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 1.24 by dl, Sat Jul 25 17:49:01 2009 UTC vs.
Revision 1.65 by dl, Wed Aug 18 14:05:27 2010 UTC

# Line 13 | Line 13 | import java.util.Arrays;
13   import java.util.Collection;
14   import java.util.Collections;
15   import java.util.List;
16 import java.util.concurrent.locks.Condition;
16   import java.util.concurrent.locks.LockSupport;
17   import java.util.concurrent.locks.ReentrantLock;
18   import java.util.concurrent.atomic.AtomicInteger;
19 < import java.util.concurrent.atomic.AtomicLong;
19 > import java.util.concurrent.CountDownLatch;
20  
21   /**
22 < * An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
23 < * ForkJoinPool provides the entry point for submissions from
24 < * non-ForkJoinTasks, as well as management and monitoring operations.
25 < * Normally a single ForkJoinPool is used for a large number of
27 < * submitted tasks. Otherwise, use would not usually outweigh the
28 < * construction and bookkeeping overhead of creating a large set of
29 < * threads.
22 > * An {@link ExecutorService} for running {@link ForkJoinTask}s.
23 > * A {@code ForkJoinPool} provides the entry point for submissions
24 > * from non-{@code ForkJoinTask} clients, as well as management and
25 > * monitoring operations.
26   *
27 < * <p>ForkJoinPools differ from other kinds of Executors mainly in
28 < * that they provide <em>work-stealing</em>: all threads in the pool
29 < * attempt to find and execute subtasks created by other active tasks
30 < * (eventually blocking if none exist). This makes them efficient when
31 < * most tasks spawn other subtasks (as do most ForkJoinTasks), as well
32 < * as the mixed execution of some plain Runnable- or Callable- based
33 < * activities along with ForkJoinTasks. When setting
34 < * {@code setAsyncMode}, a ForkJoinPools may also be appropriate for
35 < * use with fine-grained tasks that are never joined. Otherwise, other
40 < * ExecutorService implementations are typically more appropriate
41 < * choices.
27 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
28 > * ExecutorService} mainly by virtue of employing
29 > * <em>work-stealing</em>: all threads in the pool attempt to find and
30 > * execute subtasks created by other active tasks (eventually blocking
31 > * waiting for work if none exist). This enables efficient processing
32 > * when most tasks spawn other subtasks (as do most {@code
33 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
34 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
35 > * with event-style tasks that are never joined.
36   *
37 < * <p>A ForkJoinPool may be constructed with a given parallelism level
38 < * (target pool size), which it attempts to maintain by dynamically
39 < * adding, suspending, or resuming threads, even if some tasks are
40 < * waiting to join others. However, no such adjustments are performed
41 < * in the face of blocked IO or other unmanaged synchronization. The
42 < * nested {@code ManagedBlocker} interface enables extension of
43 < * the kinds of synchronization accommodated.  The target parallelism
44 < * level may also be changed dynamically ({@code setParallelism})
45 < * and thread construction can be limited using methods
52 < * {@code setMaximumPoolSize} and/or
53 < * {@code setMaintainsParallelism}.
37 > * <p>A {@code ForkJoinPool} is constructed with a given target
38 > * parallelism level; by default, equal to the number of available
39 > * processors. The pool attempts to maintain enough active (or
40 > * available) threads by dynamically adding, suspending, or resuming
41 > * internal worker threads, even if some tasks are stalled waiting to
42 > * join others. However, no such adjustments are guaranteed in the
43 > * face of blocked IO or other unmanaged synchronization. The nested
44 > * {@link ManagedBlocker} interface enables extension of the kinds of
45 > * synchronization accommodated.
46   *
47   * <p>In addition to execution and lifecycle control methods, this
48   * class provides status check methods (for example
49 < * {@code getStealCount}) that are intended to aid in developing,
49 > * {@link #getStealCount}) that are intended to aid in developing,
50   * tuning, and monitoring fork/join applications. Also, method
51 < * {@code toString} returns indications of pool state in a
51 > * {@link #toString} returns indications of pool state in a
52   * convenient form for informal monitoring.
53   *
54 + * <p> As is the case with other ExecutorServices, there are three
55 + * main task execution methods summarized in the following
56 + * table. These are designed to be used by clients not already engaged
57 + * in fork/join computations in the current pool.  The main forms of
58 + * these methods accept instances of {@code ForkJoinTask}, but
59 + * overloaded forms also allow mixed execution of plain {@code
60 + * Runnable}- or {@code Callable}- based activities as well.  However,
61 + * tasks that are already executing in a pool should normally
62 + * <em>NOT</em> use these pool execution methods, but instead use the
63 + * within-computation forms listed in the table.
64 + *
65 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
66 + *  <tr>
67 + *    <td></td>
68 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
69 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
70 + *  </tr>
71 + *  <tr>
72 + *    <td> <b>Arange async execution</td>
73 + *    <td> {@link #execute(ForkJoinTask)}</td>
74 + *    <td> {@link ForkJoinTask#fork}</td>
75 + *  </tr>
76 + *  <tr>
77 + *    <td> <b>Await and obtain result</td>
78 + *    <td> {@link #invoke(ForkJoinTask)}</td>
79 + *    <td> {@link ForkJoinTask#invoke}</td>
80 + *  </tr>
81 + *  <tr>
82 + *    <td> <b>Arrange exec and obtain Future</td>
83 + *    <td> {@link #submit(ForkJoinTask)}</td>
84 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
85 + *  </tr>
86 + * </table>
87 + *
88 + * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
89 + * used for all parallel task execution in a program or subsystem.
90 + * Otherwise, use would not usually outweigh the construction and
91 + * bookkeeping overhead of creating a large set of threads. For
92 + * example, a common pool could be used for the {@code SortTasks}
93 + * illustrated in {@link RecursiveAction}. Because {@code
94 + * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
95 + * daemon} mode, there is typically no need to explicitly {@link
96 + * #shutdown} such a pool upon program exit.
97 + *
98 + * <pre>
99 + * static final ForkJoinPool mainPool = new ForkJoinPool();
100 + * ...
101 + * public void sort(long[] array) {
102 + *   mainPool.invoke(new SortTask(array, 0, array.length));
103 + * }
104 + * </pre>
105 + *
106   * <p><b>Implementation notes</b>: This implementation restricts the
107   * maximum number of running threads to 32767. Attempts to create
108 < * pools with greater than the maximum result in
109 < * IllegalArgumentExceptions.
108 > * pools with greater than the maximum number result in
109 > * {@code IllegalArgumentException}.
110 > *
111 > * <p>This implementation rejects submitted tasks (that is, by throwing
112 > * {@link RejectedExecutionException}) only when the pool is shut down
113 > * or internal resources have been exhausted.
114   *
115   * @since 1.7
116   * @author Doug Lea
# Line 70 | Line 118 | import java.util.concurrent.atomic.Atomi
118   public class ForkJoinPool extends AbstractExecutorService {
119  
120      /*
121 <     * See the extended comments interspersed below for design,
122 <     * rationale, and walkthroughs.
123 <     */
124 <
125 <    /** Mask for packing and unpacking shorts */
126 <    private static final int  shortMask = 0xffff;
127 <
128 <    /** Max pool size -- must be a power of two minus 1 */
129 <    private static final int MAX_THREADS =  0x7FFF;
130 <
131 <    /**
132 <     * Factory for creating new ForkJoinWorkerThreads.  A
133 <     * ForkJoinWorkerThreadFactory must be defined and used for
134 <     * ForkJoinWorkerThread subclasses that extend base functionality
135 <     * or initialize threads with different contexts.
121 >     * Implementation Overview
122 >     *
123 >     * This class provides the central bookkeeping and control for a
124 >     * set of worker threads: Submissions from non-FJ threads enter
125 >     * into a submission queue. Workers take these tasks and typically
126 >     * split them into subtasks that may be stolen by other workers.
127 >     * The main work-stealing mechanics implemented in class
128 >     * ForkJoinWorkerThread give first priority to processing tasks
129 >     * from their own queues (LIFO or FIFO, depending on mode), then
130 >     * to randomized FIFO steals of tasks in other worker queues, and
131 >     * lastly to new submissions. These mechanics do not consider
132 >     * affinities, loads, cache localities, etc, so rarely provide the
133 >     * best possible performance on a given machine, but portably
134 >     * provide good throughput by averaging over these factors.
135 >     * (Further, even if we did try to use such information, we do not
136 >     * usually have a basis for exploiting it. For example, some sets
137 >     * of tasks profit from cache affinities, but others are harmed by
138 >     * cache pollution effects.)
139 >     *
140 >     * Beyond work-stealing support and essential bookkeeping, the
141 >     * main responsibility of this framework is to take actions when
142 >     * one worker is waiting to join a task stolen (or always held by)
143 >     * another.  Becauae we are multiplexing many tasks on to a pool
144 >     * of workers, we can't just let them block (as in Thread.join).
145 >     * We also cannot just reassign the joiner's run-time stack with
146 >     * another and replace it later, which would be a form of
147 >     * "continuation", that even if possible is not necessarily a good
148 >     * idea. Given that the creation costs of most threads on most
149 >     * systems mainly surrounds setting up runtime stacks, thread
150 >     * creation and switching is usually not much more expensive than
151 >     * stack creation and switching, and is more flexible). Instead we
152 >     * combine two tactics:
153 >     *
154 >     *   Helping: Arranging for the joiner to execute some task that it
155 >     *      would be running if the steal had not occurred.  Method
156 >     *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
157 >     *      links to try to find such a task.
158 >     *
159 >     *   Compensating: Unless there are already enough live threads,
160 >     *      method helpMaintainParallelism() may create or or
161 >     *      re-activate a spare thread to compensate for blocked
162 >     *      joiners until they unblock.
163 >     *
164 >     * Because the determining existence of conservatively safe
165 >     * helping targets, the availability of already-created spares,
166 >     * and the apparent need to create new spares are all racy and
167 >     * require heuristic guidance, we rely on multiple retries of
168 >     * each. Further, because it is impossible to keep exactly the
169 >     * target (parallelism) number of threads running at any given
170 >     * time, we allow compensation during joins to fail, and enlist
171 >     * all other threads to help out whenever they are not otherwise
172 >     * occupied (i.e., mainly in method preStep).
173 >     *
174 >     * The ManagedBlocker extension API can't use helping so relies
175 >     * only on compensation in method awaitBlocker.
176 >     *
177 >     * The main throughput advantages of work-stealing stem from
178 >     * decentralized control -- workers mostly steal tasks from each
179 >     * other. We do not want to negate this by creating bottlenecks
180 >     * implementing other management responsibilities. So we use a
181 >     * collection of techniques that avoid, reduce, or cope well with
182 >     * contention. These entail several instances of bit-packing into
183 >     * CASable fields to maintain only the minimally required
184 >     * atomicity. To enable such packing, we restrict maximum
185 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
186 >     * unbalanced increments and decrements) to fit into a 16 bit
187 >     * field, which is far in excess of normal operating range.  Even
188 >     * though updates to some of these bookkeeping fields do sometimes
189 >     * contend with each other, they don't normally cache-contend with
190 >     * updates to others enough to warrant memory padding or
191 >     * isolation. So they are all held as fields of ForkJoinPool
192 >     * objects.  The main capabilities are as follows:
193 >     *
194 >     * 1. Creating and removing workers. Workers are recorded in the
195 >     * "workers" array. This is an array as opposed to some other data
196 >     * structure to support index-based random steals by workers.
197 >     * Updates to the array recording new workers and unrecording
198 >     * terminated ones are protected from each other by a lock
199 >     * (workerLock) but the array is otherwise concurrently readable,
200 >     * and accessed directly by workers. To simplify index-based
201 >     * operations, the array size is always a power of two, and all
202 >     * readers must tolerate null slots. Currently, all worker thread
203 >     * creation is on-demand, triggered by task submissions,
204 >     * replacement of terminated workers, and/or compensation for
205 >     * blocked workers. However, all other support code is set up to
206 >     * work with other policies.
207 >     *
208 >     * To ensure that we do not hold on to worker references that
209 >     * would prevent GC, ALL accesses to workers are via indices into
210 >     * the workers array (which is one source of some of the unusual
211 >     * code constructions here). In essence, the workers array serves
212 >     * as a WeakReference mechanism. Thus for example the event queue
213 >     * stores worker indices, not worker references. Access to the
214 >     * workers in associated methods (for example releaseEventWaiters)
215 >     * must both index-check and null-check the IDs. All such accesses
216 >     * ignore bad IDs by returning out early from what they are doing,
217 >     * since this can only be associated with shutdown, in which case
218 >     * it is OK to give up. On termination, we just clobber these
219 >     * data structures without trying to use them.
220 >     *
221 >     * 2. Bookkeeping for dynamically adding and removing workers. We
222 >     * aim to approximately maintain the given level of parallelism.
223 >     * When some workers are known to be blocked (on joins or via
224 >     * ManagedBlocker), we may create or resume others to take their
225 >     * place until they unblock (see below). Implementing this
226 >     * requires counts of the number of "running" threads (i.e., those
227 >     * that are neither blocked nor artifically suspended) as well as
228 >     * the total number.  These two values are packed into one field,
229 >     * "workerCounts" because we need accurate snapshots when deciding
230 >     * to create, resume or suspend.  Note however that the
231 >     * correspondance of these counts to reality is not guaranteed. In
232 >     * particular updates for unblocked threads may lag until they
233 >     * actually wake up.
234 >     *
235 >     * 3. Maintaining global run state. The run state of the pool
236 >     * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
237 >     * those in other Executor implementations, as well as a count of
238 >     * "active" workers -- those that are, or soon will be, or
239 >     * recently were executing tasks. The runLevel and active count
240 >     * are packed together in order to correctly trigger shutdown and
241 >     * termination. Without care, active counts can be subject to very
242 >     * high contention.  We substantially reduce this contention by
243 >     * relaxing update rules.  A worker must claim active status
244 >     * prospectively, by activating if it sees that a submitted or
245 >     * stealable task exists (it may find after activating that the
246 >     * task no longer exists). It stays active while processing this
247 >     * task (if it exists) and any other local subtasks it produces,
248 >     * until it cannot find any other tasks. It then tries
249 >     * inactivating (see method preStep), but upon update contention
250 >     * instead scans for more tasks, later retrying inactivation if it
251 >     * doesn't find any.
252 >     *
253 >     * 4. Managing idle workers waiting for tasks. We cannot let
254 >     * workers spin indefinitely scanning for tasks when none are
255 >     * available. On the other hand, we must quickly prod them into
256 >     * action when new tasks are submitted or generated.  We
257 >     * park/unpark these idle workers using an event-count scheme.
258 >     * Field eventCount is incremented upon events that may enable
259 >     * workers that previously could not find a task to now find one:
260 >     * Submission of a new task to the pool, or another worker pushing
261 >     * a task onto a previously empty queue.  (We also use this
262 >     * mechanism for configuration and termination actions that
263 >     * require wakeups of idle workers).  Each worker maintains its
264 >     * last known event count, and blocks when a scan for work did not
265 >     * find a task AND its lastEventCount matches the current
266 >     * eventCount. Waiting idle workers are recorded in a variant of
267 >     * Treiber stack headed by field eventWaiters which, when nonzero,
268 >     * encodes the thread index and count awaited for by the worker
269 >     * thread most recently calling eventSync. This thread in turn has
270 >     * a record (field nextEventWaiter) for the next waiting worker.
271 >     * In addition to allowing simpler decisions about need for
272 >     * wakeup, the event count bits in eventWaiters serve the role of
273 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
274 >     * released threads also try to release others (but give up upon
275 >     * contention to reduce useless flailing).  The net effect is a
276 >     * tree-like diffusion of signals, where released threads (and
277 >     * possibly others) help with unparks.  To further reduce
278 >     * contention effects a bit, failed CASes to increment field
279 >     * eventCount are tolerated without retries in signalWork.
280 >     * Conceptually they are merged into the same event, which is OK
281 >     * when their only purpose is to enable workers to scan for work.
282 >     *
283 >     * 5. Managing suspension of extra workers. When a worker is about
284 >     * to block waiting for a join (or via ManagedBlockers), we may
285 >     * create a new thread to maintain parallelism level, or at least
286 >     * avoid starvation. Usually, extra threads are needed for only
287 >     * very short periods, yet join dependencies are such that we
288 >     * sometimes need them in bursts. Rather than create new threads
289 >     * each time this happens, we suspend no-longer-needed extra ones
290 >     * as "spares". For most purposes, we don't distinguish "extra"
291 >     * spare threads from normal "core" threads: On each call to
292 >     * preStep (the only point at which we can do this) a worker
293 >     * checks to see if there are now too many running workers, and if
294 >     * so, suspends itself.  Method helpMaintainParallelism looks for
295 >     * suspended threads to resume before considering creating a new
296 >     * replacement. The spares themselves are encoded on another
297 >     * variant of a Treiber Stack, headed at field "spareWaiters".
298 >     * Note that the use of spares is intrinsically racy.  One thread
299 >     * may become a spare at about the same time as another is
300 >     * needlessly being created. We counteract this and related slop
301 >     * in part by requiring resumed spares to immediately recheck (in
302 >     * preStep) to see whether they they should re-suspend.
303 >     *
304 >     * 6. Killing off unneeded workers. The Spare and Event queues use
305 >     * similar mechanisms to shed unused workers: The oldest (first)
306 >     * waiter uses a timed rather than hard wait. When this wait times
307 >     * out without a normal wakeup, it tries to shutdown any one (for
308 >     * convenience the newest) other waiter via tryShutdownSpare or
309 >     * tryShutdownWaiter, respectively. The wakeup rates for spares
310 >     * are much shorter than for waiters. Together, they will
311 >     * eventually reduce the number of worker threads to a minimum of
312 >     * one after a long enough period without use.
313 >     *
314 >     * 7. Deciding when to create new workers. The main dynamic
315 >     * control in this class is deciding when to create extra threads
316 >     * in method helpMaintainParallelism. We would like to keep
317 >     * exactly #parallelism threads running, which is an impossble
318 >     * task. We always need to create one when the number of running
319 >     * threads would become zero and all workers are busy. Beyond
320 >     * this, we must rely on heuristics that work well in the the
321 >     * presence of transients phenomena such as GC stalls, dynamic
322 >     * compilation, and wake-up lags. These transients are extremely
323 >     * common -- we are normally trying to fully saturate the CPUs on
324 >     * a machine, so almost any activity other than running tasks
325 >     * impedes accuracy. Our main defense is to allow some slack in
326 >     * creation thresholds, using rules that reflect the fact that the
327 >     * more threads we have running, the more likely that we are
328 >     * underestimating the number running threads. (We also include
329 >     * some heuristic use of Thread.yield when all workers appear to
330 >     * be busy, to improve likelihood of counts settling.) The rules
331 >     * also better cope with the fact that some of the methods in this
332 >     * class tend to never become compiled (but are interpreted), so
333 >     * some components of the entire set of controls might execute 100
334 >     * times faster than others. And similarly for cases where the
335 >     * apparent lack of work is just due to GC stalls and other
336 >     * transient system activity.
337 >     *
338 >     * Beware that there is a lot of representation-level coupling
339 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
340 >     * ForkJoinTask.  For example, direct access to "workers" array by
341 >     * workers, and direct access to ForkJoinTask.status by both
342 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
343 >     * trying to reduce this, since any associated future changes in
344 >     * representations will need to be accompanied by algorithmic
345 >     * changes anyway.
346 >     *
347 >     * Style notes: There are lots of inline assignments (of form
348 >     * "while ((local = field) != 0)") which are usually the simplest
349 >     * way to ensure the required read orderings (which are sometimes
350 >     * critical). Also several occurrences of the unusual "do {}
351 >     * while(!cas...)" which is the simplest way to force an update of
352 >     * a CAS'ed variable. There are also other coding oddities that
353 >     * help some methods perform reasonably even when interpreted (not
354 >     * compiled), at the expense of some messy constructions that
355 >     * reduce byte code counts.
356 >     *
357 >     * The order of declarations in this file is: (1) statics (2)
358 >     * fields (along with constants used when unpacking some of them)
359 >     * (3) internal control methods (4) callbacks and other support
360 >     * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
361 >     * methods (plus a few little helpers).
362 >     */
363 >
364 >    /**
365 >     * Factory for creating new {@link ForkJoinWorkerThread}s.
366 >     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
367 >     * for {@code ForkJoinWorkerThread} subclasses that extend base
368 >     * functionality or initialize threads with different contexts.
369       */
370      public static interface ForkJoinWorkerThreadFactory {
371          /**
372           * Returns a new worker thread operating in the given pool.
373           *
374           * @param pool the pool this thread works in
375 <         * @throws NullPointerException if pool is null
375 >         * @throws NullPointerException if the pool is null
376           */
377          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
378      }
# Line 100 | Line 381 | public class ForkJoinPool extends Abstra
381       * Default ForkJoinWorkerThreadFactory implementation; creates a
382       * new ForkJoinWorkerThread.
383       */
384 <    static class  DefaultForkJoinWorkerThreadFactory
384 >    static class DefaultForkJoinWorkerThreadFactory
385          implements ForkJoinWorkerThreadFactory {
386          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
387 <            try {
107 <                return new ForkJoinWorkerThread(pool);
108 <            } catch (OutOfMemoryError oom)  {
109 <                return null;
110 <            }
387 >            return new ForkJoinWorkerThread(pool);
388          }
389      }
390  
# Line 143 | Line 420 | public class ForkJoinPool extends Abstra
420          new AtomicInteger();
421  
422      /**
423 <     * Array holding all worker threads in the pool. Initialized upon
424 <     * first use. Array size must be a power of two.  Updates and
425 <     * replacements are protected by workerLock, but it is always kept
426 <     * in a consistent enough state to be randomly accessed without
427 <     * locking by workers performing work-stealing.
423 >     * The wakeup interval (in nanoseconds) for the oldest worker
424 >     * worker waiting for an event invokes tryShutdownWaiter to shrink
425 >     * the number of workers.  The exact value does not matter too
426 >     * much, but should be long enough to slowly release resources
427 >     * during long periods without use without disrupting normal use.
428 >     */
429 >    private static final long SHRINK_RATE_NANOS =
430 >        60L * 1000L * 1000L * 1000L; // one minute
431 >
432 >    /**
433 >     * Absolute bound for parallelism level. Twice this number plus
434 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
435 >     * word-packing for some counts and indices.
436 >     */
437 >    private static final int MAX_WORKERS   = 0x7fff;
438 >
439 >    /**
440 >     * Array holding all worker threads in the pool.  Array size must
441 >     * be a power of two.  Updates and replacements are protected by
442 >     * workerLock, but the array is always kept in a consistent enough
443 >     * state to be randomly accessed without locking by workers
444 >     * performing work-stealing, as well as other traversal-based
445 >     * methods in this class. All readers must tolerate that some
446 >     * array slots may be null.
447       */
448      volatile ForkJoinWorkerThread[] workers;
449  
450      /**
451 <     * Lock protecting access to workers.
451 >     * Queue for external submissions.
452       */
453 <    private final ReentrantLock workerLock;
453 >    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
454  
455      /**
456 <     * Condition for awaitTermination.
456 >     * Lock protecting updates to workers array.
457       */
458 <    private final Condition termination;
458 >    private final ReentrantLock workerLock;
459  
460      /**
461 <     * The uncaught exception handler used when any worker
166 <     * abruptly terminates
461 >     * Latch released upon termination.
462       */
463 <    private Thread.UncaughtExceptionHandler ueh;
463 >    private final Phaser termination;
464  
465      /**
466       * Creation factory for worker threads.
# Line 173 | Line 468 | public class ForkJoinPool extends Abstra
468      private final ForkJoinWorkerThreadFactory factory;
469  
470      /**
176     * Head of stack of threads that were created to maintain
177     * parallelism when other threads blocked, but have since
178     * suspended when the parallelism level rose.
179     */
180    private volatile WaitQueueNode spareStack;
181
182    /**
471       * Sum of per-thread steal counts, updated only when threads are
472       * idle or terminating.
473       */
474 <    private final AtomicLong stealCount;
474 >    private volatile long stealCount;
475  
476      /**
477 <     * Queue for external submissions.
477 >     * Encoded record of top of treiber stack of threads waiting for
478 >     * events. The top 32 bits contain the count being waited for. The
479 >     * bottom 16 bits contains one plus the pool index of waiting
480 >     * worker thread. (Bits 16-31 are unused.)
481       */
482 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
482 >    private volatile long eventWaiters;
483 >
484 >    private static final int  EVENT_COUNT_SHIFT = 32;
485 >    private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
486  
487      /**
488 <     * Head of Treiber stack for barrier sync. See below for explanation.
488 >     * A counter for events that may wake up worker threads:
489 >     *   - Submission of a new task to the pool
490 >     *   - A worker pushing a task on an empty queue
491 >     *   - termination
492       */
493 <    private volatile WaitQueueNode syncStack;
493 >    private volatile int eventCount;
494  
495      /**
496 <     * The count for event barrier
496 >     * Encoded record of top of treiber stack of spare threads waiting
497 >     * for resumption. The top 16 bits contain an arbitrary count to
498 >     * avoid ABA effects. The bottom 16bits contains one plus the pool
499 >     * index of waiting worker thread.
500       */
501 <    private volatile long eventCount;
501 >    private volatile int spareWaiters;
502 >
503 >    private static final int SPARE_COUNT_SHIFT = 16;
504 >    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
505  
506      /**
507 <     * Pool number, just for assigning useful names to worker threads
507 >     * Lifecycle control. The low word contains the number of workers
508 >     * that are (probably) executing tasks. This value is atomically
509 >     * incremented before a worker gets a task to run, and decremented
510 >     * when worker has no tasks and cannot find any.  Bits 16-18
511 >     * contain runLevel value. When all are zero, the pool is
512 >     * running. Level transitions are monotonic (running -> shutdown
513 >     * -> terminating -> terminated) so each transition adds a bit.
514 >     * These are bundled together to ensure consistent read for
515 >     * termination checks (i.e., that runLevel is at least SHUTDOWN
516 >     * and active threads is zero).
517 >     *
518 >     * Notes: Most direct CASes are dependent on these bitfield
519 >     * positions.  Also, this field is non-private to enable direct
520 >     * performance-sensitive CASes in ForkJoinWorkerThread.
521       */
522 <    private final int poolNumber;
522 >    volatile int runState;
523 >
524 >    // Note: The order among run level values matters.
525 >    private static final int RUNLEVEL_SHIFT     = 16;
526 >    private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
527 >    private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
528 >    private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
529 >    private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
530  
531      /**
532 <     * The maximum allowed pool size
532 >     * Holds number of total (i.e., created and not yet terminated)
533 >     * and running (i.e., not blocked on joins or other managed sync)
534 >     * threads, packed together to ensure consistent snapshot when
535 >     * making decisions about creating and suspending spare
536 >     * threads. Updated only by CAS. Note that adding a new worker
537 >     * requires incrementing both counts, since workers start off in
538 >     * running state.
539       */
540 <    private volatile int maxPoolSize;
540 >    private volatile int workerCounts;
541 >
542 >    private static final int TOTAL_COUNT_SHIFT  = 16;
543 >    private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
544 >    private static final int ONE_RUNNING        = 1;
545 >    private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
546  
547      /**
548 <     * The desired parallelism level, updated only under workerLock.
548 >     * The target parallelism level.
549 >     * Accessed directly by ForkJoinWorkerThreads.
550       */
551 <    private volatile int parallelism;
551 >    final int parallelism;
552  
553      /**
554       * True if use local fifo, not default lifo, for local polling
555 +     * Read by, and replicated by ForkJoinWorkerThreads
556       */
557 <    private volatile boolean locallyFifo;
557 >    final boolean locallyFifo;
558  
559      /**
560 <     * Holds number of total (i.e., created and not yet terminated)
561 <     * and running (i.e., not blocked on joins or other managed sync)
226 <     * threads, packed into one int to ensure consistent snapshot when
227 <     * making decisions about creating and suspending spare
228 <     * threads. Updated only by CAS.  Note: CASes in
229 <     * updateRunningCount and preJoin assume that running active count
230 <     * is in low word, so need to be modified if this changes.
560 >     * The uncaught exception handler used when any worker abruptly
561 >     * terminates.
562       */
563 <    private volatile int workerCounts;
563 >    private final Thread.UncaughtExceptionHandler ueh;
564 >
565 >    /**
566 >     * Pool number, just for assigning useful names to worker threads
567 >     */
568 >    private final int poolNumber;
569  
570 <    private static int totalCountOf(int s)           { return s >>> 16;  }
571 <    private static int runningCountOf(int s)         { return s & shortMask; }
572 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
570 >
571 >    // Utilities for CASing fields. Note that most of these
572 >    // are usually manually inlined by callers
573  
574      /**
575 <     * Adds delta (which may be negative) to running count.  This must
240 <     * be called before (with negative arg) and after (with positive)
241 <     * any managed synchronization (i.e., mainly, joins).
242 <     *
243 <     * @param delta the number to add
575 >     * Increments running count part of workerCounts
576       */
577 <    final void updateRunningCount(int delta) {
578 <        int s;
579 <        do {} while (!casWorkerCounts(s = workerCounts, s + delta));
577 >    final void incrementRunningCount() {
578 >        int c;
579 >        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
580 >                                               c = workerCounts,
581 >                                               c + ONE_RUNNING));
582      }
583  
584      /**
585 <     * Adds delta (which may be negative) to both total and running
252 <     * count.  This must be called upon creation and termination of
253 <     * worker threads.
254 <     *
255 <     * @param delta the number to add
585 >     * Tries to decrement running count unless already zero
586       */
587 <    private void updateWorkerCount(int delta) {
588 <        int d = delta + (delta << 16); // add to both lo and hi parts
589 <        int s;
590 <        do {} while (!casWorkerCounts(s = workerCounts, s + d));
587 >    final boolean tryDecrementRunningCount() {
588 >        int wc = workerCounts;
589 >        if ((wc & RUNNING_COUNT_MASK) == 0)
590 >            return false;
591 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
592 >                                        wc, wc - ONE_RUNNING);
593      }
594  
595      /**
596 <     * Lifecycle control. High word contains runState, low word
597 <     * contains the number of workers that are (probably) executing
598 <     * tasks. This value is atomically incremented before a worker
599 <     * gets a task to run, and decremented when worker has no tasks
600 <     * and cannot find any. These two fields are bundled together to
269 <     * support correct termination triggering.  Note: activeCount
270 <     * CAS'es cheat by assuming active count is in low word, so need
271 <     * to be modified if this changes
596 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
597 >     * (rarely) necessary when other count updates lag.
598 >     *
599 >     * @param dr -- either zero or ONE_RUNNING
600 >     * @param dt == either zero or ONE_TOTAL
601       */
602 <    private volatile int runControl;
603 <
604 <    // RunState values. Order among values matters
605 <    private static final int RUNNING     = 0;
606 <    private static final int SHUTDOWN    = 1;
607 <    private static final int TERMINATING = 2;
608 <    private static final int TERMINATED  = 3;
602 >    private void decrementWorkerCounts(int dr, int dt) {
603 >        for (;;) {
604 >            int wc = workerCounts;
605 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
606 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
607 >                if ((runState & TERMINATED) != 0)
608 >                    return; // lagging termination on a backout
609 >                Thread.yield();
610 >            }
611 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
612 >                                         wc, wc - (dr + dt)))
613 >                return;
614 >        }
615 >    }
616  
617 <    private static int runStateOf(int c)             { return c >>> 16; }
618 <    private static int activeCountOf(int c)          { return c & shortMask; }
619 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
617 >    /**
618 >     * Increments event count
619 >     */
620 >    private void advanceEventCount() {
621 >        int c;
622 >        do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset,
623 >                                              c = eventCount, c+1));
624 >    }
625  
626      /**
627       * Tries incrementing active count; fails on contention.
628 <     * Called by workers before/during executing tasks.
628 >     * Called by workers before executing tasks.
629       *
630       * @return true on success
631       */
632      final boolean tryIncrementActiveCount() {
633 <        int c = runControl;
634 <        return casRunControl(c, c+1);
633 >        int c;
634 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
635 >                                        c = runState, c + 1);
636      }
637  
638      /**
639       * Tries decrementing active count; fails on contention.
640 <     * Possibly triggers termination on success.
299 <     * Called by workers when they can't find tasks.
300 <     *
301 <     * @return true on success
640 >     * Called when workers cannot find tasks to run.
641       */
642      final boolean tryDecrementActiveCount() {
643 <        int c = runControl;
644 <        int nextc = c - 1;
645 <        if (!casRunControl(c, nextc))
307 <            return false;
308 <        if (canTerminateOnShutdown(nextc))
309 <            terminateOnShutdown();
310 <        return true;
643 >        int c;
644 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
645 >                                        c = runState, c - 1);
646      }
647  
648      /**
649 <     * Returns true if argument represents zero active count and
650 <     * nonzero runstate, which is the triggering condition for
316 <     * terminating on shutdown.
649 >     * Advances to at least the given level. Returns true if not
650 >     * already in at least the given level.
651       */
652 <    private static boolean canTerminateOnShutdown(int c) {
653 <        // i.e. least bit is nonzero runState bit
654 <        return ((c & -c) >>> 16) != 0;
652 >    private boolean advanceRunLevel(int level) {
653 >        for (;;) {
654 >            int s = runState;
655 >            if ((s & level) != 0)
656 >                return false;
657 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
658 >                return true;
659 >        }
660      }
661  
662 +    // workers array maintenance
663 +
664      /**
665 <     * Transition run state to at least the given state. Return true
325 <     * if not already at least given state.
665 >     * Records and returns a workers array index for new worker.
666       */
667 <    private boolean transitionRunStateTo(int state) {
668 <        for (;;) {
669 <            int c = runControl;
670 <            if (runStateOf(c) >= state)
671 <                return false;
672 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
673 <                return true;
667 >    private int recordWorker(ForkJoinWorkerThread w) {
668 >        // Try using slot totalCount-1. If not available, scan and/or resize
669 >        int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
670 >        final ReentrantLock lock = this.workerLock;
671 >        lock.lock();
672 >        try {
673 >            ForkJoinWorkerThread[] ws = workers;
674 >            int n = ws.length;
675 >            if (k < 0 || k >= n || ws[k] != null) {
676 >                for (k = 0; k < n && ws[k] != null; ++k)
677 >                    ;
678 >                if (k == n)
679 >                    ws = Arrays.copyOf(ws, n << 1);
680 >            }
681 >            ws[k] = w;
682 >            workers = ws; // volatile array write ensures slot visibility
683 >        } finally {
684 >            lock.unlock();
685          }
686 +        return k;
687      }
688  
689      /**
690 <     * Controls whether to add spares to maintain parallelism
690 >     * Nulls out record of worker in workers array
691       */
692 <    private volatile boolean maintainsParallelism;
692 >    private void forgetWorker(ForkJoinWorkerThread w) {
693 >        int idx = w.poolIndex;
694 >        // Locking helps method recordWorker avoid unecessary expansion
695 >        final ReentrantLock lock = this.workerLock;
696 >        lock.lock();
697 >        try {
698 >            ForkJoinWorkerThread[] ws = workers;
699 >            if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
700 >                ws[idx] = null;
701 >        } finally {
702 >            lock.unlock();
703 >        }
704 >    }
705  
706 <    // Constructors
706 >    // adding and removing workers
707  
708      /**
709 <     * Creates a ForkJoinPool with a pool size equal to the number of
710 <     * processors available on the system, using the default
711 <     * ForkJoinWorkerThreadFactory.
709 >     * Tries to create and add new worker. Assumes that worker counts
710 >     * are already updated to accommodate the worker, so adjusts on
711 >     * failure.
712       *
713 <     * @throws SecurityException if a security manager exists and
350 <     *         the caller is not permitted to modify threads
351 <     *         because it does not hold {@link
352 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
713 >     * @return the worker, or null on failure
714       */
715 <    public ForkJoinPool() {
716 <        this(Runtime.getRuntime().availableProcessors(),
717 <             defaultForkJoinWorkerThreadFactory);
715 >    private ForkJoinWorkerThread addWorker() {
716 >        ForkJoinWorkerThread w = null;
717 >        try {
718 >            w = factory.newThread(this);
719 >        } finally { // Adjust on either null or exceptional factory return
720 >            if (w == null) {
721 >                decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
722 >                tryTerminate(false); // in case of failure during shutdown
723 >            }
724 >        }
725 >        if (w != null) {
726 >            w.start(recordWorker(w), ueh);
727 >            advanceEventCount();
728 >        }
729 >        return w;
730      }
731  
732      /**
733 <     * Creates a ForkJoinPool with the indicated parallelism level
734 <     * threads and using the default ForkJoinWorkerThreadFactory.
733 >     * Final callback from terminating worker.  Removes record of
734 >     * worker from array, and adjusts counts. If pool is shutting
735 >     * down, tries to complete terminatation.
736       *
737 <     * @param parallelism the number of worker threads
364 <     * @throws IllegalArgumentException if parallelism less than or
365 <     * equal to zero
366 <     * @throws SecurityException if a security manager exists and
367 <     *         the caller is not permitted to modify threads
368 <     *         because it does not hold {@link
369 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
737 >     * @param w the worker
738       */
739 <    public ForkJoinPool(int parallelism) {
740 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
739 >    final void workerTerminated(ForkJoinWorkerThread w) {
740 >        forgetWorker(w);
741 >        decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
742 >        while (w.stealCount != 0) // collect final count
743 >            tryAccumulateStealCount(w);
744 >        tryTerminate(false);
745      }
746  
747 +    // Waiting for and signalling events
748 +
749      /**
750 <     * Creates a ForkJoinPool with parallelism equal to the number of
751 <     * processors available on the system and using the given
752 <     * ForkJoinWorkerThreadFactory.
753 <     *
380 <     * @param factory the factory for creating new threads
381 <     * @throws NullPointerException if factory is null
382 <     * @throws SecurityException if a security manager exists and
383 <     *         the caller is not permitted to modify threads
384 <     *         because it does not hold {@link
385 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
750 >     * Releases workers blocked on a count not equal to current count.
751 >     * Normally called after precheck that eventWaiters isn't zero to
752 >     * avoid wasted array checks. Gives up upon a change in count or
753 >     * contention, letting other workers take over.
754       */
755 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
756 <        this(Runtime.getRuntime().availableProcessors(), factory);
755 >    private void releaseEventWaiters() {
756 >        ForkJoinWorkerThread[] ws = workers;
757 >        int n = ws.length;
758 >        long h = eventWaiters;
759 >        int ec = eventCount;
760 >        ForkJoinWorkerThread w; int id;
761 >        while ((int)(h >>> EVENT_COUNT_SHIFT) != ec &&
762 >               (id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
763 >               id < n && (w = ws[id]) != null &&
764 >               UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
765 >                                         h,  h = w.nextWaiter)) {
766 >            LockSupport.unpark(w);
767 >            if (eventWaiters != h || eventCount != ec)
768 >                break;
769 >        }
770      }
771  
772      /**
773 <     * Creates a ForkJoinPool with the given parallelism and factory.
773 >     * Tries to advance eventCount and releases waiters. Called only
774 >     * from workers.
775 >     */
776 >    final void signalWork() {
777 >        int c; // try to increment event count -- CAS failure OK
778 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
779 >        if (eventWaiters != 0L)
780 >            releaseEventWaiters();
781 >    }
782 >
783 >    /**
784 >     * Adds the given worker to event queue and blocks until
785 >     * terminating or event count advances from the workers
786 >     * lastEventCount value
787       *
788 <     * @param parallelism the targeted number of worker threads
395 <     * @param factory the factory for creating new threads
396 <     * @throws IllegalArgumentException if parallelism less than or
397 <     * equal to zero, or greater than implementation limit
398 <     * @throws NullPointerException if factory is null
399 <     * @throws SecurityException if a security manager exists and
400 <     *         the caller is not permitted to modify threads
401 <     *         because it does not hold {@link
402 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
788 >     * @param w the calling worker thread
789       */
790 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
791 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
792 <            throw new IllegalArgumentException();
793 <        if (factory == null)
794 <            throw new NullPointerException();
795 <        checkPermission();
796 <        this.factory = factory;
797 <        this.parallelism = parallelism;
798 <        this.maxPoolSize = MAX_THREADS;
799 <        this.maintainsParallelism = true;
800 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
801 <        this.workerLock = new ReentrantLock();
802 <        this.termination = workerLock.newCondition();
803 <        this.stealCount = new AtomicLong();
418 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
419 <        // worker array and workers are lazily constructed
790 >    private void eventSync(ForkJoinWorkerThread w) {
791 >        int ec = w.lastEventCount;
792 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
793 >        long h;
794 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
795 >               (((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
796 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
797 >               eventCount == ec) {
798 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
799 >                                          w.nextWaiter = h, nh)) {
800 >                awaitEvent(w, ec);
801 >                break;
802 >            }
803 >        }
804      }
805  
806      /**
807 <     * Creates a new worker thread using factory.
807 >     * Blocks the given worker (that has already been entered as an
808 >     * event waiter) until terminating or event count advances from
809 >     * the given value. The oldest (first) waiter uses a timed wait to
810 >     * occasionally one-by-one shrink the number of workers (to a
811 >     * minumum of one) if the pool has not been used for extended
812 >     * periods.
813       *
814 <     * @param index the index to assign worker
815 <     * @return new worker, or null of factory failed
814 >     * @param w the calling worker thread
815 >     * @param ec the count
816       */
817 <    private ForkJoinWorkerThread createWorker(int index) {
818 <        Thread.UncaughtExceptionHandler h = ueh;
819 <        ForkJoinWorkerThread w = factory.newThread(this);
820 <        if (w != null) {
821 <            w.poolIndex = index;
822 <            w.setDaemon(true);
823 <            w.setAsyncMode(locallyFifo);
824 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
825 <            if (h != null)
826 <                w.setUncaughtExceptionHandler(h);
817 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
818 >        while (eventCount == ec) {
819 >            if (tryAccumulateStealCount(w)) { // transfer while idle
820 >                boolean untimed = (w.nextWaiter != 0L ||
821 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
822 >                long startTime = untimed? 0 : System.nanoTime();
823 >                Thread.interrupted();         // clear/ignore interrupt
824 >                if (eventCount != ec || !w.isRunning() ||
825 >                    runState >= TERMINATING)  // recheck after clear
826 >                    break;
827 >                if (untimed)
828 >                    LockSupport.park(w);
829 >                else {
830 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
831 >                    if (eventCount != ec || !w.isRunning() ||
832 >                        runState >= TERMINATING)
833 >                        break;
834 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
835 >                        tryShutdownWaiter(ec);
836 >                }
837 >            }
838          }
439        return w;
839      }
840  
841      /**
842 <     * Returns a good size for worker array given pool size.
843 <     * Currently requires size to be a power of two.
842 >     * Callback from the oldest waiter in awaitEvent waking up after a
843 >     * period of non-use. Tries (once) to shutdown an event waiter (or
844 >     * a spare, if one exists). Note that we don't need CAS or locks
845 >     * here because the method is called only from one thread
846 >     * occasionally waking (and even misfires are OK). Note that
847 >     * until the shutdown worker fully terminates, workerCounts
848 >     * will overestimate total count, which is tolerable.
849 >     *
850 >     * @param ec the event count waited on by caller (to abort
851 >     * attempt if count has since changed).
852 >     */
853 >    private void tryShutdownWaiter(int ec) {
854 >        if (spareWaiters != 0) { // prefer killing spares
855 >            tryShutdownSpare();
856 >            return;
857 >        }
858 >        ForkJoinWorkerThread[] ws = workers;
859 >        int n = ws.length;
860 >        long h = eventWaiters;
861 >        ForkJoinWorkerThread w; int id; long nh;
862 >        if (runState == 0 &&
863 >            submissionQueue.isEmpty() &&
864 >            eventCount == ec &&
865 >            (id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
866 >            id < n && (w = ws[id]) != null &&
867 >            (nh = w.nextWaiter) != 0L && // keep at least one worker
868 >            UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) {
869 >            w.shutdown();
870 >            LockSupport.unpark(w);
871 >        }
872 >        releaseEventWaiters();
873 >    }
874 >
875 >    // Maintaining spares
876 >
877 >    /**
878 >     * Pushes worker onto the spare stack
879       */
880 <    private static int arraySizeFor(int poolSize) {
881 <        return (poolSize <= 1) ? 1 :
882 <            (1 << (32 - Integer.numberOfLeadingZeros(poolSize-1)));
880 >    final void pushSpare(ForkJoinWorkerThread w) {
881 >        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
882 >        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
883 >                                               w.nextSpare = spareWaiters,ns));
884 >    }
885 >
886 >    /**
887 >     * Callback from oldest spare occasionally waking up.  Tries
888 >     * (once) to shutdown a spare. Same idea as tryShutdownWaiter.
889 >     */
890 >    final void tryShutdownSpare() {
891 >        int sw, id;
892 >        ForkJoinWorkerThread w;
893 >        ForkJoinWorkerThread[] ws;
894 >        if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
895 >            id < (ws = workers).length && (w = ws[id]) != null &&
896 >            (workerCounts & RUNNING_COUNT_MASK) >= parallelism &&
897 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
898 >                                     sw, w.nextSpare)) {
899 >            w.shutdown();
900 >            LockSupport.unpark(w);
901 >            advanceEventCount();
902 >        }
903      }
904  
905      /**
906 <     * Creates or resizes array if necessary to hold newLength.
907 <     * Call only under exclusion.
906 >     * Tries (once) to resume a spare if worker counts match
907 >     * the given count.
908       *
909 <     * @return the array
909 >     * @param wc workerCounts value on invocation of this method
910       */
911 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
911 >    private void tryResumeSpare(int wc) {
912          ForkJoinWorkerThread[] ws = workers;
913 <        if (ws == null)
914 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
915 <        else if (newLength > ws.length)
916 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
917 <        else
918 <            return ws;
913 >        int n = ws.length;
914 >        int sw, id, rs;  ForkJoinWorkerThread w;
915 >        if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
916 >            id < n && (w = ws[id]) != null &&
917 >            (rs = runState) < TERMINATING &&
918 >            eventWaiters == 0L && workerCounts == wc) {
919 >            // In case all workers busy, heuristically back off to let settle
920 >            Thread.yield();
921 >            if (eventWaiters == 0L && runState == rs && // recheck
922 >                workerCounts == wc && spareWaiters == sw &&
923 >                UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
924 >                                         sw, w.nextSpare)) {
925 >                int c;              // increment running count before resume
926 >                do {} while(!UNSAFE.compareAndSwapInt
927 >                            (this, workerCountsOffset,
928 >                             c = workerCounts, c + ONE_RUNNING));
929 >                if (w.tryUnsuspend())
930 >                    LockSupport.unpark(w);
931 >                else               // back out if w was shutdown
932 >                    decrementWorkerCounts(ONE_RUNNING, 0);
933 >            }
934 >        }
935 >    }
936 >
937 >    // adding workers on demand
938 >
939 >    /**
940 >     * Adds one or more workers if needed to establish target parallelism.
941 >     * Retries upon contention.
942 >     */
943 >    private void addWorkerIfBelowTarget() {
944 >        int pc = parallelism;
945 >        int wc;
946 >        while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < pc &&
947 >               runState < TERMINATING) {
948 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
949 >                                         wc + (ONE_RUNNING|ONE_TOTAL))) {
950 >                if (addWorker() == null)
951 >                    break;
952 >            }
953 >        }
954      }
955  
956      /**
957 <     * Tries to shrink workers into smaller array after one or more terminate.
957 >     * Tries (once) to add a new worker if all existing workers are
958 >     * busy, and there are either no running workers or the deficit is
959 >     * at least twice the surplus.
960 >     *
961 >     * @param wc workerCounts value on invocation of this method
962       */
963 <    private void tryShrinkWorkerArray() {
964 <        ForkJoinWorkerThread[] ws = workers;
965 <        if (ws != null) {
966 <            int len = ws.length;
967 <            int last = len - 1;
968 <            while (last >= 0 && ws[last] == null)
969 <                --last;
970 <            int newLength = arraySizeFor(last+1);
971 <            if (newLength < len)
972 <                workers = Arrays.copyOf(ws, newLength);
963 >    private void tryAddWorkerIfBusy(int wc) {
964 >        int tc, rc, rs;
965 >        int pc = parallelism;
966 >        if ((tc = wc >>> TOTAL_COUNT_SHIFT) < MAX_WORKERS &&
967 >            ((rc = wc & RUNNING_COUNT_MASK) == 0 ||
968 >             rc < pc - ((tc - pc) << 1)) &&
969 >            (rs = runState) < TERMINATING &&
970 >            (rs & ACTIVE_COUNT_MASK) == tc) {
971 >            // Since all workers busy, heuristically back off to let settle
972 >            Thread.yield();
973 >            if (eventWaiters == 0L && spareWaiters == 0 && // recheck
974 >                runState == rs && workerCounts == wc &&
975 >                UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
976 >                                         wc + (ONE_RUNNING|ONE_TOTAL)))
977 >                addWorker();
978          }
979      }
980  
981      /**
982 <     * Initializes workers if necessary.
982 >     * Does at most one of:
983 >     *
984 >     * 1. Help wake up existing workers waiting for work via
985 >     *    releaseEventWaiters. (If any exist, then it doesn't
986 >     *    matter right now if under target parallelism level.)
987 >     *
988 >     * 2. If a spare exists, try (once) to resume it via tryResumeSpare.
989 >     *
990 >     * 3. If there are not enough total workers, add some
991 >     *    via addWorkerIfBelowTarget;
992 >     *
993 >     * 4. Try (once) to add a new worker if all existing workers
994 >     *     are busy, via tryAddWorkerIfBusy
995       */
996 <    final void ensureWorkerInitialization() {
997 <        ForkJoinWorkerThread[] ws = workers;
998 <        if (ws == null) {
999 <            final ReentrantLock lock = this.workerLock;
1000 <            lock.lock();
1001 <            try {
1002 <                ws = workers;
1003 <                if (ws == null) {
1004 <                    int ps = parallelism;
1005 <                    ws = ensureWorkerArrayCapacity(ps);
1006 <                    for (int i = 0; i < ps; ++i) {
1007 <                        ForkJoinWorkerThread w = createWorker(i);
1008 <                        if (w != null) {
1009 <                            ws[i] = w;
1010 <                            w.start();
1011 <                            updateWorkerCount(1);
1012 <                        }
996 >    private void helpMaintainParallelism() {
997 >        long h; int pc, wc;
998 >        if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0) {
999 >            if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1000 >                releaseEventWaiters(); // avoid useless call
1001 >        }
1002 >        else if ((pc = parallelism) >
1003 >                 ((wc = workerCounts) & RUNNING_COUNT_MASK)) {
1004 >            if (spareWaiters != 0)
1005 >                tryResumeSpare(wc);
1006 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
1007 >                addWorkerIfBelowTarget();
1008 >            else
1009 >                tryAddWorkerIfBusy(wc);
1010 >        }
1011 >    }
1012 >
1013 >    /**
1014 >     * Callback from workers invoked upon each top-level action (i.e.,
1015 >     * stealing a task or taking a submission and running it).
1016 >     * Performs one or more of the following:
1017 >     *
1018 >     * 1. If the worker is active, try to set its active status to
1019 >     *    inactive and update activeCount. On contention, we may try
1020 >     *    again on this or subsequent call.
1021 >     *
1022 >     * 2. Release any existing event waiters that are now relesable
1023 >     *
1024 >     * 3. If there are too many running threads, suspend this worker
1025 >     *    (first forcing inactive if necessary).  If it is not
1026 >     *    needed, it may be killed while suspended via
1027 >     *    tryShutdownSpare. Otherwise, upon resume it rechecks to make
1028 >     *    sure that it is still needed.
1029 >     *
1030 >     * 4. If more than 1 miss, await the next task event via
1031 >     *    eventSync (first forcing inactivation if necessary), upon
1032 >     *    which worker may also be killed, via tryShutdownWaiter.
1033 >     *
1034 >     * 5. Help reactivate other workers via helpMaintainParallelism
1035 >     *
1036 >     * @param w the worker
1037 >     * @param misses the number of scans by caller failing to find work
1038 >     * (saturating at 2 to avoid wraparound)
1039 >     */
1040 >    final void preStep(ForkJoinWorkerThread w, int misses) {
1041 >        boolean active = w.active;
1042 >        int pc = parallelism;
1043 >        for (;;) {
1044 >            int rs, wc, rc, ec; long h;
1045 >            if (active && UNSAFE.compareAndSwapInt(this, runStateOffset,
1046 >                                                   rs = runState, rs - 1))
1047 >                active = w.active = false;
1048 >            if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0 &&
1049 >                (int)(h >>> EVENT_COUNT_SHIFT) != eventCount) {
1050 >                releaseEventWaiters();
1051 >                if (misses > 1)
1052 >                    continue;                  // clear before sync below
1053 >            }
1054 >            if ((rc = ((wc = workerCounts) & RUNNING_COUNT_MASK)) > pc) {
1055 >                if (!active &&                 // must inactivate to suspend
1056 >                    workerCounts == wc &&      // try to suspend as spare
1057 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1058 >                                             wc, wc - ONE_RUNNING)) {
1059 >                    w.suspendAsSpare();
1060 >                    if (!w.isRunning())
1061 >                        break;                 // was killed while spare
1062 >                }
1063 >                continue;
1064 >            }
1065 >            if (misses > 0) {
1066 >                if ((ec = eventCount) == w.lastEventCount && misses > 1) {
1067 >                    if (!active) {             // must inactivate to sync
1068 >                        eventSync(w);
1069 >                        if (w.isRunning())
1070 >                            misses = 1;        // don't re-sync
1071 >                        else
1072 >                            break;             // was killed while waiting
1073                      }
1074 +                    continue;
1075                  }
1076 <            } finally {
506 <                lock.unlock();
1076 >                w.lastEventCount = ec;
1077              }
1078 +            if (rc < pc)
1079 +                helpMaintainParallelism();
1080 +            break;
1081          }
1082      }
1083  
1084      /**
1085 <     * Worker creation and startup for threads added via setParallelism.
1085 >     * Helps and/or blocks awaiting join of the given task.
1086 >     * Alternates between helpJoinTask() and helpMaintainParallelism()
1087 >     * as many times as there is a deficit in running count (or longer
1088 >     * if running count would become zero), then blocks if task still
1089 >     * not done.
1090 >     *
1091 >     * @param joinMe the task to join
1092       */
1093 <    private void createAndStartAddedWorkers() {
1094 <        resumeAllSpares();  // Allow spares to convert to nonspare
1095 <        int ps = parallelism;
1096 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
1097 <        int len = ws.length;
1098 <        // Sweep through slots, to keep lowest indices most populated
1099 <        int k = 0;
1100 <        while (k < len) {
1101 <            if (ws[k] != null) {
1102 <                ++k;
1103 <                continue;
1093 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1094 >        int threshold = parallelism;         // descend blocking thresholds
1095 >        while (joinMe.status >= 0) {
1096 >            boolean block; int wc;
1097 >            worker.helpJoinTask(joinMe);
1098 >            if (joinMe.status < 0)
1099 >                break;
1100 >            if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1101 >                if (threshold > 0)
1102 >                    --threshold;
1103 >                else
1104 >                    advanceEventCount(); // force release
1105 >                block = false;
1106              }
1107 <            int s = workerCounts;
1108 <            int tc = totalCountOf(s);
1109 <            int rc = runningCountOf(s);
1110 <            if (rc >= ps || tc >= ps)
1107 >            else
1108 >                block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1109 >                                                 wc, wc - ONE_RUNNING);
1110 >            helpMaintainParallelism();
1111 >            if (block) {
1112 >                int c;
1113 >                joinMe.internalAwaitDone();
1114 >                do {} while (!UNSAFE.compareAndSwapInt
1115 >                             (this, workerCountsOffset,
1116 >                              c = workerCounts, c + ONE_RUNNING));
1117                  break;
1118 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
1119 <                ForkJoinWorkerThread w = createWorker(k);
1120 <                if (w != null) {
1121 <                    ws[k++] = w;
1122 <                    w.start();
1118 >            }
1119 >        }
1120 >    }
1121 >
1122 >    /**
1123 >     * Same idea as awaitJoin, but no helping
1124 >     */
1125 >    final void awaitBlocker(ManagedBlocker blocker)
1126 >        throws InterruptedException {
1127 >        int threshold = parallelism;
1128 >        while (!blocker.isReleasable()) {
1129 >            boolean block; int wc;
1130 >            if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1131 >                if (threshold > 0)
1132 >                    --threshold;
1133 >                else
1134 >                    advanceEventCount();
1135 >                block = false;
1136 >            }
1137 >            else
1138 >                block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1139 >                                                 wc, wc - ONE_RUNNING);
1140 >            helpMaintainParallelism();
1141 >            if (block) {
1142 >                try {
1143 >                    do {} while (!blocker.isReleasable() && !blocker.block());
1144 >                } finally {
1145 >                    int c;
1146 >                    do {} while (!UNSAFE.compareAndSwapInt
1147 >                                 (this, workerCountsOffset,
1148 >                                  c = workerCounts, c + ONE_RUNNING));
1149                  }
1150 <                else {
1151 <                    updateWorkerCount(-1); // back out on failed creation
1152 <                    break;
1150 >                break;
1151 >            }
1152 >        }
1153 >    }
1154 >
1155 >    /**
1156 >     * Possibly initiates and/or completes termination.
1157 >     *
1158 >     * @param now if true, unconditionally terminate, else only
1159 >     * if shutdown and empty queue and no active workers
1160 >     * @return true if now terminating or terminated
1161 >     */
1162 >    private boolean tryTerminate(boolean now) {
1163 >        if (now)
1164 >            advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1165 >        else if (runState < SHUTDOWN ||
1166 >                 !submissionQueue.isEmpty() ||
1167 >                 (runState & ACTIVE_COUNT_MASK) != 0)
1168 >            return false;
1169 >
1170 >        if (advanceRunLevel(TERMINATING))
1171 >            startTerminating();
1172 >
1173 >        // Finish now if all threads terminated; else in some subsequent call
1174 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1175 >            advanceRunLevel(TERMINATED);
1176 >            termination.arrive();
1177 >        }
1178 >        return true;
1179 >    }
1180 >
1181 >    /**
1182 >     * Actions on transition to TERMINATING
1183 >     *
1184 >     * Runs up to four passes through workers: (0) shutting down each
1185 >     * (without waking up if parked) to quickly spread notifications
1186 >     * without unnecessary bouncing around event queues etc (1) wake
1187 >     * up and help cancel tasks (2) interrupt (3) mop up races with
1188 >     * interrupted workers
1189 >     */
1190 >    private void startTerminating() {
1191 >        cancelSubmissions();
1192 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1193 >            advanceEventCount();
1194 >            eventWaiters = 0L; // clobber lists
1195 >            spareWaiters = 0;
1196 >            ForkJoinWorkerThread[] ws = workers;
1197 >            int n = ws.length;
1198 >            for (int i = 0; i < n; ++i) {
1199 >                ForkJoinWorkerThread w = ws[i];
1200 >                if (w != null) {
1201 >                    w.shutdown();
1202 >                    if (passes > 0 && !w.isTerminated()) {
1203 >                        w.cancelTasks();
1204 >                        LockSupport.unpark(w);
1205 >                        if (passes > 1) {
1206 >                            try {
1207 >                                w.interrupt();
1208 >                            } catch (SecurityException ignore) {
1209 >                            }
1210 >                        }
1211 >                    }
1212                  }
1213              }
1214          }
1215      }
1216  
1217 +    /**
1218 +     * Clear out and cancel submissions, ignoring exceptions
1219 +     */
1220 +    private void cancelSubmissions() {
1221 +        ForkJoinTask<?> task;
1222 +        while ((task = submissionQueue.poll()) != null) {
1223 +            try {
1224 +                task.cancel(false);
1225 +            } catch (Throwable ignore) {
1226 +            }
1227 +        }
1228 +    }
1229 +
1230 +    // misc support for ForkJoinWorkerThread
1231 +
1232 +    /**
1233 +     * Returns pool number
1234 +     */
1235 +    final int getPoolNumber() {
1236 +        return poolNumber;
1237 +    }
1238 +
1239 +    /**
1240 +     * Tries to accumulates steal count from a worker, clearing
1241 +     * the worker's value.
1242 +     *
1243 +     * @return true if worker steal count now zero
1244 +     */
1245 +    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1246 +        int sc = w.stealCount;
1247 +        long c = stealCount;
1248 +        // CAS even if zero, for fence effects
1249 +        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1250 +            if (sc != 0)
1251 +                w.stealCount = 0;
1252 +            return true;
1253 +        }
1254 +        return sc == 0;
1255 +    }
1256 +
1257 +    /**
1258 +     * Returns the approximate (non-atomic) number of idle threads per
1259 +     * active thread.
1260 +     */
1261 +    final int idlePerActive() {
1262 +        int pc = parallelism; // use parallelism, not rc
1263 +        int ac = runState;    // no mask -- artifically boosts during shutdown
1264 +        // Use exact results for small values, saturate past 4
1265 +        return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1266 +    }
1267 +
1268 +    // Public and protected methods
1269 +
1270 +    // Constructors
1271 +
1272 +    /**
1273 +     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1274 +     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1275 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1276 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1277 +     *
1278 +     * @throws SecurityException if a security manager exists and
1279 +     *         the caller is not permitted to modify threads
1280 +     *         because it does not hold {@link
1281 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1282 +     */
1283 +    public ForkJoinPool() {
1284 +        this(Runtime.getRuntime().availableProcessors(),
1285 +             defaultForkJoinWorkerThreadFactory, null, false);
1286 +    }
1287 +
1288 +    /**
1289 +     * Creates a {@code ForkJoinPool} with the indicated parallelism
1290 +     * level, the {@linkplain
1291 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1292 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1293 +     *
1294 +     * @param parallelism the parallelism level
1295 +     * @throws IllegalArgumentException if parallelism less than or
1296 +     *         equal to zero, or greater than implementation limit
1297 +     * @throws SecurityException if a security manager exists and
1298 +     *         the caller is not permitted to modify threads
1299 +     *         because it does not hold {@link
1300 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1301 +     */
1302 +    public ForkJoinPool(int parallelism) {
1303 +        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1304 +    }
1305 +
1306 +    /**
1307 +     * Creates a {@code ForkJoinPool} with the given parameters.
1308 +     *
1309 +     * @param parallelism the parallelism level. For default value,
1310 +     * use {@link java.lang.Runtime#availableProcessors}.
1311 +     * @param factory the factory for creating new threads. For default value,
1312 +     * use {@link #defaultForkJoinWorkerThreadFactory}.
1313 +     * @param handler the handler for internal worker threads that
1314 +     * terminate due to unrecoverable errors encountered while executing
1315 +     * tasks. For default value, use <code>null</code>.
1316 +     * @param asyncMode if true,
1317 +     * establishes local first-in-first-out scheduling mode for forked
1318 +     * tasks that are never joined. This mode may be more appropriate
1319 +     * than default locally stack-based mode in applications in which
1320 +     * worker threads only process event-style asynchronous tasks.
1321 +     * For default value, use <code>false</code>.
1322 +     * @throws IllegalArgumentException if parallelism less than or
1323 +     *         equal to zero, or greater than implementation limit
1324 +     * @throws NullPointerException if the factory is null
1325 +     * @throws SecurityException if a security manager exists and
1326 +     *         the caller is not permitted to modify threads
1327 +     *         because it does not hold {@link
1328 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1329 +     */
1330 +    public ForkJoinPool(int parallelism,
1331 +                        ForkJoinWorkerThreadFactory factory,
1332 +                        Thread.UncaughtExceptionHandler handler,
1333 +                        boolean asyncMode) {
1334 +        checkPermission();
1335 +        if (factory == null)
1336 +            throw new NullPointerException();
1337 +        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1338 +            throw new IllegalArgumentException();
1339 +        this.parallelism = parallelism;
1340 +        this.factory = factory;
1341 +        this.ueh = handler;
1342 +        this.locallyFifo = asyncMode;
1343 +        int arraySize = initialArraySizeFor(parallelism);
1344 +        this.workers = new ForkJoinWorkerThread[arraySize];
1345 +        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1346 +        this.workerLock = new ReentrantLock();
1347 +        this.termination = new Phaser(1);
1348 +        this.poolNumber = poolNumberGenerator.incrementAndGet();
1349 +    }
1350 +
1351 +    /**
1352 +     * Returns initial power of two size for workers array.
1353 +     * @param pc the initial parallelism level
1354 +     */
1355 +    private static int initialArraySizeFor(int pc) {
1356 +        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1357 +        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1358 +        size |= size >>> 1;
1359 +        size |= size >>> 2;
1360 +        size |= size >>> 4;
1361 +        size |= size >>> 8;
1362 +        return size + 1;
1363 +    }
1364 +
1365      // Execution methods
1366  
1367      /**
# Line 550 | Line 1370 | public class ForkJoinPool extends Abstra
1370      private <T> void doSubmit(ForkJoinTask<T> task) {
1371          if (task == null)
1372              throw new NullPointerException();
1373 <        if (isShutdown())
1373 >        if (runState >= SHUTDOWN)
1374              throw new RejectedExecutionException();
555        if (workers == null)
556            ensureWorkerInitialization();
1375          submissionQueue.offer(task);
1376 <        signalIdleWorkers();
1376 >        advanceEventCount();
1377 >        if (eventWaiters != 0L)
1378 >            releaseEventWaiters();
1379 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
1380 >            addWorkerIfBelowTarget();
1381      }
1382  
1383      /**
# Line 563 | Line 1385 | public class ForkJoinPool extends Abstra
1385       *
1386       * @param task the task
1387       * @return the task's result
1388 <     * @throws NullPointerException if task is null
1389 <     * @throws RejectedExecutionException if pool is shut down
1388 >     * @throws NullPointerException if the task is null
1389 >     * @throws RejectedExecutionException if the task cannot be
1390 >     *         scheduled for execution
1391       */
1392      public <T> T invoke(ForkJoinTask<T> task) {
1393          doSubmit(task);
# Line 575 | Line 1398 | public class ForkJoinPool extends Abstra
1398       * Arranges for (asynchronous) execution of the given task.
1399       *
1400       * @param task the task
1401 <     * @throws NullPointerException if task is null
1402 <     * @throws RejectedExecutionException if pool is shut down
1401 >     * @throws NullPointerException if the task is null
1402 >     * @throws RejectedExecutionException if the task cannot be
1403 >     *         scheduled for execution
1404       */
1405 <    public <T> void execute(ForkJoinTask<T> task) {
1405 >    public void execute(ForkJoinTask<?> task) {
1406          doSubmit(task);
1407      }
1408  
1409      // AbstractExecutorService methods
1410  
1411 +    /**
1412 +     * @throws NullPointerException if the task is null
1413 +     * @throws RejectedExecutionException if the task cannot be
1414 +     *         scheduled for execution
1415 +     */
1416      public void execute(Runnable task) {
1417          ForkJoinTask<?> job;
1418 <        if (task instanceof ForkJoinTask) // avoid re-wrap
1419 <            job = (ForkJoinTask<?>)task;
1418 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1419 >            job = (ForkJoinTask<?>) task;
1420          else
1421 <            job = new AdaptedRunnable<Void>(task, null);
1421 >            job = ForkJoinTask.adapt(task, null);
1422          doSubmit(job);
1423      }
1424  
596    public <T> ForkJoinTask<T> submit(Callable<T> task) {
597        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
598        doSubmit(job);
599        return job;
600    }
601
602    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
603        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
604        doSubmit(job);
605        return job;
606    }
607
608    public ForkJoinTask<?> submit(Runnable task) {
609        ForkJoinTask<?> job;
610        if (task instanceof ForkJoinTask) // avoid re-wrap
611            job = (ForkJoinTask<?>)task;
612        else
613            job = new AdaptedRunnable<Void>(task, null);
614        doSubmit(job);
615        return job;
616    }
617
1425      /**
1426       * Submits a ForkJoinTask for execution.
1427       *
1428       * @param task the task to submit
1429       * @return the task
1430 +     * @throws NullPointerException if the task is null
1431       * @throws RejectedExecutionException if the task cannot be
1432       *         scheduled for execution
625     * @throws NullPointerException if the task is null
1433       */
1434      public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1435          doSubmit(task);
# Line 630 | Line 1437 | public class ForkJoinPool extends Abstra
1437      }
1438  
1439      /**
1440 <     * Adaptor for Runnables. This implements RunnableFuture
1441 <     * to be compliant with AbstractExecutorService constraints.
1440 >     * @throws NullPointerException if the task is null
1441 >     * @throws RejectedExecutionException if the task cannot be
1442 >     *         scheduled for execution
1443       */
1444 <    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
1445 <        implements RunnableFuture<T> {
1446 <        final Runnable runnable;
1447 <        final T resultOnCompletion;
640 <        T result;
641 <        AdaptedRunnable(Runnable runnable, T result) {
642 <            if (runnable == null) throw new NullPointerException();
643 <            this.runnable = runnable;
644 <            this.resultOnCompletion = result;
645 <        }
646 <        public T getRawResult() { return result; }
647 <        public void setRawResult(T v) { result = v; }
648 <        public boolean exec() {
649 <            runnable.run();
650 <            result = resultOnCompletion;
651 <            return true;
652 <        }
653 <        public void run() { invoke(); }
654 <        private static final long serialVersionUID = 5232453952276885070L;
1444 >    public <T> ForkJoinTask<T> submit(Callable<T> task) {
1445 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1446 >        doSubmit(job);
1447 >        return job;
1448      }
1449  
1450      /**
1451 <     * Adaptor for Callables
1451 >     * @throws NullPointerException if the task is null
1452 >     * @throws RejectedExecutionException if the task cannot be
1453 >     *         scheduled for execution
1454       */
1455 <    static final class AdaptedCallable<T> extends ForkJoinTask<T>
1456 <        implements RunnableFuture<T> {
1457 <        final Callable<T> callable;
1458 <        T result;
1459 <        AdaptedCallable(Callable<T> callable) {
1460 <            if (callable == null) throw new NullPointerException();
1461 <            this.callable = callable;
1462 <        }
1463 <        public T getRawResult() { return result; }
1464 <        public void setRawResult(T v) { result = v; }
1465 <        public boolean exec() {
1466 <            try {
1467 <                result = callable.call();
1468 <                return true;
1469 <            } catch (Error err) {
1470 <                throw err;
1471 <            } catch (RuntimeException rex) {
1472 <                throw rex;
1473 <            } catch (Exception ex) {
679 <                throw new RuntimeException(ex);
680 <            }
681 <        }
682 <        public void run() { invoke(); }
683 <        private static final long serialVersionUID = 2838392045355241008L;
1455 >    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1456 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1457 >        doSubmit(job);
1458 >        return job;
1459 >    }
1460 >
1461 >    /**
1462 >     * @throws NullPointerException if the task is null
1463 >     * @throws RejectedExecutionException if the task cannot be
1464 >     *         scheduled for execution
1465 >     */
1466 >    public ForkJoinTask<?> submit(Runnable task) {
1467 >        ForkJoinTask<?> job;
1468 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1469 >            job = (ForkJoinTask<?>) task;
1470 >        else
1471 >            job = ForkJoinTask.adapt(task, null);
1472 >        doSubmit(job);
1473 >        return job;
1474      }
1475  
1476 +    /**
1477 +     * @throws NullPointerException       {@inheritDoc}
1478 +     * @throws RejectedExecutionException {@inheritDoc}
1479 +     */
1480      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1481          ArrayList<ForkJoinTask<T>> forkJoinTasks =
1482              new ArrayList<ForkJoinTask<T>>(tasks.size());
1483          for (Callable<T> task : tasks)
1484 <            forkJoinTasks.add(new AdaptedCallable<T>(task));
1484 >            forkJoinTasks.add(ForkJoinTask.adapt(task));
1485          invoke(new InvokeAll<T>(forkJoinTasks));
1486  
1487          @SuppressWarnings({"unchecked", "rawtypes"})
1488 <        List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1488 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1489          return futures;
1490      }
1491  
# Line 705 | Line 1499 | public class ForkJoinPool extends Abstra
1499          private static final long serialVersionUID = -7914297376763021607L;
1500      }
1501  
708    // Configuration and status settings and queries
709
1502      /**
1503       * Returns the factory used for constructing new workers.
1504       *
# Line 720 | Line 1512 | public class ForkJoinPool extends Abstra
1512       * Returns the handler for internal worker threads that terminate
1513       * due to unrecoverable errors encountered while executing tasks.
1514       *
1515 <     * @return the handler, or null if none
1515 >     * @return the handler, or {@code null} if none
1516       */
1517      public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1518 <        Thread.UncaughtExceptionHandler h;
727 <        final ReentrantLock lock = this.workerLock;
728 <        lock.lock();
729 <        try {
730 <            h = ueh;
731 <        } finally {
732 <            lock.unlock();
733 <        }
734 <        return h;
735 <    }
736 <
737 <    /**
738 <     * Sets the handler for internal worker threads that terminate due
739 <     * to unrecoverable errors encountered while executing tasks.
740 <     * Unless set, the current default or ThreadGroup handler is used
741 <     * as handler.
742 <     *
743 <     * @param h the new handler
744 <     * @return the old handler, or null if none
745 <     * @throws SecurityException if a security manager exists and
746 <     *         the caller is not permitted to modify threads
747 <     *         because it does not hold {@link
748 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
749 <     */
750 <    public Thread.UncaughtExceptionHandler
751 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
752 <        checkPermission();
753 <        Thread.UncaughtExceptionHandler old = null;
754 <        final ReentrantLock lock = this.workerLock;
755 <        lock.lock();
756 <        try {
757 <            old = ueh;
758 <            ueh = h;
759 <            ForkJoinWorkerThread[] ws = workers;
760 <            if (ws != null) {
761 <                for (int i = 0; i < ws.length; ++i) {
762 <                    ForkJoinWorkerThread w = ws[i];
763 <                    if (w != null)
764 <                        w.setUncaughtExceptionHandler(h);
765 <                }
766 <            }
767 <        } finally {
768 <            lock.unlock();
769 <        }
770 <        return old;
1518 >        return ueh;
1519      }
1520  
773
1521      /**
1522 <     * Sets the target parallelism level of this pool.
1522 >     * Returns the targeted parallelism level of this pool.
1523       *
1524 <     * @param parallelism the target parallelism
778 <     * @throws IllegalArgumentException if parallelism less than or
779 <     * equal to zero or greater than maximum size bounds
780 <     * @throws SecurityException if a security manager exists and
781 <     *         the caller is not permitted to modify threads
782 <     *         because it does not hold {@link
783 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
784 <     */
785 <    public void setParallelism(int parallelism) {
786 <        checkPermission();
787 <        if (parallelism <= 0 || parallelism > maxPoolSize)
788 <            throw new IllegalArgumentException();
789 <        final ReentrantLock lock = this.workerLock;
790 <        lock.lock();
791 <        try {
792 <            if (!isTerminating()) {
793 <                int p = this.parallelism;
794 <                this.parallelism = parallelism;
795 <                if (parallelism > p)
796 <                    createAndStartAddedWorkers();
797 <                else
798 <                    trimSpares();
799 <            }
800 <        } finally {
801 <            lock.unlock();
802 <        }
803 <        signalIdleWorkers();
804 <    }
805 <
806 <    /**
807 <     * Returns the targeted number of worker threads in this pool.
808 <     *
809 <     * @return the targeted number of worker threads in this pool
1524 >     * @return the targeted parallelism level of this pool
1525       */
1526      public int getParallelism() {
1527          return parallelism;
# Line 815 | Line 1530 | public class ForkJoinPool extends Abstra
1530      /**
1531       * Returns the number of worker threads that have started but not
1532       * yet terminated.  This result returned by this method may differ
1533 <     * from {@code getParallelism} when threads are created to
1533 >     * from {@link #getParallelism} when threads are created to
1534       * maintain parallelism when others are cooperatively blocked.
1535       *
1536       * @return the number of worker threads
1537       */
1538      public int getPoolSize() {
1539 <        return totalCountOf(workerCounts);
1539 >        return workerCounts >>> TOTAL_COUNT_SHIFT;
1540      }
1541  
1542      /**
1543 <     * Returns the maximum number of threads allowed to exist in the
829 <     * pool, even if there are insufficient unblocked running threads.
830 <     *
831 <     * @return the maximum
832 <     */
833 <    public int getMaximumPoolSize() {
834 <        return maxPoolSize;
835 <    }
836 <
837 <    /**
838 <     * Sets the maximum number of threads allowed to exist in the
839 <     * pool, even if there are insufficient unblocked running threads.
840 <     * Setting this value has no effect on current pool size. It
841 <     * controls construction of new threads.
842 <     *
843 <     * @throws IllegalArgumentException if negative or greater then
844 <     * internal implementation limit
845 <     */
846 <    public void setMaximumPoolSize(int newMax) {
847 <        if (newMax < 0 || newMax > MAX_THREADS)
848 <            throw new IllegalArgumentException();
849 <        maxPoolSize = newMax;
850 <    }
851 <
852 <
853 <    /**
854 <     * Returns true if this pool dynamically maintains its target
855 <     * parallelism level. If false, new threads are added only to
856 <     * avoid possible starvation.
857 <     * This setting is by default true.
858 <     *
859 <     * @return true if maintains parallelism
860 <     */
861 <    public boolean getMaintainsParallelism() {
862 <        return maintainsParallelism;
863 <    }
864 <
865 <    /**
866 <     * Sets whether this pool dynamically maintains its target
867 <     * parallelism level. If false, new threads are added only to
868 <     * avoid possible starvation.
869 <     *
870 <     * @param enable true to maintains parallelism
871 <     */
872 <    public void setMaintainsParallelism(boolean enable) {
873 <        maintainsParallelism = enable;
874 <    }
875 <
876 <    /**
877 <     * Establishes local first-in-first-out scheduling mode for forked
878 <     * tasks that are never joined. This mode may be more appropriate
879 <     * than default locally stack-based mode in applications in which
880 <     * worker threads only process asynchronous tasks.  This method is
881 <     * designed to be invoked only when pool is quiescent, and
882 <     * typically only before any tasks are submitted. The effects of
883 <     * invocations at other times may be unpredictable.
884 <     *
885 <     * @param async if true, use locally FIFO scheduling
886 <     * @return the previous mode
887 <     */
888 <    public boolean setAsyncMode(boolean async) {
889 <        boolean oldMode = locallyFifo;
890 <        locallyFifo = async;
891 <        ForkJoinWorkerThread[] ws = workers;
892 <        if (ws != null) {
893 <            for (int i = 0; i < ws.length; ++i) {
894 <                ForkJoinWorkerThread t = ws[i];
895 <                if (t != null)
896 <                    t.setAsyncMode(async);
897 <            }
898 <        }
899 <        return oldMode;
900 <    }
901 <
902 <    /**
903 <     * Returns true if this pool uses local first-in-first-out
1543 >     * Returns {@code true} if this pool uses local first-in-first-out
1544       * scheduling mode for forked tasks that are never joined.
1545       *
1546 <     * @return true if this pool uses async mode
1546 >     * @return {@code true} if this pool uses async mode
1547       */
1548      public boolean getAsyncMode() {
1549          return locallyFifo;
# Line 912 | Line 1552 | public class ForkJoinPool extends Abstra
1552      /**
1553       * Returns an estimate of the number of worker threads that are
1554       * not blocked waiting to join tasks or for other managed
1555 <     * synchronization.
1555 >     * synchronization. This method may overestimate the
1556 >     * number of running threads.
1557       *
1558       * @return the number of worker threads
1559       */
1560      public int getRunningThreadCount() {
1561 <        return runningCountOf(workerCounts);
1561 >        return workerCounts & RUNNING_COUNT_MASK;
1562      }
1563  
1564      /**
# Line 928 | Line 1569 | public class ForkJoinPool extends Abstra
1569       * @return the number of active threads
1570       */
1571      public int getActiveThreadCount() {
1572 <        return activeCountOf(runControl);
1572 >        return runState & ACTIVE_COUNT_MASK;
1573      }
1574  
1575      /**
1576 <     * Returns an estimate of the number of threads that are currently
1577 <     * idle waiting for tasks. This method may underestimate the
1578 <     * number of idle threads.
1576 >     * Returns {@code true} if all worker threads are currently idle.
1577 >     * An idle worker is one that cannot obtain a task to execute
1578 >     * because none are available to steal from other threads, and
1579 >     * there are no pending submissions to the pool. This method is
1580 >     * conservative; it might not return {@code true} immediately upon
1581 >     * idleness of all threads, but will eventually become true if
1582 >     * threads remain inactive.
1583       *
1584 <     * @return the number of idle threads
940 <     */
941 <    final int getIdleThreadCount() {
942 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
943 <        return (c <= 0) ? 0 : c;
944 <    }
945 <
946 <    /**
947 <     * Returns true if all worker threads are currently idle. An idle
948 <     * worker is one that cannot obtain a task to execute because none
949 <     * are available to steal from other threads, and there are no
950 <     * pending submissions to the pool. This method is conservative;
951 <     * it might not return true immediately upon idleness of all
952 <     * threads, but will eventually become true if threads remain
953 <     * inactive.
954 <     *
955 <     * @return true if all threads are currently idle
1584 >     * @return {@code true} if all threads are currently idle
1585       */
1586      public boolean isQuiescent() {
1587 <        return activeCountOf(runControl) == 0;
1587 >        return (runState & ACTIVE_COUNT_MASK) == 0;
1588      }
1589  
1590      /**
# Line 970 | Line 1599 | public class ForkJoinPool extends Abstra
1599       * @return the number of steals
1600       */
1601      public long getStealCount() {
1602 <        return stealCount.get();
974 <    }
975 <
976 <    /**
977 <     * Accumulates steal count from a worker.
978 <     * Call only when worker known to be idle.
979 <     */
980 <    private void updateStealCount(ForkJoinWorkerThread w) {
981 <        int sc = w.getAndClearStealCount();
982 <        if (sc != 0)
983 <            stealCount.addAndGet(sc);
1602 >        return stealCount;
1603      }
1604  
1605      /**
# Line 996 | Line 1615 | public class ForkJoinPool extends Abstra
1615      public long getQueuedTaskCount() {
1616          long count = 0;
1617          ForkJoinWorkerThread[] ws = workers;
1618 <        if (ws != null) {
1619 <            for (int i = 0; i < ws.length; ++i) {
1620 <                ForkJoinWorkerThread t = ws[i];
1621 <                if (t != null)
1622 <                    count += t.getQueueSize();
1004 <            }
1618 >        int n = ws.length;
1619 >        for (int i = 0; i < n; ++i) {
1620 >            ForkJoinWorkerThread w = ws[i];
1621 >            if (w != null)
1622 >                count += w.getQueueSize();
1623          }
1624          return count;
1625      }
1626  
1627      /**
1628 <     * Returns an estimate of the number tasks submitted to this pool
1629 <     * that have not yet begun executing. This method takes time
1628 >     * Returns an estimate of the number of tasks submitted to this
1629 >     * pool that have not yet begun executing.  This method takes time
1630       * proportional to the number of submissions.
1631       *
1632       * @return the number of queued submissions
# Line 1018 | Line 1636 | public class ForkJoinPool extends Abstra
1636      }
1637  
1638      /**
1639 <     * Returns true if there are any tasks submitted to this pool
1640 <     * that have not yet begun executing.
1639 >     * Returns {@code true} if there are any tasks submitted to this
1640 >     * pool that have not yet begun executing.
1641       *
1642       * @return {@code true} if there are any queued submissions
1643       */
# Line 1032 | Line 1650 | public class ForkJoinPool extends Abstra
1650       * available.  This method may be useful in extensions to this
1651       * class that re-assign work in systems with multiple pools.
1652       *
1653 <     * @return the next submission, or null if none
1653 >     * @return the next submission, or {@code null} if none
1654       */
1655      protected ForkJoinTask<?> pollSubmission() {
1656          return submissionQueue.poll();
# Line 1042 | Line 1660 | public class ForkJoinPool extends Abstra
1660       * Removes all available unexecuted submitted and forked tasks
1661       * from scheduling queues and adds them to the given collection,
1662       * without altering their execution status. These may include
1663 <     * artificially generated or wrapped tasks. This method is designed
1664 <     * to be invoked only when the pool is known to be
1663 >     * artificially generated or wrapped tasks. This method is
1664 >     * designed to be invoked only when the pool is known to be
1665       * quiescent. Invocations at other times may not remove all
1666       * tasks. A failure encountered while attempting to add elements
1667       * to collection {@code c} may result in elements being in
# Line 1055 | Line 1673 | public class ForkJoinPool extends Abstra
1673       * @param c the collection to transfer elements into
1674       * @return the number of elements transferred
1675       */
1676 <    protected int drainTasksTo(Collection<ForkJoinTask<?>> c) {
1677 <        int n = submissionQueue.drainTo(c);
1676 >    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1677 >        int count = submissionQueue.drainTo(c);
1678          ForkJoinWorkerThread[] ws = workers;
1679 <        if (ws != null) {
1680 <            for (int i = 0; i < ws.length; ++i) {
1681 <                ForkJoinWorkerThread w = ws[i];
1682 <                if (w != null)
1683 <                    n += w.drainTasksTo(c);
1066 <            }
1679 >        int n = ws.length;
1680 >        for (int i = 0; i < n; ++i) {
1681 >            ForkJoinWorkerThread w = ws[i];
1682 >            if (w != null)
1683 >                count += w.drainTasksTo(c);
1684          }
1685 <        return n;
1685 >        return count;
1686      }
1687  
1688      /**
# Line 1076 | Line 1693 | public class ForkJoinPool extends Abstra
1693       * @return a string identifying this pool, as well as its state
1694       */
1695      public String toString() {
1079        int ps = parallelism;
1080        int wc = workerCounts;
1081        int rc = runControl;
1696          long st = getStealCount();
1697          long qt = getQueuedTaskCount();
1698          long qs = getQueuedSubmissionCount();
1699 +        int wc = workerCounts;
1700 +        int tc = wc >>> TOTAL_COUNT_SHIFT;
1701 +        int rc = wc & RUNNING_COUNT_MASK;
1702 +        int pc = parallelism;
1703 +        int rs = runState;
1704 +        int ac = rs & ACTIVE_COUNT_MASK;
1705          return super.toString() +
1706 <            "[" + runStateToString(runStateOf(rc)) +
1707 <            ", parallelism = " + ps +
1708 <            ", size = " + totalCountOf(wc) +
1709 <            ", active = " + activeCountOf(rc) +
1710 <            ", running = " + runningCountOf(wc) +
1706 >            "[" + runLevelToString(rs) +
1707 >            ", parallelism = " + pc +
1708 >            ", size = " + tc +
1709 >            ", active = " + ac +
1710 >            ", running = " + rc +
1711              ", steals = " + st +
1712              ", tasks = " + qt +
1713              ", submissions = " + qs +
1714              "]";
1715      }
1716  
1717 <    private static String runStateToString(int rs) {
1718 <        switch(rs) {
1719 <        case RUNNING: return "Running";
1720 <        case SHUTDOWN: return "Shutting down";
1721 <        case TERMINATING: return "Terminating";
1102 <        case TERMINATED: return "Terminated";
1103 <        default: throw new Error("Unknown run state");
1104 <        }
1717 >    private static String runLevelToString(int s) {
1718 >        return ((s & TERMINATED) != 0 ? "Terminated" :
1719 >                ((s & TERMINATING) != 0 ? "Terminating" :
1720 >                 ((s & SHUTDOWN) != 0 ? "Shutting down" :
1721 >                  "Running")));
1722      }
1723  
1107    // lifecycle control
1108
1724      /**
1725       * Initiates an orderly shutdown in which previously submitted
1726       * tasks are executed, but no new tasks will be accepted.
# Line 1120 | Line 1735 | public class ForkJoinPool extends Abstra
1735       */
1736      public void shutdown() {
1737          checkPermission();
1738 <        transitionRunStateTo(SHUTDOWN);
1739 <        if (canTerminateOnShutdown(runControl))
1125 <            terminateOnShutdown();
1738 >        advanceRunLevel(SHUTDOWN);
1739 >        tryTerminate(false);
1740      }
1741  
1742      /**
1743 <     * Attempts to stop all actively executing tasks, and cancels all
1744 <     * waiting tasks.  Tasks that are in the process of being
1745 <     * submitted or executed concurrently during the course of this
1746 <     * method may or may not be rejected. Unlike some other executors,
1747 <     * this method cancels rather than collects non-executed tasks
1748 <     * upon termination, so always returns an empty list. However, you
1749 <     * can use method {@code drainTasksTo} before invoking this
1750 <     * method to transfer unexecuted tasks to another collection.
1743 >     * Attempts to cancel and/or stop all tasks, and reject all
1744 >     * subsequently submitted tasks.  Tasks that are in the process of
1745 >     * being submitted or executed concurrently during the course of
1746 >     * this method may or may not be rejected. This method cancels
1747 >     * both existing and unexecuted tasks, in order to permit
1748 >     * termination in the presence of task dependencies. So the method
1749 >     * always returns an empty list (unlike the case for some other
1750 >     * Executors).
1751       *
1752       * @return an empty list
1753       * @throws SecurityException if a security manager exists and
# Line 1143 | Line 1757 | public class ForkJoinPool extends Abstra
1757       */
1758      public List<Runnable> shutdownNow() {
1759          checkPermission();
1760 <        terminate();
1760 >        tryTerminate(true);
1761          return Collections.emptyList();
1762      }
1763  
# Line 1153 | Line 1767 | public class ForkJoinPool extends Abstra
1767       * @return {@code true} if all tasks have completed following shut down
1768       */
1769      public boolean isTerminated() {
1770 <        return runStateOf(runControl) == TERMINATED;
1770 >        return runState >= TERMINATED;
1771      }
1772  
1773      /**
1774       * Returns {@code true} if the process of termination has
1775 <     * commenced but possibly not yet completed.
1775 >     * commenced but not yet completed.  This method may be useful for
1776 >     * debugging. A return of {@code true} reported a sufficient
1777 >     * period after shutdown may indicate that submitted tasks have
1778 >     * ignored or suppressed interruption, causing this executor not
1779 >     * to properly terminate.
1780       *
1781 <     * @return {@code true} if terminating
1781 >     * @return {@code true} if terminating but not yet terminated
1782       */
1783      public boolean isTerminating() {
1784 <        return runStateOf(runControl) >= TERMINATING;
1784 >        return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1785      }
1786  
1787      /**
# Line 1172 | Line 1790 | public class ForkJoinPool extends Abstra
1790       * @return {@code true} if this pool has been shut down
1791       */
1792      public boolean isShutdown() {
1793 <        return runStateOf(runControl) >= SHUTDOWN;
1793 >        return runState >= SHUTDOWN;
1794      }
1795  
1796      /**
# Line 1188 | Line 1806 | public class ForkJoinPool extends Abstra
1806       */
1807      public boolean awaitTermination(long timeout, TimeUnit unit)
1808          throws InterruptedException {
1191        long nanos = unit.toNanos(timeout);
1192        final ReentrantLock lock = this.workerLock;
1193        lock.lock();
1194        try {
1195            for (;;) {
1196                if (isTerminated())
1197                    return true;
1198                if (nanos <= 0)
1199                    return false;
1200                nanos = termination.awaitNanos(nanos);
1201            }
1202        } finally {
1203            lock.unlock();
1204        }
1205    }
1206
1207    // Shutdown and termination support
1208
1209    /**
1210     * Callback from terminating worker. Nulls out the corresponding
1211     * workers slot, and if terminating, tries to terminate; else
1212     * tries to shrink workers array.
1213     *
1214     * @param w the worker
1215     */
1216    final void workerTerminated(ForkJoinWorkerThread w) {
1217        updateStealCount(w);
1218        updateWorkerCount(-1);
1219        final ReentrantLock lock = this.workerLock;
1220        lock.lock();
1221        try {
1222            ForkJoinWorkerThread[] ws = workers;
1223            if (ws != null) {
1224                int idx = w.poolIndex;
1225                if (idx >= 0 && idx < ws.length && ws[idx] == w)
1226                    ws[idx] = null;
1227                if (totalCountOf(workerCounts) == 0) {
1228                    terminate(); // no-op if already terminating
1229                    transitionRunStateTo(TERMINATED);
1230                    termination.signalAll();
1231                }
1232                else if (!isTerminating()) {
1233                    tryShrinkWorkerArray();
1234                    tryResumeSpare(true); // allow replacement
1235                }
1236            }
1237        } finally {
1238            lock.unlock();
1239        }
1240        signalIdleWorkers();
1241    }
1242
1243    /**
1244     * Initiates termination.
1245     */
1246    private void terminate() {
1247        if (transitionRunStateTo(TERMINATING)) {
1248            stopAllWorkers();
1249            resumeAllSpares();
1250            signalIdleWorkers();
1251            cancelQueuedSubmissions();
1252            cancelQueuedWorkerTasks();
1253            interruptUnterminatedWorkers();
1254            signalIdleWorkers(); // resignal after interrupt
1255        }
1256    }
1257
1258    /**
1259     * Possibly terminates when on shutdown state.
1260     */
1261    private void terminateOnShutdown() {
1262        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1263            terminate();
1264    }
1265
1266    /**
1267     * Clears out and cancels submissions.
1268     */
1269    private void cancelQueuedSubmissions() {
1270        ForkJoinTask<?> task;
1271        while ((task = pollSubmission()) != null)
1272            task.cancel(false);
1273    }
1274
1275    /**
1276     * Cleans out worker queues.
1277     */
1278    private void cancelQueuedWorkerTasks() {
1279        final ReentrantLock lock = this.workerLock;
1280        lock.lock();
1809          try {
1810 <            ForkJoinWorkerThread[] ws = workers;
1811 <            if (ws != null) {
1284 <                for (int i = 0; i < ws.length; ++i) {
1285 <                    ForkJoinWorkerThread t = ws[i];
1286 <                    if (t != null)
1287 <                        t.cancelTasks();
1288 <                }
1289 <            }
1290 <        } finally {
1291 <            lock.unlock();
1292 <        }
1293 <    }
1294 <
1295 <    /**
1296 <     * Sets each worker's status to terminating. Requires lock to avoid
1297 <     * conflicts with add/remove.
1298 <     */
1299 <    private void stopAllWorkers() {
1300 <        final ReentrantLock lock = this.workerLock;
1301 <        lock.lock();
1302 <        try {
1303 <            ForkJoinWorkerThread[] ws = workers;
1304 <            if (ws != null) {
1305 <                for (int i = 0; i < ws.length; ++i) {
1306 <                    ForkJoinWorkerThread t = ws[i];
1307 <                    if (t != null)
1308 <                        t.shutdownNow();
1309 <                }
1310 <            }
1311 <        } finally {
1312 <            lock.unlock();
1313 <        }
1314 <    }
1315 <
1316 <    /**
1317 <     * Interrupts all unterminated workers.  This is not required for
1318 <     * sake of internal control, but may help unstick user code during
1319 <     * shutdown.
1320 <     */
1321 <    private void interruptUnterminatedWorkers() {
1322 <        final ReentrantLock lock = this.workerLock;
1323 <        lock.lock();
1324 <        try {
1325 <            ForkJoinWorkerThread[] ws = workers;
1326 <            if (ws != null) {
1327 <                for (int i = 0; i < ws.length; ++i) {
1328 <                    ForkJoinWorkerThread t = ws[i];
1329 <                    if (t != null && !t.isTerminated()) {
1330 <                        try {
1331 <                            t.interrupt();
1332 <                        } catch (SecurityException ignore) {
1333 <                        }
1334 <                    }
1335 <                }
1336 <            }
1337 <        } finally {
1338 <            lock.unlock();
1339 <        }
1340 <    }
1341 <
1342 <
1343 <    /*
1344 <     * Nodes for event barrier to manage idle threads.  Queue nodes
1345 <     * are basic Treiber stack nodes, also used for spare stack.
1346 <     *
1347 <     * The event barrier has an event count and a wait queue (actually
1348 <     * a Treiber stack).  Workers are enabled to look for work when
1349 <     * the eventCount is incremented. If they fail to find work, they
1350 <     * may wait for next count. Upon release, threads help others wake
1351 <     * up.
1352 <     *
1353 <     * Synchronization events occur only in enough contexts to
1354 <     * maintain overall liveness:
1355 <     *
1356 <     *   - Submission of a new task to the pool
1357 <     *   - Resizes or other changes to the workers array
1358 <     *   - pool termination
1359 <     *   - A worker pushing a task on an empty queue
1360 <     *
1361 <     * The case of pushing a task occurs often enough, and is heavy
1362 <     * enough compared to simple stack pushes, to require special
1363 <     * handling: Method signalWork returns without advancing count if
1364 <     * the queue appears to be empty.  This would ordinarily result in
1365 <     * races causing some queued waiters not to be woken up. To avoid
1366 <     * this, the first worker enqueued in method sync (see
1367 <     * syncIsReleasable) rescans for tasks after being enqueued, and
1368 <     * helps signal if any are found. This works well because the
1369 <     * worker has nothing better to do, and so might as well help
1370 <     * alleviate the overhead and contention on the threads actually
1371 <     * doing work.  Also, since event counts increments on task
1372 <     * availability exist to maintain liveness (rather than to force
1373 <     * refreshes etc), it is OK for callers to exit early if
1374 <     * contending with another signaller.
1375 <     */
1376 <    static final class WaitQueueNode {
1377 <        WaitQueueNode next; // only written before enqueued
1378 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1379 <        final long count; // unused for spare stack
1380 <
1381 <        WaitQueueNode(long c, ForkJoinWorkerThread w) {
1382 <            count = c;
1383 <            thread = w;
1384 <        }
1385 <
1386 <        /**
1387 <         * Wakes up waiter, returning false if known to already
1388 <         */
1389 <        boolean signal() {
1390 <            ForkJoinWorkerThread t = thread;
1391 <            if (t == null)
1392 <                return false;
1393 <            thread = null;
1394 <            LockSupport.unpark(t);
1395 <            return true;
1396 <        }
1397 <
1398 <        /**
1399 <         * Awaits release on sync.
1400 <         */
1401 <        void awaitSyncRelease(ForkJoinPool p) {
1402 <            while (thread != null && !p.syncIsReleasable(this))
1403 <                LockSupport.park(this);
1404 <        }
1405 <
1406 <        /**
1407 <         * Awaits resumption as spare.
1408 <         */
1409 <        void awaitSpareRelease() {
1410 <            while (thread != null) {
1411 <                if (!Thread.interrupted())
1412 <                    LockSupport.park(this);
1413 <            }
1414 <        }
1415 <    }
1416 <
1417 <    /**
1418 <     * Ensures that no thread is waiting for count to advance from the
1419 <     * current value of eventCount read on entry to this method, by
1420 <     * releasing waiting threads if necessary.
1421 <     *
1422 <     * @return the count
1423 <     */
1424 <    final long ensureSync() {
1425 <        long c = eventCount;
1426 <        WaitQueueNode q;
1427 <        while ((q = syncStack) != null && q.count < c) {
1428 <            if (casBarrierStack(q, null)) {
1429 <                do {
1430 <                    q.signal();
1431 <                } while ((q = q.next) != null);
1432 <                break;
1433 <            }
1434 <        }
1435 <        return c;
1436 <    }
1437 <
1438 <    /**
1439 <     * Increments event count and releases waiting threads.
1440 <     */
1441 <    private void signalIdleWorkers() {
1442 <        long c;
1443 <        do {} while (!casEventCount(c = eventCount, c+1));
1444 <        ensureSync();
1445 <    }
1446 <
1447 <    /**
1448 <     * Signals threads waiting to poll a task. Because method sync
1449 <     * rechecks availability, it is OK to only proceed if queue
1450 <     * appears to be non-empty, and OK to skip under contention to
1451 <     * increment count (since some other thread succeeded).
1452 <     */
1453 <    final void signalWork() {
1454 <        long c;
1455 <        WaitQueueNode q;
1456 <        if (syncStack != null &&
1457 <            casEventCount(c = eventCount, c+1) &&
1458 <            (((q = syncStack) != null && q.count <= c) &&
1459 <             (!casBarrierStack(q, q.next) || !q.signal())))
1460 <            ensureSync();
1461 <    }
1462 <
1463 <    /**
1464 <     * Waits until event count advances from last value held by
1465 <     * caller, or if excess threads, caller is resumed as spare, or
1466 <     * caller or pool is terminating. Updates caller's event on exit.
1467 <     *
1468 <     * @param w the calling worker thread
1469 <     */
1470 <    final void sync(ForkJoinWorkerThread w) {
1471 <        updateStealCount(w); // Transfer w's count while it is idle
1472 <
1473 <        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1474 <            long prev = w.lastEventCount;
1475 <            WaitQueueNode node = null;
1476 <            WaitQueueNode h;
1477 <            while (eventCount == prev &&
1478 <                   ((h = syncStack) == null || h.count == prev)) {
1479 <                if (node == null)
1480 <                    node = new WaitQueueNode(prev, w);
1481 <                if (casBarrierStack(node.next = h, node)) {
1482 <                    node.awaitSyncRelease(this);
1483 <                    break;
1484 <                }
1485 <            }
1486 <            long ec = ensureSync();
1487 <            if (ec != prev) {
1488 <                w.lastEventCount = ec;
1489 <                break;
1490 <            }
1491 <        }
1492 <    }
1493 <
1494 <    /**
1495 <     * Returns true if worker waiting on sync can proceed:
1496 <     *  - on signal (thread == null)
1497 <     *  - on event count advance (winning race to notify vs signaller)
1498 <     *  - on interrupt
1499 <     *  - if the first queued node, we find work available
1500 <     * If node was not signalled and event count not advanced on exit,
1501 <     * then we also help advance event count.
1502 <     *
1503 <     * @return true if node can be released
1504 <     */
1505 <    final boolean syncIsReleasable(WaitQueueNode node) {
1506 <        long prev = node.count;
1507 <        if (!Thread.interrupted() && node.thread != null &&
1508 <            (node.next != null ||
1509 <             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1510 <            eventCount == prev)
1810 >            return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1811 >        } catch(TimeoutException ex) {
1812              return false;
1512        if (node.thread != null) {
1513            node.thread = null;
1514            long ec = eventCount;
1515            if (prev <= ec) // help signal
1516                casEventCount(ec, ec+1);
1517        }
1518        return true;
1519    }
1520
1521    /**
1522     * Returns true if a new sync event occurred since last call to
1523     * sync or this method, if so, updating caller's count.
1524     */
1525    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1526        long lc = w.lastEventCount;
1527        long ec = ensureSync();
1528        if (ec == lc)
1529            return false;
1530        w.lastEventCount = ec;
1531        return true;
1532    }
1533
1534    //  Parallelism maintenance
1535
1536    /**
1537     * Decrements running count; if too low, adds spare.
1538     *
1539     * Conceptually, all we need to do here is add or resume a
1540     * spare thread when one is about to block (and remove or
1541     * suspend it later when unblocked -- see suspendIfSpare).
1542     * However, implementing this idea requires coping with
1543     * several problems: we have imperfect information about the
1544     * states of threads. Some count updates can and usually do
1545     * lag run state changes, despite arrangements to keep them
1546     * accurate (for example, when possible, updating counts
1547     * before signalling or resuming), especially when running on
1548     * dynamic JVMs that don't optimize the infrequent paths that
1549     * update counts. Generating too many threads can make these
1550     * problems become worse, because excess threads are more
1551     * likely to be context-switched with others, slowing them all
1552     * down, especially if there is no work available, so all are
1553     * busy scanning or idling.  Also, excess spare threads can
1554     * only be suspended or removed when they are idle, not
1555     * immediately when they aren't needed. So adding threads will
1556     * raise parallelism level for longer than necessary.  Also,
1557     * FJ applications often encounter highly transient peaks when
1558     * many threads are blocked joining, but for less time than it
1559     * takes to create or resume spares.
1560     *
1561     * @param joinMe if non-null, return early if done
1562     * @param maintainParallelism if true, try to stay within
1563     * target counts, else create only to avoid starvation
1564     * @return true if joinMe known to be done
1565     */
1566    final boolean preJoin(ForkJoinTask<?> joinMe,
1567                          boolean maintainParallelism) {
1568        maintainParallelism &= maintainsParallelism; // overrride
1569        boolean dec = false;  // true when running count decremented
1570        while (spareStack == null || !tryResumeSpare(dec)) {
1571            int counts = workerCounts;
1572            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1573                // CAS cheat
1574                if (!needSpare(counts, maintainParallelism))
1575                    break;
1576                if (joinMe.status < 0)
1577                    return true;
1578                if (tryAddSpare(counts))
1579                    break;
1580            }
1581        }
1582        return false;
1583    }
1584
1585    /**
1586     * Same idea as preJoin
1587     */
1588    final boolean preBlock(ManagedBlocker blocker,
1589                           boolean maintainParallelism) {
1590        maintainParallelism &= maintainsParallelism;
1591        boolean dec = false;
1592        while (spareStack == null || !tryResumeSpare(dec)) {
1593            int counts = workerCounts;
1594            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1595                if (!needSpare(counts, maintainParallelism))
1596                    break;
1597                if (blocker.isReleasable())
1598                    return true;
1599                if (tryAddSpare(counts))
1600                    break;
1601            }
1602        }
1603        return false;
1604    }
1605
1606    /**
1607     * Returns true if a spare thread appears to be needed.  If
1608     * maintaining parallelism, returns true when the deficit in
1609     * running threads is more than the surplus of total threads, and
1610     * there is apparently some work to do.  This self-limiting rule
1611     * means that the more threads that have already been added, the
1612     * less parallelism we will tolerate before adding another.
1613     *
1614     * @param counts current worker counts
1615     * @param maintainParallelism try to maintain parallelism
1616     */
1617    private boolean needSpare(int counts, boolean maintainParallelism) {
1618        int ps = parallelism;
1619        int rc = runningCountOf(counts);
1620        int tc = totalCountOf(counts);
1621        int runningDeficit = ps - rc;
1622        int totalSurplus = tc - ps;
1623        return (tc < maxPoolSize &&
1624                (rc == 0 || totalSurplus < 0 ||
1625                 (maintainParallelism &&
1626                  runningDeficit > totalSurplus &&
1627                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
1628    }
1629
1630    /**
1631     * Adds a spare worker if lock available and no more than the
1632     * expected numbers of threads exist.
1633     *
1634     * @return true if successful
1635     */
1636    private boolean tryAddSpare(int expectedCounts) {
1637        final ReentrantLock lock = this.workerLock;
1638        int expectedRunning = runningCountOf(expectedCounts);
1639        int expectedTotal = totalCountOf(expectedCounts);
1640        boolean success = false;
1641        boolean locked = false;
1642        // confirm counts while locking; CAS after obtaining lock
1643        try {
1644            for (;;) {
1645                int s = workerCounts;
1646                int tc = totalCountOf(s);
1647                int rc = runningCountOf(s);
1648                if (rc > expectedRunning || tc > expectedTotal)
1649                    break;
1650                if (!locked && !(locked = lock.tryLock()))
1651                    break;
1652                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1653                    createAndStartSpare(tc);
1654                    success = true;
1655                    break;
1656                }
1657            }
1658        } finally {
1659            if (locked)
1660                lock.unlock();
1661        }
1662        return success;
1663    }
1664
1665    /**
1666     * Adds the kth spare worker. On entry, pool counts are already
1667     * adjusted to reflect addition.
1668     */
1669    private void createAndStartSpare(int k) {
1670        ForkJoinWorkerThread w = null;
1671        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1672        int len = ws.length;
1673        // Probably, we can place at slot k. If not, find empty slot
1674        if (k < len && ws[k] != null) {
1675            for (k = 0; k < len && ws[k] != null; ++k)
1676                ;
1677        }
1678        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1679            ws[k] = w;
1680            w.start();
1681        }
1682        else
1683            updateWorkerCount(-1); // adjust on failure
1684        signalIdleWorkers();
1685    }
1686
1687    /**
1688     * Suspends calling thread w if there are excess threads.  Called
1689     * only from sync.  Spares are enqueued in a Treiber stack using
1690     * the same WaitQueueNodes as barriers.  They are resumed mainly
1691     * in preJoin, but are also woken on pool events that require all
1692     * threads to check run state.
1693     *
1694     * @param w the caller
1695     */
1696    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1697        WaitQueueNode node = null;
1698        int s;
1699        while (parallelism < runningCountOf(s = workerCounts)) {
1700            if (node == null)
1701                node = new WaitQueueNode(0, w);
1702            if (casWorkerCounts(s, s-1)) { // representation-dependent
1703                // push onto stack
1704                do {} while (!casSpareStack(node.next = spareStack, node));
1705                // block until released by resumeSpare
1706                node.awaitSpareRelease();
1707                return true;
1708            }
1709        }
1710        return false;
1711    }
1712
1713    /**
1714     * Tries to pop and resume a spare thread.
1715     *
1716     * @param updateCount if true, increment running count on success
1717     * @return true if successful
1718     */
1719    private boolean tryResumeSpare(boolean updateCount) {
1720        WaitQueueNode q;
1721        while ((q = spareStack) != null) {
1722            if (casSpareStack(q, q.next)) {
1723                if (updateCount)
1724                    updateRunningCount(1);
1725                q.signal();
1726                return true;
1727            }
1728        }
1729        return false;
1730    }
1731
1732    /**
1733     * Pops and resumes all spare threads. Same idea as ensureSync.
1734     *
1735     * @return true if any spares released
1736     */
1737    private boolean resumeAllSpares() {
1738        WaitQueueNode q;
1739        while ( (q = spareStack) != null) {
1740            if (casSpareStack(q, null)) {
1741                do {
1742                    updateRunningCount(1);
1743                    q.signal();
1744                } while ((q = q.next) != null);
1745                return true;
1746            }
1747        }
1748        return false;
1749    }
1750
1751    /**
1752     * Pops and shuts down excessive spare threads. Call only while
1753     * holding lock. This is not guaranteed to eliminate all excess
1754     * threads, only those suspended as spares, which are the ones
1755     * unlikely to be needed in the future.
1756     */
1757    private void trimSpares() {
1758        int surplus = totalCountOf(workerCounts) - parallelism;
1759        WaitQueueNode q;
1760        while (surplus > 0 && (q = spareStack) != null) {
1761            if (casSpareStack(q, null)) {
1762                do {
1763                    updateRunningCount(1);
1764                    ForkJoinWorkerThread w = q.thread;
1765                    if (w != null && surplus > 0 &&
1766                        runningCountOf(workerCounts) > 0 && w.shutdown())
1767                        --surplus;
1768                    q.signal();
1769                } while ((q = q.next) != null);
1770            }
1813          }
1814      }
1815  
1816      /**
1817       * Interface for extending managed parallelism for tasks running
1818 <     * in ForkJoinPools. A ManagedBlocker provides two methods.
1819 <     * Method {@code isReleasable} must return true if blocking is not
1820 <     * necessary. Method {@code block} blocks the current thread if
1821 <     * necessary (perhaps internally invoking {@code isReleasable}
1822 <     * before actually blocking.).
1818 >     * in {@link ForkJoinPool}s.
1819 >     *
1820 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1821 >     * {@code isReleasable} must return {@code true} if blocking is
1822 >     * not necessary. Method {@code block} blocks the current thread
1823 >     * if necessary (perhaps internally invoking {@code isReleasable}
1824 >     * before actually blocking). The unusual methods in this API
1825 >     * accommodate synchronizers that may, but don't usually, block
1826 >     * for long periods. Similarly, they allow more efficient internal
1827 >     * handling of cases in which additional workers may be, but
1828 >     * usually are not, needed to ensure sufficient parallelism.
1829 >     * Toward this end, implementations of method {@code isReleasable}
1830 >     * must be amenable to repeated invocation.
1831       *
1832       * <p>For example, here is a ManagedBlocker based on a
1833       * ReentrantLock:
# Line 1795 | Line 1845 | public class ForkJoinPool extends Abstra
1845       *     return hasLock || (hasLock = lock.tryLock());
1846       *   }
1847       * }}</pre>
1848 +     *
1849 +     * <p>Here is a class that possibly blocks waiting for an
1850 +     * item on a given queue:
1851 +     *  <pre> {@code
1852 +     * class QueueTaker<E> implements ManagedBlocker {
1853 +     *   final BlockingQueue<E> queue;
1854 +     *   volatile E item = null;
1855 +     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1856 +     *   public boolean block() throws InterruptedException {
1857 +     *     if (item == null)
1858 +     *       item = queue.take();
1859 +     *     return true;
1860 +     *   }
1861 +     *   public boolean isReleasable() {
1862 +     *     return item != null || (item = queue.poll()) != null;
1863 +     *   }
1864 +     *   public E getItem() { // call after pool.managedBlock completes
1865 +     *     return item;
1866 +     *   }
1867 +     * }}</pre>
1868       */
1869      public static interface ManagedBlocker {
1870          /**
1871           * Possibly blocks the current thread, for example waiting for
1872           * a lock or condition.
1873           *
1874 <         * @return true if no additional blocking is necessary (i.e.,
1875 <         * if isReleasable would return true)
1874 >         * @return {@code true} if no additional blocking is necessary
1875 >         * (i.e., if isReleasable would return true)
1876           * @throws InterruptedException if interrupted while waiting
1877           * (the method is not required to do so, but is allowed to)
1878           */
1879          boolean block() throws InterruptedException;
1880  
1881          /**
1882 <         * Returns true if blocking is unnecessary.
1882 >         * Returns {@code true} if blocking is unnecessary.
1883           */
1884          boolean isReleasable();
1885      }
1886  
1887      /**
1888       * Blocks in accord with the given blocker.  If the current thread
1889 <     * is a ForkJoinWorkerThread, this method possibly arranges for a
1890 <     * spare thread to be activated if necessary to ensure parallelism
1891 <     * while the current thread is blocked.  If
1822 <     * {@code maintainParallelism} is true and the pool supports
1823 <     * it ({@link #getMaintainsParallelism}), this method attempts to
1824 <     * maintain the pool's nominal parallelism. Otherwise it activates
1825 <     * a thread only if necessary to avoid complete starvation. This
1826 <     * option may be preferable when blockages use timeouts, or are
1827 <     * almost always brief.
1889 >     * is a {@link ForkJoinWorkerThread}, this method possibly
1890 >     * arranges for a spare thread to be activated if necessary to
1891 >     * ensure sufficient parallelism while the current thread is blocked.
1892       *
1893 <     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
1894 <     * equivalent to
1893 >     * <p>If the caller is not a {@link ForkJoinTask}, this method is
1894 >     * behaviorally equivalent to
1895       *  <pre> {@code
1896       * while (!blocker.isReleasable())
1897       *   if (blocker.block())
1898       *     return;
1899       * }</pre>
1900 <     * If the caller is a ForkJoinTask, then the pool may first
1901 <     * be expanded to ensure parallelism, and later adjusted.
1900 >     *
1901 >     * If the caller is a {@code ForkJoinTask}, then the pool may
1902 >     * first be expanded to ensure parallelism, and later adjusted.
1903       *
1904       * @param blocker the blocker
1840     * @param maintainParallelism if true and supported by this pool,
1841     * attempt to maintain the pool's nominal parallelism; otherwise
1842     * activate a thread only if necessary to avoid complete
1843     * starvation.
1905       * @throws InterruptedException if blocker.block did so
1906       */
1907 <    public static void managedBlock(ManagedBlocker blocker,
1847 <                                    boolean maintainParallelism)
1907 >    public static void managedBlock(ManagedBlocker blocker)
1908          throws InterruptedException {
1909          Thread t = Thread.currentThread();
1910 <        ForkJoinPool pool = ((t instanceof ForkJoinWorkerThread) ?
1911 <                             ((ForkJoinWorkerThread) t).pool : null);
1912 <        if (!blocker.isReleasable()) {
1913 <            try {
1914 <                if (pool == null ||
1915 <                    !pool.preBlock(blocker, maintainParallelism))
1856 <                    awaitBlocker(blocker);
1857 <            } finally {
1858 <                if (pool != null)
1859 <                    pool.updateRunningCount(1);
1860 <            }
1910 >        if (t instanceof ForkJoinWorkerThread) {
1911 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1912 >            w.pool.awaitBlocker(blocker);
1913 >        }
1914 >        else {
1915 >            do {} while (!blocker.isReleasable() && !blocker.block());
1916          }
1917      }
1918  
1919 <    private static void awaitBlocker(ManagedBlocker blocker)
1920 <        throws InterruptedException {
1921 <        do {} while (!blocker.isReleasable() && !blocker.block());
1867 <    }
1868 <
1869 <    // AbstractExecutorService overrides
1919 >    // AbstractExecutorService overrides.  These rely on undocumented
1920 >    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
1921 >    // implement RunnableFuture.
1922  
1923      protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
1924 <        return new AdaptedRunnable<T>(runnable, value);
1924 >        return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
1925      }
1926  
1927      protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
1928 <        return new AdaptedCallable<T>(callable);
1928 >        return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
1929      }
1930  
1931 +    // Unsafe mechanics
1932  
1933 <    // Unsafe mechanics for jsr166y 3rd party package.
1933 >    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1934 >    private static final long workerCountsOffset =
1935 >        objectFieldOffset("workerCounts", ForkJoinPool.class);
1936 >    private static final long runStateOffset =
1937 >        objectFieldOffset("runState", ForkJoinPool.class);
1938 >    private static final long eventCountOffset =
1939 >        objectFieldOffset("eventCount", ForkJoinPool.class);
1940 >    private static final long eventWaitersOffset =
1941 >        objectFieldOffset("eventWaiters",ForkJoinPool.class);
1942 >    private static final long stealCountOffset =
1943 >        objectFieldOffset("stealCount",ForkJoinPool.class);
1944 >    private static final long spareWaitersOffset =
1945 >        objectFieldOffset("spareWaiters",ForkJoinPool.class);
1946 >
1947 >    private static long objectFieldOffset(String field, Class<?> klazz) {
1948 >        try {
1949 >            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1950 >        } catch (NoSuchFieldException e) {
1951 >            // Convert Exception to corresponding Error
1952 >            NoSuchFieldError error = new NoSuchFieldError(field);
1953 >            error.initCause(e);
1954 >            throw error;
1955 >        }
1956 >    }
1957 >
1958 >    /**
1959 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1960 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
1961 >     * into a jdk.
1962 >     *
1963 >     * @return a sun.misc.Unsafe
1964 >     */
1965      private static sun.misc.Unsafe getUnsafe() {
1966          try {
1967              return sun.misc.Unsafe.getUnsafe();
1968          } catch (SecurityException se) {
1969              try {
1970                  return java.security.AccessController.doPrivileged
1971 <                    (new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1971 >                    (new java.security
1972 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1973                          public sun.misc.Unsafe run() throws Exception {
1974 <                            return getUnsafeByReflection();
1974 >                            java.lang.reflect.Field f = sun.misc
1975 >                                .Unsafe.class.getDeclaredField("theUnsafe");
1976 >                            f.setAccessible(true);
1977 >                            return (sun.misc.Unsafe) f.get(null);
1978                          }});
1979              } catch (java.security.PrivilegedActionException e) {
1980                  throw new RuntimeException("Could not initialize intrinsics",
# Line 1894 | Line 1982 | public class ForkJoinPool extends Abstra
1982              }
1983          }
1984      }
1897
1898    private static sun.misc.Unsafe getUnsafeByReflection()
1899            throws NoSuchFieldException, IllegalAccessException {
1900        java.lang.reflect.Field f =
1901            sun.misc.Unsafe.class.getDeclaredField("theUnsafe");
1902        f.setAccessible(true);
1903        return (sun.misc.Unsafe) f.get(null);
1904    }
1905
1906    private static long fieldOffset(String fieldName, Class<?> klazz) {
1907        try {
1908            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(fieldName));
1909        } catch (NoSuchFieldException e) {
1910            // Convert Exception to Error
1911            NoSuchFieldError error = new NoSuchFieldError(fieldName);
1912            error.initCause(e);
1913            throw error;
1914        }
1915    }
1916
1917    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1918    static final long eventCountOffset =
1919        fieldOffset("eventCount", ForkJoinPool.class);
1920    static final long workerCountsOffset =
1921        fieldOffset("workerCounts", ForkJoinPool.class);
1922    static final long runControlOffset =
1923        fieldOffset("runControl", ForkJoinPool.class);
1924    static final long syncStackOffset =
1925        fieldOffset("syncStack",ForkJoinPool.class);
1926    static final long spareStackOffset =
1927        fieldOffset("spareStack", ForkJoinPool.class);
1928
1929    private boolean casEventCount(long cmp, long val) {
1930        return UNSAFE.compareAndSwapLong(this, eventCountOffset, cmp, val);
1931    }
1932    private boolean casWorkerCounts(int cmp, int val) {
1933        return UNSAFE.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1934    }
1935    private boolean casRunControl(int cmp, int val) {
1936        return UNSAFE.compareAndSwapInt(this, runControlOffset, cmp, val);
1937    }
1938    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1939        return UNSAFE.compareAndSwapObject(this, spareStackOffset, cmp, val);
1940    }
1941    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1942        return UNSAFE.compareAndSwapObject(this, syncStackOffset, cmp, val);
1943    }
1985   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines