ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/ForkJoinTask.java
Revision: 1.88
Committed: Sun Mar 4 19:47:08 2012 UTC (12 years, 2 months ago) by dl
Branch: MAIN
Changes since 1.87: +1 -1 lines
Log Message:
typo

File Contents

# Content
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/publicdomain/zero/1.0/
5 */
6
7 package jsr166y;
8 import java.io.Serializable;
9 import java.util.Collection;
10 import java.util.List;
11 import java.util.RandomAccess;
12 import java.lang.ref.WeakReference;
13 import java.lang.ref.ReferenceQueue;
14 import java.util.concurrent.Callable;
15 import java.util.concurrent.CancellationException;
16 import java.util.concurrent.ExecutionException;
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.locks.ReentrantLock;
23 import java.lang.reflect.Constructor;
24
25 /**
26 * Abstract base class for tasks that run within a {@link ForkJoinPool}.
27 * A {@code ForkJoinTask} is a thread-like entity that is much
28 * lighter weight than a normal thread. Huge numbers of tasks and
29 * subtasks may be hosted by a small number of actual threads in a
30 * ForkJoinPool, at the price of some usage limitations.
31 *
32 * <p>A "main" {@code ForkJoinTask} begins execution when submitted
33 * to a {@link ForkJoinPool}. Once started, it will usually in turn
34 * start other subtasks. As indicated by the name of this class,
35 * many programs using {@code ForkJoinTask} employ only methods
36 * {@link #fork} and {@link #join}, or derivatives such as {@link
37 * #invokeAll(ForkJoinTask...) invokeAll}. However, this class also
38 * provides a number of other methods that can come into play in
39 * advanced usages, as well as extension mechanics that allow
40 * support of new forms of fork/join processing.
41 *
42 * <p>A {@code ForkJoinTask} is a lightweight form of {@link Future}.
43 * The efficiency of {@code ForkJoinTask}s stems from a set of
44 * restrictions (that are only partially statically enforceable)
45 * reflecting their main use as computational tasks calculating pure
46 * functions or operating on purely isolated objects. The primary
47 * coordination mechanisms are {@link #fork}, that arranges
48 * asynchronous execution, and {@link #join}, that doesn't proceed
49 * until the task's result has been computed. Computations should
50 * ideally avoid {@code synchronized} methods or blocks, and should
51 * minimize other blocking synchronization apart from joining other
52 * tasks or using synchronizers such as Phasers that are advertised to
53 * cooperate with fork/join scheduling. Subdividable tasks should also
54 * not perform blocking IO, and should ideally access variables that
55 * are completely independent of those accessed by other running
56 * tasks. These guidelines are loosely enforced by not permitting
57 * checked exceptions such as {@code IOExceptions} to be
58 * thrown. However, computations may still encounter unchecked
59 * exceptions, that are rethrown to callers attempting to join
60 * them. These exceptions may additionally include {@link
61 * RejectedExecutionException} stemming from internal resource
62 * exhaustion, such as failure to allocate internal task
63 * queues. Rethrown exceptions behave in the same way as regular
64 * exceptions, but, when possible, contain stack traces (as displayed
65 * for example using {@code ex.printStackTrace()}) of both the thread
66 * that initiated the computation as well as the thread actually
67 * encountering the exception; minimally only the latter.
68 *
69 * <p>It is possible to define and use ForkJoinTasks that may block,
70 * but doing do requires three further considerations: (1) Completion
71 * of few if any <em>other</em> tasks should be dependent on a task
72 * that blocks on external synchronization or IO. Event-style async
73 * tasks that are never joined often fall into this category. (2) To
74 * minimize resource impact, tasks should be small; ideally performing
75 * only the (possibly) blocking action. (3) Unless the {@link
76 * ForkJoinPool.ManagedBlocker} API is used, or the number of possibly
77 * blocked tasks is known to be less than the pool's {@link
78 * ForkJoinPool#getParallelism} level, the pool cannot guarantee that
79 * enough threads will be available to ensure progress or good
80 * performance.
81 *
82 * <p>The primary method for awaiting completion and extracting
83 * results of a task is {@link #join}, but there are several variants:
84 * The {@link Future#get} methods support interruptible and/or timed
85 * waits for completion and report results using {@code Future}
86 * conventions. Method {@link #invoke} is semantically
87 * equivalent to {@code fork(); join()} but always attempts to begin
88 * execution in the current thread. The "<em>quiet</em>" forms of
89 * these methods do not extract results or report exceptions. These
90 * may be useful when a set of tasks are being executed, and you need
91 * to delay processing of results or exceptions until all complete.
92 * Method {@code invokeAll} (available in multiple versions)
93 * performs the most common form of parallel invocation: forking a set
94 * of tasks and joining them all.
95 *
96 * <p>In the most typical usages, a fork-join pair act like a call
97 * (fork) and return (join) from a parallel recursive function. As is
98 * the case with other forms of recursive calls, returns (joins)
99 * should be performed innermost-first. For example, {@code a.fork();
100 * b.fork(); b.join(); a.join();} is likely to be substantially more
101 * efficient than joining {@code a} before {@code b}.
102 *
103 * <p>The execution status of tasks may be queried at several levels
104 * of detail: {@link #isDone} is true if a task completed in any way
105 * (including the case where a task was cancelled without executing);
106 * {@link #isCompletedNormally} is true if a task completed without
107 * cancellation or encountering an exception; {@link #isCancelled} is
108 * true if the task was cancelled (in which case {@link #getException}
109 * returns a {@link java.util.concurrent.CancellationException}); and
110 * {@link #isCompletedAbnormally} is true if a task was either
111 * cancelled or encountered an exception, in which case {@link
112 * #getException} will return either the encountered exception or
113 * {@link java.util.concurrent.CancellationException}.
114 *
115 * <p>The ForkJoinTask class is not usually directly subclassed.
116 * Instead, you subclass one of the abstract classes that support a
117 * particular style of fork/join processing, typically {@link
118 * RecursiveAction} for computations that do not return results, or
119 * {@link RecursiveTask} for those that do. Normally, a concrete
120 * ForkJoinTask subclass declares fields comprising its parameters,
121 * established in a constructor, and then defines a {@code compute}
122 * method that somehow uses the control methods supplied by this base
123 * class. While these methods have {@code public} access (to allow
124 * instances of different task subclasses to call each other's
125 * methods), some of them may only be called from within other
126 * ForkJoinTasks (as may be determined using method {@link
127 * #inForkJoinPool}). Attempts to invoke them in other contexts
128 * result in exceptions or errors, possibly including
129 * {@code ClassCastException}.
130 *
131 * <p>Method {@link #join} and its variants are appropriate for use
132 * only when completion dependencies are acyclic; that is, the
133 * parallel computation can be described as a directed acyclic graph
134 * (DAG). Otherwise, executions may encounter a form of deadlock as
135 * tasks cyclically wait for each other. However, this framework
136 * supports other methods and techniques (for example the use of
137 * {@link Phaser}, {@link #helpQuiesce}, and {@link #complete}) that
138 * may be of use in constructing custom subclasses for problems that
139 * are not statically structured as DAGs. To support such usages a
140 * ForkJoinTask may be atomically <em>tagged</em> with a {@code
141 * short} value using {@link #setForkJoinTaskTag} or {@link
142 * #compareAndSetForkJoinTaskTag} and checked using {@link
143 * #getForkJoinTaskTag}. The ForkJoinTask implementation does not
144 * use these {@code protected} methods or tags for any purpose, but
145 * they may be of use in the construction of specialized subclasses.
146 * For example, parallel graph traversals can use the supplied methods
147 * to avoid revisiting nodes/tasks that have already been processed.
148 * Also, completion based designs can use them to record that subtasks
149 * have completed. (Method names for tagging are bulky in part to
150 * encourage definition of methods that reflect their usage patterns.)
151 *
152 * <p>Most base support methods are {@code final}, to prevent
153 * overriding of implementations that are intrinsically tied to the
154 * underlying lightweight task scheduling framework. Developers
155 * creating new basic styles of fork/join processing should minimally
156 * implement {@code protected} methods {@link #exec}, {@link
157 * #setRawResult}, and {@link #getRawResult}, while also introducing
158 * an abstract computational method that can be implemented in its
159 * subclasses, possibly relying on other {@code protected} methods
160 * provided by this class.
161 *
162 * <p>ForkJoinTasks should perform relatively small amounts of
163 * computation. Large tasks should be split into smaller subtasks,
164 * usually via recursive decomposition. As a very rough rule of thumb,
165 * a task should perform more than 100 and less than 10000 basic
166 * computational steps, and should avoid indefinite looping. If tasks
167 * are too big, then parallelism cannot improve throughput. If too
168 * small, then memory and internal task maintenance overhead may
169 * overwhelm processing.
170 *
171 * <p>This class provides {@code adapt} methods for {@link Runnable}
172 * and {@link Callable}, that may be of use when mixing execution of
173 * {@code ForkJoinTasks} with other kinds of tasks. When all tasks are
174 * of this form, consider using a pool constructed in <em>asyncMode</em>.
175 *
176 * <p>ForkJoinTasks are {@code Serializable}, which enables them to be
177 * used in extensions such as remote execution frameworks. It is
178 * sensible to serialize tasks only before or after, but not during,
179 * execution. Serialization is not relied on during execution itself.
180 *
181 * @since 1.7
182 * @author Doug Lea
183 */
184 public abstract class ForkJoinTask<V> implements Future<V>, Serializable {
185
186 /*
187 * See the internal documentation of class ForkJoinPool for a
188 * general implementation overview. ForkJoinTasks are mainly
189 * responsible for maintaining their "status" field amidst relays
190 * to methods in ForkJoinWorkerThread and ForkJoinPool.
191 *
192 * The methods of this class are more-or-less layered into
193 * (1) basic status maintenance
194 * (2) execution and awaiting completion
195 * (3) user-level methods that additionally report results.
196 * This is sometimes hard to see because this file orders exported
197 * methods in a way that flows well in javadocs.
198 */
199
200 /*
201 * The status field holds run control status bits packed into a
202 * single int to minimize footprint and to ensure atomicity (via
203 * CAS). Status is initially zero, and takes on nonnegative
204 * values until completed, upon which status (anded with
205 * DONE_MASK) holds value NORMAL, CANCELLED, or EXCEPTIONAL. Tasks
206 * undergoing blocking waits by other threads have the SIGNAL bit
207 * set. Completion of a stolen task with SIGNAL set awakens any
208 * waiters via notifyAll. Even though suboptimal for some
209 * purposes, we use basic builtin wait/notify to take advantage of
210 * "monitor inflation" in JVMs that we would otherwise need to
211 * emulate to avoid adding further per-task bookkeeping overhead.
212 * We want these monitors to be "fat", i.e., not use biasing or
213 * thin-lock techniques, so use some odd coding idioms that tend
214 * to avoid them, mainly by arranging that every synchronized
215 * block performs a wait, notifyAll or both.
216 *
217 * These control bits occupy only (some of) the upper half (16
218 * bits) of status field. The lower bits are used for user-defined
219 * tags.
220 */
221
222 /** The run status of this task */
223 volatile int status; // accessed directly by pool and workers
224 static final int DONE_MASK = 0xf0000000; // mask out non-completion bits
225 static final int NORMAL = 0xf0000000; // must be negative
226 static final int CANCELLED = 0xc0000000; // must be < NORMAL
227 static final int EXCEPTIONAL = 0x80000000; // must be < CANCELLED
228 static final int SIGNAL = 0x00010000; // must be >= 1 << 16
229 static final int SMASK = 0x0000ffff; // short bits for tags
230
231 /**
232 * Marks completion and wakes up threads waiting to join this
233 * task.
234 *
235 * @param completion one of NORMAL, CANCELLED, EXCEPTIONAL
236 * @return completion status on exit
237 */
238 private int setCompletion(int completion) {
239 for (int s;;) {
240 if ((s = status) < 0)
241 return s;
242 if (U.compareAndSwapInt(this, STATUS, s, s | completion)) {
243 if ((s >>> 16) != 0)
244 synchronized (this) { notifyAll(); }
245 return completion;
246 }
247 }
248 }
249
250 /**
251 * Primary execution method for stolen tasks. Unless done, calls
252 * exec and records status if completed, but doesn't wait for
253 * completion otherwise.
254 *
255 * @return status on exit from this method
256 */
257 final int doExec() {
258 int s; boolean completed;
259 if ((s = status) >= 0) {
260 try {
261 completed = exec();
262 } catch (Throwable rex) {
263 return setExceptionalCompletion(rex);
264 }
265 if (completed)
266 s = setCompletion(NORMAL);
267 }
268 return s;
269 }
270
271 /**
272 * Tries to set SIGNAL status. Used by ForkJoinPool. Other
273 * variants are directly incorporated into externalAwaitDone etc.
274 *
275 * @return true if successful
276 */
277 final boolean trySetSignal() {
278 int s;
279 return U.compareAndSwapInt(this, STATUS, s = status, s | SIGNAL);
280 }
281
282 /**
283 * Blocks a non-worker-thread until completion.
284 * @return status upon completion
285 */
286 private int externalAwaitDone() {
287 boolean interrupted = false;
288 int s;
289 while ((s = status) >= 0) {
290 if (U.compareAndSwapInt(this, STATUS, s, s | SIGNAL)) {
291 synchronized (this) {
292 if (status >= 0) {
293 try {
294 wait();
295 } catch (InterruptedException ie) {
296 interrupted = true;
297 }
298 }
299 else
300 notifyAll();
301 }
302 }
303 }
304 if (interrupted)
305 Thread.currentThread().interrupt();
306 return s;
307 }
308
309 /**
310 * Blocks a non-worker-thread until completion or interruption.
311 */
312 private int externalInterruptibleAwaitDone() throws InterruptedException {
313 int s;
314 if (Thread.interrupted())
315 throw new InterruptedException();
316 while ((s = status) >= 0) {
317 if (U.compareAndSwapInt(this, STATUS, s, s | SIGNAL)) {
318 synchronized (this) {
319 if (status >= 0)
320 wait();
321 else
322 notifyAll();
323 }
324 }
325 }
326 return s;
327 }
328
329 /**
330 * Implementation for join, get, quietlyJoin. Directly handles
331 * only cases of already-completed, external wait, and
332 * unfork+exec. Others are relayed to ForkJoinPool.awaitJoin.
333 *
334 * @return status upon completion
335 */
336 private int doJoin() {
337 int s; Thread t; ForkJoinWorkerThread wt; ForkJoinPool.WorkQueue w;
338 if ((s = status) >= 0) {
339 if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
340 if (!(w = (wt = (ForkJoinWorkerThread)t).workQueue).
341 tryUnpush(this) || (s = doExec()) >= 0)
342 s = wt.pool.awaitJoin(w, this);
343 }
344 else
345 s = externalAwaitDone();
346 }
347 return s;
348 }
349
350 /**
351 * Implementation for invoke, quietlyInvoke.
352 *
353 * @return status upon completion
354 */
355 private int doInvoke() {
356 int s; Thread t; ForkJoinWorkerThread wt;
357 if ((s = doExec()) >= 0) {
358 if ((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)
359 s = (wt = (ForkJoinWorkerThread)t).pool.awaitJoin(wt.workQueue,
360 this);
361 else
362 s = externalAwaitDone();
363 }
364 return s;
365 }
366
367 // Exception table support
368
369 /**
370 * Table of exceptions thrown by tasks, to enable reporting by
371 * callers. Because exceptions are rare, we don't directly keep
372 * them with task objects, but instead use a weak ref table. Note
373 * that cancellation exceptions don't appear in the table, but are
374 * instead recorded as status values.
375 *
376 * Note: These statics are initialized below in static block.
377 */
378 private static final ExceptionNode[] exceptionTable;
379 private static final ReentrantLock exceptionTableLock;
380 private static final ReferenceQueue<Object> exceptionTableRefQueue;
381
382 /**
383 * Fixed capacity for exceptionTable.
384 */
385 private static final int EXCEPTION_MAP_CAPACITY = 32;
386
387 /**
388 * Key-value nodes for exception table. The chained hash table
389 * uses identity comparisons, full locking, and weak references
390 * for keys. The table has a fixed capacity because it only
391 * maintains task exceptions long enough for joiners to access
392 * them, so should never become very large for sustained
393 * periods. However, since we do not know when the last joiner
394 * completes, we must use weak references and expunge them. We do
395 * so on each operation (hence full locking). Also, some thread in
396 * any ForkJoinPool will call helpExpungeStaleExceptions when its
397 * pool becomes isQuiescent.
398 */
399 static final class ExceptionNode extends WeakReference<ForkJoinTask<?>> {
400 final Throwable ex;
401 ExceptionNode next;
402 final long thrower; // use id not ref to avoid weak cycles
403 ExceptionNode(ForkJoinTask<?> task, Throwable ex, ExceptionNode next) {
404 super(task, exceptionTableRefQueue);
405 this.ex = ex;
406 this.next = next;
407 this.thrower = Thread.currentThread().getId();
408 }
409 }
410
411 /**
412 * Records exception and sets exceptional completion.
413 *
414 * @return status on exit
415 */
416 private int setExceptionalCompletion(Throwable ex) {
417 int h = System.identityHashCode(this);
418 final ReentrantLock lock = exceptionTableLock;
419 lock.lock();
420 try {
421 expungeStaleExceptions();
422 ExceptionNode[] t = exceptionTable;
423 int i = h & (t.length - 1);
424 for (ExceptionNode e = t[i]; ; e = e.next) {
425 if (e == null) {
426 t[i] = new ExceptionNode(this, ex, t[i]);
427 break;
428 }
429 if (e.get() == this) // already present
430 break;
431 }
432 } finally {
433 lock.unlock();
434 }
435 return setCompletion(EXCEPTIONAL);
436 }
437
438 /**
439 * Cancels, ignoring any exceptions thrown by cancel. Used during
440 * worker and pool shutdown. Cancel is spec'ed not to throw any
441 * exceptions, but if it does anyway, we have no recourse during
442 * shutdown, so guard against this case.
443 */
444 static final void cancelIgnoringExceptions(ForkJoinTask<?> t) {
445 if (t != null && t.status >= 0) {
446 try {
447 t.cancel(false);
448 } catch (Throwable ignore) {
449 }
450 }
451 }
452
453 /**
454 * Removes exception node and clears status
455 */
456 private void clearExceptionalCompletion() {
457 int h = System.identityHashCode(this);
458 final ReentrantLock lock = exceptionTableLock;
459 lock.lock();
460 try {
461 ExceptionNode[] t = exceptionTable;
462 int i = h & (t.length - 1);
463 ExceptionNode e = t[i];
464 ExceptionNode pred = null;
465 while (e != null) {
466 ExceptionNode next = e.next;
467 if (e.get() == this) {
468 if (pred == null)
469 t[i] = next;
470 else
471 pred.next = next;
472 break;
473 }
474 pred = e;
475 e = next;
476 }
477 expungeStaleExceptions();
478 status = 0;
479 } finally {
480 lock.unlock();
481 }
482 }
483
484 /**
485 * Returns a rethrowable exception for the given task, if
486 * available. To provide accurate stack traces, if the exception
487 * was not thrown by the current thread, we try to create a new
488 * exception of the same type as the one thrown, but with the
489 * recorded exception as its cause. If there is no such
490 * constructor, we instead try to use a no-arg constructor,
491 * followed by initCause, to the same effect. If none of these
492 * apply, or any fail due to other exceptions, we return the
493 * recorded exception, which is still correct, although it may
494 * contain a misleading stack trace.
495 *
496 * @return the exception, or null if none
497 */
498 private Throwable getThrowableException() {
499 if ((status & DONE_MASK) != EXCEPTIONAL)
500 return null;
501 int h = System.identityHashCode(this);
502 ExceptionNode e;
503 final ReentrantLock lock = exceptionTableLock;
504 lock.lock();
505 try {
506 expungeStaleExceptions();
507 ExceptionNode[] t = exceptionTable;
508 e = t[h & (t.length - 1)];
509 while (e != null && e.get() != this)
510 e = e.next;
511 } finally {
512 lock.unlock();
513 }
514 Throwable ex;
515 if (e == null || (ex = e.ex) == null)
516 return null;
517 if (e.thrower != Thread.currentThread().getId()) {
518 Class<? extends Throwable> ec = ex.getClass();
519 try {
520 Constructor<?> noArgCtor = null;
521 Constructor<?>[] cs = ec.getConstructors();// public ctors only
522 for (int i = 0; i < cs.length; ++i) {
523 Constructor<?> c = cs[i];
524 Class<?>[] ps = c.getParameterTypes();
525 if (ps.length == 0)
526 noArgCtor = c;
527 else if (ps.length == 1 && ps[0] == Throwable.class)
528 return (Throwable)(c.newInstance(ex));
529 }
530 if (noArgCtor != null) {
531 Throwable wx = (Throwable)(noArgCtor.newInstance());
532 wx.initCause(ex);
533 return wx;
534 }
535 } catch (Exception ignore) {
536 }
537 }
538 return ex;
539 }
540
541 /**
542 * Poll stale refs and remove them. Call only while holding lock.
543 */
544 private static void expungeStaleExceptions() {
545 for (Object x; (x = exceptionTableRefQueue.poll()) != null;) {
546 if (x instanceof ExceptionNode) {
547 ForkJoinTask<?> key = ((ExceptionNode)x).get();
548 ExceptionNode[] t = exceptionTable;
549 int i = System.identityHashCode(key) & (t.length - 1);
550 ExceptionNode e = t[i];
551 ExceptionNode pred = null;
552 while (e != null) {
553 ExceptionNode next = e.next;
554 if (e == x) {
555 if (pred == null)
556 t[i] = next;
557 else
558 pred.next = next;
559 break;
560 }
561 pred = e;
562 e = next;
563 }
564 }
565 }
566 }
567
568 /**
569 * If lock is available, poll stale refs and remove them.
570 * Called from ForkJoinPool when pools become quiescent.
571 */
572 static final void helpExpungeStaleExceptions() {
573 final ReentrantLock lock = exceptionTableLock;
574 if (lock.tryLock()) {
575 try {
576 expungeStaleExceptions();
577 } finally {
578 lock.unlock();
579 }
580 }
581 }
582
583 /**
584 * Throws exception, if any, associated with the given status.
585 */
586 private void reportException(int s) {
587 Throwable ex = ((s == CANCELLED) ? new CancellationException() :
588 (s == EXCEPTIONAL) ? getThrowableException() :
589 null);
590 if (ex != null)
591 U.throwException(ex);
592 }
593
594 // public methods
595
596 /**
597 * Arranges to asynchronously execute this task. While it is not
598 * necessarily enforced, it is a usage error to fork a task more
599 * than once unless it has completed and been reinitialized.
600 * Subsequent modifications to the state of this task or any data
601 * it operates on are not necessarily consistently observable by
602 * any thread other than the one executing it unless preceded by a
603 * call to {@link #join} or related methods, or a call to {@link
604 * #isDone} returning {@code true}.
605 *
606 * <p>This method may be invoked only from within {@code
607 * ForkJoinPool} computations (as may be determined using method
608 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
609 * result in exceptions or errors, possibly including {@code
610 * ClassCastException}.
611 *
612 * @return {@code this}, to simplify usage
613 */
614 public final ForkJoinTask<V> fork() {
615 ((ForkJoinWorkerThread)Thread.currentThread()).workQueue.push(this);
616 return this;
617 }
618
619 /**
620 * Returns the result of the computation when it {@link #isDone is
621 * done}. This method differs from {@link #get()} in that
622 * abnormal completion results in {@code RuntimeException} or
623 * {@code Error}, not {@code ExecutionException}, and that
624 * interrupts of the calling thread do <em>not</em> cause the
625 * method to abruptly return by throwing {@code
626 * InterruptedException}.
627 *
628 * @return the computed result
629 */
630 public final V join() {
631 int s;
632 if ((s = doJoin() & DONE_MASK) != NORMAL)
633 reportException(s);
634 return getRawResult();
635 }
636
637 /**
638 * Commences performing this task, awaits its completion if
639 * necessary, and returns its result, or throws an (unchecked)
640 * {@code RuntimeException} or {@code Error} if the underlying
641 * computation did so.
642 *
643 * @return the computed result
644 */
645 public final V invoke() {
646 int s;
647 if ((s = doInvoke() & DONE_MASK) != NORMAL)
648 reportException(s);
649 return getRawResult();
650 }
651
652 /**
653 * Forks the given tasks, returning when {@code isDone} holds for
654 * each task or an (unchecked) exception is encountered, in which
655 * case the exception is rethrown. If more than one task
656 * encounters an exception, then this method throws any one of
657 * these exceptions. If any task encounters an exception, the
658 * other may be cancelled. However, the execution status of
659 * individual tasks is not guaranteed upon exceptional return. The
660 * status of each task may be obtained using {@link
661 * #getException()} and related methods to check if they have been
662 * cancelled, completed normally or exceptionally, or left
663 * unprocessed.
664 *
665 * <p>This method may be invoked only from within {@code
666 * ForkJoinPool} computations (as may be determined using method
667 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
668 * result in exceptions or errors, possibly including {@code
669 * ClassCastException}.
670 *
671 * @param t1 the first task
672 * @param t2 the second task
673 * @throws NullPointerException if any task is null
674 */
675 public static void invokeAll(ForkJoinTask<?> t1, ForkJoinTask<?> t2) {
676 int s1, s2;
677 t2.fork();
678 if ((s1 = t1.doInvoke() & DONE_MASK) != NORMAL)
679 t1.reportException(s1);
680 if ((s2 = t2.doJoin() & DONE_MASK) != NORMAL)
681 t2.reportException(s2);
682 }
683
684 /**
685 * Forks the given tasks, returning when {@code isDone} holds for
686 * each task or an (unchecked) exception is encountered, in which
687 * case the exception is rethrown. If more than one task
688 * encounters an exception, then this method throws any one of
689 * these exceptions. If any task encounters an exception, others
690 * may be cancelled. However, the execution status of individual
691 * tasks is not guaranteed upon exceptional return. The status of
692 * each task may be obtained using {@link #getException()} and
693 * related methods to check if they have been cancelled, completed
694 * normally or exceptionally, or left unprocessed.
695 *
696 * <p>This method may be invoked only from within {@code
697 * ForkJoinPool} computations (as may be determined using method
698 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
699 * result in exceptions or errors, possibly including {@code
700 * ClassCastException}.
701 *
702 * @param tasks the tasks
703 * @throws NullPointerException if any task is null
704 */
705 public static void invokeAll(ForkJoinTask<?>... tasks) {
706 Throwable ex = null;
707 int last = tasks.length - 1;
708 for (int i = last; i >= 0; --i) {
709 ForkJoinTask<?> t = tasks[i];
710 if (t == null) {
711 if (ex == null)
712 ex = new NullPointerException();
713 }
714 else if (i != 0)
715 t.fork();
716 else if (t.doInvoke() < NORMAL && ex == null)
717 ex = t.getException();
718 }
719 for (int i = 1; i <= last; ++i) {
720 ForkJoinTask<?> t = tasks[i];
721 if (t != null) {
722 if (ex != null)
723 t.cancel(false);
724 else if (t.doJoin() < NORMAL)
725 ex = t.getException();
726 }
727 }
728 if (ex != null)
729 U.throwException(ex);
730 }
731
732 /**
733 * Forks all tasks in the specified collection, returning when
734 * {@code isDone} holds for each task or an (unchecked) exception
735 * is encountered, in which case the exception is rethrown. If
736 * more than one task encounters an exception, then this method
737 * throws any one of these exceptions. If any task encounters an
738 * exception, others may be cancelled. However, the execution
739 * status of individual tasks is not guaranteed upon exceptional
740 * return. The status of each task may be obtained using {@link
741 * #getException()} and related methods to check if they have been
742 * cancelled, completed normally or exceptionally, or left
743 * unprocessed.
744 *
745 * <p>This method may be invoked only from within {@code
746 * ForkJoinPool} computations (as may be determined using method
747 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
748 * result in exceptions or errors, possibly including {@code
749 * ClassCastException}.
750 *
751 * @param tasks the collection of tasks
752 * @return the tasks argument, to simplify usage
753 * @throws NullPointerException if tasks or any element are null
754 */
755 public static <T extends ForkJoinTask<?>> Collection<T> invokeAll(Collection<T> tasks) {
756 if (!(tasks instanceof RandomAccess) || !(tasks instanceof List<?>)) {
757 invokeAll(tasks.toArray(new ForkJoinTask<?>[tasks.size()]));
758 return tasks;
759 }
760 @SuppressWarnings("unchecked")
761 List<? extends ForkJoinTask<?>> ts =
762 (List<? extends ForkJoinTask<?>>) tasks;
763 Throwable ex = null;
764 int last = ts.size() - 1;
765 for (int i = last; i >= 0; --i) {
766 ForkJoinTask<?> t = ts.get(i);
767 if (t == null) {
768 if (ex == null)
769 ex = new NullPointerException();
770 }
771 else if (i != 0)
772 t.fork();
773 else if (t.doInvoke() < NORMAL && ex == null)
774 ex = t.getException();
775 }
776 for (int i = 1; i <= last; ++i) {
777 ForkJoinTask<?> t = ts.get(i);
778 if (t != null) {
779 if (ex != null)
780 t.cancel(false);
781 else if (t.doJoin() < NORMAL)
782 ex = t.getException();
783 }
784 }
785 if (ex != null)
786 U.throwException(ex);
787 return tasks;
788 }
789
790 /**
791 * Attempts to cancel execution of this task. This attempt will
792 * fail if the task has already completed or could not be
793 * cancelled for some other reason. If successful, and this task
794 * has not started when {@code cancel} is called, execution of
795 * this task is suppressed. After this method returns
796 * successfully, unless there is an intervening call to {@link
797 * #reinitialize}, subsequent calls to {@link #isCancelled},
798 * {@link #isDone}, and {@code cancel} will return {@code true}
799 * and calls to {@link #join} and related methods will result in
800 * {@code CancellationException}.
801 *
802 * <p>This method may be overridden in subclasses, but if so, must
803 * still ensure that these properties hold. In particular, the
804 * {@code cancel} method itself must not throw exceptions.
805 *
806 * <p>This method is designed to be invoked by <em>other</em>
807 * tasks. To terminate the current task, you can just return or
808 * throw an unchecked exception from its computation method, or
809 * invoke {@link #completeExceptionally}.
810 *
811 * @param mayInterruptIfRunning this value has no effect in the
812 * default implementation because interrupts are not used to
813 * control cancellation.
814 *
815 * @return {@code true} if this task is now cancelled
816 */
817 public boolean cancel(boolean mayInterruptIfRunning) {
818 return (setCompletion(CANCELLED) & DONE_MASK) == CANCELLED;
819 }
820
821 public final boolean isDone() {
822 return status < 0;
823 }
824
825 public final boolean isCancelled() {
826 return (status & DONE_MASK) == CANCELLED;
827 }
828
829 /**
830 * Returns {@code true} if this task threw an exception or was cancelled.
831 *
832 * @return {@code true} if this task threw an exception or was cancelled
833 */
834 public final boolean isCompletedAbnormally() {
835 return status < NORMAL;
836 }
837
838 /**
839 * Returns {@code true} if this task completed without throwing an
840 * exception and was not cancelled.
841 *
842 * @return {@code true} if this task completed without throwing an
843 * exception and was not cancelled
844 */
845 public final boolean isCompletedNormally() {
846 return (status & DONE_MASK) == NORMAL;
847 }
848
849 /**
850 * Returns the exception thrown by the base computation, or a
851 * {@code CancellationException} if cancelled, or {@code null} if
852 * none or if the method has not yet completed.
853 *
854 * @return the exception, or {@code null} if none
855 */
856 public final Throwable getException() {
857 int s = status & DONE_MASK;
858 return ((s >= NORMAL) ? null :
859 (s == CANCELLED) ? new CancellationException() :
860 getThrowableException());
861 }
862
863 /**
864 * Completes this task abnormally, and if not already aborted or
865 * cancelled, causes it to throw the given exception upon
866 * {@code join} and related operations. This method may be used
867 * to induce exceptions in asynchronous tasks, or to force
868 * completion of tasks that would not otherwise complete. Its use
869 * in other situations is discouraged. This method is
870 * overridable, but overridden versions must invoke {@code super}
871 * implementation to maintain guarantees.
872 *
873 * @param ex the exception to throw. If this exception is not a
874 * {@code RuntimeException} or {@code Error}, the actual exception
875 * thrown will be a {@code RuntimeException} with cause {@code ex}.
876 */
877 public void completeExceptionally(Throwable ex) {
878 setExceptionalCompletion((ex instanceof RuntimeException) ||
879 (ex instanceof Error) ? ex :
880 new RuntimeException(ex));
881 }
882
883 /**
884 * Completes this task, and if not already aborted or cancelled,
885 * returning the given value as the result of subsequent
886 * invocations of {@code join} and related operations. This method
887 * may be used to provide results for asynchronous tasks, or to
888 * provide alternative handling for tasks that would not otherwise
889 * complete normally. Its use in other situations is
890 * discouraged. This method is overridable, but overridden
891 * versions must invoke {@code super} implementation to maintain
892 * guarantees.
893 *
894 * @param value the result value for this task
895 */
896 public void complete(V value) {
897 try {
898 setRawResult(value);
899 } catch (Throwable rex) {
900 setExceptionalCompletion(rex);
901 return;
902 }
903 setCompletion(NORMAL);
904 }
905
906 /**
907 * Completes this task. The most recent value established by
908 * {@link #setRawResult} (or {@code null}) will be returned as the
909 * result of subsequent invocations of {@code join} and related
910 * operations. This method may be useful when processing sets of
911 * tasks when some do not otherwise complete normally. Its use in
912 * other situations is discouraged.
913 */
914 public final void quietlyComplete() {
915 setCompletion(NORMAL);
916 }
917
918 /**
919 * Waits if necessary for the computation to complete, and then
920 * retrieves its result.
921 *
922 * @return the computed result
923 * @throws CancellationException if the computation was cancelled
924 * @throws ExecutionException if the computation threw an
925 * exception
926 * @throws InterruptedException if the current thread is not a
927 * member of a ForkJoinPool and was interrupted while waiting
928 */
929 public final V get() throws InterruptedException, ExecutionException {
930 int s = (Thread.currentThread() instanceof ForkJoinWorkerThread) ?
931 doJoin() : externalInterruptibleAwaitDone();
932 Throwable ex;
933 if ((s &= DONE_MASK) == CANCELLED)
934 throw new CancellationException();
935 if (s == EXCEPTIONAL && (ex = getThrowableException()) != null)
936 throw new ExecutionException(ex);
937 return getRawResult();
938 }
939
940 /**
941 * Waits if necessary for at most the given time for the computation
942 * to complete, and then retrieves its result, if available.
943 *
944 * @param timeout the maximum time to wait
945 * @param unit the time unit of the timeout argument
946 * @return the computed result
947 * @throws CancellationException if the computation was cancelled
948 * @throws ExecutionException if the computation threw an
949 * exception
950 * @throws InterruptedException if the current thread is not a
951 * member of a ForkJoinPool and was interrupted while waiting
952 * @throws TimeoutException if the wait timed out
953 */
954 public final V get(long timeout, TimeUnit unit)
955 throws InterruptedException, ExecutionException, TimeoutException {
956 if (Thread.interrupted())
957 throw new InterruptedException();
958 // Messy in part because we measure in nanosecs, but wait in millisecs
959 int s; long ns, ms;
960 if ((s = status) >= 0 && (ns = unit.toNanos(timeout)) > 0L) {
961 long deadline = System.nanoTime() + ns;
962 ForkJoinPool p = null;
963 ForkJoinPool.WorkQueue w = null;
964 Thread t = Thread.currentThread();
965 if (t instanceof ForkJoinWorkerThread) {
966 ForkJoinWorkerThread wt = (ForkJoinWorkerThread)t;
967 p = wt.pool;
968 w = wt.workQueue;
969 s = p.helpJoinOnce(w, this); // no retries on failure
970 }
971 boolean canBlock = false;
972 boolean interrupted = false;
973 try {
974 while ((s = status) >= 0) {
975 if (w != null && w.runState < 0)
976 cancelIgnoringExceptions(this);
977 else if (!canBlock) {
978 if (p == null || p.tryCompensate(this, null))
979 canBlock = true;
980 }
981 else {
982 if ((ms = TimeUnit.NANOSECONDS.toMillis(ns)) > 0L &&
983 U.compareAndSwapInt(this, STATUS, s, s | SIGNAL)) {
984 synchronized (this) {
985 if (status >= 0) {
986 try {
987 wait(ms);
988 } catch (InterruptedException ie) {
989 if (p == null)
990 interrupted = true;
991 }
992 }
993 else
994 notifyAll();
995 }
996 }
997 if ((s = status) < 0 || interrupted ||
998 (ns = deadline - System.nanoTime()) <= 0L)
999 break;
1000 }
1001 }
1002 } finally {
1003 if (p != null && canBlock)
1004 p.incrementActiveCount();
1005 }
1006 if (interrupted)
1007 throw new InterruptedException();
1008 }
1009 if ((s &= DONE_MASK) != NORMAL) {
1010 Throwable ex;
1011 if (s == CANCELLED)
1012 throw new CancellationException();
1013 if (s != EXCEPTIONAL)
1014 throw new TimeoutException();
1015 if ((ex = getThrowableException()) != null)
1016 throw new ExecutionException(ex);
1017 }
1018 return getRawResult();
1019 }
1020
1021 /**
1022 * Joins this task, without returning its result or throwing its
1023 * exception. This method may be useful when processing
1024 * collections of tasks when some have been cancelled or otherwise
1025 * known to have aborted.
1026 */
1027 public final void quietlyJoin() {
1028 doJoin();
1029 }
1030
1031 /**
1032 * Commences performing this task and awaits its completion if
1033 * necessary, without returning its result or throwing its
1034 * exception.
1035 */
1036 public final void quietlyInvoke() {
1037 doInvoke();
1038 }
1039
1040 /**
1041 * Possibly executes tasks until the pool hosting the current task
1042 * {@link ForkJoinPool#isQuiescent is quiescent}. This method may
1043 * be of use in designs in which many tasks are forked, but none
1044 * are explicitly joined, instead executing them until all are
1045 * processed.
1046 *
1047 * <p>This method may be invoked only from within {@code
1048 * ForkJoinPool} computations (as may be determined using method
1049 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1050 * result in exceptions or errors, possibly including {@code
1051 * ClassCastException}.
1052 */
1053 public static void helpQuiesce() {
1054 ForkJoinWorkerThread wt =
1055 (ForkJoinWorkerThread)Thread.currentThread();
1056 wt.pool.helpQuiescePool(wt.workQueue);
1057 }
1058
1059 /**
1060 * Resets the internal bookkeeping state of this task, allowing a
1061 * subsequent {@code fork}. This method allows repeated reuse of
1062 * this task, but only if reuse occurs when this task has either
1063 * never been forked, or has been forked, then completed and all
1064 * outstanding joins of this task have also completed. Effects
1065 * under any other usage conditions are not guaranteed.
1066 * This method may be useful when executing
1067 * pre-constructed trees of subtasks in loops.
1068 *
1069 * <p>Upon completion of this method, {@code isDone()} reports
1070 * {@code false}, and {@code getException()} reports {@code
1071 * null}. However, the value returned by {@code getRawResult} is
1072 * unaffected. To clear this value, you can invoke {@code
1073 * setRawResult(null)}.
1074 */
1075 public void reinitialize() {
1076 if ((status & DONE_MASK) == EXCEPTIONAL)
1077 clearExceptionalCompletion();
1078 else
1079 status = 0;
1080 }
1081
1082 /**
1083 * Returns the pool hosting the current task execution, or null
1084 * if this task is executing outside of any ForkJoinPool.
1085 *
1086 * @see #inForkJoinPool
1087 * @return the pool, or {@code null} if none
1088 */
1089 public static ForkJoinPool getPool() {
1090 Thread t = Thread.currentThread();
1091 return (t instanceof ForkJoinWorkerThread) ?
1092 ((ForkJoinWorkerThread) t).pool : null;
1093 }
1094
1095 /**
1096 * Returns {@code true} if the current thread is a {@link
1097 * ForkJoinWorkerThread} executing as a ForkJoinPool computation.
1098 *
1099 * @return {@code true} if the current thread is a {@link
1100 * ForkJoinWorkerThread} executing as a ForkJoinPool computation,
1101 * or {@code false} otherwise
1102 */
1103 public static boolean inForkJoinPool() {
1104 return Thread.currentThread() instanceof ForkJoinWorkerThread;
1105 }
1106
1107 /**
1108 * Tries to unschedule this task for execution. This method will
1109 * typically succeed if this task is the most recently forked task
1110 * by the current thread, and has not commenced executing in
1111 * another thread. This method may be useful when arranging
1112 * alternative local processing of tasks that could have been, but
1113 * were not, stolen.
1114 *
1115 * <p>This method may be invoked only from within {@code
1116 * ForkJoinPool} computations (as may be determined using method
1117 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1118 * result in exceptions or errors, possibly including {@code
1119 * ClassCastException}.
1120 *
1121 * @return {@code true} if unforked
1122 */
1123 public boolean tryUnfork() {
1124 return ((ForkJoinWorkerThread)Thread.currentThread())
1125 .workQueue.tryUnpush(this);
1126 }
1127
1128 /**
1129 * Returns an estimate of the number of tasks that have been
1130 * forked by the current worker thread but not yet executed. This
1131 * value may be useful for heuristic decisions about whether to
1132 * fork other tasks.
1133 *
1134 * <p>This method may be invoked only from within {@code
1135 * ForkJoinPool} computations (as may be determined using method
1136 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1137 * result in exceptions or errors, possibly including {@code
1138 * ClassCastException}.
1139 *
1140 * @return the number of tasks
1141 */
1142 public static int getQueuedTaskCount() {
1143 return ((ForkJoinWorkerThread) Thread.currentThread())
1144 .workQueue.queueSize();
1145 }
1146
1147 /**
1148 * Returns an estimate of how many more locally queued tasks are
1149 * held by the current worker thread than there are other worker
1150 * threads that might steal them. This value may be useful for
1151 * heuristic decisions about whether to fork other tasks. In many
1152 * usages of ForkJoinTasks, at steady state, each worker should
1153 * aim to maintain a small constant surplus (for example, 3) of
1154 * tasks, and to process computations locally if this threshold is
1155 * exceeded.
1156 *
1157 * <p>This method may be invoked only from within {@code
1158 * ForkJoinPool} computations (as may be determined using method
1159 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1160 * result in exceptions or errors, possibly including {@code
1161 * ClassCastException}.
1162 *
1163 * @return the surplus number of tasks, which may be negative
1164 */
1165 public static int getSurplusQueuedTaskCount() {
1166 /*
1167 * The aim of this method is to return a cheap heuristic guide
1168 * for task partitioning when programmers, frameworks, tools,
1169 * or languages have little or no idea about task granularity.
1170 * In essence by offering this method, we ask users only about
1171 * tradeoffs in overhead vs expected throughput and its
1172 * variance, rather than how finely to partition tasks.
1173 *
1174 * In a steady state strict (tree-structured) computation,
1175 * each thread makes available for stealing enough tasks for
1176 * other threads to remain active. Inductively, if all threads
1177 * play by the same rules, each thread should make available
1178 * only a constant number of tasks.
1179 *
1180 * The minimum useful constant is just 1. But using a value of
1181 * 1 would require immediate replenishment upon each steal to
1182 * maintain enough tasks, which is infeasible. Further,
1183 * partitionings/granularities of offered tasks should
1184 * minimize steal rates, which in general means that threads
1185 * nearer the top of computation tree should generate more
1186 * than those nearer the bottom. In perfect steady state, each
1187 * thread is at approximately the same level of computation
1188 * tree. However, producing extra tasks amortizes the
1189 * uncertainty of progress and diffusion assumptions.
1190 *
1191 * So, users will want to use values larger, but not much
1192 * larger than 1 to both smooth over transient shortages and
1193 * hedge against uneven progress; as traded off against the
1194 * cost of extra task overhead. We leave the user to pick a
1195 * threshold value to compare with the results of this call to
1196 * guide decisions, but recommend values such as 3.
1197 *
1198 * When all threads are active, it is on average OK to
1199 * estimate surplus strictly locally. In steady-state, if one
1200 * thread is maintaining say 2 surplus tasks, then so are
1201 * others. So we can just use estimated queue length.
1202 * However, this strategy alone leads to serious mis-estimates
1203 * in some non-steady-state conditions (ramp-up, ramp-down,
1204 * other stalls). We can detect many of these by further
1205 * considering the number of "idle" threads, that are known to
1206 * have zero queued tasks, so compensate by a factor of
1207 * (#idle/#active) threads.
1208 */
1209 ForkJoinWorkerThread wt =
1210 (ForkJoinWorkerThread)Thread.currentThread();
1211 return wt.workQueue.queueSize() - wt.pool.idlePerActive();
1212 }
1213
1214 // Extension methods
1215
1216 /**
1217 * Returns the result that would be returned by {@link #join}, even
1218 * if this task completed abnormally, or {@code null} if this task
1219 * is not known to have been completed. This method is designed
1220 * to aid debugging, as well as to support extensions. Its use in
1221 * any other context is discouraged.
1222 *
1223 * @return the result, or {@code null} if not completed
1224 */
1225 public abstract V getRawResult();
1226
1227 /**
1228 * Forces the given value to be returned as a result. This method
1229 * is designed to support extensions, and should not in general be
1230 * called otherwise.
1231 *
1232 * @param value the value
1233 */
1234 protected abstract void setRawResult(V value);
1235
1236 /**
1237 * Immediately performs the base action of this task. This method
1238 * is designed to support extensions, and should not in general be
1239 * called otherwise. The return value controls whether this task
1240 * is considered to be done normally. It may return false in
1241 * asynchronous actions that require explicit invocations of
1242 * {@link #complete} to become joinable. It may also throw an
1243 * (unchecked) exception to indicate abnormal exit.
1244 *
1245 * @return {@code true} if completed normally
1246 */
1247 protected abstract boolean exec();
1248
1249 /**
1250 * Returns, but does not unschedule or execute, a task queued by
1251 * the current thread but not yet executed, if one is immediately
1252 * available. There is no guarantee that this task will actually
1253 * be polled or executed next. Conversely, this method may return
1254 * null even if a task exists but cannot be accessed without
1255 * contention with other threads. This method is designed
1256 * primarily to support extensions, and is unlikely to be useful
1257 * otherwise.
1258 *
1259 * <p>This method may be invoked only from within {@code
1260 * ForkJoinPool} computations (as may be determined using method
1261 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1262 * result in exceptions or errors, possibly including {@code
1263 * ClassCastException}.
1264 *
1265 * @return the next task, or {@code null} if none are available
1266 */
1267 protected static ForkJoinTask<?> peekNextLocalTask() {
1268 return ((ForkJoinWorkerThread) Thread.currentThread()).workQueue.peek();
1269 }
1270
1271 /**
1272 * Unschedules and returns, without executing, the next task
1273 * queued by the current thread but not yet executed. This method
1274 * is designed primarily to support extensions, and is unlikely to
1275 * be useful otherwise.
1276 *
1277 * <p>This method may be invoked only from within {@code
1278 * ForkJoinPool} computations (as may be determined using method
1279 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1280 * result in exceptions or errors, possibly including {@code
1281 * ClassCastException}.
1282 *
1283 * @return the next task, or {@code null} if none are available
1284 */
1285 protected static ForkJoinTask<?> pollNextLocalTask() {
1286 return ((ForkJoinWorkerThread) Thread.currentThread())
1287 .workQueue.nextLocalTask();
1288 }
1289
1290 /**
1291 * Unschedules and returns, without executing, the next task
1292 * queued by the current thread but not yet executed, if one is
1293 * available, or if not available, a task that was forked by some
1294 * other thread, if available. Availability may be transient, so a
1295 * {@code null} result does not necessarily imply quiescence
1296 * of the pool this task is operating in. This method is designed
1297 * primarily to support extensions, and is unlikely to be useful
1298 * otherwise.
1299 *
1300 * <p>This method may be invoked only from within {@code
1301 * ForkJoinPool} computations (as may be determined using method
1302 * {@link #inForkJoinPool}). Attempts to invoke in other contexts
1303 * result in exceptions or errors, possibly including {@code
1304 * ClassCastException}.
1305 *
1306 * @return a task, or {@code null} if none are available
1307 */
1308 protected static ForkJoinTask<?> pollTask() {
1309 ForkJoinWorkerThread wt =
1310 (ForkJoinWorkerThread)Thread.currentThread();
1311 return wt.pool.nextTaskFor(wt.workQueue);
1312 }
1313
1314 // tag operations
1315
1316 /**
1317 * Returns the tag for this task.
1318 *
1319 * @return the tag for this task
1320 * @since 1.8
1321 */
1322 public final short getForkJoinTaskTag() {
1323 return (short)status;
1324 }
1325
1326 /**
1327 * Atomically sets the tag value for this task.
1328 *
1329 * @param tag the tag value
1330 * @return the previous value of the tag
1331 * @since 1.8
1332 */
1333 public final short setForkJoinTaskTag(short tag) {
1334 for (int s;;) {
1335 if (U.compareAndSwapInt(this, STATUS, s = status,
1336 (s & ~SMASK) | (tag & SMASK)))
1337 return (short)s;
1338 }
1339 }
1340
1341 /**
1342 * Atomically conditionally sets the tag value for this task.
1343 * Among other applications, tags can be used as visit markers
1344 * in tasks operating on graphs, as in methods that check: {@code
1345 * if (task.compareAndSetForkJoinTaskTag((short)0, (short)1))}
1346 * before processing, otherwise exiting because the node has
1347 * already been visited.
1348 *
1349 * @param e the expected tag value
1350 * @param tag the new tag value
1351 * @return true if successful; i.e., the current value was
1352 * equal to e and is now tag.
1353 * @since 1.8
1354 */
1355 public final boolean compareAndSetForkJoinTaskTag(short e, short tag) {
1356 for (int s;;) {
1357 if ((short)(s = status) != e)
1358 return false;
1359 if (U.compareAndSwapInt(this, STATUS, s,
1360 (s & ~SMASK) | (tag & SMASK)))
1361 return true;
1362 }
1363 }
1364
1365 /**
1366 * Adaptor for Runnables. This implements RunnableFuture
1367 * to be compliant with AbstractExecutorService constraints
1368 * when used in ForkJoinPool.
1369 */
1370 static final class AdaptedRunnable<T> extends ForkJoinTask<T>
1371 implements RunnableFuture<T> {
1372 final Runnable runnable;
1373 T result;
1374 AdaptedRunnable(Runnable runnable, T result) {
1375 if (runnable == null) throw new NullPointerException();
1376 this.runnable = runnable;
1377 this.result = result; // OK to set this even before completion
1378 }
1379 public final T getRawResult() { return result; }
1380 public final void setRawResult(T v) { result = v; }
1381 public final boolean exec() { runnable.run(); return true; }
1382 public final void run() { invoke(); }
1383 private static final long serialVersionUID = 5232453952276885070L;
1384 }
1385
1386 /**
1387 * Adaptor for Runnables without results
1388 */
1389 static final class AdaptedRunnableAction extends ForkJoinTask<Void>
1390 implements RunnableFuture<Void> {
1391 final Runnable runnable;
1392 AdaptedRunnableAction(Runnable runnable) {
1393 if (runnable == null) throw new NullPointerException();
1394 this.runnable = runnable;
1395 }
1396 public final Void getRawResult() { return null; }
1397 public final void setRawResult(Void v) { }
1398 public final boolean exec() { runnable.run(); return true; }
1399 public final void run() { invoke(); }
1400 private static final long serialVersionUID = 5232453952276885070L;
1401 }
1402
1403 /**
1404 * Adaptor for Callables
1405 */
1406 static final class AdaptedCallable<T> extends ForkJoinTask<T>
1407 implements RunnableFuture<T> {
1408 final Callable<? extends T> callable;
1409 T result;
1410 AdaptedCallable(Callable<? extends T> callable) {
1411 if (callable == null) throw new NullPointerException();
1412 this.callable = callable;
1413 }
1414 public final T getRawResult() { return result; }
1415 public final void setRawResult(T v) { result = v; }
1416 public final boolean exec() {
1417 try {
1418 result = callable.call();
1419 return true;
1420 } catch (Error err) {
1421 throw err;
1422 } catch (RuntimeException rex) {
1423 throw rex;
1424 } catch (Exception ex) {
1425 throw new RuntimeException(ex);
1426 }
1427 }
1428 public final void run() { invoke(); }
1429 private static final long serialVersionUID = 2838392045355241008L;
1430 }
1431
1432 /**
1433 * Returns a new {@code ForkJoinTask} that performs the {@code run}
1434 * method of the given {@code Runnable} as its action, and returns
1435 * a null result upon {@link #join}.
1436 *
1437 * @param runnable the runnable action
1438 * @return the task
1439 */
1440 public static ForkJoinTask<?> adapt(Runnable runnable) {
1441 return new AdaptedRunnableAction(runnable);
1442 }
1443
1444 /**
1445 * Returns a new {@code ForkJoinTask} that performs the {@code run}
1446 * method of the given {@code Runnable} as its action, and returns
1447 * the given result upon {@link #join}.
1448 *
1449 * @param runnable the runnable action
1450 * @param result the result upon completion
1451 * @return the task
1452 */
1453 public static <T> ForkJoinTask<T> adapt(Runnable runnable, T result) {
1454 return new AdaptedRunnable<T>(runnable, result);
1455 }
1456
1457 /**
1458 * Returns a new {@code ForkJoinTask} that performs the {@code call}
1459 * method of the given {@code Callable} as its action, and returns
1460 * its result upon {@link #join}, translating any checked exceptions
1461 * encountered into {@code RuntimeException}.
1462 *
1463 * @param callable the callable action
1464 * @return the task
1465 */
1466 public static <T> ForkJoinTask<T> adapt(Callable<? extends T> callable) {
1467 return new AdaptedCallable<T>(callable);
1468 }
1469
1470 // Serialization support
1471
1472 private static final long serialVersionUID = -7721805057305804111L;
1473
1474 /**
1475 * Saves this task to a stream (that is, serializes it).
1476 *
1477 * @serialData the current run status and the exception thrown
1478 * during execution, or {@code null} if none
1479 */
1480 private void writeObject(java.io.ObjectOutputStream s)
1481 throws java.io.IOException {
1482 s.defaultWriteObject();
1483 s.writeObject(getException());
1484 }
1485
1486 /**
1487 * Reconstitutes this task from a stream (that is, deserializes it).
1488 */
1489 private void readObject(java.io.ObjectInputStream s)
1490 throws java.io.IOException, ClassNotFoundException {
1491 s.defaultReadObject();
1492 Object ex = s.readObject();
1493 if (ex != null)
1494 setExceptionalCompletion((Throwable)ex);
1495 }
1496
1497 // Unsafe mechanics
1498 private static final sun.misc.Unsafe U;
1499 private static final long STATUS;
1500 static {
1501 exceptionTableLock = new ReentrantLock();
1502 exceptionTableRefQueue = new ReferenceQueue<Object>();
1503 exceptionTable = new ExceptionNode[EXCEPTION_MAP_CAPACITY];
1504 try {
1505 U = getUnsafe();
1506 STATUS = U.objectFieldOffset
1507 (ForkJoinTask.class.getDeclaredField("status"));
1508 } catch (Exception e) {
1509 throw new Error(e);
1510 }
1511 }
1512
1513 /**
1514 * Returns a sun.misc.Unsafe. Suitable for use in a 3rd party package.
1515 * Replace with a simple call to Unsafe.getUnsafe when integrating
1516 * into a jdk.
1517 *
1518 * @return a sun.misc.Unsafe
1519 */
1520 private static sun.misc.Unsafe getUnsafe() {
1521 try {
1522 return sun.misc.Unsafe.getUnsafe();
1523 } catch (SecurityException se) {
1524 try {
1525 return java.security.AccessController.doPrivileged
1526 (new java.security
1527 .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1528 public sun.misc.Unsafe run() throws Exception {
1529 java.lang.reflect.Field f = sun.misc
1530 .Unsafe.class.getDeclaredField("theUnsafe");
1531 f.setAccessible(true);
1532 return (sun.misc.Unsafe) f.get(null);
1533 }});
1534 } catch (java.security.PrivilegedActionException e) {
1535 throw new RuntimeException("Could not initialize intrinsics",
1536 e.getCause());
1537 }
1538 }
1539 }
1540 }