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 1.4 by dl, Mon Jan 12 17:16:18 2009 UTC vs.
Revision 1.137 by dl, Tue Oct 30 14:23:11 2012 UTC

# Line 1 | Line 1
1   /*
2   * Written by Doug Lea with assistance from members of JCP JSR-166
3   * Expert Group and released to the public domain, as explained at
4 < * http://creativecommons.org/licenses/publicdomain
4 > * http://creativecommons.org/publicdomain/zero/1.0/
5   */
6  
7   package jsr166y;
8 < import java.util.*;
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.*;
8 >
9 > import java.util.ArrayList;
10 > import java.util.Arrays;
11 > import java.util.Collection;
12 > import java.util.Collections;
13 > import java.util.List;
14 > import java.util.Random;
15 > import java.util.concurrent.AbstractExecutorService;
16 > import java.util.concurrent.Callable;
17 > import java.util.concurrent.ExecutorService;
18 > import java.util.concurrent.Future;
19 > import java.util.concurrent.RejectedExecutionException;
20 > import java.util.concurrent.RunnableFuture;
21 > import java.util.concurrent.TimeUnit;
22 > import java.util.concurrent.atomic.AtomicInteger;
23 > import java.util.concurrent.atomic.AtomicLong;
24 > import java.util.concurrent.locks.AbstractQueuedSynchronizer;
25 > import java.util.concurrent.locks.Condition;
26  
27   /**
28 < * An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
29 < * ForkJoinPool provides the entry point for submissions from
30 < * non-ForkJoinTasks, as well as management and monitoring operations.
31 < * Normally a single ForkJoinPool is used for a large number of
32 < * submitted tasks. Otherwise, use would not usually outweigh the
33 < * construction and bookkeeping overhead of creating a large set of
34 < * threads.
28 > * An {@link ExecutorService} for running {@link ForkJoinTask}s.
29 > * A {@code ForkJoinPool} provides the entry point for submissions
30 > * from non-{@code ForkJoinTask} clients, as well as management and
31 > * monitoring operations.
32 > *
33 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
34 > * ExecutorService} mainly by virtue of employing
35 > * <em>work-stealing</em>: all threads in the pool attempt to find and
36 > * execute tasks submitted to the pool and/or created by other active
37 > * tasks (eventually blocking waiting for work if none exist). This
38 > * enables efficient processing when most tasks spawn other subtasks
39 > * (as do most {@code ForkJoinTask}s), as well as when many small
40 > * tasks are submitted to the pool from external clients.  Especially
41 > * when setting <em>asyncMode</em> to true in constructors, {@code
42 > * ForkJoinPool}s may also be appropriate for use with event-style
43 > * tasks that are never joined.
44   *
45 < * <p>ForkJoinPools differ from other kinds of Executors mainly in
46 < * that they provide <em>work-stealing</em>: all threads in the pool
47 < * attempt to find and execute subtasks created by other active tasks
48 < * (eventually blocking if none exist). This makes them efficient when
49 < * most tasks spawn other subtasks (as do most ForkJoinTasks), as well
50 < * as the mixed execution of some plain Runnable- or Callable- based
51 < * activities along with ForkJoinTasks. Otherwise, other
52 < * ExecutorService implementations are typically more appropriate
53 < * choices.
45 > * <p>A static {@link #commonPool} is available and appropriate for
46 > * most applications. The common pool is used by any ForkJoinTask that
47 > * is not explicitly submitted to a specified pool. Using the common
48 > * pool normally reduces resource usage (its threads are slowly
49 > * reclaimed during periods of non-use, and reinstated upon subsequent
50 > * use).  The common pool is by default constructed with default
51 > * parameters, but these may be controlled by setting any or all of
52 > * the three properties {@code
53 > * java.util.concurrent.ForkJoinPool.common.{parallelism,
54 > * threadFactory, exceptionHandler}}.
55   *
56 < * <p>A ForkJoinPool may be constructed with a given parallelism level
57 < * (target pool size), which it attempts to maintain by dynamically
58 < * adding, suspending, or resuming threads, even if some tasks are
59 < * waiting to join others. However, no such adjustments are performed
60 < * in the face of blocked IO or other unmanaged synchronization. The
61 < * nested <code>ManagedBlocker</code> interface enables extension of
62 < * the kinds of synchronization accommodated.  The target parallelism
63 < * level may also be changed dynamically (<code>setParallelism</code>)
64 < * and dynamically thread construction can be limited using methods
65 < * <code>setMaximumPoolSize</code> and/or
44 < * <code>setMaintainsParallelism</code>.
56 > * <p>For applications that require separate or custom pools, a {@code
57 > * ForkJoinPool} may be constructed with a given target parallelism
58 > * level; by default, equal to the number of available processors. The
59 > * pool attempts to maintain enough active (or available) threads by
60 > * dynamically adding, suspending, or resuming internal worker
61 > * threads, even if some tasks are stalled waiting to join
62 > * others. However, no such adjustments are guaranteed in the face of
63 > * blocked IO or other unmanaged synchronization. The nested {@link
64 > * ManagedBlocker} interface enables extension of the kinds of
65 > * synchronization accommodated.
66   *
67   * <p>In addition to execution and lifecycle control methods, this
68   * class provides status check methods (for example
69 < * <code>getStealCount</code>) that are intended to aid in developing,
69 > * {@link #getStealCount}) that are intended to aid in developing,
70   * tuning, and monitoring fork/join applications. Also, method
71 < * <code>toString</code> returns indications of pool state in a
71 > * {@link #toString} returns indications of pool state in a
72   * convenient form for informal monitoring.
73   *
74 + * <p> As is the case with other ExecutorServices, there are three
75 + * main task execution methods summarized in the following table.
76 + * These are designed to be used primarily by clients not already
77 + * engaged in fork/join computations in the current pool.  The main
78 + * forms of these methods accept instances of {@code ForkJoinTask},
79 + * but overloaded forms also allow mixed execution of plain {@code
80 + * Runnable}- or {@code Callable}- based activities as well.  However,
81 + * tasks that are already executing in a pool should normally instead
82 + * use the within-computation forms listed in the table unless using
83 + * async event-style tasks that are not usually joined, in which case
84 + * there is little difference among choice of methods.
85 + *
86 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
87 + *  <tr>
88 + *    <td></td>
89 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
90 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
91 + *  </tr>
92 + *  <tr>
93 + *    <td> <b>Arrange async execution</td>
94 + *    <td> {@link #execute(ForkJoinTask)}</td>
95 + *    <td> {@link ForkJoinTask#fork}</td>
96 + *  </tr>
97 + *  <tr>
98 + *    <td> <b>Await and obtain result</td>
99 + *    <td> {@link #invoke(ForkJoinTask)}</td>
100 + *    <td> {@link ForkJoinTask#invoke}</td>
101 + *  </tr>
102 + *  <tr>
103 + *    <td> <b>Arrange exec and obtain Future</td>
104 + *    <td> {@link #submit(ForkJoinTask)}</td>
105 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
106 + *  </tr>
107 + * </table>
108 + *
109   * <p><b>Implementation notes</b>: This implementation restricts the
110   * maximum number of running threads to 32767. Attempts to create
111 < * pools with greater than the maximum result in
112 < * IllegalArgumentExceptions.
111 > * pools with greater than the maximum number result in
112 > * {@code IllegalArgumentException}.
113 > *
114 > * <p>This implementation rejects submitted tasks (that is, by throwing
115 > * {@link RejectedExecutionException}) only when the pool is shut down
116 > * or internal resources have been exhausted.
117 > *
118 > * @since 1.7
119 > * @author Doug Lea
120   */
121   public class ForkJoinPool extends AbstractExecutorService {
122  
123      /*
124 <     * See the extended comments interspersed below for design,
125 <     * rationale, and walkthroughs.
124 >     * Implementation Overview
125 >     *
126 >     * This class and its nested classes provide the main
127 >     * functionality and control for a set of worker threads:
128 >     * Submissions from non-FJ threads enter into submission queues.
129 >     * Workers take these tasks and typically split them into subtasks
130 >     * that may be stolen by other workers.  Preference rules give
131 >     * first priority to processing tasks from their own queues (LIFO
132 >     * or FIFO, depending on mode), then to randomized FIFO steals of
133 >     * tasks in other queues.
134 >     *
135 >     * WorkQueues
136 >     * ==========
137 >     *
138 >     * Most operations occur within work-stealing queues (in nested
139 >     * class WorkQueue).  These are special forms of Deques that
140 >     * support only three of the four possible end-operations -- push,
141 >     * pop, and poll (aka steal), under the further constraints that
142 >     * push and pop are called only from the owning thread (or, as
143 >     * extended here, under a lock), while poll may be called from
144 >     * other threads.  (If you are unfamiliar with them, you probably
145 >     * want to read Herlihy and Shavit's book "The Art of
146 >     * Multiprocessor programming", chapter 16 describing these in
147 >     * more detail before proceeding.)  The main work-stealing queue
148 >     * design is roughly similar to those in the papers "Dynamic
149 >     * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
150 >     * (http://research.sun.com/scalable/pubs/index.html) and
151 >     * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
152 >     * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
153 >     * The main differences ultimately stem from GC requirements that
154 >     * we null out taken slots as soon as we can, to maintain as small
155 >     * a footprint as possible even in programs generating huge
156 >     * numbers of tasks. To accomplish this, we shift the CAS
157 >     * arbitrating pop vs poll (steal) from being on the indices
158 >     * ("base" and "top") to the slots themselves.  So, both a
159 >     * successful pop and poll mainly entail a CAS of a slot from
160 >     * non-null to null.  Because we rely on CASes of references, we
161 >     * do not need tag bits on base or top.  They are simple ints as
162 >     * used in any circular array-based queue (see for example
163 >     * ArrayDeque).  Updates to the indices must still be ordered in a
164 >     * way that guarantees that top == base means the queue is empty,
165 >     * but otherwise may err on the side of possibly making the queue
166 >     * appear nonempty when a push, pop, or poll have not fully
167 >     * committed. Note that this means that the poll operation,
168 >     * considered individually, is not wait-free. One thief cannot
169 >     * successfully continue until another in-progress one (or, if
170 >     * previously empty, a push) completes.  However, in the
171 >     * aggregate, we ensure at least probabilistic non-blockingness.
172 >     * If an attempted steal fails, a thief always chooses a different
173 >     * random victim target to try next. So, in order for one thief to
174 >     * progress, it suffices for any in-progress poll or new push on
175 >     * any empty queue to complete. (This is why we normally use
176 >     * method pollAt and its variants that try once at the apparent
177 >     * base index, else consider alternative actions, rather than
178 >     * method poll.)
179 >     *
180 >     * This approach also enables support of a user mode in which local
181 >     * task processing is in FIFO, not LIFO order, simply by using
182 >     * poll rather than pop.  This can be useful in message-passing
183 >     * frameworks in which tasks are never joined.  However neither
184 >     * mode considers affinities, loads, cache localities, etc, so
185 >     * rarely provide the best possible performance on a given
186 >     * machine, but portably provide good throughput by averaging over
187 >     * these factors.  (Further, even if we did try to use such
188 >     * information, we do not usually have a basis for exploiting it.
189 >     * For example, some sets of tasks profit from cache affinities,
190 >     * but others are harmed by cache pollution effects.)
191 >     *
192 >     * WorkQueues are also used in a similar way for tasks submitted
193 >     * to the pool. We cannot mix these tasks in the same queues used
194 >     * for work-stealing (this would contaminate lifo/fifo
195 >     * processing). Instead, we loosely associate submission queues
196 >     * with submitting threads, using a form of hashing.  The
197 >     * ThreadLocal Submitter class contains a value initially used as
198 >     * a hash code for choosing existing queues, but may be randomly
199 >     * repositioned upon contention with other submitters.  In
200 >     * essence, submitters act like workers except that they never
201 >     * take tasks, and they are multiplexed on to a finite number of
202 >     * shared work queues. However, classes are set up so that future
203 >     * extensions could allow submitters to optionally help perform
204 >     * tasks as well. Insertion of tasks in shared mode requires a
205 >     * lock (mainly to protect in the case of resizing) but we use
206 >     * only a simple spinlock (using bits in field runState), because
207 >     * submitters encountering a busy queue move on to try or create
208 >     * other queues -- they block only when creating and registering
209 >     * new queues.
210 >     *
211 >     * Management
212 >     * ==========
213 >     *
214 >     * The main throughput advantages of work-stealing stem from
215 >     * decentralized control -- workers mostly take tasks from
216 >     * themselves or each other. We cannot negate this in the
217 >     * implementation of other management responsibilities. The main
218 >     * tactic for avoiding bottlenecks is packing nearly all
219 >     * essentially atomic control state into two volatile variables
220 >     * that are by far most often read (not written) as status and
221 >     * consistency checks.
222 >     *
223 >     * Field "ctl" contains 64 bits holding all the information needed
224 >     * to atomically decide to add, inactivate, enqueue (on an event
225 >     * queue), dequeue, and/or re-activate workers.  To enable this
226 >     * packing, we restrict maximum parallelism to (1<<15)-1 (which is
227 >     * far in excess of normal operating range) to allow ids, counts,
228 >     * and their negations (used for thresholding) to fit into 16bit
229 >     * fields.
230 >     *
231 >     * Field "runState" contains 32 bits needed to register and
232 >     * deregister WorkQueues, as well as to enable shutdown. It is
233 >     * only modified under a lock (normally briefly held, but
234 >     * occasionally protecting allocations and resizings) but even
235 >     * when locked remains available to check consistency.
236 >     *
237 >     * Recording WorkQueues.  WorkQueues are recorded in the
238 >     * "workQueues" array that is created upon first use and expanded
239 >     * if necessary.  Updates to the array while recording new workers
240 >     * and unrecording terminated ones are protected from each other
241 >     * by a lock but the array is otherwise concurrently readable, and
242 >     * accessed directly.  To simplify index-based operations, the
243 >     * array size is always a power of two, and all readers must
244 >     * tolerate null slots. Shared (submission) queues are at even
245 >     * indices, worker queues at odd indices. Grouping them together
246 >     * in this way simplifies and speeds up task scanning.
247 >     *
248 >     * All worker thread creation is on-demand, triggered by task
249 >     * submissions, replacement of terminated workers, and/or
250 >     * compensation for blocked workers. However, all other support
251 >     * code is set up to work with other policies.  To ensure that we
252 >     * do not hold on to worker references that would prevent GC, ALL
253 >     * accesses to workQueues are via indices into the workQueues
254 >     * array (which is one source of some of the messy code
255 >     * constructions here). In essence, the workQueues array serves as
256 >     * a weak reference mechanism. Thus for example the wait queue
257 >     * field of ctl stores indices, not references.  Access to the
258 >     * workQueues in associated methods (for example signalWork) must
259 >     * both index-check and null-check the IDs. All such accesses
260 >     * ignore bad IDs by returning out early from what they are doing,
261 >     * since this can only be associated with termination, in which
262 >     * case it is OK to give up.  All uses of the workQueues array
263 >     * also check that it is non-null (even if previously
264 >     * non-null). This allows nulling during termination, which is
265 >     * currently not necessary, but remains an option for
266 >     * resource-revocation-based shutdown schemes. It also helps
267 >     * reduce JIT issuance of uncommon-trap code, which tends to
268 >     * unnecessarily complicate control flow in some methods.
269 >     *
270 >     * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
271 >     * let workers spin indefinitely scanning for tasks when none can
272 >     * be found immediately, and we cannot start/resume workers unless
273 >     * there appear to be tasks available.  On the other hand, we must
274 >     * quickly prod them into action when new tasks are submitted or
275 >     * generated. In many usages, ramp-up time to activate workers is
276 >     * the main limiting factor in overall performance (this is
277 >     * compounded at program start-up by JIT compilation and
278 >     * allocation). So we try to streamline this as much as possible.
279 >     * We park/unpark workers after placing in an event wait queue
280 >     * when they cannot find work. This "queue" is actually a simple
281 >     * Treiber stack, headed by the "id" field of ctl, plus a 15bit
282 >     * counter value (that reflects the number of times a worker has
283 >     * been inactivated) to avoid ABA effects (we need only as many
284 >     * version numbers as worker threads). Successors are held in
285 >     * field WorkQueue.nextWait.  Queuing deals with several intrinsic
286 >     * races, mainly that a task-producing thread can miss seeing (and
287 >     * signalling) another thread that gave up looking for work but
288 >     * has not yet entered the wait queue. We solve this by requiring
289 >     * a full sweep of all workers (via repeated calls to method
290 >     * scan()) both before and after a newly waiting worker is added
291 >     * to the wait queue. During a rescan, the worker might release
292 >     * some other queued worker rather than itself, which has the same
293 >     * net effect. Because enqueued workers may actually be rescanning
294 >     * rather than waiting, we set and clear the "parker" field of
295 >     * WorkQueues to reduce unnecessary calls to unpark.  (This
296 >     * requires a secondary recheck to avoid missed signals.)  Note
297 >     * the unusual conventions about Thread.interrupts surrounding
298 >     * parking and other blocking: Because interrupts are used solely
299 >     * to alert threads to check termination, which is checked anyway
300 >     * upon blocking, we clear status (using Thread.interrupted)
301 >     * before any call to park, so that park does not immediately
302 >     * return due to status being set via some other unrelated call to
303 >     * interrupt in user code.
304 >     *
305 >     * Signalling.  We create or wake up workers only when there
306 >     * appears to be at least one task they might be able to find and
307 >     * execute.  When a submission is added or another worker adds a
308 >     * task to a queue that previously had fewer than two tasks, they
309 >     * signal waiting workers (or trigger creation of new ones if
310 >     * fewer than the given parallelism level -- see signalWork).
311 >     * These primary signals are buttressed by signals during rescans;
312 >     * together these cover the signals needed in cases when more
313 >     * tasks are pushed but untaken, and improve performance compared
314 >     * to having one thread wake up all workers.
315 >     *
316 >     * Trimming workers. To release resources after periods of lack of
317 >     * use, a worker starting to wait when the pool is quiescent will
318 >     * time out and terminate if the pool has remained quiescent for a
319 >     * given period -- a short period if there are more threads than
320 >     * parallelism, longer as the number of threads decreases. This
321 >     * will slowly propagate, eventually terminating all workers after
322 >     * periods of non-use.
323 >     *
324 >     * Shutdown and Termination. A call to shutdownNow atomically sets
325 >     * a runState bit and then (non-atomically) sets each worker's
326 >     * runState status, cancels all unprocessed tasks, and wakes up
327 >     * all waiting workers.  Detecting whether termination should
328 >     * commence after a non-abrupt shutdown() call requires more work
329 >     * and bookkeeping. We need consensus about quiescence (i.e., that
330 >     * there is no more work). The active count provides a primary
331 >     * indication but non-abrupt shutdown still requires a rechecking
332 >     * scan for any workers that are inactive but not queued.
333 >     *
334 >     * Joining Tasks
335 >     * =============
336 >     *
337 >     * Any of several actions may be taken when one worker is waiting
338 >     * to join a task stolen (or always held) by another.  Because we
339 >     * are multiplexing many tasks on to a pool of workers, we can't
340 >     * just let them block (as in Thread.join).  We also cannot just
341 >     * reassign the joiner's run-time stack with another and replace
342 >     * it later, which would be a form of "continuation", that even if
343 >     * possible is not necessarily a good idea since we sometimes need
344 >     * both an unblocked task and its continuation to progress.
345 >     * Instead we combine two tactics:
346 >     *
347 >     *   Helping: Arranging for the joiner to execute some task that it
348 >     *      would be running if the steal had not occurred.
349 >     *
350 >     *   Compensating: Unless there are already enough live threads,
351 >     *      method tryCompensate() may create or re-activate a spare
352 >     *      thread to compensate for blocked joiners until they unblock.
353 >     *
354 >     * A third form (implemented in tryRemoveAndExec and
355 >     * tryPollForAndExec) amounts to helping a hypothetical
356 >     * compensator: If we can readily tell that a possible action of a
357 >     * compensator is to steal and execute the task being joined, the
358 >     * joining thread can do so directly, without the need for a
359 >     * compensation thread (although at the expense of larger run-time
360 >     * stacks, but the tradeoff is typically worthwhile).
361 >     *
362 >     * The ManagedBlocker extension API can't use helping so relies
363 >     * only on compensation in method awaitBlocker.
364 >     *
365 >     * The algorithm in tryHelpStealer entails a form of "linear"
366 >     * helping: Each worker records (in field currentSteal) the most
367 >     * recent task it stole from some other worker. Plus, it records
368 >     * (in field currentJoin) the task it is currently actively
369 >     * joining. Method tryHelpStealer uses these markers to try to
370 >     * find a worker to help (i.e., steal back a task from and execute
371 >     * it) that could hasten completion of the actively joined task.
372 >     * In essence, the joiner executes a task that would be on its own
373 >     * local deque had the to-be-joined task not been stolen. This may
374 >     * be seen as a conservative variant of the approach in Wagner &
375 >     * Calder "Leapfrogging: a portable technique for implementing
376 >     * efficient futures" SIGPLAN Notices, 1993
377 >     * (http://portal.acm.org/citation.cfm?id=155354). It differs in
378 >     * that: (1) We only maintain dependency links across workers upon
379 >     * steals, rather than use per-task bookkeeping.  This sometimes
380 >     * requires a linear scan of workQueues array to locate stealers,
381 >     * but often doesn't because stealers leave hints (that may become
382 >     * stale/wrong) of where to locate them.  A stealHint is only a
383 >     * hint because a worker might have had multiple steals and the
384 >     * hint records only one of them (usually the most current).
385 >     * Hinting isolates cost to when it is needed, rather than adding
386 >     * to per-task overhead.  (2) It is "shallow", ignoring nesting
387 >     * and potentially cyclic mutual steals.  (3) It is intentionally
388 >     * racy: field currentJoin is updated only while actively joining,
389 >     * which means that we miss links in the chain during long-lived
390 >     * tasks, GC stalls etc (which is OK since blocking in such cases
391 >     * is usually a good idea).  (4) We bound the number of attempts
392 >     * to find work (see MAX_HELP) and fall back to suspending the
393 >     * worker and if necessary replacing it with another.
394 >     *
395 >     * It is impossible to keep exactly the target parallelism number
396 >     * of threads running at any given time.  Determining the
397 >     * existence of conservatively safe helping targets, the
398 >     * availability of already-created spares, and the apparent need
399 >     * to create new spares are all racy, so we rely on multiple
400 >     * retries of each.  Compensation in the apparent absence of
401 >     * helping opportunities is challenging to control on JVMs, where
402 >     * GC and other activities can stall progress of tasks that in
403 >     * turn stall out many other dependent tasks, without us being
404 >     * able to determine whether they will ever require compensation.
405 >     * Even though work-stealing otherwise encounters little
406 >     * degradation in the presence of more threads than cores,
407 >     * aggressively adding new threads in such cases entails risk of
408 >     * unwanted positive feedback control loops in which more threads
409 >     * cause more dependent stalls (as well as delayed progress of
410 >     * unblocked threads to the point that we know they are available)
411 >     * leading to more situations requiring more threads, and so
412 >     * on. This aspect of control can be seen as an (analytically
413 >     * intractable) game with an opponent that may choose the worst
414 >     * (for us) active thread to stall at any time.  We take several
415 >     * precautions to bound losses (and thus bound gains), mainly in
416 >     * methods tryCompensate and awaitJoin: (1) We only try
417 >     * compensation after attempting enough helping steps (measured
418 >     * via counting and timing) that we have already consumed the
419 >     * estimated cost of creating and activating a new thread.  (2) We
420 >     * allow up to 50% of threads to be blocked before initially
421 >     * adding any others, and unless completely saturated, check that
422 >     * some work is available for a new worker before adding. Also, we
423 >     * create up to only 50% more threads until entering a mode that
424 >     * only adds a thread if all others are possibly blocked.  All
425 >     * together, this means that we might be half as fast to react,
426 >     * and create half as many threads as possible in the ideal case,
427 >     * but present vastly fewer anomalies in all other cases compared
428 >     * to both more aggressive and more conservative alternatives.
429 >     *
430 >     * Style notes: There is a lot of representation-level coupling
431 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
432 >     * ForkJoinTask.  The fields of WorkQueue maintain data structures
433 >     * managed by ForkJoinPool, so are directly accessed.  There is
434 >     * little point trying to reduce this, since any associated future
435 >     * changes in representations will need to be accompanied by
436 >     * algorithmic changes anyway. Several methods intrinsically
437 >     * sprawl because they must accumulate sets of consistent reads of
438 >     * volatiles held in local variables.  Methods signalWork() and
439 >     * scan() are the main bottlenecks, so are especially heavily
440 >     * micro-optimized/mangled.  There are lots of inline assignments
441 >     * (of form "while ((local = field) != 0)") which are usually the
442 >     * simplest way to ensure the required read orderings (which are
443 >     * sometimes critical). This leads to a "C"-like style of listing
444 >     * declarations of these locals at the heads of methods or blocks.
445 >     * There are several occurrences of the unusual "do {} while
446 >     * (!cas...)"  which is the simplest way to force an update of a
447 >     * CAS'ed variable. There are also other coding oddities that help
448 >     * some methods perform reasonably even when interpreted (not
449 >     * compiled).
450 >     *
451 >     * The order of declarations in this file is:
452 >     * (1) Static utility functions
453 >     * (2) Nested (static) classes
454 >     * (3) Static fields
455 >     * (4) Fields, along with constants used when unpacking some of them
456 >     * (5) Internal control methods
457 >     * (6) Callbacks and other support for ForkJoinTask methods
458 >     * (7) Exported methods
459 >     * (8) Static block initializing statics in minimally dependent order
460       */
461  
462 <    /** Mask for packing and unpacking shorts */
463 <    private static final int  shortMask = 0xffff;
462 >    // Static utilities
463 >
464 >    /**
465 >     * If there is a security manager, makes sure caller has
466 >     * permission to modify threads.
467 >     */
468 >    private static void checkPermission() {
469 >        SecurityManager security = System.getSecurityManager();
470 >        if (security != null)
471 >            security.checkPermission(modifyThreadPermission);
472 >    }
473  
474 <    /** Max pool size -- must be a power of two minus 1 */
69 <    private static final int MAX_THREADS =  0x7FFF;
474 >    // Nested classes
475  
476      /**
477 <     * Factory for creating new ForkJoinWorkerThreads.  A
478 <     * ForkJoinWorkerThreadFactory must be defined and used for
479 <     * ForkJoinWorkerThread subclasses that extend base functionality
480 <     * or initialize threads with different contexts.
477 >     * Factory for creating new {@link ForkJoinWorkerThread}s.
478 >     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
479 >     * for {@code ForkJoinWorkerThread} subclasses that extend base
480 >     * functionality or initialize threads with different contexts.
481       */
482      public static interface ForkJoinWorkerThreadFactory {
483          /**
484           * Returns a new worker thread operating in the given pool.
485           *
486           * @param pool the pool this thread works in
487 <         * @throws NullPointerException if pool is null;
487 >         * @throws NullPointerException if the pool is null
488           */
489          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
490      }
491  
492      /**
493 <     * Default ForkJoinWorkerThreadFactory implementation, creates a
493 >     * Default ForkJoinWorkerThreadFactory implementation; creates a
494       * new ForkJoinWorkerThread.
495       */
496 <    static class  DefaultForkJoinWorkerThreadFactory
496 >    static class DefaultForkJoinWorkerThreadFactory
497          implements ForkJoinWorkerThreadFactory {
498          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
499 <            try {
500 <                return new ForkJoinWorkerThread(pool);
501 <            } catch (OutOfMemoryError oom)  {
499 >            return new ForkJoinWorkerThread(pool);
500 >        }
501 >    }
502 >
503 >    /**
504 >     * Class for artificial tasks that are used to replace the target
505 >     * of local joins if they are removed from an interior queue slot
506 >     * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
507 >     * actually do anything beyond having a unique identity.
508 >     */
509 >    static final class EmptyTask extends ForkJoinTask<Void> {
510 >        EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
511 >        public final Void getRawResult() { return null; }
512 >        public final void setRawResult(Void x) {}
513 >        public final boolean exec() { return true; }
514 >    }
515 >
516 >    /**
517 >     * Queues supporting work-stealing as well as external task
518 >     * submission. See above for main rationale and algorithms.
519 >     * Implementation relies heavily on "Unsafe" intrinsics
520 >     * and selective use of "volatile":
521 >     *
522 >     * Field "base" is the index (mod array.length) of the least valid
523 >     * queue slot, which is always the next position to steal (poll)
524 >     * from if nonempty. Reads and writes require volatile orderings
525 >     * but not CAS, because updates are only performed after slot
526 >     * CASes.
527 >     *
528 >     * Field "top" is the index (mod array.length) of the next queue
529 >     * slot to push to or pop from. It is written only by owner thread
530 >     * for push, or under lock for trySharedPush, and accessed by
531 >     * other threads only after reading (volatile) base.  Both top and
532 >     * base are allowed to wrap around on overflow, but (top - base)
533 >     * (or more commonly -(base - top) to force volatile read of base
534 >     * before top) still estimates size.
535 >     *
536 >     * The array slots are read and written using the emulation of
537 >     * volatiles/atomics provided by Unsafe. Insertions must in
538 >     * general use putOrderedObject as a form of releasing store to
539 >     * ensure that all writes to the task object are ordered before
540 >     * its publication in the queue. (Although we can avoid one case
541 >     * of this when locked in trySharedPush.) All removals entail a
542 >     * CAS to null.  The array is always a power of two. To ensure
543 >     * safety of Unsafe array operations, all accesses perform
544 >     * explicit null checks and implicit bounds checks via
545 >     * power-of-two masking.
546 >     *
547 >     * In addition to basic queuing support, this class contains
548 >     * fields described elsewhere to control execution. It turns out
549 >     * to work better memory-layout-wise to include them in this
550 >     * class rather than a separate class.
551 >     *
552 >     * Performance on most platforms is very sensitive to placement of
553 >     * instances of both WorkQueues and their arrays -- we absolutely
554 >     * do not want multiple WorkQueue instances or multiple queue
555 >     * arrays sharing cache lines. (It would be best for queue objects
556 >     * and their arrays to share, but there is nothing available to
557 >     * help arrange that).  Unfortunately, because they are recorded
558 >     * in a common array, WorkQueue instances are often moved to be
559 >     * adjacent by garbage collectors. To reduce impact, we use field
560 >     * padding that works OK on common platforms; this effectively
561 >     * trades off slightly slower average field access for the sake of
562 >     * avoiding really bad worst-case access. (Until better JVM
563 >     * support is in place, this padding is dependent on transient
564 >     * properties of JVM field layout rules.)  We also take care in
565 >     * allocating, sizing and resizing the array. Non-shared queue
566 >     * arrays are initialized (via method growArray) by workers before
567 >     * use. Others are allocated on first use.
568 >     */
569 >    static final class WorkQueue {
570 >        /**
571 >         * Capacity of work-stealing queue array upon initialization.
572 >         * Must be a power of two; at least 4, but should be larger to
573 >         * reduce or eliminate cacheline sharing among queues.
574 >         * Currently, it is much larger, as a partial workaround for
575 >         * the fact that JVMs often place arrays in locations that
576 >         * share GC bookkeeping (especially cardmarks) such that
577 >         * per-write accesses encounter serious memory contention.
578 >         */
579 >        static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
580 >
581 >        /**
582 >         * Maximum size for queue arrays. Must be a power of two less
583 >         * than or equal to 1 << (31 - width of array entry) to ensure
584 >         * lack of wraparound of index calculations, but defined to a
585 >         * value a bit less than this to help users trap runaway
586 >         * programs before saturating systems.
587 >         */
588 >        static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
589 >
590 >        volatile long totalSteals; // cumulative number of steals
591 >        int seed;                  // for random scanning; initialize nonzero
592 >        volatile int eventCount;   // encoded inactivation count; < 0 if inactive
593 >        int nextWait;              // encoded record of next event waiter
594 >        int rescans;               // remaining scans until block
595 >        int nsteals;               // top-level task executions since last idle
596 >        final int mode;            // lifo, fifo, or shared
597 >        int poolIndex;             // index of this queue in pool (or 0)
598 >        int stealHint;             // index of most recent known stealer
599 >        volatile int runState;     // 1: locked, -1: terminate; else 0
600 >        volatile int base;         // index of next slot for poll
601 >        int top;                   // index of next slot for push
602 >        ForkJoinTask<?>[] array;   // the elements (initially unallocated)
603 >        final ForkJoinPool pool;   // the containing pool (may be null)
604 >        final ForkJoinWorkerThread owner; // owning thread or null if shared
605 >        volatile Thread parker;    // == owner during call to park; else null
606 >        volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
607 >        ForkJoinTask<?> currentSteal; // current non-local task being executed
608 >        // Heuristic padding to ameliorate unfortunate memory placements
609 >        Object p00, p01, p02, p03, p04, p05, p06, p07;
610 >        Object p08, p09, p0a, p0b, p0c, p0d, p0e;
611 >
612 >        WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
613 >            this.mode = mode;
614 >            this.pool = pool;
615 >            this.owner = owner;
616 >            // Place indices in the center of array (that is not yet allocated)
617 >            base = top = INITIAL_QUEUE_CAPACITY >>> 1;
618 >        }
619 >
620 >        /**
621 >         * Returns the approximate number of tasks in the queue.
622 >         */
623 >        final int queueSize() {
624 >            int n = base - top;       // non-owner callers must read base first
625 >            return (n >= 0) ? 0 : -n; // ignore transient negative
626 >        }
627 >
628 >        /**
629 >         * Provides a more accurate estimate of whether this queue has
630 >         * any tasks than does queueSize, by checking whether a
631 >         * near-empty queue has at least one unclaimed task.
632 >         */
633 >        final boolean isEmpty() {
634 >            ForkJoinTask<?>[] a; int m, s;
635 >            int n = base - (s = top);
636 >            return (n >= 0 ||
637 >                    (n == -1 &&
638 >                     ((a = array) == null ||
639 >                      (m = a.length - 1) < 0 ||
640 >                      U.getObjectVolatile
641 >                      (a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
642 >        }
643 >
644 >        /**
645 >         * Pushes a task. Call only by owner in unshared queues.
646 >         *
647 >         * @param task the task. Caller must ensure non-null.
648 >         * @throw RejectedExecutionException if array cannot be resized
649 >         */
650 >        final void push(ForkJoinTask<?> task) {
651 >            ForkJoinTask<?>[] a; ForkJoinPool p;
652 >            int s = top, m, n;
653 >            if ((a = array) != null) {    // ignore if queue removed
654 >                U.putOrderedObject
655 >                    (a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
656 >                if ((n = (top = s + 1) - base) <= 2) {
657 >                    if ((p = pool) != null)
658 >                        p.signalWork();
659 >                }
660 >                else if (n >= m)
661 >                    growArray(true);
662 >            }
663 >        }
664 >
665 >        /**
666 >         * Pushes a task if lock is free and array is either big
667 >         * enough or can be resized to be big enough.
668 >         *
669 >         * @param task the task. Caller must ensure non-null.
670 >         * @return true if submitted
671 >         */
672 >        final boolean trySharedPush(ForkJoinTask<?> task) {
673 >            boolean submitted = false;
674 >            if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
675 >                ForkJoinTask<?>[] a = array;
676 >                int s = top;
677 >                try {
678 >                    if ((a != null && a.length > s + 1 - base) ||
679 >                        (a = growArray(false)) != null) { // must presize
680 >                        int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
681 >                        U.putObject(a, (long)j, task);    // don't need "ordered"
682 >                        top = s + 1;
683 >                        submitted = true;
684 >                    }
685 >                } finally {
686 >                    runState = 0;                         // unlock
687 >                }
688 >            }
689 >            return submitted;
690 >        }
691 >
692 >        /**
693 >         * Takes next task, if one exists, in LIFO order.  Call only
694 >         * by owner in unshared queues.
695 >         */
696 >        final ForkJoinTask<?> pop() {
697 >            ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
698 >            if ((a = array) != null && (m = a.length - 1) >= 0) {
699 >                for (int s; (s = top - 1) - base >= 0;) {
700 >                    long j = ((m & s) << ASHIFT) + ABASE;
701 >                    if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
702 >                        break;
703 >                    if (U.compareAndSwapObject(a, j, t, null)) {
704 >                        top = s;
705 >                        return t;
706 >                    }
707 >                }
708 >            }
709 >            return null;
710 >        }
711 >
712 >        final ForkJoinTask<?> sharedPop() {
713 >            ForkJoinTask<?> task = null;
714 >            if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
715 >                try {
716 >                    ForkJoinTask<?>[] a; int m;
717 >                    if ((a = array) != null && (m = a.length - 1) >= 0) {
718 >                        for (int s; (s = top - 1) - base >= 0;) {
719 >                            long j = ((m & s) << ASHIFT) + ABASE;
720 >                            ForkJoinTask<?> t =
721 >                                (ForkJoinTask<?>)U.getObject(a, j);
722 >                            if (t == null)
723 >                                break;
724 >                            if (U.compareAndSwapObject(a, j, t, null)) {
725 >                                top = s;
726 >                                task = t;
727 >                                break;
728 >                            }
729 >                        }
730 >                    }
731 >                } finally {
732 >                    runState = 0;
733 >                }
734 >            }
735 >            return task;
736 >        }
737 >
738 >        /**
739 >         * Version of pop that takes top element only if it
740 >         * its root is the given CountedCompleter.
741 >         */
742 >        final ForkJoinTask<?> popCC(CountedCompleter<?> root) {
743 >            ForkJoinTask<?>[] a; int m;
744 >            if (root != null && (a = array) != null && (m = a.length - 1) >= 0) {
745 >                for (int s; (s = top - 1) - base >= 0;) {
746 >                    long j = ((m & s) << ASHIFT) + ABASE;
747 >                    ForkJoinTask<?> t =
748 >                        (ForkJoinTask<?>)U.getObject(a, j);
749 >                    if (t == null || !(t instanceof CountedCompleter) ||
750 >                        ((CountedCompleter<?>)t).getRoot() != root)
751 >                        break;
752 >                    if (U.compareAndSwapObject(a, j, t, null)) {
753 >                        top = s;
754 >                        return t;
755 >                    }
756 >                    if (root.status < 0)
757 >                        break;
758 >                }
759 >            }
760 >            return null;
761 >        }
762 >
763 >        /**
764 >         * Shared version of popCC
765 >         */
766 >        final ForkJoinTask<?> sharedPopCC(CountedCompleter<?> root) {
767 >            ForkJoinTask<?> task = null;
768 >            if (root != null &&
769 >                runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
770 >                try {
771 >                    ForkJoinTask<?>[] a; int m;
772 >                    if ((a = array) != null && (m = a.length - 1) >= 0) {
773 >                        for (int s; (s = top - 1) - base >= 0;) {
774 >                            long j = ((m & s) << ASHIFT) + ABASE;
775 >                            ForkJoinTask<?> t =
776 >                                (ForkJoinTask<?>)U.getObject(a, j);
777 >                            if (t == null || !(t instanceof CountedCompleter) ||
778 >                                ((CountedCompleter<?>)t).getRoot() != root)
779 >                                break;
780 >                            if (U.compareAndSwapObject(a, j, t, null)) {
781 >                                top = s;
782 >                                task = t;
783 >                                break;
784 >                            }
785 >                            if (root.status < 0)
786 >                                break;
787 >                        }
788 >                    }
789 >                } finally {
790 >                    runState = 0;
791 >                }
792 >            }
793 >            return task;
794 >        }
795 >
796 >        /**
797 >         * Takes a task in FIFO order if b is base of queue and a task
798 >         * can be claimed without contention. Specialized versions
799 >         * appear in ForkJoinPool methods scan and tryHelpStealer.
800 >         */
801 >        final ForkJoinTask<?> pollAt(int b) {
802 >            ForkJoinTask<?> t; ForkJoinTask<?>[] a;
803 >            if ((a = array) != null) {
804 >                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
805 >                if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
806 >                    base == b &&
807 >                    U.compareAndSwapObject(a, j, t, null)) {
808 >                    base = b + 1;
809 >                    return t;
810 >                }
811 >            }
812 >            return null;
813 >        }
814 >
815 >        /**
816 >         * Takes next task, if one exists, in FIFO order.
817 >         */
818 >        final ForkJoinTask<?> poll() {
819 >            ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
820 >            while ((b = base) - top < 0 && (a = array) != null) {
821 >                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
822 >                t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
823 >                if (t != null) {
824 >                    if (base == b &&
825 >                        U.compareAndSwapObject(a, j, t, null)) {
826 >                        base = b + 1;
827 >                        return t;
828 >                    }
829 >                }
830 >                else if (base == b) {
831 >                    if (b + 1 == top)
832 >                        break;
833 >                    Thread.yield(); // wait for lagging update
834 >                }
835 >            }
836 >            return null;
837 >        }
838 >
839 >        /**
840 >         * Takes next task, if one exists, in order specified by mode.
841 >         */
842 >        final ForkJoinTask<?> nextLocalTask() {
843 >            return mode == 0 ? pop() : poll();
844 >        }
845 >
846 >        /**
847 >         * Returns next task, if one exists, in order specified by mode.
848 >         */
849 >        final ForkJoinTask<?> peek() {
850 >            ForkJoinTask<?>[] a = array; int m;
851 >            if (a == null || (m = a.length - 1) < 0)
852                  return null;
853 +            int i = mode == 0 ? top - 1 : base;
854 +            int j = ((i & m) << ASHIFT) + ABASE;
855 +            return (ForkJoinTask<?>)U.getObjectVolatile(a, j);
856 +        }
857 +
858 +        /**
859 +         * Pops the given task only if it is at the current top.
860 +         */
861 +        final boolean tryUnpush(ForkJoinTask<?> t) {
862 +            ForkJoinTask<?>[] a; int s;
863 +            if ((a = array) != null && (s = top) != base &&
864 +                U.compareAndSwapObject
865 +                (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
866 +                top = s;
867 +                return true;
868 +            }
869 +            return false;
870 +        }
871 +
872 +        /**
873 +         * Version of tryUnpush for shared queues; called by non-FJ
874 +         * submitters after prechecking that task probably exists.
875 +         */
876 +        final boolean trySharedUnpush(ForkJoinTask<?> t) {
877 +            boolean success = false;
878 +            if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
879 +                try {
880 +                    ForkJoinTask<?>[] a; int s;
881 +                    if ((a = array) != null && (s = top) != base &&
882 +                        U.compareAndSwapObject
883 +                        (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
884 +                        top = s;
885 +                        success = true;
886 +                    }
887 +                } finally {
888 +                    runState = 0;                         // unlock
889 +                }
890              }
891 +            return success;
892 +        }
893 +
894 +        /**
895 +         * Polls the given task only if it is at the current base.
896 +         */
897 +        final boolean pollFor(ForkJoinTask<?> task) {
898 +            ForkJoinTask<?>[] a; int b;
899 +            if ((b = base) - top < 0 && (a = array) != null) {
900 +                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
901 +                if (U.getObjectVolatile(a, j) == task && base == b &&
902 +                    U.compareAndSwapObject(a, j, task, null)) {
903 +                    base = b + 1;
904 +                    return true;
905 +                }
906 +            }
907 +            return false;
908 +        }
909 +
910 +        /**
911 +         * Initializes or doubles the capacity of array. Call either
912 +         * by owner or with lock held -- it is OK for base, but not
913 +         * top, to move while resizings are in progress.
914 +         *
915 +         * @param rejectOnFailure if true, throw exception if capacity
916 +         * exceeded (relayed ultimately to user); else return null.
917 +         */
918 +        final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
919 +            ForkJoinTask<?>[] oldA = array;
920 +            int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
921 +            if (size <= MAXIMUM_QUEUE_CAPACITY) {
922 +                int oldMask, t, b;
923 +                ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
924 +                if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
925 +                    (t = top) - (b = base) > 0) {
926 +                    int mask = size - 1;
927 +                    do {
928 +                        ForkJoinTask<?> x;
929 +                        int oldj = ((b & oldMask) << ASHIFT) + ABASE;
930 +                        int j    = ((b &    mask) << ASHIFT) + ABASE;
931 +                        x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
932 +                        if (x != null &&
933 +                            U.compareAndSwapObject(oldA, oldj, x, null))
934 +                            U.putObjectVolatile(a, j, x);
935 +                    } while (++b != t);
936 +                }
937 +                return a;
938 +            }
939 +            else if (!rejectOnFailure)
940 +                return null;
941 +            else
942 +                throw new RejectedExecutionException("Queue capacity exceeded");
943 +        }
944 +
945 +        /**
946 +         * Removes and cancels all known tasks, ignoring any exceptions.
947 +         */
948 +        final void cancelAll() {
949 +            ForkJoinTask.cancelIgnoringExceptions(currentJoin);
950 +            ForkJoinTask.cancelIgnoringExceptions(currentSteal);
951 +            for (ForkJoinTask<?> t; (t = poll()) != null; )
952 +                ForkJoinTask.cancelIgnoringExceptions(t);
953 +        }
954 +
955 +        /**
956 +         * Computes next value for random probes.  Scans don't require
957 +         * a very high quality generator, but also not a crummy one.
958 +         * Marsaglia xor-shift is cheap and works well enough.  Note:
959 +         * This is manually inlined in its usages in ForkJoinPool to
960 +         * avoid writes inside busy scan loops.
961 +         */
962 +        final int nextSeed() {
963 +            int r = seed;
964 +            r ^= r << 13;
965 +            r ^= r >>> 17;
966 +            return seed = r ^= r << 5;
967 +        }
968 +
969 +        // Execution methods
970 +
971 +        /**
972 +         * Pops and runs tasks until empty.
973 +         */
974 +        private void popAndExecAll() {
975 +            // A bit faster than repeated pop calls
976 +            ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
977 +            while ((a = array) != null && (m = a.length - 1) >= 0 &&
978 +                   (s = top - 1) - base >= 0 &&
979 +                   (t = ((ForkJoinTask<?>)
980 +                         U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
981 +                   != null) {
982 +                if (U.compareAndSwapObject(a, j, t, null)) {
983 +                    top = s;
984 +                    t.doExec();
985 +                }
986 +            }
987 +        }
988 +
989 +        /**
990 +         * Polls and runs tasks until empty.
991 +         */
992 +        private void pollAndExecAll() {
993 +            for (ForkJoinTask<?> t; (t = poll()) != null;)
994 +                t.doExec();
995 +        }
996 +
997 +        /**
998 +         * If present, removes from queue and executes the given task, or
999 +         * any other cancelled task. Returns (true) immediately on any CAS
1000 +         * or consistency check failure so caller can retry.
1001 +         *
1002 +         * @return 0 if no progress can be made, else positive
1003 +         * (this unusual convention simplifies use with tryHelpStealer.)
1004 +         */
1005 +        final int tryRemoveAndExec(ForkJoinTask<?> task) {
1006 +            int stat = 1;
1007 +            boolean removed = false, empty = true;
1008 +            ForkJoinTask<?>[] a; int m, s, b, n;
1009 +            if ((a = array) != null && (m = a.length - 1) >= 0 &&
1010 +                (n = (s = top) - (b = base)) > 0) {
1011 +                for (ForkJoinTask<?> t;;) {           // traverse from s to b
1012 +                    int j = ((--s & m) << ASHIFT) + ABASE;
1013 +                    t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
1014 +                    if (t == null)                    // inconsistent length
1015 +                        break;
1016 +                    else if (t == task) {
1017 +                        if (s + 1 == top) {           // pop
1018 +                            if (!U.compareAndSwapObject(a, j, task, null))
1019 +                                break;
1020 +                            top = s;
1021 +                            removed = true;
1022 +                        }
1023 +                        else if (base == b)           // replace with proxy
1024 +                            removed = U.compareAndSwapObject(a, j, task,
1025 +                                                             new EmptyTask());
1026 +                        break;
1027 +                    }
1028 +                    else if (t.status >= 0)
1029 +                        empty = false;
1030 +                    else if (s + 1 == top) {          // pop and throw away
1031 +                        if (U.compareAndSwapObject(a, j, t, null))
1032 +                            top = s;
1033 +                        break;
1034 +                    }
1035 +                    if (--n == 0) {
1036 +                        if (!empty && base == b)
1037 +                            stat = 0;
1038 +                        break;
1039 +                    }
1040 +                }
1041 +            }
1042 +            if (removed)
1043 +                task.doExec();
1044 +            return stat;
1045 +        }
1046 +
1047 +        /**
1048 +         * Executes a top-level task and any local tasks remaining
1049 +         * after execution.
1050 +         */
1051 +        final void runTask(ForkJoinTask<?> t) {
1052 +            if (t != null) {
1053 +                currentSteal = t;
1054 +                t.doExec();
1055 +                if (top != base) {       // process remaining local tasks
1056 +                    if (mode == 0)
1057 +                        popAndExecAll();
1058 +                    else
1059 +                        pollAndExecAll();
1060 +                }
1061 +                ++nsteals;
1062 +                currentSteal = null;
1063 +            }
1064 +        }
1065 +
1066 +        /**
1067 +         * Executes a non-top-level (stolen) task.
1068 +         */
1069 +        final void runSubtask(ForkJoinTask<?> t) {
1070 +            if (t != null) {
1071 +                ForkJoinTask<?> ps = currentSteal;
1072 +                currentSteal = t;
1073 +                t.doExec();
1074 +                currentSteal = ps;
1075 +            }
1076 +        }
1077 +
1078 +        /**
1079 +         * Returns true if owned and not known to be blocked.
1080 +         */
1081 +        final boolean isApparentlyUnblocked() {
1082 +            Thread wt; Thread.State s;
1083 +            return (eventCount >= 0 &&
1084 +                    (wt = owner) != null &&
1085 +                    (s = wt.getState()) != Thread.State.BLOCKED &&
1086 +                    s != Thread.State.WAITING &&
1087 +                    s != Thread.State.TIMED_WAITING);
1088 +        }
1089 +
1090 +        /**
1091 +         * If this owned and is not already interrupted, try to
1092 +         * interrupt and/or unpark, ignoring exceptions.
1093 +         */
1094 +        final void interruptOwner() {
1095 +            Thread wt, p;
1096 +            if ((wt = owner) != null && !wt.isInterrupted()) {
1097 +                try {
1098 +                    wt.interrupt();
1099 +                } catch (SecurityException ignore) {
1100 +                }
1101 +            }
1102 +            if ((p = parker) != null)
1103 +                U.unpark(p);
1104 +        }
1105 +
1106 +        // Unsafe mechanics
1107 +        private static final sun.misc.Unsafe U;
1108 +        private static final long RUNSTATE;
1109 +        private static final int ABASE;
1110 +        private static final int ASHIFT;
1111 +        static {
1112 +            int s;
1113 +            try {
1114 +                U = getUnsafe();
1115 +                Class<?> k = WorkQueue.class;
1116 +                Class<?> ak = ForkJoinTask[].class;
1117 +                RUNSTATE = U.objectFieldOffset
1118 +                    (k.getDeclaredField("runState"));
1119 +                ABASE = U.arrayBaseOffset(ak);
1120 +                s = U.arrayIndexScale(ak);
1121 +            } catch (Exception e) {
1122 +                throw new Error(e);
1123 +            }
1124 +            if ((s & (s-1)) != 0)
1125 +                throw new Error("data type scale not a power of two");
1126 +            ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1127 +        }
1128 +    }
1129 +
1130 +    /**
1131 +     * Per-thread records for threads that submit to pools. Currently
1132 +     * holds only pseudo-random seed / index that is used to choose
1133 +     * submission queues in method doSubmit. In the future, this may
1134 +     * also incorporate a means to implement different task rejection
1135 +     * and resubmission policies.
1136 +     *
1137 +     * Seeds for submitters and workers/workQueues work in basically
1138 +     * the same way but are initialized and updated using slightly
1139 +     * different mechanics. Both are initialized using the same
1140 +     * approach as in class ThreadLocal, where successive values are
1141 +     * unlikely to collide with previous values. This is done during
1142 +     * registration for workers, but requires a separate AtomicInteger
1143 +     * for submitters. Seeds are then randomly modified upon
1144 +     * collisions using xorshifts, which requires a non-zero seed.
1145 +     */
1146 +    static final class Submitter {
1147 +        int seed;
1148 +        Submitter() {
1149 +            int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1150 +            seed = (s == 0) ? 1 : s; // ensure non-zero
1151          }
1152      }
1153  
1154 +    /** ThreadLocal class for Submitters */
1155 +    static final class ThreadSubmitter extends ThreadLocal<Submitter> {
1156 +        public Submitter initialValue() { return new Submitter(); }
1157 +    }
1158 +
1159 +    // static fields (initialized in static initializer below)
1160 +
1161      /**
1162       * Creates a new ForkJoinWorkerThread. This factory is used unless
1163       * overridden in ForkJoinPool constructors.
1164       */
1165      public static final ForkJoinWorkerThreadFactory
1166 <        defaultForkJoinWorkerThreadFactory =
1167 <        new DefaultForkJoinWorkerThreadFactory();
1166 >        defaultForkJoinWorkerThreadFactory;
1167 >
1168 >
1169 >    /** Property prefix for constructing common pool */
1170 >    private static final String propPrefix =
1171 >        "java.util.concurrent.ForkJoinPool.common.";
1172  
1173      /**
1174 <     * Permission required for callers of methods that may start or
1175 <     * kill threads.
1174 >     * Common (static) pool. Non-null for public use unless a static
1175 >     * construction exception, but internal usages must null-check on
1176 >     * use.
1177       */
1178 <    private static final RuntimePermission modifyThreadPermission =
115 <        new RuntimePermission("modifyThread");
1178 >    static final ForkJoinPool commonPool;
1179  
1180      /**
1181 <     * If there is a security manager, makes sure caller has
119 <     * permission to modify threads.
1181 >     * Common pool parallelism. Must equal commonPool.parallelism.
1182       */
1183 <    private static void checkPermission() {
122 <        SecurityManager security = System.getSecurityManager();
123 <        if (security != null)
124 <            security.checkPermission(modifyThreadPermission);
125 <    }
1183 >    static final int commonPoolParallelism;
1184  
1185      /**
1186       * Generator for assigning sequence numbers as pool names.
1187       */
1188 <    private static final AtomicInteger poolNumberGenerator =
131 <        new AtomicInteger();
1188 >    private static final AtomicInteger poolNumberGenerator;
1189  
1190      /**
1191 <     * Array holding all worker threads in the pool. Array size must
1192 <     * be a power of two.  Updates and replacements are protected by
136 <     * workerLock, but it is always kept in a consistent enough state
137 <     * to be randomly accessed without locking by workers performing
138 <     * work-stealing.
1191 >     * Generator for initial hashes/seeds for submitters. Accessed by
1192 >     * Submitter class constructor.
1193       */
1194 <    volatile ForkJoinWorkerThread[] workers;
1194 >    static final AtomicInteger nextSubmitterSeed;
1195  
1196      /**
1197 <     * Lock protecting access to workers.
1197 >     * Permission required for callers of methods that may start or
1198 >     * kill threads.
1199       */
1200 <    private final ReentrantLock workerLock;
1200 >    private static final RuntimePermission modifyThreadPermission;
1201  
1202      /**
1203 <     * Condition for awaitTermination.
1203 >     * Per-thread submission bookkeeping. Shared across all pools
1204 >     * to reduce ThreadLocal pollution and because random motion
1205 >     * to avoid contention in one pool is likely to hold for others.
1206       */
1207 <    private final Condition termination;
1207 >    private static final ThreadSubmitter submitters;
1208 >
1209 >    // static constants
1210  
1211      /**
1212 <     * The uncaught exception handler used when any worker
1213 <     * abrupty terminates
1212 >     * Initial timeout value (in nanoseconds) for the thread triggering
1213 >     * quiescence to park waiting for new work. On timeout, the thread
1214 >     * will instead try to shrink the number of workers.
1215       */
1216 <    private Thread.UncaughtExceptionHandler ueh;
1216 >    private static final long IDLE_TIMEOUT      = 1000L * 1000L * 1000L; // 1sec
1217  
1218      /**
1219 <     * Creation factory for worker threads.
1219 >     * Timeout value when there are more threads than parallelism level
1220       */
1221 <    private final ForkJoinWorkerThreadFactory factory;
1221 >    private static final long FAST_IDLE_TIMEOUT =  100L * 1000L * 1000L;
1222  
1223      /**
1224 <     * Head of stack of threads that were created to maintain
1225 <     * parallelism when other threads blocked, but have since
1226 <     * suspended when the parallelism level rose.
1224 >     * The maximum stolen->joining link depth allowed in method
1225 >     * tryHelpStealer.  Must be a power of two. This value also
1226 >     * controls the maximum number of times to try to help join a task
1227 >     * without any apparent progress or change in pool state before
1228 >     * giving up and blocking (see awaitJoin).  Depths for legitimate
1229 >     * chains are unbounded, but we use a fixed constant to avoid
1230 >     * (otherwise unchecked) cycles and to bound staleness of
1231 >     * traversal parameters at the expense of sometimes blocking when
1232 >     * we could be helping.
1233       */
1234 <    private volatile WaitQueueNode spareStack;
1234 >    private static final int MAX_HELP = 64;
1235  
1236      /**
1237 <     * Sum of per-thread steal counts, updated only when threads are
1238 <     * idle or terminating.
1237 >     * Secondary time-based bound (in nanosecs) for helping attempts
1238 >     * before trying compensated blocking in awaitJoin. Used in
1239 >     * conjunction with MAX_HELP to reduce variance due to different
1240 >     * polling rates associated with different helping options. The
1241 >     * value should roughly approximate the time required to create
1242 >     * and/or activate a worker thread.
1243       */
1244 <    private final AtomicLong stealCount;
1244 >    private static final long COMPENSATION_DELAY = 1L << 18; // ~0.25 millisec
1245  
1246      /**
1247 <     * Queue for external submissions.
1247 >     * Increment for seed generators. See class ThreadLocal for
1248 >     * explanation.
1249       */
1250 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
1250 >    private static final int SEED_INCREMENT = 0x61c88647;
1251  
1252      /**
1253 <     * Head of Treiber stack for barrier sync. See below for explanation
1254 <     */
1255 <    private volatile WaitQueueNode syncStack;
1253 >     * Bits and masks for control variables
1254 >     *
1255 >     * Field ctl is a long packed with:
1256 >     * AC: Number of active running workers minus target parallelism (16 bits)
1257 >     * TC: Number of total workers minus target parallelism (16 bits)
1258 >     * ST: true if pool is terminating (1 bit)
1259 >     * EC: the wait count of top waiting thread (15 bits)
1260 >     * ID: poolIndex of top of Treiber stack of waiters (16 bits)
1261 >     *
1262 >     * When convenient, we can extract the upper 32 bits of counts and
1263 >     * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
1264 >     * (int)ctl.  The ec field is never accessed alone, but always
1265 >     * together with id and st. The offsets of counts by the target
1266 >     * parallelism and the positionings of fields makes it possible to
1267 >     * perform the most common checks via sign tests of fields: When
1268 >     * ac is negative, there are not enough active workers, when tc is
1269 >     * negative, there are not enough total workers, and when e is
1270 >     * negative, the pool is terminating.  To deal with these possibly
1271 >     * negative fields, we use casts in and out of "short" and/or
1272 >     * signed shifts to maintain signedness.
1273 >     *
1274 >     * When a thread is queued (inactivated), its eventCount field is
1275 >     * set negative, which is the only way to tell if a worker is
1276 >     * prevented from executing tasks, even though it must continue to
1277 >     * scan for them to avoid queuing races. Note however that
1278 >     * eventCount updates lag releases so usage requires care.
1279 >     *
1280 >     * Field runState is an int packed with:
1281 >     * SHUTDOWN: true if shutdown is enabled (1 bit)
1282 >     * SEQ:  a sequence number updated upon (de)registering workers (30 bits)
1283 >     * INIT: set true after workQueues array construction (1 bit)
1284 >     *
1285 >     * The sequence number enables simple consistency checks:
1286 >     * Staleness of read-only operations on the workQueues array can
1287 >     * be checked by comparing runState before vs after the reads.
1288 >     */
1289 >
1290 >    // bit positions/shifts for fields
1291 >    private static final int  AC_SHIFT   = 48;
1292 >    private static final int  TC_SHIFT   = 32;
1293 >    private static final int  ST_SHIFT   = 31;
1294 >    private static final int  EC_SHIFT   = 16;
1295 >
1296 >    // bounds
1297 >    private static final int  SMASK      = 0xffff;  // short bits
1298 >    private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
1299 >    private static final int  SQMASK     = 0xfffe;  // even short bits
1300 >    private static final int  SHORT_SIGN = 1 << 15;
1301 >    private static final int  INT_SIGN   = 1 << 31;
1302 >
1303 >    // masks
1304 >    private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
1305 >    private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
1306 >    private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
1307 >
1308 >    // units for incrementing and decrementing
1309 >    private static final long TC_UNIT    = 1L << TC_SHIFT;
1310 >    private static final long AC_UNIT    = 1L << AC_SHIFT;
1311 >
1312 >    // masks and units for dealing with u = (int)(ctl >>> 32)
1313 >    private static final int  UAC_SHIFT  = AC_SHIFT - 32;
1314 >    private static final int  UTC_SHIFT  = TC_SHIFT - 32;
1315 >    private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
1316 >    private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
1317 >    private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
1318 >    private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
1319 >
1320 >    // masks and units for dealing with e = (int)ctl
1321 >    private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
1322 >    private static final int E_SEQ       = 1 << EC_SHIFT;
1323 >
1324 >    // runState bits
1325 >    private static final int SHUTDOWN    = 1 << 31;
1326 >
1327 >    // access mode for WorkQueue
1328 >    static final int LIFO_QUEUE          =  0;
1329 >    static final int FIFO_QUEUE          =  1;
1330 >    static final int SHARED_QUEUE        = -1;
1331 >
1332 >    // Instance fields
1333 >
1334 >    /*
1335 >     * Field layout order in this class tends to matter more than one
1336 >     * would like. Runtime layout order is only loosely related to
1337 >     * declaration order and may differ across JVMs, but the following
1338 >     * empirically works OK on current JVMs.
1339 >     */
1340 >
1341 >    volatile long stealCount;                  // collects worker counts
1342 >    volatile long ctl;                         // main pool control
1343 >    final int parallelism;                     // parallelism level
1344 >    final int localMode;                       // per-worker scheduling mode
1345 >    volatile int nextWorkerNumber;             // to create worker name string
1346 >    final int submitMask;                      // submit queue index bound
1347 >    int nextSeed;                              // for initializing worker seeds
1348 >    volatile int mainLock;                     // spinlock for array updates
1349 >    volatile int runState;                     // shutdown status and seq
1350 >    WorkQueue[] workQueues;                    // main registry
1351 >    final ForkJoinWorkerThreadFactory factory; // factory for new workers
1352 >    final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1353 >    final String workerNamePrefix;             // to create worker name string
1354 >
1355 >    /*
1356 >     * Mechanics for main lock protecting worker array updates.  Uses
1357 >     * the same strategy as ConcurrentHashMap bins -- a spinLock for
1358 >     * normal cases, but falling back to builtin lock when (rarely)
1359 >     * needed.  See internal ConcurrentHashMap documentation for
1360 >     * explanation.
1361 >     */
1362 >
1363 >    static final int LOCK_WAITING = 2; // bit to indicate need for signal
1364 >    static final int MAX_LOCK_SPINS = 1 << 8;
1365 >
1366 >    private void tryAwaitMainLock() {
1367 >        int spins = MAX_LOCK_SPINS, r = 0, h;
1368 >        while (((h = mainLock) & 1) != 0) {
1369 >            if (r == 0)
1370 >                r = ThreadLocalRandom.current().nextInt(); // randomize spins
1371 >            else if (spins >= 0) {
1372 >                r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1373 >                if (r >= 0)
1374 >                    --spins;
1375 >            }
1376 >            else if (U.compareAndSwapInt(this, MAINLOCK, h, h | LOCK_WAITING)) {
1377 >                synchronized (this) {
1378 >                    if ((mainLock & LOCK_WAITING) != 0) {
1379 >                        try {
1380 >                            wait();
1381 >                        } catch (InterruptedException ie) {
1382 >                            Thread.currentThread().interrupt();
1383 >                        }
1384 >                    }
1385 >                    else
1386 >                        notifyAll(); // possibly won race vs signaller
1387 >                }
1388 >                break;
1389 >            }
1390 >        }
1391 >    }
1392 >
1393 >    //  Creating, registering, and deregistering workers
1394  
1395      /**
1396 <     * The count for event barrier
1396 >     * Tries to create and start a worker
1397       */
1398 <    private volatile long eventCount;
1398 >    private void addWorker() {
1399 >        Throwable ex = null;
1400 >        ForkJoinWorkerThread wt = null;
1401 >        try {
1402 >            if ((wt = factory.newThread(this)) != null) {
1403 >                wt.start();
1404 >                return;
1405 >            }
1406 >        } catch (Throwable e) {
1407 >            ex = e;
1408 >        }
1409 >        deregisterWorker(wt, ex); // adjust counts etc on failure
1410 >    }
1411  
1412      /**
1413 <     * Pool number, just for assigning useful names to worker threads
1413 >     * Callback from ForkJoinWorkerThread constructor to assign a
1414 >     * public name. This must be separate from registerWorker because
1415 >     * it is called during the "super" constructor call in
1416 >     * ForkJoinWorkerThread.
1417       */
1418 <    private final int poolNumber;
1418 >    final String nextWorkerName() {
1419 >        int n;
1420 >        do {} while(!U.compareAndSwapInt(this, NEXTWORKERNUMBER,
1421 >                                         n = nextWorkerNumber, ++n));
1422 >        return workerNamePrefix.concat(Integer.toString(n));
1423 >    }
1424  
1425      /**
1426 <     * The maximum allowed pool size
1426 >     * Callback from ForkJoinWorkerThread constructor to establish its
1427 >     * poolIndex and record its WorkQueue. To avoid scanning bias due
1428 >     * to packing entries in front of the workQueues array, we treat
1429 >     * the array as a simple power-of-two hash table using per-thread
1430 >     * seed as hash, expanding as needed.
1431 >     *
1432 >     * @param w the worker's queue
1433       */
1434 <    private volatile int maxPoolSize;
1434 >    final void registerWorker(WorkQueue w) {
1435 >        while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1436 >            tryAwaitMainLock();
1437 >        try {
1438 >            WorkQueue[] ws;
1439 >            if ((ws = workQueues) == null)
1440 >                ws = workQueues = new WorkQueue[submitMask + 1];
1441 >            if (w != null) {
1442 >                int rs, n =  ws.length, m = n - 1;
1443 >                int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1444 >                w.seed = (s == 0) ? 1 : s;          // ensure non-zero seed
1445 >                int r = (s << 1) | 1;               // use odd-numbered indices
1446 >                if (ws[r &= m] != null) {           // collision
1447 >                    int probes = 0;                 // step by approx half size
1448 >                    int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1449 >                    while (ws[r = (r + step) & m] != null) {
1450 >                        if (++probes >= n) {
1451 >                            workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1452 >                            m = n - 1;
1453 >                            probes = 0;
1454 >                        }
1455 >                    }
1456 >                }
1457 >                w.eventCount = w.poolIndex = r;     // establish before recording
1458 >                ws[r] = w;                          // also update seq
1459 >                runState = ((rs = runState) & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN);
1460 >            }
1461 >        } finally {
1462 >            if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1463 >                mainLock = 0;
1464 >                synchronized (this) { notifyAll(); };
1465 >            }
1466 >        }
1467 >
1468 >    }
1469  
1470      /**
1471 <     * The desired parallelism level, updated only under workerLock.
1471 >     * Final callback from terminating worker, as well as upon failure
1472 >     * to construct or start a worker in addWorker.  Removes record of
1473 >     * worker from array, and adjusts counts. If pool is shutting
1474 >     * down, tries to complete termination.
1475 >     *
1476 >     * @param wt the worker thread or null if addWorker failed
1477 >     * @param ex the exception causing failure, or null if none
1478       */
1479 <    private volatile int parallelism;
1479 >    final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1480 >        WorkQueue w = null;
1481 >        if (wt != null && (w = wt.workQueue) != null) {
1482 >            w.runState = -1;                // ensure runState is set
1483 >            long steals = w.totalSteals + w.nsteals, sc;
1484 >            do {} while(!U.compareAndSwapLong(this, STEALCOUNT,
1485 >                                              sc = stealCount, sc + steals));
1486 >            int idx = w.poolIndex;
1487 >            while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1488 >                tryAwaitMainLock();
1489 >            try {
1490 >                WorkQueue[] ws = workQueues;
1491 >                if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1492 >                    ws[idx] = null;
1493 >            } finally {
1494 >                if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1495 >                    mainLock = 0;
1496 >                    synchronized (this) { notifyAll(); };
1497 >                }
1498 >            }
1499 >        }
1500 >
1501 >        long c;                             // adjust ctl counts
1502 >        do {} while (!U.compareAndSwapLong
1503 >                     (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1504 >                                           ((c - TC_UNIT) & TC_MASK) |
1505 >                                           (c & ~(AC_MASK|TC_MASK)))));
1506 >
1507 >        if (!tryTerminate(false, false) && w != null) {
1508 >            w.cancelAll();                  // cancel remaining tasks
1509 >            if (w.array != null)            // suppress signal if never ran
1510 >                signalWork();               // wake up or create replacement
1511 >            if (ex == null)                 // help clean refs on way out
1512 >                ForkJoinTask.helpExpungeStaleExceptions();
1513 >        }
1514 >
1515 >        if (ex != null)                     // rethrow
1516 >            U.throwException(ex);
1517 >    }
1518 >
1519 >    // Submissions
1520  
1521      /**
1522 <     * Holds number of total (i.e., created and not yet terminated)
1523 <     * and running (i.e., not blocked on joins or other managed sync)
1524 <     * threads, packed into one int to ensure consistent snapshot when
1525 <     * making decisions about creating and suspending spare
1526 <     * threads. Updated only by CAS.  Note: CASes in
1527 <     * updateRunningCount and preJoin running active count is in low
1528 <     * word, so need to be modified if this changes
1522 >     * Unless shutting down, adds the given task to a submission queue
1523 >     * at submitter's current queue index (modulo submission
1524 >     * range). If no queue exists at the index, one is created.  If
1525 >     * the queue is busy, another index is randomly chosen. The
1526 >     * submitMask bounds the effective number of queues to the
1527 >     * (nearest power of two for) parallelism level.
1528 >     *
1529 >     * @param task the task. Caller must ensure non-null.
1530       */
1531 <    private volatile int workerCounts;
1532 <
1533 <    private static int totalCountOf(int s)           { return s >>> 16;  }
1534 <    private static int runningCountOf(int s)         { return s & shortMask; }
1535 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
1531 >    private void doSubmit(ForkJoinTask<?> task) {
1532 >        Submitter s = submitters.get();
1533 >        for (int r = s.seed, m = submitMask;;) {
1534 >            WorkQueue[] ws; WorkQueue q;
1535 >            int k = r & m & SQMASK;          // use only even indices
1536 >            if (runState < 0)
1537 >                throw new RejectedExecutionException(); // shutting down
1538 >            else if ((ws = workQueues) == null || ws.length <= k) {
1539 >                while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1540 >                    tryAwaitMainLock();
1541 >                try {
1542 >                    if (workQueues == null)
1543 >                        workQueues = new WorkQueue[submitMask + 1];
1544 >                } finally {
1545 >                    if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1546 >                        mainLock = 0;
1547 >                        synchronized (this) { notifyAll(); };
1548 >                    }
1549 >                }
1550 >            }
1551 >            else if ((q = ws[k]) == null) {  // create new queue
1552 >                WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1553 >                while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
1554 >                    tryAwaitMainLock();
1555 >                try {
1556 >                    int rs = runState;       // to update seq
1557 >                    if (ws == workQueues && ws[k] == null) {
1558 >                        ws[k] = nq;
1559 >                        runState = ((rs & SHUTDOWN) | ((rs + 2) & ~SHUTDOWN));
1560 >                    }
1561 >                } finally {
1562 >                    if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
1563 >                        mainLock = 0;
1564 >                        synchronized (this) { notifyAll(); };
1565 >                    }
1566 >                }
1567 >            }
1568 >            else if (q.trySharedPush(task)) {
1569 >                signalWork();
1570 >                return;
1571 >            }
1572 >            else if (m > 1) {                // move to a different index
1573 >                r ^= r << 13;                // same xorshift as WorkQueues
1574 >                r ^= r >>> 17;
1575 >                s.seed = r ^= r << 5;
1576 >            }
1577 >            else
1578 >                Thread.yield();              // yield if no alternatives
1579 >        }
1580 >    }
1581  
1582      /**
1583 <     * Add delta (which may be negative) to running count.  This must
223 <     * 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
1583 >     * Submits the given (non-null) task to the common pool, if possible.
1584       */
1585 <    final void updateRunningCount(int delta) {
1586 <        int s;
1587 <        do;while (!casWorkerCounts(s = workerCounts, s + delta));
1585 >    static void submitToCommonPool(ForkJoinTask<?> task) {
1586 >        ForkJoinPool p;
1587 >        if ((p = commonPool) == null)
1588 >            throw new RejectedExecutionException("Common Pool Unavailable");
1589 >        p.doSubmit(task);
1590      }
1591  
1592      /**
1593 <     * Add delta (which may be negative) to both total and running
1594 <     * count.  This must be called upon creation and termination of
1595 <     * worker threads.
236 <     * @param delta the number to add
1593 >     * Returns true if caller is (or may be) submitter to the common
1594 >     * pool, and not all workers are active, and there appear to be
1595 >     * tasks in the associated submission queue.
1596       */
1597 <    private void updateWorkerCount(int delta) {
1598 <        int d = delta + (delta << 16); // add to both lo and hi parts
1599 <        int s;
1600 <        do;while (!casWorkerCounts(s = workerCounts, s + d));
1597 >    static boolean canHelpCommonPool() {
1598 >        ForkJoinPool p; WorkQueue[] ws; WorkQueue q;
1599 >        int k = submitters.get().seed & SQMASK;
1600 >        return ((p = commonPool) != null &&
1601 >                (int)(p.ctl >> AC_SHIFT) < 0 &&
1602 >                (ws = p.workQueues) != null &&
1603 >                ws.length > (k &= p.submitMask) &&
1604 >                (q = ws[k]) != null &&
1605 >                q.top - q.base > 0);
1606      }
1607  
1608      /**
1609 <     * Lifecycle control. High word contains runState, low word
1610 <     * contains the number of workers that are (probably) executing
1611 <     * tasks. This value is atomically incremented before a worker
1612 <     * gets a task to run, and decremented when worker has no tasks
1613 <     * and cannot find any. These two fields are bundled together to
1614 <     * support correct termination triggering.  Note: activeCount
1615 <     * CAS'es cheat by assuming active count is in low word, so need
252 <     * to be modified if this changes
1609 >     * Returns true if the given task was submitted to common pool
1610 >     * and has not yet commenced execution, and is available for
1611 >     * removal according to execution policies; if so removing the
1612 >     * submission from the pool.
1613 >     *
1614 >     * @param task the task
1615 >     * @return true if successful
1616       */
1617 <    private volatile int runControl;
1617 >    static boolean tryUnsubmitFromCommonPool(ForkJoinTask<?> task) {
1618 >        // Peek, looking for task and eligibility before
1619 >        // using trySharedUnpush to actually take it under lock
1620 >        ForkJoinPool p; WorkQueue[] ws; WorkQueue q;
1621 >        ForkJoinTask<?>[] a; int s;
1622 >        int k = submitters.get().seed & SQMASK;
1623 >        return ((p = commonPool) != null &&
1624 >                (int)(p.ctl >> AC_SHIFT) < 0 &&
1625 >                (ws = p.workQueues) != null &&
1626 >                ws.length > (k &= p.submitMask) &&
1627 >                (q = ws[k]) != null &&
1628 >                (a = q.array) != null &&
1629 >                (s = q.top - 1) - q.base >= 0 &&
1630 >                s >= 0 && s < a.length &&
1631 >                a[s] == task &&
1632 >                q.trySharedUnpush(task));
1633 >    }
1634 >
1635 >    /**
1636 >     * Tries to pop a task from common pool with given root
1637 >     */
1638 >    static ForkJoinTask<?> popCCFromCommonPool(CountedCompleter<?> root) {
1639 >        ForkJoinPool p; WorkQueue[] ws; WorkQueue q;
1640 >        ForkJoinTask<?> t;
1641 >        int k = submitters.get().seed & SQMASK;
1642 >        if (root != null &&
1643 >            (p = commonPool) != null &&
1644 >            (int)(p.ctl >> AC_SHIFT) < 0 &&
1645 >            (ws = p.workQueues) != null &&
1646 >            ws.length > (k &= p.submitMask) &&
1647 >            (q = ws[k]) != null && q.top - q.base > 0 &&
1648 >            root.status < 0 &&
1649 >            (t = q.sharedPopCC(root)) != null)
1650 >            return t;
1651 >        return null;
1652 >    }
1653  
256    // RunState values. Order among values matters
257    private static final int RUNNING     = 0;
258    private static final int SHUTDOWN    = 1;
259    private static final int TERMINATING = 2;
260    private static final int TERMINATED  = 3;
1654  
1655 <    private static int runStateOf(int c)             { return c >>> 16; }
263 <    private static int activeCountOf(int c)          { return c & shortMask; }
264 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
1655 >    // Maintaining ctl counts
1656  
1657      /**
1658 <     * Try incrementing active count; fail on contention. Called by
268 <     * workers before/during executing tasks.
269 <     * @return true on success;
1658 >     * Increments active count; mainly called upon return from blocking.
1659       */
1660 <    final boolean tryIncrementActiveCount() {
1661 <        int c = runControl;
1662 <        return casRunControl(c, c+1);
1660 >    final void incrementActiveCount() {
1661 >        long c;
1662 >        do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
1663      }
1664  
1665      /**
1666 <     * Try decrementing active count; fail on contention.
278 <     * Possibly trigger termination on success
279 <     * Called by workers when they can't find tasks.
280 <     * @return true on success
1666 >     * Tries to create one or activate one or more workers if too few are active.
1667       */
1668 <    final boolean tryDecrementActiveCount() {
1669 <        int c = runControl;
1670 <        int nextc = c - 1;
1671 <        if (!casRunControl(c, nextc))
1672 <            return false;
1673 <        if (canTerminateOnShutdown(nextc))
1674 <            terminateOnShutdown();
1675 <        return true;
1668 >    final void signalWork() {
1669 >        long c; int u;
1670 >        while ((u = (int)((c = ctl) >>> 32)) < 0) {     // too few active
1671 >            WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1672 >            if ((e = (int)c) > 0) {                     // at least one waiting
1673 >                if (ws != null && (i = e & SMASK) < ws.length &&
1674 >                    (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1675 >                    long nc = (((long)(w.nextWait & E_MASK)) |
1676 >                               ((long)(u + UAC_UNIT) << 32));
1677 >                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
1678 >                        w.eventCount = (e + E_SEQ) & E_MASK;
1679 >                        if ((p = w.parker) != null)
1680 >                            U.unpark(p);                // activate and release
1681 >                        break;
1682 >                    }
1683 >                }
1684 >                else
1685 >                    break;
1686 >            }
1687 >            else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1688 >                long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1689 >                                 ((u + UAC_UNIT) & UAC_MASK)) << 32;
1690 >                if (U.compareAndSwapLong(this, CTL, c, nc)) {
1691 >                    addWorker();
1692 >                    break;
1693 >                }
1694 >            }
1695 >            else
1696 >                break;
1697 >        }
1698      }
1699  
1700 +    // Scanning for tasks
1701 +
1702      /**
1703 <     * Return true if argument represents zero active count and
294 <     * nonzero runstate, which is the triggering condition for
295 <     * terminating on shutdown.
1703 >     * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1704       */
1705 <    private static boolean canTerminateOnShutdown(int c) {
1706 <        return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
1705 >    final void runWorker(WorkQueue w) {
1706 >        w.growArray(false);         // initialize queue array in this thread
1707 >        do { w.runTask(scan(w)); } while (w.runState >= 0);
1708      }
1709  
1710      /**
1711 <     * Transition run state to at least the given state. Return true
1712 <     * if not already at least given state.
1711 >     * Scans for and, if found, returns one task, else possibly
1712 >     * inactivates the worker. This method operates on single reads of
1713 >     * volatile state and is designed to be re-invoked continuously,
1714 >     * in part because it returns upon detecting inconsistencies,
1715 >     * contention, or state changes that indicate possible success on
1716 >     * re-invocation.
1717 >     *
1718 >     * The scan searches for tasks across a random permutation of
1719 >     * queues (starting at a random index and stepping by a random
1720 >     * relative prime, checking each at least once).  The scan
1721 >     * terminates upon either finding a non-empty queue, or completing
1722 >     * the sweep. If the worker is not inactivated, it takes and
1723 >     * returns a task from this queue.  On failure to find a task, we
1724 >     * take one of the following actions, after which the caller will
1725 >     * retry calling this method unless terminated.
1726 >     *
1727 >     * * If pool is terminating, terminate the worker.
1728 >     *
1729 >     * * If not a complete sweep, try to release a waiting worker.  If
1730 >     * the scan terminated because the worker is inactivated, then the
1731 >     * released worker will often be the calling worker, and it can
1732 >     * succeed obtaining a task on the next call. Or maybe it is
1733 >     * another worker, but with same net effect. Releasing in other
1734 >     * cases as well ensures that we have enough workers running.
1735 >     *
1736 >     * * If not already enqueued, try to inactivate and enqueue the
1737 >     * worker on wait queue. Or, if inactivating has caused the pool
1738 >     * to be quiescent, relay to idleAwaitWork to check for
1739 >     * termination and possibly shrink pool.
1740 >     *
1741 >     * * If already inactive, and the caller has run a task since the
1742 >     * last empty scan, return (to allow rescan) unless others are
1743 >     * also inactivated.  Field WorkQueue.rescans counts down on each
1744 >     * scan to ensure eventual inactivation and blocking.
1745 >     *
1746 >     * * If already enqueued and none of the above apply, park
1747 >     * awaiting signal,
1748 >     *
1749 >     * @param w the worker (via its WorkQueue)
1750 >     * @return a task or null if none found
1751       */
1752 <    private boolean transitionRunStateTo(int state) {
1753 <        for (;;) {
1754 <            int c = runControl;
1755 <            if (runStateOf(c) >= state)
1756 <                return false;
1757 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
1758 <                return true;
1752 >    private final ForkJoinTask<?> scan(WorkQueue w) {
1753 >        WorkQueue[] ws;                       // first update random seed
1754 >        int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1755 >        int rs = runState, m;                 // volatile read order matters
1756 >        if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1757 >            int ec = w.eventCount;            // ec is negative if inactive
1758 >            int step = (r >>> 16) | 1;        // relative prime
1759 >            for (int j = (m + 1) << 2; ; r += step) {
1760 >                WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1761 >                if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1762 >                    (a = q.array) != null) {  // probably nonempty
1763 >                    int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1764 >                    t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1765 >                    if (q.base == b && ec >= 0 && t != null &&
1766 >                        U.compareAndSwapObject(a, i, t, null)) {
1767 >                        if (q.top - (q.base = b + 1) > 0)
1768 >                            signalWork();    // help pushes signal
1769 >                        return t;
1770 >                    }
1771 >                    else if (ec < 0 || j <= m) {
1772 >                        rs = 0;               // mark scan as imcomplete
1773 >                        break;                // caller can retry after release
1774 >                    }
1775 >                }
1776 >                if (--j < 0)
1777 >                    break;
1778 >            }
1779 >
1780 >            long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1781 >            if (e < 0)                        // decode ctl on empty scan
1782 >                w.runState = -1;              // pool is terminating
1783 >            else if (rs == 0 || rs != runState) { // incomplete scan
1784 >                WorkQueue v; Thread p;        // try to release a waiter
1785 >                if (e > 0 && a < 0 && w.eventCount == ec &&
1786 >                    (v = ws[e & m]) != null && v.eventCount == (e | INT_SIGN)) {
1787 >                    long nc = ((long)(v.nextWait & E_MASK) |
1788 >                               ((c + AC_UNIT) & (AC_MASK|TC_MASK)));
1789 >                    if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1790 >                        v.eventCount = (e + E_SEQ) & E_MASK;
1791 >                        if ((p = v.parker) != null)
1792 >                            U.unpark(p);
1793 >                    }
1794 >                }
1795 >            }
1796 >            else if (ec >= 0) {               // try to enqueue/inactivate
1797 >                long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
1798 >                w.nextWait = e;
1799 >                w.eventCount = ec | INT_SIGN; // mark as inactive
1800 >                if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc))
1801 >                    w.eventCount = ec;        // unmark on CAS failure
1802 >                else {
1803 >                    if ((ns = w.nsteals) != 0) {
1804 >                        w.nsteals = 0;        // set rescans if ran task
1805 >                        w.rescans = (a > 0) ? 0 : a + parallelism;
1806 >                        w.totalSteals += ns;
1807 >                    }
1808 >                    if (a == 1 - parallelism) // quiescent
1809 >                        idleAwaitWork(w, nc, c);
1810 >                }
1811 >            }
1812 >            else if (w.eventCount < 0) {      // already queued
1813 >                int ac = a + parallelism;
1814 >                if ((nr = w.rescans) > 0)     // continue rescanning
1815 >                    w.rescans = (ac < nr) ? ac : nr - 1;
1816 >                else if (((w.seed >>> 16) & ac) == 0) { // randomize park
1817 >                    Thread.interrupted();     // clear status
1818 >                    Thread wt = Thread.currentThread();
1819 >                    U.putObject(wt, PARKBLOCKER, this);
1820 >                    w.parker = wt;            // emulate LockSupport.park
1821 >                    if (w.eventCount < 0)     // recheck
1822 >                        U.park(false, 0L);
1823 >                    w.parker = null;
1824 >                    U.putObject(wt, PARKBLOCKER, null);
1825 >                }
1826 >            }
1827          }
1828 +        return null;
1829      }
1830  
1831      /**
1832 <     * Controls whether to add spares to maintain parallelism
1833 <     */
1834 <    private volatile boolean maintainsParallelism;
1832 >     * If inactivating worker w has caused the pool to become
1833 >     * quiescent, checks for pool termination, and, so long as this is
1834 >     * not the only worker, waits for event for up to a given
1835 >     * duration.  On timeout, if ctl has not changed, terminates the
1836 >     * worker, which will in turn wake up another worker to possibly
1837 >     * repeat this process.
1838 >     *
1839 >     * @param w the calling worker
1840 >     * @param currentCtl the ctl value triggering possible quiescence
1841 >     * @param prevCtl the ctl value to restore if thread is terminated
1842 >     */
1843 >    private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1844 >        if (w.eventCount < 0 && !tryTerminate(false, false) &&
1845 >            (int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1846 >            int dc = -(short)(currentCtl >>> TC_SHIFT);
1847 >            long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1848 >            long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1849 >            Thread wt = Thread.currentThread();
1850 >            while (ctl == currentCtl) {
1851 >                Thread.interrupted();  // timed variant of version in scan()
1852 >                U.putObject(wt, PARKBLOCKER, this);
1853 >                w.parker = wt;
1854 >                if (ctl == currentCtl)
1855 >                    U.park(false, parkTime);
1856 >                w.parker = null;
1857 >                U.putObject(wt, PARKBLOCKER, null);
1858 >                if (ctl != currentCtl)
1859 >                    break;
1860 >                if (deadline - System.nanoTime() <= 0L &&
1861 >                    U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1862 >                    w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1863 >                    w.runState = -1;   // shrink
1864 >                    break;
1865 >                }
1866 >            }
1867 >        }
1868 >    }
1869  
1870 <    // Constructors
1870 >    /**
1871 >     * Tries to locate and execute tasks for a stealer of the given
1872 >     * task, or in turn one of its stealers, Traces currentSteal ->
1873 >     * currentJoin links looking for a thread working on a descendant
1874 >     * of the given task and with a non-empty queue to steal back and
1875 >     * execute tasks from. The first call to this method upon a
1876 >     * waiting join will often entail scanning/search, (which is OK
1877 >     * because the joiner has nothing better to do), but this method
1878 >     * leaves hints in workers to speed up subsequent calls. The
1879 >     * implementation is very branchy to cope with potential
1880 >     * inconsistencies or loops encountering chains that are stale,
1881 >     * unknown, or so long that they are likely cyclic.
1882 >     *
1883 >     * @param joiner the joining worker
1884 >     * @param task the task to join
1885 >     * @return 0 if no progress can be made, negative if task
1886 >     * known complete, else positive
1887 >     */
1888 >    private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1889 >        int stat = 0, steps = 0;                    // bound to avoid cycles
1890 >        if (joiner != null && task != null) {       // hoist null checks
1891 >            restart: for (;;) {
1892 >                ForkJoinTask<?> subtask = task;     // current target
1893 >                for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
1894 >                    WorkQueue[] ws; int m, s, h;
1895 >                    if ((s = task.status) < 0) {
1896 >                        stat = s;
1897 >                        break restart;
1898 >                    }
1899 >                    if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1900 >                        break restart;              // shutting down
1901 >                    if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
1902 >                        v.currentSteal != subtask) {
1903 >                        for (int origin = h;;) {    // find stealer
1904 >                            if (((h = (h + 2) & m) & 15) == 1 &&
1905 >                                (subtask.status < 0 || j.currentJoin != subtask))
1906 >                                continue restart;   // occasional staleness check
1907 >                            if ((v = ws[h]) != null &&
1908 >                                v.currentSteal == subtask) {
1909 >                                j.stealHint = h;    // save hint
1910 >                                break;
1911 >                            }
1912 >                            if (h == origin)
1913 >                                break restart;      // cannot find stealer
1914 >                        }
1915 >                    }
1916 >                    for (;;) { // help stealer or descend to its stealer
1917 >                        ForkJoinTask[] a;  int b;
1918 >                        if (subtask.status < 0)     // surround probes with
1919 >                            continue restart;       //   consistency checks
1920 >                        if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
1921 >                            int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1922 >                            ForkJoinTask<?> t =
1923 >                                (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1924 >                            if (subtask.status < 0 || j.currentJoin != subtask ||
1925 >                                v.currentSteal != subtask)
1926 >                                continue restart;   // stale
1927 >                            stat = 1;               // apparent progress
1928 >                            if (t != null && v.base == b &&
1929 >                                U.compareAndSwapObject(a, i, t, null)) {
1930 >                                v.base = b + 1;     // help stealer
1931 >                                joiner.runSubtask(t);
1932 >                            }
1933 >                            else if (v.base == b && ++steps == MAX_HELP)
1934 >                                break restart;      // v apparently stalled
1935 >                        }
1936 >                        else {                      // empty -- try to descend
1937 >                            ForkJoinTask<?> next = v.currentJoin;
1938 >                            if (subtask.status < 0 || j.currentJoin != subtask ||
1939 >                                v.currentSteal != subtask)
1940 >                                continue restart;   // stale
1941 >                            else if (next == null || ++steps == MAX_HELP)
1942 >                                break restart;      // dead-end or maybe cyclic
1943 >                            else {
1944 >                                subtask = next;
1945 >                                j = v;
1946 >                                break;
1947 >                            }
1948 >                        }
1949 >                    }
1950 >                }
1951 >            }
1952 >        }
1953 >        return stat;
1954 >    }
1955  
1956      /**
1957 <     * Creates a ForkJoinPool with a pool size equal to the number of
1958 <     * processors available on the system and using the default
1959 <     * ForkJoinWorkerThreadFactory,
1960 <     * @throws SecurityException if a security manager exists and
327 <     *         the caller is not permitted to modify threads
328 <     *         because it does not hold {@link
329 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1957 >     * If task is at base of some steal queue, steals and executes it.
1958 >     *
1959 >     * @param joiner the joining worker
1960 >     * @param task the task
1961       */
1962 <    public ForkJoinPool() {
1963 <        this(Runtime.getRuntime().availableProcessors(),
1964 <             defaultForkJoinWorkerThreadFactory);
1962 >    private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1963 >        WorkQueue[] ws;
1964 >        if ((ws = workQueues) != null) {
1965 >            for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1966 >                WorkQueue q = ws[j];
1967 >                if (q != null && q.pollFor(task)) {
1968 >                    joiner.runSubtask(task);
1969 >                    break;
1970 >                }
1971 >            }
1972 >        }
1973      }
1974  
1975      /**
1976 <     * Creates a ForkJoinPool with the indicated parellelism level
1977 <     * threads, and using the default ForkJoinWorkerThreadFactory,
1978 <     * @param parallelism the number of worker threads
1979 <     * @throws IllegalArgumentException if parallelism less than or
1980 <     * equal to zero
1981 <     * @throws SecurityException if a security manager exists and
1982 <     *         the caller is not permitted to modify threads
1983 <     *         because it does not hold {@link
1984 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1985 <     */
1986 <    public ForkJoinPool(int parallelism) {
1987 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
1976 >     * Tries to decrement active count (sometimes implicitly) and
1977 >     * possibly release or create a compensating worker in preparation
1978 >     * for blocking. Fails on contention or termination. Otherwise,
1979 >     * adds a new thread if no idle workers are available and either
1980 >     * pool would become completely starved or: (at least half
1981 >     * starved, and fewer than 50% spares exist, and there is at least
1982 >     * one task apparently available). Even though the availability
1983 >     * check requires a full scan, it is worthwhile in reducing false
1984 >     * alarms.
1985 >     *
1986 >     * @param task if non-null, a task being waited for
1987 >     * @param blocker if non-null, a blocker being waited for
1988 >     * @return true if the caller can block, else should recheck and retry
1989 >     */
1990 >    final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1991 >        int pc = parallelism, e;
1992 >        long c = ctl;
1993 >        WorkQueue[] ws = workQueues;
1994 >        if ((e = (int)c) >= 0 && ws != null) {
1995 >            int u, a, ac, hc;
1996 >            int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1997 >            boolean replace = false;
1998 >            if ((a = u >> UAC_SHIFT) <= 0) {
1999 >                if ((ac = a + pc) <= 1)
2000 >                    replace = true;
2001 >                else if ((e > 0 || (task != null &&
2002 >                                    ac <= (hc = pc >>> 1) && tc < pc + hc))) {
2003 >                    WorkQueue w;
2004 >                    for (int j = 0; j < ws.length; ++j) {
2005 >                        if ((w = ws[j]) != null && !w.isEmpty()) {
2006 >                            replace = true;
2007 >                            break;   // in compensation range and tasks available
2008 >                        }
2009 >                    }
2010 >                }
2011 >            }
2012 >            if ((task == null || task.status >= 0) && // recheck need to block
2013 >                (blocker == null || !blocker.isReleasable()) && ctl == c) {
2014 >                if (!replace) {          // no compensation
2015 >                    long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
2016 >                    if (U.compareAndSwapLong(this, CTL, c, nc))
2017 >                        return true;
2018 >                }
2019 >                else if (e != 0) {       // release an idle worker
2020 >                    WorkQueue w; Thread p; int i;
2021 >                    if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
2022 >                        long nc = ((long)(w.nextWait & E_MASK) |
2023 >                                   (c & (AC_MASK|TC_MASK)));
2024 >                        if (w.eventCount == (e | INT_SIGN) &&
2025 >                            U.compareAndSwapLong(this, CTL, c, nc)) {
2026 >                            w.eventCount = (e + E_SEQ) & E_MASK;
2027 >                            if ((p = w.parker) != null)
2028 >                                U.unpark(p);
2029 >                            return true;
2030 >                        }
2031 >                    }
2032 >                }
2033 >                else if (tc < MAX_CAP) { // create replacement
2034 >                    long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
2035 >                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
2036 >                        addWorker();
2037 >                        return true;
2038 >                    }
2039 >                }
2040 >            }
2041 >        }
2042 >        return false;
2043      }
2044  
2045      /**
2046 <     * Creates a ForkJoinPool with parallelism equal to the number of
2047 <     * processors available on the system and using the given
2048 <     * ForkJoinWorkerThreadFactory,
2049 <     * @param factory the factory for creating new threads
2050 <     * @throws NullPointerException if factory is null
357 <     * @throws SecurityException if a security manager exists and
358 <     *         the caller is not permitted to modify threads
359 <     *         because it does not hold {@link
360 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
2046 >     * Helps and/or blocks until the given task is done.
2047 >     *
2048 >     * @param joiner the joining worker
2049 >     * @param task the task
2050 >     * @return task status on exit
2051       */
2052 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
2053 <        this(Runtime.getRuntime().availableProcessors(), factory);
2052 >    final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
2053 >        int s;
2054 >        if ((s = task.status) >= 0) {
2055 >            ForkJoinTask<?> prevJoin = joiner.currentJoin;
2056 >            joiner.currentJoin = task;
2057 >            long startTime = 0L;
2058 >            for (int k = 0;;) {
2059 >                if ((s = (joiner.isEmpty() ?           // try to help
2060 >                          tryHelpStealer(joiner, task) :
2061 >                          joiner.tryRemoveAndExec(task))) == 0 &&
2062 >                    (s = task.status) >= 0) {
2063 >                    if (k == 0) {
2064 >                        startTime = System.nanoTime();
2065 >                        tryPollForAndExec(joiner, task); // check uncommon case
2066 >                    }
2067 >                    else if ((k & (MAX_HELP - 1)) == 0 &&
2068 >                             System.nanoTime() - startTime >=
2069 >                             COMPENSATION_DELAY &&
2070 >                             tryCompensate(task, null)) {
2071 >                        if (task.trySetSignal()) {
2072 >                            synchronized (task) {
2073 >                                if (task.status >= 0) {
2074 >                                    try {                // see ForkJoinTask
2075 >                                        task.wait();     //  for explanation
2076 >                                    } catch (InterruptedException ie) {
2077 >                                    }
2078 >                                }
2079 >                                else
2080 >                                    task.notifyAll();
2081 >                            }
2082 >                        }
2083 >                        long c;                          // re-activate
2084 >                        do {} while (!U.compareAndSwapLong
2085 >                                     (this, CTL, c = ctl, c + AC_UNIT));
2086 >                    }
2087 >                }
2088 >                if (s < 0 || (s = task.status) < 0) {
2089 >                    joiner.currentJoin = prevJoin;
2090 >                    break;
2091 >                }
2092 >                else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
2093 >                    Thread.yield();                     // for politeness
2094 >            }
2095 >        }
2096 >        return s;
2097      }
2098  
2099      /**
2100 <     * Creates a ForkJoinPool with the given parallelism and factory.
2100 >     * Stripped-down variant of awaitJoin used by timed joins. Tries
2101 >     * to help join only while there is continuous progress. (Caller
2102 >     * will then enter a timed wait.)
2103       *
2104 <     * @param parallelism the targeted number of worker threads
2105 <     * @param factory the factory for creating new threads
2106 <     * @throws IllegalArgumentException if parallelism less than or
372 <     * equal to zero, or greater than implementation limit.
373 <     * @throws NullPointerException if factory is null
374 <     * @throws SecurityException if a security manager exists and
375 <     *         the caller is not permitted to modify threads
376 <     *         because it does not hold {@link
377 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
2104 >     * @param joiner the joining worker
2105 >     * @param task the task
2106 >     * @return task status on exit
2107       */
2108 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
2109 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
2110 <            throw new IllegalArgumentException();
2111 <        if (factory == null)
2112 <            throw new NullPointerException();
2113 <        checkPermission();
2114 <        this.factory = factory;
2115 <        this.parallelism = parallelism;
387 <        this.maxPoolSize = MAX_THREADS;
388 <        this.maintainsParallelism = true;
389 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
390 <        this.workerLock = new ReentrantLock();
391 <        this.termination = workerLock.newCondition();
392 <        this.stealCount = new AtomicLong();
393 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
394 <        createAndStartInitialWorkers(parallelism);
2108 >    final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
2109 >        int s;
2110 >        while ((s = task.status) >= 0 &&
2111 >               (joiner.isEmpty() ?
2112 >                tryHelpStealer(joiner, task) :
2113 >                joiner.tryRemoveAndExec(task)) != 0)
2114 >            ;
2115 >        return s;
2116      }
2117  
2118      /**
2119 <     * Create new worker using factory.
2120 <     * @param index the index to assign worker
2121 <     * @return new worker, or null of factory failed
2122 <     */
2123 <    private ForkJoinWorkerThread createWorker(int index) {
2124 <        Thread.UncaughtExceptionHandler h = ueh;
2125 <        ForkJoinWorkerThread w = factory.newThread(this);
2126 <        if (w != null) {
2127 <            w.poolIndex = index;
2128 <            w.setDaemon(true);
2129 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
2130 <            if (h != null)
2131 <                w.setUncaughtExceptionHandler(h);
2119 >     * Returns a (probably) non-empty steal queue, if one is found
2120 >     * during a random, then cyclic scan, else null.  This method must
2121 >     * be retried by caller if, by the time it tries to use the queue,
2122 >     * it is empty.
2123 >     */
2124 >    private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
2125 >        // Similar to loop in scan(), but ignoring submissions
2126 >        int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
2127 >        int step = (r >>> 16) | 1;
2128 >        for (WorkQueue[] ws;;) {
2129 >            int rs = runState, m;
2130 >            if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2131 >                return null;
2132 >            for (int j = (m + 1) << 2; ; r += step) {
2133 >                WorkQueue q = ws[((r << 1) | 1) & m];
2134 >                if (q != null && !q.isEmpty())
2135 >                    return q;
2136 >                else if (--j < 0) {
2137 >                    if (runState == rs)
2138 >                        return null;
2139 >                    break;
2140 >                }
2141 >            }
2142          }
412        return w;
2143      }
2144  
2145      /**
2146 <     * Return a good size for worker array given pool size.
2147 <     * Currently requires size to be a power of two.
2148 <     */
2149 <    private static int arraySizeFor(int ps) {
2150 <        return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
2146 >     * Runs tasks until {@code isQuiescent()}. We piggyback on
2147 >     * active count ctl maintenance, but rather than blocking
2148 >     * when tasks cannot be found, we rescan until all others cannot
2149 >     * find tasks either.
2150 >     */
2151 >    final void helpQuiescePool(WorkQueue w) {
2152 >        for (boolean active = true;;) {
2153 >            ForkJoinTask<?> localTask; // exhaust local queue
2154 >            while ((localTask = w.nextLocalTask()) != null)
2155 >                localTask.doExec();
2156 >            WorkQueue q = findNonEmptyStealQueue(w);
2157 >            if (q != null) {
2158 >                ForkJoinTask<?> t; int b;
2159 >                if (!active) {      // re-establish active count
2160 >                    long c;
2161 >                    active = true;
2162 >                    do {} while (!U.compareAndSwapLong
2163 >                                 (this, CTL, c = ctl, c + AC_UNIT));
2164 >                }
2165 >                if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2166 >                    w.runSubtask(t);
2167 >            }
2168 >            else {
2169 >                long c;
2170 >                if (active) {       // decrement active count without queuing
2171 >                    active = false;
2172 >                    do {} while (!U.compareAndSwapLong
2173 >                                 (this, CTL, c = ctl, c -= AC_UNIT));
2174 >                }
2175 >                else
2176 >                    c = ctl;        // re-increment on exit
2177 >                if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2178 >                    do {} while (!U.compareAndSwapLong
2179 >                                 (this, CTL, c = ctl, c + AC_UNIT));
2180 >                    break;
2181 >                }
2182 >            }
2183 >        }
2184      }
2185  
2186      /**
2187 <     * Create or resize array if necessary to hold newLength
425 <     * @return the array
2187 >     * Restricted version of helpQuiescePool for non-FJ callers
2188       */
2189 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
2190 <        ForkJoinWorkerThread[] ws = workers;
2191 <        if (ws == null)
2192 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
2193 <        else if (newLength > ws.length)
2194 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
2195 <        else
2196 <            return ws;
2189 >    static void externalHelpQuiescePool() {
2190 >        ForkJoinPool p; WorkQueue[] ws; WorkQueue q, sq;
2191 >        ForkJoinTask<?>[] a; int b;
2192 >        ForkJoinTask<?> t = null;
2193 >        int k = submitters.get().seed & SQMASK;
2194 >        if ((p = commonPool) != null &&
2195 >            (int)(p.ctl >> AC_SHIFT) < 0 &&
2196 >            (ws = p.workQueues) != null &&
2197 >            ws.length > (k &= p.submitMask) &&
2198 >            (q = ws[k]) != null) {
2199 >            while (q.top - q.base > 0) {
2200 >                if ((t = q.sharedPop()) != null)
2201 >                    break;
2202 >            }
2203 >            if (t == null && (sq = p.findNonEmptyStealQueue(q)) != null &&
2204 >                (b = sq.base) - sq.top < 0)
2205 >                t = sq.pollAt(b);
2206 >            if (t != null)
2207 >                t.doExec();
2208 >        }
2209      }
2210  
2211      /**
2212 <     * Try to shrink workers into smaller array after one or more terminate
2212 >     * Gets and removes a local or stolen task for the given worker.
2213 >     *
2214 >     * @return a task, if available
2215       */
2216 <    private void tryShrinkWorkerArray() {
2217 <        ForkJoinWorkerThread[] ws = workers;
2218 <        int len = ws.length;
2219 <        int last = len - 1;
2220 <        while (last >= 0 && ws[last] == null)
2221 <            --last;
2222 <        int newLength = arraySizeFor(last+1);
2223 <        if (newLength < len)
2224 <            workers = Arrays.copyOf(ws, newLength);
2216 >    final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
2217 >        for (ForkJoinTask<?> t;;) {
2218 >            WorkQueue q; int b;
2219 >            if ((t = w.nextLocalTask()) != null)
2220 >                return t;
2221 >            if ((q = findNonEmptyStealQueue(w)) == null)
2222 >                return null;
2223 >            if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2224 >                return t;
2225 >        }
2226      }
2227  
2228      /**
2229 <     * Initial worker array and worker creation and startup. (This
2230 <     * must be done under lock to avoid interference by some of the
2231 <     * newly started threads while creating others.)
2229 >     * Returns the approximate (non-atomic) number of idle threads per
2230 >     * active thread to offset steal queue size for method
2231 >     * ForkJoinTask.getSurplusQueuedTaskCount().
2232       */
2233 <    private void createAndStartInitialWorkers(int ps) {
2234 <        final ReentrantLock lock = this.workerLock;
2235 <        lock.lock();
2236 <        try {
2237 <            ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
2238 <            for (int i = 0; i < ps; ++i) {
2239 <                ForkJoinWorkerThread w = createWorker(i);
2240 <                if (w != null) {
2241 <                    ws[i] = w;
465 <                    w.start();
466 <                    updateWorkerCount(1);
467 <                }
468 <            }
469 <        } finally {
470 <            lock.unlock();
471 <        }
2233 >    final int idlePerActive() {
2234 >        // Approximate at powers of two for small values, saturate past 4
2235 >        int p = parallelism;
2236 >        int a = p + (int)(ctl >> AC_SHIFT);
2237 >        return (a > (p >>>= 1) ? 0 :
2238 >                a > (p >>>= 1) ? 1 :
2239 >                a > (p >>>= 1) ? 2 :
2240 >                a > (p >>>= 1) ? 4 :
2241 >                8);
2242      }
2243  
2244      /**
2245 <     * Worker creation and startup for threads added via setParallelism.
2245 >     * Returns approximate submission queue length for the given caller
2246       */
2247 <    private void createAndStartAddedWorkers() {
2248 <        resumeAllSpares();  // Allow spares to convert to nonspare
2249 <        int ps = parallelism;
2250 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
2251 <        int len = ws.length;
2252 <        // Sweep through slots, to keep lowest indices most populated
2253 <        int k = 0;
2254 <        while (k < len) {
2255 <            if (ws[k] != null) {
2256 <                ++k;
2257 <                continue;
2247 >    static int getEstimatedSubmitterQueueLength() {
2248 >        ForkJoinPool p; WorkQueue[] ws; WorkQueue q;
2249 >        int k = submitters.get().seed & SQMASK;
2250 >        return ((p = commonPool) != null &&
2251 >                p.runState >= 0 &&
2252 >                (ws = p.workQueues) != null &&
2253 >                ws.length > (k &= p.submitMask) &&
2254 >                (q = ws[k]) != null) ?
2255 >            q.queueSize() : 0;
2256 >    }
2257 >
2258 >    //  Termination
2259 >
2260 >    /**
2261 >     * Possibly initiates and/or completes termination.  The caller
2262 >     * triggering termination runs three passes through workQueues:
2263 >     * (0) Setting termination status, followed by wakeups of queued
2264 >     * workers; (1) cancelling all tasks; (2) interrupting lagging
2265 >     * threads (likely in external tasks, but possibly also blocked in
2266 >     * joins).  Each pass repeats previous steps because of potential
2267 >     * lagging thread creation.
2268 >     *
2269 >     * @param now if true, unconditionally terminate, else only
2270 >     * if no work and no active workers
2271 >     * @param enable if true, enable shutdown when next possible
2272 >     * @return true if now terminating or terminated
2273 >     */
2274 >    private boolean tryTerminate(boolean now, boolean enable) {
2275 >        for (long c;;) {
2276 >            if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2277 >                if ((short)(c >>> TC_SHIFT) == -parallelism) {
2278 >                    synchronized(this) {
2279 >                        notifyAll();                // signal when 0 workers
2280 >                    }
2281 >                }
2282 >                return true;
2283              }
2284 <            int s = workerCounts;
2285 <            int tc = totalCountOf(s);
2286 <            int rc = runningCountOf(s);
2287 <            if (rc >= ps || tc >= ps)
2288 <                break;
2289 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
2290 <                ForkJoinWorkerThread w = createWorker(k);
2291 <                if (w != null) {
2292 <                    ws[k++] = w;
2293 <                    w.start();
2284 >            if (runState >= 0) {                    // not yet enabled
2285 >                if (!enable)
2286 >                    return false;
2287 >                while (!U.compareAndSwapInt(this, MAINLOCK, 0, 1))
2288 >                    tryAwaitMainLock();
2289 >                try {
2290 >                    runState |= SHUTDOWN;
2291 >                } finally {
2292 >                    if (!U.compareAndSwapInt(this, MAINLOCK, 1, 0)) {
2293 >                        mainLock = 0;
2294 >                        synchronized (this) { notifyAll(); };
2295 >                    }
2296                  }
2297 <                else {
2298 <                    updateWorkerCount(-1); // back out on failed creation
2299 <                    break;
2297 >            }
2298 >            if (!now) {                             // check if idle & no tasks
2299 >                if ((int)(c >> AC_SHIFT) != -parallelism ||
2300 >                    hasQueuedSubmissions())
2301 >                    return false;
2302 >                // Check for unqueued inactive workers. One pass suffices.
2303 >                WorkQueue[] ws = workQueues; WorkQueue w;
2304 >                if (ws != null) {
2305 >                    for (int i = 1; i < ws.length; i += 2) {
2306 >                        if ((w = ws[i]) != null && w.eventCount >= 0)
2307 >                            return false;
2308 >                    }
2309 >                }
2310 >            }
2311 >            if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2312 >                for (int pass = 0; pass < 3; ++pass) {
2313 >                    WorkQueue[] ws = workQueues;
2314 >                    if (ws != null) {
2315 >                        WorkQueue w;
2316 >                        int n = ws.length;
2317 >                        for (int i = 0; i < n; ++i) {
2318 >                            if ((w = ws[i]) != null) {
2319 >                                w.runState = -1;
2320 >                                if (pass > 0) {
2321 >                                    w.cancelAll();
2322 >                                    if (pass > 1)
2323 >                                        w.interruptOwner();
2324 >                                }
2325 >                            }
2326 >                        }
2327 >                        // Wake up workers parked on event queue
2328 >                        int i, e; long cc; Thread p;
2329 >                        while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2330 >                               (i = e & SMASK) < n &&
2331 >                               (w = ws[i]) != null) {
2332 >                            long nc = ((long)(w.nextWait & E_MASK) |
2333 >                                       ((cc + AC_UNIT) & AC_MASK) |
2334 >                                       (cc & (TC_MASK|STOP_BIT)));
2335 >                            if (w.eventCount == (e | INT_SIGN) &&
2336 >                                U.compareAndSwapLong(this, CTL, cc, nc)) {
2337 >                                w.eventCount = (e + E_SEQ) & E_MASK;
2338 >                                w.runState = -1;
2339 >                                if ((p = w.parker) != null)
2340 >                                    U.unpark(p);
2341 >                            }
2342 >                        }
2343 >                    }
2344                  }
2345              }
2346          }
2347      }
2348  
2349 <    // Execution methods
2349 >    // Exported methods
2350 >
2351 >    // Constructors
2352  
2353      /**
2354 <     * Common code for execute, invoke and submit
2354 >     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
2355 >     * java.lang.Runtime#availableProcessors}, using the {@linkplain
2356 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
2357 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
2358 >     *
2359 >     * @throws SecurityException if a security manager exists and
2360 >     *         the caller is not permitted to modify threads
2361 >     *         because it does not hold {@link
2362 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
2363       */
2364 <    private <T> void doSubmit(ForkJoinTask<T> task) {
2365 <        if (isShutdown())
2366 <            throw new RejectedExecutionException();
516 <        submissionQueue.offer(task);
517 <        signalIdleWorkers();
2364 >    public ForkJoinPool() {
2365 >        this(Runtime.getRuntime().availableProcessors(),
2366 >             defaultForkJoinWorkerThreadFactory, null, false);
2367      }
2368  
2369      /**
2370 <     * Performs the given task; returning its result upon completion
2370 >     * Creates a {@code ForkJoinPool} with the indicated parallelism
2371 >     * level, the {@linkplain
2372 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
2373 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
2374 >     *
2375 >     * @param parallelism the parallelism level
2376 >     * @throws IllegalArgumentException if parallelism less than or
2377 >     *         equal to zero, or greater than implementation limit
2378 >     * @throws SecurityException if a security manager exists and
2379 >     *         the caller is not permitted to modify threads
2380 >     *         because it does not hold {@link
2381 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
2382 >     */
2383 >    public ForkJoinPool(int parallelism) {
2384 >        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
2385 >    }
2386 >
2387 >    /**
2388 >     * Creates a {@code ForkJoinPool} with the given parameters.
2389 >     *
2390 >     * @param parallelism the parallelism level. For default value,
2391 >     * use {@link java.lang.Runtime#availableProcessors}.
2392 >     * @param factory the factory for creating new threads. For default value,
2393 >     * use {@link #defaultForkJoinWorkerThreadFactory}.
2394 >     * @param handler the handler for internal worker threads that
2395 >     * terminate due to unrecoverable errors encountered while executing
2396 >     * tasks. For default value, use {@code null}.
2397 >     * @param asyncMode if true,
2398 >     * establishes local first-in-first-out scheduling mode for forked
2399 >     * tasks that are never joined. This mode may be more appropriate
2400 >     * than default locally stack-based mode in applications in which
2401 >     * worker threads only process event-style asynchronous tasks.
2402 >     * For default value, use {@code false}.
2403 >     * @throws IllegalArgumentException if parallelism less than or
2404 >     *         equal to zero, or greater than implementation limit
2405 >     * @throws NullPointerException if the factory is null
2406 >     * @throws SecurityException if a security manager exists and
2407 >     *         the caller is not permitted to modify threads
2408 >     *         because it does not hold {@link
2409 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
2410 >     */
2411 >    public ForkJoinPool(int parallelism,
2412 >                        ForkJoinWorkerThreadFactory factory,
2413 >                        Thread.UncaughtExceptionHandler handler,
2414 >                        boolean asyncMode) {
2415 >        checkPermission();
2416 >        if (factory == null)
2417 >            throw new NullPointerException();
2418 >        if (parallelism <= 0 || parallelism > MAX_CAP)
2419 >            throw new IllegalArgumentException();
2420 >        this.parallelism = parallelism;
2421 >        this.factory = factory;
2422 >        this.ueh = handler;
2423 >        this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2424 >        long np = (long)(-parallelism); // offset ctl counts
2425 >        this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2426 >        // Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2427 >        int n = parallelism - 1;
2428 >        n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
2429 >        this.submitMask = ((n + 1) << 1) - 1;
2430 >        int pn = poolNumberGenerator.incrementAndGet();
2431 >        StringBuilder sb = new StringBuilder("ForkJoinPool-");
2432 >        sb.append(Integer.toString(pn));
2433 >        sb.append("-worker-");
2434 >        this.workerNamePrefix = sb.toString();
2435 >        this.runState = 1;              // set init flag
2436 >    }
2437 >
2438 >    /**
2439 >     * Constructor for common pool, suitable only for static initialization.
2440 >     * Basically the same as above, but uses smallest possible initial footprint.
2441 >     */
2442 >    ForkJoinPool(int parallelism, int submitMask,
2443 >                 ForkJoinWorkerThreadFactory factory,
2444 >                 Thread.UncaughtExceptionHandler handler) {
2445 >        this.factory = factory;
2446 >        this.ueh = handler;
2447 >        this.submitMask = submitMask;
2448 >        this.parallelism = parallelism;
2449 >        long np = (long)(-parallelism);
2450 >        this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2451 >        this.localMode = LIFO_QUEUE;
2452 >        this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2453 >        this.runState = 1;
2454 >    }
2455 >
2456 >    /**
2457 >     * Returns the common pool instance.
2458 >     *
2459 >     * @return the common pool instance
2460 >     */
2461 >    public static ForkJoinPool commonPool() {
2462 >        ForkJoinPool p;
2463 >        if ((p = commonPool) == null)
2464 >            throw new Error("Common Pool Unavailable");
2465 >        return p;
2466 >    }
2467 >
2468 >    // Execution methods
2469 >
2470 >    /**
2471 >     * Performs the given task, returning its result upon completion.
2472 >     * If the computation encounters an unchecked Exception or Error,
2473 >     * it is rethrown as the outcome of this invocation.  Rethrown
2474 >     * exceptions behave in the same way as regular exceptions, but,
2475 >     * when possible, contain stack traces (as displayed for example
2476 >     * using {@code ex.printStackTrace()}) of both the current thread
2477 >     * as well as the thread actually encountering the exception;
2478 >     * minimally only the latter.
2479 >     *
2480       * @param task the task
2481       * @return the task's result
2482 <     * @throws NullPointerException if task is null
2483 <     * @throws RejectedExecutionException if pool is shut down
2482 >     * @throws NullPointerException if the task is null
2483 >     * @throws RejectedExecutionException if the task cannot be
2484 >     *         scheduled for execution
2485       */
2486      public <T> T invoke(ForkJoinTask<T> task) {
2487 +        if (task == null)
2488 +            throw new NullPointerException();
2489          doSubmit(task);
2490          return task.join();
2491      }
2492  
2493      /**
2494       * Arranges for (asynchronous) execution of the given task.
2495 +     *
2496       * @param task the task
2497 <     * @throws NullPointerException if task is null
2498 <     * @throws RejectedExecutionException if pool is shut down
2497 >     * @throws NullPointerException if the task is null
2498 >     * @throws RejectedExecutionException if the task cannot be
2499 >     *         scheduled for execution
2500       */
2501 <    public <T> void execute(ForkJoinTask<T> task) {
2501 >    public void execute(ForkJoinTask<?> task) {
2502 >        if (task == null)
2503 >            throw new NullPointerException();
2504          doSubmit(task);
2505      }
2506  
2507      // AbstractExecutorService methods
2508  
2509 +    /**
2510 +     * @throws NullPointerException if the task is null
2511 +     * @throws RejectedExecutionException if the task cannot be
2512 +     *         scheduled for execution
2513 +     */
2514      public void execute(Runnable task) {
2515 <        doSubmit(new AdaptedRunnable<Void>(task, null));
2515 >        if (task == null)
2516 >            throw new NullPointerException();
2517 >        ForkJoinTask<?> job;
2518 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2519 >            job = (ForkJoinTask<?>) task;
2520 >        else
2521 >            job = new ForkJoinTask.AdaptedRunnableAction(task);
2522 >        doSubmit(job);
2523 >    }
2524 >
2525 >    /**
2526 >     * Submits a ForkJoinTask for execution.
2527 >     *
2528 >     * @param task the task to submit
2529 >     * @return the task
2530 >     * @throws NullPointerException if the task is null
2531 >     * @throws RejectedExecutionException if the task cannot be
2532 >     *         scheduled for execution
2533 >     */
2534 >    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2535 >        if (task == null)
2536 >            throw new NullPointerException();
2537 >        doSubmit(task);
2538 >        return task;
2539      }
2540  
2541 +    /**
2542 +     * @throws NullPointerException if the task is null
2543 +     * @throws RejectedExecutionException if the task cannot be
2544 +     *         scheduled for execution
2545 +     */
2546      public <T> ForkJoinTask<T> submit(Callable<T> task) {
2547 <        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
2547 >        ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2548          doSubmit(job);
2549          return job;
2550      }
2551  
2552 +    /**
2553 +     * @throws NullPointerException if the task is null
2554 +     * @throws RejectedExecutionException if the task cannot be
2555 +     *         scheduled for execution
2556 +     */
2557      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2558 <        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
2558 >        ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2559          doSubmit(job);
2560          return job;
2561      }
2562  
2563 +    /**
2564 +     * @throws NullPointerException if the task is null
2565 +     * @throws RejectedExecutionException if the task cannot be
2566 +     *         scheduled for execution
2567 +     */
2568      public ForkJoinTask<?> submit(Runnable task) {
2569 <        ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
2569 >        if (task == null)
2570 >            throw new NullPointerException();
2571 >        ForkJoinTask<?> job;
2572 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2573 >            job = (ForkJoinTask<?>) task;
2574 >        else
2575 >            job = new ForkJoinTask.AdaptedRunnableAction(task);
2576          doSubmit(job);
2577          return job;
2578      }
2579  
2580      /**
2581 <     * Adaptor for Runnables. This implements RunnableFuture
2582 <     * to be compliant with AbstractExecutorService constraints
2581 >     * @throws NullPointerException       {@inheritDoc}
2582 >     * @throws RejectedExecutionException {@inheritDoc}
2583       */
570    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
571        implements RunnableFuture<T> {
572        final Runnable runnable;
573        final T resultOnCompletion;
574        T result;
575        AdaptedRunnable(Runnable runnable, T result) {
576            if (runnable == null) throw new NullPointerException();
577            this.runnable = runnable;
578            this.resultOnCompletion = result;
579        }
580        public T getRawResult() { return result; }
581        public void setRawResult(T v) { result = v; }
582        public boolean exec() {
583            runnable.run();
584            result = resultOnCompletion;
585            return true;
586        }
587        public void run() { invoke(); }
588    }
589
590    /**
591     * Adaptor for Callables
592     */
593    static final class AdaptedCallable<T> extends ForkJoinTask<T>
594        implements RunnableFuture<T> {
595        final Callable<T> callable;
596        T result;
597        AdaptedCallable(Callable<T> callable) {
598            if (callable == null) throw new NullPointerException();
599            this.callable = callable;
600        }
601        public T getRawResult() { return result; }
602        public void setRawResult(T v) { result = v; }
603        public boolean exec() {
604            try {
605                result = callable.call();
606                return true;
607            } catch (Error err) {
608                throw err;
609            } catch (RuntimeException rex) {
610                throw rex;
611            } catch (Exception ex) {
612                throw new RuntimeException(ex);
613            }
614        }
615        public void run() { invoke(); }
616    }
617
2584      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2585 <        ArrayList<ForkJoinTask<T>> ts =
2586 <            new ArrayList<ForkJoinTask<T>>(tasks.size());
2587 <        for (Callable<T> c : tasks)
2588 <            ts.add(new AdaptedCallable<T>(c));
2589 <        invoke(new InvokeAll<T>(ts));
2590 <        return (List<Future<T>>)(List)ts;
2591 <    }
2585 >        // In previous versions of this class, this method constructed
2586 >        // a task to run ForkJoinTask.invokeAll, but now external
2587 >        // invocation of multiple tasks is at least as efficient.
2588 >        List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2589 >        // Workaround needed because method wasn't declared with
2590 >        // wildcards in return type but should have been.
2591 >        @SuppressWarnings({"unchecked", "rawtypes"})
2592 >            List<Future<T>> futures = (List<Future<T>>) (List) fs;
2593  
2594 <    static final class InvokeAll<T> extends RecursiveAction {
2595 <        final ArrayList<ForkJoinTask<T>> tasks;
2596 <        InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
2597 <        public void compute() {
2598 <            try { invokeAll(tasks); } catch(Exception ignore) {}
2594 >        boolean done = false;
2595 >        try {
2596 >            for (Callable<T> t : tasks) {
2597 >                ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2598 >                doSubmit(f);
2599 >                fs.add(f);
2600 >            }
2601 >            for (ForkJoinTask<T> f : fs)
2602 >                f.quietlyJoin();
2603 >            done = true;
2604 >            return futures;
2605 >        } finally {
2606 >            if (!done)
2607 >                for (ForkJoinTask<T> f : fs)
2608 >                    f.cancel(false);
2609          }
2610      }
2611  
635    // Configuration and status settings and queries
636
2612      /**
2613 <     * Returns the factory used for constructing new workers
2613 >     * Returns the factory used for constructing new workers.
2614       *
2615       * @return the factory used for constructing new workers
2616       */
# Line 646 | Line 2621 | public class ForkJoinPool extends Abstra
2621      /**
2622       * Returns the handler for internal worker threads that terminate
2623       * due to unrecoverable errors encountered while executing tasks.
2624 <     * @return the handler, or null if none
2624 >     *
2625 >     * @return the handler, or {@code null} if none
2626       */
2627      public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
2628 <        Thread.UncaughtExceptionHandler h;
653 <        final ReentrantLock lock = this.workerLock;
654 <        lock.lock();
655 <        try {
656 <            h = ueh;
657 <        } finally {
658 <            lock.unlock();
659 <        }
660 <        return h;
2628 >        return ueh;
2629      }
2630  
2631      /**
2632 <     * Sets the handler for internal worker threads that terminate due
665 <     * to unrecoverable errors encountered while executing tasks.
666 <     * Unless set, the current default or ThreadGroup handler is used
667 <     * as handler.
2632 >     * Returns the targeted parallelism level of this pool.
2633       *
2634 <     * @param h the new handler
670 <     * @return the old handler, or null if none
671 <     * @throws SecurityException if a security manager exists and
672 <     *         the caller is not permitted to modify threads
673 <     *         because it does not hold {@link
674 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
675 <     */
676 <    public Thread.UncaughtExceptionHandler
677 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
678 <        checkPermission();
679 <        Thread.UncaughtExceptionHandler old = null;
680 <        final ReentrantLock lock = this.workerLock;
681 <        lock.lock();
682 <        try {
683 <            old = ueh;
684 <            ueh = h;
685 <            ForkJoinWorkerThread[] ws = workers;
686 <            for (int i = 0; i < ws.length; ++i) {
687 <                ForkJoinWorkerThread w = ws[i];
688 <                if (w != null)
689 <                    w.setUncaughtExceptionHandler(h);
690 <            }
691 <        } finally {
692 <            lock.unlock();
693 <        }
694 <        return old;
695 <    }
696 <
697 <
698 <    /**
699 <     * Sets the target paralleism level of this pool.
700 <     * @param parallelism the target parallelism
701 <     * @throws IllegalArgumentException if parallelism less than or
702 <     * equal to zero or greater than maximum size bounds.
703 <     * @throws SecurityException if a security manager exists and
704 <     *         the caller is not permitted to modify threads
705 <     *         because it does not hold {@link
706 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
2634 >     * @return the targeted parallelism level of this pool
2635       */
2636 <    public void setParallelism(int parallelism) {
2637 <        checkPermission();
710 <        if (parallelism <= 0 || parallelism > maxPoolSize)
711 <            throw new IllegalArgumentException();
712 <        final ReentrantLock lock = this.workerLock;
713 <        lock.lock();
714 <        try {
715 <            if (!isTerminating()) {
716 <                int p = this.parallelism;
717 <                this.parallelism = parallelism;
718 <                if (parallelism > p)
719 <                    createAndStartAddedWorkers();
720 <                else
721 <                    trimSpares();
722 <            }
723 <        } finally {
724 <            lock.unlock();
725 <        }
726 <        signalIdleWorkers();
2636 >    public int getParallelism() {
2637 >        return parallelism;
2638      }
2639  
2640      /**
2641 <     * Returns the targeted number of worker threads in this pool.
2641 >     * Returns the targeted parallelism level of the common pool.
2642       *
2643 <     * @return the targeted number of worker threads in this pool
2643 >     * @return the targeted parallelism level of the common pool
2644       */
2645 <    public int getParallelism() {
2646 <        return parallelism;
2645 >    public static int getCommonPoolParallelism() {
2646 >        return commonPoolParallelism;
2647      }
2648  
2649      /**
2650       * Returns the number of worker threads that have started but not
2651 <     * yet terminated.  This result returned by this method may differ
2652 <     * from <code>getParallelism</code> when threads are created to
2651 >     * yet terminated.  The result returned by this method may differ
2652 >     * from {@link #getParallelism} when threads are created to
2653       * maintain parallelism when others are cooperatively blocked.
2654       *
2655       * @return the number of worker threads
2656       */
2657      public int getPoolSize() {
2658 <        return totalCountOf(workerCounts);
2658 >        return parallelism + (short)(ctl >>> TC_SHIFT);
2659      }
2660  
2661      /**
2662 <     * Returns the maximum number of threads allowed to exist in the
2663 <     * pool, even if there are insufficient unblocked running threads.
2664 <     * @return the maximum
2665 <     */
755 <    public int getMaximumPoolSize() {
756 <        return maxPoolSize;
757 <    }
758 <
759 <    /**
760 <     * Sets the maximum number of threads allowed to exist in the
761 <     * pool, even if there are insufficient unblocked running threads.
762 <     * Setting this value has no effect on current pool size. It
763 <     * controls construction of new threads.
764 <     * @throws IllegalArgumentException if negative or greater then
765 <     * internal implementation limit.
766 <     */
767 <    public void setMaximumPoolSize(int newMax) {
768 <        if (newMax < 0 || newMax > MAX_THREADS)
769 <            throw new IllegalArgumentException();
770 <        maxPoolSize = newMax;
771 <    }
772 <
773 <
774 <    /**
775 <     * Returns true if this pool dynamically maintains its target
776 <     * parallelism level. If false, new threads are added only to
777 <     * avoid possible starvation.
778 <     * This setting is by default true;
779 <     * @return true if maintains parallelism
780 <     */
781 <    public boolean getMaintainsParallelism() {
782 <        return maintainsParallelism;
783 <    }
784 <
785 <    /**
786 <     * Sets whether this pool dynamically maintains its target
787 <     * parallelism level. If false, new threads are added only to
788 <     * avoid possible starvation.
789 <     * @param enable true to maintains parallelism
2662 >     * Returns {@code true} if this pool uses local first-in-first-out
2663 >     * scheduling mode for forked tasks that are never joined.
2664 >     *
2665 >     * @return {@code true} if this pool uses async mode
2666       */
2667 <    public void setMaintainsParallelism(boolean enable) {
2668 <        maintainsParallelism = enable;
2667 >    public boolean getAsyncMode() {
2668 >        return localMode != 0;
2669      }
2670  
2671      /**
2672       * Returns an estimate of the number of worker threads that are
2673       * not blocked waiting to join tasks or for other managed
2674 <     * synchronization.
2674 >     * synchronization. This method may overestimate the
2675 >     * number of running threads.
2676       *
2677       * @return the number of worker threads
2678       */
2679      public int getRunningThreadCount() {
2680 <        return runningCountOf(workerCounts);
2680 >        int rc = 0;
2681 >        WorkQueue[] ws; WorkQueue w;
2682 >        if ((ws = workQueues) != null) {
2683 >            for (int i = 1; i < ws.length; i += 2) {
2684 >                if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2685 >                    ++rc;
2686 >            }
2687 >        }
2688 >        return rc;
2689      }
2690  
2691      /**
2692       * Returns an estimate of the number of threads that are currently
2693       * stealing or executing tasks. This method may overestimate the
2694       * number of active threads.
2695 <     * @return the number of active threads.
2695 >     *
2696 >     * @return the number of active threads
2697       */
2698      public int getActiveThreadCount() {
2699 <        return activeCountOf(runControl);
2699 >        int r = parallelism + (int)(ctl >> AC_SHIFT);
2700 >        return (r <= 0) ? 0 : r; // suppress momentarily negative values
2701      }
2702  
2703      /**
2704 <     * Returns an estimate of the number of threads that are currently
2705 <     * idle waiting for tasks. This method may underestimate the
2706 <     * number of idle threads.
2707 <     * @return the number of idle threads.
2708 <     */
2709 <    final int getIdleThreadCount() {
2710 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
2711 <        return (c <= 0)? 0 : c;
2712 <    }
826 <
827 <    /**
828 <     * Returns true if all worker threads are currently idle. An idle
829 <     * worker is one that cannot obtain a task to execute because none
830 <     * are available to steal from other threads, and there are no
831 <     * pending submissions to the pool. This method is conservative:
832 <     * It might not return true immediately upon idleness of all
833 <     * threads, but will eventually become true if threads remain
834 <     * inactive.
835 <     * @return true if all threads are currently idle
2704 >     * Returns {@code true} if all worker threads are currently idle.
2705 >     * An idle worker is one that cannot obtain a task to execute
2706 >     * because none are available to steal from other threads, and
2707 >     * there are no pending submissions to the pool. This method is
2708 >     * conservative; it might not return {@code true} immediately upon
2709 >     * idleness of all threads, but will eventually become true if
2710 >     * threads remain inactive.
2711 >     *
2712 >     * @return {@code true} if all threads are currently idle
2713       */
2714      public boolean isQuiescent() {
2715 <        return activeCountOf(runControl) == 0;
2715 >        return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2716      }
2717  
2718      /**
# Line 843 | Line 2720 | public class ForkJoinPool extends Abstra
2720       * one thread's work queue by another. The reported value
2721       * underestimates the actual total number of steals when the pool
2722       * is not quiescent. This value may be useful for monitoring and
2723 <     * tuning fork/join programs: In general, steal counts should be
2723 >     * tuning fork/join programs: in general, steal counts should be
2724       * high enough to keep threads busy, but low enough to avoid
2725       * overhead and contention across threads.
2726 <     * @return the number of steals.
2726 >     *
2727 >     * @return the number of steals
2728       */
2729      public long getStealCount() {
2730 <        return stealCount.get();
2731 <    }
2732 <
2733 <    /**
2734 <     * Accumulate steal count from a worker. Call only
2735 <     * when worker known to be idle.
2736 <     */
2737 <    private void updateStealCount(ForkJoinWorkerThread w) {
2738 <        int sc = w.getAndClearStealCount();
861 <        if (sc != 0)
862 <            stealCount.addAndGet(sc);
2730 >        long count = stealCount;
2731 >        WorkQueue[] ws; WorkQueue w;
2732 >        if ((ws = workQueues) != null) {
2733 >            for (int i = 1; i < ws.length; i += 2) {
2734 >                if ((w = ws[i]) != null)
2735 >                    count += w.totalSteals;
2736 >            }
2737 >        }
2738 >        return count;
2739      }
2740  
2741      /**
# Line 869 | Line 2745 | public class ForkJoinPool extends Abstra
2745       * an approximation, obtained by iterating across all threads in
2746       * the pool. This method may be useful for tuning task
2747       * granularities.
2748 <     * @return the number of queued tasks.
2748 >     *
2749 >     * @return the number of queued tasks
2750       */
2751      public long getQueuedTaskCount() {
2752          long count = 0;
2753 <        ForkJoinWorkerThread[] ws = workers;
2754 <        for (int i = 0; i < ws.length; ++i) {
2755 <            ForkJoinWorkerThread t = ws[i];
2756 <            if (t != null)
2757 <                count += t.getQueueSize();
2753 >        WorkQueue[] ws; WorkQueue w;
2754 >        if ((ws = workQueues) != null) {
2755 >            for (int i = 1; i < ws.length; i += 2) {
2756 >                if ((w = ws[i]) != null)
2757 >                    count += w.queueSize();
2758 >            }
2759          }
2760          return count;
2761      }
2762  
2763      /**
2764 <     * Returns an estimate of the number tasks submitted to this pool
2765 <     * that have not yet begun executing. This method takes time
2766 <     * proportional to the number of submissions.
2767 <     * @return the number of queued submissions.
2764 >     * Returns an estimate of the number of tasks submitted to this
2765 >     * pool that have not yet begun executing.  This method may take
2766 >     * time proportional to the number of submissions.
2767 >     *
2768 >     * @return the number of queued submissions
2769       */
2770      public int getQueuedSubmissionCount() {
2771 <        return submissionQueue.size();
2771 >        int count = 0;
2772 >        WorkQueue[] ws; WorkQueue w;
2773 >        if ((ws = workQueues) != null) {
2774 >            for (int i = 0; i < ws.length; i += 2) {
2775 >                if ((w = ws[i]) != null)
2776 >                    count += w.queueSize();
2777 >            }
2778 >        }
2779 >        return count;
2780      }
2781  
2782      /**
2783 <     * Returns true if there are any tasks submitted to this pool
2784 <     * that have not yet begun executing.
2785 <     * @return <code>true</code> if there are any queued submissions.
2783 >     * Returns {@code true} if there are any tasks submitted to this
2784 >     * pool that have not yet begun executing.
2785 >     *
2786 >     * @return {@code true} if there are any queued submissions
2787       */
2788      public boolean hasQueuedSubmissions() {
2789 <        return !submissionQueue.isEmpty();
2789 >        WorkQueue[] ws; WorkQueue w;
2790 >        if ((ws = workQueues) != null) {
2791 >            for (int i = 0; i < ws.length; i += 2) {
2792 >                if ((w = ws[i]) != null && !w.isEmpty())
2793 >                    return true;
2794 >            }
2795 >        }
2796 >        return false;
2797      }
2798  
2799      /**
2800       * Removes and returns the next unexecuted submission if one is
2801       * available.  This method may be useful in extensions to this
2802       * class that re-assign work in systems with multiple pools.
2803 <     * @return the next submission, or null if none
2803 >     *
2804 >     * @return the next submission, or {@code null} if none
2805       */
2806      protected ForkJoinTask<?> pollSubmission() {
2807 <        return submissionQueue.poll();
2807 >        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2808 >        if ((ws = workQueues) != null) {
2809 >            for (int i = 0; i < ws.length; i += 2) {
2810 >                if ((w = ws[i]) != null && (t = w.poll()) != null)
2811 >                    return t;
2812 >            }
2813 >        }
2814 >        return null;
2815 >    }
2816 >
2817 >    /**
2818 >     * Removes all available unexecuted submitted and forked tasks
2819 >     * from scheduling queues and adds them to the given collection,
2820 >     * without altering their execution status. These may include
2821 >     * artificially generated or wrapped tasks. This method is
2822 >     * designed to be invoked only when the pool is known to be
2823 >     * quiescent. Invocations at other times may not remove all
2824 >     * tasks. A failure encountered while attempting to add elements
2825 >     * to collection {@code c} may result in elements being in
2826 >     * neither, either or both collections when the associated
2827 >     * exception is thrown.  The behavior of this operation is
2828 >     * undefined if the specified collection is modified while the
2829 >     * operation is in progress.
2830 >     *
2831 >     * @param c the collection to transfer elements into
2832 >     * @return the number of elements transferred
2833 >     */
2834 >    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
2835 >        int count = 0;
2836 >        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2837 >        if ((ws = workQueues) != null) {
2838 >            for (int i = 0; i < ws.length; ++i) {
2839 >                if ((w = ws[i]) != null) {
2840 >                    while ((t = w.poll()) != null) {
2841 >                        c.add(t);
2842 >                        ++count;
2843 >                    }
2844 >                }
2845 >            }
2846 >        }
2847 >        return count;
2848      }
2849  
2850      /**
# Line 919 | Line 2855 | public class ForkJoinPool extends Abstra
2855       * @return a string identifying this pool, as well as its state
2856       */
2857      public String toString() {
2858 <        int ps = parallelism;
2859 <        int wc = workerCounts;
2860 <        int rc = runControl;
2861 <        long st = getStealCount();
2862 <        long qt = getQueuedTaskCount();
2863 <        long qs = getQueuedSubmissionCount();
2858 >        // Use a single pass through workQueues to collect counts
2859 >        long qt = 0L, qs = 0L; int rc = 0;
2860 >        long st = stealCount;
2861 >        long c = ctl;
2862 >        WorkQueue[] ws; WorkQueue w;
2863 >        if ((ws = workQueues) != null) {
2864 >            for (int i = 0; i < ws.length; ++i) {
2865 >                if ((w = ws[i]) != null) {
2866 >                    int size = w.queueSize();
2867 >                    if ((i & 1) == 0)
2868 >                        qs += size;
2869 >                    else {
2870 >                        qt += size;
2871 >                        st += w.totalSteals;
2872 >                        if (w.isApparentlyUnblocked())
2873 >                            ++rc;
2874 >                    }
2875 >                }
2876 >            }
2877 >        }
2878 >        int pc = parallelism;
2879 >        int tc = pc + (short)(c >>> TC_SHIFT);
2880 >        int ac = pc + (int)(c >> AC_SHIFT);
2881 >        if (ac < 0) // ignore transient negative
2882 >            ac = 0;
2883 >        String level;
2884 >        if ((c & STOP_BIT) != 0)
2885 >            level = (tc == 0) ? "Terminated" : "Terminating";
2886 >        else
2887 >            level = runState < 0 ? "Shutting down" : "Running";
2888          return super.toString() +
2889 <            "[" + runStateToString(runStateOf(rc)) +
2890 <            ", parallelism = " + ps +
2891 <            ", size = " + totalCountOf(wc) +
2892 <            ", active = " + activeCountOf(rc) +
2893 <            ", running = " + runningCountOf(wc) +
2889 >            "[" + level +
2890 >            ", parallelism = " + pc +
2891 >            ", size = " + tc +
2892 >            ", active = " + ac +
2893 >            ", running = " + rc +
2894              ", steals = " + st +
2895              ", tasks = " + qt +
2896              ", submissions = " + qs +
2897              "]";
2898      }
2899  
940    private static String runStateToString(int rs) {
941        switch(rs) {
942        case RUNNING: return "Running";
943        case SHUTDOWN: return "Shutting down";
944        case TERMINATING: return "Terminating";
945        case TERMINATED: return "Terminated";
946        default: throw new Error("Unknown run state");
947        }
948    }
949
950    // lifecycle control
951
2900      /**
2901 <     * Initiates an orderly shutdown in which previously submitted
2902 <     * tasks are executed, but no new tasks will be accepted.
2903 <     * Invocation has no additional effect if already shut down.
2904 <     * Tasks that are in the process of being submitted concurrently
2905 <     * during the course of this method may or may not be rejected.
2901 >     * Possibly initiates an orderly shutdown in which previously
2902 >     * submitted tasks are executed, but no new tasks will be
2903 >     * accepted. Invocation has no effect on execution state if this
2904 >     * is the {@link #commonPool}, and no additional effect if
2905 >     * already shut down.  Tasks that are in the process of being
2906 >     * submitted concurrently during the course of this method may or
2907 >     * may not be rejected.
2908 >     *
2909       * @throws SecurityException if a security manager exists and
2910       *         the caller is not permitted to modify threads
2911       *         because it does not hold {@link
2912 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
2912 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
2913       */
2914      public void shutdown() {
2915          checkPermission();
2916 <        transitionRunStateTo(SHUTDOWN);
2917 <        if (canTerminateOnShutdown(runControl))
967 <            terminateOnShutdown();
2916 >        if (this != commonPool)
2917 >            tryTerminate(false, true);
2918      }
2919  
2920      /**
2921 <     * Attempts to stop all actively executing tasks, and cancels all
2922 <     * waiting tasks.  Tasks that are in the process of being
2923 <     * submitted or executed concurrently during the course of this
2924 <     * method may or may not be rejected. Unlike some other executors,
2925 <     * this method cancels rather than collects non-executed tasks,
2926 <     * so always returns an empty list.
2921 >     * Possibly attempts to cancel and/or stop all tasks, and reject
2922 >     * all subsequently submitted tasks.  Invocation has no effect on
2923 >     * execution state if this is the {@link #commonPool}, and no
2924 >     * additional effect if already shut down. Otherwise, tasks that
2925 >     * are in the process of being submitted or executed concurrently
2926 >     * during the course of this method may or may not be
2927 >     * rejected. This method cancels both existing and unexecuted
2928 >     * tasks, in order to permit termination in the presence of task
2929 >     * dependencies. So the method always returns an empty list
2930 >     * (unlike the case for some other Executors).
2931 >     *
2932       * @return an empty list
2933       * @throws SecurityException if a security manager exists and
2934       *         the caller is not permitted to modify threads
2935       *         because it does not hold {@link
2936 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
2936 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
2937       */
2938      public List<Runnable> shutdownNow() {
2939          checkPermission();
2940 <        terminate();
2940 >        if (this != commonPool)
2941 >            tryTerminate(true, true);
2942          return Collections.emptyList();
2943      }
2944  
2945      /**
2946 <     * Returns <code>true</code> if all tasks have completed following shut down.
2946 >     * Returns {@code true} if all tasks have completed following shut down.
2947       *
2948 <     * @return <code>true</code> if all tasks have completed following shut down
2948 >     * @return {@code true} if all tasks have completed following shut down
2949       */
2950      public boolean isTerminated() {
2951 <        return runStateOf(runControl) == TERMINATED;
2951 >        long c = ctl;
2952 >        return ((c & STOP_BIT) != 0L &&
2953 >                (short)(c >>> TC_SHIFT) == -parallelism);
2954      }
2955  
2956      /**
2957 <     * Returns <code>true</code> if the process of termination has
2958 <     * commenced but possibly not yet completed.
2957 >     * Returns {@code true} if the process of termination has
2958 >     * commenced but not yet completed.  This method may be useful for
2959 >     * debugging. A return of {@code true} reported a sufficient
2960 >     * period after shutdown may indicate that submitted tasks have
2961 >     * ignored or suppressed interruption, or are waiting for IO,
2962 >     * causing this executor not to properly terminate. (See the
2963 >     * advisory notes for class {@link ForkJoinTask} stating that
2964 >     * tasks should not normally entail blocking operations.  But if
2965 >     * they do, they must abort them on interrupt.)
2966       *
2967 <     * @return <code>true</code> if terminating
2967 >     * @return {@code true} if terminating but not yet terminated
2968       */
2969      public boolean isTerminating() {
2970 <        return runStateOf(runControl) >= TERMINATING;
2970 >        long c = ctl;
2971 >        return ((c & STOP_BIT) != 0L &&
2972 >                (short)(c >>> TC_SHIFT) != -parallelism);
2973      }
2974  
2975      /**
2976 <     * Returns <code>true</code> if this pool has been shut down.
2976 >     * Returns {@code true} if this pool has been shut down.
2977       *
2978 <     * @return <code>true</code> if this pool has been shut down
2978 >     * @return {@code true} if this pool has been shut down
2979       */
2980      public boolean isShutdown() {
2981 <        return runStateOf(runControl) >= SHUTDOWN;
2981 >        return runState < 0;
2982      }
2983  
2984      /**
# Line 1021 | Line 2988 | public class ForkJoinPool extends Abstra
2988       *
2989       * @param timeout the maximum time to wait
2990       * @param unit the time unit of the timeout argument
2991 <     * @return <code>true</code> if this executor terminated and
2992 <     *         <code>false</code> if the timeout elapsed before termination
2991 >     * @return {@code true} if this executor terminated and
2992 >     *         {@code false} if the timeout elapsed before termination
2993       * @throws InterruptedException if interrupted while waiting
2994       */
2995      public boolean awaitTermination(long timeout, TimeUnit unit)
2996          throws InterruptedException {
2997          long nanos = unit.toNanos(timeout);
2998 <        final ReentrantLock lock = this.workerLock;
1032 <        lock.lock();
1033 <        try {
1034 <            for (;;) {
1035 <                if (isTerminated())
1036 <                    return true;
1037 <                if (nanos <= 0)
1038 <                    return false;
1039 <                nanos = termination.awaitNanos(nanos);
1040 <            }
1041 <        } finally {
1042 <            lock.unlock();
1043 <        }
1044 <    }
1045 <
1046 <    // Shutdown and termination support
1047 <
1048 <    /**
1049 <     * Callback from terminating worker. Null out the corresponding
1050 <     * workers slot, and if terminating, try to terminate, else try to
1051 <     * shrink workers array.
1052 <     * @param w the worker
1053 <     */
1054 <    final void workerTerminated(ForkJoinWorkerThread w) {
1055 <        updateStealCount(w);
1056 <        updateWorkerCount(-1);
1057 <        final ReentrantLock lock = this.workerLock;
1058 <        lock.lock();
1059 <        try {
1060 <            ForkJoinWorkerThread[] ws = workers;
1061 <            int idx = w.poolIndex;
1062 <            if (idx >= 0 && idx < ws.length && ws[idx] == w)
1063 <                ws[idx] = null;
1064 <            if (totalCountOf(workerCounts) == 0) {
1065 <                terminate(); // no-op if already terminating
1066 <                transitionRunStateTo(TERMINATED);
1067 <                termination.signalAll();
1068 <            }
1069 <            else if (!isTerminating()) {
1070 <                tryShrinkWorkerArray();
1071 <                tryResumeSpare(true); // allow replacement
1072 <            }
1073 <        } finally {
1074 <            lock.unlock();
1075 <        }
1076 <        signalIdleWorkers();
1077 <    }
1078 <
1079 <    /**
1080 <     * Initiate termination.
1081 <     */
1082 <    private void terminate() {
1083 <        if (transitionRunStateTo(TERMINATING)) {
1084 <            stopAllWorkers();
1085 <            resumeAllSpares();
1086 <            signalIdleWorkers();
1087 <            cancelQueuedSubmissions();
1088 <            cancelQueuedWorkerTasks();
1089 <            interruptUnterminatedWorkers();
1090 <            signalIdleWorkers(); // resignal after interrupt
1091 <        }
1092 <    }
1093 <
1094 <    /**
1095 <     * Possibly terminate when on shutdown state
1096 <     */
1097 <    private void terminateOnShutdown() {
1098 <        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1099 <            terminate();
1100 <    }
1101 <
1102 <    /**
1103 <     * Clear out and cancel submissions
1104 <     */
1105 <    private void cancelQueuedSubmissions() {
1106 <        ForkJoinTask<?> task;
1107 <        while ((task = pollSubmission()) != null)
1108 <            task.cancel(false);
1109 <    }
1110 <
1111 <    /**
1112 <     * Clean out worker queues.
1113 <     */
1114 <    private void cancelQueuedWorkerTasks() {
1115 <        final ReentrantLock lock = this.workerLock;
1116 <        lock.lock();
1117 <        try {
1118 <            ForkJoinWorkerThread[] ws = workers;
1119 <            for (int i = 0; i < ws.length; ++i) {
1120 <                ForkJoinWorkerThread t = ws[i];
1121 <                if (t != null)
1122 <                    t.cancelTasks();
1123 <            }
1124 <        } finally {
1125 <            lock.unlock();
1126 <        }
1127 <    }
1128 <
1129 <    /**
1130 <     * Set each worker's status to terminating. Requires lock to avoid
1131 <     * conflicts with add/remove
1132 <     */
1133 <    private void stopAllWorkers() {
1134 <        final ReentrantLock lock = this.workerLock;
1135 <        lock.lock();
1136 <        try {
1137 <            ForkJoinWorkerThread[] ws = workers;
1138 <            for (int i = 0; i < ws.length; ++i) {
1139 <                ForkJoinWorkerThread t = ws[i];
1140 <                if (t != null)
1141 <                    t.shutdownNow();
1142 <            }
1143 <        } finally {
1144 <            lock.unlock();
1145 <        }
1146 <    }
1147 <
1148 <    /**
1149 <     * Interrupt all unterminated workers.  This is not required for
1150 <     * sake of internal control, but may help unstick user code during
1151 <     * shutdown.
1152 <     */
1153 <    private void interruptUnterminatedWorkers() {
1154 <        final ReentrantLock lock = this.workerLock;
1155 <        lock.lock();
1156 <        try {
1157 <            ForkJoinWorkerThread[] ws = workers;
1158 <            for (int i = 0; i < ws.length; ++i) {
1159 <                ForkJoinWorkerThread t = ws[i];
1160 <                if (t != null && !t.isTerminated()) {
1161 <                    try {
1162 <                        t.interrupt();
1163 <                    } catch (SecurityException ignore) {
1164 <                    }
1165 <                }
1166 <            }
1167 <        } finally {
1168 <            lock.unlock();
1169 <        }
1170 <    }
1171 <
1172 <
1173 <    /*
1174 <     * Nodes for event barrier to manage idle threads.  Queue nodes
1175 <     * are basic Treiber stack nodes, also used for spare stack.
1176 <     *
1177 <     * The event barrier has an event count and a wait queue (actually
1178 <     * a Treiber stack).  Workers are enabled to look for work when
1179 <     * the eventCount is incremented. If they fail to find work, they
1180 <     * may wait for next count. Upon release, threads help others wake
1181 <     * up.
1182 <     *
1183 <     * Synchronization events occur only in enough contexts to
1184 <     * maintain overall liveness:
1185 <     *
1186 <     *   - Submission of a new task to the pool
1187 <     *   - Resizes or other changes to the workers array
1188 <     *   - pool termination
1189 <     *   - A worker pushing a task on an empty queue
1190 <     *
1191 <     * The case of pushing a task occurs often enough, and is heavy
1192 <     * enough compared to simple stack pushes, to require special
1193 <     * handling: Method signalWork returns without advancing count if
1194 <     * the queue appears to be empty.  This would ordinarily result in
1195 <     * races causing some queued waiters not to be woken up. To avoid
1196 <     * this, the first worker enqueued in method sync (see
1197 <     * syncIsReleasable) rescans for tasks after being enqueued, and
1198 <     * helps signal if any are found. This works well because the
1199 <     * worker has nothing better to do, and so might as well help
1200 <     * alleviate the overhead and contention on the threads actually
1201 <     * doing work.  Also, since event counts increments on task
1202 <     * availability exist to maintain liveness (rather than to force
1203 <     * refreshes etc), it is OK for callers to exit early if
1204 <     * contending with another signaller.
1205 <     */
1206 <    static final class WaitQueueNode {
1207 <        WaitQueueNode next; // only written before enqueued
1208 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1209 <        final long count; // unused for spare stack
1210 <
1211 <        WaitQueueNode(long c, ForkJoinWorkerThread w) {
1212 <            count = c;
1213 <            thread = w;
1214 <        }
1215 <
1216 <        /**
1217 <         * Wake up waiter, returning false if known to already
1218 <         */
1219 <        boolean signal() {
1220 <            ForkJoinWorkerThread t = thread;
1221 <            if (t == null)
1222 <                return false;
1223 <            thread = null;
1224 <            LockSupport.unpark(t);
2998 >        if (isTerminated())
2999              return true;
3000 <        }
3001 <
3002 <        /**
3003 <         * Await release on sync
3004 <         */
3005 <        void awaitSyncRelease(ForkJoinPool p) {
3006 <            while (thread != null && !p.syncIsReleasable(this))
1233 <                LockSupport.park(this);
1234 <        }
1235 <
1236 <        /**
1237 <         * Await resumption as spare
1238 <         */
1239 <        void awaitSpareRelease() {
1240 <            while (thread != null) {
1241 <                if (!Thread.interrupted())
1242 <                    LockSupport.park(this);
1243 <            }
1244 <        }
1245 <    }
1246 <
1247 <    /**
1248 <     * Ensures that no thread is waiting for count to advance from the
1249 <     * current value of eventCount read on entry to this method, by
1250 <     * releasing waiting threads if necessary.
1251 <     * @return the count
1252 <     */
1253 <    final long ensureSync() {
1254 <        long c = eventCount;
1255 <        WaitQueueNode q;
1256 <        while ((q = syncStack) != null && q.count < c) {
1257 <            if (casBarrierStack(q, null)) {
1258 <                do {
1259 <                    q.signal();
1260 <                } while ((q = q.next) != null);
1261 <                break;
1262 <            }
1263 <        }
1264 <        return c;
1265 <    }
1266 <
1267 <    /**
1268 <     * Increments event count and releases waiting threads.
1269 <     */
1270 <    private void signalIdleWorkers() {
1271 <        long c;
1272 <        do;while (!casEventCount(c = eventCount, c+1));
1273 <        ensureSync();
1274 <    }
1275 <
1276 <    /**
1277 <     * Signal threads waiting to poll a task. Because method sync
1278 <     * rechecks availability, it is OK to only proceed if queue
1279 <     * appears to be non-empty, and OK to skip under contention to
1280 <     * increment count (since some other thread succeeded).
1281 <     */
1282 <    final void signalWork() {
1283 <        long c;
1284 <        WaitQueueNode q;
1285 <        if (syncStack != null &&
1286 <            casEventCount(c = eventCount, c+1) &&
1287 <            (((q = syncStack) != null && q.count <= c) &&
1288 <             (!casBarrierStack(q, q.next) || !q.signal())))
1289 <            ensureSync();
1290 <    }
1291 <
1292 <    /**
1293 <     * Waits until event count advances from last value held by
1294 <     * caller, or if excess threads, caller is resumed as spare, or
1295 <     * caller or pool is terminating. Updates caller's event on exit.
1296 <     * @param w the calling worker thread
1297 <     */
1298 <    final void sync(ForkJoinWorkerThread w) {
1299 <        updateStealCount(w); // Transfer w's count while it is idle
1300 <
1301 <        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1302 <            long prev = w.lastEventCount;
1303 <            WaitQueueNode node = null;
1304 <            WaitQueueNode h;
1305 <            while (eventCount == prev &&
1306 <                   ((h = syncStack) == null || h.count == prev)) {
1307 <                if (node == null)
1308 <                    node = new WaitQueueNode(prev, w);
1309 <                if (casBarrierStack(node.next = h, node)) {
1310 <                    node.awaitSyncRelease(this);
1311 <                    break;
1312 <                }
1313 <            }
1314 <            long ec = ensureSync();
1315 <            if (ec != prev) {
1316 <                w.lastEventCount = ec;
1317 <                break;
1318 <            }
1319 <        }
1320 <    }
1321 <
1322 <    /**
1323 <     * Returns true if worker waiting on sync can proceed:
1324 <     *  - on signal (thread == null)
1325 <     *  - on event count advance (winning race to notify vs signaller)
1326 <     *  - on Interrupt
1327 <     *  - if the first queued node, we find work available
1328 <     * If node was not signalled and event count not advanced on exit,
1329 <     * then we also help advance event count.
1330 <     * @return true if node can be released
1331 <     */
1332 <    final boolean syncIsReleasable(WaitQueueNode node) {
1333 <        long prev = node.count;
1334 <        if (!Thread.interrupted() && node.thread != null &&
1335 <            (node.next != null ||
1336 <             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1337 <            eventCount == prev)
1338 <            return false;
1339 <        if (node.thread != null) {
1340 <            node.thread = null;
1341 <            long ec = eventCount;
1342 <            if (prev <= ec) // help signal
1343 <                casEventCount(ec, ec+1);
1344 <        }
1345 <        return true;
1346 <    }
1347 <
1348 <    /**
1349 <     * Returns true if a new sync event occurred since last call to
1350 <     * sync or this method, if so, updating caller's count.
1351 <     */
1352 <    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1353 <        long lc = w.lastEventCount;
1354 <        long ec = ensureSync();
1355 <        if (ec == lc)
1356 <            return false;
1357 <        w.lastEventCount = ec;
1358 <        return true;
1359 <    }
1360 <
1361 <    //  Parallelism maintenance
1362 <
1363 <    /**
1364 <     * Decrement running count; if too low, add spare.
1365 <     *
1366 <     * Conceptually, all we need to do here is add or resume a
1367 <     * spare thread when one is about to block (and remove or
1368 <     * suspend it later when unblocked -- see suspendIfSpare).
1369 <     * However, implementing this idea requires coping with
1370 <     * several problems: We have imperfect information about the
1371 <     * states of threads. Some count updates can and usually do
1372 <     * lag run state changes, despite arrangements to keep them
1373 <     * accurate (for example, when possible, updating counts
1374 <     * before signalling or resuming), especially when running on
1375 <     * dynamic JVMs that don't optimize the infrequent paths that
1376 <     * update counts. Generating too many threads can make these
1377 <     * problems become worse, because excess threads are more
1378 <     * likely to be context-switched with others, slowing them all
1379 <     * down, especially if there is no work available, so all are
1380 <     * busy scanning or idling.  Also, excess spare threads can
1381 <     * only be suspended or removed when they are idle, not
1382 <     * immediately when they aren't needed. So adding threads will
1383 <     * raise parallelism level for longer than necessary.  Also,
1384 <     * FJ applications often enounter highly transient peaks when
1385 <     * many threads are blocked joining, but for less time than it
1386 <     * takes to create or resume spares.
1387 <     *
1388 <     * @param joinMe if non-null, return early if done
1389 <     * @param maintainParallelism if true, try to stay within
1390 <     * target counts, else create only to avoid starvation
1391 <     * @return true if joinMe known to be done
1392 <     */
1393 <    final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
1394 <        maintainParallelism &= maintainsParallelism; // overrride
1395 <        boolean dec = false;  // true when running count decremented
1396 <        while (spareStack == null || !tryResumeSpare(dec)) {
1397 <            int counts = workerCounts;
1398 <            if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
1399 <                if (!needSpare(counts, maintainParallelism))
1400 <                    break;
1401 <                if (joinMe.status < 0)
1402 <                    return true;
1403 <                if (tryAddSpare(counts))
1404 <                    break;
1405 <            }
1406 <        }
1407 <        return false;
1408 <    }
1409 <
1410 <    /**
1411 <     * Same idea as preJoin
1412 <     */
1413 <    final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
1414 <        maintainParallelism &= maintainsParallelism;
1415 <        boolean dec = false;
1416 <        while (spareStack == null || !tryResumeSpare(dec)) {
1417 <            int counts = workerCounts;
1418 <            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1419 <                if (!needSpare(counts, maintainParallelism))
1420 <                    break;
1421 <                if (blocker.isReleasable())
1422 <                    return true;
1423 <                if (tryAddSpare(counts))
3000 >        long startTime = System.nanoTime();
3001 >        boolean terminated = false;
3002 >        synchronized(this) {
3003 >            for (long waitTime = nanos, millis = 0L;;) {
3004 >                if (terminated = isTerminated() ||
3005 >                    waitTime <= 0L ||
3006 >                    (millis = unit.toMillis(waitTime)) <= 0L)
3007                      break;
3008 +                wait(millis);
3009 +                waitTime = nanos - (System.nanoTime() - startTime);
3010              }
3011          }
3012 <        return false;
1428 <    }
1429 <
1430 <    /**
1431 <     * Returns true if a spare thread appears to be needed.  If
1432 <     * maintaining parallelism, returns true when the deficit in
1433 <     * running threads is more than the surplus of total threads, and
1434 <     * there is apparently some work to do.  This self-limiting rule
1435 <     * means that the more threads that have already been added, the
1436 <     * less parallelism we will tolerate before adding another.
1437 <     * @param counts current worker counts
1438 <     * @param maintainParallelism try to maintain parallelism
1439 <     */
1440 <    private boolean needSpare(int counts, boolean maintainParallelism) {
1441 <        int ps = parallelism;
1442 <        int rc = runningCountOf(counts);
1443 <        int tc = totalCountOf(counts);
1444 <        int runningDeficit = ps - rc;
1445 <        int totalSurplus = tc - ps;
1446 <        return (tc < maxPoolSize &&
1447 <                (rc == 0 || totalSurplus < 0 ||
1448 <                 (maintainParallelism &&
1449 <                  runningDeficit > totalSurplus &&
1450 <                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
1451 <    }
1452 <    
1453 <    /**
1454 <     * Add a spare worker if lock available and no more than the
1455 <     * expected numbers of threads exist
1456 <     * @return true if successful
1457 <     */
1458 <    private boolean tryAddSpare(int expectedCounts) {
1459 <        final ReentrantLock lock = this.workerLock;
1460 <        int expectedRunning = runningCountOf(expectedCounts);
1461 <        int expectedTotal = totalCountOf(expectedCounts);
1462 <        boolean success = false;
1463 <        boolean locked = false;
1464 <        // confirm counts while locking; CAS after obtaining lock
1465 <        try {
1466 <            for (;;) {
1467 <                int s = workerCounts;
1468 <                int tc = totalCountOf(s);
1469 <                int rc = runningCountOf(s);
1470 <                if (rc > expectedRunning || tc > expectedTotal)
1471 <                    break;
1472 <                if (!locked && !(locked = lock.tryLock()))
1473 <                    break;
1474 <                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1475 <                    createAndStartSpare(tc);
1476 <                    success = true;
1477 <                    break;
1478 <                }
1479 <            }
1480 <        } finally {
1481 <            if (locked)
1482 <                lock.unlock();
1483 <        }
1484 <        return success;
1485 <    }
1486 <
1487 <    /**
1488 <     * Add the kth spare worker. On entry, pool coounts are already
1489 <     * adjusted to reflect addition.
1490 <     */
1491 <    private void createAndStartSpare(int k) {
1492 <        ForkJoinWorkerThread w = null;
1493 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1494 <        int len = ws.length;
1495 <        // Probably, we can place at slot k. If not, find empty slot
1496 <        if (k < len && ws[k] != null) {
1497 <            for (k = 0; k < len && ws[k] != null; ++k)
1498 <                ;
1499 <        }
1500 <        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1501 <            ws[k] = w;
1502 <            w.start();
1503 <        }
1504 <        else
1505 <            updateWorkerCount(-1); // adjust on failure
1506 <        signalIdleWorkers();
1507 <    }
1508 <
1509 <    /**
1510 <     * Suspend calling thread w if there are excess threads.  Called
1511 <     * only from sync.  Spares are enqueued in a Treiber stack
1512 <     * using the same WaitQueueNodes as barriers.  They are resumed
1513 <     * mainly in preJoin, but are also woken on pool events that
1514 <     * require all threads to check run state.
1515 <     * @param w the caller
1516 <     */
1517 <    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1518 <        WaitQueueNode node = null;
1519 <        int s;
1520 <        while (parallelism < runningCountOf(s = workerCounts)) {
1521 <            if (node == null)
1522 <                node = new WaitQueueNode(0, w);
1523 <            if (casWorkerCounts(s, s-1)) { // representation-dependent
1524 <                // push onto stack
1525 <                do;while (!casSpareStack(node.next = spareStack, node));
1526 <                // block until released by resumeSpare
1527 <                node.awaitSpareRelease();
1528 <                return true;
1529 <            }
1530 <        }
1531 <        return false;
1532 <    }
1533 <
1534 <    /**
1535 <     * Try to pop and resume a spare thread.
1536 <     * @param updateCount if true, increment running count on success
1537 <     * @return true if successful
1538 <     */
1539 <    private boolean tryResumeSpare(boolean updateCount) {
1540 <        WaitQueueNode q;
1541 <        while ((q = spareStack) != null) {
1542 <            if (casSpareStack(q, q.next)) {
1543 <                if (updateCount)
1544 <                    updateRunningCount(1);
1545 <                q.signal();
1546 <                return true;
1547 <            }
1548 <        }
1549 <        return false;
1550 <    }
1551 <
1552 <    /**
1553 <     * Pop and resume all spare threads. Same idea as ensureSync.
1554 <     * @return true if any spares released
1555 <     */
1556 <    private boolean resumeAllSpares() {
1557 <        WaitQueueNode q;
1558 <        while ( (q = spareStack) != null) {
1559 <            if (casSpareStack(q, null)) {
1560 <                do {
1561 <                    updateRunningCount(1);
1562 <                    q.signal();
1563 <                } while ((q = q.next) != null);
1564 <                return true;
1565 <            }
1566 <        }
1567 <        return false;
1568 <    }
1569 <
1570 <    /**
1571 <     * Pop and shutdown excessive spare threads. Call only while
1572 <     * holding lock. This is not guaranteed to eliminate all excess
1573 <     * threads, only those suspended as spares, which are the ones
1574 <     * unlikely to be needed in the future.
1575 <     */
1576 <    private void trimSpares() {
1577 <        int surplus = totalCountOf(workerCounts) - parallelism;
1578 <        WaitQueueNode q;
1579 <        while (surplus > 0 && (q = spareStack) != null) {
1580 <            if (casSpareStack(q, null)) {
1581 <                do {
1582 <                    updateRunningCount(1);
1583 <                    ForkJoinWorkerThread w = q.thread;
1584 <                    if (w != null && surplus > 0 &&
1585 <                        runningCountOf(workerCounts) > 0 && w.shutdown())
1586 <                        --surplus;
1587 <                    q.signal();
1588 <                } while ((q = q.next) != null);
1589 <            }
1590 <        }
3012 >        return terminated;
3013      }
3014  
3015      /**
3016       * Interface for extending managed parallelism for tasks running
3017 <     * in ForkJoinPools. A ManagedBlocker provides two methods.
3018 <     * Method <code>isReleasable</code> must return true if blocking is not
3019 <     * necessary. Method <code>block</code> blocks the current thread
3020 <     * if necessary (perhaps internally invoking isReleasable before
3021 <     * actually blocking.).
3017 >     * in {@link ForkJoinPool}s.
3018 >     *
3019 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
3020 >     * {@code isReleasable} must return {@code true} if blocking is
3021 >     * not necessary. Method {@code block} blocks the current thread
3022 >     * if necessary (perhaps internally invoking {@code isReleasable}
3023 >     * before actually blocking). These actions are performed by any
3024 >     * thread invoking {@link ForkJoinPool#managedBlock}.  The
3025 >     * unusual methods in this API accommodate synchronizers that may,
3026 >     * but don't usually, block for long periods. Similarly, they
3027 >     * allow more efficient internal handling of cases in which
3028 >     * additional workers may be, but usually are not, needed to
3029 >     * ensure sufficient parallelism.  Toward this end,
3030 >     * implementations of method {@code isReleasable} must be amenable
3031 >     * to repeated invocation.
3032 >     *
3033       * <p>For example, here is a ManagedBlocker based on a
3034       * ReentrantLock:
3035 <     * <pre>
3036 <     *   class ManagedLocker implements ManagedBlocker {
3037 <     *     final ReentrantLock lock;
3038 <     *     boolean hasLock = false;
3039 <     *     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
3040 <     *     public boolean block() {
3041 <     *        if (!hasLock)
3042 <     *           lock.lock();
3043 <     *        return true;
3044 <     *     }
3045 <     *     public boolean isReleasable() {
3046 <     *        return hasLock || (hasLock = lock.tryLock());
1614 <     *     }
3035 >     *  <pre> {@code
3036 >     * class ManagedLocker implements ManagedBlocker {
3037 >     *   final ReentrantLock lock;
3038 >     *   boolean hasLock = false;
3039 >     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
3040 >     *   public boolean block() {
3041 >     *     if (!hasLock)
3042 >     *       lock.lock();
3043 >     *     return true;
3044 >     *   }
3045 >     *   public boolean isReleasable() {
3046 >     *     return hasLock || (hasLock = lock.tryLock());
3047       *   }
3048 <     * </pre>
3048 >     * }}</pre>
3049 >     *
3050 >     * <p>Here is a class that possibly blocks waiting for an
3051 >     * item on a given queue:
3052 >     *  <pre> {@code
3053 >     * class QueueTaker<E> implements ManagedBlocker {
3054 >     *   final BlockingQueue<E> queue;
3055 >     *   volatile E item = null;
3056 >     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
3057 >     *   public boolean block() throws InterruptedException {
3058 >     *     if (item == null)
3059 >     *       item = queue.take();
3060 >     *     return true;
3061 >     *   }
3062 >     *   public boolean isReleasable() {
3063 >     *     return item != null || (item = queue.poll()) != null;
3064 >     *   }
3065 >     *   public E getItem() { // call after pool.managedBlock completes
3066 >     *     return item;
3067 >     *   }
3068 >     * }}</pre>
3069       */
3070      public static interface ManagedBlocker {
3071          /**
3072           * Possibly blocks the current thread, for example waiting for
3073           * a lock or condition.
3074 <         * @return true if no additional blocking is necessary (i.e.,
3075 <         * if isReleasable would return true).
3074 >         *
3075 >         * @return {@code true} if no additional blocking is necessary
3076 >         * (i.e., if isReleasable would return true)
3077           * @throws InterruptedException if interrupted while waiting
3078 <         * (the method is not required to do so, but is allowe to).
3078 >         * (the method is not required to do so, but is allowed to)
3079           */
3080          boolean block() throws InterruptedException;
3081  
3082          /**
3083 <         * Returns true if blocking is unnecessary.
3083 >         * Returns {@code true} if blocking is unnecessary.
3084           */
3085          boolean isReleasable();
3086      }
3087  
3088      /**
3089       * Blocks in accord with the given blocker.  If the current thread
3090 <     * is a ForkJoinWorkerThread, this method possibly arranges for a
3091 <     * spare thread to be activated if necessary to ensure parallelism
3092 <     * while the current thread is blocked.  If
3093 <     * <code>maintainParallelism</code> is true and the pool supports
3094 <     * it ({@link #getMaintainsParallelism}), this method attempts to
3095 <     * maintain the pool's nominal parallelism. Otherwise if activates
3096 <     * a thread only if necessary to avoid complete starvation. This
3097 <     * option may be preferable when blockages use timeouts, or are
3098 <     * almost always brief.
3099 <     *
3100 <     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
3101 <     * equivalent to
3102 <     * <pre>
3103 <     *   while (!blocker.isReleasable())
1651 <     *      if (blocker.block())
1652 <     *         return;
1653 <     * </pre>
1654 <     * If the caller is a ForkJoinTask, then the pool may first
1655 <     * be expanded to ensure parallelism, and later adjusted.
3090 >     * is a {@link ForkJoinWorkerThread}, this method possibly
3091 >     * arranges for a spare thread to be activated if necessary to
3092 >     * ensure sufficient parallelism while the current thread is blocked.
3093 >     *
3094 >     * <p>If the caller is not a {@link ForkJoinTask}, this method is
3095 >     * behaviorally equivalent to
3096 >     *  <pre> {@code
3097 >     * while (!blocker.isReleasable())
3098 >     *   if (blocker.block())
3099 >     *     return;
3100 >     * }</pre>
3101 >     *
3102 >     * If the caller is a {@code ForkJoinTask}, then the pool may
3103 >     * first be expanded to ensure parallelism, and later adjusted.
3104       *
3105       * @param blocker the blocker
3106 <     * @param maintainParallelism if true and supported by this pool,
1659 <     * attempt to maintain the pool's nominal parallelism; otherwise
1660 <     * activate a thread only if necessary to avoid complete
1661 <     * starvation.
1662 <     * @throws InterruptedException if blocker.block did so.
3106 >     * @throws InterruptedException if blocker.block did so
3107       */
3108 <    public static void managedBlock(ManagedBlocker blocker,
1665 <                                    boolean maintainParallelism)
3108 >    public static void managedBlock(ManagedBlocker blocker)
3109          throws InterruptedException {
3110          Thread t = Thread.currentThread();
3111 <        ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
3112 <                             ((ForkJoinWorkerThread)t).pool : null);
3113 <        if (!blocker.isReleasable()) {
3114 <            try {
3115 <                if (pool == null ||
3116 <                    !pool.preBlock(blocker, maintainParallelism))
3117 <                    awaitBlocker(blocker);
3118 <            } finally {
3119 <                if (pool != null)
3120 <                    pool.updateRunningCount(1);
3111 >        ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
3112 >                          ((ForkJoinWorkerThread)t).pool : null);
3113 >        while (!blocker.isReleasable()) {
3114 >            if (p == null || p.tryCompensate(null, blocker)) {
3115 >                try {
3116 >                    do {} while (!blocker.isReleasable() && !blocker.block());
3117 >                } finally {
3118 >                    if (p != null)
3119 >                        p.incrementActiveCount();
3120 >                }
3121 >                break;
3122              }
3123          }
3124      }
3125  
3126 <    private static void awaitBlocker(ManagedBlocker blocker)
3127 <        throws InterruptedException {
3128 <        do;while (!blocker.isReleasable() && !blocker.block());
1685 <    }
1686 <
1687 <    // AbstractExecutorService overrides
3126 >    // AbstractExecutorService overrides.  These rely on undocumented
3127 >    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
3128 >    // implement RunnableFuture.
3129  
3130      protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3131 <        return new AdaptedRunnable(runnable, value);
3131 >        return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3132      }
3133  
3134      protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3135 <        return new AdaptedCallable(callable);
3135 >        return new ForkJoinTask.AdaptedCallable<T>(callable);
3136      }
3137  
3138 <
3139 <    // Temporary Unsafe mechanics for preliminary release
3140 <
3141 <    static final Unsafe _unsafe;
3142 <    static final long eventCountOffset;
3143 <    static final long workerCountsOffset;
3144 <    static final long runControlOffset;
3145 <    static final long syncStackOffset;
3146 <    static final long spareStackOffset;
3138 >    // Unsafe mechanics
3139 >    private static final sun.misc.Unsafe U;
3140 >    private static final long CTL;
3141 >    private static final long PARKBLOCKER;
3142 >    private static final int ABASE;
3143 >    private static final int ASHIFT;
3144 >    private static final long NEXTWORKERNUMBER;
3145 >    private static final long STEALCOUNT;
3146 >    private static final long MAINLOCK;
3147  
3148      static {
3149 +        poolNumberGenerator = new AtomicInteger();
3150 +        nextSubmitterSeed = new AtomicInteger(0x55555555);
3151 +        modifyThreadPermission = new RuntimePermission("modifyThread");
3152 +        defaultForkJoinWorkerThreadFactory =
3153 +            new DefaultForkJoinWorkerThreadFactory();
3154 +        submitters = new ThreadSubmitter();
3155 +        int s;
3156          try {
3157 <            if (ForkJoinPool.class.getClassLoader() != null) {
3158 <                Field f = Unsafe.class.getDeclaredField("theUnsafe");
3159 <                f.setAccessible(true);
3160 <                _unsafe = (Unsafe)f.get(null);
3161 <            }
3162 <            else
3163 <                _unsafe = Unsafe.getUnsafe();
3164 <            eventCountOffset = _unsafe.objectFieldOffset
3165 <                (ForkJoinPool.class.getDeclaredField("eventCount"));
3166 <            workerCountsOffset = _unsafe.objectFieldOffset
3167 <                (ForkJoinPool.class.getDeclaredField("workerCounts"));
3168 <            runControlOffset = _unsafe.objectFieldOffset
3169 <                (ForkJoinPool.class.getDeclaredField("runControl"));
3170 <            syncStackOffset = _unsafe.objectFieldOffset
3171 <                (ForkJoinPool.class.getDeclaredField("syncStack"));
3172 <            spareStackOffset = _unsafe.objectFieldOffset
3173 <                (ForkJoinPool.class.getDeclaredField("spareStack"));
3157 >            U = getUnsafe();
3158 >            Class<?> k = ForkJoinPool.class;
3159 >            Class<?> ak = ForkJoinTask[].class;
3160 >            CTL = U.objectFieldOffset
3161 >                (k.getDeclaredField("ctl"));
3162 >            NEXTWORKERNUMBER = U.objectFieldOffset
3163 >                (k.getDeclaredField("nextWorkerNumber"));
3164 >            STEALCOUNT = U.objectFieldOffset
3165 >                (k.getDeclaredField("stealCount"));
3166 >            MAINLOCK = U.objectFieldOffset
3167 >                (k.getDeclaredField("mainLock"));
3168 >            Class<?> tk = Thread.class;
3169 >            PARKBLOCKER = U.objectFieldOffset
3170 >                (tk.getDeclaredField("parkBlocker"));
3171 >            ABASE = U.arrayBaseOffset(ak);
3172 >            s = U.arrayIndexScale(ak);
3173 >            ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3174          } catch (Exception e) {
3175 <            throw new RuntimeException("Could not initialize intrinsics", e);
3175 >            throw new Error(e);
3176 >        }
3177 >        if ((s & (s-1)) != 0)
3178 >            throw new Error("data type scale not a power of two");
3179 >        try { // Establish common pool
3180 >            String pp = System.getProperty(propPrefix + "parallelism");
3181 >            String fp = System.getProperty(propPrefix + "threadFactory");
3182 >            String up = System.getProperty(propPrefix + "exceptionHandler");
3183 >            ForkJoinWorkerThreadFactory fac = (fp == null) ?
3184 >                defaultForkJoinWorkerThreadFactory :
3185 >                ((ForkJoinWorkerThreadFactory)ClassLoader.
3186 >                 getSystemClassLoader().loadClass(fp).newInstance());
3187 >            Thread.UncaughtExceptionHandler ueh = (up == null)? null :
3188 >                ((Thread.UncaughtExceptionHandler)ClassLoader.
3189 >                 getSystemClassLoader().loadClass(up).newInstance());
3190 >            int par;
3191 >            if ((pp == null || (par = Integer.parseInt(pp)) <= 0))
3192 >                par = Runtime.getRuntime().availableProcessors();
3193 >            if (par > MAX_CAP)
3194 >                par = MAX_CAP;
3195 >            commonPoolParallelism = par;
3196 >            int n = par - 1; // precompute submit mask
3197 >            n |= n >>> 1; n |= n >>> 2; n |= n >>> 4;
3198 >            n |= n >>> 8; n |= n >>> 16;
3199 >            int mask = ((n + 1) << 1) - 1;
3200 >            commonPool = new ForkJoinPool(par, mask, fac, ueh);
3201 >        } catch (Exception e) {
3202 >            throw new Error(e);
3203          }
3204      }
3205  
3206 <    private boolean casEventCount(long cmp, long val) {
3207 <        return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
3208 <    }
3209 <    private boolean casWorkerCounts(int cmp, int val) {
3210 <        return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
3211 <    }
3212 <    private boolean casRunControl(int cmp, int val) {
3213 <        return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
3214 <    }
3215 <    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
3216 <        return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
3217 <    }
3218 <    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
3219 <        return _unsafe.compareAndSwapObject(this, syncStackOffset, cmp, val);
3206 >    /**
3207 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
3208 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
3209 >     * into a jdk.
3210 >     *
3211 >     * @return a sun.misc.Unsafe
3212 >     */
3213 >    private static sun.misc.Unsafe getUnsafe() {
3214 >        try {
3215 >            return sun.misc.Unsafe.getUnsafe();
3216 >        } catch (SecurityException se) {
3217 >            try {
3218 >                return java.security.AccessController.doPrivileged
3219 >                    (new java.security
3220 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
3221 >                        public sun.misc.Unsafe run() throws Exception {
3222 >                            java.lang.reflect.Field f = sun.misc
3223 >                                .Unsafe.class.getDeclaredField("theUnsafe");
3224 >                            f.setAccessible(true);
3225 >                            return (sun.misc.Unsafe) f.get(null);
3226 >                        }});
3227 >            } catch (java.security.PrivilegedActionException e) {
3228 >                throw new RuntimeException("Could not initialize intrinsics",
3229 >                                           e.getCause());
3230 >            }
3231 >        }
3232      }
3233 +
3234   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines