();
1685	–	} while ((q = q.next) != null);
1686	–	}
1687	–	}
1688	–	}
1689	–
1690	–	/**
1691	–	* Interface for extending managed parallelism for tasks running
1692	–	* in ForkJoinPools. A ManagedBlocker provides two methods.
1693	–	* Method {@code isReleasable} must return true if blocking is not
1694	–	* necessary. Method {@code block} blocks the current thread
1695	–	* if necessary (perhaps internally invoking isReleasable before
1696	–	* actually blocking.).
2722		* <p>For example, here is a ManagedBlocker based on a
2723		* ReentrantLock:
2724	<	* <pre>
2725	<	*   class ManagedLocker implements ManagedBlocker {
2726	<	*     final ReentrantLock lock;
2727	<	*     boolean hasLock = false;
2728	<	*     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
2729	<	*     public boolean block() {
2730	<	*        if (!hasLock)
2731	<	*           lock.lock();
2732	<	*        return true;
2733	<	*     }
2734	<	*     public boolean isReleasable() {
2735	<	*        return hasLock || (hasLock = lock.tryLock());
1711	<	*     }
2724	>	*  <pre> {@code
2725	>	* class ManagedLocker implements ManagedBlocker {
2726	>	*   final ReentrantLock lock;
2727	>	*   boolean hasLock = false;
2728	>	*   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
2729	>	*   public boolean block() {
2730	>	*     if (!hasLock)
2731	>	*       lock.lock();
2732	>	*     return true;
2733	>	*   }
2734	>	*   public boolean isReleasable() {
2735	>	*     return hasLock || (hasLock = lock.tryLock());
2736		*   }
2737	<	* </pre>
2737	>	* }}</pre>
2738	>	*
2739	>	* <p>Here is a class that possibly blocks waiting for an
2740	>	* item on a given queue:
2741	>	*  <pre> {@code
2742	>	* class QueueTaker<E> implements ManagedBlocker {
2743	>	*   final BlockingQueue<E> queue;
2744	>	*   volatile E item = null;
2745	>	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
2746	>	*   public boolean block() throws InterruptedException {
2747	>	*     if (item == null)
2748	>	*       item = queue.take();
2749	>	*     return true;
2750	>	*   }
2751	>	*   public boolean isReleasable() {
2752	>	*     return item != null || (item = queue.poll()) != null;
2753	>	*   }
2754	>	*   public E getItem() { // call after pool.managedBlock completes
2755	>	*     return item;
2756	>	*   }
2757	>	* }}</pre>
2758		*/
2759		public static interface ManagedBlocker {
2760		/**
2761		* Possibly blocks the current thread, for example waiting for
2762		* a lock or condition.
2763	<	* @return true if no additional blocking is necessary (i.e.,
2764	<	* if isReleasable would return true).
2763	>	*
2764	>	* @return {@code true} if no additional blocking is necessary
2765	>	* (i.e., if isReleasable would return true)
2766		* @throws InterruptedException if interrupted while waiting
2767	<	* (the method is not required to do so, but is allowed to).
2767	>	* (the method is not required to do so, but is allowed to)
2768		*/
2769		boolean block() throws InterruptedException;
2770
2771		/**
2772	<	* Returns true if blocking is unnecessary.
2772	>	* Returns {@code true} if blocking is unnecessary.
2773		*/
2774		boolean isReleasable();
2775		}
2776
2777		/**
2778		* Blocks in accord with the given blocker.  If the current thread
2779	<	* is a ForkJoinWorkerThread, this method possibly arranges for a
2780	<	* spare thread to be activated if necessary to ensure parallelism
2781	<	* while the current thread is blocked.  If
2782	<	* {@code maintainParallelism} is true and the pool supports
2783	<	* it ({@link #getMaintainsParallelism}), this method attempts to
2784	<	* maintain the pool's nominal parallelism. Otherwise if activates
2785	<	* a thread only if necessary to avoid complete starvation. This
2786	<	* option may be preferable when blockages use timeouts, or are
2787	<	* almost always brief.
2788	<	*
2789	<	* <p> If the caller is not a ForkJoinTask, this method is behaviorally
2790	<	* equivalent to
2791	<	* <pre>
2792	<	*   while (!blocker.isReleasable())
1748	<	*      if (blocker.block())
1749	<	*         return;
1750	<	* </pre>
1751	<	* If the caller is a ForkJoinTask, then the pool may first
1752	<	* be expanded to ensure parallelism, and later adjusted.
2779	>	* is a {@link ForkJoinWorkerThread}, this method possibly
2780	>	* arranges for a spare thread to be activated if necessary to
2781	>	* ensure sufficient parallelism while the current thread is blocked.
2782	>	*
2783	>	* <p>If the caller is not a {@link ForkJoinTask}, this method is
2784	>	* behaviorally equivalent to
2785	>	*  <pre> {@code
2786	>	* while (!blocker.isReleasable())
2787	>	*   if (blocker.block())
2788	>	*     return;
2789	>	* }</pre>
2790	>	*
2791	>	* If the caller is a {@code ForkJoinTask}, then the pool may
2792	>	* first be expanded to ensure parallelism, and later adjusted.
2793		*
2794		* @param blocker the blocker
2795	<	* @param maintainParallelism if true and supported by this pool,
1756	<	* attempt to maintain the pool's nominal parallelism; otherwise
1757	<	* activate a thread only if necessary to avoid complete
1758	<	* starvation.
1759	<	* @throws InterruptedException if blocker.block did so.
2795	>	* @throws InterruptedException if blocker.block did so
2796		*/
2797	<	public static void managedBlock(ManagedBlocker blocker,
1762	<	boolean maintainParallelism)
2797	>	public static void managedBlock(ManagedBlocker blocker)
2798		throws InterruptedException {
2799		Thread t = Thread.currentThread();
2800	<	ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
2801	<	((ForkJoinWorkerThread)t).pool : null);
2802	<	if (!blocker.isReleasable()) {
2803	<	try {
2804	<	if (pool == null ||
2805	<	!pool.preBlock(blocker, maintainParallelism))
2806	<	awaitBlocker(blocker);
2807	<	} finally {
2808	<	if (pool != null)
2809	<	pool.updateRunningCount(1);
2800	>	ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
2801	>	((ForkJoinWorkerThread)t).pool : null);
2802	>	while (!blocker.isReleasable()) {
2803	>	if (p == null || p.tryCompensate(null, blocker)) {
2804	>	try {
2805	>	do {} while (!blocker.isReleasable() && !blocker.block());
2806	>	} finally {
2807	>	if (p != null)
2808	>	p.incrementActiveCount();
2809	>	}
2810	>	break;
2811		}
2812		}
2813		}
2814
2815	<	private static void awaitBlocker(ManagedBlocker blocker)
2816	<	throws InterruptedException {
2817	<	do;while (!blocker.isReleasable() && !blocker.block());
1782	<	}
1783	<
1784	<	// AbstractExecutorService overrides
2815	>	// AbstractExecutorService overrides.  These rely on undocumented
2816	>	// fact that ForkJoinTask.adapt returns ForkJoinTasks that also
2817	>	// implement RunnableFuture.
2818
2819		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
2820	<	return new AdaptedRunnable(runnable, value);
2820	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
2821		}
2822
2823		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
2824	<	return new AdaptedCallable(callable);
2824	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
2825		}
2826
2827	+	// Unsafe mechanics
2828	+	private static final sun.misc.Unsafe U;
2829	+	private static final long CTL;
2830	+	private static final long PARKBLOCKER;
2831	+	private static final int ABASE;
2832	+	private static final int ASHIFT;
2833
2834	<	// Temporary Unsafe mechanics for preliminary release
2835	<	private static Unsafe getUnsafe() throws Throwable {
2834	>	static {
2835	>	poolNumberGenerator = new AtomicInteger();
2836	>	nextSubmitterSeed = new AtomicInteger(0x55555555);
2837	>	modifyThreadPermission = new RuntimePermission("modifyThread");
2838	>	defaultForkJoinWorkerThreadFactory =
2839	>	new DefaultForkJoinWorkerThreadFactory();
2840	>	submitters = new ThreadSubmitter();
2841	>	int s;
2842		try {
2843	<	return Unsafe.getUnsafe();
2843	>	U = getUnsafe();
2844	>	Class<?> k = ForkJoinPool.class;
2845	>	Class<?> ak = ForkJoinTask[].class;
2846	>	CTL = U.objectFieldOffset
2847	>	(k.getDeclaredField("ctl"));
2848	>	Class<?> tk = Thread.class;
2849	>	PARKBLOCKER = U.objectFieldOffset
2850	>	(tk.getDeclaredField("parkBlocker"));
2851	>	ABASE = U.arrayBaseOffset(ak);
2852	>	s = U.arrayIndexScale(ak);
2853	>	} catch (Exception e) {
2854	>	throw new Error(e);
2855	>	}
2856	>	if ((s & (s-1)) != 0)
2857	>	throw new Error("data type scale not a power of two");
2858	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
2859	>	}
2860	>
2861	>	/**
2862	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
2863	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
2864	>	* into a jdk.
2865	>	*
2866	>	* @return a sun.misc.Unsafe
2867	>	*/
2868	>	private static sun.misc.Unsafe getUnsafe() {
2869	>	try {
2870	>	return sun.misc.Unsafe.getUnsafe();
2871		} catch (SecurityException se) {
2872		try {
2873		return java.security.AccessController.doPrivileged
2874	<	(new java.security.PrivilegedExceptionAction<Unsafe>() {
2875	<	public Unsafe run() throws Exception {
2876	<	return getUnsafePrivileged();
2874	>	(new java.security
2875	>	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
2876	>	public sun.misc.Unsafe run() throws Exception {
2877	>	java.lang.reflect.Field f = sun.misc
2878	>	.Unsafe.class.getDeclaredField("theUnsafe");
2879	>	f.setAccessible(true);
2880	>	return (sun.misc.Unsafe) f.get(null);
2881		}});
2882		} catch (java.security.PrivilegedActionException e) {
2883	<	throw e.getCause();
2883	>	throw new RuntimeException("Could not initialize intrinsics",
2884	>	e.getCause());
2885		}
2886		}
2887		}
2888
1812	–	private static Unsafe getUnsafePrivileged()
1813	–	throws NoSuchFieldException, IllegalAccessException {
1814	–	Field f = Unsafe.class.getDeclaredField("theUnsafe");
1815	–	f.setAccessible(true);
1816	–	return (Unsafe) f.get(null);
1817	–	}
1818	–
1819	–	private static long fieldOffset(String fieldName)
1820	–	throws NoSuchFieldException {
1821	–	return _unsafe.objectFieldOffset
1822	–	(ForkJoinPool.class.getDeclaredField(fieldName));
1823	–	}
1824	–
1825	–	static final Unsafe _unsafe;
1826	–	static final long eventCountOffset;
1827	–	static final long workerCountsOffset;
1828	–	static final long runControlOffset;
1829	–	static final long syncStackOffset;
1830	–	static final long spareStackOffset;
1831	–
1832	–	static {
1833	–	try {
1834	–	_unsafe = getUnsafe();
1835	–	eventCountOffset = fieldOffset("eventCount");
1836	–	workerCountsOffset = fieldOffset("workerCounts");
1837	–	runControlOffset = fieldOffset("runControl");
1838	–	syncStackOffset = fieldOffset("syncStack");
1839	–	spareStackOffset = fieldOffset("spareStack");
1840	–	} catch (Throwable e) {
1841	–	throw new RuntimeException("Could not initialize intrinsics", e);
1842	–	}
1843	–	}
1844	–
1845	–	private boolean casEventCount(long cmp, long val) {
1846	–	return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
1847	–	}
1848	–	private boolean casWorkerCounts(int cmp, int val) {
1849	–	return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1850	–	}
1851	–	private boolean casRunControl(int cmp, int val) {
1852	–	return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
1853	–	}
1854	–	private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1855	–	return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
1856	–	}
1857	–	private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1858	–	return _unsafe.compareAndSwapObject(this, syncStackOffset, cmp, val);
1859	–	}
2889		}

Diff Legend

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