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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.41 by jsr166, Mon Aug 3 01:11:58 2009 UTC vs.
Revision 1.60 by dl, Sat Jul 24 20:28:18 2010 UTC

# Line 13 | Line 13 | import java.util.Arrays;
13   import java.util.Collection;
14   import java.util.Collections;
15   import java.util.List;
16 import java.util.concurrent.locks.Condition;
16   import java.util.concurrent.locks.LockSupport;
17   import java.util.concurrent.locks.ReentrantLock;
18   import java.util.concurrent.atomic.AtomicInteger;
19 < import java.util.concurrent.atomic.AtomicLong;
19 > import java.util.concurrent.CountDownLatch;
20  
21   /**
22   * An {@link ExecutorService} for running {@link ForkJoinTask}s.
23   * A {@code ForkJoinPool} provides the entry point for submissions
24 < * from non-{@code ForkJoinTask}s, as well as management and
25 < * monitoring operations.  Normally a single {@code ForkJoinPool} is
27 < * used for a large number of submitted tasks. Otherwise, use would
28 < * not usually outweigh the construction and bookkeeping overhead of
29 < * creating a large set of threads.
24 > * from non-{@code ForkJoinTask} clients, as well as management and
25 > * monitoring operations.
26   *
27 < * <p>{@code ForkJoinPool}s differ from other kinds of {@link
28 < * Executor}s mainly in that they provide <em>work-stealing</em>: all
29 < * threads in the pool attempt to find and execute subtasks created by
30 < * other active tasks (eventually blocking if none exist). This makes
31 < * them efficient when most tasks spawn other subtasks (as do most
32 < * {@code ForkJoinTask}s), as well as the mixed execution of some
33 < * plain {@code Runnable}- or {@code Callable}- based activities along
34 < * with {@code ForkJoinTask}s. When setting {@linkplain #setAsyncMode
35 < * async mode}, a {@code ForkJoinPool} may also be appropriate for use
40 < * with fine-grained tasks that are never joined. Otherwise, other
41 < * {@code ExecutorService} implementations are typically more
42 < * appropriate choices.
27 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
28 > * ExecutorService} mainly by virtue of employing
29 > * <em>work-stealing</em>: all threads in the pool attempt to find and
30 > * execute subtasks created by other active tasks (eventually blocking
31 > * waiting for work if none exist). This enables efficient processing
32 > * when most tasks spawn other subtasks (as do most {@code
33 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
34 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
35 > * with event-style tasks that are never joined.
36   *
37 < * <p>A {@code ForkJoinPool} may be constructed with a given
38 < * parallelism level (target pool size), which it attempts to maintain
39 < * by dynamically adding, suspending, or resuming threads, even if
40 < * some tasks are waiting to join others. However, no such adjustments
41 < * are performed in the face of blocked IO or other unmanaged
42 < * synchronization. The nested {@link ManagedBlocker} interface
43 < * enables extension of the kinds of synchronization accommodated.
44 < * The target parallelism level may also be changed dynamically
45 < * ({@link #setParallelism}) and thread construction can be limited
53 < * using methods {@link #setMaximumPoolSize} and/or {@link
54 < * #setMaintainsParallelism}.
37 > * <p>A {@code ForkJoinPool} is constructed with a given target
38 > * parallelism level; by default, equal to the number of available
39 > * processors. The pool attempts to maintain enough active (or
40 > * available) threads by dynamically adding, suspending, or resuming
41 > * internal worker threads, even if some tasks are stalled waiting to
42 > * join others. However, no such adjustments are guaranteed in the
43 > * face of blocked IO or other unmanaged synchronization. The nested
44 > * {@link ManagedBlocker} interface enables extension of the kinds of
45 > * synchronization accommodated.
46   *
47   * <p>In addition to execution and lifecycle control methods, this
48   * class provides status check methods (for example
# Line 60 | Line 51 | import java.util.concurrent.atomic.Atomi
51   * {@link #toString} returns indications of pool state in a
52   * convenient form for informal monitoring.
53   *
54 + * <p> As is the case with other ExecutorServices, there are three
55 + * main task execution methods summarized in the following
56 + * table. These are designed to be used by clients not already engaged
57 + * in fork/join computations in the current pool.  The main forms of
58 + * these methods accept instances of {@code ForkJoinTask}, but
59 + * overloaded forms also allow mixed execution of plain {@code
60 + * Runnable}- or {@code Callable}- based activities as well.  However,
61 + * tasks that are already executing in a pool should normally
62 + * <em>NOT</em> use these pool execution methods, but instead use the
63 + * within-computation forms listed in the table.
64 + *
65 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
66 + *  <tr>
67 + *    <td></td>
68 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
69 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
70 + *  </tr>
71 + *  <tr>
72 + *    <td> <b>Arange async execution</td>
73 + *    <td> {@link #execute(ForkJoinTask)}</td>
74 + *    <td> {@link ForkJoinTask#fork}</td>
75 + *  </tr>
76 + *  <tr>
77 + *    <td> <b>Await and obtain result</td>
78 + *    <td> {@link #invoke(ForkJoinTask)}</td>
79 + *    <td> {@link ForkJoinTask#invoke}</td>
80 + *  </tr>
81 + *  <tr>
82 + *    <td> <b>Arrange exec and obtain Future</td>
83 + *    <td> {@link #submit(ForkJoinTask)}</td>
84 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
85 + *  </tr>
86 + * </table>
87 + *
88 + * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
89 + * used for all parallel task execution in a program or subsystem.
90 + * Otherwise, use would not usually outweigh the construction and
91 + * bookkeeping overhead of creating a large set of threads. For
92 + * example, a common pool could be used for the {@code SortTasks}
93 + * illustrated in {@link RecursiveAction}. Because {@code
94 + * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
95 + * daemon} mode, there is typically no need to explicitly {@link
96 + * #shutdown} such a pool upon program exit.
97 + *
98 + * <pre>
99 + * static final ForkJoinPool mainPool = new ForkJoinPool();
100 + * ...
101 + * public void sort(long[] array) {
102 + *   mainPool.invoke(new SortTask(array, 0, array.length));
103 + * }
104 + * </pre>
105 + *
106   * <p><b>Implementation notes</b>: This implementation restricts the
107   * maximum number of running threads to 32767. Attempts to create
108 < * pools with greater than the maximum result in
108 > * pools with greater than the maximum number result in
109   * {@code IllegalArgumentException}.
110   *
111 + * <p>This implementation rejects submitted tasks (that is, by throwing
112 + * {@link RejectedExecutionException}) only when the pool is shut down
113 + * or internal resources have been exhuasted.
114 + *
115   * @since 1.7
116   * @author Doug Lea
117   */
118   public class ForkJoinPool extends AbstractExecutorService {
119  
120      /*
121 <     * See the extended comments interspersed below for design,
122 <     * rationale, and walkthroughs.
121 >     * Implementation Overview
122 >     *
123 >     * This class provides the central bookkeeping and control for a
124 >     * set of worker threads: Submissions from non-FJ threads enter
125 >     * into a submission queue. Workers take these tasks and typically
126 >     * split them into subtasks that may be stolen by other workers.
127 >     * The main work-stealing mechanics implemented in class
128 >     * ForkJoinWorkerThread give first priority to processing tasks
129 >     * from their own queues (LIFO or FIFO, depending on mode), then
130 >     * to randomized FIFO steals of tasks in other worker queues, and
131 >     * lastly to new submissions. These mechanics do not consider
132 >     * affinities, loads, cache localities, etc, so rarely provide the
133 >     * best possible performance on a given machine, but portably
134 >     * provide good throughput by averaging over these factors.
135 >     * (Further, even if we did try to use such information, we do not
136 >     * usually have a basis for exploiting it. For example, some sets
137 >     * of tasks profit from cache affinities, but others are harmed by
138 >     * cache pollution effects.)
139 >     *
140 >     * Beyond work-stealing support and essential bookkeeping, the
141 >     * main responsibility of this framework is to take actions when
142 >     * one worker is waiting to join a task stolen (or always held by)
143 >     * another.  Becauae we are multiplexing many tasks on to a pool
144 >     * of workers, we can't just let them block (as in Thread.join).
145 >     * We also cannot just reassign the joiner's run-time stack with
146 >     * another and replace it later, which would be a form of
147 >     * "continuation", that even if possible is not necessarily a good
148 >     * idea. Given that the creation costs of most threads on most
149 >     * systems mainly surrounds setting up runtime stacks, thread
150 >     * creation and switching is usually not much more expensive than
151 >     * stack creation and switching, and is more flexible). Instead we
152 >     * combine two tactics:
153 >     *
154 >     *   Helping: Arranging for the joiner to execute some task that it
155 >     *      would be running if the steal had not occurred.  Method
156 >     *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
157 >     *      links to try to find such a task.
158 >     *
159 >     *   Compensating: Unless there are already enough live threads,
160 >     *      creating or or re-activating a spare thread to compensate
161 >     *      for the (blocked) joiner until it unblocks.  Spares then
162 >     *      suspend at their next opportunity or eventually die if
163 >     *      unused for too long.  See below and the internal
164 >     *      documentation for tryAwaitJoin for more details about
165 >     *      compensation rules.
166 >     *
167 >     * Because the determining existence of conservatively safe
168 >     * helping targets, the availability of already-created spares,
169 >     * and the apparent need to create new spares are all racy and
170 >     * require heuristic guidance, joins (in
171 >     * ForkJoinWorkerThread.joinTask) interleave these options until
172 >     * successful.  Creating a new spare always succeeds, but also
173 >     * increases application footprint, so we try to avoid it, within
174 >     * reason.
175 >     *
176 >     * The ManagedBlocker extension API can't use helping so uses a
177 >     * special version of compensation in method awaitBlocker.
178 >     *
179 >     * The main throughput advantages of work-stealing stem from
180 >     * decentralized control -- workers mostly steal tasks from each
181 >     * other. We do not want to negate this by creating bottlenecks
182 >     * implementing other management responsibilities. So we use a
183 >     * collection of techniques that avoid, reduce, or cope well with
184 >     * contention. These entail several instances of bit-packing into
185 >     * CASable fields to maintain only the minimally required
186 >     * atomicity. To enable such packing, we restrict maximum
187 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
188 >     * unbalanced increments and decrements) to fit into a 16 bit
189 >     * field, which is far in excess of normal operating range.  Even
190 >     * though updates to some of these bookkeeping fields do sometimes
191 >     * contend with each other, they don't normally cache-contend with
192 >     * updates to others enough to warrant memory padding or
193 >     * isolation. So they are all held as fields of ForkJoinPool
194 >     * objects.  The main capabilities are as follows:
195 >     *
196 >     * 1. Creating and removing workers. Workers are recorded in the
197 >     * "workers" array. This is an array as opposed to some other data
198 >     * structure to support index-based random steals by workers.
199 >     * Updates to the array recording new workers and unrecording
200 >     * terminated ones are protected from each other by a lock
201 >     * (workerLock) but the array is otherwise concurrently readable,
202 >     * and accessed directly by workers. To simplify index-based
203 >     * operations, the array size is always a power of two, and all
204 >     * readers must tolerate null slots. Currently, all worker thread
205 >     * creation is on-demand, triggered by task submissions,
206 >     * replacement of terminated workers, and/or compensation for
207 >     * blocked workers. However, all other support code is set up to
208 >     * work with other policies.
209 >     *
210 >     * 2. Bookkeeping for dynamically adding and removing workers. We
211 >     * aim to approximately maintain the given level of parallelism.
212 >     * When some workers are known to be blocked (on joins or via
213 >     * ManagedBlocker), we may create or resume others to take their
214 >     * place until they unblock (see below). Implementing this
215 >     * requires counts of the number of "running" threads (i.e., those
216 >     * that are neither blocked nor artifically suspended) as well as
217 >     * the total number.  These two values are packed into one field,
218 >     * "workerCounts" because we need accurate snapshots when deciding
219 >     * to create, resume or suspend.  Note however that the
220 >     * correspondance of these counts to reality is not guaranteed. In
221 >     * particular updates for unblocked threads may lag until they
222 >     * actually wake up.
223 >     *
224 >     * 3. Maintaining global run state. The run state of the pool
225 >     * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
226 >     * those in other Executor implementations, as well as a count of
227 >     * "active" workers -- those that are, or soon will be, or
228 >     * recently were executing tasks. The runLevel and active count
229 >     * are packed together in order to correctly trigger shutdown and
230 >     * termination. Without care, active counts can be subject to very
231 >     * high contention.  We substantially reduce this contention by
232 >     * relaxing update rules.  A worker must claim active status
233 >     * prospectively, by activating if it sees that a submitted or
234 >     * stealable task exists (it may find after activating that the
235 >     * task no longer exists). It stays active while processing this
236 >     * task (if it exists) and any other local subtasks it produces,
237 >     * until it cannot find any other tasks. It then tries
238 >     * inactivating (see method preStep), but upon update contention
239 >     * instead scans for more tasks, later retrying inactivation if it
240 >     * doesn't find any.
241 >     *
242 >     * 4. Managing idle workers waiting for tasks. We cannot let
243 >     * workers spin indefinitely scanning for tasks when none are
244 >     * available. On the other hand, we must quickly prod them into
245 >     * action when new tasks are submitted or generated.  We
246 >     * park/unpark these idle workers using an event-count scheme.
247 >     * Field eventCount is incremented upon events that may enable
248 >     * workers that previously could not find a task to now find one:
249 >     * Submission of a new task to the pool, or another worker pushing
250 >     * a task onto a previously empty queue.  (We also use this
251 >     * mechanism for termination and reconfiguration actions that
252 >     * require wakeups of idle workers).  Each worker maintains its
253 >     * last known event count, and blocks when a scan for work did not
254 >     * find a task AND its lastEventCount matches the current
255 >     * eventCount. Waiting idle workers are recorded in a variant of
256 >     * Treiber stack headed by field eventWaiters which, when nonzero,
257 >     * encodes the thread index and count awaited for by the worker
258 >     * thread most recently calling eventSync. This thread in turn has
259 >     * a record (field nextEventWaiter) for the next waiting worker.
260 >     * In addition to allowing simpler decisions about need for
261 >     * wakeup, the event count bits in eventWaiters serve the role of
262 >     * tags to avoid ABA errors in Treiber stacks.  To reduce delays
263 >     * in task diffusion, workers not otherwise occupied may invoke
264 >     * method releaseWaiters, that removes and signals (unparks)
265 >     * workers not waiting on current count. To minimize task
266 >     * production stalls associate with signalling, any worker pushing
267 >     * a task on an empty queue invokes the weaker method signalWork,
268 >     * that only releases idle workers until it detects interference
269 >     * by other threads trying to release, and lets them take
270 >     * over. The net effect is a tree-like diffusion of signals, where
271 >     * released threads (and possibly others) help with unparks.  To
272 >     * further reduce contention effects a bit, failed CASes to
273 >     * increment field eventCount are tolerated without retries.
274 >     * Conceptually they are merged into the same event, which is OK
275 >     * when their only purpose is to enable workers to scan for work.
276 >     *
277 >     * 5. Managing suspension of extra workers. When a worker is about
278 >     * to block waiting for a join (or via ManagedBlockers), we may
279 >     * create a new thread to maintain parallelism level, or at least
280 >     * avoid starvation. Usually, extra threads are needed for only
281 >     * very short periods, yet join dependencies are such that we
282 >     * sometimes need them in bursts. Rather than create new threads
283 >     * each time this happens, we suspend no-longer-needed extra ones
284 >     * as "spares". For most purposes, we don't distinguish "extra"
285 >     * spare threads from normal "core" threads: On each call to
286 >     * preStep (the only point at which we can do this) a worker
287 >     * checks to see if there are now too many running workers, and if
288 >     * so, suspends itself.  Methods tryAwaitJoin and awaitBlocker
289 >     * look for suspended threads to resume before considering
290 >     * creating a new replacement. We don't need a special data
291 >     * structure to maintain spares; simply scanning the workers array
292 >     * looking for worker.isSuspended() is fine because the calling
293 >     * thread is otherwise not doing anything useful anyway; we are at
294 >     * least as happy if after locating a spare, the caller doesn't
295 >     * actually block because the join is ready before we try to
296 >     * adjust and compensate.  Note that this is intrinsically racy.
297 >     * One thread may become a spare at about the same time as another
298 >     * is needlessly being created. We counteract this and related
299 >     * slop in part by requiring resumed spares to immediately recheck
300 >     * (in preStep) to see whether they they should re-suspend. The
301 >     * only effective difference between "extra" and "core" threads is
302 >     * that we allow the "extra" ones to time out and die if they are
303 >     * not resumed within a keep-alive interval of a few seconds. This
304 >     * is implemented mainly within ForkJoinWorkerThread, but requires
305 >     * some coordination (isTrimmed() -- meaning killed while
306 >     * suspended) to correctly maintain pool counts.
307 >     *
308 >     * 6. Deciding when to create new workers. The main dynamic
309 >     * control in this class is deciding when to create extra threads,
310 >     * in methods awaitJoin and awaitBlocker. We always need to create
311 >     * one when the number of running threads would become zero and
312 >     * all workers are busy. However, this is not easy to detect
313 >     * reliably in the presence of transients so we use retries and
314 >     * allow slack (in tryAwaitJoin) to reduce false alarms.  These
315 >     * effectively reduce churn at the price of systematically
316 >     * undershooting target parallelism when many threads are blocked.
317 >     * However, biasing toward undeshooting partially compensates for
318 >     * the above mechanics to suspend extra threads, that normally
319 >     * lead to overshoot because we can only suspend workers
320 >     * in-between top-level actions. It also better copes with the
321 >     * fact that some of the methods in this class tend to never
322 >     * become compiled (but are interpreted), so some components of
323 >     * the entire set of controls might execute many times faster than
324 >     * others. And similarly for cases where the apparent lack of work
325 >     * is just due to GC stalls and other transient system activity.
326 >     *
327 >     * Beware that there is a lot of representation-level coupling
328 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
329 >     * ForkJoinTask.  For example, direct access to "workers" array by
330 >     * workers, and direct access to ForkJoinTask.status by both
331 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
332 >     * trying to reduce this, since any associated future changes in
333 >     * representations will need to be accompanied by algorithmic
334 >     * changes anyway.
335 >     *
336 >     * Style notes: There are lots of inline assignments (of form
337 >     * "while ((local = field) != 0)") which are usually the simplest
338 >     * way to ensure read orderings. Also several occurrences of the
339 >     * unusual "do {} while(!cas...)" which is the simplest way to
340 >     * force an update of a CAS'ed variable. There are also other
341 >     * coding oddities that help some methods perform reasonably even
342 >     * when interpreted (not compiled), at the expense of messiness.
343 >     *
344 >     * The order of declarations in this file is: (1) statics (2)
345 >     * fields (along with constants used when unpacking some of them)
346 >     * (3) internal control methods (4) callbacks and other support
347 >     * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
348 >     * methods (plus a few little helpers).
349       */
350  
78    /** Mask for packing and unpacking shorts */
79    private static final int  shortMask = 0xffff;
80
81    /** Max pool size -- must be a power of two minus 1 */
82    private static final int MAX_THREADS =  0x7FFF;
83
351      /**
352       * Factory for creating new {@link ForkJoinWorkerThread}s.
353       * A {@code ForkJoinWorkerThreadFactory} must be defined and used
# Line 92 | Line 359 | public class ForkJoinPool extends Abstra
359           * Returns a new worker thread operating in the given pool.
360           *
361           * @param pool the pool this thread works in
362 <         * @throws NullPointerException if pool is null
362 >         * @throws NullPointerException if the pool is null
363           */
364          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
365      }
# Line 101 | Line 368 | public class ForkJoinPool extends Abstra
368       * Default ForkJoinWorkerThreadFactory implementation; creates a
369       * new ForkJoinWorkerThread.
370       */
371 <    static class  DefaultForkJoinWorkerThreadFactory
371 >    static class DefaultForkJoinWorkerThreadFactory
372          implements ForkJoinWorkerThreadFactory {
373          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
374 <            try {
108 <                return new ForkJoinWorkerThread(pool);
109 <            } catch (OutOfMemoryError oom)  {
110 <                return null;
111 <            }
374 >            return new ForkJoinWorkerThread(pool);
375          }
376      }
377  
# Line 144 | Line 407 | public class ForkJoinPool extends Abstra
407          new AtomicInteger();
408  
409      /**
410 <     * Array holding all worker threads in the pool. Initialized upon
411 <     * first use. Array size must be a power of two.  Updates and
149 <     * replacements are protected by workerLock, but it is always kept
150 <     * in a consistent enough state to be randomly accessed without
151 <     * locking by workers performing work-stealing.
410 >     * Absolute bound for parallelism level. Twice this number must
411 >     * fit into a 16bit field to enable word-packing for some counts.
412       */
413 <    volatile ForkJoinWorkerThread[] workers;
413 >    private static final int MAX_THREADS = 0x7fff;
414  
415      /**
416 <     * Lock protecting access to workers.
416 >     * Array holding all worker threads in the pool.  Array size must
417 >     * be a power of two.  Updates and replacements are protected by
418 >     * workerLock, but the array is always kept in a consistent enough
419 >     * state to be randomly accessed without locking by workers
420 >     * performing work-stealing, as well as other traversal-based
421 >     * methods in this class. All readers must tolerate that some
422 >     * array slots may be null.
423       */
424 <    private final ReentrantLock workerLock;
424 >    volatile ForkJoinWorkerThread[] workers;
425  
426      /**
427 <     * Condition for awaitTermination.
427 >     * Queue for external submissions.
428       */
429 <    private final Condition termination;
429 >    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
430  
431      /**
432 <     * The uncaught exception handler used when any worker
167 <     * abruptly terminates
432 >     * Lock protecting updates to workers array.
433       */
434 <    private Thread.UncaughtExceptionHandler ueh;
434 >    private final ReentrantLock workerLock;
435  
436      /**
437 <     * Creation factory for worker threads.
437 >     * Latch released upon termination.
438       */
439 <    private final ForkJoinWorkerThreadFactory factory;
439 >    private final Phaser termination;
440  
441      /**
442 <     * Head of stack of threads that were created to maintain
178 <     * parallelism when other threads blocked, but have since
179 <     * suspended when the parallelism level rose.
442 >     * Creation factory for worker threads.
443       */
444 <    private volatile WaitQueueNode spareStack;
444 >    private final ForkJoinWorkerThreadFactory factory;
445  
446      /**
447       * Sum of per-thread steal counts, updated only when threads are
448       * idle or terminating.
449       */
450 <    private final AtomicLong stealCount;
450 >    private volatile long stealCount;
451  
452      /**
453 <     * Queue for external submissions.
453 >     * Encoded record of top of treiber stack of threads waiting for
454 >     * events. The top 32 bits contain the count being waited for. The
455 >     * bottom word contains one plus the pool index of waiting worker
456 >     * thread.
457       */
458 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
458 >    private volatile long eventWaiters;
459 >
460 >    private static final int  EVENT_COUNT_SHIFT = 32;
461 >    private static final long WAITER_ID_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
462  
463      /**
464 <     * Head of Treiber stack for barrier sync. See below for explanation.
464 >     * A counter for events that may wake up worker threads:
465 >     *   - Submission of a new task to the pool
466 >     *   - A worker pushing a task on an empty queue
467 >     *   - termination and reconfiguration
468       */
469 <    private volatile WaitQueueNode syncStack;
469 >    private volatile int eventCount;
470  
471      /**
472 <     * The count for event barrier
473 <     */
474 <    private volatile long eventCount;
472 >     * Lifecycle control. The low word contains the number of workers
473 >     * that are (probably) executing tasks. This value is atomically
474 >     * incremented before a worker gets a task to run, and decremented
475 >     * when worker has no tasks and cannot find any.  Bits 16-18
476 >     * contain runLevel value. When all are zero, the pool is
477 >     * running. Level transitions are monotonic (running -> shutdown
478 >     * -> terminating -> terminated) so each transition adds a bit.
479 >     * These are bundled together to ensure consistent read for
480 >     * termination checks (i.e., that runLevel is at least SHUTDOWN
481 >     * and active threads is zero).
482 >     */
483 >    private volatile int runState;
484 >
485 >    // Note: The order among run level values matters.
486 >    private static final int RUNLEVEL_SHIFT     = 16;
487 >    private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
488 >    private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
489 >    private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
490 >    private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
491 >    private static final int ONE_ACTIVE         = 1; // active update delta
492  
493      /**
494 <     * Pool number, just for assigning useful names to worker threads
494 >     * Holds number of total (i.e., created and not yet terminated)
495 >     * and running (i.e., not blocked on joins or other managed sync)
496 >     * threads, packed together to ensure consistent snapshot when
497 >     * making decisions about creating and suspending spare
498 >     * threads. Updated only by CAS. Note that adding a new worker
499 >     * requires incrementing both counts, since workers start off in
500 >     * running state.
501       */
502 <    private final int poolNumber;
502 >    private volatile int workerCounts;
503 >
504 >    private static final int TOTAL_COUNT_SHIFT  = 16;
505 >    private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
506 >    private static final int ONE_RUNNING        = 1;
507 >    private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
508  
509      /**
510 <     * The maximum allowed pool size
510 >     * The target parallelism level.
511 >     * Accessed directly by ForkJoinWorkerThreads.
512       */
513 <    private volatile int maxPoolSize;
513 >    final int parallelism;
514  
515      /**
516 <     * The desired parallelism level, updated only under workerLock.
516 >     * True if use local fifo, not default lifo, for local polling
517 >     * Read by, and replicated by ForkJoinWorkerThreads
518       */
519 <    private volatile int parallelism;
519 >    final boolean locallyFifo;
520  
521      /**
522 <     * True if use local fifo, not default lifo, for local polling
522 >     * The uncaught exception handler used when any worker abruptly
523 >     * terminates.
524       */
525 <    private volatile boolean locallyFifo;
525 >    private final Thread.UncaughtExceptionHandler ueh;
526  
527      /**
528 <     * Holds number of total (i.e., created and not yet terminated)
226 <     * and running (i.e., not blocked on joins or other managed sync)
227 <     * threads, packed into one int to ensure consistent snapshot when
228 <     * making decisions about creating and suspending spare
229 <     * threads. Updated only by CAS.  Note: CASes in
230 <     * updateRunningCount and preJoin assume that running active count
231 <     * is in low word, so need to be modified if this changes.
528 >     * Pool number, just for assigning useful names to worker threads
529       */
530 <    private volatile int workerCounts;
530 >    private final int poolNumber;
531  
532 <    private static int totalCountOf(int s)           { return s >>> 16;  }
533 <    private static int runningCountOf(int s)         { return s & shortMask; }
237 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
532 >    // Utilities for CASing fields. Note that several of these
533 >    // are manually inlined by callers
534  
535      /**
536 <     * Adds delta (which may be negative) to running count.  This must
241 <     * be called before (with negative arg) and after (with positive)
242 <     * any managed synchronization (i.e., mainly, joins).
243 <     *
244 <     * @param delta the number to add
536 >     * Increments running count.  Also used by ForkJoinTask.
537       */
538 <    final void updateRunningCount(int delta) {
539 <        int s;
540 <        do {} while (!casWorkerCounts(s = workerCounts, s + delta));
538 >    final void incrementRunningCount() {
539 >        int c;
540 >        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
541 >                                               c = workerCounts,
542 >                                               c + ONE_RUNNING));
543      }
544  
545      /**
546 <     * Adds delta (which may be negative) to both total and running
253 <     * count.  This must be called upon creation and termination of
254 <     * worker threads.
255 <     *
256 <     * @param delta the number to add
546 >     * Tries to decrement running count unless already zero
547       */
548 <    private void updateWorkerCount(int delta) {
549 <        int d = delta + (delta << 16); // add to both lo and hi parts
550 <        int s;
551 <        do {} while (!casWorkerCounts(s = workerCounts, s + d));
548 >    final boolean tryDecrementRunningCount() {
549 >        int wc = workerCounts;
550 >        if ((wc & RUNNING_COUNT_MASK) == 0)
551 >            return false;
552 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
553 >                                        wc, wc - ONE_RUNNING);
554      }
555  
556      /**
557 <     * Lifecycle control. High word contains runState, low word
266 <     * contains the number of workers that are (probably) executing
267 <     * tasks. This value is atomically incremented before a worker
268 <     * gets a task to run, and decremented when worker has no tasks
269 <     * and cannot find any. These two fields are bundled together to
270 <     * support correct termination triggering.  Note: activeCount
271 <     * CAS'es cheat by assuming active count is in low word, so need
272 <     * to be modified if this changes
557 >     * Tries to increment running count
558       */
559 <    private volatile int runControl;
560 <
561 <    // RunState values. Order among values matters
562 <    private static final int RUNNING     = 0;
563 <    private static final int SHUTDOWN    = 1;
279 <    private static final int TERMINATING = 2;
280 <    private static final int TERMINATED  = 3;
281 <
282 <    private static int runStateOf(int c)             { return c >>> 16; }
283 <    private static int activeCountOf(int c)          { return c & shortMask; }
284 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
559 >    final boolean tryIncrementRunningCount() {
560 >        int wc;
561 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
562 >                                        wc = workerCounts, wc + ONE_RUNNING);
563 >    }
564  
565      /**
566       * Tries incrementing active count; fails on contention.
567 <     * Called by workers before/during executing tasks.
567 >     * Called by workers before executing tasks.
568       *
569       * @return true on success
570       */
571      final boolean tryIncrementActiveCount() {
572 <        int c = runControl;
573 <        return casRunControl(c, c+1);
572 >        int c;
573 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
574 >                                        c = runState, c + ONE_ACTIVE);
575      }
576  
577      /**
578       * Tries decrementing active count; fails on contention.
579 <     * Possibly triggers termination on success.
300 <     * Called by workers when they can't find tasks.
301 <     *
302 <     * @return true on success
579 >     * Called when workers cannot find tasks to run.
580       */
581      final boolean tryDecrementActiveCount() {
582 <        int c = runControl;
583 <        int nextc = c - 1;
584 <        if (!casRunControl(c, nextc))
308 <            return false;
309 <        if (canTerminateOnShutdown(nextc))
310 <            terminateOnShutdown();
311 <        return true;
582 >        int c;
583 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
584 >                                        c = runState, c - ONE_ACTIVE);
585      }
586  
587      /**
588 <     * Returns {@code true} if argument represents zero active count
589 <     * and nonzero runstate, which is the triggering condition for
317 <     * terminating on shutdown.
588 >     * Advances to at least the given level. Returns true if not
589 >     * already in at least the given level.
590       */
591 <    private static boolean canTerminateOnShutdown(int c) {
592 <        // i.e. least bit is nonzero runState bit
593 <        return ((c & -c) >>> 16) != 0;
591 >    private boolean advanceRunLevel(int level) {
592 >        for (;;) {
593 >            int s = runState;
594 >            if ((s & level) != 0)
595 >                return false;
596 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
597 >                return true;
598 >        }
599      }
600  
601 +    // workers array maintenance
602 +
603      /**
604 <     * Transition run state to at least the given state. Return true
326 <     * if not already at least given state.
604 >     * Records and returns a workers array index for new worker.
605       */
606 <    private boolean transitionRunStateTo(int state) {
607 <        for (;;) {
608 <            int c = runControl;
609 <            if (runStateOf(c) >= state)
610 <                return false;
611 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
612 <                return true;
606 >    private int recordWorker(ForkJoinWorkerThread w) {
607 >        // Try using slot totalCount-1. If not available, scan and/or resize
608 >        int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
609 >        final ReentrantLock lock = this.workerLock;
610 >        lock.lock();
611 >        try {
612 >            ForkJoinWorkerThread[] ws = workers;
613 >            int nws = ws.length;
614 >            if (k < 0 || k >= nws || ws[k] != null) {
615 >                for (k = 0; k < nws && ws[k] != null; ++k)
616 >                    ;
617 >                if (k == nws)
618 >                    ws = Arrays.copyOf(ws, nws << 1);
619 >            }
620 >            ws[k] = w;
621 >            workers = ws; // volatile array write ensures slot visibility
622 >        } finally {
623 >            lock.unlock();
624          }
625 +        return k;
626      }
627  
628      /**
629 <     * Controls whether to add spares to maintain parallelism
629 >     * Nulls out record of worker in workers array
630       */
631 <    private volatile boolean maintainsParallelism;
631 >    private void forgetWorker(ForkJoinWorkerThread w) {
632 >        int idx = w.poolIndex;
633 >        // Locking helps method recordWorker avoid unecessary expansion
634 >        final ReentrantLock lock = this.workerLock;
635 >        lock.lock();
636 >        try {
637 >            ForkJoinWorkerThread[] ws = workers;
638 >            if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
639 >                ws[idx] = null;
640 >        } finally {
641 >            lock.unlock();
642 >        }
643 >    }
644  
645 <    // Constructors
645 >    // adding and removing workers
646  
647      /**
648 <     * Creates a {@code ForkJoinPool} with a pool size equal to the
649 <     * number of processors available on the system, using the
650 <     * {@linkplain #defaultForkJoinWorkerThreadFactory default thread factory}.
648 >     * Tries to create and add new worker. Assumes that worker counts
649 >     * are already updated to accommodate the worker, so adjusts on
650 >     * failure.
651       *
652 <     * @throws SecurityException if a security manager exists and
351 <     *         the caller is not permitted to modify threads
352 <     *         because it does not hold {@link
353 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
652 >     * @return new worker or null if creation failed
653       */
654 <    public ForkJoinPool() {
655 <        this(Runtime.getRuntime().availableProcessors(),
656 <             defaultForkJoinWorkerThreadFactory);
654 >    private ForkJoinWorkerThread addWorker() {
655 >        ForkJoinWorkerThread w = null;
656 >        try {
657 >            w = factory.newThread(this);
658 >        } finally { // Adjust on either null or exceptional factory return
659 >            if (w == null)
660 >                onWorkerCreationFailure();
661 >        }
662 >        if (w != null)
663 >            w.start(recordWorker(w), ueh);
664 >        return w;
665      }
666  
667      /**
668 <     * Creates a {@code ForkJoinPool} with the indicated parallelism level
362 <     * threads and using the
363 <     * {@linkplain #defaultForkJoinWorkerThreadFactory default thread factory}.
364 <     *
365 <     * @param parallelism the number of worker threads
366 <     * @throws IllegalArgumentException if parallelism less than or
367 <     * equal to zero
368 <     * @throws SecurityException if a security manager exists and
369 <     *         the caller is not permitted to modify threads
370 <     *         because it does not hold {@link
371 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
668 >     * Adjusts counts upon failure to create worker
669       */
670 <    public ForkJoinPool(int parallelism) {
671 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
670 >    private void onWorkerCreationFailure() {
671 >        for (;;) {
672 >            int wc = workerCounts;
673 >            int rc = wc & RUNNING_COUNT_MASK;
674 >            int tc = wc >>> TOTAL_COUNT_SHIFT;
675 >            if (rc == 0 || wc == 0)
676 >                Thread.yield(); // must wait for other counts to settle
677 >            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
678 >                                              wc - (ONE_RUNNING|ONE_TOTAL)))
679 >                break;
680 >        }
681 >        tryTerminate(false); // in case of failure during shutdown
682      }
683  
684      /**
685 <     * Creates a {@code ForkJoinPool} with parallelism equal to the
686 <     * number of processors available on the system and using the
380 <     * given thread factory.
381 <     *
382 <     * @param factory the factory for creating new threads
383 <     * @throws NullPointerException if factory is null
384 <     * @throws SecurityException if a security manager exists and
385 <     *         the caller is not permitted to modify threads
386 <     *         because it does not hold {@link
387 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
685 >     * Creates enough total workers to establish target parallelism,
686 >     * giving up if terminating or addWorker fails
687       */
688 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
689 <        this(Runtime.getRuntime().availableProcessors(), factory);
688 >    private void ensureEnoughTotalWorkers() {
689 >        int wc;
690 >        while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
691 >               runState < TERMINATING) {
692 >            if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
693 >                                          wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
694 >                 addWorker() == null))
695 >                break;
696 >        }
697      }
698  
699      /**
700 <     * Creates a {@code ForkJoinPool} with the given parallelism and
701 <     * thread factory.
700 >     * Final callback from terminating worker.  Removes record of
701 >     * worker from array, and adjusts counts. If pool is shutting
702 >     * down, tries to complete terminatation, else possibly replaces
703 >     * the worker.
704       *
705 <     * @param parallelism the targeted number of worker threads
398 <     * @param factory the factory for creating new threads
399 <     * @throws IllegalArgumentException if parallelism less than or
400 <     * equal to zero, or greater than implementation limit
401 <     * @throws NullPointerException if factory is null
402 <     * @throws SecurityException if a security manager exists and
403 <     *         the caller is not permitted to modify threads
404 <     *         because it does not hold {@link
405 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
705 >     * @param w the worker
706       */
707 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
708 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
709 <            throw new IllegalArgumentException();
710 <        if (factory == null)
711 <            throw new NullPointerException();
712 <        checkPermission();
713 <        this.factory = factory;
714 <        this.parallelism = parallelism;
715 <        this.maxPoolSize = MAX_THREADS;
716 <        this.maintainsParallelism = true;
717 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
718 <        this.workerLock = new ReentrantLock();
719 <        this.termination = workerLock.newCondition();
720 <        this.stealCount = new AtomicLong();
721 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
722 <        // worker array and workers are lazily constructed
707 >    final void workerTerminated(ForkJoinWorkerThread w) {
708 >        if (w.active) { // force inactive
709 >            w.active = false;
710 >            do {} while (!tryDecrementActiveCount());
711 >        }
712 >        forgetWorker(w);
713 >
714 >        // Decrement total count, and if was running, running count
715 >        // Spin (waiting for other updates) if either would be negative
716 >        int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
717 >        int unit = ONE_TOTAL + nr;
718 >        for (;;) {
719 >            int wc = workerCounts;
720 >            int rc = wc & RUNNING_COUNT_MASK;
721 >            int tc = wc >>> TOTAL_COUNT_SHIFT;
722 >            if (rc - nr < 0 || tc == 0)
723 >                Thread.yield(); // back off if waiting for other updates
724 >            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
725 >                                              wc, wc - unit))
726 >                break;
727 >        }
728 >
729 >        accumulateStealCount(w); // collect final count
730 >        if (!tryTerminate(false))
731 >            ensureEnoughTotalWorkers();
732      }
733  
734 +    // Waiting for and signalling events
735 +
736      /**
737 <     * Creates a new worker thread using factory.
738 <     *
428 <     * @param index the index to assign worker
429 <     * @return new worker, or null if factory failed
737 >     * Releases workers blocked on a count not equal to current count.
738 >     * @return true if any released
739       */
740 <    private ForkJoinWorkerThread createWorker(int index) {
741 <        Thread.UncaughtExceptionHandler h = ueh;
742 <        ForkJoinWorkerThread w = factory.newThread(this);
743 <        if (w != null) {
744 <            w.poolIndex = index;
745 <            w.setDaemon(true);
746 <            w.setAsyncMode(locallyFifo);
747 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
748 <            if (h != null)
749 <                w.setUncaughtExceptionHandler(h);
740 >    private void releaseWaiters() {
741 >        long top;
742 >        while ((top = eventWaiters) != 0L) {
743 >            ForkJoinWorkerThread[] ws = workers;
744 >            int n = ws.length;
745 >            for (;;) {
746 >                int i = ((int)(top & WAITER_ID_MASK)) - 1;
747 >                int e = (int)(top >>> EVENT_COUNT_SHIFT);
748 >                if (i < 0 || e == eventCount)
749 >                    return;
750 >                ForkJoinWorkerThread w;
751 >                if (i < n && (w = ws[i]) != null &&
752 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
753 >                                              top, w.nextWaiter)) {
754 >                    LockSupport.unpark(w);
755 >                    top = eventWaiters;
756 >                }
757 >                else
758 >                    break;      // possibly stale; reread
759 >            }
760          }
442        return w;
761      }
762  
763      /**
764 <     * Returns a good size for worker array given pool size.
765 <     * Currently requires size to be a power of two.
764 >     * Ensures eventCount on exit is different (mod 2^32) than on
765 >     * entry and wakes up all waiters
766       */
767 <    private static int arraySizeFor(int poolSize) {
768 <        return (poolSize <= 1) ? 1 :
769 <            (1 << (32 - Integer.numberOfLeadingZeros(poolSize-1)));
767 >    private void signalEvent() {
768 >        int c;
769 >        do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
770 >                                               c = eventCount, c+1));
771 >        releaseWaiters();
772      }
773  
774      /**
775 <     * Creates or resizes array if necessary to hold newLength.
776 <     * Call only under exclusion.
457 <     *
458 <     * @return the array
775 >     * Advances eventCount and releases waiters until interference by
776 >     * other releasing threads is detected.
777       */
778 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
779 <        ForkJoinWorkerThread[] ws = workers;
780 <        if (ws == null)
781 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
782 <        else if (newLength > ws.length)
783 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
784 <        else
785 <            return ws;
778 >    final void signalWork() {
779 >        int c;
780 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c=eventCount, c+1);
781 >        long top;
782 >        while ((top = eventWaiters) != 0L) {
783 >            int ec = eventCount;
784 >            ForkJoinWorkerThread[] ws = workers;
785 >            int n = ws.length;
786 >            for (;;) {
787 >                int i = ((int)(top & WAITER_ID_MASK)) - 1;
788 >                int e = (int)(top >>> EVENT_COUNT_SHIFT);
789 >                if (i < 0 || e == ec)
790 >                    return;
791 >                ForkJoinWorkerThread w;
792 >                if (i < n && (w = ws[i]) != null &&
793 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
794 >                                              top, top = w.nextWaiter)) {
795 >                    LockSupport.unpark(w);
796 >                    if (top != eventWaiters) // let someone else take over
797 >                        return;
798 >                }
799 >                else
800 >                    break;      // possibly stale; reread
801 >            }
802 >        }
803      }
804  
805      /**
806 <     * Tries to shrink workers into smaller array after one or more terminate.
806 >     * Blockss worker until terminating or event count
807 >     * advances from last value held by worker
808 >     *
809 >     * @param w the calling worker thread
810       */
811 <    private void tryShrinkWorkerArray() {
812 <        ForkJoinWorkerThread[] ws = workers;
813 <        if (ws != null) {
814 <            int len = ws.length;
815 <            int last = len - 1;
816 <            while (last >= 0 && ws[last] == null)
817 <                --last;
818 <            int newLength = arraySizeFor(last+1);
819 <            if (newLength < len)
820 <                workers = Arrays.copyOf(ws, newLength);
811 >    private void eventSync(ForkJoinWorkerThread w) {
812 >        int wec = w.lastEventCount;
813 >        long nextTop = (((long)wec << EVENT_COUNT_SHIFT) |
814 >                        ((long)(w.poolIndex + 1)));
815 >        long top;
816 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
817 >               (((int)(top = eventWaiters) & WAITER_ID_MASK) == 0 ||
818 >                (int)(top >>> EVENT_COUNT_SHIFT) == wec) &&
819 >               eventCount == wec) {
820 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
821 >                                          w.nextWaiter = top, nextTop)) {
822 >                accumulateStealCount(w); // transfer steals while idle
823 >                Thread.interrupted();    // clear/ignore interrupt
824 >                while (eventCount == wec)
825 >                    w.doPark();
826 >                break;
827 >            }
828          }
829 +        w.lastEventCount = eventCount;
830      }
831  
832      /**
833 <     * Initializes workers if necessary.
833 >     * Callback from workers invoked upon each top-level action (i.e.,
834 >     * stealing a task or taking a submission and running
835 >     * it). Performs one or both of the following:
836 >     *
837 >     * * If the worker cannot find work, updates its active status to
838 >     * inactive and updates activeCount unless there is contention, in
839 >     * which case it may try again (either in this or a subsequent
840 >     * call).  Additionally, awaits the next task event and/or helps
841 >     * wake up other releasable waiters.
842 >     *
843 >     * * If there are too many running threads, suspends this worker
844 >     * (first forcing inactivation if necessary).  If it is not
845 >     * resumed before a keepAlive elapses, the worker may be "trimmed"
846 >     * -- killed while suspended within suspendAsSpare. Otherwise,
847 >     * upon resume it rechecks to make sure that it is still needed.
848 >     *
849 >     * @param w the worker
850 >     * @param retries the number of scans by caller failing to find work
851 >     * find any (in which case it may block waiting for work).
852       */
853 <    final void ensureWorkerInitialization() {
854 <        ForkJoinWorkerThread[] ws = workers;
855 <        if (ws == null) {
856 <            final ReentrantLock lock = this.workerLock;
857 <            lock.lock();
858 <            try {
859 <                ws = workers;
860 <                if (ws == null) {
861 <                    int ps = parallelism;
862 <                    ws = ensureWorkerArrayCapacity(ps);
863 <                    for (int i = 0; i < ps; ++i) {
864 <                        ForkJoinWorkerThread w = createWorker(i);
865 <                        if (w != null) {
866 <                            ws[i] = w;
867 <                            w.start();
868 <                            updateWorkerCount(1);
853 >    final void preStep(ForkJoinWorkerThread w, int retries) {
854 >        boolean active = w.active;
855 >        boolean inactivate = active && retries > 0;
856 >        for (;;) {
857 >            int rs, wc;
858 >            if (inactivate &&
859 >                UNSAFE.compareAndSwapInt(this, runStateOffset,
860 >                                         rs = runState, rs - ONE_ACTIVE))
861 >                inactivate = active = w.active = false;
862 >            if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= parallelism) {
863 >                if (retries > 0) {
864 >                    if (retries > 1 && !active)
865 >                        eventSync(w);
866 >                    releaseWaiters();
867 >                }
868 >                break;
869 >            }
870 >            if (!(inactivate |= active) &&  // must inactivate to suspend
871 >                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
872 >                                         wc, wc - ONE_RUNNING) &&
873 >                !w.suspendAsSpare())             // false if trimmed
874 >                break;
875 >        }
876 >    }
877 >
878 >    /**
879 >     * Awaits join of the given task if enough threads, or can resume
880 >     * or create a spare. Fails (in which case the given task might
881 >     * not be done) upon contention or lack of decision about
882 >     * blocking.
883 >     *
884 >     * We allow blocking if:
885 >     *
886 >     * 1. There would still be at least as many running threads as
887 >     *    parallelism level if this thread blocks.
888 >     *
889 >     * 2. A spare is resumed to replace this worker. We tolerate
890 >     *    races in the decision to replace when a spare is found.
891 >     *    This may release too many, but if so, the superfluous ones
892 >     *    will re-suspend via preStep().
893 >     *
894 >     * 3. After #spares repeated retries, there are fewer than #spare
895 >     *    threads not running. We allow this slack to avoid hysteresis
896 >     *    and as a hedge against lag/uncertainty of running count
897 >     *    estimates when signalling or unblocking stalls.
898 >     *
899 >     * 4. All existing workers are busy (as rechecked via #spares
900 >     *    repeated retries by caller) and a new spare is created.
901 >     *
902 >     * If none of the above hold, we escape out by re-incrementing
903 >     * count and returning to caller, which can retry later.
904 >     *
905 >     * @param joinMe the task to join
906 >     * @param retries the number of calls to this method for this join
907 >     */
908 >    final void tryAwaitJoin(ForkJoinTask<?> joinMe, int retries) {
909 >        int pc = parallelism;
910 >        boolean running = true; // false when running count decremented
911 >        outer:while (joinMe.status >= 0) {
912 >            int wc = workerCounts;
913 >            int rc = wc & RUNNING_COUNT_MASK;
914 >            int tc = wc >>> TOTAL_COUNT_SHIFT;
915 >            if (running) { // replace with spare or decrement count
916 >                if (rc <= pc && tc > pc &&
917 >                    (retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
918 >                    ForkJoinWorkerThread[] ws = workers; // search for spare
919 >                    int nws = ws.length;
920 >                    for (int i = 0; i < nws; ++i) {
921 >                        ForkJoinWorkerThread w = ws[i];
922 >                        if (w != null && w.isSuspended()) {
923 >                            if ((workerCounts & RUNNING_COUNT_MASK) > pc)
924 >                                continue outer;
925 >                            if (joinMe.status < 0)
926 >                                break outer;
927 >                            if (w.tryResumeSpare()) {
928 >                                running = false;
929 >                                break outer;
930 >                            }
931 >                            continue outer; // rescan on failure to resume
932                          }
933                      }
934                  }
935 <            } finally {
936 <                lock.unlock();
935 >                if ((rc <= pc && (rc == 0 || --retries < 0)) || // no retry
936 >                    joinMe.status < 0)
937 >                    break;
938 >                if (workerCounts == wc &&
939 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
940 >                                             wc, wc - ONE_RUNNING))
941 >                    running = false;
942 >            }
943 >            else { // allow blocking if enough threads
944 >                int sc = tc - pc + 1;          // = spares, plus the one to add
945 >                if (sc > 0 && rc > 0 && rc >= pc - sc && rc > pc - retries)
946 >                    break;  
947 >                if (--retries > sc && tc < MAX_THREADS &&
948 >                    tc == (runState & ACTIVE_COUNT_MASK) &&
949 >                    workerCounts == wc &&
950 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
951 >                                             wc + (ONE_RUNNING|ONE_TOTAL))) {
952 >                    addWorker();
953 >                    break;
954 >                }
955 >                if (workerCounts == wc &&
956 >                    UNSAFE.compareAndSwapInt (this, workerCountsOffset,
957 >                                              wc, wc + ONE_RUNNING)) {
958 >                    running = true;            // back out; allow retry
959 >                    break;
960 >                }
961              }
962          }
963 +        if (!running) { // can block
964 +            int c;                      // to inline incrementRunningCount
965 +            joinMe.internalAwaitDone();
966 +            do {} while (!UNSAFE.compareAndSwapInt
967 +                         (this, workerCountsOffset,
968 +                          c = workerCounts, c + ONE_RUNNING));
969 +        }
970      }
971  
972      /**
973 <     * Worker creation and startup for threads added via setParallelism.
973 >     * Same idea as (and shares many code snippets with) tryAwaitJoin,
974 >     * but self-contained because there are no caller retries.
975 >     * TODO: Rework to use simpler API.
976       */
977 <    private void createAndStartAddedWorkers() {
978 <        resumeAllSpares();  // Allow spares to convert to nonspare
979 <        int ps = parallelism;
980 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
981 <        int len = ws.length;
982 <        // Sweep through slots, to keep lowest indices most populated
983 <        int k = 0;
984 <        while (k < len) {
985 <            if (ws[k] != null) {
986 <                ++k;
987 <                continue;
977 >    final void awaitBlocker(ManagedBlocker blocker)
978 >        throws InterruptedException {
979 >        int pc = parallelism;
980 >        boolean running = true;
981 >        int retries = 0;
982 >        boolean done;
983 >        outer:while (!(done = blocker.isReleasable())) {
984 >            int wc = workerCounts;
985 >            int rc = wc & RUNNING_COUNT_MASK;
986 >            int tc = wc >>> TOTAL_COUNT_SHIFT;
987 >            if (running) {
988 >                if (rc <= pc && tc > pc &&
989 >                    (retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
990 >                    ForkJoinWorkerThread[] ws = workers;
991 >                    int nws = ws.length;
992 >                    for (int i = 0; i < nws; ++i) {
993 >                        ForkJoinWorkerThread w = ws[i];
994 >                        if (w != null && w.isSuspended()) {
995 >                            if ((workerCounts & RUNNING_COUNT_MASK) > pc)
996 >                                continue outer;
997 >                            if (done = blocker.isReleasable())
998 >                                break outer;
999 >                            if (w.tryResumeSpare()) {
1000 >                                running = false;
1001 >                                break outer;
1002 >                            }
1003 >                            continue outer;
1004 >                        }
1005 >                    }
1006 >                    if (done = blocker.isReleasable())
1007 >                        break;
1008 >                }
1009 >                if (rc > 0 && workerCounts == wc &&
1010 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1011 >                                             wc, wc - ONE_RUNNING)) {
1012 >                    running = false;
1013 >                    if (rc > pc)
1014 >                        break;
1015 >                }
1016              }
1017 <            int s = workerCounts;
530 <            int tc = totalCountOf(s);
531 <            int rc = runningCountOf(s);
532 <            if (rc >= ps || tc >= ps)
1017 >            else if (rc >= pc)
1018                  break;
1019 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
1020 <                ForkJoinWorkerThread w = createWorker(k);
1021 <                if (w != null) {
1022 <                    ws[k++] = w;
1023 <                    w.start();
1024 <                }
1025 <                else {
1026 <                    updateWorkerCount(-1); // back out on failed creation
1027 <                    break;
1019 >            else if (tc < MAX_THREADS &&
1020 >                     tc == (runState & ACTIVE_COUNT_MASK) &&
1021 >                     workerCounts == wc &&
1022 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1023 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
1024 >                addWorker();
1025 >                break;
1026 >            }
1027 >            else if (workerCounts == wc &&
1028 >                     UNSAFE.compareAndSwapInt (this, workerCountsOffset,
1029 >                                              wc, wc + ONE_RUNNING)) {
1030 >                Thread.yield();
1031 >                ++retries;
1032 >                running = true;            // allow rescan
1033 >            }
1034 >        }
1035 >
1036 >        try {
1037 >            if (!done)
1038 >                do {} while (!blocker.isReleasable() && !blocker.block());
1039 >        } finally {
1040 >            if (!running) {
1041 >                int c;
1042 >                do {} while (!UNSAFE.compareAndSwapInt
1043 >                             (this, workerCountsOffset,
1044 >                              c = workerCounts, c + ONE_RUNNING));
1045 >            }
1046 >        }
1047 >    }
1048 >
1049 >    /**
1050 >     * Possibly initiates and/or completes termination.
1051 >     *
1052 >     * @param now if true, unconditionally terminate, else only
1053 >     * if shutdown and empty queue and no active workers
1054 >     * @return true if now terminating or terminated
1055 >     */
1056 >    private boolean tryTerminate(boolean now) {
1057 >        if (now)
1058 >            advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1059 >        else if (runState < SHUTDOWN ||
1060 >                 !submissionQueue.isEmpty() ||
1061 >                 (runState & ACTIVE_COUNT_MASK) != 0)
1062 >            return false;
1063 >
1064 >        if (advanceRunLevel(TERMINATING))
1065 >            startTerminating();
1066 >
1067 >        // Finish now if all threads terminated; else in some subsequent call
1068 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1069 >            advanceRunLevel(TERMINATED);
1070 >            termination.arrive();
1071 >        }
1072 >        return true;
1073 >    }
1074 >
1075 >    /**
1076 >     * Actions on transition to TERMINATING
1077 >     */
1078 >    private void startTerminating() {
1079 >        for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1080 >            cancelSubmissions();
1081 >            shutdownWorkers();
1082 >            cancelWorkerTasks();
1083 >            signalEvent();
1084 >            interruptWorkers();
1085 >        }
1086 >    }
1087 >
1088 >    /**
1089 >     * Clear out and cancel submissions, ignoring exceptions
1090 >     */
1091 >    private void cancelSubmissions() {
1092 >        ForkJoinTask<?> task;
1093 >        while ((task = submissionQueue.poll()) != null) {
1094 >            try {
1095 >                task.cancel(false);
1096 >            } catch (Throwable ignore) {
1097 >            }
1098 >        }
1099 >    }
1100 >
1101 >    /**
1102 >     * Sets all worker run states to at least shutdown,
1103 >     * also resuming suspended workers
1104 >     */
1105 >    private void shutdownWorkers() {
1106 >        ForkJoinWorkerThread[] ws = workers;
1107 >        int nws = ws.length;
1108 >        for (int i = 0; i < nws; ++i) {
1109 >            ForkJoinWorkerThread w = ws[i];
1110 >            if (w != null)
1111 >                w.shutdown();
1112 >        }
1113 >    }
1114 >
1115 >    /**
1116 >     * Clears out and cancels all locally queued tasks
1117 >     */
1118 >    private void cancelWorkerTasks() {
1119 >        ForkJoinWorkerThread[] ws = workers;
1120 >        int nws = ws.length;
1121 >        for (int i = 0; i < nws; ++i) {
1122 >            ForkJoinWorkerThread w = ws[i];
1123 >            if (w != null)
1124 >                w.cancelTasks();
1125 >        }
1126 >    }
1127 >
1128 >    /**
1129 >     * Unsticks all workers blocked on joins etc
1130 >     */
1131 >    private void interruptWorkers() {
1132 >        ForkJoinWorkerThread[] ws = workers;
1133 >        int nws = ws.length;
1134 >        for (int i = 0; i < nws; ++i) {
1135 >            ForkJoinWorkerThread w = ws[i];
1136 >            if (w != null && !w.isTerminated()) {
1137 >                try {
1138 >                    w.interrupt();
1139 >                } catch (SecurityException ignore) {
1140                  }
1141              }
1142          }
1143      }
1144  
1145 +    // misc support for ForkJoinWorkerThread
1146 +
1147 +    /**
1148 +     * Returns pool number
1149 +     */
1150 +    final int getPoolNumber() {
1151 +        return poolNumber;
1152 +    }
1153 +
1154 +    /**
1155 +     * Accumulates steal count from a worker, clearing
1156 +     * the worker's value
1157 +     */
1158 +    final void accumulateStealCount(ForkJoinWorkerThread w) {
1159 +        int sc = w.stealCount;
1160 +        if (sc != 0) {
1161 +            long c;
1162 +            w.stealCount = 0;
1163 +            do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1164 +                                                    c = stealCount, c + sc));
1165 +        }
1166 +    }
1167 +
1168 +    /**
1169 +     * Returns the approximate (non-atomic) number of idle threads per
1170 +     * active thread.
1171 +     */
1172 +    final int idlePerActive() {
1173 +        int pc = parallelism; // use parallelism, not rc
1174 +        int ac = runState;    // no mask -- artifically boosts during shutdown
1175 +        // Use exact results for small values, saturate past 4
1176 +        return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1177 +    }
1178 +
1179 +    // Public and protected methods
1180 +
1181 +    // Constructors
1182 +
1183 +    /**
1184 +     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1185 +     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1186 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1187 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1188 +     *
1189 +     * @throws SecurityException if a security manager exists and
1190 +     *         the caller is not permitted to modify threads
1191 +     *         because it does not hold {@link
1192 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1193 +     */
1194 +    public ForkJoinPool() {
1195 +        this(Runtime.getRuntime().availableProcessors(),
1196 +             defaultForkJoinWorkerThreadFactory, null, false);
1197 +    }
1198 +
1199 +    /**
1200 +     * Creates a {@code ForkJoinPool} with the indicated parallelism
1201 +     * level, the {@linkplain
1202 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1203 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1204 +     *
1205 +     * @param parallelism the parallelism level
1206 +     * @throws IllegalArgumentException if parallelism less than or
1207 +     *         equal to zero, or greater than implementation limit
1208 +     * @throws SecurityException if a security manager exists and
1209 +     *         the caller is not permitted to modify threads
1210 +     *         because it does not hold {@link
1211 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1212 +     */
1213 +    public ForkJoinPool(int parallelism) {
1214 +        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1215 +    }
1216 +
1217 +    /**
1218 +     * Creates a {@code ForkJoinPool} with the given parameters.
1219 +     *
1220 +     * @param parallelism the parallelism level. For default value,
1221 +     * use {@link java.lang.Runtime#availableProcessors}.
1222 +     * @param factory the factory for creating new threads. For default value,
1223 +     * use {@link #defaultForkJoinWorkerThreadFactory}.
1224 +     * @param handler the handler for internal worker threads that
1225 +     * terminate due to unrecoverable errors encountered while executing
1226 +     * tasks. For default value, use <code>null</code>.
1227 +     * @param asyncMode if true,
1228 +     * establishes local first-in-first-out scheduling mode for forked
1229 +     * tasks that are never joined. This mode may be more appropriate
1230 +     * than default locally stack-based mode in applications in which
1231 +     * worker threads only process event-style asynchronous tasks.
1232 +     * For default value, use <code>false</code>.
1233 +     * @throws IllegalArgumentException if parallelism less than or
1234 +     *         equal to zero, or greater than implementation limit
1235 +     * @throws NullPointerException if the factory is null
1236 +     * @throws SecurityException if a security manager exists and
1237 +     *         the caller is not permitted to modify threads
1238 +     *         because it does not hold {@link
1239 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1240 +     */
1241 +    public ForkJoinPool(int parallelism,
1242 +                        ForkJoinWorkerThreadFactory factory,
1243 +                        Thread.UncaughtExceptionHandler handler,
1244 +                        boolean asyncMode) {
1245 +        checkPermission();
1246 +        if (factory == null)
1247 +            throw new NullPointerException();
1248 +        if (parallelism <= 0 || parallelism > MAX_THREADS)
1249 +            throw new IllegalArgumentException();
1250 +        this.parallelism = parallelism;
1251 +        this.factory = factory;
1252 +        this.ueh = handler;
1253 +        this.locallyFifo = asyncMode;
1254 +        int arraySize = initialArraySizeFor(parallelism);
1255 +        this.workers = new ForkJoinWorkerThread[arraySize];
1256 +        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1257 +        this.workerLock = new ReentrantLock();
1258 +        this.termination = new Phaser(1);
1259 +        this.poolNumber = poolNumberGenerator.incrementAndGet();
1260 +    }
1261 +
1262 +    /**
1263 +     * Returns initial power of two size for workers array.
1264 +     * @param pc the initial parallelism level
1265 +     */
1266 +    private static int initialArraySizeFor(int pc) {
1267 +        // See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1268 +        int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1269 +        size |= size >>> 1;
1270 +        size |= size >>> 2;
1271 +        size |= size >>> 4;
1272 +        size |= size >>> 8;
1273 +        return size + 1;
1274 +    }
1275 +
1276      // Execution methods
1277  
1278      /**
# Line 553 | Line 1281 | public class ForkJoinPool extends Abstra
1281      private <T> void doSubmit(ForkJoinTask<T> task) {
1282          if (task == null)
1283              throw new NullPointerException();
1284 <        if (isShutdown())
1284 >        if (runState >= SHUTDOWN)
1285              throw new RejectedExecutionException();
558        if (workers == null)
559            ensureWorkerInitialization();
1286          submissionQueue.offer(task);
1287 <        signalIdleWorkers();
1287 >        signalEvent();
1288 >        ensureEnoughTotalWorkers();
1289      }
1290  
1291      /**
1292       * Performs the given task, returning its result upon completion.
1293 +     * If the caller is already engaged in a fork/join computation in
1294 +     * the current pool, this method is equivalent in effect to
1295 +     * {@link ForkJoinTask#invoke}.
1296       *
1297       * @param task the task
1298       * @return the task's result
1299 <     * @throws NullPointerException if task is null
1300 <     * @throws RejectedExecutionException if pool is shut down
1299 >     * @throws NullPointerException if the task is null
1300 >     * @throws RejectedExecutionException if the task cannot be
1301 >     *         scheduled for execution
1302       */
1303      public <T> T invoke(ForkJoinTask<T> task) {
1304          doSubmit(task);
# Line 576 | Line 1307 | public class ForkJoinPool extends Abstra
1307  
1308      /**
1309       * Arranges for (asynchronous) execution of the given task.
1310 +     * If the caller is already engaged in a fork/join computation in
1311 +     * the current pool, this method is equivalent in effect to
1312 +     * {@link ForkJoinTask#fork}.
1313       *
1314       * @param task the task
1315 <     * @throws NullPointerException if task is null
1316 <     * @throws RejectedExecutionException if pool is shut down
1315 >     * @throws NullPointerException if the task is null
1316 >     * @throws RejectedExecutionException if the task cannot be
1317 >     *         scheduled for execution
1318       */
1319      public void execute(ForkJoinTask<?> task) {
1320          doSubmit(task);
# Line 587 | Line 1322 | public class ForkJoinPool extends Abstra
1322  
1323      // AbstractExecutorService methods
1324  
1325 +    /**
1326 +     * @throws NullPointerException if the task is null
1327 +     * @throws RejectedExecutionException if the task cannot be
1328 +     *         scheduled for execution
1329 +     */
1330      public void execute(Runnable task) {
1331          ForkJoinTask<?> job;
1332          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
# Line 596 | Line 1336 | public class ForkJoinPool extends Abstra
1336          doSubmit(job);
1337      }
1338  
1339 +    /**
1340 +     * Submits a ForkJoinTask for execution.
1341 +     * If the caller is already engaged in a fork/join computation in
1342 +     * the current pool, this method is equivalent in effect to
1343 +     * {@link ForkJoinTask#fork}.
1344 +     *
1345 +     * @param task the task to submit
1346 +     * @return the task
1347 +     * @throws NullPointerException if the task is null
1348 +     * @throws RejectedExecutionException if the task cannot be
1349 +     *         scheduled for execution
1350 +     */
1351 +    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1352 +        doSubmit(task);
1353 +        return task;
1354 +    }
1355 +
1356 +    /**
1357 +     * @throws NullPointerException if the task is null
1358 +     * @throws RejectedExecutionException if the task cannot be
1359 +     *         scheduled for execution
1360 +     */
1361      public <T> ForkJoinTask<T> submit(Callable<T> task) {
1362          ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1363          doSubmit(job);
1364          return job;
1365      }
1366  
1367 +    /**
1368 +     * @throws NullPointerException if the task is null
1369 +     * @throws RejectedExecutionException if the task cannot be
1370 +     *         scheduled for execution
1371 +     */
1372      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1373          ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1374          doSubmit(job);
1375          return job;
1376      }
1377  
1378 +    /**
1379 +     * @throws NullPointerException if the task is null
1380 +     * @throws RejectedExecutionException if the task cannot be
1381 +     *         scheduled for execution
1382 +     */
1383      public ForkJoinTask<?> submit(Runnable task) {
1384          ForkJoinTask<?> job;
1385          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
# Line 619 | Line 1391 | public class ForkJoinPool extends Abstra
1391      }
1392  
1393      /**
1394 <     * Submits a ForkJoinTask for execution.
1395 <     *
624 <     * @param task the task to submit
625 <     * @return the task
626 <     * @throws RejectedExecutionException if the task cannot be
627 <     *         scheduled for execution
628 <     * @throws NullPointerException if the task is null
1394 >     * @throws NullPointerException       {@inheritDoc}
1395 >     * @throws RejectedExecutionException {@inheritDoc}
1396       */
630    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
631        doSubmit(task);
632        return task;
633    }
634
635
1397      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1398          ArrayList<ForkJoinTask<T>> forkJoinTasks =
1399              new ArrayList<ForkJoinTask<T>>(tasks.size());
# Line 641 | Line 1402 | public class ForkJoinPool extends Abstra
1402          invoke(new InvokeAll<T>(forkJoinTasks));
1403  
1404          @SuppressWarnings({"unchecked", "rawtypes"})
1405 <        List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1405 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1406          return futures;
1407      }
1408  
# Line 655 | Line 1416 | public class ForkJoinPool extends Abstra
1416          private static final long serialVersionUID = -7914297376763021607L;
1417      }
1418  
658    // Configuration and status settings and queries
659
1419      /**
1420       * Returns the factory used for constructing new workers.
1421       *
# Line 673 | Line 1432 | public class ForkJoinPool extends Abstra
1432       * @return the handler, or {@code null} if none
1433       */
1434      public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1435 <        Thread.UncaughtExceptionHandler h;
677 <        final ReentrantLock lock = this.workerLock;
678 <        lock.lock();
679 <        try {
680 <            h = ueh;
681 <        } finally {
682 <            lock.unlock();
683 <        }
684 <        return h;
1435 >        return ueh;
1436      }
1437  
1438      /**
1439 <     * Sets the handler for internal worker threads that terminate due
689 <     * to unrecoverable errors encountered while executing tasks.
690 <     * Unless set, the current default or ThreadGroup handler is used
691 <     * as handler.
1439 >     * Returns the targeted parallelism level of this pool.
1440       *
1441 <     * @param h the new handler
694 <     * @return the old handler, or {@code null} if none
695 <     * @throws SecurityException if a security manager exists and
696 <     *         the caller is not permitted to modify threads
697 <     *         because it does not hold {@link
698 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
699 <     */
700 <    public Thread.UncaughtExceptionHandler
701 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
702 <        checkPermission();
703 <        Thread.UncaughtExceptionHandler old = null;
704 <        final ReentrantLock lock = this.workerLock;
705 <        lock.lock();
706 <        try {
707 <            old = ueh;
708 <            ueh = h;
709 <            ForkJoinWorkerThread[] ws = workers;
710 <            if (ws != null) {
711 <                for (int i = 0; i < ws.length; ++i) {
712 <                    ForkJoinWorkerThread w = ws[i];
713 <                    if (w != null)
714 <                        w.setUncaughtExceptionHandler(h);
715 <                }
716 <            }
717 <        } finally {
718 <            lock.unlock();
719 <        }
720 <        return old;
721 <    }
722 <
723 <
724 <    /**
725 <     * Sets the target parallelism level of this pool.
726 <     *
727 <     * @param parallelism the target parallelism
728 <     * @throws IllegalArgumentException if parallelism less than or
729 <     * equal to zero or greater than maximum size bounds
730 <     * @throws SecurityException if a security manager exists and
731 <     *         the caller is not permitted to modify threads
732 <     *         because it does not hold {@link
733 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
734 <     */
735 <    public void setParallelism(int parallelism) {
736 <        checkPermission();
737 <        if (parallelism <= 0 || parallelism > maxPoolSize)
738 <            throw new IllegalArgumentException();
739 <        final ReentrantLock lock = this.workerLock;
740 <        lock.lock();
741 <        try {
742 <            if (!isTerminating()) {
743 <                int p = this.parallelism;
744 <                this.parallelism = parallelism;
745 <                if (parallelism > p)
746 <                    createAndStartAddedWorkers();
747 <                else
748 <                    trimSpares();
749 <            }
750 <        } finally {
751 <            lock.unlock();
752 <        }
753 <        signalIdleWorkers();
754 <    }
755 <
756 <    /**
757 <     * Returns the targeted number of worker threads in this pool.
758 <     *
759 <     * @return the targeted number of worker threads in this pool
1441 >     * @return the targeted parallelism level of this pool
1442       */
1443      public int getParallelism() {
1444          return parallelism;
# Line 771 | Line 1453 | public class ForkJoinPool extends Abstra
1453       * @return the number of worker threads
1454       */
1455      public int getPoolSize() {
1456 <        return totalCountOf(workerCounts);
775 <    }
776 <
777 <    /**
778 <     * Returns the maximum number of threads allowed to exist in the
779 <     * pool, even if there are insufficient unblocked running threads.
780 <     *
781 <     * @return the maximum
782 <     */
783 <    public int getMaximumPoolSize() {
784 <        return maxPoolSize;
785 <    }
786 <
787 <    /**
788 <     * Sets the maximum number of threads allowed to exist in the
789 <     * pool, even if there are insufficient unblocked running threads.
790 <     * Setting this value has no effect on current pool size. It
791 <     * controls construction of new threads.
792 <     *
793 <     * @throws IllegalArgumentException if negative or greater than
794 <     * internal implementation limit
795 <     */
796 <    public void setMaximumPoolSize(int newMax) {
797 <        if (newMax < 0 || newMax > MAX_THREADS)
798 <            throw new IllegalArgumentException();
799 <        maxPoolSize = newMax;
800 <    }
801 <
802 <
803 <    /**
804 <     * Returns {@code true} if this pool dynamically maintains its
805 <     * target parallelism level. If false, new threads are added only
806 <     * to avoid possible starvation.  This setting is by default true.
807 <     *
808 <     * @return {@code true} if maintains parallelism
809 <     */
810 <    public boolean getMaintainsParallelism() {
811 <        return maintainsParallelism;
812 <    }
813 <
814 <    /**
815 <     * Sets whether this pool dynamically maintains its target
816 <     * parallelism level. If false, new threads are added only to
817 <     * avoid possible starvation.
818 <     *
819 <     * @param enable {@code true} to maintain parallelism
820 <     */
821 <    public void setMaintainsParallelism(boolean enable) {
822 <        maintainsParallelism = enable;
823 <    }
824 <
825 <    /**
826 <     * Establishes local first-in-first-out scheduling mode for forked
827 <     * tasks that are never joined. This mode may be more appropriate
828 <     * than default locally stack-based mode in applications in which
829 <     * worker threads only process asynchronous tasks.  This method is
830 <     * designed to be invoked only when the pool is quiescent, and
831 <     * typically only before any tasks are submitted. The effects of
832 <     * invocations at other times may be unpredictable.
833 <     *
834 <     * @param async if {@code true}, use locally FIFO scheduling
835 <     * @return the previous mode
836 <     * @see #getAsyncMode
837 <     */
838 <    public boolean setAsyncMode(boolean async) {
839 <        boolean oldMode = locallyFifo;
840 <        locallyFifo = async;
841 <        ForkJoinWorkerThread[] ws = workers;
842 <        if (ws != null) {
843 <            for (int i = 0; i < ws.length; ++i) {
844 <                ForkJoinWorkerThread t = ws[i];
845 <                if (t != null)
846 <                    t.setAsyncMode(async);
847 <            }
848 <        }
849 <        return oldMode;
1456 >        return workerCounts >>> TOTAL_COUNT_SHIFT;
1457      }
1458  
1459      /**
# Line 854 | Line 1461 | public class ForkJoinPool extends Abstra
1461       * scheduling mode for forked tasks that are never joined.
1462       *
1463       * @return {@code true} if this pool uses async mode
857     * @see #setAsyncMode
1464       */
1465      public boolean getAsyncMode() {
1466          return locallyFifo;
# Line 863 | Line 1469 | public class ForkJoinPool extends Abstra
1469      /**
1470       * Returns an estimate of the number of worker threads that are
1471       * not blocked waiting to join tasks or for other managed
1472 <     * synchronization.
1472 >     * synchronization. This method may overestimate the
1473 >     * number of running threads.
1474       *
1475       * @return the number of worker threads
1476       */
1477      public int getRunningThreadCount() {
1478 <        return runningCountOf(workerCounts);
1478 >        return workerCounts & RUNNING_COUNT_MASK;
1479      }
1480  
1481      /**
# Line 879 | Line 1486 | public class ForkJoinPool extends Abstra
1486       * @return the number of active threads
1487       */
1488      public int getActiveThreadCount() {
1489 <        return activeCountOf(runControl);
883 <    }
884 <
885 <    /**
886 <     * Returns an estimate of the number of threads that are currently
887 <     * idle waiting for tasks. This method may underestimate the
888 <     * number of idle threads.
889 <     *
890 <     * @return the number of idle threads
891 <     */
892 <    final int getIdleThreadCount() {
893 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
894 <        return (c <= 0) ? 0 : c;
1489 >        return runState & ACTIVE_COUNT_MASK;
1490      }
1491  
1492      /**
# Line 906 | Line 1501 | public class ForkJoinPool extends Abstra
1501       * @return {@code true} if all threads are currently idle
1502       */
1503      public boolean isQuiescent() {
1504 <        return activeCountOf(runControl) == 0;
1504 >        return (runState & ACTIVE_COUNT_MASK) == 0;
1505      }
1506  
1507      /**
# Line 921 | Line 1516 | public class ForkJoinPool extends Abstra
1516       * @return the number of steals
1517       */
1518      public long getStealCount() {
1519 <        return stealCount.get();
925 <    }
926 <
927 <    /**
928 <     * Accumulates steal count from a worker.
929 <     * Call only when worker known to be idle.
930 <     */
931 <    private void updateStealCount(ForkJoinWorkerThread w) {
932 <        int sc = w.getAndClearStealCount();
933 <        if (sc != 0)
934 <            stealCount.addAndGet(sc);
1519 >        return stealCount;
1520      }
1521  
1522      /**
# Line 947 | Line 1532 | public class ForkJoinPool extends Abstra
1532      public long getQueuedTaskCount() {
1533          long count = 0;
1534          ForkJoinWorkerThread[] ws = workers;
1535 <        if (ws != null) {
1536 <            for (int i = 0; i < ws.length; ++i) {
1537 <                ForkJoinWorkerThread t = ws[i];
1538 <                if (t != null)
1539 <                    count += t.getQueueSize();
955 <            }
1535 >        int nws = ws.length;
1536 >        for (int i = 0; i < nws; ++i) {
1537 >            ForkJoinWorkerThread w = ws[i];
1538 >            if (w != null)
1539 >                count += w.getQueueSize();
1540          }
1541          return count;
1542      }
# Line 1009 | Line 1593 | public class ForkJoinPool extends Abstra
1593      protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1594          int n = submissionQueue.drainTo(c);
1595          ForkJoinWorkerThread[] ws = workers;
1596 <        if (ws != null) {
1597 <            for (int i = 0; i < ws.length; ++i) {
1598 <                ForkJoinWorkerThread w = ws[i];
1599 <                if (w != null)
1600 <                    n += w.drainTasksTo(c);
1017 <            }
1596 >        int nws = ws.length;
1597 >        for (int i = 0; i < nws; ++i) {
1598 >            ForkJoinWorkerThread w = ws[i];
1599 >            if (w != null)
1600 >                n += w.drainTasksTo(c);
1601          }
1602          return n;
1603      }
1604  
1605      /**
1606 +     * Returns count of total parks by existing workers.
1607 +     * Used during development only since not meaningful to users.
1608 +     */
1609 +    private int collectParkCount() {
1610 +        int count = 0;
1611 +        ForkJoinWorkerThread[] ws = workers;
1612 +        int nws = ws.length;
1613 +        for (int i = 0; i < nws; ++i) {
1614 +            ForkJoinWorkerThread w = ws[i];
1615 +            if (w != null)
1616 +                count += w.parkCount;
1617 +        }
1618 +        return count;
1619 +    }
1620 +
1621 +    /**
1622       * Returns a string identifying this pool, as well as its state,
1623       * including indications of run state, parallelism level, and
1624       * worker and task counts.
# Line 1027 | Line 1626 | public class ForkJoinPool extends Abstra
1626       * @return a string identifying this pool, as well as its state
1627       */
1628      public String toString() {
1030        int ps = parallelism;
1031        int wc = workerCounts;
1032        int rc = runControl;
1629          long st = getStealCount();
1630          long qt = getQueuedTaskCount();
1631          long qs = getQueuedSubmissionCount();
1632 +        int wc = workerCounts;
1633 +        int tc = wc >>> TOTAL_COUNT_SHIFT;
1634 +        int rc = wc & RUNNING_COUNT_MASK;
1635 +        int pc = parallelism;
1636 +        int rs = runState;
1637 +        int ac = rs & ACTIVE_COUNT_MASK;
1638 +        //        int pk = collectParkCount();
1639          return super.toString() +
1640 <            "[" + runStateToString(runStateOf(rc)) +
1641 <            ", parallelism = " + ps +
1642 <            ", size = " + totalCountOf(wc) +
1643 <            ", active = " + activeCountOf(rc) +
1644 <            ", running = " + runningCountOf(wc) +
1640 >            "[" + runLevelToString(rs) +
1641 >            ", parallelism = " + pc +
1642 >            ", size = " + tc +
1643 >            ", active = " + ac +
1644 >            ", running = " + rc +
1645              ", steals = " + st +
1646              ", tasks = " + qt +
1647              ", submissions = " + qs +
1648 +            //            ", parks = " + pk +
1649              "]";
1650      }
1651  
1652 <    private static String runStateToString(int rs) {
1653 <        switch(rs) {
1654 <        case RUNNING: return "Running";
1655 <        case SHUTDOWN: return "Shutting down";
1656 <        case TERMINATING: return "Terminating";
1053 <        case TERMINATED: return "Terminated";
1054 <        default: throw new Error("Unknown run state");
1055 <        }
1652 >    private static String runLevelToString(int s) {
1653 >        return ((s & TERMINATED) != 0 ? "Terminated" :
1654 >                ((s & TERMINATING) != 0 ? "Terminating" :
1655 >                 ((s & SHUTDOWN) != 0 ? "Shutting down" :
1656 >                  "Running")));
1657      }
1658  
1058    // lifecycle control
1059
1659      /**
1660       * Initiates an orderly shutdown in which previously submitted
1661       * tasks are executed, but no new tasks will be accepted.
# Line 1071 | Line 1670 | public class ForkJoinPool extends Abstra
1670       */
1671      public void shutdown() {
1672          checkPermission();
1673 <        transitionRunStateTo(SHUTDOWN);
1674 <        if (canTerminateOnShutdown(runControl)) {
1076 <            if (workers == null) { // shutting down before workers created
1077 <                final ReentrantLock lock = this.workerLock;
1078 <                lock.lock();
1079 <                try {
1080 <                    if (workers == null) {
1081 <                        terminate();
1082 <                        transitionRunStateTo(TERMINATED);
1083 <                        termination.signalAll();
1084 <                    }
1085 <                } finally {
1086 <                    lock.unlock();
1087 <                }
1088 <            }
1089 <            terminateOnShutdown();
1090 <        }
1673 >        advanceRunLevel(SHUTDOWN);
1674 >        tryTerminate(false);
1675      }
1676  
1677      /**
1678 <     * Attempts to stop all actively executing tasks, and cancels all
1679 <     * waiting tasks.  Tasks that are in the process of being
1680 <     * submitted or executed concurrently during the course of this
1681 <     * method may or may not be rejected. Unlike some other executors,
1682 <     * this method cancels rather than collects non-executed tasks
1683 <     * upon termination, so always returns an empty list. However, you
1684 <     * can use method {@link #drainTasksTo} before invoking this
1685 <     * method to transfer unexecuted tasks to another collection.
1678 >     * Attempts to cancel and/or stop all tasks, and reject all
1679 >     * subsequently submitted tasks.  Tasks that are in the process of
1680 >     * being submitted or executed concurrently during the course of
1681 >     * this method may or may not be rejected. This method cancels
1682 >     * both existing and unexecuted tasks, in order to permit
1683 >     * termination in the presence of task dependencies. So the method
1684 >     * always returns an empty list (unlike the case for some other
1685 >     * Executors).
1686       *
1687       * @return an empty list
1688       * @throws SecurityException if a security manager exists and
# Line 1108 | Line 1692 | public class ForkJoinPool extends Abstra
1692       */
1693      public List<Runnable> shutdownNow() {
1694          checkPermission();
1695 <        terminate();
1695 >        tryTerminate(true);
1696          return Collections.emptyList();
1697      }
1698  
# Line 1118 | Line 1702 | public class ForkJoinPool extends Abstra
1702       * @return {@code true} if all tasks have completed following shut down
1703       */
1704      public boolean isTerminated() {
1705 <        return runStateOf(runControl) == TERMINATED;
1705 >        return runState >= TERMINATED;
1706      }
1707  
1708      /**
1709       * Returns {@code true} if the process of termination has
1710 <     * commenced but possibly not yet completed.
1710 >     * commenced but not yet completed.  This method may be useful for
1711 >     * debugging. A return of {@code true} reported a sufficient
1712 >     * period after shutdown may indicate that submitted tasks have
1713 >     * ignored or suppressed interruption, causing this executor not
1714 >     * to properly terminate.
1715       *
1716 <     * @return {@code true} if terminating
1716 >     * @return {@code true} if terminating but not yet terminated
1717       */
1718      public boolean isTerminating() {
1719 <        return runStateOf(runControl) >= TERMINATING;
1719 >        return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1720      }
1721  
1722      /**
# Line 1137 | Line 1725 | public class ForkJoinPool extends Abstra
1725       * @return {@code true} if this pool has been shut down
1726       */
1727      public boolean isShutdown() {
1728 <        return runStateOf(runControl) >= SHUTDOWN;
1728 >        return runState >= SHUTDOWN;
1729      }
1730  
1731      /**
# Line 1153 | Line 1741 | public class ForkJoinPool extends Abstra
1741       */
1742      public boolean awaitTermination(long timeout, TimeUnit unit)
1743          throws InterruptedException {
1156        long nanos = unit.toNanos(timeout);
1157        final ReentrantLock lock = this.workerLock;
1158        lock.lock();
1744          try {
1745 <            for (;;) {
1746 <                if (isTerminated())
1162 <                    return true;
1163 <                if (nanos <= 0)
1164 <                    return false;
1165 <                nanos = termination.awaitNanos(nanos);
1166 <            }
1167 <        } finally {
1168 <            lock.unlock();
1169 <        }
1170 <    }
1171 <
1172 <    // Shutdown and termination support
1173 <
1174 <    /**
1175 <     * Callback from terminating worker. Nulls out the corresponding
1176 <     * workers slot, and if terminating, tries to terminate; else
1177 <     * tries to shrink workers array.
1178 <     *
1179 <     * @param w the worker
1180 <     */
1181 <    final void workerTerminated(ForkJoinWorkerThread w) {
1182 <        updateStealCount(w);
1183 <        updateWorkerCount(-1);
1184 <        final ReentrantLock lock = this.workerLock;
1185 <        lock.lock();
1186 <        try {
1187 <            ForkJoinWorkerThread[] ws = workers;
1188 <            if (ws != null) {
1189 <                int idx = w.poolIndex;
1190 <                if (idx >= 0 && idx < ws.length && ws[idx] == w)
1191 <                    ws[idx] = null;
1192 <                if (totalCountOf(workerCounts) == 0) {
1193 <                    terminate(); // no-op if already terminating
1194 <                    transitionRunStateTo(TERMINATED);
1195 <                    termination.signalAll();
1196 <                }
1197 <                else if (!isTerminating()) {
1198 <                    tryShrinkWorkerArray();
1199 <                    tryResumeSpare(true); // allow replacement
1200 <                }
1201 <            }
1202 <        } finally {
1203 <            lock.unlock();
1204 <        }
1205 <        signalIdleWorkers();
1206 <    }
1207 <
1208 <    /**
1209 <     * Initiates termination.
1210 <     */
1211 <    private void terminate() {
1212 <        if (transitionRunStateTo(TERMINATING)) {
1213 <            stopAllWorkers();
1214 <            resumeAllSpares();
1215 <            signalIdleWorkers();
1216 <            cancelQueuedSubmissions();
1217 <            cancelQueuedWorkerTasks();
1218 <            interruptUnterminatedWorkers();
1219 <            signalIdleWorkers(); // resignal after interrupt
1220 <        }
1221 <    }
1222 <
1223 <    /**
1224 <     * Possibly terminates when on shutdown state.
1225 <     */
1226 <    private void terminateOnShutdown() {
1227 <        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1228 <            terminate();
1229 <    }
1230 <
1231 <    /**
1232 <     * Clears out and cancels submissions.
1233 <     */
1234 <    private void cancelQueuedSubmissions() {
1235 <        ForkJoinTask<?> task;
1236 <        while ((task = pollSubmission()) != null)
1237 <            task.cancel(false);
1238 <    }
1239 <
1240 <    /**
1241 <     * Cleans out worker queues.
1242 <     */
1243 <    private void cancelQueuedWorkerTasks() {
1244 <        final ReentrantLock lock = this.workerLock;
1245 <        lock.lock();
1246 <        try {
1247 <            ForkJoinWorkerThread[] ws = workers;
1248 <            if (ws != null) {
1249 <                for (int i = 0; i < ws.length; ++i) {
1250 <                    ForkJoinWorkerThread t = ws[i];
1251 <                    if (t != null)
1252 <                        t.cancelTasks();
1253 <                }
1254 <            }
1255 <        } finally {
1256 <            lock.unlock();
1257 <        }
1258 <    }
1259 <
1260 <    /**
1261 <     * Sets each worker's status to terminating. Requires lock to avoid
1262 <     * conflicts with add/remove.
1263 <     */
1264 <    private void stopAllWorkers() {
1265 <        final ReentrantLock lock = this.workerLock;
1266 <        lock.lock();
1267 <        try {
1268 <            ForkJoinWorkerThread[] ws = workers;
1269 <            if (ws != null) {
1270 <                for (int i = 0; i < ws.length; ++i) {
1271 <                    ForkJoinWorkerThread t = ws[i];
1272 <                    if (t != null)
1273 <                        t.shutdownNow();
1274 <                }
1275 <            }
1276 <        } finally {
1277 <            lock.unlock();
1278 <        }
1279 <    }
1280 <
1281 <    /**
1282 <     * Interrupts all unterminated workers.  This is not required for
1283 <     * sake of internal control, but may help unstick user code during
1284 <     * shutdown.
1285 <     */
1286 <    private void interruptUnterminatedWorkers() {
1287 <        final ReentrantLock lock = this.workerLock;
1288 <        lock.lock();
1289 <        try {
1290 <            ForkJoinWorkerThread[] ws = workers;
1291 <            if (ws != null) {
1292 <                for (int i = 0; i < ws.length; ++i) {
1293 <                    ForkJoinWorkerThread t = ws[i];
1294 <                    if (t != null && !t.isTerminated()) {
1295 <                        try {
1296 <                            t.interrupt();
1297 <                        } catch (SecurityException ignore) {
1298 <                        }
1299 <                    }
1300 <                }
1301 <            }
1302 <        } finally {
1303 <            lock.unlock();
1304 <        }
1305 <    }
1306 <
1307 <
1308 <    /*
1309 <     * Nodes for event barrier to manage idle threads.  Queue nodes
1310 <     * are basic Treiber stack nodes, also used for spare stack.
1311 <     *
1312 <     * The event barrier has an event count and a wait queue (actually
1313 <     * a Treiber stack).  Workers are enabled to look for work when
1314 <     * the eventCount is incremented. If they fail to find work, they
1315 <     * may wait for next count. Upon release, threads help others wake
1316 <     * up.
1317 <     *
1318 <     * Synchronization events occur only in enough contexts to
1319 <     * maintain overall liveness:
1320 <     *
1321 <     *   - Submission of a new task to the pool
1322 <     *   - Resizes or other changes to the workers array
1323 <     *   - pool termination
1324 <     *   - A worker pushing a task on an empty queue
1325 <     *
1326 <     * The case of pushing a task occurs often enough, and is heavy
1327 <     * enough compared to simple stack pushes, to require special
1328 <     * handling: Method signalWork returns without advancing count if
1329 <     * the queue appears to be empty.  This would ordinarily result in
1330 <     * races causing some queued waiters not to be woken up. To avoid
1331 <     * this, the first worker enqueued in method sync (see
1332 <     * syncIsReleasable) rescans for tasks after being enqueued, and
1333 <     * helps signal if any are found. This works well because the
1334 <     * worker has nothing better to do, and so might as well help
1335 <     * alleviate the overhead and contention on the threads actually
1336 <     * doing work.  Also, since event counts increments on task
1337 <     * availability exist to maintain liveness (rather than to force
1338 <     * refreshes etc), it is OK for callers to exit early if
1339 <     * contending with another signaller.
1340 <     */
1341 <    static final class WaitQueueNode {
1342 <        WaitQueueNode next; // only written before enqueued
1343 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1344 <        final long count; // unused for spare stack
1345 <
1346 <        WaitQueueNode(long c, ForkJoinWorkerThread w) {
1347 <            count = c;
1348 <            thread = w;
1349 <        }
1350 <
1351 <        /**
1352 <         * Wakes up waiter, returning false if known to already
1353 <         */
1354 <        boolean signal() {
1355 <            ForkJoinWorkerThread t = thread;
1356 <            if (t == null)
1357 <                return false;
1358 <            thread = null;
1359 <            LockSupport.unpark(t);
1360 <            return true;
1361 <        }
1362 <
1363 <        /**
1364 <         * Awaits release on sync.
1365 <         */
1366 <        void awaitSyncRelease(ForkJoinPool p) {
1367 <            while (thread != null && !p.syncIsReleasable(this))
1368 <                LockSupport.park(this);
1369 <        }
1370 <
1371 <        /**
1372 <         * Awaits resumption as spare.
1373 <         */
1374 <        void awaitSpareRelease() {
1375 <            while (thread != null) {
1376 <                if (!Thread.interrupted())
1377 <                    LockSupport.park(this);
1378 <            }
1379 <        }
1380 <    }
1381 <
1382 <    /**
1383 <     * Ensures that no thread is waiting for count to advance from the
1384 <     * current value of eventCount read on entry to this method, by
1385 <     * releasing waiting threads if necessary.
1386 <     *
1387 <     * @return the count
1388 <     */
1389 <    final long ensureSync() {
1390 <        long c = eventCount;
1391 <        WaitQueueNode q;
1392 <        while ((q = syncStack) != null && q.count < c) {
1393 <            if (casBarrierStack(q, null)) {
1394 <                do {
1395 <                    q.signal();
1396 <                } while ((q = q.next) != null);
1397 <                break;
1398 <            }
1399 <        }
1400 <        return c;
1401 <    }
1402 <
1403 <    /**
1404 <     * Increments event count and releases waiting threads.
1405 <     */
1406 <    private void signalIdleWorkers() {
1407 <        long c;
1408 <        do {} while (!casEventCount(c = eventCount, c+1));
1409 <        ensureSync();
1410 <    }
1411 <
1412 <    /**
1413 <     * Signals threads waiting to poll a task. Because method sync
1414 <     * rechecks availability, it is OK to only proceed if queue
1415 <     * appears to be non-empty, and OK to skip under contention to
1416 <     * increment count (since some other thread succeeded).
1417 <     */
1418 <    final void signalWork() {
1419 <        long c;
1420 <        WaitQueueNode q;
1421 <        if (syncStack != null &&
1422 <            casEventCount(c = eventCount, c+1) &&
1423 <            (((q = syncStack) != null && q.count <= c) &&
1424 <             (!casBarrierStack(q, q.next) || !q.signal())))
1425 <            ensureSync();
1426 <    }
1427 <
1428 <    /**
1429 <     * Waits until event count advances from last value held by
1430 <     * caller, or if excess threads, caller is resumed as spare, or
1431 <     * caller or pool is terminating. Updates caller's event on exit.
1432 <     *
1433 <     * @param w the calling worker thread
1434 <     */
1435 <    final void sync(ForkJoinWorkerThread w) {
1436 <        updateStealCount(w); // Transfer w's count while it is idle
1437 <
1438 <        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1439 <            long prev = w.lastEventCount;
1440 <            WaitQueueNode node = null;
1441 <            WaitQueueNode h;
1442 <            while (eventCount == prev &&
1443 <                   ((h = syncStack) == null || h.count == prev)) {
1444 <                if (node == null)
1445 <                    node = new WaitQueueNode(prev, w);
1446 <                if (casBarrierStack(node.next = h, node)) {
1447 <                    node.awaitSyncRelease(this);
1448 <                    break;
1449 <                }
1450 <            }
1451 <            long ec = ensureSync();
1452 <            if (ec != prev) {
1453 <                w.lastEventCount = ec;
1454 <                break;
1455 <            }
1456 <        }
1457 <    }
1458 <
1459 <    /**
1460 <     * Returns {@code true} if worker waiting on sync can proceed:
1461 <     *  - on signal (thread == null)
1462 <     *  - on event count advance (winning race to notify vs signaller)
1463 <     *  - on interrupt
1464 <     *  - if the first queued node, we find work available
1465 <     * If node was not signalled and event count not advanced on exit,
1466 <     * then we also help advance event count.
1467 <     *
1468 <     * @return {@code true} if node can be released
1469 <     */
1470 <    final boolean syncIsReleasable(WaitQueueNode node) {
1471 <        long prev = node.count;
1472 <        if (!Thread.interrupted() && node.thread != null &&
1473 <            (node.next != null ||
1474 <             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1475 <            eventCount == prev)
1745 >            return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1746 >        } catch(TimeoutException ex) {
1747              return false;
1477        if (node.thread != null) {
1478            node.thread = null;
1479            long ec = eventCount;
1480            if (prev <= ec) // help signal
1481                casEventCount(ec, ec+1);
1482        }
1483        return true;
1484    }
1485
1486    /**
1487     * Returns {@code true} if a new sync event occurred since last
1488     * call to sync or this method, if so, updating caller's count.
1489     */
1490    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1491        long lc = w.lastEventCount;
1492        long ec = ensureSync();
1493        if (ec == lc)
1494            return false;
1495        w.lastEventCount = ec;
1496        return true;
1497    }
1498
1499    //  Parallelism maintenance
1500
1501    /**
1502     * Decrements running count; if too low, adds spare.
1503     *
1504     * Conceptually, all we need to do here is add or resume a
1505     * spare thread when one is about to block (and remove or
1506     * suspend it later when unblocked -- see suspendIfSpare).
1507     * However, implementing this idea requires coping with
1508     * several problems: we have imperfect information about the
1509     * states of threads. Some count updates can and usually do
1510     * lag run state changes, despite arrangements to keep them
1511     * accurate (for example, when possible, updating counts
1512     * before signalling or resuming), especially when running on
1513     * dynamic JVMs that don't optimize the infrequent paths that
1514     * update counts. Generating too many threads can make these
1515     * problems become worse, because excess threads are more
1516     * likely to be context-switched with others, slowing them all
1517     * down, especially if there is no work available, so all are
1518     * busy scanning or idling.  Also, excess spare threads can
1519     * only be suspended or removed when they are idle, not
1520     * immediately when they aren't needed. So adding threads will
1521     * raise parallelism level for longer than necessary.  Also,
1522     * FJ applications often encounter highly transient peaks when
1523     * many threads are blocked joining, but for less time than it
1524     * takes to create or resume spares.
1525     *
1526     * @param joinMe if non-null, return early if done
1527     * @param maintainParallelism if true, try to stay within
1528     * target counts, else create only to avoid starvation
1529     * @return true if joinMe known to be done
1530     */
1531    final boolean preJoin(ForkJoinTask<?> joinMe,
1532                          boolean maintainParallelism) {
1533        maintainParallelism &= maintainsParallelism; // overrride
1534        boolean dec = false;  // true when running count decremented
1535        while (spareStack == null || !tryResumeSpare(dec)) {
1536            int counts = workerCounts;
1537            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1538                // CAS cheat
1539                if (!needSpare(counts, maintainParallelism))
1540                    break;
1541                if (joinMe.status < 0)
1542                    return true;
1543                if (tryAddSpare(counts))
1544                    break;
1545            }
1546        }
1547        return false;
1548    }
1549
1550    /**
1551     * Same idea as preJoin
1552     */
1553    final boolean preBlock(ManagedBlocker blocker,
1554                           boolean maintainParallelism) {
1555        maintainParallelism &= maintainsParallelism;
1556        boolean dec = false;
1557        while (spareStack == null || !tryResumeSpare(dec)) {
1558            int counts = workerCounts;
1559            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1560                if (!needSpare(counts, maintainParallelism))
1561                    break;
1562                if (blocker.isReleasable())
1563                    return true;
1564                if (tryAddSpare(counts))
1565                    break;
1566            }
1567        }
1568        return false;
1569    }
1570
1571    /**
1572     * Returns {@code true} if a spare thread appears to be needed.
1573     * If maintaining parallelism, returns true when the deficit in
1574     * running threads is more than the surplus of total threads, and
1575     * there is apparently some work to do.  This self-limiting rule
1576     * means that the more threads that have already been added, the
1577     * less parallelism we will tolerate before adding another.
1578     *
1579     * @param counts current worker counts
1580     * @param maintainParallelism try to maintain parallelism
1581     */
1582    private boolean needSpare(int counts, boolean maintainParallelism) {
1583        int ps = parallelism;
1584        int rc = runningCountOf(counts);
1585        int tc = totalCountOf(counts);
1586        int runningDeficit = ps - rc;
1587        int totalSurplus = tc - ps;
1588        return (tc < maxPoolSize &&
1589                (rc == 0 || totalSurplus < 0 ||
1590                 (maintainParallelism &&
1591                  runningDeficit > totalSurplus &&
1592                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
1593    }
1594
1595    /**
1596     * Adds a spare worker if lock available and no more than the
1597     * expected numbers of threads exist.
1598     *
1599     * @return true if successful
1600     */
1601    private boolean tryAddSpare(int expectedCounts) {
1602        final ReentrantLock lock = this.workerLock;
1603        int expectedRunning = runningCountOf(expectedCounts);
1604        int expectedTotal = totalCountOf(expectedCounts);
1605        boolean success = false;
1606        boolean locked = false;
1607        // confirm counts while locking; CAS after obtaining lock
1608        try {
1609            for (;;) {
1610                int s = workerCounts;
1611                int tc = totalCountOf(s);
1612                int rc = runningCountOf(s);
1613                if (rc > expectedRunning || tc > expectedTotal)
1614                    break;
1615                if (!locked && !(locked = lock.tryLock()))
1616                    break;
1617                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1618                    createAndStartSpare(tc);
1619                    success = true;
1620                    break;
1621                }
1622            }
1623        } finally {
1624            if (locked)
1625                lock.unlock();
1626        }
1627        return success;
1628    }
1629
1630    /**
1631     * Adds the kth spare worker. On entry, pool counts are already
1632     * adjusted to reflect addition.
1633     */
1634    private void createAndStartSpare(int k) {
1635        ForkJoinWorkerThread w = null;
1636        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1637        int len = ws.length;
1638        // Probably, we can place at slot k. If not, find empty slot
1639        if (k < len && ws[k] != null) {
1640            for (k = 0; k < len && ws[k] != null; ++k)
1641                ;
1642        }
1643        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1644            ws[k] = w;
1645            w.start();
1646        }
1647        else
1648            updateWorkerCount(-1); // adjust on failure
1649        signalIdleWorkers();
1650    }
1651
1652    /**
1653     * Suspends calling thread w if there are excess threads.  Called
1654     * only from sync.  Spares are enqueued in a Treiber stack using
1655     * the same WaitQueueNodes as barriers.  They are resumed mainly
1656     * in preJoin, but are also woken on pool events that require all
1657     * threads to check run state.
1658     *
1659     * @param w the caller
1660     */
1661    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1662        WaitQueueNode node = null;
1663        int s;
1664        while (parallelism < runningCountOf(s = workerCounts)) {
1665            if (node == null)
1666                node = new WaitQueueNode(0, w);
1667            if (casWorkerCounts(s, s-1)) { // representation-dependent
1668                // push onto stack
1669                do {} while (!casSpareStack(node.next = spareStack, node));
1670                // block until released by resumeSpare
1671                node.awaitSpareRelease();
1672                return true;
1673            }
1674        }
1675        return false;
1676    }
1677
1678    /**
1679     * Tries to pop and resume a spare thread.
1680     *
1681     * @param updateCount if true, increment running count on success
1682     * @return true if successful
1683     */
1684    private boolean tryResumeSpare(boolean updateCount) {
1685        WaitQueueNode q;
1686        while ((q = spareStack) != null) {
1687            if (casSpareStack(q, q.next)) {
1688                if (updateCount)
1689                    updateRunningCount(1);
1690                q.signal();
1691                return true;
1692            }
1693        }
1694        return false;
1695    }
1696
1697    /**
1698     * Pops and resumes all spare threads. Same idea as ensureSync.
1699     *
1700     * @return true if any spares released
1701     */
1702    private boolean resumeAllSpares() {
1703        WaitQueueNode q;
1704        while ( (q = spareStack) != null) {
1705            if (casSpareStack(q, null)) {
1706                do {
1707                    updateRunningCount(1);
1708                    q.signal();
1709                } while ((q = q.next) != null);
1710                return true;
1711            }
1712        }
1713        return false;
1714    }
1715
1716    /**
1717     * Pops and shuts down excessive spare threads. Call only while
1718     * holding lock. This is not guaranteed to eliminate all excess
1719     * threads, only those suspended as spares, which are the ones
1720     * unlikely to be needed in the future.
1721     */
1722    private void trimSpares() {
1723        int surplus = totalCountOf(workerCounts) - parallelism;
1724        WaitQueueNode q;
1725        while (surplus > 0 && (q = spareStack) != null) {
1726            if (casSpareStack(q, null)) {
1727                do {
1728                    updateRunningCount(1);
1729                    ForkJoinWorkerThread w = q.thread;
1730                    if (w != null && surplus > 0 &&
1731                        runningCountOf(workerCounts) > 0 && w.shutdown())
1732                        --surplus;
1733                    q.signal();
1734                } while ((q = q.next) != null);
1735            }
1748          }
1749      }
1750  
# Line 1785 | Line 1797 | public class ForkJoinPool extends Abstra
1797       * Blocks in accord with the given blocker.  If the current thread
1798       * is a {@link ForkJoinWorkerThread}, this method possibly
1799       * arranges for a spare thread to be activated if necessary to
1800 <     * ensure parallelism while the current thread is blocked.
1789 <     *
1790 <     * <p>If {@code maintainParallelism} is {@code true} and the pool
1791 <     * supports it ({@link #getMaintainsParallelism}), this method
1792 <     * attempts to maintain the pool's nominal parallelism. Otherwise
1793 <     * it activates a thread only if necessary to avoid complete
1794 <     * starvation. This option may be preferable when blockages use
1795 <     * timeouts, or are almost always brief.
1800 >     * ensure sufficient parallelism while the current thread is blocked.
1801       *
1802       * <p>If the caller is not a {@link ForkJoinTask}, this method is
1803       * behaviorally equivalent to
# Line 1806 | Line 1811 | public class ForkJoinPool extends Abstra
1811       * first be expanded to ensure parallelism, and later adjusted.
1812       *
1813       * @param blocker the blocker
1809     * @param maintainParallelism if {@code true} and supported by
1810     * this pool, attempt to maintain the pool's nominal parallelism;
1811     * otherwise activate a thread only if necessary to avoid
1812     * complete starvation.
1814       * @throws InterruptedException if blocker.block did so
1815       */
1816 <    public static void managedBlock(ManagedBlocker blocker,
1816 <                                    boolean maintainParallelism)
1816 >    public static void managedBlock(ManagedBlocker blocker)
1817          throws InterruptedException {
1818          Thread t = Thread.currentThread();
1819 <        ForkJoinPool pool = ((t instanceof ForkJoinWorkerThread) ?
1820 <                             ((ForkJoinWorkerThread) t).pool : null);
1821 <        if (!blocker.isReleasable()) {
1822 <            try {
1823 <                if (pool == null ||
1824 <                    !pool.preBlock(blocker, maintainParallelism))
1825 <                    awaitBlocker(blocker);
1826 <            } finally {
1827 <                if (pool != null)
1828 <                    pool.updateRunningCount(1);
1829 <            }
1819 >        if (t instanceof ForkJoinWorkerThread)
1820 >            ((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
1821 >        else {
1822 >            do {} while (!blocker.isReleasable() && !blocker.block());
1823          }
1824      }
1825  
1833    private static void awaitBlocker(ManagedBlocker blocker)
1834        throws InterruptedException {
1835        do {} while (!blocker.isReleasable() && !blocker.block());
1836    }
1837
1826      // AbstractExecutorService overrides.  These rely on undocumented
1827      // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
1828      // implement RunnableFuture.
# Line 1850 | Line 1838 | public class ForkJoinPool extends Abstra
1838      // Unsafe mechanics
1839  
1840      private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1853    private static final long eventCountOffset =
1854        objectFieldOffset("eventCount", ForkJoinPool.class);
1841      private static final long workerCountsOffset =
1842          objectFieldOffset("workerCounts", ForkJoinPool.class);
1843 <    private static final long runControlOffset =
1844 <        objectFieldOffset("runControl", ForkJoinPool.class);
1845 <    private static final long syncStackOffset =
1846 <        objectFieldOffset("syncStack",ForkJoinPool.class);
1847 <    private static final long spareStackOffset =
1848 <        objectFieldOffset("spareStack", ForkJoinPool.class);
1849 <
1850 <    private boolean casEventCount(long cmp, long val) {
1865 <        return UNSAFE.compareAndSwapLong(this, eventCountOffset, cmp, val);
1866 <    }
1867 <    private boolean casWorkerCounts(int cmp, int val) {
1868 <        return UNSAFE.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1869 <    }
1870 <    private boolean casRunControl(int cmp, int val) {
1871 <        return UNSAFE.compareAndSwapInt(this, runControlOffset, cmp, val);
1872 <    }
1873 <    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1874 <        return UNSAFE.compareAndSwapObject(this, spareStackOffset, cmp, val);
1875 <    }
1876 <    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1877 <        return UNSAFE.compareAndSwapObject(this, syncStackOffset, cmp, val);
1878 <    }
1843 >    private static final long runStateOffset =
1844 >        objectFieldOffset("runState", ForkJoinPool.class);
1845 >    private static final long eventCountOffset =
1846 >        objectFieldOffset("eventCount", ForkJoinPool.class);
1847 >    private static final long eventWaitersOffset =
1848 >        objectFieldOffset("eventWaiters",ForkJoinPool.class);
1849 >    private static final long stealCountOffset =
1850 >        objectFieldOffset("stealCount",ForkJoinPool.class);
1851  
1852      private static long objectFieldOffset(String field, Class<?> klazz) {
1853          try {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines