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 1.2 by dl, Wed Jan 7 16:07:37 2009 UTC vs.
Revision 1.69 by jsr166, Wed Sep 1 20:12:39 2010 UTC

# Line 5 | Line 5
5   */
6  
7   package jsr166y;
8 < import java.util.*;
8 >
9   import java.util.concurrent.*;
10 < import java.util.concurrent.locks.*;
11 < import java.util.concurrent.atomic.*;
12 < import sun.misc.Unsafe;
13 < import java.lang.reflect.*;
10 >
11 > import java.util.ArrayList;
12 > 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.LockSupport;
17 > import java.util.concurrent.locks.ReentrantLock;
18 > import java.util.concurrent.atomic.AtomicInteger;
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
20 < * submitted tasks. Otherwise, use would not usually outweigh the
21 < * construction and bookkeeping overhead of creating a large set of
22 < * 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. Otherwise, other
34 < * ExecutorService implementations are typically more appropriate
35 < * 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</code> interface enables extension of
43 < * the kinds of synchronization accommodated.  The target parallelism
44 < * level may also be changed dynamically (<code>setParallelism</code>)
45 < * and dynamically thread construction can be limited using methods
43 < * <code>setMaximumPoolSize</code> and/or
44 < * <code>setMaintainsParallelism</code>.
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</code>) 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</code> 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>Arrange 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
117   */
118   public class ForkJoinPool extends AbstractExecutorService {
119  
120      /*
121 <     * See the extended comments interspersed below for design,
122 <     * rationale, and walkthroughs.
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.  Because 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
161 >     *      re-activate a spare thread to compensate for blocked
162 >     *      joiners until they unblock.
163 >     *
164 >     * It is impossible to keep exactly the target (parallelism)
165 >     * number of threads running at any given time.  Determining
166 >     * existence of conservatively safe helping targets, the
167 >     * availability of already-created spares, and the apparent need
168 >     * to create new spares are all racy and require heuristic
169 >     * guidance, so we rely on multiple retries of each.  Compensation
170 >     * occurs in slow-motion. It is triggered only upon timeouts of
171 >     * Object.wait used for joins. This reduces poor decisions that
172 >     * would otherwise be made when threads are waiting for others
173 >     * that are stalled because of unrelated activities such as
174 >     * garbage collection.
175 >     *
176 >     * The ManagedBlocker extension API can't use helping so relies
177 >     * only on compensation in method awaitBlocker.
178 >     *
179 >     * The main throughput advantages of work-stealing stem from
180 >     * decentralized control -- workers mostly steal tasks from each
181 >     * other. We do not want to negate this by creating bottlenecks
182 >     * implementing other management responsibilities. So we use a
183 >     * collection of techniques that avoid, reduce, or cope well with
184 >     * contention. These entail several instances of bit-packing into
185 >     * CASable fields to maintain only the minimally required
186 >     * atomicity. To enable such packing, we restrict maximum
187 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
188 >     * unbalanced increments and decrements) to fit into a 16 bit
189 >     * field, which is far in excess of normal operating range.  Even
190 >     * though updates to some of these bookkeeping fields do sometimes
191 >     * contend with each other, they don't normally cache-contend with
192 >     * updates to others enough to warrant memory padding or
193 >     * isolation. So they are all held as fields of ForkJoinPool
194 >     * objects.  The main capabilities are as follows:
195 >     *
196 >     * 1. Creating and removing workers. Workers are recorded in the
197 >     * "workers" array. This is an array as opposed to some other data
198 >     * structure to support index-based random steals by workers.
199 >     * Updates to the array recording new workers and unrecording
200 >     * terminated ones are protected from each other by a lock
201 >     * (workerLock) but the array is otherwise concurrently readable,
202 >     * and accessed directly by workers. To simplify index-based
203 >     * operations, the array size is always a power of two, and all
204 >     * readers must tolerate null slots. Currently, all worker thread
205 >     * creation is on-demand, triggered by task submissions,
206 >     * replacement of terminated workers, and/or compensation for
207 >     * blocked workers. However, all other support code is set up to
208 >     * work with other policies.
209 >     *
210 >     * To ensure that we do not hold on to worker references that
211 >     * would prevent GC, ALL accesses to workers are via indices into
212 >     * the workers array (which is one source of some of the unusual
213 >     * code constructions here). In essence, the workers array serves
214 >     * as a WeakReference mechanism. Thus for example the event queue
215 >     * stores worker indices, not worker references. Access to the
216 >     * workers in associated methods (for example releaseEventWaiters)
217 >     * must both index-check and null-check the IDs. All such accesses
218 >     * ignore bad IDs by returning out early from what they are doing,
219 >     * since this can only be associated with shutdown, in which case
220 >     * it is OK to give up. On termination, we just clobber these
221 >     * data structures without trying to use them.
222 >     *
223 >     * 2. Bookkeeping for dynamically adding and removing workers. We
224 >     * aim to approximately maintain the given level of parallelism.
225 >     * When some workers are known to be blocked (on joins or via
226 >     * ManagedBlocker), we may create or resume others to take their
227 >     * place until they unblock (see below). Implementing this
228 >     * requires counts of the number of "running" threads (i.e., those
229 >     * that are neither blocked nor artificially suspended) as well as
230 >     * the total number.  These two values are packed into one field,
231 >     * "workerCounts" because we need accurate snapshots when deciding
232 >     * to create, resume or suspend.  Note however that the
233 >     * correspondence of these counts to reality is not guaranteed. In
234 >     * particular updates for unblocked threads may lag until they
235 >     * actually wake up.
236 >     *
237 >     * 3. Maintaining global run state. The run state of the pool
238 >     * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
239 >     * those in other Executor implementations, as well as a count of
240 >     * "active" workers -- those that are, or soon will be, or
241 >     * recently were executing tasks. The runLevel and active count
242 >     * are packed together in order to correctly trigger shutdown and
243 >     * termination. Without care, active counts can be subject to very
244 >     * high contention.  We substantially reduce this contention by
245 >     * relaxing update rules.  A worker must claim active status
246 >     * prospectively, by activating if it sees that a submitted or
247 >     * stealable task exists (it may find after activating that the
248 >     * task no longer exists). It stays active while processing this
249 >     * task (if it exists) and any other local subtasks it produces,
250 >     * until it cannot find any other tasks. It then tries
251 >     * inactivating (see method preStep), but upon update contention
252 >     * instead scans for more tasks, later retrying inactivation if it
253 >     * doesn't find any.
254 >     *
255 >     * 4. Managing idle workers waiting for tasks. We cannot let
256 >     * workers spin indefinitely scanning for tasks when none are
257 >     * available. On the other hand, we must quickly prod them into
258 >     * action when new tasks are submitted or generated.  We
259 >     * park/unpark these idle workers using an event-count scheme.
260 >     * Field eventCount is incremented upon events that may enable
261 >     * workers that previously could not find a task to now find one:
262 >     * Submission of a new task to the pool, or another worker pushing
263 >     * a task onto a previously empty queue.  (We also use this
264 >     * mechanism for configuration and termination actions that
265 >     * require wakeups of idle workers).  Each worker maintains its
266 >     * last known event count, and blocks when a scan for work did not
267 >     * find a task AND its lastEventCount matches the current
268 >     * eventCount. Waiting idle workers are recorded in a variant of
269 >     * Treiber stack headed by field eventWaiters which, when nonzero,
270 >     * encodes the thread index and count awaited for by the worker
271 >     * thread most recently calling eventSync. This thread in turn has
272 >     * a record (field nextEventWaiter) for the next waiting worker.
273 >     * In addition to allowing simpler decisions about need for
274 >     * wakeup, the event count bits in eventWaiters serve the role of
275 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
276 >     * released threads also try to release at most two others.  The
277 >     * net effect is a tree-like diffusion of signals, where released
278 >     * threads (and possibly others) help with unparks.  To further
279 >     * reduce contention effects a bit, failed CASes to increment
280 >     * field eventCount are tolerated without retries in signalWork.
281 >     * Conceptually they are merged into the same event, which is OK
282 >     * when their only purpose is to enable workers to scan for work.
283 >     *
284 >     * 5. Managing suspension of extra workers. When a worker notices
285 >     * (usually upon timeout of a wait()) that there are too few
286 >     * running threads, we may create a new thread to maintain
287 >     * parallelism level, or at least avoid starvation. Usually, extra
288 >     * threads are needed for only very short periods, yet join
289 >     * dependencies are such that we sometimes need them in
290 >     * bursts. Rather than create new threads each time this happens,
291 >     * we suspend no-longer-needed extra ones as "spares". For most
292 >     * purposes, we don't distinguish "extra" spare threads from
293 >     * normal "core" threads: On each call to preStep (the only point
294 >     * at which we can do this) a worker checks to see if there are
295 >     * now too many running workers, and if so, suspends itself.
296 >     * Method helpMaintainParallelism looks for suspended threads to
297 >     * resume before considering creating a new replacement. The
298 >     * spares themselves are encoded on another variant of a Treiber
299 >     * Stack, headed at field "spareWaiters".  Note that the use of
300 >     * spares is intrinsically racy.  One thread may become a spare at
301 >     * about the same time as another is needlessly being created. We
302 >     * counteract this and related slop in part by requiring resumed
303 >     * spares to immediately recheck (in preStep) to see whether they
304 >     * they should re-suspend.
305 >     *
306 >     * 6. Killing off unneeded workers. A timeout mechanism is used to
307 >     * shed unused workers: The oldest (first) event queue waiter uses
308 >     * a timed rather than hard wait. When this wait times out without
309 >     * a normal wakeup, it tries to shutdown any one (for convenience
310 >     * the newest) other spare or event waiter via
311 >     * tryShutdownUnusedWorker. This eventually reduces the number of
312 >     * worker threads to a minimum of one after a long enough period
313 >     * without use.
314 >     *
315 >     * 7. Deciding when to create new workers. The main dynamic
316 >     * control in this class is deciding when to create extra threads
317 >     * in method helpMaintainParallelism. We would like to keep
318 >     * exactly #parallelism threads running, which is an impossible
319 >     * task. We always need to create one when the number of running
320 >     * threads would become zero and all workers are busy. Beyond
321 >     * this, we must rely on heuristics that work well in the
322 >     * presence of transient phenomena such as GC stalls, dynamic
323 >     * compilation, and wake-up lags. These transients are extremely
324 >     * common -- we are normally trying to fully saturate the CPUs on
325 >     * a machine, so almost any activity other than running tasks
326 >     * impedes accuracy. Our main defense is to allow parallelism to
327 >     * lapse for a while during joins, and use a timeout to see if,
328 >     * after the resulting settling, there is still a need for
329 >     * additional workers.  This also better copes with the fact that
330 >     * some of the methods in this class tend to never become compiled
331 >     * (but are interpreted), so some components of the entire set of
332 >     * controls might execute 100 times faster than others. And
333 >     * similarly for cases where the apparent lack of work is just due
334 >     * to GC stalls and other transient system activity.
335 >     *
336 >     * Beware that there is a lot of representation-level coupling
337 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
338 >     * ForkJoinTask.  For example, direct access to "workers" array by
339 >     * workers, and direct access to ForkJoinTask.status by both
340 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
341 >     * trying to reduce this, since any associated future changes in
342 >     * representations will need to be accompanied by algorithmic
343 >     * changes anyway.
344 >     *
345 >     * Style notes: There are lots of inline assignments (of form
346 >     * "while ((local = field) != 0)") which are usually the simplest
347 >     * way to ensure the required read orderings (which are sometimes
348 >     * critical). Also several occurrences of the unusual "do {}
349 >     * while (!cas...)" which is the simplest way to force an update of
350 >     * a CAS'ed variable. There are also other coding oddities that
351 >     * help some methods perform reasonably even when interpreted (not
352 >     * compiled), at the expense of some messy constructions that
353 >     * reduce byte code counts.
354 >     *
355 >     * The order of declarations in this file is: (1) statics (2)
356 >     * fields (along with constants used when unpacking some of them)
357 >     * (3) internal control methods (4) callbacks and other support
358 >     * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
359 >     * methods (plus a few little helpers).
360       */
361  
65    /** Mask for packing and unpacking shorts */
66    private static final int  shortMask = 0xffff;
67
68    /** Max pool size -- must be a power of two minus 1 */
69    private static final int MAX_THREADS =  0x7FFF;
70
362      /**
363 <     * Factory for creating new ForkJoinWorkerThreads.  A
364 <     * ForkJoinWorkerThreadFactory must be defined and used for
365 <     * ForkJoinWorkerThread subclasses that extend base functionality
366 <     * or initialize threads with different contexts.
363 >     * Factory for creating new {@link ForkJoinWorkerThread}s.
364 >     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
365 >     * for {@code ForkJoinWorkerThread} subclasses that extend base
366 >     * functionality or initialize threads with different contexts.
367       */
368      public static interface ForkJoinWorkerThreadFactory {
369          /**
370           * Returns a new worker thread operating in the given pool.
371           *
372           * @param pool the pool this thread works in
373 <         * @throws NullPointerException if pool is null;
373 >         * @throws NullPointerException if the pool is null
374           */
375          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
376      }
377  
378      /**
379 <     * Default ForkJoinWorkerThreadFactory implementation, creates a
379 >     * Default ForkJoinWorkerThreadFactory implementation; creates a
380       * new ForkJoinWorkerThread.
381       */
382 <    static class  DefaultForkJoinWorkerThreadFactory
382 >    static class DefaultForkJoinWorkerThreadFactory
383          implements ForkJoinWorkerThreadFactory {
384          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
385 <            try {
95 <                return new ForkJoinWorkerThread(pool);
96 <            } catch (OutOfMemoryError oom)  {
97 <                return null;
98 <            }
385 >            return new ForkJoinWorkerThread(pool);
386          }
387      }
388  
# Line 131 | Line 418 | public class ForkJoinPool extends Abstra
418          new AtomicInteger();
419  
420      /**
421 <     * Array holding all worker threads in the pool. Array size must
422 <     * be a power of two.  Updates and replacements are protected by
423 <     * workerLock, but it is always kept in a consistent enough state
424 <     * to be randomly accessed without locking by workers performing
425 <     * work-stealing.
421 >     * The time to block in a join (see awaitJoin) before checking if
422 >     * a new worker should be (re)started to maintain parallelism
423 >     * level. The value should be short enough to maintain global
424 >     * responsiveness and progress but long enough to avoid
425 >     * counterproductive firings during GC stalls or unrelated system
426 >     * activity, and to not bog down systems with continual re-firings
427 >     * on GCs or legitimately long waits.
428       */
429 <    volatile ForkJoinWorkerThread[] workers;
429 >    private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
430  
431      /**
432 <     * Lock protecting access to workers.
432 >     * The wakeup interval (in nanoseconds) for the oldest worker
433 >     * worker waiting for an event invokes tryShutdownUnusedWorker to shrink
434 >     * the number of workers.  The exact value does not matter too
435 >     * much, but should be long enough to slowly release resources
436 >     * during long periods without use without disrupting normal use.
437       */
438 <    private final ReentrantLock workerLock;
438 >    private static final long SHRINK_RATE_NANOS =
439 >        30L * 1000L * 1000L * 1000L; // 2 per minute
440  
441      /**
442 <     * Condition for awaitTermination.
442 >     * Absolute bound for parallelism level. Twice this number plus
443 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
444 >     * word-packing for some counts and indices.
445       */
446 <    private final Condition termination;
446 >    private static final int MAX_WORKERS   = 0x7fff;
447  
448      /**
449 <     * The uncaught exception handler used when any worker
450 <     * abrupty terminates
449 >     * Array holding all worker threads in the pool.  Array size must
450 >     * be a power of two.  Updates and replacements are protected by
451 >     * workerLock, but the array is always kept in a consistent enough
452 >     * state to be randomly accessed without locking by workers
453 >     * performing work-stealing, as well as other traversal-based
454 >     * methods in this class. All readers must tolerate that some
455 >     * array slots may be null.
456       */
457 <    private Thread.UncaughtExceptionHandler ueh;
457 >    volatile ForkJoinWorkerThread[] workers;
458  
459      /**
460 <     * Creation factory for worker threads.
460 >     * Queue for external submissions.
461       */
462 <    private final ForkJoinWorkerThreadFactory factory;
462 >    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
463  
464      /**
465 <     * Head of stack of threads that were created to maintain
165 <     * parallelism when other threads blocked, but have since
166 <     * suspended when the parallelism level rose.
465 >     * Lock protecting updates to workers array.
466       */
467 <    private volatile WaitQueueNode spareStack;
467 >    private final ReentrantLock workerLock;
468  
469      /**
470 <     * Sum of per-thread steal counts, updated only when threads are
172 <     * idle or terminating.
470 >     * Latch released upon termination.
471       */
472 <    private final AtomicLong stealCount;
472 >    private final Phaser termination;
473  
474      /**
475 <     * Queue for external submissions.
475 >     * Creation factory for worker threads.
476       */
477 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
477 >    private final ForkJoinWorkerThreadFactory factory;
478  
479      /**
480 <     * Head of Treiber stack for barrier sync. See below for explanation
480 >     * Sum of per-thread steal counts, updated only when threads are
481 >     * idle or terminating.
482       */
483 <    private volatile WaitQueueNode barrierStack;
483 >    private volatile long stealCount;
484  
485      /**
486 <     * The count for event barrier
486 >     * Encoded record of top of Treiber stack of threads waiting for
487 >     * events. The top 32 bits contain the count being waited for. The
488 >     * bottom 16 bits contains one plus the pool index of waiting
489 >     * worker thread. (Bits 16-31 are unused.)
490       */
491 <    private volatile long eventCount;
491 >    private volatile long eventWaiters;
492 >
493 >    private static final int  EVENT_COUNT_SHIFT = 32;
494 >    private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
495  
496      /**
497 <     * Pool number, just for assigning useful names to worker threads
497 >     * A counter for events that may wake up worker threads:
498 >     *   - Submission of a new task to the pool
499 >     *   - A worker pushing a task on an empty queue
500 >     *   - termination
501       */
502 <    private final int poolNumber;
502 >    private volatile int eventCount;
503  
504      /**
505 <     * The maximum allowed pool size
505 >     * Encoded record of top of Treiber stack of spare threads waiting
506 >     * for resumption. The top 16 bits contain an arbitrary count to
507 >     * avoid ABA effects. The bottom 16bits contains one plus the pool
508 >     * index of waiting worker thread.
509       */
510 <    private volatile int maxPoolSize;
510 >    private volatile int spareWaiters;
511 >
512 >    private static final int SPARE_COUNT_SHIFT = 16;
513 >    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
514  
515      /**
516 <     * The desired parallelism level, updated only under workerLock.
517 <     */
518 <    private volatile int parallelism;
516 >     * Lifecycle control. The low word contains the number of workers
517 >     * that are (probably) executing tasks. This value is atomically
518 >     * incremented before a worker gets a task to run, and decremented
519 >     * when worker has no tasks and cannot find any.  Bits 16-18
520 >     * contain runLevel value. When all are zero, the pool is
521 >     * running. Level transitions are monotonic (running -> shutdown
522 >     * -> terminating -> terminated) so each transition adds a bit.
523 >     * These are bundled together to ensure consistent read for
524 >     * termination checks (i.e., that runLevel is at least SHUTDOWN
525 >     * and active threads is zero).
526 >     *
527 >     * Notes: Most direct CASes are dependent on these bitfield
528 >     * positions.  Also, this field is non-private to enable direct
529 >     * performance-sensitive CASes in ForkJoinWorkerThread.
530 >     */
531 >    volatile int runState;
532 >
533 >    // Note: The order among run level values matters.
534 >    private static final int RUNLEVEL_SHIFT     = 16;
535 >    private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
536 >    private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
537 >    private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
538 >    private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
539  
540      /**
541       * Holds number of total (i.e., created and not yet terminated)
542       * and running (i.e., not blocked on joins or other managed sync)
543 <     * threads, packed into one int to ensure consistent snapshot when
543 >     * threads, packed together to ensure consistent snapshot when
544       * making decisions about creating and suspending spare
545 <     * threads. Updated only by CAS.  Note: CASes in
546 <     * updateRunningCount and preJoin running active count is in low
547 <     * word, so need to be modified if this changes
545 >     * threads. Updated only by CAS. Note that adding a new worker
546 >     * requires incrementing both counts, since workers start off in
547 >     * running state.
548       */
549      private volatile int workerCounts;
550  
551 <    private static int totalCountOf(int s)           { return s >>> 16;  }
552 <    private static int runningCountOf(int s)         { return s & shortMask; }
553 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
551 >    private static final int TOTAL_COUNT_SHIFT  = 16;
552 >    private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
553 >    private static final int ONE_RUNNING        = 1;
554 >    private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
555  
556      /**
557 <     * Add delta (which may be negative) to running count.  This must
558 <     * be called before (with negative arg) and after (with positive)
224 <     * any managed synchronization (i.e., mainly, joins)
225 <     * @param delta the number to add
557 >     * The target parallelism level.
558 >     * Accessed directly by ForkJoinWorkerThreads.
559       */
560 <    final void updateRunningCount(int delta) {
228 <        int s;
229 <        do;while (!casWorkerCounts(s = workerCounts, s + delta));
230 <    }
560 >    final int parallelism;
561  
562      /**
563 <     * Add delta (which may be negative) to both total and running
564 <     * count.  This must be called upon creation and termination of
235 <     * worker threads.
236 <     * @param delta the number to add
563 >     * True if use local fifo, not default lifo, for local polling
564 >     * Read by, and replicated by ForkJoinWorkerThreads
565       */
566 <    private void updateWorkerCount(int delta) {
239 <        int d = delta + (delta << 16); // add to both lo and hi parts
240 <        int s;
241 <        do;while (!casWorkerCounts(s = workerCounts, s + d));
242 <    }
566 >    final boolean locallyFifo;
567  
568      /**
569 <     * Lifecycle control. High word contains runState, low word
570 <     * contains the number of workers that are (probably) executing
247 <     * tasks. This value is atomically incremented before a worker
248 <     * gets a task to run, and decremented when worker has no tasks
249 <     * and cannot find any. These two fields are bundled together to
250 <     * support correct termination triggering.  Note: activeCount
251 <     * CAS'es cheat by assuming active count is in low word, so need
252 <     * to be modified if this changes
569 >     * The uncaught exception handler used when any worker abruptly
570 >     * terminates.
571       */
572 <    private volatile int runControl;
572 >    private final Thread.UncaughtExceptionHandler ueh;
573  
574 <    // RunState values. Order among values matters
575 <    private static final int RUNNING     = 0;
576 <    private static final int SHUTDOWN    = 1;
577 <    private static final int TERMINATING = 2;
260 <    private static final int TERMINATED  = 3;
574 >    /**
575 >     * Pool number, just for assigning useful names to worker threads
576 >     */
577 >    private final int poolNumber;
578  
579 <    private static int runStateOf(int c)             { return c >>> 16; }
580 <    private static int activeCountOf(int c)          { return c & shortMask; }
264 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
579 >    // Utilities for CASing fields. Note that most of these
580 >    // are usually manually inlined by callers
581  
582      /**
583 <     * Increment active count. Called by workers before/during
268 <     * executing tasks.
583 >     * Increments running count part of workerCounts
584       */
585 <    final void incrementActiveCount() {
585 >    final void incrementRunningCount() {
586          int c;
587 <        do;while (!casRunControl(c = runControl, c+1));
587 >        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
588 >                                               c = workerCounts,
589 >                                               c + ONE_RUNNING));
590      }
591  
592      /**
593 <     * Decrement active count; possibly trigger termination.
277 <     * Called by workers when they can't find tasks.
593 >     * Tries to decrement running count unless already zero
594       */
595 <    final void decrementActiveCount() {
596 <        int c, nextc;
597 <        do;while (!casRunControl(c = runControl, nextc = c-1));
598 <        if (canTerminateOnShutdown(nextc))
599 <            terminateOnShutdown();
595 >    final boolean tryDecrementRunningCount() {
596 >        int wc = workerCounts;
597 >        if ((wc & RUNNING_COUNT_MASK) == 0)
598 >            return false;
599 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
600 >                                        wc, wc - ONE_RUNNING);
601      }
602  
603      /**
604 <     * Return true if argument represents zero active count and
605 <     * nonzero runstate, which is the triggering condition for
606 <     * terminating on shutdown.
604 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
605 >     * (rarely) necessary when other count updates lag.
606 >     *
607 >     * @param dr -- either zero or ONE_RUNNING
608 >     * @param dt == either zero or ONE_TOTAL
609       */
610 <    private static boolean canTerminateOnShutdown(int c) {
611 <        return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
610 >    private void decrementWorkerCounts(int dr, int dt) {
611 >        for (;;) {
612 >            int wc = workerCounts;
613 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
614 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
615 >                if ((runState & TERMINATED) != 0)
616 >                    return; // lagging termination on a backout
617 >                Thread.yield();
618 >            }
619 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
620 >                                         wc, wc - (dr + dt)))
621 >                return;
622 >        }
623      }
624  
625      /**
626 <     * Transition run state to at least the given state. Return true
627 <     * if not already at least given state.
626 >     * Tries decrementing active count; fails on contention.
627 >     * Called when workers cannot find tasks to run.
628       */
629 <    private boolean transitionRunStateTo(int state) {
629 >    final boolean tryDecrementActiveCount() {
630 >        int c;
631 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
632 >                                        c = runState, c - 1);
633 >    }
634 >
635 >    /**
636 >     * Advances to at least the given level. Returns true if not
637 >     * already in at least the given level.
638 >     */
639 >    private boolean advanceRunLevel(int level) {
640          for (;;) {
641 <            int c = runControl;
642 <            if (runStateOf(c) >= state)
641 >            int s = runState;
642 >            if ((s & level) != 0)
643                  return false;
644 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
644 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
645                  return true;
646          }
647      }
648  
649 +    // workers array maintenance
650 +
651      /**
652 <     * Controls whether to add spares to maintain parallelism
652 >     * Records and returns a workers array index for new worker.
653       */
654 <    private volatile boolean maintainsParallelism;
654 >    private int recordWorker(ForkJoinWorkerThread w) {
655 >        // Try using slot totalCount-1. If not available, scan and/or resize
656 >        int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
657 >        final ReentrantLock lock = this.workerLock;
658 >        lock.lock();
659 >        try {
660 >            ForkJoinWorkerThread[] ws = workers;
661 >            int n = ws.length;
662 >            if (k < 0 || k >= n || ws[k] != null) {
663 >                for (k = 0; k < n && ws[k] != null; ++k)
664 >                    ;
665 >                if (k == n)
666 >                    ws = Arrays.copyOf(ws, n << 1);
667 >            }
668 >            ws[k] = w;
669 >            workers = ws; // volatile array write ensures slot visibility
670 >        } finally {
671 >            lock.unlock();
672 >        }
673 >        return k;
674 >    }
675  
676 <    // Constructors
676 >    /**
677 >     * Nulls out record of worker in workers array
678 >     */
679 >    private void forgetWorker(ForkJoinWorkerThread w) {
680 >        int idx = w.poolIndex;
681 >        // Locking helps method recordWorker avoid unnecessary expansion
682 >        final ReentrantLock lock = this.workerLock;
683 >        lock.lock();
684 >        try {
685 >            ForkJoinWorkerThread[] ws = workers;
686 >            if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
687 >                ws[idx] = null;
688 >        } finally {
689 >            lock.unlock();
690 >        }
691 >    }
692  
693      /**
694 <     * Creates a ForkJoinPool with a pool size equal to the number of
695 <     * processors available on the system and using the default
696 <     * ForkJoinWorkerThreadFactory,
697 <     * @throws SecurityException if a security manager exists and
698 <     *         the caller is not permitted to modify threads
322 <     *         because it does not hold {@link
323 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
694 >     * Final callback from terminating worker.  Removes record of
695 >     * worker from array, and adjusts counts. If pool is shutting
696 >     * down, tries to complete termination.
697 >     *
698 >     * @param w the worker
699       */
700 <    public ForkJoinPool() {
701 <        this(Runtime.getRuntime().availableProcessors(),
702 <             defaultForkJoinWorkerThreadFactory);
700 >    final void workerTerminated(ForkJoinWorkerThread w) {
701 >        forgetWorker(w);
702 >        decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
703 >        while (w.stealCount != 0) // collect final count
704 >            tryAccumulateStealCount(w);
705 >        tryTerminate(false);
706      }
707  
708 +    // Waiting for and signalling events
709 +
710      /**
711 <     * Creates a ForkJoinPool with the indicated parellelism level
712 <     * threads, and using the default ForkJoinWorkerThreadFactory,
713 <     * @param parallelism the number of worker threads
714 <     * @throws IllegalArgumentException if parallelism less than or
335 <     * equal to zero
336 <     * @throws SecurityException if a security manager exists and
337 <     *         the caller is not permitted to modify threads
338 <     *         because it does not hold {@link
339 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
711 >     * Releases workers blocked on a count not equal to current count.
712 >     * Normally called after precheck that eventWaiters isn't zero to
713 >     * avoid wasted array checks. Gives up upon a change in count or
714 >     * upon releasing two workers, letting others take over.
715       */
716 <    public ForkJoinPool(int parallelism) {
717 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
716 >    private void releaseEventWaiters() {
717 >        ForkJoinWorkerThread[] ws = workers;
718 >        int n = ws.length;
719 >        long h = eventWaiters;
720 >        int ec = eventCount;
721 >        boolean releasedOne = false;
722 >        ForkJoinWorkerThread w; int id;
723 >        while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
724 >               (int)(h >>> EVENT_COUNT_SHIFT) != ec &&
725 >               id < n && (w = ws[id]) != null) {
726 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
727 >                                          h,  w.nextWaiter)) {
728 >                LockSupport.unpark(w);
729 >                if (releasedOne) // exit on second release
730 >                    break;
731 >                releasedOne = true;
732 >            }
733 >            if (eventCount != ec)
734 >                break;
735 >            h = eventWaiters;
736 >        }
737      }
738  
739      /**
740 <     * Creates a ForkJoinPool with parallelism equal to the number of
741 <     * processors available on the system and using the given
348 <     * ForkJoinWorkerThreadFactory,
349 <     * @param factory the factory for creating new threads
350 <     * @throws NullPointerException if factory is null
351 <     * @throws SecurityException if a security manager exists and
352 <     *         the caller is not permitted to modify threads
353 <     *         because it does not hold {@link
354 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
740 >     * Tries to advance eventCount and releases waiters. Called only
741 >     * from workers.
742       */
743 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
744 <        this(Runtime.getRuntime().availableProcessors(), factory);
743 >    final void signalWork() {
744 >        int c; // try to increment event count -- CAS failure OK
745 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
746 >        if (eventWaiters != 0L)
747 >            releaseEventWaiters();
748      }
749  
750      /**
751 <     * Creates a ForkJoinPool with the given parallelism and factory.
751 >     * Adds the given worker to event queue and blocks until
752 >     * terminating or event count advances from the given value
753       *
754 <     * @param parallelism the targeted number of worker threads
755 <     * @param factory the factory for creating new threads
365 <     * @throws IllegalArgumentException if parallelism less than or
366 <     * equal to zero, or greater than implementation limit.
367 <     * @throws NullPointerException if factory is null
368 <     * @throws SecurityException if a security manager exists and
369 <     *         the caller is not permitted to modify threads
370 <     *         because it does not hold {@link
371 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
754 >     * @param w the calling worker thread
755 >     * @param ec the count
756       */
757 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
758 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
759 <            throw new IllegalArgumentException();
760 <        if (factory == null)
761 <            throw new NullPointerException();
762 <        checkPermission();
763 <        this.factory = factory;
764 <        this.parallelism = parallelism;
765 <        this.maxPoolSize = MAX_THREADS;
766 <        this.maintainsParallelism = true;
767 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
768 <        this.workerLock = new ReentrantLock();
769 <        this.termination = workerLock.newCondition();
386 <        this.stealCount = new AtomicLong();
387 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
388 <        createAndStartInitialWorkers(parallelism);
757 >    private void eventSync(ForkJoinWorkerThread w, int ec) {
758 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
759 >        long h;
760 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
761 >               (((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
762 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
763 >               eventCount == ec) {
764 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
765 >                                          w.nextWaiter = h, nh)) {
766 >                awaitEvent(w, ec);
767 >                break;
768 >            }
769 >        }
770      }
771  
772      /**
773 <     * Create new worker using factory.
774 <     * @param index the index to assign worker
775 <     * @return new worker, or null of factory failed
773 >     * Blocks the given worker (that has already been entered as an
774 >     * event waiter) until terminating or event count advances from
775 >     * the given value. The oldest (first) waiter uses a timed wait to
776 >     * occasionally one-by-one shrink the number of workers (to a
777 >     * minimum of one) if the pool has not been used for extended
778 >     * periods.
779 >     *
780 >     * @param w the calling worker thread
781 >     * @param ec the count
782       */
783 <    private ForkJoinWorkerThread createWorker(int index) {
784 <        Thread.UncaughtExceptionHandler h = ueh;
785 <        ForkJoinWorkerThread w = factory.newThread(this);
786 <        if (w != null) {
787 <            w.poolIndex = index;
788 <            w.setDaemon(true);
789 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
790 <            if (h != null)
791 <                w.setUncaughtExceptionHandler(h);
783 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
784 >        while (eventCount == ec) {
785 >            if (tryAccumulateStealCount(w)) { // transfer while idle
786 >                boolean untimed = (w.nextWaiter != 0L ||
787 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
788 >                long startTime = untimed? 0 : System.nanoTime();
789 >                Thread.interrupted();         // clear/ignore interrupt
790 >                if (eventCount != ec || w.runState != 0 ||
791 >                    runState >= TERMINATING)  // recheck after clear
792 >                    break;
793 >                if (untimed)
794 >                    LockSupport.park(w);
795 >                else {
796 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
797 >                    if (eventCount != ec || w.runState != 0 ||
798 >                        runState >= TERMINATING)
799 >                        break;
800 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
801 >                        tryShutdownUnusedWorker(ec);
802 >                }
803 >            }
804          }
406        return w;
805      }
806  
807 +    // Maintaining parallelism
808 +
809      /**
810 <     * Return a good size for worker array given pool size.
411 <     * Currently requires size to be a power of two.
810 >     * Pushes worker onto the spare stack
811       */
812 <    private static int arraySizeFor(int ps) {
813 <        return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
812 >    final void pushSpare(ForkJoinWorkerThread w) {
813 >        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
814 >        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
815 >                                               w.nextSpare = spareWaiters,ns));
816      }
817  
818      /**
819 <     * Create or resize array if necessary to hold newLength
820 <     * @return the array
819 >     * Tries (once) to resume a spare if the number of running
820 >     * threads is less than target.
821       */
822 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
822 >    private void tryResumeSpare() {
823 >        int sw, id;
824          ForkJoinWorkerThread[] ws = workers;
825 <        if (ws == null)
826 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
827 <        else if (newLength > ws.length)
828 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
829 <        else
830 <            return ws;
825 >        int n = ws.length;
826 >        ForkJoinWorkerThread w;
827 >        if ((sw = spareWaiters) != 0 &&
828 >            (id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
829 >            id < n && (w = ws[id]) != null &&
830 >            (workerCounts & RUNNING_COUNT_MASK) < parallelism &&
831 >            spareWaiters == sw &&
832 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
833 >                                     sw, w.nextSpare)) {
834 >            int c; // increment running count before resume
835 >            do {} while (!UNSAFE.compareAndSwapInt
836 >                         (this, workerCountsOffset,
837 >                          c = workerCounts, c + ONE_RUNNING));
838 >            if (w.tryUnsuspend())
839 >                LockSupport.unpark(w);
840 >            else   // back out if w was shutdown
841 >                decrementWorkerCounts(ONE_RUNNING, 0);
842 >        }
843 >    }
844 >
845 >    /**
846 >     * Tries to increase the number of running workers if below target
847 >     * parallelism: If a spare exists tries to resume it via
848 >     * tryResumeSpare.  Otherwise, if not enough total workers or all
849 >     * existing workers are busy, adds a new worker. In all cases also
850 >     * helps wake up releasable workers waiting for work.
851 >     */
852 >    private void helpMaintainParallelism() {
853 >        int pc = parallelism;
854 >        int wc, rs, tc;
855 >        while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
856 >               (rs = runState) < TERMINATING) {
857 >            if (spareWaiters != 0)
858 >                tryResumeSpare();
859 >            else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
860 >                     (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
861 >                break;   // enough total
862 >            else if (runState == rs && workerCounts == wc &&
863 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
864 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
865 >                ForkJoinWorkerThread w = null;
866 >                try {
867 >                    w = factory.newThread(this);
868 >                } finally { // adjust on null or exceptional factory return
869 >                    if (w == null) {
870 >                        decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
871 >                        tryTerminate(false); // handle failure during shutdown
872 >                    }
873 >                }
874 >                if (w == null)
875 >                    break;
876 >                w.start(recordWorker(w), ueh);
877 >                if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
878 >                    int c; // advance event count
879 >                    UNSAFE.compareAndSwapInt(this, eventCountOffset,
880 >                                             c = eventCount, c+1);
881 >                    break; // add at most one unless total below target
882 >                }
883 >            }
884 >        }
885 >        if (eventWaiters != 0L)
886 >            releaseEventWaiters();
887      }
888  
889      /**
890 <     * Try to shrink workers into smaller array after one or more terminate
890 >     * Callback from the oldest waiter in awaitEvent waking up after a
891 >     * period of non-use. If all workers are idle, tries (once) to
892 >     * shutdown an event waiter or a spare, if one exists. Note that
893 >     * we don't need CAS or locks here because the method is called
894 >     * only from one thread occasionally waking (and even misfires are
895 >     * OK). Note that until the shutdown worker fully terminates,
896 >     * workerCounts will overestimate total count, which is tolerable.
897 >     *
898 >     * @param ec the event count waited on by caller (to abort
899 >     * attempt if count has since changed).
900       */
901 <    private void tryShrinkWorkerArray() {
902 <        ForkJoinWorkerThread[] ws = workers;
903 <        int len = ws.length;
904 <        int last = len - 1;
905 <        while (last >= 0 && ws[last] == null)
906 <            --last;
907 <        int newLength = arraySizeFor(last+1);
908 <        if (newLength < len)
909 <            workers = Arrays.copyOf(ws, newLength);
901 >    private void tryShutdownUnusedWorker(int ec) {
902 >        if (runState == 0 && eventCount == ec) { // only trigger if all idle
903 >            ForkJoinWorkerThread[] ws = workers;
904 >            int n = ws.length;
905 >            ForkJoinWorkerThread w = null;
906 >            boolean shutdown = false;
907 >            int sw;
908 >            long h;
909 >            if ((sw = spareWaiters) != 0) { // prefer killing spares
910 >                int id = (sw & SPARE_ID_MASK) - 1;
911 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
912 >                    UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
913 >                                             sw, w.nextSpare))
914 >                    shutdown = true;
915 >            }
916 >            else if ((h = eventWaiters) != 0L) {
917 >                long nh;
918 >                int id = ((int)(h & WAITER_ID_MASK)) - 1;
919 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
920 >                    (nh = w.nextWaiter) != 0L && // keep at least one worker
921 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
922 >                    shutdown = true;
923 >            }
924 >            if (w != null && shutdown) {
925 >                w.shutdown();
926 >                LockSupport.unpark(w);
927 >            }
928 >        }
929 >        releaseEventWaiters(); // in case of interference
930      }
931  
932      /**
933 <     * Initial worker array and worker creation and startup. (This
934 <     * must be done under lock to avoid interference by some of the
935 <     * newly started threads while creating others.)
933 >     * Callback from workers invoked upon each top-level action (i.e.,
934 >     * stealing a task or taking a submission and running it).
935 >     * Performs one or more of the following:
936 >     *
937 >     * 1. If the worker is active and either did not run a task
938 >     *    or there are too many workers, try to set its active status
939 >     *    to inactive and update activeCount. On contention, we may
940 >     *    try again in this or a subsequent call.
941 >     *
942 >     * 2. If not enough total workers, help create some.
943 >     *
944 >     * 3. If there are too many running workers, suspend this worker
945 >     *    (first forcing inactive if necessary).  If it is not needed,
946 >     *    it may be shutdown while suspended (via
947 >     *    tryShutdownUnusedWorker).  Otherwise, upon resume it
948 >     *    rechecks running thread count and need for event sync.
949 >     *
950 >     * 4. If worker did not run a task, await the next task event via
951 >     *    eventSync if necessary (first forcing inactivation), upon
952 >     *    which the worker may be shutdown via
953 >     *    tryShutdownUnusedWorker.  Otherwise, help release any
954 >     *    existing event waiters that are now releasable,
955 >     *
956 >     * @param w the worker
957 >     * @param ran true if worker ran a task since last call to this method
958       */
959 <    private void createAndStartInitialWorkers(int ps) {
960 <        final ReentrantLock lock = this.workerLock;
961 <        lock.lock();
962 <        try {
963 <            ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
964 <            for (int i = 0; i < ps; ++i) {
965 <                ForkJoinWorkerThread w = createWorker(i);
966 <                if (w != null) {
967 <                    ws[i] = w;
968 <                    w.start();
969 <                    updateWorkerCount(1);
959 >    final void preStep(ForkJoinWorkerThread w, boolean ran) {
960 >        int wec = w.lastEventCount;
961 >        boolean active = w.active;
962 >        boolean inactivate = false;
963 >        int pc = parallelism;
964 >        int rs;
965 >        while (w.runState == 0 && (rs = runState) < TERMINATING) {
966 >            if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
967 >                UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
968 >                inactivate = active = w.active = false;
969 >            int wc = workerCounts;
970 >            if ((wc & RUNNING_COUNT_MASK) > pc) {
971 >                if (!(inactivate |= active) && // must inactivate to suspend
972 >                    workerCounts == wc &&      // try to suspend as spare
973 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
974 >                                             wc, wc - ONE_RUNNING))
975 >                    w.suspendAsSpare();
976 >            }
977 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
978 >                helpMaintainParallelism();     // not enough workers
979 >            else if (!ran) {
980 >                long h = eventWaiters;
981 >                int ec = eventCount;
982 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
983 >                    releaseEventWaiters();     // release others before waiting
984 >                else if (ec != wec) {
985 >                    w.lastEventCount = ec;     // no need to wait
986 >                    break;
987                  }
988 +                else if (!(inactivate |= active))
989 +                    eventSync(w, wec);         // must inactivate before sync
990              }
991 <        } finally {
992 <            lock.unlock();
991 >            else
992 >                break;
993          }
994      }
995  
996      /**
997 <     * Worker creation and startup for threads added via setParallelism.
997 >     * Helps and/or blocks awaiting join of the given task.
998 >     * See above for explanation.
999 >     *
1000 >     * @param joinMe the task to join
1001 >     * @param worker the current worker thread
1002       */
1003 <    private void createAndStartAddedWorkers() {
1004 <        resumeAllSpares();  // Allow spares to convert to nonspare
1005 <        int ps = parallelism;
1006 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
1007 <        int len = ws.length;
1008 <        // Sweep through slots, to keep lowest indices most populated
1009 <        int k = 0;
1010 <        while (k < len) {
1011 <            if (ws[k] != null) {
1012 <                ++k;
1013 <                continue;
1003 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1004 >        int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1005 >        while (joinMe.status >= 0) {
1006 >            int wc;
1007 >            worker.helpJoinTask(joinMe);
1008 >            if (joinMe.status < 0)
1009 >                break;
1010 >            else if (retries > 0)
1011 >                --retries;
1012 >            else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1013 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1014 >                                              wc, wc - ONE_RUNNING)) {
1015 >                int stat, c; long h;
1016 >                while ((stat = joinMe.status) >= 0 &&
1017 >                       (h = eventWaiters) != 0L && // help release others
1018 >                       (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1019 >                    releaseEventWaiters();
1020 >                if (stat >= 0 &&
1021 >                    ((workerCounts & RUNNING_COUNT_MASK) == 0 ||
1022 >                     (stat =
1023 >                      joinMe.internalAwaitDone(JOIN_TIMEOUT_MILLIS)) >= 0))
1024 >                    helpMaintainParallelism(); // timeout or no running workers
1025 >                do {} while (!UNSAFE.compareAndSwapInt
1026 >                             (this, workerCountsOffset,
1027 >                              c = workerCounts, c + ONE_RUNNING));
1028 >                if (stat < 0)
1029 >                    break;   // else restart
1030              }
1031 <            int s = workerCounts;
1032 <            int tc = totalCountOf(s);
1033 <            int rc = runningCountOf(s);
1034 <            if (rc >= ps || tc >= ps)
1031 >        }
1032 >    }
1033 >
1034 >    /**
1035 >     * Same idea as awaitJoin, but no helping, retries, or timeouts.
1036 >     */
1037 >    final void awaitBlocker(ManagedBlocker blocker)
1038 >        throws InterruptedException {
1039 >        while (!blocker.isReleasable()) {
1040 >            int wc = workerCounts;
1041 >            if ((wc & RUNNING_COUNT_MASK) != 0 &&
1042 >                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1043 >                                         wc, wc - ONE_RUNNING)) {
1044 >                try {
1045 >                    while (!blocker.isReleasable()) {
1046 >                        long h = eventWaiters;
1047 >                        if (h != 0L &&
1048 >                            (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1049 >                            releaseEventWaiters();
1050 >                        else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1051 >                                 runState < TERMINATING)
1052 >                            helpMaintainParallelism();
1053 >                        else if (blocker.block())
1054 >                            break;
1055 >                    }
1056 >                } finally {
1057 >                    int c;
1058 >                    do {} while (!UNSAFE.compareAndSwapInt
1059 >                                 (this, workerCountsOffset,
1060 >                                  c = workerCounts, c + ONE_RUNNING));
1061 >                }
1062                  break;
1063 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
1064 <                ForkJoinWorkerThread w = createWorker(k);
1063 >            }
1064 >        }
1065 >    }
1066 >
1067 >    /**
1068 >     * Possibly initiates and/or completes termination.
1069 >     *
1070 >     * @param now if true, unconditionally terminate, else only
1071 >     * if shutdown and empty queue and no active workers
1072 >     * @return true if now terminating or terminated
1073 >     */
1074 >    private boolean tryTerminate(boolean now) {
1075 >        if (now)
1076 >            advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1077 >        else if (runState < SHUTDOWN ||
1078 >                 !submissionQueue.isEmpty() ||
1079 >                 (runState & ACTIVE_COUNT_MASK) != 0)
1080 >            return false;
1081 >
1082 >        if (advanceRunLevel(TERMINATING))
1083 >            startTerminating();
1084 >
1085 >        // Finish now if all threads terminated; else in some subsequent call
1086 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1087 >            advanceRunLevel(TERMINATED);
1088 >            termination.arrive();
1089 >        }
1090 >        return true;
1091 >    }
1092 >
1093 >    /**
1094 >     * Actions on transition to TERMINATING
1095 >     *
1096 >     * Runs up to four passes through workers: (0) shutting down each
1097 >     * (without waking up if parked) to quickly spread notifications
1098 >     * without unnecessary bouncing around event queues etc (1) wake
1099 >     * up and help cancel tasks (2) interrupt (3) mop up races with
1100 >     * interrupted workers
1101 >     */
1102 >    private void startTerminating() {
1103 >        cancelSubmissions();
1104 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1105 >            int c; // advance event count
1106 >            UNSAFE.compareAndSwapInt(this, eventCountOffset,
1107 >                                     c = eventCount, c+1);
1108 >            eventWaiters = 0L; // clobber lists
1109 >            spareWaiters = 0;
1110 >            ForkJoinWorkerThread[] ws = workers;
1111 >            int n = ws.length;
1112 >            for (int i = 0; i < n; ++i) {
1113 >                ForkJoinWorkerThread w = ws[i];
1114                  if (w != null) {
1115 <                    ws[k++] = w;
1116 <                    w.start();
1117 <                }
1118 <                else {
1119 <                    updateWorkerCount(-1); // back out on failed creation
1120 <                    break;
1115 >                    w.shutdown();
1116 >                    if (passes > 0 && !w.isTerminated()) {
1117 >                        w.cancelTasks();
1118 >                        LockSupport.unpark(w);
1119 >                        if (passes > 1) {
1120 >                            try {
1121 >                                w.interrupt();
1122 >                            } catch (SecurityException ignore) {
1123 >                            }
1124 >                        }
1125 >                    }
1126                  }
1127              }
1128          }
1129      }
1130  
1131 +    /**
1132 +     * Clear out and cancel submissions, ignoring exceptions
1133 +     */
1134 +    private void cancelSubmissions() {
1135 +        ForkJoinTask<?> task;
1136 +        while ((task = submissionQueue.poll()) != null) {
1137 +            try {
1138 +                task.cancel(false);
1139 +            } catch (Throwable ignore) {
1140 +            }
1141 +        }
1142 +    }
1143 +
1144 +    // misc support for ForkJoinWorkerThread
1145 +
1146 +    /**
1147 +     * Returns pool number
1148 +     */
1149 +    final int getPoolNumber() {
1150 +        return poolNumber;
1151 +    }
1152 +
1153 +    /**
1154 +     * Tries to accumulates steal count from a worker, clearing
1155 +     * the worker's value.
1156 +     *
1157 +     * @return true if worker steal count now zero
1158 +     */
1159 +    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1160 +        int sc = w.stealCount;
1161 +        long c = stealCount;
1162 +        // CAS even if zero, for fence effects
1163 +        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1164 +            if (sc != 0)
1165 +                w.stealCount = 0;
1166 +            return true;
1167 +        }
1168 +        return sc == 0;
1169 +    }
1170 +
1171 +    /**
1172 +     * Returns the approximate (non-atomic) number of idle threads per
1173 +     * active thread.
1174 +     */
1175 +    final int idlePerActive() {
1176 +        int pc = parallelism; // use parallelism, not rc
1177 +        int ac = runState;    // no mask -- artificially boosts during shutdown
1178 +        // Use exact results for small values, saturate past 4
1179 +        return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1180 +    }
1181 +
1182 +    // Public and protected methods
1183 +
1184 +    // Constructors
1185 +
1186 +    /**
1187 +     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1188 +     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1189 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1190 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1191 +     *
1192 +     * @throws SecurityException if a security manager exists and
1193 +     *         the caller is not permitted to modify threads
1194 +     *         because it does not hold {@link
1195 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1196 +     */
1197 +    public ForkJoinPool() {
1198 +        this(Runtime.getRuntime().availableProcessors(),
1199 +             defaultForkJoinWorkerThreadFactory, null, false);
1200 +    }
1201 +
1202 +    /**
1203 +     * Creates a {@code ForkJoinPool} with the indicated parallelism
1204 +     * level, the {@linkplain
1205 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1206 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1207 +     *
1208 +     * @param parallelism the parallelism level
1209 +     * @throws IllegalArgumentException if parallelism less than or
1210 +     *         equal to zero, or greater than implementation limit
1211 +     * @throws SecurityException if a security manager exists and
1212 +     *         the caller is not permitted to modify threads
1213 +     *         because it does not hold {@link
1214 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1215 +     */
1216 +    public ForkJoinPool(int parallelism) {
1217 +        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1218 +    }
1219 +
1220 +    /**
1221 +     * Creates a {@code ForkJoinPool} with the given parameters.
1222 +     *
1223 +     * @param parallelism the parallelism level. For default value,
1224 +     * use {@link java.lang.Runtime#availableProcessors}.
1225 +     * @param factory the factory for creating new threads. For default value,
1226 +     * use {@link #defaultForkJoinWorkerThreadFactory}.
1227 +     * @param handler the handler for internal worker threads that
1228 +     * terminate due to unrecoverable errors encountered while executing
1229 +     * tasks. For default value, use <code>null</code>.
1230 +     * @param asyncMode if true,
1231 +     * establishes local first-in-first-out scheduling mode for forked
1232 +     * tasks that are never joined. This mode may be more appropriate
1233 +     * than default locally stack-based mode in applications in which
1234 +     * worker threads only process event-style asynchronous tasks.
1235 +     * For default value, use <code>false</code>.
1236 +     * @throws IllegalArgumentException if parallelism less than or
1237 +     *         equal to zero, or greater than implementation limit
1238 +     * @throws NullPointerException if the factory is null
1239 +     * @throws SecurityException if a security manager exists and
1240 +     *         the caller is not permitted to modify threads
1241 +     *         because it does not hold {@link
1242 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1243 +     */
1244 +    public ForkJoinPool(int parallelism,
1245 +                        ForkJoinWorkerThreadFactory factory,
1246 +                        Thread.UncaughtExceptionHandler handler,
1247 +                        boolean asyncMode) {
1248 +        checkPermission();
1249 +        if (factory == null)
1250 +            throw new NullPointerException();
1251 +        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1252 +            throw new IllegalArgumentException();
1253 +        this.parallelism = parallelism;
1254 +        this.factory = factory;
1255 +        this.ueh = handler;
1256 +        this.locallyFifo = asyncMode;
1257 +        int arraySize = initialArraySizeFor(parallelism);
1258 +        this.workers = new ForkJoinWorkerThread[arraySize];
1259 +        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1260 +        this.workerLock = new ReentrantLock();
1261 +        this.termination = new Phaser(1);
1262 +        this.poolNumber = poolNumberGenerator.incrementAndGet();
1263 +    }
1264 +
1265 +    /**
1266 +     * Returns initial power of two size for workers array.
1267 +     * @param pc the initial parallelism level
1268 +     */
1269 +    private static int initialArraySizeFor(int pc) {
1270 +        // If possible, initially allocate enough space for one spare
1271 +        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1272 +        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1273 +        size |= size >>> 1;
1274 +        size |= size >>> 2;
1275 +        size |= size >>> 4;
1276 +        size |= size >>> 8;
1277 +        return size + 1;
1278 +    }
1279 +
1280      // Execution methods
1281  
1282      /**
1283       * Common code for execute, invoke and submit
1284       */
1285      private <T> void doSubmit(ForkJoinTask<T> task) {
1286 <        if (isShutdown())
1286 >        if (task == null)
1287 >            throw new NullPointerException();
1288 >        if (runState >= SHUTDOWN)
1289              throw new RejectedExecutionException();
1290          submissionQueue.offer(task);
1291 <        signalIdleWorkers(true);
1291 >        int c; // try to increment event count -- CAS failure OK
1292 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1293 >        helpMaintainParallelism(); // create, start, or resume some workers
1294      }
1295  
1296      /**
1297 <     * Performs the given task; returning its result upon completion
1297 >     * Performs the given task, returning its result upon completion.
1298 >     *
1299       * @param task the task
1300       * @return the task's result
1301 <     * @throws NullPointerException if task is null
1302 <     * @throws RejectedExecutionException if pool is shut down
1301 >     * @throws NullPointerException if the task is null
1302 >     * @throws RejectedExecutionException if the task cannot be
1303 >     *         scheduled for execution
1304       */
1305      public <T> T invoke(ForkJoinTask<T> task) {
1306          doSubmit(task);
# Line 525 | Line 1309 | public class ForkJoinPool extends Abstra
1309  
1310      /**
1311       * Arranges for (asynchronous) execution of the given task.
1312 +     *
1313       * @param task the task
1314 <     * @throws NullPointerException if task is null
1315 <     * @throws RejectedExecutionException if pool is shut down
1314 >     * @throws NullPointerException if the task is null
1315 >     * @throws RejectedExecutionException if the task cannot be
1316 >     *         scheduled for execution
1317       */
1318 <    public <T> void execute(ForkJoinTask<T> task) {
1318 >    public void execute(ForkJoinTask<?> task) {
1319          doSubmit(task);
1320      }
1321  
1322      // AbstractExecutorService methods
1323  
1324 +    /**
1325 +     * @throws NullPointerException if the task is null
1326 +     * @throws RejectedExecutionException if the task cannot be
1327 +     *         scheduled for execution
1328 +     */
1329      public void execute(Runnable task) {
1330 <        doSubmit(new AdaptedRunnable<Void>(task, null));
1330 >        ForkJoinTask<?> job;
1331 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1332 >            job = (ForkJoinTask<?>) task;
1333 >        else
1334 >            job = ForkJoinTask.adapt(task, null);
1335 >        doSubmit(job);
1336      }
1337  
1338 +    /**
1339 +     * Submits a ForkJoinTask for execution.
1340 +     *
1341 +     * @param task the task to submit
1342 +     * @return the task
1343 +     * @throws NullPointerException if the task is null
1344 +     * @throws RejectedExecutionException if the task cannot be
1345 +     *         scheduled for execution
1346 +     */
1347 +    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1348 +        doSubmit(task);
1349 +        return task;
1350 +    }
1351 +
1352 +    /**
1353 +     * @throws NullPointerException if the task is null
1354 +     * @throws RejectedExecutionException if the task cannot be
1355 +     *         scheduled for execution
1356 +     */
1357      public <T> ForkJoinTask<T> submit(Callable<T> task) {
1358 <        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
1358 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1359          doSubmit(job);
1360          return job;
1361      }
1362  
1363 +    /**
1364 +     * @throws NullPointerException if the task is null
1365 +     * @throws RejectedExecutionException if the task cannot be
1366 +     *         scheduled for execution
1367 +     */
1368      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1369 <        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
1369 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1370          doSubmit(job);
1371          return job;
1372      }
1373  
1374 +    /**
1375 +     * @throws NullPointerException if the task is null
1376 +     * @throws RejectedExecutionException if the task cannot be
1377 +     *         scheduled for execution
1378 +     */
1379      public ForkJoinTask<?> submit(Runnable task) {
1380 <        ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
1380 >        ForkJoinTask<?> job;
1381 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1382 >            job = (ForkJoinTask<?>) task;
1383 >        else
1384 >            job = ForkJoinTask.adapt(task, null);
1385          doSubmit(job);
1386          return job;
1387      }
1388  
1389      /**
1390 <     * Adaptor for Runnables. This implements RunnableFuture
1391 <     * to be compliant with AbstractExecutorService constraints
563 <     */
564 <    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
565 <        implements RunnableFuture<T> {
566 <        final Runnable runnable;
567 <        final T resultOnCompletion;
568 <        T result;
569 <        AdaptedRunnable(Runnable runnable, T result) {
570 <            if (runnable == null) throw new NullPointerException();
571 <            this.runnable = runnable;
572 <            this.resultOnCompletion = result;
573 <        }
574 <        public T getRawResult() { return result; }
575 <        public void setRawResult(T v) { result = v; }
576 <        public boolean exec() {
577 <            runnable.run();
578 <            result = resultOnCompletion;
579 <            return true;
580 <        }
581 <        public void run() { invoke(); }
582 <    }
583 <
584 <    /**
585 <     * Adaptor for Callables
1390 >     * @throws NullPointerException       {@inheritDoc}
1391 >     * @throws RejectedExecutionException {@inheritDoc}
1392       */
587    static final class AdaptedCallable<T> extends ForkJoinTask<T>
588        implements RunnableFuture<T> {
589        final Callable<T> callable;
590        T result;
591        AdaptedCallable(Callable<T> callable) {
592            if (callable == null) throw new NullPointerException();
593            this.callable = callable;
594        }
595        public T getRawResult() { return result; }
596        public void setRawResult(T v) { result = v; }
597        public boolean exec() {
598            try {
599                result = callable.call();
600                return true;
601            } catch (Error err) {
602                throw err;
603            } catch (RuntimeException rex) {
604                throw rex;
605            } catch (Exception ex) {
606                throw new RuntimeException(ex);
607            }
608        }
609        public void run() { invoke(); }
610    }
611
1393      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1394 <        ArrayList<ForkJoinTask<T>> ts =
1394 >        ArrayList<ForkJoinTask<T>> forkJoinTasks =
1395              new ArrayList<ForkJoinTask<T>>(tasks.size());
1396 <        for (Callable<T> c : tasks)
1397 <            ts.add(new AdaptedCallable<T>(c));
1398 <        invoke(new InvokeAll<T>(ts));
1399 <        return (List<Future<T>>)(List)ts;
1396 >        for (Callable<T> task : tasks)
1397 >            forkJoinTasks.add(ForkJoinTask.adapt(task));
1398 >        invoke(new InvokeAll<T>(forkJoinTasks));
1399 >
1400 >        @SuppressWarnings({"unchecked", "rawtypes"})
1401 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1402 >        return futures;
1403      }
1404  
1405      static final class InvokeAll<T> extends RecursiveAction {
1406          final ArrayList<ForkJoinTask<T>> tasks;
1407          InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
1408          public void compute() {
1409 <            try { invokeAll(tasks); } catch(Exception ignore) {}
1409 >            try { invokeAll(tasks); }
1410 >            catch (Exception ignore) {}
1411          }
1412 +        private static final long serialVersionUID = -7914297376763021607L;
1413      }
1414  
629    // Configuration and status settings and queries
630
1415      /**
1416 <     * Returns the factory used for constructing new workers
1416 >     * Returns the factory used for constructing new workers.
1417       *
1418       * @return the factory used for constructing new workers
1419       */
# Line 640 | Line 1424 | public class ForkJoinPool extends Abstra
1424      /**
1425       * Returns the handler for internal worker threads that terminate
1426       * due to unrecoverable errors encountered while executing tasks.
643     * @return the handler, or null if none
644     */
645    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
646        Thread.UncaughtExceptionHandler h;
647        final ReentrantLock lock = this.workerLock;
648        lock.lock();
649        try {
650            h = ueh;
651        } finally {
652            lock.unlock();
653        }
654        return h;
655    }
656
657    /**
658     * Sets the handler for internal worker threads that terminate due
659     * to unrecoverable errors encountered while executing tasks.
660     * Unless set, the current default or ThreadGroup handler is used
661     * as handler.
1427       *
1428 <     * @param h the new handler
664 <     * @return the old handler, or null if none
665 <     * @throws SecurityException if a security manager exists and
666 <     *         the caller is not permitted to modify threads
667 <     *         because it does not hold {@link
668 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1428 >     * @return the handler, or {@code null} if none
1429       */
1430 <    public Thread.UncaughtExceptionHandler
1431 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
672 <        checkPermission();
673 <        Thread.UncaughtExceptionHandler old = null;
674 <        final ReentrantLock lock = this.workerLock;
675 <        lock.lock();
676 <        try {
677 <            old = ueh;
678 <            ueh = h;
679 <            ForkJoinWorkerThread[] ws = workers;
680 <            for (int i = 0; i < ws.length; ++i) {
681 <                ForkJoinWorkerThread w = ws[i];
682 <                if (w != null)
683 <                    w.setUncaughtExceptionHandler(h);
684 <            }
685 <        } finally {
686 <            lock.unlock();
687 <        }
688 <        return old;
689 <    }
690 <
691 <
692 <    /**
693 <     * Sets the target paralleism level of this pool.
694 <     * @param parallelism the target parallelism
695 <     * @throws IllegalArgumentException if parallelism less than or
696 <     * equal to zero or greater than maximum size bounds.
697 <     * @throws SecurityException if a security manager exists and
698 <     *         the caller is not permitted to modify threads
699 <     *         because it does not hold {@link
700 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
701 <     */
702 <    public void setParallelism(int parallelism) {
703 <        checkPermission();
704 <        if (parallelism <= 0 || parallelism > maxPoolSize)
705 <            throw new IllegalArgumentException();
706 <        final ReentrantLock lock = this.workerLock;
707 <        lock.lock();
708 <        try {
709 <            if (!isTerminating()) {
710 <                int p = this.parallelism;
711 <                this.parallelism = parallelism;
712 <                if (parallelism > p)
713 <                    createAndStartAddedWorkers();
714 <                else
715 <                    trimSpares();
716 <            }
717 <        } finally {
718 <            lock.unlock();
719 <        }
720 <        signalIdleWorkers(false);
1430 >    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1431 >        return ueh;
1432      }
1433  
1434      /**
1435 <     * Returns the targeted number of worker threads in this pool.
1435 >     * Returns the targeted parallelism level of this pool.
1436       *
1437 <     * @return the targeted number of worker threads in this pool
1437 >     * @return the targeted parallelism level of this pool
1438       */
1439      public int getParallelism() {
1440          return parallelism;
# Line 732 | Line 1443 | public class ForkJoinPool extends Abstra
1443      /**
1444       * Returns the number of worker threads that have started but not
1445       * yet terminated.  This result returned by this method may differ
1446 <     * from <code>getParallelism</code> when threads are created to
1446 >     * from {@link #getParallelism} when threads are created to
1447       * maintain parallelism when others are cooperatively blocked.
1448       *
1449       * @return the number of worker threads
1450       */
1451      public int getPoolSize() {
1452 <        return totalCountOf(workerCounts);
742 <    }
743 <
744 <    /**
745 <     * Returns the maximum number of threads allowed to exist in the
746 <     * pool, even if there are insufficient unblocked running threads.
747 <     * @return the maximum
748 <     */
749 <    public int getMaximumPoolSize() {
750 <        return maxPoolSize;
1452 >        return workerCounts >>> TOTAL_COUNT_SHIFT;
1453      }
1454  
1455      /**
1456 <     * Sets the maximum number of threads allowed to exist in the
1457 <     * pool, even if there are insufficient unblocked running threads.
1458 <     * Setting this value has no effect on current pool size. It
1459 <     * controls construction of new threads.
758 <     * @throws IllegalArgumentException if negative or greater then
759 <     * internal implementation limit.
760 <     */
761 <    public void setMaximumPoolSize(int newMax) {
762 <        if (newMax < 0 || newMax > MAX_THREADS)
763 <            throw new IllegalArgumentException();
764 <        maxPoolSize = newMax;
765 <    }
766 <
767 <
768 <    /**
769 <     * Returns true if this pool dynamically maintains its target
770 <     * parallelism level. If false, new threads are added only to
771 <     * avoid possible starvation.
772 <     * This setting is by default true;
773 <     * @return true if maintains parallelism
774 <     */
775 <    public boolean getMaintainsParallelism() {
776 <        return maintainsParallelism;
777 <    }
778 <
779 <    /**
780 <     * Sets whether this pool dynamically maintains its target
781 <     * parallelism level. If false, new threads are added only to
782 <     * avoid possible starvation.
783 <     * @param enable true to maintains parallelism
1456 >     * Returns {@code true} if this pool uses local first-in-first-out
1457 >     * scheduling mode for forked tasks that are never joined.
1458 >     *
1459 >     * @return {@code true} if this pool uses async mode
1460       */
1461 <    public void setMaintainsParallelism(boolean enable) {
1462 <        maintainsParallelism = enable;
1461 >    public boolean getAsyncMode() {
1462 >        return locallyFifo;
1463      }
1464  
1465      /**
1466       * Returns an estimate of the number of worker threads that are
1467       * not blocked waiting to join tasks or for other managed
1468 <     * synchronization.
1468 >     * synchronization. This method may overestimate the
1469 >     * number of running threads.
1470       *
1471       * @return the number of worker threads
1472       */
1473      public int getRunningThreadCount() {
1474 <        return runningCountOf(workerCounts);
1474 >        return workerCounts & RUNNING_COUNT_MASK;
1475      }
1476  
1477      /**
1478       * Returns an estimate of the number of threads that are currently
1479       * stealing or executing tasks. This method may overestimate the
1480       * number of active threads.
1481 <     * @return the number of active threads.
1481 >     *
1482 >     * @return the number of active threads
1483       */
1484      public int getActiveThreadCount() {
1485 <        return activeCountOf(runControl);
808 <    }
809 <
810 <    /**
811 <     * Returns an estimate of the number of threads that are currently
812 <     * idle waiting for tasks. This method may underestimate the
813 <     * number of idle threads.
814 <     * @return the number of idle threads.
815 <     */
816 <    final int getIdleThreadCount() {
817 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
818 <        return (c <= 0)? 0 : c;
1485 >        return runState & ACTIVE_COUNT_MASK;
1486      }
1487  
1488      /**
1489 <     * Returns true if all worker threads are currently idle. An idle
1490 <     * worker is one that cannot obtain a task to execute because none
1491 <     * are available to steal from other threads, and there are no
1492 <     * pending submissions to the pool. This method is conservative:
1493 <     * It might not return true immediately upon idleness of all
1494 <     * threads, but will eventually become true if threads remain
1495 <     * inactive.
1496 <     * @return true if all threads are currently idle
1489 >     * Returns {@code true} if all worker threads are currently idle.
1490 >     * An idle worker is one that cannot obtain a task to execute
1491 >     * because none are available to steal from other threads, and
1492 >     * there are no pending submissions to the pool. This method is
1493 >     * conservative; it might not return {@code true} immediately upon
1494 >     * idleness of all threads, but will eventually become true if
1495 >     * threads remain inactive.
1496 >     *
1497 >     * @return {@code true} if all threads are currently idle
1498       */
1499      public boolean isQuiescent() {
1500 <        return activeCountOf(runControl) == 0;
1500 >        return (runState & ACTIVE_COUNT_MASK) == 0;
1501      }
1502  
1503      /**
# Line 837 | Line 1505 | public class ForkJoinPool extends Abstra
1505       * one thread's work queue by another. The reported value
1506       * underestimates the actual total number of steals when the pool
1507       * is not quiescent. This value may be useful for monitoring and
1508 <     * tuning fork/join programs: In general, steal counts should be
1508 >     * tuning fork/join programs: in general, steal counts should be
1509       * high enough to keep threads busy, but low enough to avoid
1510       * overhead and contention across threads.
1511 <     * @return the number of steals.
1511 >     *
1512 >     * @return the number of steals
1513       */
1514      public long getStealCount() {
1515 <        return stealCount.get();
847 <    }
848 <
849 <    /**
850 <     * Accumulate steal count from a worker. Call only
851 <     * when worker known to be idle.
852 <     */
853 <    private void updateStealCount(ForkJoinWorkerThread w) {
854 <        int sc = w.getAndClearStealCount();
855 <        if (sc != 0)
856 <            stealCount.addAndGet(sc);
1515 >        return stealCount;
1516      }
1517  
1518      /**
# Line 863 | Line 1522 | public class ForkJoinPool extends Abstra
1522       * an approximation, obtained by iterating across all threads in
1523       * the pool. This method may be useful for tuning task
1524       * granularities.
1525 <     * @return the number of queued tasks.
1525 >     *
1526 >     * @return the number of queued tasks
1527       */
1528      public long getQueuedTaskCount() {
1529          long count = 0;
1530          ForkJoinWorkerThread[] ws = workers;
1531 <        for (int i = 0; i < ws.length; ++i) {
1532 <            ForkJoinWorkerThread t = ws[i];
1533 <            if (t != null)
1534 <                count += t.getQueueSize();
1531 >        int n = ws.length;
1532 >        for (int i = 0; i < n; ++i) {
1533 >            ForkJoinWorkerThread w = ws[i];
1534 >            if (w != null)
1535 >                count += w.getQueueSize();
1536          }
1537          return count;
1538      }
1539  
1540      /**
1541 <     * Returns an estimate of the number tasks submitted to this pool
1542 <     * that have not yet begun executing. This method takes time
1541 >     * Returns an estimate of the number of tasks submitted to this
1542 >     * pool that have not yet begun executing.  This method takes time
1543       * proportional to the number of submissions.
1544 <     * @return the number of queued submissions.
1544 >     *
1545 >     * @return the number of queued submissions
1546       */
1547      public int getQueuedSubmissionCount() {
1548          return submissionQueue.size();
1549      }
1550  
1551      /**
1552 <     * Returns true if there are any tasks submitted to this pool
1553 <     * that have not yet begun executing.
1554 <     * @return <code>true</code> if there are any queued submissions.
1552 >     * Returns {@code true} if there are any tasks submitted to this
1553 >     * pool that have not yet begun executing.
1554 >     *
1555 >     * @return {@code true} if there are any queued submissions
1556       */
1557      public boolean hasQueuedSubmissions() {
1558          return !submissionQueue.isEmpty();
# Line 899 | Line 1562 | public class ForkJoinPool extends Abstra
1562       * Removes and returns the next unexecuted submission if one is
1563       * available.  This method may be useful in extensions to this
1564       * class that re-assign work in systems with multiple pools.
1565 <     * @return the next submission, or null if none
1565 >     *
1566 >     * @return the next submission, or {@code null} if none
1567       */
1568      protected ForkJoinTask<?> pollSubmission() {
1569          return submissionQueue.poll();
1570      }
1571  
1572      /**
1573 +     * Removes all available unexecuted submitted and forked tasks
1574 +     * from scheduling queues and adds them to the given collection,
1575 +     * without altering their execution status. These may include
1576 +     * artificially generated or wrapped tasks. This method is
1577 +     * designed to be invoked only when the pool is known to be
1578 +     * quiescent. Invocations at other times may not remove all
1579 +     * tasks. A failure encountered while attempting to add elements
1580 +     * to collection {@code c} may result in elements being in
1581 +     * neither, either or both collections when the associated
1582 +     * exception is thrown.  The behavior of this operation is
1583 +     * undefined if the specified collection is modified while the
1584 +     * operation is in progress.
1585 +     *
1586 +     * @param c the collection to transfer elements into
1587 +     * @return the number of elements transferred
1588 +     */
1589 +    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1590 +        int count = submissionQueue.drainTo(c);
1591 +        ForkJoinWorkerThread[] ws = workers;
1592 +        int n = ws.length;
1593 +        for (int i = 0; i < n; ++i) {
1594 +            ForkJoinWorkerThread w = ws[i];
1595 +            if (w != null)
1596 +                count += w.drainTasksTo(c);
1597 +        }
1598 +        return count;
1599 +    }
1600 +
1601 +    /**
1602       * Returns a string identifying this pool, as well as its state,
1603       * including indications of run state, parallelism level, and
1604       * worker and task counts.
# Line 913 | Line 1606 | public class ForkJoinPool extends Abstra
1606       * @return a string identifying this pool, as well as its state
1607       */
1608      public String toString() {
916        int ps = parallelism;
917        int wc = workerCounts;
918        int rc = runControl;
1609          long st = getStealCount();
1610          long qt = getQueuedTaskCount();
1611          long qs = getQueuedSubmissionCount();
1612 +        int wc = workerCounts;
1613 +        int tc = wc >>> TOTAL_COUNT_SHIFT;
1614 +        int rc = wc & RUNNING_COUNT_MASK;
1615 +        int pc = parallelism;
1616 +        int rs = runState;
1617 +        int ac = rs & ACTIVE_COUNT_MASK;
1618          return super.toString() +
1619 <            "[" + runStateToString(runStateOf(rc)) +
1620 <            ", parallelism = " + ps +
1621 <            ", size = " + totalCountOf(wc) +
1622 <            ", active = " + activeCountOf(rc) +
1623 <            ", running = " + runningCountOf(wc) +
1619 >            "[" + runLevelToString(rs) +
1620 >            ", parallelism = " + pc +
1621 >            ", size = " + tc +
1622 >            ", active = " + ac +
1623 >            ", running = " + rc +
1624              ", steals = " + st +
1625              ", tasks = " + qt +
1626              ", submissions = " + qs +
1627              "]";
1628      }
1629  
1630 <    private static String runStateToString(int rs) {
1631 <        switch(rs) {
1632 <        case RUNNING: return "Running";
1633 <        case SHUTDOWN: return "Shutting down";
1634 <        case TERMINATING: return "Terminating";
939 <        case TERMINATED: return "Terminated";
940 <        default: throw new Error("Unknown run state");
941 <        }
1630 >    private static String runLevelToString(int s) {
1631 >        return ((s & TERMINATED) != 0 ? "Terminated" :
1632 >                ((s & TERMINATING) != 0 ? "Terminating" :
1633 >                 ((s & SHUTDOWN) != 0 ? "Shutting down" :
1634 >                  "Running")));
1635      }
1636  
944    // lifecycle control
945
1637      /**
1638       * Initiates an orderly shutdown in which previously submitted
1639       * tasks are executed, but no new tasks will be accepted.
1640       * Invocation has no additional effect if already shut down.
1641       * Tasks that are in the process of being submitted concurrently
1642       * during the course of this method may or may not be rejected.
1643 +     *
1644       * @throws SecurityException if a security manager exists and
1645       *         the caller is not permitted to modify threads
1646       *         because it does not hold {@link
1647 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1647 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1648       */
1649      public void shutdown() {
1650          checkPermission();
1651 <        transitionRunStateTo(SHUTDOWN);
1652 <        if (canTerminateOnShutdown(runControl))
961 <            terminateOnShutdown();
1651 >        advanceRunLevel(SHUTDOWN);
1652 >        tryTerminate(false);
1653      }
1654  
1655      /**
1656 <     * Attempts to stop all actively executing tasks, and cancels all
1657 <     * waiting tasks.  Tasks that are in the process of being
1658 <     * submitted or executed concurrently during the course of this
1659 <     * method may or may not be rejected. Unlike some other executors,
1660 <     * this method cancels rather than collects non-executed tasks,
1661 <     * so always returns an empty list.
1656 >     * Attempts to cancel and/or stop all tasks, and reject all
1657 >     * subsequently submitted tasks.  Tasks that are in the process of
1658 >     * being submitted or executed concurrently during the course of
1659 >     * this method may or may not be rejected. This method cancels
1660 >     * both existing and unexecuted tasks, in order to permit
1661 >     * termination in the presence of task dependencies. So the method
1662 >     * always returns an empty list (unlike the case for some other
1663 >     * Executors).
1664 >     *
1665       * @return an empty list
1666       * @throws SecurityException if a security manager exists and
1667       *         the caller is not permitted to modify threads
1668       *         because it does not hold {@link
1669 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1669 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1670       */
1671      public List<Runnable> shutdownNow() {
1672          checkPermission();
1673 <        terminate();
1673 >        tryTerminate(true);
1674          return Collections.emptyList();
1675      }
1676  
1677      /**
1678 <     * Returns <code>true</code> if all tasks have completed following shut down.
1678 >     * Returns {@code true} if all tasks have completed following shut down.
1679       *
1680 <     * @return <code>true</code> if all tasks have completed following shut down
1680 >     * @return {@code true} if all tasks have completed following shut down
1681       */
1682      public boolean isTerminated() {
1683 <        return runStateOf(runControl) == TERMINATED;
1683 >        return runState >= TERMINATED;
1684      }
1685  
1686      /**
1687 <     * Returns <code>true</code> if the process of termination has
1688 <     * commenced but possibly not yet completed.
1687 >     * Returns {@code true} if the process of termination has
1688 >     * commenced but not yet completed.  This method may be useful for
1689 >     * debugging. A return of {@code true} reported a sufficient
1690 >     * period after shutdown may indicate that submitted tasks have
1691 >     * ignored or suppressed interruption, causing this executor not
1692 >     * to properly terminate.
1693       *
1694 <     * @return <code>true</code> if terminating
1694 >     * @return {@code true} if terminating but not yet terminated
1695       */
1696      public boolean isTerminating() {
1697 <        return runStateOf(runControl) >= TERMINATING;
1697 >        return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1698      }
1699  
1700      /**
1701 <     * Returns <code>true</code> if this pool has been shut down.
1701 >     * Returns {@code true} if this pool has been shut down.
1702       *
1703 <     * @return <code>true</code> if this pool has been shut down
1703 >     * @return {@code true} if this pool has been shut down
1704       */
1705      public boolean isShutdown() {
1706 <        return runStateOf(runControl) >= SHUTDOWN;
1706 >        return runState >= SHUTDOWN;
1707      }
1708  
1709      /**
# Line 1015 | Line 1713 | public class ForkJoinPool extends Abstra
1713       *
1714       * @param timeout the maximum time to wait
1715       * @param unit the time unit of the timeout argument
1716 <     * @return <code>true</code> if this executor terminated and
1717 <     *         <code>false</code> if the timeout elapsed before termination
1716 >     * @return {@code true} if this executor terminated and
1717 >     *         {@code false} if the timeout elapsed before termination
1718       * @throws InterruptedException if interrupted while waiting
1719       */
1720      public boolean awaitTermination(long timeout, TimeUnit unit)
1721          throws InterruptedException {
1024        long nanos = unit.toNanos(timeout);
1025        final ReentrantLock lock = this.workerLock;
1026        lock.lock();
1722          try {
1723 <            for (;;) {
1724 <                if (isTerminated())
1030 <                    return true;
1031 <                if (nanos <= 0)
1032 <                    return false;
1033 <                nanos = termination.awaitNanos(nanos);
1034 <            }
1035 <        } finally {
1036 <            lock.unlock();
1037 <        }
1038 <    }
1039 <
1040 <    // Shutdown and termination support
1041 <
1042 <    /**
1043 <     * Callback from terminating worker. Null out the corresponding
1044 <     * workers slot, and if terminating, try to terminate, else try to
1045 <     * shrink workers array.
1046 <     * @param w the worker
1047 <     */
1048 <    final void workerTerminated(ForkJoinWorkerThread w) {
1049 <        updateStealCount(w);
1050 <        updateWorkerCount(-1);
1051 <        final ReentrantLock lock = this.workerLock;
1052 <        lock.lock();
1053 <        try {
1054 <            ForkJoinWorkerThread[] ws = workers;
1055 <            int idx = w.poolIndex;
1056 <            if (idx >= 0 && idx < ws.length && ws[idx] == w)
1057 <                ws[idx] = null;
1058 <            if (totalCountOf(workerCounts) == 0) {
1059 <                terminate(); // no-op if already terminating
1060 <                transitionRunStateTo(TERMINATED);
1061 <                termination.signalAll();
1062 <            }
1063 <            else if (!isTerminating()) {
1064 <                tryShrinkWorkerArray();
1065 <                tryResumeSpare(true); // allow replacement
1066 <            }
1067 <        } finally {
1068 <            lock.unlock();
1069 <        }
1070 <        signalIdleWorkers(false);
1071 <    }
1072 <
1073 <    /**
1074 <     * Initiate termination.
1075 <     */
1076 <    private void terminate() {
1077 <        if (transitionRunStateTo(TERMINATING)) {
1078 <            stopAllWorkers();
1079 <            resumeAllSpares();
1080 <            signalIdleWorkers(true);
1081 <            cancelQueuedSubmissions();
1082 <            cancelQueuedWorkerTasks();
1083 <            interruptUnterminatedWorkers();
1084 <            signalIdleWorkers(true); // resignal after interrupt
1085 <        }
1086 <    }
1087 <
1088 <    /**
1089 <     * Possibly terminate when on shutdown state
1090 <     */
1091 <    private void terminateOnShutdown() {
1092 <        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1093 <            terminate();
1094 <    }
1095 <
1096 <    /**
1097 <     * Clear out and cancel submissions
1098 <     */
1099 <    private void cancelQueuedSubmissions() {
1100 <        ForkJoinTask<?> task;
1101 <        while ((task = pollSubmission()) != null)
1102 <            task.cancel(false);
1103 <    }
1104 <
1105 <    /**
1106 <     * Clean out worker queues.
1107 <     */
1108 <    private void cancelQueuedWorkerTasks() {
1109 <        final ReentrantLock lock = this.workerLock;
1110 <        lock.lock();
1111 <        try {
1112 <            ForkJoinWorkerThread[] ws = workers;
1113 <            for (int i = 0; i < ws.length; ++i) {
1114 <                ForkJoinWorkerThread t = ws[i];
1115 <                if (t != null)
1116 <                    t.cancelTasks();
1117 <            }
1118 <        } finally {
1119 <            lock.unlock();
1120 <        }
1121 <    }
1122 <
1123 <    /**
1124 <     * Set each worker's status to terminating. Requires lock to avoid
1125 <     * conflicts with add/remove
1126 <     */
1127 <    private void stopAllWorkers() {
1128 <        final ReentrantLock lock = this.workerLock;
1129 <        lock.lock();
1130 <        try {
1131 <            ForkJoinWorkerThread[] ws = workers;
1132 <            for (int i = 0; i < ws.length; ++i) {
1133 <                ForkJoinWorkerThread t = ws[i];
1134 <                if (t != null)
1135 <                    t.shutdownNow();
1136 <            }
1137 <        } finally {
1138 <            lock.unlock();
1139 <        }
1140 <    }
1141 <
1142 <    /**
1143 <     * Interrupt all unterminated workers.  This is not required for
1144 <     * sake of internal control, but may help unstick user code during
1145 <     * shutdown.
1146 <     */
1147 <    private void interruptUnterminatedWorkers() {
1148 <        final ReentrantLock lock = this.workerLock;
1149 <        lock.lock();
1150 <        try {
1151 <            ForkJoinWorkerThread[] ws = workers;
1152 <            for (int i = 0; i < ws.length; ++i) {
1153 <                ForkJoinWorkerThread t = ws[i];
1154 <                if (t != null && !t.isTerminated()) {
1155 <                    try {
1156 <                        t.interrupt();
1157 <                    } catch (SecurityException ignore) {
1158 <                    }
1159 <                }
1160 <            }
1161 <        } finally {
1162 <            lock.unlock();
1163 <        }
1164 <    }
1165 <
1166 <
1167 <    /*
1168 <     * Nodes for event barrier to manage idle threads.
1169 <     *
1170 <     * The event barrier has an event count and a wait queue (actually
1171 <     * a Treiber stack).  Workers are enabled to look for work when
1172 <     * the eventCount is incremented. If they fail to find some,
1173 <     * they may wait for next count. Synchronization events occur only
1174 <     * in enough contexts to maintain overall liveness:
1175 <     *
1176 <     *   - Submission of a new task to the pool
1177 <     *   - Creation or termination of a worker
1178 <     *   - pool termination
1179 <     *   - A worker pushing a task on an empty queue
1180 <     *
1181 <     * The last case (pushing a task) occurs often enough, and is
1182 <     * heavy enough compared to simple stack pushes to require some
1183 <     * special handling: Method signalNonEmptyWorkerQueue returns
1184 <     * without advancing count if the queue appears to be empty.  This
1185 <     * would ordinarily result in races causing some queued waiters
1186 <     * not to be woken up. To avoid this, a worker in sync
1187 <     * rescans for tasks after being enqueued if it was the first to
1188 <     * enqueue, and aborts the wait if finding one, also helping to
1189 <     * signal others. This works well because the worker has nothing
1190 <     * better to do anyway, and so might as well help alleviate the
1191 <     * overhead and contention on the threads actually doing work.
1192 <     *
1193 <     * Queue nodes are basic Treiber stack nodes, also used for spare
1194 <     * stack.
1195 <     */
1196 <    static final class WaitQueueNode {
1197 <        WaitQueueNode next; // only written before enqueued
1198 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1199 <        final long count; // unused for spare stack
1200 <        WaitQueueNode(ForkJoinWorkerThread w, long c) {
1201 <            count = c;
1202 <            thread = w;
1203 <        }
1204 <        final boolean signal() {
1205 <            ForkJoinWorkerThread t = thread;
1206 <            thread = null;
1207 <            if (t != null) {
1208 <                LockSupport.unpark(t);
1209 <                return true;
1210 <            }
1723 >            return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1724 >        } catch (TimeoutException ex) {
1725              return false;
1726          }
1727      }
1728  
1729      /**
1216     * Release at least one thread waiting for event count to advance,
1217     * if one exists. If initial attempt fails, release all threads.
1218     * @param all if false, at first try to only release one thread
1219     * @return current event
1220     */
1221    private long releaseIdleWorkers(boolean all) {
1222        long c;
1223        for (;;) {
1224            WaitQueueNode q = barrierStack;
1225            c = eventCount;
1226            long qc;
1227            if (q == null || (qc = q.count) >= c)
1228                break;
1229            if (!all) {
1230                if (casBarrierStack(q, q.next) && q.signal())
1231                    break;
1232                all = true;
1233            }
1234            else if (casBarrierStack(q, null)) {
1235                do {
1236                 q.signal();
1237                } while ((q = q.next) != null);
1238                break;
1239            }
1240        }
1241        return c;
1242    }
1243
1244    /**
1245     * Returns current barrier event count
1246     * @return current barrier event count
1247     */
1248    final long getEventCount() {
1249        long ec = eventCount;
1250        releaseIdleWorkers(true); // release to ensure accurate result
1251        return ec;
1252    }
1253
1254    /**
1255     * Increment event count and release at least one waiting thread,
1256     * if one exists (released threads will in turn wake up others).
1257     * @param all if true, try to wake up all
1258     */
1259    final void signalIdleWorkers(boolean all) {
1260        long c;
1261        do;while (!casEventCount(c = eventCount, c+1));
1262        releaseIdleWorkers(all);
1263    }
1264
1265    /**
1266     * Wake up threads waiting to steal a task. Because method
1267     * sync rechecks availability, it is OK to only proceed if
1268     * queue appears to be non-empty.
1269     */
1270    final void signalNonEmptyWorkerQueue() {
1271        // If CAS fails another signaller must have succeeded
1272        long c;
1273        if (barrierStack != null && casEventCount(c = eventCount, c+1))
1274            releaseIdleWorkers(false);
1275    }
1276
1277    /**
1278     * Waits until event count advances from count, or some thread is
1279     * waiting on a previous count, or there is stealable work
1280     * available. Help wake up others on release.
1281     * @param w the calling worker thread
1282     * @param prev previous value returned by sync (or 0)
1283     * @return current event count
1284     */
1285    final long sync(ForkJoinWorkerThread w, long prev) {
1286        updateStealCount(w);
1287
1288        while (!w.isShutdown() && !isTerminating() &&
1289               (parallelism >= runningCountOf(workerCounts) ||
1290                !suspendIfSpare(w))) { // prefer suspend to waiting here
1291            WaitQueueNode node = null;
1292            boolean queued = false;
1293            for (;;) {
1294                if (!queued) {
1295                    if (eventCount != prev)
1296                        break;
1297                    WaitQueueNode h = barrierStack;
1298                    if (h != null && h.count != prev)
1299                        break; // release below and maybe retry
1300                    if (node == null)
1301                        node = new WaitQueueNode(w, prev);
1302                    queued = casBarrierStack(node.next = h, node);
1303                }
1304                else if (Thread.interrupted() ||
1305                         node.thread == null ||
1306                         (node.next == null && w.prescan()) ||
1307                         eventCount != prev) {
1308                    node.thread = null;
1309                    if (eventCount == prev) // help trigger
1310                        casEventCount(prev, prev+1);
1311                    break;
1312                }
1313                else
1314                    LockSupport.park(this);
1315            }
1316            long ec = eventCount;
1317            if (releaseIdleWorkers(false) != prev)
1318                return ec;
1319        }
1320        return prev; // return old count if aborted
1321    }
1322
1323    //  Parallelism maintenance
1324
1325    /**
1326     * Decrement running count; if too low, add spare.
1327     *
1328     * Conceptually, all we need to do here is add or resume a
1329     * spare thread when one is about to block (and remove or
1330     * suspend it later when unblocked -- see suspendIfSpare).
1331     * However, implementing this idea requires coping with
1332     * several problems: We have imperfect information about the
1333     * states of threads. Some count updates can and usually do
1334     * lag run state changes, despite arrangements to keep them
1335     * accurate (for example, when possible, updating counts
1336     * before signalling or resuming), especially when running on
1337     * dynamic JVMs that don't optimize the infrequent paths that
1338     * update counts. Generating too many threads can make these
1339     * problems become worse, because excess threads are more
1340     * likely to be context-switched with others, slowing them all
1341     * down, especially if there is no work available, so all are
1342     * busy scanning or idling.  Also, excess spare threads can
1343     * only be suspended or removed when they are idle, not
1344     * immediately when they aren't needed. So adding threads will
1345     * raise parallelism level for longer than necessary.  Also,
1346     * FJ applications often enounter highly transient peaks when
1347     * many threads are blocked joining, but for less time than it
1348     * takes to create or resume spares.
1349     *
1350     * @param joinMe if non-null, return early if done
1351     * @param maintainParallelism if true, try to stay within
1352     * target counts, else create only to avoid starvation
1353     * @return true if joinMe known to be done
1354     */
1355    final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
1356        maintainParallelism &= maintainsParallelism; // overrride
1357        boolean dec = false;  // true when running count decremented
1358        while (spareStack == null || !tryResumeSpare(dec)) {
1359            int counts = workerCounts;
1360            if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
1361                if (!needSpare(counts, maintainParallelism))
1362                    break;
1363                if (joinMe.status < 0)
1364                    return true;
1365                if (tryAddSpare(counts))
1366                    break;
1367            }
1368        }
1369        return false;
1370    }
1371
1372    /**
1373     * Same idea as preJoin
1374     */
1375    final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
1376        maintainParallelism &= maintainsParallelism;
1377        boolean dec = false;
1378        while (spareStack == null || !tryResumeSpare(dec)) {
1379            int counts = workerCounts;
1380            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1381                if (!needSpare(counts, maintainParallelism))
1382                    break;
1383                if (blocker.isReleasable())
1384                    return true;
1385                if (tryAddSpare(counts))
1386                    break;
1387            }
1388        }
1389        return false;
1390    }
1391
1392    /**
1393     * Returns true if a spare thread appears to be needed.  If
1394     * maintaining parallelism, returns true when the deficit in
1395     * running threads is more than the surplus of total threads, and
1396     * there is apparently some work to do.  This self-limiting rule
1397     * means that the more threads that have already been added, the
1398     * less parallelism we will tolerate before adding another.
1399     * @param counts current worker counts
1400     * @param maintainParallelism try to maintain parallelism
1401     */
1402    private boolean needSpare(int counts, boolean maintainParallelism) {
1403        int ps = parallelism;
1404        int rc = runningCountOf(counts);
1405        int tc = totalCountOf(counts);
1406        int runningDeficit = ps - rc;
1407        int totalSurplus = tc - ps;
1408        return (tc < maxPoolSize &&
1409                (rc == 0 || totalSurplus < 0 ||
1410                 (maintainParallelism &&
1411                  runningDeficit > totalSurplus && mayHaveQueuedWork())));
1412    }
1413
1414    /**
1415     * Returns true if at least one worker queue appears to be
1416     * nonempty. This is expensive but not often called. It is not
1417     * critical that this be accurate, but if not, more or fewer
1418     * running threads than desired might be maintained.
1419     */
1420    private boolean mayHaveQueuedWork() {
1421        ForkJoinWorkerThread[] ws = workers;
1422        int len = ws.length;
1423        ForkJoinWorkerThread v;
1424        for (int i = 0; i < len; ++i) {
1425            if ((v = ws[i]) != null && v.getRawQueueSize() > 0) {
1426                releaseIdleWorkers(false); // help wake up stragglers
1427                return true;
1428            }
1429        }
1430        return false;
1431    }
1432
1433    /**
1434     * Add a spare worker if lock available and no more than the
1435     * expected numbers of threads exist
1436     * @return true if successful
1437     */
1438    private boolean tryAddSpare(int expectedCounts) {
1439        final ReentrantLock lock = this.workerLock;
1440        int expectedRunning = runningCountOf(expectedCounts);
1441        int expectedTotal = totalCountOf(expectedCounts);
1442        boolean success = false;
1443        boolean locked = false;
1444        // confirm counts while locking; CAS after obtaining lock
1445        try {
1446            for (;;) {
1447                int s = workerCounts;
1448                int tc = totalCountOf(s);
1449                int rc = runningCountOf(s);
1450                if (rc > expectedRunning || tc > expectedTotal)
1451                    break;
1452                if (!locked && !(locked = lock.tryLock()))
1453                    break;
1454                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1455                    createAndStartSpare(tc);
1456                    success = true;
1457                    break;
1458                }
1459            }
1460        } finally {
1461            if (locked)
1462                lock.unlock();
1463        }
1464        return success;
1465    }
1466
1467    /**
1468     * Add the kth spare worker. On entry, pool coounts are already
1469     * adjusted to reflect addition.
1470     */
1471    private void createAndStartSpare(int k) {
1472        ForkJoinWorkerThread w = null;
1473        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1474        int len = ws.length;
1475        // Probably, we can place at slot k. If not, find empty slot
1476        if (k < len && ws[k] != null) {
1477            for (k = 0; k < len && ws[k] != null; ++k)
1478                ;
1479        }
1480        if (k < len && (w = createWorker(k)) != null) {
1481            ws[k] = w;
1482            w.start();
1483        }
1484        else
1485            updateWorkerCount(-1); // adjust on failure
1486        signalIdleWorkers(false);
1487    }
1488
1489    /**
1490     * Suspend calling thread w if there are excess threads.  Called
1491     * only from sync.  Spares are enqueued in a Treiber stack
1492     * using the same WaitQueueNodes as barriers.  They are resumed
1493     * mainly in preJoin, but are also woken on pool events that
1494     * require all threads to check run state.
1495     * @param w the caller
1496     */
1497    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1498        WaitQueueNode node = null;
1499        int s;
1500        while (parallelism < runningCountOf(s = workerCounts)) {
1501            if (node == null)
1502                node = new WaitQueueNode(w, 0);
1503            if (casWorkerCounts(s, s-1)) { // representation-dependent
1504                // push onto stack
1505                do;while (!casSpareStack(node.next = spareStack, node));
1506
1507                // block until released by resumeSpare
1508                while (node.thread != null) {
1509                    if (!Thread.interrupted())
1510                        LockSupport.park(this);
1511                }
1512                w.activate(); // help warm up
1513                return true;
1514            }
1515        }
1516        return false;
1517    }
1518
1519    /**
1520     * Try to pop and resume a spare thread.
1521     * @param updateCount if true, increment running count on success
1522     * @return true if successful
1523     */
1524    private boolean tryResumeSpare(boolean updateCount) {
1525        WaitQueueNode q;
1526        while ((q = spareStack) != null) {
1527            if (casSpareStack(q, q.next)) {
1528                if (updateCount)
1529                    updateRunningCount(1);
1530                q.signal();
1531                return true;
1532            }
1533        }
1534        return false;
1535    }
1536
1537    /**
1538     * Pop and resume all spare threads. Same idea as
1539     * releaseIdleWorkers.
1540     * @return true if any spares released
1541     */
1542    private boolean resumeAllSpares() {
1543        WaitQueueNode q;
1544        while ( (q = spareStack) != null) {
1545            if (casSpareStack(q, null)) {
1546                do {
1547                    updateRunningCount(1);
1548                    q.signal();
1549                } while ((q = q.next) != null);
1550                return true;
1551            }
1552        }
1553        return false;
1554    }
1555
1556    /**
1557     * Pop and shutdown excessive spare threads. Call only while
1558     * holding lock. This is not guaranteed to eliminate all excess
1559     * threads, only those suspended as spares, which are the ones
1560     * unlikely to be needed in the future.
1561     */
1562    private void trimSpares() {
1563        int surplus = totalCountOf(workerCounts) - parallelism;
1564        WaitQueueNode q;
1565        while (surplus > 0 && (q = spareStack) != null) {
1566            if (casSpareStack(q, null)) {
1567                do {
1568                    updateRunningCount(1);
1569                    ForkJoinWorkerThread w = q.thread;
1570                    if (w != null && surplus > 0 &&
1571                        runningCountOf(workerCounts) > 0 && w.shutdown())
1572                        --surplus;
1573                    q.signal();
1574                } while ((q = q.next) != null);
1575            }
1576        }
1577    }
1578
1579    /**
1580     * Returns approximate number of spares, just for diagnostics.
1581     */
1582    private int countSpares() {
1583        int sum = 0;
1584        for (WaitQueueNode q = spareStack; q != null; q = q.next)
1585            ++sum;
1586        return sum;
1587    }
1588
1589    /**
1730       * Interface for extending managed parallelism for tasks running
1731 <     * in ForkJoinPools. A ManagedBlocker provides two methods.
1732 <     * Method <code>isReleasable</code> must return true if blocking is not
1733 <     * necessary. Method <code>block</code> blocks the current thread
1734 <     * if necessary (perhaps internally invoking isReleasable before
1735 <     * actually blocking.).
1731 >     * in {@link ForkJoinPool}s.
1732 >     *
1733 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1734 >     * {@code isReleasable} must return {@code true} if blocking is
1735 >     * not necessary. Method {@code block} blocks the current thread
1736 >     * if necessary (perhaps internally invoking {@code isReleasable}
1737 >     * before actually blocking). The unusual methods in this API
1738 >     * accommodate synchronizers that may, but don't usually, block
1739 >     * for long periods. Similarly, they allow more efficient internal
1740 >     * handling of cases in which additional workers may be, but
1741 >     * usually are not, needed to ensure sufficient parallelism.
1742 >     * Toward this end, implementations of method {@code isReleasable}
1743 >     * must be amenable to repeated invocation.
1744 >     *
1745       * <p>For example, here is a ManagedBlocker based on a
1746       * ReentrantLock:
1747 <     * <pre>
1748 <     *   class ManagedLocker implements ManagedBlocker {
1749 <     *     final ReentrantLock lock;
1750 <     *     boolean hasLock = false;
1751 <     *     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
1752 <     *     public boolean block() {
1753 <     *        if (!hasLock)
1754 <     *           lock.lock();
1755 <     *        return true;
1756 <     *     }
1757 <     *     public boolean isReleasable() {
1758 <     *        return hasLock || (hasLock = lock.tryLock());
1759 <     *     }
1747 >     *  <pre> {@code
1748 >     * class ManagedLocker implements ManagedBlocker {
1749 >     *   final ReentrantLock lock;
1750 >     *   boolean hasLock = false;
1751 >     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
1752 >     *   public boolean block() {
1753 >     *     if (!hasLock)
1754 >     *       lock.lock();
1755 >     *     return true;
1756 >     *   }
1757 >     *   public boolean isReleasable() {
1758 >     *     return hasLock || (hasLock = lock.tryLock());
1759 >     *   }
1760 >     * }}</pre>
1761 >     *
1762 >     * <p>Here is a class that possibly blocks waiting for an
1763 >     * item on a given queue:
1764 >     *  <pre> {@code
1765 >     * class QueueTaker<E> implements ManagedBlocker {
1766 >     *   final BlockingQueue<E> queue;
1767 >     *   volatile E item = null;
1768 >     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1769 >     *   public boolean block() throws InterruptedException {
1770 >     *     if (item == null)
1771 >     *       item = queue.take();
1772 >     *     return true;
1773 >     *   }
1774 >     *   public boolean isReleasable() {
1775 >     *     return item != null || (item = queue.poll()) != null;
1776       *   }
1777 <     * </pre>
1777 >     *   public E getItem() { // call after pool.managedBlock completes
1778 >     *     return item;
1779 >     *   }
1780 >     * }}</pre>
1781       */
1782      public static interface ManagedBlocker {
1783          /**
1784           * Possibly blocks the current thread, for example waiting for
1785           * a lock or condition.
1786 <         * @return true if no additional blocking is necessary (i.e.,
1787 <         * if isReleasable would return true).
1786 >         *
1787 >         * @return {@code true} if no additional blocking is necessary
1788 >         * (i.e., if isReleasable would return true)
1789           * @throws InterruptedException if interrupted while waiting
1790 <         * (the method is not required to do so, but is allowe to).
1790 >         * (the method is not required to do so, but is allowed to)
1791           */
1792          boolean block() throws InterruptedException;
1793  
1794          /**
1795 <         * Returns true if blocking is unnecessary.
1795 >         * Returns {@code true} if blocking is unnecessary.
1796           */
1797          boolean isReleasable();
1798      }
1799  
1800      /**
1801       * Blocks in accord with the given blocker.  If the current thread
1802 <     * is a ForkJoinWorkerThread, this method possibly arranges for a
1803 <     * spare thread to be activated if necessary to ensure parallelism
1804 <     * while the current thread is blocked.  If
1805 <     * <code>maintainParallelism</code> is true and the pool supports
1806 <     * it ({@link #getMaintainsParallelism}), this method attempts to
1807 <     * maintain the pool's nominal parallelism. Otherwise if activates
1808 <     * a thread only if necessary to avoid complete starvation. This
1809 <     * option may be preferable when blockages use timeouts, or are
1810 <     * almost always brief.
1811 <     *
1812 <     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
1813 <     * equivalent to
1814 <     * <pre>
1815 <     *   while (!blocker.isReleasable())
1647 <     *      if (blocker.block())
1648 <     *         return;
1649 <     * </pre>
1650 <     * If the caller is a ForkJoinTask, then the pool may first
1651 <     * be expanded to ensure parallelism, and later adjusted.
1802 >     * is a {@link ForkJoinWorkerThread}, this method possibly
1803 >     * arranges for a spare thread to be activated if necessary to
1804 >     * ensure sufficient parallelism while the current thread is blocked.
1805 >     *
1806 >     * <p>If the caller is not a {@link ForkJoinTask}, this method is
1807 >     * behaviorally equivalent to
1808 >     *  <pre> {@code
1809 >     * while (!blocker.isReleasable())
1810 >     *   if (blocker.block())
1811 >     *     return;
1812 >     * }</pre>
1813 >     *
1814 >     * If the caller is a {@code ForkJoinTask}, then the pool may
1815 >     * first be expanded to ensure parallelism, and later adjusted.
1816       *
1817       * @param blocker the blocker
1818 <     * @param maintainParallelism if true and supported by this pool,
1655 <     * attempt to maintain the pool's nominal parallelism; otherwise
1656 <     * activate a thread only if necessary to avoid complete
1657 <     * starvation.
1658 <     * @throws InterruptedException if blocker.block did so.
1818 >     * @throws InterruptedException if blocker.block did so
1819       */
1820 <    public static void managedBlock(ManagedBlocker blocker,
1661 <                                    boolean maintainParallelism)
1820 >    public static void managedBlock(ManagedBlocker blocker)
1821          throws InterruptedException {
1822          Thread t = Thread.currentThread();
1823 <        ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
1824 <                             ((ForkJoinWorkerThread)t).pool : null);
1825 <        if (!blocker.isReleasable()) {
1826 <            try {
1827 <                if (pool == null ||
1828 <                    !pool.preBlock(blocker, maintainParallelism))
1670 <                    awaitBlocker(blocker);
1671 <            } finally {
1672 <                if (pool != null)
1673 <                    pool.updateRunningCount(1);
1674 <            }
1823 >        if (t instanceof ForkJoinWorkerThread) {
1824 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1825 >            w.pool.awaitBlocker(blocker);
1826 >        }
1827 >        else {
1828 >            do {} while (!blocker.isReleasable() && !blocker.block());
1829          }
1830      }
1831  
1832 <    private static void awaitBlocker(ManagedBlocker blocker)
1833 <        throws InterruptedException {
1834 <        do;while (!blocker.isReleasable() && !blocker.block());
1681 <    }
1682 <
1683 <    // AbstractExecutorService overrides
1832 >    // AbstractExecutorService overrides.  These rely on undocumented
1833 >    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
1834 >    // implement RunnableFuture.
1835  
1836      protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
1837 <        return new AdaptedRunnable(runnable, value);
1837 >        return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
1838      }
1839  
1840      protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
1841 <        return new AdaptedCallable(callable);
1841 >        return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
1842      }
1843  
1844 +    // Unsafe mechanics
1845  
1846 <    // Temporary Unsafe mechanics for preliminary release
1847 <
1848 <    static final Unsafe _unsafe;
1849 <    static final long eventCountOffset;
1850 <    static final long workerCountsOffset;
1851 <    static final long runControlOffset;
1852 <    static final long barrierStackOffset;
1853 <    static final long spareStackOffset;
1846 >    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1847 >    private static final long workerCountsOffset =
1848 >        objectFieldOffset("workerCounts", ForkJoinPool.class);
1849 >    private static final long runStateOffset =
1850 >        objectFieldOffset("runState", ForkJoinPool.class);
1851 >    private static final long eventCountOffset =
1852 >        objectFieldOffset("eventCount", ForkJoinPool.class);
1853 >    private static final long eventWaitersOffset =
1854 >        objectFieldOffset("eventWaiters",ForkJoinPool.class);
1855 >    private static final long stealCountOffset =
1856 >        objectFieldOffset("stealCount",ForkJoinPool.class);
1857 >    private static final long spareWaitersOffset =
1858 >        objectFieldOffset("spareWaiters",ForkJoinPool.class);
1859  
1860 <    static {
1860 >    private static long objectFieldOffset(String field, Class<?> klazz) {
1861          try {
1862 <            if (ForkJoinPool.class.getClassLoader() != null) {
1863 <                Field f = Unsafe.class.getDeclaredField("theUnsafe");
1864 <                f.setAccessible(true);
1865 <                _unsafe = (Unsafe)f.get(null);
1866 <            }
1867 <            else
1711 <                _unsafe = Unsafe.getUnsafe();
1712 <            eventCountOffset = _unsafe.objectFieldOffset
1713 <                (ForkJoinPool.class.getDeclaredField("eventCount"));
1714 <            workerCountsOffset = _unsafe.objectFieldOffset
1715 <                (ForkJoinPool.class.getDeclaredField("workerCounts"));
1716 <            runControlOffset = _unsafe.objectFieldOffset
1717 <                (ForkJoinPool.class.getDeclaredField("runControl"));
1718 <            barrierStackOffset = _unsafe.objectFieldOffset
1719 <                (ForkJoinPool.class.getDeclaredField("barrierStack"));
1720 <            spareStackOffset = _unsafe.objectFieldOffset
1721 <                (ForkJoinPool.class.getDeclaredField("spareStack"));
1722 <        } catch (Exception e) {
1723 <            throw new RuntimeException("Could not initialize intrinsics", e);
1862 >            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1863 >        } catch (NoSuchFieldException e) {
1864 >            // Convert Exception to corresponding Error
1865 >            NoSuchFieldError error = new NoSuchFieldError(field);
1866 >            error.initCause(e);
1867 >            throw error;
1868          }
1869      }
1870  
1871 <    private boolean casEventCount(long cmp, long val) {
1872 <        return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
1873 <    }
1874 <    private boolean casWorkerCounts(int cmp, int val) {
1875 <        return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1876 <    }
1877 <    private boolean casRunControl(int cmp, int val) {
1878 <        return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
1879 <    }
1880 <    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1881 <        return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
1882 <    }
1883 <    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1884 <        return _unsafe.compareAndSwapObject(this, barrierStackOffset, cmp, val);
1871 >    /**
1872 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1873 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
1874 >     * into a jdk.
1875 >     *
1876 >     * @return a sun.misc.Unsafe
1877 >     */
1878 >    private static sun.misc.Unsafe getUnsafe() {
1879 >        try {
1880 >            return sun.misc.Unsafe.getUnsafe();
1881 >        } catch (SecurityException se) {
1882 >            try {
1883 >                return java.security.AccessController.doPrivileged
1884 >                    (new java.security
1885 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1886 >                        public sun.misc.Unsafe run() throws Exception {
1887 >                            java.lang.reflect.Field f = sun.misc
1888 >                                .Unsafe.class.getDeclaredField("theUnsafe");
1889 >                            f.setAccessible(true);
1890 >                            return (sun.misc.Unsafe) f.get(null);
1891 >                        }});
1892 >            } catch (java.security.PrivilegedActionException e) {
1893 >                throw new RuntimeException("Could not initialize intrinsics",
1894 >                                           e.getCause());
1895 >            }
1896 >        }
1897      }
1898   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines