ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/ForkJoinPool.java
(Generate patch)

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.41 by jsr166, Mon Aug 3 01:11:58 2009 UTC vs.
Revision 1.80 by dl, Fri Sep 17 14:24:56 2010 UTC

# Line 6 | Line 6
6  
7   package jsr166y;
8  
9 import java.util.concurrent.*;
10
9   import java.util.ArrayList;
10   import java.util.Arrays;
11   import java.util.Collection;
12   import java.util.Collections;
13   import java.util.List;
14 < import java.util.concurrent.locks.Condition;
14 > import java.util.concurrent.AbstractExecutorService;
15 > import java.util.concurrent.Callable;
16 > import java.util.concurrent.ExecutorService;
17 > import java.util.concurrent.Future;
18 > import java.util.concurrent.RejectedExecutionException;
19 > import java.util.concurrent.RunnableFuture;
20 > import java.util.concurrent.TimeUnit;
21 > import java.util.concurrent.TimeoutException;
22 > import java.util.concurrent.atomic.AtomicInteger;
23   import java.util.concurrent.locks.LockSupport;
24   import java.util.concurrent.locks.ReentrantLock;
19 import java.util.concurrent.atomic.AtomicInteger;
20 import java.util.concurrent.atomic.AtomicLong;
25  
26   /**
27   * An {@link ExecutorService} for running {@link ForkJoinTask}s.
28   * A {@code ForkJoinPool} provides the entry point for submissions
29 < * from non-{@code ForkJoinTask}s, as well as management and
30 < * monitoring operations.  Normally a single {@code ForkJoinPool} is
27 < * used for a large number of submitted tasks. Otherwise, use would
28 < * not usually outweigh the construction and bookkeeping overhead of
29 < * creating a large set of threads.
29 > * from non-{@code ForkJoinTask} clients, as well as management and
30 > * monitoring operations.
31   *
32 < * <p>{@code ForkJoinPool}s differ from other kinds of {@link
33 < * Executor}s mainly in that they provide <em>work-stealing</em>: all
34 < * threads in the pool attempt to find and execute subtasks created by
35 < * other active tasks (eventually blocking if none exist). This makes
36 < * them efficient when most tasks spawn other subtasks (as do most
37 < * {@code ForkJoinTask}s), as well as the mixed execution of some
38 < * plain {@code Runnable}- or {@code Callable}- based activities along
39 < * with {@code ForkJoinTask}s. When setting {@linkplain #setAsyncMode
40 < * async mode}, a {@code ForkJoinPool} may also be appropriate for use
40 < * with fine-grained tasks that are never joined. Otherwise, other
41 < * {@code ExecutorService} implementations are typically more
42 < * appropriate choices.
32 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
33 > * ExecutorService} mainly by virtue of employing
34 > * <em>work-stealing</em>: all threads in the pool attempt to find and
35 > * execute subtasks created by other active tasks (eventually blocking
36 > * waiting for work if none exist). This enables efficient processing
37 > * when most tasks spawn other subtasks (as do most {@code
38 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
39 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
40 > * with event-style tasks that are never joined.
41   *
42 < * <p>A {@code ForkJoinPool} may be constructed with a given
43 < * parallelism level (target pool size), which it attempts to maintain
44 < * by dynamically adding, suspending, or resuming threads, even if
45 < * some tasks are waiting to join others. However, no such adjustments
46 < * are performed in the face of blocked IO or other unmanaged
47 < * synchronization. The nested {@link ManagedBlocker} interface
48 < * enables extension of the kinds of synchronization accommodated.
49 < * The target parallelism level may also be changed dynamically
50 < * ({@link #setParallelism}) and thread construction can be limited
53 < * using methods {@link #setMaximumPoolSize} and/or {@link
54 < * #setMaintainsParallelism}.
42 > * <p>A {@code ForkJoinPool} is constructed with a given target
43 > * parallelism level; by default, equal to the number of available
44 > * processors. The pool attempts to maintain enough active (or
45 > * available) threads by dynamically adding, suspending, or resuming
46 > * internal worker threads, even if some tasks are stalled waiting to
47 > * join others. However, no such adjustments are guaranteed in the
48 > * face of blocked IO or other unmanaged synchronization. The nested
49 > * {@link ManagedBlocker} interface enables extension of the kinds of
50 > * synchronization accommodated.
51   *
52   * <p>In addition to execution and lifecycle control methods, this
53   * class provides status check methods (for example
# Line 60 | Line 56 | import java.util.concurrent.atomic.Atomi
56   * {@link #toString} returns indications of pool state in a
57   * convenient form for informal monitoring.
58   *
59 + * <p> As is the case with other ExecutorServices, there are three
60 + * main task execution methods summarized in the following
61 + * table. These are designed to be used by clients not already engaged
62 + * in fork/join computations in the current pool.  The main forms of
63 + * these methods accept instances of {@code ForkJoinTask}, but
64 + * overloaded forms also allow mixed execution of plain {@code
65 + * Runnable}- or {@code Callable}- based activities as well.  However,
66 + * tasks that are already executing in a pool should normally
67 + * <em>NOT</em> use these pool execution methods, but instead use the
68 + * within-computation forms listed in the table.
69 + *
70 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
71 + *  <tr>
72 + *    <td></td>
73 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
74 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
75 + *  </tr>
76 + *  <tr>
77 + *    <td> <b>Arrange async execution</td>
78 + *    <td> {@link #execute(ForkJoinTask)}</td>
79 + *    <td> {@link ForkJoinTask#fork}</td>
80 + *  </tr>
81 + *  <tr>
82 + *    <td> <b>Await and obtain result</td>
83 + *    <td> {@link #invoke(ForkJoinTask)}</td>
84 + *    <td> {@link ForkJoinTask#invoke}</td>
85 + *  </tr>
86 + *  <tr>
87 + *    <td> <b>Arrange exec and obtain Future</td>
88 + *    <td> {@link #submit(ForkJoinTask)}</td>
89 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
90 + *  </tr>
91 + * </table>
92 + *
93 + * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
94 + * used for all parallel task execution in a program or subsystem.
95 + * Otherwise, use would not usually outweigh the construction and
96 + * bookkeeping overhead of creating a large set of threads. For
97 + * example, a common pool could be used for the {@code SortTasks}
98 + * illustrated in {@link RecursiveAction}. Because {@code
99 + * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
100 + * daemon} mode, there is typically no need to explicitly {@link
101 + * #shutdown} such a pool upon program exit.
102 + *
103 + * <pre>
104 + * static final ForkJoinPool mainPool = new ForkJoinPool();
105 + * ...
106 + * public void sort(long[] array) {
107 + *   mainPool.invoke(new SortTask(array, 0, array.length));
108 + * }
109 + * </pre>
110 + *
111   * <p><b>Implementation notes</b>: This implementation restricts the
112   * maximum number of running threads to 32767. Attempts to create
113 < * pools with greater than the maximum result in
113 > * pools with greater than the maximum number result in
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
118 + * or internal resources have been exhausted.
119 + *
120   * @since 1.7
121   * @author Doug Lea
122   */
123   public class ForkJoinPool extends AbstractExecutorService {
124  
125      /*
126 <     * See the extended comments interspersed below for design,
127 <     * rationale, and walkthroughs.
126 >     * Implementation Overview
127 >     *
128 >     * This class provides the central bookkeeping and control for a
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
136 >     * lastly to new submissions. These mechanics do not consider
137 >     * affinities, loads, cache localities, etc, so rarely provide the
138 >     * best possible performance on a given machine, but portably
139 >     * provide good throughput by averaging over these factors.
140 >     * (Further, even if we did try to use such information, we do not
141 >     * usually have a basis for exploiting it. For example, some sets
142 >     * of tasks profit from cache affinities, but others are harmed by
143 >     * cache pollution effects.)
144 >     *
145 >     * Beyond work-stealing support and essential bookkeeping, the
146 >     * main responsibility of this framework is to take actions when
147 >     * one worker is waiting to join a task stolen (or always held by)
148 >     * another.  Because we are multiplexing many tasks on to a pool
149 >     * of workers, we can't just let them block (as in Thread.join).
150 >     * We also cannot just reassign the joiner's run-time stack with
151 >     * another and replace it later, which would be a form of
152 >     * "continuation", that even if possible is not necessarily a good
153 >     * idea. Given that the creation costs of most threads on most
154 >     * systems mainly surrounds setting up runtime stacks, thread
155 >     * creation and switching is usually not much more expensive than
156 >     * stack creation and switching, and is more flexible). Instead we
157 >     * combine two tactics:
158 >     *
159 >     *   Helping: Arranging for the joiner to execute some task that it
160 >     *      would be running if the steal had not occurred.  Method
161 >     *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
162 >     *      links to try to find such a task.
163 >     *
164 >     *   Compensating: Unless there are already enough live threads,
165 >     *      method helpMaintainParallelism() may create or
166 >     *      re-activate a spare thread to compensate for blocked
167 >     *      joiners until they unblock.
168 >     *
169 >     * It is impossible to keep exactly the target (parallelism)
170 >     * number of threads running at any given time.  Determining
171 >     * existence of conservatively safe helping targets, the
172 >     * availability of already-created spares, and the apparent need
173 >     * to create new spares are all racy and require heuristic
174 >     * guidance, so we rely on multiple retries of each.  Compensation
175 >     * occurs in slow-motion. It is triggered only upon timeouts of
176 >     * Object.wait used for joins. This reduces poor decisions that
177 >     * would otherwise be made when threads are waiting for others
178 >     * that are stalled because of unrelated activities such as
179 >     * garbage collection.
180 >     *
181 >     * The ManagedBlocker extension API can't use helping so relies
182 >     * only on compensation in method awaitBlocker.
183 >     *
184 >     * The main throughput advantages of work-stealing stem from
185 >     * decentralized control -- workers mostly steal tasks from each
186 >     * other. We do not want to negate this by creating bottlenecks
187 >     * implementing other management responsibilities. So we use a
188 >     * collection of techniques that avoid, reduce, or cope well with
189 >     * contention. These entail several instances of bit-packing into
190 >     * CASable fields to maintain only the minimally required
191 >     * atomicity. To enable such packing, we restrict maximum
192 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
193 >     * unbalanced increments and decrements) to fit into a 16 bit
194 >     * field, which is far in excess of normal operating range.  Even
195 >     * though updates to some of these bookkeeping fields do sometimes
196 >     * contend with each other, they don't normally cache-contend with
197 >     * updates to others enough to warrant memory padding or
198 >     * isolation. So they are all held as fields of ForkJoinPool
199 >     * objects.  The main capabilities are as follows:
200 >     *
201 >     * 1. Creating and removing workers. Workers are recorded in the
202 >     * "workers" array. This is an array as opposed to some other data
203 >     * structure to support index-based random steals by workers.
204 >     * Updates to the array recording new workers and unrecording
205 >     * terminated ones are protected from each other by a lock
206 >     * (workerLock) but the array is otherwise concurrently readable,
207 >     * and accessed directly by workers. To simplify index-based
208 >     * operations, the array size is always a power of two, and all
209 >     * readers must tolerate null slots. Currently, all worker thread
210 >     * creation is on-demand, triggered by task submissions,
211 >     * replacement of terminated workers, and/or compensation for
212 >     * blocked workers. However, all other support code is set up to
213 >     * work with other policies.
214 >     *
215 >     * To ensure that we do not hold on to worker references that
216 >     * would prevent GC, ALL accesses to workers are via indices into
217 >     * the workers array (which is one source of some of the unusual
218 >     * code constructions here). In essence, the workers array serves
219 >     * as a WeakReference mechanism. Thus for example the event queue
220 >     * stores worker indices, not worker references. Access to the
221 >     * workers in associated methods (for example releaseEventWaiters)
222 >     * must both index-check and null-check the IDs. All such accesses
223 >     * ignore bad IDs by returning out early from what they are doing,
224 >     * since this can only be associated with shutdown, in which case
225 >     * it is OK to give up. On termination, we just clobber these
226 >     * data structures without trying to use them.
227 >     *
228 >     * 2. Bookkeeping for dynamically adding and removing workers. We
229 >     * aim to approximately maintain the given level of parallelism.
230 >     * When some workers are known to be blocked (on joins or via
231 >     * ManagedBlocker), we may create or resume others to take their
232 >     * place until they unblock (see below). Implementing this
233 >     * requires counts of the number of "running" threads (i.e., those
234 >     * that are neither blocked nor artificially suspended) as well as
235 >     * the total number.  These two values are packed into one field,
236 >     * "workerCounts" because we need accurate snapshots when deciding
237 >     * to create, resume or suspend.  Note however that the
238 >     * correspondence of these counts to reality is not guaranteed. In
239 >     * particular updates for unblocked threads may lag until they
240 >     * actually wake up.
241 >     *
242 >     * 3. Maintaining global run state. The run state of the pool
243 >     * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
244 >     * those in other Executor implementations, as well as a count of
245 >     * "active" workers -- those that are, or soon will be, or
246 >     * recently were executing tasks. The runLevel and active count
247 >     * are packed together in order to correctly trigger shutdown and
248 >     * termination. Without care, active counts can be subject to very
249 >     * high contention.  We substantially reduce this contention by
250 >     * relaxing update rules.  A worker must claim active status
251 >     * prospectively, by activating if it sees that a submitted or
252 >     * stealable task exists (it may find after activating that the
253 >     * task no longer exists). It stays active while processing this
254 >     * task (if it exists) and any other local subtasks it produces,
255 >     * until it cannot find any other tasks. It then tries
256 >     * inactivating (see method preStep), but upon update contention
257 >     * instead scans for more tasks, later retrying inactivation if it
258 >     * doesn't find any.
259 >     *
260 >     * 4. Managing idle workers waiting for tasks. We cannot let
261 >     * workers spin indefinitely scanning for tasks when none are
262 >     * available. On the other hand, we must quickly prod them into
263 >     * action when new tasks are submitted or generated.  We
264 >     * park/unpark these idle workers using an event-count scheme.
265 >     * Field eventCount is incremented upon events that may enable
266 >     * workers that previously could not find a task to now find one:
267 >     * Submission of a new task to the pool, or another worker pushing
268 >     * a task onto a previously empty queue.  (We also use this
269 >     * mechanism for configuration and termination actions that
270 >     * require wakeups of idle workers).  Each worker maintains its
271 >     * last known event count, and blocks when a scan for work did not
272 >     * find a task AND its lastEventCount matches the current
273 >     * eventCount. Waiting idle workers are recorded in a variant of
274 >     * Treiber stack headed by field eventWaiters which, when nonzero,
275 >     * encodes the thread index and count awaited for by the worker
276 >     * thread most recently calling eventSync. This thread in turn has
277 >     * a record (field nextEventWaiter) for the next waiting worker.
278 >     * In addition to allowing simpler decisions about need for
279 >     * wakeup, the event count bits in eventWaiters serve the role of
280 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
281 >     * released threads also try to release at most two others.  The
282 >     * net effect is a tree-like diffusion of signals, where released
283 >     * threads (and possibly others) help with unparks.  To further
284 >     * reduce contention effects a bit, failed CASes to increment
285 >     * field eventCount are tolerated without retries in signalWork.
286 >     * Conceptually they are merged into the same event, which is OK
287 >     * when their only purpose is to enable workers to scan for work.
288 >     *
289 >     * 5. Managing suspension of extra workers. When a worker notices
290 >     * (usually upon timeout of a wait()) that there are too few
291 >     * running threads, we may create a new thread to maintain
292 >     * parallelism level, or at least avoid starvation. Usually, extra
293 >     * threads are needed for only very short periods, yet join
294 >     * dependencies are such that we sometimes need them in
295 >     * bursts. Rather than create new threads each time this happens,
296 >     * we suspend no-longer-needed extra ones as "spares". For most
297 >     * purposes, we don't distinguish "extra" spare threads from
298 >     * normal "core" threads: On each call to preStep (the only point
299 >     * at which we can do this) a worker checks to see if there are
300 >     * now too many running workers, and if so, suspends itself.
301 >     * Method helpMaintainParallelism looks for suspended threads to
302 >     * resume before considering creating a new replacement. The
303 >     * spares themselves are encoded on another variant of a Treiber
304 >     * Stack, headed at field "spareWaiters".  Note that the use of
305 >     * spares is intrinsically racy.  One thread may become a spare at
306 >     * about the same time as another is needlessly being created. We
307 >     * counteract this and related slop in part by requiring resumed
308 >     * spares to immediately recheck (in preStep) to see whether they
309 >     * should re-suspend.
310 >     *
311 >     * 6. Killing off unneeded workers. A timeout mechanism is used to
312 >     * shed unused workers: The oldest (first) event queue waiter uses
313 >     * a timed rather than hard wait. When this wait times out without
314 >     * a normal wakeup, it tries to shutdown any one (for convenience
315 >     * the newest) other spare or event waiter via
316 >     * tryShutdownUnusedWorker. This eventually reduces the number of
317 >     * worker threads to a minimum of one after a long enough period
318 >     * without use.
319 >     *
320 >     * 7. Deciding when to create new workers. The main dynamic
321 >     * control in this class is deciding when to create extra threads
322 >     * in method helpMaintainParallelism. We would like to keep
323 >     * exactly #parallelism threads running, which is an impossible
324 >     * task. We always need to create one when the number of running
325 >     * threads would become zero and all workers are busy. Beyond
326 >     * this, we must rely on heuristics that work well in the
327 >     * presence of transient phenomena such as GC stalls, dynamic
328 >     * compilation, and wake-up lags. These transients are extremely
329 >     * common -- we are normally trying to fully saturate the CPUs on
330 >     * a machine, so almost any activity other than running tasks
331 >     * impedes accuracy. Our main defense is to allow parallelism to
332 >     * lapse for a while during joins, and use a timeout to see if,
333 >     * after the resulting settling, there is still a need for
334 >     * additional workers.  This also better copes with the fact that
335 >     * some of the methods in this class tend to never become compiled
336 >     * (but are interpreted), so some components of the entire set of
337 >     * controls might execute 100 times faster than others. And
338 >     * similarly for cases where the apparent lack of work is just due
339 >     * to GC stalls and other transient system activity.
340 >     *
341 >     * Beware that there is a lot of representation-level coupling
342 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
343 >     * ForkJoinTask.  For example, direct access to "workers" array by
344 >     * workers, and direct access to ForkJoinTask.status by both
345 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
346 >     * trying to reduce this, since any associated future changes in
347 >     * representations will need to be accompanied by algorithmic
348 >     * changes anyway.
349 >     *
350 >     * Style notes: There are lots of inline assignments (of form
351 >     * "while ((local = field) != 0)") which are usually the simplest
352 >     * way to ensure the required read orderings (which are sometimes
353 >     * critical). Also several occurrences of the unusual "do {}
354 >     * while (!cas...)" which is the simplest way to force an update of
355 >     * a CAS'ed variable. There are also other coding oddities that
356 >     * help some methods perform reasonably even when interpreted (not
357 >     * compiled), at the expense of some messy constructions that
358 >     * reduce byte code counts.
359 >     *
360 >     * The order of declarations in this file is: (1) statics (2)
361 >     * fields (along with constants used when unpacking some of them)
362 >     * (3) internal control methods (4) callbacks and other support
363 >     * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
364 >     * methods (plus a few little helpers).
365       */
366  
78    /** Mask for packing and unpacking shorts */
79    private static final int  shortMask = 0xffff;
80
81    /** Max pool size -- must be a power of two minus 1 */
82    private static final int MAX_THREADS =  0x7FFF;
83
367      /**
368       * Factory for creating new {@link ForkJoinWorkerThread}s.
369       * A {@code ForkJoinWorkerThreadFactory} must be defined and used
# Line 92 | Line 375 | public class ForkJoinPool extends Abstra
375           * Returns a new worker thread operating in the given pool.
376           *
377           * @param pool the pool this thread works in
378 <         * @throws NullPointerException if pool is null
378 >         * @throws NullPointerException if the pool is null
379           */
380          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
381      }
# Line 101 | Line 384 | public class ForkJoinPool extends Abstra
384       * Default ForkJoinWorkerThreadFactory implementation; creates a
385       * new ForkJoinWorkerThread.
386       */
387 <    static class  DefaultForkJoinWorkerThreadFactory
387 >    static class DefaultForkJoinWorkerThreadFactory
388          implements ForkJoinWorkerThreadFactory {
389          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
390 <            try {
108 <                return new ForkJoinWorkerThread(pool);
109 <            } catch (OutOfMemoryError oom)  {
110 <                return null;
111 <            }
390 >            return new ForkJoinWorkerThread(pool);
391          }
392      }
393  
# Line 144 | Line 423 | public class ForkJoinPool extends Abstra
423          new AtomicInteger();
424  
425      /**
426 <     * Array holding all worker threads in the pool. Initialized upon
427 <     * first use. Array size must be a power of two.  Updates and
428 <     * replacements are protected by workerLock, but it is always kept
429 <     * in a consistent enough state to be randomly accessed without
430 <     * locking by workers performing work-stealing.
426 >     * The time to block in a join (see awaitJoin) before checking if
427 >     * a new worker should be (re)started to maintain parallelism
428 >     * level. The value should be short enough to maintain global
429 >     * responsiveness and progress but long enough to avoid
430 >     * counterproductive firings during GC stalls or unrelated system
431 >     * activity, and to not bog down systems with continual re-firings
432 >     * on GCs or legitimately long waits.
433 >     */
434 >    private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
435 >
436 >    /**
437 >     * The wakeup interval (in nanoseconds) for the oldest worker
438 >     * waiting for an event to invoke tryShutdownUnusedWorker to
439 >     * shrink the number of workers.  The exact value does not matter
440 >     * too much. It must be short enough to release resources during
441 >     * sustained periods of idleness, but not so short that threads
442 >     * are continually re-created.
443 >     */
444 >    private static final long SHRINK_RATE_NANOS =
445 >        30L * 1000L * 1000L * 1000L; // 2 per minute
446 >
447 >    /**
448 >     * Absolute bound for parallelism level. Twice this number plus
449 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
450 >     * word-packing for some counts and indices.
451 >     */
452 >    private static final int MAX_WORKERS   = 0x7fff;
453 >
454 >    /**
455 >     * Array holding all worker threads in the pool.  Array size must
456 >     * be a power of two.  Updates and replacements are protected by
457 >     * workerLock, but the array is always kept in a consistent enough
458 >     * state to be randomly accessed without locking by workers
459 >     * performing work-stealing, as well as other traversal-based
460 >     * methods in this class. All readers must tolerate that some
461 >     * array slots may be null.
462       */
463      volatile ForkJoinWorkerThread[] workers;
464  
465      /**
466 <     * Lock protecting access to workers.
466 >     * Queue for external submissions.
467       */
468 <    private final ReentrantLock workerLock;
468 >    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
469  
470      /**
471 <     * Condition for awaitTermination.
471 >     * Lock protecting updates to workers array.
472       */
473 <    private final Condition termination;
473 >    private final ReentrantLock workerLock;
474  
475      /**
476 <     * The uncaught exception handler used when any worker
167 <     * abruptly terminates
476 >     * Latch released upon termination.
477       */
478 <    private Thread.UncaughtExceptionHandler ueh;
478 >    private final Phaser termination;
479  
480      /**
481       * Creation factory for worker threads.
# Line 174 | Line 483 | public class ForkJoinPool extends Abstra
483      private final ForkJoinWorkerThreadFactory factory;
484  
485      /**
177     * Head of stack of threads that were created to maintain
178     * parallelism when other threads blocked, but have since
179     * suspended when the parallelism level rose.
180     */
181    private volatile WaitQueueNode spareStack;
182
183    /**
486       * Sum of per-thread steal counts, updated only when threads are
487       * idle or terminating.
488       */
489 <    private final AtomicLong stealCount;
489 >    private volatile long stealCount;
490  
491      /**
492 <     * Queue for external submissions.
492 >     * Encoded record of top of Treiber stack of threads waiting for
493 >     * events. The top 32 bits contain the count being waited for. The
494 >     * bottom 16 bits contains one plus the pool index of waiting
495 >     * worker thread. (Bits 16-31 are unused.)
496       */
497 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
497 >    private volatile long eventWaiters;
498  
499 <    /**
500 <     * Head of Treiber stack for barrier sync. See below for explanation.
196 <     */
197 <    private volatile WaitQueueNode syncStack;
499 >    private static final int  EVENT_COUNT_SHIFT = 32;
500 >    private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
501  
502      /**
503 <     * The count for event barrier
503 >     * A counter for events that may wake up worker threads:
504 >     *   - Submission of a new task to the pool
505 >     *   - A worker pushing a task on an empty queue
506 >     *   - termination
507       */
508 <    private volatile long eventCount;
508 >    private volatile int eventCount;
509  
510      /**
511 <     * Pool number, just for assigning useful names to worker threads
511 >     * Encoded record of top of Treiber stack of spare threads waiting
512 >     * for resumption. The top 16 bits contain an arbitrary count to
513 >     * avoid ABA effects. The bottom 16bits contains one plus the pool
514 >     * index of waiting worker thread.
515       */
516 <    private final int poolNumber;
516 >    private volatile int spareWaiters;
517  
518 <    /**
519 <     * The maximum allowed pool size
211 <     */
212 <    private volatile int maxPoolSize;
518 >    private static final int SPARE_COUNT_SHIFT = 16;
519 >    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
520  
521      /**
522 <     * The desired parallelism level, updated only under workerLock.
522 >     * Lifecycle control. The low word contains the number of workers
523 >     * that are (probably) executing tasks. This value is atomically
524 >     * incremented before a worker gets a task to run, and decremented
525 >     * when a worker has no tasks and cannot find any.  Bits 16-18
526 >     * contain runLevel value. When all are zero, the pool is
527 >     * running. Level transitions are monotonic (running -> shutdown
528 >     * -> terminating -> terminated) so each transition adds a bit.
529 >     * These are bundled together to ensure consistent read for
530 >     * termination checks (i.e., that runLevel is at least SHUTDOWN
531 >     * and active threads is zero).
532 >     *
533 >     * Notes: Most direct CASes are dependent on these bitfield
534 >     * positions.  Also, this field is non-private to enable direct
535 >     * performance-sensitive CASes in ForkJoinWorkerThread.
536       */
537 <    private volatile int parallelism;
537 >    volatile int runState;
538  
539 <    /**
540 <     * True if use local fifo, not default lifo, for local polling
541 <     */
542 <    private volatile boolean locallyFifo;
539 >    // Note: The order among run level values matters.
540 >    private static final int RUNLEVEL_SHIFT     = 16;
541 >    private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
542 >    private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
543 >    private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
544 >    private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
545  
546      /**
547       * Holds number of total (i.e., created and not yet terminated)
548       * and running (i.e., not blocked on joins or other managed sync)
549 <     * threads, packed into one int to ensure consistent snapshot when
549 >     * threads, packed together to ensure consistent snapshot when
550       * making decisions about creating and suspending spare
551 <     * threads. Updated only by CAS.  Note: CASes in
552 <     * updateRunningCount and preJoin assume that running active count
553 <     * is in low word, so need to be modified if this changes.
551 >     * threads. Updated only by CAS. Note that adding a new worker
552 >     * requires incrementing both counts, since workers start off in
553 >     * running state.
554       */
555      private volatile int workerCounts;
556  
557 <    private static int totalCountOf(int s)           { return s >>> 16;  }
558 <    private static int runningCountOf(int s)         { return s & shortMask; }
559 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
557 >    private static final int TOTAL_COUNT_SHIFT  = 16;
558 >    private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
559 >    private static final int ONE_RUNNING        = 1;
560 >    private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
561  
562      /**
563 <     * Adds delta (which may be negative) to running count.  This must
564 <     * be called before (with negative arg) and after (with positive)
242 <     * any managed synchronization (i.e., mainly, joins).
243 <     *
244 <     * @param delta the number to add
563 >     * The target parallelism level.
564 >     * Accessed directly by ForkJoinWorkerThreads.
565       */
566 <    final void updateRunningCount(int delta) {
247 <        int s;
248 <        do {} while (!casWorkerCounts(s = workerCounts, s + delta));
249 <    }
566 >    final int parallelism;
567  
568      /**
569 <     * Adds delta (which may be negative) to both total and running
570 <     * count.  This must be called upon creation and termination of
254 <     * worker threads.
255 <     *
256 <     * @param delta the number to add
569 >     * True if use local fifo, not default lifo, for local polling
570 >     * Read by, and replicated by ForkJoinWorkerThreads
571       */
572 <    private void updateWorkerCount(int delta) {
259 <        int d = delta + (delta << 16); // add to both lo and hi parts
260 <        int s;
261 <        do {} while (!casWorkerCounts(s = workerCounts, s + d));
262 <    }
572 >    final boolean locallyFifo;
573  
574      /**
575 <     * Lifecycle control. High word contains runState, low word
576 <     * contains the number of workers that are (probably) executing
267 <     * tasks. This value is atomically incremented before a worker
268 <     * gets a task to run, and decremented when worker has no tasks
269 <     * and cannot find any. These two fields are bundled together to
270 <     * support correct termination triggering.  Note: activeCount
271 <     * CAS'es cheat by assuming active count is in low word, so need
272 <     * to be modified if this changes
575 >     * The uncaught exception handler used when any worker abruptly
576 >     * terminates.
577       */
578 <    private volatile int runControl;
578 >    private final Thread.UncaughtExceptionHandler ueh;
579  
580 <    // RunState values. Order among values matters
581 <    private static final int RUNNING     = 0;
582 <    private static final int SHUTDOWN    = 1;
583 <    private static final int TERMINATING = 2;
280 <    private static final int TERMINATED  = 3;
580 >    /**
581 >     * Pool number, just for assigning useful names to worker threads
582 >     */
583 >    private final int poolNumber;
584  
585 <    private static int runStateOf(int c)             { return c >>> 16; }
586 <    private static int activeCountOf(int c)          { return c & shortMask; }
284 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
585 >    // Utilities for CASing fields. Note that most of these
586 >    // are usually manually inlined by callers
587  
588      /**
589 <     * Tries incrementing active count; fails on contention.
288 <     * Called by workers before/during executing tasks.
289 <     *
290 <     * @return true on success
589 >     * Increments running count part of workerCounts
590       */
591 <    final boolean tryIncrementActiveCount() {
592 <        int c = runControl;
593 <        return casRunControl(c, c+1);
591 >    final void incrementRunningCount() {
592 >        int c;
593 >        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
594 >                                               c = workerCounts,
595 >                                               c + ONE_RUNNING));
596      }
597  
598      /**
599 <     * Tries decrementing active count; fails on contention.
299 <     * Possibly triggers termination on success.
300 <     * Called by workers when they can't find tasks.
301 <     *
302 <     * @return true on success
599 >     * Tries to decrement running count unless already zero
600       */
601 <    final boolean tryDecrementActiveCount() {
602 <        int c = runControl;
603 <        int nextc = c - 1;
307 <        if (!casRunControl(c, nextc))
601 >    final boolean tryDecrementRunningCount() {
602 >        int wc = workerCounts;
603 >        if ((wc & RUNNING_COUNT_MASK) == 0)
604              return false;
605 <        if (canTerminateOnShutdown(nextc))
606 <            terminateOnShutdown();
311 <        return true;
605 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
606 >                                        wc, wc - ONE_RUNNING);
607      }
608  
609      /**
610 <     * Returns {@code true} if argument represents zero active count
611 <     * and nonzero runstate, which is the triggering condition for
612 <     * terminating on shutdown.
610 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
611 >     * (rarely) necessary when other count updates lag.
612 >     *
613 >     * @param dr -- either zero or ONE_RUNNING
614 >     * @param dt -- either zero or ONE_TOTAL
615       */
616 <    private static boolean canTerminateOnShutdown(int c) {
617 <        // i.e. least bit is nonzero runState bit
618 <        return ((c & -c) >>> 16) != 0;
616 >    private void decrementWorkerCounts(int dr, int dt) {
617 >        for (;;) {
618 >            int wc = workerCounts;
619 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
620 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
621 >                if ((runState & TERMINATED) != 0)
622 >                    return; // lagging termination on a backout
623 >                Thread.yield();
624 >            }
625 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
626 >                                         wc, wc - (dr + dt)))
627 >                return;
628 >        }
629      }
630  
631      /**
632 <     * Transition run state to at least the given state. Return true
633 <     * if not already at least given state.
632 >     * Tries decrementing active count; fails on contention.
633 >     * Called when workers cannot find tasks to run.
634       */
635 <    private boolean transitionRunStateTo(int state) {
635 >    final boolean tryDecrementActiveCount() {
636 >        int c;
637 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
638 >                                        c = runState, c - 1);
639 >    }
640 >
641 >    /**
642 >     * Advances to at least the given level. Returns true if not
643 >     * already in at least the given level.
644 >     */
645 >    private boolean advanceRunLevel(int level) {
646          for (;;) {
647 <            int c = runControl;
648 <            if (runStateOf(c) >= state)
647 >            int s = runState;
648 >            if ((s & level) != 0)
649                  return false;
650 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
650 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
651                  return true;
652          }
653      }
654  
655 +    // workers array maintenance
656 +
657      /**
658 <     * Controls whether to add spares to maintain parallelism
658 >     * Records and returns a workers array index for new worker.
659       */
660 <    private volatile boolean maintainsParallelism;
660 >    private int recordWorker(ForkJoinWorkerThread w) {
661 >        // Try using slot totalCount-1. If not available, scan and/or resize
662 >        int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
663 >        final ReentrantLock lock = this.workerLock;
664 >        lock.lock();
665 >        try {
666 >            ForkJoinWorkerThread[] ws = workers;
667 >            int n = ws.length;
668 >            if (k < 0 || k >= n || ws[k] != null) {
669 >                for (k = 0; k < n && ws[k] != null; ++k)
670 >                    ;
671 >                if (k == n)
672 >                    ws = Arrays.copyOf(ws, n << 1);
673 >            }
674 >            ws[k] = w;
675 >            workers = ws; // volatile array write ensures slot visibility
676 >        } finally {
677 >            lock.unlock();
678 >        }
679 >        return k;
680 >    }
681  
682 <    // Constructors
682 >    /**
683 >     * Nulls out record of worker in workers array.
684 >     */
685 >    private void forgetWorker(ForkJoinWorkerThread w) {
686 >        int idx = w.poolIndex;
687 >        // Locking helps method recordWorker avoid unnecessary expansion
688 >        final ReentrantLock lock = this.workerLock;
689 >        lock.lock();
690 >        try {
691 >            ForkJoinWorkerThread[] ws = workers;
692 >            if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
693 >                ws[idx] = null;
694 >        } finally {
695 >            lock.unlock();
696 >        }
697 >    }
698  
699      /**
700 <     * Creates a {@code ForkJoinPool} with a pool size equal to the
701 <     * number of processors available on the system, using the
702 <     * {@linkplain #defaultForkJoinWorkerThreadFactory default thread factory}.
700 >     * Final callback from terminating worker.  Removes record of
701 >     * worker from array, and adjusts counts. If pool is shutting
702 >     * down, tries to complete termination.
703       *
704 <     * @throws SecurityException if a security manager exists and
351 <     *         the caller is not permitted to modify threads
352 <     *         because it does not hold {@link
353 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
704 >     * @param w the worker
705       */
706 <    public ForkJoinPool() {
707 <        this(Runtime.getRuntime().availableProcessors(),
708 <             defaultForkJoinWorkerThreadFactory);
706 >    final void workerTerminated(ForkJoinWorkerThread w) {
707 >        forgetWorker(w);
708 >        decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
709 >        while (w.stealCount != 0) // collect final count
710 >            tryAccumulateStealCount(w);
711 >        tryTerminate(false);
712      }
713  
714 +    // Waiting for and signalling events
715 +
716      /**
717 <     * Creates a {@code ForkJoinPool} with the indicated parallelism level
718 <     * threads and using the
719 <     * {@linkplain #defaultForkJoinWorkerThreadFactory default thread factory}.
720 <     *
365 <     * @param parallelism the number of worker threads
366 <     * @throws IllegalArgumentException if parallelism less than or
367 <     * equal to zero
368 <     * @throws SecurityException if a security manager exists and
369 <     *         the caller is not permitted to modify threads
370 <     *         because it does not hold {@link
371 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
717 >     * Releases workers blocked on a count not equal to current count.
718 >     * Normally called after precheck that eventWaiters isn't zero to
719 >     * avoid wasted array checks. Gives up upon a change in count or
720 >     * upon releasing two workers, letting others take over.
721       */
722 <    public ForkJoinPool(int parallelism) {
723 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
722 >    private void releaseEventWaiters() {
723 >        ForkJoinWorkerThread[] ws = workers;
724 >        int n = ws.length;
725 >        long h = eventWaiters;
726 >        int ec = eventCount;
727 >        boolean releasedOne = false;
728 >        ForkJoinWorkerThread w; int id;
729 >        while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
730 >               (int)(h >>> EVENT_COUNT_SHIFT) != ec &&
731 >               id < n && (w = ws[id]) != null) {
732 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
733 >                                          h,  w.nextWaiter)) {
734 >                LockSupport.unpark(w);
735 >                if (releasedOne) // exit on second release
736 >                    break;
737 >                releasedOne = true;
738 >            }
739 >            if (eventCount != ec)
740 >                break;
741 >            h = eventWaiters;
742 >        }
743      }
744  
745      /**
746 <     * Creates a {@code ForkJoinPool} with parallelism equal to the
747 <     * number of processors available on the system and using the
380 <     * given thread factory.
381 <     *
382 <     * @param factory the factory for creating new threads
383 <     * @throws NullPointerException if factory is null
384 <     * @throws SecurityException if a security manager exists and
385 <     *         the caller is not permitted to modify threads
386 <     *         because it does not hold {@link
387 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
746 >     * Tries to advance eventCount and releases waiters. Called only
747 >     * from workers.
748       */
749 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
750 <        this(Runtime.getRuntime().availableProcessors(), factory);
749 >    final void signalWork() {
750 >        int c; // try to increment event count -- CAS failure OK
751 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
752 >        if (eventWaiters != 0L)
753 >            releaseEventWaiters();
754      }
755  
756      /**
757 <     * Creates a {@code ForkJoinPool} with the given parallelism and
758 <     * thread factory.
757 >     * Adds the given worker to event queue and blocks until
758 >     * terminating or event count advances from the given value
759       *
760 <     * @param parallelism the targeted number of worker threads
761 <     * @param factory the factory for creating new threads
399 <     * @throws IllegalArgumentException if parallelism less than or
400 <     * equal to zero, or greater than implementation limit
401 <     * @throws NullPointerException if factory is null
402 <     * @throws SecurityException if a security manager exists and
403 <     *         the caller is not permitted to modify threads
404 <     *         because it does not hold {@link
405 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
760 >     * @param w the calling worker thread
761 >     * @param ec the count
762       */
763 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
764 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
765 <            throw new IllegalArgumentException();
766 <        if (factory == null)
767 <            throw new NullPointerException();
768 <        checkPermission();
769 <        this.factory = factory;
770 <        this.parallelism = parallelism;
771 <        this.maxPoolSize = MAX_THREADS;
772 <        this.maintainsParallelism = true;
773 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
774 <        this.workerLock = new ReentrantLock();
775 <        this.termination = workerLock.newCondition();
420 <        this.stealCount = new AtomicLong();
421 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
422 <        // worker array and workers are lazily constructed
763 >    private void eventSync(ForkJoinWorkerThread w, int ec) {
764 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
765 >        long h;
766 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
767 >               (((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
768 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
769 >               eventCount == ec) {
770 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
771 >                                          w.nextWaiter = h, nh)) {
772 >                awaitEvent(w, ec);
773 >                break;
774 >            }
775 >        }
776      }
777  
778      /**
779 <     * Creates a new worker thread using factory.
779 >     * Blocks the given worker (that has already been entered as an
780 >     * event waiter) until terminating or event count advances from
781 >     * the given value. The oldest (first) waiter uses a timed wait to
782 >     * occasionally one-by-one shrink the number of workers (to a
783 >     * minimum of one) if the pool has not been used for extended
784 >     * periods.
785       *
786 <     * @param index the index to assign worker
787 <     * @return new worker, or null if factory failed
786 >     * @param w the calling worker thread
787 >     * @param ec the count
788       */
789 <    private ForkJoinWorkerThread createWorker(int index) {
790 <        Thread.UncaughtExceptionHandler h = ueh;
791 <        ForkJoinWorkerThread w = factory.newThread(this);
792 <        if (w != null) {
793 <            w.poolIndex = index;
794 <            w.setDaemon(true);
795 <            w.setAsyncMode(locallyFifo);
796 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
797 <            if (h != null)
798 <                w.setUncaughtExceptionHandler(h);
789 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
790 >        while (eventCount == ec) {
791 >            if (tryAccumulateStealCount(w)) { // transfer while idle
792 >                boolean untimed = (w.nextWaiter != 0L ||
793 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
794 >                long startTime = untimed? 0 : System.nanoTime();
795 >                Thread.interrupted();         // clear/ignore interrupt
796 >                if (eventCount != ec || w.isTerminating())
797 >                    break;                    // recheck after clear
798 >                if (untimed)
799 >                    LockSupport.park(w);
800 >                else {
801 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
802 >                    if (eventCount != ec || w.isTerminating())
803 >                        break;
804 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
805 >                        tryShutdownUnusedWorker(ec);
806 >                }
807 >            }
808          }
442        return w;
809      }
810  
811 +    // Maintaining parallelism
812 +
813      /**
814 <     * Returns a good size for worker array given pool size.
447 <     * Currently requires size to be a power of two.
814 >     * Pushes worker onto the spare stack.
815       */
816 <    private static int arraySizeFor(int poolSize) {
817 <        return (poolSize <= 1) ? 1 :
818 <            (1 << (32 - Integer.numberOfLeadingZeros(poolSize-1)));
816 >    final void pushSpare(ForkJoinWorkerThread w) {
817 >        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
818 >        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
819 >                                               w.nextSpare = spareWaiters,ns));
820      }
821  
822      /**
823 <     * Creates or resizes array if necessary to hold newLength.
824 <     * Call only under exclusion.
457 <     *
458 <     * @return the array
823 >     * Tries (once) to resume a spare if the number of running
824 >     * threads is less than target.
825       */
826 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
826 >    private void tryResumeSpare() {
827 >        int sw, id;
828          ForkJoinWorkerThread[] ws = workers;
829 <        if (ws == null)
830 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
831 <        else if (newLength > ws.length)
832 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
833 <        else
834 <            return ws;
829 >        int n = ws.length;
830 >        ForkJoinWorkerThread w;
831 >        if ((sw = spareWaiters) != 0 &&
832 >            (id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
833 >            id < n && (w = ws[id]) != null &&
834 >            (workerCounts & RUNNING_COUNT_MASK) < parallelism &&
835 >            spareWaiters == sw &&
836 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
837 >                                     sw, w.nextSpare)) {
838 >            int c; // increment running count before resume
839 >            do {} while (!UNSAFE.compareAndSwapInt
840 >                         (this, workerCountsOffset,
841 >                          c = workerCounts, c + ONE_RUNNING));
842 >            if (w.tryUnsuspend())
843 >                LockSupport.unpark(w);
844 >            else   // back out if w was shutdown
845 >                decrementWorkerCounts(ONE_RUNNING, 0);
846 >        }
847      }
848  
849      /**
850 <     * Tries to shrink workers into smaller array after one or more terminate.
851 <     */
852 <    private void tryShrinkWorkerArray() {
853 <        ForkJoinWorkerThread[] ws = workers;
854 <        if (ws != null) {
855 <            int len = ws.length;
856 <            int last = len - 1;
857 <            while (last >= 0 && ws[last] == null)
858 <                --last;
859 <            int newLength = arraySizeFor(last+1);
860 <            if (newLength < len)
861 <                workers = Arrays.copyOf(ws, newLength);
850 >     * Tries to increase the number of running workers if below target
851 >     * parallelism: If a spare exists tries to resume it via
852 >     * tryResumeSpare.  Otherwise, if not enough total workers or all
853 >     * existing workers are busy, adds a new worker. In all cases also
854 >     * helps wake up releasable workers waiting for work.
855 >     */
856 >    private void helpMaintainParallelism() {
857 >        int pc = parallelism;
858 >        int wc, rs, tc;
859 >        while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
860 >               (rs = runState) < TERMINATING) {
861 >            if (spareWaiters != 0)
862 >                tryResumeSpare();
863 >            else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
864 >                     (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
865 >                break;   // enough total
866 >            else if (runState == rs && workerCounts == wc &&
867 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
868 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
869 >                ForkJoinWorkerThread w = null;
870 >                Throwable fail = null;
871 >                try {
872 >                    w = factory.newThread(this);
873 >                } catch (Throwable ex) {
874 >                    fail = ex;
875 >                }
876 >                if (w == null) { // null or exceptional factory return
877 >                    decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
878 >                    tryTerminate(false); // handle failure during shutdown
879 >                    // If originating from an external caller,
880 >                    // propagate exception, else ignore
881 >                    if (fail != null && runState < TERMINATING &&
882 >                        !(Thread.currentThread() instanceof
883 >                          ForkJoinWorkerThread))
884 >                        UNSAFE.throwException(fail);
885 >                    break;
886 >                }
887 >                w.start(recordWorker(w), ueh);
888 >                if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
889 >                    int c; // advance event count
890 >                    UNSAFE.compareAndSwapInt(this, eventCountOffset,
891 >                                             c = eventCount, c+1);
892 >                    break; // add at most one unless total below target
893 >                }
894 >            }
895          }
896 +        if (eventWaiters != 0L)
897 +            releaseEventWaiters();
898      }
899  
900      /**
901 <     * Initializes workers if necessary.
901 >     * Callback from the oldest waiter in awaitEvent waking up after a
902 >     * period of non-use. If all workers are idle, tries (once) to
903 >     * shutdown an event waiter or a spare, if one exists. Note that
904 >     * we don't need CAS or locks here because the method is called
905 >     * only from one thread occasionally waking (and even misfires are
906 >     * OK). Note that until the shutdown worker fully terminates,
907 >     * workerCounts will overestimate total count, which is tolerable.
908 >     *
909 >     * @param ec the event count waited on by caller (to abort
910 >     * attempt if count has since changed).
911       */
912 <    final void ensureWorkerInitialization() {
913 <        ForkJoinWorkerThread[] ws = workers;
914 <        if (ws == null) {
915 <            final ReentrantLock lock = this.workerLock;
916 <            lock.lock();
917 <            try {
918 <                ws = workers;
919 <                if (ws == null) {
920 <                    int ps = parallelism;
921 <                    ws = ensureWorkerArrayCapacity(ps);
922 <                    for (int i = 0; i < ps; ++i) {
923 <                        ForkJoinWorkerThread w = createWorker(i);
924 <                        if (w != null) {
925 <                            ws[i] = w;
926 <                            w.start();
927 <                            updateWorkerCount(1);
928 <                        }
929 <                    }
912 >    private void tryShutdownUnusedWorker(int ec) {
913 >        if (runState == 0 && eventCount == ec) { // only trigger if all idle
914 >            ForkJoinWorkerThread[] ws = workers;
915 >            int n = ws.length;
916 >            ForkJoinWorkerThread w = null;
917 >            boolean shutdown = false;
918 >            int sw;
919 >            long h;
920 >            if ((sw = spareWaiters) != 0) { // prefer killing spares
921 >                int id = (sw & SPARE_ID_MASK) - 1;
922 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
923 >                    UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
924 >                                             sw, w.nextSpare))
925 >                    shutdown = true;
926 >            }
927 >            else if ((h = eventWaiters) != 0L) {
928 >                long nh;
929 >                int id = ((int)(h & WAITER_ID_MASK)) - 1;
930 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
931 >                    (nh = w.nextWaiter) != 0L && // keep at least one worker
932 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
933 >                    shutdown = true;
934 >            }
935 >            if (w != null && shutdown) {
936 >                w.shutdown();
937 >                LockSupport.unpark(w);
938 >            }
939 >        }
940 >        releaseEventWaiters(); // in case of interference
941 >    }
942 >
943 >    /**
944 >     * Callback from workers invoked upon each top-level action (i.e.,
945 >     * stealing a task or taking a submission and running it).
946 >     * Performs one or more of the following:
947 >     *
948 >     * 1. If the worker is active and either did not run a task
949 >     *    or there are too many workers, try to set its active status
950 >     *    to inactive and update activeCount. On contention, we may
951 >     *    try again in this or a subsequent call.
952 >     *
953 >     * 2. If not enough total workers, help create some.
954 >     *
955 >     * 3. If there are too many running workers, suspend this worker
956 >     *    (first forcing inactive if necessary).  If it is not needed,
957 >     *    it may be shutdown while suspended (via
958 >     *    tryShutdownUnusedWorker).  Otherwise, upon resume it
959 >     *    rechecks running thread count and need for event sync.
960 >     *
961 >     * 4. If worker did not run a task, await the next task event via
962 >     *    eventSync if necessary (first forcing inactivation), upon
963 >     *    which the worker may be shutdown via
964 >     *    tryShutdownUnusedWorker.  Otherwise, help release any
965 >     *    existing event waiters that are now releasable,
966 >     *
967 >     * @param w the worker
968 >     * @param ran true if worker ran a task since last call to this method
969 >     */
970 >    final void preStep(ForkJoinWorkerThread w, boolean ran) {
971 >        int wec = w.lastEventCount;
972 >        boolean active = w.active;
973 >        boolean inactivate = false;
974 >        int pc = parallelism;
975 >        while (w.runState == 0) {
976 >            int rs = runState;
977 >            if (rs >= TERMINATING) { // propagate shutdown
978 >                w.shutdown();
979 >                break;
980 >            }
981 >            if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
982 >                UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
983 >                inactivate = active = w.active = false;
984 >            int wc = workerCounts;
985 >            if ((wc & RUNNING_COUNT_MASK) > pc) {
986 >                if (!(inactivate |= active) && // must inactivate to suspend
987 >                    workerCounts == wc &&      // try to suspend as spare
988 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
989 >                                             wc, wc - ONE_RUNNING))
990 >                    w.suspendAsSpare();
991 >            }
992 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
993 >                helpMaintainParallelism();     // not enough workers
994 >            else if (!ran) {
995 >                long h = eventWaiters;
996 >                int ec = eventCount;
997 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
998 >                    releaseEventWaiters();     // release others before waiting
999 >                else if (ec != wec) {
1000 >                    w.lastEventCount = ec;     // no need to wait
1001 >                    break;
1002                  }
1003 <            } finally {
1004 <                lock.unlock();
1003 >                else if (!(inactivate |= active))
1004 >                    eventSync(w, wec);         // must inactivate before sync
1005              }
1006 +            else
1007 +                break;
1008          }
1009      }
1010  
1011      /**
1012 <     * Worker creation and startup for threads added via setParallelism.
1012 >     * Helps and/or blocks awaiting join of the given task.
1013 >     * See above for explanation.
1014 >     *
1015 >     * @param joinMe the task to join
1016 >     * @param worker the current worker thread
1017       */
1018 <    private void createAndStartAddedWorkers() {
1019 <        resumeAllSpares();  // Allow spares to convert to nonspare
1020 <        int ps = parallelism;
1021 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
1022 <        int len = ws.length;
1023 <        // Sweep through slots, to keep lowest indices most populated
1024 <        int k = 0;
1025 <        while (k < len) {
1026 <            if (ws[k] != null) {
1027 <                ++k;
1028 <                continue;
1018 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1019 >        int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1020 >        while (joinMe.status >= 0) {
1021 >            int wc;
1022 >            worker.helpJoinTask(joinMe);
1023 >            if (joinMe.status < 0)
1024 >                break;
1025 >            else if (retries > 0)
1026 >                --retries;
1027 >            else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1028 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1029 >                                              wc, wc - ONE_RUNNING)) {
1030 >                int stat, c; long h;
1031 >                while ((stat = joinMe.status) >= 0 &&
1032 >                       (h = eventWaiters) != 0L && // help release others
1033 >                       (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1034 >                    releaseEventWaiters();
1035 >                if (stat >= 0 &&
1036 >                    ((workerCounts & RUNNING_COUNT_MASK) == 0 ||
1037 >                     (stat =
1038 >                      joinMe.internalAwaitDone(JOIN_TIMEOUT_MILLIS)) >= 0))
1039 >                    helpMaintainParallelism(); // timeout or no running workers
1040 >                do {} while (!UNSAFE.compareAndSwapInt
1041 >                             (this, workerCountsOffset,
1042 >                              c = workerCounts, c + ONE_RUNNING));
1043 >                if (stat < 0)
1044 >                    break;   // else restart
1045              }
1046 <            int s = workerCounts;
1047 <            int tc = totalCountOf(s);
1048 <            int rc = runningCountOf(s);
1049 <            if (rc >= ps || tc >= ps)
1046 >        }
1047 >    }
1048 >
1049 >    /**
1050 >     * Same idea as awaitJoin, but no helping, retries, or timeouts.
1051 >     */
1052 >    final void awaitBlocker(ManagedBlocker blocker)
1053 >        throws InterruptedException {
1054 >        while (!blocker.isReleasable()) {
1055 >            int wc = workerCounts;
1056 >            if ((wc & RUNNING_COUNT_MASK) != 0 &&
1057 >                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1058 >                                         wc, wc - ONE_RUNNING)) {
1059 >                try {
1060 >                    while (!blocker.isReleasable()) {
1061 >                        long h = eventWaiters;
1062 >                        if (h != 0L &&
1063 >                            (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1064 >                            releaseEventWaiters();
1065 >                        else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1066 >                                 runState < TERMINATING)
1067 >                            helpMaintainParallelism();
1068 >                        else if (blocker.block())
1069 >                            break;
1070 >                    }
1071 >                } finally {
1072 >                    int c;
1073 >                    do {} while (!UNSAFE.compareAndSwapInt
1074 >                                 (this, workerCountsOffset,
1075 >                                  c = workerCounts, c + ONE_RUNNING));
1076 >                }
1077                  break;
1078 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
1079 <                ForkJoinWorkerThread w = createWorker(k);
1078 >            }
1079 >        }
1080 >    }
1081 >
1082 >    /**
1083 >     * Possibly initiates and/or completes termination.
1084 >     *
1085 >     * @param now if true, unconditionally terminate, else only
1086 >     * if shutdown and empty queue and no active workers
1087 >     * @return true if now terminating or terminated
1088 >     */
1089 >    private boolean tryTerminate(boolean now) {
1090 >        if (now)
1091 >            advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1092 >        else if (runState < SHUTDOWN ||
1093 >                 !submissionQueue.isEmpty() ||
1094 >                 (runState & ACTIVE_COUNT_MASK) != 0)
1095 >            return false;
1096 >
1097 >        if (advanceRunLevel(TERMINATING))
1098 >            startTerminating();
1099 >
1100 >        // Finish now if all threads terminated; else in some subsequent call
1101 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1102 >            advanceRunLevel(TERMINATED);
1103 >            termination.arrive();
1104 >        }
1105 >        return true;
1106 >    }
1107 >
1108 >
1109 >    /**
1110 >     * Actions on transition to TERMINATING
1111 >     *
1112 >     * Runs up to four passes through workers: (0) shutting down each
1113 >     * (without waking up if parked) to quickly spread notifications
1114 >     * without unnecessary bouncing around event queues etc (1) wake
1115 >     * up and help cancel tasks (2) interrupt (3) mop up races with
1116 >     * interrupted workers
1117 >     */
1118 >    private void startTerminating() {
1119 >        cancelSubmissions();
1120 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1121 >            int c; // advance event count
1122 >            UNSAFE.compareAndSwapInt(this, eventCountOffset,
1123 >                                     c = eventCount, c+1);
1124 >            eventWaiters = 0L; // clobber lists
1125 >            spareWaiters = 0;
1126 >            for (ForkJoinWorkerThread w : workers) {
1127                  if (w != null) {
1128 <                    ws[k++] = w;
1129 <                    w.start();
1130 <                }
1131 <                else {
1132 <                    updateWorkerCount(-1); // back out on failed creation
1133 <                    break;
1128 >                    w.shutdown();
1129 >                    if (passes > 0 && !w.isTerminated()) {
1130 >                        w.cancelTasks();
1131 >                        LockSupport.unpark(w);
1132 >                        if (passes > 1 && !w.isInterrupted()) {
1133 >                            try {
1134 >                                w.interrupt();
1135 >                            } catch (SecurityException ignore) {
1136 >                            }
1137 >                        }
1138 >                    }
1139                  }
1140              }
1141          }
1142      }
1143  
1144 +    /**
1145 +     * Clears out and cancels submissions, ignoring exceptions.
1146 +     */
1147 +    private void cancelSubmissions() {
1148 +        ForkJoinTask<?> task;
1149 +        while ((task = submissionQueue.poll()) != null) {
1150 +            try {
1151 +                task.cancel(false);
1152 +            } catch (Throwable ignore) {
1153 +            }
1154 +        }
1155 +    }
1156 +
1157 +    // misc support for ForkJoinWorkerThread
1158 +
1159 +    /**
1160 +     * Returns pool number.
1161 +     */
1162 +    final int getPoolNumber() {
1163 +        return poolNumber;
1164 +    }
1165 +
1166 +    /**
1167 +     * Tries to accumulate steal count from a worker, clearing
1168 +     * the worker's value if successful.
1169 +     *
1170 +     * @return true if worker steal count now zero
1171 +     */
1172 +    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1173 +        int sc = w.stealCount;
1174 +        long c = stealCount;
1175 +        // CAS even if zero, for fence effects
1176 +        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1177 +            if (sc != 0)
1178 +                w.stealCount = 0;
1179 +            return true;
1180 +        }
1181 +        return sc == 0;
1182 +    }
1183 +
1184 +    /**
1185 +     * Returns the approximate (non-atomic) number of idle threads per
1186 +     * active thread.
1187 +     */
1188 +    final int idlePerActive() {
1189 +        int pc = parallelism; // use parallelism, not rc
1190 +        int ac = runState;    // no mask -- artificially boosts during shutdown
1191 +        // Use exact results for small values, saturate past 4
1192 +        return ((pc <= ac) ? 0 :
1193 +                (pc >>> 1 <= ac) ? 1 :
1194 +                (pc >>> 2 <= ac) ? 3 :
1195 +                pc >>> 3);
1196 +    }
1197 +
1198 +    // Public and protected methods
1199 +
1200 +    // Constructors
1201 +
1202 +    /**
1203 +     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1204 +     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1205 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1206 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1207 +     *
1208 +     * @throws SecurityException if a security manager exists and
1209 +     *         the caller is not permitted to modify threads
1210 +     *         because it does not hold {@link
1211 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1212 +     */
1213 +    public ForkJoinPool() {
1214 +        this(Runtime.getRuntime().availableProcessors(),
1215 +             defaultForkJoinWorkerThreadFactory, null, false);
1216 +    }
1217 +
1218 +    /**
1219 +     * Creates a {@code ForkJoinPool} with the indicated parallelism
1220 +     * level, the {@linkplain
1221 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1222 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1223 +     *
1224 +     * @param parallelism the parallelism level
1225 +     * @throws IllegalArgumentException if parallelism less than or
1226 +     *         equal to zero, or greater than implementation limit
1227 +     * @throws SecurityException if a security manager exists and
1228 +     *         the caller is not permitted to modify threads
1229 +     *         because it does not hold {@link
1230 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1231 +     */
1232 +    public ForkJoinPool(int parallelism) {
1233 +        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1234 +    }
1235 +
1236 +    /**
1237 +     * Creates a {@code ForkJoinPool} with the given parameters.
1238 +     *
1239 +     * @param parallelism the parallelism level. For default value,
1240 +     * use {@link java.lang.Runtime#availableProcessors}.
1241 +     * @param factory the factory for creating new threads. For default value,
1242 +     * use {@link #defaultForkJoinWorkerThreadFactory}.
1243 +     * @param handler the handler for internal worker threads that
1244 +     * terminate due to unrecoverable errors encountered while executing
1245 +     * tasks. For default value, use {@code null}.
1246 +     * @param asyncMode if true,
1247 +     * establishes local first-in-first-out scheduling mode for forked
1248 +     * tasks that are never joined. This mode may be more appropriate
1249 +     * than default locally stack-based mode in applications in which
1250 +     * worker threads only process event-style asynchronous tasks.
1251 +     * For default value, use {@code false}.
1252 +     * @throws IllegalArgumentException if parallelism less than or
1253 +     *         equal to zero, or greater than implementation limit
1254 +     * @throws NullPointerException if the factory is null
1255 +     * @throws SecurityException if a security manager exists and
1256 +     *         the caller is not permitted to modify threads
1257 +     *         because it does not hold {@link
1258 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1259 +     */
1260 +    public ForkJoinPool(int parallelism,
1261 +                        ForkJoinWorkerThreadFactory factory,
1262 +                        Thread.UncaughtExceptionHandler handler,
1263 +                        boolean asyncMode) {
1264 +        checkPermission();
1265 +        if (factory == null)
1266 +            throw new NullPointerException();
1267 +        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1268 +            throw new IllegalArgumentException();
1269 +        this.parallelism = parallelism;
1270 +        this.factory = factory;
1271 +        this.ueh = handler;
1272 +        this.locallyFifo = asyncMode;
1273 +        int arraySize = initialArraySizeFor(parallelism);
1274 +        this.workers = new ForkJoinWorkerThread[arraySize];
1275 +        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1276 +        this.workerLock = new ReentrantLock();
1277 +        this.termination = new Phaser(1);
1278 +        this.poolNumber = poolNumberGenerator.incrementAndGet();
1279 +    }
1280 +
1281 +    /**
1282 +     * Returns initial power of two size for workers array.
1283 +     * @param pc the initial parallelism level
1284 +     */
1285 +    private static int initialArraySizeFor(int pc) {
1286 +        // If possible, initially allocate enough space for one spare
1287 +        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1288 +        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1289 +        size |= size >>> 1;
1290 +        size |= size >>> 2;
1291 +        size |= size >>> 4;
1292 +        size |= size >>> 8;
1293 +        return size + 1;
1294 +    }
1295 +
1296      // Execution methods
1297  
1298      /**
# Line 553 | Line 1301 | public class ForkJoinPool extends Abstra
1301      private <T> void doSubmit(ForkJoinTask<T> task) {
1302          if (task == null)
1303              throw new NullPointerException();
1304 <        if (isShutdown())
1304 >        if (runState >= SHUTDOWN)
1305              throw new RejectedExecutionException();
558        if (workers == null)
559            ensureWorkerInitialization();
1306          submissionQueue.offer(task);
1307 <        signalIdleWorkers();
1307 >        int c; // try to increment event count -- CAS failure OK
1308 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1309 >        helpMaintainParallelism(); // create, start, or resume some workers
1310      }
1311  
1312      /**
# Line 566 | Line 1314 | public class ForkJoinPool extends Abstra
1314       *
1315       * @param task the task
1316       * @return the task's result
1317 <     * @throws NullPointerException if task is null
1318 <     * @throws RejectedExecutionException if pool is shut down
1317 >     * @throws NullPointerException if the task is null
1318 >     * @throws RejectedExecutionException if the task cannot be
1319 >     *         scheduled for execution
1320       */
1321      public <T> T invoke(ForkJoinTask<T> task) {
1322          doSubmit(task);
# Line 578 | Line 1327 | public class ForkJoinPool extends Abstra
1327       * Arranges for (asynchronous) execution of the given task.
1328       *
1329       * @param task the task
1330 <     * @throws NullPointerException if task is null
1331 <     * @throws RejectedExecutionException if pool is shut down
1330 >     * @throws NullPointerException if the task is null
1331 >     * @throws RejectedExecutionException if the task cannot be
1332 >     *         scheduled for execution
1333       */
1334      public void execute(ForkJoinTask<?> task) {
1335          doSubmit(task);
# Line 587 | Line 1337 | public class ForkJoinPool extends Abstra
1337  
1338      // AbstractExecutorService methods
1339  
1340 +    /**
1341 +     * @throws NullPointerException if the task is null
1342 +     * @throws RejectedExecutionException if the task cannot be
1343 +     *         scheduled for execution
1344 +     */
1345      public void execute(Runnable task) {
1346          ForkJoinTask<?> job;
1347          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
# Line 596 | Line 1351 | public class ForkJoinPool extends Abstra
1351          doSubmit(job);
1352      }
1353  
1354 +    /**
1355 +     * Submits a ForkJoinTask for execution.
1356 +     *
1357 +     * @param task the task to submit
1358 +     * @return the task
1359 +     * @throws NullPointerException if the task is null
1360 +     * @throws RejectedExecutionException if the task cannot be
1361 +     *         scheduled for execution
1362 +     */
1363 +    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1364 +        doSubmit(task);
1365 +        return task;
1366 +    }
1367 +
1368 +    /**
1369 +     * @throws NullPointerException if the task is null
1370 +     * @throws RejectedExecutionException if the task cannot be
1371 +     *         scheduled for execution
1372 +     */
1373      public <T> ForkJoinTask<T> submit(Callable<T> task) {
1374          ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1375          doSubmit(job);
1376          return job;
1377      }
1378  
1379 +    /**
1380 +     * @throws NullPointerException if the task is null
1381 +     * @throws RejectedExecutionException if the task cannot be
1382 +     *         scheduled for execution
1383 +     */
1384      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1385          ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1386          doSubmit(job);
1387          return job;
1388      }
1389  
1390 +    /**
1391 +     * @throws NullPointerException if the task is null
1392 +     * @throws RejectedExecutionException if the task cannot be
1393 +     *         scheduled for execution
1394 +     */
1395      public ForkJoinTask<?> submit(Runnable task) {
1396          ForkJoinTask<?> job;
1397          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
# Line 619 | Line 1403 | public class ForkJoinPool extends Abstra
1403      }
1404  
1405      /**
1406 <     * Submits a ForkJoinTask for execution.
1407 <     *
624 <     * @param task the task to submit
625 <     * @return the task
626 <     * @throws RejectedExecutionException if the task cannot be
627 <     *         scheduled for execution
628 <     * @throws NullPointerException if the task is null
1406 >     * @throws NullPointerException       {@inheritDoc}
1407 >     * @throws RejectedExecutionException {@inheritDoc}
1408       */
630    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
631        doSubmit(task);
632        return task;
633    }
634
635
1409      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1410          ArrayList<ForkJoinTask<T>> forkJoinTasks =
1411              new ArrayList<ForkJoinTask<T>>(tasks.size());
# Line 641 | Line 1414 | public class ForkJoinPool extends Abstra
1414          invoke(new InvokeAll<T>(forkJoinTasks));
1415  
1416          @SuppressWarnings({"unchecked", "rawtypes"})
1417 <        List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1417 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1418          return futures;
1419      }
1420  
# Line 655 | Line 1428 | public class ForkJoinPool extends Abstra
1428          private static final long serialVersionUID = -7914297376763021607L;
1429      }
1430  
658    // Configuration and status settings and queries
659
1431      /**
1432       * Returns the factory used for constructing new workers.
1433       *
# Line 673 | Line 1444 | public class ForkJoinPool extends Abstra
1444       * @return the handler, or {@code null} if none
1445       */
1446      public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1447 <        Thread.UncaughtExceptionHandler h;
677 <        final ReentrantLock lock = this.workerLock;
678 <        lock.lock();
679 <        try {
680 <            h = ueh;
681 <        } finally {
682 <            lock.unlock();
683 <        }
684 <        return h;
685 <    }
686 <
687 <    /**
688 <     * Sets the handler for internal worker threads that terminate due
689 <     * to unrecoverable errors encountered while executing tasks.
690 <     * Unless set, the current default or ThreadGroup handler is used
691 <     * as handler.
692 <     *
693 <     * @param h the new handler
694 <     * @return the old handler, or {@code null} if none
695 <     * @throws SecurityException if a security manager exists and
696 <     *         the caller is not permitted to modify threads
697 <     *         because it does not hold {@link
698 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
699 <     */
700 <    public Thread.UncaughtExceptionHandler
701 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
702 <        checkPermission();
703 <        Thread.UncaughtExceptionHandler old = null;
704 <        final ReentrantLock lock = this.workerLock;
705 <        lock.lock();
706 <        try {
707 <            old = ueh;
708 <            ueh = h;
709 <            ForkJoinWorkerThread[] ws = workers;
710 <            if (ws != null) {
711 <                for (int i = 0; i < ws.length; ++i) {
712 <                    ForkJoinWorkerThread w = ws[i];
713 <                    if (w != null)
714 <                        w.setUncaughtExceptionHandler(h);
715 <                }
716 <            }
717 <        } finally {
718 <            lock.unlock();
719 <        }
720 <        return old;
721 <    }
722 <
723 <
724 <    /**
725 <     * Sets the target parallelism level of this pool.
726 <     *
727 <     * @param parallelism the target parallelism
728 <     * @throws IllegalArgumentException if parallelism less than or
729 <     * equal to zero or greater than maximum size bounds
730 <     * @throws SecurityException if a security manager exists and
731 <     *         the caller is not permitted to modify threads
732 <     *         because it does not hold {@link
733 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
734 <     */
735 <    public void setParallelism(int parallelism) {
736 <        checkPermission();
737 <        if (parallelism <= 0 || parallelism > maxPoolSize)
738 <            throw new IllegalArgumentException();
739 <        final ReentrantLock lock = this.workerLock;
740 <        lock.lock();
741 <        try {
742 <            if (!isTerminating()) {
743 <                int p = this.parallelism;
744 <                this.parallelism = parallelism;
745 <                if (parallelism > p)
746 <                    createAndStartAddedWorkers();
747 <                else
748 <                    trimSpares();
749 <            }
750 <        } finally {
751 <            lock.unlock();
752 <        }
753 <        signalIdleWorkers();
1447 >        return ueh;
1448      }
1449  
1450      /**
1451 <     * Returns the targeted number of worker threads in this pool.
1451 >     * Returns the targeted parallelism level of this pool.
1452       *
1453 <     * @return the targeted number of worker threads in this pool
1453 >     * @return the targeted parallelism level of this pool
1454       */
1455      public int getParallelism() {
1456          return parallelism;
# Line 764 | Line 1458 | public class ForkJoinPool extends Abstra
1458  
1459      /**
1460       * Returns the number of worker threads that have started but not
1461 <     * yet terminated.  This result returned by this method may differ
1461 >     * yet terminated.  The result returned by this method may differ
1462       * from {@link #getParallelism} when threads are created to
1463       * maintain parallelism when others are cooperatively blocked.
1464       *
1465       * @return the number of worker threads
1466       */
1467      public int getPoolSize() {
1468 <        return totalCountOf(workerCounts);
775 <    }
776 <
777 <    /**
778 <     * Returns the maximum number of threads allowed to exist in the
779 <     * pool, even if there are insufficient unblocked running threads.
780 <     *
781 <     * @return the maximum
782 <     */
783 <    public int getMaximumPoolSize() {
784 <        return maxPoolSize;
785 <    }
786 <
787 <    /**
788 <     * Sets the maximum number of threads allowed to exist in the
789 <     * pool, even if there are insufficient unblocked running threads.
790 <     * Setting this value has no effect on current pool size. It
791 <     * controls construction of new threads.
792 <     *
793 <     * @throws IllegalArgumentException if negative or greater than
794 <     * internal implementation limit
795 <     */
796 <    public void setMaximumPoolSize(int newMax) {
797 <        if (newMax < 0 || newMax > MAX_THREADS)
798 <            throw new IllegalArgumentException();
799 <        maxPoolSize = newMax;
800 <    }
801 <
802 <
803 <    /**
804 <     * Returns {@code true} if this pool dynamically maintains its
805 <     * target parallelism level. If false, new threads are added only
806 <     * to avoid possible starvation.  This setting is by default true.
807 <     *
808 <     * @return {@code true} if maintains parallelism
809 <     */
810 <    public boolean getMaintainsParallelism() {
811 <        return maintainsParallelism;
812 <    }
813 <
814 <    /**
815 <     * Sets whether this pool dynamically maintains its target
816 <     * parallelism level. If false, new threads are added only to
817 <     * avoid possible starvation.
818 <     *
819 <     * @param enable {@code true} to maintain parallelism
820 <     */
821 <    public void setMaintainsParallelism(boolean enable) {
822 <        maintainsParallelism = enable;
823 <    }
824 <
825 <    /**
826 <     * Establishes local first-in-first-out scheduling mode for forked
827 <     * tasks that are never joined. This mode may be more appropriate
828 <     * than default locally stack-based mode in applications in which
829 <     * worker threads only process asynchronous tasks.  This method is
830 <     * designed to be invoked only when the pool is quiescent, and
831 <     * typically only before any tasks are submitted. The effects of
832 <     * invocations at other times may be unpredictable.
833 <     *
834 <     * @param async if {@code true}, use locally FIFO scheduling
835 <     * @return the previous mode
836 <     * @see #getAsyncMode
837 <     */
838 <    public boolean setAsyncMode(boolean async) {
839 <        boolean oldMode = locallyFifo;
840 <        locallyFifo = async;
841 <        ForkJoinWorkerThread[] ws = workers;
842 <        if (ws != null) {
843 <            for (int i = 0; i < ws.length; ++i) {
844 <                ForkJoinWorkerThread t = ws[i];
845 <                if (t != null)
846 <                    t.setAsyncMode(async);
847 <            }
848 <        }
849 <        return oldMode;
1468 >        return workerCounts >>> TOTAL_COUNT_SHIFT;
1469      }
1470  
1471      /**
# Line 854 | Line 1473 | public class ForkJoinPool extends Abstra
1473       * scheduling mode for forked tasks that are never joined.
1474       *
1475       * @return {@code true} if this pool uses async mode
857     * @see #setAsyncMode
1476       */
1477      public boolean getAsyncMode() {
1478          return locallyFifo;
# Line 863 | Line 1481 | public class ForkJoinPool extends Abstra
1481      /**
1482       * Returns an estimate of the number of worker threads that are
1483       * not blocked waiting to join tasks or for other managed
1484 <     * synchronization.
1484 >     * synchronization. This method may overestimate the
1485 >     * number of running threads.
1486       *
1487       * @return the number of worker threads
1488       */
1489      public int getRunningThreadCount() {
1490 <        return runningCountOf(workerCounts);
1490 >        return workerCounts & RUNNING_COUNT_MASK;
1491      }
1492  
1493      /**
# Line 879 | Line 1498 | public class ForkJoinPool extends Abstra
1498       * @return the number of active threads
1499       */
1500      public int getActiveThreadCount() {
1501 <        return activeCountOf(runControl);
883 <    }
884 <
885 <    /**
886 <     * Returns an estimate of the number of threads that are currently
887 <     * idle waiting for tasks. This method may underestimate the
888 <     * number of idle threads.
889 <     *
890 <     * @return the number of idle threads
891 <     */
892 <    final int getIdleThreadCount() {
893 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
894 <        return (c <= 0) ? 0 : c;
1501 >        return runState & ACTIVE_COUNT_MASK;
1502      }
1503  
1504      /**
# Line 906 | Line 1513 | public class ForkJoinPool extends Abstra
1513       * @return {@code true} if all threads are currently idle
1514       */
1515      public boolean isQuiescent() {
1516 <        return activeCountOf(runControl) == 0;
1516 >        return (runState & ACTIVE_COUNT_MASK) == 0;
1517      }
1518  
1519      /**
# Line 921 | Line 1528 | public class ForkJoinPool extends Abstra
1528       * @return the number of steals
1529       */
1530      public long getStealCount() {
1531 <        return stealCount.get();
925 <    }
926 <
927 <    /**
928 <     * Accumulates steal count from a worker.
929 <     * Call only when worker known to be idle.
930 <     */
931 <    private void updateStealCount(ForkJoinWorkerThread w) {
932 <        int sc = w.getAndClearStealCount();
933 <        if (sc != 0)
934 <            stealCount.addAndGet(sc);
1531 >        return stealCount;
1532      }
1533  
1534      /**
# Line 946 | Line 1543 | public class ForkJoinPool extends Abstra
1543       */
1544      public long getQueuedTaskCount() {
1545          long count = 0;
1546 <        ForkJoinWorkerThread[] ws = workers;
1547 <        if (ws != null) {
1548 <            for (int i = 0; i < ws.length; ++i) {
952 <                ForkJoinWorkerThread t = ws[i];
953 <                if (t != null)
954 <                    count += t.getQueueSize();
955 <            }
956 <        }
1546 >        for (ForkJoinWorkerThread w : workers)
1547 >            if (w != null)
1548 >                count += w.getQueueSize();
1549          return count;
1550      }
1551  
# Line 1007 | Line 1599 | public class ForkJoinPool extends Abstra
1599       * @return the number of elements transferred
1600       */
1601      protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1602 <        int n = submissionQueue.drainTo(c);
1603 <        ForkJoinWorkerThread[] ws = workers;
1604 <        if (ws != null) {
1605 <            for (int i = 0; i < ws.length; ++i) {
1606 <                ForkJoinWorkerThread w = ws[i];
1015 <                if (w != null)
1016 <                    n += w.drainTasksTo(c);
1017 <            }
1018 <        }
1019 <        return n;
1602 >        int count = submissionQueue.drainTo(c);
1603 >        for (ForkJoinWorkerThread w : workers)
1604 >            if (w != null)
1605 >                count += w.drainTasksTo(c);
1606 >        return count;
1607      }
1608  
1609      /**
# Line 1027 | Line 1614 | public class ForkJoinPool extends Abstra
1614       * @return a string identifying this pool, as well as its state
1615       */
1616      public String toString() {
1030        int ps = parallelism;
1031        int wc = workerCounts;
1032        int rc = runControl;
1617          long st = getStealCount();
1618          long qt = getQueuedTaskCount();
1619          long qs = getQueuedSubmissionCount();
1620 +        int wc = workerCounts;
1621 +        int tc = wc >>> TOTAL_COUNT_SHIFT;
1622 +        int rc = wc & RUNNING_COUNT_MASK;
1623 +        int pc = parallelism;
1624 +        int rs = runState;
1625 +        int ac = rs & ACTIVE_COUNT_MASK;
1626          return super.toString() +
1627 <            "[" + runStateToString(runStateOf(rc)) +
1628 <            ", parallelism = " + ps +
1629 <            ", size = " + totalCountOf(wc) +
1630 <            ", active = " + activeCountOf(rc) +
1631 <            ", running = " + runningCountOf(wc) +
1627 >            "[" + runLevelToString(rs) +
1628 >            ", parallelism = " + pc +
1629 >            ", size = " + tc +
1630 >            ", active = " + ac +
1631 >            ", running = " + rc +
1632              ", steals = " + st +
1633              ", tasks = " + qt +
1634              ", submissions = " + qs +
1635              "]";
1636      }
1637  
1638 <    private static String runStateToString(int rs) {
1639 <        switch(rs) {
1640 <        case RUNNING: return "Running";
1641 <        case SHUTDOWN: return "Shutting down";
1642 <        case TERMINATING: return "Terminating";
1053 <        case TERMINATED: return "Terminated";
1054 <        default: throw new Error("Unknown run state");
1055 <        }
1638 >    private static String runLevelToString(int s) {
1639 >        return ((s & TERMINATED) != 0 ? "Terminated" :
1640 >                ((s & TERMINATING) != 0 ? "Terminating" :
1641 >                 ((s & SHUTDOWN) != 0 ? "Shutting down" :
1642 >                  "Running")));
1643      }
1644  
1058    // lifecycle control
1059
1645      /**
1646       * Initiates an orderly shutdown in which previously submitted
1647       * tasks are executed, but no new tasks will be accepted.
# Line 1071 | Line 1656 | public class ForkJoinPool extends Abstra
1656       */
1657      public void shutdown() {
1658          checkPermission();
1659 <        transitionRunStateTo(SHUTDOWN);
1660 <        if (canTerminateOnShutdown(runControl)) {
1076 <            if (workers == null) { // shutting down before workers created
1077 <                final ReentrantLock lock = this.workerLock;
1078 <                lock.lock();
1079 <                try {
1080 <                    if (workers == null) {
1081 <                        terminate();
1082 <                        transitionRunStateTo(TERMINATED);
1083 <                        termination.signalAll();
1084 <                    }
1085 <                } finally {
1086 <                    lock.unlock();
1087 <                }
1088 <            }
1089 <            terminateOnShutdown();
1090 <        }
1659 >        advanceRunLevel(SHUTDOWN);
1660 >        tryTerminate(false);
1661      }
1662  
1663      /**
1664 <     * Attempts to stop all actively executing tasks, and cancels all
1665 <     * waiting tasks.  Tasks that are in the process of being
1666 <     * submitted or executed concurrently during the course of this
1667 <     * method may or may not be rejected. Unlike some other executors,
1668 <     * this method cancels rather than collects non-executed tasks
1669 <     * upon termination, so always returns an empty list. However, you
1670 <     * can use method {@link #drainTasksTo} before invoking this
1671 <     * method to transfer unexecuted tasks to another collection.
1664 >     * Attempts to cancel and/or stop all tasks, and reject all
1665 >     * subsequently submitted tasks.  Tasks that are in the process of
1666 >     * being submitted or executed concurrently during the course of
1667 >     * this method may or may not be rejected. This method cancels
1668 >     * both existing and unexecuted tasks, in order to permit
1669 >     * termination in the presence of task dependencies. So the method
1670 >     * always returns an empty list (unlike the case for some other
1671 >     * Executors).
1672       *
1673       * @return an empty list
1674       * @throws SecurityException if a security manager exists and
# Line 1108 | Line 1678 | public class ForkJoinPool extends Abstra
1678       */
1679      public List<Runnable> shutdownNow() {
1680          checkPermission();
1681 <        terminate();
1681 >        tryTerminate(true);
1682          return Collections.emptyList();
1683      }
1684  
# Line 1118 | Line 1688 | public class ForkJoinPool extends Abstra
1688       * @return {@code true} if all tasks have completed following shut down
1689       */
1690      public boolean isTerminated() {
1691 <        return runStateOf(runControl) == TERMINATED;
1691 >        return runState >= TERMINATED;
1692      }
1693  
1694      /**
1695       * Returns {@code true} if the process of termination has
1696 <     * commenced but possibly not yet completed.
1696 >     * commenced but not yet completed.  This method may be useful for
1697 >     * debugging. A return of {@code true} reported a sufficient
1698 >     * period after shutdown may indicate that submitted tasks have
1699 >     * ignored or suppressed interruption, causing this executor not
1700 >     * to properly terminate.
1701       *
1702 <     * @return {@code true} if terminating
1702 >     * @return {@code true} if terminating but not yet terminated
1703       */
1704      public boolean isTerminating() {
1705 <        return runStateOf(runControl) >= TERMINATING;
1705 >        return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1706      }
1707  
1708      /**
1709 +     * Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1710 +     */
1711 +    final boolean isAtLeastTerminating() {
1712 +        return runState >= TERMINATING;
1713 +    }
1714 +        
1715 +    /**
1716       * Returns {@code true} if this pool has been shut down.
1717       *
1718       * @return {@code true} if this pool has been shut down
1719       */
1720      public boolean isShutdown() {
1721 <        return runStateOf(runControl) >= SHUTDOWN;
1721 >        return runState >= SHUTDOWN;
1722      }
1723  
1724      /**
# Line 1153 | Line 1734 | public class ForkJoinPool extends Abstra
1734       */
1735      public boolean awaitTermination(long timeout, TimeUnit unit)
1736          throws InterruptedException {
1156        long nanos = unit.toNanos(timeout);
1157        final ReentrantLock lock = this.workerLock;
1158        lock.lock();
1159        try {
1160            for (;;) {
1161                if (isTerminated())
1162                    return true;
1163                if (nanos <= 0)
1164                    return false;
1165                nanos = termination.awaitNanos(nanos);
1166            }
1167        } finally {
1168            lock.unlock();
1169        }
1170    }
1171
1172    // Shutdown and termination support
1173
1174    /**
1175     * Callback from terminating worker. Nulls out the corresponding
1176     * workers slot, and if terminating, tries to terminate; else
1177     * tries to shrink workers array.
1178     *
1179     * @param w the worker
1180     */
1181    final void workerTerminated(ForkJoinWorkerThread w) {
1182        updateStealCount(w);
1183        updateWorkerCount(-1);
1184        final ReentrantLock lock = this.workerLock;
1185        lock.lock();
1186        try {
1187            ForkJoinWorkerThread[] ws = workers;
1188            if (ws != null) {
1189                int idx = w.poolIndex;
1190                if (idx >= 0 && idx < ws.length && ws[idx] == w)
1191                    ws[idx] = null;
1192                if (totalCountOf(workerCounts) == 0) {
1193                    terminate(); // no-op if already terminating
1194                    transitionRunStateTo(TERMINATED);
1195                    termination.signalAll();
1196                }
1197                else if (!isTerminating()) {
1198                    tryShrinkWorkerArray();
1199                    tryResumeSpare(true); // allow replacement
1200                }
1201            }
1202        } finally {
1203            lock.unlock();
1204        }
1205        signalIdleWorkers();
1206    }
1207
1208    /**
1209     * Initiates termination.
1210     */
1211    private void terminate() {
1212        if (transitionRunStateTo(TERMINATING)) {
1213            stopAllWorkers();
1214            resumeAllSpares();
1215            signalIdleWorkers();
1216            cancelQueuedSubmissions();
1217            cancelQueuedWorkerTasks();
1218            interruptUnterminatedWorkers();
1219            signalIdleWorkers(); // resignal after interrupt
1220        }
1221    }
1222
1223    /**
1224     * Possibly terminates when on shutdown state.
1225     */
1226    private void terminateOnShutdown() {
1227        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1228            terminate();
1229    }
1230
1231    /**
1232     * Clears out and cancels submissions.
1233     */
1234    private void cancelQueuedSubmissions() {
1235        ForkJoinTask<?> task;
1236        while ((task = pollSubmission()) != null)
1237            task.cancel(false);
1238    }
1239
1240    /**
1241     * Cleans out worker queues.
1242     */
1243    private void cancelQueuedWorkerTasks() {
1244        final ReentrantLock lock = this.workerLock;
1245        lock.lock();
1737          try {
1738 <            ForkJoinWorkerThread[] ws = workers;
1739 <            if (ws != null) {
1249 <                for (int i = 0; i < ws.length; ++i) {
1250 <                    ForkJoinWorkerThread t = ws[i];
1251 <                    if (t != null)
1252 <                        t.cancelTasks();
1253 <                }
1254 <            }
1255 <        } finally {
1256 <            lock.unlock();
1257 <        }
1258 <    }
1259 <
1260 <    /**
1261 <     * Sets each worker's status to terminating. Requires lock to avoid
1262 <     * conflicts with add/remove.
1263 <     */
1264 <    private void stopAllWorkers() {
1265 <        final ReentrantLock lock = this.workerLock;
1266 <        lock.lock();
1267 <        try {
1268 <            ForkJoinWorkerThread[] ws = workers;
1269 <            if (ws != null) {
1270 <                for (int i = 0; i < ws.length; ++i) {
1271 <                    ForkJoinWorkerThread t = ws[i];
1272 <                    if (t != null)
1273 <                        t.shutdownNow();
1274 <                }
1275 <            }
1276 <        } finally {
1277 <            lock.unlock();
1278 <        }
1279 <    }
1280 <
1281 <    /**
1282 <     * Interrupts all unterminated workers.  This is not required for
1283 <     * sake of internal control, but may help unstick user code during
1284 <     * shutdown.
1285 <     */
1286 <    private void interruptUnterminatedWorkers() {
1287 <        final ReentrantLock lock = this.workerLock;
1288 <        lock.lock();
1289 <        try {
1290 <            ForkJoinWorkerThread[] ws = workers;
1291 <            if (ws != null) {
1292 <                for (int i = 0; i < ws.length; ++i) {
1293 <                    ForkJoinWorkerThread t = ws[i];
1294 <                    if (t != null && !t.isTerminated()) {
1295 <                        try {
1296 <                            t.interrupt();
1297 <                        } catch (SecurityException ignore) {
1298 <                        }
1299 <                    }
1300 <                }
1301 <            }
1302 <        } finally {
1303 <            lock.unlock();
1304 <        }
1305 <    }
1306 <
1307 <
1308 <    /*
1309 <     * Nodes for event barrier to manage idle threads.  Queue nodes
1310 <     * are basic Treiber stack nodes, also used for spare stack.
1311 <     *
1312 <     * The event barrier has an event count and a wait queue (actually
1313 <     * a Treiber stack).  Workers are enabled to look for work when
1314 <     * the eventCount is incremented. If they fail to find work, they
1315 <     * may wait for next count. Upon release, threads help others wake
1316 <     * up.
1317 <     *
1318 <     * Synchronization events occur only in enough contexts to
1319 <     * maintain overall liveness:
1320 <     *
1321 <     *   - Submission of a new task to the pool
1322 <     *   - Resizes or other changes to the workers array
1323 <     *   - pool termination
1324 <     *   - A worker pushing a task on an empty queue
1325 <     *
1326 <     * The case of pushing a task occurs often enough, and is heavy
1327 <     * enough compared to simple stack pushes, to require special
1328 <     * handling: Method signalWork returns without advancing count if
1329 <     * the queue appears to be empty.  This would ordinarily result in
1330 <     * races causing some queued waiters not to be woken up. To avoid
1331 <     * this, the first worker enqueued in method sync (see
1332 <     * syncIsReleasable) rescans for tasks after being enqueued, and
1333 <     * helps signal if any are found. This works well because the
1334 <     * worker has nothing better to do, and so might as well help
1335 <     * alleviate the overhead and contention on the threads actually
1336 <     * doing work.  Also, since event counts increments on task
1337 <     * availability exist to maintain liveness (rather than to force
1338 <     * refreshes etc), it is OK for callers to exit early if
1339 <     * contending with another signaller.
1340 <     */
1341 <    static final class WaitQueueNode {
1342 <        WaitQueueNode next; // only written before enqueued
1343 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1344 <        final long count; // unused for spare stack
1345 <
1346 <        WaitQueueNode(long c, ForkJoinWorkerThread w) {
1347 <            count = c;
1348 <            thread = w;
1349 <        }
1350 <
1351 <        /**
1352 <         * Wakes up waiter, returning false if known to already
1353 <         */
1354 <        boolean signal() {
1355 <            ForkJoinWorkerThread t = thread;
1356 <            if (t == null)
1357 <                return false;
1358 <            thread = null;
1359 <            LockSupport.unpark(t);
1360 <            return true;
1361 <        }
1362 <
1363 <        /**
1364 <         * Awaits release on sync.
1365 <         */
1366 <        void awaitSyncRelease(ForkJoinPool p) {
1367 <            while (thread != null && !p.syncIsReleasable(this))
1368 <                LockSupport.park(this);
1369 <        }
1370 <
1371 <        /**
1372 <         * Awaits resumption as spare.
1373 <         */
1374 <        void awaitSpareRelease() {
1375 <            while (thread != null) {
1376 <                if (!Thread.interrupted())
1377 <                    LockSupport.park(this);
1378 <            }
1379 <        }
1380 <    }
1381 <
1382 <    /**
1383 <     * Ensures that no thread is waiting for count to advance from the
1384 <     * current value of eventCount read on entry to this method, by
1385 <     * releasing waiting threads if necessary.
1386 <     *
1387 <     * @return the count
1388 <     */
1389 <    final long ensureSync() {
1390 <        long c = eventCount;
1391 <        WaitQueueNode q;
1392 <        while ((q = syncStack) != null && q.count < c) {
1393 <            if (casBarrierStack(q, null)) {
1394 <                do {
1395 <                    q.signal();
1396 <                } while ((q = q.next) != null);
1397 <                break;
1398 <            }
1399 <        }
1400 <        return c;
1401 <    }
1402 <
1403 <    /**
1404 <     * Increments event count and releases waiting threads.
1405 <     */
1406 <    private void signalIdleWorkers() {
1407 <        long c;
1408 <        do {} while (!casEventCount(c = eventCount, c+1));
1409 <        ensureSync();
1410 <    }
1411 <
1412 <    /**
1413 <     * Signals threads waiting to poll a task. Because method sync
1414 <     * rechecks availability, it is OK to only proceed if queue
1415 <     * appears to be non-empty, and OK to skip under contention to
1416 <     * increment count (since some other thread succeeded).
1417 <     */
1418 <    final void signalWork() {
1419 <        long c;
1420 <        WaitQueueNode q;
1421 <        if (syncStack != null &&
1422 <            casEventCount(c = eventCount, c+1) &&
1423 <            (((q = syncStack) != null && q.count <= c) &&
1424 <             (!casBarrierStack(q, q.next) || !q.signal())))
1425 <            ensureSync();
1426 <    }
1427 <
1428 <    /**
1429 <     * Waits until event count advances from last value held by
1430 <     * caller, or if excess threads, caller is resumed as spare, or
1431 <     * caller or pool is terminating. Updates caller's event on exit.
1432 <     *
1433 <     * @param w the calling worker thread
1434 <     */
1435 <    final void sync(ForkJoinWorkerThread w) {
1436 <        updateStealCount(w); // Transfer w's count while it is idle
1437 <
1438 <        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1439 <            long prev = w.lastEventCount;
1440 <            WaitQueueNode node = null;
1441 <            WaitQueueNode h;
1442 <            while (eventCount == prev &&
1443 <                   ((h = syncStack) == null || h.count == prev)) {
1444 <                if (node == null)
1445 <                    node = new WaitQueueNode(prev, w);
1446 <                if (casBarrierStack(node.next = h, node)) {
1447 <                    node.awaitSyncRelease(this);
1448 <                    break;
1449 <                }
1450 <            }
1451 <            long ec = ensureSync();
1452 <            if (ec != prev) {
1453 <                w.lastEventCount = ec;
1454 <                break;
1455 <            }
1456 <        }
1457 <    }
1458 <
1459 <    /**
1460 <     * Returns {@code true} if worker waiting on sync can proceed:
1461 <     *  - on signal (thread == null)
1462 <     *  - on event count advance (winning race to notify vs signaller)
1463 <     *  - on interrupt
1464 <     *  - if the first queued node, we find work available
1465 <     * If node was not signalled and event count not advanced on exit,
1466 <     * then we also help advance event count.
1467 <     *
1468 <     * @return {@code true} if node can be released
1469 <     */
1470 <    final boolean syncIsReleasable(WaitQueueNode node) {
1471 <        long prev = node.count;
1472 <        if (!Thread.interrupted() && node.thread != null &&
1473 <            (node.next != null ||
1474 <             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1475 <            eventCount == prev)
1476 <            return false;
1477 <        if (node.thread != null) {
1478 <            node.thread = null;
1479 <            long ec = eventCount;
1480 <            if (prev <= ec) // help signal
1481 <                casEventCount(ec, ec+1);
1482 <        }
1483 <        return true;
1484 <    }
1485 <
1486 <    /**
1487 <     * Returns {@code true} if a new sync event occurred since last
1488 <     * call to sync or this method, if so, updating caller's count.
1489 <     */
1490 <    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1491 <        long lc = w.lastEventCount;
1492 <        long ec = ensureSync();
1493 <        if (ec == lc)
1738 >            return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1739 >        } catch (TimeoutException ex) {
1740              return false;
1495        w.lastEventCount = ec;
1496        return true;
1497    }
1498
1499    //  Parallelism maintenance
1500
1501    /**
1502     * Decrements running count; if too low, adds spare.
1503     *
1504     * Conceptually, all we need to do here is add or resume a
1505     * spare thread when one is about to block (and remove or
1506     * suspend it later when unblocked -- see suspendIfSpare).
1507     * However, implementing this idea requires coping with
1508     * several problems: we have imperfect information about the
1509     * states of threads. Some count updates can and usually do
1510     * lag run state changes, despite arrangements to keep them
1511     * accurate (for example, when possible, updating counts
1512     * before signalling or resuming), especially when running on
1513     * dynamic JVMs that don't optimize the infrequent paths that
1514     * update counts. Generating too many threads can make these
1515     * problems become worse, because excess threads are more
1516     * likely to be context-switched with others, slowing them all
1517     * down, especially if there is no work available, so all are
1518     * busy scanning or idling.  Also, excess spare threads can
1519     * only be suspended or removed when they are idle, not
1520     * immediately when they aren't needed. So adding threads will
1521     * raise parallelism level for longer than necessary.  Also,
1522     * FJ applications often encounter highly transient peaks when
1523     * many threads are blocked joining, but for less time than it
1524     * takes to create or resume spares.
1525     *
1526     * @param joinMe if non-null, return early if done
1527     * @param maintainParallelism if true, try to stay within
1528     * target counts, else create only to avoid starvation
1529     * @return true if joinMe known to be done
1530     */
1531    final boolean preJoin(ForkJoinTask<?> joinMe,
1532                          boolean maintainParallelism) {
1533        maintainParallelism &= maintainsParallelism; // overrride
1534        boolean dec = false;  // true when running count decremented
1535        while (spareStack == null || !tryResumeSpare(dec)) {
1536            int counts = workerCounts;
1537            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1538                // CAS cheat
1539                if (!needSpare(counts, maintainParallelism))
1540                    break;
1541                if (joinMe.status < 0)
1542                    return true;
1543                if (tryAddSpare(counts))
1544                    break;
1545            }
1546        }
1547        return false;
1548    }
1549
1550    /**
1551     * Same idea as preJoin
1552     */
1553    final boolean preBlock(ManagedBlocker blocker,
1554                           boolean maintainParallelism) {
1555        maintainParallelism &= maintainsParallelism;
1556        boolean dec = false;
1557        while (spareStack == null || !tryResumeSpare(dec)) {
1558            int counts = workerCounts;
1559            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1560                if (!needSpare(counts, maintainParallelism))
1561                    break;
1562                if (blocker.isReleasable())
1563                    return true;
1564                if (tryAddSpare(counts))
1565                    break;
1566            }
1567        }
1568        return false;
1569    }
1570
1571    /**
1572     * Returns {@code true} if a spare thread appears to be needed.
1573     * If maintaining parallelism, returns true when the deficit in
1574     * running threads is more than the surplus of total threads, and
1575     * there is apparently some work to do.  This self-limiting rule
1576     * means that the more threads that have already been added, the
1577     * less parallelism we will tolerate before adding another.
1578     *
1579     * @param counts current worker counts
1580     * @param maintainParallelism try to maintain parallelism
1581     */
1582    private boolean needSpare(int counts, boolean maintainParallelism) {
1583        int ps = parallelism;
1584        int rc = runningCountOf(counts);
1585        int tc = totalCountOf(counts);
1586        int runningDeficit = ps - rc;
1587        int totalSurplus = tc - ps;
1588        return (tc < maxPoolSize &&
1589                (rc == 0 || totalSurplus < 0 ||
1590                 (maintainParallelism &&
1591                  runningDeficit > totalSurplus &&
1592                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
1593    }
1594
1595    /**
1596     * Adds a spare worker if lock available and no more than the
1597     * expected numbers of threads exist.
1598     *
1599     * @return true if successful
1600     */
1601    private boolean tryAddSpare(int expectedCounts) {
1602        final ReentrantLock lock = this.workerLock;
1603        int expectedRunning = runningCountOf(expectedCounts);
1604        int expectedTotal = totalCountOf(expectedCounts);
1605        boolean success = false;
1606        boolean locked = false;
1607        // confirm counts while locking; CAS after obtaining lock
1608        try {
1609            for (;;) {
1610                int s = workerCounts;
1611                int tc = totalCountOf(s);
1612                int rc = runningCountOf(s);
1613                if (rc > expectedRunning || tc > expectedTotal)
1614                    break;
1615                if (!locked && !(locked = lock.tryLock()))
1616                    break;
1617                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1618                    createAndStartSpare(tc);
1619                    success = true;
1620                    break;
1621                }
1622            }
1623        } finally {
1624            if (locked)
1625                lock.unlock();
1626        }
1627        return success;
1628    }
1629
1630    /**
1631     * Adds the kth spare worker. On entry, pool counts are already
1632     * adjusted to reflect addition.
1633     */
1634    private void createAndStartSpare(int k) {
1635        ForkJoinWorkerThread w = null;
1636        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1637        int len = ws.length;
1638        // Probably, we can place at slot k. If not, find empty slot
1639        if (k < len && ws[k] != null) {
1640            for (k = 0; k < len && ws[k] != null; ++k)
1641                ;
1642        }
1643        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1644            ws[k] = w;
1645            w.start();
1646        }
1647        else
1648            updateWorkerCount(-1); // adjust on failure
1649        signalIdleWorkers();
1650    }
1651
1652    /**
1653     * Suspends calling thread w if there are excess threads.  Called
1654     * only from sync.  Spares are enqueued in a Treiber stack using
1655     * the same WaitQueueNodes as barriers.  They are resumed mainly
1656     * in preJoin, but are also woken on pool events that require all
1657     * threads to check run state.
1658     *
1659     * @param w the caller
1660     */
1661    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1662        WaitQueueNode node = null;
1663        int s;
1664        while (parallelism < runningCountOf(s = workerCounts)) {
1665            if (node == null)
1666                node = new WaitQueueNode(0, w);
1667            if (casWorkerCounts(s, s-1)) { // representation-dependent
1668                // push onto stack
1669                do {} while (!casSpareStack(node.next = spareStack, node));
1670                // block until released by resumeSpare
1671                node.awaitSpareRelease();
1672                return true;
1673            }
1674        }
1675        return false;
1676    }
1677
1678    /**
1679     * Tries to pop and resume a spare thread.
1680     *
1681     * @param updateCount if true, increment running count on success
1682     * @return true if successful
1683     */
1684    private boolean tryResumeSpare(boolean updateCount) {
1685        WaitQueueNode q;
1686        while ((q = spareStack) != null) {
1687            if (casSpareStack(q, q.next)) {
1688                if (updateCount)
1689                    updateRunningCount(1);
1690                q.signal();
1691                return true;
1692            }
1693        }
1694        return false;
1695    }
1696
1697    /**
1698     * Pops and resumes all spare threads. Same idea as ensureSync.
1699     *
1700     * @return true if any spares released
1701     */
1702    private boolean resumeAllSpares() {
1703        WaitQueueNode q;
1704        while ( (q = spareStack) != null) {
1705            if (casSpareStack(q, null)) {
1706                do {
1707                    updateRunningCount(1);
1708                    q.signal();
1709                } while ((q = q.next) != null);
1710                return true;
1711            }
1712        }
1713        return false;
1714    }
1715
1716    /**
1717     * Pops and shuts down excessive spare threads. Call only while
1718     * holding lock. This is not guaranteed to eliminate all excess
1719     * threads, only those suspended as spares, which are the ones
1720     * unlikely to be needed in the future.
1721     */
1722    private void trimSpares() {
1723        int surplus = totalCountOf(workerCounts) - parallelism;
1724        WaitQueueNode q;
1725        while (surplus > 0 && (q = spareStack) != null) {
1726            if (casSpareStack(q, null)) {
1727                do {
1728                    updateRunningCount(1);
1729                    ForkJoinWorkerThread w = q.thread;
1730                    if (w != null && surplus > 0 &&
1731                        runningCountOf(workerCounts) > 0 && w.shutdown())
1732                        --surplus;
1733                    q.signal();
1734                } while ((q = q.next) != null);
1735            }
1741          }
1742      }
1743  
# Line 1740 | Line 1745 | public class ForkJoinPool extends Abstra
1745       * Interface for extending managed parallelism for tasks running
1746       * in {@link ForkJoinPool}s.
1747       *
1748 <     * <p>A {@code ManagedBlocker} provides two methods.
1749 <     * Method {@code isReleasable} must return {@code true} if
1750 <     * blocking is not necessary. Method {@code block} blocks the
1751 <     * current thread if necessary (perhaps internally invoking
1752 <     * {@code isReleasable} before actually blocking).
1748 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1749 >     * {@code isReleasable} must return {@code true} if blocking is
1750 >     * not necessary. Method {@code block} blocks the current thread
1751 >     * if necessary (perhaps internally invoking {@code isReleasable}
1752 >     * before actually blocking). The unusual methods in this API
1753 >     * accommodate synchronizers that may, but don't usually, block
1754 >     * for long periods. Similarly, they allow more efficient internal
1755 >     * handling of cases in which additional workers may be, but
1756 >     * usually are not, needed to ensure sufficient parallelism.
1757 >     * Toward this end, implementations of method {@code isReleasable}
1758 >     * must be amenable to repeated invocation.
1759       *
1760       * <p>For example, here is a ManagedBlocker based on a
1761       * ReentrantLock:
# Line 1762 | Line 1773 | public class ForkJoinPool extends Abstra
1773       *     return hasLock || (hasLock = lock.tryLock());
1774       *   }
1775       * }}</pre>
1776 +     *
1777 +     * <p>Here is a class that possibly blocks waiting for an
1778 +     * item on a given queue:
1779 +     *  <pre> {@code
1780 +     * class QueueTaker<E> implements ManagedBlocker {
1781 +     *   final BlockingQueue<E> queue;
1782 +     *   volatile E item = null;
1783 +     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1784 +     *   public boolean block() throws InterruptedException {
1785 +     *     if (item == null)
1786 +     *       item = queue.take();
1787 +     *     return true;
1788 +     *   }
1789 +     *   public boolean isReleasable() {
1790 +     *     return item != null || (item = queue.poll()) != null;
1791 +     *   }
1792 +     *   public E getItem() { // call after pool.managedBlock completes
1793 +     *     return item;
1794 +     *   }
1795 +     * }}</pre>
1796       */
1797      public static interface ManagedBlocker {
1798          /**
# Line 1785 | Line 1816 | public class ForkJoinPool extends Abstra
1816       * Blocks in accord with the given blocker.  If the current thread
1817       * is a {@link ForkJoinWorkerThread}, this method possibly
1818       * arranges for a spare thread to be activated if necessary to
1819 <     * ensure parallelism while the current thread is blocked.
1789 <     *
1790 <     * <p>If {@code maintainParallelism} is {@code true} and the pool
1791 <     * supports it ({@link #getMaintainsParallelism}), this method
1792 <     * attempts to maintain the pool's nominal parallelism. Otherwise
1793 <     * it activates a thread only if necessary to avoid complete
1794 <     * starvation. This option may be preferable when blockages use
1795 <     * timeouts, or are almost always brief.
1819 >     * ensure sufficient parallelism while the current thread is blocked.
1820       *
1821       * <p>If the caller is not a {@link ForkJoinTask}, this method is
1822       * behaviorally equivalent to
# Line 1806 | Line 1830 | public class ForkJoinPool extends Abstra
1830       * first be expanded to ensure parallelism, and later adjusted.
1831       *
1832       * @param blocker the blocker
1809     * @param maintainParallelism if {@code true} and supported by
1810     * this pool, attempt to maintain the pool's nominal parallelism;
1811     * otherwise activate a thread only if necessary to avoid
1812     * complete starvation.
1833       * @throws InterruptedException if blocker.block did so
1834       */
1835 <    public static void managedBlock(ManagedBlocker blocker,
1816 <                                    boolean maintainParallelism)
1835 >    public static void managedBlock(ManagedBlocker blocker)
1836          throws InterruptedException {
1837          Thread t = Thread.currentThread();
1838 <        ForkJoinPool pool = ((t instanceof ForkJoinWorkerThread) ?
1839 <                             ((ForkJoinWorkerThread) t).pool : null);
1840 <        if (!blocker.isReleasable()) {
1841 <            try {
1842 <                if (pool == null ||
1843 <                    !pool.preBlock(blocker, maintainParallelism))
1825 <                    awaitBlocker(blocker);
1826 <            } finally {
1827 <                if (pool != null)
1828 <                    pool.updateRunningCount(1);
1829 <            }
1838 >        if (t instanceof ForkJoinWorkerThread) {
1839 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1840 >            w.pool.awaitBlocker(blocker);
1841 >        }
1842 >        else {
1843 >            do {} while (!blocker.isReleasable() && !blocker.block());
1844          }
1831    }
1832
1833    private static void awaitBlocker(ManagedBlocker blocker)
1834        throws InterruptedException {
1835        do {} while (!blocker.isReleasable() && !blocker.block());
1845      }
1846  
1847      // AbstractExecutorService overrides.  These rely on undocumented
# Line 1850 | Line 1859 | public class ForkJoinPool extends Abstra
1859      // Unsafe mechanics
1860  
1861      private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1853    private static final long eventCountOffset =
1854        objectFieldOffset("eventCount", ForkJoinPool.class);
1862      private static final long workerCountsOffset =
1863          objectFieldOffset("workerCounts", ForkJoinPool.class);
1864 <    private static final long runControlOffset =
1865 <        objectFieldOffset("runControl", ForkJoinPool.class);
1866 <    private static final long syncStackOffset =
1867 <        objectFieldOffset("syncStack",ForkJoinPool.class);
1868 <    private static final long spareStackOffset =
1869 <        objectFieldOffset("spareStack", ForkJoinPool.class);
1870 <
1871 <    private boolean casEventCount(long cmp, long val) {
1872 <        return UNSAFE.compareAndSwapLong(this, eventCountOffset, cmp, val);
1873 <    }
1867 <    private boolean casWorkerCounts(int cmp, int val) {
1868 <        return UNSAFE.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1869 <    }
1870 <    private boolean casRunControl(int cmp, int val) {
1871 <        return UNSAFE.compareAndSwapInt(this, runControlOffset, cmp, val);
1872 <    }
1873 <    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1874 <        return UNSAFE.compareAndSwapObject(this, spareStackOffset, cmp, val);
1875 <    }
1876 <    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1877 <        return UNSAFE.compareAndSwapObject(this, syncStackOffset, cmp, val);
1878 <    }
1864 >    private static final long runStateOffset =
1865 >        objectFieldOffset("runState", ForkJoinPool.class);
1866 >    private static final long eventCountOffset =
1867 >        objectFieldOffset("eventCount", ForkJoinPool.class);
1868 >    private static final long eventWaitersOffset =
1869 >        objectFieldOffset("eventWaiters", ForkJoinPool.class);
1870 >    private static final long stealCountOffset =
1871 >        objectFieldOffset("stealCount", ForkJoinPool.class);
1872 >    private static final long spareWaitersOffset =
1873 >        objectFieldOffset("spareWaiters", ForkJoinPool.class);
1874  
1875      private static long objectFieldOffset(String field, Class<?> klazz) {
1876          try {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines