--- jsr166/src/jsr166y/ForkJoinPool.java 2009/07/25 17:49:01 1.24
+++ jsr166/src/jsr166y/ForkJoinPool.java 2012/11/18 06:28:18 1.143
@@ -1,68 +1,121 @@
/*
* 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.concurrent.*;
-
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
import java.util.List;
-import java.util.concurrent.locks.Condition;
-import java.util.concurrent.locks.LockSupport;
-import java.util.concurrent.locks.ReentrantLock;
-import java.util.concurrent.atomic.AtomicInteger;
-import java.util.concurrent.atomic.AtomicLong;
+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;
/**
- * 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.
+ *
+ *
A {@code ForkJoinPool} differs from other kinds of {@link
+ * ExecutorService} mainly by virtue of employing
+ * work-stealing: 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 asyncMode to true in constructors, {@code
+ * ForkJoinPool}s may also be appropriate for use with event-style
+ * tasks that are never joined.
*
- *
ForkJoinPools differ from other kinds of Executors mainly in
- * that they provide work-stealing: 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. When setting
- * {@code setAsyncMode}, a ForkJoinPools may also be appropriate for
- * use with fine-grained tasks that are never joined. Otherwise, other
- * ExecutorService implementations are typically more appropriate
- * choices.
+ *
A static {@link #commonPool} is available and appropriate for
+ * most applications. The common pool is used by any ForkJoinTask that
+ * is not explicitly submitted to a specified pool. Using the common
+ * pool normally reduces resource usage (its threads are slowly
+ * reclaimed during periods of non-use, and reinstated upon subsequent
+ * use).
*
- *
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} interface enables extension of
- * the kinds of synchronization accommodated. The target parallelism
- * level may also be changed dynamically ({@code setParallelism})
- * and thread construction can be limited using methods
- * {@code setMaximumPoolSize} and/or
- * {@code setMaintainsParallelism}.
+ *
For applications that require separate or custom pools, a {@code
+ * ForkJoinPool} may be 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.
*
*
In addition to execution and lifecycle control methods, this
* class provides status check methods (for example
- * {@code getStealCount}) 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} returns indications of pool state in a
+ * {@link #toString} returns indications of pool state in a
* convenient form for informal monitoring.
*
+ *
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.
+ *
+ *
+ *
+ * |
+ * Call from non-fork/join clients |
+ * Call from within fork/join computations |
+ *
+ *
+ * Arrange async execution |
+ * {@link #execute(ForkJoinTask)} |
+ * {@link ForkJoinTask#fork} |
+ *
+ *
+ * Await and obtain result |
+ * {@link #invoke(ForkJoinTask)} |
+ * {@link ForkJoinTask#invoke} |
+ *
+ *
+ * Arrange exec and obtain Future |
+ * {@link #submit(ForkJoinTask)} |
+ * {@link ForkJoinTask#fork} (ForkJoinTasks are Futures) |
+ *
+ *
+ *
+ * The common pool is by default constructed with default
+ * parameters, but these may be controlled by setting three {@link
+ * System#getProperty properties} with prefix {@code
+ * java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
+ * an integer greater than zero, {@code threadFactory} -- the class
+ * name of a {@link ForkJoinWorkerThreadFactory}, and {@code
+ * exceptionHandler} -- the class name of a {@link
+ * java.lang.Thread.UncaughtExceptionHandler
+ * Thread.UncaughtExceptionHandler}. Upon any error in establishing
+ * these settings, default parameters are used.
+ *
*
Implementation notes: 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}.
+ *
+ *
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
@@ -70,28 +123,400 @@ import java.util.concurrent.atomic.Atomi
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 is why we normally use
+ * method pollAt and its variants that try once at the apparent
+ * base index, else consider alternative actions, rather than
+ * method poll.)
+ *
+ * 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 randomly 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 are
+ * restricted to executing local tasks that they submitted (or in
+ * the case of CountedCompleters, others with the same root task).
+ * However, because most shared/external queue operations are more
+ * expensive than internal, and because, at steady state, external
+ * submitters will compete for CPU with workers, ForkJoinTask.join
+ * and related methods disable them from repeatedly helping to
+ * process tasks if all workers are active. 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 qlock), because submitters encountering a busy queue move
+ * on to try or create other queues -- they block only when
+ * creating and registering new queues.
+ *
+ * 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 "plock" is a form of sequence lock with a saturating
+ * shutdown bit (similarly for per-queue "qlocks"), mainly
+ * protecting updates to the workQueues array, as well as to
+ * enable shutdown. When used as a lock, it is normally only very
+ * briefly held, so is nearly always available after at most a
+ * brief spin, but we use a monitor-based backup strategy to
+ * blocking when needed.
+ *
+ * Recording WorkQueues. WorkQueues are recorded in the
+ * "workQueues" array that is created upon first use 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. Worker queues are at odd indices Shared
+ * (submission) queues are at even indices, up to a maximum of 64
+ * slots, to limit growth even if array needs to expand to add
+ * more workers. Grouping them together in this way simplifies and
+ * speeds up task scanning.
+ *
+ * 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. However, many other threads may notice the same task
+ * and each signal to wake up a thread that might take it. So in
+ * general, pools will be over-signalled. When a submission is
+ * added or another worker adds a task to a queue that is
+ * apparently empty, 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 whenever other threads scan for work or do not have a
+ * task to process. On most platforms, signalling (unpark)
+ * overhead time is noticeably long, and the time between
+ * signalling a thread and it actually making progress can be very
+ * noticeably long, so it is worth offloading these delays from
+ * critical paths as much as possible.
+ *
+ * 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 a
+ * given period -- a short period if there are more threads than
+ * parallelism, longer as the number of threads decreases. This
+ * will slowly propagate, eventually terminating all workers after
+ * periods of non-use.
+ *
+ * Shutdown and Termination. A call to shutdownNow atomically sets
+ * a plock bit and then (non-atomically) sets each worker's
+ * qlock 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) 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) and fall back to suspending the
+ * worker and if necessary replacing it with another.
+ *
+ * Helping actions for CountedCompleters are much simpler: Method
+ * helpComplete can take and execute any task with the same root
+ * as the task being waited on. However, this still entails some
+ * traversal of completer chains, so is less efficient than using
+ * CountedCompleters without explicit joins.
+ *
+ * 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. Compensation in the apparent absence of
+ * helping opportunities is challenging to control on JVMs, where
+ * GC and other activities can stall progress of tasks that in
+ * turn stall out many other dependent tasks, without us being
+ * able to determine whether they will ever require compensation.
+ * Even though work-stealing otherwise encounters little
+ * degradation in the presence of more threads than cores,
+ * aggressively adding new threads in such cases entails risk of
+ * unwanted positive feedback control loops in which more threads
+ * cause more dependent stalls (as well as delayed progress of
+ * unblocked threads to the point that we know they are available)
+ * leading to more situations requiring more threads, and so
+ * on. This aspect of control can be seen as an (analytically
+ * intractable) game with an opponent that may choose the worst
+ * (for us) active thread to stall at any time. We take several
+ * precautions to bound losses (and thus bound gains), mainly in
+ * methods tryCompensate and awaitJoin.
+ *
+ * Common Pool
+ * ===========
+ *
+ * The static commonPool always exists after static
+ * initialization. Since it (or any other created pool) need
+ * never be used, we minimize initial construction overhead and
+ * footprint to the setup of about a dozen fields, with no nested
+ * allocation. Most bootstrapping occurs within method
+ * fullExternalPush during the first submission to the pool.
+ *
+ * When external threads submit to the common pool, they can
+ * perform some subtask processing (see externalHelpJoin and
+ * related methods). We do not need to record whether these
+ * submissions are to the common pool -- if not, externalHelpJoin
+ * returns quickly (at the most helping to signal some common pool
+ * workers). These submitters would otherwise be blocked waiting
+ * for completion, so the extra effort (with liberally sprinkled
+ * task status checks) in inapplicable cases amounts to an odd
+ * form of limited spin-wait before blocking in ForkJoinTask.join.
+ *
+ * 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 (including
+ * several unnecessary-looking hoisted null checks) 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
*/
- /** Mask for packing and unpacking shorts */
- private static final int shortMask = 0xffff;
+ // Static utilities
- /** Max pool size -- must be a power of two minus 1 */
- private static final int MAX_THREADS = 0x7FFF;
+ /**
+ * 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);
+ }
+
+ // 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);
}
@@ -100,251 +525,1861 @@ public class ForkJoinPool extends Abstra
* Default ForkJoinWorkerThreadFactory implementation; creates a
* new ForkJoinWorkerThread.
*/
- static class DefaultForkJoinWorkerThreadFactory
+ static class DefaultForkJoinWorkerThreadFactory
implements ForkJoinWorkerThreadFactory {
public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
+ return new ForkJoinWorkerThread(pool);
+ }
+ }
+
+ /**
+ * 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.
+ */
+ static final class EmptyTask extends ForkJoinTask {
+ private static final long serialVersionUID = -7721805057305804111L;
+ EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
+ public final Void getRawResult() { return null; }
+ public final void setRawResult(Void x) {}
+ public final boolean exec() { return true; }
+ }
+
+ /**
+ * 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 external/shared push, 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 lock ("qlock") is
+ * forced to -1 on termination, causing all further lock attempts
+ * to fail. (Note: we don't need CAS for termination state because
+ * upon pool shutdown, all shared-queues will stop being used
+ * anyway.) Nearly all lock bodies are set up so that exceptions
+ * within lock bodies are "impossible" (modulo JVM errors that
+ * would cause failure anyway.)
+ *
+ * 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. 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 by workers before use. Others are
+ * allocated on first use.
+ */
+ static final class WorkQueue {
+ /**
+ * Capacity of work-stealing queue array upon initialization.
+ * Must be a power of two; at least 4, but should be larger to
+ * reduce or eliminate cacheline sharing among queues.
+ * Currently, it is much larger, as a partial workaround for
+ * the fact that JVMs often place arrays in locations that
+ * share GC bookkeeping (especially cardmarks) such that
+ * per-write accesses encounter serious memory contention.
+ */
+ static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
+
+ /**
+ * 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
+
+ int seed; // for random scanning; initialize nonzero
+ volatile int eventCount; // encoded inactivation count; < 0 if inactive
+ int nextWait; // encoded record of next event waiter
+ final int mode; // lifo, fifo, or shared
+ int nsteals; // cumulative number of steals
+ int poolIndex; // index of this queue in pool (or 0)
+ int stealHint; // index of most recent known stealer
+ volatile int qlock; // 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 ForkJoinPool pool; // the containing pool (may be null)
+ final ForkJoinWorkerThread owner; // owning thread or null if shared
+ volatile Thread parker; // == owner during call to park; else null
+ volatile 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;
+ Object p08, p09, p0a, p0b, p0c, p0d, p0e;
+
+ WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
+ this.mode = mode;
+ this.pool = pool;
+ this.owner = owner;
+ // Place indices in the center of array (that is not yet allocated)
+ base = top = INITIAL_QUEUE_CAPACITY >>> 1;
+ }
+
+ /**
+ * Pushes a task. Call only by owner in unshared queues.
+ * Cases needing resizing or rejection are relayed to fullPush
+ * (that also handles shared queues).
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @throw RejectedExecutionException if array cannot be resized
+ */
+ final void push(ForkJoinTask> task) {
+ ForkJoinPool p; ForkJoinTask>[] a;
+ int s = top, n;
+ if ((a = array) != null && a.length > (n = s + 1 - base)) {
+ U.putOrderedObject
+ (a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
+ top = s + 1;
+ if (n <= 1 && (p = pool) != null)
+ p.signalWork(this, 1);
+ }
+ else
+ fullPush(task, true);
+ }
+
+ /**
+ * Pushes a task if lock is free and array is either big
+ * enough or can be resized to be big enough. Note: a
+ * specialization of a common fast path of this method is in
+ * ForkJoinPool.externalPush. When called from a FJWT queue,
+ * this can fail only if the pool has been shut down or
+ * an out of memory error.
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @param owned if true, throw RJE on failure
+ */
+ final boolean fullPush(ForkJoinTask> task, boolean owned) {
+ ForkJoinPool p; ForkJoinTask>[] a;
+ if (owned) {
+ if (qlock < 0) // must be shutting down
+ throw new RejectedExecutionException();
+ }
+ else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
+ return false;
try {
- return new ForkJoinWorkerThread(pool);
- } catch (OutOfMemoryError oom) {
+ int s = top, oldLen, len;
+ if ((a = array) == null)
+ a = array = new ForkJoinTask>[len=INITIAL_QUEUE_CAPACITY];
+ else if ((oldLen = a.length) > s + 1 - base)
+ len = oldLen;
+ else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
+ throw new RejectedExecutionException("Capacity exceeded");
+ else {
+ int oldMask, b;
+ ForkJoinTask>[] oldA = a;
+ a = array = new ForkJoinTask>[len];
+ if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
+ int mask = len - 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 != s);
+ }
+ }
+ U.putOrderedObject
+ (a, (((len - 1) & s) << ASHIFT) + ABASE, task);
+ top = s + 1;
+ } finally {
+ if (!owned)
+ qlock = 0;
+ }
+ if ((p = pool) != null)
+ p.signalWork(this, 1);
+ return true;
+ }
+
+ /**
+ * Takes next task, if one exists, in LIFO order. Call only
+ * by owner in unshared queues.
+ */
+ final ForkJoinTask> pop() {
+ ForkJoinTask>[] a; ForkJoinTask> t; int m;
+ if ((a = array) != null && (m = a.length - 1) >= 0) {
+ for (int s; (s = top - 1) - base >= 0;) {
+ long j = ((m & s) << ASHIFT) + ABASE;
+ if ((t = (ForkJoinTask>)U.getObject(a, j)) == null)
+ break;
+ if (U.compareAndSwapObject(a, j, t, null)) {
+ top = s;
+ return t;
+ }
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Takes a task in FIFO order if b is base of queue and a task
+ * can be claimed without contention. Specialized versions
+ * appear in ForkJoinPool methods scan and tryHelpStealer.
+ */
+ 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;
+ }
+
+ /**
+ * 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;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t != null) {
+ if (base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ return t;
+ }
+ }
+ else if (base == b) {
+ if (b + 1 == top)
+ break;
+ Thread.yield(); // wait for lagging update (very rare)
+ }
+ }
+ 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);
+ }
+
+ /**
+ * Pops the given task only if it is at the current top.
+ * (A shared version is available only via FJP.tryExternalUnpush)
+ */
+ 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;
+ }
+
+ /**
+ * 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 its 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;
+ }
+
+ /**
+ * Provides a more accurate estimate of size than (top - base)
+ * by ordering reads and checking whether a near-empty queue
+ * has at least one unclaimed task.
+ */
+ final int queueSize() {
+ ForkJoinTask>[] a; int k, s, n;
+ return ((n = base - (s = top)) < 0 &&
+ (n != -1 ||
+ ((a = array) != null && (k = a.length) > 0 &&
+ U.getObject
+ (a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
+ -n : 0;
+ }
+
+ // Specialized execution methods
+
+ /**
+ * Pops and runs tasks until empty.
+ */
+ private void popAndExecAll() {
+ // A bit faster than repeated pop calls
+ ForkJoinTask>[] a; int m, s; long j; ForkJoinTask> t;
+ while ((a = array) != null && (m = a.length - 1) >= 0 &&
+ (s = top - 1) - base >= 0 &&
+ (t = ((ForkJoinTask>)
+ U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
+ != null) {
+ if (U.compareAndSwapObject(a, j, t, null)) {
+ top = s;
+ t.doExec();
+ }
}
}
+
+ /**
+ * Polls and runs tasks until empty.
+ */
+ private void pollAndExecAll() {
+ for (ForkJoinTask> t; (t = poll()) != null;)
+ t.doExec();
+ }
+
+ /**
+ * If present, removes from queue and executes the given task,
+ * or any other cancelled task. Returns (true) on any CAS
+ * or consistency check failure so caller can retry.
+ *
+ * @return false if no progress can be made, else true;
+ */
+ final boolean tryRemoveAndExec(ForkJoinTask> task) {
+ boolean stat = true, removed = false, empty = 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)
+ stat = false;
+ break;
+ }
+ }
+ }
+ if (removed)
+ task.doExec();
+ return stat;
+ }
+
+ /**
+ * Polls for and executes the given task or any other task in
+ * its CountedCompleter computation
+ */
+ final boolean pollAndExecCC(ForkJoinTask> root) {
+ ForkJoinTask>[] a; int b; Object o;
+ outer: while ((b = base) - top < 0 && (a = array) != null) {
+ long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ if ((o = U.getObject(a, j)) == null ||
+ !(o instanceof CountedCompleter))
+ break;
+ for (CountedCompleter> t = (CountedCompleter>)o, r = t;;) {
+ if (r == root) {
+ if (base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ t.doExec();
+ return true;
+ }
+ else
+ break; // restart
+ }
+ if ((r = r.completer) == null)
+ break outer; // not part of root computation
+ }
+ }
+ return false;
+ }
+
+ /**
+ * Executes a top-level task and any local tasks remaining
+ * after execution.
+ */
+ final void runTask(ForkJoinTask> t) {
+ if (t != null) {
+ (currentSteal = t).doExec();
+ currentSteal = null;
+ if (++nsteals < 0) { // spill on overflow
+ ForkJoinPool p;
+ if ((p = pool) != null)
+ p.collectStealCount(this);
+ }
+ if (top != base) { // process remaining local tasks
+ if (mode == 0)
+ popAndExecAll();
+ else
+ pollAndExecAll();
+ }
+ }
+ }
+
+ /**
+ * Executes a non-top-level (stolen) task.
+ */
+ final void runSubtask(ForkJoinTask> t) {
+ if (t != null) {
+ ForkJoinTask> ps = currentSteal;
+ (currentSteal = 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 QLOCK;
+ private static final int ABASE;
+ private static final int ASHIFT;
+ static {
+ int s;
+ try {
+ U = getUnsafe();
+ Class> k = WorkQueue.class;
+ Class> ak = ForkJoinTask[].class;
+ QLOCK = U.objectFieldOffset
+ (k.getDeclaredField("qlock"));
+ 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);
+ }
}
/**
+ * 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 externalPush. In the future, this may
+ * also incorporate a means to implement different task rejection
+ * and resubmission policies.
+ *
+ * Seeds for submitters and workers/workQueues work in basically
+ * the same way but are initialized and updated using slightly
+ * different mechanics. Both are initialized using the same
+ * approach as in class ThreadLocal, where successive values are
+ * unlikely to collide with previous values. Seeds are then
+ * randomly modified upon collisions using xorshifts, which
+ * requires a non-zero seed.
+ */
+ static final class Submitter {
+ int seed;
+ Submitter(int s) { seed = s; }
+ }
+
+ /** Property prefix for constructing common pool */
+ private static final String propPrefix =
+ "java.util.concurrent.ForkJoinPool.common.";
+
+ // static fields (initialized in static initializer below)
+
+ /**
* Creates a new ForkJoinWorkerThread. This factory is used unless
* overridden in ForkJoinPool constructors.
*/
public static final ForkJoinWorkerThreadFactory
- defaultForkJoinWorkerThreadFactory =
- new DefaultForkJoinWorkerThreadFactory();
+ defaultForkJoinWorkerThreadFactory;
+
+ /**
+ * Common (static) pool. Non-null for public use unless a static
+ * construction exception, but internal usages null-check on use
+ * to paranoically avoid potential initialization circularities
+ * as well as to simplify generated code.
+ */
+ static final ForkJoinPool commonPool;
/**
* Permission required for callers of methods that may start or
* kill threads.
*/
- private static final RuntimePermission modifyThreadPermission =
- new RuntimePermission("modifyThread");
+ private static final RuntimePermission modifyThreadPermission;
/**
- * If there is a security manager, makes sure caller has
- * permission to modify threads.
+ * Per-thread submission bookkeeping. Shared across all pools
+ * to reduce ThreadLocal pollution and because random motion
+ * to avoid contention in one pool is likely to hold for others.
+ * Lazily initialized on first submission (but null-checked
+ * in other contexts to avoid unnecessary initialization).
*/
- private static void checkPermission() {
- SecurityManager security = System.getSecurityManager();
- if (security != null)
- security.checkPermission(modifyThreadPermission);
- }
+ static final ThreadLocal submitters;
/**
- * Generator for assigning sequence numbers as pool names.
+ * Common pool parallelism. Must equal commonPool.parallelism.
*/
- private static final AtomicInteger poolNumberGenerator =
- new AtomicInteger();
+ static final int commonPoolParallelism;
/**
- * Array holding all worker threads in the pool. Initialized upon
- * first use. 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.
+ * Sequence number for creating workerNamePrefix.
*/
- volatile ForkJoinWorkerThread[] workers;
+ private static int poolNumberSequence;
/**
- * Lock protecting access to workers.
+ * Return the next sequence number. We don't expect this to
+ * ever contend so use simple builtin sync.
*/
- private final ReentrantLock workerLock;
+ private static final synchronized int nextPoolId() {
+ return ++poolNumberSequence;
+ }
+
+ // static constants
/**
- * Condition for awaitTermination.
+ * Initial timeout value (in nanoseconds) for the thread
+ * triggering quiescence to park waiting for new work. On timeout,
+ * the thread will instead try to shrink the number of
+ * workers. The value should be large enough to avoid overly
+ * aggressive shrinkage during most transient stalls (long GCs
+ * etc).
*/
- private final Condition termination;
+ private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
/**
- * The uncaught exception handler used when any worker
- * abruptly terminates
+ * Timeout value when there are more threads than parallelism level
*/
- private Thread.UncaughtExceptionHandler ueh;
+ private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
/**
- * Creation factory for worker threads.
+ * The maximum stolen->joining link depth allowed in method
+ * tryHelpStealer. Must be a power of two. 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 final ForkJoinWorkerThreadFactory factory;
+ private static final int MAX_HELP = 64;
/**
- * Head of stack of threads that were created to maintain
- * parallelism when other threads blocked, but have since
- * suspended when the parallelism level rose.
+ * Increment for seed generators. See class ThreadLocal for
+ * explanation.
*/
- private volatile WaitQueueNode spareStack;
+ private static final int SEED_INCREMENT = 0x61c88647;
/**
- * Sum of per-thread steal counts, updated only when threads are
- * idle or terminating.
+ * 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 plock is an int packed with:
+ * SHUTDOWN: true if shutdown is enabled (1 bit)
+ * SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
+ * SIGNAL: set when threads may be waiting on the lock (1 bit)
+ *
+ * The sequence number enables simple consistency checks:
+ * Staleness of read-only operations on the workQueues array can
+ * be checked by comparing plock before vs after the reads.
*/
- private final AtomicLong stealCount;
- /**
- * Queue for external submissions.
+ // 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 SMASK = 0xffff; // short bits
+ private static final int MAX_CAP = 0x7fff; // max #workers - 1
+ private static final int EVENMASK = 0xfffe; // even short bits
+ private static final int SQMASK = 0x007e; // max 64 (even) slots
+ 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;
+
+ // plock bits
+ private static final int SHUTDOWN = 1 << 31;
+ private static final int PL_LOCK = 2;
+ private static final int PL_SIGNAL = 1;
+ private static final int PL_SPINS = 1 << 8;
+
+ // access mode for WorkQueue
+ static final int LIFO_QUEUE = 0;
+ static final int FIFO_QUEUE = 1;
+ static final int SHARED_QUEUE = -1;
+
+ // Instance fields
+
+ /*
+ * 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 stealCount; // collects worker counts
+ volatile long ctl; // main pool control
+ final int parallelism; // parallelism level
+ final int localMode; // per-worker scheduling mode
+ volatile int indexSeed; // worker/submitter index seed
+ volatile int plock; // shutdown status and seqLock
+ WorkQueue[] workQueues; // main registry
+ final ForkJoinWorkerThreadFactory factory; // factory for new workers
+ final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
+ final String workerNamePrefix; // to create worker name string
+
+ /*
+ * Acquires the plock lock to protect worker array and related
+ * updates. This method is called only if an initial CAS on plock
+ * fails. This acts as a spinLock for normal cases, but falls back
+ * to builtin monitor to block when (rarely) needed. This would be
+ * a terrible idea for a highly contended lock, but works fine as
+ * a more conservative alternative to a pure spinlock. See
+ * internal ConcurrentHashMap documentation for further
+ * explanation of nearly the same construction.
*/
- private final LinkedTransferQueue> submissionQueue;
+ private int acquirePlock() {
+ int spins = PL_SPINS, r = 0, ps, nps;
+ for (;;) {
+ if (((ps = plock) & PL_LOCK) == 0 &&
+ U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
+ return nps;
+ else if (r == 0)
+ r = ThreadLocalRandom.current().nextInt(); // randomize spins
+ else if (spins >= 0) {
+ r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
+ if (r >= 0)
+ --spins;
+ }
+ else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
+ synchronized (this) {
+ if ((plock & PL_SIGNAL) != 0) {
+ try {
+ wait();
+ } catch (InterruptedException ie) {
+ try {
+ Thread.currentThread().interrupt();
+ } catch (SecurityException ignore) {
+ }
+ }
+ }
+ else
+ notifyAll();
+ }
+ }
+ }
+ }
/**
- * Head of Treiber stack for barrier sync. See below for explanation.
+ * Unlocks and signals any thread waiting for plock. Called only
+ * when CAS of seq value for unlock fails.
*/
- private volatile WaitQueueNode syncStack;
+ private void releasePlock(int ps) {
+ plock = ps;
+ synchronized (this) { notifyAll(); }
+ }
+
+ // Registering and deregistering workers
+
+ /**
+ * Callback from ForkJoinWorkerThread constructor to establish its
+ * poolIndex and record its WorkQueue. To avoid scanning bias due
+ * to packing entries in front of the workQueues array, we treat
+ * the array as a simple power-of-two hash table using per-thread
+ * seed as hash, expanding as needed.
+ *
+ * @param w the worker's queue
+ */
+ final void registerWorker(WorkQueue w) {
+ int s, ps; // generate a rarely colliding candidate index seed
+ do {} while (!U.compareAndSwapInt(this, INDEXSEED,
+ s = indexSeed, s += SEED_INCREMENT) ||
+ s == 0); // skip 0
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ try {
+ WorkQueue[] ws;
+ if (w != null && (ws = workQueues) != null) {
+ w.seed = s;
+ int n = ws.length, m = n - 1;
+ int r = (s << 1) | 1; // use odd-numbered indices
+ if (ws[r &= m] != null) { // collision
+ int probes = 0; // step by approx half size
+ int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
+ while (ws[r = (r + step) & m] != null) {
+ if (++probes >= n) {
+ workQueues = ws = Arrays.copyOf(ws, n <<= 1);
+ m = n - 1;
+ probes = 0;
+ }
+ }
+ }
+ w.eventCount = w.poolIndex = r; // establish before recording
+ ws[r] = w;
+ }
+ } finally {
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ }
/**
- * The count for event barrier
- */
- private volatile long eventCount;
+ * Final callback from terminating worker, as well as upon failure
+ * to construct or start a worker. 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 construction failed
+ * @param ex the exception causing failure, or null if none
+ */
+ final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
+ WorkQueue w = null;
+ if (wt != null && (w = wt.workQueue) != null) {
+ int ps;
+ collectStealCount(w);
+ w.qlock = -1; // ensure set
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ try {
+ int idx = w.poolIndex;
+ WorkQueue[] ws = workQueues;
+ if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
+ ws[idx] = null;
+ } finally {
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ }
+
+ 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)))));
+
+ if (!tryTerminate(false, false) && w != null) {
+ w.cancelAll(); // cancel remaining tasks
+ if (w.array != null) // suppress signal if never ran
+ signalWork(null, 1); // wake up or create replacement
+ if (ex == null) // help clean refs on way out
+ ForkJoinTask.helpExpungeStaleExceptions();
+ }
+
+ if (ex != null) // rethrow
+ ForkJoinTask.rethrow(ex);
+ }
/**
- * Pool number, just for assigning useful names to worker threads
+ * Collect worker steal count into total. Called on termination
+ * and upon int overflow of local count. (There is a possible race
+ * in the latter case vs any caller of getStealCount, which can
+ * make its results less accurate than usual.)
*/
- private final int poolNumber;
+ final void collectStealCount(WorkQueue w) {
+ if (w != null) {
+ long sc;
+ int ns = w.nsteals;
+ w.nsteals = 0; // handle overflow
+ long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
+ do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
+ sc = stealCount, sc + steals));
+ }
+ }
+
+ // Submissions
+
+ /**
+ * Unless shutting down, adds the given task to a submission queue
+ * at submitter's current queue index (modulo submission
+ * range). Only the most common path is directly handled in this
+ * method. All others are relayed to fullExternalPush.
+ *
+ * @param task the task. Caller must ensure non-null.
+ */
+ final void externalPush(ForkJoinTask> task) {
+ WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask>[] a;
+ if ((z = submitters.get()) != null && plock > 0 &&
+ (ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
+ int s = q.top, n;
+ if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
+ U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
+ task);
+ q.top = s + 1; // push on to deque
+ q.qlock = 0;
+ if (n <= 1)
+ signalWork(q, 1);
+ return;
+ }
+ q.qlock = 0;
+ }
+ fullExternalPush(task);
+ }
+
+ /**
+ * Full version of externalPush. This method is called, among
+ * other times, upon the first submission of the first task to the
+ * pool, so must perform secondary initialization: creating
+ * workQueue array and setting plock to a valid value. It also
+ * detects first submission by an external thread by looking up
+ * its ThreadLocal, and creates a new shared queue if the one at
+ * index if empty or contended. The lock bodies must be
+ * exception-free (so no try/finally) so we optimistically
+ * allocate new queues/arrays outside the locks and throw them
+ * away if (very rarely) not needed. Note that the plock seq value
+ * can eventually wrap around zero, but if so harmlessly fails to
+ * reinitialize.
+ */
+ private void fullExternalPush(ForkJoinTask> task) {
+ for (Submitter z = null;;) {
+ WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
+ if ((ps = plock) < 0)
+ throw new RejectedExecutionException();
+ else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) {
+ int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2;
+ n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
+ WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
+ if ((ps & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ if ((ws = workQueues) == null)
+ workQueues = nws;
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ else if (z == null && (z = submitters.get()) == null) {
+ if (U.compareAndSwapInt(this, INDEXSEED,
+ s = indexSeed, s += SEED_INCREMENT) &&
+ s != 0) // skip 0
+ submitters.set(z = new Submitter(s));
+ }
+ else {
+ int k = (r = z.seed) & m & SQMASK;
+ if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
+ (q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ WorkQueue w = null;
+ if ((ws = workQueues) != null && k < ws.length &&
+ (w = ws[k]) == null)
+ ws[k] = q;
+ else
+ q = w;
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ if (q != null && q.qlock == 0 && q.fullPush(task, false))
+ return;
+ r ^= r << 13; // same xorshift as WorkQueues
+ r ^= r >>> 17;
+ z.seed = r ^= r << 5; // move to a different index
+ }
+ }
+ }
+
+ // Maintaining ctl counts
/**
- * The maximum allowed pool size
+ * Increments active count; mainly called upon return from blocking.
*/
- private volatile int maxPoolSize;
+ final void incrementActiveCount() {
+ long c;
+ do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
+ }
/**
- * The desired parallelism level, updated only under workerLock.
- */
- private volatile int parallelism;
+ * Tries to create (at most one) or activate (possibly several)
+ * workers if too few are active. On contention failure, continues
+ * until at least one worker is signalled or the given queue is
+ * empty or all workers are active.
+ *
+ * @param q if non-null, the queue holding tasks to be signalled
+ * @param signals the target number of signals.
+ */
+ final void signalWork(WorkQueue q, int signals) {
+ long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
+ while ((u = (int)((c = ctl) >>> 32)) < 0) {
+ if ((e = (int)c) > 0) {
+ if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
+ (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);
+ if (--signals <= 0)
+ break;
+ }
+ else
+ signals = 1;
+ if ((q != null && q.queueSize() == 0))
+ break;
+ }
+ else
+ break;
+ }
+ else if (e == 0 && (u & SHORT_SIGN) != 0) {
+ long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
+ ((u + UAC_UNIT) & UAC_MASK)) << 32;
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ ForkJoinWorkerThread wt = null;
+ Throwable ex = null;
+ boolean started = false;
+ try {
+ ForkJoinWorkerThreadFactory fac;
+ if ((fac = factory) != null &&
+ (wt = fac.newThread(this)) != null) {
+ wt.start();
+ started = true;
+ }
+ } catch (Throwable rex) {
+ ex = rex;
+ }
+ if (!started)
+ deregisterWorker(wt, ex); // adjust counts on failure
+ break;
+ }
+ }
+ else
+ break;
+ }
+ }
+
+ // Scanning for tasks
/**
- * True if use local fifo, not default lifo, for local polling
+ * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
*/
- private volatile boolean locallyFifo;
+ final void runWorker(WorkQueue w) {
+ // initialize queue array in this thread
+ w.array = new ForkJoinTask>[WorkQueue.INITIAL_QUEUE_CAPACITY];
+ do { w.runTask(scan(w)); } while (w.qlock >= 0);
+ }
/**
- * 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 assume that running active count
- * is in low word, so need to be modified if this changes.
+ * 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 a random permutation of
+ * queues (starting at a random index and stepping by a random
+ * relative prime, checking each at least once). The scan
+ * terminates upon either finding a non-empty queue, or completing
+ * the sweep. If the worker is not inactivated, it takes and
+ * returns a task from this queue. Otherwise, if not activated, it
+ * signals workers (that may include itself) and returns so caller
+ * can retry. Also returns for trtry if the worker array may have
+ * changed during an empty scan. 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 already enqueued, try to inactivate and enqueue the
+ * worker on wait queue. Or, if inactivating has caused the pool
+ * to be quiescent, relay to idleAwaitWork to check for
+ * termination and possibly shrink pool.
+ *
+ * * If already enqueued and none of the above apply, possibly
+ * (with 1/2 probability) park awaiting signal, else lingering to
+ * help scan and signal.
+ *
+ * @param w the worker (via its WorkQueue)
+ * @return a task or null if none found
+ */
+ private final ForkJoinTask> scan(WorkQueue w) {
+ WorkQueue[] ws; WorkQueue q; // first update random seed
+ int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
+ int ps = plock, m; // volatile read order matters
+ if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
+ int ec = w.eventCount; // ec is negative if inactive
+ int step = (r >>> 16) | 1; // relatively prime
+ for (int j = (m + 1) << 2; ; --j, r += step) {
+ ForkJoinTask> t; ForkJoinTask>[] a; int b, n;
+ if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
+ (a = q.array) != null) { // probably nonempty
+ int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, i);
+ if (q.base == b && ec >= 0 && t != null &&
+ U.compareAndSwapObject(a, i, t, null)) {
+ if ((n = q.top - (q.base = b + 1)) > 0)
+ signalWork(q, n);
+ return t; // taken
+ }
+ if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) {
+ if ((n = q.queueSize() - 1) > 0)
+ signalWork(q, n);
+ break; // let caller retry after signal
+ }
+ }
+ else if (j < 0) { // end of scan
+ long c = ctl; int e;
+ if (plock != ps) // incomplete sweep
+ break;
+ if ((e = (int)c) < 0) // pool is terminating
+ w.qlock = -1;
+ else if (ec >= 0) { // try to enqueue/inactivate
+ long nc = ((long)ec |
+ ((c - AC_UNIT) & (AC_MASK|TC_MASK)));
+ w.nextWait = e;
+ w.eventCount = ec | INT_SIGN; // mark as inactive
+ if (ctl != c ||
+ !U.compareAndSwapLong(this, CTL, c, nc))
+ w.eventCount = ec; // unmark on CAS failure
+ else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
+ idleAwaitWork(w, nc, c); // quiescent
+ }
+ else if (w.seed >= 0 && w.eventCount < 0) {
+ Thread wt = Thread.currentThread();
+ Thread.interrupted(); // clear status
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt; // emulate LockSupport.park
+ if (w.eventCount < 0) // recheck
+ U.park(false, 0L);
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ }
+ break;
+ }
+ }
+ }
+ return null;
+ }
+
+ /**
+ * If inactivating worker w has caused the pool to become
+ * quiescent, checks for pool termination, and, so long as this is
+ * not the only worker, waits for event for up to a given
+ * duration. 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
+ * @param currentCtl the ctl value triggering possible quiescence
+ * @param prevCtl the ctl value to restore if thread is terminated
+ */
+ private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
+ if (w.eventCount < 0 &&
+ (this == commonPool || !tryTerminate(false, false)) &&
+ (int)prevCtl != 0) {
+ int dc = -(short)(currentCtl >>> TC_SHIFT);
+ long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
+ long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
+ Thread wt = Thread.currentThread();
+ while (ctl == currentCtl) {
+ Thread.interrupted(); // timed variant of version in scan()
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt;
+ if (ctl == currentCtl)
+ U.park(false, parkTime);
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ if (ctl != currentCtl)
+ break;
+ if (deadline - System.nanoTime() <= 0L &&
+ U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
+ w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
+ w.qlock = -1; // shrink
+ break;
+ }
+ }
+ }
+ }
+
+ /**
+ * Scans through queues looking for work while joining a task;
+ * if any are present, signals.
+ *
+ * @param task to return early if done
+ * @param origin an index to start scan
*/
- private volatile int workerCounts;
+ final int helpSignal(ForkJoinTask> task, int origin) {
+ WorkQueue[] ws; WorkQueue q; int m, n, s;
+ if (task != null && (ws = workQueues) != null &&
+ (m = ws.length - 1) >= 0) {
+ for (int i = 0; i <= m; ++i) {
+ if ((s = task.status) < 0)
+ return s;
+ if ((q = ws[(i + origin) & m]) != null &&
+ (n = q.queueSize()) > 0) {
+ signalWork(q, n);
+ if ((int)(ctl >> AC_SHIFT) >= 0)
+ break;
+ }
+ }
+ }
+ return 0;
+ }
- 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; }
+ /**
+ * 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 so long that they are likely cyclic.
+ *
+ * @param joiner the joining worker
+ * @param task the task to join
+ * @return 0 if no progress can be made, negative if task
+ * known complete, else positive
+ */
+ private int tryHelpStealer(WorkQueue joiner, ForkJoinTask> task) {
+ int stat = 0, steps = 0; // bound to avoid cycles
+ if (joiner != null && task != null) { // hoist null checks
+ restart: for (;;) {
+ ForkJoinTask> subtask = task; // current target
+ for (WorkQueue j = joiner, v;;) { // v is stealer of subtask
+ WorkQueue[] ws; int m, s, h;
+ if ((s = task.status) < 0) {
+ stat = s;
+ break restart;
+ }
+ if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
+ break restart; // shutting down
+ if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
+ v.currentSteal != subtask) {
+ for (int origin = h;;) { // find stealer
+ if (((h = (h + 2) & m) & 15) == 1 &&
+ (subtask.status < 0 || j.currentJoin != subtask))
+ continue restart; // occasional staleness check
+ if ((v = ws[h]) != null &&
+ v.currentSteal == subtask) {
+ j.stealHint = h; // save hint
+ break;
+ }
+ if (h == origin)
+ break restart; // cannot find stealer
+ }
+ }
+ for (;;) { // help stealer or descend to its stealer
+ ForkJoinTask[] a; int b;
+ if (subtask.status < 0) // surround probes with
+ continue restart; // consistency checks
+ if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
+ int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ ForkJoinTask> t =
+ (ForkJoinTask>)U.getObjectVolatile(a, i);
+ if (subtask.status < 0 || j.currentJoin != subtask ||
+ v.currentSteal != subtask)
+ continue restart; // stale
+ stat = 1; // apparent progress
+ if (t != null && v.base == b &&
+ U.compareAndSwapObject(a, i, t, null)) {
+ v.base = b + 1; // help stealer
+ joiner.runSubtask(t);
+ }
+ else if (v.base == b && ++steps == MAX_HELP)
+ break restart; // v apparently stalled
+ }
+ else { // empty -- try to descend
+ ForkJoinTask> next = v.currentJoin;
+ if (subtask.status < 0 || j.currentJoin != subtask ||
+ v.currentSteal != subtask)
+ continue restart; // stale
+ else if (next == null || ++steps == MAX_HELP)
+ break restart; // dead-end or maybe cyclic
+ else {
+ subtask = next;
+ j = v;
+ break;
+ }
+ }
+ }
+ }
+ }
+ }
+ return stat;
+ }
/**
- * Adds 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).
+ * Analog of tryHelpStealer for CountedCompleters. Tries to steal
+ * and run tasks within the target's computation
+ *
+ * @param task the task to join
+ * @param mode if shared, exit upon completing any task
+ * if all workers are active
*
- * @param delta the number to add
*/
- final void updateRunningCount(int delta) {
- int s;
- do {} while (!casWorkerCounts(s = workerCounts, s + delta));
+ private int helpComplete(ForkJoinTask> task, int mode) {
+ WorkQueue[] ws; WorkQueue q; int m, n, s;
+ if (task != null && (ws = workQueues) != null &&
+ (m = ws.length - 1) >= 0) {
+ for (int j = 1, origin = j;;) {
+ if ((s = task.status) < 0)
+ return s;
+ if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
+ origin = j;
+ if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
+ break;
+ }
+ else if ((j = (j + 2) & m) == origin)
+ break;
+ }
+ }
+ return 0;
+ }
+
+ /**
+ * Tries to decrement active count (sometimes implicitly) and
+ * possibly release or create a compensating worker in preparation
+ * for blocking. Fails on contention or termination. Otherwise,
+ * adds a new thread if no idle workers are available and pool
+ * may become starved.
+ */
+ final boolean tryCompensate() {
+ int pc = parallelism, e, u, i, tc; long c;
+ WorkQueue[] ws; WorkQueue w; Thread p;
+ if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
+ if (e != 0 && (i = e & SMASK) < ws.length &&
+ (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
+ long nc = ((long)(w.nextWait & E_MASK) |
+ (c & (AC_MASK|TC_MASK)));
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ return true; // replace with idle worker
+ }
+ }
+ else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
+ (u >> UAC_SHIFT) + pc > 1) {
+ long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc))
+ return true; // no compensation
+ }
+ else if ((tc = u + pc) < MAX_CAP) {
+ long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ Throwable ex = null;
+ ForkJoinWorkerThread wt = null;
+ try {
+ ForkJoinWorkerThreadFactory fac;
+ if ((fac = factory) != null &&
+ (wt = fac.newThread(this)) != null) {
+ wt.start();
+ return true;
+ }
+ } catch (Throwable rex) {
+ ex = rex;
+ }
+ deregisterWorker(wt, ex); // adjust counts etc
+ }
+ }
+ }
+ return false;
}
/**
- * Adds delta (which may be negative) to both total and running
- * count. This must be called upon creation and termination of
- * worker threads.
+ * Helps and/or blocks until the given task is done.
*
- * @param delta the number to add
+ * @param joiner the joining worker
+ * @param task the task
+ * @return task status on exit
*/
- 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 int awaitJoin(WorkQueue joiner, ForkJoinTask> task) {
+ int s = 0;
+ if (joiner != null && task != null && (s = task.status) >= 0) {
+ ForkJoinTask> prevJoin = joiner.currentJoin;
+ joiner.currentJoin = task;
+ do {} while ((s = task.status) >= 0 &&
+ joiner.queueSize() > 0 &&
+ joiner.tryRemoveAndExec(task)); // process local tasks
+ if (s >= 0 && (s = task.status) >= 0 &&
+ (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
+ (task instanceof CountedCompleter))
+ s = helpComplete(task, LIFO_QUEUE);
+ while (s >= 0 && (s = task.status) >= 0) {
+ if ((joiner.queueSize() > 0 || // try helping
+ (s = tryHelpStealer(joiner, task)) == 0) &&
+ (s = task.status) >= 0 && tryCompensate()) {
+ if (task.trySetSignal() && (s = task.status) >= 0) {
+ synchronized (task) {
+ if (task.status >= 0) {
+ try { // see ForkJoinTask
+ task.wait(); // for explanation
+ } catch (InterruptedException ie) {
+ }
+ }
+ else
+ task.notifyAll();
+ }
+ }
+ long c; // re-activate
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ }
+ }
+ joiner.currentJoin = prevJoin;
+ }
+ return s;
}
/**
- * 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
+ * Stripped-down variant of awaitJoin used by timed joins. Tries
+ * to help join only while there is continuous progress. (Caller
+ * will then enter a timed wait.)
+ *
+ * @param joiner the joining worker
+ * @param task the task
*/
- private volatile int runControl;
+ final void helpJoinOnce(WorkQueue joiner, ForkJoinTask> task) {
+ int s;
+ if (joiner != null && task != null && (s = task.status) >= 0) {
+ ForkJoinTask> prevJoin = joiner.currentJoin;
+ joiner.currentJoin = task;
+ do {} while ((s = task.status) >= 0 &&
+ joiner.queueSize() > 0 &&
+ joiner.tryRemoveAndExec(task));
+ if (s >= 0 && (s = task.status) >= 0 &&
+ (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
+ (task instanceof CountedCompleter))
+ s = helpComplete(task, LIFO_QUEUE);
+ if (s >= 0 && joiner.queueSize() == 0) {
+ do {} while (task.status >= 0 &&
+ tryHelpStealer(joiner, task) > 0);
+ }
+ joiner.currentJoin = prevJoin;
+ }
+ }
- // 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;
+ /**
+ * Returns a (probably) 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.
+ * @param r a (random) seed for scanning
+ */
+ private WorkQueue findNonEmptyStealQueue(int r) {
+ int step = (r >>> 16) | 1;
+ for (WorkQueue[] ws;;) {
+ int ps = plock, m;
+ if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
+ return null;
+ for (int j = (m + 1) << 2; ; r += step) {
+ WorkQueue q = ws[((r << 1) | 1) & m];
+ if (q != null && q.queueSize() > 0)
+ return q;
+ else if (--j < 0) {
+ if (plock == ps)
+ return null;
+ break;
+ }
+ }
+ }
+ }
- 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; }
+ /**
+ * 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;;) {
+ ForkJoinTask> localTask; // exhaust local queue
+ while ((localTask = w.nextLocalTask()) != null)
+ localTask.doExec();
+ // Similar to loop in scan(), but ignoring submissions
+ WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
+ if (q != null) {
+ ForkJoinTask> t; int b;
+ if (!active) { // re-establish active count
+ long c;
+ active = true;
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ }
+ if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != 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;
+ }
+ }
+ }
+ }
/**
- * Tries incrementing active count; fails on contention.
- * Called by workers before/during executing tasks.
+ * Gets and removes a local or stolen task for the given worker.
*
- * @return true on success
+ * @return a task, if available
*/
- final boolean tryIncrementActiveCount() {
- int c = runControl;
- return casRunControl(c, c+1);
+ final ForkJoinTask> nextTaskFor(WorkQueue w) {
+ for (ForkJoinTask> t;;) {
+ WorkQueue q; int b;
+ if ((t = w.nextLocalTask()) != null)
+ return t;
+ if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
+ return null;
+ if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
+ return t;
+ }
}
/**
- * Tries decrementing active count; fails on contention.
- * Possibly triggers termination on success.
- * Called by workers when they can't find tasks.
- *
- * @return true on success
+ * Returns a cheap heuristic guide for task partitioning when
+ * programmers, frameworks, tools, or languages have little or no
+ * idea about task granularity. In essence by offering this
+ * method, we ask users only about tradeoffs in overhead vs
+ * expected throughput and its variance, rather than how finely to
+ * partition tasks.
+ *
+ * In a steady state strict (tree-structured) computation, each
+ * thread makes available for stealing enough tasks for other
+ * threads to remain active. Inductively, if all threads play by
+ * the same rules, each thread should make available only a
+ * constant number of tasks.
+ *
+ * The minimum useful constant is just 1. But using a value of 1
+ * would require immediate replenishment upon each steal to
+ * maintain enough tasks, which is infeasible. Further,
+ * partitionings/granularities of offered tasks should minimize
+ * steal rates, which in general means that threads nearer the top
+ * of computation tree should generate more than those nearer the
+ * bottom. In perfect steady state, each thread is at
+ * approximately the same level of computation tree. However,
+ * producing extra tasks amortizes the uncertainty of progress and
+ * diffusion assumptions.
+ *
+ * So, users will want to use values larger, but not much larger
+ * than 1 to both smooth over transient shortages and hedge
+ * against uneven progress; as traded off against the cost of
+ * extra task overhead. We leave the user to pick a threshold
+ * value to compare with the results of this call to guide
+ * decisions, but recommend values such as 3.
+ *
+ * When all threads are active, it is on average OK to estimate
+ * surplus strictly locally. In steady-state, if one thread is
+ * maintaining say 2 surplus tasks, then so are others. So we can
+ * just use estimated queue length. However, this strategy alone
+ * leads to serious mis-estimates in some non-steady-state
+ * conditions (ramp-up, ramp-down, other stalls). We can detect
+ * many of these by further considering the number of "idle"
+ * threads, that are known to have zero queued tasks, so
+ * compensate by a factor of (#idle/#active) threads.
+ *
+ * Note: The approximation of #busy workers as #active workers is
+ * not very good under current signalling scheme, and should be
+ * improved.
+ */
+ static int getSurplusQueuedTaskCount() {
+ Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
+ if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
+ int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
+ int p = (pool = wt.pool).parallelism;
+ int a = (int)(pool.ctl >> AC_SHIFT) + p;
+ return q.top - b - (a > (p >>>= 1) ? 0 :
+ a > (p >>>= 1) ? 1 :
+ a > (p >>>= 1) ? 2 :
+ a > (p >>>= 1) ? 4 :
+ 8);
+ }
+ return 0;
+ }
+
+ // Termination
+
+ /**
+ * 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
*/
- final boolean tryDecrementActiveCount() {
- int c = runControl;
- int nextc = c - 1;
- if (!casRunControl(c, nextc))
+ private boolean tryTerminate(boolean now, boolean enable) {
+ if (this == commonPool) // cannot shut down
return false;
- if (canTerminateOnShutdown(nextc))
- terminateOnShutdown();
- return true;
+ for (long c;;) {
+ if (((c = ctl) & STOP_BIT) != 0) { // already terminating
+ if ((short)(c >>> TC_SHIFT) == -parallelism) {
+ synchronized (this) {
+ notifyAll(); // signal when 0 workers
+ }
+ }
+ return true;
+ }
+ if (plock >= 0) { // not yet enabled
+ int ps;
+ if (!enable)
+ return false;
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = SHUTDOWN;
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ 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.qlock = -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.qlock = -1;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ }
+ }
+ }
+ }
+ }
+ }
}
+ // external operations on common pool
+
/**
- * Returns true if argument represents zero active count and
- * nonzero runstate, which is the triggering condition for
- * terminating on shutdown.
+ * Returns common pool queue for a thread that has submitted at
+ * least one task.
*/
- private static boolean canTerminateOnShutdown(int c) {
- // i.e. least bit is nonzero runState bit
- return ((c & -c) >>> 16) != 0;
+ static WorkQueue commonSubmitterQueue() {
+ ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
+ return ((z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0) ?
+ ws[m & z.seed & SQMASK] : null;
}
/**
- * Transition run state to at least the given state. Return true
- * if not already at least given state.
- */
- private boolean transitionRunStateTo(int state) {
- for (;;) {
- int c = runControl;
- if (runStateOf(c) >= state)
- return false;
- if (casRunControl(c, runControlFor(state, activeCountOf(c))))
+ * Tries to pop the given task from submitter's queue in common pool.
+ */
+ static boolean tryExternalUnpush(ForkJoinTask> t) {
+ ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
+ ForkJoinTask>[] a; int m, s; long j;
+ if ((z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ (s = q.top) != q.base &&
+ (a = q.array) != null &&
+ U.getObjectVolatile
+ (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s && // recheck
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ q.qlock = 0;
return true;
+ }
+ q.qlock = 0;
+ }
+ return false;
+ }
+
+ /**
+ * Tries to pop and run local tasks within the same computation
+ * as the given root. On failure, tries to help complete from
+ * other queues via helpComplete.
+ */
+ private void externalHelpComplete(WorkQueue q, ForkJoinTask> root) {
+ ForkJoinTask>[] a; int m;
+ if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
+ root != null && root.status >= 0) {
+ for (;;) {
+ int s; Object o; CountedCompleter> task = null;
+ if ((s = q.top) - q.base > 0) {
+ long j = ((m & (s - 1)) << ASHIFT) + ABASE;
+ if ((o = U.getObject(a, j)) != null &&
+ (o instanceof CountedCompleter)) {
+ CountedCompleter> t = (CountedCompleter>)o, r = t;
+ do {
+ if (r == root) {
+ if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ task = t;
+ }
+ q.qlock = 0;
+ }
+ break;
+ }
+ } while ((r = r.completer) != null);
+ }
+ }
+ if (task != null)
+ task.doExec();
+ if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0)
+ break;
+ if (task == null) {
+ if (helpSignal(root, q.poolIndex) >= 0)
+ helpComplete(root, SHARED_QUEUE);
+ break;
+ }
+ }
}
}
/**
- * Controls whether to add spares to maintain parallelism
+ * Tries to help execute or signal availability of the given task
+ * from submitter's queue in common pool.
*/
- private volatile boolean maintainsParallelism;
+ static void externalHelpJoin(ForkJoinTask> t) {
+ // Some hard-to-avoid overlap with tryExternalUnpush
+ ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
+ ForkJoinTask>[] a; int m, s, n; long j;
+ if (t != null && t.status >= 0 &&
+ (z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ (a = q.array) != null) {
+ if ((s = q.top) != q.base &&
+ U.getObjectVolatile
+ (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ q.qlock = 0;
+ t.doExec();
+ }
+ else
+ q.qlock = 0;
+ }
+ if (t.status >= 0) {
+ if (t instanceof CountedCompleter)
+ p.externalHelpComplete(q, t);
+ else
+ p.helpSignal(t, q.poolIndex);
+ }
+ }
+ }
+
+ /**
+ * Restricted version of helpQuiescePool for external callers
+ */
+ static void externalHelpQuiescePool() {
+ ForkJoinPool p; ForkJoinTask> t; WorkQueue q; int b;
+ int r = ThreadLocalRandom.current().nextInt();
+ if ((p = commonPool) != null &&
+ (q = p.findNonEmptyStealQueue(r)) != null &&
+ (b = q.base) - q.top < 0 &&
+ (t = q.pollAt(b)) != null)
+ t.doExec();
+ }
+
+ // Exported methods
// Constructors
/**
- * Creates a ForkJoinPool with a pool size equal to the number of
- * processors available on the system, using the default
- * ForkJoinWorkerThreadFactory.
+ * 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
@@ -353,221 +2388,119 @@ public class ForkJoinPool extends Abstra
*/
public ForkJoinPool() {
this(Runtime.getRuntime().availableProcessors(),
- defaultForkJoinWorkerThreadFactory);
+ defaultForkJoinWorkerThreadFactory, null, false);
}
/**
- * Creates a ForkJoinPool with the indicated parallelism level
- * threads and using the default ForkJoinWorkerThreadFactory.
+ * 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 number of worker threads
+ * @param parallelism the parallelism level
* @throws IllegalArgumentException if parallelism less than or
- * equal to zero
+ * 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);
- }
-
- /**
- * 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")}
- */
- public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
- this(Runtime.getRuntime().availableProcessors(), factory);
+ this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
}
/**
- * Creates a ForkJoinPool with the given parallelism and factory.
+ * Creates a {@code ForkJoinPool} with the given parameters.
*
- * @param parallelism the targeted number of worker threads
- * @param factory the factory for creating new threads
+ * @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 factory is null
+ * 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")}
*/
- public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
- if (parallelism <= 0 || parallelism > MAX_THREADS)
- throw new IllegalArgumentException();
+ public ForkJoinPool(int parallelism,
+ ForkJoinWorkerThreadFactory factory,
+ Thread.UncaughtExceptionHandler handler,
+ boolean asyncMode) {
+ checkPermission();
if (factory == null)
throw new NullPointerException();
- checkPermission();
- this.factory = factory;
+ if (parallelism <= 0 || parallelism > MAX_CAP)
+ throw new IllegalArgumentException();
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>();
- // worker array and workers are lazily constructed
- }
-
- /**
- * Creates a new worker thread 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.setAsyncMode(locallyFifo);
- w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
- if (h != null)
- w.setUncaughtExceptionHandler(h);
- }
- return w;
+ this.factory = factory;
+ this.ueh = handler;
+ this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
+ long np = (long)(-parallelism); // offset ctl counts
+ this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
+ int pn = nextPoolId();
+ StringBuilder sb = new StringBuilder("ForkJoinPool-");
+ sb.append(Integer.toString(pn));
+ sb.append("-worker-");
+ this.workerNamePrefix = sb.toString();
}
/**
- * Returns a good size for worker array given pool size.
- * Currently requires size to be a power of two.
- */
- private static int arraySizeFor(int poolSize) {
- return (poolSize <= 1) ? 1 :
- (1 << (32 - Integer.numberOfLeadingZeros(poolSize-1)));
+ * Constructor for common pool, suitable only for static initialization.
+ * Basically the same as above, but uses smallest possible initial footprint.
+ */
+ ForkJoinPool(int parallelism, long ctl,
+ ForkJoinWorkerThreadFactory factory,
+ Thread.UncaughtExceptionHandler handler) {
+ this.parallelism = parallelism;
+ this.ctl = ctl;
+ this.factory = factory;
+ this.ueh = handler;
+ this.localMode = LIFO_QUEUE;
+ this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
}
/**
- * Creates or resizes array if necessary to hold newLength.
- * Call only under exclusion.
+ * Returns the common pool instance.
*
- * @return the array
+ * @return the common pool instance
*/
- 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;
- }
-
- /**
- * Tries to shrink workers into smaller array after one or more terminate.
- */
- private void tryShrinkWorkerArray() {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- 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);
- }
- }
-
- /**
- * Initializes workers if necessary.
- */
- final void ensureWorkerInitialization() {
- ForkJoinWorkerThread[] ws = workers;
- if (ws == null) {
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ws = workers;
- if (ws == null) {
- int ps = parallelism;
- ws = ensureWorkerArrayCapacity(ps);
- for (int i = 0; i < ps; ++i) {
- ForkJoinWorkerThread w = createWorker(i);
- if (w != null) {
- ws[i] = w;
- w.start();
- updateWorkerCount(1);
- }
- }
- }
- } finally {
- lock.unlock();
- }
- }
- }
-
- /**
- * Worker creation and startup for threads added via setParallelism.
- */
- 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 static ForkJoinPool commonPool() {
+ return commonPool; // cannot be null (if so, a static init error)
}
// Execution methods
/**
- * Common code for execute, invoke and submit
- */
- private void doSubmit(ForkJoinTask task) {
- if (task == null)
- throw new NullPointerException();
- if (isShutdown())
- throw new RejectedExecutionException();
- if (workers == null)
- ensureWorkerInitialization();
- submissionQueue.offer(task);
- signalIdleWorkers();
- }
-
- /**
* 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 invoke(ForkJoinTask task) {
- doSubmit(task);
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(task);
return task.join();
}
@@ -575,44 +2508,32 @@ public class ForkJoinPool extends Abstra
* 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 void execute(ForkJoinTask task) {
- doSubmit(task);
+ public void execute(ForkJoinTask> task) {
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(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) {
+ if (task == null)
+ throw new NullPointerException();
ForkJoinTask> job;
- if (task instanceof ForkJoinTask) // avoid re-wrap
- job = (ForkJoinTask>)task;
- else
- job = new AdaptedRunnable(task, null);
- doSubmit(job);
- }
-
- public ForkJoinTask submit(Callable task) {
- ForkJoinTask job = new AdaptedCallable(task);
- doSubmit(job);
- return job;
- }
-
- public ForkJoinTask submit(Runnable task, T result) {
- ForkJoinTask job = new AdaptedRunnable(task, result);
- doSubmit(job);
- return job;
- }
-
- public ForkJoinTask> submit(Runnable task) {
- ForkJoinTask> job;
- if (task instanceof ForkJoinTask) // avoid re-wrap
- job = (ForkJoinTask>)task;
+ if (task instanceof ForkJoinTask>) // avoid re-wrap
+ job = (ForkJoinTask>) task;
else
- job = new AdaptedRunnable(task, null);
- doSubmit(job);
- return job;
+ job = new ForkJoinTask.AdaptedRunnableAction(task);
+ externalPush(job);
}
/**
@@ -620,93 +2541,88 @@ public class ForkJoinPool extends Abstra
*
* @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
- * @throws NullPointerException if the task is null
*/
public ForkJoinTask submit(ForkJoinTask task) {
- doSubmit(task);
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(task);
return task;
}
/**
- * Adaptor for Runnables. This implements RunnableFuture
- * to be compliant with AbstractExecutorService constraints.
+ * @throws NullPointerException if the task is null
+ * @throws RejectedExecutionException if the task cannot be
+ * scheduled for execution
*/
- static final class AdaptedRunnable extends ForkJoinTask
- implements RunnableFuture {
- 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(); }
- private static final long serialVersionUID = 5232453952276885070L;
+ public ForkJoinTask submit(Callable task) {
+ ForkJoinTask job = new ForkJoinTask.AdaptedCallable(task);
+ externalPush(job);
+ return job;
}
/**
- * Adaptor for Callables
+ * @throws NullPointerException if the task is null
+ * @throws RejectedExecutionException if the task cannot be
+ * scheduled for execution
*/
- static final class AdaptedCallable extends ForkJoinTask
- implements RunnableFuture {
- final Callable callable;
- T result;
- AdaptedCallable(Callable 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(); }
- private static final long serialVersionUID = 2838392045355241008L;
+ public ForkJoinTask submit(Runnable task, T result) {
+ ForkJoinTask job = new ForkJoinTask.AdaptedRunnable(task, result);
+ externalPush(job);
+ return job;
}
- public List> invokeAll(Collection extends Callable> tasks) {
- ArrayList> forkJoinTasks =
- new ArrayList>(tasks.size());
- for (Callable task : tasks)
- forkJoinTasks.add(new AdaptedCallable(task));
- invoke(new InvokeAll(forkJoinTasks));
+ /**
+ * @throws NullPointerException if the task is null
+ * @throws RejectedExecutionException if the task cannot be
+ * scheduled for execution
+ */
+ public ForkJoinTask> submit(Runnable task) {
+ if (task == null)
+ throw new NullPointerException();
+ ForkJoinTask> job;
+ if (task instanceof ForkJoinTask>) // avoid re-wrap
+ job = (ForkJoinTask>) task;
+ else
+ job = new ForkJoinTask.AdaptedRunnableAction(task);
+ externalPush(job);
+ return job;
+ }
+ /**
+ * @throws NullPointerException {@inheritDoc}
+ * @throws RejectedExecutionException {@inheritDoc}
+ */
+ public List> invokeAll(Collection extends Callable> tasks) {
+ // 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> fs = new ArrayList>(tasks.size());
+ // Workaround needed because method wasn't declared with
+ // wildcards in return type but should have been.
@SuppressWarnings({"unchecked", "rawtypes"})
- List> futures = (List>) (List) forkJoinTasks;
- return futures;
- }
+ List> futures = (List>) (List) fs;
- static final class InvokeAll extends RecursiveAction {
- final ArrayList> tasks;
- InvokeAll(ArrayList> tasks) { this.tasks = tasks; }
- public void compute() {
- try { invokeAll(tasks); }
- catch (Exception ignore) {}
+ boolean done = false;
+ try {
+ for (Callable t : tasks) {
+ ForkJoinTask f = new ForkJoinTask.AdaptedCallable(t);
+ externalPush(f);
+ fs.add(f);
+ }
+ for (ForkJoinTask f : fs)
+ f.quietlyJoin();
+ done = true;
+ return futures;
+ } finally {
+ if (!done)
+ for (ForkJoinTask f : fs)
+ f.cancel(false);
}
- private static final long serialVersionUID = -7914297376763021607L;
}
- // Configuration and status settings and queries
-
/**
* Returns the factory used for constructing new workers.
*
@@ -720,204 +2636,70 @@ public class ForkJoinPool extends Abstra
* Returns the handler for internal worker threads that terminate
* due to unrecoverable errors encountered while executing tasks.
*
- * @return the handler, or null if none
+ * @return the handler, or {@code 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")}
- */
- 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;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- w.setUncaughtExceptionHandler(h);
- }
- }
- } finally {
- lock.unlock();
- }
- return old;
+ return ueh;
}
-
/**
- * Sets the target parallelism level of this pool.
+ * Returns the targeted parallelism 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")}
+ * @return the targeted parallelism level of this pool
*/
- 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 int getParallelism() {
+ return parallelism;
}
/**
- * Returns the targeted number of worker threads in this pool.
+ * Returns the targeted parallelism level of the common pool.
*
- * @return the targeted number of worker threads in this pool
+ * @return the targeted parallelism level of the common pool
*/
- public int getParallelism() {
- return parallelism;
+ public static int getCommonPoolParallelism() {
+ return commonPoolParallelism;
}
/**
* Returns the number of worker threads that have started but not
- * yet terminated. This result returned by this method may differ
- * from {@code getParallelism} 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);
- }
-
- /**
- * 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
- */
- public void setMaintainsParallelism(boolean enable) {
- maintainsParallelism = enable;
- }
-
- /**
- * 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 asynchronous tasks. This method is
- * designed to be invoked only when pool is quiescent, and
- * typically only before any tasks are submitted. The effects of
- * invocations at other times may be unpredictable.
- *
- * @param async if true, use locally FIFO scheduling
- * @return the previous mode
- */
- public boolean setAsyncMode(boolean async) {
- boolean oldMode = locallyFifo;
- locallyFifo = async;
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.setAsyncMode(async);
- }
- }
- return oldMode;
+ return parallelism + (short)(ctl >>> TC_SHIFT);
}
/**
- * Returns true if this pool uses local first-in-first-out
+ * Returns {@code true} if this pool uses local first-in-first-out
* scheduling mode for forked tasks that are never joined.
*
- * @return true if this pool uses async mode
+ * @return {@code true} if this pool uses async mode
*/
public boolean getAsyncMode() {
- return locallyFifo;
+ 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;
}
/**
@@ -928,34 +2710,23 @@ public class ForkJoinPool extends Abstra
* @return the number of active threads
*/
public int getActiveThreadCount() {
- return activeCountOf(runControl);
+ int r = parallelism + (int)(ctl >> AC_SHIFT);
+ return (r <= 0) ? 0 : r; // suppress momentarily negative values
}
/**
- * Returns an estimate of the number of threads that are currently
- * idle waiting for tasks. This method may underestimate the
- * number of idle threads.
+ * 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 the number of idle threads
- */
- final int getIdleThreadCount() {
- int c = runningCountOf(workerCounts) - activeCountOf(runControl);
- return (c <= 0) ? 0 : c;
- }
-
- /**
- * 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
+ * @return {@code true} if all threads are currently idle
*/
public boolean isQuiescent() {
- return activeCountOf(runControl) == 0;
+ return (int)(ctl >> AC_SHIFT) + parallelism == 0;
}
/**
@@ -970,17 +2741,15 @@ public class ForkJoinPool extends Abstra
* @return the number of steals
*/
public long getStealCount() {
- return stealCount.get();
- }
-
- /**
- * Accumulates 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;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.nsteals;
+ }
+ }
+ return count;
}
/**
@@ -995,36 +2764,50 @@ public class ForkJoinPool extends Abstra
*/
public long getQueuedTaskCount() {
long count = 0;
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- 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.
+ * 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.
+ * 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;
}
/**
@@ -1032,18 +2815,25 @@ public class ForkJoinPool extends Abstra
* 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
+ * 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
@@ -1055,17 +2845,20 @@ public class ForkJoinPool extends Abstra
* @param c the collection to transfer elements into
* @return the number of elements transferred
*/
- protected int drainTasksTo(Collection> c) {
- int n = submissionQueue.drainTo(c);
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
+ 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) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- n += w.drainTasksTo(c);
+ if ((w = ws[i]) != null) {
+ while ((t = w.poll()) != null) {
+ c.add(t);
+ ++count;
+ }
+ }
}
}
- return n;
+ return count;
}
/**
@@ -1076,42 +2869,56 @@ public class ForkJoinPool extends Abstra
* @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;
+ 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.nsteals;
+ 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 = plock < 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.
+ * Possibly initiates an orderly shutdown in which previously
+ * submitted tasks are executed, but no new tasks will be
+ * accepted. Invocation has no effect on execution state if this
+ * is the {@link #commonPool}, and 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
@@ -1120,20 +2927,20 @@ public class ForkJoinPool extends Abstra
*/
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
- * upon termination, so always returns an empty list. However, you
- * can use method {@code drainTasksTo} before invoking this
- * method to transfer unexecuted tasks to another collection.
+ * Possibly attempts to cancel and/or stop all tasks, and reject
+ * all subsequently submitted tasks. Invocation has no effect on
+ * execution state if this is the {@link #commonPool}, and no
+ * additional effect if already shut down. Otherwise, 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
@@ -1143,7 +2950,7 @@ public class ForkJoinPool extends Abstra
*/
public List shutdownNow() {
checkPermission();
- terminate();
+ tryTerminate(true, true);
return Collections.emptyList();
}
@@ -1153,17 +2960,28 @@ public class ForkJoinPool extends Abstra
* @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} if the process of termination has
- * commenced but possibly not yet completed.
+ * 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} 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);
}
/**
@@ -1172,13 +2990,15 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if this pool has been shut down
*/
public boolean isShutdown() {
- return runStateOf(runControl) >= SHUTDOWN;
+ return plock < 0;
}
/**
- * Blocks until all tasks have completed execution after a shutdown
- * request, or the timeout occurs, or the current thread is
- * interrupted, whichever happens first.
+ * Blocks until all tasks have completed execution after a
+ * shutdown request, or the timeout occurs, or the current thread
+ * is interrupted, whichever happens first. Note that the {@link
+ * #commonPool()} never terminates until program shutdown so
+ * this method will always time out.
*
* @param timeout the maximum time to wait
* @param unit the time unit of the timeout argument
@@ -1189,595 +3009,40 @@ public class ForkJoinPool extends Abstra
public boolean awaitTermination(long timeout, TimeUnit unit)
throws InterruptedException {
long nanos = unit.toNanos(timeout);
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- for (;;) {
- if (isTerminated())
- return true;
- if (nanos <= 0)
- return false;
- nanos = termination.awaitNanos(nanos);
- }
- } finally {
- lock.unlock();
- }
- }
-
- // Shutdown and termination support
-
- /**
- * Callback from terminating worker. Nulls out the corresponding
- * workers slot, and if terminating, tries to terminate; else
- * tries 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;
- if (ws != null) {
- 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();
- }
-
- /**
- * Initiates termination.
- */
- private void terminate() {
- if (transitionRunStateTo(TERMINATING)) {
- stopAllWorkers();
- resumeAllSpares();
- signalIdleWorkers();
- cancelQueuedSubmissions();
- cancelQueuedWorkerTasks();
- interruptUnterminatedWorkers();
- signalIdleWorkers(); // resignal after interrupt
- }
- }
-
- /**
- * Possibly terminates when on shutdown state.
- */
- private void terminateOnShutdown() {
- if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
- terminate();
- }
-
- /**
- * Clears out and cancels submissions.
- */
- private void cancelQueuedSubmissions() {
- ForkJoinTask> task;
- while ((task = pollSubmission()) != null)
- task.cancel(false);
- }
-
- /**
- * Cleans out worker queues.
- */
- private void cancelQueuedWorkerTasks() {
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.cancelTasks();
- }
- }
- } finally {
- lock.unlock();
- }
- }
-
- /**
- * Sets 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;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.shutdownNow();
- }
- }
- } finally {
- lock.unlock();
- }
- }
-
- /**
- * Interrupts 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;
- if (ws != null) {
- 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.
- *
- * 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;
- }
-
- /**
- * Wakes up waiter, returning false if known to already
- */
- boolean signal() {
- ForkJoinWorkerThread t = thread;
- if (t == null)
- return false;
- thread = null;
- LockSupport.unpark(t);
+ if (isTerminated())
return true;
- }
-
- /**
- * Awaits release on sync.
- */
- void awaitSyncRelease(ForkJoinPool p) {
- while (thread != null && !p.syncIsReleasable(this))
- LockSupport.park(this);
- }
-
- /**
- * Awaits 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();
- }
-
- /**
- * Signals 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
-
- /**
- * Decrements running count; if too low, adds spare.
- *
- * 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 encounter 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))));
- }
-
- /**
- * Adds 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()))
+ long startTime = System.nanoTime();
+ boolean terminated = false;
+ synchronized (this) {
+ for (long waitTime = nanos, millis = 0L;;) {
+ if (terminated = isTerminated() ||
+ waitTime <= 0L ||
+ (millis = unit.toMillis(waitTime)) <= 0L)
break;
- if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
- createAndStartSpare(tc);
- success = true;
- break;
- }
- }
- } finally {
- if (locked)
- lock.unlock();
- }
- return success;
- }
-
- /**
- * Adds the kth spare worker. On entry, pool counts 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();
- }
-
- /**
- * Suspends 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;
- }
-
- /**
- * Tries 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;
- }
-
- /**
- * Pops and resumes 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;
- }
-
- /**
- * Pops and shuts down 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);
+ wait(millis);
+ waitTime = nanos - (System.nanoTime() - startTime);
}
}
+ return terminated;
}
/**
* Interface for extending managed parallelism for tasks running
- * in ForkJoinPools. A ManagedBlocker provides two methods.
- * Method {@code isReleasable} must return true if blocking is not
- * necessary. Method {@code block} blocks the current thread if
- * necessary (perhaps internally invoking {@code isReleasable}
- * before actually blocking.).
+ * in {@link ForkJoinPool}s.
+ *
+ * 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.
*
*
For example, here is a ManagedBlocker based on a
* ReentrantLock:
@@ -1795,98 +3060,215 @@ public class ForkJoinPool extends Abstra
* return hasLock || (hasLock = lock.tryLock());
* }
* }}
+ *
+ *
Here is a class that possibly blocks waiting for an
+ * item on a given queue:
+ *
{@code
+ * class QueueTaker implements ManagedBlocker {
+ * final BlockingQueue queue;
+ * volatile E item = null;
+ * QueueTaker(BlockingQueue 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;
+ * }
+ * public E getItem() { // call after pool.managedBlock completes
+ * return item;
+ * }
+ * }}
*/
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 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} is true and the pool supports
- * it ({@link #getMaintainsParallelism}), this method attempts to
- * maintain the pool's nominal parallelism. Otherwise it activates
- * a thread only if necessary to avoid complete starvation. This
- * option may be preferable when blockages use timeouts, or are
- * almost always brief.
+ * 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.
*
- * If the caller is not a ForkJoinTask, this method is behaviorally
- * equivalent to
+ *
If the caller is not a {@link ForkJoinTask}, this method is
+ * behaviorally equivalent to
*
{@code
* while (!blocker.isReleasable())
* if (blocker.block())
* return;
* }
- * If the caller is a ForkJoinTask, then the pool may first
- * be expanded to ensure parallelism, and later adjusted.
+ *
+ * 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
*/
- 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);
+ if (t instanceof ForkJoinWorkerThread) {
+ ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
+ while (!blocker.isReleasable()) { // variant of helpSignal
+ WorkQueue[] ws; WorkQueue q; int m, n;
+ if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
+ for (int i = 0; i <= m; ++i) {
+ if (blocker.isReleasable())
+ return;
+ if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
+ p.signalWork(q, n);
+ if ((int)(p.ctl >> AC_SHIFT) >= 0)
+ break;
+ }
+ }
+ }
+ if (p.tryCompensate()) {
+ try {
+ do {} while (!blocker.isReleasable() &&
+ !blocker.block());
+ } finally {
+ p.incrementActiveCount();
+ }
+ break;
+ }
}
}
+ else {
+ do {} while (!blocker.isReleasable() &&
+ !blocker.block());
+ }
}
- 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 RunnableFuture newTaskFor(Runnable runnable, T value) {
- return new AdaptedRunnable(runnable, value);
+ return new ForkJoinTask.AdaptedRunnable(runnable, value);
}
protected RunnableFuture newTaskFor(Callable callable) {
- return new AdaptedCallable(callable);
+ return new ForkJoinTask.AdaptedCallable(callable);
}
+ // Unsafe mechanics
+ private static final sun.misc.Unsafe U;
+ private static final long CTL;
+ private static final long PARKBLOCKER;
+ private static final int ABASE;
+ private static final int ASHIFT;
+ private static final long STEALCOUNT;
+ private static final long PLOCK;
+ private static final long INDEXSEED;
+ private static final long QLOCK;
+
+ static {
+ // Establish common pool parameters
+ // TBD: limit or report ignored exceptions?
+
+ int par = 0;
+ ForkJoinWorkerThreadFactory fac = null;
+ Thread.UncaughtExceptionHandler handler = null;
+ try {
+ String pp = System.getProperty(propPrefix + "parallelism");
+ String hp = System.getProperty(propPrefix + "exceptionHandler");
+ String fp = System.getProperty(propPrefix + "threadFactory");
+ if (fp != null)
+ fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
+ getSystemClassLoader().loadClass(fp).newInstance());
+ if (hp != null)
+ handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
+ getSystemClassLoader().loadClass(hp).newInstance());
+ if (pp != null)
+ par = Integer.parseInt(pp);
+ } catch (Exception ignore) {
+ }
+
+ int s; // initialize field offsets for CAS etc
+ try {
+ U = getUnsafe();
+ Class> k = ForkJoinPool.class;
+ CTL = U.objectFieldOffset
+ (k.getDeclaredField("ctl"));
+ STEALCOUNT = U.objectFieldOffset
+ (k.getDeclaredField("stealCount"));
+ PLOCK = U.objectFieldOffset
+ (k.getDeclaredField("plock"));
+ INDEXSEED = U.objectFieldOffset
+ (k.getDeclaredField("indexSeed"));
+ Class> tk = Thread.class;
+ PARKBLOCKER = U.objectFieldOffset
+ (tk.getDeclaredField("parkBlocker"));
+ Class> wk = WorkQueue.class;
+ QLOCK = U.objectFieldOffset
+ (wk.getDeclaredField("qlock"));
+ Class> ak = ForkJoinTask[].class;
+ ABASE = U.arrayBaseOffset(ak);
+ s = U.arrayIndexScale(ak);
+ ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
+ } catch (Exception e) {
+ throw new Error(e);
+ }
+ if ((s & (s-1)) != 0)
+ throw new Error("data type scale not a power of two");
+
+ /*
+ * For extra caution, computations to set up pool state are
+ * here; the constructor just assigns these values to fields.
+ */
+ ForkJoinWorkerThreadFactory defaultFac =
+ defaultForkJoinWorkerThreadFactory =
+ new DefaultForkJoinWorkerThreadFactory();
+ if (fac == null)
+ fac = defaultFac;
+ if (par <= 0)
+ par = Runtime.getRuntime().availableProcessors();
+ if (par > MAX_CAP)
+ par = MAX_CAP;
+ long np = (long)(-par); // precompute initial ctl value
+ long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
+
+ commonPoolParallelism = par;
+ commonPool = new ForkJoinPool(par, ct, fac, handler);
+ modifyThreadPermission = new RuntimePermission("modifyThread");
+ submitters = new ThreadLocal();
+ }
- // Unsafe mechanics for jsr166y 3rd party package.
+ /**
+ * 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 sun.misc.Unsafe.getUnsafe();
} catch (SecurityException se) {
try {
return java.security.AccessController.doPrivileged
- (new java.security.PrivilegedExceptionAction() {
+ (new java.security
+ .PrivilegedExceptionAction() {
public sun.misc.Unsafe run() throws Exception {
- return getUnsafeByReflection();
+ 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 new RuntimeException("Could not initialize intrinsics",
@@ -1895,50 +3277,4 @@ public class ForkJoinPool extends Abstra
}
}
- private static sun.misc.Unsafe getUnsafeByReflection()
- throws NoSuchFieldException, IllegalAccessException {
- java.lang.reflect.Field f =
- sun.misc.Unsafe.class.getDeclaredField("theUnsafe");
- f.setAccessible(true);
- return (sun.misc.Unsafe) f.get(null);
- }
-
- private static long fieldOffset(String fieldName, Class> klazz) {
- try {
- return UNSAFE.objectFieldOffset(klazz.getDeclaredField(fieldName));
- } catch (NoSuchFieldException e) {
- // Convert Exception to Error
- NoSuchFieldError error = new NoSuchFieldError(fieldName);
- error.initCause(e);
- throw error;
- }
- }
-
- private static final sun.misc.Unsafe UNSAFE = getUnsafe();
- static final long eventCountOffset =
- fieldOffset("eventCount", ForkJoinPool.class);
- static final long workerCountsOffset =
- fieldOffset("workerCounts", ForkJoinPool.class);
- static final long runControlOffset =
- fieldOffset("runControl", ForkJoinPool.class);
- static final long syncStackOffset =
- fieldOffset("syncStack",ForkJoinPool.class);
- static final long spareStackOffset =
- fieldOffset("spareStack", ForkJoinPool.class);
-
- 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);
- }
}