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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.60 by dl, Sat Jul 24 20:28:18 2010 UTC vs.
Revision 1.83 by dl, Sun Oct 24 19:37:26 2010 UTC

# Line 6 | Line 6
6  
7   package jsr166y;
8  
9 import java.util.concurrent.*;
10
9   import java.util.ArrayList;
10   import java.util.Arrays;
11   import java.util.Collection;
12   import java.util.Collections;
13   import java.util.List;
14 + import java.util.concurrent.AbstractExecutorService;
15 + import java.util.concurrent.Callable;
16 + import java.util.concurrent.ExecutorService;
17 + import java.util.concurrent.Future;
18 + import java.util.concurrent.RejectedExecutionException;
19 + import java.util.concurrent.RunnableFuture;
20 + import java.util.concurrent.TimeUnit;
21 + import java.util.concurrent.TimeoutException;
22 + import java.util.concurrent.atomic.AtomicInteger;
23   import java.util.concurrent.locks.LockSupport;
24   import java.util.concurrent.locks.ReentrantLock;
18 import java.util.concurrent.atomic.AtomicInteger;
19 import java.util.concurrent.CountDownLatch;
25  
26   /**
27   * An {@link ExecutorService} for running {@link ForkJoinTask}s.
# Line 69 | Line 74 | import java.util.concurrent.CountDownLat
74   *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
75   *  </tr>
76   *  <tr>
77 < *    <td> <b>Arange async execution</td>
77 > *    <td> <b>Arrange async execution</td>
78   *    <td> {@link #execute(ForkJoinTask)}</td>
79   *    <td> {@link ForkJoinTask#fork}</td>
80   *  </tr>
# Line 110 | Line 115 | import java.util.concurrent.CountDownLat
115   *
116   * <p>This implementation rejects submitted tasks (that is, by throwing
117   * {@link RejectedExecutionException}) only when the pool is shut down
118 < * or internal resources have been exhuasted.
118 > * or internal resources have been exhausted.
119   *
120   * @since 1.7
121   * @author Doug Lea
# Line 140 | Line 145 | public class ForkJoinPool extends Abstra
145       * Beyond work-stealing support and essential bookkeeping, the
146       * main responsibility of this framework is to take actions when
147       * one worker is waiting to join a task stolen (or always held by)
148 <     * another.  Becauae we are multiplexing many tasks on to a pool
148 >     * another.  Because we are multiplexing many tasks on to a pool
149       * of workers, we can't just let them block (as in Thread.join).
150       * We also cannot just reassign the joiner's run-time stack with
151       * another and replace it later, which would be a form of
# Line 156 | Line 161 | public class ForkJoinPool extends Abstra
161       *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
162       *      links to try to find such a task.
163       *
164 <     *   Compensating: Unless there are already enough live threads,
165 <     *      creating or or re-activating a spare thread to compensate
166 <     *      for the (blocked) joiner until it unblocks.  Spares then
167 <     *      suspend at their next opportunity or eventually die if
168 <     *      unused for too long.  See below and the internal
169 <     *      documentation for tryAwaitJoin for more details about
170 <     *      compensation rules.
171 <     *
172 <     * Because the determining existence of conservatively safe
173 <     * helping targets, the availability of already-created spares,
174 <     * and the apparent need to create new spares are all racy and
175 <     * require heuristic guidance, joins (in
176 <     * ForkJoinWorkerThread.joinTask) interleave these options until
177 <     * successful.  Creating a new spare always succeeds, but also
178 <     * increases application footprint, so we try to avoid it, within
179 <     * reason.
164 >     *   Compensating: Unless there are already enough live threads,
165 >     *      method helpMaintainParallelism() may create or
166 >     *      re-activate a spare thread to compensate for blocked
167 >     *      joiners until they unblock.
168 >     *
169 >     * It is impossible to keep exactly the target (parallelism)
170 >     * number of threads running at any given time.  Determining
171 >     * existence of conservatively safe helping targets, the
172 >     * availability of already-created spares, and the apparent need
173 >     * to create new spares are all racy and require heuristic
174 >     * guidance, so we rely on multiple retries of each.  Compensation
175 >     * occurs in slow-motion. It is triggered only upon timeouts of
176 >     * Object.wait used for joins. This reduces poor decisions that
177 >     * would otherwise be made when threads are waiting for others
178 >     * that are stalled because of unrelated activities such as
179 >     * garbage collection.
180       *
181 <     * The ManagedBlocker extension API can't use helping so uses a
182 <     * special version of compensation in method awaitBlocker.
181 >     * The ManagedBlocker extension API can't use helping so relies
182 >     * only on compensation in method awaitBlocker.
183       *
184       * The main throughput advantages of work-stealing stem from
185       * decentralized control -- workers mostly steal tasks from each
# Line 207 | Line 212 | public class ForkJoinPool extends Abstra
212       * blocked workers. However, all other support code is set up to
213       * work with other policies.
214       *
215 +     * To ensure that we do not hold on to worker references that
216 +     * would prevent GC, ALL accesses to workers are via indices into
217 +     * the workers array (which is one source of some of the unusual
218 +     * code constructions here). In essence, the workers array serves
219 +     * as a WeakReference mechanism. Thus for example the event queue
220 +     * stores worker indices, not worker references. Access to the
221 +     * workers in associated methods (for example releaseEventWaiters)
222 +     * must both index-check and null-check the IDs. All such accesses
223 +     * ignore bad IDs by returning out early from what they are doing,
224 +     * since this can only be associated with shutdown, in which case
225 +     * it is OK to give up. On termination, we just clobber these
226 +     * data structures without trying to use them.
227 +     *
228       * 2. Bookkeeping for dynamically adding and removing workers. We
229       * aim to approximately maintain the given level of parallelism.
230       * When some workers are known to be blocked (on joins or via
231       * ManagedBlocker), we may create or resume others to take their
232       * place until they unblock (see below). Implementing this
233       * requires counts of the number of "running" threads (i.e., those
234 <     * that are neither blocked nor artifically suspended) as well as
234 >     * that are neither blocked nor artificially suspended) as well as
235       * the total number.  These two values are packed into one field,
236       * "workerCounts" because we need accurate snapshots when deciding
237       * to create, resume or suspend.  Note however that the
238 <     * correspondance of these counts to reality is not guaranteed. In
238 >     * correspondence of these counts to reality is not guaranteed. In
239       * particular updates for unblocked threads may lag until they
240       * actually wake up.
241       *
# Line 248 | Line 266 | public class ForkJoinPool extends Abstra
266       * workers that previously could not find a task to now find one:
267       * Submission of a new task to the pool, or another worker pushing
268       * a task onto a previously empty queue.  (We also use this
269 <     * mechanism for termination and reconfiguration actions that
269 >     * mechanism for configuration and termination actions that
270       * require wakeups of idle workers).  Each worker maintains its
271       * last known event count, and blocks when a scan for work did not
272       * find a task AND its lastEventCount matches the current
# Line 259 | Line 277 | public class ForkJoinPool extends Abstra
277       * a record (field nextEventWaiter) for the next waiting worker.
278       * In addition to allowing simpler decisions about need for
279       * wakeup, the event count bits in eventWaiters serve the role of
280 <     * tags to avoid ABA errors in Treiber stacks.  To reduce delays
281 <     * in task diffusion, workers not otherwise occupied may invoke
282 <     * method releaseWaiters, that removes and signals (unparks)
283 <     * workers not waiting on current count. To minimize task
284 <     * production stalls associate with signalling, any worker pushing
285 <     * a task on an empty queue invokes the weaker method signalWork,
268 <     * that only releases idle workers until it detects interference
269 <     * by other threads trying to release, and lets them take
270 <     * over. The net effect is a tree-like diffusion of signals, where
271 <     * released threads (and possibly others) help with unparks.  To
272 <     * further reduce contention effects a bit, failed CASes to
273 <     * increment field eventCount are tolerated without retries.
280 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
281 >     * released threads also try to release at most two others.  The
282 >     * net effect is a tree-like diffusion of signals, where released
283 >     * threads (and possibly others) help with unparks.  To further
284 >     * reduce contention effects a bit, failed CASes to increment
285 >     * field eventCount are tolerated without retries in signalWork.
286       * Conceptually they are merged into the same event, which is OK
287       * when their only purpose is to enable workers to scan for work.
288       *
289 <     * 5. Managing suspension of extra workers. When a worker is about
290 <     * to block waiting for a join (or via ManagedBlockers), we may
291 <     * create a new thread to maintain parallelism level, or at least
292 <     * avoid starvation. Usually, extra threads are needed for only
293 <     * very short periods, yet join dependencies are such that we
294 <     * sometimes need them in bursts. Rather than create new threads
295 <     * each time this happens, we suspend no-longer-needed extra ones
296 <     * as "spares". For most purposes, we don't distinguish "extra"
297 <     * spare threads from normal "core" threads: On each call to
298 <     * preStep (the only point at which we can do this) a worker
299 <     * checks to see if there are now too many running workers, and if
300 <     * so, suspends itself.  Methods tryAwaitJoin and awaitBlocker
301 <     * look for suspended threads to resume before considering
302 <     * creating a new replacement. We don't need a special data
303 <     * structure to maintain spares; simply scanning the workers array
304 <     * looking for worker.isSuspended() is fine because the calling
305 <     * thread is otherwise not doing anything useful anyway; we are at
306 <     * least as happy if after locating a spare, the caller doesn't
307 <     * actually block because the join is ready before we try to
308 <     * adjust and compensate.  Note that this is intrinsically racy.
309 <     * One thread may become a spare at about the same time as another
310 <     * is needlessly being created. We counteract this and related
311 <     * slop in part by requiring resumed spares to immediately recheck
312 <     * (in preStep) to see whether they they should re-suspend. The
313 <     * only effective difference between "extra" and "core" threads is
314 <     * that we allow the "extra" ones to time out and die if they are
315 <     * not resumed within a keep-alive interval of a few seconds. This
316 <     * is implemented mainly within ForkJoinWorkerThread, but requires
317 <     * some coordination (isTrimmed() -- meaning killed while
318 <     * suspended) to correctly maintain pool counts.
319 <     *
320 <     * 6. Deciding when to create new workers. The main dynamic
321 <     * control in this class is deciding when to create extra threads,
322 <     * in methods awaitJoin and awaitBlocker. We always need to create
323 <     * one when the number of running threads would become zero and
324 <     * all workers are busy. However, this is not easy to detect
325 <     * reliably in the presence of transients so we use retries and
326 <     * allow slack (in tryAwaitJoin) to reduce false alarms.  These
327 <     * effectively reduce churn at the price of systematically
328 <     * undershooting target parallelism when many threads are blocked.
329 <     * However, biasing toward undeshooting partially compensates for
330 <     * the above mechanics to suspend extra threads, that normally
331 <     * lead to overshoot because we can only suspend workers
332 <     * in-between top-level actions. It also better copes with the
333 <     * fact that some of the methods in this class tend to never
334 <     * become compiled (but are interpreted), so some components of
335 <     * the entire set of controls might execute many times faster than
336 <     * others. And similarly for cases where the apparent lack of work
337 <     * is just due to GC stalls and other transient system activity.
289 >     * 5. Managing suspension of extra workers. When a worker notices
290 >     * (usually upon timeout of a wait()) that there are too few
291 >     * running threads, we may create a new thread to maintain
292 >     * parallelism level, or at least avoid starvation. Usually, extra
293 >     * threads are needed for only very short periods, yet join
294 >     * dependencies are such that we sometimes need them in
295 >     * bursts. Rather than create new threads each time this happens,
296 >     * we suspend no-longer-needed extra ones as "spares". For most
297 >     * purposes, we don't distinguish "extra" spare threads from
298 >     * normal "core" threads: On each call to preStep (the only point
299 >     * at which we can do this) a worker checks to see if there are
300 >     * now too many running workers, and if so, suspends itself.
301 >     * Method helpMaintainParallelism looks for suspended threads to
302 >     * resume before considering creating a new replacement. The
303 >     * spares themselves are encoded on another variant of a Treiber
304 >     * Stack, headed at field "spareWaiters".  Note that the use of
305 >     * spares is intrinsically racy.  One thread may become a spare at
306 >     * about the same time as another is needlessly being created. We
307 >     * counteract this and related slop in part by requiring resumed
308 >     * spares to immediately recheck (in preStep) to see whether they
309 >     * should re-suspend.
310 >     *
311 >     * 6. Killing off unneeded workers. A timeout mechanism is used to
312 >     * shed unused workers: The oldest (first) event queue waiter uses
313 >     * a timed rather than hard wait. When this wait times out without
314 >     * a normal wakeup, it tries to shutdown any one (for convenience
315 >     * the newest) other spare or event waiter via
316 >     * tryShutdownUnusedWorker. This eventually reduces the number of
317 >     * worker threads to a minimum of one after a long enough period
318 >     * without use.
319 >     *
320 >     * 7. Deciding when to create new workers. The main dynamic
321 >     * control in this class is deciding when to create extra threads
322 >     * in method helpMaintainParallelism. We would like to keep
323 >     * exactly #parallelism threads running, which is an impossible
324 >     * task. We always need to create one when the number of running
325 >     * threads would become zero and all workers are busy. Beyond
326 >     * this, we must rely on heuristics that work well in the
327 >     * presence of transient phenomena such as GC stalls, dynamic
328 >     * compilation, and wake-up lags. These transients are extremely
329 >     * common -- we are normally trying to fully saturate the CPUs on
330 >     * a machine, so almost any activity other than running tasks
331 >     * impedes accuracy. Our main defense is to allow parallelism to
332 >     * lapse for a while during joins, and use a timeout to see if,
333 >     * after the resulting settling, there is still a need for
334 >     * additional workers.  This also better copes with the fact that
335 >     * some of the methods in this class tend to never become compiled
336 >     * (but are interpreted), so some components of the entire set of
337 >     * controls might execute 100 times faster than others. And
338 >     * similarly for cases where the apparent lack of work is just due
339 >     * to GC stalls and other transient system activity.
340       *
341       * Beware that there is a lot of representation-level coupling
342       * among classes ForkJoinPool, ForkJoinWorkerThread, and
# Line 335 | Line 349 | public class ForkJoinPool extends Abstra
349       *
350       * Style notes: There are lots of inline assignments (of form
351       * "while ((local = field) != 0)") which are usually the simplest
352 <     * way to ensure read orderings. Also several occurrences of the
353 <     * unusual "do {} while(!cas...)" which is the simplest way to
354 <     * force an update of a CAS'ed variable. There are also other
355 <     * coding oddities that help some methods perform reasonably even
356 <     * when interpreted (not compiled), at the expense of messiness.
352 >     * way to ensure the required read orderings (which are sometimes
353 >     * critical). Also several occurrences of the unusual "do {}
354 >     * while (!cas...)" which is the simplest way to force an update of
355 >     * a CAS'ed variable. There are also other coding oddities that
356 >     * help some methods perform reasonably even when interpreted (not
357 >     * compiled), at the expense of some messy constructions that
358 >     * reduce byte code counts.
359       *
360       * The order of declarations in this file is: (1) statics (2)
361       * fields (along with constants used when unpacking some of them)
# Line 407 | Line 423 | public class ForkJoinPool extends Abstra
423          new AtomicInteger();
424  
425      /**
426 <     * Absolute bound for parallelism level. Twice this number must
427 <     * fit into a 16bit field to enable word-packing for some counts.
426 >     * The time to block in a join (see awaitJoin) before checking if
427 >     * a new worker should be (re)started to maintain parallelism
428 >     * level. The value should be short enough to maintain global
429 >     * responsiveness and progress but long enough to avoid
430 >     * counterproductive firings during GC stalls or unrelated system
431 >     * activity, and to not bog down systems with continual re-firings
432 >     * on GCs or legitimately long waits.
433 >     */
434 >    private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
435 >
436 >    /**
437 >     * The wakeup interval (in nanoseconds) for the oldest worker
438 >     * waiting for an event to invoke tryShutdownUnusedWorker to
439 >     * shrink the number of workers.  The exact value does not matter
440 >     * too much. It must be short enough to release resources during
441 >     * sustained periods of idleness, but not so short that threads
442 >     * are continually re-created.
443 >     */
444 >    private static final long SHRINK_RATE_NANOS =
445 >        30L * 1000L * 1000L * 1000L; // 2 per minute
446 >
447 >    /**
448 >     * Absolute bound for parallelism level. Twice this number plus
449 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
450 >     * word-packing for some counts and indices.
451       */
452 <    private static final int MAX_THREADS = 0x7fff;
452 >    private static final int MAX_WORKERS   = 0x7fff;
453  
454      /**
455       * Array holding all worker threads in the pool.  Array size must
# Line 450 | Line 489 | public class ForkJoinPool extends Abstra
489      private volatile long stealCount;
490  
491      /**
492 <     * Encoded record of top of treiber stack of threads waiting for
492 >     * Encoded record of top of Treiber stack of threads waiting for
493       * events. The top 32 bits contain the count being waited for. The
494 <     * bottom word contains one plus the pool index of waiting worker
495 <     * thread.
494 >     * bottom 16 bits contains one plus the pool index of waiting
495 >     * worker thread. (Bits 16-31 are unused.)
496       */
497      private volatile long eventWaiters;
498  
499      private static final int  EVENT_COUNT_SHIFT = 32;
500 <    private static final long WAITER_ID_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
500 >    private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
501  
502      /**
503       * A counter for events that may wake up worker threads:
504       *   - Submission of a new task to the pool
505       *   - A worker pushing a task on an empty queue
506 <     *   - termination and reconfiguration
506 >     *   - termination
507       */
508      private volatile int eventCount;
509  
510      /**
511 +     * Encoded record of top of Treiber stack of spare threads waiting
512 +     * for resumption. The top 16 bits contain an arbitrary count to
513 +     * avoid ABA effects. The bottom 16bits contains one plus the pool
514 +     * index of waiting worker thread.
515 +     */
516 +    private volatile int spareWaiters;
517 +
518 +    private static final int SPARE_COUNT_SHIFT = 16;
519 +    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
520 +
521 +    /**
522       * Lifecycle control. The low word contains the number of workers
523       * that are (probably) executing tasks. This value is atomically
524       * incremented before a worker gets a task to run, and decremented
525 <     * when worker has no tasks and cannot find any.  Bits 16-18
525 >     * when a worker has no tasks and cannot find any.  Bits 16-18
526       * contain runLevel value. When all are zero, the pool is
527       * running. Level transitions are monotonic (running -> shutdown
528       * -> terminating -> terminated) so each transition adds a bit.
529       * These are bundled together to ensure consistent read for
530       * termination checks (i.e., that runLevel is at least SHUTDOWN
531       * and active threads is zero).
532 +     *
533 +     * Notes: Most direct CASes are dependent on these bitfield
534 +     * positions.  Also, this field is non-private to enable direct
535 +     * performance-sensitive CASes in ForkJoinWorkerThread.
536       */
537 <    private volatile int runState;
537 >    volatile int runState;
538  
539      // Note: The order among run level values matters.
540      private static final int RUNLEVEL_SHIFT     = 16;
# Line 488 | Line 542 | public class ForkJoinPool extends Abstra
542      private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
543      private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
544      private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
491    private static final int ONE_ACTIVE         = 1; // active update delta
545  
546      /**
547       * Holds number of total (i.e., created and not yet terminated)
# Line 529 | Line 582 | public class ForkJoinPool extends Abstra
582       */
583      private final int poolNumber;
584  
585 <    // Utilities for CASing fields. Note that several of these
586 <    // are manually inlined by callers
585 >    // Utilities for CASing fields. Note that most of these
586 >    // are usually manually inlined by callers
587  
588      /**
589 <     * Increments running count.  Also used by ForkJoinTask.
589 >     * Increments running count part of workerCounts
590       */
591      final void incrementRunningCount() {
592          int c;
# Line 554 | Line 607 | public class ForkJoinPool extends Abstra
607      }
608  
609      /**
610 <     * Tries to increment running count
611 <     */
559 <    final boolean tryIncrementRunningCount() {
560 <        int wc;
561 <        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
562 <                                        wc = workerCounts, wc + ONE_RUNNING);
563 <    }
564 <
565 <    /**
566 <     * Tries incrementing active count; fails on contention.
567 <     * Called by workers before executing tasks.
610 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
611 >     * (rarely) necessary when other count updates lag.
612       *
613 <     * @return true on success
613 >     * @param dr -- either zero or ONE_RUNNING
614 >     * @param dt -- either zero or ONE_TOTAL
615       */
616 <    final boolean tryIncrementActiveCount() {
617 <        int c;
618 <        return UNSAFE.compareAndSwapInt(this, runStateOffset,
619 <                                        c = runState, c + ONE_ACTIVE);
616 >    private void decrementWorkerCounts(int dr, int dt) {
617 >        for (;;) {
618 >            int wc = workerCounts;
619 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
620 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
621 >                if ((runState & TERMINATED) != 0)
622 >                    return; // lagging termination on a backout
623 >                Thread.yield();
624 >            }
625 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
626 >                                         wc, wc - (dr + dt)))
627 >                return;
628 >        }
629      }
630  
631      /**
# Line 581 | Line 635 | public class ForkJoinPool extends Abstra
635      final boolean tryDecrementActiveCount() {
636          int c;
637          return UNSAFE.compareAndSwapInt(this, runStateOffset,
638 <                                        c = runState, c - ONE_ACTIVE);
638 >                                        c = runState, c - 1);
639      }
640  
641      /**
# Line 610 | Line 664 | public class ForkJoinPool extends Abstra
664          lock.lock();
665          try {
666              ForkJoinWorkerThread[] ws = workers;
667 <            int nws = ws.length;
668 <            if (k < 0 || k >= nws || ws[k] != null) {
669 <                for (k = 0; k < nws && ws[k] != null; ++k)
667 >            int n = ws.length;
668 >            if (k < 0 || k >= n || ws[k] != null) {
669 >                for (k = 0; k < n && ws[k] != null; ++k)
670                      ;
671 <                if (k == nws)
672 <                    ws = Arrays.copyOf(ws, nws << 1);
671 >                if (k == n)
672 >                    ws = Arrays.copyOf(ws, n << 1);
673              }
674              ws[k] = w;
675              workers = ws; // volatile array write ensures slot visibility
# Line 626 | Line 680 | public class ForkJoinPool extends Abstra
680      }
681  
682      /**
683 <     * Nulls out record of worker in workers array
683 >     * Nulls out record of worker in workers array.
684       */
685      private void forgetWorker(ForkJoinWorkerThread w) {
686          int idx = w.poolIndex;
687 <        // Locking helps method recordWorker avoid unecessary expansion
687 >        // Locking helps method recordWorker avoid unnecessary expansion
688          final ReentrantLock lock = this.workerLock;
689          lock.lock();
690          try {
# Line 642 | Line 696 | public class ForkJoinPool extends Abstra
696          }
697      }
698  
645    // adding and removing workers
646
699      /**
700 <     * Tries to create and add new worker. Assumes that worker counts
701 <     * are already updated to accommodate the worker, so adjusts on
702 <     * failure.
700 >     * Final callback from terminating worker.  Removes record of
701 >     * worker from array, and adjusts counts. If pool is shutting
702 >     * down, tries to complete termination.
703       *
704 <     * @return new worker or null if creation failed
704 >     * @param w the worker
705       */
706 <    private ForkJoinWorkerThread addWorker() {
707 <        ForkJoinWorkerThread w = null;
708 <        try {
709 <            w = factory.newThread(this);
710 <        } finally { // Adjust on either null or exceptional factory return
711 <            if (w == null)
660 <                onWorkerCreationFailure();
661 <        }
662 <        if (w != null)
663 <            w.start(recordWorker(w), ueh);
664 <        return w;
706 >    final void workerTerminated(ForkJoinWorkerThread w) {
707 >        forgetWorker(w);
708 >        decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL);
709 >        while (w.stealCount != 0) // collect final count
710 >            tryAccumulateStealCount(w);
711 >        tryTerminate(false);
712      }
713  
714 +    // Waiting for and signalling events
715 +
716      /**
717 <     * Adjusts counts upon failure to create worker
717 >     * Releases workers blocked on a count not equal to current count.
718 >     * Normally called after precheck that eventWaiters isn't zero to
719 >     * avoid wasted array checks. Gives up upon a change in count or
720 >     * upon releasing two workers, letting others take over.
721       */
722 <    private void onWorkerCreationFailure() {
723 <        for (;;) {
724 <            int wc = workerCounts;
725 <            int rc = wc & RUNNING_COUNT_MASK;
726 <            int tc = wc >>> TOTAL_COUNT_SHIFT;
727 <            if (rc == 0 || wc == 0)
728 <                Thread.yield(); // must wait for other counts to settle
729 <            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
730 <                                              wc - (ONE_RUNNING|ONE_TOTAL)))
722 >    private void releaseEventWaiters() {
723 >        ForkJoinWorkerThread[] ws = workers;
724 >        int n = ws.length;
725 >        long h = eventWaiters;
726 >        int ec = eventCount;
727 >        boolean releasedOne = false;
728 >        ForkJoinWorkerThread w; int id;
729 >        while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
730 >               (int)(h >>> EVENT_COUNT_SHIFT) != ec &&
731 >               id < n && (w = ws[id]) != null) {
732 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
733 >                                          h,  w.nextWaiter)) {
734 >                LockSupport.unpark(w);
735 >                if (releasedOne) // exit on second release
736 >                    break;
737 >                releasedOne = true;
738 >            }
739 >            if (eventCount != ec)
740                  break;
741 +            h = eventWaiters;
742          }
681        tryTerminate(false); // in case of failure during shutdown
743      }
744  
745      /**
746 <     * Creates enough total workers to establish target parallelism,
747 <     * giving up if terminating or addWorker fails
746 >     * Tries to advance eventCount and releases waiters. Called only
747 >     * from workers.
748       */
749 <    private void ensureEnoughTotalWorkers() {
750 <        int wc;
751 <        while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
752 <               runState < TERMINATING) {
753 <            if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
693 <                                          wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
694 <                 addWorker() == null))
695 <                break;
696 <        }
749 >    final void signalWork() {
750 >        int c; // try to increment event count -- CAS failure OK
751 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
752 >        if (eventWaiters != 0L)
753 >            releaseEventWaiters();
754      }
755  
756      /**
757 <     * Final callback from terminating worker.  Removes record of
758 <     * worker from array, and adjusts counts. If pool is shutting
702 <     * down, tries to complete terminatation, else possibly replaces
703 <     * the worker.
757 >     * Adds the given worker to event queue and blocks until
758 >     * terminating or event count advances from the given value
759       *
760 <     * @param w the worker
760 >     * @param w the calling worker thread
761 >     * @param ec the count
762       */
763 <    final void workerTerminated(ForkJoinWorkerThread w) {
764 <        if (w.active) { // force inactive
765 <            w.active = false;
766 <            do {} while (!tryDecrementActiveCount());
767 <        }
768 <        forgetWorker(w);
769 <
770 <        // Decrement total count, and if was running, running count
771 <        // Spin (waiting for other updates) if either would be negative
772 <        int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
717 <        int unit = ONE_TOTAL + nr;
718 <        for (;;) {
719 <            int wc = workerCounts;
720 <            int rc = wc & RUNNING_COUNT_MASK;
721 <            int tc = wc >>> TOTAL_COUNT_SHIFT;
722 <            if (rc - nr < 0 || tc == 0)
723 <                Thread.yield(); // back off if waiting for other updates
724 <            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
725 <                                              wc, wc - unit))
763 >    private void eventSync(ForkJoinWorkerThread w, int ec) {
764 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
765 >        long h;
766 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
767 >               (((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
768 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
769 >               eventCount == ec) {
770 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
771 >                                          w.nextWaiter = h, nh)) {
772 >                awaitEvent(w, ec);
773                  break;
774 +            }
775          }
728
729        accumulateStealCount(w); // collect final count
730        if (!tryTerminate(false))
731            ensureEnoughTotalWorkers();
776      }
777  
734    // Waiting for and signalling events
735
778      /**
779 <     * Releases workers blocked on a count not equal to current count.
780 <     * @return true if any released
779 >     * Blocks the given worker (that has already been entered as an
780 >     * event waiter) until terminating or event count advances from
781 >     * the given value. The oldest (first) waiter uses a timed wait to
782 >     * occasionally one-by-one shrink the number of workers (to a
783 >     * minimum of one) if the pool has not been used for extended
784 >     * periods.
785 >     *
786 >     * @param w the calling worker thread
787 >     * @param ec the count
788       */
789 <    private void releaseWaiters() {
790 <        long top;
791 <        while ((top = eventWaiters) != 0L) {
792 <            ForkJoinWorkerThread[] ws = workers;
793 <            int n = ws.length;
794 <            for (;;) {
795 <                int i = ((int)(top & WAITER_ID_MASK)) - 1;
796 <                int e = (int)(top >>> EVENT_COUNT_SHIFT);
797 <                if (i < 0 || e == eventCount)
798 <                    return;
799 <                ForkJoinWorkerThread w;
800 <                if (i < n && (w = ws[i]) != null &&
801 <                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
802 <                                              top, w.nextWaiter)) {
803 <                    LockSupport.unpark(w);
804 <                    top = eventWaiters;
789 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
790 >        while (eventCount == ec) {
791 >            if (tryAccumulateStealCount(w)) { // transfer while idle
792 >                boolean untimed = (w.nextWaiter != 0L ||
793 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
794 >                long startTime = untimed ? 0 : System.nanoTime();
795 >                Thread.interrupted();         // clear/ignore interrupt
796 >                if (eventCount != ec || w.isTerminating())
797 >                    break;                    // recheck after clear
798 >                if (untimed)
799 >                    LockSupport.park(w);
800 >                else {
801 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
802 >                    if (eventCount != ec || w.isTerminating())
803 >                        break;
804 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
805 >                        tryShutdownUnusedWorker(ec);
806                  }
757                else
758                    break;      // possibly stale; reread
807              }
808          }
809      }
810  
811 +    // Maintaining parallelism
812 +
813      /**
814 <     * Ensures eventCount on exit is different (mod 2^32) than on
765 <     * entry and wakes up all waiters
814 >     * Pushes worker onto the spare stack.
815       */
816 <    private void signalEvent() {
817 <        int c;
818 <        do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
819 <                                               c = eventCount, c+1));
771 <        releaseWaiters();
816 >    final void pushSpare(ForkJoinWorkerThread w) {
817 >        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
818 >        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
819 >                                               w.nextSpare = spareWaiters,ns));
820      }
821  
822      /**
823 <     * Advances eventCount and releases waiters until interference by
824 <     * other releasing threads is detected.
823 >     * Tries (once) to resume a spare if the number of running
824 >     * threads is less than target.
825       */
826 <    final void signalWork() {
827 <        int c;
828 <        UNSAFE.compareAndSwapInt(this, eventCountOffset, c=eventCount, c+1);
829 <        long top;
830 <        while ((top = eventWaiters) != 0L) {
831 <            int ec = eventCount;
832 <            ForkJoinWorkerThread[] ws = workers;
833 <            int n = ws.length;
834 <            for (;;) {
835 <                int i = ((int)(top & WAITER_ID_MASK)) - 1;
836 <                int e = (int)(top >>> EVENT_COUNT_SHIFT);
837 <                if (i < 0 || e == ec)
838 <                    return;
839 <                ForkJoinWorkerThread w;
840 <                if (i < n && (w = ws[i]) != null &&
841 <                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
842 <                                              top, top = w.nextWaiter)) {
843 <                    LockSupport.unpark(w);
844 <                    if (top != eventWaiters) // let someone else take over
845 <                        return;
826 >    private void tryResumeSpare() {
827 >        int sw, id;
828 >        ForkJoinWorkerThread[] ws = workers;
829 >        int n = ws.length;
830 >        ForkJoinWorkerThread w;
831 >        if ((sw = spareWaiters) != 0 &&
832 >            (id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
833 >            id < n && (w = ws[id]) != null &&
834 >            (workerCounts & RUNNING_COUNT_MASK) < parallelism &&
835 >            spareWaiters == sw &&
836 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
837 >                                     sw, w.nextSpare)) {
838 >            int c; // increment running count before resume
839 >            do {} while (!UNSAFE.compareAndSwapInt
840 >                         (this, workerCountsOffset,
841 >                          c = workerCounts, c + ONE_RUNNING));
842 >            if (w.tryUnsuspend())
843 >                LockSupport.unpark(w);
844 >            else   // back out if w was shutdown
845 >                decrementWorkerCounts(ONE_RUNNING, 0);
846 >        }
847 >    }
848 >
849 >    /**
850 >     * Tries to increase the number of running workers if below target
851 >     * parallelism: If a spare exists tries to resume it via
852 >     * tryResumeSpare.  Otherwise, if not enough total workers or all
853 >     * existing workers are busy, adds a new worker. In all cases also
854 >     * helps wake up releasable workers waiting for work.
855 >     */
856 >    private void helpMaintainParallelism() {
857 >        int pc = parallelism;
858 >        int wc, rs, tc;
859 >        while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
860 >               (rs = runState) < TERMINATING) {
861 >            if (spareWaiters != 0)
862 >                tryResumeSpare();
863 >            else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
864 >                     (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
865 >                break;   // enough total
866 >            else if (runState == rs && workerCounts == wc &&
867 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
868 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
869 >                ForkJoinWorkerThread w = null;
870 >                Throwable fail = null;
871 >                try {
872 >                    w = factory.newThread(this);
873 >                } catch (Throwable ex) {
874 >                    fail = ex;
875 >                }
876 >                if (w == null) { // null or exceptional factory return
877 >                    decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
878 >                    tryTerminate(false); // handle failure during shutdown
879 >                    // If originating from an external caller,
880 >                    // propagate exception, else ignore
881 >                    if (fail != null && runState < TERMINATING &&
882 >                        !(Thread.currentThread() instanceof
883 >                          ForkJoinWorkerThread))
884 >                        UNSAFE.throwException(fail);
885 >                    break;
886 >                }
887 >                w.start(recordWorker(w), ueh);
888 >                if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
889 >                    int c; // advance event count
890 >                    UNSAFE.compareAndSwapInt(this, eventCountOffset,
891 >                                             c = eventCount, c+1);
892 >                    break; // add at most one unless total below target
893                  }
799                else
800                    break;      // possibly stale; reread
894              }
895          }
896 +        if (eventWaiters != 0L)
897 +            releaseEventWaiters();
898      }
899  
900      /**
901 <     * Blockss worker until terminating or event count
902 <     * advances from last value held by worker
901 >     * Callback from the oldest waiter in awaitEvent waking up after a
902 >     * period of non-use. If all workers are idle, tries (once) to
903 >     * shutdown an event waiter or a spare, if one exists. Note that
904 >     * we don't need CAS or locks here because the method is called
905 >     * only from one thread occasionally waking (and even misfires are
906 >     * OK). Note that until the shutdown worker fully terminates,
907 >     * workerCounts will overestimate total count, which is tolerable.
908       *
909 <     * @param w the calling worker thread
909 >     * @param ec the event count waited on by caller (to abort
910 >     * attempt if count has since changed).
911       */
912 <    private void eventSync(ForkJoinWorkerThread w) {
913 <        int wec = w.lastEventCount;
914 <        long nextTop = (((long)wec << EVENT_COUNT_SHIFT) |
915 <                        ((long)(w.poolIndex + 1)));
916 <        long top;
917 <        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
918 <               (((int)(top = eventWaiters) & WAITER_ID_MASK) == 0 ||
919 <                (int)(top >>> EVENT_COUNT_SHIFT) == wec) &&
920 <               eventCount == wec) {
921 <            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
922 <                                          w.nextWaiter = top, nextTop)) {
923 <                accumulateStealCount(w); // transfer steals while idle
924 <                Thread.interrupted();    // clear/ignore interrupt
925 <                while (eventCount == wec)
926 <                    w.doPark();
927 <                break;
912 >    private void tryShutdownUnusedWorker(int ec) {
913 >        if (runState == 0 && eventCount == ec) { // only trigger if all idle
914 >            ForkJoinWorkerThread[] ws = workers;
915 >            int n = ws.length;
916 >            ForkJoinWorkerThread w = null;
917 >            boolean shutdown = false;
918 >            int sw;
919 >            long h;
920 >            if ((sw = spareWaiters) != 0) { // prefer killing spares
921 >                int id = (sw & SPARE_ID_MASK) - 1;
922 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
923 >                    UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
924 >                                             sw, w.nextSpare))
925 >                    shutdown = true;
926 >            }
927 >            else if ((h = eventWaiters) != 0L) {
928 >                long nh;
929 >                int id = ((int)(h & WAITER_ID_MASK)) - 1;
930 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
931 >                    (nh = w.nextWaiter) != 0L && // keep at least one worker
932 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
933 >                    shutdown = true;
934 >            }
935 >            if (w != null && shutdown) {
936 >                w.shutdown();
937 >                LockSupport.unpark(w);
938              }
939          }
940 <        w.lastEventCount = eventCount;
940 >        releaseEventWaiters(); // in case of interference
941      }
942  
943      /**
944       * Callback from workers invoked upon each top-level action (i.e.,
945 <     * stealing a task or taking a submission and running
946 <     * it). Performs one or both of the following:
945 >     * stealing a task or taking a submission and running it).
946 >     * Performs one or more of the following:
947       *
948 <     * * If the worker cannot find work, updates its active status to
949 <     * inactive and updates activeCount unless there is contention, in
950 <     * which case it may try again (either in this or a subsequent
951 <     * call).  Additionally, awaits the next task event and/or helps
952 <     * wake up other releasable waiters.
953 <     *
954 <     * * If there are too many running threads, suspends this worker
955 <     * (first forcing inactivation if necessary).  If it is not
956 <     * resumed before a keepAlive elapses, the worker may be "trimmed"
957 <     * -- killed while suspended within suspendAsSpare. Otherwise,
958 <     * upon resume it rechecks to make sure that it is still needed.
948 >     * 1. If the worker is active and either did not run a task
949 >     *    or there are too many workers, try to set its active status
950 >     *    to inactive and update activeCount. On contention, we may
951 >     *    try again in this or a subsequent call.
952 >     *
953 >     * 2. If not enough total workers, help create some.
954 >     *
955 >     * 3. If there are too many running workers, suspend this worker
956 >     *    (first forcing inactive if necessary).  If it is not needed,
957 >     *    it may be shutdown while suspended (via
958 >     *    tryShutdownUnusedWorker).  Otherwise, upon resume it
959 >     *    rechecks running thread count and need for event sync.
960 >     *
961 >     * 4. If worker did not run a task, await the next task event via
962 >     *    eventSync if necessary (first forcing inactivation), upon
963 >     *    which the worker may be shutdown via
964 >     *    tryShutdownUnusedWorker.  Otherwise, help release any
965 >     *    existing event waiters that are now releasable,
966       *
967       * @param w the worker
968 <     * @param retries the number of scans by caller failing to find work
851 <     * find any (in which case it may block waiting for work).
968 >     * @param ran true if worker ran a task since last call to this method
969       */
970 <    final void preStep(ForkJoinWorkerThread w, int retries) {
970 >    final void preStep(ForkJoinWorkerThread w, boolean ran) {
971 >        int wec = w.lastEventCount;
972          boolean active = w.active;
973 <        boolean inactivate = active && retries > 0;
974 <        for (;;) {
975 <            int rs, wc;
976 <            if (inactivate &&
977 <                UNSAFE.compareAndSwapInt(this, runStateOffset,
978 <                                         rs = runState, rs - ONE_ACTIVE))
973 >        boolean inactivate = false;
974 >        int pc = parallelism;
975 >        while (w.runState == 0) {
976 >            int rs = runState;
977 >            if (rs >= TERMINATING) { // propagate shutdown
978 >                w.shutdown();
979 >                break;
980 >            }
981 >            if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
982 >                UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
983                  inactivate = active = w.active = false;
984 <            if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= parallelism) {
985 <                if (retries > 0) {
986 <                    if (retries > 1 && !active)
987 <                        eventSync(w);
988 <                    releaseWaiters();
984 >            int wc = workerCounts;
985 >            if ((wc & RUNNING_COUNT_MASK) > pc) {
986 >                if (!(inactivate |= active) && // must inactivate to suspend
987 >                    workerCounts == wc &&      // try to suspend as spare
988 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
989 >                                             wc, wc - ONE_RUNNING))
990 >                    w.suspendAsSpare();
991 >            }
992 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
993 >                helpMaintainParallelism();     // not enough workers
994 >            else if (!ran) {
995 >                long h = eventWaiters;
996 >                int ec = eventCount;
997 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
998 >                    releaseEventWaiters();     // release others before waiting
999 >                else if (ec != wec) {
1000 >                    w.lastEventCount = ec;     // no need to wait
1001 >                    break;
1002                  }
1003 <                break;
1003 >                else if (!(inactivate |= active))
1004 >                    eventSync(w, wec);         // must inactivate before sync
1005              }
1006 <            if (!(inactivate |= active) &&  // must inactivate to suspend
871 <                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
872 <                                         wc, wc - ONE_RUNNING) &&
873 <                !w.suspendAsSpare())             // false if trimmed
1006 >            else
1007                  break;
1008          }
1009      }
1010  
1011      /**
1012 <     * Awaits join of the given task if enough threads, or can resume
1013 <     * or create a spare. Fails (in which case the given task might
881 <     * not be done) upon contention or lack of decision about
882 <     * blocking.
883 <     *
884 <     * We allow blocking if:
885 <     *
886 <     * 1. There would still be at least as many running threads as
887 <     *    parallelism level if this thread blocks.
888 <     *
889 <     * 2. A spare is resumed to replace this worker. We tolerate
890 <     *    races in the decision to replace when a spare is found.
891 <     *    This may release too many, but if so, the superfluous ones
892 <     *    will re-suspend via preStep().
893 <     *
894 <     * 3. After #spares repeated retries, there are fewer than #spare
895 <     *    threads not running. We allow this slack to avoid hysteresis
896 <     *    and as a hedge against lag/uncertainty of running count
897 <     *    estimates when signalling or unblocking stalls.
898 <     *
899 <     * 4. All existing workers are busy (as rechecked via #spares
900 <     *    repeated retries by caller) and a new spare is created.
901 <     *
902 <     * If none of the above hold, we escape out by re-incrementing
903 <     * count and returning to caller, which can retry later.
1012 >     * Helps and/or blocks awaiting join of the given task.
1013 >     * See above for explanation.
1014       *
1015       * @param joinMe the task to join
1016 <     * @param retries the number of calls to this method for this join
1017 <     */
1018 <    final void tryAwaitJoin(ForkJoinTask<?> joinMe, int retries) {
1019 <        int pc = parallelism;
1020 <        boolean running = true; // false when running count decremented
1021 <        outer:while (joinMe.status >= 0) {
1022 <            int wc = workerCounts;
1023 <            int rc = wc & RUNNING_COUNT_MASK;
1024 <            int tc = wc >>> TOTAL_COUNT_SHIFT;
1025 <            if (running) { // replace with spare or decrement count
1026 <                if (rc <= pc && tc > pc &&
1027 <                    (retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1028 <                    ForkJoinWorkerThread[] ws = workers; // search for spare
1029 <                    int nws = ws.length;
1030 <                    for (int i = 0; i < nws; ++i) {
1031 <                        ForkJoinWorkerThread w = ws[i];
1032 <                        if (w != null && w.isSuspended()) {
1033 <                            if ((workerCounts & RUNNING_COUNT_MASK) > pc)
1034 <                                continue outer;
1035 <                            if (joinMe.status < 0)
1036 <                                break outer;
1037 <                            if (w.tryResumeSpare()) {
1038 <                                running = false;
1039 <                                break outer;
1016 >     * @param worker the current worker thread
1017 >     * @param timed true if wait should time out
1018 >     * @param nanos timeout value if timed
1019 >     */
1020 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker,
1021 >                         boolean timed, long nanos) {
1022 >        long startTime = timed? System.nanoTime() : 0L;
1023 >        int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1024 >        while (joinMe.status >= 0) {
1025 >            int wc;
1026 >            long nt = 0L;
1027 >            if (runState >= TERMINATING) {
1028 >                joinMe.cancelIgnoringExceptions();
1029 >                break;
1030 >            }
1031 >            worker.helpJoinTask(joinMe);
1032 >            if (joinMe.status < 0)
1033 >                break;
1034 >            else if (retries > 0)
1035 >                --retries;
1036 >            else if (timed &&
1037 >                     (nt = nanos - (System.nanoTime() - startTime)) <= 0L)
1038 >                break;
1039 >            else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1040 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1041 >                                              wc, wc - ONE_RUNNING)) {
1042 >                int stat, c; long h;
1043 >                while ((stat = joinMe.status) >= 0 &&
1044 >                       (h = eventWaiters) != 0L && // help release others
1045 >                       (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1046 >                    releaseEventWaiters();
1047 >                if (stat >= 0) {
1048 >                    if ((workerCounts & RUNNING_COUNT_MASK) != 0) {
1049 >                        long ms; int ns;
1050 >                        if (!timed) {
1051 >                            ms = JOIN_TIMEOUT_MILLIS;
1052 >                            ns = 0;
1053 >                        }
1054 >                        else { // at most JOIN_TIMEOUT_MILLIS per wait
1055 >                            ms = nt / 1000000;
1056 >                            if (ms > JOIN_TIMEOUT_MILLIS) {
1057 >                                ms = JOIN_TIMEOUT_MILLIS;
1058 >                                ns = 0;
1059                              }
1060 <                            continue outer; // rescan on failure to resume
1060 >                            else
1061 >                                ns = (int) (nt % 1000000);
1062                          }
1063 +                        stat = joinMe.internalAwaitDone(ms, ns);
1064                      }
1065 +                    if (stat >= 0) // timeout or no running workers
1066 +                        helpMaintainParallelism();
1067                  }
1068 <                if ((rc <= pc && (rc == 0 || --retries < 0)) || // no retry
1069 <                    joinMe.status < 0)
1070 <                    break;
1071 <                if (workerCounts == wc &&
1072 <                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
940 <                                             wc, wc - ONE_RUNNING))
941 <                    running = false;
942 <            }
943 <            else { // allow blocking if enough threads
944 <                int sc = tc - pc + 1;          // = spares, plus the one to add
945 <                if (sc > 0 && rc > 0 && rc >= pc - sc && rc > pc - retries)
946 <                    break;  
947 <                if (--retries > sc && tc < MAX_THREADS &&
948 <                    tc == (runState & ACTIVE_COUNT_MASK) &&
949 <                    workerCounts == wc &&
950 <                    UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
951 <                                             wc + (ONE_RUNNING|ONE_TOTAL))) {
952 <                    addWorker();
953 <                    break;
954 <                }
955 <                if (workerCounts == wc &&
956 <                    UNSAFE.compareAndSwapInt (this, workerCountsOffset,
957 <                                              wc, wc + ONE_RUNNING)) {
958 <                    running = true;            // back out; allow retry
959 <                    break;
960 <                }
1068 >                do {} while (!UNSAFE.compareAndSwapInt
1069 >                             (this, workerCountsOffset,
1070 >                              c = workerCounts, c + ONE_RUNNING));
1071 >                if (stat < 0)
1072 >                    break;   // else restart
1073              }
1074          }
963        if (!running) { // can block
964            int c;                      // to inline incrementRunningCount
965            joinMe.internalAwaitDone();
966            do {} while (!UNSAFE.compareAndSwapInt
967                         (this, workerCountsOffset,
968                          c = workerCounts, c + ONE_RUNNING));
969        }
1075      }
1076  
1077      /**
1078 <     * Same idea as (and shares many code snippets with) tryAwaitJoin,
974 <     * but self-contained because there are no caller retries.
975 <     * TODO: Rework to use simpler API.
1078 >     * Same idea as awaitJoin, but no helping, retries, or timeouts.
1079       */
1080      final void awaitBlocker(ManagedBlocker blocker)
1081          throws InterruptedException {
1082 <        int pc = parallelism;
980 <        boolean running = true;
981 <        int retries = 0;
982 <        boolean done;
983 <        outer:while (!(done = blocker.isReleasable())) {
1082 >        while (!blocker.isReleasable()) {
1083              int wc = workerCounts;
1084 <            int rc = wc & RUNNING_COUNT_MASK;
1085 <            int tc = wc >>> TOTAL_COUNT_SHIFT;
1086 <            if (running) {
1087 <                if (rc <= pc && tc > pc &&
1088 <                    (retries > 0 || tc > (runState & ACTIVE_COUNT_MASK))) {
1089 <                    ForkJoinWorkerThread[] ws = workers;
1090 <                    int nws = ws.length;
1091 <                    for (int i = 0; i < nws; ++i) {
1092 <                        ForkJoinWorkerThread w = ws[i];
1093 <                        if (w != null && w.isSuspended()) {
1094 <                            if ((workerCounts & RUNNING_COUNT_MASK) > pc)
1095 <                                continue outer;
1096 <                            if (done = blocker.isReleasable())
1097 <                                break outer;
999 <                            if (w.tryResumeSpare()) {
1000 <                                running = false;
1001 <                                break outer;
1002 <                            }
1003 <                            continue outer;
1004 <                        }
1084 >            if ((wc & RUNNING_COUNT_MASK) != 0 &&
1085 >                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1086 >                                         wc, wc - ONE_RUNNING)) {
1087 >                try {
1088 >                    while (!blocker.isReleasable()) {
1089 >                        long h = eventWaiters;
1090 >                        if (h != 0L &&
1091 >                            (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1092 >                            releaseEventWaiters();
1093 >                        else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1094 >                                 runState < TERMINATING)
1095 >                            helpMaintainParallelism();
1096 >                        else if (blocker.block())
1097 >                            break;
1098                      }
1099 <                    if (done = blocker.isReleasable())
1100 <                        break;
1101 <                }
1102 <                if (rc > 0 && workerCounts == wc &&
1103 <                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1011 <                                             wc, wc - ONE_RUNNING)) {
1012 <                    running = false;
1013 <                    if (rc > pc)
1014 <                        break;
1099 >                } finally {
1100 >                    int c;
1101 >                    do {} while (!UNSAFE.compareAndSwapInt
1102 >                                 (this, workerCountsOffset,
1103 >                                  c = workerCounts, c + ONE_RUNNING));
1104                  }
1016            }
1017            else if (rc >= pc)
1018                break;
1019            else if (tc < MAX_THREADS &&
1020                     tc == (runState & ACTIVE_COUNT_MASK) &&
1021                     workerCounts == wc &&
1022                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1023                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
1024                addWorker();
1105                  break;
1106              }
1027            else if (workerCounts == wc &&
1028                     UNSAFE.compareAndSwapInt (this, workerCountsOffset,
1029                                              wc, wc + ONE_RUNNING)) {
1030                Thread.yield();
1031                ++retries;
1032                running = true;            // allow rescan
1033            }
1034        }
1035
1036        try {
1037            if (!done)
1038                do {} while (!blocker.isReleasable() && !blocker.block());
1039        } finally {
1040            if (!running) {
1041                int c;
1042                do {} while (!UNSAFE.compareAndSwapInt
1043                             (this, workerCountsOffset,
1044                              c = workerCounts, c + ONE_RUNNING));
1045            }
1107          }
1108      }
1109  
# Line 1072 | Line 1133 | public class ForkJoinPool extends Abstra
1133          return true;
1134      }
1135  
1136 +
1137      /**
1138       * Actions on transition to TERMINATING
1139 +     *
1140 +     * Runs up to four passes through workers: (0) shutting down each
1141 +     * (without waking up if parked) to quickly spread notifications
1142 +     * without unnecessary bouncing around event queues etc (1) wake
1143 +     * up and help cancel tasks (2) interrupt (3) mop up races with
1144 +     * interrupted workers
1145       */
1146      private void startTerminating() {
1147 <        for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1148 <            cancelSubmissions();
1149 <            shutdownWorkers();
1150 <            cancelWorkerTasks();
1151 <            signalEvent();
1152 <            interruptWorkers();
1147 >        cancelSubmissions();
1148 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1149 >            int c; // advance event count
1150 >            UNSAFE.compareAndSwapInt(this, eventCountOffset,
1151 >                                     c = eventCount, c+1);
1152 >            eventWaiters = 0L; // clobber lists
1153 >            spareWaiters = 0;
1154 >            for (ForkJoinWorkerThread w : workers) {
1155 >                if (w != null) {
1156 >                    w.shutdown();
1157 >                    if (passes > 0 && !w.isTerminated()) {
1158 >                        w.cancelTasks();
1159 >                        LockSupport.unpark(w);
1160 >                        if (passes > 1 && !w.isInterrupted()) {
1161 >                            try {
1162 >                                w.interrupt();
1163 >                            } catch (SecurityException ignore) {
1164 >                            }
1165 >                        }
1166 >                    }
1167 >                }
1168 >            }
1169          }
1170      }
1171  
1172      /**
1173 <     * Clear out and cancel submissions, ignoring exceptions
1173 >     * Clears out and cancels submissions, ignoring exceptions.
1174       */
1175      private void cancelSubmissions() {
1176          ForkJoinTask<?> task;
# Line 1098 | Line 1182 | public class ForkJoinPool extends Abstra
1182          }
1183      }
1184  
1101    /**
1102     * Sets all worker run states to at least shutdown,
1103     * also resuming suspended workers
1104     */
1105    private void shutdownWorkers() {
1106        ForkJoinWorkerThread[] ws = workers;
1107        int nws = ws.length;
1108        for (int i = 0; i < nws; ++i) {
1109            ForkJoinWorkerThread w = ws[i];
1110            if (w != null)
1111                w.shutdown();
1112        }
1113    }
1114
1115    /**
1116     * Clears out and cancels all locally queued tasks
1117     */
1118    private void cancelWorkerTasks() {
1119        ForkJoinWorkerThread[] ws = workers;
1120        int nws = ws.length;
1121        for (int i = 0; i < nws; ++i) {
1122            ForkJoinWorkerThread w = ws[i];
1123            if (w != null)
1124                w.cancelTasks();
1125        }
1126    }
1127
1128    /**
1129     * Unsticks all workers blocked on joins etc
1130     */
1131    private void interruptWorkers() {
1132        ForkJoinWorkerThread[] ws = workers;
1133        int nws = ws.length;
1134        for (int i = 0; i < nws; ++i) {
1135            ForkJoinWorkerThread w = ws[i];
1136            if (w != null && !w.isTerminated()) {
1137                try {
1138                    w.interrupt();
1139                } catch (SecurityException ignore) {
1140                }
1141            }
1142        }
1143    }
1144
1185      // misc support for ForkJoinWorkerThread
1186  
1187      /**
1188 <     * Returns pool number
1188 >     * Returns pool number.
1189       */
1190      final int getPoolNumber() {
1191          return poolNumber;
1192      }
1193  
1194      /**
1195 <     * Accumulates steal count from a worker, clearing
1196 <     * the worker's value
1195 >     * Tries to accumulate steal count from a worker, clearing
1196 >     * the worker's value if successful.
1197 >     *
1198 >     * @return true if worker steal count now zero
1199       */
1200 <    final void accumulateStealCount(ForkJoinWorkerThread w) {
1200 >    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1201          int sc = w.stealCount;
1202 <        if (sc != 0) {
1203 <            long c;
1204 <            w.stealCount = 0;
1205 <            do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1206 <                                                    c = stealCount, c + sc));
1202 >        long c = stealCount;
1203 >        // CAS even if zero, for fence effects
1204 >        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1205 >            if (sc != 0)
1206 >                w.stealCount = 0;
1207 >            return true;
1208          }
1209 +        return sc == 0;
1210      }
1211  
1212      /**
# Line 1171 | Line 1215 | public class ForkJoinPool extends Abstra
1215       */
1216      final int idlePerActive() {
1217          int pc = parallelism; // use parallelism, not rc
1218 <        int ac = runState;    // no mask -- artifically boosts during shutdown
1218 >        int ac = runState;    // no mask -- artificially boosts during shutdown
1219          // Use exact results for small values, saturate past 4
1220 <        return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1220 >        return ((pc <= ac) ? 0 :
1221 >                (pc >>> 1 <= ac) ? 1 :
1222 >                (pc >>> 2 <= ac) ? 3 :
1223 >                pc >>> 3);
1224      }
1225  
1226      // Public and protected methods
# Line 1223 | Line 1270 | public class ForkJoinPool extends Abstra
1270       * use {@link #defaultForkJoinWorkerThreadFactory}.
1271       * @param handler the handler for internal worker threads that
1272       * terminate due to unrecoverable errors encountered while executing
1273 <     * tasks. For default value, use <code>null</code>.
1273 >     * tasks. For default value, use {@code null}.
1274       * @param asyncMode if true,
1275       * establishes local first-in-first-out scheduling mode for forked
1276       * tasks that are never joined. This mode may be more appropriate
1277       * than default locally stack-based mode in applications in which
1278       * worker threads only process event-style asynchronous tasks.
1279 <     * For default value, use <code>false</code>.
1279 >     * For default value, use {@code false}.
1280       * @throws IllegalArgumentException if parallelism less than or
1281       *         equal to zero, or greater than implementation limit
1282       * @throws NullPointerException if the factory is null
# Line 1245 | Line 1292 | public class ForkJoinPool extends Abstra
1292          checkPermission();
1293          if (factory == null)
1294              throw new NullPointerException();
1295 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
1295 >        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1296              throw new IllegalArgumentException();
1297          this.parallelism = parallelism;
1298          this.factory = factory;
# Line 1264 | Line 1311 | public class ForkJoinPool extends Abstra
1311       * @param pc the initial parallelism level
1312       */
1313      private static int initialArraySizeFor(int pc) {
1314 <        // See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1315 <        int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1314 >        // If possible, initially allocate enough space for one spare
1315 >        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1316 >        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1317          size |= size >>> 1;
1318          size |= size >>> 2;
1319          size |= size >>> 4;
# Line 1276 | Line 1324 | public class ForkJoinPool extends Abstra
1324      // Execution methods
1325  
1326      /**
1327 <     * Common code for execute, invoke and submit
1327 >     * Submits task and creates, starts, or resumes some workers if necessary
1328       */
1329      private <T> void doSubmit(ForkJoinTask<T> task) {
1282        if (task == null)
1283            throw new NullPointerException();
1284        if (runState >= SHUTDOWN)
1285            throw new RejectedExecutionException();
1330          submissionQueue.offer(task);
1331 <        signalEvent();
1332 <        ensureEnoughTotalWorkers();
1331 >        int c; // try to increment event count -- CAS failure OK
1332 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1333 >        helpMaintainParallelism();
1334      }
1335  
1336      /**
1337       * Performs the given task, returning its result upon completion.
1293     * If the caller is already engaged in a fork/join computation in
1294     * the current pool, this method is equivalent in effect to
1295     * {@link ForkJoinTask#invoke}.
1338       *
1339       * @param task the task
1340       * @return the task's result
# Line 1301 | Line 1343 | public class ForkJoinPool extends Abstra
1343       *         scheduled for execution
1344       */
1345      public <T> T invoke(ForkJoinTask<T> task) {
1346 <        doSubmit(task);
1347 <        return task.join();
1346 >        if (task == null)
1347 >            throw new NullPointerException();
1348 >        if (runState >= SHUTDOWN)
1349 >            throw new RejectedExecutionException();
1350 >        Thread t = Thread.currentThread();
1351 >        if ((t instanceof ForkJoinWorkerThread) &&
1352 >            ((ForkJoinWorkerThread)t).pool == this)
1353 >            return task.invoke();  // bypass submit if in same pool
1354 >        else {
1355 >            doSubmit(task);
1356 >            return task.join();
1357 >        }
1358 >    }
1359 >
1360 >    /**
1361 >     * Unless terminating, forks task if within an ongoing FJ
1362 >     * computation in the current pool, else submits as external task.
1363 >     */
1364 >    private <T> void forkOrSubmit(ForkJoinTask<T> task) {
1365 >        if (runState >= SHUTDOWN)
1366 >            throw new RejectedExecutionException();
1367 >        Thread t = Thread.currentThread();
1368 >        if ((t instanceof ForkJoinWorkerThread) &&
1369 >            ((ForkJoinWorkerThread)t).pool == this)
1370 >            task.fork();
1371 >        else
1372 >            doSubmit(task);
1373      }
1374  
1375      /**
1376       * Arranges for (asynchronous) execution of the given task.
1310     * If the caller is already engaged in a fork/join computation in
1311     * the current pool, this method is equivalent in effect to
1312     * {@link ForkJoinTask#fork}.
1377       *
1378       * @param task the task
1379       * @throws NullPointerException if the task is null
# Line 1317 | Line 1381 | public class ForkJoinPool extends Abstra
1381       *         scheduled for execution
1382       */
1383      public void execute(ForkJoinTask<?> task) {
1384 <        doSubmit(task);
1384 >        if (task == null)
1385 >            throw new NullPointerException();
1386 >        forkOrSubmit(task);
1387      }
1388  
1389      // AbstractExecutorService methods
# Line 1328 | Line 1394 | public class ForkJoinPool extends Abstra
1394       *         scheduled for execution
1395       */
1396      public void execute(Runnable task) {
1397 +        if (task == null)
1398 +            throw new NullPointerException();
1399          ForkJoinTask<?> job;
1400          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1401              job = (ForkJoinTask<?>) task;
1402          else
1403              job = ForkJoinTask.adapt(task, null);
1404 <        doSubmit(job);
1404 >        forkOrSubmit(job);
1405      }
1406  
1407      /**
1408       * Submits a ForkJoinTask for execution.
1341     * If the caller is already engaged in a fork/join computation in
1342     * the current pool, this method is equivalent in effect to
1343     * {@link ForkJoinTask#fork}.
1409       *
1410       * @param task the task to submit
1411       * @return the task
# Line 1349 | Line 1414 | public class ForkJoinPool extends Abstra
1414       *         scheduled for execution
1415       */
1416      public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1417 <        doSubmit(task);
1417 >        if (task == null)
1418 >            throw new NullPointerException();
1419 >        forkOrSubmit(task);
1420          return task;
1421      }
1422  
# Line 1359 | Line 1426 | public class ForkJoinPool extends Abstra
1426       *         scheduled for execution
1427       */
1428      public <T> ForkJoinTask<T> submit(Callable<T> task) {
1429 +        if (task == null)
1430 +            throw new NullPointerException();
1431          ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1432 <        doSubmit(job);
1432 >        forkOrSubmit(job);
1433          return job;
1434      }
1435  
# Line 1370 | Line 1439 | public class ForkJoinPool extends Abstra
1439       *         scheduled for execution
1440       */
1441      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1442 +        if (task == null)
1443 +            throw new NullPointerException();
1444          ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1445 <        doSubmit(job);
1445 >        forkOrSubmit(job);
1446          return job;
1447      }
1448  
# Line 1381 | Line 1452 | public class ForkJoinPool extends Abstra
1452       *         scheduled for execution
1453       */
1454      public ForkJoinTask<?> submit(Runnable task) {
1455 +        if (task == null)
1456 +            throw new NullPointerException();
1457          ForkJoinTask<?> job;
1458          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1459              job = (ForkJoinTask<?>) task;
1460          else
1461              job = ForkJoinTask.adapt(task, null);
1462 <        doSubmit(job);
1462 >        forkOrSubmit(job);
1463          return job;
1464      }
1465  
# Line 1446 | Line 1519 | public class ForkJoinPool extends Abstra
1519  
1520      /**
1521       * Returns the number of worker threads that have started but not
1522 <     * yet terminated.  This result returned by this method may differ
1522 >     * yet terminated.  The result returned by this method may differ
1523       * from {@link #getParallelism} when threads are created to
1524       * maintain parallelism when others are cooperatively blocked.
1525       *
# Line 1531 | Line 1604 | public class ForkJoinPool extends Abstra
1604       */
1605      public long getQueuedTaskCount() {
1606          long count = 0;
1607 <        ForkJoinWorkerThread[] ws = workers;
1535 <        int nws = ws.length;
1536 <        for (int i = 0; i < nws; ++i) {
1537 <            ForkJoinWorkerThread w = ws[i];
1607 >        for (ForkJoinWorkerThread w : workers)
1608              if (w != null)
1609                  count += w.getQueueSize();
1540        }
1610          return count;
1611      }
1612  
# Line 1591 | Line 1660 | public class ForkJoinPool extends Abstra
1660       * @return the number of elements transferred
1661       */
1662      protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1663 <        int n = submissionQueue.drainTo(c);
1664 <        ForkJoinWorkerThread[] ws = workers;
1596 <        int nws = ws.length;
1597 <        for (int i = 0; i < nws; ++i) {
1598 <            ForkJoinWorkerThread w = ws[i];
1663 >        int count = submissionQueue.drainTo(c);
1664 >        for (ForkJoinWorkerThread w : workers)
1665              if (w != null)
1666 <                n += w.drainTasksTo(c);
1601 <        }
1602 <        return n;
1603 <    }
1604 <
1605 <    /**
1606 <     * Returns count of total parks by existing workers.
1607 <     * Used during development only since not meaningful to users.
1608 <     */
1609 <    private int collectParkCount() {
1610 <        int count = 0;
1611 <        ForkJoinWorkerThread[] ws = workers;
1612 <        int nws = ws.length;
1613 <        for (int i = 0; i < nws; ++i) {
1614 <            ForkJoinWorkerThread w = ws[i];
1615 <            if (w != null)
1616 <                count += w.parkCount;
1617 <        }
1666 >                count += w.drainTasksTo(c);
1667          return count;
1668      }
1669  
# Line 1635 | Line 1684 | public class ForkJoinPool extends Abstra
1684          int pc = parallelism;
1685          int rs = runState;
1686          int ac = rs & ACTIVE_COUNT_MASK;
1638        //        int pk = collectParkCount();
1687          return super.toString() +
1688              "[" + runLevelToString(rs) +
1689              ", parallelism = " + pc +
# Line 1645 | Line 1693 | public class ForkJoinPool extends Abstra
1693              ", steals = " + st +
1694              ", tasks = " + qt +
1695              ", submissions = " + qs +
1648            //            ", parks = " + pk +
1696              "]";
1697      }
1698  
# Line 1720 | Line 1767 | public class ForkJoinPool extends Abstra
1767      }
1768  
1769      /**
1770 +     * Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1771 +     */
1772 +    final boolean isAtLeastTerminating() {
1773 +        return runState >= TERMINATING;
1774 +    }
1775 +
1776 +    /**
1777       * Returns {@code true} if this pool has been shut down.
1778       *
1779       * @return {@code true} if this pool has been shut down
# Line 1743 | Line 1797 | public class ForkJoinPool extends Abstra
1797          throws InterruptedException {
1798          try {
1799              return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
1800 <        } catch(TimeoutException ex) {
1800 >        } catch (TimeoutException ex) {
1801              return false;
1802          }
1803      }
# Line 1752 | Line 1806 | public class ForkJoinPool extends Abstra
1806       * Interface for extending managed parallelism for tasks running
1807       * in {@link ForkJoinPool}s.
1808       *
1809 <     * <p>A {@code ManagedBlocker} provides two methods.
1810 <     * Method {@code isReleasable} must return {@code true} if
1811 <     * blocking is not necessary. Method {@code block} blocks the
1812 <     * current thread if necessary (perhaps internally invoking
1813 <     * {@code isReleasable} before actually blocking).
1809 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1810 >     * {@code isReleasable} must return {@code true} if blocking is
1811 >     * not necessary. Method {@code block} blocks the current thread
1812 >     * if necessary (perhaps internally invoking {@code isReleasable}
1813 >     * before actually blocking). The unusual methods in this API
1814 >     * accommodate synchronizers that may, but don't usually, block
1815 >     * for long periods. Similarly, they allow more efficient internal
1816 >     * handling of cases in which additional workers may be, but
1817 >     * usually are not, needed to ensure sufficient parallelism.
1818 >     * Toward this end, implementations of method {@code isReleasable}
1819 >     * must be amenable to repeated invocation.
1820       *
1821       * <p>For example, here is a ManagedBlocker based on a
1822       * ReentrantLock:
# Line 1774 | Line 1834 | public class ForkJoinPool extends Abstra
1834       *     return hasLock || (hasLock = lock.tryLock());
1835       *   }
1836       * }}</pre>
1837 +     *
1838 +     * <p>Here is a class that possibly blocks waiting for an
1839 +     * item on a given queue:
1840 +     *  <pre> {@code
1841 +     * class QueueTaker<E> implements ManagedBlocker {
1842 +     *   final BlockingQueue<E> queue;
1843 +     *   volatile E item = null;
1844 +     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1845 +     *   public boolean block() throws InterruptedException {
1846 +     *     if (item == null)
1847 +     *       item = queue.take();
1848 +     *     return true;
1849 +     *   }
1850 +     *   public boolean isReleasable() {
1851 +     *     return item != null || (item = queue.poll()) != null;
1852 +     *   }
1853 +     *   public E getItem() { // call after pool.managedBlock completes
1854 +     *     return item;
1855 +     *   }
1856 +     * }}</pre>
1857       */
1858      public static interface ManagedBlocker {
1859          /**
# Line 1816 | Line 1896 | public class ForkJoinPool extends Abstra
1896      public static void managedBlock(ManagedBlocker blocker)
1897          throws InterruptedException {
1898          Thread t = Thread.currentThread();
1899 <        if (t instanceof ForkJoinWorkerThread)
1900 <            ((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
1899 >        if (t instanceof ForkJoinWorkerThread) {
1900 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1901 >            w.pool.awaitBlocker(blocker);
1902 >        }
1903          else {
1904              do {} while (!blocker.isReleasable() && !blocker.block());
1905          }
# Line 1845 | Line 1927 | public class ForkJoinPool extends Abstra
1927      private static final long eventCountOffset =
1928          objectFieldOffset("eventCount", ForkJoinPool.class);
1929      private static final long eventWaitersOffset =
1930 <        objectFieldOffset("eventWaiters",ForkJoinPool.class);
1930 >        objectFieldOffset("eventWaiters", ForkJoinPool.class);
1931      private static final long stealCountOffset =
1932 <        objectFieldOffset("stealCount",ForkJoinPool.class);
1932 >        objectFieldOffset("stealCount", ForkJoinPool.class);
1933 >    private static final long spareWaitersOffset =
1934 >        objectFieldOffset("spareWaiters", ForkJoinPool.class);
1935  
1936      private static long objectFieldOffset(String field, Class<?> klazz) {
1937          try {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines