ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/main/java/util/concurrent/ForkJoinPool.java
Revision: 1.13
Committed: Sat Dec 5 11:43:01 2009 UTC (14 years, 6 months ago) by dl
Branch: MAIN
Changes since 1.12: +78 -105 lines
Log Message:
Sync with jsr166y version

File Contents

# User Rev Content
1 jsr166 1.1 /*
2     * Written by Doug Lea with assistance from members of JCP JSR-166
3     * Expert Group and released to the public domain, as explained at
4     * http://creativecommons.org/licenses/publicdomain
5     */
6    
7     package java.util.concurrent;
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.locks.Condition;
15     import java.util.concurrent.locks.LockSupport;
16     import java.util.concurrent.locks.ReentrantLock;
17     import java.util.concurrent.atomic.AtomicInteger;
18     import java.util.concurrent.atomic.AtomicLong;
19    
20     /**
21 jsr166 1.4 * An {@link ExecutorService} for running {@link ForkJoinTask}s.
22 jsr166 1.8 * A {@code ForkJoinPool} provides the entry point for submissions
23     * from non-{@code ForkJoinTask}s, as well as management and
24 jsr166 1.11 * monitoring operations.
25 jsr166 1.1 *
26 jsr166 1.9 * <p>A {@code ForkJoinPool} differs from other kinds of {@link
27     * ExecutorService} mainly by virtue of employing
28     * <em>work-stealing</em>: all threads in the pool attempt to find and
29     * execute subtasks created by other active tasks (eventually blocking
30     * waiting for work if none exist). This enables efficient processing
31     * when most tasks spawn other subtasks (as do most {@code
32     * ForkJoinTask}s). A {@code ForkJoinPool} may also be used for mixed
33     * execution of some plain {@code Runnable}- or {@code Callable}-
34     * based activities along with {@code ForkJoinTask}s. When setting
35     * {@linkplain #setAsyncMode async mode}, a {@code ForkJoinPool} may
36     * also be appropriate for use with fine-grained tasks of any form
37     * that are never joined. Otherwise, other {@code ExecutorService}
38     * implementations are typically more appropriate choices.
39 jsr166 1.1 *
40 jsr166 1.9 * <p>A {@code ForkJoinPool} is constructed with a given target
41     * parallelism level; by default, equal to the number of available
42     * processors. Unless configured otherwise via {@link
43     * #setMaintainsParallelism}, the pool attempts to maintain this
44     * number of active (or available) threads by dynamically adding,
45     * suspending, or resuming internal worker threads, even if some tasks
46 jsr166 1.10 * are stalled waiting to join others. However, no such adjustments
47     * are performed in the face of blocked IO or other unmanaged
48 jsr166 1.8 * synchronization. The nested {@link ManagedBlocker} interface
49     * enables extension of the kinds of synchronization accommodated.
50     * The target parallelism level may also be changed dynamically
51 jsr166 1.9 * ({@link #setParallelism}). The total number of threads may be
52     * limited using method {@link #setMaximumPoolSize}, in which case it
53     * may become possible for the activities of a pool to stall due to
54     * the lack of available threads to process new tasks.
55 jsr166 1.1 *
56     * <p>In addition to execution and lifecycle control methods, this
57     * class provides status check methods (for example
58 jsr166 1.4 * {@link #getStealCount}) that are intended to aid in developing,
59 jsr166 1.1 * tuning, and monitoring fork/join applications. Also, method
60 jsr166 1.4 * {@link #toString} returns indications of pool state in a
61 jsr166 1.1 * convenient form for informal monitoring.
62     *
63 jsr166 1.9 * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
64     * used for all parallel task execution in a program or subsystem.
65     * Otherwise, use would not usually outweigh the construction and
66     * bookkeeping overhead of creating a large set of threads. For
67     * example, a common pool could be used for the {@code SortTasks}
68     * illustrated in {@link RecursiveAction}. Because {@code
69     * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
70     * daemon} mode, there is typically no need to explicitly {@link
71     * #shutdown} such a pool upon program exit.
72     *
73     * <pre>
74     * static final ForkJoinPool mainPool = new ForkJoinPool();
75     * ...
76     * public void sort(long[] array) {
77     * mainPool.invoke(new SortTask(array, 0, array.length));
78     * }
79     * </pre>
80     *
81 jsr166 1.1 * <p><b>Implementation notes</b>: This implementation restricts the
82     * maximum number of running threads to 32767. Attempts to create
83 jsr166 1.11 * pools with greater than the maximum number result in
84 jsr166 1.8 * {@code IllegalArgumentException}.
85 jsr166 1.1 *
86 jsr166 1.11 * <p>This implementation rejects submitted tasks (that is, by throwing
87     * {@link RejectedExecutionException}) only when the pool is shut down.
88     *
89 jsr166 1.1 * @since 1.7
90     * @author Doug Lea
91     */
92     public class ForkJoinPool extends AbstractExecutorService {
93    
94     /*
95     * See the extended comments interspersed below for design,
96     * rationale, and walkthroughs.
97     */
98    
99     /** Mask for packing and unpacking shorts */
100     private static final int shortMask = 0xffff;
101    
102     /** Max pool size -- must be a power of two minus 1 */
103     private static final int MAX_THREADS = 0x7FFF;
104    
105     /**
106 jsr166 1.8 * Factory for creating new {@link ForkJoinWorkerThread}s.
107     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
108     * for {@code ForkJoinWorkerThread} subclasses that extend base
109     * functionality or initialize threads with different contexts.
110 jsr166 1.1 */
111     public static interface ForkJoinWorkerThreadFactory {
112     /**
113     * Returns a new worker thread operating in the given pool.
114     *
115     * @param pool the pool this thread works in
116 jsr166 1.11 * @throws NullPointerException if the pool is null
117 jsr166 1.1 */
118     public ForkJoinWorkerThread newThread(ForkJoinPool pool);
119     }
120    
121     /**
122     * Default ForkJoinWorkerThreadFactory implementation; creates a
123     * new ForkJoinWorkerThread.
124     */
125     static class DefaultForkJoinWorkerThreadFactory
126     implements ForkJoinWorkerThreadFactory {
127     public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
128     try {
129     return new ForkJoinWorkerThread(pool);
130     } catch (OutOfMemoryError oom) {
131     return null;
132     }
133     }
134     }
135    
136     /**
137     * Creates a new ForkJoinWorkerThread. This factory is used unless
138     * overridden in ForkJoinPool constructors.
139     */
140     public static final ForkJoinWorkerThreadFactory
141     defaultForkJoinWorkerThreadFactory =
142     new DefaultForkJoinWorkerThreadFactory();
143    
144     /**
145     * Permission required for callers of methods that may start or
146     * kill threads.
147     */
148     private static final RuntimePermission modifyThreadPermission =
149     new RuntimePermission("modifyThread");
150    
151     /**
152     * If there is a security manager, makes sure caller has
153     * permission to modify threads.
154     */
155     private static void checkPermission() {
156     SecurityManager security = System.getSecurityManager();
157     if (security != null)
158     security.checkPermission(modifyThreadPermission);
159     }
160    
161     /**
162     * Generator for assigning sequence numbers as pool names.
163     */
164     private static final AtomicInteger poolNumberGenerator =
165     new AtomicInteger();
166    
167     /**
168     * Array holding all worker threads in the pool. Initialized upon
169     * first use. Array size must be a power of two. Updates and
170     * replacements are protected by workerLock, but it is always kept
171     * in a consistent enough state to be randomly accessed without
172     * locking by workers performing work-stealing.
173     */
174     volatile ForkJoinWorkerThread[] workers;
175    
176     /**
177     * Lock protecting access to workers.
178     */
179     private final ReentrantLock workerLock;
180    
181     /**
182     * Condition for awaitTermination.
183     */
184     private final Condition termination;
185    
186     /**
187     * The uncaught exception handler used when any worker
188     * abruptly terminates
189     */
190     private Thread.UncaughtExceptionHandler ueh;
191    
192     /**
193     * Creation factory for worker threads.
194     */
195     private final ForkJoinWorkerThreadFactory factory;
196    
197     /**
198     * Head of stack of threads that were created to maintain
199     * parallelism when other threads blocked, but have since
200     * suspended when the parallelism level rose.
201     */
202     private volatile WaitQueueNode spareStack;
203    
204     /**
205     * Sum of per-thread steal counts, updated only when threads are
206     * idle or terminating.
207     */
208     private final AtomicLong stealCount;
209    
210     /**
211     * Queue for external submissions.
212     */
213     private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
214    
215     /**
216     * Head of Treiber stack for barrier sync. See below for explanation.
217     */
218     private volatile WaitQueueNode syncStack;
219    
220     /**
221     * The count for event barrier
222     */
223     private volatile long eventCount;
224    
225     /**
226     * Pool number, just for assigning useful names to worker threads
227     */
228     private final int poolNumber;
229    
230     /**
231     * The maximum allowed pool size
232     */
233     private volatile int maxPoolSize;
234    
235     /**
236     * The desired parallelism level, updated only under workerLock.
237     */
238     private volatile int parallelism;
239    
240     /**
241     * True if use local fifo, not default lifo, for local polling
242     */
243     private volatile boolean locallyFifo;
244    
245     /**
246     * Holds number of total (i.e., created and not yet terminated)
247     * and running (i.e., not blocked on joins or other managed sync)
248     * threads, packed into one int to ensure consistent snapshot when
249     * making decisions about creating and suspending spare
250     * threads. Updated only by CAS. Note: CASes in
251     * updateRunningCount and preJoin assume that running active count
252     * is in low word, so need to be modified if this changes.
253     */
254     private volatile int workerCounts;
255    
256     private static int totalCountOf(int s) { return s >>> 16; }
257     private static int runningCountOf(int s) { return s & shortMask; }
258     private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
259    
260     /**
261     * Adds delta (which may be negative) to running count. This must
262     * be called before (with negative arg) and after (with positive)
263     * any managed synchronization (i.e., mainly, joins).
264     *
265     * @param delta the number to add
266     */
267     final void updateRunningCount(int delta) {
268     int s;
269     do {} while (!casWorkerCounts(s = workerCounts, s + delta));
270     }
271    
272     /**
273     * Adds delta (which may be negative) to both total and running
274     * count. This must be called upon creation and termination of
275     * worker threads.
276     *
277     * @param delta the number to add
278     */
279     private void updateWorkerCount(int delta) {
280     int d = delta + (delta << 16); // add to both lo and hi parts
281     int s;
282     do {} while (!casWorkerCounts(s = workerCounts, s + d));
283     }
284    
285     /**
286     * Lifecycle control. High word contains runState, low word
287     * contains the number of workers that are (probably) executing
288     * tasks. This value is atomically incremented before a worker
289     * gets a task to run, and decremented when worker has no tasks
290     * and cannot find any. These two fields are bundled together to
291     * support correct termination triggering. Note: activeCount
292     * CAS'es cheat by assuming active count is in low word, so need
293     * to be modified if this changes
294     */
295     private volatile int runControl;
296    
297     // RunState values. Order among values matters
298     private static final int RUNNING = 0;
299     private static final int SHUTDOWN = 1;
300     private static final int TERMINATING = 2;
301     private static final int TERMINATED = 3;
302    
303     private static int runStateOf(int c) { return c >>> 16; }
304     private static int activeCountOf(int c) { return c & shortMask; }
305     private static int runControlFor(int r, int a) { return (r << 16) + a; }
306    
307     /**
308     * Tries incrementing active count; fails on contention.
309     * Called by workers before/during executing tasks.
310     *
311     * @return true on success
312     */
313     final boolean tryIncrementActiveCount() {
314     int c = runControl;
315     return casRunControl(c, c+1);
316     }
317    
318     /**
319     * Tries decrementing active count; fails on contention.
320     * Possibly triggers termination on success.
321     * Called by workers when they can't find tasks.
322     *
323     * @return true on success
324     */
325     final boolean tryDecrementActiveCount() {
326     int c = runControl;
327     int nextc = c - 1;
328     if (!casRunControl(c, nextc))
329     return false;
330     if (canTerminateOnShutdown(nextc))
331     terminateOnShutdown();
332     return true;
333     }
334    
335     /**
336 jsr166 1.4 * Returns {@code true} if argument represents zero active count
337     * and nonzero runstate, which is the triggering condition for
338 jsr166 1.1 * terminating on shutdown.
339     */
340     private static boolean canTerminateOnShutdown(int c) {
341     // i.e. least bit is nonzero runState bit
342     return ((c & -c) >>> 16) != 0;
343     }
344    
345     /**
346     * Transition run state to at least the given state. Return true
347     * if not already at least given state.
348     */
349     private boolean transitionRunStateTo(int state) {
350     for (;;) {
351     int c = runControl;
352     if (runStateOf(c) >= state)
353     return false;
354     if (casRunControl(c, runControlFor(state, activeCountOf(c))))
355     return true;
356     }
357     }
358    
359     /**
360     * Controls whether to add spares to maintain parallelism
361     */
362     private volatile boolean maintainsParallelism;
363    
364     // Constructors
365    
366     /**
367 jsr166 1.9 * Creates a {@code ForkJoinPool} with parallelism equal to {@link
368     * java.lang.Runtime#availableProcessors}, and using the {@linkplain
369     * #defaultForkJoinWorkerThreadFactory default thread factory}.
370 jsr166 1.1 *
371     * @throws SecurityException if a security manager exists and
372     * the caller is not permitted to modify threads
373     * because it does not hold {@link
374     * java.lang.RuntimePermission}{@code ("modifyThread")}
375     */
376     public ForkJoinPool() {
377     this(Runtime.getRuntime().availableProcessors(),
378     defaultForkJoinWorkerThreadFactory);
379     }
380    
381     /**
382 jsr166 1.9 * Creates a {@code ForkJoinPool} with the indicated parallelism
383     * level and using the {@linkplain
384     * #defaultForkJoinWorkerThreadFactory default thread factory}.
385 jsr166 1.1 *
386 jsr166 1.9 * @param parallelism the parallelism level
387 jsr166 1.1 * @throws IllegalArgumentException if parallelism less than or
388 jsr166 1.11 * equal to zero, or greater than implementation limit
389 jsr166 1.1 * @throws SecurityException if a security manager exists and
390     * the caller is not permitted to modify threads
391     * because it does not hold {@link
392     * java.lang.RuntimePermission}{@code ("modifyThread")}
393     */
394     public ForkJoinPool(int parallelism) {
395     this(parallelism, defaultForkJoinWorkerThreadFactory);
396     }
397    
398     /**
399 jsr166 1.9 * Creates a {@code ForkJoinPool} with parallelism equal to {@link
400     * java.lang.Runtime#availableProcessors}, and using the given
401     * thread factory.
402 jsr166 1.1 *
403     * @param factory the factory for creating new threads
404 jsr166 1.11 * @throws NullPointerException if the factory is null
405 jsr166 1.1 * @throws SecurityException if a security manager exists and
406     * the caller is not permitted to modify threads
407     * because it does not hold {@link
408     * java.lang.RuntimePermission}{@code ("modifyThread")}
409     */
410     public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
411     this(Runtime.getRuntime().availableProcessors(), factory);
412     }
413    
414     /**
415 jsr166 1.8 * Creates a {@code ForkJoinPool} with the given parallelism and
416     * thread factory.
417 jsr166 1.1 *
418 jsr166 1.9 * @param parallelism the parallelism level
419 jsr166 1.1 * @param factory the factory for creating new threads
420     * @throws IllegalArgumentException if parallelism less than or
421 jsr166 1.11 * equal to zero, or greater than implementation limit
422     * @throws NullPointerException if the factory is null
423 jsr166 1.1 * @throws SecurityException if a security manager exists and
424     * the caller is not permitted to modify threads
425     * because it does not hold {@link
426     * java.lang.RuntimePermission}{@code ("modifyThread")}
427     */
428     public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
429     if (parallelism <= 0 || parallelism > MAX_THREADS)
430     throw new IllegalArgumentException();
431     if (factory == null)
432     throw new NullPointerException();
433     checkPermission();
434     this.factory = factory;
435     this.parallelism = parallelism;
436     this.maxPoolSize = MAX_THREADS;
437     this.maintainsParallelism = true;
438     this.poolNumber = poolNumberGenerator.incrementAndGet();
439     this.workerLock = new ReentrantLock();
440     this.termination = workerLock.newCondition();
441     this.stealCount = new AtomicLong();
442     this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
443     // worker array and workers are lazily constructed
444     }
445    
446     /**
447     * Creates a new worker thread using factory.
448     *
449     * @param index the index to assign worker
450 jsr166 1.8 * @return new worker, or null if factory failed
451 jsr166 1.1 */
452     private ForkJoinWorkerThread createWorker(int index) {
453     Thread.UncaughtExceptionHandler h = ueh;
454     ForkJoinWorkerThread w = factory.newThread(this);
455     if (w != null) {
456     w.poolIndex = index;
457     w.setDaemon(true);
458     w.setAsyncMode(locallyFifo);
459     w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
460     if (h != null)
461     w.setUncaughtExceptionHandler(h);
462     }
463     return w;
464     }
465    
466     /**
467     * Returns a good size for worker array given pool size.
468     * Currently requires size to be a power of two.
469     */
470     private static int arraySizeFor(int poolSize) {
471 jsr166 1.9 if (poolSize <= 1)
472     return 1;
473     // See Hackers Delight, sec 3.2
474     int c = poolSize >= MAX_THREADS ? MAX_THREADS : (poolSize - 1);
475     c |= c >>> 1;
476     c |= c >>> 2;
477     c |= c >>> 4;
478     c |= c >>> 8;
479     c |= c >>> 16;
480     return c + 1;
481 jsr166 1.1 }
482    
483     /**
484     * Creates or resizes array if necessary to hold newLength.
485     * Call only under exclusion.
486     *
487     * @return the array
488     */
489     private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
490     ForkJoinWorkerThread[] ws = workers;
491     if (ws == null)
492     return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
493     else if (newLength > ws.length)
494     return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
495     else
496     return ws;
497     }
498    
499     /**
500     * Tries to shrink workers into smaller array after one or more terminate.
501     */
502     private void tryShrinkWorkerArray() {
503     ForkJoinWorkerThread[] ws = workers;
504     if (ws != null) {
505     int len = ws.length;
506     int last = len - 1;
507     while (last >= 0 && ws[last] == null)
508     --last;
509     int newLength = arraySizeFor(last+1);
510     if (newLength < len)
511     workers = Arrays.copyOf(ws, newLength);
512     }
513     }
514    
515     /**
516     * Initializes workers if necessary.
517     */
518     final void ensureWorkerInitialization() {
519     ForkJoinWorkerThread[] ws = workers;
520     if (ws == null) {
521     final ReentrantLock lock = this.workerLock;
522     lock.lock();
523     try {
524     ws = workers;
525     if (ws == null) {
526     int ps = parallelism;
527 dl 1.13 updateWorkerCount(ps);
528 jsr166 1.1 ws = ensureWorkerArrayCapacity(ps);
529     for (int i = 0; i < ps; ++i) {
530     ForkJoinWorkerThread w = createWorker(i);
531     if (w != null) {
532     ws[i] = w;
533     w.start();
534     }
535 dl 1.13 else
536     updateWorkerCount(-1);
537 jsr166 1.1 }
538     }
539     } finally {
540     lock.unlock();
541     }
542     }
543     }
544    
545     /**
546     * Worker creation and startup for threads added via setParallelism.
547     */
548     private void createAndStartAddedWorkers() {
549     resumeAllSpares(); // Allow spares to convert to nonspare
550     int ps = parallelism;
551     ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
552     int len = ws.length;
553     // Sweep through slots, to keep lowest indices most populated
554     int k = 0;
555     while (k < len) {
556     if (ws[k] != null) {
557     ++k;
558     continue;
559     }
560     int s = workerCounts;
561     int tc = totalCountOf(s);
562     int rc = runningCountOf(s);
563     if (rc >= ps || tc >= ps)
564     break;
565     if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
566     ForkJoinWorkerThread w = createWorker(k);
567     if (w != null) {
568     ws[k++] = w;
569     w.start();
570     }
571     else {
572     updateWorkerCount(-1); // back out on failed creation
573     break;
574     }
575     }
576     }
577     }
578    
579     // Execution methods
580    
581     /**
582     * Common code for execute, invoke and submit
583     */
584     private <T> void doSubmit(ForkJoinTask<T> task) {
585 jsr166 1.2 if (task == null)
586     throw new NullPointerException();
587 jsr166 1.1 if (isShutdown())
588     throw new RejectedExecutionException();
589     if (workers == null)
590     ensureWorkerInitialization();
591     submissionQueue.offer(task);
592     signalIdleWorkers();
593     }
594    
595     /**
596     * Performs the given task, returning its result upon completion.
597     *
598     * @param task the task
599     * @return the task's result
600 jsr166 1.11 * @throws NullPointerException if the task is null
601     * @throws RejectedExecutionException if the task cannot be
602     * scheduled for execution
603 jsr166 1.1 */
604     public <T> T invoke(ForkJoinTask<T> task) {
605     doSubmit(task);
606     return task.join();
607     }
608    
609     /**
610     * Arranges for (asynchronous) execution of the given task.
611     *
612     * @param task the task
613 jsr166 1.11 * @throws NullPointerException if the task is null
614     * @throws RejectedExecutionException if the task cannot be
615     * scheduled for execution
616 jsr166 1.1 */
617 jsr166 1.8 public void execute(ForkJoinTask<?> task) {
618 jsr166 1.1 doSubmit(task);
619     }
620    
621     // AbstractExecutorService methods
622    
623 jsr166 1.11 /**
624     * @throws NullPointerException if the task is null
625     * @throws RejectedExecutionException if the task cannot be
626     * scheduled for execution
627     */
628 jsr166 1.1 public void execute(Runnable task) {
629 jsr166 1.2 ForkJoinTask<?> job;
630 jsr166 1.3 if (task instanceof ForkJoinTask<?>) // avoid re-wrap
631     job = (ForkJoinTask<?>) task;
632 jsr166 1.2 else
633 jsr166 1.7 job = ForkJoinTask.adapt(task, null);
634 jsr166 1.2 doSubmit(job);
635 jsr166 1.1 }
636    
637 jsr166 1.11 /**
638     * @throws NullPointerException if the task is null
639     * @throws RejectedExecutionException if the task cannot be
640     * scheduled for execution
641     */
642 jsr166 1.1 public <T> ForkJoinTask<T> submit(Callable<T> task) {
643 jsr166 1.7 ForkJoinTask<T> job = ForkJoinTask.adapt(task);
644 jsr166 1.1 doSubmit(job);
645     return job;
646     }
647    
648 jsr166 1.11 /**
649     * @throws NullPointerException if the task is null
650     * @throws RejectedExecutionException if the task cannot be
651     * scheduled for execution
652     */
653 jsr166 1.1 public <T> ForkJoinTask<T> submit(Runnable task, T result) {
654 jsr166 1.7 ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
655 jsr166 1.1 doSubmit(job);
656     return job;
657     }
658    
659 jsr166 1.11 /**
660     * @throws NullPointerException if the task is null
661     * @throws RejectedExecutionException if the task cannot be
662     * scheduled for execution
663     */
664 jsr166 1.1 public ForkJoinTask<?> submit(Runnable task) {
665 jsr166 1.2 ForkJoinTask<?> job;
666 jsr166 1.3 if (task instanceof ForkJoinTask<?>) // avoid re-wrap
667     job = (ForkJoinTask<?>) task;
668 jsr166 1.2 else
669 jsr166 1.7 job = ForkJoinTask.adapt(task, null);
670 jsr166 1.1 doSubmit(job);
671     return job;
672     }
673    
674     /**
675 jsr166 1.2 * Submits a ForkJoinTask for execution.
676     *
677     * @param task the task to submit
678     * @return the task
679 jsr166 1.11 * @throws NullPointerException if the task is null
680 jsr166 1.2 * @throws RejectedExecutionException if the task cannot be
681     * scheduled for execution
682     */
683     public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
684     doSubmit(task);
685     return task;
686     }
687    
688 jsr166 1.1
689 jsr166 1.11 /**
690     * @throws NullPointerException {@inheritDoc}
691     * @throws RejectedExecutionException {@inheritDoc}
692     */
693 jsr166 1.1 public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
694     ArrayList<ForkJoinTask<T>> forkJoinTasks =
695     new ArrayList<ForkJoinTask<T>>(tasks.size());
696     for (Callable<T> task : tasks)
697 jsr166 1.7 forkJoinTasks.add(ForkJoinTask.adapt(task));
698 jsr166 1.1 invoke(new InvokeAll<T>(forkJoinTasks));
699    
700     @SuppressWarnings({"unchecked", "rawtypes"})
701     List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
702     return futures;
703     }
704    
705     static final class InvokeAll<T> extends RecursiveAction {
706     final ArrayList<ForkJoinTask<T>> tasks;
707     InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
708     public void compute() {
709     try { invokeAll(tasks); }
710     catch (Exception ignore) {}
711     }
712     private static final long serialVersionUID = -7914297376763021607L;
713     }
714    
715     // Configuration and status settings and queries
716    
717     /**
718     * Returns the factory used for constructing new workers.
719     *
720     * @return the factory used for constructing new workers
721     */
722     public ForkJoinWorkerThreadFactory getFactory() {
723     return factory;
724     }
725    
726     /**
727     * Returns the handler for internal worker threads that terminate
728     * due to unrecoverable errors encountered while executing tasks.
729     *
730 jsr166 1.4 * @return the handler, or {@code null} if none
731 jsr166 1.1 */
732     public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
733     Thread.UncaughtExceptionHandler h;
734     final ReentrantLock lock = this.workerLock;
735     lock.lock();
736     try {
737     h = ueh;
738     } finally {
739     lock.unlock();
740     }
741     return h;
742     }
743    
744     /**
745     * Sets the handler for internal worker threads that terminate due
746     * to unrecoverable errors encountered while executing tasks.
747     * Unless set, the current default or ThreadGroup handler is used
748     * as handler.
749     *
750     * @param h the new handler
751 jsr166 1.4 * @return the old handler, or {@code null} if none
752 jsr166 1.1 * @throws SecurityException if a security manager exists and
753     * the caller is not permitted to modify threads
754     * because it does not hold {@link
755     * java.lang.RuntimePermission}{@code ("modifyThread")}
756     */
757     public Thread.UncaughtExceptionHandler
758     setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
759     checkPermission();
760     Thread.UncaughtExceptionHandler old = null;
761     final ReentrantLock lock = this.workerLock;
762     lock.lock();
763     try {
764     old = ueh;
765     ueh = h;
766     ForkJoinWorkerThread[] ws = workers;
767     if (ws != null) {
768     for (int i = 0; i < ws.length; ++i) {
769     ForkJoinWorkerThread w = ws[i];
770     if (w != null)
771     w.setUncaughtExceptionHandler(h);
772     }
773     }
774     } finally {
775     lock.unlock();
776     }
777     return old;
778     }
779    
780    
781     /**
782     * Sets the target parallelism level of this pool.
783     *
784     * @param parallelism the target parallelism
785     * @throws IllegalArgumentException if parallelism less than or
786     * equal to zero or greater than maximum size bounds
787     * @throws SecurityException if a security manager exists and
788     * the caller is not permitted to modify threads
789     * because it does not hold {@link
790     * java.lang.RuntimePermission}{@code ("modifyThread")}
791     */
792     public void setParallelism(int parallelism) {
793     checkPermission();
794     if (parallelism <= 0 || parallelism > maxPoolSize)
795     throw new IllegalArgumentException();
796     final ReentrantLock lock = this.workerLock;
797     lock.lock();
798     try {
799 jsr166 1.9 if (isProcessingTasks()) {
800 jsr166 1.1 int p = this.parallelism;
801     this.parallelism = parallelism;
802 dl 1.13 if (workers != null) {
803     if (parallelism > p)
804     createAndStartAddedWorkers();
805     else
806     trimSpares();
807     }
808 jsr166 1.1 }
809     } finally {
810     lock.unlock();
811     }
812     signalIdleWorkers();
813     }
814    
815     /**
816 jsr166 1.9 * Returns the targeted parallelism level of this pool.
817 jsr166 1.1 *
818 jsr166 1.9 * @return the targeted parallelism level of this pool
819 jsr166 1.1 */
820     public int getParallelism() {
821     return parallelism;
822     }
823    
824     /**
825     * Returns the number of worker threads that have started but not
826     * yet terminated. This result returned by this method may differ
827 jsr166 1.4 * from {@link #getParallelism} when threads are created to
828 jsr166 1.1 * maintain parallelism when others are cooperatively blocked.
829     *
830     * @return the number of worker threads
831     */
832     public int getPoolSize() {
833     return totalCountOf(workerCounts);
834     }
835    
836     /**
837     * Returns the maximum number of threads allowed to exist in the
838 jsr166 1.10 * pool. Unless set using {@link #setMaximumPoolSize}, the
839 jsr166 1.9 * maximum is an implementation-defined value designed only to
840     * prevent runaway growth.
841 jsr166 1.1 *
842     * @return the maximum
843     */
844     public int getMaximumPoolSize() {
845     return maxPoolSize;
846     }
847    
848     /**
849     * Sets the maximum number of threads allowed to exist in the
850 jsr166 1.10 * pool. The given value should normally be greater than or equal
851     * to the {@link #getParallelism parallelism} level. Setting this
852     * value has no effect on current pool size. It controls
853     * construction of new threads.
854 jsr166 1.1 *
855 jsr166 1.8 * @throws IllegalArgumentException if negative or greater than
856 jsr166 1.1 * internal implementation limit
857     */
858     public void setMaximumPoolSize(int newMax) {
859     if (newMax < 0 || newMax > MAX_THREADS)
860     throw new IllegalArgumentException();
861     maxPoolSize = newMax;
862     }
863    
864    
865     /**
866 jsr166 1.4 * Returns {@code true} if this pool dynamically maintains its
867     * target parallelism level. If false, new threads are added only
868     * to avoid possible starvation. This setting is by default true.
869 jsr166 1.1 *
870 jsr166 1.4 * @return {@code true} if maintains parallelism
871 jsr166 1.1 */
872     public boolean getMaintainsParallelism() {
873     return maintainsParallelism;
874     }
875    
876     /**
877     * Sets whether this pool dynamically maintains its target
878     * parallelism level. If false, new threads are added only to
879     * avoid possible starvation.
880     *
881 jsr166 1.4 * @param enable {@code true} to maintain parallelism
882 jsr166 1.1 */
883     public void setMaintainsParallelism(boolean enable) {
884     maintainsParallelism = enable;
885     }
886    
887     /**
888     * Establishes local first-in-first-out scheduling mode for forked
889     * tasks that are never joined. This mode may be more appropriate
890     * than default locally stack-based mode in applications in which
891     * worker threads only process asynchronous tasks. This method is
892 jsr166 1.4 * designed to be invoked only when the pool is quiescent, and
893 jsr166 1.1 * typically only before any tasks are submitted. The effects of
894     * invocations at other times may be unpredictable.
895     *
896 jsr166 1.4 * @param async if {@code true}, use locally FIFO scheduling
897 jsr166 1.1 * @return the previous mode
898 jsr166 1.4 * @see #getAsyncMode
899 jsr166 1.1 */
900     public boolean setAsyncMode(boolean async) {
901     boolean oldMode = locallyFifo;
902     locallyFifo = async;
903     ForkJoinWorkerThread[] ws = workers;
904     if (ws != null) {
905     for (int i = 0; i < ws.length; ++i) {
906     ForkJoinWorkerThread t = ws[i];
907     if (t != null)
908     t.setAsyncMode(async);
909     }
910     }
911     return oldMode;
912     }
913    
914     /**
915 jsr166 1.4 * Returns {@code true} if this pool uses local first-in-first-out
916 jsr166 1.1 * scheduling mode for forked tasks that are never joined.
917     *
918 jsr166 1.4 * @return {@code true} if this pool uses async mode
919     * @see #setAsyncMode
920 jsr166 1.1 */
921     public boolean getAsyncMode() {
922     return locallyFifo;
923     }
924    
925     /**
926     * Returns an estimate of the number of worker threads that are
927     * not blocked waiting to join tasks or for other managed
928     * synchronization.
929     *
930     * @return the number of worker threads
931     */
932     public int getRunningThreadCount() {
933     return runningCountOf(workerCounts);
934     }
935    
936     /**
937     * Returns an estimate of the number of threads that are currently
938     * stealing or executing tasks. This method may overestimate the
939     * number of active threads.
940     *
941     * @return the number of active threads
942     */
943     public int getActiveThreadCount() {
944     return activeCountOf(runControl);
945     }
946    
947     /**
948     * Returns an estimate of the number of threads that are currently
949     * idle waiting for tasks. This method may underestimate the
950     * number of idle threads.
951     *
952     * @return the number of idle threads
953     */
954     final int getIdleThreadCount() {
955     int c = runningCountOf(workerCounts) - activeCountOf(runControl);
956     return (c <= 0) ? 0 : c;
957     }
958    
959     /**
960 jsr166 1.4 * Returns {@code true} if all worker threads are currently idle.
961     * An idle worker is one that cannot obtain a task to execute
962     * because none are available to steal from other threads, and
963     * there are no pending submissions to the pool. This method is
964     * conservative; it might not return {@code true} immediately upon
965     * idleness of all threads, but will eventually become true if
966     * threads remain inactive.
967 jsr166 1.1 *
968 jsr166 1.4 * @return {@code true} if all threads are currently idle
969 jsr166 1.1 */
970     public boolean isQuiescent() {
971     return activeCountOf(runControl) == 0;
972     }
973    
974     /**
975     * Returns an estimate of the total number of tasks stolen from
976     * one thread's work queue by another. The reported value
977     * underestimates the actual total number of steals when the pool
978     * is not quiescent. This value may be useful for monitoring and
979     * tuning fork/join programs: in general, steal counts should be
980     * high enough to keep threads busy, but low enough to avoid
981     * overhead and contention across threads.
982     *
983     * @return the number of steals
984     */
985     public long getStealCount() {
986     return stealCount.get();
987     }
988    
989     /**
990     * Accumulates steal count from a worker.
991     * Call only when worker known to be idle.
992     */
993     private void updateStealCount(ForkJoinWorkerThread w) {
994     int sc = w.getAndClearStealCount();
995     if (sc != 0)
996     stealCount.addAndGet(sc);
997     }
998    
999     /**
1000     * Returns an estimate of the total number of tasks currently held
1001     * in queues by worker threads (but not including tasks submitted
1002     * to the pool that have not begun executing). This value is only
1003     * an approximation, obtained by iterating across all threads in
1004     * the pool. This method may be useful for tuning task
1005     * granularities.
1006     *
1007     * @return the number of queued tasks
1008     */
1009     public long getQueuedTaskCount() {
1010     long count = 0;
1011     ForkJoinWorkerThread[] ws = workers;
1012     if (ws != null) {
1013     for (int i = 0; i < ws.length; ++i) {
1014     ForkJoinWorkerThread t = ws[i];
1015     if (t != null)
1016     count += t.getQueueSize();
1017     }
1018     }
1019     return count;
1020     }
1021    
1022     /**
1023 jsr166 1.8 * Returns an estimate of the number of tasks submitted to this
1024     * pool that have not yet begun executing. This method takes time
1025 jsr166 1.1 * proportional to the number of submissions.
1026     *
1027     * @return the number of queued submissions
1028     */
1029     public int getQueuedSubmissionCount() {
1030     return submissionQueue.size();
1031     }
1032    
1033     /**
1034 jsr166 1.4 * Returns {@code true} if there are any tasks submitted to this
1035     * pool that have not yet begun executing.
1036 jsr166 1.1 *
1037     * @return {@code true} if there are any queued submissions
1038     */
1039     public boolean hasQueuedSubmissions() {
1040     return !submissionQueue.isEmpty();
1041     }
1042    
1043     /**
1044     * Removes and returns the next unexecuted submission if one is
1045     * available. This method may be useful in extensions to this
1046     * class that re-assign work in systems with multiple pools.
1047     *
1048 jsr166 1.4 * @return the next submission, or {@code null} if none
1049 jsr166 1.1 */
1050     protected ForkJoinTask<?> pollSubmission() {
1051     return submissionQueue.poll();
1052     }
1053    
1054     /**
1055     * Removes all available unexecuted submitted and forked tasks
1056     * from scheduling queues and adds them to the given collection,
1057     * without altering their execution status. These may include
1058 jsr166 1.8 * artificially generated or wrapped tasks. This method is
1059     * designed to be invoked only when the pool is known to be
1060 jsr166 1.1 * quiescent. Invocations at other times may not remove all
1061     * tasks. A failure encountered while attempting to add elements
1062     * to collection {@code c} may result in elements being in
1063     * neither, either or both collections when the associated
1064     * exception is thrown. The behavior of this operation is
1065     * undefined if the specified collection is modified while the
1066     * operation is in progress.
1067     *
1068     * @param c the collection to transfer elements into
1069     * @return the number of elements transferred
1070     */
1071 jsr166 1.5 protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1072 jsr166 1.1 int n = submissionQueue.drainTo(c);
1073     ForkJoinWorkerThread[] ws = workers;
1074     if (ws != null) {
1075     for (int i = 0; i < ws.length; ++i) {
1076     ForkJoinWorkerThread w = ws[i];
1077     if (w != null)
1078     n += w.drainTasksTo(c);
1079     }
1080     }
1081     return n;
1082     }
1083    
1084     /**
1085     * Returns a string identifying this pool, as well as its state,
1086     * including indications of run state, parallelism level, and
1087     * worker and task counts.
1088     *
1089     * @return a string identifying this pool, as well as its state
1090     */
1091     public String toString() {
1092     int ps = parallelism;
1093     int wc = workerCounts;
1094     int rc = runControl;
1095     long st = getStealCount();
1096     long qt = getQueuedTaskCount();
1097     long qs = getQueuedSubmissionCount();
1098     return super.toString() +
1099     "[" + runStateToString(runStateOf(rc)) +
1100     ", parallelism = " + ps +
1101     ", size = " + totalCountOf(wc) +
1102     ", active = " + activeCountOf(rc) +
1103     ", running = " + runningCountOf(wc) +
1104     ", steals = " + st +
1105     ", tasks = " + qt +
1106     ", submissions = " + qs +
1107     "]";
1108     }
1109    
1110     private static String runStateToString(int rs) {
1111 jsr166 1.12 switch (rs) {
1112 jsr166 1.1 case RUNNING: return "Running";
1113     case SHUTDOWN: return "Shutting down";
1114     case TERMINATING: return "Terminating";
1115     case TERMINATED: return "Terminated";
1116     default: throw new Error("Unknown run state");
1117     }
1118     }
1119    
1120     // lifecycle control
1121    
1122     /**
1123     * Initiates an orderly shutdown in which previously submitted
1124     * tasks are executed, but no new tasks will be accepted.
1125     * Invocation has no additional effect if already shut down.
1126     * Tasks that are in the process of being submitted concurrently
1127     * during the course of this method may or may not be rejected.
1128     *
1129     * @throws SecurityException if a security manager exists and
1130     * the caller is not permitted to modify threads
1131     * because it does not hold {@link
1132     * java.lang.RuntimePermission}{@code ("modifyThread")}
1133     */
1134     public void shutdown() {
1135     checkPermission();
1136     transitionRunStateTo(SHUTDOWN);
1137 jsr166 1.6 if (canTerminateOnShutdown(runControl)) {
1138     if (workers == null) { // shutting down before workers created
1139     final ReentrantLock lock = this.workerLock;
1140     lock.lock();
1141     try {
1142     if (workers == null) {
1143     terminate();
1144     transitionRunStateTo(TERMINATED);
1145     termination.signalAll();
1146     }
1147     } finally {
1148     lock.unlock();
1149     }
1150     }
1151 jsr166 1.1 terminateOnShutdown();
1152 jsr166 1.6 }
1153 jsr166 1.1 }
1154    
1155     /**
1156 jsr166 1.9 * Attempts to cancel and/or stop all tasks, and reject all
1157     * subsequently submitted tasks. Tasks that are in the process of
1158     * being submitted or executed concurrently during the course of
1159     * this method may or may not be rejected. This method cancels
1160     * both existing and unexecuted tasks, in order to permit
1161     * termination in the presence of task dependencies. So the method
1162     * always returns an empty list (unlike the case for some other
1163     * Executors).
1164 jsr166 1.1 *
1165     * @return an empty list
1166     * @throws SecurityException if a security manager exists and
1167     * the caller is not permitted to modify threads
1168     * because it does not hold {@link
1169     * java.lang.RuntimePermission}{@code ("modifyThread")}
1170     */
1171     public List<Runnable> shutdownNow() {
1172     checkPermission();
1173     terminate();
1174     return Collections.emptyList();
1175     }
1176    
1177     /**
1178     * Returns {@code true} if all tasks have completed following shut down.
1179     *
1180     * @return {@code true} if all tasks have completed following shut down
1181     */
1182     public boolean isTerminated() {
1183     return runStateOf(runControl) == TERMINATED;
1184     }
1185    
1186     /**
1187     * Returns {@code true} if the process of termination has
1188 jsr166 1.9 * commenced but not yet completed. This method may be useful for
1189     * debugging. A return of {@code true} reported a sufficient
1190     * period after shutdown may indicate that submitted tasks have
1191     * ignored or suppressed interruption, causing this executor not
1192     * to properly terminate.
1193 jsr166 1.1 *
1194 jsr166 1.9 * @return {@code true} if terminating but not yet terminated
1195 jsr166 1.1 */
1196     public boolean isTerminating() {
1197 jsr166 1.9 return runStateOf(runControl) == TERMINATING;
1198 jsr166 1.1 }
1199    
1200     /**
1201     * Returns {@code true} if this pool has been shut down.
1202     *
1203     * @return {@code true} if this pool has been shut down
1204     */
1205     public boolean isShutdown() {
1206     return runStateOf(runControl) >= SHUTDOWN;
1207     }
1208    
1209     /**
1210 jsr166 1.9 * Returns true if pool is not terminating or terminated.
1211     * Used internally to suppress execution when terminating.
1212     */
1213     final boolean isProcessingTasks() {
1214     return runStateOf(runControl) < TERMINATING;
1215     }
1216    
1217     /**
1218 jsr166 1.1 * Blocks until all tasks have completed execution after a shutdown
1219     * request, or the timeout occurs, or the current thread is
1220     * interrupted, whichever happens first.
1221     *
1222     * @param timeout the maximum time to wait
1223     * @param unit the time unit of the timeout argument
1224     * @return {@code true} if this executor terminated and
1225     * {@code false} if the timeout elapsed before termination
1226     * @throws InterruptedException if interrupted while waiting
1227     */
1228     public boolean awaitTermination(long timeout, TimeUnit unit)
1229     throws InterruptedException {
1230     long nanos = unit.toNanos(timeout);
1231     final ReentrantLock lock = this.workerLock;
1232     lock.lock();
1233     try {
1234     for (;;) {
1235     if (isTerminated())
1236     return true;
1237     if (nanos <= 0)
1238     return false;
1239     nanos = termination.awaitNanos(nanos);
1240     }
1241     } finally {
1242     lock.unlock();
1243     }
1244     }
1245    
1246     // Shutdown and termination support
1247    
1248     /**
1249     * Callback from terminating worker. Nulls out the corresponding
1250     * workers slot, and if terminating, tries to terminate; else
1251     * tries to shrink workers array.
1252     *
1253     * @param w the worker
1254     */
1255     final void workerTerminated(ForkJoinWorkerThread w) {
1256     updateStealCount(w);
1257     updateWorkerCount(-1);
1258     final ReentrantLock lock = this.workerLock;
1259     lock.lock();
1260     try {
1261     ForkJoinWorkerThread[] ws = workers;
1262     if (ws != null) {
1263     int idx = w.poolIndex;
1264     if (idx >= 0 && idx < ws.length && ws[idx] == w)
1265     ws[idx] = null;
1266     if (totalCountOf(workerCounts) == 0) {
1267     terminate(); // no-op if already terminating
1268     transitionRunStateTo(TERMINATED);
1269     termination.signalAll();
1270     }
1271 jsr166 1.9 else if (isProcessingTasks()) {
1272 jsr166 1.1 tryShrinkWorkerArray();
1273     tryResumeSpare(true); // allow replacement
1274     }
1275     }
1276     } finally {
1277     lock.unlock();
1278     }
1279     signalIdleWorkers();
1280     }
1281    
1282     /**
1283     * Initiates termination.
1284     */
1285     private void terminate() {
1286     if (transitionRunStateTo(TERMINATING)) {
1287     stopAllWorkers();
1288     resumeAllSpares();
1289     signalIdleWorkers();
1290     cancelQueuedSubmissions();
1291     cancelQueuedWorkerTasks();
1292     interruptUnterminatedWorkers();
1293     signalIdleWorkers(); // resignal after interrupt
1294     }
1295     }
1296    
1297     /**
1298     * Possibly terminates when on shutdown state.
1299     */
1300     private void terminateOnShutdown() {
1301     if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1302     terminate();
1303     }
1304    
1305     /**
1306     * Clears out and cancels submissions.
1307     */
1308     private void cancelQueuedSubmissions() {
1309     ForkJoinTask<?> task;
1310     while ((task = pollSubmission()) != null)
1311     task.cancel(false);
1312     }
1313    
1314     /**
1315     * Cleans out worker queues.
1316     */
1317     private void cancelQueuedWorkerTasks() {
1318     final ReentrantLock lock = this.workerLock;
1319     lock.lock();
1320     try {
1321     ForkJoinWorkerThread[] ws = workers;
1322     if (ws != null) {
1323     for (int i = 0; i < ws.length; ++i) {
1324     ForkJoinWorkerThread t = ws[i];
1325     if (t != null)
1326     t.cancelTasks();
1327     }
1328     }
1329     } finally {
1330     lock.unlock();
1331     }
1332     }
1333    
1334     /**
1335     * Sets each worker's status to terminating. Requires lock to avoid
1336     * conflicts with add/remove.
1337     */
1338     private void stopAllWorkers() {
1339     final ReentrantLock lock = this.workerLock;
1340     lock.lock();
1341     try {
1342     ForkJoinWorkerThread[] ws = workers;
1343     if (ws != null) {
1344     for (int i = 0; i < ws.length; ++i) {
1345     ForkJoinWorkerThread t = ws[i];
1346     if (t != null)
1347     t.shutdownNow();
1348     }
1349     }
1350     } finally {
1351     lock.unlock();
1352     }
1353     }
1354    
1355     /**
1356     * Interrupts all unterminated workers. This is not required for
1357     * sake of internal control, but may help unstick user code during
1358     * shutdown.
1359     */
1360     private void interruptUnterminatedWorkers() {
1361     final ReentrantLock lock = this.workerLock;
1362     lock.lock();
1363     try {
1364     ForkJoinWorkerThread[] ws = workers;
1365     if (ws != null) {
1366     for (int i = 0; i < ws.length; ++i) {
1367     ForkJoinWorkerThread t = ws[i];
1368     if (t != null && !t.isTerminated()) {
1369     try {
1370     t.interrupt();
1371     } catch (SecurityException ignore) {
1372     }
1373     }
1374     }
1375     }
1376     } finally {
1377     lock.unlock();
1378     }
1379     }
1380    
1381     /*
1382     * Nodes for event barrier to manage idle threads. Queue nodes
1383     * are basic Treiber stack nodes, also used for spare stack.
1384     *
1385     * The event barrier has an event count and a wait queue (actually
1386     * a Treiber stack). Workers are enabled to look for work when
1387     * the eventCount is incremented. If they fail to find work, they
1388     * may wait for next count. Upon release, threads help others wake
1389     * up.
1390     *
1391     * Synchronization events occur only in enough contexts to
1392     * maintain overall liveness:
1393     *
1394     * - Submission of a new task to the pool
1395     * - Resizes or other changes to the workers array
1396     * - pool termination
1397     * - A worker pushing a task on an empty queue
1398     *
1399     * The case of pushing a task occurs often enough, and is heavy
1400     * enough compared to simple stack pushes, to require special
1401     * handling: Method signalWork returns without advancing count if
1402     * the queue appears to be empty. This would ordinarily result in
1403     * races causing some queued waiters not to be woken up. To avoid
1404 dl 1.13 * this, the first worker enqueued in method sync rescans for
1405     * tasks after being enqueued, and helps signal if any are
1406     * found. This works well because the worker has nothing better to
1407     * do, and so might as well help alleviate the overhead and
1408     * contention on the threads actually doing work. Also, since
1409     * event counts increments on task availability exist to maintain
1410     * liveness (rather than to force refreshes etc), it is OK for
1411     * callers to exit early if contending with another signaller.
1412 jsr166 1.1 */
1413     static final class WaitQueueNode {
1414     WaitQueueNode next; // only written before enqueued
1415     volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1416     final long count; // unused for spare stack
1417    
1418     WaitQueueNode(long c, ForkJoinWorkerThread w) {
1419     count = c;
1420     thread = w;
1421     }
1422    
1423     /**
1424 dl 1.13 * Wakes up waiter, also clearing thread field
1425 jsr166 1.1 */
1426 dl 1.13 void signal() {
1427 jsr166 1.1 ForkJoinWorkerThread t = thread;
1428 dl 1.13 if (t != null) {
1429     thread = null;
1430     LockSupport.unpark(t);
1431 jsr166 1.1 }
1432     }
1433     }
1434    
1435     /**
1436     * Ensures that no thread is waiting for count to advance from the
1437     * current value of eventCount read on entry to this method, by
1438     * releasing waiting threads if necessary.
1439     */
1440 dl 1.13 final void ensureSync() {
1441 jsr166 1.1 long c = eventCount;
1442     WaitQueueNode q;
1443     while ((q = syncStack) != null && q.count < c) {
1444     if (casBarrierStack(q, null)) {
1445     do {
1446     q.signal();
1447     } while ((q = q.next) != null);
1448     break;
1449     }
1450     }
1451     }
1452    
1453     /**
1454     * Increments event count and releases waiting threads.
1455     */
1456     private void signalIdleWorkers() {
1457     long c;
1458     do {} while (!casEventCount(c = eventCount, c+1));
1459     ensureSync();
1460     }
1461    
1462     /**
1463     * Signals threads waiting to poll a task. Because method sync
1464     * rechecks availability, it is OK to only proceed if queue
1465 dl 1.13 * appears to be non-empty, and OK if CAS to increment count
1466     * fails (since some other thread succeeded).
1467 jsr166 1.1 */
1468     final void signalWork() {
1469 dl 1.13 if (syncStack != null) {
1470     long c = eventCount;
1471     casEventCount(c, c+1);
1472     WaitQueueNode q = syncStack;
1473     if (q != null && q.count <= c) {
1474     if (casBarrierStack(q, q.next))
1475     q.signal();
1476     else
1477     ensureSync(); // awaken all on contention
1478     }
1479     }
1480 jsr166 1.1 }
1481    
1482     /**
1483 dl 1.13 * Possibly blocks until event count advances from last value held
1484     * by caller, or if excess threads, caller is resumed as spare, or
1485 jsr166 1.1 * caller or pool is terminating. Updates caller's event on exit.
1486     *
1487     * @param w the calling worker thread
1488     */
1489     final void sync(ForkJoinWorkerThread w) {
1490     updateStealCount(w); // Transfer w's count while it is idle
1491    
1492 dl 1.13 if (!w.isShutdown() && isProcessingTasks() && !suspendIfSpare(w)) {
1493 jsr166 1.1 long prev = w.lastEventCount;
1494     WaitQueueNode node = null;
1495     WaitQueueNode h;
1496 dl 1.13 long c;
1497     while ((c = eventCount) == prev &&
1498 jsr166 1.1 ((h = syncStack) == null || h.count == prev)) {
1499     if (node == null)
1500     node = new WaitQueueNode(prev, w);
1501     if (casBarrierStack(node.next = h, node)) {
1502 dl 1.13 if (!Thread.interrupted() &&
1503     node.thread != null &&
1504     eventCount == prev &&
1505     (h != null || // cover signalWork race
1506     (!ForkJoinWorkerThread.hasQueuedTasks(workers) &&
1507     eventCount == prev)))
1508     LockSupport.park(this);
1509     c = eventCount;
1510     if (node.thread != null) { // help signal if not unparked
1511     node.thread = null;
1512     if (c == prev)
1513     casEventCount(prev, prev + 1);
1514     }
1515 jsr166 1.1 break;
1516     }
1517     }
1518 dl 1.13 w.lastEventCount = c;
1519     ensureSync();
1520 jsr166 1.1 }
1521     }
1522    
1523     /**
1524 jsr166 1.4 * Returns {@code true} if a new sync event occurred since last
1525     * call to sync or this method, if so, updating caller's count.
1526 jsr166 1.1 */
1527     final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
1528 dl 1.13 long wc = w.lastEventCount;
1529     long c = eventCount;
1530     if (wc != c)
1531     w.lastEventCount = c;
1532     ensureSync();
1533     return wc != c || wc != eventCount;
1534 jsr166 1.1 }
1535    
1536     // Parallelism maintenance
1537    
1538     /**
1539     * Decrements running count; if too low, adds spare.
1540     *
1541     * Conceptually, all we need to do here is add or resume a
1542     * spare thread when one is about to block (and remove or
1543     * suspend it later when unblocked -- see suspendIfSpare).
1544     * However, implementing this idea requires coping with
1545     * several problems: we have imperfect information about the
1546     * states of threads. Some count updates can and usually do
1547     * lag run state changes, despite arrangements to keep them
1548     * accurate (for example, when possible, updating counts
1549     * before signalling or resuming), especially when running on
1550     * dynamic JVMs that don't optimize the infrequent paths that
1551     * update counts. Generating too many threads can make these
1552     * problems become worse, because excess threads are more
1553     * likely to be context-switched with others, slowing them all
1554     * down, especially if there is no work available, so all are
1555     * busy scanning or idling. Also, excess spare threads can
1556     * only be suspended or removed when they are idle, not
1557     * immediately when they aren't needed. So adding threads will
1558     * raise parallelism level for longer than necessary. Also,
1559     * FJ applications often encounter highly transient peaks when
1560     * many threads are blocked joining, but for less time than it
1561     * takes to create or resume spares.
1562     *
1563     * @param joinMe if non-null, return early if done
1564     * @param maintainParallelism if true, try to stay within
1565     * target counts, else create only to avoid starvation
1566     * @return true if joinMe known to be done
1567     */
1568     final boolean preJoin(ForkJoinTask<?> joinMe,
1569     boolean maintainParallelism) {
1570     maintainParallelism &= maintainsParallelism; // overrride
1571     boolean dec = false; // true when running count decremented
1572     while (spareStack == null || !tryResumeSpare(dec)) {
1573     int counts = workerCounts;
1574     if (dec || (dec = casWorkerCounts(counts, --counts))) {
1575     if (!needSpare(counts, maintainParallelism))
1576     break;
1577     if (joinMe.status < 0)
1578     return true;
1579     if (tryAddSpare(counts))
1580     break;
1581     }
1582     }
1583     return false;
1584     }
1585    
1586     /**
1587     * Same idea as preJoin
1588     */
1589     final boolean preBlock(ManagedBlocker blocker,
1590     boolean maintainParallelism) {
1591     maintainParallelism &= maintainsParallelism;
1592     boolean dec = false;
1593     while (spareStack == null || !tryResumeSpare(dec)) {
1594     int counts = workerCounts;
1595     if (dec || (dec = casWorkerCounts(counts, --counts))) {
1596     if (!needSpare(counts, maintainParallelism))
1597     break;
1598     if (blocker.isReleasable())
1599     return true;
1600     if (tryAddSpare(counts))
1601     break;
1602     }
1603     }
1604     return false;
1605     }
1606    
1607     /**
1608 jsr166 1.4 * Returns {@code true} if a spare thread appears to be needed.
1609     * If maintaining parallelism, returns true when the deficit in
1610 jsr166 1.1 * running threads is more than the surplus of total threads, and
1611     * there is apparently some work to do. This self-limiting rule
1612     * means that the more threads that have already been added, the
1613     * less parallelism we will tolerate before adding another.
1614     *
1615     * @param counts current worker counts
1616     * @param maintainParallelism try to maintain parallelism
1617     */
1618     private boolean needSpare(int counts, boolean maintainParallelism) {
1619     int ps = parallelism;
1620     int rc = runningCountOf(counts);
1621     int tc = totalCountOf(counts);
1622     int runningDeficit = ps - rc;
1623     int totalSurplus = tc - ps;
1624     return (tc < maxPoolSize &&
1625     (rc == 0 || totalSurplus < 0 ||
1626     (maintainParallelism &&
1627     runningDeficit > totalSurplus &&
1628     ForkJoinWorkerThread.hasQueuedTasks(workers))));
1629     }
1630    
1631     /**
1632     * Adds a spare worker if lock available and no more than the
1633     * expected numbers of threads exist.
1634     *
1635     * @return true if successful
1636     */
1637     private boolean tryAddSpare(int expectedCounts) {
1638     final ReentrantLock lock = this.workerLock;
1639     int expectedRunning = runningCountOf(expectedCounts);
1640     int expectedTotal = totalCountOf(expectedCounts);
1641     boolean success = false;
1642     boolean locked = false;
1643     // confirm counts while locking; CAS after obtaining lock
1644     try {
1645     for (;;) {
1646     int s = workerCounts;
1647     int tc = totalCountOf(s);
1648     int rc = runningCountOf(s);
1649     if (rc > expectedRunning || tc > expectedTotal)
1650     break;
1651     if (!locked && !(locked = lock.tryLock()))
1652     break;
1653     if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1654     createAndStartSpare(tc);
1655     success = true;
1656     break;
1657     }
1658     }
1659     } finally {
1660     if (locked)
1661     lock.unlock();
1662     }
1663     return success;
1664     }
1665    
1666     /**
1667     * Adds the kth spare worker. On entry, pool counts are already
1668     * adjusted to reflect addition.
1669     */
1670     private void createAndStartSpare(int k) {
1671     ForkJoinWorkerThread w = null;
1672     ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1673     int len = ws.length;
1674     // Probably, we can place at slot k. If not, find empty slot
1675     if (k < len && ws[k] != null) {
1676     for (k = 0; k < len && ws[k] != null; ++k)
1677     ;
1678     }
1679 jsr166 1.9 if (k < len && isProcessingTasks() && (w = createWorker(k)) != null) {
1680 jsr166 1.1 ws[k] = w;
1681     w.start();
1682     }
1683     else
1684     updateWorkerCount(-1); // adjust on failure
1685     signalIdleWorkers();
1686     }
1687    
1688     /**
1689     * Suspends calling thread w if there are excess threads. Called
1690     * only from sync. Spares are enqueued in a Treiber stack using
1691     * the same WaitQueueNodes as barriers. They are resumed mainly
1692     * in preJoin, but are also woken on pool events that require all
1693     * threads to check run state.
1694     *
1695     * @param w the caller
1696     */
1697     private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1698     WaitQueueNode node = null;
1699 dl 1.13 for (;;) {
1700     int s = workerCounts;
1701     int rc = runningCountOf(s);
1702     int tc = totalCountOf(s);
1703     int ps = parallelism;
1704     // use tc as bound if rc transiently out of sync
1705     if (tc <= ps || rc <= ps)
1706     return false; // not a spare
1707 jsr166 1.1 if (node == null)
1708     node = new WaitQueueNode(0, w);
1709 dl 1.13 if (casWorkerCounts(s, workerCountsFor(tc, rc - 1)))
1710     break;
1711 jsr166 1.1 }
1712 dl 1.13 // push onto stack
1713     do {} while (!casSpareStack(node.next = spareStack, node));
1714     // block until released by resumeSpare
1715     while (!Thread.interrupted() && node.thread != null)
1716     LockSupport.park(this);
1717     return true;
1718 jsr166 1.1 }
1719    
1720     /**
1721     * Tries to pop and resume a spare thread.
1722     *
1723     * @param updateCount if true, increment running count on success
1724     * @return true if successful
1725     */
1726     private boolean tryResumeSpare(boolean updateCount) {
1727     WaitQueueNode q;
1728     while ((q = spareStack) != null) {
1729     if (casSpareStack(q, q.next)) {
1730     if (updateCount)
1731     updateRunningCount(1);
1732     q.signal();
1733     return true;
1734     }
1735     }
1736     return false;
1737     }
1738    
1739     /**
1740     * Pops and resumes all spare threads. Same idea as ensureSync.
1741     *
1742     * @return true if any spares released
1743     */
1744     private boolean resumeAllSpares() {
1745     WaitQueueNode q;
1746     while ( (q = spareStack) != null) {
1747     if (casSpareStack(q, null)) {
1748     do {
1749     updateRunningCount(1);
1750     q.signal();
1751     } while ((q = q.next) != null);
1752     return true;
1753     }
1754     }
1755     return false;
1756     }
1757    
1758     /**
1759     * Pops and shuts down excessive spare threads. Call only while
1760     * holding lock. This is not guaranteed to eliminate all excess
1761     * threads, only those suspended as spares, which are the ones
1762     * unlikely to be needed in the future.
1763     */
1764     private void trimSpares() {
1765     int surplus = totalCountOf(workerCounts) - parallelism;
1766     WaitQueueNode q;
1767     while (surplus > 0 && (q = spareStack) != null) {
1768     if (casSpareStack(q, null)) {
1769     do {
1770     updateRunningCount(1);
1771     ForkJoinWorkerThread w = q.thread;
1772     if (w != null && surplus > 0 &&
1773     runningCountOf(workerCounts) > 0 && w.shutdown())
1774     --surplus;
1775     q.signal();
1776     } while ((q = q.next) != null);
1777     }
1778     }
1779     }
1780    
1781     /**
1782     * Interface for extending managed parallelism for tasks running
1783 jsr166 1.8 * in {@link ForkJoinPool}s.
1784     *
1785     * <p>A {@code ManagedBlocker} provides two methods.
1786 jsr166 1.4 * Method {@code isReleasable} must return {@code true} if
1787     * blocking is not necessary. Method {@code block} blocks the
1788     * current thread if necessary (perhaps internally invoking
1789 jsr166 1.8 * {@code isReleasable} before actually blocking).
1790 jsr166 1.1 *
1791     * <p>For example, here is a ManagedBlocker based on a
1792     * ReentrantLock:
1793     * <pre> {@code
1794     * class ManagedLocker implements ManagedBlocker {
1795     * final ReentrantLock lock;
1796     * boolean hasLock = false;
1797     * ManagedLocker(ReentrantLock lock) { this.lock = lock; }
1798     * public boolean block() {
1799     * if (!hasLock)
1800     * lock.lock();
1801     * return true;
1802     * }
1803     * public boolean isReleasable() {
1804     * return hasLock || (hasLock = lock.tryLock());
1805     * }
1806     * }}</pre>
1807     */
1808     public static interface ManagedBlocker {
1809     /**
1810     * Possibly blocks the current thread, for example waiting for
1811     * a lock or condition.
1812     *
1813 jsr166 1.4 * @return {@code true} if no additional blocking is necessary
1814     * (i.e., if isReleasable would return true)
1815 jsr166 1.1 * @throws InterruptedException if interrupted while waiting
1816     * (the method is not required to do so, but is allowed to)
1817     */
1818     boolean block() throws InterruptedException;
1819    
1820     /**
1821 jsr166 1.4 * Returns {@code true} if blocking is unnecessary.
1822 jsr166 1.1 */
1823     boolean isReleasable();
1824     }
1825    
1826     /**
1827     * Blocks in accord with the given blocker. If the current thread
1828 jsr166 1.8 * is a {@link ForkJoinWorkerThread}, this method possibly
1829     * arranges for a spare thread to be activated if necessary to
1830     * ensure parallelism while the current thread is blocked.
1831     *
1832     * <p>If {@code maintainParallelism} is {@code true} and the pool
1833     * supports it ({@link #getMaintainsParallelism}), this method
1834     * attempts to maintain the pool's nominal parallelism. Otherwise
1835     * it activates a thread only if necessary to avoid complete
1836     * starvation. This option may be preferable when blockages use
1837     * timeouts, or are almost always brief.
1838 jsr166 1.1 *
1839 jsr166 1.8 * <p>If the caller is not a {@link ForkJoinTask}, this method is
1840     * behaviorally equivalent to
1841 jsr166 1.1 * <pre> {@code
1842     * while (!blocker.isReleasable())
1843     * if (blocker.block())
1844     * return;
1845     * }</pre>
1846 jsr166 1.8 *
1847     * If the caller is a {@code ForkJoinTask}, then the pool may
1848     * first be expanded to ensure parallelism, and later adjusted.
1849 jsr166 1.1 *
1850     * @param blocker the blocker
1851 jsr166 1.4 * @param maintainParallelism if {@code true} and supported by
1852     * this pool, attempt to maintain the pool's nominal parallelism;
1853     * otherwise activate a thread only if necessary to avoid
1854     * complete starvation.
1855 jsr166 1.1 * @throws InterruptedException if blocker.block did so
1856     */
1857     public static void managedBlock(ManagedBlocker blocker,
1858     boolean maintainParallelism)
1859     throws InterruptedException {
1860     Thread t = Thread.currentThread();
1861     ForkJoinPool pool = ((t instanceof ForkJoinWorkerThread) ?
1862     ((ForkJoinWorkerThread) t).pool : null);
1863     if (!blocker.isReleasable()) {
1864     try {
1865     if (pool == null ||
1866     !pool.preBlock(blocker, maintainParallelism))
1867     awaitBlocker(blocker);
1868     } finally {
1869     if (pool != null)
1870     pool.updateRunningCount(1);
1871     }
1872     }
1873     }
1874    
1875     private static void awaitBlocker(ManagedBlocker blocker)
1876     throws InterruptedException {
1877     do {} while (!blocker.isReleasable() && !blocker.block());
1878     }
1879    
1880 jsr166 1.7 // AbstractExecutorService overrides. These rely on undocumented
1881     // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
1882     // implement RunnableFuture.
1883 jsr166 1.1
1884     protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
1885 jsr166 1.7 return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
1886 jsr166 1.1 }
1887    
1888     protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
1889 jsr166 1.7 return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
1890 jsr166 1.1 }
1891    
1892     // Unsafe mechanics
1893    
1894     private static final sun.misc.Unsafe UNSAFE = sun.misc.Unsafe.getUnsafe();
1895 jsr166 1.2 private static final long eventCountOffset =
1896 jsr166 1.3 objectFieldOffset("eventCount", ForkJoinPool.class);
1897 jsr166 1.2 private static final long workerCountsOffset =
1898 jsr166 1.3 objectFieldOffset("workerCounts", ForkJoinPool.class);
1899 jsr166 1.2 private static final long runControlOffset =
1900 jsr166 1.3 objectFieldOffset("runControl", ForkJoinPool.class);
1901 jsr166 1.2 private static final long syncStackOffset =
1902 jsr166 1.3 objectFieldOffset("syncStack",ForkJoinPool.class);
1903 jsr166 1.2 private static final long spareStackOffset =
1904 jsr166 1.3 objectFieldOffset("spareStack", ForkJoinPool.class);
1905 jsr166 1.1
1906     private boolean casEventCount(long cmp, long val) {
1907     return UNSAFE.compareAndSwapLong(this, eventCountOffset, cmp, val);
1908     }
1909     private boolean casWorkerCounts(int cmp, int val) {
1910     return UNSAFE.compareAndSwapInt(this, workerCountsOffset, cmp, val);
1911     }
1912     private boolean casRunControl(int cmp, int val) {
1913     return UNSAFE.compareAndSwapInt(this, runControlOffset, cmp, val);
1914     }
1915     private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
1916     return UNSAFE.compareAndSwapObject(this, spareStackOffset, cmp, val);
1917     }
1918     private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
1919     return UNSAFE.compareAndSwapObject(this, syncStackOffset, cmp, val);
1920     }
1921 jsr166 1.3
1922     private static long objectFieldOffset(String field, Class<?> klazz) {
1923     try {
1924     return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1925     } catch (NoSuchFieldException e) {
1926     // Convert Exception to corresponding Error
1927     NoSuchFieldError error = new NoSuchFieldError(field);
1928     error.initCause(e);
1929     throw error;
1930     }
1931     }
1932 jsr166 1.1 }