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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.56 by dl, Thu May 27 16:46:48 2010 UTC vs.
Revision 1.78 by dl, Tue Sep 7 14:52:24 2010 UTC

#	Line 6 \| Line 6
6
7		package jsr166y;
8
9	–	import java.util.concurrent.*;
10	–
9		import java.util.ArrayList;
10		import java.util.Arrays;
11		import java.util.Collection;
12		import java.util.Collections;
13		import java.util.List;
14	+	import java.util.concurrent.AbstractExecutorService;
15	+	import java.util.concurrent.Callable;
16	+	import java.util.concurrent.CountDownLatch;
17	+	import java.util.concurrent.ExecutorService;
18	+	import java.util.concurrent.Future;
19	+	import java.util.concurrent.RejectedExecutionException;
20	+	import java.util.concurrent.RunnableFuture;
21	+	import java.util.concurrent.TimeUnit;
22	+	import java.util.concurrent.TimeoutException;
23	+	import java.util.concurrent.atomic.AtomicInteger;
24		import java.util.concurrent.locks.LockSupport;
25		import java.util.concurrent.locks.ReentrantLock;
18	–	import java.util.concurrent.atomic.AtomicInteger;
19	–	import java.util.concurrent.CountDownLatch;
26
27		/**
28		* An {@link ExecutorService} for running {@link ForkJoinTask}s.
29		* A {@code ForkJoinPool} provides the entry point for submissions
30	<	* from non-{@code ForkJoinTask}s, as well as management and
30	>	* from non-{@code ForkJoinTask} clients, as well as management and
31		* monitoring operations.
32		*
33		* <p>A {@code ForkJoinPool} differs from other kinds of {@link
#	Line 30 \| Line 36 \| import java.util.concurrent.CountDownLat
36		* execute subtasks created by other active tasks (eventually blocking
37		* waiting for work if none exist). This enables efficient processing
38		* when most tasks spawn other subtasks (as do most {@code
39	<	* ForkJoinTask}s). A {@code ForkJoinPool} may also be used for mixed
40	<	* execution of some plain {@code Runnable}- or {@code Callable}-
41	<	* based activities along with {@code ForkJoinTask}s. When setting
36	<	* {@linkplain #setAsyncMode async mode}, a {@code ForkJoinPool} may
37	<	* also be appropriate for use with fine-grained tasks of any form
38	<	* that are never joined. Otherwise, other {@code ExecutorService}
39	<	* implementations are typically more appropriate choices.
39	>	* ForkJoinTask}s). When setting <em>asyncMode</em> to true in
40	>	* constructors, {@code ForkJoinPool}s may also be appropriate for use
41	>	* with event-style tasks that are never joined.
42		*
43		* <p>A {@code ForkJoinPool} is constructed with a given target
44		* parallelism level; by default, equal to the number of available
45	<	* processors. Unless configured otherwise via {@link
46	<	* #setMaintainsParallelism}, the pool attempts to maintain this
47	<	* number of active (or available) threads by dynamically adding,
48	<	* suspending, or resuming internal worker threads, even if some tasks
49	<	* are stalled waiting to join others. However, no such adjustments
50	<	* are performed in the face of blocked IO or other unmanaged
51	<	* synchronization. The nested {@link ManagedBlocker} interface
50	<	* enables extension of the kinds of synchronization accommodated.
51	<	* The target parallelism level may also be changed dynamically
52	<	* ({@link #setParallelism}). The total number of threads may be
53	<	* limited using method {@link #setMaximumPoolSize}, in which case it
54	<	* may become possible for the activities of a pool to stall due to
55	<	* the lack of available threads to process new tasks. When the pool
56	<	* is executing tasks, these and other configuration setting methods
57	<	* may only gradually affect actual pool sizes. It is normally best
58	<	* practice to invoke these methods only when the pool is known to be
59	<	* quiescent.
45	>	* processors. The pool attempts to maintain enough active (or
46	>	* available) threads by dynamically adding, suspending, or resuming
47	>	* internal worker threads, even if some tasks are stalled waiting to
48	>	* join others. However, no such adjustments are guaranteed in the
49	>	* face of blocked IO or other unmanaged synchronization. The nested
50	>	* {@link ManagedBlocker} interface enables extension of the kinds of
51	>	* synchronization accommodated.
52		*
53		* <p>In addition to execution and lifecycle control methods, this
54		* class provides status check methods (for example
#	Line 65 \| Line 57 \| import java.util.concurrent.CountDownLat
57		* {@link #toString} returns indications of pool state in a
58		* convenient form for informal monitoring.
59		*
60	+	* <p> As is the case with other ExecutorServices, there are three
61	+	* main task execution methods summarized in the following
62	+	* table. These are designed to be used by clients not already engaged
63	+	* in fork/join computations in the current pool. The main forms of
64	+	* these methods accept instances of {@code ForkJoinTask}, but
65	+	* overloaded forms also allow mixed execution of plain {@code
66	+	* Runnable}- or {@code Callable}- based activities as well. However,
67	+	* tasks that are already executing in a pool should normally
68	+	* <em>NOT</em> use these pool execution methods, but instead use the
69	+	* within-computation forms listed in the table.
70	+	*
71	+	* <table BORDER CELLPADDING=3 CELLSPACING=1>
72	+	* <tr>
73	+	* <td></td>
74	+	* <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
75	+	* <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
76	+	* </tr>
77	+	* <tr>
78	+	* <td> <b>Arrange async execution</td>
79	+	* <td> {@link #execute(ForkJoinTask)}</td>
80	+	* <td> {@link ForkJoinTask#fork}</td>
81	+	* </tr>
82	+	* <tr>
83	+	* <td> <b>Await and obtain result</td>
84	+	* <td> {@link #invoke(ForkJoinTask)}</td>
85	+	* <td> {@link ForkJoinTask#invoke}</td>
86	+	* </tr>
87	+	* <tr>
88	+	* <td> <b>Arrange exec and obtain Future</td>
89	+	* <td> {@link #submit(ForkJoinTask)}</td>
90	+	* <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
91	+	* </tr>
92	+	* </table>
93	+	*
94		* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
95		* used for all parallel task execution in a program or subsystem.
96		* Otherwise, use would not usually outweigh the construction and
#	Line 89 \| Line 115 \| import java.util.concurrent.CountDownLat
115		* {@code IllegalArgumentException}.
116		*
117		* <p>This implementation rejects submitted tasks (that is, by throwing
118	<	* {@link RejectedExecutionException}) only when the pool is shut down.
118	>	* {@link RejectedExecutionException}) only when the pool is shut down
119	>	* or internal resources have been exhausted.
120		*
121		* @since 1.7
122		* @author Doug Lea
#	Line 116 \| Line 143 \| public class ForkJoinPool extends Abstra
143		* of tasks profit from cache affinities, but others are harmed by
144		* cache pollution effects.)
145		*
146	+	* Beyond work-stealing support and essential bookkeeping, the
147	+	* main responsibility of this framework is to take actions when
148	+	* one worker is waiting to join a task stolen (or always held by)
149	+	* another. Because we are multiplexing many tasks on to a pool
150	+	* of workers, we can't just let them block (as in Thread.join).
151	+	* We also cannot just reassign the joiner's run-time stack with
152	+	* another and replace it later, which would be a form of
153	+	* "continuation", that even if possible is not necessarily a good
154	+	* idea. Given that the creation costs of most threads on most
155	+	* systems mainly surrounds setting up runtime stacks, thread
156	+	* creation and switching is usually not much more expensive than
157	+	* stack creation and switching, and is more flexible). Instead we
158	+	* combine two tactics:
159	+	*
160	+	* Helping: Arranging for the joiner to execute some task that it
161	+	* would be running if the steal had not occurred. Method
162	+	* ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
163	+	* links to try to find such a task.
164	+	*
165	+	* Compensating: Unless there are already enough live threads,
166	+	* method helpMaintainParallelism() may create or
167	+	* re-activate a spare thread to compensate for blocked
168	+	* joiners until they unblock.
169	+	*
170	+	* It is impossible to keep exactly the target (parallelism)
171	+	* number of threads running at any given time. Determining
172	+	* existence of conservatively safe helping targets, the
173	+	* availability of already-created spares, and the apparent need
174	+	* to create new spares are all racy and require heuristic
175	+	* guidance, so we rely on multiple retries of each. Compensation
176	+	* occurs in slow-motion. It is triggered only upon timeouts of
177	+	* Object.wait used for joins. This reduces poor decisions that
178	+	* would otherwise be made when threads are waiting for others
179	+	* that are stalled because of unrelated activities such as
180	+	* garbage collection.
181	+	*
182	+	* The ManagedBlocker extension API can't use helping so relies
183	+	* only on compensation in method awaitBlocker.
184	+	*
185		* The main throughput advantages of work-stealing stem from
186		* decentralized control -- workers mostly steal tasks from each
187		* other. We do not want to negate this by creating bottlenecks
188	<	* implementing the management responsibilities of this class. So
189	<	* we use a collection of techniques that avoid, reduce, or cope
190	<	* well with contention. These entail several instances of
191	<	* bit-packing into CASable fields to maintain only the minimally
192	<	* required atomicity. To enable such packing, we restrict maximum
193	<	* parallelism to (1<<15)-1 (enabling twice this to fit into a 16
194	<	* bit field), which is far in excess of normal operating range.
195	<	* Even though updates to some of these bookkeeping fields do
196	<	* sometimes contend with each other, they don't normally
197	<	* cache-contend with updates to others enough to warrant memory
198	<	* padding or isolation. So they are all held as fields of
199	<	* ForkJoinPool objects. The main capabilities are as follows:
188	>	* implementing other management responsibilities. So we use a
189	>	* collection of techniques that avoid, reduce, or cope well with
190	>	* contention. These entail several instances of bit-packing into
191	>	* CASable fields to maintain only the minimally required
192	>	* atomicity. To enable such packing, we restrict maximum
193	>	* parallelism to (1<<15)-1 (enabling twice this (to accommodate
194	>	* unbalanced increments and decrements) to fit into a 16 bit
195	>	* field, which is far in excess of normal operating range. Even
196	>	* though updates to some of these bookkeeping fields do sometimes
197	>	* contend with each other, they don't normally cache-contend with
198	>	* updates to others enough to warrant memory padding or
199	>	* isolation. So they are all held as fields of ForkJoinPool
200	>	* objects. The main capabilities are as follows:
201		*
202		* 1. Creating and removing workers. Workers are recorded in the
203		* "workers" array. This is an array as opposed to some other data
#	Line 146 \| Line 213 \| public class ForkJoinPool extends Abstra
213		* blocked workers. However, all other support code is set up to
214		* work with other policies.
215		*
216	+	* To ensure that we do not hold on to worker references that
217	+	* would prevent GC, ALL accesses to workers are via indices into
218	+	* the workers array (which is one source of some of the unusual
219	+	* code constructions here). In essence, the workers array serves
220	+	* as a WeakReference mechanism. Thus for example the event queue
221	+	* stores worker indices, not worker references. Access to the
222	+	* workers in associated methods (for example releaseEventWaiters)
223	+	* must both index-check and null-check the IDs. All such accesses
224	+	* ignore bad IDs by returning out early from what they are doing,
225	+	* since this can only be associated with shutdown, in which case
226	+	* it is OK to give up. On termination, we just clobber these
227	+	* data structures without trying to use them.
228	+	*
229		* 2. Bookkeeping for dynamically adding and removing workers. We
230	<	* maintain a given level of parallelism (or, if
231	<	* maintainsParallelism is false, at least avoid starvation). When
152	<	* some workers are known to be blocked (on joins or via
230	>	* aim to approximately maintain the given level of parallelism.
231	>	* When some workers are known to be blocked (on joins or via
232		* ManagedBlocker), we may create or resume others to take their
233		* place until they unblock (see below). Implementing this
234		* requires counts of the number of "running" threads (i.e., those
235	<	* that are neither blocked nor artifically suspended) as well as
235	>	* that are neither blocked nor artificially suspended) as well as
236		* the total number. These two values are packed into one field,
237		* "workerCounts" because we need accurate snapshots when deciding
238	<	* to create, resume or suspend. To support these decisions,
239	<	* updates to spare counts must be prospective (not
240	<	* retrospective). For example, the running count is decremented
241	<	* before blocking by a thread about to block as a spare, but
163	<	* incremented by the thread about to unblock it. Updates upon
164	<	* resumption ofr threads blocking in awaitJoin or awaitBlocker
165	<	* cannot usually be prospective, so the running count is in
166	<	* general an upper bound of the number of productively running
167	<	* threads Updates to the workerCounts field sometimes transiently
168	<	* encounter a fair amount of contention when join dependencies
169	<	* are such that many threads block or unblock at about the same
170	<	* time. We alleviate this by sometimes bundling updates (for
171	<	* example blocking one thread on join and resuming a spare cancel
172	<	* each other out), and in most other cases performing an
173	<	* alternative action like releasing waiters or locating spares.
238	>	* to create, resume or suspend. Note however that the
239	>	* correspondence of these counts to reality is not guaranteed. In
240	>	* particular updates for unblocked threads may lag until they
241	>	* actually wake up.
242		*
243		* 3. Maintaining global run state. The run state of the pool
244		* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
#	Line 199 \| Line 267 \| public class ForkJoinPool extends Abstra
267		* workers that previously could not find a task to now find one:
268		* Submission of a new task to the pool, or another worker pushing
269		* a task onto a previously empty queue. (We also use this
270	<	* mechanism for termination and reconfiguration actions that
270	>	* mechanism for configuration and termination actions that
271		* require wakeups of idle workers). Each worker maintains its
272		* last known event count, and blocks when a scan for work did not
273		* find a task AND its lastEventCount matches the current
#	Line 210 \| Line 278 \| public class ForkJoinPool extends Abstra
278		* a record (field nextEventWaiter) for the next waiting worker.
279		* In addition to allowing simpler decisions about need for
280		* wakeup, the event count bits in eventWaiters serve the role of
281	<	* tags to avoid ABA errors in Treiber stacks. To reduce delays
282	<	* in task diffusion, workers not otherwise occupied may invoke
283	<	* method releaseWaiters, that removes and signals (unparks)
284	<	* workers not waiting on current count. To minimize task
285	<	* production stalls associate with signalling, any worker pushing
286	<	* a task on an empty queue invokes the weaker method signalWork,
219	<	* that only releases idle workers until it detects interference
220	<	* by other threads trying to release, and lets them take
221	<	* over. The net effect is a tree-like diffusion of signals, where
222	<	* released threads (and possibly others) help with unparks. To
223	<	* further reduce contention effects a bit, failed CASes to
224	<	* increment field eventCount are tolerated without retries.
281	>	* tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
282	>	* released threads also try to release at most two others. The
283	>	* net effect is a tree-like diffusion of signals, where released
284	>	* threads (and possibly others) help with unparks. To further
285	>	* reduce contention effects a bit, failed CASes to increment
286	>	* field eventCount are tolerated without retries in signalWork.
287		* Conceptually they are merged into the same event, which is OK
288		* when their only purpose is to enable workers to scan for work.
289		*
290	<	* 5. Managing suspension of extra workers. When a worker is about
291	<	* to block waiting for a join (or via ManagedBlockers), we may
292	<	* create a new thread to maintain parallelism level, or at least
293	<	* avoid starvation (see below). Usually, extra threads are needed
294	<	* for only very short periods, yet join dependencies are such
295	<	* that we sometimes need them in bursts. Rather than create new
296	<	* threads each time this happens, we suspend no-longer-needed
297	<	* extra ones as "spares". For most purposes, we don't distinguish
298	<	* "extra" spare threads from normal "core" threads: On each call
299	<	* to preStep (the only point at which we can do this) a worker
300	<	* checks to see if there are now too many running workers, and if
301	<	* so, suspends itself. Methods awaitJoin and awaitBlocker look
302	<	* for suspended threads to resume before considering creating a
303	<	* new replacement. We don't need a special data structure to
304	<	* maintain spares; simply scanning the workers array looking for
305	<	* worker.isSuspended() is fine because the calling thread is
306	<	* otherwise not doing anything useful anyway; we are at least as
307	<	* happy if after locating a spare, the caller doesn't actually
308	<	* block because the join is ready before we try to adjust and
309	<	* compensate. Note that this is intrinsically racy. One thread
310	<	* may become a spare at about the same time as another is
311	<	* needlessly being created. We counteract this and related slop
312	<	* in part by requiring resumed spares to immediately recheck (in
313	<	* preStep) to see whether they they should re-suspend. The only
314	<	* effective difference between "extra" and "core" threads is that
315	<	* we allow the "extra" ones to time out and die if they are not
316	<	* resumed within a keep-alive interval of a few seconds. This is
317	<	* implemented mainly within ForkJoinWorkerThread, but requires
318	<	* some coordination (isTrimmed() -- meaning killed while
319	<	* suspended) to correctly maintain pool counts.
320	<	*
321	<	* 6. Deciding when to create new workers. The main dynamic
322	<	* control in this class is deciding when to create extra threads,
323	<	* in methods awaitJoin and awaitBlocker. We always
324	<	* need to create one when the number of running threads becomes
325	<	* zero. But because blocked joins are typically dependent, we
326	<	* don't necessarily need or want one-to-one replacement. Using a
327	<	* one-to-one compensation rule often leads to enough useless
328	<	* overhead creating, suspending, resuming, and/or killing threads
329	<	* to signficantly degrade throughput. We use a rule reflecting
330	<	* the idea that, the more spare threads you already have, the
331	<	* more evidence you need to create another one. The "evidence"
332	<	* here takes two forms: (1) Using a creation threshold expressed
333	<	* in terms of the current deficit -- target minus running
334	<	* threads. To reduce flickering and drift around target values,
335	<	* the relation is quadratic: adding a spare if (dcdc)>=(scpc)
336	<	* (where dc is deficit, sc is number of spare threads and pc is
337	<	* target parallelism.) (2) Using a form of adaptive
338	<	* spionning. requiring a number of threshold checks proportional
339	<	* to the number of spare threads. This effectively reduces churn
340	<	* at the price of systematically undershooting target parallelism
279	<	* when many threads are blocked. However, biasing toward
280	<	* undeshooting partially compensates for the above mechanics to
281	<	* suspend extra threads, that normally lead to overshoot because
282	<	* we can only suspend workers in-between top-level actions. It
283	<	* also better copes with the fact that some of the methods in
284	<	* this class tend to never become compiled (but are interpreted),
285	<	* so some components of the entire set of controls might execute
286	<	* many times faster than others. And similarly for cases where
287	<	* the apparent lack of work is just due to GC stalls and other
288	<	* transient system activity.
289	<	*
290	<	* 7. Maintaining other configuration parameters and monitoring
291	<	* statistics. Updates to fields controlling parallelism level,
292	<	* max size, etc can only meaningfully take effect for individual
293	<	* threads upon their next top-level actions; i.e., between
294	<	* stealing/running tasks/submission, which are separated by calls
295	<	* to preStep. Memory ordering for these (assumed infrequent)
296	<	* reconfiguration calls is ensured by using reads and writes to
297	<	* volatile field workerCounts (that must be read in preStep anyway)
298	<	* as "fences" -- user-level reads are preceded by reads of
299	<	* workCounts, and writes are followed by no-op CAS to
300	<	* workerCounts. The values reported by other management and
301	<	* monitoring methods are either computed on demand, or are kept
302	<	* in fields that are only updated when threads are otherwise
303	<	* idle.
290	>	* 5. Managing suspension of extra workers. When a worker notices
291	>	* (usually upon timeout of a wait()) that there are too few
292	>	* running threads, we may create a new thread to maintain
293	>	* parallelism level, or at least avoid starvation. Usually, extra
294	>	* threads are needed for only very short periods, yet join
295	>	* dependencies are such that we sometimes need them in
296	>	* bursts. Rather than create new threads each time this happens,
297	>	* we suspend no-longer-needed extra ones as "spares". For most
298	>	* purposes, we don't distinguish "extra" spare threads from
299	>	* normal "core" threads: On each call to preStep (the only point
300	>	* at which we can do this) a worker checks to see if there are
301	>	* now too many running workers, and if so, suspends itself.
302	>	* Method helpMaintainParallelism looks for suspended threads to
303	>	* resume before considering creating a new replacement. The
304	>	* spares themselves are encoded on another variant of a Treiber
305	>	* Stack, headed at field "spareWaiters". Note that the use of
306	>	* spares is intrinsically racy. One thread may become a spare at
307	>	* about the same time as another is needlessly being created. We
308	>	* counteract this and related slop in part by requiring resumed
309	>	* spares to immediately recheck (in preStep) to see whether they
310	>	* should re-suspend.
311	>	*
312	>	* 6. Killing off unneeded workers. A timeout mechanism is used to
313	>	* shed unused workers: The oldest (first) event queue waiter uses
314	>	* a timed rather than hard wait. When this wait times out without
315	>	* a normal wakeup, it tries to shutdown any one (for convenience
316	>	* the newest) other spare or event waiter via
317	>	* tryShutdownUnusedWorker. This eventually reduces the number of
318	>	* worker threads to a minimum of one after a long enough period
319	>	* without use.
320	>	*
321	>	* 7. Deciding when to create new workers. The main dynamic
322	>	* control in this class is deciding when to create extra threads
323	>	* in method helpMaintainParallelism. We would like to keep
324	>	* exactly #parallelism threads running, which is an impossible
325	>	* task. We always need to create one when the number of running
326	>	* threads would become zero and all workers are busy. Beyond
327	>	* this, we must rely on heuristics that work well in the
328	>	* presence of transient phenomena such as GC stalls, dynamic
329	>	* compilation, and wake-up lags. These transients are extremely
330	>	* common -- we are normally trying to fully saturate the CPUs on
331	>	* a machine, so almost any activity other than running tasks
332	>	* impedes accuracy. Our main defense is to allow parallelism to
333	>	* lapse for a while during joins, and use a timeout to see if,
334	>	* after the resulting settling, there is still a need for
335	>	* additional workers. This also better copes with the fact that
336	>	* some of the methods in this class tend to never become compiled
337	>	* (but are interpreted), so some components of the entire set of
338	>	* controls might execute 100 times faster than others. And
339	>	* similarly for cases where the apparent lack of work is just due
340	>	* to GC stalls and other transient system activity.
341		*
342		* Beware that there is a lot of representation-level coupling
343		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 313 \| Line 350 \| public class ForkJoinPool extends Abstra
350		*
351		* Style notes: There are lots of inline assignments (of form
352		* "while ((local = field) != 0)") which are usually the simplest
353	<	* way to ensure read orderings. Also several occurrences of the
354	<	* unusual "do {} while(!cas...)" which is the simplest way to
355	<	* force an update of a CAS'ed variable. There are also a few
356	<	* other coding oddities that help some methods perform reasonably
357	<	* even when interpreted (not compiled).
353	>	* way to ensure the required read orderings (which are sometimes
354	>	* critical). Also several occurrences of the unusual "do {}
355	>	* while (!cas...)" which is the simplest way to force an update of
356	>	* a CAS'ed variable. There are also other coding oddities that
357	>	* help some methods perform reasonably even when interpreted (not
358	>	* compiled), at the expense of some messy constructions that
359	>	* reduce byte code counts.
360		*
361		* The order of declarations in this file is: (1) statics (2)
362		* fields (along with constants used when unpacking some of them)
#	Line 346 \| Line 385 \| public class ForkJoinPool extends Abstra
385		* Default ForkJoinWorkerThreadFactory implementation; creates a
386		* new ForkJoinWorkerThread.
387		*/
388	<	static class DefaultForkJoinWorkerThreadFactory
388	>	static class DefaultForkJoinWorkerThreadFactory
389		implements ForkJoinWorkerThreadFactory {
390		public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
391		return new ForkJoinWorkerThread(pool);
#	Line 385 \| Line 424 \| public class ForkJoinPool extends Abstra
424		new AtomicInteger();
425
426		/**
427	<	* Absolute bound for parallelism level. Twice this number must
428	<	* fit into a 16bit field to enable word-packing for some counts.
427	>	* The time to block in a join (see awaitJoin) before checking if
428	>	* a new worker should be (re)started to maintain parallelism
429	>	* level. The value should be short enough to maintain global
430	>	* responsiveness and progress but long enough to avoid
431	>	* counterproductive firings during GC stalls or unrelated system
432	>	* activity, and to not bog down systems with continual re-firings
433	>	* on GCs or legitimately long waits.
434		*/
435	<	private static final int MAX_THREADS = 0x7fff;
435	>	private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
436	>
437	>	/**
438	>	* The wakeup interval (in nanoseconds) for the oldest worker
439	>	* waiting for an event to invoke tryShutdownUnusedWorker to
440	>	* shrink the number of workers. The exact value does not matter
441	>	* too much. It must be short enough to release resources during
442	>	* sustained periods of idleness, but not so short that threads
443	>	* are continually re-created.
444	>	*/
445	>	private static final long SHRINK_RATE_NANOS =
446	>	30L * 1000L * 1000L * 1000L; // 2 per minute
447	>
448	>	/**
449	>	* Absolute bound for parallelism level. Twice this number plus
450	>	* one (i.e., 0xfff) must fit into a 16bit field to enable
451	>	* word-packing for some counts and indices.
452	>	*/
453	>	private static final int MAX_WORKERS = 0x7fff;
454
455		/**
456		* Array holding all worker threads in the pool. Array size must
#	Line 414 \| Line 476 \| public class ForkJoinPool extends Abstra
476		/**
477		* Latch released upon termination.
478		*/
479	<	private final CountDownLatch terminationLatch;
479	>	private final Phaser termination;
480
481		/**
482		* Creation factory for worker threads.
#	Line 428 \| Line 490 \| public class ForkJoinPool extends Abstra
490		private volatile long stealCount;
491
492		/**
493	<	* Encoded record of top of treiber stack of threads waiting for
493	>	* Encoded record of top of Treiber stack of threads waiting for
494		* events. The top 32 bits contain the count being waited for. The
495	<	* bottom word contains one plus the pool index of waiting worker
496	<	* thread.
495	>	* bottom 16 bits contains one plus the pool index of waiting
496	>	* worker thread. (Bits 16-31 are unused.)
497		*/
498		private volatile long eventWaiters;
499
500		private static final int EVENT_COUNT_SHIFT = 32;
501	<	private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
501	>	private static final long WAITER_ID_MASK = (1L << 16) - 1L;
502
503		/**
504		* A counter for events that may wake up worker threads:
505		* - Submission of a new task to the pool
506		* - A worker pushing a task on an empty queue
507	<	* - termination and reconfiguration
507	>	* - termination
508		*/
509		private volatile int eventCount;
510
511		/**
512	+	* Encoded record of top of Treiber stack of spare threads waiting
513	+	* for resumption. The top 16 bits contain an arbitrary count to
514	+	* avoid ABA effects. The bottom 16bits contains one plus the pool
515	+	* index of waiting worker thread.
516	+	*/
517	+	private volatile int spareWaiters;
518	+
519	+	private static final int SPARE_COUNT_SHIFT = 16;
520	+	private static final int SPARE_ID_MASK = (1 << 16) - 1;
521	+
522	+	/**
523		* Lifecycle control. The low word contains the number of workers
524		* that are (probably) executing tasks. This value is atomically
525		* incremented before a worker gets a task to run, and decremented
#	Line 457 \| Line 530 \| public class ForkJoinPool extends Abstra
530		* These are bundled together to ensure consistent read for
531		* termination checks (i.e., that runLevel is at least SHUTDOWN
532		* and active threads is zero).
533	+	*
534	+	* Notes: Most direct CASes are dependent on these bitfield
535	+	* positions. Also, this field is non-private to enable direct
536	+	* performance-sensitive CASes in ForkJoinWorkerThread.
537		*/
538	<	private volatile int runState;
538	>	volatile int runState;
539
540		// Note: The order among run level values matters.
541		private static final int RUNLEVEL_SHIFT = 16;
#	Line 466 \| Line 543 \| public class ForkJoinPool extends Abstra
543		private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1);
544		private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2);
545		private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1;
469	–	private static final int ONE_ACTIVE = 1; // active update delta
546
547		/**
548		* Holds number of total (i.e., created and not yet terminated)
#	Line 475 \| Line 551 \| public class ForkJoinPool extends Abstra
551		* making decisions about creating and suspending spare
552		* threads. Updated only by CAS. Note that adding a new worker
553		* requires incrementing both counts, since workers start off in
554	<	* running state. This field is also used for memory-fencing
479	<	* configuration parameters.
554	>	* running state.
555		*/
556		private volatile int workerCounts;
557
#	Line 485 \| Line 560 \| public class ForkJoinPool extends Abstra
560		private static final int ONE_RUNNING = 1;
561		private static final int ONE_TOTAL = 1 << TOTAL_COUNT_SHIFT;
562
488	–	/*
489	–	* Fields parallelism. maxPoolSize, and maintainsParallelism are
490	–	* non-volatile, but external reads/writes use workerCount fences
491	–	* to ensure visability.
492	–	*/
493	–
563		/**
564		* The target parallelism level.
565	+	* Accessed directly by ForkJoinWorkerThreads.
566		*/
567	<	private int parallelism;
498	<
499	<	/**
500	<	* The maximum allowed pool size.
501	<	*/
502	<	private int maxPoolSize;
567	>	final int parallelism;
568
569		/**
570		* True if use local fifo, not default lifo, for local polling
571	<	* Replicated by ForkJoinWorkerThreads
507	<	*/
508	<	private volatile boolean locallyFifo;
509	<
510	<	/**
511	<	* Controls whether to add spares to maintain parallelism
571	>	* Read by, and replicated by ForkJoinWorkerThreads
572		*/
573	<	private boolean maintainsParallelism;
573	>	final boolean locallyFifo;
574
575		/**
576	<	* The uncaught exception handler used when any worker
577	<	* abruptly terminates
576	>	* The uncaught exception handler used when any worker abruptly
577	>	* terminates.
578		*/
579	<	private volatile Thread.UncaughtExceptionHandler ueh;
579	>	private final Thread.UncaughtExceptionHandler ueh;
580
581		/**
582		* Pool number, just for assigning useful names to worker threads
583		*/
584		private final int poolNumber;
585
586	<	// utilities for updating fields
586	>	// Utilities for CASing fields. Note that most of these
587	>	// are usually manually inlined by callers
588
589		/**
590	<	* Adds delta to running count. Used mainly by ForkJoinTask.
590	>	* Increments running count part of workerCounts
591		*/
592	<	final void updateRunningCount(int delta) {
593	<	int wc;
592	>	final void incrementRunningCount() {
593	>	int c;
594		do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
595	<	wc = workerCounts,
596	<	wc + delta));
595	>	c = workerCounts,
596	>	c + ONE_RUNNING));
597		}
598
599		/**
600	<	* Decrements running count unless already zero
600	>	* Tries to decrement running count unless already zero
601		*/
602		final boolean tryDecrementRunningCount() {
603		int wc = workerCounts;
#	Line 547 \| Line 608 \| public class ForkJoinPool extends Abstra
608		}
609
610		/**
611	<	* Write fence for user modifications of pool parameters
612	<	* (parallelism. etc). Note that it doesn't matter if CAS fails.
552	<	*/
553	<	private void workerCountWriteFence() {
554	<	int wc;
555	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
556	<	wc = workerCounts, wc);
557	<	}
558	<
559	<	/**
560	<	* Read fence for external reads of pool parameters
561	<	* (parallelism. maxPoolSize, etc).
562	<	*/
563	<	private void workerCountReadFence() {
564	<	int ignore = workerCounts;
565	<	}
566	<
567	<	/**
568	<	* Tries incrementing active count; fails on contention.
569	<	* Called by workers before executing tasks.
611	>	* Forces decrement of encoded workerCounts, awaiting nonzero if
612	>	* (rarely) necessary when other count updates lag.
613		*
614	<	* @return true on success
614	>	* @param dr -- either zero or ONE_RUNNING
615	>	* @param dt -- either zero or ONE_TOTAL
616		*/
617	<	final boolean tryIncrementActiveCount() {
618	<	int c;
619	<	return UNSAFE.compareAndSwapInt(this, runStateOffset,
620	<	c = runState, c + ONE_ACTIVE);
617	>	private void decrementWorkerCounts(int dr, int dt) {
618	>	for (;;) {
619	>	int wc = workerCounts;
620	>	if ((wc & RUNNING_COUNT_MASK) - dr < 0 \|\|
621	>	(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
622	>	if ((runState & TERMINATED) != 0)
623	>	return; // lagging termination on a backout
624	>	Thread.yield();
625	>	}
626	>	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
627	>	wc, wc - (dr + dt)))
628	>	return;
629	>	}
630		}
631
632		/**
#	Line 583 \| Line 636 \| public class ForkJoinPool extends Abstra
636		final boolean tryDecrementActiveCount() {
637		int c;
638		return UNSAFE.compareAndSwapInt(this, runStateOffset,
639	<	c = runState, c - ONE_ACTIVE);
639	>	c = runState, c - 1);
640		}
641
642		/**
#	Line 612 \| Line 665 \| public class ForkJoinPool extends Abstra
665		lock.lock();
666		try {
667		ForkJoinWorkerThread[] ws = workers;
668	<	int nws = ws.length;
669	<	if (k < 0 \|\| k >= nws \|\| ws[k] != null) {
670	<	for (k = 0; k < nws && ws[k] != null; ++k)
668	>	int n = ws.length;
669	>	if (k < 0 \|\| k >= n \|\| ws[k] != null) {
670	>	for (k = 0; k < n && ws[k] != null; ++k)
671		;
672	<	if (k == nws)
673	<	ws = Arrays.copyOf(ws, nws << 1);
672	>	if (k == n)
673	>	ws = Arrays.copyOf(ws, n << 1);
674		}
675		ws[k] = w;
676		workers = ws; // volatile array write ensures slot visibility
#	Line 628 \| Line 681 \| public class ForkJoinPool extends Abstra
681		}
682
683		/**
684	<	* Nulls out record of worker in workers array
684	>	* Nulls out record of worker in workers array.
685		*/
686		private void forgetWorker(ForkJoinWorkerThread w) {
687		int idx = w.poolIndex;
688	<	// Locking helps method recordWorker avoid unecessary expansion
688	>	// Locking helps method recordWorker avoid unnecessary expansion
689		final ReentrantLock lock = this.workerLock;
690		lock.lock();
691		try {
#	Line 644 \| Line 697 \| public class ForkJoinPool extends Abstra
697		}
698		}
699
647	–	// adding and removing workers
648	–
700		/**
701	<	* Tries to create and add new worker. Assumes that worker counts
702	<	* are already updated to accommodate the worker, so adjusts on
703	<	* failure.
701	>	* Final callback from terminating worker. Removes record of
702	>	* worker from array, and adjusts counts. If pool is shutting
703	>	* down, tries to complete termination.
704		*
705	<	* @return new worker or null if creation failed
705	>	* @param w the worker
706		*/
707	<	private ForkJoinWorkerThread addWorker() {
708	<	ForkJoinWorkerThread w = null;
709	<	try {
710	<	w = factory.newThread(this);
711	<	} finally { // Adjust on either null or exceptional factory return
712	<	if (w == null) {
662	<	onWorkerCreationFailure();
663	<	return null;
664	<	}
665	<	}
666	<	w.start(recordWorker(w), locallyFifo, ueh);
667	<	return w;
707	>	final void workerTerminated(ForkJoinWorkerThread w) {
708	>	forgetWorker(w);
709	>	decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
710	>	while (w.stealCount != 0) // collect final count
711	>	tryAccumulateStealCount(w);
712	>	tryTerminate(false);
713		}
714
715	+	// Waiting for and signalling events
716	+
717		/**
718	<	* Adjusts counts upon failure to create worker
718	>	* Releases workers blocked on a count not equal to current count.
719	>	* Normally called after precheck that eventWaiters isn't zero to
720	>	* avoid wasted array checks. Gives up upon a change in count or
721	>	* upon releasing two workers, letting others take over.
722		*/
723	<	private void onWorkerCreationFailure() {
724	<	for (;;) {
725	<	int wc = workerCounts;
726	<	if ((wc >>> TOTAL_COUNT_SHIFT) > 0 &&
727	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
728	<	wc, wc - (ONE_RUNNING\|ONE_TOTAL)))
723	>	private void releaseEventWaiters() {
724	>	ForkJoinWorkerThread[] ws = workers;
725	>	int n = ws.length;
726	>	long h = eventWaiters;
727	>	int ec = eventCount;
728	>	boolean releasedOne = false;
729	>	ForkJoinWorkerThread w; int id;
730	>	while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
731	>	(int)(h >>> EVENT_COUNT_SHIFT) != ec &&
732	>	id < n && (w = ws[id]) != null) {
733	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
734	>	h, w.nextWaiter)) {
735	>	LockSupport.unpark(w);
736	>	if (releasedOne) // exit on second release
737	>	break;
738	>	releasedOne = true;
739	>	}
740	>	if (eventCount != ec)
741		break;
742	+	h = eventWaiters;
743		}
681	–	tryTerminate(false); // in case of failure during shutdown
744		}
745
746		/**
747	<	* Create enough total workers to establish target parallelism,
748	<	* giving up if terminating or addWorker fails
747	>	* Tries to advance eventCount and releases waiters. Called only
748	>	* from workers.
749		*/
750	<	private void ensureEnoughTotalWorkers() {
751	<	int wc;
752	<	while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
753	<	runState < TERMINATING) {
754	<	if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
693	<	wc, wc + (ONE_RUNNING\|ONE_TOTAL)) &&
694	<	addWorker() == null))
695	<	break;
696	<	}
750	>	final void signalWork() {
751	>	int c; // try to increment event count -- CAS failure OK
752	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
753	>	if (eventWaiters != 0L)
754	>	releaseEventWaiters();
755		}
756
757		/**
758	<	* Final callback from terminating worker. Removes record of
759	<	* worker from array, and adjusts counts. If pool is shutting
702	<	* down, tries to complete terminatation, else possibly replaces
703	<	* the worker.
758	>	* Adds the given worker to event queue and blocks until
759	>	* terminating or event count advances from the given value
760		*
761	<	* @param w the worker
761	>	* @param w the calling worker thread
762	>	* @param ec the count
763		*/
764	<	final void workerTerminated(ForkJoinWorkerThread w) {
765	<	if (w.active) { // force inactive
766	<	w.active = false;
767	<	do {} while (!tryDecrementActiveCount());
768	<	}
769	<	forgetWorker(w);
770	<
771	<	// Decrement total count, and if was running, running count
772	<	// Spin (waiting for other updates) if either would be negative
773	<	int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
717	<	int unit = ONE_TOTAL + nr;
718	<	for (;;) {
719	<	int wc = workerCounts;
720	<	int rc = wc & RUNNING_COUNT_MASK;
721	<	if (rc - nr < 0 \|\| (wc >>> TOTAL_COUNT_SHIFT) == 0)
722	<	Thread.yield(); // back off if waiting for other updates
723	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
724	<	wc, wc - unit))
764	>	private void eventSync(ForkJoinWorkerThread w, int ec) {
765	>	long nh = (((long)ec) << EVENT_COUNT_SHIFT) \| ((long)(w.poolIndex+1));
766	>	long h;
767	>	while ((runState < SHUTDOWN \|\| !tryTerminate(false)) &&
768	>	(((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 \|\|
769	>	(int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
770	>	eventCount == ec) {
771	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
772	>	w.nextWaiter = h, nh)) {
773	>	awaitEvent(w, ec);
774		break;
775	+	}
776		}
727	–
728	–	accumulateStealCount(w); // collect final count
729	–	if (!tryTerminate(false))
730	–	ensureEnoughTotalWorkers();
777		}
778
733	–	// Waiting for and signalling events
734	–
779		/**
780	<	* Ensures eventCount on exit is different (mod 2^32) than on
781	<	* entry. CAS failures are OK -- any change in count suffices.
780	>	* Blocks the given worker (that has already been entered as an
781	>	* event waiter) until terminating or event count advances from
782	>	* the given value. The oldest (first) waiter uses a timed wait to
783	>	* occasionally one-by-one shrink the number of workers (to a
784	>	* minimum of one) if the pool has not been used for extended
785	>	* periods.
786	>	*
787	>	* @param w the calling worker thread
788	>	* @param ec the count
789		*/
790	<	private void advanceEventCount() {
791	<	int c;
792	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
790	>	private void awaitEvent(ForkJoinWorkerThread w, int ec) {
791	>	while (eventCount == ec) {
792	>	if (tryAccumulateStealCount(w)) { // transfer while idle
793	>	boolean untimed = (w.nextWaiter != 0L \|\|
794	>	(workerCounts & RUNNING_COUNT_MASK) <= 1);
795	>	long startTime = untimed? 0 : System.nanoTime();
796	>	Thread.interrupted(); // clear/ignore interrupt
797	>	if (eventCount != ec \|\| w.runState != 0 \|\|
798	>	runState >= TERMINATING) // recheck after clear
799	>	break;
800	>	if (untimed)
801	>	LockSupport.park(w);
802	>	else {
803	>	LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
804	>	if (eventCount != ec \|\| w.runState != 0 \|\|
805	>	runState >= TERMINATING)
806	>	break;
807	>	if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
808	>	tryShutdownUnusedWorker(ec);
809	>	}
810	>	}
811	>	}
812		}
813
814	+	// Maintaining parallelism
815	+
816		/**
817	<	* Releases workers blocked on a count not equal to current count.
817	>	* Pushes worker onto the spare stack.
818		*/
819	<	final void releaseWaiters() {
820	<	long top;
821	<	int id;
822	<	while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 &&
751	<	(int)(top >>> EVENT_COUNT_SHIFT) != eventCount) {
752	<	ForkJoinWorkerThread[] ws = workers;
753	<	ForkJoinWorkerThread w;
754	<	if (ws.length >= id && (w = ws[id - 1]) != null &&
755	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
756	<	top, w.nextWaiter))
757	<	LockSupport.unpark(w);
758	<	}
819	>	final void pushSpare(ForkJoinWorkerThread w) {
820	>	int ns = (++w.spareCount << SPARE_COUNT_SHIFT) \| (w.poolIndex + 1);
821	>	do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
822	>	w.nextSpare = spareWaiters,ns));
823		}
824
825		/**
826	<	* Advances eventCount and releases waiters until interference by
827	<	* other releasing threads is detected.
826	>	* Tries (once) to resume a spare if the number of running
827	>	* threads is less than target.
828		*/
829	<	final void signalWork() {
830	<	int ec;
831	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1);
832	<	outer:for (;;) {
833	<	long top = eventWaiters;
834	<	ec = eventCount;
835	<	for (;;) {
836	<	ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
837	<	int id = (int)(top & WAITER_INDEX_MASK);
838	<	if (id <= 0 \|\| (int)(top >>> EVENT_COUNT_SHIFT) == ec)
839	<	return;
840	<	if ((ws = workers).length < id \|\| (w = ws[id - 1]) == null \|\|
841	<	!UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
842	<	top, top = w.nextWaiter))
843	<	continue outer; // possibly stale; reread
829	>	private void tryResumeSpare() {
830	>	int sw, id;
831	>	ForkJoinWorkerThread[] ws = workers;
832	>	int n = ws.length;
833	>	ForkJoinWorkerThread w;
834	>	if ((sw = spareWaiters) != 0 &&
835	>	(id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
836	>	id < n && (w = ws[id]) != null &&
837	>	(workerCounts & RUNNING_COUNT_MASK) < parallelism &&
838	>	spareWaiters == sw &&
839	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
840	>	sw, w.nextSpare)) {
841	>	int c; // increment running count before resume
842	>	do {} while (!UNSAFE.compareAndSwapInt
843	>	(this, workerCountsOffset,
844	>	c = workerCounts, c + ONE_RUNNING));
845	>	if (w.tryUnsuspend())
846		LockSupport.unpark(w);
847	<	if (top != eventWaiters) // let someone else take over
848	<	return;
783	<	}
847	>	else // back out if w was shutdown
848	>	decrementWorkerCounts(ONE_RUNNING, 0);
849		}
850		}
851
852		/**
853	<	* If worker is inactive, blocks until terminating or event count
854	<	* advances from last value held by worker; in any case helps
855	<	* release others.
856	<	*
857	<	* @param w the calling worker thread
853	>	* Tries to increase the number of running workers if below target
854	>	* parallelism: If a spare exists tries to resume it via
855	>	* tryResumeSpare. Otherwise, if not enough total workers or all
856	>	* existing workers are busy, adds a new worker. In all cases also
857	>	* helps wake up releasable workers waiting for work.
858		*/
859	<	private void eventSync(ForkJoinWorkerThread w) {
860	<	if (!w.active) {
861	<	int prev = w.lastEventCount;
862	<	long nextTop = (((long)prev << EVENT_COUNT_SHIFT) \|
863	<	((long)(w.poolIndex + 1)));
864	<	long top;
865	<	while ((runState < SHUTDOWN \|\| !tryTerminate(false)) &&
866	<	(((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 \|\|
867	<	(int)(top >>> EVENT_COUNT_SHIFT) == prev) &&
868	<	eventCount == prev) {
869	<	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
870	<	w.nextWaiter = top, nextTop)) {
871	<	accumulateStealCount(w); // transfer steals while idle
872	<	Thread.interrupted(); // clear/ignore interrupt
873	<	while (eventCount == prev)
874	<	w.doPark();
859	>	private void helpMaintainParallelism() {
860	>	int pc = parallelism;
861	>	int wc, rs, tc;
862	>	while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
863	>	(rs = runState) < TERMINATING) {
864	>	if (spareWaiters != 0)
865	>	tryResumeSpare();
866	>	else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS \|\|
867	>	(tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
868	>	break; // enough total
869	>	else if (runState == rs && workerCounts == wc &&
870	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
871	>	wc + (ONE_RUNNING\|ONE_TOTAL))) {
872	>	ForkJoinWorkerThread w = null;
873	>	try {
874	>	w = factory.newThread(this);
875	>	} finally { // adjust on null or exceptional factory return
876	>	if (w == null) {
877	>	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
878	>	tryTerminate(false); // handle failure during shutdown
879	>	}
880	>	}
881	>	if (w == null)
882		break;
883	+	w.start(recordWorker(w), ueh);
884	+	if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
885	+	int c; // advance event count
886	+	UNSAFE.compareAndSwapInt(this, eventCountOffset,
887	+	c = eventCount, c+1);
888	+	break; // add at most one unless total below target
889		}
890		}
813	–	w.lastEventCount = eventCount;
891		}
892	<	releaseWaiters();
892	>	if (eventWaiters != 0L)
893	>	releaseEventWaiters();
894		}
895
896		/**
897	<	* Callback from workers invoked upon each top-level action (i.e.,
898	<	* stealing a task or taking a submission and running
899	<	* it). Performs one or both of the following:
897	>	* Callback from the oldest waiter in awaitEvent waking up after a
898	>	* period of non-use. If all workers are idle, tries (once) to
899	>	* shutdown an event waiter or a spare, if one exists. Note that
900	>	* we don't need CAS or locks here because the method is called
901	>	* only from one thread occasionally waking (and even misfires are
902	>	* OK). Note that until the shutdown worker fully terminates,
903	>	* workerCounts will overestimate total count, which is tolerable.
904		*
905	<	* * If the worker cannot find work, updates its active status to
906	<	* inactive and updates activeCount unless there is contention, in
825	<	* which case it may try again (either in this or a subsequent
826	<	* call). Additionally, awaits the next task event and/or helps
827	<	* wake up other releasable waiters.
828	<	*
829	<	* * If there are too many running threads, suspends this worker
830	<	* (first forcing inactivation if necessary). If it is not
831	<	* resumed before a keepAlive elapses, the worker may be "trimmed"
832	<	* -- killed while suspended within suspendAsSpare. Otherwise,
833	<	* upon resume it rechecks to make sure that it is still needed.
834	<	*
835	<	* @param w the worker
836	<	* @param worked false if the worker scanned for work but didn't
837	<	* find any (in which case it may block waiting for work).
905	>	* @param ec the event count waited on by caller (to abort
906	>	* attempt if count has since changed).
907		*/
908	<	final void preStep(ForkJoinWorkerThread w, boolean worked) {
909	<	boolean active = w.active;
910	<	boolean inactivate = !worked & active;
911	<	for (;;) {
912	<	if (inactivate) {
913	<	int c = runState;
914	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset,
915	<	c, c - ONE_ACTIVE))
916	<	inactivate = active = w.active = false;
908	>	private void tryShutdownUnusedWorker(int ec) {
909	>	if (runState == 0 && eventCount == ec) { // only trigger if all idle
910	>	ForkJoinWorkerThread[] ws = workers;
911	>	int n = ws.length;
912	>	ForkJoinWorkerThread w = null;
913	>	boolean shutdown = false;
914	>	int sw;
915	>	long h;
916	>	if ((sw = spareWaiters) != 0) { // prefer killing spares
917	>	int id = (sw & SPARE_ID_MASK) - 1;
918	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
919	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
920	>	sw, w.nextSpare))
921	>	shutdown = true;
922		}
923	<	int wc = workerCounts;
924	<	if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
925	<	if (!worked)
926	<	eventSync(w);
927	<	return;
923	>	else if ((h = eventWaiters) != 0L) {
924	>	long nh;
925	>	int id = ((int)(h & WAITER_ID_MASK)) - 1;
926	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
927	>	(nh = w.nextWaiter) != 0L && // keep at least one worker
928	>	UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
929	>	shutdown = true;
930	>	}
931	>	if (w != null && shutdown) {
932	>	w.shutdown();
933	>	LockSupport.unpark(w);
934		}
855	–	if (!(inactivate \|= active) && // must inactivate to suspend
856	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
857	–	wc, wc - ONE_RUNNING) &&
858	–	!w.suspendAsSpare()) // false if trimmed
859	–	return;
935		}
936	+	releaseEventWaiters(); // in case of interference
937		}
938
939		/**
940	<	* Adjusts counts and creates or resumes compensating threads for
941	<	* a worker blocking on task joinMe. First tries resuming an
942	<	* existing spare (which usually also avoids any count
867	<	* adjustment), but must then decrement running count to determine
868	<	* whether a new thread is needed. See above for fuller
869	<	* explanation. This code is sprawled out non-modularly mainly
870	<	* because adaptive spinning works best if the entire method is
871	<	* either interpreted or compiled vs having only some pieces of it
872	<	* compiled.
940	>	* Callback from workers invoked upon each top-level action (i.e.,
941	>	* stealing a task or taking a submission and running it).
942	>	* Performs one or more of the following:
943		*
944	<	* @param joinMe the task to join
945	<	* @return task status on exit (to simplify usage by callers)
944	>	* 1. If the worker is active and either did not run a task
945	>	* or there are too many workers, try to set its active status
946	>	* to inactive and update activeCount. On contention, we may
947	>	* try again in this or a subsequent call.
948	>	*
949	>	* 2. If not enough total workers, help create some.
950	>	*
951	>	* 3. If there are too many running workers, suspend this worker
952	>	* (first forcing inactive if necessary). If it is not needed,
953	>	* it may be shutdown while suspended (via
954	>	* tryShutdownUnusedWorker). Otherwise, upon resume it
955	>	* rechecks running thread count and need for event sync.
956	>	*
957	>	* 4. If worker did not run a task, await the next task event via
958	>	* eventSync if necessary (first forcing inactivation), upon
959	>	* which the worker may be shutdown via
960	>	* tryShutdownUnusedWorker. Otherwise, help release any
961	>	* existing event waiters that are now releasable,
962	>	*
963	>	* @param w the worker
964	>	* @param ran true if worker ran a task since last call to this method
965		*/
966	<	final int awaitJoin(ForkJoinTask<?> joinMe) {
966	>	final void preStep(ForkJoinWorkerThread w, boolean ran) {
967	>	int wec = w.lastEventCount;
968	>	boolean active = w.active;
969	>	boolean inactivate = false;
970		int pc = parallelism;
971	<	boolean adj = false; // true when running count adjusted
972	<	int scans = 0;
973	<
974	<	while (joinMe.status >= 0) {
975	<	ForkJoinWorkerThread spare = null;
884	<	if ((workerCounts & RUNNING_COUNT_MASK) < pc) {
885	<	ForkJoinWorkerThread[] ws = workers;
886	<	int nws = ws.length;
887	<	for (int i = 0; i < nws; ++i) {
888	<	ForkJoinWorkerThread w = ws[i];
889	<	if (w != null && w.isSuspended()) {
890	<	spare = w;
891	<	break;
892	<	}
893	<	}
894	<	if (joinMe.status < 0)
895	<	break;
896	<	}
971	>	int rs;
972	>	while (w.runState == 0 && (rs = runState) < TERMINATING) {
973	>	if ((inactivate \|\| (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
974	>	UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
975	>	inactivate = active = w.active = false;
976		int wc = workerCounts;
977	<	int rc = wc & RUNNING_COUNT_MASK;
978	<	int dc = pc - rc;
979	<	if (dc > 0 && spare != null && spare.tryUnsuspend()) {
980	<	if (adj) {
981	<	int c;
982	<	do {} while (!UNSAFE.compareAndSwapInt
904	<	(this, workerCountsOffset,
905	<	c = workerCounts, c + ONE_RUNNING));
906	<	}
907	<	adj = true;
908	<	LockSupport.unpark(spare);
977	>	if ((wc & RUNNING_COUNT_MASK) > pc) {
978	>	if (!(inactivate \|= active) && // must inactivate to suspend
979	>	workerCounts == wc && // try to suspend as spare
980	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
981	>	wc, wc - ONE_RUNNING))
982	>	w.suspendAsSpare();
983		}
984	<	else if (adj) {
985	<	if (dc <= 0)
984	>	else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
985	>	helpMaintainParallelism(); // not enough workers
986	>	else if (!ran) {
987	>	long h = eventWaiters;
988	>	int ec = eventCount;
989	>	if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
990	>	releaseEventWaiters(); // release others before waiting
991	>	else if (ec != wec) {
992	>	w.lastEventCount = ec; // no need to wait
993		break;
913	–	int tc = wc >>> TOTAL_COUNT_SHIFT;
914	–	if (scans > tc) {
915	–	int ts = (tc - pc) * pc;
916	–	if (rc != 0 && (dc * dc < ts \|\| !maintainsParallelism))
917	–	break;
918	–	if (scans > ts && tc < maxPoolSize &&
919	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
920	–	wc+(ONE_RUNNING\|ONE_TOTAL))){
921	–	addWorker();
922	–	break;
923	–	}
994		}
995	+	else if (!(inactivate \|= active))
996	+	eventSync(w, wec); // must inactivate before sync
997		}
926	–	else if (rc != 0)
927	–	adj = UNSAFE.compareAndSwapInt (this, workerCountsOffset,
928	–	wc, wc - ONE_RUNNING);
929	–	if ((scans++ & 1) == 0)
930	–	releaseWaiters(); // help others progress
998		else
999	<	Thread.yield(); // avoid starving productive threads
933	<	}
934	<
935	<	if (adj) {
936	<	joinMe.internalAwaitDone();
937	<	int c;
938	<	do {} while (!UNSAFE.compareAndSwapInt
939	<	(this, workerCountsOffset,
940	<	c = workerCounts, c + ONE_RUNNING));
999	>	break;
1000		}
942	–	return joinMe.status;
1001		}
1002
1003		/**
1004	<	* Same idea as awaitJoin
1004	>	* Helps and/or blocks awaiting join of the given task.
1005	>	* See above for explanation.
1006	>	*
1007	>	* @param joinMe the task to join
1008	>	* @param worker the current worker thread
1009		*/
1010	<	final void awaitBlocker(ManagedBlocker blocker, boolean maintainPar)
1011	<	throws InterruptedException {
1012	<	maintainPar &= maintainsParallelism;
1013	<	int pc = parallelism;
1014	<	boolean adj = false; // true when running count adjusted
1015	<	int scans = 0;
954	<	boolean done;
955	<
956	<	for (;;) {
957	<	if (done = blocker.isReleasable())
1010	>	final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1011	>	int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1012	>	while (joinMe.status >= 0) {
1013	>	int wc;
1014	>	worker.helpJoinTask(joinMe);
1015	>	if (joinMe.status < 0)
1016		break;
1017	<	ForkJoinWorkerThread spare = null;
1018	<	if ((workerCounts & RUNNING_COUNT_MASK) < pc) {
1019	<	ForkJoinWorkerThread[] ws = workers;
1020	<	int nws = ws.length;
1021	<	for (int i = 0; i < nws; ++i) {
1022	<	ForkJoinWorkerThread w = ws[i];
1023	<	if (w != null && w.isSuspended()) {
1024	<	spare = w;
1025	<	break;
1026	<	}
1027	<	}
1028	<	if (done = blocker.isReleasable())
1029	<	break;
1030	<	}
1031	<	int wc = workerCounts;
974	<	int rc = wc & RUNNING_COUNT_MASK;
975	<	int dc = pc - rc;
976	<	if (dc > 0 && spare != null && spare.tryUnsuspend()) {
977	<	if (adj) {
978	<	int c;
979	<	do {} while (!UNSAFE.compareAndSwapInt
980	<	(this, workerCountsOffset,
981	<	c = workerCounts, c + ONE_RUNNING));
982	<	}
983	<	adj = true;
984	<	LockSupport.unpark(spare);
985	<	}
986	<	else if (adj) {
987	<	if (dc <= 0)
988	<	break;
989	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
990	<	if (scans > tc) {
991	<	int ts = (tc - pc) * pc;
992	<	if (rc != 0 && (dc * dc < ts \|\| !maintainPar))
993	<	break;
994	<	if (scans > ts && tc < maxPoolSize &&
995	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
996	<	wc+(ONE_RUNNING\|ONE_TOTAL))){
997	<	addWorker();
998	<	break;
999	<	}
1000	<	}
1001	<	}
1002	<	else if (rc != 0)
1003	<	adj = UNSAFE.compareAndSwapInt (this, workerCountsOffset,
1004	<	wc, wc - ONE_RUNNING);
1005	<	if ((++scans & 1) == 0)
1006	<	releaseWaiters(); // help others progress
1007	<	else
1008	<	Thread.yield(); // avoid starving productive threads
1009	<	}
1010	<
1011	<	try {
1012	<	if (!done)
1013	<	do {} while (!blocker.isReleasable() && !blocker.block());
1014	<	} finally {
1015	<	if (adj) {
1016	<	int c;
1017	>	else if (retries > 0)
1018	>	--retries;
1019	>	else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1020	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1021	>	wc, wc - ONE_RUNNING)) {
1022	>	int stat, c; long h;
1023	>	while ((stat = joinMe.status) >= 0 &&
1024	>	(h = eventWaiters) != 0L && // help release others
1025	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1026	>	releaseEventWaiters();
1027	>	if (stat >= 0 &&
1028	>	((workerCounts & RUNNING_COUNT_MASK) == 0 \|\|
1029	>	(stat =
1030	>	joinMe.internalAwaitDone(JOIN_TIMEOUT_MILLIS)) >= 0))
1031	>	helpMaintainParallelism(); // timeout or no running workers
1032		do {} while (!UNSAFE.compareAndSwapInt
1033		(this, workerCountsOffset,
1034		c = workerCounts, c + ONE_RUNNING));
1035	+	if (stat < 0)
1036	+	break; // else restart
1037		}
1038		}
1039		}
1040
1041		/**
1042	<	* Unless there are not enough other running threads, adjusts
1026	<	* counts and blocks a worker performing helpJoin that cannot find
1027	<	* any work.
1028	<	*
1029	<	* @return true if joinMe now done
1042	>	* Same idea as awaitJoin, but no helping, retries, or timeouts.
1043		*/
1044	<	final boolean tryAwaitBusyJoin(ForkJoinTask<?> joinMe) {
1045	<	int pc = parallelism;
1046	<	outer:for (;;) {
1047	<	releaseWaiters();
1048	<	if ((workerCounts & RUNNING_COUNT_MASK) < pc) {
1049	<	ForkJoinWorkerThread[] ws = workers;
1050	<	int nws = ws.length;
1051	<	for (int i = 0; i < nws; ++i) {
1052	<	ForkJoinWorkerThread w = ws[i];
1053	<	if (w != null && w.isSuspended()) {
1054	<	if (joinMe.status < 0)
1055	<	return true;
1056	<	if ((workerCounts & RUNNING_COUNT_MASK) > pc)
1044	>	final void awaitBlocker(ManagedBlocker blocker)
1045	>	throws InterruptedException {
1046	>	while (!blocker.isReleasable()) {
1047	>	int wc = workerCounts;
1048	>	if ((wc & RUNNING_COUNT_MASK) != 0 &&
1049	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1050	>	wc, wc - ONE_RUNNING)) {
1051	>	try {
1052	>	while (!blocker.isReleasable()) {
1053	>	long h = eventWaiters;
1054	>	if (h != 0L &&
1055	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1056	>	releaseEventWaiters();
1057	>	else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1058	>	runState < TERMINATING)
1059	>	helpMaintainParallelism();
1060	>	else if (blocker.block())
1061		break;
1045	–	if (w.tryUnsuspend()) {
1046	–	LockSupport.unpark(w);
1047	–	break outer;
1048	–	}
1049	–	continue outer;
1062		}
1063	+	} finally {
1064	+	int c;
1065	+	do {} while (!UNSAFE.compareAndSwapInt
1066	+	(this, workerCountsOffset,
1067	+	c = workerCounts, c + ONE_RUNNING));
1068		}
1052	–	}
1053	–	if (joinMe.status < 0)
1054	–	return true;
1055	–	int wc = workerCounts;
1056	–	if ((wc & RUNNING_COUNT_MASK) <= 2 \|\|
1057	–	(wc >>> TOTAL_COUNT_SHIFT) < pc)
1058	–	return false; // keep this thread alive
1059	–	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1060	–	wc, wc - ONE_RUNNING))
1069		break;
1070	+	}
1071		}
1063	–
1064	–	joinMe.internalAwaitDone();
1065	–	int c;
1066	–	do {} while (!UNSAFE.compareAndSwapInt
1067	–	(this, workerCountsOffset,
1068	–	c = workerCounts, c + ONE_RUNNING));
1069	–	return true;
1072		}
1073
1074		/**
#	Line 1090 \| Line 1092 \| public class ForkJoinPool extends Abstra
1092		// Finish now if all threads terminated; else in some subsequent call
1093		if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1094		advanceRunLevel(TERMINATED);
1095	<	terminationLatch.countDown();
1095	>	termination.arrive();
1096		}
1097		return true;
1098		}
1099
1100		/**
1101		* Actions on transition to TERMINATING
1102	+	*
1103	+	* Runs up to four passes through workers: (0) shutting down each
1104	+	* (without waking up if parked) to quickly spread notifications
1105	+	* without unnecessary bouncing around event queues etc (1) wake
1106	+	* up and help cancel tasks (2) interrupt (3) mop up races with
1107	+	* interrupted workers
1108		*/
1109		private void startTerminating() {
1110	<	for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1111	<	cancelSubmissions();
1112	<	shutdownWorkers();
1113	<	cancelWorkerTasks();
1114	<	advanceEventCount();
1115	<	releaseWaiters();
1116	<	interruptWorkers();
1110	>	cancelSubmissions();
1111	>	for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1112	>	int c; // advance event count
1113	>	UNSAFE.compareAndSwapInt(this, eventCountOffset,
1114	>	c = eventCount, c+1);
1115	>	eventWaiters = 0L; // clobber lists
1116	>	spareWaiters = 0;
1117	>	for (ForkJoinWorkerThread w : workers) {
1118	>	if (w != null) {
1119	>	w.shutdown();
1120	>	if (passes > 0 && !w.isTerminated()) {
1121	>	w.cancelTasks();
1122	>	LockSupport.unpark(w);
1123	>	if (passes > 1) {
1124	>	try {
1125	>	w.interrupt();
1126	>	} catch (SecurityException ignore) {
1127	>	}
1128	>	}
1129	>	}
1130	>	}
1131	>	}
1132		}
1133		}
1134
1135		/**
1136	<	* Clear out and cancel submissions, ignoring exceptions
1136	>	* Clears out and cancels submissions, ignoring exceptions.
1137		*/
1138		private void cancelSubmissions() {
1139		ForkJoinTask<?> task;
#	Line 1122 \| Line 1145 \| public class ForkJoinPool extends Abstra
1145		}
1146		}
1147
1125	–	/**
1126	–	* Sets all worker run states to at least shutdown,
1127	–	* also resuming suspended workers
1128	–	*/
1129	–	private void shutdownWorkers() {
1130	–	ForkJoinWorkerThread[] ws = workers;
1131	–	int nws = ws.length;
1132	–	for (int i = 0; i < nws; ++i) {
1133	–	ForkJoinWorkerThread w = ws[i];
1134	–	if (w != null)
1135	–	w.shutdown();
1136	–	}
1137	–	}
1138	–
1139	–	/**
1140	–	* Clears out and cancels all locally queued tasks
1141	–	*/
1142	–	private void cancelWorkerTasks() {
1143	–	ForkJoinWorkerThread[] ws = workers;
1144	–	int nws = ws.length;
1145	–	for (int i = 0; i < nws; ++i) {
1146	–	ForkJoinWorkerThread w = ws[i];
1147	–	if (w != null)
1148	–	w.cancelTasks();
1149	–	}
1150	–	}
1151	–
1152	–	/**
1153	–	* Unsticks all workers blocked on joins etc
1154	–	*/
1155	–	private void interruptWorkers() {
1156	–	ForkJoinWorkerThread[] ws = workers;
1157	–	int nws = ws.length;
1158	–	for (int i = 0; i < nws; ++i) {
1159	–	ForkJoinWorkerThread w = ws[i];
1160	–	if (w != null && !w.isTerminated()) {
1161	–	try {
1162	–	w.interrupt();
1163	–	} catch (SecurityException ignore) {
1164	–	}
1165	–	}
1166	–	}
1167	–	}
1168	–
1148		// misc support for ForkJoinWorkerThread
1149
1150		/**
1151	<	* Returns pool number
1151	>	* Returns pool number.
1152		*/
1153		final int getPoolNumber() {
1154		return poolNumber;
1155		}
1156
1157		/**
1158	<	* Accumulates steal count from a worker, clearing
1159	<	* the worker's value
1158	>	* Tries to accumulate steal count from a worker, clearing
1159	>	* the worker's value if successful.
1160	>	*
1161	>	* @return true if worker steal count now zero
1162		*/
1163	<	final void accumulateStealCount(ForkJoinWorkerThread w) {
1163	>	final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1164		int sc = w.stealCount;
1165	<	if (sc != 0) {
1166	<	long c;
1167	<	w.stealCount = 0;
1168	<	do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1169	<	c = stealCount, c + sc));
1165	>	long c = stealCount;
1166	>	// CAS even if zero, for fence effects
1167	>	if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1168	>	if (sc != 0)
1169	>	w.stealCount = 0;
1170	>	return true;
1171		}
1172	+	return sc == 0;
1173		}
1174
1175		/**
#	Line 1194 \| Line 1177 \| public class ForkJoinPool extends Abstra
1177		* active thread.
1178		*/
1179		final int idlePerActive() {
1180	<	int ac = runState; // no mask -- artifically boosts during shutdown
1181	<	int pc = parallelism; // use targeted parallelism, not rc
1180	>	int pc = parallelism; // use parallelism, not rc
1181	>	int ac = runState; // no mask -- artificially boosts during shutdown
1182		// Use exact results for small values, saturate past 4
1183	<	return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1183	>	return ((pc <= ac) ? 0 :
1184	>	(pc >>> 1 <= ac) ? 1 :
1185	>	(pc >>> 2 <= ac) ? 3 :
1186	>	pc >>> 3);
1187		}
1188
1189		// Public and protected methods
#	Line 1206 \| Line 1192 \| public class ForkJoinPool extends Abstra
1192
1193		/**
1194		* Creates a {@code ForkJoinPool} with parallelism equal to {@link
1195	<	* java.lang.Runtime#availableProcessors}, and using the {@linkplain
1196	<	* #defaultForkJoinWorkerThreadFactory default thread factory}.
1195	>	* java.lang.Runtime#availableProcessors}, using the {@linkplain
1196	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
1197	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
1198		*
1199		* @throws SecurityException if a security manager exists and
1200		* the caller is not permitted to modify threads
#	Line 1216 \| Line 1203 \| public class ForkJoinPool extends Abstra
1203		*/
1204		public ForkJoinPool() {
1205		this(Runtime.getRuntime().availableProcessors(),
1206	<	defaultForkJoinWorkerThreadFactory);
1206	>	defaultForkJoinWorkerThreadFactory, null, false);
1207		}
1208
1209		/**
1210		* Creates a {@code ForkJoinPool} with the indicated parallelism
1211	<	* level and using the {@linkplain
1212	<	* #defaultForkJoinWorkerThreadFactory default thread factory}.
1211	>	* level, the {@linkplain
1212	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
1213	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
1214		*
1215		* @param parallelism the parallelism level
1216		* @throws IllegalArgumentException if parallelism less than or
#	Line 1233 \| Line 1221 \| public class ForkJoinPool extends Abstra
1221		* java.lang.RuntimePermission}{@code ("modifyThread")}
1222		*/
1223		public ForkJoinPool(int parallelism) {
1224	<	this(parallelism, defaultForkJoinWorkerThreadFactory);
1224	>	this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1225		}
1226
1227		/**
1228	<	* Creates a {@code ForkJoinPool} with parallelism equal to {@link
1241	<	* java.lang.Runtime#availableProcessors}, and using the given
1242	<	* thread factory.
1228	>	* Creates a {@code ForkJoinPool} with the given parameters.
1229		*
1230	<	* @param factory the factory for creating new threads
1231	<	* @throws NullPointerException if the factory is null
1232	<	* @throws SecurityException if a security manager exists and
1233	<	* the caller is not permitted to modify threads
1234	<	* because it does not hold {@link
1235	<	* java.lang.RuntimePermission}{@code ("modifyThread")}
1236	<	*/
1237	<	public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
1238	<	this(Runtime.getRuntime().availableProcessors(), factory);
1239	<	}
1240	<
1241	<	/**
1242	<	* Creates a {@code ForkJoinPool} with the given parallelism and
1257	<	* thread factory.
1258	<	*
1259	<	* @param parallelism the parallelism level
1260	<	* @param factory the factory for creating new threads
1230	>	* @param parallelism the parallelism level. For default value,
1231	>	* use {@link java.lang.Runtime#availableProcessors}.
1232	>	* @param factory the factory for creating new threads. For default value,
1233	>	* use {@link #defaultForkJoinWorkerThreadFactory}.
1234	>	* @param handler the handler for internal worker threads that
1235	>	* terminate due to unrecoverable errors encountered while executing
1236	>	* tasks. For default value, use {@code null}.
1237	>	* @param asyncMode if true,
1238	>	* establishes local first-in-first-out scheduling mode for forked
1239	>	* tasks that are never joined. This mode may be more appropriate
1240	>	* than default locally stack-based mode in applications in which
1241	>	* worker threads only process event-style asynchronous tasks.
1242	>	* For default value, use {@code false}.
1243		* @throws IllegalArgumentException if parallelism less than or
1244		* equal to zero, or greater than implementation limit
1245		* @throws NullPointerException if the factory is null
#	Line 1266 \| Line 1248 \| public class ForkJoinPool extends Abstra
1248		* because it does not hold {@link
1249		* java.lang.RuntimePermission}{@code ("modifyThread")}
1250		*/
1251	<	public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
1251	>	public ForkJoinPool(int parallelism,
1252	>	ForkJoinWorkerThreadFactory factory,
1253	>	Thread.UncaughtExceptionHandler handler,
1254	>	boolean asyncMode) {
1255		checkPermission();
1256		if (factory == null)
1257		throw new NullPointerException();
1258	<	if (parallelism <= 0 \|\| parallelism > MAX_THREADS)
1258	>	if (parallelism <= 0 \|\| parallelism > MAX_WORKERS)
1259		throw new IllegalArgumentException();
1275	–	this.poolNumber = poolNumberGenerator.incrementAndGet();
1276	–	int arraySize = initialArraySizeFor(parallelism);
1260		this.parallelism = parallelism;
1261		this.factory = factory;
1262	<	this.maxPoolSize = MAX_THREADS;
1263	<	this.maintainsParallelism = true;
1262	>	this.ueh = handler;
1263	>	this.locallyFifo = asyncMode;
1264	>	int arraySize = initialArraySizeFor(parallelism);
1265		this.workers = new ForkJoinWorkerThread[arraySize];
1266		this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1267		this.workerLock = new ReentrantLock();
1268	<	this.terminationLatch = new CountDownLatch(1);
1268	>	this.termination = new Phaser(1);
1269	>	this.poolNumber = poolNumberGenerator.incrementAndGet();
1270		}
1271
1272		/**
#	Line 1289 \| Line 1274 \| public class ForkJoinPool extends Abstra
1274		* @param pc the initial parallelism level
1275		*/
1276		private static int initialArraySizeFor(int pc) {
1277	<	// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1278	<	int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1277	>	// If possible, initially allocate enough space for one spare
1278	>	int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1279	>	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1280		size \|= size >>> 1;
1281		size \|= size >>> 2;
1282		size \|= size >>> 4;
#	Line 1309 \| Line 1295 \| public class ForkJoinPool extends Abstra
1295		if (runState >= SHUTDOWN)
1296		throw new RejectedExecutionException();
1297		submissionQueue.offer(task);
1298	<	advanceEventCount();
1299	<	releaseWaiters();
1300	<	ensureEnoughTotalWorkers();
1298	>	int c; // try to increment event count -- CAS failure OK
1299	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1300	>	helpMaintainParallelism(); // create, start, or resume some workers
1301		}
1302
1303		/**
#	Line 1357 \| Line 1343 \| public class ForkJoinPool extends Abstra
1343		}
1344
1345		/**
1346	+	* Submits a ForkJoinTask for execution.
1347	+	*
1348	+	* @param task the task to submit
1349	+	* @return the task
1350	+	* @throws NullPointerException if the task is null
1351	+	* @throws RejectedExecutionException if the task cannot be
1352	+	* scheduled for execution
1353	+	*/
1354	+	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1355	+	doSubmit(task);
1356	+	return task;
1357	+	}
1358	+
1359	+	/**
1360		* @throws NullPointerException if the task is null
1361		* @throws RejectedExecutionException if the task cannot be
1362		* scheduled for execution
#	Line 1394 \| Line 1394 \| public class ForkJoinPool extends Abstra
1394		}
1395
1396		/**
1397	–	* Submits a ForkJoinTask for execution.
1398	–	*
1399	–	* @param task the task to submit
1400	–	* @return the task
1401	–	* @throws NullPointerException if the task is null
1402	–	* @throws RejectedExecutionException if the task cannot be
1403	–	* scheduled for execution
1404	–	*/
1405	–	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1406	–	doSubmit(task);
1407	–	return task;
1408	–	}
1409	–
1410	–	/**
1397		* @throws NullPointerException {@inheritDoc}
1398		* @throws RejectedExecutionException {@inheritDoc}
1399		*/
#	Line 1449 \| Line 1435 \| public class ForkJoinPool extends Abstra
1435		* @return the handler, or {@code null} if none
1436		*/
1437		public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1452	–	workerCountReadFence();
1438		return ueh;
1439		}
1440
1441		/**
1457	–	* Sets the handler for internal worker threads that terminate due
1458	–	* to unrecoverable errors encountered while executing tasks.
1459	–	* Unless set, the current default or ThreadGroup handler is used
1460	–	* as handler.
1461	–	*
1462	–	* @param h the new handler
1463	–	* @return the old handler, or {@code null} if none
1464	–	* @throws SecurityException if a security manager exists and
1465	–	* the caller is not permitted to modify threads
1466	–	* because it does not hold {@link
1467	–	* java.lang.RuntimePermission}{@code ("modifyThread")}
1468	–	*/
1469	–	public Thread.UncaughtExceptionHandler
1470	–	setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
1471	–	checkPermission();
1472	–	Thread.UncaughtExceptionHandler old = ueh;
1473	–	if (h != old) {
1474	–	ueh = h;
1475	–	ForkJoinWorkerThread[] ws = workers;
1476	–	int nws = ws.length;
1477	–	for (int i = 0; i < nws; ++i) {
1478	–	ForkJoinWorkerThread w = ws[i];
1479	–	if (w != null)
1480	–	w.setUncaughtExceptionHandler(h);
1481	–	}
1482	–	}
1483	–	return old;
1484	–	}
1485	–
1486	–	/**
1487	–	* Sets the target parallelism level of this pool.
1488	–	*
1489	–	* @param parallelism the target parallelism
1490	–	* @throws IllegalArgumentException if parallelism less than or
1491	–	* equal to zero or greater than maximum size bounds
1492	–	* @throws SecurityException if a security manager exists and
1493	–	* the caller is not permitted to modify threads
1494	–	* because it does not hold {@link
1495	–	* java.lang.RuntimePermission}{@code ("modifyThread")}
1496	–	*/
1497	–	public void setParallelism(int parallelism) {
1498	–	checkPermission();
1499	–	if (parallelism <= 0 \|\| parallelism > maxPoolSize)
1500	–	throw new IllegalArgumentException();
1501	–	workerCountReadFence();
1502	–	int pc = this.parallelism;
1503	–	if (pc != parallelism) {
1504	–	this.parallelism = parallelism;
1505	–	workerCountWriteFence();
1506	–	// Release spares. If too many, some will die after re-suspend
1507	–	ForkJoinWorkerThread[] ws = workers;
1508	–	int nws = ws.length;
1509	–	for (int i = 0; i < nws; ++i) {
1510	–	ForkJoinWorkerThread w = ws[i];
1511	–	if (w != null && w.tryUnsuspend()) {
1512	–	int c;
1513	–	do {} while (!UNSAFE.compareAndSwapInt
1514	–	(this, workerCountsOffset,
1515	–	c = workerCounts, c + ONE_RUNNING));
1516	–	LockSupport.unpark(w);
1517	–	}
1518	–	}
1519	–	ensureEnoughTotalWorkers();
1520	–	advanceEventCount();
1521	–	releaseWaiters(); // force config recheck by existing workers
1522	–	}
1523	–	}
1524	–
1525	–	/**
1442		* Returns the targeted parallelism level of this pool.
1443		*
1444		* @return the targeted parallelism level of this pool
1445		*/
1446		public int getParallelism() {
1531	–	// workerCountReadFence(); // inlined below
1532	–	int ignore = workerCounts;
1447		return parallelism;
1448		}
1449
1450		/**
1451		* Returns the number of worker threads that have started but not
1452	<	* yet terminated. This result returned by this method may differ
1452	>	* yet terminated. The result returned by this method may differ
1453		* from {@link #getParallelism} when threads are created to
1454		* maintain parallelism when others are cooperatively blocked.
1455		*
#	Line 1546 \| Line 1460 \| public class ForkJoinPool extends Abstra
1460		}
1461
1462		/**
1549	–	* Returns the maximum number of threads allowed to exist in the
1550	–	* pool. Unless set using {@link #setMaximumPoolSize}, the
1551	–	* maximum is an implementation-defined value designed only to
1552	–	* prevent runaway growth.
1553	–	*
1554	–	* @return the maximum
1555	–	*/
1556	–	public int getMaximumPoolSize() {
1557	–	workerCountReadFence();
1558	–	return maxPoolSize;
1559	–	}
1560	–
1561	–	/**
1562	–	* Sets the maximum number of threads allowed to exist in the
1563	–	* pool. The given value should normally be greater than or equal
1564	–	* to the {@link #getParallelism parallelism} level. Setting this
1565	–	* value has no effect on current pool size. It controls
1566	–	* construction of new threads. The use of this method may cause
1567	–	* tasks that intrinsically require extra threads for dependent
1568	–	* computations to indefinitely stall. If you are instead trying
1569	–	* to minimize internal thread creation, consider setting {@link
1570	–	* #setMaintainsParallelism} as false.
1571	–	*
1572	–	* @throws IllegalArgumentException if negative or greater than
1573	–	* internal implementation limit
1574	–	*/
1575	–	public void setMaximumPoolSize(int newMax) {
1576	–	if (newMax < 0 \|\| newMax > MAX_THREADS)
1577	–	throw new IllegalArgumentException();
1578	–	maxPoolSize = newMax;
1579	–	workerCountWriteFence();
1580	–	}
1581	–
1582	–	/**
1583	–	* Returns {@code true} if this pool dynamically maintains its
1584	–	* target parallelism level. If false, new threads are added only
1585	–	* to avoid possible starvation. This setting is by default true.
1586	–	*
1587	–	* @return {@code true} if maintains parallelism
1588	–	*/
1589	–	public boolean getMaintainsParallelism() {
1590	–	workerCountReadFence();
1591	–	return maintainsParallelism;
1592	–	}
1593	–
1594	–	/**
1595	–	* Sets whether this pool dynamically maintains its target
1596	–	* parallelism level. If false, new threads are added only to
1597	–	* avoid possible starvation.
1598	–	*
1599	–	* @param enable {@code true} to maintain parallelism
1600	–	*/
1601	–	public void setMaintainsParallelism(boolean enable) {
1602	–	maintainsParallelism = enable;
1603	–	workerCountWriteFence();
1604	–	}
1605	–
1606	–	/**
1607	–	* Establishes local first-in-first-out scheduling mode for forked
1608	–	* tasks that are never joined. This mode may be more appropriate
1609	–	* than default locally stack-based mode in applications in which
1610	–	* worker threads only process asynchronous tasks. This method is
1611	–	* designed to be invoked only when the pool is quiescent, and
1612	–	* typically only before any tasks are submitted. The effects of
1613	–	* invocations at other times may be unpredictable.
1614	–	*
1615	–	* @param async if {@code true}, use locally FIFO scheduling
1616	–	* @return the previous mode
1617	–	* @see #getAsyncMode
1618	–	*/
1619	–	public boolean setAsyncMode(boolean async) {
1620	–	workerCountReadFence();
1621	–	boolean oldMode = locallyFifo;
1622	–	if (oldMode != async) {
1623	–	locallyFifo = async;
1624	–	workerCountWriteFence();
1625	–	ForkJoinWorkerThread[] ws = workers;
1626	–	int nws = ws.length;
1627	–	for (int i = 0; i < nws; ++i) {
1628	–	ForkJoinWorkerThread w = ws[i];
1629	–	if (w != null)
1630	–	w.setAsyncMode(async);
1631	–	}
1632	–	}
1633	–	return oldMode;
1634	–	}
1635	–
1636	–	/**
1463		* Returns {@code true} if this pool uses local first-in-first-out
1464		* scheduling mode for forked tasks that are never joined.
1465		*
1466		* @return {@code true} if this pool uses async mode
1641	–	* @see #setAsyncMode
1467		*/
1468		public boolean getAsyncMode() {
1644	–	workerCountReadFence();
1469		return locallyFifo;
1470		}
1471
#	Line 1710 \| Line 1534 \| public class ForkJoinPool extends Abstra
1534		*/
1535		public long getQueuedTaskCount() {
1536		long count = 0;
1537	<	ForkJoinWorkerThread[] ws = workers;
1714	<	int nws = ws.length;
1715	<	for (int i = 0; i < nws; ++i) {
1716	<	ForkJoinWorkerThread w = ws[i];
1537	>	for (ForkJoinWorkerThread w : workers)
1538		if (w != null)
1539		count += w.getQueueSize();
1719	–	}
1540		return count;
1541		}
1542
#	Line 1770 \| Line 1590 \| public class ForkJoinPool extends Abstra
1590		* @return the number of elements transferred
1591		*/
1592		protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1593	<	int n = submissionQueue.drainTo(c);
1594	<	ForkJoinWorkerThread[] ws = workers;
1775	<	int nws = ws.length;
1776	<	for (int i = 0; i < nws; ++i) {
1777	<	ForkJoinWorkerThread w = ws[i];
1593	>	int count = submissionQueue.drainTo(c);
1594	>	for (ForkJoinWorkerThread w : workers)
1595		if (w != null)
1596	<	n += w.drainTasksTo(c);
1597	<	}
1781	<	return n;
1596	>	count += w.drainTasksTo(c);
1597	>	return count;
1598		}
1599
1600		/**
#	Line 1902 \| Line 1718 \| public class ForkJoinPool extends Abstra
1718		*/
1719		public boolean awaitTermination(long timeout, TimeUnit unit)
1720		throws InterruptedException {
1721	<	return terminationLatch.await(timeout, unit);
1721	>	try {
1722	>	return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1723	>	} catch (TimeoutException ex) {
1724	>	return false;
1725	>	}
1726		}
1727
1728		/**
1729		* Interface for extending managed parallelism for tasks running
1730		* in {@link ForkJoinPool}s.
1731		*
1732	<	* <p>A {@code ManagedBlocker} provides two methods.
1733	<	* Method {@code isReleasable} must return {@code true} if
1734	<	* blocking is not necessary. Method {@code block} blocks the
1735	<	* current thread if necessary (perhaps internally invoking
1736	<	* {@code isReleasable} before actually blocking).
1732	>	* <p>A {@code ManagedBlocker} provides two methods. Method
1733	>	* {@code isReleasable} must return {@code true} if blocking is
1734	>	* not necessary. Method {@code block} blocks the current thread
1735	>	* if necessary (perhaps internally invoking {@code isReleasable}
1736	>	* before actually blocking). The unusual methods in this API
1737	>	* accommodate synchronizers that may, but don't usually, block
1738	>	* for long periods. Similarly, they allow more efficient internal
1739	>	* handling of cases in which additional workers may be, but
1740	>	* usually are not, needed to ensure sufficient parallelism.
1741	>	* Toward this end, implementations of method {@code isReleasable}
1742	>	* must be amenable to repeated invocation.
1743		*
1744		* <p>For example, here is a ManagedBlocker based on a
1745		* ReentrantLock:
#	Line 1931 \| Line 1757 \| public class ForkJoinPool extends Abstra
1757		* return hasLock \|\| (hasLock = lock.tryLock());
1758		* }
1759		* }}</pre>
1760	+	*
1761	+	* <p>Here is a class that possibly blocks waiting for an
1762	+	* item on a given queue:
1763	+	* <pre> {@code
1764	+	* class QueueTaker<E> implements ManagedBlocker {
1765	+	* final BlockingQueue<E> queue;
1766	+	* volatile E item = null;
1767	+	* QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1768	+	* public boolean block() throws InterruptedException {
1769	+	* if (item == null)
1770	+	* item = queue.take();
1771	+	* return true;
1772	+	* }
1773	+	* public boolean isReleasable() {
1774	+	* return item != null \|\| (item = queue.poll()) != null;
1775	+	* }
1776	+	* public E getItem() { // call after pool.managedBlock completes
1777	+	* return item;
1778	+	* }
1779	+	* }}</pre>
1780		*/
1781		public static interface ManagedBlocker {
1782		/**
#	Line 1954 \| Line 1800 \| public class ForkJoinPool extends Abstra
1800		* Blocks in accord with the given blocker. If the current thread
1801		* is a {@link ForkJoinWorkerThread}, this method possibly
1802		* arranges for a spare thread to be activated if necessary to
1803	<	* ensure parallelism while the current thread is blocked.
1958	<	*
1959	<	* <p>If {@code maintainParallelism} is {@code true} and the pool
1960	<	* supports it ({@link #getMaintainsParallelism}), this method
1961	<	* attempts to maintain the pool's nominal parallelism. Otherwise
1962	<	* it activates a thread only if necessary to avoid complete
1963	<	* starvation. This option may be preferable when blockages use
1964	<	* timeouts, or are almost always brief.
1803	>	* ensure sufficient parallelism while the current thread is blocked.
1804		*
1805		* <p>If the caller is not a {@link ForkJoinTask}, this method is
1806		* behaviorally equivalent to
#	Line 1975 \| Line 1814 \| public class ForkJoinPool extends Abstra
1814		* first be expanded to ensure parallelism, and later adjusted.
1815		*
1816		* @param blocker the blocker
1978	–	* @param maintainParallelism if {@code true} and supported by
1979	–	* this pool, attempt to maintain the pool's nominal parallelism;
1980	–	* otherwise activate a thread only if necessary to avoid
1981	–	* complete starvation.
1817		* @throws InterruptedException if blocker.block did so
1818		*/
1819	<	public static void managedBlock(ManagedBlocker blocker,
1985	<	boolean maintainParallelism)
1819	>	public static void managedBlock(ManagedBlocker blocker)
1820		throws InterruptedException {
1821		Thread t = Thread.currentThread();
1822	<	if (t instanceof ForkJoinWorkerThread)
1823	<	((ForkJoinWorkerThread) t).pool.
1824	<	awaitBlocker(blocker, maintainParallelism);
1825	<	else
1826	<	awaitBlocker(blocker);
1827	<	}
1828	<
1995	<	/**
1996	<	* Performs Non-FJ blocking
1997	<	*/
1998	<	private static void awaitBlocker(ManagedBlocker blocker)
1999	<	throws InterruptedException {
2000	<	do {} while (!blocker.isReleasable() && !blocker.block());
1822	>	if (t instanceof ForkJoinWorkerThread) {
1823	>	ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1824	>	w.pool.awaitBlocker(blocker);
1825	>	}
1826	>	else {
1827	>	do {} while (!blocker.isReleasable() && !blocker.block());
1828	>	}
1829		}
1830
1831		// AbstractExecutorService overrides. These rely on undocumented
#	Line 2022 \| Line 1850 \| public class ForkJoinPool extends Abstra
1850		private static final long eventCountOffset =
1851		objectFieldOffset("eventCount", ForkJoinPool.class);
1852		private static final long eventWaitersOffset =
1853	<	objectFieldOffset("eventWaiters",ForkJoinPool.class);
1853	>	objectFieldOffset("eventWaiters", ForkJoinPool.class);
1854		private static final long stealCountOffset =
1855	<	objectFieldOffset("stealCount",ForkJoinPool.class);
1856	<
1855	>	objectFieldOffset("stealCount", ForkJoinPool.class);
1856	>	private static final long spareWaitersOffset =
1857	>	objectFieldOffset("spareWaiters", ForkJoinPool.class);
1858
1859		private static long objectFieldOffset(String field, Class<?> klazz) {
1860		try {

Diff Legend

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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents): Revision 1.56 by dl, Thu May 27 16:46:48 2010 UTC vs. Revision 1.78 by dl, Tue Sep 7 14:52:24 2010 UTC

Diff Legend

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.56 by dl, Thu May 27 16:46:48 2010 UTC vs.
Revision 1.78 by dl, Tue Sep 7 14:52:24 2010 UTC