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 1.55 by dl, Sun Apr 18 13:59:57 2010 UTC vs.
Revision 1.107 by jsr166, Fri Jul 1 03:09:02 2011 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
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.Random;
15	+	import java.util.concurrent.AbstractExecutorService;
16	+	import java.util.concurrent.Callable;
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.atomic.AtomicInteger;
23		import java.util.concurrent.locks.LockSupport;
24		import java.util.concurrent.locks.ReentrantLock;
25	<	import java.util.concurrent.atomic.AtomicInteger;
19	<	import java.util.concurrent.CountDownLatch;
25	>	import java.util.concurrent.locks.Condition;
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 75 | Line 101 | import java.util.concurrent.CountDownLat
101		* daemon} mode, there is typically no need to explicitly {@link
102		* #shutdown} such a pool upon program exit.
103		*
104	<	* <pre>
104	>	*  <pre> {@code
105		* static final ForkJoinPool mainPool = new ForkJoinPool();
106		* ...
107		* public void sort(long[] array) {
108		*   mainPool.invoke(new SortTask(array, 0, array.length));
109	<	* }
84	<	* </pre>
109	>	* }}</pre>
110		*
111		* <p><b>Implementation notes</b>: This implementation restricts the
112		* maximum number of running threads to 32767. Attempts to create
#	Line 89 | Line 114 | import java.util.concurrent.CountDownLat
114		* {@code IllegalArgumentException}.
115		*
116		* <p>This implementation rejects submitted tasks (that is, by throwing
117	<	* {@link RejectedExecutionException}) only when the pool is shut down.
117	>	* {@link RejectedExecutionException}) only when the pool is shut down
118	>	* or internal resources have been exhausted.
119		*
120		* @since 1.7
121		* @author Doug Lea
#	Line 103 | Line 129 | public class ForkJoinPool extends Abstra
129		* set of worker threads: Submissions from non-FJ threads enter
130		* into a submission queue. Workers take these tasks and typically
131		* split them into subtasks that may be stolen by other workers.
132	<	* The main work-stealing mechanics implemented in class
133	<	* ForkJoinWorkerThread give first priority to processing tasks
134	<	* from their own queues (LIFO or FIFO, depending on mode), then
135	<	* to randomized FIFO steals of tasks in other worker queues, and
110	<	* lastly to new submissions. These mechanics do not consider
111	<	* affinities, loads, cache localities, etc, so rarely provide the
112	<	* best possible performance on a given machine, but portably
113	<	* provide good throughput by averaging over these factors.
114	<	* (Further, even if we did try to use such information, we do not
115	<	* usually have a basis for exploiting it. For example, some sets
116	<	* of tasks profit from cache affinities, but others are harmed by
117	<	* cache pollution effects.)
132	>	* Preference rules give first priority to processing tasks from
133	>	* their own queues (LIFO or FIFO, depending on mode), then to
134	>	* randomized FIFO steals of tasks in other worker queues, and
135	>	* lastly to new submissions.
136		*
137		* The main throughput advantages of work-stealing stem from
138	<	* decentralized control -- workers mostly steal tasks from each
139	<	* other. We do not want to negate this by creating bottlenecks
140	<	* implementing the management responsibilities of this class. So
141	<	* we use a collection of techniques that avoid, reduce, or cope
142	<	* well with contention. These entail several instances of
143	<	* bit-packing into CASable fields to maintain only the minimally
144	<	* required atomicity. To enable such packing, we restrict maximum
145	<	* parallelism to (1<<15)-1 (enabling twice this to fit into a 16
146	<	* bit field), which is far in excess of normal operating range.
147	<	* Even though updates to some of these bookkeeping fields do
148	<	* sometimes contend with each other, they don't normally
149	<	* cache-contend with updates to others enough to warrant memory
150	<	* padding or isolation. So they are all held as fields of
151	<	* ForkJoinPool objects.  The main capabilities are as follows:
152	<	*
153	<	* 1. Creating and removing workers. Workers are recorded in the
154	<	* "workers" array. This is an array as opposed to some other data
138	>	* decentralized control -- workers mostly take tasks from
139	>	* themselves or each other. We cannot negate this in the
140	>	* implementation of other management responsibilities. The main
141	>	* tactic for avoiding bottlenecks is packing nearly all
142	>	* essentially atomic control state into a single 64bit volatile
143	>	* variable ("ctl"). This variable is read on the order of 10-100
144	>	* times as often as it is modified (always via CAS). (There is
145	>	* some additional control state, for example variable "shutdown"
146	>	* for which we can cope with uncoordinated updates.)  This
147	>	* streamlines synchronization and control at the expense of messy
148	>	* constructions needed to repack status bits upon updates.
149	>	* Updates tend not to contend with each other except during
150	>	* bursts while submitted tasks begin or end.  In some cases when
151	>	* they do contend, threads can instead do something else
152	>	* (usually, scan for tasks) until contention subsides.
153	>	*
154	>	* To enable packing, we restrict maximum parallelism to (1<<15)-1
155	>	* (which is far in excess of normal operating range) to allow
156	>	* ids, counts, and their negations (used for thresholding) to fit
157	>	* into 16bit fields.
158	>	*
159	>	* Recording Workers.  Workers are recorded in the "workers" array
160	>	* that is created upon pool construction and expanded if (rarely)
161	>	* necessary.  This is an array as opposed to some other data
162		* structure to support index-based random steals by workers.
163		* Updates to the array recording new workers and unrecording
164	<	* terminated ones are protected from each other by a lock
165	<	* (workerLock) but the array is otherwise concurrently readable,
164	>	* terminated ones are protected from each other by a seqLock
165	>	* (scanGuard) but the array is otherwise concurrently readable,
166		* and accessed directly by workers. To simplify index-based
167		* operations, the array size is always a power of two, and all
168	<	* readers must tolerate null slots. Currently, all but the first
169	<	* worker thread creation is on-demand, triggered by task
170	<	* submissions, replacement of terminated workers, and/or
171	<	* compensation for blocked workers. However, all other support
172	<	* code is set up to work with other policies.
173	<	*
174	<	* 2. Bookkeeping for dynamically adding and removing workers. We
175	<	* maintain a given level of parallelism (or, if
176	<	* maintainsParallelism is false, at least avoid starvation). When
177	<	* some workers are known to be blocked (on joins or via
178	<	* ManagedBlocker), we may create or resume others to take their
179	<	* place until they unblock (see below). Implementing this
180	<	* requires counts of the number of "running" threads (i.e., those
181	<	* that are neither blocked nor artifically suspended) as well as
182	<	* the total number.  These two values are packed into one field,
183	<	* "workerCounts" because we need accurate snapshots when deciding
184	<	* to create, resume or suspend.  To support these decisions,
185	<	* updates must be prospective (not retrospective).  For example,
186	<	* the running count is decremented before blocking by a thread
187	<	* about to block, but incremented by the thread about to unblock
188	<	* it. (In a few cases, these prospective updates may need to be
189	<	* rolled back, for example when deciding to create a new worker
190	<	* but the thread factory fails or returns null. In these cases,
191	<	* we are no worse off wrt other decisions than we would be
192	<	* otherwise.)  Updates to the workerCounts field sometimes
193	<	* transiently encounter a fair amount of contention when join
194	<	* dependencies are such that many threads block or unblock at
195	<	* about the same time. We alleviate this by sometimes bundling
196	<	* updates (for example blocking one thread on join and resuming a
197	<	* spare cancel each other out), and in most other cases
198	<	* performing an alternative action (like releasing waiters and
199	<	* finding spares; see below) as a more productive form of
200	<	* backoff.
201	<	*
202	<	* 3. Maintaining global run state. The run state of the pool
203	<	* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
204	<	* those in other Executor implementations, as well as a count of
205	<	* "active" workers -- those that are, or soon will be, or
206	<	* recently were executing tasks. The runLevel and active count
207	<	* are packed together in order to correctly trigger shutdown and
208	<	* termination. Without care, active counts can be subject to very
209	<	* high contention.  We substantially reduce this contention by
210	<	* relaxing update rules.  A worker must claim active status
211	<	* prospectively, by activating if it sees that a submitted or
212	<	* stealable task exists (it may find after activating that the
213	<	* task no longer exists). It stays active while processing this
214	<	* task (if it exists) and any other local subtasks it produces,
215	<	* until it cannot find any other tasks. It then tries
216	<	* inactivating (see method preStep), but upon update contention
217	<	* instead scans for more tasks, later retrying inactivation if it
218	<	* doesn't find any.
219	<	*
220	<	* 4. Managing idle workers waiting for tasks. We cannot let
221	<	* workers spin indefinitely scanning for tasks when none are
222	<	* available. On the other hand, we must quickly prod them into
223	<	* action when new tasks are submitted or generated.  We
224	<	* park/unpark these idle workers using an event-count scheme.
225	<	* Field eventCount is incremented upon events that may enable
226	<	* workers that previously could not find a task to now find one:
227	<	* Submission of a new task to the pool, or another worker pushing
228	<	* a task onto a previously empty queue.  (We also use this
229	<	* mechanism for termination and reconfiguration actions that
230	<	* require wakeups of idle workers).  Each worker maintains its
231	<	* last known event count, and blocks when a scan for work did not
232	<	* find a task AND its lastEventCount matches the current
233	<	* eventCount. Waiting idle workers are recorded in a variant of
234	<	* Treiber stack headed by field eventWaiters which, when nonzero,
235	<	* encodes the thread index and count awaited for by the worker
236	<	* thread most recently calling eventSync. This thread in turn has
237	<	* a record (field nextEventWaiter) for the next waiting worker.
238	<	* In addition to allowing simpler decisions about need for
239	<	* wakeup, the event count bits in eventWaiters serve the role of
240	<	* tags to avoid ABA errors in Treiber stacks.  To reduce delays
241	<	* in task diffusion, workers not otherwise occupied may invoke
242	<	* method releaseWaiters, that removes and signals (unparks)
243	<	* workers not waiting on current count. To minimize task
244	<	* production stalls associate with signalling, any worker pushing
245	<	* a task on an empty queue invokes the weaker method signalWork,
246	<	* that only releases idle workers until it detects interference
247	<	* by other threads trying to release, and lets them take
248	<	* over. The net effect is a tree-like diffusion of signals, where
249	<	* released threads and possibly others) help with unparks.  To
250	<	* further reduce contention effects a bit, failed CASes to
251	<	* increment field eventCount are tolerated without retries.
252	<	* Conceptually they are merged into the same event, which is OK
253	<	* when their only purpose is to enable workers to scan for work.
254	<	*
255	<	* 5. Managing suspension of extra workers. When a worker is about
256	<	* to block waiting for a join (or via ManagedBlockers), we may
257	<	* create a new thread to maintain parallelism level, or at least
258	<	* avoid starvation (see below). Usually, extra threads are needed
259	<	* for only very short periods, yet join dependencies are such
260	<	* that we sometimes need them in bursts. Rather than create new
261	<	* threads each time this happens, we suspend no-longer-needed
262	<	* extra ones as "spares". For most purposes, we don't distinguish
263	<	* "extra" spare threads from normal "core" threads: On each call
264	<	* to preStep (the only point at which we can do this) a worker
265	<	* checks to see if there are now too many running workers, and if
266	<	* so, suspends itself.  Methods preJoin and doBlock look for
267	<	* suspended threads to resume before considering creating a new
268	<	* replacement. We don't need a special data structure to maintain
269	<	* spares; simply scanning the workers array looking for
270	<	* worker.isSuspended() is fine because the calling thread is
271	<	* otherwise not doing anything useful anyway; we are at least as
272	<	* happy if after locating a spare, the caller doesn't actually
273	<	* block because the join is ready before we try to adjust and
274	<	* compensate.  Note that this is intrinsically racy.  One thread
275	<	* may become a spare at about the same time as another is
276	<	* needlessly being created. We counteract this and related slop
277	<	* in part by requiring resumed spares to immediately recheck (in
278	<	* preStep) to see whether they they should re-suspend. The only
279	<	* effective difference between "extra" and "core" threads is that
280	<	* we allow the "extra" ones to time out and die if they are not
281	<	* resumed within a keep-alive interval of a few seconds. This is
282	<	* implemented mainly within ForkJoinWorkerThread, but requires
283	<	* some coordination (isTrimmed() -- meaning killed while
284	<	* suspended) to correctly maintain pool counts.
285	<	*
286	<	* 6. Deciding when to create new workers. The main dynamic
287	<	* control in this class is deciding when to create extra threads,
288	<	* in methods preJoin and doBlock. We always need to create one
289	<	* when the number of running threads becomes zero. But because
290	<	* blocked joins are typically dependent, we don't necessarily
291	<	* need or want one-to-one replacement. Using a one-to-one
292	<	* compensation rule often leads to enough useless overhead
293	<	* creating, suspending, resuming, and/or killing threads to
294	<	* signficantly degrade throughput.  We use a rule reflecting the
295	<	* idea that, the more spare threads you already have, the more
296	<	* evidence you need to create another one; where "evidence" is
297	<	* expressed as the current deficit -- target minus running
273	<	* threads. To reduce flickering and drift around target values,
274	<	* the relation is quadratic: adding a spare if (dc*dc)>=(sc*pc)
275	<	* (where dc is deficit, sc is number of spare threads and pc is
276	<	* target parallelism.)  This effectively reduces churn at the
277	<	* price of systematically undershooting target parallelism when
278	<	* many threads are blocked.  However, biasing toward undeshooting
279	<	* partially compensates for the above mechanics to suspend extra
280	<	* threads, that normally lead to overshoot because we can only
281	<	* suspend workers in-between top-level actions. It also better
282	<	* copes with the fact that some of the methods in this class tend
283	<	* to never become compiled (but are interpreted), so some
284	<	* components of the entire set of controls might execute many
285	<	* times faster than others. And similarly for cases where the
286	<	* apparent lack of work is just due to GC stalls and other
287	<	* transient system activity.
288	<	*
289	<	* 7. Maintaining other configuration parameters and monitoring
290	<	* statistics. Updates to fields controlling parallelism level,
291	<	* max size, etc can only meaningfully take effect for individual
292	<	* threads upon their next top-level actions; i.e., between
293	<	* stealing/running tasks/submission, which are separated by calls
294	<	* to preStep.  Memory ordering for these (assumed infrequent)
295	<	* reconfiguration calls is ensured by using reads and writes to
296	<	* volatile field workerCounts (that must be read in preStep anyway)
297	<	* as "fences" -- user-level reads are preceded by reads of
298	<	* workCounts, and writes are followed by no-op CAS to
299	<	* workerCounts. The values reported by other management and
300	<	* monitoring methods are either computed on demand, or are kept
301	<	* in fields that are only updated when threads are otherwise
302	<	* idle.
168	>	* readers must tolerate null slots. To avoid flailing during
169	>	* start-up, the array is presized to hold twice #parallelism
170	>	* workers (which is unlikely to need further resizing during
171	>	* execution). But to avoid dealing with so many null slots,
172	>	* variable scanGuard includes a mask for the nearest power of two
173	>	* that contains all current workers.  All worker thread creation
174	>	* is on-demand, triggered by task submissions, replacement of
175	>	* terminated workers, and/or compensation for blocked
176	>	* workers. However, all other support code is set up to work with
177	>	* other policies.  To ensure that we do not hold on to worker
178	>	* references that would prevent GC, ALL accesses to workers are
179	>	* via indices into the workers array (which is one source of some
180	>	* of the messy code constructions here). In essence, the workers
181	>	* array serves as a weak reference mechanism. Thus for example
182	>	* the wait queue field of ctl stores worker indices, not worker
183	>	* references.  Access to the workers in associated methods (for
184	>	* example signalWork) must both index-check and null-check the
185	>	* IDs. All such accesses ignore bad IDs by returning out early
186	>	* from what they are doing, since this can only be associated
187	>	* with termination, in which case it is OK to give up.
188	>	*
189	>	* All uses of the workers array, as well as queue arrays, check
190	>	* that the array is non-null (even if previously non-null). This
191	>	* allows nulling during termination, which is currently not
192	>	* necessary, but remains an option for resource-revocation-based
193	>	* shutdown schemes.
194	>	*
195	>	* Wait Queuing. Unlike HPC work-stealing frameworks, we cannot
196	>	* let workers spin indefinitely scanning for tasks when none can
197	>	* be found immediately, and we cannot start/resume workers unless
198	>	* there appear to be tasks available.  On the other hand, we must
199	>	* quickly prod them into action when new tasks are submitted or
200	>	* generated.  We park/unpark workers after placing in an event
201	>	* wait queue when they cannot find work. This "queue" is actually
202	>	* a simple Treiber stack, headed by the "id" field of ctl, plus a
203	>	* 15bit counter value to both wake up waiters (by advancing their
204	>	* count) and avoid ABA effects. Successors are held in worker
205	>	* field "nextWait".  Queuing deals with several intrinsic races,
206	>	* mainly that a task-producing thread can miss seeing (and
207	>	* signalling) another thread that gave up looking for work but
208	>	* has not yet entered the wait queue. We solve this by requiring
209	>	* a full sweep of all workers both before (in scan()) and after
210	>	* (in tryAwaitWork()) a newly waiting worker is added to the wait
211	>	* queue. During a rescan, the worker might release some other
212	>	* queued worker rather than itself, which has the same net
213	>	* effect. Because enqueued workers may actually be rescanning
214	>	* rather than waiting, we set and clear the "parked" field of
215	>	* ForkJoinWorkerThread to reduce unnecessary calls to unpark.
216	>	* (Use of the parked field requires a secondary recheck to avoid
217	>	* missed signals.)
218	>	*
219	>	* Signalling.  We create or wake up workers only when there
220	>	* appears to be at least one task they might be able to find and
221	>	* execute.  When a submission is added or another worker adds a
222	>	* task to a queue that previously had two or fewer tasks, they
223	>	* signal waiting workers (or trigger creation of new ones if
224	>	* fewer than the given parallelism level -- see signalWork).
225	>	* These primary signals are buttressed by signals during rescans
226	>	* as well as those performed when a worker steals a task and
227	>	* notices that there are more tasks too; together these cover the
228	>	* signals needed in cases when more than two tasks are pushed
229	>	* but untaken.
230	>	*
231	>	* Trimming workers. To release resources after periods of lack of
232	>	* use, a worker starting to wait when the pool is quiescent will
233	>	* time out and terminate if the pool has remained quiescent for
234	>	* SHRINK_RATE nanosecs. This will slowly propagate, eventually
235	>	* terminating all workers after long periods of non-use.
236	>	*
237	>	* Submissions. External submissions are maintained in an
238	>	* array-based queue that is structured identically to
239	>	* ForkJoinWorkerThread queues except for the use of
240	>	* submissionLock in method addSubmission. Unlike the case for
241	>	* worker queues, multiple external threads can add new
242	>	* submissions, so adding requires a lock.
243	>	*
244	>	* Compensation. Beyond work-stealing support and lifecycle
245	>	* control, the main responsibility of this framework is to take
246	>	* actions when one worker is waiting to join a task stolen (or
247	>	* always held by) another.  Because we are multiplexing many
248	>	* tasks on to a pool of workers, we can't just let them block (as
249	>	* in Thread.join).  We also cannot just reassign the joiner's
250	>	* run-time stack with another and replace it later, which would
251	>	* be a form of "continuation", that even if possible is not
252	>	* necessarily a good idea since we sometimes need both an
253	>	* unblocked task and its continuation to progress. Instead we
254	>	* combine two tactics:
255	>	*
256	>	*   Helping: Arranging for the joiner to execute some task that it
257	>	*      would be running if the steal had not occurred.  Method
258	>	*      ForkJoinWorkerThread.joinTask tracks joining->stealing
259	>	*      links to try to find such a task.
260	>	*
261	>	*   Compensating: Unless there are already enough live threads,
262	>	*      method tryPreBlock() may create or re-activate a spare
263	>	*      thread to compensate for blocked joiners until they
264	>	*      unblock.
265	>	*
266	>	* The ManagedBlocker extension API can't use helping so relies
267	>	* only on compensation in method awaitBlocker.
268	>	*
269	>	* It is impossible to keep exactly the target parallelism number
270	>	* of threads running at any given time.  Determining the
271	>	* existence of conservatively safe helping targets, the
272	>	* availability of already-created spares, and the apparent need
273	>	* to create new spares are all racy and require heuristic
274	>	* guidance, so we rely on multiple retries of each.  Currently,
275	>	* in keeping with on-demand signalling policy, we compensate only
276	>	* if blocking would leave less than one active (non-waiting,
277	>	* non-blocked) worker. Additionally, to avoid some false alarms
278	>	* due to GC, lagging counters, system activity, etc, compensated
279	>	* blocking for joins is only attempted after rechecks stabilize
280	>	* (retries are interspersed with Thread.yield, for good
281	>	* citizenship).  The variable blockedCount, incremented before
282	>	* blocking and decremented after, is sometimes needed to
283	>	* distinguish cases of waiting for work vs blocking on joins or
284	>	* other managed sync. Both cases are equivalent for most pool
285	>	* control, so we can update non-atomically. (Additionally,
286	>	* contention on blockedCount alleviates some contention on ctl).
287	>	*
288	>	* Shutdown and Termination. A call to shutdownNow atomically sets
289	>	* the ctl stop bit and then (non-atomically) sets each workers
290	>	* "terminate" status, cancels all unprocessed tasks, and wakes up
291	>	* all waiting workers.  Detecting whether termination should
292	>	* commence after a non-abrupt shutdown() call requires more work
293	>	* and bookkeeping. We need consensus about quiescence (i.e., that
294	>	* there is no more work) which is reflected in active counts so
295	>	* long as there are no current blockers, as well as possible
296	>	* re-evaluations during independent changes in blocking or
297	>	* quiescing workers.
298		*
299	<	* Beware that there is a lot of representation-level coupling
299	>	* Style notes: There is a lot of representation-level coupling
300		* among classes ForkJoinPool, ForkJoinWorkerThread, and
301	<	* ForkJoinTask.  For example, direct access to "workers" array by
301	>	* ForkJoinTask.  Most fields of ForkJoinWorkerThread maintain
302	>	* data structures managed by ForkJoinPool, so are directly
303	>	* accessed.  Conversely we allow access to "workers" array by
304		* workers, and direct access to ForkJoinTask.status by both
305		* ForkJoinPool and ForkJoinWorkerThread.  There is little point
306		* trying to reduce this, since any associated future changes in
307		* representations will need to be accompanied by algorithmic
308	<	* changes anyway.
309	<	*
310	<	* Style notes: There are lots of inline assignments (of form
311	<	* "while ((local = field) != 0)") which are usually the simplest
312	<	* way to ensure read orderings. Also several occurrences of the
313	<	* unusual "do {} while(!cas...)" which is the simplest way to
314	<	* force an update of a CAS'ed variable. There are also a few
315	<	* other coding oddities that help some methods perform reasonably
316	<	* even when interpreted (not compiled).
317	<	*
318	<	* The order of declarations in this file is: (1) statics (2)
319	<	* fields (along with constants used when unpacking some of them)
320	<	* (3) internal control methods (4) callbacks and other support
321	<	* for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
322	<	* methods (plus a few little helpers).
308	>	* changes anyway. All together, these low-level implementation
309	>	* choices produce as much as a factor of 4 performance
310	>	* improvement compared to naive implementations, and enable the
311	>	* processing of billions of tasks per second, at the expense of
312	>	* some ugliness.
313	>	*
314	>	* Methods signalWork() and scan() are the main bottlenecks so are
315	>	* especially heavily micro-optimized/mangled.  There are lots of
316	>	* inline assignments (of form "while ((local = field) != 0)")
317	>	* which are usually the simplest way to ensure the required read
318	>	* orderings (which are sometimes critical). This leads to a
319	>	* "C"-like style of listing declarations of these locals at the
320	>	* heads of methods or blocks.  There are several occurrences of
321	>	* the unusual "do {} while (!cas...)"  which is the simplest way
322	>	* to force an update of a CAS'ed variable. There are also other
323	>	* coding oddities that help some methods perform reasonably even
324	>	* when interpreted (not compiled).
325	>	*
326	>	* The order of declarations in this file is: (1) declarations of
327	>	* statics (2) fields (along with constants used when unpacking
328	>	* some of them), listed in an order that tends to reduce
329	>	* contention among them a bit under most JVMs.  (3) internal
330	>	* control methods (4) callbacks and other support for
331	>	* ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
332	>	* methods (plus a few little helpers). (6) static block
333	>	* initializing all statics in a minimally dependent order.
334		*/
335
336		/**
#	Line 345 | Line 353 | public class ForkJoinPool extends Abstra
353		* Default ForkJoinWorkerThreadFactory implementation; creates a
354		* new ForkJoinWorkerThread.
355		*/
356	<	static class  DefaultForkJoinWorkerThreadFactory
356	>	static class DefaultForkJoinWorkerThreadFactory
357		implements ForkJoinWorkerThreadFactory {
358		public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
359		return new ForkJoinWorkerThread(pool);
#	Line 357 | Line 365 | public class ForkJoinPool extends Abstra
365		* overridden in ForkJoinPool constructors.
366		*/
367		public static final ForkJoinWorkerThreadFactory
368	<	defaultForkJoinWorkerThreadFactory =
361	<	new DefaultForkJoinWorkerThreadFactory();
368	>	defaultForkJoinWorkerThreadFactory;
369
370		/**
371		* Permission required for callers of methods that may start or
372		* kill threads.
373		*/
374	<	private static final RuntimePermission modifyThreadPermission =
368	<	new RuntimePermission("modifyThread");
374	>	private static final RuntimePermission modifyThreadPermission;
375
376		/**
377		* If there is a security manager, makes sure caller has
#	Line 380 | Line 386 | public class ForkJoinPool extends Abstra
386		/**
387		* Generator for assigning sequence numbers as pool names.
388		*/
389	<	private static final AtomicInteger poolNumberGenerator =
390	<	new AtomicInteger();
389	>	private static final AtomicInteger poolNumberGenerator;
390	>
391	>	/**
392	>	* Generator for initial random seeds for worker victim
393	>	* selection. This is used only to create initial seeds. Random
394	>	* steals use a cheaper xorshift generator per steal attempt. We
395	>	* don't expect much contention on seedGenerator, so just use a
396	>	* plain Random.
397	>	*/
398	>	static final Random workerSeedGenerator;
399
400		/**
401	<	* Absolute bound for parallelism level. Twice this number must
402	<	* fit into a 16bit field to enable word-packing for some counts.
401	>	* Array holding all worker threads in the pool.  Initialized upon
402	>	* construction. Array size must be a power of two.  Updates and
403	>	* replacements are protected by scanGuard, but the array is
404	>	* always kept in a consistent enough state to be randomly
405	>	* accessed without locking by workers performing work-stealing,
406	>	* as well as other traversal-based methods in this class, so long
407	>	* as reads memory-acquire by first reading ctl. All readers must
408	>	* tolerate that some array slots may be null.
409		*/
410	<	private static final int MAX_THREADS = 0x7fff;
410	>	ForkJoinWorkerThread[] workers;
411
412		/**
413	<	* Array holding all worker threads in the pool.  Array size must
414	<	* be a power of two.  Updates and replacements are protected by
415	<	* workerLock, but the array is always kept in a consistent enough
396	<	* state to be randomly accessed without locking by workers
397	<	* performing work-stealing, as well as other traversal-based
398	<	* methods in this class. All readers must tolerate that some
399	<	* array slots may be null.
413	>	* Initial size for submission queue array. Must be a power of
414	>	* two.  In many applications, these always stay small so we use a
415	>	* small initial cap.
416		*/
417	<	volatile ForkJoinWorkerThread[] workers;
417	>	private static final int INITIAL_QUEUE_CAPACITY = 8;
418
419		/**
420	<	* Queue for external submissions.
420	>	* Maximum size for submission queue array. Must be a power of two
421	>	* less than or equal to 1 << (31 - width of array entry) to
422	>	* ensure lack of index wraparound, but is capped at a lower
423	>	* value to help users trap runaway computations.
424		*/
425	<	private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
425	>	private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 24; // 16M
426
427		/**
428	<	* Lock protecting updates to workers array.
428	>	* Array serving as submission queue. Initialized upon construction.
429		*/
430	<	private final ReentrantLock workerLock;
430	>	private ForkJoinTask<?>[] submissionQueue;
431
432		/**
433	<	* Latch released upon termination.
433	>	* Lock protecting submissions array for addSubmission
434		*/
435	<	private final CountDownLatch terminationLatch;
435	>	private final ReentrantLock submissionLock;
436	>
437	>	/**
438	>	* Condition for awaitTermination, using submissionLock for
439	>	* convenience.
440	>	*/
441	>	private final Condition termination;
442
443		/**
444		* Creation factory for worker threads.
#	Line 421 | Line 446 | public class ForkJoinPool extends Abstra
446		private final ForkJoinWorkerThreadFactory factory;
447
448		/**
449	+	* The uncaught exception handler used when any worker abruptly
450	+	* terminates.
451	+	*/
452	+	final Thread.UncaughtExceptionHandler ueh;
453	+
454	+	/**
455	+	* Prefix for assigning names to worker threads
456	+	*/
457	+	private final String workerNamePrefix;
458	+
459	+	/**
460		* Sum of per-thread steal counts, updated only when threads are
461		* idle or terminating.
462		*/
463		private volatile long stealCount;
464
465		/**
466	<	* Encoded record of top of treiber stack of threads waiting for
467	<	* events. The top 32 bits contain the count being waited for. The
468	<	* bottom word contains one plus the pool index of waiting worker
469	<	* thread.
470	<	*/
471	<	private volatile long eventWaiters;
472	<
473	<	private static final int  EVENT_COUNT_SHIFT = 32;
474	<	private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
475	<
476	<	/**
477	<	* A counter for events that may wake up worker threads:
478	<	*   - Submission of a new task to the pool
479	<	*   - A worker pushing a task on an empty queue
480	<	*   - termination and reconfiguration
481	<	*/
482	<	private volatile int eventCount;
483	<
484	<	/**
485	<	* Lifecycle control. The low word contains the number of workers
486	<	* that are (probably) executing tasks. This value is atomically
487	<	* incremented before a worker gets a task to run, and decremented
488	<	* when worker has no tasks and cannot find any.  Bits 16-18
489	<	* contain runLevel value. When all are zero, the pool is
490	<	* running. Level transitions are monotonic (running -> shutdown
491	<	* -> terminating -> terminated) so each transition adds a bit.
492	<	* These are bundled together to ensure consistent read for
493	<	* termination checks (i.e., that runLevel is at least SHUTDOWN
494	<	* and active threads is zero).
495	<	*/
496	<	private volatile int runState;
497	<
498	<	// Note: The order among run level values matters.
499	<	private static final int RUNLEVEL_SHIFT     = 16;
500	<	private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
501	<	private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
502	<	private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
503	<	private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
504	<	private static final int ONE_ACTIVE         = 1; // active update delta
505	<
506	<	/**
507	<	* Holds number of total (i.e., created and not yet terminated)
508	<	* and running (i.e., not blocked on joins or other managed sync)
509	<	* threads, packed together to ensure consistent snapshot when
510	<	* making decisions about creating and suspending spare
511	<	* threads. Updated only by CAS. Note that adding a new worker
512	<	* requires incrementing both counts, since workers start off in
513	<	* running state.  This field is also used for memory-fencing
514	<	* configuration parameters.
515	<	*/
516	<	private volatile int workerCounts;
517	<
518	<	private static final int TOTAL_COUNT_SHIFT  = 16;
519	<	private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
484	<	private static final int ONE_RUNNING        = 1;
485	<	private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
486	<
487	<	/*
488	<	* Fields parallelism. maxPoolSize, locallyFifo,
489	<	* maintainsParallelism, and ueh are non-volatile, but external
490	<	* reads/writes use workerCount fences to ensure visability.
491	<	*/
466	>	* Main pool control -- a long packed with:
467	>	* AC: Number of active running workers minus target parallelism (16 bits)
468	>	* TC: Number of total workers minus target parallelism (16 bits)
469	>	* ST: true if pool is terminating (1 bit)
470	>	* EC: the wait count of top waiting thread (15 bits)
471	>	* ID: ~poolIndex of top of Treiber stack of waiting threads (16 bits)
472	>	*
473	>	* When convenient, we can extract the upper 32 bits of counts and
474	>	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
475	>	* (int)ctl.  The ec field is never accessed alone, but always
476	>	* together with id and st. The offsets of counts by the target
477	>	* parallelism and the positionings of fields makes it possible to
478	>	* perform the most common checks via sign tests of fields: When
479	>	* ac is negative, there are not enough active workers, when tc is
480	>	* negative, there are not enough total workers, when id is
481	>	* negative, there is at least one waiting worker, and when e is
482	>	* negative, the pool is terminating.  To deal with these possibly
483	>	* negative fields, we use casts in and out of "short" and/or
484	>	* signed shifts to maintain signedness.
485	>	*/
486	>	volatile long ctl;
487	>
488	>	// bit positions/shifts for fields
489	>	private static final int  AC_SHIFT   = 48;
490	>	private static final int  TC_SHIFT   = 32;
491	>	private static final int  ST_SHIFT   = 31;
492	>	private static final int  EC_SHIFT   = 16;
493	>
494	>	// bounds
495	>	private static final int  MAX_ID     = 0x7fff;  // max poolIndex
496	>	private static final int  SMASK      = 0xffff;  // mask short bits
497	>	private static final int  SHORT_SIGN = 1 << 15;
498	>	private static final int  INT_SIGN   = 1 << 31;
499	>
500	>	// masks
501	>	private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
502	>	private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
503	>	private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
504	>
505	>	// units for incrementing and decrementing
506	>	private static final long TC_UNIT    = 1L << TC_SHIFT;
507	>	private static final long AC_UNIT    = 1L << AC_SHIFT;
508	>
509	>	// masks and units for dealing with u = (int)(ctl >>> 32)
510	>	private static final int  UAC_SHIFT  = AC_SHIFT - 32;
511	>	private static final int  UTC_SHIFT  = TC_SHIFT - 32;
512	>	private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
513	>	private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
514	>	private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
515	>	private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
516	>
517	>	// masks and units for dealing with e = (int)ctl
518	>	private static final int  E_MASK     = 0x7fffffff; // no STOP_BIT
519	>	private static final int  EC_UNIT    = 1 << EC_SHIFT;
520
521		/**
522		* The target parallelism level.
523		*/
524	<	private int parallelism;
524	>	final int parallelism;
525
526		/**
527	<	* The maximum allowed pool size.
527	>	* Index (mod submission queue length) of next element to take
528	>	* from submission queue. Usage is identical to that for
529	>	* per-worker queues -- see ForkJoinWorkerThread internal
530	>	* documentation.
531		*/
532	<	private int maxPoolSize;
532	>	volatile int queueBase;
533
534		/**
535	<	* True if use local fifo, not default lifo, for local polling
536	<	* Replicated by ForkJoinWorkerThreads
535	>	* Index (mod submission queue length) of next element to add
536	>	* in submission queue. Usage is identical to that for
537	>	* per-worker queues -- see ForkJoinWorkerThread internal
538	>	* documentation.
539		*/
540	<	private boolean locallyFifo;
540	>	int queueTop;
541
542		/**
543	<	* Controls whether to add spares to maintain parallelism
543	>	* True when shutdown() has been called.
544		*/
545	<	private boolean maintainsParallelism;
545	>	volatile boolean shutdown;
546
547		/**
548	<	* The uncaught exception handler used when any worker
549	<	* abruptly terminates
548	>	* True if use local fifo, not default lifo, for local polling.
549	>	* Read by, and replicated by ForkJoinWorkerThreads.
550		*/
551	<	private Thread.UncaughtExceptionHandler ueh;
551	>	final boolean locallyFifo;
552
553		/**
554	<	* Pool number, just for assigning useful names to worker threads
554	>	* The number of threads in ForkJoinWorkerThreads.helpQuiescePool.
555	>	* When non-zero, suppresses automatic shutdown when active
556	>	* counts become zero.
557		*/
558	<	private final int poolNumber;
524	<
525	<	// utilities for updating fields
558	>	volatile int quiescerCount;
559
560		/**
561	<	* Adds delta to running count.  Used mainly by ForkJoinTask.
529	<	*
530	<	* @param delta the number to add
561	>	* The number of threads blocked in join.
562		*/
563	<	final void updateRunningCount(int delta) {
533	<	int wc;
534	<	do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
535	<	wc = workerCounts,
536	<	wc + delta));
537	<	}
563	>	volatile int blockedCount;
564
565		/**
566	<	* Write fence for user modifications of pool parameters
541	<	* (parallelism. etc).  Note that it doesn't matter if CAS fails.
566	>	* Counter for worker Thread names (unrelated to their poolIndex)
567		*/
568	<	private void workerCountWriteFence() {
544	<	int wc;
545	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
546	<	wc = workerCounts, wc);
547	<	}
568	>	private volatile int nextWorkerNumber;
569
570		/**
571	<	* Read fence for external reads of pool parameters
551	<	* (parallelism. maxPoolSize, etc).
571	>	* The index for the next created worker. Accessed under scanGuard.
572		*/
573	<	private void workerCountReadFence() {
554	<	int ignore = workerCounts;
555	<	}
573	>	private int nextWorkerIndex;
574
575		/**
576	<	* Tries incrementing active count; fails on contention.
577	<	* Called by workers before executing tasks.
578	<	*
579	<	* @return true on success
576	>	* SeqLock and index masking for updates to workers array.  Locked
577	>	* when SG_UNIT is set. Unlocking clears bit by adding
578	>	* SG_UNIT. Staleness of read-only operations can be checked by
579	>	* comparing scanGuard to value before the reads. The low 16 bits
580	>	* (i.e, anding with SMASK) hold (the smallest power of two
581	>	* covering all worker indices, minus one, and is used to avoid
582	>	* dealing with large numbers of null slots when the workers array
583	>	* is overallocated.
584		*/
585	<	final boolean tryIncrementActiveCount() {
586	<	int c;
587	<	return UNSAFE.compareAndSwapInt(this, runStateOffset,
566	<	c = runState, c + ONE_ACTIVE);
567	<	}
585	>	volatile int scanGuard;
586	>
587	>	private static final int SG_UNIT = 1 << 16;
588
589		/**
590	<	* Tries decrementing active count; fails on contention.
591	<	* Called when workers cannot find tasks to run.
590	>	* The wakeup interval (in nanoseconds) for a worker waiting for a
591	>	* task when the pool is quiescent to instead try to shrink the
592	>	* number of workers.  The exact value does not matter too
593	>	* much. It must be short enough to release resources during
594	>	* sustained periods of idleness, but not so short that threads
595	>	* are continually re-created.
596		*/
597	<	final boolean tryDecrementActiveCount() {
598	<	int c;
575	<	return UNSAFE.compareAndSwapInt(this, runStateOffset,
576	<	c = runState, c - ONE_ACTIVE);
577	<	}
597	>	private static final long SHRINK_RATE =
598	>	4L * 1000L * 1000L * 1000L; // 4 seconds
599
600		/**
601	<	* Advances to at least the given level. Returns true if not
602	<	* already in at least the given level.
601	>	* Top-level loop for worker threads: On each step: if the
602	>	* previous step swept through all queues and found no tasks, or
603	>	* there are excess threads, then possibly blocks. Otherwise,
604	>	* scans for and, if found, executes a task. Returns when pool
605	>	* and/or worker terminate.
606	>	*
607	>	* @param w the worker
608		*/
609	<	private boolean advanceRunLevel(int level) {
610	<	for (;;) {
611	<	int s = runState;
612	<	if ((s & level) != 0)
613	<	return false;
614	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
615	<	return true;
609	>	final void work(ForkJoinWorkerThread w) {
610	>	boolean swept = false;                // true on empty scans
611	>	long c;
612	>	while (!w.terminate && (int)(c = ctl) >= 0) {
613	>	int a;                            // active count
614	>	if (!swept && (a = (int)(c >> AC_SHIFT)) <= 0)
615	>	swept = scan(w, a);
616	>	else if (tryAwaitWork(w, c))
617	>	swept = false;
618		}
619		}
620
621	<	// workers array maintenance
621	>	// Signalling
622
623		/**
624	<	* Records and returns a workers array index for new worker.
624	>	* Wakes up or creates a worker.
625		*/
626	<	private int recordWorker(ForkJoinWorkerThread w) {
627	<	// Try using slot totalCount-1. If not available, scan and/or resize
628	<	int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
629	<	final ReentrantLock lock = this.workerLock;
630	<	lock.lock();
631	<	try {
632	<	ForkJoinWorkerThread[] ws = workers;
633	<	int len = ws.length;
634	<	if (k < 0 || k >= len || ws[k] != null) {
635	<	for (k = 0; k < len && ws[k] != null; ++k)
636	<	;
637	<	if (k == len)
638	<	ws = Arrays.copyOf(ws, len << 1);
626	>	final void signalWork() {
627	>	/*
628	>	* The while condition is true if: (there is are too few total
629	>	* workers OR there is at least one waiter) AND (there are too
630	>	* few active workers OR the pool is terminating).  The value
631	>	* of e distinguishes the remaining cases: zero (no waiters)
632	>	* for create, negative if terminating (in which case do
633	>	* nothing), else release a waiter. The secondary checks for
634	>	* release (non-null array etc) can fail if the pool begins
635	>	* terminating after the test, and don't impose any added cost
636	>	* because JVMs must perform null and bounds checks anyway.
637	>	*/
638	>	long c; int e, u;
639	>	while ((((e = (int)(c = ctl)) | (u = (int)(c >>> 32))) &
640	>	(INT_SIGN|SHORT_SIGN)) == (INT_SIGN|SHORT_SIGN) && e >= 0) {
641	>	if (e > 0) {                         // release a waiting worker
642	>	int i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws;
643	>	if ((ws = workers) == null ||
644	>	(i = ~e & SMASK) >= ws.length ||
645	>	(w = ws[i]) == null)
646	>	break;
647	>	long nc = (((long)(w.nextWait & E_MASK)) |
648	>	((long)(u + UAC_UNIT) << 32));
649	>	if (w.eventCount == e &&
650	>	UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
651	>	w.eventCount = (e + EC_UNIT) & E_MASK;
652	>	if (w.parked)
653	>	UNSAFE.unpark(w);
654	>	break;
655	>	}
656	>	}
657	>	else if (UNSAFE.compareAndSwapLong
658	>	(this, ctlOffset, c,
659	>	(long)(((u + UTC_UNIT) & UTC_MASK) |
660	>	((u + UAC_UNIT) & UAC_MASK)) << 32)) {
661	>	addWorker();
662	>	break;
663		}
612	–	ws[k] = w;
613	–	workers = ws; // volatile array write ensures slot visibility
614	–	} finally {
615	–	lock.unlock();
664		}
617	–	return k;
665		}
666
667		/**
668	<	* Nulls out record of worker in workers array
668	>	* Variant of signalWork to help release waiters on rescans.
669	>	* Tries once to release a waiter if active count < 0.
670	>	*
671	>	* @return false if failed due to contention, else true
672		*/
673	<	private void forgetWorker(ForkJoinWorkerThread w) {
674	<	int idx = w.poolIndex;
675	<	// Locking helps method recordWorker avoid unecessary expansion
676	<	final ReentrantLock lock = this.workerLock;
677	<	lock.lock();
678	<	try {
679	<	ForkJoinWorkerThread[] ws = workers;
680	<	if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
681	<	ws[idx] = null;
682	<	} finally {
683	<	lock.unlock();
673	>	private boolean tryReleaseWaiter() {
674	>	long c; int e, i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws;
675	>	if ((e = (int)(c = ctl)) > 0 &&
676	>	(int)(c >> AC_SHIFT) < 0 &&
677	>	(ws = workers) != null &&
678	>	(i = ~e & SMASK) < ws.length &&
679	>	(w = ws[i]) != null) {
680	>	long nc = ((long)(w.nextWait & E_MASK) |
681	>	((c + AC_UNIT) & (AC_MASK|TC_MASK)));
682	>	if (w.eventCount != e ||
683	>	!UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc))
684	>	return false;
685	>	w.eventCount = (e + EC_UNIT) & E_MASK;
686	>	if (w.parked)
687	>	UNSAFE.unpark(w);
688		}
689	+	return true;
690		}
691
692	<	// adding and removing workers
692	>	// Scanning for tasks
693
694		/**
695	<	* Tries to create and add new worker. Assumes that worker counts
696	<	* are already updated to accommodate the worker, so adjusts on
697	<	* failure.
695	>	* Scans for and, if found, executes one task. Scans start at a
696	>	* random index of workers array, and randomly select the first
697	>	* (2*#workers)-1 probes, and then, if all empty, resort to 2
698	>	* circular sweeps, which is necessary to check quiescence. and
699	>	* taking a submission only if no stealable tasks were found.  The
700	>	* steal code inside the loop is a specialized form of
701	>	* ForkJoinWorkerThread.deqTask, followed bookkeeping to support
702	>	* helpJoinTask and signal propagation. The code for submission
703	>	* queues is almost identical. On each steal, the worker completes
704	>	* not only the task, but also all local tasks that this task may
705	>	* have generated. On detecting staleness or contention when
706	>	* trying to take a task, this method returns without finishing
707	>	* sweep, which allows global state rechecks before retry.
708		*
709	<	* @return new worker or null if creation failed
709	>	* @param w the worker
710	>	* @param a the number of active workers
711	>	* @return true if swept all queues without finding a task
712		*/
713	<	private ForkJoinWorkerThread addWorker() {
714	<	ForkJoinWorkerThread w = null;
715	<	try {
716	<	w = factory.newThread(this);
717	<	} finally { // Adjust on either null or exceptional factory return
718	<	if (w == null) {
719	<	onWorkerCreationFailure();
720	<	return null;
713	>	private boolean scan(ForkJoinWorkerThread w, int a) {
714	>	int g = scanGuard; // mask 0 avoids useless scans if only one active
715	>	int m = (parallelism == 1 - a && blockedCount == 0) ? 0 : g & SMASK;
716	>	ForkJoinWorkerThread[] ws = workers;
717	>	if (ws == null || ws.length <= m)         // staleness check
718	>	return false;
719	>	for (int r = w.seed, k = r, j = -(m + m); j <= m + m; ++j) {
720	>	ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
721	>	ForkJoinWorkerThread v = ws[k & m];
722	>	if (v != null && (b = v.queueBase) != v.queueTop &&
723	>	(q = v.queue) != null && (i = (q.length - 1) & b) >= 0) {
724	>	long u = (i << ASHIFT) + ABASE;
725	>	if ((t = q[i]) != null && v.queueBase == b &&
726	>	UNSAFE.compareAndSwapObject(q, u, t, null)) {
727	>	int d = (v.queueBase = b + 1) - v.queueTop;
728	>	v.stealHint = w.poolIndex;
729	>	if (d != 0)
730	>	signalWork();             // propagate if nonempty
731	>	w.execTask(t);
732	>	}
733	>	r ^= r << 13; r ^= r >>> 17; w.seed = r ^ (r << 5);
734	>	return false;                     // store next seed
735	>	}
736	>	else if (j < 0) {                     // xorshift
737	>	r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5;
738	>	}
739	>	else
740	>	++k;
741	>	}
742	>	if (scanGuard != g)                       // staleness check
743	>	return false;
744	>	else {                                    // try to take submission
745	>	ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
746	>	if ((b = queueBase) != queueTop &&
747	>	(q = submissionQueue) != null &&
748	>	(i = (q.length - 1) & b) >= 0) {
749	>	long u = (i << ASHIFT) + ABASE;
750	>	if ((t = q[i]) != null && queueBase == b &&
751	>	UNSAFE.compareAndSwapObject(q, u, t, null)) {
752	>	queueBase = b + 1;
753	>	w.execTask(t);
754	>	}
755	>	return false;
756		}
757	+	return true;                         // all queues empty
758		}
656	–	w.start(recordWorker(w), locallyFifo, ueh);
657	–	return w;
759		}
760
761		/**
762	<	* Adjusts counts upon failure to create worker
763	<	*/
764	<	private void onWorkerCreationFailure() {
765	<	int c;
766	<	do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
767	<	c = workerCounts,
768	<	c - (ONE_RUNNING|ONE_TOTAL)));
769	<	tryTerminate(false); // in case of failure during shutdown
762	>	* Tries to enqueue worker w in wait queue and await change in
763	>	* worker's eventCount.  If the pool is quiescent and there is
764	>	* more than one worker, possibly terminates worker upon exit.
765	>	* Otherwise, before blocking, rescans queues to avoid missed
766	>	* signals.  Upon finding work, releases at least one worker
767	>	* (which may be the current worker). Rescans restart upon
768	>	* detected staleness or failure to release due to
769	>	* contention. Note the unusual conventions about Thread.interrupt
770	>	* here and elsewhere: Because interrupts are used solely to alert
771	>	* threads to check termination, which is checked here anyway, we
772	>	* clear status (using Thread.interrupted) before any call to
773	>	* park, so that park does not immediately return due to status
774	>	* being set via some other unrelated call to interrupt in user
775	>	* code.
776	>	*
777	>	* @param w the calling worker
778	>	* @param c the ctl value on entry
779	>	* @return true if waited or another thread was released upon enq
780	>	*/
781	>	private boolean tryAwaitWork(ForkJoinWorkerThread w, long c) {
782	>	int v = w.eventCount;
783	>	w.nextWait = (int)c;                      // w's successor record
784	>	long nc = (long)(v & E_MASK) | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
785	>	if (ctl != c || !UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
786	>	long d = ctl; // return true if lost to a deq, to force scan
787	>	return (int)d != (int)c && ((d - c) & AC_MASK) >= 0L;
788	>	}
789	>	for (int sc = w.stealCount; sc != 0;) {   // accumulate stealCount
790	>	long s = stealCount;
791	>	if (UNSAFE.compareAndSwapLong(this, stealCountOffset, s, s + sc))
792	>	sc = w.stealCount = 0;
793	>	else if (w.eventCount != v)
794	>	return true;                      // update next time
795	>	}
796	>	if ((!shutdown || !tryTerminate(false)) &&
797	>	(int)c != 0 && parallelism + (int)(nc >> AC_SHIFT) == 0 &&
798	>	blockedCount == 0 && quiescerCount == 0)
799	>	idleAwaitWork(w, nc, c, v);           // quiescent
800	>	for (boolean rescanned = false;;) {
801	>	if (w.eventCount != v)
802	>	return true;
803	>	if (!rescanned) {
804	>	int g = scanGuard, m = g & SMASK;
805	>	ForkJoinWorkerThread[] ws = workers;
806	>	if (ws != null && m < ws.length) {
807	>	rescanned = true;
808	>	for (int i = 0; i <= m; ++i) {
809	>	ForkJoinWorkerThread u = ws[i];
810	>	if (u != null) {
811	>	if (u.queueBase != u.queueTop &&
812	>	!tryReleaseWaiter())
813	>	rescanned = false; // contended
814	>	if (w.eventCount != v)
815	>	return true;
816	>	}
817	>	}
818	>	}
819	>	if (scanGuard != g ||              // stale
820	>	(queueBase != queueTop && !tryReleaseWaiter()))
821	>	rescanned = false;
822	>	if (!rescanned)
823	>	Thread.yield();                // reduce contention
824	>	else
825	>	Thread.interrupted();          // clear before park
826	>	}
827	>	else {
828	>	w.parked = true;                   // must recheck
829	>	if (w.eventCount != v) {
830	>	w.parked = false;
831	>	return true;
832	>	}
833	>	LockSupport.park(this);
834	>	rescanned = w.parked = false;
835	>	}
836	>	}
837		}
838
839		/**
840	<	* Create enough total workers to establish target parallelism,
841	<	* giving up if terminating or addWorker fails
842	<	*/
843	<	private void ensureEnoughTotalWorkers() {
844	<	int wc;
845	<	while (runState < TERMINATING &&
846	<	((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism) {
847	<	if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
848	<	wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
849	<	addWorker() == null))
850	<	break;
840	>	* If inactivating worker w has caused pool to become
841	>	* quiescent, check for pool termination, and wait for event
842	>	* for up to SHRINK_RATE nanosecs (rescans are unnecessary in
843	>	* this case because quiescence reflects consensus about lack
844	>	* of work). On timeout, if ctl has not changed, terminate the
845	>	* worker. Upon its termination (see deregisterWorker), it may
846	>	* wake up another worker to possibly repeat this process.
847	>	*
848	>	* @param w the calling worker
849	>	* @param currentCtl the ctl value after enqueuing w
850	>	* @param prevCtl the ctl value if w terminated
851	>	* @param v the eventCount w awaits change
852	>	*/
853	>	private void idleAwaitWork(ForkJoinWorkerThread w, long currentCtl,
854	>	long prevCtl, int v) {
855	>	if (w.eventCount == v) {
856	>	if (shutdown)
857	>	tryTerminate(false);
858	>	ForkJoinTask.helpExpungeStaleExceptions(); // help clean weak refs
859	>	while (ctl == currentCtl) {
860	>	long startTime = System.nanoTime();
861	>	w.parked = true;
862	>	if (w.eventCount == v)             // must recheck
863	>	LockSupport.parkNanos(this, SHRINK_RATE);
864	>	w.parked = false;
865	>	if (w.eventCount != v)
866	>	break;
867	>	else if (System.nanoTime() - startTime <
868	>	SHRINK_RATE - (SHRINK_RATE / 10)) // timing slop
869	>	Thread.interrupted();          // spurious wakeup
870	>	else if (UNSAFE.compareAndSwapLong(this, ctlOffset,
871	>	currentCtl, prevCtl)) {
872	>	w.terminate = true;            // restore previous
873	>	w.eventCount = ((int)currentCtl + EC_UNIT) & E_MASK;
874	>	break;
875	>	}
876	>	}
877		}
878		}
879
880	+	// Submissions
881	+
882		/**
883	<	* Final callback from terminating worker.  Removes record of
884	<	* worker from array, and adjusts counts. If pool is shutting
689	<	* down, tries to complete terminatation, else possibly replaces
690	<	* the worker.
883	>	* Enqueues the given task in the submissionQueue.  Same idea as
884	>	* ForkJoinWorkerThread.pushTask except for use of submissionLock.
885		*
886	<	* @param w the worker
886	>	* @param t the task
887		*/
888	<	final void workerTerminated(ForkJoinWorkerThread w) {
889	<	if (w.active) { // force inactive
890	<	w.active = false;
891	<	do {} while (!tryDecrementActiveCount());
888	>	private void addSubmission(ForkJoinTask<?> t) {
889	>	final ReentrantLock lock = this.submissionLock;
890	>	lock.lock();
891	>	try {
892	>	ForkJoinTask<?>[] q; int s, m;
893	>	if ((q = submissionQueue) != null) {    // ignore if queue removed
894	>	long u = (((s = queueTop) & (m = q.length-1)) << ASHIFT)+ABASE;
895	>	UNSAFE.putOrderedObject(q, u, t);
896	>	queueTop = s + 1;
897	>	if (s - queueBase == m)
898	>	growSubmissionQueue();
899	>	}
900	>	} finally {
901	>	lock.unlock();
902		}
903	<	forgetWorker(w);
700	<
701	<	// decrement total count, and if was running, running count
702	<	int unit = w.isTrimmed()? ONE_TOTAL : (ONE_RUNNING|ONE_TOTAL);
703	<	int wc;
704	<	do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
705	<	wc = workerCounts, wc - unit));
706	<
707	<	accumulateStealCount(w); // collect final count
708	<	if (!tryTerminate(false))
709	<	ensureEnoughTotalWorkers();
903	>	signalWork();
904		}
905
906	<	// Waiting for and signalling events
713	<
714	<	/**
715	<	* Ensures eventCount on exit is different (mod 2^32) than on
716	<	* entry.  CAS failures are OK -- any change in count suffices.
717	<	*/
718	<	private void advanceEventCount() {
719	<	int c;
720	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
721	<	}
906	>	//  (pollSubmission is defined below with exported methods)
907
908		/**
909	<	* Releases workers blocked on a count not equal to current count.
909	>	* Creates or doubles submissionQueue array.
910	>	* Basically identical to ForkJoinWorkerThread version.
911		*/
912	<	final void releaseWaiters() {
913	<	long top;
914	<	int id;
915	<	while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 &&
916	<	(int)(top >>> EVENT_COUNT_SHIFT) != eventCount) {
917	<	ForkJoinWorkerThread[] ws = workers;
918	<	ForkJoinWorkerThread w;
919	<	if (ws.length >= id && (w = ws[id - 1]) != null &&
920	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
921	<	top, w.nextWaiter))
922	<	LockSupport.unpark(w);
912	>	private void growSubmissionQueue() {
913	>	ForkJoinTask<?>[] oldQ = submissionQueue;
914	>	int size = oldQ != null ? oldQ.length << 1 : INITIAL_QUEUE_CAPACITY;
915	>	if (size > MAXIMUM_QUEUE_CAPACITY)
916	>	throw new RejectedExecutionException("Queue capacity exceeded");
917	>	if (size < INITIAL_QUEUE_CAPACITY)
918	>	size = INITIAL_QUEUE_CAPACITY;
919	>	ForkJoinTask<?>[] q = submissionQueue = new ForkJoinTask<?>[size];
920	>	int mask = size - 1;
921	>	int top = queueTop;
922	>	int oldMask;
923	>	if (oldQ != null && (oldMask = oldQ.length - 1) >= 0) {
924	>	for (int b = queueBase; b != top; ++b) {
925	>	long u = ((b & oldMask) << ASHIFT) + ABASE;
926	>	Object x = UNSAFE.getObjectVolatile(oldQ, u);
927	>	if (x != null && UNSAFE.compareAndSwapObject(oldQ, u, x, null))
928	>	UNSAFE.putObjectVolatile
929	>	(q, ((b & mask) << ASHIFT) + ABASE, x);
930	>	}
931		}
932		}
933
934	+	// Blocking support
935	+
936		/**
937	<	* Advances eventCount and releases waiters until interference by
938	<	* other releasing threads is detected.
937	>	* Tries to increment blockedCount, decrement active count
938	>	* (sometimes implicitly) and possibly release or create a
939	>	* compensating worker in preparation for blocking. Fails
940	>	* on contention or termination.
941	>	*
942	>	* @return true if the caller can block, else should recheck and retry
943		*/
944	<	final void signalWork() {
945	<	int ec;
946	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1);
947	<	outer:for (;;) {
948	<	long top = eventWaiters;
749	<	ec = eventCount;
750	<	for (;;) {
944	>	private boolean tryPreBlock() {
945	>	int b = blockedCount;
946	>	if (UNSAFE.compareAndSwapInt(this, blockedCountOffset, b, b + 1)) {
947	>	int pc = parallelism;
948	>	do {
949		ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
950	<	int id = (int)(top & WAITER_INDEX_MASK);
951	<	if (id <= 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec)
952	<	return;
953	<	if ((ws = workers).length < id || (w = ws[id - 1]) == null ||
954	<	!UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
955	<	top, top = w.nextWaiter))
956	<	continue outer;      // possibly stale; reread
957	<	LockSupport.unpark(w);
958	<	if (top != eventWaiters) // let someone else take over
959	<	return;
960	<	}
950	>	int e, ac, tc, i;
951	>	long c = ctl;
952	>	int u = (int)(c >>> 32);
953	>	if ((e = (int)c) < 0) {
954	>	// skip -- terminating
955	>	}
956	>	else if ((ac = (u >> UAC_SHIFT)) <= 0 && e != 0 &&
957	>	(ws = workers) != null &&
958	>	(i = ~e & SMASK) < ws.length &&
959	>	(w = ws[i]) != null) {
960	>	long nc = ((long)(w.nextWait & E_MASK) |
961	>	(c & (AC_MASK|TC_MASK)));
962	>	if (w.eventCount == e &&
963	>	UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
964	>	w.eventCount = (e + EC_UNIT) & E_MASK;
965	>	if (w.parked)
966	>	UNSAFE.unpark(w);
967	>	return true;             // release an idle worker
968	>	}
969	>	}
970	>	else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
971	>	long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
972	>	if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc))
973	>	return true;             // no compensation needed
974	>	}
975	>	else if (tc + pc < MAX_ID) {
976	>	long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
977	>	if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
978	>	addWorker();
979	>	return true;            // create a replacement
980	>	}
981	>	}
982	>	// try to back out on any failure and let caller retry
983	>	} while (!UNSAFE.compareAndSwapInt(this, blockedCountOffset,
984	>	b = blockedCount, b - 1));
985		}
986	+	return false;
987		}
988
989		/**
990	<	* If worker is inactive, blocks until terminating or event count
991	<	* advances from last value held by worker; in any case helps
992	<	* release others.
993	<	*
994	<	* @param w the calling worker thread
995	<	*/
996	<	private void eventSync(ForkJoinWorkerThread w) {
997	<	if (!w.active) {
998	<	int prev = w.lastEventCount;
776	<	long nextTop = (((long)prev << EVENT_COUNT_SHIFT) |
777	<	((long)(w.poolIndex + 1)));
778	<	long top;
779	<	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
780	<	(((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 ||
781	<	(int)(top >>> EVENT_COUNT_SHIFT) == prev) &&
782	<	eventCount == prev) {
783	<	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
784	<	w.nextWaiter = top, nextTop)) {
785	<	accumulateStealCount(w); // transfer steals while idle
786	<	Thread.interrupted();    // clear/ignore interrupt
787	<	while (eventCount == prev)
788	<	w.doPark();
789	<	break;
790	<	}
791	<	}
792	<	w.lastEventCount = eventCount;
793	<	}
794	<	releaseWaiters();
990	>	* Decrements blockedCount and increments active count.
991	>	*/
992	>	private void postBlock() {
993	>	long c;
994	>	do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset,  // no mask
995	>	c = ctl, c + AC_UNIT));
996	>	int b;
997	>	do {} while (!UNSAFE.compareAndSwapInt(this, blockedCountOffset,
998	>	b = blockedCount, b - 1));
999		}
1000
1001		/**
1002	<	* Callback from workers invoked upon each top-level action (i.e.,
1003	<	* stealing a task or taking a submission and running
800	<	* it). Performs one or both of the following:
1002	>	* Possibly blocks waiting for the given task to complete, or
1003	>	* cancels the task if terminating.  Fails to wait if contended.
1004		*
1005	<	* * If the worker cannot find work, updates its active status to
803	<	* inactive and updates activeCount unless there is contention, in
804	<	* which case it may try again (either in this or a subsequent
805	<	* call).  Additionally, awaits the next task event and/or helps
806	<	* wake up other releasable waiters.
807	<	*
808	<	* * If there are too many running threads, suspends this worker
809	<	* (first forcing inactivation if necessary).  If it is not
810	<	* resumed before a keepAlive elapses, the worker may be "trimmed"
811	<	* -- killed while suspended within suspendAsSpare. Otherwise,
812	<	* upon resume it rechecks to make sure that it is still needed.
813	<	*
814	<	* @param w the worker
815	<	* @param worked false if the worker scanned for work but didn't
816	<	* find any (in which case it may block waiting for work).
1005	>	* @param joinMe the task
1006		*/
1007	<	final void preStep(ForkJoinWorkerThread w, boolean worked) {
1008	<	boolean active = w.active;
1009	<	boolean inactivate = !worked & active;
1010	<	for (;;) {
1011	<	if (inactivate) {
1012	<	int c = runState;
1013	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset,
1014	<	c, c - ONE_ACTIVE))
1015	<	inactivate = active = w.active = false;
827	<	}
828	<	int wc = workerCounts;
829	<	if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
830	<	if (!worked)
831	<	eventSync(w);
832	<	return;
833	<	}
834	<	if (!(inactivate |= active) &&  // must inactivate to suspend
835	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
836	<	wc, wc - ONE_RUNNING) &&
837	<	!w.suspendAsSpare())        // false if trimmed
838	<	return;
1007	>	final void tryAwaitJoin(ForkJoinTask<?> joinMe) {
1008	>	Thread.interrupted(); // clear interrupts before checking termination
1009	>	if (joinMe.status >= 0) {
1010	>	if (tryPreBlock()) {
1011	>	joinMe.tryAwaitDone(0L);
1012	>	postBlock();
1013	>	}
1014	>	else if ((ctl & STOP_BIT) != 0L)
1015	>	joinMe.cancelIgnoringExceptions();
1016		}
1017		}
1018
1019		/**
1020	<	* Adjusts counts and creates or resumes compensating threads for
1021	<	* a worker about to block on task joinMe, returning early if
1022	<	* joinMe becomes ready. First tries resuming an existing spare
1023	<	* (which usually also avoids any count adjustment), but must then
1024	<	* decrement running count to determine whether a new thread is
1025	<	* needed. See above for fuller explanation.
1026	<	*/
1027	<	final void preJoin(ForkJoinTask<?> joinMe) {
1028	<	boolean dec = false;       // true when running count decremented
1029	<	for (;;) {
1030	<	releaseWaiters();      // help other threads progress
1031	<
855	<	if (joinMe.status < 0) // surround spare search with done checks
856	<	return;
857	<	ForkJoinWorkerThread spare = null;
858	<	for (ForkJoinWorkerThread w : workers) {
859	<	if (w != null && w.isSuspended()) {
860	<	spare = w;
861	<	break;
862	<	}
1020	>	* Possibly blocks the given worker waiting for joinMe to
1021	>	* complete or timeout.
1022	>	*
1023	>	* @param joinMe the task
1024	>	* @param millis the wait time for underlying Object.wait
1025	>	*/
1026	>	final void timedAwaitJoin(ForkJoinTask<?> joinMe, long nanos) {
1027	>	while (joinMe.status >= 0) {
1028	>	Thread.interrupted();
1029	>	if ((ctl & STOP_BIT) != 0L) {
1030	>	joinMe.cancelIgnoringExceptions();
1031	>	break;
1032		}
1033	<	if (joinMe.status < 0)
1034	<	return;
1035	<
1036	<	if (spare != null && spare.tryUnsuspend()) {
1037	<	if (dec || joinMe.requestSignal() < 0) {
1038	<	int c;
1039	<	do {} while (!UNSAFE.compareAndSwapInt(this,
1040	<	workerCountsOffset,
1041	<	c = workerCounts,
1042	<	c + ONE_RUNNING));
1043	<	} // else no net count change
1044	<	LockSupport.unpark(spare);
1045	<	return;
1046	<	}
1047	<
1048	<	int wc = workerCounts; // decrement running count
880	<	if (!dec && (wc & RUNNING_COUNT_MASK) != 0 &&
881	<	(dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
882	<	wc, wc -= ONE_RUNNING)) &&
883	<	joinMe.requestSignal() < 0) { // cannot block
884	<	int c;                        // back out
885	<	do {} while (!UNSAFE.compareAndSwapInt(this,
886	<	workerCountsOffset,
887	<	c = workerCounts,
888	<	c + ONE_RUNNING));
889	<	return;
890	<	}
891	<
892	<	if (dec) {
893	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
894	<	int pc = parallelism;
895	<	int dc = pc - (wc & RUNNING_COUNT_MASK); // deficit count
896	<	if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
897	<	!maintainsParallelism)) ||
898	<	tc >= maxPoolSize) // cannot add
899	<	return;
900	<	if (spare == null &&
901	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
902	<	wc + (ONE_RUNNING|ONE_TOTAL))) {
903	<	addWorker();
904	<	return;
1033	>	if (tryPreBlock()) {
1034	>	long last = System.nanoTime();
1035	>	while (joinMe.status >= 0) {
1036	>	long millis = TimeUnit.NANOSECONDS.toMillis(nanos);
1037	>	if (millis <= 0)
1038	>	break;
1039	>	joinMe.tryAwaitDone(millis);
1040	>	if (joinMe.status < 0)
1041	>	break;
1042	>	if ((ctl & STOP_BIT) != 0L) {
1043	>	joinMe.cancelIgnoringExceptions();
1044	>	break;
1045	>	}
1046	>	long now = System.nanoTime();
1047	>	nanos -= now - last;
1048	>	last = now;
1049		}
1050	+	postBlock();
1051	+	break;
1052		}
1053		}
1054		}
1055
1056		/**
1057	<	* Same idea as preJoin but with too many differing details to
912	<	* integrate: There are no task-based signal counts, and only one
913	<	* way to do the actual blocking. So for simplicity it is directly
914	<	* incorporated into this method.
1057	>	* If necessary, compensates for blocker, and blocks.
1058		*/
1059	<	final void doBlock(ManagedBlocker blocker, boolean maintainPar)
1059	>	private void awaitBlocker(ManagedBlocker blocker)
1060		throws InterruptedException {
1061	<	maintainPar &= maintainsParallelism; // override
1062	<	boolean dec = false;
1063	<	boolean done = false;
1064	<	for (;;) {
1065	<	releaseWaiters();
1066	<	if (done = blocker.isReleasable())
924	<	break;
925	<	ForkJoinWorkerThread spare = null;
926	<	for (ForkJoinWorkerThread w : workers) {
927	<	if (w != null && w.isSuspended()) {
928	<	spare = w;
929	<	break;
930	<	}
931	<	}
932	<	if (done = blocker.isReleasable())
933	<	break;
934	<	if (spare != null && spare.tryUnsuspend()) {
935	<	if (dec) {
936	<	int c;
937	<	do {} while (!UNSAFE.compareAndSwapInt(this,
938	<	workerCountsOffset,
939	<	c = workerCounts,
940	<	c + ONE_RUNNING));
1061	>	while (!blocker.isReleasable()) {
1062	>	if (tryPreBlock()) {
1063	>	try {
1064	>	do {} while (!blocker.isReleasable() && !blocker.block());
1065	>	} finally {
1066	>	postBlock();
1067		}
942	–	LockSupport.unpark(spare);
1068		break;
1069		}
945	–	int wc = workerCounts;
946	–	if (!dec && (wc & RUNNING_COUNT_MASK) != 0)
947	–	dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
948	–	wc, wc -= ONE_RUNNING);
949	–	if (dec) {
950	–	int tc = wc >>> TOTAL_COUNT_SHIFT;
951	–	int pc = parallelism;
952	–	int dc = pc - (wc & RUNNING_COUNT_MASK);
953	–	if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
954	–	!maintainPar)) ||
955	–	tc >= maxPoolSize)
956	–	break;
957	–	if (spare == null &&
958	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
959	–	wc + (ONE_RUNNING|ONE_TOTAL))){
960	–	addWorker();
961	–	break;
962	–	}
963	–	}
1070		}
1071	+	}
1072
1073	+	// Creating, registering and deregistring workers
1074	+
1075	+	/**
1076	+	* Tries to create and start a worker; minimally rolls back counts
1077	+	* on failure.
1078	+	*/
1079	+	private void addWorker() {
1080	+	Throwable ex = null;
1081	+	ForkJoinWorkerThread t = null;
1082		try {
1083	<	if (!done)
1084	<	do {} while (!blocker.isReleasable() && !blocker.block());
1085	<	} finally {
1086	<	if (dec) {
1087	<	int c;
1088	<	do {} while (!UNSAFE.compareAndSwapInt(this,
1089	<	workerCountsOffset,
1090	<	c = workerCounts,
1091	<	c + ONE_RUNNING));
1092	<	}
1083	>	t = factory.newThread(this);
1084	>	} catch (Throwable e) {
1085	>	ex = e;
1086	>	}
1087	>	if (t == null) {  // null or exceptional factory return
1088	>	long c;       // adjust counts
1089	>	do {} while (!UNSAFE.compareAndSwapLong
1090	>	(this, ctlOffset, c = ctl,
1091	>	(((c - AC_UNIT) & AC_MASK) |
1092	>	((c - TC_UNIT) & TC_MASK) |
1093	>	(c & ~(AC_MASK|TC_MASK)))));
1094	>	// Propagate exception if originating from an external caller
1095	>	if (!tryTerminate(false) && ex != null &&
1096	>	!(Thread.currentThread() instanceof ForkJoinWorkerThread))
1097	>	UNSAFE.throwException(ex);
1098		}
1099	+	else
1100	+	t.start();
1101		}
1102
1103		/**
1104	<	* Unless there are not enough other running threads, adjusts
1105	<	* counts for a a worker in performing helpJoin that cannot find
983	<	* any work, so that this worker can now block.
984	<	*
985	<	* @return true if worker may block
1104	>	* Callback from ForkJoinWorkerThread constructor to assign a
1105	>	* public name
1106		*/
1107	<	final boolean preBlockHelpingJoin(ForkJoinTask<?> joinMe) {
1108	<	while (joinMe.status >= 0) {
1109	<	releaseWaiters(); // help other threads progress
1107	>	final String nextWorkerName() {
1108	>	for (int n;;) {
1109	>	if (UNSAFE.compareAndSwapInt(this, nextWorkerNumberOffset,
1110	>	n = nextWorkerNumber, ++n))
1111	>	return workerNamePrefix + n;
1112	>	}
1113	>	}
1114
1115	<	// if a spare exists, resume it to maintain parallelism level
1116	<	if ((workerCounts & RUNNING_COUNT_MASK) <= parallelism) {
1117	<	ForkJoinWorkerThread spare = null;
1118	<	for (ForkJoinWorkerThread w : workers) {
1119	<	if (w != null && w.isSuspended()) {
1120	<	spare = w;
1121	<	break;
1115	>	/**
1116	>	* Callback from ForkJoinWorkerThread constructor to
1117	>	* determine its poolIndex and record in workers array.
1118	>	*
1119	>	* @param w the worker
1120	>	* @return the worker's pool index
1121	>	*/
1122	>	final int registerWorker(ForkJoinWorkerThread w) {
1123	>	/*
1124	>	* In the typical case, a new worker acquires the lock, uses
1125	>	* next available index and returns quickly.  Since we should
1126	>	* not block callers (ultimately from signalWork or
1127	>	* tryPreBlock) waiting for the lock needed to do this, we
1128	>	* instead help release other workers while waiting for the
1129	>	* lock.
1130	>	*/
1131	>	for (int g;;) {
1132	>	ForkJoinWorkerThread[] ws;
1133	>	if (((g = scanGuard) & SG_UNIT) == 0 &&
1134	>	UNSAFE.compareAndSwapInt(this, scanGuardOffset,
1135	>	g, g | SG_UNIT)) {
1136	>	int k = nextWorkerIndex;
1137	>	try {
1138	>	if ((ws = workers) != null) { // ignore on shutdown
1139	>	int n = ws.length;
1140	>	if (k < 0 || k >= n || ws[k] != null) {
1141	>	for (k = 0; k < n && ws[k] != null; ++k)
1142	>	;
1143	>	if (k == n)
1144	>	ws = workers = Arrays.copyOf(ws, n << 1);
1145	>	}
1146	>	ws[k] = w;
1147	>	nextWorkerIndex = k + 1;
1148	>	int m = g & SMASK;
1149	>	g = (k > m) ? ((m << 1) + 1) & SMASK : g + (SG_UNIT<<1);
1150		}
1151	+	} finally {
1152	+	scanGuard = g;
1153		}
1154	<	if (joinMe.status < 0)
1155	<	break;
1156	<	if (spare != null) {
1157	<	if (spare.tryUnsuspend()) {
1158	<	boolean canBlock = true;
1159	<	if (joinMe.requestSignal() < 0) {
1160	<	canBlock = false; // already done
1007	<	int c;
1008	<	do {} while (!UNSAFE.compareAndSwapInt
1009	<	(this, workerCountsOffset,
1010	<	c = workerCounts, c + ONE_RUNNING));
1011	<	}
1012	<	LockSupport.unpark(spare);
1013	<	return canBlock;
1154	>	return k;
1155	>	}
1156	>	else if ((ws = workers) != null) { // help release others
1157	>	for (ForkJoinWorkerThread u : ws) {
1158	>	if (u != null && u.queueBase != u.queueTop) {
1159	>	if (tryReleaseWaiter())
1160	>	break;
1161		}
1015	–	continue; // recheck -- another spare may exist
1162		}
1163		}
1164	+	}
1165	+	}
1166
1167	<	int wc = workerCounts; // reread to shorten CAS window
1168	<	int rc = wc & RUNNING_COUNT_MASK;
1169	<	if (rc <= 2) // keep this and at most one other thread alive
1170	<	break;
1171	<
1172	<	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1173	<	wc, wc - ONE_RUNNING)) {
1174	<	if (joinMe.requestSignal() >= 0)
1175	<	return true;
1176	<	int c;                        // back out
1177	<	do {} while (!UNSAFE.compareAndSwapInt
1178	<	(this, workerCountsOffset,
1179	<	c = workerCounts, c + ONE_RUNNING));
1180	<	break;
1181	<	}
1167	>	/**
1168	>	* Final callback from terminating worker.  Removes record of
1169	>	* worker from array, and adjusts counts. If pool is shutting
1170	>	* down, tries to complete termination.
1171	>	*
1172	>	* @param w the worker
1173	>	*/
1174	>	final void deregisterWorker(ForkJoinWorkerThread w, Throwable ex) {
1175	>	int idx = w.poolIndex;
1176	>	int sc = w.stealCount;
1177	>	int steps = 0;
1178	>	// Remove from array, adjust worker counts and collect steal count.
1179	>	// We can intermix failed removes or adjusts with steal updates
1180	>	do {
1181	>	long s, c;
1182	>	int g;
1183	>	if (steps == 0 && ((g = scanGuard) & SG_UNIT) == 0 &&
1184	>	UNSAFE.compareAndSwapInt(this, scanGuardOffset,
1185	>	g, g |= SG_UNIT)) {
1186	>	ForkJoinWorkerThread[] ws = workers;
1187	>	if (ws != null && idx >= 0 &&
1188	>	idx < ws.length && ws[idx] == w)
1189	>	ws[idx] = null;    // verify
1190	>	nextWorkerIndex = idx;
1191	>	scanGuard = g + SG_UNIT;
1192	>	steps = 1;
1193	>	}
1194	>	if (steps == 1 &&
1195	>	UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl,
1196	>	(((c - AC_UNIT) & AC_MASK) |
1197	>	((c - TC_UNIT) & TC_MASK) |
1198	>	(c & ~(AC_MASK|TC_MASK)))))
1199	>	steps = 2;
1200	>	if (sc != 0 &&
1201	>	UNSAFE.compareAndSwapLong(this, stealCountOffset,
1202	>	s = stealCount, s + sc))
1203	>	sc = 0;
1204	>	} while (steps != 2 || sc != 0);
1205	>	if (!tryTerminate(false)) {
1206	>	if (ex != null)   // possibly replace if died abnormally
1207	>	signalWork();
1208	>	else
1209	>	tryReleaseWaiter();
1210		}
1035	–	return false;
1211		}
1212
1213	+	// Shutdown and termination
1214	+
1215		/**
1216		* Possibly initiates and/or completes termination.
1217		*
#	Line 1043 | Line 1220 | public class ForkJoinPool extends Abstra
1220		* @return true if now terminating or terminated
1221		*/
1222		private boolean tryTerminate(boolean now) {
1223	<	if (now)
1224	<	advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1225	<	else if (runState < SHUTDOWN ||
1226	<	!submissionQueue.isEmpty() ||
1227	<	(runState & ACTIVE_COUNT_MASK) != 0)
1228	<	return false;
1229	<
1230	<	if (advanceRunLevel(TERMINATING))
1231	<	startTerminating();
1232	<
1233	<	// Finish now if all threads terminated; else in some subsequent call
1234	<	if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1235	<	advanceRunLevel(TERMINATED);
1236	<	terminationLatch.countDown();
1223	>	long c;
1224	>	while (((c = ctl) & STOP_BIT) == 0) {
1225	>	if (!now) {
1226	>	if ((int)(c >> AC_SHIFT) != -parallelism)
1227	>	return false;
1228	>	if (!shutdown || blockedCount != 0 || quiescerCount != 0 ||
1229	>	queueBase != queueTop) {
1230	>	if (ctl == c) // staleness check
1231	>	return false;
1232	>	continue;
1233	>	}
1234	>	}
1235	>	if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, c | STOP_BIT))
1236	>	startTerminating();
1237	>	}
1238	>	if ((short)(c >>> TC_SHIFT) == -parallelism) { // signal when 0 workers
1239	>	final ReentrantLock lock = this.submissionLock;
1240	>	lock.lock();
1241	>	try {
1242	>	termination.signalAll();
1243	>	} finally {
1244	>	lock.unlock();
1245	>	}
1246		}
1247		return true;
1248		}
1249
1250		/**
1251	<	* Actions on transition to TERMINATING
1251	>	* Runs up to three passes through workers: (0) Setting
1252	>	* termination status for each worker, followed by wakeups up to
1253	>	* queued workers; (1) helping cancel tasks; (2) interrupting
1254	>	* lagging threads (likely in external tasks, but possibly also
1255	>	* blocked in joins).  Each pass repeats previous steps because of
1256	>	* potential lagging thread creation.
1257		*/
1258		private void startTerminating() {
1259	<	// Clear out and cancel submissions, ignoring exceptions
1260	<	ForkJoinTask<?> task;
1261	<	while ((task = submissionQueue.poll()) != null) {
1262	<	try {
1263	<	task.cancel(false);
1264	<	} catch (Throwable ignore) {
1259	>	cancelSubmissions();
1260	>	for (int pass = 0; pass < 3; ++pass) {
1261	>	ForkJoinWorkerThread[] ws = workers;
1262	>	if (ws != null) {
1263	>	for (ForkJoinWorkerThread w : ws) {
1264	>	if (w != null) {
1265	>	w.terminate = true;
1266	>	if (pass > 0) {
1267	>	w.cancelTasks();
1268	>	if (pass > 1 && !w.isInterrupted()) {
1269	>	try {
1270	>	w.interrupt();
1271	>	} catch (SecurityException ignore) {
1272	>	}
1273	>	}
1274	>	}
1275	>	}
1276	>	}
1277	>	terminateWaiters();
1278		}
1279		}
1280	<	// Propagate run level
1281	<	for (ForkJoinWorkerThread w : workers) {
1282	<	if (w != null)
1283	<	w.shutdown();    // also resumes suspended workers
1284	<	}
1285	<	// Ensure no straggling local tasks
1286	<	for (ForkJoinWorkerThread w : workers) {
1287	<	if (w != null)
1288	<	w.cancelTasks();
1085	<	}
1086	<	// Wake up idle workers
1087	<	advanceEventCount();
1088	<	releaseWaiters();
1089	<	// Unstick pending joins
1090	<	for (ForkJoinWorkerThread w : workers) {
1091	<	if (w != null && !w.isTerminated()) {
1280	>	}
1281	>
1282	>	/**
1283	>	* Polls and cancels all submissions. Called only during termination.
1284	>	*/
1285	>	private void cancelSubmissions() {
1286	>	while (queueBase != queueTop) {
1287	>	ForkJoinTask<?> task = pollSubmission();
1288	>	if (task != null) {
1289		try {
1290	<	w.interrupt();
1291	<	} catch (SecurityException ignore) {
1290	>	task.cancel(false);
1291	>	} catch (Throwable ignore) {
1292		}
1293		}
1294		}
1295		}
1296
1297	<	// misc support for ForkJoinWorkerThread
1297	>	/**
1298	>	* Tries to set the termination status of waiting workers, and
1299	>	* then wakes them up (after which they will terminate).
1300	>	*/
1301	>	private void terminateWaiters() {
1302	>	ForkJoinWorkerThread[] ws = workers;
1303	>	if (ws != null) {
1304	>	ForkJoinWorkerThread w; long c; int i, e;
1305	>	int n = ws.length;
1306	>	while ((i = ~(e = (int)(c = ctl)) & SMASK) < n &&
1307	>	(w = ws[i]) != null && w.eventCount == (e & E_MASK)) {
1308	>	if (UNSAFE.compareAndSwapLong(this, ctlOffset, c,
1309	>	(long)(w.nextWait & E_MASK) |
1310	>	((c + AC_UNIT) & AC_MASK) |
1311	>	(c & (TC_MASK|STOP_BIT)))) {
1312	>	w.terminate = true;
1313	>	w.eventCount = e + EC_UNIT;
1314	>	if (w.parked)
1315	>	UNSAFE.unpark(w);
1316	>	}
1317	>	}
1318	>	}
1319	>	}
1320	>
1321	>	// misc ForkJoinWorkerThread support
1322
1323		/**
1324	<	* Returns pool number
1324	>	* Increments or decrements quiescerCount. Needed only to prevent
1325	>	* triggering shutdown if a worker is transiently inactive while
1326	>	* checking quiescence.
1327	>	*
1328	>	* @param delta 1 for increment, -1 for decrement
1329		*/
1330	<	final int getPoolNumber() {
1331	<	return poolNumber;
1330	>	final void addQuiescerCount(int delta) {
1331	>	int c;
1332	>	do {} while (!UNSAFE.compareAndSwapInt(this, quiescerCountOffset,
1333	>	c = quiescerCount, c + delta));
1334		}
1335
1336		/**
1337	<	* Accumulates steal count from a worker, clearing
1338	<	* the worker's value
1337	>	* Directly increments or decrements active count without queuing.
1338	>	* This method is used to transiently assert inactivation while
1339	>	* checking quiescence.
1340	>	*
1341	>	* @param delta 1 for increment, -1 for decrement
1342		*/
1343	<	final void accumulateStealCount(ForkJoinWorkerThread w) {
1344	<	int sc = w.stealCount;
1345	<	if (sc != 0) {
1346	<	long c;
1347	<	w.stealCount = 0;
1348	<	do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1119	<	c = stealCount, c + sc));
1120	<	}
1343	>	final void addActiveCount(int delta) {
1344	>	long d = delta < 0 ? -AC_UNIT : AC_UNIT;
1345	>	long c;
1346	>	do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl,
1347	>	((c + d) & AC_MASK) |
1348	>	(c & ~AC_MASK)));
1349		}
1350
1351		/**
#	Line 1125 | Line 1353 | public class ForkJoinPool extends Abstra
1353		* active thread.
1354		*/
1355		final int idlePerActive() {
1356	<	int ac = runState;    // no mask -- artifically boosts during shutdown
1357	<	int pc = parallelism; // use targeted parallelism, not rc
1358	<	// Use exact results for small values, saturate past 4
1359	<	return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1356	>	// Approximate at powers of two for small values, saturate past 4
1357	>	int p = parallelism;
1358	>	int a = p + (int)(ctl >> AC_SHIFT);
1359	>	return (a > (p >>>= 1) ? 0 :
1360	>	a > (p >>>= 1) ? 1 :
1361	>	a > (p >>>= 1) ? 2 :
1362	>	a > (p >>>= 1) ? 4 :
1363	>	8);
1364		}
1365
1366	<	// Public and protected methods
1366	>	// Exported methods
1367
1368		// Constructors
1369
1370		/**
1371		* Creates a {@code ForkJoinPool} with parallelism equal to {@link
1372	<	* java.lang.Runtime#availableProcessors}, and using the {@linkplain
1373	<	* #defaultForkJoinWorkerThreadFactory default thread factory}.
1372	>	* java.lang.Runtime#availableProcessors}, using the {@linkplain
1373	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
1374	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
1375		*
1376		* @throws SecurityException if a security manager exists and
1377		*         the caller is not permitted to modify threads
#	Line 1147 | Line 1380 | public class ForkJoinPool extends Abstra
1380		*/
1381		public ForkJoinPool() {
1382		this(Runtime.getRuntime().availableProcessors(),
1383	<	defaultForkJoinWorkerThreadFactory);
1383	>	defaultForkJoinWorkerThreadFactory, null, false);
1384		}
1385
1386		/**
1387		* Creates a {@code ForkJoinPool} with the indicated parallelism
1388	<	* level and using the {@linkplain
1389	<	* #defaultForkJoinWorkerThreadFactory default thread factory}.
1388	>	* level, the {@linkplain
1389	>	* #defaultForkJoinWorkerThreadFactory default thread factory},
1390	>	* no UncaughtExceptionHandler, and non-async LIFO processing mode.
1391		*
1392		* @param parallelism the parallelism level
1393		* @throws IllegalArgumentException if parallelism less than or
#	Line 1164 | Line 1398 | public class ForkJoinPool extends Abstra
1398		*         java.lang.RuntimePermission}{@code ("modifyThread")}
1399		*/
1400		public ForkJoinPool(int parallelism) {
1401	<	this(parallelism, defaultForkJoinWorkerThreadFactory);
1401	>	this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1402		}
1403
1404		/**
1405	<	* Creates a {@code ForkJoinPool} with parallelism equal to {@link
1172	<	* java.lang.Runtime#availableProcessors}, and using the given
1173	<	* thread factory.
1405	>	* Creates a {@code ForkJoinPool} with the given parameters.
1406		*
1407	<	* @param factory the factory for creating new threads
1408	<	* @throws NullPointerException if the factory is null
1409	<	* @throws SecurityException if a security manager exists and
1410	<	*         the caller is not permitted to modify threads
1411	<	*         because it does not hold {@link
1412	<	*         java.lang.RuntimePermission}{@code ("modifyThread")}
1413	<	*/
1414	<	public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
1415	<	this(Runtime.getRuntime().availableProcessors(), factory);
1416	<	}
1417	<
1418	<	/**
1419	<	* Creates a {@code ForkJoinPool} with the given parallelism and
1188	<	* thread factory.
1189	<	*
1190	<	* @param parallelism the parallelism level
1191	<	* @param factory the factory for creating new threads
1407	>	* @param parallelism the parallelism level. For default value,
1408	>	* use {@link java.lang.Runtime#availableProcessors}.
1409	>	* @param factory the factory for creating new threads. For default value,
1410	>	* use {@link #defaultForkJoinWorkerThreadFactory}.
1411	>	* @param handler the handler for internal worker threads that
1412	>	* terminate due to unrecoverable errors encountered while executing
1413	>	* tasks. For default value, use {@code null}.
1414	>	* @param asyncMode if true,
1415	>	* establishes local first-in-first-out scheduling mode for forked
1416	>	* tasks that are never joined. This mode may be more appropriate
1417	>	* than default locally stack-based mode in applications in which
1418	>	* worker threads only process event-style asynchronous tasks.
1419	>	* For default value, use {@code false}.
1420		* @throws IllegalArgumentException if parallelism less than or
1421		*         equal to zero, or greater than implementation limit
1422		* @throws NullPointerException if the factory is null
#	Line 1197 | Line 1425 | public class ForkJoinPool extends Abstra
1425		*         because it does not hold {@link
1426		*         java.lang.RuntimePermission}{@code ("modifyThread")}
1427		*/
1428	<	public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
1428	>	public ForkJoinPool(int parallelism,
1429	>	ForkJoinWorkerThreadFactory factory,
1430	>	Thread.UncaughtExceptionHandler handler,
1431	>	boolean asyncMode) {
1432		checkPermission();
1433		if (factory == null)
1434		throw new NullPointerException();
1435	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
1435	>	if (parallelism <= 0 || parallelism > MAX_ID)
1436		throw new IllegalArgumentException();
1206	–	this.poolNumber = poolNumberGenerator.incrementAndGet();
1207	–	int arraySize = initialArraySizeFor(parallelism);
1437		this.parallelism = parallelism;
1438		this.factory = factory;
1439	<	this.maxPoolSize = MAX_THREADS;
1440	<	this.maintainsParallelism = true;
1441	<	this.workers = new ForkJoinWorkerThread[arraySize];
1442	<	this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1443	<	this.workerLock = new ReentrantLock();
1444	<	this.terminationLatch = new CountDownLatch(1);
1445	<	// Start first worker; remaining workers added upon first submission
1446	<	workerCounts = ONE_RUNNING | ONE_TOTAL;
1447	<	addWorker();
1448	<	}
1449	<
1450	<	/**
1451	<	* Returns initial power of two size for workers array.
1452	<	* @param pc the initial parallelism level
1453	<	*/
1454	<	private static int initialArraySizeFor(int pc) {
1455	<	// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1456	<	int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1457	<	size |= size >>> 1;
1229	<	size |= size >>> 2;
1230	<	size |= size >>> 4;
1231	<	size |= size >>> 8;
1232	<	return size + 1;
1439	>	this.ueh = handler;
1440	>	this.locallyFifo = asyncMode;
1441	>	long np = (long)(-parallelism); // offset ctl counts
1442	>	this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
1443	>	this.submissionQueue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
1444	>	// initialize workers array with room for 2*parallelism if possible
1445	>	int n = parallelism << 1;
1446	>	if (n >= MAX_ID)
1447	>	n = MAX_ID;
1448	>	else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
1449	>	n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8;
1450	>	}
1451	>	workers = new ForkJoinWorkerThread[n + 1];
1452	>	this.submissionLock = new ReentrantLock();
1453	>	this.termination = submissionLock.newCondition();
1454	>	StringBuilder sb = new StringBuilder("ForkJoinPool-");
1455	>	sb.append(poolNumberGenerator.incrementAndGet());
1456	>	sb.append("-worker-");
1457	>	this.workerNamePrefix = sb.toString();
1458		}
1459
1460		// Execution methods
1461
1462		/**
1238	–	* Common code for execute, invoke and submit
1239	–	*/
1240	–	private <T> void doSubmit(ForkJoinTask<T> task) {
1241	–	if (task == null)
1242	–	throw new NullPointerException();
1243	–	if (runState >= SHUTDOWN)
1244	–	throw new RejectedExecutionException();
1245	–	submissionQueue.offer(task);
1246	–	advanceEventCount();
1247	–	releaseWaiters();
1248	–	if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
1249	–	ensureEnoughTotalWorkers();
1250	–	}
1251	–
1252	–	/**
1463		* Performs the given task, returning its result upon completion.
1464	+	* If the computation encounters an unchecked Exception or Error,
1465	+	* it is rethrown as the outcome of this invocation.  Rethrown
1466	+	* exceptions behave in the same way as regular exceptions, but,
1467	+	* when possible, contain stack traces (as displayed for example
1468	+	* using {@code ex.printStackTrace()}) of both the current thread
1469	+	* as well as the thread actually encountering the exception;
1470	+	* minimally only the latter.
1471		*
1472		* @param task the task
1473		* @return the task's result
#	Line 1259 | Line 1476 | public class ForkJoinPool extends Abstra
1476		*         scheduled for execution
1477		*/
1478		public <T> T invoke(ForkJoinTask<T> task) {
1479	<	doSubmit(task);
1480	<	return task.join();
1479	>	Thread t = Thread.currentThread();
1480	>	if (task == null)
1481	>	throw new NullPointerException();
1482	>	if (shutdown)
1483	>	throw new RejectedExecutionException();
1484	>	if ((t instanceof ForkJoinWorkerThread) &&
1485	>	((ForkJoinWorkerThread)t).pool == this)
1486	>	return task.invoke();  // bypass submit if in same pool
1487	>	else {
1488	>	addSubmission(task);
1489	>	return task.join();
1490	>	}
1491	>	}
1492	>
1493	>	/**
1494	>	* Unless terminating, forks task if within an ongoing FJ
1495	>	* computation in the current pool, else submits as external task.
1496	>	*/
1497	>	private <T> void forkOrSubmit(ForkJoinTask<T> task) {
1498	>	ForkJoinWorkerThread w;
1499	>	Thread t = Thread.currentThread();
1500	>	if (shutdown)
1501	>	throw new RejectedExecutionException();
1502	>	if ((t instanceof ForkJoinWorkerThread) &&
1503	>	(w = (ForkJoinWorkerThread)t).pool == this)
1504	>	w.pushTask(task);
1505	>	else
1506	>	addSubmission(task);
1507		}
1508
1509		/**
#	Line 1272 | Line 1515 | public class ForkJoinPool extends Abstra
1515		*         scheduled for execution
1516		*/
1517		public void execute(ForkJoinTask<?> task) {
1518	<	doSubmit(task);
1518	>	if (task == null)
1519	>	throw new NullPointerException();
1520	>	forkOrSubmit(task);
1521		}
1522
1523		// AbstractExecutorService methods
#	Line 1283 | Line 1528 | public class ForkJoinPool extends Abstra
1528		*         scheduled for execution
1529		*/
1530		public void execute(Runnable task) {
1531	+	if (task == null)
1532	+	throw new NullPointerException();
1533		ForkJoinTask<?> job;
1534		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1535		job = (ForkJoinTask<?>) task;
1536		else
1537		job = ForkJoinTask.adapt(task, null);
1538	<	doSubmit(job);
1538	>	forkOrSubmit(job);
1539	>	}
1540	>
1541	>	/**
1542	>	* Submits a ForkJoinTask for execution.
1543	>	*
1544	>	* @param task the task to submit
1545	>	* @return the task
1546	>	* @throws NullPointerException if the task is null
1547	>	* @throws RejectedExecutionException if the task cannot be
1548	>	*         scheduled for execution
1549	>	*/
1550	>	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1551	>	if (task == null)
1552	>	throw new NullPointerException();
1553	>	forkOrSubmit(task);
1554	>	return task;
1555		}
1556
1557		/**
#	Line 1297 | Line 1560 | public class ForkJoinPool extends Abstra
1560		*         scheduled for execution
1561		*/
1562		public <T> ForkJoinTask<T> submit(Callable<T> task) {
1563	+	if (task == null)
1564	+	throw new NullPointerException();
1565		ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1566	<	doSubmit(job);
1566	>	forkOrSubmit(job);
1567		return job;
1568		}
1569
#	Line 1308 | Line 1573 | public class ForkJoinPool extends Abstra
1573		*         scheduled for execution
1574		*/
1575		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1576	+	if (task == null)
1577	+	throw new NullPointerException();
1578		ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1579	<	doSubmit(job);
1579	>	forkOrSubmit(job);
1580		return job;
1581		}
1582
#	Line 1319 | Line 1586 | public class ForkJoinPool extends Abstra
1586		*         scheduled for execution
1587		*/
1588		public ForkJoinTask<?> submit(Runnable task) {
1589	+	if (task == null)
1590	+	throw new NullPointerException();
1591		ForkJoinTask<?> job;
1592		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1593		job = (ForkJoinTask<?>) task;
1594		else
1595		job = ForkJoinTask.adapt(task, null);
1596	<	doSubmit(job);
1596	>	forkOrSubmit(job);
1597		return job;
1598		}
1599
1600		/**
1332	–	* Submits a ForkJoinTask for execution.
1333	–	*
1334	–	* @param task the task to submit
1335	–	* @return the task
1336	–	* @throws NullPointerException if the task is null
1337	–	* @throws RejectedExecutionException if the task cannot be
1338	–	*         scheduled for execution
1339	–	*/
1340	–	public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1341	–	doSubmit(task);
1342	–	return task;
1343	–	}
1344	–
1345	–	/**
1601		* @throws NullPointerException       {@inheritDoc}
1602		* @throws RejectedExecutionException {@inheritDoc}
1603		*/
#	Line 1384 | Line 1639 | public class ForkJoinPool extends Abstra
1639		* @return the handler, or {@code null} if none
1640		*/
1641		public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1387	–	workerCountReadFence();
1642		return ueh;
1643		}
1644
1645		/**
1392	–	* Sets the handler for internal worker threads that terminate due
1393	–	* to unrecoverable errors encountered while executing tasks.
1394	–	* Unless set, the current default or ThreadGroup handler is used
1395	–	* as handler.
1396	–	*
1397	–	* @param h the new handler
1398	–	* @return the old handler, or {@code null} if none
1399	–	* @throws SecurityException if a security manager exists and
1400	–	*         the caller is not permitted to modify threads
1401	–	*         because it does not hold {@link
1402	–	*         java.lang.RuntimePermission}{@code ("modifyThread")}
1403	–	*/
1404	–	public Thread.UncaughtExceptionHandler
1405	–	setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
1406	–	checkPermission();
1407	–	workerCountReadFence();
1408	–	Thread.UncaughtExceptionHandler old = ueh;
1409	–	if (h != old) {
1410	–	ueh = h;
1411	–	workerCountWriteFence();
1412	–	for (ForkJoinWorkerThread w : workers) {
1413	–	if (w != null)
1414	–	w.setUncaughtExceptionHandler(h);
1415	–	}
1416	–	}
1417	–	return old;
1418	–	}
1419	–
1420	–	/**
1421	–	* Sets the target parallelism level of this pool.
1422	–	*
1423	–	* @param parallelism the target parallelism
1424	–	* @throws IllegalArgumentException if parallelism less than or
1425	–	* equal to zero or greater than maximum size bounds
1426	–	* @throws SecurityException if a security manager exists and
1427	–	*         the caller is not permitted to modify threads
1428	–	*         because it does not hold {@link
1429	–	*         java.lang.RuntimePermission}{@code ("modifyThread")}
1430	–	*/
1431	–	public void setParallelism(int parallelism) {
1432	–	checkPermission();
1433	–	if (parallelism <= 0 || parallelism > maxPoolSize)
1434	–	throw new IllegalArgumentException();
1435	–	workerCountReadFence();
1436	–	int pc = this.parallelism;
1437	–	if (pc != parallelism) {
1438	–	this.parallelism = parallelism;
1439	–	workerCountWriteFence();
1440	–	// Release spares. If too many, some will die after re-suspend
1441	–	for (ForkJoinWorkerThread w : workers) {
1442	–	if (w != null && w.tryUnsuspend()) {
1443	–	updateRunningCount(1);
1444	–	LockSupport.unpark(w);
1445	–	}
1446	–	}
1447	–	ensureEnoughTotalWorkers();
1448	–	advanceEventCount();
1449	–	releaseWaiters(); // force config recheck by existing workers
1450	–	}
1451	–	}
1452	–
1453	–	/**
1646		* Returns the targeted parallelism level of this pool.
1647		*
1648		* @return the targeted parallelism level of this pool
1649		*/
1650		public int getParallelism() {
1459	–	//        workerCountReadFence(); // inlined below
1460	–	int ignore = workerCounts;
1651		return parallelism;
1652		}
1653
1654		/**
1655		* Returns the number of worker threads that have started but not
1656	<	* yet terminated.  This result returned by this method may differ
1656	>	* yet terminated.  The result returned by this method may differ
1657		* from {@link #getParallelism} when threads are created to
1658		* maintain parallelism when others are cooperatively blocked.
1659		*
1660		* @return the number of worker threads
1661		*/
1662		public int getPoolSize() {
1663	<	return workerCounts >>> TOTAL_COUNT_SHIFT;
1474	<	}
1475	<
1476	<	/**
1477	<	* Returns the maximum number of threads allowed to exist in the
1478	<	* pool. Unless set using {@link #setMaximumPoolSize}, the
1479	<	* maximum is an implementation-defined value designed only to
1480	<	* prevent runaway growth.
1481	<	*
1482	<	* @return the maximum
1483	<	*/
1484	<	public int getMaximumPoolSize() {
1485	<	workerCountReadFence();
1486	<	return maxPoolSize;
1487	<	}
1488	<
1489	<	/**
1490	<	* Sets the maximum number of threads allowed to exist in the
1491	<	* pool. The given value should normally be greater than or equal
1492	<	* to the {@link #getParallelism parallelism} level. Setting this
1493	<	* value has no effect on current pool size. It controls
1494	<	* construction of new threads. The use of this method may cause
1495	<	* tasks that intrinsically require extra threads for dependent
1496	<	* computations to indefinitely stall. If you are instead trying
1497	<	* to minimize internal thread creation, consider setting {@link
1498	<	* #setMaintainsParallelism} as false.
1499	<	*
1500	<	* @throws IllegalArgumentException if negative or greater than
1501	<	* internal implementation limit
1502	<	*/
1503	<	public void setMaximumPoolSize(int newMax) {
1504	<	if (newMax < 0 || newMax > MAX_THREADS)
1505	<	throw new IllegalArgumentException();
1506	<	maxPoolSize = newMax;
1507	<	workerCountWriteFence();
1508	<	}
1509	<
1510	<	/**
1511	<	* Returns {@code true} if this pool dynamically maintains its
1512	<	* target parallelism level. If false, new threads are added only
1513	<	* to avoid possible starvation.  This setting is by default true.
1514	<	*
1515	<	* @return {@code true} if maintains parallelism
1516	<	*/
1517	<	public boolean getMaintainsParallelism() {
1518	<	workerCountReadFence();
1519	<	return maintainsParallelism;
1520	<	}
1521	<
1522	<	/**
1523	<	* Sets whether this pool dynamically maintains its target
1524	<	* parallelism level. If false, new threads are added only to
1525	<	* avoid possible starvation.
1526	<	*
1527	<	* @param enable {@code true} to maintain parallelism
1528	<	*/
1529	<	public void setMaintainsParallelism(boolean enable) {
1530	<	maintainsParallelism = enable;
1531	<	workerCountWriteFence();
1532	<	}
1533	<
1534	<	/**
1535	<	* Establishes local first-in-first-out scheduling mode for forked
1536	<	* tasks that are never joined. This mode may be more appropriate
1537	<	* than default locally stack-based mode in applications in which
1538	<	* worker threads only process asynchronous tasks.  This method is
1539	<	* designed to be invoked only when the pool is quiescent, and
1540	<	* typically only before any tasks are submitted. The effects of
1541	<	* invocations at other times may be unpredictable.
1542	<	*
1543	<	* @param async if {@code true}, use locally FIFO scheduling
1544	<	* @return the previous mode
1545	<	* @see #getAsyncMode
1546	<	*/
1547	<	public boolean setAsyncMode(boolean async) {
1548	<	workerCountReadFence();
1549	<	boolean oldMode = locallyFifo;
1550	<	if (oldMode != async) {
1551	<	locallyFifo = async;
1552	<	workerCountWriteFence();
1553	<	for (ForkJoinWorkerThread w : workers) {
1554	<	if (w != null)
1555	<	w.setAsyncMode(async);
1556	<	}
1557	<	}
1558	<	return oldMode;
1663	>	return parallelism + (short)(ctl >>> TC_SHIFT);
1664		}
1665
1666		/**
#	Line 1563 | Line 1668 | public class ForkJoinPool extends Abstra
1668		* scheduling mode for forked tasks that are never joined.
1669		*
1670		* @return {@code true} if this pool uses async mode
1566	–	* @see #setAsyncMode
1671		*/
1672		public boolean getAsyncMode() {
1569	–	workerCountReadFence();
1673		return locallyFifo;
1674		}
1675
#	Line 1579 | Line 1682 | public class ForkJoinPool extends Abstra
1682		* @return the number of worker threads
1683		*/
1684		public int getRunningThreadCount() {
1685	<	return workerCounts & RUNNING_COUNT_MASK;
1685	>	int r = parallelism + (int)(ctl >> AC_SHIFT);
1686	>	return (r <= 0) ? 0 : r; // suppress momentarily negative values
1687		}
1688
1689		/**
#	Line 1590 | Line 1694 | public class ForkJoinPool extends Abstra
1694		* @return the number of active threads
1695		*/
1696		public int getActiveThreadCount() {
1697	<	return runState & ACTIVE_COUNT_MASK;
1697	>	int r = parallelism + (int)(ctl >> AC_SHIFT) + blockedCount;
1698	>	return (r <= 0) ? 0 : r; // suppress momentarily negative values
1699		}
1700
1701		/**
#	Line 1605 | Line 1710 | public class ForkJoinPool extends Abstra
1710		* @return {@code true} if all threads are currently idle
1711		*/
1712		public boolean isQuiescent() {
1713	<	return (runState & ACTIVE_COUNT_MASK) == 0;
1713	>	return parallelism + (int)(ctl >> AC_SHIFT) + blockedCount == 0;
1714		}
1715
1716		/**
#	Line 1635 | Line 1740 | public class ForkJoinPool extends Abstra
1740		*/
1741		public long getQueuedTaskCount() {
1742		long count = 0;
1743	<	for (ForkJoinWorkerThread w : workers) {
1744	<	if (w != null)
1745	<	count += w.getQueueSize();
1743	>	ForkJoinWorkerThread[] ws;
1744	>	if ((short)(ctl >>> TC_SHIFT) > -parallelism &&
1745	>	(ws = workers) != null) {
1746	>	for (ForkJoinWorkerThread w : ws)
1747	>	if (w != null)
1748	>	count -= w.queueBase - w.queueTop; // must read base first
1749		}
1750		return count;
1751		}
1752
1753		/**
1754		* Returns an estimate of the number of tasks submitted to this
1755	<	* pool that have not yet begun executing.  This method takes time
1756	<	* proportional to the number of submissions.
1755	>	* pool that have not yet begun executing.  This method may take
1756	>	* time proportional to the number of submissions.
1757		*
1758		* @return the number of queued submissions
1759		*/
1760		public int getQueuedSubmissionCount() {
1761	<	return submissionQueue.size();
1761	>	return -queueBase + queueTop;
1762		}
1763
1764		/**
#	Line 1660 | Line 1768 | public class ForkJoinPool extends Abstra
1768		* @return {@code true} if there are any queued submissions
1769		*/
1770		public boolean hasQueuedSubmissions() {
1771	<	return !submissionQueue.isEmpty();
1771	>	return queueBase != queueTop;
1772		}
1773
1774		/**
#	Line 1671 | Line 1779 | public class ForkJoinPool extends Abstra
1779		* @return the next submission, or {@code null} if none
1780		*/
1781		protected ForkJoinTask<?> pollSubmission() {
1782	<	return submissionQueue.poll();
1782	>	ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
1783	>	while ((b = queueBase) != queueTop &&
1784	>	(q = submissionQueue) != null &&
1785	>	(i = (q.length - 1) & b) >= 0) {
1786	>	long u = (i << ASHIFT) + ABASE;
1787	>	if ((t = q[i]) != null &&
1788	>	queueBase == b &&
1789	>	UNSAFE.compareAndSwapObject(q, u, t, null)) {
1790	>	queueBase = b + 1;
1791	>	return t;
1792	>	}
1793	>	}
1794	>	return null;
1795		}
1796
1797		/**
#	Line 1692 | Line 1812 | public class ForkJoinPool extends Abstra
1812		* @return the number of elements transferred
1813		*/
1814		protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1815	<	int n = submissionQueue.drainTo(c);
1816	<	for (ForkJoinWorkerThread w : workers) {
1817	<	if (w != null)
1818	<	n += w.drainTasksTo(c);
1815	>	int count = 0;
1816	>	while (queueBase != queueTop) {
1817	>	ForkJoinTask<?> t = pollSubmission();
1818	>	if (t != null) {
1819	>	c.add(t);
1820	>	++count;
1821	>	}
1822	>	}
1823	>	ForkJoinWorkerThread[] ws;
1824	>	if ((short)(ctl >>> TC_SHIFT) > -parallelism &&
1825	>	(ws = workers) != null) {
1826	>	for (ForkJoinWorkerThread w : ws)
1827	>	if (w != null)
1828	>	count += w.drainTasksTo(c);
1829		}
1830	<	return n;
1830	>	return count;
1831		}
1832
1833		/**
#	Line 1711 | Line 1841 | public class ForkJoinPool extends Abstra
1841		long st = getStealCount();
1842		long qt = getQueuedTaskCount();
1843		long qs = getQueuedSubmissionCount();
1714	–	int wc = workerCounts;
1715	–	int tc = wc >>> TOTAL_COUNT_SHIFT;
1716	–	int rc = wc & RUNNING_COUNT_MASK;
1844		int pc = parallelism;
1845	<	int rs = runState;
1846	<	int ac = rs & ACTIVE_COUNT_MASK;
1845	>	long c = ctl;
1846	>	int tc = pc + (short)(c >>> TC_SHIFT);
1847	>	int rc = pc + (int)(c >> AC_SHIFT);
1848	>	if (rc < 0) // ignore transient negative
1849	>	rc = 0;
1850	>	int ac = rc + blockedCount;
1851	>	String level;
1852	>	if ((c & STOP_BIT) != 0)
1853	>	level = (tc == 0) ? "Terminated" : "Terminating";
1854	>	else
1855	>	level = shutdown ? "Shutting down" : "Running";
1856		return super.toString() +
1857	<	"[" + runLevelToString(rs) +
1857	>	"[" + level +
1858		", parallelism = " + pc +
1859		", size = " + tc +
1860		", active = " + ac +
#	Line 1729 | Line 1865 | public class ForkJoinPool extends Abstra
1865		"]";
1866		}
1867
1732	–	private static String runLevelToString(int s) {
1733	–	return ((s & TERMINATED) != 0 ? "Terminated" :
1734	–	((s & TERMINATING) != 0 ? "Terminating" :
1735	–	((s & SHUTDOWN) != 0 ? "Shutting down" :
1736	–	"Running")));
1737	–	}
1738	–
1868		/**
1869		* Initiates an orderly shutdown in which previously submitted
1870		* tasks are executed, but no new tasks will be accepted.
#	Line 1750 | Line 1879 | public class ForkJoinPool extends Abstra
1879		*/
1880		public void shutdown() {
1881		checkPermission();
1882	<	advanceRunLevel(SHUTDOWN);
1882	>	shutdown = true;
1883		tryTerminate(false);
1884		}
1885
#	Line 1772 | Line 1901 | public class ForkJoinPool extends Abstra
1901		*/
1902		public List<Runnable> shutdownNow() {
1903		checkPermission();
1904	+	shutdown = true;
1905		tryTerminate(true);
1906		return Collections.emptyList();
1907		}
#	Line 1782 | Line 1912 | public class ForkJoinPool extends Abstra
1912		* @return {@code true} if all tasks have completed following shut down
1913		*/
1914		public boolean isTerminated() {
1915	<	return runState >= TERMINATED;
1915	>	long c = ctl;
1916	>	return ((c & STOP_BIT) != 0L &&
1917	>	(short)(c >>> TC_SHIFT) == -parallelism);
1918		}
1919
1920		/**
#	Line 1790 | Line 1922 | public class ForkJoinPool extends Abstra
1922		* commenced but not yet completed.  This method may be useful for
1923		* debugging. A return of {@code true} reported a sufficient
1924		* period after shutdown may indicate that submitted tasks have
1925	<	* ignored or suppressed interruption, causing this executor not
1926	<	* to properly terminate.
1925	>	* ignored or suppressed interruption, or are waiting for IO,
1926	>	* causing this executor not to properly terminate. (See the
1927	>	* advisory notes for class {@link ForkJoinTask} stating that
1928	>	* tasks should not normally entail blocking operations.  But if
1929	>	* they do, they must abort them on interrupt.)
1930		*
1931		* @return {@code true} if terminating but not yet terminated
1932		*/
1933		public boolean isTerminating() {
1934	<	return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1934	>	long c = ctl;
1935	>	return ((c & STOP_BIT) != 0L &&
1936	>	(short)(c >>> TC_SHIFT) != -parallelism);
1937	>	}
1938	>
1939	>	/**
1940	>	* Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1941	>	*/
1942	>	final boolean isAtLeastTerminating() {
1943	>	return (ctl & STOP_BIT) != 0L;
1944		}
1945
1946		/**
#	Line 1805 | Line 1949 | public class ForkJoinPool extends Abstra
1949		* @return {@code true} if this pool has been shut down
1950		*/
1951		public boolean isShutdown() {
1952	<	return runState >= SHUTDOWN;
1952	>	return shutdown;
1953		}
1954
1955		/**
#	Line 1821 | Line 1965 | public class ForkJoinPool extends Abstra
1965		*/
1966		public boolean awaitTermination(long timeout, TimeUnit unit)
1967		throws InterruptedException {
1968	<	return terminationLatch.await(timeout, unit);
1968	>	long nanos = unit.toNanos(timeout);
1969	>	final ReentrantLock lock = this.submissionLock;
1970	>	lock.lock();
1971	>	try {
1972	>	for (;;) {
1973	>	if (isTerminated())
1974	>	return true;
1975	>	if (nanos <= 0)
1976	>	return false;
1977	>	nanos = termination.awaitNanos(nanos);
1978	>	}
1979	>	} finally {
1980	>	lock.unlock();
1981	>	}
1982		}
1983
1984		/**
1985		* Interface for extending managed parallelism for tasks running
1986		* in {@link ForkJoinPool}s.
1987		*
1988	<	* <p>A {@code ManagedBlocker} provides two methods.
1989	<	* Method {@code isReleasable} must return {@code true} if
1990	<	* blocking is not necessary. Method {@code block} blocks the
1991	<	* current thread if necessary (perhaps internally invoking
1992	<	* {@code isReleasable} before actually blocking).
1988	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
1989	>	* {@code isReleasable} must return {@code true} if blocking is
1990	>	* not necessary. Method {@code block} blocks the current thread
1991	>	* if necessary (perhaps internally invoking {@code isReleasable}
1992	>	* before actually blocking). These actions are performed by any
1993	>	* thread invoking {@link ForkJoinPool#managedBlock}.  The
1994	>	* unusual methods in this API accommodate synchronizers that may,
1995	>	* but don't usually, block for long periods. Similarly, they
1996	>	* allow more efficient internal handling of cases in which
1997	>	* additional workers may be, but usually are not, needed to
1998	>	* ensure sufficient parallelism.  Toward this end,
1999	>	* implementations of method {@code isReleasable} must be amenable
2000	>	* to repeated invocation.
2001		*
2002		* <p>For example, here is a ManagedBlocker based on a
2003		* ReentrantLock:
#	Line 1850 | Line 2015 | public class ForkJoinPool extends Abstra
2015		*     return hasLock || (hasLock = lock.tryLock());
2016		*   }
2017		* }}</pre>
2018	+	*
2019	+	* <p>Here is a class that possibly blocks waiting for an
2020	+	* item on a given queue:
2021	+	*  <pre> {@code
2022	+	* class QueueTaker<E> implements ManagedBlocker {
2023	+	*   final BlockingQueue<E> queue;
2024	+	*   volatile E item = null;
2025	+	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
2026	+	*   public boolean block() throws InterruptedException {
2027	+	*     if (item == null)
2028	+	*       item = queue.take();
2029	+	*     return true;
2030	+	*   }
2031	+	*   public boolean isReleasable() {
2032	+	*     return item != null || (item = queue.poll()) != null;
2033	+	*   }
2034	+	*   public E getItem() { // call after pool.managedBlock completes
2035	+	*     return item;
2036	+	*   }
2037	+	* }}</pre>
2038		*/
2039		public static interface ManagedBlocker {
2040		/**
#	Line 1873 | Line 2058 | public class ForkJoinPool extends Abstra
2058		* Blocks in accord with the given blocker.  If the current thread
2059		* is a {@link ForkJoinWorkerThread}, this method possibly
2060		* arranges for a spare thread to be activated if necessary to
2061	<	* ensure parallelism while the current thread is blocked.
1877	<	*
1878	<	* <p>If {@code maintainParallelism} is {@code true} and the pool
1879	<	* supports it ({@link #getMaintainsParallelism}), this method
1880	<	* attempts to maintain the pool's nominal parallelism. Otherwise
1881	<	* it activates a thread only if necessary to avoid complete
1882	<	* starvation. This option may be preferable when blockages use
1883	<	* timeouts, or are almost always brief.
2061	>	* ensure sufficient parallelism while the current thread is blocked.
2062		*
2063		* <p>If the caller is not a {@link ForkJoinTask}, this method is
2064		* behaviorally equivalent to
#	Line 1894 | Line 2072 | public class ForkJoinPool extends Abstra
2072		* first be expanded to ensure parallelism, and later adjusted.
2073		*
2074		* @param blocker the blocker
1897	–	* @param maintainParallelism if {@code true} and supported by
1898	–	* this pool, attempt to maintain the pool's nominal parallelism;
1899	–	* otherwise activate a thread only if necessary to avoid
1900	–	* complete starvation.
2075		* @throws InterruptedException if blocker.block did so
2076		*/
2077	<	public static void managedBlock(ManagedBlocker blocker,
1904	<	boolean maintainParallelism)
2077	>	public static void managedBlock(ManagedBlocker blocker)
2078		throws InterruptedException {
2079		Thread t = Thread.currentThread();
2080	<	if (t instanceof ForkJoinWorkerThread)
2081	<	((ForkJoinWorkerThread) t).pool.
2082	<	doBlock(blocker, maintainParallelism);
2083	<	else
2084	<	awaitBlocker(blocker);
2085	<	}
2086	<
1914	<	/**
1915	<	* Performs Non-FJ blocking
1916	<	*/
1917	<	private static void awaitBlocker(ManagedBlocker blocker)
1918	<	throws InterruptedException {
1919	<	do {} while (!blocker.isReleasable() && !blocker.block());
2080	>	if (t instanceof ForkJoinWorkerThread) {
2081	>	ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
2082	>	w.pool.awaitBlocker(blocker);
2083	>	}
2084	>	else {
2085	>	do {} while (!blocker.isReleasable() && !blocker.block());
2086	>	}
2087		}
2088
2089		// AbstractExecutorService overrides.  These rely on undocumented
#	Line 1932 | Line 2099 | public class ForkJoinPool extends Abstra
2099		}
2100
2101		// Unsafe mechanics
2102	<
2103	<	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
2104	<	private static final long workerCountsOffset =
2105	<	objectFieldOffset("workerCounts", ForkJoinPool.class);
2106	<	private static final long runStateOffset =
2107	<	objectFieldOffset("runState", ForkJoinPool.class);
2108	<	private static final long eventCountOffset =
2109	<	objectFieldOffset("eventCount", ForkJoinPool.class);
2110	<	private static final long eventWaitersOffset =
2111	<	objectFieldOffset("eventWaiters",ForkJoinPool.class);
2112	<	private static final long stealCountOffset =
2113	<	objectFieldOffset("stealCount",ForkJoinPool.class);
2114	<
2115	<
2116	<	private static long objectFieldOffset(String field, Class<?> klazz) {
2102	>	private static final sun.misc.Unsafe UNSAFE;
2103	>	private static final long ctlOffset;
2104	>	private static final long stealCountOffset;
2105	>	private static final long blockedCountOffset;
2106	>	private static final long quiescerCountOffset;
2107	>	private static final long scanGuardOffset;
2108	>	private static final long nextWorkerNumberOffset;
2109	>	private static final long ABASE;
2110	>	private static final int ASHIFT;
2111	>
2112	>	static {
2113	>	poolNumberGenerator = new AtomicInteger();
2114	>	workerSeedGenerator = new Random();
2115	>	modifyThreadPermission = new RuntimePermission("modifyThread");
2116	>	defaultForkJoinWorkerThreadFactory =
2117	>	new DefaultForkJoinWorkerThreadFactory();
2118		try {
2119	<	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
2120	<	} catch (NoSuchFieldException e) {
2121	<	// Convert Exception to corresponding Error
2122	<	NoSuchFieldError error = new NoSuchFieldError(field);
2123	<	error.initCause(e);
2124	<	throw error;
2125	<	}
2119	>	UNSAFE = getUnsafe();
2120	>	Class<?> k = ForkJoinPool.class;
2121	>	ctlOffset = UNSAFE.objectFieldOffset
2122	>	(k.getDeclaredField("ctl"));
2123	>	stealCountOffset = UNSAFE.objectFieldOffset
2124	>	(k.getDeclaredField("stealCount"));
2125	>	blockedCountOffset = UNSAFE.objectFieldOffset
2126	>	(k.getDeclaredField("blockedCount"));
2127	>	quiescerCountOffset = UNSAFE.objectFieldOffset
2128	>	(k.getDeclaredField("quiescerCount"));
2129	>	scanGuardOffset = UNSAFE.objectFieldOffset
2130	>	(k.getDeclaredField("scanGuard"));
2131	>	nextWorkerNumberOffset = UNSAFE.objectFieldOffset
2132	>	(k.getDeclaredField("nextWorkerNumber"));
2133	>	} catch (Exception e) {
2134	>	throw new Error(e);
2135	>	}
2136	>	Class<?> a = ForkJoinTask[].class;
2137	>	ABASE = UNSAFE.arrayBaseOffset(a);
2138	>	int s = UNSAFE.arrayIndexScale(a);
2139	>	if ((s & (s-1)) != 0)
2140	>	throw new Error("data type scale not a power of two");
2141	>	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
2142		}
2143
2144		/**

Diff Legend

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