[ViewVC] Diff of: jsr166/jsr166/src/jsr166y/ForkJoinPool.java

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.5 by jsr166, Thu Mar 19 05:10:42 2009 UTC vs.
Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC

+/*
+ * Written by Doug Lea with assistance from members of JCP JSR-166
+ * Expert Group and released to the public domain, as explained at
-<
+ * http://creativecommons.org/licenses/publicdomain
->
+ * http://creativecommons.org/publicdomain/zero/1.0/
+ */
+package jsr166y;
-<
+import java.util.*;
-<
+import java.util.concurrent.*;
-<
+import java.util.concurrent.locks.*;
-<
+import java.util.concurrent.atomic.*;
-<
+import sun.misc.Unsafe;
-<
+import java.lang.reflect.*;
->
->
+import java.util.ArrayList;
->
+import java.util.Arrays;
->
+import java.util.Collection;
->
+import java.util.Collections;
->
+import java.util.List;
->
+import java.util.Random;
->
+import java.util.concurrent.AbstractExecutorService;
->
+import java.util.concurrent.Callable;
->
+import java.util.concurrent.ExecutorService;
->
+import java.util.concurrent.Future;
->
+import java.util.concurrent.RejectedExecutionException;
->
+import java.util.concurrent.RunnableFuture;
->
+import java.util.concurrent.TimeUnit;
->
+import java.util.concurrent.atomic.AtomicInteger;
->
+import java.util.concurrent.atomic.AtomicLong;
->
+import java.util.concurrent.locks.AbstractQueuedSynchronizer;
->
+import java.util.concurrent.locks.Condition;
+/**
-<
+ * An {@link ExecutorService} for running {@link ForkJoinTask}s.  A
-<
+ * ForkJoinPool provides the entry point for submissions from
-<
+ * non-ForkJoinTasks, as well as management and monitoring operations.
-<
+ * Normally a single ForkJoinPool is used for a large number of
-<
+ * submitted tasks. Otherwise, use would not usually outweigh the
-<
+ * construction and bookkeeping overhead of creating a large set of
-<
+ * threads.
->
+ * An {@link ExecutorService} for running {@link ForkJoinTask}s.
->
+ * A {@code ForkJoinPool} provides the entry point for submissions
->
+ * from non-{@code ForkJoinTask} clients, as well as management and
->
+ * monitoring operations.
-<
+ * <p>ForkJoinPools differ from other kinds of Executors mainly in
-<
+ * that they provide <em>work-stealing</em>: all threads in the pool
-<
+ * attempt to find and execute subtasks created by other active tasks
-<
+ * (eventually blocking if none exist). This makes them efficient when
-<
+ * most tasks spawn other subtasks (as do most ForkJoinTasks), as well
-<
+ * as the mixed execution of some plain Runnable- or Callable- based
-<
+ * activities along with ForkJoinTasks. Otherwise, other
-<
+ * ExecutorService implementations are typically more appropriate
-<
+ * choices.
->
+ * <p>A {@code ForkJoinPool} differs from other kinds of {@link
->
+ * ExecutorService} mainly by virtue of employing
->
+ * <em>work-stealing</em>: all threads in the pool attempt to find and
->
+ * execute tasks submitted to the pool and/or created by other active
->
+ * tasks (eventually blocking waiting for work if none exist). This
->
+ * enables efficient processing when most tasks spawn other subtasks
->
+ * (as do most {@code ForkJoinTask}s), as well as when many small
->
+ * tasks are submitted to the pool from external clients.  Especially
->
+ * when setting <em>asyncMode</em> to true in constructors, {@code
->
+ * ForkJoinPool}s may also be appropriate for use with event-style
->
+ * tasks that are never joined.
-<
+ * <p>A ForkJoinPool may be constructed with a given parallelism level
-<
+ * (target pool size), which it attempts to maintain by dynamically
-<
+ * adding, suspending, or resuming threads, even if some tasks are
-<
+ * waiting to join others. However, no such adjustments are performed
-<
+ * in the face of blocked IO or other unmanaged synchronization. The
-<
+ * nested <code>ManagedBlocker</code> interface enables extension of
-<
+ * the kinds of synchronization accommodated.  The target parallelism
-<
+ * level may also be changed dynamically (<code>setParallelism</code>)
-<
+ * and dynamically thread construction can be limited using methods
-<
+ * <code>setMaximumPoolSize</code> and/or
-<
+ * <code>setMaintainsParallelism</code>.
->
+ * <p>A {@code ForkJoinPool} is constructed with a given target
->
+ * parallelism level; by default, equal to the number of available
->
+ * processors. The pool attempts to maintain enough active (or
->
+ * available) threads by dynamically adding, suspending, or resuming
->
+ * internal worker threads, even if some tasks are stalled waiting to
->
+ * join others. However, no such adjustments are guaranteed in the
->
+ * face of blocked IO or other unmanaged synchronization. The nested
->
+ * {@link ManagedBlocker} interface enables extension of the kinds of
->
+ * synchronization accommodated.
+ * <p>In addition to execution and lifecycle control methods, this
+ * class provides status check methods (for example
-<
+ * <code>getStealCount</code>) that are intended to aid in developing,
->
+ * {@link #getStealCount}) that are intended to aid in developing,
+ * tuning, and monitoring fork/join applications. Also, method
-<
+ * <code>toString</code> returns indications of pool state in a
->
+ * {@link #toString} returns indications of pool state in a
+ * convenient form for informal monitoring.
-+
+ * <p> As is the case with other ExecutorServices, there are three
-+
+ * main task execution methods summarized in the following table.
-+
+ * These are designed to be used primarily by clients not already
-+
+ * engaged in fork/join computations in the current pool.  The main
-+
+ * forms of these methods accept instances of {@code ForkJoinTask},
-+
+ * but overloaded forms also allow mixed execution of plain {@code
-+
+ * Runnable}- or {@code Callable}- based activities as well.  However,
-+
+ * tasks that are already executing in a pool should normally instead
-+
+ * use the within-computation forms listed in the table unless using
-+
+ * async event-style tasks that are not usually joined, in which case
-+
+ * there is little difference among choice of methods.
-+
+ *
-+
+ * <table BORDER CELLPADDING=3 CELLSPACING=1>
-+
+ *  <tr>
-+
+ *    <td></td>
-+
+ *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
-+
+ *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
-+
+ *  </tr>
-+
+ *  <tr>
-+
+ *    <td> <b>Arrange async execution</td>
-+
+ *    <td> {@link #execute(ForkJoinTask)}</td>
-+
+ *    <td> {@link ForkJoinTask#fork}</td>
-+
+ *  </tr>
-+
+ *  <tr>
-+
+ *    <td> <b>Await and obtain result</td>
-+
+ *    <td> {@link #invoke(ForkJoinTask)}</td>
-+
+ *    <td> {@link ForkJoinTask#invoke}</td>
-+
+ *  </tr>
-+
+ *  <tr>
-+
+ *    <td> <b>Arrange exec and obtain Future</td>
-+
+ *    <td> {@link #submit(ForkJoinTask)}</td>
-+
+ *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
-+
+ *  </tr>
-+
+ * </table>
-+
+ *
-+
+ * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
-+
+ * used for all parallel task execution in a program or subsystem.
-+
+ * Otherwise, use would not usually outweigh the construction and
-+
+ * bookkeeping overhead of creating a large set of threads. For
-+
+ * example, a common pool could be used for the {@code SortTasks}
-+
+ * illustrated in {@link RecursiveAction}. Because {@code
-+
+ * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
-+
+ * daemon} mode, there is typically no need to explicitly {@link
-+
+ * #shutdown} such a pool upon program exit.
-+
+ *
-+
+ *  <pre> {@code
-+
+ * static final ForkJoinPool mainPool = new ForkJoinPool();
-+
+ * ...
-+
+ * public void sort(long[] array) {
-+
+ *   mainPool.invoke(new SortTask(array, 0, array.length));
-+
+ * }}</pre>
-+
+ *
+ * <p><b>Implementation notes</b>: This implementation restricts the
+ * maximum number of running threads to 32767. Attempts to create
-<
+ * pools with greater than the maximum result in
-<
+ * IllegalArgumentExceptions.
->
+ * pools with greater than the maximum number result in
->
+ * {@code IllegalArgumentException}.
->
+ *
->
+ * <p>This implementation rejects submitted tasks (that is, by throwing
->
+ * {@link RejectedExecutionException}) only when the pool is shut down
->
+ * or internal resources have been exhausted.
->
+ *
->
+ * @since 1.7
->
+ * @author Doug Lea
+ */
+public class ForkJoinPool extends AbstractExecutorService {
+    /*
-<
+     * See the extended comments interspersed below for design,
-<
+     * rationale, and walkthroughs.
-<
+     */
->
+     * Implementation Overview
->
+     *
->
+     * This class and its nested classes provide the main
->
+     * functionality and control for a set of worker threads:
->
+     * Submissions from non-FJ threads enter into submission queues.
->
+     * Workers take these tasks and typically split them into subtasks
->
+     * that may be stolen by other workers.  Preference rules give
->
+     * first priority to processing tasks from their own queues (LIFO
->
+     * or FIFO, depending on mode), then to randomized FIFO steals of
->
+     * tasks in other queues.
->
+     *
->
+     * WorkQueues
->
+     * ==========
->
+     *
->
+     * Most operations occur within work-stealing queues (in nested
->
+     * class WorkQueue).  These are special forms of Deques that
->
+     * support only three of the four possible end-operations -- push,
->
+     * pop, and poll (aka steal), under the further constraints that
->
+     * push and pop are called only from the owning thread (or, as
->
+     * extended here, under a lock), while poll may be called from
->
+     * other threads.  (If you are unfamiliar with them, you probably
->
+     * want to read Herlihy and Shavit's book "The Art of
->
+     * Multiprocessor programming", chapter 16 describing these in
->
+     * more detail before proceeding.)  The main work-stealing queue
->
+     * design is roughly similar to those in the papers "Dynamic
->
+     * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
->
+     * (http://research.sun.com/scalable/pubs/index.html) and
->
+     * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
->
+     * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
->
+     * The main differences ultimately stem from GC requirements that
->
+     * we null out taken slots as soon as we can, to maintain as small
->
+     * a footprint as possible even in programs generating huge
->
+     * numbers of tasks. To accomplish this, we shift the CAS
->
+     * arbitrating pop vs poll (steal) from being on the indices
->
+     * ("base" and "top") to the slots themselves.  So, both a
->
+     * successful pop and poll mainly entail a CAS of a slot from
->
+     * non-null to null.  Because we rely on CASes of references, we
->
+     * do not need tag bits on base or top.  They are simple ints as
->
+     * used in any circular array-based queue (see for example
->
+     * ArrayDeque).  Updates to the indices must still be ordered in a
->
+     * way that guarantees that top == base means the queue is empty,
->
+     * but otherwise may err on the side of possibly making the queue
->
+     * appear nonempty when a push, pop, or poll have not fully
->
+     * committed. Note that this means that the poll operation,
->
+     * considered individually, is not wait-free. One thief cannot
->
+     * successfully continue until another in-progress one (or, if
->
+     * previously empty, a push) completes.  However, in the
->
+     * aggregate, we ensure at least probabilistic non-blockingness.
->
+     * If an attempted steal fails, a thief always chooses a different
->
+     * random victim target to try next. So, in order for one thief to
->
+     * progress, it suffices for any in-progress poll or new push on
->
+     * any empty queue to complete.
->
+     *
->
+     * This approach also enables support of a user mode in which local
->
+     * task processing is in FIFO, not LIFO order, simply by using
->
+     * poll rather than pop.  This can be useful in message-passing
->
+     * frameworks in which tasks are never joined.  However neither
->
+     * mode considers affinities, loads, cache localities, etc, so
->
+     * rarely provide the best possible performance on a given
->
+     * machine, but portably provide good throughput by averaging over
->
+     * these factors.  (Further, even if we did try to use such
->
+     * information, we do not usually have a basis for exploiting it.
->
+     * For example, some sets of tasks profit from cache affinities,
->
+     * but others are harmed by cache pollution effects.)
->
+     *
->
+     * WorkQueues are also used in a similar way for tasks submitted
->
+     * to the pool. We cannot mix these tasks in the same queues used
->
+     * for work-stealing (this would contaminate lifo/fifo
->
+     * processing). Instead, we loosely associate submission queues
->
+     * with submitting threads, using a form of hashing.  The
->
+     * ThreadLocal Submitter class contains a value initially used as
->
+     * a hash code for choosing existing queues, but may be randomly
->
+     * repositioned upon contention with other submitters.  In
->
+     * essence, submitters act like workers except that they never
->
+     * take tasks, and they are multiplexed on to a finite number of
->
+     * shared work queues. However, classes are set up so that future
->
+     * extensions could allow submitters to optionally help perform
->
+     * tasks as well. Insertion of tasks in shared mode requires a
->
+     * lock (mainly to protect in the case of resizing) but we use
->
+     * only a simple spinlock (using bits in field runState), because
->
+     * submitters encountering a busy queue move on to try or create
->
+     * other queues, so never block.
->
+     *
->
+     * Management
->
+     * ==========
->
+     *
->
+     * The main throughput advantages of work-stealing stem from
->
+     * decentralized control -- workers mostly take tasks from
->
+     * themselves or each other. We cannot negate this in the
->
+     * implementation of other management responsibilities. The main
->
+     * tactic for avoiding bottlenecks is packing nearly all
->
+     * essentially atomic control state into two volatile variables
->
+     * that are by far most often read (not written) as status and
->
+     * consistency checks.
->
+     *
->
+     * Field "ctl" contains 64 bits holding all the information needed
->
+     * to atomically decide to add, inactivate, enqueue (on an event
->
+     * queue), dequeue, and/or re-activate workers.  To enable this
->
+     * packing, we restrict maximum parallelism to (1<<15)-1 (which is
->
+     * far in excess of normal operating range) to allow ids, counts,
->
+     * and their negations (used for thresholding) to fit into 16bit
->
+     * fields.
->
+     *
->
+     * Field "runState" contains 32 bits needed to register and
->
+     * deregister WorkQueues, as well as to enable shutdown. It is
->
+     * only modified under a lock (normally briefly held, but
->
+     * occasionally protecting allocations and resizings) but even
->
+     * when locked remains available to check consistency. An
->
+     * auxiliary field "growHints", also only modified under lock,
->
+     * contains a candidate index for the next WorkQueue and
->
+     * a mask for submission queue indices.
->
+     *
->
+     * Recording WorkQueues.  WorkQueues are recorded in the
->
+     * "workQueues" array that is created upon pool construction and
->
+     * expanded if necessary.  Updates to the array while recording
->
+     * new workers and unrecording terminated ones are protected from
->
+     * each other by a lock but the array is otherwise concurrently
->
+     * readable, and accessed directly.  To simplify index-based
->
+     * operations, the array size is always a power of two, and all
->
+     * readers must tolerate null slots. Shared (submission) queues
->
+     * are at even indices, worker queues at odd indices. Grouping
->
+     * them together in this way simplifies and speeds up task
->
+     * scanning. To avoid flailing during start-up, the array is
->
+     * presized to hold twice #parallelism workers (which is unlikely
->
+     * to need further resizing during execution). But to avoid
->
+     * dealing with so many null slots, variable runState includes a
->
+     * mask for the nearest power of two that contains all currently
->
+     * used indices.
->
+     *
->
+     * All worker thread creation is on-demand, triggered by task
->
+     * submissions, replacement of terminated workers, and/or
->
+     * compensation for blocked workers. However, all other support
->
+     * code is set up to work with other policies.  To ensure that we
->
+     * do not hold on to worker references that would prevent GC, ALL
->
+     * accesses to workQueues are via indices into the workQueues
->
+     * array (which is one source of some of the messy code
->
+     * constructions here). In essence, the workQueues array serves as
->
+     * a weak reference mechanism. Thus for example the wait queue
->
+     * field of ctl stores indices, not references.  Access to the
->
+     * workQueues in associated methods (for example signalWork) must
->
+     * both index-check and null-check the IDs. All such accesses
->
+     * ignore bad IDs by returning out early from what they are doing,
->
+     * since this can only be associated with termination, in which
->
+     * case it is OK to give up.  All uses of the workQueues array
->
+     * also check that it is non-null (even if previously
->
+     * non-null). This allows nulling during termination, which is
->
+     * currently not necessary, but remains an option for
->
+     * resource-revocation-based shutdown schemes. It also helps
->
+     * reduce JIT issuance of uncommon-trap code, which tends to
->
+     * unnecessarily complicate control flow in some methods.
->
+     *
->
+     * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
->
+     * let workers spin indefinitely scanning for tasks when none can
->
+     * be found immediately, and we cannot start/resume workers unless
->
+     * there appear to be tasks available.  On the other hand, we must
->
+     * quickly prod them into action when new tasks are submitted or
->
+     * generated. In many usages, ramp-up time to activate workers is
->
+     * the main limiting factor in overall performance (this is
->
+     * compounded at program start-up by JIT compilation and
->
+     * allocation). So we try to streamline this as much as possible.
->
+     * We park/unpark workers after placing in an event wait queue
->
+     * when they cannot find work. This "queue" is actually a simple
->
+     * Treiber stack, headed by the "id" field of ctl, plus a 15bit
->
+     * counter value (that reflects the number of times a worker has
->
+     * been inactivated) to avoid ABA effects (we need only as many
->
+     * version numbers as worker threads). Successors are held in
->
+     * field WorkQueue.nextWait.  Queuing deals with several intrinsic
->
+     * races, mainly that a task-producing thread can miss seeing (and
->
+     * signalling) another thread that gave up looking for work but
->
+     * has not yet entered the wait queue. We solve this by requiring
->
+     * a full sweep of all workers (via repeated calls to method
->
+     * scan()) both before and after a newly waiting worker is added
->
+     * to the wait queue. During a rescan, the worker might release
->
+     * some other queued worker rather than itself, which has the same
->
+     * net effect. Because enqueued workers may actually be rescanning
->
+     * rather than waiting, we set and clear the "parker" field of
->
+     * WorkQueues to reduce unnecessary calls to unpark.  (This
->
+     * requires a secondary recheck to avoid missed signals.)  Note
->
+     * the unusual conventions about Thread.interrupts surrounding
->
+     * parking and other blocking: Because interrupts are used solely
->
+     * to alert threads to check termination, which is checked anyway
->
+     * upon blocking, we clear status (using Thread.interrupted)
->
+     * before any call to park, so that park does not immediately
->
+     * return due to status being set via some other unrelated call to
->
+     * interrupt in user code.
->
+     *
->
+     * Signalling.  We create or wake up workers only when there
->
+     * appears to be at least one task they might be able to find and
->
+     * execute.  When a submission is added or another worker adds a
->
+     * task to a queue that previously had fewer than two tasks, they
->
+     * signal waiting workers (or trigger creation of new ones if
->
+     * fewer than the given parallelism level -- see signalWork).
->
+     * These primary signals are buttressed by signals during rescans;
->
+     * together these cover the signals needed in cases when more
->
+     * tasks are pushed but untaken, and improve performance compared
->
+     * to having one thread wake up all workers.
->
+     *
->
+     * Trimming workers. To release resources after periods of lack of
->
+     * use, a worker starting to wait when the pool is quiescent will
->
+     * time out and terminate if the pool has remained quiescent for
->
+     * SHRINK_RATE nanosecs. This will slowly propagate, eventually
->
+     * terminating all workers after long periods of non-use.
->
+     *
->
+     * Shutdown and Termination. A call to shutdownNow atomically sets
->
+     * a runState bit and then (non-atomically) sets each worker's
->
+     * runState status, cancels all unprocessed tasks, and wakes up
->
+     * all waiting workers.  Detecting whether termination should
->
+     * commence after a non-abrupt shutdown() call requires more work
->
+     * and bookkeeping. We need consensus about quiescence (i.e., that
->
+     * there is no more work). The active count provides a primary
->
+     * indication but non-abrupt shutdown still requires a rechecking
->
+     * scan for any workers that are inactive but not queued.
->
+     *
->
+     * Joining Tasks
->
+     * =============
->
+     *
->
+     * Any of several actions may be taken when one worker is waiting
->
+     * to join a task stolen (or always held) by another.  Because we
->
+     * are multiplexing many tasks on to a pool of workers, we can't
->
+     * just let them block (as in Thread.join).  We also cannot just
->
+     * reassign the joiner's run-time stack with another and replace
->
+     * it later, which would be a form of "continuation", that even if
->
+     * possible is not necessarily a good idea since we sometimes need
->
+     * both an unblocked task and its continuation to progress.
->
+     * Instead we combine two tactics:
->
+     *
->
+     *   Helping: Arranging for the joiner to execute some task that it
->
+     *      would be running if the steal had not occurred.
->
+     *
->
+     *   Compensating: Unless there are already enough live threads,
->
+     *      method tryCompensate() may create or re-activate a spare
->
+     *      thread to compensate for blocked joiners until they unblock.
->
+     *
->
+     * A third form (implemented in tryRemoveAndExec and
->
+     * tryPollForAndExec) amounts to helping a hypothetical
->
+     * compensator: If we can readily tell that a possible action of a
->
+     * compensator is to steal and execute the task being joined, the
->
+     * joining thread can do so directly, without the need for a
->
+     * compensation thread (although at the expense of larger run-time
->
+     * stacks, but the tradeoff is typically worthwhile).
->
+     *
->
+     * The ManagedBlocker extension API can't use helping so relies
->
+     * only on compensation in method awaitBlocker.
->
+     *
->
+     * The algorithm in tryHelpStealer entails a form of "linear"
->
+     * helping: Each worker records (in field currentSteal) the most
->
+     * recent task it stole from some other worker. Plus, it records
->
+     * (in field currentJoin) the task it is currently actively
->
+     * joining. Method tryHelpStealer uses these markers to try to
->
+     * find a worker to help (i.e., steal back a task from and execute
->
+     * it) that could hasten completion of the actively joined task.
->
+     * In essence, the joiner executes a task that would be on its own
->
+     * local deque had the to-be-joined task not been stolen. This may
->
+     * be seen as a conservative variant of the approach in Wagner &
->
+     * Calder "Leapfrogging: a portable technique for implementing
->
+     * efficient futures" SIGPLAN Notices, 1993
->
+     * (http://portal.acm.org/citation.cfm?id=155354). It differs in
->
+     * that: (1) We only maintain dependency links across workers upon
->
+     * steals, rather than use per-task bookkeeping.  This sometimes
->
+     * requires a linear scan of workQueues array to locate stealers, but
->
+     * often doesn't because stealers leave hints (that may become
->
+     * stale/wrong) of where to locate them.  A stealHint is only a
->
+     * hint because a worker might have had multiple steals and the
->
+     * hint records only one of them (usually the most current).
->
+     * Hinting isolates cost to when it is needed, rather than adding
->
+     * to per-task overhead.  (2) It is "shallow", ignoring nesting
->
+     * and potentially cyclic mutual steals.  (3) It is intentionally
->
+     * racy: field currentJoin is updated only while actively joining,
->
+     * which means that we miss links in the chain during long-lived
->
+     * tasks, GC stalls etc (which is OK since blocking in such cases
->
+     * is usually a good idea).  (4) We bound the number of attempts
->
+     * to find work (see MAX_HELP_DEPTH) and fall back to suspending
->
+     * the worker and if necessary replacing it with another.
->
+     *
->
+     * It is impossible to keep exactly the target parallelism number
->
+     * of threads running at any given time.  Determining the
->
+     * existence of conservatively safe helping targets, the
->
+     * availability of already-created spares, and the apparent need
->
+     * to create new spares are all racy, so we rely on multiple
->
+     * retries of each.  Currently, in keeping with on-demand
->
+     * signalling policy, we compensate only if blocking would leave
->
+     * less than one active (non-waiting, non-blocked) worker.
->
+     * Additionally, to avoid some false alarms due to GC, lagging
->
+     * counters, system activity, etc, compensated blocking for joins
->
+     * is only attempted after rechecks stabilize in
->
+     * ForkJoinTask.awaitJoin. (Retries are interspersed with
->
+     * Thread.yield, for good citizenship.)
->
+     *
->
+     * Style notes: There is a lot of representation-level coupling
->
+     * among classes ForkJoinPool, ForkJoinWorkerThread, and
->
+     * ForkJoinTask.  The fields of WorkQueue maintain data structures
->
+     * managed by ForkJoinPool, so are directly accessed.  There is
->
+     * little point trying to reduce this, since any associated future
->
+     * changes in representations will need to be accompanied by
->
+     * algorithmic changes anyway. Several methods intrinsically
->
+     * sprawl because they must accumulate sets of consistent reads of
->
+     * volatiles held in local variables.  Methods signalWork() and
->
+     * scan() are the main bottlenecks, so are especially heavily
->
+     * micro-optimized/mangled.  There are lots of inline assignments
->
+     * (of form "while ((local = field) != 0)") which are usually the
->
+     * simplest way to ensure the required read orderings (which are
->
+     * sometimes critical). This leads to a "C"-like style of listing
->
+     * declarations of these locals at the heads of methods or blocks.
->
+     * There are several occurrences of the unusual "do {} while
->
+     * (!cas...)"  which is the simplest way to force an update of a
->
+     * CAS'ed variable. There are also other coding oddities that help
->
+     * some methods perform reasonably even when interpreted (not
->
+     * compiled).
->
+     *
->
+     * The order of declarations in this file is:
->
+     * (1) Static utility functions
->
+     * (2) Nested (static) classes
->
+     * (3) Static fields
->
+     * (4) Fields, along with constants used when unpacking some of them
->
+     * (5) Internal control methods
->
+     * (6) Callbacks and other support for ForkJoinTask methods
->
+     * (7) Exported methods
->
+     * (8) Static block initializing statics in minimally dependent order
->
+     */
->
->
+    // Static utilities
->
->
+    /**
->
+     * Computes an initial hash code (also serving as a non-zero
->
+     * random seed) for a thread id. This method is expected to
->
+     * provide higher-quality hash codes than using method hashCode().
->
+     */
->
+    static final int hashId(long id) {
->
+        int h = (int)id ^ (int)(id >>> 32); // Use MurmurHash of thread id
->
+        h ^= h >>> 16; h *= 0x85ebca6b;
->
+        h ^= h >>> 13; h *= 0xc2b2ae35;
->
+        h ^= h >>> 16;
->
+        return (h == 0)? 1 : h; // ensure nonzero
->
+    }
-<
+    /** Mask for packing and unpacking shorts */
-<
+    private static final int  shortMask = 0xffff;
->
+    /**
->
+     * If there is a security manager, makes sure caller has
->
+     * permission to modify threads.
->
+     */
->
+    private static void checkPermission() {
->
+        SecurityManager security = System.getSecurityManager();
->
+        if (security != null)
->
+            security.checkPermission(modifyThreadPermission);
->
+    }
-<
+    /** Max pool size -- must be a power of two minus 1 */
-<
+    private static final int MAX_THREADS =  0x7FFF;
->
+    // Nested classes
+    /**
-<
+     * Factory for creating new ForkJoinWorkerThreads.  A
-<
+     * ForkJoinWorkerThreadFactory must be defined and used for
-<
+     * ForkJoinWorkerThread subclasses that extend base functionality
-<
+     * or initialize threads with different contexts.
->
+     * Factory for creating new {@link ForkJoinWorkerThread}s.
->
+     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
->
+     * for {@code ForkJoinWorkerThread} subclasses that extend base
->
+     * functionality or initialize threads with different contexts.
+     */
+    public static interface ForkJoinWorkerThreadFactory {
+        /**
+         * Returns a new worker thread operating in the given pool.
+         * @param pool the pool this thread works in
-<
+         * @throws NullPointerException if pool is null;
->
+         * @throws NullPointerException if the pool is null
+         */
+        public ForkJoinWorkerThread newThread(ForkJoinPool pool);
+    /**
-<
+     * Default ForkJoinWorkerThreadFactory implementation, creates a
->
+     * Default ForkJoinWorkerThreadFactory implementation; creates a
+     * new ForkJoinWorkerThread.
+     */
-<
+    static class  DefaultForkJoinWorkerThreadFactory
->
+    static class DefaultForkJoinWorkerThreadFactory
+        implements ForkJoinWorkerThreadFactory {
+        public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
-<
+            try {
-<
+                return new ForkJoinWorkerThread(pool);
-<
+            } catch (OutOfMemoryError oom)  {
-<
+                return null;
-<
+            }
->
+            return new ForkJoinWorkerThread(pool);
+    /**
-<
+     * Creates a new ForkJoinWorkerThread. This factory is used unless
-<
+     * overridden in ForkJoinPool constructors.
-<
+     */
-<
+    public static final ForkJoinWorkerThreadFactory
-<
+        defaultForkJoinWorkerThreadFactory =
-<
+        new DefaultForkJoinWorkerThreadFactory();
-<
-<
+    /**
-<
+     * Permission required for callers of methods that may start or
-<
+     * kill threads.
->
+     * A simple non-reentrant lock used for exclusion when managing
->
+     * queues and workers. We use a custom lock so that we can readily
->
+     * probe lock state in constructions that check among alternative
->
+     * actions. The lock is normally only very briefly held, and
->
+     * sometimes treated as a spinlock, but other usages block to
->
+     * reduce overall contention in those cases where locked code
->
+     * bodies perform allocation/resizing.
+     */
-<
+    private static final RuntimePermission modifyThreadPermission =
-<
+        new RuntimePermission("modifyThread");
->
+    static final class Mutex extends AbstractQueuedSynchronizer {
->
+        public final boolean tryAcquire(int ignore) {
->
+            return compareAndSetState(0, 1);
->
+        }
->
+        public final boolean tryRelease(int ignore) {
->
+            setState(0);
->
+            return true;
->
+        }
->
+        public final void lock() { acquire(0); }
->
+        public final void unlock() { release(0); }
->
+        public final boolean isHeldExclusively() { return getState() == 1; }
->
+        public final Condition newCondition() { return new ConditionObject(); }
->
+    }
+    /**
-<
+     * If there is a security manager, makes sure caller has
-<
+     * permission to modify threads.
->
+     * Class for artificial tasks that are used to replace the target
->
+     * of local joins if they are removed from an interior queue slot
->
+     * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
->
+     * actually do anything beyond having a unique identity.
+     */
-<
+    private static void checkPermission() {
-<
+        SecurityManager security = System.getSecurityManager();
-<
+        if (security != null)
-<
+            security.checkPermission(modifyThreadPermission);
->
+    static final class EmptyTask extends ForkJoinTask<Void> {
->
+        EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
->
+        public final Void getRawResult() { return null; }
->
+        public final void setRawResult(Void x) {}
->
+        public final boolean exec() { return true; }
+    /**
-<
+     * Generator for assigning sequence numbers as pool names.
->
+     * Queues supporting work-stealing as well as external task
->
+     * submission. See above for main rationale and algorithms.
->
+     * Implementation relies heavily on "Unsafe" intrinsics
->
+     * and selective use of "volatile":
->
+     *
->
+     * Field "base" is the index (mod array.length) of the least valid
->
+     * queue slot, which is always the next position to steal (poll)
->
+     * from if nonempty. Reads and writes require volatile orderings
->
+     * but not CAS, because updates are only performed after slot
->
+     * CASes.
->
+     *
->
+     * Field "top" is the index (mod array.length) of the next queue
->
+     * slot to push to or pop from. It is written only by owner thread
->
+     * for push, or under lock for trySharedPush, and accessed by
->
+     * other threads only after reading (volatile) base.  Both top and
->
+     * base are allowed to wrap around on overflow, but (top - base)
->
+     * (or more commonly -(base - top) to force volatile read of base
->
+     * before top) still estimates size.
->
+     *
->
+     * The array slots are read and written using the emulation of
->
+     * volatiles/atomics provided by Unsafe. Insertions must in
->
+     * general use putOrderedObject as a form of releasing store to
->
+     * ensure that all writes to the task object are ordered before
->
+     * its publication in the queue. (Although we can avoid one case
->
+     * of this when locked in trySharedPush.) All removals entail a
->
+     * CAS to null.  The array is always a power of two. To ensure
->
+     * safety of Unsafe array operations, all accesses perform
->
+     * explicit null checks and implicit bounds checks via
->
+     * power-of-two masking.
->
+     *
->
+     * In addition to basic queuing support, this class contains
->
+     * fields described elsewhere to control execution. It turns out
->
+     * to work better memory-layout-wise to include them in this
->
+     * class rather than a separate class.
->
+     *
->
+     * Performance on most platforms is very sensitive to placement of
->
+     * instances of both WorkQueues and their arrays -- we absolutely
->
+     * do not want multiple WorkQueue instances or multiple queue
->
+     * arrays sharing cache lines. (It would be best for queue objects
->
+     * and their arrays to share, but there is nothing available to
->
+     * help arrange that).  Unfortunately, because they are recorded
->
+     * in a common array, WorkQueue instances are often moved to be
->
+     * adjacent by garbage collectors. To reduce impact, we use field
->
+     * padding that works OK on common platforms; this effectively
->
+     * trades off slightly slower average field access for the sake of
->
+     * avoiding really bad worst-case access. (Until better JVM
->
+     * support is in place, this padding is dependent on transient
->
+     * properties of JVM field layout rules.)  We also take care in
->
+     * allocating, sizing and resizing the array. Non-shared queue
->
+     * arrays are initialized (via method growArray) by workers before
->
+     * use. Others are allocated on first use.
+     */
-<
+    private static final AtomicInteger poolNumberGenerator =
-<
+        new AtomicInteger();
->
+    static final class WorkQueue {
->
+        /**
->
+         * Capacity of work-stealing queue array upon initialization.
->
+         * Must be a power of two; at least 4, but set larger to
->
+         * reduce cacheline sharing among queues.
->
+         */
->
+        static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
-<
+    /**
-<
+     * Array holding all worker threads in the pool. Array size must
-<
+     * be a power of two.  Updates and replacements are protected by
-<
+     * workerLock, but it is always kept in a consistent enough state
-<
+     * to be randomly accessed without locking by workers performing
-<
+     * work-stealing.
-<
+     */
-<
+    volatile ForkJoinWorkerThread[] workers;
->
+        /**
->
+         * Maximum size for queue arrays. Must be a power of two less
->
+         * than or equal to 1 << (31 - width of array entry) to ensure
->
+         * lack of wraparound of index calculations, but defined to a
->
+         * value a bit less than this to help users trap runaway
->
+         * programs before saturating systems.
->
+         */
->
+        static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
-<
+    /**
-<
+     * Lock protecting access to workers.
-<
+     */
-<
+    private final ReentrantLock workerLock;
->
+        volatile long totalSteals; // cumulative number of steals
->
+        int seed;                  // for random scanning; initialize nonzero
->
+        volatile int eventCount;   // encoded inactivation count; < 0 if inactive
->
+        int nextWait;              // encoded record of next event waiter
->
+        int rescans;               // remaining scans until block
->
+        int nsteals;               // top-level task executions since last idle
->
+        final int mode;            // lifo, fifo, or shared
->
+        int poolIndex;             // index of this queue in pool (or 0)
->
+        int stealHint;             // index of most recent known stealer
->
+        volatile int runState;     // 1: locked, -1: terminate; else 0
->
+        volatile int base;         // index of next slot for poll
->
+        int top;                   // index of next slot for push
->
+        ForkJoinTask<?>[] array;   // the elements (initially unallocated)
->
+        final ForkJoinWorkerThread owner; // owning thread or null if shared
->
+        volatile Thread parker;    // == owner during call to park; else null
->
+        ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
->
+        ForkJoinTask<?> currentSteal; // current non-local task being executed
->
+        // Heuristic padding to ameliorate unfortunate memory placements
->
+        Object p00, p01, p02, p03, p04, p05, p06, p07, p08, p09, p0a;
->
->
+        WorkQueue(ForkJoinWorkerThread owner, int mode) {
->
+            this.owner = owner;
->
+            this.mode = mode;
->
+            // Place indices in the center of array (that is not yet allocated)
->
+            base = top = INITIAL_QUEUE_CAPACITY >>> 1;
->
+        }
-<
+    /**
-<
+     * Condition for awaitTermination.
-<
+     */
-<
+    private final Condition termination;
->
+        /**
->
+         * Returns number of tasks in the queue.
->
+         */
->
+        final int queueSize() {
->
+            int n = base - top; // non-owner callers must read base first
->
+            return (n >= 0) ? 0 : -n;
->
+        }
-<
+    /**
-<
+     * The uncaught exception handler used when any worker
-<
+     * abrupty terminates
-<
+     */
-<
+    private Thread.UncaughtExceptionHandler ueh;
->
+        /**
->
+         * Pushes a task. Call only by owner in unshared queues.
->
+         *
->
+         * @param task the task. Caller must ensure non-null.
->
+         * @param p if non-null, pool to signal if necessary
->
+         * @throw RejectedExecutionException if array cannot be resized
->
+         */
->
+        final void push(ForkJoinTask<?> task, ForkJoinPool p) {
->
+            ForkJoinTask<?>[] a;
->
+            int s = top, m, n;
->
+            if ((a = array) != null) {    // ignore if queue removed
->
+                U.putOrderedObject
->
+                    (a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
->
+                if ((n = (top = s + 1) - base) <= 2) {
->
+                    if (p != null)
->
+                        p.signalWork();
->
+                }
->
+                else if (n >= m)
->
+                    growArray(true);
->
+            }
->
+        }
-<
+    /**
-<
+     * Creation factory for worker threads.
-<
+     */
-<
+    private final ForkJoinWorkerThreadFactory factory;
->
+        /**
->
+         * Pushes a task if lock is free and array is either big
->
+         * enough or can be resized to be big enough.
->
+         *
->
+         * @param task the task. Caller must ensure non-null.
->
+         * @return true if submitted
->
+         */
->
+        final boolean trySharedPush(ForkJoinTask<?> task) {
->
+            boolean submitted = false;
->
+            if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
->
+                ForkJoinTask<?>[] a = array;
->
+                int s = top;
->
+                try {
->
+                    if ((a != null && a.length > s + 1 - base) ||
->
+                        (a = growArray(false)) != null) { // must presize
->
+                        int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
->
+                        U.putObject(a, (long)j, task);    // don't need "ordered"
->
+                        top = s + 1;
->
+                        submitted = true;
->
+                    }
->
+                } finally {
->
+                    runState = 0;                         // unlock
->
+                }
->
+            }
->
+            return submitted;
->
+        }
->
->
+        /**
->
+         * Takes next task, if one exists, in FIFO order.
->
+         */
->
+        final ForkJoinTask<?> poll() {
->
+            ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
->
+            while ((b = base) - top < 0 && (a = array) != null) {
->
+                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
->
+                if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
->
+                    base == b &&
->
+                    U.compareAndSwapObject(a, j, t, null)) {
->
+                    base = b + 1;
->
+                    return t;
->
+                }
->
+            }
->
+            return null;
->
+        }
->
->
+        /**
->
+         * Takes next task, if one exists, in LIFO order.  Call only
->
+         * by owner in unshared queues. (We do not have a shared
->
+         * version of this method because it is never needed.)
->
+         */
->
+        final ForkJoinTask<?> pop() {
->
+            ForkJoinTask<?> t; int m;
->
+            ForkJoinTask<?>[] a = array;
->
+            if (a != null && (m = a.length - 1) >= 0) {
->
+                for (int s; (s = top - 1) - base >= 0;) {
->
+                    int j = ((m & s) << ASHIFT) + ABASE;
->
+                    if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) == null)
->
+                        break;
->
+                    if (U.compareAndSwapObject(a, j, t, null)) {
->
+                        top = s;
->
+                        return t;
->
+                    }
->
+                }
->
+            }
->
+            return null;
->
+        }
->
->
+        /**
->
+         * Takes next task, if one exists, in order specified by mode.
->
+         */
->
+        final ForkJoinTask<?> nextLocalTask() {
->
+            return mode == 0 ? pop() : poll();
->
+        }
->
->
+        /**
->
+         * Returns next task, if one exists, in order specified by mode.
->
+         */
->
+        final ForkJoinTask<?> peek() {
->
+            ForkJoinTask<?>[] a = array; int m;
->
+            if (a == null || (m = a.length - 1) < 0)
->
+                return null;
->
+            int i = mode == 0 ? top - 1 : base;
->
+            int j = ((i & m) << ASHIFT) + ABASE;
->
+            return (ForkJoinTask<?>)U.getObjectVolatile(a, j);
->
+        }
->
->
+        /**
->
+         * Returns task at index b if b is current base of queue.
->
+         */
->
+        final ForkJoinTask<?> pollAt(int b) {
->
+            ForkJoinTask<?> t; ForkJoinTask<?>[] a;
->
+            if ((a = array) != null) {
->
+                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
->
+                if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
->
+                    base == b &&
->
+                    U.compareAndSwapObject(a, j, t, null)) {
->
+                    base = b + 1;
->
+                    return t;
->
+                }
->
+            }
->
+            return null;
->
+        }
->
->
+        /**
->
+         * Pops the given task only if it is at the current top.
->
+         */
->
+        final boolean tryUnpush(ForkJoinTask<?> t) {
->
+            ForkJoinTask<?>[] a; int s;
->
+            if ((a = array) != null && (s = top) != base &&
->
+                U.compareAndSwapObject
->
+                (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
->
+                top = s;
->
+                return true;
->
+            }
->
+            return false;
->
+        }
->
->
+        /**
->
+         * Polls the given task only if it is at the current base.
->
+         */
->
+        final boolean pollFor(ForkJoinTask<?> task) {
->
+            ForkJoinTask<?>[] a; int b;
->
+            if ((b = base) - top < 0 && (a = array) != null) {
->
+                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
->
+                if (U.getObjectVolatile(a, j) == task && base == b &&
->
+                    U.compareAndSwapObject(a, j, task, null)) {
->
+                    base = b + 1;
->
+                    return true;
->
+                }
->
+            }
->
+            return false;
->
+        }
->
->
+        /**
->
+         * If present, removes from queue and executes the given task, or
->
+         * any other cancelled task. Returns (true) immediately on any CAS
->
+         * or consistency check failure so caller can retry.
->
+         *
->
+         * @return false if no progress can be made
->
+         */
->
+        final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
->
+            boolean removed = false, empty = true, progress = true;
->
+            ForkJoinTask<?>[] a; int m, s, b, n;
->
+            if ((a = array) != null && (m = a.length - 1) >= 0 &&
->
+                (n = (s = top) - (b = base)) > 0) {
->
+                for (ForkJoinTask<?> t;;) {           // traverse from s to b
->
+                    int j = ((--s & m) << ASHIFT) + ABASE;
->
+                    t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
->
+                    if (t == null)                    // inconsistent length
->
+                        break;
->
+                    else if (t == task) {
->
+                        if (s + 1 == top) {           // pop
->
+                            if (!U.compareAndSwapObject(a, j, task, null))
->
+                                break;
->
+                            top = s;
->
+                            removed = true;
->
+                        }
->
+                        else if (base == b)           // replace with proxy
->
+                            removed = U.compareAndSwapObject(a, j, task,
->
+                                                             new EmptyTask());
->
+                        break;
->
+                    }
->
+                    else if (t.status >= 0)
->
+                        empty = false;
->
+                    else if (s + 1 == top) {          // pop and throw away
->
+                        if (U.compareAndSwapObject(a, j, t, null))
->
+                            top = s;
->
+                        break;
->
+                    }
->
+                    if (--n == 0) {
->
+                        if (!empty && base == b)
->
+                            progress = false;
->
+                        break;
->
+                    }
->
+                }
->
+            }
->
+            if (removed)
->
+                task.doExec();
->
+            return progress;
->
+        }
->
->
+        /**
->
+         * Initializes or doubles the capacity of array. Call either
->
+         * by owner or with lock held -- it is OK for base, but not
->
+         * top, to move while resizings are in progress.
->
+         *
->
+         * @param rejectOnFailure if true, throw exception if capacity
->
+         * exceeded (relayed ultimately to user); else return null.
->
+         */
->
+        final ForkJoinTask<?>[] growArray(boolean rejectOnFailure) {
->
+            ForkJoinTask<?>[] oldA = array;
->
+            int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
->
+            if (size <= MAXIMUM_QUEUE_CAPACITY) {
->
+                int oldMask, t, b;
->
+                ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
->
+                if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
->
+                    (t = top) - (b = base) > 0) {
->
+                    int mask = size - 1;
->
+                    do {
->
+                        ForkJoinTask<?> x;
->
+                        int oldj = ((b & oldMask) << ASHIFT) + ABASE;
->
+                        int j    = ((b &    mask) << ASHIFT) + ABASE;
->
+                        x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
->
+                        if (x != null &&
->
+                            U.compareAndSwapObject(oldA, oldj, x, null))
->
+                            U.putObjectVolatile(a, j, x);
->
+                    } while (++b != t);
->
+                }
->
+                return a;
->
+            }
->
+            else if (!rejectOnFailure)
->
+                return null;
->
+            else
->
+                throw new RejectedExecutionException("Queue capacity exceeded");
->
+        }
->
->
+        /**
->
+         * Removes and cancels all known tasks, ignoring any exceptions.
->
+         */
->
+        final void cancelAll() {
->
+            ForkJoinTask.cancelIgnoringExceptions(currentJoin);
->
+            ForkJoinTask.cancelIgnoringExceptions(currentSteal);
->
+            for (ForkJoinTask<?> t; (t = poll()) != null; )
->
+                ForkJoinTask.cancelIgnoringExceptions(t);
->
+        }
->
->
+        /**
->
+         * Computes next value for random probes.  Scans don't require
->
+         * a very high quality generator, but also not a crummy one.
->
+         * Marsaglia xor-shift is cheap and works well enough.  Note:
->
+         * This is manually inlined in several usages in ForkJoinPool
->
+         * to avoid writes inside busy scan loops.
->
+         */
->
+        final int nextSeed() {
->
+            int r = seed;
->
+            r ^= r << 13;
->
+            r ^= r >>> 17;
->
+            return seed = r ^= r << 5;
->
+        }
->
->
+        // Execution methods
->
->
+        /**
->
+         * Removes and runs tasks until empty, using local mode
->
+         * ordering.
->
+         */
->
+        final void runLocalTasks() {
->
+            if (base - top < 0) {
->
+                for (ForkJoinTask<?> t; (t = nextLocalTask()) != null; )
->
+                    t.doExec();
->
+            }
->
+        }
->
->
+        /**
->
+         * Executes a top-level task and any local tasks remaining
->
+         * after execution.
->
+         *
->
+         * @return true unless terminating
->
+         */
->
+        final boolean runTask(ForkJoinTask<?> t) {
->
+            boolean alive = true;
->
+            if (t != null) {
->
+                currentSteal = t;
->
+                t.doExec();
->
+                runLocalTasks();
->
+                ++nsteals;
->
+                currentSteal = null;
->
+            }
->
+            else if (runState < 0)            // terminating
->
+                alive = false;
->
+            return alive;
->
+        }
->
->
+        /**
->
+         * Executes a non-top-level (stolen) task.
->
+         */
->
+        final void runSubtask(ForkJoinTask<?> t) {
->
+            if (t != null) {
->
+                ForkJoinTask<?> ps = currentSteal;
->
+                currentSteal = t;
->
+                t.doExec();
->
+                currentSteal = ps;
->
+            }
->
+        }
->
->
+        /**
->
+         * Returns true if owned and not known to be blocked.
->
+         */
->
+        final boolean isApparentlyUnblocked() {
->
+            Thread wt; Thread.State s;
->
+            return (eventCount >= 0 &&
->
+                    (wt = owner) != null &&
->
+                    (s = wt.getState()) != Thread.State.BLOCKED &&
->
+                    s != Thread.State.WAITING &&
->
+                    s != Thread.State.TIMED_WAITING);
->
+        }
->
->
+        /**
->
+         * If this owned and is not already interrupted, try to
->
+         * interrupt and/or unpark, ignoring exceptions.
->
+         */
->
+        final void interruptOwner() {
->
+            Thread wt, p;
->
+            if ((wt = owner) != null && !wt.isInterrupted()) {
->
+                try {
->
+                    wt.interrupt();
->
+                } catch (SecurityException ignore) {
->
+                }
->
+            }
->
+            if ((p = parker) != null)
->
+                U.unpark(p);
->
+        }
->
->
+        // Unsafe mechanics
->
+        private static final sun.misc.Unsafe U;
->
+        private static final long RUNSTATE;
->
+        private static final int ABASE;
->
+        private static final int ASHIFT;
->
+        static {
->
+            int s;
->
+            try {
->
+                U = getUnsafe();
->
+                Class<?> k = WorkQueue.class;
->
+                Class<?> ak = ForkJoinTask[].class;
->
+                RUNSTATE = U.objectFieldOffset
->
+                    (k.getDeclaredField("runState"));
->
+                ABASE = U.arrayBaseOffset(ak);
->
+                s = U.arrayIndexScale(ak);
->
+            } catch (Exception e) {
->
+                throw new Error(e);
->
+            }
->
+            if ((s & (s-1)) != 0)
->
+                throw new Error("data type scale not a power of two");
->
+            ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
->
+        }
->
+    }
+    /**
-<
+     * Head of stack of threads that were created to maintain
-<
+     * parallelism when other threads blocked, but have since
-<
+     * suspended when the parallelism level rose.
->
+     * Per-thread records for threads that submit to pools. Currently
->
+     * holds only pseudo-random seed / index that is used to choose
->
+     * submission queues in method doSubmit. In the future, this may
->
+     * also incorporate a means to implement different task rejection
->
+     * and resubmission policies.
+     */
-<
+    private volatile WaitQueueNode spareStack;
->
+    static final class Submitter {
->
+        int seed;
->
+        Submitter() { seed = hashId(Thread.currentThread().getId()); }
->
+    }
->
->
+    /** ThreadLocal class for Submitters */
->
+    static final class ThreadSubmitter extends ThreadLocal<Submitter> {
->
+        public Submitter initialValue() { return new Submitter(); }
->
+    }
->
->
+    // static fields (initialized in static initializer below)
+    /**
-<
+     * Sum of per-thread steal counts, updated only when threads are
-<
+     * idle or terminating.
->
+     * Creates a new ForkJoinWorkerThread. This factory is used unless
->
+     * overridden in ForkJoinPool constructors.
+     */
-<
+    private final AtomicLong stealCount;
->
+    public static final ForkJoinWorkerThreadFactory
->
+        defaultForkJoinWorkerThreadFactory;
+    /**
-<
+     * Queue for external submissions.
->
+     * Generator for assigning sequence numbers as pool names.
+     */
-<
+    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
->
+    private static final AtomicInteger poolNumberGenerator;
+    /**
-<
+     * Head of Treiber stack for barrier sync. See below for explanation
->
+     * Permission required for callers of methods that may start or
->
+     * kill threads.
+     */
-<
+    private volatile WaitQueueNode syncStack;
->
+    private static final RuntimePermission modifyThreadPermission;
+    /**
-<
+     * The count for event barrier
->
+     * Per-thread submission bookeeping. Shared across all pools
->
+     * to reduce ThreadLocal pollution and because random motion
->
+     * to avoid contention in one pool is likely to hold for others.
+     */
-<
+    private volatile long eventCount;
->
+    private static final ThreadSubmitter submitters;
->
->
+    // static constants
+    /**
-<
+     * Pool number, just for assigning useful names to worker threads
->
+     * The wakeup interval (in nanoseconds) for a worker waiting for a
->
+     * task when the pool is quiescent to instead try to shrink the
->
+     * number of workers.  The exact value does not matter too
->
+     * much. It must be short enough to release resources during
->
+     * sustained periods of idleness, but not so short that threads
->
+     * are continually re-created.
+     */
-<
+    private final int poolNumber;
->
+    private static final long SHRINK_RATE =
->
+L * 1000L * 1000L * 1000L; // 4 seconds
+    /**
-<
+     * The maximum allowed pool size
->
+     * The timeout value for attempted shrinkage, includes
->
+     * some slop to cope with system timer imprecision.
+     */
-<
+    private volatile int maxPoolSize;
->
+    private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
+    /**
-<
+     * The desired parallelism level, updated only under workerLock.
->
+     * The maximum stolen->joining link depth allowed in tryHelpStealer.
->
+     * Depths for legitimate chains are unbounded, but we use a fixed
->
+     * constant to avoid (otherwise unchecked) cycles and to bound
->
+     * staleness of traversal parameters at the expense of sometimes
->
+     * blocking when we could be helping.
+     */
-<
+    private volatile int parallelism;
->
+    private static final int MAX_HELP_DEPTH = 16;
+    /**
-<
+     * Holds number of total (i.e., created and not yet terminated)
-<
+     * and running (i.e., not blocked on joins or other managed sync)
-<
+     * threads, packed into one int to ensure consistent snapshot when
-<
+     * making decisions about creating and suspending spare
-<
+     * threads. Updated only by CAS.  Note: CASes in
-<
+     * updateRunningCount and preJoin running active count is in low
-<
+     * word, so need to be modified if this changes
-<
+     */
-<
+    private volatile int workerCounts;
->
+     * Bits and masks for control variables
->
+     *
->
+     * Field ctl is a long packed with:
->
+     * AC: Number of active running workers minus target parallelism (16 bits)
->
+     * TC: Number of total workers minus target parallelism (16 bits)
->
+     * ST: true if pool is terminating (1 bit)
->
+     * EC: the wait count of top waiting thread (15 bits)
->
+     * ID: poolIndex of top of Treiber stack of waiters (16 bits)
->
+     *
->
+     * When convenient, we can extract the upper 32 bits of counts and
->
+     * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
->
+     * (int)ctl.  The ec field is never accessed alone, but always
->
+     * together with id and st. The offsets of counts by the target
->
+     * parallelism and the positionings of fields makes it possible to
->
+     * perform the most common checks via sign tests of fields: When
->
+     * ac is negative, there are not enough active workers, when tc is
->
+     * negative, there are not enough total workers, and when e is
->
+     * negative, the pool is terminating.  To deal with these possibly
->
+     * negative fields, we use casts in and out of "short" and/or
->
+     * signed shifts to maintain signedness.
->
+     *
->
+     * When a thread is queued (inactivated), its eventCount field is
->
+     * set negative, which is the only way to tell if a worker is
->
+     * prevented from executing tasks, even though it must continue to
->
+     * scan for them to avoid queuing races. Note however that
->
+     * eventCount updates lag releases so usage requires care.
->
+     *
->
+     * Field runState is an int packed with:
->
+     * SHUTDOWN: true if shutdown is enabled (1 bit)
->
+     * SEQ:  a sequence number updated upon (de)registering workers (15 bits)
->
+     * MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
->
+     *
->
+     * The combination of mask and sequence number enables simple
->
+     * consistency checks: Staleness of read-only operations on the
->
+     * workQueues array can be checked by comparing runState before vs
->
+     * after the reads. The low 16 bits (i.e, anding with SMASK) hold
->
+     * the smallest power of two covering all indices, minus
->
+     * one.
->
+     */
->
->
+    // bit positions/shifts for fields
->
+    private static final int  AC_SHIFT   = 48;
->
+    private static final int  TC_SHIFT   = 32;
->
+    private static final int  ST_SHIFT   = 31;
->
+    private static final int  EC_SHIFT   = 16;
->
->
+    // bounds
->
+    private static final int  POOL_MAX   = 0x7fff;  // max #workers - 1
->
+    private static final int  SMASK      = 0xffff;  // short bits
->
+    private static final int  SQMASK     = 0xfffe;  // even short bits
->
+    private static final int  SHORT_SIGN = 1 << 15;
->
+    private static final int  INT_SIGN   = 1 << 31;
->
->
+    // masks
->
+    private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
->
+    private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
->
+    private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
->
->
+    // units for incrementing and decrementing
->
+    private static final long TC_UNIT    = 1L << TC_SHIFT;
->
+    private static final long AC_UNIT    = 1L << AC_SHIFT;
->
->
+    // masks and units for dealing with u = (int)(ctl >>> 32)
->
+    private static final int  UAC_SHIFT  = AC_SHIFT - 32;
->
+    private static final int  UTC_SHIFT  = TC_SHIFT - 32;
->
+    private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
->
+    private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
->
+    private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
->
+    private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
->
->
+    // masks and units for dealing with e = (int)ctl
->
+    private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
->
+    private static final int E_SEQ       = 1 << EC_SHIFT;
->
->
+    // runState bits
->
+    private static final int SHUTDOWN    = 1 << 31;
->
+    private static final int RS_SEQ      = 1 << 16;
->
+    private static final int RS_SEQ_MASK = 0x7fff0000;
->
->
+    // access mode for WorkQueue
->
+    static final int LIFO_QUEUE          =  0;
->
+    static final int FIFO_QUEUE          =  1;
->
+    static final int SHARED_QUEUE        = -1;
->
->
+    // Instance fields
-<
+    private static int totalCountOf(int s)           { return s >>> 16;  }
-<
+    private static int runningCountOf(int s)         { return s & shortMask; }
-<
+    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
->
+    /*
->
+     * Field layout order in this class tends to matter more than one
->
+     * would like. Runtime layout order is only loosely related to
->
+     * declaration order and may differ across JVMs, but the following
->
+     * empirically works OK on current JVMs.
->
+     */
->
->
+    volatile long ctl;                         // main pool control
->
+    final int parallelism;                     // parallelism level
->
+    final int localMode;                       // per-worker scheduling mode
->
+    int growHints;                             // for expanding indices/ranges
->
+    volatile int runState;                     // shutdown status, seq, and mask
->
+    WorkQueue[] workQueues;                    // main registry
->
+    final Mutex lock;                          // for registration
->
+    final Condition termination;               // for awaitTermination
->
+    final ForkJoinWorkerThreadFactory factory; // factory for new workers
->
+    final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
->
+    final AtomicLong stealCount;               // collect counts when terminated
->
+    final AtomicInteger nextWorkerNumber;      // to create worker name string
->
+    final String workerNamePrefix;             // to create worker name string
->
->
+    //  Creating, registering, deregistering and running workers
->
->
+    /**
->
+     * Tries to create and start a worker
->
+     */
->
+    private void addWorker() {
->
+        Throwable ex = null;
->
+        ForkJoinWorkerThread wt = null;
->
+        try {
->
+            if ((wt = factory.newThread(this)) != null) {
->
+                wt.start();
->
+                return;
->
+            }
->
+        } catch (Throwable e) {
->
+            ex = e;
->
+        }
->
+        deregisterWorker(wt, ex); // adjust counts etc on failure
->
+    }
+    /**
-<
+     * Add delta (which may be negative) to running count.  This must
-<
+     * be called before (with negative arg) and after (with positive)
-<
+     * any managed synchronization (i.e., mainly, joins)
-<
+     * @param delta the number to add
->
+     * Callback from ForkJoinWorkerThread constructor to assign a
->
+     * public name. This must be separate from registerWorker because
->
+     * it is called during the "super" constructor call in
->
+     * ForkJoinWorkerThread.
+     */
-<
+    final void updateRunningCount(int delta) {
-<
+        int s;
-<
+        do;while (!casWorkerCounts(s = workerCounts, s + delta));
->
+    final String nextWorkerName() {
->
+        return workerNamePrefix.concat
->
+            (Integer.toString(nextWorkerNumber.addAndGet(1)));
+    /**
-<
+     * Add delta (which may be negative) to both total and running
-<
+     * count.  This must be called upon creation and termination of
-<
+     * worker threads.
-<
+     * @param delta the number to add
->
+     * Callback from ForkJoinWorkerThread constructor to establish and
->
+     * record its WorkQueue.
->
+     *
->
+     * @param wt the worker thread
+     */
-<
+    private void updateWorkerCount(int delta) {
-<
+        int d = delta + (delta << 16); // add to both lo and hi parts
-<
+        int s;
-<
+        do;while (!casWorkerCounts(s = workerCounts, s + d));
->
+    final void registerWorker(ForkJoinWorkerThread wt) {
->
+        WorkQueue w = wt.workQueue;
->
+        Mutex lock = this.lock;
->
+        lock.lock();
->
+        try {
->
+            int g = growHints, k = g & SMASK;
->
+            WorkQueue[] ws = workQueues;
->
+            if (ws != null) {                       // ignore on shutdown
->
+                int n = ws.length;
->
+                if ((k & 1) == 0 || k >= n || ws[k] != null) {
->
+                    for (k = 1; k < n && ws[k] != null; k += 2)
->
+                        ;                           // workers are at odd indices
->
+                    if (k >= n)                     // resize
->
+                        workQueues = ws = Arrays.copyOf(ws, n << 1);
->
+                }
->
+                w.eventCount = w.poolIndex = k;     // establish before recording
->
+                ws[k] = w;
->
+                growHints = (g & ~SMASK) | ((k + 2) & SMASK);
->
+                int rs = runState;
->
+                int m = rs & SMASK;                 // recalculate runState mask
->
+                if (k > m)
->
+                    m = (m << 1) + 1;
->
+                runState = (rs & SHUTDOWN) | ((rs + RS_SEQ) & RS_SEQ_MASK) | m;
->
+            }
->
+        } finally {
->
+            lock.unlock();
->
+        }
+    /**
-<
+     * Lifecycle control. High word contains runState, low word
-<
+     * contains the number of workers that are (probably) executing
-<
+     * tasks. This value is atomically incremented before a worker
-<
+     * gets a task to run, and decremented when worker has no tasks
-<
+     * and cannot find any. These two fields are bundled together to
-<
+     * support correct termination triggering.  Note: activeCount
-<
+     * CAS'es cheat by assuming active count is in low word, so need
-<
+     * to be modified if this changes
->
+     * Final callback from terminating worker, as well as upon failure
->
+     * to construct or start a worker in addWorker.  Removes record of
->
+     * worker from array, and adjusts counts. If pool is shutting
->
+     * down, tries to complete termination.
->
+     *
->
+     * @param wt the worker thread or null if addWorker failed
->
+     * @param ex the exception causing failure, or null if none
+     */
-<
+    private volatile int runControl;
->
+    final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
->
+        WorkQueue w = null;
->
+        if (wt != null && (w = wt.workQueue) != null) {
->
+            w.runState = -1;                // ensure runState is set
->
+            stealCount.getAndAdd(w.totalSteals + w.nsteals);
->
+            int idx = w.poolIndex;
->
+            Mutex lock = this.lock;
->
+            lock.lock();
->
+            try {                           // remove record from array
->
+                WorkQueue[] ws = workQueues;
->
+                if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w) {
->
+                    ws[idx] = null;
->
+                    growHints = (growHints & ~SMASK) | idx;
->
+                }
->
+            } finally {
->
+                lock.unlock();
->
+            }
->
+        }
->
->
+        long c;                             // adjust ctl counts
->
+        do {} while (!U.compareAndSwapLong
->
+                     (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
->
+                                           ((c - TC_UNIT) & TC_MASK) |
->
+                                           (c & ~(AC_MASK|TC_MASK)))));
-<
+    // RunState values. Order among values matters
-<
+    private static final int RUNNING     = 0;
-<
+    private static final int SHUTDOWN    = 1;
-<
+    private static final int TERMINATING = 2;
-<
+    private static final int TERMINATED  = 3;
->
+        if (!tryTerminate(false, false) && w != null) {
->
+            w.cancelAll();                  // cancel remaining tasks
->
+            if (w.array != null)            // suppress signal if never ran
->
+                signalWork();               // wake up or create replacement
->
+            if (ex == null)                 // help clean refs on way out
->
+                ForkJoinTask.helpExpungeStaleExceptions();
->
+        }
-<
+    private static int runStateOf(int c)             { return c >>> 16; }
-<
+    private static int activeCountOf(int c)          { return c & shortMask; }
-<
+    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
->
+        if (ex != null)                     // rethrow
->
+            U.throwException(ex);
->
+    }
+    /**
-<
+     * Try incrementing active count; fail on contention. Called by
-<
+     * workers before/during executing tasks.
-<
+     * @return true on success;
->
+     * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
+     */
-<
+    final boolean tryIncrementActiveCount() {
-<
+        int c = runControl;
-<
+        return casRunControl(c, c+1);
->
+    final void runWorker(ForkJoinWorkerThread wt) {
->
+        // Initialize queue array and seed in this thread
->
+        WorkQueue w = wt.workQueue;
->
+        w.growArray(false);
->
+        w.seed = hashId(Thread.currentThread().getId());
->
->
+        do {} while (w.runTask(scan(w)));
-+
+    // Submissions
-+
+    /**
-<
+     * Try decrementing active count; fail on contention.
-<
+     * Possibly trigger termination on success
-<
+     * Called by workers when they can't find tasks.
-<
+     * @return true on success
->
+     * Unless shutting down, adds the given task to a submission queue
->
+     * at submitter's current queue index (modulo submission
->
+     * range). If no queue exists at the index, one is created unless
->
+     * pool lock is busy.  If the queue and/or lock are busy, another
->
+     * index is randomly chosen. The mask in growHints controls the
->
+     * effective index range of queues considered. The mask is
->
+     * expanded, up to the current workerQueue mask, upon any detected
->
+     * contention but otherwise remains small to avoid needlessly
->
+     * creating queues when there is no contention.
+     */
-<
+    final boolean tryDecrementActiveCount() {
-<
+        int c = runControl;
-<
+        int nextc = c - 1;
-<
+        if (!casRunControl(c, nextc))
-<
+            return false;
-<
+        if (canTerminateOnShutdown(nextc))
-<
+            terminateOnShutdown();
-<
+        return true;
->
+    private void doSubmit(ForkJoinTask<?> task) {
->
+        if (task == null)
->
+            throw new NullPointerException();
->
+        Submitter s = submitters.get();
->
+        for (int r = s.seed, m = growHints >>> 16;;) {
->
+            WorkQueue[] ws; WorkQueue q; Mutex lk;
->
+            int k = r & m & SQMASK;          // use only even indices
->
+            if (runState < 0 || (ws = workQueues) == null || ws.length <= k)
->
+                throw new RejectedExecutionException(); // shutting down
->
+            if ((q = ws[k]) == null && (lk = lock).tryAcquire(0)) {
->
+                try {                        // try to create new queue
->
+                    if (ws == workQueues && (q = ws[k]) == null) {
->
+                        int rs;              // update runState seq
->
+                        ws[k] = q = new WorkQueue(null, SHARED_QUEUE);
->
+                        runState = (((rs = runState) & SHUTDOWN) |
->
+                                    ((rs + RS_SEQ) & ~SHUTDOWN));
->
+                    }
->
+                } finally {
->
+                    lk.unlock();
->
+                }
->
+            }
->
+            if (q != null) {
->
+                if (q.trySharedPush(task)) {
->
+                    signalWork();
->
+                    return;
->
+                }
->
+                else if (m < parallelism - 1 && m < (runState & SMASK)) {
->
+                    Mutex lock = this.lock;
->
+                    lock.lock();             // block until lock free
->
+                    int g = growHints;
->
+                    if (g >>> 16 == m)       // expand range
->
+                        growHints = (((m << 1) + 1) << 16) | (g & SMASK);
->
+                    lock.unlock();           // no need for try/finally
->
+                }
->
+                else if ((r & m) == 0)
->
+                    Thread.yield();          // occasionally yield if busy
->
+            }
->
+            if (m == (m = growHints >>> 16)) {
->
+                r ^= r << 13;                // update seed unless new range
->
+                r ^= r >>> 17;               // same xorshift as WorkQueues
->
+                s.seed = r ^= r << 5;
->
+            }
->
+        }
-+
+    // Maintaining ctl counts
-+
+    /**
-<
+     * Return true if argument represents zero active count and
-<
+     * nonzero runstate, which is the triggering condition for
-<
+     * terminating on shutdown.
->
+     * Increments active count; mainly called upon return from blocking.
+     */
-<
+    private static boolean canTerminateOnShutdown(int c) {
-<
+        return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
->
+    final void incrementActiveCount() {
->
+        long c;
->
+        do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
+    /**
-<
+     * Transition run state to at least the given state. Return true
-<
+     * if not already at least given state.
->
+     * Tries to activate or create a worker if too few are active.
+     */
-<
+    private boolean transitionRunStateTo(int state) {
-<
+        for (;;) {
-<
+            int c = runControl;
-<
+            if (runStateOf(c) >= state)
-<
+                return false;
-<
+            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
-<
+                return true;
->
+    final void signalWork() {
->
+        long c; int u;
->
+        while ((u = (int)((c = ctl) >>> 32)) < 0) {     // too few active
->
+            WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
->
+            if ((e = (int)c) > 0) {                     // at least one waiting
->
+                if (ws != null && (i = e & SMASK) < ws.length &&
->
+                    (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
->
+                    long nc = (((long)(w.nextWait & E_MASK)) |
->
+                               ((long)(u + UAC_UNIT) << 32));
->
+                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                        w.eventCount = (e + E_SEQ) & E_MASK;
->
+                        if ((p = w.parker) != null)
->
+                            U.unpark(p);                // activate and release
->
+                        break;
->
+                    }
->
+                }
->
+                else
->
+                    break;
->
+            }
->
+            else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
->
+                long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
->
+                                 ((u + UAC_UNIT) & UAC_MASK)) << 32;
->
+                if (U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                    addWorker();
->
+                    break;
->
+                }
->
+            }
->
+            else
->
+                break;
+    /**
-<
+     * Controls whether to add spares to maintain parallelism
->
+     * Tries to decrement active count (sometimes implicitly) and
->
+     * possibly release or create a compensating worker in preparation
->
+     * for blocking. Fails on contention or termination.
->
+     *
->
+     * @return true if the caller can block, else should recheck and retry
+     */
-<
+    private volatile boolean maintainsParallelism;
->
+    final boolean tryCompensate() {
->
+        WorkQueue w; Thread p;
->
+        int pc = parallelism, e, u, ac, tc, i;
->
+        long c = ctl;
->
+        WorkQueue[] ws = workQueues;
->
+        if ((e = (int)c) >= 0) {
->
+            if ((ac = ((u = (int)(c >>> 32)) >> UAC_SHIFT)) <= 0 &&
->
+                e != 0 && ws != null && (i = e & SMASK) < ws.length &&
->
+                (w = ws[i]) != null) {
->
+                long nc = (long)(w.nextWait & E_MASK) | (c & (AC_MASK|TC_MASK));
->
+                if (w.eventCount == (e | INT_SIGN) &&
->
+                    U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                    w.eventCount = (e + E_SEQ) & E_MASK;
->
+                    if ((p = w.parker) != null)
->
+                        U.unpark(p);
->
+                    return true;             // release an idle worker
->
+                }
->
+            }
->
+            else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
->
+                long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
->
+                if (U.compareAndSwapLong(this, CTL, c, nc))
->
+                    return true;             // no compensation needed
->
+            }
->
+            else if (tc + pc < POOL_MAX) {
->
+                long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
->
+                if (U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                    addWorker();
->
+                    return true;             // create replacement
->
+                }
->
+            }
->
+        }
->
+        return false;
->
+    }
-<
+    // Constructors
->
+    // Scanning for tasks
+    /**
-<
+     * Creates a ForkJoinPool with a pool size equal to the number of
-<
+     * processors available on the system and using the default
-<
+     * ForkJoinWorkerThreadFactory,
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     * Scans for and, if found, returns one task, else possibly
->
+     * inactivates the worker. This method operates on single reads of
->
+     * volatile state and is designed to be re-invoked continuously in
->
+     * part because it returns upon detecting inconsistencies,
->
+     * contention, or state changes that indicate possible success on
->
+     * re-invocation.
->
+     *
->
+     * The scan searches for tasks across queues, randomly selecting
->
+     * the first #queues probes, favoring steals over submissions
->
+     * (by exploiting even/odd indexing), and then performing a
->
+     * circular sweep of all queues.  The scan terminates upon either
->
+     * finding a non-empty queue, or completing a full sweep. If the
->
+     * worker is not inactivated, it takes and returns a task from
->
+     * this queue.  On failure to find a task, we take one of the
->
+     * following actions, after which the caller will retry calling
->
+     * this method unless terminated.
->
+     *
->
+     * * If pool is terminating, terminate the worker.
->
+     *
->
+     * * If not a complete sweep, try to release a waiting worker.  If
->
+     * the scan terminated because the worker is inactivated, then the
->
+     * released worker will often be the calling worker, and it can
->
+     * succeed obtaining a task on the next call. Or maybe it is
->
+     * another worker, but with same net effect. Releasing in other
->
+     * cases as well ensures that we have enough workers running.
->
+     *
->
+     * * If the caller has run a task since the last empty scan,
->
+     * return (to allow rescan) if other workers are not also yet
->
+     * enqueued.  Field WorkQueue.rescans counts down on each scan to
->
+     * ensure eventual inactivation and blocking.
->
+     *
->
+     * * If not already enqueued, try to inactivate and enqueue the
->
+     * worker on wait queue.
->
+     *
->
+     * * If already enqueued and none of the above apply, either park
->
+     * awaiting signal, or if this is the most recent waiter and pool
->
+     * is quiescent, relay to idleAwaitWork to check for termination
->
+     * and possibly shrink pool.
->
+     *
->
+     * @param w the worker (via its WorkQueue)
->
+     * @return a task or null of none found
+     */
-<
+    public ForkJoinPool() {
-<
+        this(Runtime.getRuntime().availableProcessors(),
-<
+             defaultForkJoinWorkerThreadFactory);
->
+    private final ForkJoinTask<?> scan(WorkQueue w) {
->
+        boolean swept = false;               // true after full empty scan
->
+        WorkQueue[] ws;                      // volatile read order matters
->
+        int r = w.seed, ec = w.eventCount;   // ec is negative if inactive
->
+        int rs = runState, m = rs & SMASK;
->
+        if ((ws = workQueues) != null && ws.length > m) { // consistency check
->
+            for (int k = 0, j = -1 - m; ; ++j) {
->
+                WorkQueue q; int b;
->
+                if (j < 0) {                 // random probes while j negative
->
+                    r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) | (j & 1);
->
+                }                            // worker (not submit) for odd j
->
+                else                         // cyclic scan when j >= 0
->
+                    k += 7;                  // step 7 reduces array packing bias
->
+                if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
->
+                    ForkJoinTask<?> t = (ec >= 0) ? q.pollAt(b) : null;
->
+                    w.seed = r;              // save seed for next scan
->
+                    if (t != null)
->
+                        return t;
->
+                    break;
->
+                }
->
+                else if (j - m > m) {
->
+                    if (rs == runState)      // staleness check
->
+                        swept = true;
->
+                    break;
->
+                }
->
+            }
->
->
+            // Decode ctl on empty scan
->
+            long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
->
+            if (e < 0)                       // pool is terminating
->
+                w.runState = -1;
->
+            else if (!swept) {               // try to release a waiter
->
+                WorkQueue v; Thread p;
->
+                if (e > 0 && a < 0 && (v = ws[e & m]) != null &&
->
+                    v.eventCount == (e | INT_SIGN)) {
->
+                    long nc = ((long)(v.nextWait & E_MASK) |
->
+                               ((c + AC_UNIT) & (AC_MASK|TC_MASK)));
->
+                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                        v.eventCount = (e + E_SEQ) & E_MASK;
->
+                        if ((p = v.parker) != null)
->
+                            U.unpark(p);
->
+                    }
->
+                }
->
+            }
->
+            else if ((nr = w.rescans) > 0) { // continue rescanning
->
+                int ac = a + parallelism;
->
+                if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0 &&
->
+                    w.eventCount == ec)
->
+                    Thread.yield();          // occasionally yield
->
+            }
->
+            else if (ec >= 0) {              // try to enqueue
->
+                long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
->
+                w.nextWait = e;
->
+                w.eventCount = ec | INT_SIGN;// mark as inactive
->
+                if (!U.compareAndSwapLong(this, CTL, c, nc))
->
+                    w.eventCount = ec;       // unmark on CAS failure
->
+                else if ((ns = w.nsteals) != 0) {
->
+                    w.nsteals = 0;           // set rescans if ran task
->
+                    w.rescans = a + parallelism;
->
+                    w.totalSteals += ns;
->
+                }
->
+            }
->
+            else{                            // already queued
->
+                if (parallelism == -a)
->
+                    idleAwaitWork(w);        // quiescent
->
+                if (w.eventCount == ec) {
->
+                    Thread.interrupted();    // clear status
->
+                    ForkJoinWorkerThread wt = w.owner;
->
+                    U.putObject(wt, PARKBLOCKER, this);
->
+                    w.parker = wt;           // emulate LockSupport.park
->
+                    if (w.eventCount == ec)  // recheck
->
+                        U.park(false, 0L);   // block
->
+                    w.parker = null;
->
+                    U.putObject(wt, PARKBLOCKER, null);
->
+                }
->
+            }
->
+        }
->
+        return null;
+    /**
-<
+     * Creates a ForkJoinPool with the indicated parellelism level
-<
+     * threads, and using the default ForkJoinWorkerThreadFactory,
-<
+     * @param parallelism the number of worker threads
-<
+     * @throws IllegalArgumentException if parallelism less than or
-<
+     * equal to zero
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     * If inactivating worker w has caused pool to become quiescent,
->
+     * checks for pool termination, and, so long as this is not the
->
+     * only worker, waits for event for up to SHRINK_RATE nanosecs.
->
+     * On timeout, if ctl has not changed, terminates the worker,
->
+     * which will in turn wake up another worker to possibly repeat
->
+     * this process.
->
+     *
->
+     * @param w the calling worker
+     */
-<
+    public ForkJoinPool(int parallelism) {
-<
+        this(parallelism, defaultForkJoinWorkerThreadFactory);
->
+    private void idleAwaitWork(WorkQueue w) {
->
+        long c; int nw, ec;
->
+        if (!tryTerminate(false, false) &&
->
+            (int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
->
+            (ec = w.eventCount) == ((int)c | INT_SIGN) &&
->
+            (nw = w.nextWait) != 0) {
->
+            long nc = ((long)(nw & E_MASK) | // ctl to restore on timeout
->
+                       ((c + AC_UNIT) & AC_MASK) | (c & TC_MASK));
->
+            ForkJoinWorkerThread wt = w.owner;
->
+            while (ctl == c) {
->
+                long startTime = System.nanoTime();
->
+                Thread.interrupted();  // timed variant of version in scan()
->
+                U.putObject(wt, PARKBLOCKER, this);
->
+                w.parker = wt;
->
+                if (ctl == c)
->
+                    U.park(false, SHRINK_RATE);
->
+                w.parker = null;
->
+                U.putObject(wt, PARKBLOCKER, null);
->
+                if (ctl != c)
->
+                    break;
->
+                if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
->
+                    U.compareAndSwapLong(this, CTL, c, nc)) {
->
+                    w.eventCount = (ec + E_SEQ) | E_MASK;
->
+                    w.runState = -1;          // shrink
->
+                    break;
->
+                }
->
+            }
->
+        }
+    /**
-<
+     * Creates a ForkJoinPool with parallelism equal to the number of
-<
+     * processors available on the system and using the given
-<
+     * ForkJoinWorkerThreadFactory,
-<
+     * @param factory the factory for creating new threads
-<
+     * @throws NullPointerException if factory is null
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
-<
+     */
-<
+    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
-<
+        this(Runtime.getRuntime().availableProcessors(), factory);
->
+     * Tries to locate and execute tasks for a stealer of the given
->
+     * task, or in turn one of its stealers, Traces currentSteal ->
->
+     * currentJoin links looking for a thread working on a descendant
->
+     * of the given task and with a non-empty queue to steal back and
->
+     * execute tasks from. The first call to this method upon a
->
+     * waiting join will often entail scanning/search, (which is OK
->
+     * because the joiner has nothing better to do), but this method
->
+     * leaves hints in workers to speed up subsequent calls. The
->
+     * implementation is very branchy to cope with potential
->
+     * inconsistencies or loops encountering chains that are stale,
->
+     * unknown, or of length greater than MAX_HELP_DEPTH links.  All
->
+     * of these cases are dealt with by just retrying by caller.
->
+     *
->
+     * @param joiner the joining worker
->
+     * @param task the task to join
->
+     * @return true if found or ran a task (and so is immediately retryable)
->
+     */
->
+    final boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
->
+        ForkJoinTask<?> subtask;    // current target
->
+        boolean progress = false;
->
+        int depth = 0;              // current chain depth
->
+        int m = runState & SMASK;
->
+        WorkQueue[] ws = workQueues;
->
->
+        if (ws != null && ws.length > m && (subtask = task).status >= 0) {
->
+            outer:for (WorkQueue j = joiner;;) {
->
+                // Try to find the stealer of subtask, by first using hint
->
+                WorkQueue stealer = null;
->
+                WorkQueue v = ws[j.stealHint & m];
->
+                if (v != null && v.currentSteal == subtask)
->
+                    stealer = v;
->
+                else {
->
+                    for (int i = 1; i <= m; i += 2) {
->
+                        if ((v = ws[i]) != null && v.currentSteal == subtask) {
->
+                            stealer = v;
->
+                            j.stealHint = i; // save hint
->
+                            break;
->
+                        }
->
+                    }
->
+                    if (stealer == null)
->
+                        break;
->
+                }
->
->
+                for (WorkQueue q = stealer;;) { // Try to help stealer
->
+                    ForkJoinTask<?> t; int b;
->
+                    if (task.status < 0)
->
+                        break outer;
->
+                    if ((b = q.base) - q.top < 0) {
->
+                        progress = true;
->
+                        if (subtask.status < 0)
->
+                            break outer;               // stale
->
+                        if ((t = q.pollAt(b)) != null) {
->
+                            stealer.stealHint = joiner.poolIndex;
->
+                            joiner.runSubtask(t);
->
+                        }
->
+                    }
->
+                    else { // empty - try to descend to find stealer's stealer
->
+                        ForkJoinTask<?> next = stealer.currentJoin;
->
+                        if (++depth == MAX_HELP_DEPTH || subtask.status < 0 ||
->
+                            next == null || next == subtask)
->
+                            break outer;  // max depth, stale, dead-end, cyclic
->
+                        subtask = next;
->
+                        j = stealer;
->
+                        break;
->
+                    }
->
+                }
->
+            }
->
+        }
->
+        return progress;
+    /**
-<
+     * Creates a ForkJoinPool with the given parallelism and factory.
->
+     * If task is at base of some steal queue, steals and executes it.
-<
+     * @param parallelism the targeted number of worker threads
-<
+     * @param factory the factory for creating new threads
-<
+     * @throws IllegalArgumentException if parallelism less than or
-<
+     * equal to zero, or greater than implementation limit.
-<
+     * @throws NullPointerException if factory is null
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     * @param joiner the joining worker
->
+     * @param task the task
+     */
-<
+    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
-<
+        if (parallelism <= 0 || parallelism > MAX_THREADS)
-<
+            throw new IllegalArgumentException();
-<
+        if (factory == null)
-<
+            throw new NullPointerException();
-<
+        checkPermission();
-<
+        this.factory = factory;
-<
+        this.parallelism = parallelism;
-<
+        this.maxPoolSize = MAX_THREADS;
-<
+        this.maintainsParallelism = true;
-<
+        this.poolNumber = poolNumberGenerator.incrementAndGet();
-<
+        this.workerLock = new ReentrantLock();
-<
+        this.termination = workerLock.newCondition();
-<
+        this.stealCount = new AtomicLong();
-<
+        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
-<
+        createAndStartInitialWorkers(parallelism);
->
+    final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
->
+        WorkQueue[] ws;
->
+        int m = runState & SMASK;
->
+        if ((ws = workQueues) != null && ws.length > m) {
->
+            for (int j = 1; j <= m && task.status >= 0; j += 2) {
->
+                WorkQueue q = ws[j];
->
+                if (q != null && q.pollFor(task)) {
->
+                    joiner.runSubtask(task);
->
+                    break;
->
+                }
->
+            }
->
+        }
+    /**
-<
+     * Create new worker using factory.
-<
+     * @param index the index to assign worker
-<
+     * @return new worker, or null of factory failed
-<
+     */
-<
+    private ForkJoinWorkerThread createWorker(int index) {
-<
+        Thread.UncaughtExceptionHandler h = ueh;
-<
+        ForkJoinWorkerThread w = factory.newThread(this);
-<
+        if (w != null) {
-<
+            w.poolIndex = index;
-<
+            w.setDaemon(true);
-<
+            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
-<
+            if (h != null)
-<
+                w.setUncaughtExceptionHandler(h);
->
+     * Returns a non-empty steal queue, if one is found during a random,
->
+     * then cyclic scan, else null.  This method must be retried by
->
+     * caller if, by the time it tries to use the queue, it is empty.
->
+     */
->
+    private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
->
+        int r = w.seed;    // Same idea as scan(), but ignoring submissions
->
+        for (WorkQueue[] ws;;) {
->
+            int m = runState & SMASK;
->
+            if ((ws = workQueues) == null)
->
+                return null;
->
+            if (ws.length > m) {
->
+                WorkQueue q;
->
+                for (int k = 0, j = -1 - m;; ++j) {
->
+                    if (j < 0) {
->
+                        r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5;
->
+                    }
->
+                    else
->
+                        k += 7;
->
+                    if ((q = ws[(k | 1) & m]) != null && q.base - q.top < 0) {
->
+                        w.seed = r;
->
+                        return q;
->
+                    }
->
+                    else if (j - m > m)
->
+                        return null;
->
+                }
->
+            }
-–
+        return w;
+    /**
-<
+     * Return a good size for worker array given pool size.
-<
+     * Currently requires size to be a power of two.
-<
+     */
-<
+    private static int arraySizeFor(int ps) {
-<
+        return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
->
+     * Runs tasks until {@code isQuiescent()}. We piggyback on
->
+     * active count ctl maintenance, but rather than blocking
->
+     * when tasks cannot be found, we rescan until all others cannot
->
+     * find tasks either.
->
+     */
->
+    final void helpQuiescePool(WorkQueue w) {
->
+        for (boolean active = true;;) {
->
+            w.runLocalTasks();      // exhaust local queue
->
+            WorkQueue q = findNonEmptyStealQueue(w);
->
+            if (q != null) {
->
+                ForkJoinTask<?> t;
->
+                if (!active) {      // re-establish active count
->
+                    long c;
->
+                    active = true;
->
+                    do {} while (!U.compareAndSwapLong
->
+                                 (this, CTL, c = ctl, c + AC_UNIT));
->
+                }
->
+                if ((t = q.poll()) != null)
->
+                    w.runSubtask(t);
->
+            }
->
+            else {
->
+                long c;
->
+                if (active) {       // decrement active count without queuing
->
+                    active = false;
->
+                    do {} while (!U.compareAndSwapLong
->
+                                 (this, CTL, c = ctl, c -= AC_UNIT));
->
+                }
->
+                else
->
+                    c = ctl;        // re-increment on exit
->
+                if ((int)(c >> AC_SHIFT) + parallelism == 0) {
->
+                    do {} while (!U.compareAndSwapLong
->
+                                 (this, CTL, c = ctl, c + AC_UNIT));
->
+                    break;
->
+                }
->
+            }
->
+        }
+    /**
-<
+     * Create or resize array if necessary to hold newLength
-<
+     * @return the array
->
+     * Gets and removes a local or stolen task for the given worker.
->
+     *
->
+     * @return a task, if available
+     */
-<
+    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
-<
+        ForkJoinWorkerThread[] ws = workers;
-<
+        if (ws == null)
-<
+            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
-<
+        else if (newLength > ws.length)
-<
+            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
-<
+        else
-<
+            return ws;
->
+    final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
->
+        for (ForkJoinTask<?> t;;) {
->
+            WorkQueue q;
->
+            if ((t = w.nextLocalTask()) != null)
->
+                return t;
->
+            if ((q = findNonEmptyStealQueue(w)) == null)
->
+                return null;
->
+            if ((t = q.poll()) != null)
->
+                return t;
->
+        }
+    /**
-<
+     * Try to shrink workers into smaller array after one or more terminate
->
+     * Returns the approximate (non-atomic) number of idle threads per
->
+     * active thread to offset steal queue size for method
->
+     * ForkJoinTask.getSurplusQueuedTaskCount().
+     */
-<
+    private void tryShrinkWorkerArray() {
-<
+        ForkJoinWorkerThread[] ws = workers;
-<
+        int len = ws.length;
-<
+        int last = len - 1;
-<
+        while (last >= 0 && ws[last] == null)
-<
+            --last;
-<
+        int newLength = arraySizeFor(last+1);
-<
+        if (newLength < len)
-<
+            workers = Arrays.copyOf(ws, newLength);
->
+    final int idlePerActive() {
->
+        // Approximate at powers of two for small values, saturate past 4
->
+        int p = parallelism;
->
+        int a = p + (int)(ctl >> AC_SHIFT);
->
+        return (a > (p >>>= 1) ? 0 :
->
+                a > (p >>>= 1) ? 1 :
->
+                a > (p >>>= 1) ? 2 :
->
+                a > (p >>>= 1) ? 4 :
->
+);
-+
+    //  Termination
-+
+    /**
-<
+     * Initial worker array and worker creation and startup. (This
-<
+     * must be done under lock to avoid interference by some of the
-<
+     * newly started threads while creating others.)
-<
+     */
-<
+    private void createAndStartInitialWorkers(int ps) {
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
-<
+            for (int i = 0; i < ps; ++i) {
-<
+                ForkJoinWorkerThread w = createWorker(i);
-<
+                if (w != null) {
-<
+                    ws[i] = w;
-<
+                    w.start();
-<
+                    updateWorkerCount(1);
->
+     * Possibly initiates and/or completes termination.  The caller
->
+     * triggering termination runs three passes through workQueues:
->
+     * (0) Setting termination status, followed by wakeups of queued
->
+     * workers; (1) cancelling all tasks; (2) interrupting lagging
->
+     * threads (likely in external tasks, but possibly also blocked in
->
+     * joins).  Each pass repeats previous steps because of potential
->
+     * lagging thread creation.
->
+     *
->
+     * @param now if true, unconditionally terminate, else only
->
+     * if no work and no active workers
->
+     * @param enable if true, enable shutdown when next possible
->
+     * @return true if now terminating or terminated
->
+     */
->
+    private boolean tryTerminate(boolean now, boolean enable) {
->
+        Mutex lock = this.lock;
->
+        for (long c;;) {
->
+            if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
->
+                if ((short)(c >>> TC_SHIFT) == -parallelism) {
->
+                    lock.lock();                    // don't need try/finally
->
+                    termination.signalAll();        // signal when 0 workers
->
+                    lock.unlock();
->
+                }
->
+                return true;
->
+            }
->
+            if (runState >= 0) {                    // not yet enabled
->
+                if (!enable)
->
+                    return false;
->
+                lock.lock();
->
+                runState |= SHUTDOWN;
->
+                lock.unlock();
->
+            }
->
+            if (!now) {                             // check if idle & no tasks
->
+                if ((int)(c >> AC_SHIFT) != -parallelism ||
->
+                    hasQueuedSubmissions())
->
+                    return false;
->
+                // Check for unqueued inactive workers. One pass suffices.
->
+                WorkQueue[] ws = workQueues; WorkQueue w;
->
+                if (ws != null) {
->
+                    for (int i = 1; i < ws.length; i += 2) {
->
+                        if ((w = ws[i]) != null && w.eventCount >= 0)
->
+                            return false;
->
+                    }
->
+                }
->
+            }
->
+            if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
->
+                for (int pass = 0; pass < 3; ++pass) {
->
+                    WorkQueue[] ws = workQueues;
->
+                    if (ws != null) {
->
+                        WorkQueue w;
->
+                        int n = ws.length;
->
+                        for (int i = 0; i < n; ++i) {
->
+                            if ((w = ws[i]) != null) {
->
+                                w.runState = -1;
->
+                                if (pass > 0) {
->
+                                    w.cancelAll();
->
+                                    if (pass > 1)
->
+                                        w.interruptOwner();
->
+                                }
->
+                            }
->
+                        }
->
+                        // Wake up workers parked on event queue
->
+                        int i, e; long cc; Thread p;
->
+                        while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
->
+                               (i = e & SMASK) < n &&
->
+                               (w = ws[i]) != null) {
->
+                            long nc = ((long)(w.nextWait & E_MASK) |
->
+                                       ((cc + AC_UNIT) & AC_MASK) |
->
+                                       (cc & (TC_MASK|STOP_BIT)));
->
+                            if (w.eventCount == (e | INT_SIGN) &&
->
+                                U.compareAndSwapLong(this, CTL, cc, nc)) {
->
+                                w.eventCount = (e + E_SEQ) & E_MASK;
->
+                                w.runState = -1;
->
+                                if ((p = w.parker) != null)
->
+                                    U.unpark(p);
->
+                            }
->
+                        }
->
+                    }
-–
+        } finally {
-–
+            lock.unlock();
-+
+    // Exported methods
-+
-+
+    // Constructors
-+
+    /**
-<
+     * Worker creation and startup for threads added via setParallelism.
->
+     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
->
+     * java.lang.Runtime#availableProcessors}, using the {@linkplain
->
+     * #defaultForkJoinWorkerThreadFactory default thread factory},
->
+     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
->
+     *
->
+     * @throws SecurityException if a security manager exists and
->
+     *         the caller is not permitted to modify threads
->
+     *         because it does not hold {@link
->
+     *         java.lang.RuntimePermission}{@code ("modifyThread")}
+     */
-<
+    private void createAndStartAddedWorkers() {
-<
+        resumeAllSpares();  // Allow spares to convert to nonspare
-<
+        int ps = parallelism;
-<
+        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
-<
+        int len = ws.length;
-<
+        // Sweep through slots, to keep lowest indices most populated
-<
+        int k = 0;
-<
+        while (k < len) {
-<
+            if (ws[k] != null) {
-<
+                ++k;
-<
+                continue;
-<
+            }
-<
+            int s = workerCounts;
-<
+            int tc = totalCountOf(s);
-<
+            int rc = runningCountOf(s);
-<
+            if (rc >= ps || tc >= ps)
-<
+                break;
-<
+            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
-<
+                ForkJoinWorkerThread w = createWorker(k);
-<
+                if (w != null) {
-<
+                    ws[k++] = w;
-<
+                    w.start();
-<
+                }
-<
+                else {
-<
+                    updateWorkerCount(-1); // back out on failed creation
-<
+                    break;
-<
+                }
-<
+            }
-<
+        }
->
+    public ForkJoinPool() {
->
+        this(Runtime.getRuntime().availableProcessors(),
->
+             defaultForkJoinWorkerThreadFactory, null, false);
-<
+    // Execution methods
->
+    /**
->
+     * Creates a {@code ForkJoinPool} with the indicated parallelism
->
+     * level, the {@linkplain
->
+     * #defaultForkJoinWorkerThreadFactory default thread factory},
->
+     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
->
+     *
->
+     * @param parallelism the parallelism level
->
+     * @throws IllegalArgumentException if parallelism less than or
->
+     *         equal to zero, or greater than implementation limit
->
+     * @throws SecurityException if a security manager exists and
->
+     *         the caller is not permitted to modify threads
->
+     *         because it does not hold {@link
->
+     *         java.lang.RuntimePermission}{@code ("modifyThread")}
->
+     */
->
+    public ForkJoinPool(int parallelism) {
->
+        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
->
+    }
+    /**
-<
+     * Common code for execute, invoke and submit
->
+     * Creates a {@code ForkJoinPool} with the given parameters.
->
+     *
->
+     * @param parallelism the parallelism level. For default value,
->
+     * use {@link java.lang.Runtime#availableProcessors}.
->
+     * @param factory the factory for creating new threads. For default value,
->
+     * use {@link #defaultForkJoinWorkerThreadFactory}.
->
+     * @param handler the handler for internal worker threads that
->
+     * terminate due to unrecoverable errors encountered while executing
->
+     * tasks. For default value, use {@code null}.
->
+     * @param asyncMode if true,
->
+     * establishes local first-in-first-out scheduling mode for forked
->
+     * tasks that are never joined. This mode may be more appropriate
->
+     * than default locally stack-based mode in applications in which
->
+     * worker threads only process event-style asynchronous tasks.
->
+     * For default value, use {@code false}.
->
+     * @throws IllegalArgumentException if parallelism less than or
->
+     *         equal to zero, or greater than implementation limit
->
+     * @throws NullPointerException if the factory is null
->
+     * @throws SecurityException if a security manager exists and
->
+     *         the caller is not permitted to modify threads
->
+     *         because it does not hold {@link
->
+     *         java.lang.RuntimePermission}{@code ("modifyThread")}
+     */
-<
+    private <T> void doSubmit(ForkJoinTask<T> task) {
-<
+        if (isShutdown())
-<
+            throw new RejectedExecutionException();
-<
+        submissionQueue.offer(task);
-<
+        signalIdleWorkers();
->
+    public ForkJoinPool(int parallelism,
->
+                        ForkJoinWorkerThreadFactory factory,
->
+                        Thread.UncaughtExceptionHandler handler,
->
+                        boolean asyncMode) {
->
+        checkPermission();
->
+        if (factory == null)
->
+            throw new NullPointerException();
->
+        if (parallelism <= 0 || parallelism > POOL_MAX)
->
+            throw new IllegalArgumentException();
->
+        this.parallelism = parallelism;
->
+        this.factory = factory;
->
+        this.ueh = handler;
->
+        this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
->
+        this.growHints = 1;
->
+        long np = (long)(-parallelism); // offset ctl counts
->
+        this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
->
+        // initialize workQueues array with room for 2*parallelism if possible
->
+        int n = parallelism << 1;
->
+        if (n >= POOL_MAX)
->
+            n = POOL_MAX;
->
+        else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
->
+            n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8;
->
+        }
->
+        this.workQueues = new WorkQueue[(n + 1) << 1]; // #slots = 2 * #workers
->
+        this.termination = (this.lock = new Mutex()).newCondition();
->
+        this.stealCount = new AtomicLong();
->
+        this.nextWorkerNumber = new AtomicInteger();
->
+        StringBuilder sb = new StringBuilder("ForkJoinPool-");
->
+        sb.append(poolNumberGenerator.incrementAndGet());
->
+        sb.append("-worker-");
->
+        this.workerNamePrefix = sb.toString();
-+
+    // Execution methods
-+
+    /**
-<
+     * Performs the given task; returning its result upon completion
->
+     * Performs the given task, returning its result upon completion.
->
+     * If the computation encounters an unchecked Exception or Error,
->
+     * it is rethrown as the outcome of this invocation.  Rethrown
->
+     * exceptions behave in the same way as regular exceptions, but,
->
+     * when possible, contain stack traces (as displayed for example
->
+     * using {@code ex.printStackTrace()}) of both the current thread
->
+     * as well as the thread actually encountering the exception;
->
+     * minimally only the latter.
->
+     *
+     * @param task the task
+     * @return the task's result
-<
+     * @throws NullPointerException if task is null
-<
+     * @throws RejectedExecutionException if pool is shut down
->
+     * @throws NullPointerException if the task is null
->
+     * @throws RejectedExecutionException if the task cannot be
->
+     *         scheduled for execution
+     */
+    public <T> T invoke(ForkJoinTask<T> task) {
+        doSubmit(task);
+    /**
+     * Arranges for (asynchronous) execution of the given task.
-+
+     *
+     * @param task the task
-<
+     * @throws NullPointerException if task is null
-<
+     * @throws RejectedExecutionException if pool is shut down
->
+     * @throws NullPointerException if the task is null
->
+     * @throws RejectedExecutionException if the task cannot be
->
+     *         scheduled for execution
+     */
-<
+    public <T> void execute(ForkJoinTask<T> task) {
->
+    public void execute(ForkJoinTask<?> task) {
+        doSubmit(task);
+    // AbstractExecutorService methods
-+
+    /**
-+
+     * @throws NullPointerException if the task is null
-+
+     * @throws RejectedExecutionException if the task cannot be
-+
+     *         scheduled for execution
-+
+     */
+    public void execute(Runnable task) {
-<
+        doSubmit(new AdaptedRunnable<Void>(task, null));
->
+        if (task == null)
->
+            throw new NullPointerException();
->
+        ForkJoinTask<?> job;
->
+        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
->
+            job = (ForkJoinTask<?>) task;
->
+        else
->
+            job = ForkJoinTask.adapt(task, null);
->
+        doSubmit(job);
->
+    }
->
->
+    /**
->
+     * Submits a ForkJoinTask for execution.
->
+     *
->
+     * @param task the task to submit
->
+     * @return the task
->
+     * @throws NullPointerException if the task is null
->
+     * @throws RejectedExecutionException if the task cannot be
->
+     *         scheduled for execution
->
+     */
->
+    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
->
+        doSubmit(task);
->
+        return task;
-+
+    /**
-+
+     * @throws NullPointerException if the task is null
-+
+     * @throws RejectedExecutionException if the task cannot be
-+
+     *         scheduled for execution
-+
+     */
+    public <T> ForkJoinTask<T> submit(Callable<T> task) {
-<
+        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
->
+        if (task == null)
->
+            throw new NullPointerException();
->
+        ForkJoinTask<T> job = ForkJoinTask.adapt(task);
+        doSubmit(job);
+        return job;
-+
+    /**
-+
+     * @throws NullPointerException if the task is null
-+
+     * @throws RejectedExecutionException if the task cannot be
-+
+     *         scheduled for execution
-+
+     */
+    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
-<
+        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
->
+        if (task == null)
->
+            throw new NullPointerException();
->
+        ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
+        doSubmit(job);
+        return job;
-+
+    /**
-+
+     * @throws NullPointerException if the task is null
-+
+     * @throws RejectedExecutionException if the task cannot be
-+
+     *         scheduled for execution
-+
+     */
+    public ForkJoinTask<?> submit(Runnable task) {
-<
+        ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
->
+        if (task == null)
->
+            throw new NullPointerException();
->
+        ForkJoinTask<?> job;
->
+        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
->
+            job = (ForkJoinTask<?>) task;
->
+        else
->
+            job = ForkJoinTask.adapt(task, null);
+        doSubmit(job);
+        return job;
+    /**
-<
+     * Adaptor for Runnables. This implements RunnableFuture
-<
+     * to be compliant with AbstractExecutorService constraints
-<
+     */
-<
+    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
-<
+        implements RunnableFuture<T> {
-<
+        final Runnable runnable;
-<
+        final T resultOnCompletion;
-<
+        T result;
-<
+        AdaptedRunnable(Runnable runnable, T result) {
-<
+            if (runnable == null) throw new NullPointerException();
-<
+            this.runnable = runnable;
-<
+            this.resultOnCompletion = result;
-<
+        }
-<
+        public T getRawResult() { return result; }
-<
+        public void setRawResult(T v) { result = v; }
-<
+        public boolean exec() {
-<
+            runnable.run();
-<
+            result = resultOnCompletion;
-<
+            return true;
-<
+        }
-<
+        public void run() { invoke(); }
-<
+    }
-<
-<
+    /**
-<
+     * Adaptor for Callables
->
+     * @throws NullPointerException       {@inheritDoc}
->
+     * @throws RejectedExecutionException {@inheritDoc}
+     */
-–
+    static final class AdaptedCallable<T> extends ForkJoinTask<T>
-–
+        implements RunnableFuture<T> {
-–
+        final Callable<T> callable;
-–
+        T result;
-–
+        AdaptedCallable(Callable<T> callable) {
-–
+            if (callable == null) throw new NullPointerException();
-–
+            this.callable = callable;
-–
+        }
-–
+        public T getRawResult() { return result; }
-–
+        public void setRawResult(T v) { result = v; }
-–
+        public boolean exec() {
-–
+            try {
-–
+                result = callable.call();
-–
+                return true;
-–
+            } catch (Error err) {
-–
+                throw err;
-–
+            } catch (RuntimeException rex) {
-–
+                throw rex;
-–
+            } catch (Exception ex) {
-–
+                throw new RuntimeException(ex);
-–
+            }
-–
+        }
-–
+        public void run() { invoke(); }
-–
+    }
-–
+    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
-<
+        ArrayList<ForkJoinTask<T>> ts =
-<
+            new ArrayList<ForkJoinTask<T>>(tasks.size());
-<
+        for (Callable<T> c : tasks)
-<
+            ts.add(new AdaptedCallable<T>(c));
-<
+        invoke(new InvokeAll<T>(ts));
-<
+        return (List<Future<T>>)(List)ts;
-<
+    }
->
+        // In previous versions of this class, this method constructed
->
+        // a task to run ForkJoinTask.invokeAll, but now external
->
+        // invocation of multiple tasks is at least as efficient.
->
+        List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
->
+        // Workaround needed because method wasn't declared with
->
+        // wildcards in return type but should have been.
->
+        @SuppressWarnings({"unchecked", "rawtypes"})
->
+            List<Future<T>> futures = (List<Future<T>>) (List) fs;
-<
+    static final class InvokeAll<T> extends RecursiveAction {
-<
+        final ArrayList<ForkJoinTask<T>> tasks;
-<
+        InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
-<
+        public void compute() {
-<
+            try { invokeAll(tasks); } catch(Exception ignore) {}
->
+        boolean done = false;
->
+        try {
->
+            for (Callable<T> t : tasks) {
->
+                ForkJoinTask<T> f = ForkJoinTask.adapt(t);
->
+                doSubmit(f);
->
+                fs.add(f);
->
+            }
->
+            for (ForkJoinTask<T> f : fs)
->
+                f.quietlyJoin();
->
+            done = true;
->
+            return futures;
->
+        } finally {
->
+            if (!done)
->
+                for (ForkJoinTask<T> f : fs)
->
+                    f.cancel(false);
-–
+    // Configuration and status settings and queries
-–
+    /**
-<
+     * Returns the factory used for constructing new workers
->
+     * Returns the factory used for constructing new workers.
+     * @return the factory used for constructing new workers
+     */
+    /**
+     * Returns the handler for internal worker threads that terminate
+     * due to unrecoverable errors encountered while executing tasks.
-–
+     * @return the handler, or null if none
-–
+     */
-–
+    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
-–
+        Thread.UncaughtExceptionHandler h;
-–
+        final ReentrantLock lock = this.workerLock;
-–
+        lock.lock();
-–
+        try {
-–
+            h = ueh;
-–
+        } finally {
-–
+            lock.unlock();
-–
+        }
-–
+        return h;
-–
+    }
-–
-–
+    /**
-–
+     * Sets the handler for internal worker threads that terminate due
-–
+     * to unrecoverable errors encountered while executing tasks.
-–
+     * Unless set, the current default or ThreadGroup handler is used
-–
+     * as handler.
-<
+     * @param h the new handler
-<
+     * @return the old handler, or null if none
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     * @return the handler, or {@code null} if none
+     */
-<
+    public Thread.UncaughtExceptionHandler
-<
+        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
-<
+        checkPermission();
-<
+        Thread.UncaughtExceptionHandler old = null;
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            old = ueh;
-<
+            ueh = h;
-<
+            ForkJoinWorkerThread[] ws = workers;
-<
+            for (int i = 0; i < ws.length; ++i) {
-<
+                ForkJoinWorkerThread w = ws[i];
-<
+                if (w != null)
-<
+                    w.setUncaughtExceptionHandler(h);
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+        return old;
-<
+    }
-<
-<
-<
+    /**
-<
+     * Sets the target paralleism level of this pool.
-<
+     * @param parallelism the target parallelism
-<
+     * @throws IllegalArgumentException if parallelism less than or
-<
+     * equal to zero or greater than maximum size bounds.
-<
+     * @throws SecurityException if a security manager exists and
-<
+     *         the caller is not permitted to modify threads
-<
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
-<
+     */
-<
+    public void setParallelism(int parallelism) {
-<
+        checkPermission();
-<
+        if (parallelism <= 0 || parallelism > maxPoolSize)
-<
+            throw new IllegalArgumentException();
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            if (!isTerminating()) {
-<
+                int p = this.parallelism;
-<
+                this.parallelism = parallelism;
-<
+                if (parallelism > p)
-<
+                    createAndStartAddedWorkers();
-<
+                else
-<
+                    trimSpares();
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+        signalIdleWorkers();
->
+    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
->
+        return ueh;
+    /**
-<
+     * Returns the targeted number of worker threads in this pool.
->
+     * Returns the targeted parallelism level of this pool.
-<
+     * @return the targeted number of worker threads in this pool
->
+     * @return the targeted parallelism level of this pool
+     */
+    public int getParallelism() {
+        return parallelism;
+    /**
+     * Returns the number of worker threads that have started but not
-<
+     * yet terminated.  This result returned by this method may differ
-<
+     * from <code>getParallelism</code> when threads are created to
->
+     * yet terminated.  The result returned by this method may differ
->
+     * from {@link #getParallelism} when threads are created to
+     * maintain parallelism when others are cooperatively blocked.
+     * @return the number of worker threads
+     */
+    public int getPoolSize() {
-<
+        return totalCountOf(workerCounts);
->
+        return parallelism + (short)(ctl >>> TC_SHIFT);
+    /**
-<
+     * Returns the maximum number of threads allowed to exist in the
-<
+     * pool, even if there are insufficient unblocked running threads.
-<
+     * @return the maximum
-<
+     */
-<
+    public int getMaximumPoolSize() {
-<
+        return maxPoolSize;
-<
+    }
-<
-<
+    /**
-<
+     * Sets the maximum number of threads allowed to exist in the
-<
+     * pool, even if there are insufficient unblocked running threads.
-<
+     * Setting this value has no effect on current pool size. It
-<
+     * controls construction of new threads.
-<
+     * @throws IllegalArgumentException if negative or greater then
-<
+     * internal implementation limit.
-<
+     */
-<
+    public void setMaximumPoolSize(int newMax) {
-<
+        if (newMax < 0 || newMax > MAX_THREADS)
-<
+            throw new IllegalArgumentException();
-<
+        maxPoolSize = newMax;
-<
+    }
-<
-<
-<
+    /**
-<
+     * Returns true if this pool dynamically maintains its target
-<
+     * parallelism level. If false, new threads are added only to
-<
+     * avoid possible starvation.
-<
+     * This setting is by default true;
-<
+     * @return true if maintains parallelism
-<
+     */
-<
+    public boolean getMaintainsParallelism() {
-<
+        return maintainsParallelism;
-<
+    }
-<
-<
+    /**
-<
+     * Sets whether this pool dynamically maintains its target
-<
+     * parallelism level. If false, new threads are added only to
-<
+     * avoid possible starvation.
-<
+     * @param enable true to maintains parallelism
->
+     * Returns {@code true} if this pool uses local first-in-first-out
->
+     * scheduling mode for forked tasks that are never joined.
->
+     *
->
+     * @return {@code true} if this pool uses async mode
+     */
-<
+    public void setMaintainsParallelism(boolean enable) {
-<
+        maintainsParallelism = enable;
->
+    public boolean getAsyncMode() {
->
+        return localMode != 0;
+    /**
+     * Returns an estimate of the number of worker threads that are
+     * not blocked waiting to join tasks or for other managed
-<
+     * synchronization.
->
+     * synchronization. This method may overestimate the
->
+     * number of running threads.
+     * @return the number of worker threads
+     */
+    public int getRunningThreadCount() {
-<
+        return runningCountOf(workerCounts);
->
+        int rc = 0;
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 1; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null && w.isApparentlyUnblocked())
->
+                    ++rc;
->
+            }
->
+        }
->
+        return rc;
+    /**
+     * Returns an estimate of the number of threads that are currently
+     * stealing or executing tasks. This method may overestimate the
+     * number of active threads.
-<
+     * @return the number of active threads.
->
+     *
->
+     * @return the number of active threads
+     */
+    public int getActiveThreadCount() {
-<
+        return activeCountOf(runControl);
-<
+    }
-<
-<
+    /**
-<
+     * Returns an estimate of the number of threads that are currently
-<
+     * idle waiting for tasks. This method may underestimate the
-<
+     * number of idle threads.
-<
+     * @return the number of idle threads.
-<
+     */
-<
+    final int getIdleThreadCount() {
-<
+        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
-<
+        return (c <= 0)? 0 : c;
->
+        int r = parallelism + (int)(ctl >> AC_SHIFT);
->
+        return (r <= 0) ? 0 : r; // suppress momentarily negative values
+    /**
-<
+     * Returns true if all worker threads are currently idle. An idle
-<
+     * worker is one that cannot obtain a task to execute because none
-<
+     * are available to steal from other threads, and there are no
-<
+     * pending submissions to the pool. This method is conservative:
-<
+     * It might not return true immediately upon idleness of all
-<
+     * threads, but will eventually become true if threads remain
-<
+     * inactive.
-<
+     * @return true if all threads are currently idle
->
+     * Returns {@code true} if all worker threads are currently idle.
->
+     * An idle worker is one that cannot obtain a task to execute
->
+     * because none are available to steal from other threads, and
->
+     * there are no pending submissions to the pool. This method is
->
+     * conservative; it might not return {@code true} immediately upon
->
+     * idleness of all threads, but will eventually become true if
->
+     * threads remain inactive.
->
+     *
->
+     * @return {@code true} if all threads are currently idle
+     */
+    public boolean isQuiescent() {
-<
+        return activeCountOf(runControl) == 0;
->
+        return (int)(ctl >> AC_SHIFT) + parallelism == 0;
+    /**
+     * one thread's work queue by another. The reported value
+     * underestimates the actual total number of steals when the pool
+     * is not quiescent. This value may be useful for monitoring and
-<
+     * tuning fork/join programs: In general, steal counts should be
->
+     * tuning fork/join programs: in general, steal counts should be
+     * high enough to keep threads busy, but low enough to avoid
+     * overhead and contention across threads.
-<
+     * @return the number of steals.
->
+     *
->
+     * @return the number of steals
+     */
+    public long getStealCount() {
-<
+        return stealCount.get();
-<
+    }
-<
-<
+    /**
-<
+     * Accumulate steal count from a worker. Call only
-<
+     * when worker known to be idle.
-<
+     */
-<
+    private void updateStealCount(ForkJoinWorkerThread w) {
-<
+        int sc = w.getAndClearStealCount();
-<
+        if (sc != 0)
-<
+            stealCount.addAndGet(sc);
->
+        long count = stealCount.get();
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 1; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null)
->
+                    count += w.totalSteals;
->
+            }
->
+        }
->
+        return count;
+    /**
+     * an approximation, obtained by iterating across all threads in
+     * the pool. This method may be useful for tuning task
+     * granularities.
-<
+     * @return the number of queued tasks.
->
+     *
->
+     * @return the number of queued tasks
+     */
+    public long getQueuedTaskCount() {
+        long count = 0;
-<
+        ForkJoinWorkerThread[] ws = workers;
-<
+        for (int i = 0; i < ws.length; ++i) {
-<
+            ForkJoinWorkerThread t = ws[i];
-<
+            if (t != null)
-<
+                count += t.getQueueSize();
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 1; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null)
->
+                    count += w.queueSize();
->
+            }
+        return count;
+    /**
-<
+     * Returns an estimate of the number tasks submitted to this pool
-<
+     * that have not yet begun executing. This method takes time
-<
+     * proportional to the number of submissions.
-<
+     * @return the number of queued submissions.
->
+     * Returns an estimate of the number of tasks submitted to this
->
+     * pool that have not yet begun executing.  This method may take
->
+     * time proportional to the number of submissions.
->
+     *
->
+     * @return the number of queued submissions
+     */
+    public int getQueuedSubmissionCount() {
-<
+        return submissionQueue.size();
->
+        int count = 0;
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 0; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null)
->
+                    count += w.queueSize();
->
+            }
->
+        }
->
+        return count;
+    /**
-<
+     * Returns true if there are any tasks submitted to this pool
-<
+     * that have not yet begun executing.
-<
+     * @return <code>true</code> if there are any queued submissions.
->
+     * Returns {@code true} if there are any tasks submitted to this
->
+     * pool that have not yet begun executing.
->
+     *
->
+     * @return {@code true} if there are any queued submissions
+     */
+    public boolean hasQueuedSubmissions() {
-<
+        return !submissionQueue.isEmpty();
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 0; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null && w.queueSize() != 0)
->
+                    return true;
->
+            }
->
+        }
->
+        return false;
+    /**
+     * Removes and returns the next unexecuted submission if one is
+     * available.  This method may be useful in extensions to this
+     * class that re-assign work in systems with multiple pools.
-<
+     * @return the next submission, or null if none
->
+     *
->
+     * @return the next submission, or {@code null} if none
+     */
+    protected ForkJoinTask<?> pollSubmission() {
-<
+        return submissionQueue.poll();
->
+        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 0; i < ws.length; i += 2) {
->
+                if ((w = ws[i]) != null && (t = w.poll()) != null)
->
+                    return t;
->
+            }
->
+        }
->
+        return null;
->
+    }
->
->
+    /**
->
+     * Removes all available unexecuted submitted and forked tasks
->
+     * from scheduling queues and adds them to the given collection,
->
+     * without altering their execution status. These may include
->
+     * artificially generated or wrapped tasks. This method is
->
+     * designed to be invoked only when the pool is known to be
->
+     * quiescent. Invocations at other times may not remove all
->
+     * tasks. A failure encountered while attempting to add elements
->
+     * to collection {@code c} may result in elements being in
->
+     * neither, either or both collections when the associated
->
+     * exception is thrown.  The behavior of this operation is
->
+     * undefined if the specified collection is modified while the
->
+     * operation is in progress.
->
+     *
->
+     * @param c the collection to transfer elements into
->
+     * @return the number of elements transferred
->
+     */
->
+    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
->
+        int count = 0;
->
+        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 0; i < ws.length; ++i) {
->
+                if ((w = ws[i]) != null) {
->
+                    while ((t = w.poll()) != null) {
->
+                        c.add(t);
->
+                        ++count;
->
+                    }
->
+                }
->
+            }
->
+        }
->
+        return count;
+    /**
+     * @return a string identifying this pool, as well as its state
+     */
+    public String toString() {
-<
+        int ps = parallelism;
-<
+        int wc = workerCounts;
-<
+        int rc = runControl;
-<
+        long st = getStealCount();
-<
+        long qt = getQueuedTaskCount();
-<
+        long qs = getQueuedSubmissionCount();
->
+        // Use a single pass through workQueues to collect counts
->
+        long qt = 0L, qs = 0L; int rc = 0;
->
+        long st = stealCount.get();
->
+        long c = ctl;
->
+        WorkQueue[] ws; WorkQueue w;
->
+        if ((ws = workQueues) != null) {
->
+            for (int i = 0; i < ws.length; ++i) {
->
+                if ((w = ws[i]) != null) {
->
+                    int size = w.queueSize();
->
+                    if ((i & 1) == 0)
->
+                        qs += size;
->
+                    else {
->
+                        qt += size;
->
+                        st += w.totalSteals;
->
+                        if (w.isApparentlyUnblocked())
->
+                            ++rc;
->
+                    }
->
+                }
->
+            }
->
+        }
->
+        int pc = parallelism;
->
+        int tc = pc + (short)(c >>> TC_SHIFT);
->
+        int ac = pc + (int)(c >> AC_SHIFT);
->
+        if (ac < 0) // ignore transient negative
->
+            ac = 0;
->
+        String level;
->
+        if ((c & STOP_BIT) != 0)
->
+            level = (tc == 0) ? "Terminated" : "Terminating";
->
+        else
->
+            level = runState < 0 ? "Shutting down" : "Running";
+        return super.toString() +
-<
+            "[" + runStateToString(runStateOf(rc)) +
-<
+            ", parallelism = " + ps +
-<
+            ", size = " + totalCountOf(wc) +
-<
+            ", active = " + activeCountOf(rc) +
-<
+            ", running = " + runningCountOf(wc) +
->
+            "[" + level +
->
+            ", parallelism = " + pc +
->
+            ", size = " + tc +
->
+            ", active = " + ac +
->
+            ", running = " + rc +
+            ", steals = " + st +
+            ", tasks = " + qt +
+            ", submissions = " + qs +
+            "]";
-–
+    private static String runStateToString(int rs) {
-–
+        switch(rs) {
-–
+        case RUNNING: return "Running";
-–
+        case SHUTDOWN: return "Shutting down";
-–
+        case TERMINATING: return "Terminating";
-–
+        case TERMINATED: return "Terminated";
-–
+        default: throw new Error("Unknown run state");
-–
+        }
-–
+    }
-–
-–
+    // lifecycle control
-–
+    /**
+     * Initiates an orderly shutdown in which previously submitted
+     * tasks are executed, but no new tasks will be accepted.
+     * Invocation has no additional effect if already shut down.
+     * Tasks that are in the process of being submitted concurrently
+     * during the course of this method may or may not be rejected.
-+
+     *
+     * @throws SecurityException if a security manager exists and
+     *         the caller is not permitted to modify threads
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     *         java.lang.RuntimePermission}{@code ("modifyThread")}
+     */
+    public void shutdown() {
+        checkPermission();
-<
+        transitionRunStateTo(SHUTDOWN);
-<
+        if (canTerminateOnShutdown(runControl))
-<
+            terminateOnShutdown();
->
+        tryTerminate(false, true);
+    /**
-<
+     * Attempts to stop all actively executing tasks, and cancels all
-<
+     * waiting tasks.  Tasks that are in the process of being
-<
+     * submitted or executed concurrently during the course of this
-<
+     * method may or may not be rejected. Unlike some other executors,
-<
+     * this method cancels rather than collects non-executed tasks,
-<
+     * so always returns an empty list.
->
+     * Attempts to cancel and/or stop all tasks, and reject all
->
+     * subsequently submitted tasks.  Tasks that are in the process of
->
+     * being submitted or executed concurrently during the course of
->
+     * this method may or may not be rejected. This method cancels
->
+     * both existing and unexecuted tasks, in order to permit
->
+     * termination in the presence of task dependencies. So the method
->
+     * always returns an empty list (unlike the case for some other
->
+     * Executors).
->
+     *
+     * @return an empty list
+     * @throws SecurityException if a security manager exists and
+     *         the caller is not permitted to modify threads
+     *         because it does not hold {@link
-<
+     *         java.lang.RuntimePermission}<code>("modifyThread")</code>,
->
+     *         java.lang.RuntimePermission}{@code ("modifyThread")}
+     */
+    public List<Runnable> shutdownNow() {
+        checkPermission();
-<
+        terminate();
->
+        tryTerminate(true, true);
+        return Collections.emptyList();
+    /**
-<
+     * Returns <code>true</code> if all tasks have completed following shut down.
->
+     * Returns {@code true} if all tasks have completed following shut down.
-<
+     * @return <code>true</code> if all tasks have completed following shut down
->
+     * @return {@code true} if all tasks have completed following shut down
+     */
+    public boolean isTerminated() {
-<
+        return runStateOf(runControl) == TERMINATED;
->
+        long c = ctl;
->
+        return ((c & STOP_BIT) != 0L &&
->
+                (short)(c >>> TC_SHIFT) == -parallelism);
+    /**
-<
+     * Returns <code>true</code> if the process of termination has
-<
+     * commenced but possibly not yet completed.
->
+     * Returns {@code true} if the process of termination has
->
+     * commenced but not yet completed.  This method may be useful for
->
+     * debugging. A return of {@code true} reported a sufficient
->
+     * period after shutdown may indicate that submitted tasks have
->
+     * ignored or suppressed interruption, or are waiting for IO,
->
+     * causing this executor not to properly terminate. (See the
->
+     * advisory notes for class {@link ForkJoinTask} stating that
->
+     * tasks should not normally entail blocking operations.  But if
->
+     * they do, they must abort them on interrupt.)
-<
+     * @return <code>true</code> if terminating
->
+     * @return {@code true} if terminating but not yet terminated
+     */
+    public boolean isTerminating() {
-<
+        return runStateOf(runControl) >= TERMINATING;
->
+        long c = ctl;
->
+        return ((c & STOP_BIT) != 0L &&
->
+                (short)(c >>> TC_SHIFT) != -parallelism);
+    /**
-<
+     * Returns <code>true</code> if this pool has been shut down.
->
+     * Returns {@code true} if this pool has been shut down.
-<
+     * @return <code>true</code> if this pool has been shut down
->
+     * @return {@code true} if this pool has been shut down
+     */
+    public boolean isShutdown() {
-<
+        return runStateOf(runControl) >= SHUTDOWN;
->
+        return runState < 0;
+    /**
+     * @param timeout the maximum time to wait
+     * @param unit the time unit of the timeout argument
-<
+     * @return <code>true</code> if this executor terminated and
-<
+     *         <code>false</code> if the timeout elapsed before termination
->
+     * @return {@code true} if this executor terminated and
->
+     *         {@code false} if the timeout elapsed before termination
+     * @throws InterruptedException if interrupted while waiting
+     */
+    public boolean awaitTermination(long timeout, TimeUnit unit)
+        throws InterruptedException {
+        long nanos = unit.toNanos(timeout);
-<
+        final ReentrantLock lock = this.workerLock;
->
+        final Mutex lock = this.lock;
+        lock.lock();
+        try {
+            for (;;) {
-–
+    // Shutdown and termination support
-–
+    /**
-<
+     * Callback from terminating worker. Null out the corresponding
-<
+     * workers slot, and if terminating, try to terminate, else try to
-<
+     * shrink workers array.
-<
+     * @param w the worker
-<
+     */
-<
+    final void workerTerminated(ForkJoinWorkerThread w) {
-<
+        updateStealCount(w);
-<
+        updateWorkerCount(-1);
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            ForkJoinWorkerThread[] ws = workers;
-<
+            int idx = w.poolIndex;
-<
+            if (idx >= 0 && idx < ws.length && ws[idx] == w)
-<
+                ws[idx] = null;
-<
+            if (totalCountOf(workerCounts) == 0) {
-<
+                terminate(); // no-op if already terminating
-<
+                transitionRunStateTo(TERMINATED);
-<
+                termination.signalAll();
-<
+            }
-<
+            else if (!isTerminating()) {
-<
+                tryShrinkWorkerArray();
-<
+                tryResumeSpare(true); // allow replacement
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+        signalIdleWorkers();
-<
+    }
-<
-<
+    /**
-<
+     * Initiate termination.
-<
+     */
-<
+    private void terminate() {
-<
+        if (transitionRunStateTo(TERMINATING)) {
-<
+            stopAllWorkers();
-<
+            resumeAllSpares();
-<
+            signalIdleWorkers();
-<
+            cancelQueuedSubmissions();
-<
+            cancelQueuedWorkerTasks();
-<
+            interruptUnterminatedWorkers();
-<
+            signalIdleWorkers(); // resignal after interrupt
-<
+        }
-<
+    }
-<
-<
+    /**
-<
+     * Possibly terminate when on shutdown state
-<
+     */
-<
+    private void terminateOnShutdown() {
-<
+        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
-<
+            terminate();
-<
+    }
-<
-<
+    /**
-<
+     * Clear out and cancel submissions
-<
+     */
-<
+    private void cancelQueuedSubmissions() {
-<
+        ForkJoinTask<?> task;
-<
+        while ((task = pollSubmission()) != null)
-<
+            task.cancel(false);
-<
+    }
-<
-<
+    /**
-<
+     * Clean out worker queues.
-<
+     */
-<
+    private void cancelQueuedWorkerTasks() {
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            ForkJoinWorkerThread[] ws = workers;
-<
+            for (int i = 0; i < ws.length; ++i) {
-<
+                ForkJoinWorkerThread t = ws[i];
-<
+                if (t != null)
-<
+                    t.cancelTasks();
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+    }
-<
-<
+    /**
-<
+     * Set each worker's status to terminating. Requires lock to avoid
-<
+     * conflicts with add/remove
-<
+     */
-<
+    private void stopAllWorkers() {
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            ForkJoinWorkerThread[] ws = workers;
-<
+            for (int i = 0; i < ws.length; ++i) {
-<
+                ForkJoinWorkerThread t = ws[i];
-<
+                if (t != null)
-<
+                    t.shutdownNow();
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+    }
-<
-<
+    /**
-<
+     * Interrupt all unterminated workers.  This is not required for
-<
+     * sake of internal control, but may help unstick user code during
-<
+     * shutdown.
-<
+     */
-<
+    private void interruptUnterminatedWorkers() {
-<
+        final ReentrantLock lock = this.workerLock;
-<
+        lock.lock();
-<
+        try {
-<
+            ForkJoinWorkerThread[] ws = workers;
-<
+            for (int i = 0; i < ws.length; ++i) {
-<
+                ForkJoinWorkerThread t = ws[i];
-<
+                if (t != null && !t.isTerminated()) {
-<
+                    try {
-<
+                        t.interrupt();
-<
+                    } catch (SecurityException ignore) {
-<
+                    }
-<
+                }
-<
+            }
-<
+        } finally {
-<
+            lock.unlock();
-<
+        }
-<
+    }
-<
-<
-<
+    /*
-<
+     * Nodes for event barrier to manage idle threads.  Queue nodes
-<
+     * are basic Treiber stack nodes, also used for spare stack.
->
+     * Interface for extending managed parallelism for tasks running
->
+     * in {@link ForkJoinPool}s.
-<
+     * The event barrier has an event count and a wait queue (actually
-<
+     * a Treiber stack).  Workers are enabled to look for work when
-<
+     * the eventCount is incremented. If they fail to find work, they
-<
+     * may wait for next count. Upon release, threads help others wake
-<
+     * up.
-<
+     *
-<
+     * Synchronization events occur only in enough contexts to
-<
+     * maintain overall liveness:
-<
+     *
-<
+     *   - Submission of a new task to the pool
-<
+     *   - Resizes or other changes to the workers array
-<
+     *   - pool termination
-<
+     *   - A worker pushing a task on an empty queue
-<
+     *
-<
+     * The case of pushing a task occurs often enough, and is heavy
-<
+     * enough compared to simple stack pushes, to require special
-<
+     * handling: Method signalWork returns without advancing count if
-<
+     * the queue appears to be empty.  This would ordinarily result in
-<
+     * races causing some queued waiters not to be woken up. To avoid
-<
+     * this, the first worker enqueued in method sync (see
-<
+     * syncIsReleasable) rescans for tasks after being enqueued, and
-<
+     * helps signal if any are found. This works well because the
-<
+     * worker has nothing better to do, and so might as well help
-<
+     * alleviate the overhead and contention on the threads actually
-<
+     * doing work.  Also, since event counts increments on task
-<
+     * availability exist to maintain liveness (rather than to force
-<
+     * refreshes etc), it is OK for callers to exit early if
-<
+     * contending with another signaller.
-<
+     */
-<
+    static final class WaitQueueNode {
-<
+        WaitQueueNode next; // only written before enqueued
-<
+        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
-<
+        final long count; // unused for spare stack
-<
-<
+        WaitQueueNode(long c, ForkJoinWorkerThread w) {
-<
+            count = c;
-<
+            thread = w;
-<
+        }
-<
-<
+        /**
-<
+         * Wake up waiter, returning false if known to already
-<
+         */
-<
+        boolean signal() {
-<
+            ForkJoinWorkerThread t = thread;
-<
+            if (t == null)
-<
+                return false;
-<
+            thread = null;
-<
+            LockSupport.unpark(t);
-<
+            return true;
-<
+        }
-<
-<
+        /**
-<
+         * Await release on sync
-<
+         */
-<
+        void awaitSyncRelease(ForkJoinPool p) {
-<
+            while (thread != null && !p.syncIsReleasable(this))
-<
+                LockSupport.park(this);
-<
+        }
-<
-<
+        /**
-<
+         * Await resumption as spare
-<
+         */
-<
+        void awaitSpareRelease() {
-<
+            while (thread != null) {
-<
+                if (!Thread.interrupted())
-<
+                    LockSupport.park(this);
-<
+            }
-<
+        }
-<
+    }
-<
-<
+    /**
-<
+     * Ensures that no thread is waiting for count to advance from the
-<
+     * current value of eventCount read on entry to this method, by
-<
+     * releasing waiting threads if necessary.
-<
+     * @return the count
-<
+     */
-<
+    final long ensureSync() {
-<
+        long c = eventCount;
-<
+        WaitQueueNode q;
-<
+        while ((q = syncStack) != null && q.count < c) {
-<
+            if (casBarrierStack(q, null)) {
-<
+                do {
-<
+                    q.signal();
-<
+                } while ((q = q.next) != null);
-<
+                break;
-<
+            }
-<
+        }
-<
+        return c;
-<
+    }
-<
-<
+    /**
-<
+     * Increments event count and releases waiting threads.
-<
+     */
-<
+    private void signalIdleWorkers() {
-<
+        long c;
-<
+        do;while (!casEventCount(c = eventCount, c+1));
-<
+        ensureSync();
-<
+    }
-<
-<
+    /**
-<
+     * Signal threads waiting to poll a task. Because method sync
-<
+     * rechecks availability, it is OK to only proceed if queue
-<
+     * appears to be non-empty, and OK to skip under contention to
-<
+     * increment count (since some other thread succeeded).
-<
+     */
-<
+    final void signalWork() {
-<
+        long c;
-<
+        WaitQueueNode q;
-<
+        if (syncStack != null &&
-<
+            casEventCount(c = eventCount, c+1) &&
-<
+            (((q = syncStack) != null && q.count <= c) &&
-<
+             (!casBarrierStack(q, q.next) || !q.signal())))
-<
+            ensureSync();
-<
+    }
-<
-<
+    /**
-<
+     * Waits until event count advances from last value held by
-<
+     * caller, or if excess threads, caller is resumed as spare, or
-<
+     * caller or pool is terminating. Updates caller's event on exit.
-<
+     * @param w the calling worker thread
-<
+     */
-<
+    final void sync(ForkJoinWorkerThread w) {
-<
+        updateStealCount(w); // Transfer w's count while it is idle
-<
-<
+        while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
-<
+            long prev = w.lastEventCount;
-<
+            WaitQueueNode node = null;
-<
+            WaitQueueNode h;
-<
+            while (eventCount == prev &&
-<
+                   ((h = syncStack) == null || h.count == prev)) {
-<
+                if (node == null)
-<
+                    node = new WaitQueueNode(prev, w);
-<
+                if (casBarrierStack(node.next = h, node)) {
-<
+                    node.awaitSyncRelease(this);
-<
+                    break;
-<
+                }
-<
+            }
-<
+            long ec = ensureSync();
-<
+            if (ec != prev) {
-<
+                w.lastEventCount = ec;
-<
+                break;
-<
+            }
-<
+        }
-<
+    }
-<
-<
+    /**
-<
+     * Returns true if worker waiting on sync can proceed:
-<
+     *  - on signal (thread == null)
-<
+     *  - on event count advance (winning race to notify vs signaller)
-<
+     *  - on Interrupt
-<
+     *  - if the first queued node, we find work available
-<
+     * If node was not signalled and event count not advanced on exit,
-<
+     * then we also help advance event count.
-<
+     * @return true if node can be released
-<
+     */
-<
+    final boolean syncIsReleasable(WaitQueueNode node) {
-<
+        long prev = node.count;
-<
+        if (!Thread.interrupted() && node.thread != null &&
-<
+            (node.next != null ||
-<
+             !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
-<
+            eventCount == prev)
-<
+            return false;
-<
+        if (node.thread != null) {
-<
+            node.thread = null;
-<
+            long ec = eventCount;
-<
+            if (prev <= ec) // help signal
-<
+                casEventCount(ec, ec+1);
-<
+        }
-<
+        return true;
-<
+    }
-<
-<
+    /**
-<
+     * Returns true if a new sync event occurred since last call to
-<
+     * sync or this method, if so, updating caller's count.
-<
+     */
-<
+    final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
-<
+        long lc = w.lastEventCount;
-<
+        long ec = ensureSync();
-<
+        if (ec == lc)
-<
+            return false;
-<
+        w.lastEventCount = ec;
-<
+        return true;
-<
+    }
-<
-<
+    //  Parallelism maintenance
-<
-<
+    /**
-<
+     * Decrement running count; if too low, add spare.
->
+     * <p>A {@code ManagedBlocker} provides two methods.  Method
->
+     * {@code isReleasable} must return {@code true} if blocking is
->
+     * not necessary. Method {@code block} blocks the current thread
->
+     * if necessary (perhaps internally invoking {@code isReleasable}
->
+     * before actually blocking). These actions are performed by any
->
+     * thread invoking {@link ForkJoinPool#managedBlock}.  The
->
+     * unusual methods in this API accommodate synchronizers that may,
->
+     * but don't usually, block for long periods. Similarly, they
->
+     * allow more efficient internal handling of cases in which
->
+     * additional workers may be, but usually are not, needed to
->
+     * ensure sufficient parallelism.  Toward this end,
->
+     * implementations of method {@code isReleasable} must be amenable
->
+     * to repeated invocation.
-–
+     * Conceptually, all we need to do here is add or resume a
-–
+     * spare thread when one is about to block (and remove or
-–
+     * suspend it later when unblocked -- see suspendIfSpare).
-–
+     * However, implementing this idea requires coping with
-–
+     * several problems: We have imperfect information about the
-–
+     * states of threads. Some count updates can and usually do
-–
+     * lag run state changes, despite arrangements to keep them
-–
+     * accurate (for example, when possible, updating counts
-–
+     * before signalling or resuming), especially when running on
-–
+     * dynamic JVMs that don't optimize the infrequent paths that
-–
+     * update counts. Generating too many threads can make these
-–
+     * problems become worse, because excess threads are more
-–
+     * likely to be context-switched with others, slowing them all
-–
+     * down, especially if there is no work available, so all are
-–
+     * busy scanning or idling.  Also, excess spare threads can
-–
+     * only be suspended or removed when they are idle, not
-–
+     * immediately when they aren't needed. So adding threads will
-–
+     * raise parallelism level for longer than necessary.  Also,
-–
+     * FJ applications often enounter highly transient peaks when
-–
+     * many threads are blocked joining, but for less time than it
-–
+     * takes to create or resume spares.
-–
+     *
-–
+     * @param joinMe if non-null, return early if done
-–
+     * @param maintainParallelism if true, try to stay within
-–
+     * target counts, else create only to avoid starvation
-–
+     * @return true if joinMe known to be done
-–
+     */
-–
+    final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
-–
+        maintainParallelism &= maintainsParallelism; // overrride
-–
+        boolean dec = false;  // true when running count decremented
-–
+        while (spareStack == null || !tryResumeSpare(dec)) {
-–
+            int counts = workerCounts;
-–
+            if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
-–
+                if (!needSpare(counts, maintainParallelism))
-–
+                    break;
-–
+                if (joinMe.status < 0)
-–
+                    return true;
-–
+                if (tryAddSpare(counts))
-–
+                    break;
-–
+            }
-–
+        }
-–
+        return false;
-–
+    }
-–
-–
+    /**
-–
+     * Same idea as preJoin
-–
+     */
-–
+    final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
-–
+        maintainParallelism &= maintainsParallelism;
-–
+        boolean dec = false;
-–
+        while (spareStack == null || !tryResumeSpare(dec)) {
-–
+            int counts = workerCounts;
-–
+            if (dec || (dec = casWorkerCounts(counts, --counts))) {
-–
+                if (!needSpare(counts, maintainParallelism))
-–
+                    break;
-–
+                if (blocker.isReleasable())
-–
+                    return true;
-–
+                if (tryAddSpare(counts))
-–
+                    break;
-–
+            }
-–
+        }
-–
+        return false;
-–
+    }
-–
-–
+    /**
-–
+     * Returns true if a spare thread appears to be needed.  If
-–
+     * maintaining parallelism, returns true when the deficit in
-–
+     * running threads is more than the surplus of total threads, and
-–
+     * there is apparently some work to do.  This self-limiting rule
-–
+     * means that the more threads that have already been added, the
-–
+     * less parallelism we will tolerate before adding another.
-–
+     * @param counts current worker counts
-–
+     * @param maintainParallelism try to maintain parallelism
-–
+     */
-–
+    private boolean needSpare(int counts, boolean maintainParallelism) {
-–
+        int ps = parallelism;
-–
+        int rc = runningCountOf(counts);
-–
+        int tc = totalCountOf(counts);
-–
+        int runningDeficit = ps - rc;
-–
+        int totalSurplus = tc - ps;
-–
+        return (tc < maxPoolSize &&
-–
+                (rc == 0 || totalSurplus < 0 ||
-–
+                 (maintainParallelism &&
-–
+                  runningDeficit > totalSurplus &&
-–
+                  ForkJoinWorkerThread.hasQueuedTasks(workers))));
-–
+    }
-–
-–
+    /**
-–
+     * Add a spare worker if lock available and no more than the
-–
+     * expected numbers of threads exist
-–
+     * @return true if successful
-–
+     */
-–
+    private boolean tryAddSpare(int expectedCounts) {
-–
+        final ReentrantLock lock = this.workerLock;
-–
+        int expectedRunning = runningCountOf(expectedCounts);
-–
+        int expectedTotal = totalCountOf(expectedCounts);
-–
+        boolean success = false;
-–
+        boolean locked = false;
-–
+        // confirm counts while locking; CAS after obtaining lock
-–
+        try {
-–
+            for (;;) {
-–
+                int s = workerCounts;
-–
+                int tc = totalCountOf(s);
-–
+                int rc = runningCountOf(s);
-–
+                if (rc > expectedRunning || tc > expectedTotal)
-–
+                    break;
-–
+                if (!locked && !(locked = lock.tryLock()))
-–
+                    break;
-–
+                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
-–
+                    createAndStartSpare(tc);
-–
+                    success = true;
-–
+                    break;
-–
+                }
-–
+            }
-–
+        } finally {
-–
+            if (locked)
-–
+                lock.unlock();
-–
+        }
-–
+        return success;
-–
+    }
-–
-–
+    /**
-–
+     * Add the kth spare worker. On entry, pool coounts are already
-–
+     * adjusted to reflect addition.
-–
+     */
-–
+    private void createAndStartSpare(int k) {
-–
+        ForkJoinWorkerThread w = null;
-–
+        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
-–
+        int len = ws.length;
-–
+        // Probably, we can place at slot k. If not, find empty slot
-–
+        if (k < len && ws[k] != null) {
-–
+            for (k = 0; k < len && ws[k] != null; ++k)
-–
+                ;
-–
+        }
-–
+        if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
-–
+            ws[k] = w;
-–
+            w.start();
-–
+        }
-–
+        else
-–
+            updateWorkerCount(-1); // adjust on failure
-–
+        signalIdleWorkers();
-–
+    }
-–
-–
+    /**
-–
+     * Suspend calling thread w if there are excess threads.  Called
-–
+     * only from sync.  Spares are enqueued in a Treiber stack
-–
+     * using the same WaitQueueNodes as barriers.  They are resumed
-–
+     * mainly in preJoin, but are also woken on pool events that
-–
+     * require all threads to check run state.
-–
+     * @param w the caller
-–
+     */
-–
+    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
-–
+        WaitQueueNode node = null;
-–
+        int s;
-–
+        while (parallelism < runningCountOf(s = workerCounts)) {
-–
+            if (node == null)
-–
+                node = new WaitQueueNode(0, w);
-–
+            if (casWorkerCounts(s, s-1)) { // representation-dependent
-–
+                // push onto stack
-–
+                do;while (!casSpareStack(node.next = spareStack, node));
-–
+                // block until released by resumeSpare
-–
+                node.awaitSpareRelease();
-–
+                return true;
-–
+            }
-–
+        }
-–
+        return false;
-–
+    }
-–
-–
+    /**
-–
+     * Try to pop and resume a spare thread.
-–
+     * @param updateCount if true, increment running count on success
-–
+     * @return true if successful
-–
+     */
-–
+    private boolean tryResumeSpare(boolean updateCount) {
-–
+        WaitQueueNode q;
-–
+        while ((q = spareStack) != null) {
-–
+            if (casSpareStack(q, q.next)) {
-–
+                if (updateCount)
-–
+                    updateRunningCount(1);
-–
+                q.signal();
-–
+                return true;
-–
+            }
-–
+        }
-–
+        return false;
-–
+    }
-–
-–
+    /**
-–
+     * Pop and resume all spare threads. Same idea as ensureSync.
-–
+     * @return true if any spares released
-–
+     */
-–
+    private boolean resumeAllSpares() {
-–
+        WaitQueueNode q;
-–
+        while ( (q = spareStack) != null) {
-–
+            if (casSpareStack(q, null)) {
-–
+                do {
-–
+                    updateRunningCount(1);
-–
+                    q.signal();
-–
+                } while ((q = q.next) != null);
-–
+                return true;
-–
+            }
-–
+        }
-–
+        return false;
-–
+    }
-–
-–
+    /**
-–
+     * Pop and shutdown excessive spare threads. Call only while
-–
+     * holding lock. This is not guaranteed to eliminate all excess
-–
+     * threads, only those suspended as spares, which are the ones
-–
+     * unlikely to be needed in the future.
-–
+     */
-–
+    private void trimSpares() {
-–
+        int surplus = totalCountOf(workerCounts) - parallelism;
-–
+        WaitQueueNode q;
-–
+        while (surplus > 0 && (q = spareStack) != null) {
-–
+            if (casSpareStack(q, null)) {
-–
+                do {
-–
+                    updateRunningCount(1);
-–
+                    ForkJoinWorkerThread w = q.thread;
-–
+                    if (w != null && surplus > 0 &&
-–
+                        runningCountOf(workerCounts) > 0 && w.shutdown())
-–
+                        --surplus;
-–
+                    q.signal();
-–
+                } while ((q = q.next) != null);
-–
+            }
-–
+        }
-–
+    }
-–
-–
+    /**
-–
+     * Interface for extending managed parallelism for tasks running
-–
+     * in ForkJoinPools. A ManagedBlocker provides two methods.
-–
+     * Method <code>isReleasable</code> must return true if blocking is not
-–
+     * necessary. Method <code>block</code> blocks the current thread
-–
+     * if necessary (perhaps internally invoking isReleasable before
-–
+     * actually blocking.).
+     * <p>For example, here is a ManagedBlocker based on a
+     * ReentrantLock:
-<
+     * <pre>
-<
+     *   class ManagedLocker implements ManagedBlocker {
-<
+     *     final ReentrantLock lock;
-<
+     *     boolean hasLock = false;
-<
+     *     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
-<
+     *     public boolean block() {
-<
+     *        if (!hasLock)
-<
+     *           lock.lock();
-<
+     *        return true;
-<
+     *     }
-<
+     *     public boolean isReleasable() {
-<
+     *        return hasLock || (hasLock = lock.tryLock());
-<
+     *     }
->
+     *  <pre> {@code
->
+     * class ManagedLocker implements ManagedBlocker {
->
+     *   final ReentrantLock lock;
->
+     *   boolean hasLock = false;
->
+     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
->
+     *   public boolean block() {
->
+     *     if (!hasLock)
->
+     *       lock.lock();
->
+     *     return true;
->
+     *   }
->
+     *   public boolean isReleasable() {
->
+     *     return hasLock || (hasLock = lock.tryLock());
->
+     *   }
->
+     * }}</pre>
->
+     *
->
+     * <p>Here is a class that possibly blocks waiting for an
->
+     * item on a given queue:
->
+     *  <pre> {@code
->
+     * class QueueTaker<E> implements ManagedBlocker {
->
+     *   final BlockingQueue<E> queue;
->
+     *   volatile E item = null;
->
+     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
->
+     *   public boolean block() throws InterruptedException {
->
+     *     if (item == null)
->
+     *       item = queue.take();
->
+     *     return true;
->
+     *   }
->
+     *   public boolean isReleasable() {
->
+     *     return item != null || (item = queue.poll()) != null;
+     *   }
-<
+     * </pre>
->
+     *   public E getItem() { // call after pool.managedBlock completes
->
+     *     return item;
->
+     *   }
->
+     * }}</pre>
+     */
+    public static interface ManagedBlocker {
+        /**
+         * Possibly blocks the current thread, for example waiting for
+         * a lock or condition.
-<
+         * @return true if no additional blocking is necessary (i.e.,
-<
+         * if isReleasable would return true).
->
+         *
->
+         * @return {@code true} if no additional blocking is necessary
->
+         * (i.e., if isReleasable would return true)
+         * @throws InterruptedException if interrupted while waiting
-<
+         * (the method is not required to do so, but is allowe to).
->
+         * (the method is not required to do so, but is allowed to)
+         */
+        boolean block() throws InterruptedException;
+        /**
-<
+         * Returns true if blocking is unnecessary.
->
+         * Returns {@code true} if blocking is unnecessary.
+         */
+        boolean isReleasable();
+    /**
+     * Blocks in accord with the given blocker.  If the current thread
-<
+     * is a ForkJoinWorkerThread, this method possibly arranges for a
-<
+     * spare thread to be activated if necessary to ensure parallelism
-<
+     * while the current thread is blocked.  If
-<
+     * <code>maintainParallelism</code> is true and the pool supports
-<
+     * it ({@link #getMaintainsParallelism}), this method attempts to
-<
+     * maintain the pool's nominal parallelism. Otherwise if activates
-<
+     * a thread only if necessary to avoid complete starvation. This
-<
+     * option may be preferable when blockages use timeouts, or are
-<
+     * almost always brief.
-<
+     *
-<
+     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
-<
+     * equivalent to
-<
+     * <pre>
-<
+     *   while (!blocker.isReleasable())
-<
+     *      if (blocker.block())
-<
+     *         return;
-<
+     * </pre>
-<
+     * If the caller is a ForkJoinTask, then the pool may first
-<
+     * be expanded to ensure parallelism, and later adjusted.
->
+     * is a {@link ForkJoinWorkerThread}, this method possibly
->
+     * arranges for a spare thread to be activated if necessary to
->
+     * ensure sufficient parallelism while the current thread is blocked.
->
+     *
->
+     * <p>If the caller is not a {@link ForkJoinTask}, this method is
->
+     * behaviorally equivalent to
->
+     *  <pre> {@code
->
+     * while (!blocker.isReleasable())
->
+     *   if (blocker.block())
->
+     *     return;
->
+     * }</pre>
->
+     *
->
+     * If the caller is a {@code ForkJoinTask}, then the pool may
->
+     * first be expanded to ensure parallelism, and later adjusted.
+     * @param blocker the blocker
-<
+     * @param maintainParallelism if true and supported by this pool,
-<
+     * attempt to maintain the pool's nominal parallelism; otherwise
-<
+     * activate a thread only if necessary to avoid complete
-<
+     * starvation.
-<
+     * @throws InterruptedException if blocker.block did so.
->
+     * @throws InterruptedException if blocker.block did so
+     */
-<
+    public static void managedBlock(ManagedBlocker blocker,
-<
+                                    boolean maintainParallelism)
->
+    public static void managedBlock(ManagedBlocker blocker)
+        throws InterruptedException {
+        Thread t = Thread.currentThread();
-<
+        ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
-<
+                             ((ForkJoinWorkerThread)t).pool : null);
-<
+        if (!blocker.isReleasable()) {
-<
+            try {
-<
+                if (pool == null ||
-<
+                    !pool.preBlock(blocker, maintainParallelism))
-<
+                    awaitBlocker(blocker);
-<
+            } finally {
-<
+                if (pool != null)
-<
+                    pool.updateRunningCount(1);
->
+        ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
->
+                          ((ForkJoinWorkerThread)t).pool : null);
->
+        while (!blocker.isReleasable()) {
->
+            if (p == null || p.tryCompensate()) {
->
+                try {
->
+                    do {} while (!blocker.isReleasable() && !blocker.block());
->
+                } finally {
->
+                    if (p != null)
->
+                        p.incrementActiveCount();
->
+                }
->
+                break;
-<
+    private static void awaitBlocker(ManagedBlocker blocker)
-<
+        throws InterruptedException {
-<
+        do;while (!blocker.isReleasable() && !blocker.block());
-<
+    }
-<
-<
+    // AbstractExecutorService overrides
->
+    // AbstractExecutorService overrides.  These rely on undocumented
->
+    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
->
+    // implement RunnableFuture.
+    protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
-<
+        return new AdaptedRunnable(runnable, value);
->
+        return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
+    protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
-<
+        return new AdaptedCallable(callable);
->
+        return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
-+
+    // Unsafe mechanics
-+
+    private static final sun.misc.Unsafe U;
-+
+    private static final long CTL;
-+
+    private static final long PARKBLOCKER;
-<
+    // Temporary Unsafe mechanics for preliminary release
-<
+    private static Unsafe getUnsafe() throws Throwable {
->
+    static {
->
+        poolNumberGenerator = new AtomicInteger();
->
+        modifyThreadPermission = new RuntimePermission("modifyThread");
->
+        defaultForkJoinWorkerThreadFactory =
->
+            new DefaultForkJoinWorkerThreadFactory();
->
+        submitters = new ThreadSubmitter();
->
+        try {
->
+            U = getUnsafe();
->
+            Class<?> k = ForkJoinPool.class;
->
+            CTL = U.objectFieldOffset
->
+                (k.getDeclaredField("ctl"));
->
+            Class<?> tk = Thread.class;
->
+            PARKBLOCKER = U.objectFieldOffset
->
+                (tk.getDeclaredField("parkBlocker"));
->
+        } catch (Exception e) {
->
+            throw new Error(e);
->
+        }
->
+    }
->
->
+    /**
->
+     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
->
+     * Replace with a simple call to Unsafe.getUnsafe when integrating
->
+     * into a jdk.
->
+     *
->
+     * @return a sun.misc.Unsafe
->
+     */
->
+    private static sun.misc.Unsafe getUnsafe() {
+        try {
-<
+            return Unsafe.getUnsafe();
->
+            return sun.misc.Unsafe.getUnsafe();
+        } catch (SecurityException se) {
+            try {
+                return java.security.AccessController.doPrivileged
-<
+                    (new java.security.PrivilegedExceptionAction<Unsafe>() {
-<
+                        public Unsafe run() throws Exception {
-<
+                            return getUnsafePrivileged();
->
+                    (new java.security
->
+                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
->
+                        public sun.misc.Unsafe run() throws Exception {
->
+                            java.lang.reflect.Field f = sun.misc
->
+                                .Unsafe.class.getDeclaredField("theUnsafe");
->
+                            f.setAccessible(true);
->
+                            return (sun.misc.Unsafe) f.get(null);
+                        }});
+            } catch (java.security.PrivilegedActionException e) {
-<
+                throw e.getCause();
->
+                throw new RuntimeException("Could not initialize intrinsics",
->
+                                           e.getCause());
-–
+    private static Unsafe getUnsafePrivileged()
-–
+            throws NoSuchFieldException, IllegalAccessException {
-–
+        Field f = Unsafe.class.getDeclaredField("theUnsafe");
-–
+        f.setAccessible(true);
-–
+        return (Unsafe) f.get(null);
-–
+    }
-–
-–
+    private static long fieldOffset(String fieldName)
-–
+            throws NoSuchFieldException {
-–
+        return _unsafe.objectFieldOffset
-–
+            (ForkJoinPool.class.getDeclaredField(fieldName));
-–
+    }
-–
-–
+    static final Unsafe _unsafe;
-–
+    static final long eventCountOffset;
-–
+    static final long workerCountsOffset;
-–
+    static final long runControlOffset;
-–
+    static final long syncStackOffset;
-–
+    static final long spareStackOffset;
-–
-–
+    static {
-–
+        try {
-–
+            _unsafe = getUnsafe();
-–
+            eventCountOffset = fieldOffset("eventCount");
-–
+            workerCountsOffset = fieldOffset("workerCounts");
-–
+            runControlOffset = fieldOffset("runControl");
-–
+            syncStackOffset = fieldOffset("syncStack");
-–
+            spareStackOffset = fieldOffset("spareStack");
-–
+        } catch (Throwable e) {
-–
+            throw new RuntimeException("Could not initialize intrinsics", e);
-–
+        }
-–
+    }
-–
-–
+    private boolean casEventCount(long cmp, long val) {
-–
+        return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
-–
+    }
-–
+    private boolean casWorkerCounts(int cmp, int val) {
-–
+        return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
-–
+    }
-–
+    private boolean casRunControl(int cmp, int val) {
-–
+        return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
-–
+    }
-–
+    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
-–
+        return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
-–
+    }
-–
+    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
-–
+        return _unsafe.compareAndSwapObject(this, syncStackOffset, cmp, val);
-–
+    }

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents): Revision 1.5 by jsr166, Thu Mar 19 05:10:42 2009 UTC vs. Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC

Diff Legend

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.5 by jsr166, Thu Mar 19 05:10:42 2009 UTC vs.
Revision 1.120 by jsr166, Tue Jan 31 01:32:25 2012 UTC