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 1.7 by jsr166, Mon Jul 20 21:45:06 2009 UTC vs.
Revision 1.88 by dl, Tue Nov 23 01:06:00 2010 UTC

# Line 5 | Line 5
5   */
6  
7   package jsr166y;
8 < import java.util.*;
9 < import java.util.concurrent.*;
10 < import java.util.concurrent.locks.*;
11 < import java.util.concurrent.atomic.*;
12 < import sun.misc.Unsafe;
13 < import java.lang.reflect.*;
8 >
9 > import java.util.ArrayList;
10 > import java.util.Arrays;
11 > import java.util.Collection;
12 > import java.util.Collections;
13 > import java.util.List;
14 > import java.util.concurrent.AbstractExecutorService;
15 > import java.util.concurrent.Callable;
16 > import java.util.concurrent.ExecutorService;
17 > import java.util.concurrent.Future;
18 > import java.util.concurrent.RejectedExecutionException;
19 > import java.util.concurrent.RunnableFuture;
20 > import java.util.concurrent.TimeUnit;
21 > import java.util.concurrent.TimeoutException;
22 > import java.util.concurrent.atomic.AtomicInteger;
23 > import java.util.concurrent.locks.LockSupport;
24 > import java.util.concurrent.locks.ReentrantLock;
25  
26   /**
27 < * An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
28 < * ForkJoinPool provides the entry point for submissions from
29 < * non-ForkJoinTasks, as well as management and monitoring operations.
30 < * Normally a single ForkJoinPool is used for a large number of
20 < * submitted tasks. Otherwise, use would not usually outweigh the
21 < * construction and bookkeeping overhead of creating a large set of
22 < * threads.
27 > * An {@link ExecutorService} for running {@link ForkJoinTask}s.
28 > * A {@code ForkJoinPool} provides the entry point for submissions
29 > * from non-{@code ForkJoinTask} clients, as well as management and
30 > * monitoring operations.
31   *
32 < * <p>ForkJoinPools differ from other kinds of Executors mainly in
33 < * that they provide <em>work-stealing</em>: all threads in the pool
34 < * attempt to find and execute subtasks created by other active tasks
35 < * (eventually blocking if none exist). This makes them efficient when
36 < * most tasks spawn other subtasks (as do most ForkJoinTasks), as well
37 < * as the mixed execution of some plain Runnable- or Callable- based
38 < * activities along with ForkJoinTasks. When setting
39 < * <tt>setAsyncMode</tt>, a ForkJoinPools may also be appropriate for
40 < * use with fine-grained tasks that are never joined. Otherwise, other
33 < * ExecutorService implementations are typically more appropriate
34 < * choices.
32 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
33 > * ExecutorService} mainly by virtue of employing
34 > * <em>work-stealing</em>: all threads in the pool attempt to find and
35 > * execute subtasks created by other active tasks (eventually blocking
36 > * waiting for work if none exist). This enables efficient processing
37 > * when most tasks spawn other subtasks (as do most {@code
38 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
39 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
40 > * with event-style tasks that are never joined.
41   *
42 < * <p>A ForkJoinPool may be constructed with a given parallelism level
43 < * (target pool size), which it attempts to maintain by dynamically
44 < * adding, suspending, or resuming threads, even if some tasks are
45 < * waiting to join others. However, no such adjustments are performed
46 < * in the face of blocked IO or other unmanaged synchronization. The
47 < * nested <code>ManagedBlocker</code> interface enables extension of
48 < * the kinds of synchronization accommodated.  The target parallelism
49 < * level may also be changed dynamically (<code>setParallelism</code>)
50 < * and thread construction can be limited using methods
45 < * <code>setMaximumPoolSize</code> and/or
46 < * <code>setMaintainsParallelism</code>.
42 > * <p>A {@code ForkJoinPool} is constructed with a given target
43 > * parallelism level; by default, equal to the number of available
44 > * processors. The pool attempts to maintain enough active (or
45 > * available) threads by dynamically adding, suspending, or resuming
46 > * internal worker threads, even if some tasks are stalled waiting to
47 > * join others. However, no such adjustments are guaranteed in the
48 > * face of blocked IO or other unmanaged synchronization. The nested
49 > * {@link ManagedBlocker} interface enables extension of the kinds of
50 > * synchronization accommodated.
51   *
52   * <p>In addition to execution and lifecycle control methods, this
53   * class provides status check methods (for example
54 < * <code>getStealCount</code>) that are intended to aid in developing,
54 > * {@link #getStealCount}) that are intended to aid in developing,
55   * tuning, and monitoring fork/join applications. Also, method
56 < * <code>toString</code> returns indications of pool state in a
56 > * {@link #toString} returns indications of pool state in a
57   * convenient form for informal monitoring.
58   *
59 + * <p> As is the case with other ExecutorServices, there are three
60 + * main task execution methods summarized in the following
61 + * table. These are designed to be used by clients not already engaged
62 + * in fork/join computations in the current pool.  The main forms of
63 + * these methods accept instances of {@code ForkJoinTask}, but
64 + * overloaded forms also allow mixed execution of plain {@code
65 + * Runnable}- or {@code Callable}- based activities as well.  However,
66 + * tasks that are already executing in a pool should normally
67 + * <em>NOT</em> use these pool execution methods, but instead use the
68 + * within-computation forms listed in the table.
69 + *
70 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
71 + *  <tr>
72 + *    <td></td>
73 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
74 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
75 + *  </tr>
76 + *  <tr>
77 + *    <td> <b>Arrange async execution</td>
78 + *    <td> {@link #execute(ForkJoinTask)}</td>
79 + *    <td> {@link ForkJoinTask#fork}</td>
80 + *  </tr>
81 + *  <tr>
82 + *    <td> <b>Await and obtain result</td>
83 + *    <td> {@link #invoke(ForkJoinTask)}</td>
84 + *    <td> {@link ForkJoinTask#invoke}</td>
85 + *  </tr>
86 + *  <tr>
87 + *    <td> <b>Arrange exec and obtain Future</td>
88 + *    <td> {@link #submit(ForkJoinTask)}</td>
89 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
90 + *  </tr>
91 + * </table>
92 + *
93 + * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
94 + * used for all parallel task execution in a program or subsystem.
95 + * Otherwise, use would not usually outweigh the construction and
96 + * bookkeeping overhead of creating a large set of threads. For
97 + * example, a common pool could be used for the {@code SortTasks}
98 + * illustrated in {@link RecursiveAction}. Because {@code
99 + * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
100 + * daemon} mode, there is typically no need to explicitly {@link
101 + * #shutdown} such a pool upon program exit.
102 + *
103 + * <pre>
104 + * static final ForkJoinPool mainPool = new ForkJoinPool();
105 + * ...
106 + * public void sort(long[] array) {
107 + *   mainPool.invoke(new SortTask(array, 0, array.length));
108 + * }
109 + * </pre>
110 + *
111   * <p><b>Implementation notes</b>: This implementation restricts the
112   * maximum number of running threads to 32767. Attempts to create
113 < * pools with greater than the maximum result in
114 < * IllegalArgumentExceptions.
113 > * pools with greater than the maximum number result in
114 > * {@code IllegalArgumentException}.
115 > *
116 > * <p>This implementation rejects submitted tasks (that is, by throwing
117 > * {@link RejectedExecutionException}) only when the pool is shut down
118 > * or internal resources have been exhausted.
119 > *
120 > * @since 1.7
121 > * @author Doug Lea
122   */
123   public class ForkJoinPool extends AbstractExecutorService {
124  
125      /*
126 <     * See the extended comments interspersed below for design,
127 <     * rationale, and walkthroughs.
128 <     */
129 <
130 <    /** Mask for packing and unpacking shorts */
131 <    private static final int  shortMask = 0xffff;
132 <
133 <    /** Max pool size -- must be a power of two minus 1 */
134 <    private static final int MAX_THREADS =  0x7FFF;
135 <
136 <    /**
137 <     * Factory for creating new ForkJoinWorkerThreads.  A
138 <     * ForkJoinWorkerThreadFactory must be defined and used for
139 <     * ForkJoinWorkerThread subclasses that extend base functionality
140 <     * or initialize threads with different contexts.
126 >     * Implementation Overview
127 >     *
128 >     * This class provides the central bookkeeping and control for a
129 >     * set of worker threads: Submissions from non-FJ threads enter
130 >     * into a submission queue. Workers take these tasks and typically
131 >     * split them into subtasks that may be stolen by other workers.
132 >     * The main work-stealing mechanics implemented in class
133 >     * ForkJoinWorkerThread give first priority to processing tasks
134 >     * from their own queues (LIFO or FIFO, depending on mode), then
135 >     * to randomized FIFO steals of tasks in other worker queues, and
136 >     * lastly to new submissions. These mechanics do not consider
137 >     * affinities, loads, cache localities, etc, so rarely provide the
138 >     * best possible performance on a given machine, but portably
139 >     * provide good throughput by averaging over these factors.
140 >     * (Further, even if we did try to use such information, we do not
141 >     * usually have a basis for exploiting it. For example, some sets
142 >     * of tasks profit from cache affinities, but others are harmed by
143 >     * cache pollution effects.)
144 >     *
145 >     * Beyond work-stealing support and essential bookkeeping, the
146 >     * main responsibility of this framework is to take actions when
147 >     * one worker is waiting to join a task stolen (or always held by)
148 >     * another.  Because we are multiplexing many tasks on to a pool
149 >     * of workers, we can't just let them block (as in Thread.join).
150 >     * We also cannot just reassign the joiner's run-time stack with
151 >     * another and replace it later, which would be a form of
152 >     * "continuation", that even if possible is not necessarily a good
153 >     * idea. Given that the creation costs of most threads on most
154 >     * systems mainly surrounds setting up runtime stacks, thread
155 >     * creation and switching is usually not much more expensive than
156 >     * stack creation and switching, and is more flexible). Instead we
157 >     * combine two tactics:
158 >     *
159 >     *   Helping: Arranging for the joiner to execute some task that it
160 >     *      would be running if the steal had not occurred.  Method
161 >     *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
162 >     *      links to try to find such a task.
163 >     *
164 >     *   Compensating: Unless there are already enough live threads,
165 >     *      method helpMaintainParallelism() may create or
166 >     *      re-activate a spare thread to compensate for blocked
167 >     *      joiners until they unblock.
168 >     *
169 >     * It is impossible to keep exactly the target (parallelism)
170 >     * number of threads running at any given time.  Determining
171 >     * existence of conservatively safe helping targets, the
172 >     * availability of already-created spares, and the apparent need
173 >     * to create new spares are all racy and require heuristic
174 >     * guidance, so we rely on multiple retries of each.  Compensation
175 >     * occurs in slow-motion. It is triggered only upon timeouts of
176 >     * Object.wait used for joins. This reduces poor decisions that
177 >     * would otherwise be made when threads are waiting for others
178 >     * that are stalled because of unrelated activities such as
179 >     * garbage collection.
180 >     *
181 >     * The ManagedBlocker extension API can't use helping so relies
182 >     * only on compensation in method awaitBlocker.
183 >     *
184 >     * The main throughput advantages of work-stealing stem from
185 >     * decentralized control -- workers mostly steal tasks from each
186 >     * other. We do not want to negate this by creating bottlenecks
187 >     * implementing other management responsibilities. So we use a
188 >     * collection of techniques that avoid, reduce, or cope well with
189 >     * contention. These entail several instances of bit-packing into
190 >     * CASable fields to maintain only the minimally required
191 >     * atomicity. To enable such packing, we restrict maximum
192 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
193 >     * unbalanced increments and decrements) to fit into a 16 bit
194 >     * field, which is far in excess of normal operating range.  Even
195 >     * though updates to some of these bookkeeping fields do sometimes
196 >     * contend with each other, they don't normally cache-contend with
197 >     * updates to others enough to warrant memory padding or
198 >     * isolation. So they are all held as fields of ForkJoinPool
199 >     * objects.  The main capabilities are as follows:
200 >     *
201 >     * 1. Creating and removing workers. Workers are recorded in the
202 >     * "workers" array. This is an array as opposed to some other data
203 >     * structure to support index-based random steals by workers.
204 >     * Updates to the array recording new workers and unrecording
205 >     * terminated ones are protected from each other by a lock
206 >     * (workerLock) but the array is otherwise concurrently readable,
207 >     * and accessed directly by workers. To simplify index-based
208 >     * operations, the array size is always a power of two, and all
209 >     * readers must tolerate null slots. Currently, all worker thread
210 >     * creation is on-demand, triggered by task submissions,
211 >     * replacement of terminated workers, and/or compensation for
212 >     * blocked workers. However, all other support code is set up to
213 >     * work with other policies.
214 >     *
215 >     * To ensure that we do not hold on to worker references that
216 >     * would prevent GC, ALL accesses to workers are via indices into
217 >     * the workers array (which is one source of some of the unusual
218 >     * code constructions here). In essence, the workers array serves
219 >     * as a WeakReference mechanism. Thus for example the event queue
220 >     * stores worker indices, not worker references. Access to the
221 >     * workers in associated methods (for example releaseEventWaiters)
222 >     * must both index-check and null-check the IDs. All such accesses
223 >     * ignore bad IDs by returning out early from what they are doing,
224 >     * since this can only be associated with shutdown, in which case
225 >     * it is OK to give up. On termination, we just clobber these
226 >     * data structures without trying to use them.
227 >     *
228 >     * 2. Bookkeeping for dynamically adding and removing workers. We
229 >     * aim to approximately maintain the given level of parallelism.
230 >     * When some workers are known to be blocked (on joins or via
231 >     * ManagedBlocker), we may create or resume others to take their
232 >     * place until they unblock (see below). Implementing this
233 >     * requires counts of the number of "running" threads (i.e., those
234 >     * that are neither blocked nor artificially suspended) as well as
235 >     * the total number.  These two values are packed into one field,
236 >     * "workerCounts" because we need accurate snapshots when deciding
237 >     * to create, resume or suspend.  Note however that the
238 >     * correspondence of these counts to reality is not guaranteed. In
239 >     * particular updates for unblocked threads may lag until they
240 >     * actually wake up.
241 >     *
242 >     * 3. Maintaining global run state. The run state of the pool
243 >     * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
244 >     * those in other Executor implementations, as well as a count of
245 >     * "active" workers -- those that are, or soon will be, or
246 >     * recently were executing tasks. The runLevel and active count
247 >     * are packed together in order to correctly trigger shutdown and
248 >     * termination. Without care, active counts can be subject to very
249 >     * high contention.  We substantially reduce this contention by
250 >     * relaxing update rules.  A worker must claim active status
251 >     * prospectively, by activating if it sees that a submitted or
252 >     * stealable task exists (it may find after activating that the
253 >     * task no longer exists). It stays active while processing this
254 >     * task (if it exists) and any other local subtasks it produces,
255 >     * until it cannot find any other tasks. It then tries
256 >     * inactivating (see method preStep), but upon update contention
257 >     * instead scans for more tasks, later retrying inactivation if it
258 >     * doesn't find any.
259 >     *
260 >     * 4. Managing idle workers waiting for tasks. We cannot let
261 >     * workers spin indefinitely scanning for tasks when none are
262 >     * available. On the other hand, we must quickly prod them into
263 >     * action when new tasks are submitted or generated.  We
264 >     * park/unpark these idle workers using an event-count scheme.
265 >     * Field eventCount is incremented upon events that may enable
266 >     * workers that previously could not find a task to now find one:
267 >     * Submission of a new task to the pool, or another worker pushing
268 >     * a task onto a previously empty queue.  (We also use this
269 >     * mechanism for configuration and termination actions that
270 >     * require wakeups of idle workers).  Each worker maintains its
271 >     * last known event count, and blocks when a scan for work did not
272 >     * find a task AND its lastEventCount matches the current
273 >     * eventCount. Waiting idle workers are recorded in a variant of
274 >     * Treiber stack headed by field eventWaiters which, when nonzero,
275 >     * encodes the thread index and count awaited for by the worker
276 >     * thread most recently calling eventSync. This thread in turn has
277 >     * a record (field nextEventWaiter) for the next waiting worker.
278 >     * In addition to allowing simpler decisions about need for
279 >     * wakeup, the event count bits in eventWaiters serve the role of
280 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
281 >     * released threads also try to release at most two others.  The
282 >     * net effect is a tree-like diffusion of signals, where released
283 >     * threads (and possibly others) help with unparks.  To further
284 >     * reduce contention effects a bit, failed CASes to increment
285 >     * field eventCount are tolerated without retries in signalWork.
286 >     * Conceptually they are merged into the same event, which is OK
287 >     * when their only purpose is to enable workers to scan for work.
288 >     *
289 >     * 5. Managing suspension of extra workers. When a worker notices
290 >     * (usually upon timeout of a wait()) that there are too few
291 >     * running threads, we may create a new thread to maintain
292 >     * parallelism level, or at least avoid starvation. Usually, extra
293 >     * threads are needed for only very short periods, yet join
294 >     * dependencies are such that we sometimes need them in
295 >     * bursts. Rather than create new threads each time this happens,
296 >     * we suspend no-longer-needed extra ones as "spares". For most
297 >     * purposes, we don't distinguish "extra" spare threads from
298 >     * normal "core" threads: On each call to preStep (the only point
299 >     * at which we can do this) a worker checks to see if there are
300 >     * now too many running workers, and if so, suspends itself.
301 >     * Method helpMaintainParallelism looks for suspended threads to
302 >     * resume before considering creating a new replacement. The
303 >     * spares themselves are encoded on another variant of a Treiber
304 >     * Stack, headed at field "spareWaiters".  Note that the use of
305 >     * spares is intrinsically racy.  One thread may become a spare at
306 >     * about the same time as another is needlessly being created. We
307 >     * counteract this and related slop in part by requiring resumed
308 >     * spares to immediately recheck (in preStep) to see whether they
309 >     * should re-suspend.
310 >     *
311 >     * 6. Killing off unneeded workers. A timeout mechanism is used to
312 >     * shed unused workers: The oldest (first) event queue waiter uses
313 >     * a timed rather than hard wait. When this wait times out without
314 >     * a normal wakeup, it tries to shutdown any one (for convenience
315 >     * the newest) other spare or event waiter via
316 >     * tryShutdownUnusedWorker. This eventually reduces the number of
317 >     * worker threads to a minimum of one after a long enough period
318 >     * without use.
319 >     *
320 >     * 7. Deciding when to create new workers. The main dynamic
321 >     * control in this class is deciding when to create extra threads
322 >     * in method helpMaintainParallelism. We would like to keep
323 >     * exactly #parallelism threads running, which is an impossible
324 >     * task. We always need to create one when the number of running
325 >     * threads would become zero and all workers are busy. Beyond
326 >     * this, we must rely on heuristics that work well in the
327 >     * presence of transient phenomena such as GC stalls, dynamic
328 >     * compilation, and wake-up lags. These transients are extremely
329 >     * common -- we are normally trying to fully saturate the CPUs on
330 >     * a machine, so almost any activity other than running tasks
331 >     * impedes accuracy. Our main defense is to allow parallelism to
332 >     * lapse for a while during joins, and use a timeout to see if,
333 >     * after the resulting settling, there is still a need for
334 >     * additional workers.  This also better copes with the fact that
335 >     * some of the methods in this class tend to never become compiled
336 >     * (but are interpreted), so some components of the entire set of
337 >     * controls might execute 100 times faster than others. And
338 >     * similarly for cases where the apparent lack of work is just due
339 >     * to GC stalls and other transient system activity.
340 >     *
341 >     * Beware that there is a lot of representation-level coupling
342 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
343 >     * ForkJoinTask.  For example, direct access to "workers" array by
344 >     * workers, and direct access to ForkJoinTask.status by both
345 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
346 >     * trying to reduce this, since any associated future changes in
347 >     * representations will need to be accompanied by algorithmic
348 >     * changes anyway.
349 >     *
350 >     * Style notes: There are lots of inline assignments (of form
351 >     * "while ((local = field) != 0)") which are usually the simplest
352 >     * way to ensure the required read orderings (which are sometimes
353 >     * critical). Also several occurrences of the unusual "do {}
354 >     * while (!cas...)" which is the simplest way to force an update of
355 >     * a CAS'ed variable. There are also other coding oddities that
356 >     * help some methods perform reasonably even when interpreted (not
357 >     * compiled), at the expense of some messy constructions that
358 >     * reduce byte code counts.
359 >     *
360 >     * The order of declarations in this file is: (1) statics (2)
361 >     * fields (along with constants used when unpacking some of them)
362 >     * (3) internal control methods (4) callbacks and other support
363 >     * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
364 >     * methods (plus a few little helpers).
365 >     */
366 >
367 >    /**
368 >     * Factory for creating new {@link ForkJoinWorkerThread}s.
369 >     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
370 >     * for {@code ForkJoinWorkerThread} subclasses that extend base
371 >     * functionality or initialize threads with different contexts.
372       */
373      public static interface ForkJoinWorkerThreadFactory {
374          /**
375           * Returns a new worker thread operating in the given pool.
376           *
377           * @param pool the pool this thread works in
378 <         * @throws NullPointerException if pool is null;
378 >         * @throws NullPointerException if the pool is null
379           */
380          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
381      }
382  
383      /**
384 <     * Default ForkJoinWorkerThreadFactory implementation, creates a
384 >     * Default ForkJoinWorkerThreadFactory implementation; creates a
385       * new ForkJoinWorkerThread.
386       */
387 <    static class  DefaultForkJoinWorkerThreadFactory
387 >    static class DefaultForkJoinWorkerThreadFactory
388          implements ForkJoinWorkerThreadFactory {
389          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
390 <            try {
97 <                return new ForkJoinWorkerThread(pool);
98 <            } catch (OutOfMemoryError oom)  {
99 <                return null;
100 <            }
390 >            return new ForkJoinWorkerThread(pool);
391          }
392      }
393  
# Line 133 | Line 423 | public class ForkJoinPool extends Abstra
423          new AtomicInteger();
424  
425      /**
426 <     * Array holding all worker threads in the pool. Initialized upon
427 <     * first use. Array size must be a power of two.  Updates and
428 <     * replacements are protected by workerLock, but it is always kept
429 <     * in a consistent enough state to be randomly accessed without
430 <     * locking by workers performing work-stealing.
426 >     * The time to block in a join (see awaitJoin) before checking if
427 >     * a new worker should be (re)started to maintain parallelism
428 >     * level. The value should be short enough to maintain global
429 >     * responsiveness and progress but long enough to avoid
430 >     * counterproductive firings during GC stalls or unrelated system
431 >     * activity, and to not bog down systems with continual re-firings
432 >     * on GCs or legitimately long waits.
433 >     */
434 >    private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
435 >
436 >    /**
437 >     * The wakeup interval (in nanoseconds) for the oldest worker
438 >     * waiting for an event to invoke tryShutdownUnusedWorker to
439 >     * shrink the number of workers.  The exact value does not matter
440 >     * too much. It must be short enough to release resources during
441 >     * sustained periods of idleness, but not so short that threads
442 >     * are continually re-created.
443 >     */
444 >    private static final long SHRINK_RATE_NANOS =
445 >        30L * 1000L * 1000L * 1000L; // 2 per minute
446 >
447 >    /**
448 >     * Absolute bound for parallelism level. Twice this number plus
449 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
450 >     * word-packing for some counts and indices.
451 >     */
452 >    private static final int MAX_WORKERS   = 0x7fff;
453 >
454 >    /**
455 >     * Array holding all worker threads in the pool.  Array size must
456 >     * be a power of two.  Updates and replacements are protected by
457 >     * workerLock, but the array is always kept in a consistent enough
458 >     * state to be randomly accessed without locking by workers
459 >     * performing work-stealing, as well as other traversal-based
460 >     * methods in this class. All readers must tolerate that some
461 >     * array slots may be null.
462       */
463      volatile ForkJoinWorkerThread[] workers;
464  
465      /**
466 <     * Lock protecting access to workers.
466 >     * Queue for external submissions.
467       */
468 <    private final ReentrantLock workerLock;
468 >    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
469  
470      /**
471 <     * Condition for awaitTermination.
471 >     * Lock protecting updates to workers array.
472       */
473 <    private final Condition termination;
473 >    private final ReentrantLock workerLock;
474  
475      /**
476 <     * The uncaught exception handler used when any worker
156 <     * abrupty terminates
476 >     * Latch released upon termination.
477       */
478 <    private Thread.UncaughtExceptionHandler ueh;
478 >    private final Phaser termination;
479  
480      /**
481       * Creation factory for worker threads.
# Line 163 | Line 483 | public class ForkJoinPool extends Abstra
483      private final ForkJoinWorkerThreadFactory factory;
484  
485      /**
166     * Head of stack of threads that were created to maintain
167     * parallelism when other threads blocked, but have since
168     * suspended when the parallelism level rose.
169     */
170    private volatile WaitQueueNode spareStack;
171
172    /**
486       * Sum of per-thread steal counts, updated only when threads are
487       * idle or terminating.
488       */
489 <    private final AtomicLong stealCount;
489 >    private volatile long stealCount;
490  
491      /**
492 <     * Queue for external submissions.
492 >     * Encoded record of top of Treiber stack of threads waiting for
493 >     * events. The top 32 bits contain the count being waited for. The
494 >     * bottom 16 bits contains one plus the pool index of waiting
495 >     * worker thread. (Bits 16-31 are unused.)
496       */
497 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
497 >    private volatile long eventWaiters;
498  
499 <    /**
500 <     * Head of Treiber stack for barrier sync. See below for explanation
185 <     */
186 <    private volatile WaitQueueNode syncStack;
499 >    private static final int EVENT_COUNT_SHIFT = 32;
500 >    private static final int WAITER_ID_MASK    = (1 << 16) - 1;
501  
502      /**
503 <     * The count for event barrier
503 >     * A counter for events that may wake up worker threads:
504 >     *   - Submission of a new task to the pool
505 >     *   - A worker pushing a task on an empty queue
506 >     *   - termination
507       */
508 <    private volatile long eventCount;
508 >    private volatile int eventCount;
509  
510      /**
511 <     * Pool number, just for assigning useful names to worker threads
511 >     * Encoded record of top of Treiber stack of spare threads waiting
512 >     * for resumption. The top 16 bits contain an arbitrary count to
513 >     * avoid ABA effects. The bottom 16bits contains one plus the pool
514 >     * index of waiting worker thread.
515       */
516 <    private final int poolNumber;
516 >    private volatile int spareWaiters;
517  
518 <    /**
519 <     * The maximum allowed pool size
200 <     */
201 <    private volatile int maxPoolSize;
518 >    private static final int SPARE_COUNT_SHIFT = 16;
519 >    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
520  
521      /**
522 <     * The desired parallelism level, updated only under workerLock.
522 >     * Lifecycle control. The low word contains the number of workers
523 >     * that are (probably) executing tasks. This value is atomically
524 >     * incremented before a worker gets a task to run, and decremented
525 >     * when a worker has no tasks and cannot find any.  Bits 16-18
526 >     * contain runLevel value. When all are zero, the pool is
527 >     * running. Level transitions are monotonic (running -> shutdown
528 >     * -> terminating -> terminated) so each transition adds a bit.
529 >     * These are bundled together to ensure consistent read for
530 >     * termination checks (i.e., that runLevel is at least SHUTDOWN
531 >     * and active threads is zero).
532 >     *
533 >     * Notes: Most direct CASes are dependent on these bitfield
534 >     * positions.  Also, this field is non-private to enable direct
535 >     * performance-sensitive CASes in ForkJoinWorkerThread.
536       */
537 <    private volatile int parallelism;
537 >    volatile int runState;
538  
539 <    /**
540 <     * True if use local fifo, not default lifo, for local polling
541 <     */
542 <    private volatile boolean locallyFifo;
539 >    // Note: The order among run level values matters.
540 >    private static final int RUNLEVEL_SHIFT     = 16;
541 >    private static final int SHUTDOWN           = 1 << RUNLEVEL_SHIFT;
542 >    private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
543 >    private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
544 >    private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
545  
546      /**
547       * Holds number of total (i.e., created and not yet terminated)
548       * and running (i.e., not blocked on joins or other managed sync)
549 <     * threads, packed into one int to ensure consistent snapshot when
549 >     * threads, packed together to ensure consistent snapshot when
550       * making decisions about creating and suspending spare
551 <     * threads. Updated only by CAS.  Note: CASes in
552 <     * updateRunningCount and preJoin running active count is in low
553 <     * word, so need to be modified if this changes
551 >     * threads. Updated only by CAS. Note that adding a new worker
552 >     * requires incrementing both counts, since workers start off in
553 >     * running state.
554       */
555      private volatile int workerCounts;
556  
557 <    private static int totalCountOf(int s)           { return s >>> 16;  }
558 <    private static int runningCountOf(int s)         { return s & shortMask; }
559 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
557 >    private static final int TOTAL_COUNT_SHIFT  = 16;
558 >    private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
559 >    private static final int ONE_RUNNING        = 1;
560 >    private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
561  
562      /**
563 <     * Add delta (which may be negative) to running count.  This must
564 <     * be called before (with negative arg) and after (with positive)
231 <     * any managed synchronization (i.e., mainly, joins)
232 <     * @param delta the number to add
563 >     * The target parallelism level.
564 >     * Accessed directly by ForkJoinWorkerThreads.
565       */
566 <    final void updateRunningCount(int delta) {
235 <        int s;
236 <        do;while (!casWorkerCounts(s = workerCounts, s + delta));
237 <    }
566 >    final int parallelism;
567  
568      /**
569 <     * Add delta (which may be negative) to both total and running
570 <     * count.  This must be called upon creation and termination of
242 <     * worker threads.
243 <     * @param delta the number to add
569 >     * True if use local fifo, not default lifo, for local polling
570 >     * Read by, and replicated by ForkJoinWorkerThreads
571       */
572 <    private void updateWorkerCount(int delta) {
246 <        int d = delta + (delta << 16); // add to both lo and hi parts
247 <        int s;
248 <        do;while (!casWorkerCounts(s = workerCounts, s + d));
249 <    }
572 >    final boolean locallyFifo;
573  
574      /**
575 <     * Lifecycle control. High word contains runState, low word
576 <     * contains the number of workers that are (probably) executing
254 <     * tasks. This value is atomically incremented before a worker
255 <     * gets a task to run, and decremented when worker has no tasks
256 <     * and cannot find any. These two fields are bundled together to
257 <     * support correct termination triggering.  Note: activeCount
258 <     * CAS'es cheat by assuming active count is in low word, so need
259 <     * to be modified if this changes
575 >     * The uncaught exception handler used when any worker abruptly
576 >     * terminates.
577       */
578 <    private volatile int runControl;
578 >    private final Thread.UncaughtExceptionHandler ueh;
579  
580 <    // RunState values. Order among values matters
581 <    private static final int RUNNING     = 0;
582 <    private static final int SHUTDOWN    = 1;
583 <    private static final int TERMINATING = 2;
267 <    private static final int TERMINATED  = 3;
580 >    /**
581 >     * Pool number, just for assigning useful names to worker threads
582 >     */
583 >    private final int poolNumber;
584  
585 <    private static int runStateOf(int c)             { return c >>> 16; }
586 <    private static int activeCountOf(int c)          { return c & shortMask; }
271 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
585 >    // Utilities for CASing fields. Note that most of these
586 >    // are usually manually inlined by callers
587  
588      /**
589 <     * Try incrementing active count; fail on contention. Called by
275 <     * workers before/during executing tasks.
276 <     * @return true on success;
589 >     * Increments running count part of workerCounts.
590       */
591 <    final boolean tryIncrementActiveCount() {
592 <        int c = runControl;
593 <        return casRunControl(c, c+1);
591 >    final void incrementRunningCount() {
592 >        int c;
593 >        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
594 >                                               c = workerCounts,
595 >                                               c + ONE_RUNNING));
596      }
597  
598      /**
599 <     * Try decrementing active count; fail on contention.
285 <     * Possibly trigger termination on success
286 <     * Called by workers when they can't find tasks.
287 <     * @return true on success
599 >     * Tries to increment running count part of workerCounts.
600       */
601 <    final boolean tryDecrementActiveCount() {
602 <        int c = runControl;
603 <        int nextc = c - 1;
604 <        if (!casRunControl(c, nextc))
605 <            return false;
294 <        if (canTerminateOnShutdown(nextc))
295 <            terminateOnShutdown();
296 <        return true;
601 >    final boolean tryIncrementRunningCount() {
602 >        int c;
603 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
604 >                                        c = workerCounts,
605 >                                        c + ONE_RUNNING);
606      }
607  
608      /**
609 <     * Return true if argument represents zero active count and
301 <     * nonzero runstate, which is the triggering condition for
302 <     * terminating on shutdown.
609 >     * Tries to decrement running count unless already zero.
610       */
611 <    private static boolean canTerminateOnShutdown(int c) {
612 <        return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
611 >    final boolean tryDecrementRunningCount() {
612 >        int wc = workerCounts;
613 >        if ((wc & RUNNING_COUNT_MASK) == 0)
614 >            return false;
615 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
616 >                                        wc, wc - ONE_RUNNING);
617      }
618  
619      /**
620 <     * Transition run state to at least the given state. Return true
621 <     * if not already at least given state.
620 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
621 >     * (rarely) necessary when other count updates lag.
622 >     *
623 >     * @param dr -- either zero or ONE_RUNNING
624 >     * @param dt -- either zero or ONE_TOTAL
625       */
626 <    private boolean transitionRunStateTo(int state) {
626 >    private void decrementWorkerCounts(int dr, int dt) {
627          for (;;) {
628 <            int c = runControl;
629 <            if (runStateOf(c) >= state)
630 <                return false;
631 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
632 <                return true;
628 >            int wc = workerCounts;
629 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
630 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
631 >                if ((runState & TERMINATED) != 0)
632 >                    return; // lagging termination on a backout
633 >                Thread.yield();
634 >            }
635 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
636 >                                         wc, wc - (dr + dt)))
637 >                return;
638          }
639      }
640  
641      /**
642 <     * Controls whether to add spares to maintain parallelism
642 >     * Tries decrementing active count; fails on contention.
643 >     * Called when workers cannot find tasks to run.
644       */
645 <    private volatile boolean maintainsParallelism;
646 <
647 <    // Constructors
645 >    final boolean tryDecrementActiveCount() {
646 >        int c;
647 >        return UNSAFE.compareAndSwapInt(this, runStateOffset,
648 >                                        c = runState, c - 1);
649 >    }
650  
651      /**
652 <     * Creates a ForkJoinPool with a pool size equal to the number of
653 <     * processors available on the system and using the default
332 <     * ForkJoinWorkerThreadFactory,
333 <     * @throws SecurityException if a security manager exists and
334 <     *         the caller is not permitted to modify threads
335 <     *         because it does not hold {@link
336 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
652 >     * Advances to at least the given level. Returns true if not
653 >     * already in at least the given level.
654       */
655 <    public ForkJoinPool() {
656 <        this(Runtime.getRuntime().availableProcessors(),
657 <             defaultForkJoinWorkerThreadFactory);
655 >    private boolean advanceRunLevel(int level) {
656 >        for (;;) {
657 >            int s = runState;
658 >            if ((s & level) != 0)
659 >                return false;
660 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
661 >                return true;
662 >        }
663      }
664  
665 +    // workers array maintenance
666 +
667      /**
668 <     * Creates a ForkJoinPool with the indicated parellelism level
345 <     * threads, and using the default ForkJoinWorkerThreadFactory,
346 <     * @param parallelism the number of worker threads
347 <     * @throws IllegalArgumentException if parallelism less than or
348 <     * equal to zero
349 <     * @throws SecurityException if a security manager exists and
350 <     *         the caller is not permitted to modify threads
351 <     *         because it does not hold {@link
352 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
668 >     * Records and returns a workers array index for new worker.
669       */
670 <    public ForkJoinPool(int parallelism) {
671 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
670 >    private int recordWorker(ForkJoinWorkerThread w) {
671 >        // Try using slot totalCount-1. If not available, scan and/or resize
672 >        int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
673 >        final ReentrantLock lock = this.workerLock;
674 >        lock.lock();
675 >        try {
676 >            ForkJoinWorkerThread[] ws = workers;
677 >            int n = ws.length;
678 >            if (k < 0 || k >= n || ws[k] != null) {
679 >                for (k = 0; k < n && ws[k] != null; ++k)
680 >                    ;
681 >                if (k == n)
682 >                    ws = workers = Arrays.copyOf(ws, n << 1);
683 >            }
684 >            ws[k] = w;
685 >            int c = eventCount; // advance event count to ensure visibility
686 >            UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1);
687 >        } finally {
688 >            lock.unlock();
689 >        }
690 >        return k;
691      }
692  
693      /**
694 <     * Creates a ForkJoinPool with parallelism equal to the number of
360 <     * processors available on the system and using the given
361 <     * ForkJoinWorkerThreadFactory,
362 <     * @param factory the factory for creating new threads
363 <     * @throws NullPointerException if factory is null
364 <     * @throws SecurityException if a security manager exists and
365 <     *         the caller is not permitted to modify threads
366 <     *         because it does not hold {@link
367 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
694 >     * Nulls out record of worker in workers array.
695       */
696 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
697 <        this(Runtime.getRuntime().availableProcessors(), factory);
696 >    private void forgetWorker(ForkJoinWorkerThread w) {
697 >        int idx = w.poolIndex;
698 >        // Locking helps method recordWorker avoid unnecessary expansion
699 >        final ReentrantLock lock = this.workerLock;
700 >        lock.lock();
701 >        try {
702 >            ForkJoinWorkerThread[] ws = workers;
703 >            if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
704 >                ws[idx] = null;
705 >        } finally {
706 >            lock.unlock();
707 >        }
708      }
709  
710      /**
711 <     * Creates a ForkJoinPool with the given parallelism and factory.
711 >     * Final callback from terminating worker.  Removes record of
712 >     * worker from array, and adjusts counts. If pool is shutting
713 >     * down, tries to complete termination.
714       *
715 <     * @param parallelism the targeted number of worker threads
377 <     * @param factory the factory for creating new threads
378 <     * @throws IllegalArgumentException if parallelism less than or
379 <     * equal to zero, or greater than implementation limit.
380 <     * @throws NullPointerException if factory is null
381 <     * @throws SecurityException if a security manager exists and
382 <     *         the caller is not permitted to modify threads
383 <     *         because it does not hold {@link
384 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
715 >     * @param w the worker
716       */
717 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
718 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
719 <            throw new IllegalArgumentException();
720 <        if (factory == null)
721 <            throw new NullPointerException();
722 <        checkPermission();
392 <        this.factory = factory;
393 <        this.parallelism = parallelism;
394 <        this.maxPoolSize = MAX_THREADS;
395 <        this.maintainsParallelism = true;
396 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
397 <        this.workerLock = new ReentrantLock();
398 <        this.termination = workerLock.newCondition();
399 <        this.stealCount = new AtomicLong();
400 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
401 <        // worker array and workers are lazily constructed
717 >    final void workerTerminated(ForkJoinWorkerThread w) {
718 >        forgetWorker(w);
719 >        decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL);
720 >        while (w.stealCount != 0) // collect final count
721 >            tryAccumulateStealCount(w);
722 >        tryTerminate(false);
723      }
724  
725 +    // Waiting for and signalling events
726 +
727      /**
728 <     * Create new worker using factory.
729 <     * @param index the index to assign worker
730 <     * @return new worker, or null of factory failed
728 >     * Releases workers blocked on a count not equal to current count.
729 >     * Normally called after precheck that eventWaiters isn't zero to
730 >     * avoid wasted array checks. Gives up upon a change in count or
731 >     * upon releasing two workers, letting others take over.
732       */
733 <    private ForkJoinWorkerThread createWorker(int index) {
734 <        Thread.UncaughtExceptionHandler h = ueh;
735 <        ForkJoinWorkerThread w = factory.newThread(this);
736 <        if (w != null) {
737 <            w.poolIndex = index;
738 <            w.setDaemon(true);
739 <            w.setAsyncMode(locallyFifo);
740 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
741 <            if (h != null)
742 <                w.setUncaughtExceptionHandler(h);
733 >    private void releaseEventWaiters() {
734 >        ForkJoinWorkerThread[] ws = workers;
735 >        int n = ws.length;
736 >        long h = eventWaiters;
737 >        int ec = eventCount;
738 >        boolean releasedOne = false;
739 >        ForkJoinWorkerThread w; int id;
740 >        while ((id = (((int)h) & WAITER_ID_MASK) - 1) >= 0 &&
741 >               (int)(h >>> EVENT_COUNT_SHIFT) != ec &&
742 >               id < n && (w = ws[id]) != null) {
743 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
744 >                                          h,  w.nextWaiter)) {
745 >                LockSupport.unpark(w);
746 >                if (releasedOne) // exit on second release
747 >                    break;
748 >                releasedOne = true;
749 >            }
750 >            if (eventCount != ec)
751 >                break;
752 >            h = eventWaiters;
753          }
420        return w;
754      }
755  
756      /**
757 <     * Return a good size for worker array given pool size.
758 <     * Currently requires size to be a power of two.
757 >     * Tries to advance eventCount and releases waiters. Called only
758 >     * from workers.
759       */
760 <    private static int arraySizeFor(int ps) {
761 <        return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
760 >    final void signalWork() {
761 >        int c; // try to increment event count -- CAS failure OK
762 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
763 >        if (eventWaiters != 0L)
764 >            releaseEventWaiters();
765      }
766  
767      /**
768 <     * Create or resize array if necessary to hold newLength.
769 <     * Call only under exlusion or lock
770 <     * @return the array
768 >     * Adds the given worker to event queue and blocks until
769 >     * terminating or event count advances from the given value
770 >     *
771 >     * @param w the calling worker thread
772 >     * @param ec the count
773       */
774 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
775 <        ForkJoinWorkerThread[] ws = workers;
776 <        if (ws == null)
777 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
778 <        else if (newLength > ws.length)
779 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
780 <        else
781 <            return ws;
774 >    private void eventSync(ForkJoinWorkerThread w, int ec) {
775 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
776 >        long h;
777 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
778 >               (((int)(h = eventWaiters) & WAITER_ID_MASK) == 0 ||
779 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
780 >               eventCount == ec) {
781 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
782 >                                          w.nextWaiter = h, nh)) {
783 >                awaitEvent(w, ec);
784 >                break;
785 >            }
786 >        }
787      }
788  
789      /**
790 <     * Try to shrink workers into smaller array after one or more terminate
790 >     * Blocks the given worker (that has already been entered as an
791 >     * event waiter) until terminating or event count advances from
792 >     * the given value. The oldest (first) waiter uses a timed wait to
793 >     * occasionally one-by-one shrink the number of workers (to a
794 >     * minimum of one) if the pool has not been used for extended
795 >     * periods.
796 >     *
797 >     * @param w the calling worker thread
798 >     * @param ec the count
799       */
800 <    private void tryShrinkWorkerArray() {
801 <        ForkJoinWorkerThread[] ws = workers;
802 <        if (ws != null) {
803 <            int len = ws.length;
804 <            int last = len - 1;
805 <            while (last >= 0 && ws[last] == null)
806 <                --last;
807 <            int newLength = arraySizeFor(last+1);
808 <            if (newLength < len)
809 <                workers = Arrays.copyOf(ws, newLength);
800 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
801 >        while (eventCount == ec) {
802 >            if (tryAccumulateStealCount(w)) { // transfer while idle
803 >                boolean untimed = (w.nextWaiter != 0L ||
804 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
805 >                long startTime = untimed ? 0 : System.nanoTime();
806 >                Thread.interrupted();         // clear/ignore interrupt
807 >                if (w.isTerminating() || eventCount != ec)
808 >                    break;                    // recheck after clear
809 >                if (untimed)
810 >                    LockSupport.park(w);
811 >                else {
812 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
813 >                    if (eventCount != ec || w.isTerminating())
814 >                        break;
815 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
816 >                        tryShutdownUnusedWorker(ec);
817 >                }
818 >            }
819          }
820      }
821  
822 +    // Maintaining parallelism
823 +
824 +    /**
825 +     * Pushes worker onto the spare stack.
826 +     */
827 +    final void pushSpare(ForkJoinWorkerThread w) {
828 +        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
829 +        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
830 +                                               w.nextSpare = spareWaiters,ns));
831 +    }
832 +
833      /**
834 <     * Initialize workers if necessary
834 >     * Tries (once) to resume a spare if the number of running
835 >     * threads is less than target.
836       */
837 <    final void ensureWorkerInitialization() {
837 >    private void tryResumeSpare() {
838 >        int sw, id;
839          ForkJoinWorkerThread[] ws = workers;
840 <        if (ws == null) {
841 <            final ReentrantLock lock = this.workerLock;
842 <            lock.lock();
843 <            try {
844 <                ws = workers;
845 <                if (ws == null) {
846 <                    int ps = parallelism;
847 <                    ws = ensureWorkerArrayCapacity(ps);
848 <                    for (int i = 0; i < ps; ++i) {
849 <                        ForkJoinWorkerThread w = createWorker(i);
850 <                        if (w != null) {
851 <                            ws[i] = w;
852 <                            w.start();
853 <                            updateWorkerCount(1);
854 <                        }
855 <                    }
856 <                }
857 <            } finally {
485 <                lock.unlock();
486 <            }
840 >        int n = ws.length;
841 >        ForkJoinWorkerThread w;
842 >        if ((sw = spareWaiters) != 0 &&
843 >            (id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
844 >            id < n && (w = ws[id]) != null &&
845 >            (runState >= TERMINATING ||
846 >             (workerCounts & RUNNING_COUNT_MASK) < parallelism) &&
847 >            spareWaiters == sw &&
848 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
849 >                                     sw, w.nextSpare)) {
850 >            int c; // increment running count before resume
851 >            do {} while (!UNSAFE.compareAndSwapInt
852 >                         (this, workerCountsOffset,
853 >                          c = workerCounts, c + ONE_RUNNING));
854 >            if (w.tryUnsuspend())
855 >                LockSupport.unpark(w);
856 >            else   // back out if w was shutdown
857 >                decrementWorkerCounts(ONE_RUNNING, 0);
858          }
859      }
860  
861      /**
862 <     * Worker creation and startup for threads added via setParallelism.
863 <     */
864 <    private void createAndStartAddedWorkers() {
865 <        resumeAllSpares();  // Allow spares to convert to nonspare
866 <        int ps = parallelism;
867 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
868 <        int len = ws.length;
869 <        // Sweep through slots, to keep lowest indices most populated
870 <        int k = 0;
871 <        while (k < len) {
872 <            if (ws[k] != null) {
873 <                ++k;
874 <                continue;
875 <            }
876 <            int s = workerCounts;
877 <            int tc = totalCountOf(s);
878 <            int rc = runningCountOf(s);
879 <            if (rc >= ps || tc >= ps)
880 <                break;
881 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
882 <                ForkJoinWorkerThread w = createWorker(k);
883 <                if (w != null) {
884 <                    ws[k++] = w;
885 <                    w.start();
862 >     * Tries to increase the number of running workers if below target
863 >     * parallelism: If a spare exists tries to resume it via
864 >     * tryResumeSpare.  Otherwise, if not enough total workers or all
865 >     * existing workers are busy, adds a new worker. In all cases also
866 >     * helps wake up releasable workers waiting for work.
867 >     */
868 >    private void helpMaintainParallelism() {
869 >        int pc = parallelism;
870 >        int wc, rs, tc;
871 >        while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
872 >               (rs = runState) < TERMINATING) {
873 >            if (spareWaiters != 0)
874 >                tryResumeSpare();
875 >            else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
876 >                     (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
877 >                break;   // enough total
878 >            else if (runState == rs && workerCounts == wc &&
879 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
880 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
881 >                ForkJoinWorkerThread w = null;
882 >                Throwable fail = null;
883 >                try {
884 >                    w = factory.newThread(this);
885 >                } catch (Throwable ex) {
886 >                    fail = ex;
887                  }
888 <                else {
889 <                    updateWorkerCount(-1); // back out on failed creation
888 >                if (w == null) { // null or exceptional factory return
889 >                    decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
890 >                    tryTerminate(false); // handle failure during shutdown
891 >                    // If originating from an external caller,
892 >                    // propagate exception, else ignore
893 >                    if (fail != null && runState < TERMINATING &&
894 >                        !(Thread.currentThread() instanceof
895 >                          ForkJoinWorkerThread))
896 >                        UNSAFE.throwException(fail);
897                      break;
898                  }
899 +                w.start(recordWorker(w), ueh);
900 +                if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc)
901 +                    break; // add at most one unless total below target
902              }
903          }
904 +        if (eventWaiters != 0L)
905 +            releaseEventWaiters();
906      }
907  
524    // Execution methods
525
908      /**
909 <     * Common code for execute, invoke and submit
909 >     * Callback from the oldest waiter in awaitEvent waking up after a
910 >     * period of non-use. If all workers are idle, tries (once) to
911 >     * shutdown an event waiter or a spare, if one exists. Note that
912 >     * we don't need CAS or locks here because the method is called
913 >     * only from one thread occasionally waking (and even misfires are
914 >     * OK). Note that until the shutdown worker fully terminates,
915 >     * workerCounts will overestimate total count, which is tolerable.
916 >     *
917 >     * @param ec the event count waited on by caller (to abort
918 >     * attempt if count has since changed).
919       */
920 <    private <T> void doSubmit(ForkJoinTask<T> task) {
921 <        if (isShutdown())
922 <            throw new RejectedExecutionException();
923 <        if (workers == null)
924 <            ensureWorkerInitialization();
925 <        submissionQueue.offer(task);
926 <        signalIdleWorkers();
920 >    private void tryShutdownUnusedWorker(int ec) {
921 >        if (runState == 0 && eventCount == ec) { // only trigger if all idle
922 >            ForkJoinWorkerThread[] ws = workers;
923 >            int n = ws.length;
924 >            ForkJoinWorkerThread w = null;
925 >            boolean shutdown = false;
926 >            int sw;
927 >            long h;
928 >            if ((sw = spareWaiters) != 0) { // prefer killing spares
929 >                int id = (sw & SPARE_ID_MASK) - 1;
930 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
931 >                    UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
932 >                                             sw, w.nextSpare))
933 >                    shutdown = true;
934 >            }
935 >            else if ((h = eventWaiters) != 0L) {
936 >                long nh;
937 >                int id = (((int)h) & WAITER_ID_MASK) - 1;
938 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
939 >                    (nh = w.nextWaiter) != 0L && // keep at least one worker
940 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
941 >                    shutdown = true;
942 >            }
943 >            if (w != null && shutdown) {
944 >                w.shutdown();
945 >                LockSupport.unpark(w);
946 >            }
947 >        }
948 >        releaseEventWaiters(); // in case of interference
949      }
950  
951      /**
952 <     * Performs the given task; returning its result upon completion
953 <     * @param task the task
954 <     * @return the task's result
955 <     * @throws NullPointerException if task is null
956 <     * @throws RejectedExecutionException if pool is shut down
952 >     * Callback from workers invoked upon each top-level action (i.e.,
953 >     * stealing a task or taking a submission and running it).
954 >     * Performs one or more of the following:
955 >     *
956 >     * 1. If the worker is active and either did not run a task
957 >     *    or there are too many workers, try to set its active status
958 >     *    to inactive and update activeCount. On contention, we may
959 >     *    try again in this or a subsequent call.
960 >     *
961 >     * 2. If not enough total workers, help create some.
962 >     *
963 >     * 3. If there are too many running workers, suspend this worker
964 >     *    (first forcing inactive if necessary).  If it is not needed,
965 >     *    it may be shutdown while suspended (via
966 >     *    tryShutdownUnusedWorker).  Otherwise, upon resume it
967 >     *    rechecks running thread count and need for event sync.
968 >     *
969 >     * 4. If worker did not run a task, await the next task event via
970 >     *    eventSync if necessary (first forcing inactivation), upon
971 >     *    which the worker may be shutdown via
972 >     *    tryShutdownUnusedWorker.  Otherwise, help release any
973 >     *    existing event waiters that are now releasable,
974 >     *
975 >     * @param w the worker
976 >     * @param ran true if worker ran a task since last call to this method
977       */
978 <    public <T> T invoke(ForkJoinTask<T> task) {
979 <        doSubmit(task);
980 <        return task.join();
978 >    final void preStep(ForkJoinWorkerThread w, boolean ran) {
979 >        int wec = w.lastEventCount;
980 >        boolean active = w.active;
981 >        boolean inactivate = false;
982 >        int pc = parallelism;
983 >        while (w.runState == 0) {
984 >            int rs = runState;
985 >            if (rs >= TERMINATING) {           // propagate shutdown
986 >                w.shutdown();
987 >                break;
988 >            }
989 >            if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
990 >                UNSAFE.compareAndSwapInt(this, runStateOffset, rs, --rs)) {
991 >                inactivate = active = w.active = false;
992 >                if (rs == SHUTDOWN) {          // all inactive and shut down
993 >                    tryTerminate(false);
994 >                    continue;
995 >                }
996 >            }
997 >            int wc = workerCounts;             // try to suspend as spare
998 >            if ((wc & RUNNING_COUNT_MASK) > pc) {
999 >                if (!(inactivate |= active) && // must inactivate to suspend
1000 >                    workerCounts == wc &&
1001 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1002 >                                             wc, wc - ONE_RUNNING))
1003 >                    w.suspendAsSpare();
1004 >            }
1005 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
1006 >                helpMaintainParallelism();     // not enough workers
1007 >            else if (ran)
1008 >                break;
1009 >            else {
1010 >                long h = eventWaiters;
1011 >                int ec = eventCount;
1012 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
1013 >                    releaseEventWaiters();     // release others before waiting
1014 >                else if (ec != wec) {
1015 >                    w.lastEventCount = ec;     // no need to wait
1016 >                    break;
1017 >                }
1018 >                else if (!(inactivate |= active))
1019 >                    eventSync(w, wec);         // must inactivate before sync
1020 >            }
1021 >        }
1022      }
1023  
1024      /**
1025 <     * Arranges for (asynchronous) execution of the given task.
1026 <     * @param task the task
1027 <     * @throws NullPointerException if task is null
1028 <     * @throws RejectedExecutionException if pool is shut down
1025 >     * Helps and/or blocks awaiting join of the given task.
1026 >     * See above for explanation.
1027 >     *
1028 >     * @param joinMe the task to join
1029 >     * @param worker the current worker thread
1030 >     * @param timed true if wait should time out
1031 >     * @param nanos timeout value if timed
1032       */
1033 <    public <T> void execute(ForkJoinTask<T> task) {
1034 <        doSubmit(task);
1033 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker,
1034 >                         boolean timed, long nanos) {
1035 >        long startTime = timed? System.nanoTime() : 0L;
1036 >        int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1037 >        boolean running = true;               // false when count decremented
1038 >        while (joinMe.status >= 0) {
1039 >            if (runState >= TERMINATING) {
1040 >                joinMe.cancelIgnoringExceptions();
1041 >                break;
1042 >            }
1043 >            running = worker.helpJoinTask(joinMe, running);
1044 >            if (joinMe.status < 0)
1045 >                break;
1046 >            if (retries > 0) {
1047 >                --retries;
1048 >                continue;
1049 >            }
1050 >            int wc = workerCounts;
1051 >            if ((wc & RUNNING_COUNT_MASK) != 0) {
1052 >                if (running) {
1053 >                    if (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1054 >                                                  wc, wc - ONE_RUNNING))
1055 >                        continue;
1056 >                    running = false;
1057 >                }
1058 >                long h = eventWaiters;
1059 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1060 >                    releaseEventWaiters();
1061 >                if ((workerCounts & RUNNING_COUNT_MASK) != 0) {
1062 >                    long ms; int ns;
1063 >                    if (!timed) {
1064 >                        ms = JOIN_TIMEOUT_MILLIS;
1065 >                        ns = 0;
1066 >                    }
1067 >                    else { // at most JOIN_TIMEOUT_MILLIS per wait
1068 >                        long nt = nanos - (System.nanoTime() - startTime);
1069 >                        if (nt <= 0L)
1070 >                            break;
1071 >                        ms = nt / 1000000;
1072 >                        if (ms > JOIN_TIMEOUT_MILLIS) {
1073 >                            ms = JOIN_TIMEOUT_MILLIS;
1074 >                            ns = 0;
1075 >                        }
1076 >                        else
1077 >                            ns = (int) (nt % 1000000);
1078 >                    }
1079 >                    joinMe.internalAwaitDone(ms, ns);
1080 >                }
1081 >                if (joinMe.status < 0)
1082 >                    break;
1083 >            }
1084 >            helpMaintainParallelism();
1085 >        }
1086 >        if (!running) {
1087 >            int c;
1088 >            do {} while (!UNSAFE.compareAndSwapInt
1089 >                         (this, workerCountsOffset,
1090 >                          c = workerCounts, c + ONE_RUNNING));
1091 >        }
1092      }
1093  
1094 <    // AbstractExecutorService methods
1095 <
1096 <    public void execute(Runnable task) {
1097 <        doSubmit(new AdaptedRunnable<Void>(task, null));
1094 >    /**
1095 >     * Same idea as awaitJoin, but no helping, retries, or timeouts.
1096 >     */
1097 >    final void awaitBlocker(ManagedBlocker blocker)
1098 >        throws InterruptedException {
1099 >        while (!blocker.isReleasable()) {
1100 >            int wc = workerCounts;
1101 >            if ((wc & RUNNING_COUNT_MASK) == 0)
1102 >                helpMaintainParallelism();
1103 >            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1104 >                                              wc, wc - ONE_RUNNING)) {
1105 >                try {
1106 >                    while (!blocker.isReleasable()) {
1107 >                        long h = eventWaiters;
1108 >                        if (h != 0L &&
1109 >                            (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1110 >                            releaseEventWaiters();
1111 >                        else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1112 >                                 runState < TERMINATING)
1113 >                            helpMaintainParallelism();
1114 >                        else if (blocker.block())
1115 >                            break;
1116 >                    }
1117 >                } finally {
1118 >                    int c;
1119 >                    do {} while (!UNSAFE.compareAndSwapInt
1120 >                                 (this, workerCountsOffset,
1121 >                                  c = workerCounts, c + ONE_RUNNING));
1122 >                }
1123 >                break;
1124 >            }
1125 >        }
1126      }
1127  
1128 <    public <T> ForkJoinTask<T> submit(Callable<T> task) {
1129 <        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
1130 <        doSubmit(job);
1131 <        return job;
1132 <    }
1128 >    /**
1129 >     * Possibly initiates and/or completes termination.
1130 >     *
1131 >     * @param now if true, unconditionally terminate, else only
1132 >     * if shutdown and empty queue and no active workers
1133 >     * @return true if now terminating or terminated
1134 >     */
1135 >    private boolean tryTerminate(boolean now) {
1136 >        if (now)
1137 >            advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
1138 >        else if (runState < SHUTDOWN ||
1139 >                 !submissionQueue.isEmpty() ||
1140 >                 (runState & ACTIVE_COUNT_MASK) != 0)
1141 >            return false;
1142  
1143 <    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1144 <        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
574 <        doSubmit(job);
575 <        return job;
576 <    }
1143 >        if (advanceRunLevel(TERMINATING))
1144 >            startTerminating();
1145  
1146 <    public ForkJoinTask<?> submit(Runnable task) {
1147 <        ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
1148 <        doSubmit(job);
1149 <        return job;
1146 >        // Finish now if all threads terminated; else in some subsequent call
1147 >        if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1148 >            advanceRunLevel(TERMINATED);
1149 >            termination.forceTermination();
1150 >        }
1151 >        return true;
1152      }
1153  
1154      /**
1155 <     * Adaptor for Runnables. This implements RunnableFuture
1156 <     * to be compliant with AbstractExecutorService constraints
1157 <     */
1158 <    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
1159 <        implements RunnableFuture<T> {
1160 <        final Runnable runnable;
1161 <        final T resultOnCompletion;
1162 <        T result;
1163 <        AdaptedRunnable(Runnable runnable, T result) {
1164 <            if (runnable == null) throw new NullPointerException();
1165 <            this.runnable = runnable;
1166 <            this.resultOnCompletion = result;
1167 <        }
1168 <        public T getRawResult() { return result; }
1169 <        public void setRawResult(T v) { result = v; }
1170 <        public boolean exec() {
1171 <            runnable.run();
1172 <            result = resultOnCompletion;
1173 <            return true;
1155 >     * Actions on transition to TERMINATING
1156 >     *
1157 >     * Runs up to four passes through workers: (0) shutting down each
1158 >     * (without waking up if parked) to quickly spread notifications
1159 >     * without unnecessary bouncing around event queues etc (1) wake
1160 >     * up and help cancel tasks (2) interrupt (3) mop up races with
1161 >     * interrupted workers
1162 >     */
1163 >    private void startTerminating() {
1164 >        cancelSubmissions();
1165 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1166 >            int c; // advance event count
1167 >            UNSAFE.compareAndSwapInt(this, eventCountOffset,
1168 >                                     c = eventCount, c+1);
1169 >            eventWaiters = 0L; // clobber lists
1170 >            spareWaiters = 0;
1171 >            for (ForkJoinWorkerThread w : workers) {
1172 >                if (w != null) {
1173 >                    w.shutdown();
1174 >                    if (passes > 0 && !w.isTerminated()) {
1175 >                        w.cancelTasks();
1176 >                        LockSupport.unpark(w);
1177 >                        if (passes > 1 && !w.isInterrupted()) {
1178 >                            try {
1179 >                                w.interrupt();
1180 >                            } catch (SecurityException ignore) {
1181 >                            }
1182 >                        }
1183 >                    }
1184 >                }
1185 >            }
1186          }
605        public void run() { invoke(); }
1187      }
1188  
1189      /**
1190 <     * Adaptor for Callables
1190 >     * Clears out and cancels submissions, ignoring exceptions.
1191       */
1192 <    static final class AdaptedCallable<T> extends ForkJoinTask<T>
1193 <        implements RunnableFuture<T> {
1194 <        final Callable<T> callable;
614 <        T result;
615 <        AdaptedCallable(Callable<T> callable) {
616 <            if (callable == null) throw new NullPointerException();
617 <            this.callable = callable;
618 <        }
619 <        public T getRawResult() { return result; }
620 <        public void setRawResult(T v) { result = v; }
621 <        public boolean exec() {
1192 >    private void cancelSubmissions() {
1193 >        ForkJoinTask<?> task;
1194 >        while ((task = submissionQueue.poll()) != null) {
1195              try {
1196 <                result = callable.call();
1197 <                return true;
625 <            } catch (Error err) {
626 <                throw err;
627 <            } catch (RuntimeException rex) {
628 <                throw rex;
629 <            } catch (Exception ex) {
630 <                throw new RuntimeException(ex);
1196 >                task.cancel(false);
1197 >            } catch (Throwable ignore) {
1198              }
1199          }
633        public void run() { invoke(); }
1200      }
1201  
1202 <    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
637 <        ArrayList<ForkJoinTask<T>> ts =
638 <            new ArrayList<ForkJoinTask<T>>(tasks.size());
639 <        for (Callable<T> c : tasks)
640 <            ts.add(new AdaptedCallable<T>(c));
641 <        invoke(new InvokeAll<T>(ts));
642 <        return (List<Future<T>>)(List)ts;
643 <    }
1202 >    // misc support for ForkJoinWorkerThread
1203  
1204 <    static final class InvokeAll<T> extends RecursiveAction {
1205 <        final ArrayList<ForkJoinTask<T>> tasks;
1206 <        InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
1207 <        public void compute() {
1208 <            try { invokeAll(tasks); } catch(Exception ignore) {}
650 <        }
1204 >    /**
1205 >     * Returns pool number.
1206 >     */
1207 >    final int getPoolNumber() {
1208 >        return poolNumber;
1209      }
1210  
653    // Configuration and status settings and queries
654
1211      /**
1212 <     * Returns the factory used for constructing new workers
1212 >     * Tries to accumulate steal count from a worker, clearing
1213 >     * the worker's value if successful.
1214       *
1215 <     * @return the factory used for constructing new workers
1215 >     * @return true if worker steal count now zero
1216       */
1217 <    public ForkJoinWorkerThreadFactory getFactory() {
1218 <        return factory;
1217 >    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1218 >        int sc = w.stealCount;
1219 >        long c = stealCount;
1220 >        // CAS even if zero, for fence effects
1221 >        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1222 >            if (sc != 0)
1223 >                w.stealCount = 0;
1224 >            return true;
1225 >        }
1226 >        return sc == 0;
1227      }
1228  
1229      /**
1230 <     * Returns the handler for internal worker threads that terminate
1231 <     * due to unrecoverable errors encountered while executing tasks.
667 <     * @return the handler, or null if none
1230 >     * Returns the approximate (non-atomic) number of idle threads per
1231 >     * active thread.
1232       */
1233 <    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1234 <        Thread.UncaughtExceptionHandler h;
1235 <        final ReentrantLock lock = this.workerLock;
1236 <        lock.lock();
1237 <        try {
1238 <            h = ueh;
1239 <        } finally {
1240 <            lock.unlock();
677 <        }
678 <        return h;
1233 >    final int idlePerActive() {
1234 >        int pc = parallelism; // use parallelism, not rc
1235 >        int ac = runState;    // no mask -- artificially boosts during shutdown
1236 >        // Use exact results for small values, saturate past 4
1237 >        return ((pc <= ac) ? 0 :
1238 >                (pc >>> 1 <= ac) ? 1 :
1239 >                (pc >>> 2 <= ac) ? 3 :
1240 >                pc >>> 3);
1241      }
1242  
1243 +    // Public and protected methods
1244 +
1245 +    // Constructors
1246 +
1247      /**
1248 <     * Sets the handler for internal worker threads that terminate due
1249 <     * to unrecoverable errors encountered while executing tasks.
1250 <     * Unless set, the current default or ThreadGroup handler is used
1251 <     * as handler.
1248 >     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1249 >     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1250 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
1251 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1252       *
687     * @param h the new handler
688     * @return the old handler, or null if none
1253       * @throws SecurityException if a security manager exists and
1254       *         the caller is not permitted to modify threads
1255       *         because it does not hold {@link
1256 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1256 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1257       */
1258 <    public Thread.UncaughtExceptionHandler
1259 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
1260 <        checkPermission();
697 <        Thread.UncaughtExceptionHandler old = null;
698 <        final ReentrantLock lock = this.workerLock;
699 <        lock.lock();
700 <        try {
701 <            old = ueh;
702 <            ueh = h;
703 <            ForkJoinWorkerThread[] ws = workers;
704 <            if (ws != null) {
705 <                for (int i = 0; i < ws.length; ++i) {
706 <                    ForkJoinWorkerThread w = ws[i];
707 <                    if (w != null)
708 <                        w.setUncaughtExceptionHandler(h);
709 <                }
710 <            }
711 <        } finally {
712 <            lock.unlock();
713 <        }
714 <        return old;
1258 >    public ForkJoinPool() {
1259 >        this(Runtime.getRuntime().availableProcessors(),
1260 >             defaultForkJoinWorkerThreadFactory, null, false);
1261      }
1262  
1263 +    /**
1264 +     * Creates a {@code ForkJoinPool} with the indicated parallelism
1265 +     * level, the {@linkplain
1266 +     * #defaultForkJoinWorkerThreadFactory default thread factory},
1267 +     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1268 +     *
1269 +     * @param parallelism the parallelism level
1270 +     * @throws IllegalArgumentException if parallelism less than or
1271 +     *         equal to zero, or greater than implementation limit
1272 +     * @throws SecurityException if a security manager exists and
1273 +     *         the caller is not permitted to modify threads
1274 +     *         because it does not hold {@link
1275 +     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1276 +     */
1277 +    public ForkJoinPool(int parallelism) {
1278 +        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1279 +    }
1280  
1281      /**
1282 <     * Sets the target paralleism level of this pool.
1283 <     * @param parallelism the target parallelism
1282 >     * Creates a {@code ForkJoinPool} with the given parameters.
1283 >     *
1284 >     * @param parallelism the parallelism level. For default value,
1285 >     * use {@link java.lang.Runtime#availableProcessors}.
1286 >     * @param factory the factory for creating new threads. For default value,
1287 >     * use {@link #defaultForkJoinWorkerThreadFactory}.
1288 >     * @param handler the handler for internal worker threads that
1289 >     * terminate due to unrecoverable errors encountered while executing
1290 >     * tasks. For default value, use {@code null}.
1291 >     * @param asyncMode if true,
1292 >     * establishes local first-in-first-out scheduling mode for forked
1293 >     * tasks that are never joined. This mode may be more appropriate
1294 >     * than default locally stack-based mode in applications in which
1295 >     * worker threads only process event-style asynchronous tasks.
1296 >     * For default value, use {@code false}.
1297       * @throws IllegalArgumentException if parallelism less than or
1298 <     * equal to zero or greater than maximum size bounds.
1298 >     *         equal to zero, or greater than implementation limit
1299 >     * @throws NullPointerException if the factory is null
1300       * @throws SecurityException if a security manager exists and
1301       *         the caller is not permitted to modify threads
1302       *         because it does not hold {@link
1303 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1303 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1304       */
1305 <    public void setParallelism(int parallelism) {
1305 >    public ForkJoinPool(int parallelism,
1306 >                        ForkJoinWorkerThreadFactory factory,
1307 >                        Thread.UncaughtExceptionHandler handler,
1308 >                        boolean asyncMode) {
1309          checkPermission();
1310 <        if (parallelism <= 0 || parallelism > maxPoolSize)
1310 >        if (factory == null)
1311 >            throw new NullPointerException();
1312 >        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1313              throw new IllegalArgumentException();
1314 <        final ReentrantLock lock = this.workerLock;
1315 <        lock.lock();
1316 <        try {
1317 <            if (!isTerminating()) {
1318 <                int p = this.parallelism;
1319 <                this.parallelism = parallelism;
1320 <                if (parallelism > p)
1321 <                    createAndStartAddedWorkers();
1322 <                else
1323 <                    trimSpares();
1324 <            }
1325 <        } finally {
1326 <            lock.unlock();
1327 <        }
1328 <        signalIdleWorkers();
1314 >        this.parallelism = parallelism;
1315 >        this.factory = factory;
1316 >        this.ueh = handler;
1317 >        this.locallyFifo = asyncMode;
1318 >        int arraySize = initialArraySizeFor(parallelism);
1319 >        this.workers = new ForkJoinWorkerThread[arraySize];
1320 >        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1321 >        this.workerLock = new ReentrantLock();
1322 >        this.termination = new Phaser(1);
1323 >        this.poolNumber = poolNumberGenerator.incrementAndGet();
1324 >    }
1325 >
1326 >    /**
1327 >     * Returns initial power of two size for workers array.
1328 >     * @param pc the initial parallelism level
1329 >     */
1330 >    private static int initialArraySizeFor(int pc) {
1331 >        // If possible, initially allocate enough space for one spare
1332 >        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1333 >        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1334 >        size |= size >>> 1;
1335 >        size |= size >>> 2;
1336 >        size |= size >>> 4;
1337 >        size |= size >>> 8;
1338 >        return size + 1;
1339 >    }
1340 >
1341 >    // Execution methods
1342 >
1343 >    /**
1344 >     * Submits task and creates, starts, or resumes some workers if necessary
1345 >     */
1346 >    private <T> void doSubmit(ForkJoinTask<T> task) {
1347 >        submissionQueue.offer(task);
1348 >        int c; // try to increment event count -- CAS failure OK
1349 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1350 >        helpMaintainParallelism();
1351      }
1352  
1353      /**
1354 <     * Returns the targeted number of worker threads in this pool.
1354 >     * Performs the given task, returning its result upon completion.
1355       *
1356 <     * @return the targeted number of worker threads in this pool
1356 >     * @param task the task
1357 >     * @return the task's result
1358 >     * @throws NullPointerException if the task is null
1359 >     * @throws RejectedExecutionException if the task cannot be
1360 >     *         scheduled for execution
1361       */
1362 <    public int getParallelism() {
1363 <        return parallelism;
1362 >    public <T> T invoke(ForkJoinTask<T> task) {
1363 >        if (task == null)
1364 >            throw new NullPointerException();
1365 >        if (runState >= SHUTDOWN)
1366 >            throw new RejectedExecutionException();
1367 >        Thread t = Thread.currentThread();
1368 >        if ((t instanceof ForkJoinWorkerThread) &&
1369 >            ((ForkJoinWorkerThread)t).pool == this)
1370 >            return task.invoke();  // bypass submit if in same pool
1371 >        else {
1372 >            doSubmit(task);
1373 >            return task.join();
1374 >        }
1375      }
1376  
1377      /**
1378 <     * Returns the number of worker threads that have started but not
1379 <     * yet terminated.  This result returned by this method may differ
1380 <     * from <code>getParallelism</code> when threads are created to
1381 <     * maintain parallelism when others are cooperatively blocked.
1378 >     * Unless terminating, forks task if within an ongoing FJ
1379 >     * computation in the current pool, else submits as external task.
1380 >     */
1381 >    private <T> void forkOrSubmit(ForkJoinTask<T> task) {
1382 >        if (runState >= SHUTDOWN)
1383 >            throw new RejectedExecutionException();
1384 >        Thread t = Thread.currentThread();
1385 >        if ((t instanceof ForkJoinWorkerThread) &&
1386 >            ((ForkJoinWorkerThread)t).pool == this)
1387 >            task.fork();
1388 >        else
1389 >            doSubmit(task);
1390 >    }
1391 >
1392 >    /**
1393 >     * Arranges for (asynchronous) execution of the given task.
1394       *
1395 <     * @return the number of worker threads
1395 >     * @param task the task
1396 >     * @throws NullPointerException if the task is null
1397 >     * @throws RejectedExecutionException if the task cannot be
1398 >     *         scheduled for execution
1399       */
1400 <    public int getPoolSize() {
1401 <        return totalCountOf(workerCounts);
1400 >    public void execute(ForkJoinTask<?> task) {
1401 >        if (task == null)
1402 >            throw new NullPointerException();
1403 >        forkOrSubmit(task);
1404      }
1405  
1406 +    // AbstractExecutorService methods
1407 +
1408      /**
1409 <     * Returns the maximum number of threads allowed to exist in the
1410 <     * pool, even if there are insufficient unblocked running threads.
1411 <     * @return the maximum
1409 >     * @throws NullPointerException if the task is null
1410 >     * @throws RejectedExecutionException if the task cannot be
1411 >     *         scheduled for execution
1412       */
1413 <    public int getMaximumPoolSize() {
1414 <        return maxPoolSize;
1413 >    public void execute(Runnable task) {
1414 >        if (task == null)
1415 >            throw new NullPointerException();
1416 >        ForkJoinTask<?> job;
1417 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1418 >            job = (ForkJoinTask<?>) task;
1419 >        else
1420 >            job = ForkJoinTask.adapt(task, null);
1421 >        forkOrSubmit(job);
1422      }
1423  
1424      /**
1425 <     * Sets the maximum number of threads allowed to exist in the
1426 <     * pool, even if there are insufficient unblocked running threads.
1427 <     * Setting this value has no effect on current pool size. It
1428 <     * controls construction of new threads.
1429 <     * @throws IllegalArgumentException if negative or greater then
1430 <     * internal implementation limit.
1425 >     * Submits a ForkJoinTask for execution.
1426 >     *
1427 >     * @param task the task to submit
1428 >     * @return the task
1429 >     * @throws NullPointerException if the task is null
1430 >     * @throws RejectedExecutionException if the task cannot be
1431 >     *         scheduled for execution
1432       */
1433 <    public void setMaximumPoolSize(int newMax) {
1434 <        if (newMax < 0 || newMax > MAX_THREADS)
1435 <            throw new IllegalArgumentException();
1436 <        maxPoolSize = newMax;
1433 >    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1434 >        if (task == null)
1435 >            throw new NullPointerException();
1436 >        forkOrSubmit(task);
1437 >        return task;
1438      }
1439  
1440 +    /**
1441 +     * @throws NullPointerException if the task is null
1442 +     * @throws RejectedExecutionException if the task cannot be
1443 +     *         scheduled for execution
1444 +     */
1445 +    public <T> ForkJoinTask<T> submit(Callable<T> task) {
1446 +        if (task == null)
1447 +            throw new NullPointerException();
1448 +        ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1449 +        forkOrSubmit(job);
1450 +        return job;
1451 +    }
1452  
1453      /**
1454 <     * Returns true if this pool dynamically maintains its target
1455 <     * parallelism level. If false, new threads are added only to
1456 <     * avoid possible starvation.
798 <     * This setting is by default true;
799 <     * @return true if maintains parallelism
1454 >     * @throws NullPointerException if the task is null
1455 >     * @throws RejectedExecutionException if the task cannot be
1456 >     *         scheduled for execution
1457       */
1458 <    public boolean getMaintainsParallelism() {
1459 <        return maintainsParallelism;
1458 >    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1459 >        if (task == null)
1460 >            throw new NullPointerException();
1461 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1462 >        forkOrSubmit(job);
1463 >        return job;
1464      }
1465  
1466      /**
1467 <     * Sets whether this pool dynamically maintains its target
1468 <     * parallelism level. If false, new threads are added only to
1469 <     * avoid possible starvation.
809 <     * @param enable true to maintains parallelism
1467 >     * @throws NullPointerException if the task is null
1468 >     * @throws RejectedExecutionException if the task cannot be
1469 >     *         scheduled for execution
1470       */
1471 <    public void setMaintainsParallelism(boolean enable) {
1472 <        maintainsParallelism = enable;
1471 >    public ForkJoinTask<?> submit(Runnable task) {
1472 >        if (task == null)
1473 >            throw new NullPointerException();
1474 >        ForkJoinTask<?> job;
1475 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1476 >            job = (ForkJoinTask<?>) task;
1477 >        else
1478 >            job = ForkJoinTask.adapt(task, null);
1479 >        forkOrSubmit(job);
1480 >        return job;
1481      }
1482  
1483      /**
1484 <     * Establishes local first-in-first-out scheduling mode for forked
1485 <     * tasks that are never joined. This mode may be more appropriate
1486 <     * than default locally stack-based mode in applications in which
1487 <     * worker threads only process asynchronous tasks.  This method is
1488 <     * designed to be invoked only when pool is quiescent, and
1489 <     * typically only before any tasks are submitted. The effects of
1490 <     * invocations at ather times may be unpredictable.
1491 <     *
1492 <     * @param async if true, use locally FIFO scheduling
1493 <     * @return the previous mode.
1494 <     */
1495 <    public boolean setAsyncMode(boolean async) {
1496 <        boolean oldMode = locallyFifo;
1497 <        locallyFifo = async;
1498 <        ForkJoinWorkerThread[] ws = workers;
1499 <        if (ws != null) {
1500 <            for (int i = 0; i < ws.length; ++i) {
1501 <                ForkJoinWorkerThread t = ws[i];
1502 <                if (t != null)
1503 <                    t.setAsyncMode(async);
1504 <            }
1484 >     * @throws NullPointerException       {@inheritDoc}
1485 >     * @throws RejectedExecutionException {@inheritDoc}
1486 >     */
1487 >    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1488 >        ArrayList<ForkJoinTask<T>> forkJoinTasks =
1489 >            new ArrayList<ForkJoinTask<T>>(tasks.size());
1490 >        for (Callable<T> task : tasks)
1491 >            forkJoinTasks.add(ForkJoinTask.adapt(task));
1492 >        invoke(new InvokeAll<T>(forkJoinTasks));
1493 >
1494 >        @SuppressWarnings({"unchecked", "rawtypes"})
1495 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1496 >        return futures;
1497 >    }
1498 >
1499 >    static final class InvokeAll<T> extends RecursiveAction {
1500 >        final ArrayList<ForkJoinTask<T>> tasks;
1501 >        InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
1502 >        public void compute() {
1503 >            try { invokeAll(tasks); }
1504 >            catch (Exception ignore) {}
1505          }
1506 <        return oldMode;
1506 >        private static final long serialVersionUID = -7914297376763021607L;
1507 >    }
1508 >
1509 >    /**
1510 >     * Returns the factory used for constructing new workers.
1511 >     *
1512 >     * @return the factory used for constructing new workers
1513 >     */
1514 >    public ForkJoinWorkerThreadFactory getFactory() {
1515 >        return factory;
1516 >    }
1517 >
1518 >    /**
1519 >     * Returns the handler for internal worker threads that terminate
1520 >     * due to unrecoverable errors encountered while executing tasks.
1521 >     *
1522 >     * @return the handler, or {@code null} if none
1523 >     */
1524 >    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1525 >        return ueh;
1526 >    }
1527 >
1528 >    /**
1529 >     * Returns the targeted parallelism level of this pool.
1530 >     *
1531 >     * @return the targeted parallelism level of this pool
1532 >     */
1533 >    public int getParallelism() {
1534 >        return parallelism;
1535      }
1536  
1537      /**
1538 <     * Returns true if this pool uses local first-in-first-out
1538 >     * Returns the number of worker threads that have started but not
1539 >     * yet terminated.  The result returned by this method may differ
1540 >     * from {@link #getParallelism} when threads are created to
1541 >     * maintain parallelism when others are cooperatively blocked.
1542 >     *
1543 >     * @return the number of worker threads
1544 >     */
1545 >    public int getPoolSize() {
1546 >        return workerCounts >>> TOTAL_COUNT_SHIFT;
1547 >    }
1548 >
1549 >    /**
1550 >     * Returns {@code true} if this pool uses local first-in-first-out
1551       * scheduling mode for forked tasks that are never joined.
1552       *
1553 <     * @return true if this pool uses async mode.
1553 >     * @return {@code true} if this pool uses async mode
1554       */
1555      public boolean getAsyncMode() {
1556          return locallyFifo;
# Line 851 | Line 1559 | public class ForkJoinPool extends Abstra
1559      /**
1560       * Returns an estimate of the number of worker threads that are
1561       * not blocked waiting to join tasks or for other managed
1562 <     * synchronization.
1562 >     * synchronization. This method may overestimate the
1563 >     * number of running threads.
1564       *
1565       * @return the number of worker threads
1566       */
1567      public int getRunningThreadCount() {
1568 <        return runningCountOf(workerCounts);
1568 >        return workerCounts & RUNNING_COUNT_MASK;
1569      }
1570  
1571      /**
1572       * Returns an estimate of the number of threads that are currently
1573       * stealing or executing tasks. This method may overestimate the
1574       * number of active threads.
1575 <     * @return the number of active threads.
1575 >     *
1576 >     * @return the number of active threads
1577       */
1578      public int getActiveThreadCount() {
1579 <        return activeCountOf(runControl);
870 <    }
871 <
872 <    /**
873 <     * Returns an estimate of the number of threads that are currently
874 <     * idle waiting for tasks. This method may underestimate the
875 <     * number of idle threads.
876 <     * @return the number of idle threads.
877 <     */
878 <    final int getIdleThreadCount() {
879 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
880 <        return (c <= 0)? 0 : c;
1579 >        return runState & ACTIVE_COUNT_MASK;
1580      }
1581  
1582      /**
1583 <     * Returns true if all worker threads are currently idle. An idle
1584 <     * worker is one that cannot obtain a task to execute because none
1585 <     * are available to steal from other threads, and there are no
1586 <     * pending submissions to the pool. This method is conservative:
1587 <     * It might not return true immediately upon idleness of all
1588 <     * threads, but will eventually become true if threads remain
1589 <     * inactive.
1590 <     * @return true if all threads are currently idle
1583 >     * Returns {@code true} if all worker threads are currently idle.
1584 >     * An idle worker is one that cannot obtain a task to execute
1585 >     * because none are available to steal from other threads, and
1586 >     * there are no pending submissions to the pool. This method is
1587 >     * conservative; it might not return {@code true} immediately upon
1588 >     * idleness of all threads, but will eventually become true if
1589 >     * threads remain inactive.
1590 >     *
1591 >     * @return {@code true} if all threads are currently idle
1592       */
1593      public boolean isQuiescent() {
1594 <        return activeCountOf(runControl) == 0;
1594 >        return (runState & ACTIVE_COUNT_MASK) == 0;
1595      }
1596  
1597      /**
# Line 899 | Line 1599 | public class ForkJoinPool extends Abstra
1599       * one thread's work queue by another. The reported value
1600       * underestimates the actual total number of steals when the pool
1601       * is not quiescent. This value may be useful for monitoring and
1602 <     * tuning fork/join programs: In general, steal counts should be
1602 >     * tuning fork/join programs: in general, steal counts should be
1603       * high enough to keep threads busy, but low enough to avoid
1604       * overhead and contention across threads.
1605 <     * @return the number of steals.
1605 >     *
1606 >     * @return the number of steals
1607       */
1608      public long getStealCount() {
1609 <        return stealCount.get();
909 <    }
910 <
911 <    /**
912 <     * Accumulate steal count from a worker. Call only
913 <     * when worker known to be idle.
914 <     */
915 <    private void updateStealCount(ForkJoinWorkerThread w) {
916 <        int sc = w.getAndClearStealCount();
917 <        if (sc != 0)
918 <            stealCount.addAndGet(sc);
1609 >        return stealCount;
1610      }
1611  
1612      /**
# Line 925 | Line 1616 | public class ForkJoinPool extends Abstra
1616       * an approximation, obtained by iterating across all threads in
1617       * the pool. This method may be useful for tuning task
1618       * granularities.
1619 <     * @return the number of queued tasks.
1619 >     *
1620 >     * @return the number of queued tasks
1621       */
1622      public long getQueuedTaskCount() {
1623          long count = 0;
1624 <        ForkJoinWorkerThread[] ws = workers;
1625 <        if (ws != null) {
1626 <            for (int i = 0; i < ws.length; ++i) {
935 <                ForkJoinWorkerThread t = ws[i];
936 <                if (t != null)
937 <                    count += t.getQueueSize();
938 <            }
939 <        }
1624 >        for (ForkJoinWorkerThread w : workers)
1625 >            if (w != null)
1626 >                count += w.getQueueSize();
1627          return count;
1628      }
1629  
1630      /**
1631 <     * Returns an estimate of the number tasks submitted to this pool
1632 <     * that have not yet begun executing. This method takes time
1631 >     * Returns an estimate of the number of tasks submitted to this
1632 >     * pool that have not yet begun executing.  This method takes time
1633       * proportional to the number of submissions.
1634 <     * @return the number of queued submissions.
1634 >     *
1635 >     * @return the number of queued submissions
1636       */
1637      public int getQueuedSubmissionCount() {
1638          return submissionQueue.size();
1639      }
1640  
1641      /**
1642 <     * Returns true if there are any tasks submitted to this pool
1643 <     * that have not yet begun executing.
1644 <     * @return <code>true</code> if there are any queued submissions.
1642 >     * Returns {@code true} if there are any tasks submitted to this
1643 >     * pool that have not yet begun executing.
1644 >     *
1645 >     * @return {@code true} if there are any queued submissions
1646       */
1647      public boolean hasQueuedSubmissions() {
1648          return !submissionQueue.isEmpty();
# Line 963 | Line 1652 | public class ForkJoinPool extends Abstra
1652       * Removes and returns the next unexecuted submission if one is
1653       * available.  This method may be useful in extensions to this
1654       * class that re-assign work in systems with multiple pools.
1655 <     * @return the next submission, or null if none
1655 >     *
1656 >     * @return the next submission, or {@code null} if none
1657       */
1658      protected ForkJoinTask<?> pollSubmission() {
1659          return submissionQueue.poll();
# Line 973 | Line 1663 | public class ForkJoinPool extends Abstra
1663       * Removes all available unexecuted submitted and forked tasks
1664       * from scheduling queues and adds them to the given collection,
1665       * without altering their execution status. These may include
1666 <     * artifically generated or wrapped tasks. This method id designed
1667 <     * to be invoked only when the pool is known to be
1666 >     * artificially generated or wrapped tasks. This method is
1667 >     * designed to be invoked only when the pool is known to be
1668       * quiescent. Invocations at other times may not remove all
1669       * tasks. A failure encountered while attempting to add elements
1670 <     * to collection <tt>c</tt> may result in elements being in
1670 >     * to collection {@code c} may result in elements being in
1671       * neither, either or both collections when the associated
1672       * exception is thrown.  The behavior of this operation is
1673       * undefined if the specified collection is modified while the
1674       * operation is in progress.
1675 +     *
1676       * @param c the collection to transfer elements into
1677       * @return the number of elements transferred
1678       */
1679 <    protected int drainTasksTo(Collection<ForkJoinTask<?>> c) {
1680 <        int n = submissionQueue.drainTo(c);
1681 <        ForkJoinWorkerThread[] ws = workers;
1682 <        if (ws != null) {
1683 <            for (int i = 0; i < ws.length; ++i) {
1684 <                ForkJoinWorkerThread w = ws[i];
994 <                if (w != null)
995 <                    n += w.drainTasksTo(c);
996 <            }
997 <        }
998 <        return n;
1679 >    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1680 >        int count = submissionQueue.drainTo(c);
1681 >        for (ForkJoinWorkerThread w : workers)
1682 >            if (w != null)
1683 >                count += w.drainTasksTo(c);
1684 >        return count;
1685      }
1686  
1687      /**
# Line 1006 | Line 1692 | public class ForkJoinPool extends Abstra
1692       * @return a string identifying this pool, as well as its state
1693       */
1694      public String toString() {
1009        int ps = parallelism;
1010        int wc = workerCounts;
1011        int rc = runControl;
1695          long st = getStealCount();
1696          long qt = getQueuedTaskCount();
1697          long qs = getQueuedSubmissionCount();
1698 +        int wc = workerCounts;
1699 +        int tc = wc >>> TOTAL_COUNT_SHIFT;
1700 +        int rc = wc & RUNNING_COUNT_MASK;
1701 +        int pc = parallelism;
1702 +        int rs = runState;
1703 +        int ac = rs & ACTIVE_COUNT_MASK;
1704          return super.toString() +
1705 <            "[" + runStateToString(runStateOf(rc)) +
1706 <            ", parallelism = " + ps +
1707 <            ", size = " + totalCountOf(wc) +
1708 <            ", active = " + activeCountOf(rc) +
1709 <            ", running = " + runningCountOf(wc) +
1705 >            "[" + runLevelToString(rs) +
1706 >            ", parallelism = " + pc +
1707 >            ", size = " + tc +
1708 >            ", active = " + ac +
1709 >            ", running = " + rc +
1710              ", steals = " + st +
1711              ", tasks = " + qt +
1712              ", submissions = " + qs +
1713              "]";
1714      }
1715  
1716 <    private static String runStateToString(int rs) {
1717 <        switch(rs) {
1718 <        case RUNNING: return "Running";
1719 <        case SHUTDOWN: return "Shutting down";
1720 <        case TERMINATING: return "Terminating";
1032 <        case TERMINATED: return "Terminated";
1033 <        default: throw new Error("Unknown run state");
1034 <        }
1716 >    private static String runLevelToString(int s) {
1717 >        return ((s & TERMINATED) != 0 ? "Terminated" :
1718 >                ((s & TERMINATING) != 0 ? "Terminating" :
1719 >                 ((s & SHUTDOWN) != 0 ? "Shutting down" :
1720 >                  "Running")));
1721      }
1722  
1037    // lifecycle control
1038
1723      /**
1724       * Initiates an orderly shutdown in which previously submitted
1725       * tasks are executed, but no new tasks will be accepted.
1726       * Invocation has no additional effect if already shut down.
1727       * Tasks that are in the process of being submitted concurrently
1728       * during the course of this method may or may not be rejected.
1729 +     *
1730       * @throws SecurityException if a security manager exists and
1731       *         the caller is not permitted to modify threads
1732       *         because it does not hold {@link
1733 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1733 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1734       */
1735      public void shutdown() {
1736          checkPermission();
1737 <        transitionRunStateTo(SHUTDOWN);
1738 <        if (canTerminateOnShutdown(runControl))
1054 <            terminateOnShutdown();
1737 >        advanceRunLevel(SHUTDOWN);
1738 >        tryTerminate(false);
1739      }
1740  
1741      /**
1742 <     * Attempts to stop all actively executing tasks, and cancels all
1743 <     * waiting tasks.  Tasks that are in the process of being
1744 <     * submitted or executed concurrently during the course of this
1745 <     * method may or may not be rejected. Unlike some other executors,
1746 <     * this method cancels rather than collects non-executed tasks
1747 <     * upon termination, so always returns an empty list. However, you
1748 <     * can use method <code>drainTasksTo</code> before invoking this
1749 <     * method to transfer unexecuted tasks to another collection.
1742 >     * Attempts to cancel and/or stop all tasks, and reject all
1743 >     * subsequently submitted tasks.  Tasks that are in the process of
1744 >     * being submitted or executed concurrently during the course of
1745 >     * this method may or may not be rejected. This method cancels
1746 >     * both existing and unexecuted tasks, in order to permit
1747 >     * termination in the presence of task dependencies. So the method
1748 >     * always returns an empty list (unlike the case for some other
1749 >     * Executors).
1750 >     *
1751       * @return an empty list
1752       * @throws SecurityException if a security manager exists and
1753       *         the caller is not permitted to modify threads
1754       *         because it does not hold {@link
1755 <     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
1755 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1756       */
1757      public List<Runnable> shutdownNow() {
1758          checkPermission();
1759 <        terminate();
1759 >        tryTerminate(true);
1760          return Collections.emptyList();
1761      }
1762  
1763      /**
1764 <     * Returns <code>true</code> if all tasks have completed following shut down.
1764 >     * Returns {@code true} if all tasks have completed following shut down.
1765       *
1766 <     * @return <code>true</code> if all tasks have completed following shut down
1766 >     * @return {@code true} if all tasks have completed following shut down
1767       */
1768      public boolean isTerminated() {
1769 <        return runStateOf(runControl) == TERMINATED;
1769 >        return runState >= TERMINATED;
1770      }
1771  
1772      /**
1773 <     * Returns <code>true</code> if the process of termination has
1774 <     * commenced but possibly not yet completed.
1773 >     * Returns {@code true} if the process of termination has
1774 >     * commenced but not yet completed.  This method may be useful for
1775 >     * debugging. A return of {@code true} reported a sufficient
1776 >     * period after shutdown may indicate that submitted tasks have
1777 >     * ignored or suppressed interruption, or are waiting for IO,
1778 >     * causing this executor not to properly terminate. (See the
1779 >     * advisory notes for class {@link ForkJoinTask} stating that
1780 >     * tasks should not normally entail blocking operations.  But if
1781 >     * they do, they must abort them on interrupt.)
1782       *
1783 <     * @return <code>true</code> if terminating
1783 >     * @return {@code true} if terminating but not yet terminated
1784       */
1785      public boolean isTerminating() {
1786 <        return runStateOf(runControl) >= TERMINATING;
1786 >        return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
1787      }
1788  
1789      /**
1790 <     * Returns <code>true</code> if this pool has been shut down.
1790 >     * Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1791 >     */
1792 >    final boolean isAtLeastTerminating() {
1793 >        return runState >= TERMINATING;
1794 >    }
1795 >
1796 >    /**
1797 >     * Returns {@code true} if this pool has been shut down.
1798       *
1799 <     * @return <code>true</code> if this pool has been shut down
1799 >     * @return {@code true} if this pool has been shut down
1800       */
1801      public boolean isShutdown() {
1802 <        return runStateOf(runControl) >= SHUTDOWN;
1802 >        return runState >= SHUTDOWN;
1803      }
1804  
1805      /**
# Line 1110 | Line 1809 | public class ForkJoinPool extends Abstra
1809       *
1810       * @param timeout the maximum time to wait
1811       * @param unit the time unit of the timeout argument
1812 <     * @return <code>true</code> if this executor terminated and
1813 <     *         <code>false</code> if the timeout elapsed before termination
1812 >     * @return {@code true} if this executor terminated and
1813 >     *         {@code false} if the timeout elapsed before termination
1814       * @throws InterruptedException if interrupted while waiting
1815       */
1816      public boolean awaitTermination(long timeout, TimeUnit unit)
1817          throws InterruptedException {
1119        long nanos = unit.toNanos(timeout);
1120        final ReentrantLock lock = this.workerLock;
1121        lock.lock();
1122        try {
1123            for (;;) {
1124                if (isTerminated())
1125                    return true;
1126                if (nanos <= 0)
1127                    return false;
1128                nanos = termination.awaitNanos(nanos);
1129            }
1130        } finally {
1131            lock.unlock();
1132        }
1133    }
1134
1135    // Shutdown and termination support
1136
1137    /**
1138     * Callback from terminating worker. Null out the corresponding
1139     * workers slot, and if terminating, try to terminate, else try to
1140     * shrink workers array.
1141     * @param w the worker
1142     */
1143    final void workerTerminated(ForkJoinWorkerThread w) {
1144        updateStealCount(w);
1145        updateWorkerCount(-1);
1146        final ReentrantLock lock = this.workerLock;
1147        lock.lock();
1148        try {
1149            ForkJoinWorkerThread[] ws = workers;
1150            if (ws != null) {
1151                int idx = w.poolIndex;
1152                if (idx >= 0 && idx < ws.length && ws[idx] == w)
1153                    ws[idx] = null;
1154                if (totalCountOf(workerCounts) == 0) {
1155                    terminate(); // no-op if already terminating
1156                    transitionRunStateTo(TERMINATED);
1157                    termination.signalAll();
1158                }
1159                else if (!isTerminating()) {
1160                    tryShrinkWorkerArray();
1161                    tryResumeSpare(true); // allow replacement
1162                }
1163            }
1164        } finally {
1165            lock.unlock();
1166        }
1167        signalIdleWorkers();
1168    }
1169
1170    /**
1171     * Initiate termination.
1172     */
1173    private void terminate() {
1174        if (transitionRunStateTo(TERMINATING)) {
1175            stopAllWorkers();
1176            resumeAllSpares();
1177            signalIdleWorkers();
1178            cancelQueuedSubmissions();
1179            cancelQueuedWorkerTasks();
1180            interruptUnterminatedWorkers();
1181            signalIdleWorkers(); // resignal after interrupt
1182        }
1183    }
1184
1185    /**
1186     * Possibly terminate when on shutdown state
1187     */
1188    private void terminateOnShutdown() {
1189        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1190            terminate();
1191    }
1192
1193    /**
1194     * Clear out and cancel submissions
1195     */
1196    private void cancelQueuedSubmissions() {
1197        ForkJoinTask<?> task;
1198        while ((task = pollSubmission()) != null)
1199            task.cancel(false);
1200    }
1201
1202    /**
1203     * Clean out worker queues.
1204     */
1205    private void cancelQueuedWorkerTasks() {
1206        final ReentrantLock lock = this.workerLock;
1207        lock.lock();
1818          try {
1819 <            ForkJoinWorkerThread[] ws = workers;
1820 <            if (ws != null) {
1211 <                for (int i = 0; i < ws.length; ++i) {
1212 <                    ForkJoinWorkerThread t = ws[i];
1213 <                    if (t != null)
1214 <                        t.cancelTasks();
1215 <                }
1216 <            }
1217 <        } finally {
1218 <            lock.unlock();
1219 <        }
1220 <    }
1221 <
1222 <    /**
1223 <     * Set each worker's status to terminating. Requires lock to avoid
1224 <     * conflicts with add/remove
1225 <     */
1226 <    private void stopAllWorkers() {
1227 <        final ReentrantLock lock = this.workerLock;
1228 <        lock.lock();
1229 <        try {
1230 <            ForkJoinWorkerThread[] ws = workers;
1231 <            if (ws != null) {
1232 <                for (int i = 0; i < ws.length; ++i) {
1233 <                    ForkJoinWorkerThread t = ws[i];
1234 <                    if (t != null)
1235 <                        t.shutdownNow();
1236 <                }
1237 <            }
1238 <        } finally {
1239 <            lock.unlock();
1240 <        }
1241 <    }
1242 <
1243 <    /**
1244 <     * Interrupt all unterminated workers.  This is not required for
1245 <     * sake of internal control, but may help unstick user code during
1246 <     * shutdown.
1247 <     */
1248 <    private void interruptUnterminatedWorkers() {
1249 <        final ReentrantLock lock = this.workerLock;
1250 <        lock.lock();
1251 <        try {
1252 <            ForkJoinWorkerThread[] ws = workers;
1253 <            if (ws != null) {
1254 <                for (int i = 0; i < ws.length; ++i) {
1255 <                    ForkJoinWorkerThread t = ws[i];
1256 <                    if (t != null && !t.isTerminated()) {
1257 <                        try {
1258 <                            t.interrupt();
1259 <                        } catch (SecurityException ignore) {
1260 <                        }
1261 <                    }
1262 <                }
1263 <            }
1264 <        } finally {
1265 <            lock.unlock();
1266 <        }
1267 <    }
1268 <
1269 <
1270 <    /*
1271 <     * Nodes for event barrier to manage idle threads.  Queue nodes
1272 <     * are basic Treiber stack nodes, also used for spare stack.
1273 <     *
1274 <     * The event barrier has an event count and a wait queue (actually
1275 <     * a Treiber stack).  Workers are enabled to look for work when
1276 <     * the eventCount is incremented. If they fail to find work, they
1277 <     * may wait for next count. Upon release, threads help others wake
1278 <     * up.
1279 <     *
1280 <     * Synchronization events occur only in enough contexts to
1281 <     * maintain overall liveness:
1282 <     *
1283 <     *   - Submission of a new task to the pool
1284 <     *   - Resizes or other changes to the workers array
1285 <     *   - pool termination
1286 <     *   - A worker pushing a task on an empty queue
1287 <     *
1288 <     * The case of pushing a task occurs often enough, and is heavy
1289 <     * enough compared to simple stack pushes, to require special
1290 <     * handling: Method signalWork returns without advancing count if
1291 <     * the queue appears to be empty.  This would ordinarily result in
1292 <     * races causing some queued waiters not to be woken up. To avoid
1293 <     * this, the first worker enqueued in method sync (see
1294 <     * syncIsReleasable) rescans for tasks after being enqueued, and
1295 <     * helps signal if any are found. This works well because the
1296 <     * worker has nothing better to do, and so might as well help
1297 <     * alleviate the overhead and contention on the threads actually
1298 <     * doing work.  Also, since event counts increments on task
1299 <     * availability exist to maintain liveness (rather than to force
1300 <     * refreshes etc), it is OK for callers to exit early if
1301 <     * contending with another signaller.
1302 <     */
1303 <    static final class WaitQueueNode {
1304 <        WaitQueueNode next; // only written before enqueued
1305 <        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1306 <        final long count; // unused for spare stack
1307 <
1308 <        WaitQueueNode(long c, ForkJoinWorkerThread w) {
1309 <            count = c;
1310 <            thread = w;
1311 <        }
1312 <
1313 <        /**
1314 <         * Wake up waiter, returning false if known to already
1315 <         */
1316 <        boolean signal() {
1317 <            ForkJoinWorkerThread t = thread;
1318 <            if (t == null)
1319 <                return false;
1320 <            thread = null;
1321 <            LockSupport.unpark(t);
1322 <            return true;
1323 <        }
1324 <
1325 <        /**
1326 <         * Await release on sync
1327 <         */
1328 <        void awaitSyncRelease(ForkJoinPool p) {
1329 <            while (thread != null && !p.syncIsReleasable(this))
1330 <                LockSupport.park(this);
1331 <        }
1332 <
1333 <        /**
1334 <         * Await resumption as spare
1335 <         */
1336 <        void awaitSpareRelease() {
1337 <            while (thread != null) {
1338 <                if (!Thread.interrupted())
1339 <                    LockSupport.park(this);
1340 <            }
1341 <        }
1342 <    }
1343 <
1344 <    /**
1345 <     * Ensures that no thread is waiting for count to advance from the
1346 <     * current value of eventCount read on entry to this method, by
1347 <     * releasing waiting threads if necessary.
1348 <     * @return the count
1349 <     */
1350 <    final long ensureSync() {
1351 <        long c = eventCount;
1352 <        WaitQueueNode q;
1353 <        while ((q = syncStack) != null && q.count < c) {
1354 <            if (casBarrierStack(q, null)) {
1355 <                do {
1356 <                    q.signal();
1357 <                } while ((q = q.next) != null);
1358 <                break;
1359 <            }
1360 <        }
1361 <        return c;
1362 <    }
1363 <
1364 <    /**
1365 <     * Increments event count and releases waiting threads.
1366 <     */
1367 <    private void signalIdleWorkers() {
1368 <        long c;
1369 <        do;while (!casEventCount(c = eventCount, c+1));
1370 <        ensureSync();
1371 <    }
1372 <
1373 <    /**
1374 <     * Signal threads waiting to poll a task. Because method sync
1375 <     * rechecks availability, it is OK to only proceed if queue
1376 <     * appears to be non-empty, and OK to skip under contention to
1377 <     * increment count (since some other thread succeeded).
1378 <     */
1379 <    final void signalWork() {
1380 <        long c;
1381 <        WaitQueueNode q;
1382 <        if (syncStack != null &&
1383 <            casEventCount(c = eventCount, c+1) &&
1384 <            (((q = syncStack) != null && q.count <= c) &&
1385 <             (!casBarrierStack(q, q.next) || !q.signal())))
1386 <            ensureSync();
1387 <    }
1388 <
1389 <    /**
1390 <     * Waits until event count advances from last value held by
1391 <     * caller, or if excess threads, caller is resumed as spare, or
1392 <     * caller or pool is terminating. Updates caller's event on exit.
1393 <     * @param w the calling worker thread
1394 <     */
1395 <    final void sync(ForkJoinWorkerThread w) {
1396 <        updateStealCount(w); // Transfer w's count while it is idle
1397 <
1398 <        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
1399 <            long prev = w.lastEventCount;
1400 <            WaitQueueNode node = null;
1401 <            WaitQueueNode h;
1402 <            while (eventCount == prev &&
1403 <                   ((h = syncStack) == null || h.count == prev)) {
1404 <                if (node == null)
1405 <                    node = new WaitQueueNode(prev, w);
1406 <                if (casBarrierStack(node.next = h, node)) {
1407 <                    node.awaitSyncRelease(this);
1408 <                    break;
1409 <                }
1410 <            }
1411 <            long ec = ensureSync();
1412 <            if (ec != prev) {
1413 <                w.lastEventCount = ec;
1414 <                break;
1415 <            }
1416 <        }
1417 <    }
1418 <
1419 <    /**
1420 <     * Returns true if worker waiting on sync can proceed:
1421 <     *  - on signal (thread == null)
1422 <     *  - on event count advance (winning race to notify vs signaller)
1423 <     *  - on Interrupt
1424 <     *  - if the first queued node, we find work available
1425 <     * If node was not signalled and event count not advanced on exit,
1426 <     * then we also help advance event count.
1427 <     * @return true if node can be released
1428 <     */
1429 <    final boolean syncIsReleasable(WaitQueueNode node) {
1430 <        long prev = node.count;
1431 <        if (!Thread.interrupted() && node.thread != null &&
1432 <            (node.next != null ||
1433 <             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
1434 <            eventCount == prev)
1819 >            termination.awaitAdvanceInterruptibly(0, timeout, unit);
1820 >        } catch (TimeoutException ex) {
1821              return false;
1436        if (node.thread != null) {
1437            node.thread = null;
1438            long ec = eventCount;
1439            if (prev <= ec) // help signal
1440                casEventCount(ec, ec+1);
1822          }
1823          return true;
1824      }
1825  
1826      /**
1446     * Returns true if a new sync event occurred since last call to
1447     * sync or this method, if so, updating caller's count.
1448     */
1449    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1450        long lc = w.lastEventCount;
1451        long ec = ensureSync();
1452        if (ec == lc)
1453            return false;
1454        w.lastEventCount = ec;
1455        return true;
1456    }
1457
1458    //  Parallelism maintenance
1459
1460    /**
1461     * Decrement running count; if too low, add spare.
1462     *
1463     * Conceptually, all we need to do here is add or resume a
1464     * spare thread when one is about to block (and remove or
1465     * suspend it later when unblocked -- see suspendIfSpare).
1466     * However, implementing this idea requires coping with
1467     * several problems: We have imperfect information about the
1468     * states of threads. Some count updates can and usually do
1469     * lag run state changes, despite arrangements to keep them
1470     * accurate (for example, when possible, updating counts
1471     * before signalling or resuming), especially when running on
1472     * dynamic JVMs that don't optimize the infrequent paths that
1473     * update counts. Generating too many threads can make these
1474     * problems become worse, because excess threads are more
1475     * likely to be context-switched with others, slowing them all
1476     * down, especially if there is no work available, so all are
1477     * busy scanning or idling.  Also, excess spare threads can
1478     * only be suspended or removed when they are idle, not
1479     * immediately when they aren't needed. So adding threads will
1480     * raise parallelism level for longer than necessary.  Also,
1481     * FJ applications often enounter highly transient peaks when
1482     * many threads are blocked joining, but for less time than it
1483     * takes to create or resume spares.
1484     *
1485     * @param joinMe if non-null, return early if done
1486     * @param maintainParallelism if true, try to stay within
1487     * target counts, else create only to avoid starvation
1488     * @return true if joinMe known to be done
1489     */
1490    final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
1491        maintainParallelism &= maintainsParallelism; // overrride
1492        boolean dec = false;  // true when running count decremented
1493        while (spareStack == null || !tryResumeSpare(dec)) {
1494            int counts = workerCounts;
1495            if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
1496                if (!needSpare(counts, maintainParallelism))
1497                    break;
1498                if (joinMe.status < 0)
1499                    return true;
1500                if (tryAddSpare(counts))
1501                    break;
1502            }
1503        }
1504        return false;
1505    }
1506
1507    /**
1508     * Same idea as preJoin
1509     */
1510    final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
1511        maintainParallelism &= maintainsParallelism;
1512        boolean dec = false;
1513        while (spareStack == null || !tryResumeSpare(dec)) {
1514            int counts = workerCounts;
1515            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1516                if (!needSpare(counts, maintainParallelism))
1517                    break;
1518                if (blocker.isReleasable())
1519                    return true;
1520                if (tryAddSpare(counts))
1521                    break;
1522            }
1523        }
1524        return false;
1525    }
1526
1527    /**
1528     * Returns true if a spare thread appears to be needed.  If
1529     * maintaining parallelism, returns true when the deficit in
1530     * running threads is more than the surplus of total threads, and
1531     * there is apparently some work to do.  This self-limiting rule
1532     * means that the more threads that have already been added, the
1533     * less parallelism we will tolerate before adding another.
1534     * @param counts current worker counts
1535     * @param maintainParallelism try to maintain parallelism
1536     */
1537    private boolean needSpare(int counts, boolean maintainParallelism) {
1538        int ps = parallelism;
1539        int rc = runningCountOf(counts);
1540        int tc = totalCountOf(counts);
1541        int runningDeficit = ps - rc;
1542        int totalSurplus = tc - ps;
1543        return (tc < maxPoolSize &&
1544                (rc == 0 || totalSurplus < 0 ||
1545                 (maintainParallelism &&
1546                  runningDeficit > totalSurplus &&
1547                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
1548    }
1549
1550    /**
1551     * Add a spare worker if lock available and no more than the
1552     * expected numbers of threads exist
1553     * @return true if successful
1554     */
1555    private boolean tryAddSpare(int expectedCounts) {
1556        final ReentrantLock lock = this.workerLock;
1557        int expectedRunning = runningCountOf(expectedCounts);
1558        int expectedTotal = totalCountOf(expectedCounts);
1559        boolean success = false;
1560        boolean locked = false;
1561        // confirm counts while locking; CAS after obtaining lock
1562        try {
1563            for (;;) {
1564                int s = workerCounts;
1565                int tc = totalCountOf(s);
1566                int rc = runningCountOf(s);
1567                if (rc > expectedRunning || tc > expectedTotal)
1568                    break;
1569                if (!locked && !(locked = lock.tryLock()))
1570                    break;
1571                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1572                    createAndStartSpare(tc);
1573                    success = true;
1574                    break;
1575                }
1576            }
1577        } finally {
1578            if (locked)
1579                lock.unlock();
1580        }
1581        return success;
1582    }
1583
1584    /**
1585     * Add the kth spare worker. On entry, pool coounts are already
1586     * adjusted to reflect addition.
1587     */
1588    private void createAndStartSpare(int k) {
1589        ForkJoinWorkerThread w = null;
1590        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1591        int len = ws.length;
1592        // Probably, we can place at slot k. If not, find empty slot
1593        if (k < len && ws[k] != null) {
1594            for (k = 0; k < len && ws[k] != null; ++k)
1595                ;
1596        }
1597        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
1598            ws[k] = w;
1599            w.start();
1600        }
1601        else
1602            updateWorkerCount(-1); // adjust on failure
1603        signalIdleWorkers();
1604    }
1605
1606    /**
1607     * Suspend calling thread w if there are excess threads.  Called
1608     * only from sync.  Spares are enqueued in a Treiber stack
1609     * using the same WaitQueueNodes as barriers.  They are resumed
1610     * mainly in preJoin, but are also woken on pool events that
1611     * require all threads to check run state.
1612     * @param w the caller
1613     */
1614    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1615        WaitQueueNode node = null;
1616        int s;
1617        while (parallelism < runningCountOf(s = workerCounts)) {
1618            if (node == null)
1619                node = new WaitQueueNode(0, w);
1620            if (casWorkerCounts(s, s-1)) { // representation-dependent
1621                // push onto stack
1622                do;while (!casSpareStack(node.next = spareStack, node));
1623                // block until released by resumeSpare
1624                node.awaitSpareRelease();
1625                return true;
1626            }
1627        }
1628        return false;
1629    }
1630
1631    /**
1632     * Try to pop and resume a spare thread.
1633     * @param updateCount if true, increment running count on success
1634     * @return true if successful
1635     */
1636    private boolean tryResumeSpare(boolean updateCount) {
1637        WaitQueueNode q;
1638        while ((q = spareStack) != null) {
1639            if (casSpareStack(q, q.next)) {
1640                if (updateCount)
1641                    updateRunningCount(1);
1642                q.signal();
1643                return true;
1644            }
1645        }
1646        return false;
1647    }
1648
1649    /**
1650     * Pop and resume all spare threads. Same idea as ensureSync.
1651     * @return true if any spares released
1652     */
1653    private boolean resumeAllSpares() {
1654        WaitQueueNode q;
1655        while ( (q = spareStack) != null) {
1656            if (casSpareStack(q, null)) {
1657                do {
1658                    updateRunningCount(1);
1659                    q.signal();
1660                } while ((q = q.next) != null);
1661                return true;
1662            }
1663        }
1664        return false;
1665    }
1666
1667    /**
1668     * Pop and shutdown excessive spare threads. Call only while
1669     * holding lock. This is not guaranteed to eliminate all excess
1670     * threads, only those suspended as spares, which are the ones
1671     * unlikely to be needed in the future.
1672     */
1673    private void trimSpares() {
1674        int surplus = totalCountOf(workerCounts) - parallelism;
1675        WaitQueueNode q;
1676        while (surplus > 0 && (q = spareStack) != null) {
1677            if (casSpareStack(q, null)) {
1678                do {
1679                    updateRunningCount(1);
1680                    ForkJoinWorkerThread w = q.thread;
1681                    if (w != null && surplus > 0 &&
1682                        runningCountOf(workerCounts) > 0 && w.shutdown())
1683                        --surplus;
1684                    q.signal();
1685                } while ((q = q.next) != null);
1686            }
1687        }
1688    }
1689
1690    /**
1827       * Interface for extending managed parallelism for tasks running
1828 <     * in ForkJoinPools. A ManagedBlocker provides two methods.
1829 <     * Method <code>isReleasable</code> must return true if blocking is not
1830 <     * necessary. Method <code>block</code> blocks the current thread
1831 <     * if necessary (perhaps internally invoking isReleasable before
1832 <     * actually blocking.).
1828 >     * in {@link ForkJoinPool}s.
1829 >     *
1830 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1831 >     * {@code isReleasable} must return {@code true} if blocking is
1832 >     * not necessary. Method {@code block} blocks the current thread
1833 >     * if necessary (perhaps internally invoking {@code isReleasable}
1834 >     * before actually blocking). The unusual methods in this API
1835 >     * accommodate synchronizers that may, but don't usually, block
1836 >     * for long periods. Similarly, they allow more efficient internal
1837 >     * handling of cases in which additional workers may be, but
1838 >     * usually are not, needed to ensure sufficient parallelism.
1839 >     * Toward this end, implementations of method {@code isReleasable}
1840 >     * must be amenable to repeated invocation.
1841 >     *
1842       * <p>For example, here is a ManagedBlocker based on a
1843       * ReentrantLock:
1844 <     * <pre>
1845 <     *   class ManagedLocker implements ManagedBlocker {
1846 <     *     final ReentrantLock lock;
1847 <     *     boolean hasLock = false;
1848 <     *     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
1849 <     *     public boolean block() {
1850 <     *        if (!hasLock)
1851 <     *           lock.lock();
1852 <     *        return true;
1853 <     *     }
1854 <     *     public boolean isReleasable() {
1855 <     *        return hasLock || (hasLock = lock.tryLock());
1856 <     *     }
1844 >     *  <pre> {@code
1845 >     * class ManagedLocker implements ManagedBlocker {
1846 >     *   final ReentrantLock lock;
1847 >     *   boolean hasLock = false;
1848 >     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
1849 >     *   public boolean block() {
1850 >     *     if (!hasLock)
1851 >     *       lock.lock();
1852 >     *     return true;
1853 >     *   }
1854 >     *   public boolean isReleasable() {
1855 >     *     return hasLock || (hasLock = lock.tryLock());
1856 >     *   }
1857 >     * }}</pre>
1858 >     *
1859 >     * <p>Here is a class that possibly blocks waiting for an
1860 >     * item on a given queue:
1861 >     *  <pre> {@code
1862 >     * class QueueTaker<E> implements ManagedBlocker {
1863 >     *   final BlockingQueue<E> queue;
1864 >     *   volatile E item = null;
1865 >     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1866 >     *   public boolean block() throws InterruptedException {
1867 >     *     if (item == null)
1868 >     *       item = queue.take();
1869 >     *     return true;
1870 >     *   }
1871 >     *   public boolean isReleasable() {
1872 >     *     return item != null || (item = queue.poll()) != null;
1873 >     *   }
1874 >     *   public E getItem() { // call after pool.managedBlock completes
1875 >     *     return item;
1876       *   }
1877 <     * </pre>
1877 >     * }}</pre>
1878       */
1879      public static interface ManagedBlocker {
1880          /**
1881           * Possibly blocks the current thread, for example waiting for
1882           * a lock or condition.
1883 <         * @return true if no additional blocking is necessary (i.e.,
1884 <         * if isReleasable would return true).
1883 >         *
1884 >         * @return {@code true} if no additional blocking is necessary
1885 >         * (i.e., if isReleasable would return true)
1886           * @throws InterruptedException if interrupted while waiting
1887 <         * (the method is not required to do so, but is allowe to).
1887 >         * (the method is not required to do so, but is allowed to)
1888           */
1889          boolean block() throws InterruptedException;
1890  
1891          /**
1892 <         * Returns true if blocking is unnecessary.
1892 >         * Returns {@code true} if blocking is unnecessary.
1893           */
1894          boolean isReleasable();
1895      }
1896  
1897      /**
1898       * Blocks in accord with the given blocker.  If the current thread
1899 <     * is a ForkJoinWorkerThread, this method possibly arranges for a
1900 <     * spare thread to be activated if necessary to ensure parallelism
1901 <     * while the current thread is blocked.  If
1902 <     * <code>maintainParallelism</code> is true and the pool supports
1903 <     * it ({@link #getMaintainsParallelism}), this method attempts to
1904 <     * maintain the pool's nominal parallelism. Otherwise if activates
1905 <     * a thread only if necessary to avoid complete starvation. This
1906 <     * option may be preferable when blockages use timeouts, or are
1907 <     * almost always brief.
1908 <     *
1909 <     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
1910 <     * equivalent to
1911 <     * <pre>
1912 <     *   while (!blocker.isReleasable())
1748 <     *      if (blocker.block())
1749 <     *         return;
1750 <     * </pre>
1751 <     * If the caller is a ForkJoinTask, then the pool may first
1752 <     * be expanded to ensure parallelism, and later adjusted.
1899 >     * is a {@link ForkJoinWorkerThread}, this method possibly
1900 >     * arranges for a spare thread to be activated if necessary to
1901 >     * ensure sufficient parallelism while the current thread is blocked.
1902 >     *
1903 >     * <p>If the caller is not a {@link ForkJoinTask}, this method is
1904 >     * behaviorally equivalent to
1905 >     *  <pre> {@code
1906 >     * while (!blocker.isReleasable())
1907 >     *   if (blocker.block())
1908 >     *     return;
1909 >     * }</pre>
1910 >     *
1911 >     * If the caller is a {@code ForkJoinTask}, then the pool may
1912 >     * first be expanded to ensure parallelism, and later adjusted.
1913       *
1914       * @param blocker the blocker
1915 <     * @param maintainParallelism if true and supported by this pool,
1756 <     * attempt to maintain the pool's nominal parallelism; otherwise
1757 <     * activate a thread only if necessary to avoid complete
1758 <     * starvation.
1759 <     * @throws InterruptedException if blocker.block did so.
1915 >     * @throws InterruptedException if blocker.block did so
1916       */
1917 <    public static void managedBlock(ManagedBlocker blocker,
1762 <                                    boolean maintainParallelism)
1917 >    public static void managedBlock(ManagedBlocker blocker)
1918          throws InterruptedException {
1919          Thread t = Thread.currentThread();
1920 <        ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
1921 <                             ((ForkJoinWorkerThread)t).pool : null);
1922 <        if (!blocker.isReleasable()) {
1923 <            try {
1924 <                if (pool == null ||
1925 <                    !pool.preBlock(blocker, maintainParallelism))
1771 <                    awaitBlocker(blocker);
1772 <            } finally {
1773 <                if (pool != null)
1774 <                    pool.updateRunningCount(1);
1775 <            }
1920 >        if (t instanceof ForkJoinWorkerThread) {
1921 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1922 >            w.pool.awaitBlocker(blocker);
1923 >        }
1924 >        else {
1925 >            do {} while (!blocker.isReleasable() && !blocker.block());
1926          }
1927      }
1928  
1929 <    private static void awaitBlocker(ManagedBlocker blocker)
1930 <        throws InterruptedException {
1931 <        do;while (!blocker.isReleasable() && !blocker.block());
1782 <    }
1783 <
1784 <    // AbstractExecutorService overrides
1929 >    // AbstractExecutorService overrides.  These rely on undocumented
1930 >    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
1931 >    // implement RunnableFuture.
1932  
1933      protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
1934 <        return new AdaptedRunnable(runnable, value);
1934 >        return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
1935      }
1936  
1937      protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
1938 <        return new AdaptedCallable(callable);
1938 >        return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
1939      }
1940  
1941 +    // Unsafe mechanics
1942 +
1943 +    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1944 +    private static final long workerCountsOffset =
1945 +        objectFieldOffset("workerCounts", ForkJoinPool.class);
1946 +    private static final long runStateOffset =
1947 +        objectFieldOffset("runState", ForkJoinPool.class);
1948 +    private static final long eventCountOffset =
1949 +        objectFieldOffset("eventCount", ForkJoinPool.class);
1950 +    private static final long eventWaitersOffset =
1951 +        objectFieldOffset("eventWaiters", ForkJoinPool.class);
1952 +    private static final long stealCountOffset =
1953 +        objectFieldOffset("stealCount", ForkJoinPool.class);
1954 +    private static final long spareWaitersOffset =
1955 +        objectFieldOffset("spareWaiters", ForkJoinPool.class);
1956 +
1957 +    private static long objectFieldOffset(String field, Class<?> klazz) {
1958 +        try {
1959 +            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1960 +        } catch (NoSuchFieldException e) {
1961 +            // Convert Exception to corresponding Error
1962 +            NoSuchFieldError error = new NoSuchFieldError(field);
1963 +            error.initCause(e);
1964 +            throw error;
1965 +        }
1966 +    }
1967  
1968 <    // Temporary Unsafe mechanics for preliminary release
1969 <    private static Unsafe getUnsafe() throws Throwable {
1968 >    /**
1969 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1970 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
1971 >     * into a jdk.
1972 >     *
1973 >     * @return a sun.misc.Unsafe
1974 >     */
1975 >    private static sun.misc.Unsafe getUnsafe() {
1976          try {
1977 <            return Unsafe.getUnsafe();
1977 >            return sun.misc.Unsafe.getUnsafe();
1978          } catch (SecurityException se) {
1979              try {
1980                  return java.security.AccessController.doPrivileged
1981 <                    (new java.security.PrivilegedExceptionAction<Unsafe>() {
1982 <                        public Unsafe run() throws Exception {
1983 <                            return getUnsafePrivileged();
1981 >                    (new java.security
1982 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1983 >                        public sun.misc.Unsafe run() throws Exception {
1984 >                            java.lang.reflect.Field f = sun.misc
1985 >                                .Unsafe.class.getDeclaredField("theUnsafe");
1986 >                            f.setAccessible(true);
1987 >                            return (sun.misc.Unsafe) f.get(null);
1988                          }});
1989              } catch (java.security.PrivilegedActionException e) {
1990 <                throw e.getCause();
1990 >                throw new RuntimeException("Could not initialize intrinsics",
1991 >                                           e.getCause());
1992              }
1993          }
1994      }
1811
1812    private static Unsafe getUnsafePrivileged()
1813            throws NoSuchFieldException, IllegalAccessException {
1814        Field f = Unsafe.class.getDeclaredField("theUnsafe");
1815        f.setAccessible(true);
1816        return (Unsafe) f.get(null);
1817    }
1818
1819    private static long fieldOffset(String fieldName)
1820            throws NoSuchFieldException {
1821        return _unsafe.objectFieldOffset
1822            (ForkJoinPool.class.getDeclaredField(fieldName));
1823    }
1824
1825    static final Unsafe _unsafe;
1826    static final long eventCountOffset;
1827    static final long workerCountsOffset;
1828    static final long runControlOffset;
1829    static final long syncStackOffset;
1830    static final long spareStackOffset;
1831
1832    static {
1833        try {
1834            _unsafe = getUnsafe();
1835            eventCountOffset = fieldOffset("eventCount");
1836            workerCountsOffset = fieldOffset("workerCounts");
1837            runControlOffset = fieldOffset("runControl");
1838            syncStackOffset = fieldOffset("syncStack");
1839            spareStackOffset = fieldOffset("spareStack");
1840        } catch (Throwable e) {
1841            throw new RuntimeException("Could not initialize intrinsics", e);
1842        }
1843    }
1844
1845    private boolean casEventCount(long cmp, long val) {
1846        return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
1847    }
1848    private boolean casWorkerCounts(int cmp, int val) {
1849        return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1850    }
1851    private boolean casRunControl(int cmp, int val) {
1852        return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
1853    }
1854    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1855        return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
1856    }
1857    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1858        return _unsafe.compareAndSwapObject(this, syncStackOffset, cmp, val);
1859    }
1995   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines