--- jsr166/src/jsr166y/ForkJoinPool.java 2010/04/18 13:59:57 1.55
+++ jsr166/src/jsr166y/ForkJoinPool.java 2012/01/27 17:27:28 1.116
@@ -1,62 +1,56 @@
/*
* 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.LockSupport;
-import java.util.concurrent.locks.ReentrantLock;
+import java.util.Random;
+import java.util.concurrent.AbstractExecutorService;
+import java.util.concurrent.Callable;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Future;
+import java.util.concurrent.RejectedExecutionException;
+import java.util.concurrent.RunnableFuture;
+import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
-import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.atomic.AtomicLong;
+import java.util.concurrent.locks.ReentrantLock;
+import java.util.concurrent.locks.Condition;
/**
* An {@link ExecutorService} for running {@link ForkJoinTask}s.
* A {@code ForkJoinPool} provides the entry point for submissions
- * from non-{@code ForkJoinTask}s, as well as management and
+ * 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 subtasks 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). A {@code ForkJoinPool} may also be used for mixed
- * execution of some plain {@code Runnable}- or {@code Callable}-
- * based activities along with {@code ForkJoinTask}s. When setting
- * {@linkplain #setAsyncMode async mode}, a {@code ForkJoinPool} may
- * also be appropriate for use with fine-grained tasks of any form
- * that are never joined. Otherwise, other {@code ExecutorService}
- * implementations are typically more appropriate choices.
+ * 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.
*
*
A {@code ForkJoinPool} is constructed with a given target
* parallelism level; by default, equal to the number of available
- * processors. Unless configured otherwise via {@link
- * #setMaintainsParallelism}, the pool attempts to maintain this
- * number of 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 performed in the face of blocked IO or other unmanaged
- * synchronization. The nested {@link ManagedBlocker} interface
- * enables extension of the kinds of synchronization accommodated.
- * The target parallelism level may also be changed dynamically
- * ({@link #setParallelism}). The total number of threads may be
- * limited using method {@link #setMaximumPoolSize}, in which case it
- * may become possible for the activities of a pool to stall due to
- * the lack of available threads to process new tasks. When the pool
- * is executing tasks, these and other configuration setting methods
- * may only gradually affect actual pool sizes. It is normally best
- * practice to invoke these methods only when the pool is known to be
- * quiescent.
+ * 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
@@ -65,6 +59,42 @@ import java.util.concurrent.CountDownLat
* {@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) |
+ *
+ *
+ *
* Sample Usage. Normally a single {@code ForkJoinPool} is
* used for all parallel task execution in a program or subsystem.
* Otherwise, use would not usually outweigh the construction and
@@ -75,13 +105,12 @@ import java.util.concurrent.CountDownLat
* daemon} mode, there is typically no need to explicitly {@link
* #shutdown} such a pool upon program exit.
*
- *
+ * {@code
* static final ForkJoinPool mainPool = new ForkJoinPool();
* ...
* public void sort(long[] array) {
* mainPool.invoke(new SortTask(array, 0, array.length));
- * }
- *
+ * }}
*
* Implementation notes: This implementation restricts the
* maximum number of running threads to 32767. Attempts to create
@@ -89,7 +118,8 @@ import java.util.concurrent.CountDownLat
* {@code IllegalArgumentException}.
*
*
This implementation rejects submitted tasks (that is, by throwing
- * {@link RejectedExecutionException}) only when the pool is shut down.
+ * {@link RejectedExecutionException}) only when the pool is shut down
+ * or internal resources have been exhausted.
*
* @since 1.7
* @author Doug Lea
@@ -99,230 +129,322 @@ public class ForkJoinPool extends Abstra
/*
* Implementation Overview
*
- * This class provides the central bookkeeping and control for a
- * set of worker threads: Submissions from non-FJ threads enter
- * into a submission queue. Workers take these tasks and typically
- * split them into subtasks that may be stolen by other workers.
- * The main work-stealing mechanics implemented in class
- * ForkJoinWorkerThread 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 worker queues, and
- * lastly to new submissions. These mechanics do not consider
- * 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.)
+ * This class and its nested classes provide the main
+ * functionality and control for a set of worker threads:
+ * Submissions from non-FJ threads enter into submission
+ * queues. Workers take these tasks and typically split them into
+ * subtasks that may be stolen by other workers. Preference rules
+ * give first priority to processing tasks from their own queues
+ * (LIFO or FIFO, depending on mode), then to randomized FIFO
+ * steals of tasks in other queues.
+ *
+ * WorkQueues.
+ * ==========
+ *
+ * Most operations occur within work-stealing queues (in nested
+ * class WorkQueue). These are special forms of Deques that
+ * support only three of the four possible end-operations -- push,
+ * pop, and poll (aka steal), under the further constraints that
+ * push and pop are called only from the owning thread (or, as
+ * extended here, under a lock), while poll may be called from
+ * other threads. (If you are unfamiliar with them, you probably
+ * want to read Herlihy and Shavit's book "The Art of
+ * Multiprocessor programming", chapter 16 describing these in
+ * more detail before proceeding.) The main work-stealing queue
+ * design is roughly similar to those in the papers "Dynamic
+ * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
+ * (http://research.sun.com/scalable/pubs/index.html) and
+ * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
+ * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
+ * The main differences ultimately stem from gc requirements that
+ * we null out taken slots as soon as we can, to maintain as small
+ * a footprint as possible even in programs generating huge
+ * numbers of tasks. To accomplish this, we shift the CAS
+ * arbitrating pop vs poll (steal) from being on the indices
+ * ("base" and "top") to the slots themselves. So, both a
+ * successful pop and poll mainly entail a CAS of a slot from
+ * non-null to null. Because we rely on CASes of references, we
+ * do not need tag bits on base or top. They are simple ints as
+ * used in any circular array-based queue (see for example
+ * ArrayDeque). Updates to the indices must still be ordered in a
+ * way that guarantees that top == base means the queue is empty,
+ * but otherwise may err on the side of possibly making the queue
+ * appear nonempty when a push, pop, or poll have not fully
+ * committed. Note that this means that the poll operation,
+ * considered individually, is not wait-free. One thief cannot
+ * successfully continue until another in-progress one (or, if
+ * previously empty, a push) completes. However, in the
+ * aggregate, we ensure at least probabilistic non-blockingness.
+ * If an attempted steal fails, a thief always chooses a different
+ * random victim target to try next. So, in order for one thief to
+ * progress, it suffices for any in-progress poll or new push on
+ * any empty queue to complete.
+ *
+ * This approach also enables support of a user mode in which local
+ * task processing is in FIFO, not LIFO order, simply by using
+ * poll rather than pop. This can be useful in message-passing
+ * frameworks in which tasks are never joined. However neither
+ * mode considers affinities, loads, cache localities, etc, so
+ * rarely provide the best possible performance on a given
+ * machine, but portably provide good throughput by averaging over
+ * these factors. (Further, even if we did try to use such
+ * information, we do not usually have a basis for exploiting
+ * it. For example, some sets of tasks profit from cache
+ * affinities, but others are harmed by cache pollution effects.)
+ *
+ * WorkQueues are also used in a similar way for tasks submitted
+ * to the pool. We cannot mix these tasks in the same queues used
+ * for work-stealing (this would contaminate lifo/fifo
+ * processing). Instead, we loosely associate submission queues
+ * with submitting threads, using a form of hashing. The
+ * ThreadLocal Submitter class contains a value initially used as
+ * a hash code for choosing existing queues, but may be randomly
+ * repositioned upon contention with other submitters. In
+ * essence, submitters act like workers except that they never
+ * take tasks, and they are multiplexed on to a finite number of
+ * shared work queues. However, classes are set up so that future
+ * extensions could allow submitters to optionally help perform
+ * tasks as well. Pool submissions from internal workers are also
+ * allowed, but use randomized rather than thread-hashed queue
+ * indices to avoid imbalance. Insertion of tasks in shared mode
+ * requires a lock (mainly to protect in the case of resizing) but
+ * we use only a simple spinlock (using bits in field runState),
+ * because submitters encountering a busy queue try or create
+ * others so never block.
+ *
+ * Management.
+ * ==========
*
* The main throughput advantages of work-stealing stem from
- * decentralized control -- workers mostly steal tasks from each
- * other. We do not want to negate this by creating bottlenecks
- * implementing the management responsibilities of this class. So
- * we use a collection of techniques that avoid, reduce, or cope
- * well with contention. These entail several instances of
- * bit-packing into CASable fields to maintain only the minimally
- * required atomicity. To enable such packing, we restrict maximum
- * parallelism to (1<<15)-1 (enabling twice this to fit into a 16
- * bit field), which is far in excess of normal operating range.
- * Even though updates to some of these bookkeeping fields do
- * sometimes contend with each other, they don't normally
- * cache-contend with updates to others enough to warrant memory
- * padding or isolation. So they are all held as fields of
- * ForkJoinPool objects. The main capabilities are as follows:
- *
- * 1. Creating and removing workers. Workers are recorded in the
- * "workers" array. This is an array as opposed to some other data
- * structure to support index-based random steals by workers.
- * Updates to the array recording new workers and unrecording
- * terminated ones are protected from each other by a lock
- * (workerLock) but the array is otherwise concurrently readable,
- * and accessed directly by workers. To simplify index-based
+ * decentralized control -- workers mostly take tasks from
+ * themselves or each other. We cannot negate this in the
+ * implementation of other management responsibilities. The main
+ * tactic for avoiding bottlenecks is packing nearly all
+ * essentially atomic control state into two volatile variables
+ * that are by far most often read (not written) as status and
+ * consistency checks
+ *
+ * Field "ctl" contains 64 bits holding all the information needed
+ * to atomically decide to add, inactivate, enqueue (on an event
+ * queue), dequeue, and/or re-activate workers. To enable this
+ * packing, we restrict maximum parallelism to (1<<15)-1 (which is
+ * far in excess of normal operating range) to allow ids, counts,
+ * and their negations (used for thresholding) to fit into 16bit
+ * fields.
+ *
+ * Field "runState" contains 32 bits needed to register and
+ * deregister WorkQueues, as well as to enable shutdown. It is
+ * only modified under a lock (normally briefly held, but
+ * occasionally protecting allocations and resizings) but even
+ * when locked remains available to check consistency.
+ *
+ * Recording WorkQueues. WorkQueues are recorded in the
+ * "workQueues" array that is created upon pool construction and
+ * expanded if necessary. Updates to the array while recording
+ * new workers and unrecording terminated ones are protected from
+ * each other by a lock but the array is otherwise concurrently
+ * readable, and accessed directly. To simplify index-based
* operations, the array size is always a power of two, and all
- * readers must tolerate null slots. Currently, all but the first
- * worker thread creation is on-demand, triggered by task
- * submissions, replacement of terminated workers, and/or
+ * readers must tolerate null slots. Shared (submission) queues
+ * are at even indices, worker queues at odd indices. Grouping
+ * them together in this way simplifies and speeds up task
+ * scanning. To avoid flailing during start-up, the array is
+ * presized to hold twice #parallelism workers (which is unlikely
+ * to need further resizing during execution). But to avoid
+ * dealing with so many null slots, variable runState includes a
+ * mask for the nearest power of two that contains all current
+ * workers. 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.
+ * 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 check that it is non-null
+ * (even if previously non-null). This allows nulling during
+ * termination, which is currently not necessary, but remains an
+ * option for resource-revocation-based shutdown schemes. It also
+ * helps reduce JIT issuance of uncommon-trap code, which tends to
+ * unnecessarily complicate control flow in some methods.
+ *
+ * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
+ * let workers spin indefinitely scanning for tasks when none can
+ * be found immediately, and we cannot start/resume workers unless
+ * there appear to be tasks available. On the other hand, we must
+ * quickly prod them into action when new tasks are submitted or
+ * generated. In many usages, ramp-up time to activate workers is
+ * the main limiting factor in overall performance (this is
+ * compounded at program start-up by JIT compilation and
+ * allocation). So we try to streamline this as much as possible.
+ * We park/unpark workers after placing in an event wait queue
+ * when they cannot find work. This "queue" is actually a simple
+ * Treiber stack, headed by the "id" field of ctl, plus a 15bit
+ * counter value (that reflects the number of times a worker has
+ * been inactivated) to avoid ABA effects (we need only as many
+ * version numbers as worker threads). Successors are held in
+ * field WorkQueue.nextWait. Queuing deals with several intrinsic
+ * races, mainly that a task-producing thread can miss seeing (and
+ * signalling) another thread that gave up looking for work but
+ * has not yet entered the wait queue. We solve this by requiring
+ * a full sweep of all workers (via repeated calls to method
+ * scan()) both before and after a newly waiting worker is added
+ * to the wait queue. During a rescan, the worker might release
+ * some other queued worker rather than itself, which has the same
+ * net effect. Because enqueued workers may actually be rescanning
+ * rather than waiting, we set and clear the "parker" field of
+ * Workqueues to reduce unnecessary calls to unpark. (This
+ * requires a secondary recheck to avoid missed signals.) Note
+ * the unusual conventions about Thread.interrupts surrounding
+ * parking and other blocking: Because interrupts are used solely
+ * to alert threads to check termination, which is checked anyway
+ * upon blocking, we clear status (using Thread.interrupted)
+ * before any call to park, so that park does not immediately
+ * return due to status being set via some other unrelated call to
+ * interrupt in user code.
+ *
+ * Signalling. We create or wake up workers only when there
+ * appears to be at least one task they might be able to find and
+ * execute. When a submission is added or another worker adds a
+ * task to a queue that previously had fewer than two tasks, they
+ * signal waiting workers (or trigger creation of new ones if
+ * fewer than the given parallelism level -- see signalWork).
+ * These primary signals are buttressed by signals during rescans;
+ * together these cover the signals needed in cases when more
+ * tasks are pushed but untaken, and improve performance compared
+ * to having one thread wake up all workers.
+ *
+ * Trimming workers. To release resources after periods of lack of
+ * use, a worker starting to wait when the pool is quiescent will
+ * time out and terminate if the pool has remained quiescent for
+ * SHRINK_RATE nanosecs. This will slowly propagate, eventually
+ * terminating all workers after long periods of non-use.
+ *
+ * Shutdown and Termination. A call to shutdownNow atomically sets
+ * a runState bit and then (non-atomically) sets each workers
+ * runState status, cancels all unprocessed tasks, and wakes up
+ * all waiting workers. Detecting whether termination should
+ * commence after a non-abrupt shutdown() call requires more work
+ * and bookkeeping. We need consensus about quiescence (i.e., that
+ * there is no more work). The active count provides a primary
+ * indication but non-abrupt shutdown still requires a rechecking
+ * scan for any workers that are inactive but not queued.
+ *
+ * Joining Tasks.
+ * ==============
+ *
+ * Any of several actions may be taken when one worker is waiting
+ * to join a task stolen (or always held by) another. Because we
+ * are multiplexing many tasks on to a pool of workers, we can't
+ * just let them block (as in Thread.join). We also cannot just
+ * reassign the joiner's run-time stack with another and replace
+ * it later, which would be a form of "continuation", that even if
+ * possible is not necessarily a good idea since we sometimes need
+ * both an unblocked task and its continuation to
+ * progress. Instead we combine two tactics:
+ *
+ * Helping: Arranging for the joiner to execute some task that it
+ * would be running if the steal had not occurred.
+ *
+ * Compensating: Unless there are already enough live threads,
+ * method tryCompensate() may create or re-activate a spare
+ * thread to compensate for blocked joiners until they unblock.
+ *
+ * A third form (implemented in tryRemoveAndExec and
+ * tryPollForAndExec) amounts to helping a hypothetical
+ * compensator: If we can readily tell that a possible action of a
+ * compensator is to steal and execute the task being joined, the
+ * joining thread can do so directly, without the need for a
+ * compensation thread (although at the expense of larger run-time
+ * stacks, but the tradeoff is typically worthwhile).
+ *
+ * The ManagedBlocker extension API can't use helping so relies
+ * only on compensation in method awaitBlocker.
+ *
+ * The algorithm in tryHelpStealer entails a form of "linear"
+ * helping: Each worker records (in field currentSteal) the most
+ * recent task it stole from some other worker. Plus, it records
+ * (in field currentJoin) the task it is currently actively
+ * joining. Method tryHelpStealer uses these markers to try to
+ * find a worker to help (i.e., steal back a task from and execute
+ * it) that could hasten completion of the actively joined task.
+ * In essence, the joiner executes a task that would be on its own
+ * local deque had the to-be-joined task not been stolen. This may
+ * be seen as a conservative variant of the approach in Wagner &
+ * Calder "Leapfrogging: a portable technique for implementing
+ * efficient futures" SIGPLAN Notices, 1993
+ * (http://portal.acm.org/citation.cfm?id=155354). It differs in
+ * that: (1) We only maintain dependency links across workers upon
+ * steals, rather than use per-task bookkeeping. This sometimes
+ * requires a linear scan of workers array to locate stealers, but
+ * often doesn't because stealers leave hints (that may become
+ * stale/wrong) of where to locate them. A stealHint is only a
+ * hint because a worker might have had multiple steals and the
+ * hint records only one of them (usually the most current).
+ * Hinting isolates cost to when it is needed, rather than adding
+ * to per-task overhead. (2) It is "shallow", ignoring nesting
+ * and potentially cyclic mutual steals. (3) It is intentionally
+ * racy: field currentJoin is updated only while actively joining,
+ * which means that we miss links in the chain during long-lived
+ * tasks, GC stalls etc (which is OK since blocking in such cases
+ * is usually a good idea). (4) We bound the number of attempts
+ * to find work (see MAX_HELP_DEPTH) and fall back to suspending
+ * the worker and if necessary replacing it with another.
+ *
+ * It is impossible to keep exactly the target parallelism number
+ * of threads running at any given time. Determining the
+ * existence of conservatively safe helping targets, the
+ * availability of already-created spares, and the apparent need
+ * to create new spares are all racy, so we rely on multiple
+ * retries of each. Currently, in keeping with on-demand
+ * signalling policy, we compensate only if blocking would leave
+ * less than one active (non-waiting, non-blocked) worker.
+ * Additionally, to avoid some false alarms due to GC, lagging
+ * counters, system activity, etc, compensated blocking for joins
+ * is only attempted after rechecks stabilize in
+ * ForkJoinTask.awaitJoin. (Retries are interspersed with
+ * Thread.yield, for good citizenship.)
*
- * 2. Bookkeeping for dynamically adding and removing workers. We
- * maintain a given level of parallelism (or, if
- * maintainsParallelism is false, at least avoid starvation). When
- * some workers are known to be blocked (on joins or via
- * ManagedBlocker), we may create or resume others to take their
- * place until they unblock (see below). Implementing this
- * requires counts of the number of "running" threads (i.e., those
- * that are neither blocked nor artifically suspended) as well as
- * the total number. These two values are packed into one field,
- * "workerCounts" because we need accurate snapshots when deciding
- * to create, resume or suspend. To support these decisions,
- * updates must be prospective (not retrospective). For example,
- * the running count is decremented before blocking by a thread
- * about to block, but incremented by the thread about to unblock
- * it. (In a few cases, these prospective updates may need to be
- * rolled back, for example when deciding to create a new worker
- * but the thread factory fails or returns null. In these cases,
- * we are no worse off wrt other decisions than we would be
- * otherwise.) Updates to the workerCounts field sometimes
- * transiently encounter a fair amount of contention when join
- * dependencies are such that many threads block or unblock at
- * about the same time. We alleviate this by sometimes bundling
- * updates (for example blocking one thread on join and resuming a
- * spare cancel each other out), and in most other cases
- * performing an alternative action (like releasing waiters and
- * finding spares; see below) as a more productive form of
- * backoff.
- *
- * 3. Maintaining global run state. The run state of the pool
- * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
- * those in other Executor implementations, as well as a count of
- * "active" workers -- those that are, or soon will be, or
- * recently were executing tasks. The runLevel and active count
- * are packed together in order to correctly trigger shutdown and
- * termination. Without care, active counts can be subject to very
- * high contention. We substantially reduce this contention by
- * relaxing update rules. A worker must claim active status
- * prospectively, by activating if it sees that a submitted or
- * stealable task exists (it may find after activating that the
- * task no longer exists). It stays active while processing this
- * task (if it exists) and any other local subtasks it produces,
- * until it cannot find any other tasks. It then tries
- * inactivating (see method preStep), but upon update contention
- * instead scans for more tasks, later retrying inactivation if it
- * doesn't find any.
- *
- * 4. Managing idle workers waiting for tasks. We cannot let
- * workers spin indefinitely scanning for tasks when none are
- * available. On the other hand, we must quickly prod them into
- * action when new tasks are submitted or generated. We
- * park/unpark these idle workers using an event-count scheme.
- * Field eventCount is incremented upon events that may enable
- * workers that previously could not find a task to now find one:
- * Submission of a new task to the pool, or another worker pushing
- * a task onto a previously empty queue. (We also use this
- * mechanism for termination and reconfiguration actions that
- * require wakeups of idle workers). Each worker maintains its
- * last known event count, and blocks when a scan for work did not
- * find a task AND its lastEventCount matches the current
- * eventCount. Waiting idle workers are recorded in a variant of
- * Treiber stack headed by field eventWaiters which, when nonzero,
- * encodes the thread index and count awaited for by the worker
- * thread most recently calling eventSync. This thread in turn has
- * a record (field nextEventWaiter) for the next waiting worker.
- * In addition to allowing simpler decisions about need for
- * wakeup, the event count bits in eventWaiters serve the role of
- * tags to avoid ABA errors in Treiber stacks. To reduce delays
- * in task diffusion, workers not otherwise occupied may invoke
- * method releaseWaiters, that removes and signals (unparks)
- * workers not waiting on current count. To minimize task
- * production stalls associate with signalling, any worker pushing
- * a task on an empty queue invokes the weaker method signalWork,
- * that only releases idle workers until it detects interference
- * by other threads trying to release, and lets them take
- * over. The net effect is a tree-like diffusion of signals, where
- * released threads and possibly others) help with unparks. To
- * further reduce contention effects a bit, failed CASes to
- * increment field eventCount are tolerated without retries.
- * Conceptually they are merged into the same event, which is OK
- * when their only purpose is to enable workers to scan for work.
- *
- * 5. Managing suspension of extra workers. When a worker is about
- * to block waiting for a join (or via ManagedBlockers), we may
- * create a new thread to maintain parallelism level, or at least
- * avoid starvation (see below). Usually, extra threads are needed
- * for only very short periods, yet join dependencies are such
- * that we sometimes need them in bursts. Rather than create new
- * threads each time this happens, we suspend no-longer-needed
- * extra ones as "spares". For most purposes, we don't distinguish
- * "extra" spare threads from normal "core" threads: On each call
- * to preStep (the only point at which we can do this) a worker
- * checks to see if there are now too many running workers, and if
- * so, suspends itself. Methods preJoin and doBlock look for
- * suspended threads to resume before considering creating a new
- * replacement. We don't need a special data structure to maintain
- * spares; simply scanning the workers array looking for
- * worker.isSuspended() is fine because the calling thread is
- * otherwise not doing anything useful anyway; we are at least as
- * happy if after locating a spare, the caller doesn't actually
- * block because the join is ready before we try to adjust and
- * compensate. Note that this is intrinsically racy. One thread
- * may become a spare at about the same time as another is
- * needlessly being created. We counteract this and related slop
- * in part by requiring resumed spares to immediately recheck (in
- * preStep) to see whether they they should re-suspend. The only
- * effective difference between "extra" and "core" threads is that
- * we allow the "extra" ones to time out and die if they are not
- * resumed within a keep-alive interval of a few seconds. This is
- * implemented mainly within ForkJoinWorkerThread, but requires
- * some coordination (isTrimmed() -- meaning killed while
- * suspended) to correctly maintain pool counts.
- *
- * 6. Deciding when to create new workers. The main dynamic
- * control in this class is deciding when to create extra threads,
- * in methods preJoin and doBlock. We always need to create one
- * when the number of running threads becomes zero. But because
- * blocked joins are typically dependent, we don't necessarily
- * need or want one-to-one replacement. Using a one-to-one
- * compensation rule often leads to enough useless overhead
- * creating, suspending, resuming, and/or killing threads to
- * signficantly degrade throughput. We use a rule reflecting the
- * idea that, the more spare threads you already have, the more
- * evidence you need to create another one; where "evidence" is
- * expressed as the current deficit -- target minus running
- * threads. To reduce flickering and drift around target values,
- * the relation is quadratic: adding a spare if (dc*dc)>=(sc*pc)
- * (where dc is deficit, sc is number of spare threads and pc is
- * target parallelism.) This effectively reduces churn at the
- * price of systematically undershooting target parallelism when
- * many threads are blocked. However, biasing toward undeshooting
- * partially compensates for the above mechanics to suspend extra
- * threads, that normally lead to overshoot because we can only
- * suspend workers in-between top-level actions. It also better
- * copes with the fact that some of the methods in this class tend
- * to never become compiled (but are interpreted), so some
- * components of the entire set of controls might execute many
- * times faster than others. And similarly for cases where the
- * apparent lack of work is just due to GC stalls and other
- * transient system activity.
- *
- * 7. Maintaining other configuration parameters and monitoring
- * statistics. Updates to fields controlling parallelism level,
- * max size, etc can only meaningfully take effect for individual
- * threads upon their next top-level actions; i.e., between
- * stealing/running tasks/submission, which are separated by calls
- * to preStep. Memory ordering for these (assumed infrequent)
- * reconfiguration calls is ensured by using reads and writes to
- * volatile field workerCounts (that must be read in preStep anyway)
- * as "fences" -- user-level reads are preceded by reads of
- * workCounts, and writes are followed by no-op CAS to
- * workerCounts. The values reported by other management and
- * monitoring methods are either computed on demand, or are kept
- * in fields that are only updated when threads are otherwise
- * idle.
- *
- * Beware that there is a lot of representation-level coupling
+ * Style notes: There is a lot of representation-level coupling
* among classes ForkJoinPool, ForkJoinWorkerThread, and
- * ForkJoinTask. For example, direct access to "workers" array by
- * workers, and direct access to ForkJoinTask.status by both
- * ForkJoinPool and ForkJoinWorkerThread. There is little point
- * trying to reduce this, since any associated future changes in
- * representations will need to be accompanied by algorithmic
- * changes anyway.
- *
- * Style notes: There are lots of inline assignments (of form
- * "while ((local = field) != 0)") which are usually the simplest
- * way to ensure read orderings. Also 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 a few
- * other coding oddities that help some methods perform reasonably
- * even when interpreted (not compiled).
- *
- * The order of declarations in this file is: (1) statics (2)
- * fields (along with constants used when unpacking some of them)
- * (3) internal control methods (4) callbacks and other support
- * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
- * methods (plus a few little helpers).
+ * 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. All together, these low-level
+ * implementation choices produce as much as a factor of 4
+ * performance improvement compared to naive implementations, and
+ * enable the processing of billions of tasks per second, at the
+ * expense of some ugliness.
+ *
+ * Methods signalWork() and scan() are the main bottlenecks so are
+ * especially heavily micro-optimized/mangled. There are lots of
+ * inline assignments (of form "while ((local = field) != 0)")
+ * which are usually the simplest way to ensure the required read
+ * orderings (which are sometimes critical). This leads to a
+ * "C"-like style of listing declarations of these locals at the
+ * heads of methods or blocks. There are several occurrences of
+ * the unusual "do {} while (!cas...)" which is the simplest way
+ * to force an update of a CAS'ed variable. There are also other
+ * coding oddities that help some methods perform reasonably even
+ * when interpreted (not compiled).
+ *
+ * The order of declarations in this file is: (1) declarations of
+ * statics (2) fields (along with constants used when unpacking
+ * some of them), listed in an order that tends to reduce
+ * contention among them a bit under most JVMs; (3) nested
+ * classes; (4) internal control methods; (5) callbacks and other
+ * support for ForkJoinTask methods; (6) exported methods (plus a
+ * few little helpers); (7) static block initializing all statics
+ * in a minimally dependent order.
*/
/**
@@ -345,7 +467,7 @@ 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);
@@ -357,15 +479,13 @@ public class ForkJoinPool extends Abstra
* overridden in ForkJoinPool constructors.
*/
public static final ForkJoinWorkerThreadFactory
- defaultForkJoinWorkerThreadFactory =
- new DefaultForkJoinWorkerThreadFactory();
+ defaultForkJoinWorkerThreadFactory;
/**
* 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
@@ -380,765 +500,1392 @@ public class ForkJoinPool extends Abstra
/**
* Generator for assigning sequence numbers as pool names.
*/
- private static final AtomicInteger poolNumberGenerator =
- new AtomicInteger();
+ private static final AtomicInteger poolNumberGenerator;
/**
- * Absolute bound for parallelism level. Twice this number must
- * fit into a 16bit field to enable word-packing for some counts.
+ * 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 >>> 1) 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, when id is
+ * negative, there is at least one waiting worker, 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
+ * 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.
+ *
+ * Field runState is an int packed with:
+ * SHUTDOWN: true if shutdown is enabled (1 bit)
+ * SEQ: a sequence number updated upon (de)registering workers (15 bits)
+ * MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
+ *
+ * The combination of mask and sequence number enables simple
+ * consistency checks: Staleness of read-only operations on the
+ * workers and queues arrays can be checked by comparing runState
+ * before vs after the reads. The low 16 bits (i.e, anding with
+ * SMASK) hold (the smallest power of two covering all worker
+ * indices, minus one. The mask for queues (vs workers) is twice
+ * this value plus 1.
+ */
+
+ // 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 MAX_ID = 0x7fff; // max poolIndex
+ private static final int SMASK = 0xffff; // mask short bits
+ private static final int SHORT_SIGN = 1 << 15;
+ private static final int INT_SIGN = 1 << 31;
+
+ // masks
+ private static final long STOP_BIT = 0x0001L << ST_SHIFT;
+ private static final long AC_MASK = ((long)SMASK) << AC_SHIFT;
+ private static final long TC_MASK = ((long)SMASK) << TC_SHIFT;
+
+ // units for incrementing and decrementing
+ private static final long TC_UNIT = 1L << TC_SHIFT;
+ private static final long AC_UNIT = 1L << AC_SHIFT;
+
+ // masks and units for dealing with u = (int)(ctl >>> 32)
+ private static final int UAC_SHIFT = AC_SHIFT - 32;
+ private static final int UTC_SHIFT = TC_SHIFT - 32;
+ private static final int UAC_MASK = SMASK << UAC_SHIFT;
+ private static final int UTC_MASK = SMASK << UTC_SHIFT;
+ private static final int UAC_UNIT = 1 << UAC_SHIFT;
+ private static final int UTC_UNIT = 1 << UTC_SHIFT;
+
+ // masks and units for dealing with e = (int)ctl
+ private static final int E_MASK = 0x7fffffff; // no STOP_BIT
+ private static final int E_SEQ = 1 << EC_SHIFT;
+
+ // runState bits
+ private static final int SHUTDOWN = 1 << 31;
+ private static final int RS_SEQ = 1 << 16;
+ private static final int RS_SEQ_MASK = 0x7fff0000;
+
+ // access mode for WorkQueue
+ static final int LIFO_QUEUE = 0;
+ static final int FIFO_QUEUE = 1;
+ static final int SHARED_QUEUE = -1;
+
+ /**
+ * The wakeup interval (in nanoseconds) for a worker waiting for a
+ * task when the pool is quiescent to instead try to shrink the
+ * number of workers. The exact value does not matter too
+ * much. It must be short enough to release resources during
+ * sustained periods of idleness, but not so short that threads
+ * are continually re-created.
+ */
+ private static final long SHRINK_RATE =
+ 4L * 1000L * 1000L * 1000L; // 4 seconds
+
+ /**
+ * The timeout value for attempted shrinkage, includes
+ * some slop to cope with system timer imprecision.
+ */
+ private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
+
+ /**
+ * The maximum stolen->joining link depth allowed in tryHelpStealer.
+ * Depths for legitimate chains are unbounded, but we use a fixed
+ * constant to avoid (otherwise unchecked) cycles and to bound
+ * staleness of traversal parameters at the expense of sometimes
+ * blocking when we could be helping.
*/
- private static final int MAX_THREADS = 0x7fff;
+ private static final int MAX_HELP_DEPTH = 16;
- /**
- * Array holding all worker threads in the pool. Array size must
- * be a power of two. Updates and replacements are protected by
- * workerLock, but the array is always kept in a consistent enough
- * state to be randomly accessed without locking by workers
- * performing work-stealing, as well as other traversal-based
- * methods in this class. All readers must tolerate that some
- * array slots may be null.
+ /*
+ * Field layout order in this class tends to matter more than one
+ * would like. Runtime layout order is only loosely related to
+ * declaration order and may differ across JVMs, but the following
+ * empirically works OK on current JVMs.
+ */
+
+ volatile long ctl; // main pool control
+ final int parallelism; // parallelism level
+ final int localMode; // per-worker scheduling mode
+ int nextPoolIndex; // hint used in registerWorker
+ volatile int runState; // shutdown status, seq, and mask
+ WorkQueue[] workQueues; // main registry
+ final ReentrantLock lock; // for registration
+ final Condition termination; // for awaitTermination
+ final ForkJoinWorkerThreadFactory factory; // factory for new workers
+ final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
+ final AtomicLong stealCount; // collect counts when terminated
+ final AtomicInteger nextWorkerNumber; // to create worker name string
+ final String workerNamePrefix; // Prefix for assigning worker names
+
+ /**
+ * Queues supporting work-stealing as well as external task
+ * submission. See above for main rationale and algorithms.
+ * Implementation relies heavily on "Unsafe" intrinsics
+ * and selective use of "volatile":
+ *
+ * Field "base" is the index (mod array.length) of the least valid
+ * queue slot, which is always the next position to steal (poll)
+ * from if nonempty. Reads and writes require volatile orderings
+ * but not CAS, because updates are only performed after slot
+ * CASes.
+ *
+ * Field "top" is the index (mod array.length) of the next queue
+ * slot to push to or pop from. It is written only by owner thread
+ * for push, or under lock for trySharedPush, and accessed by
+ * other threads only after reading (volatile) base. Both top and
+ * base are allowed to wrap around on overflow, but (top - base)
+ * (or more commonly -(base - top) to force volatile read of base
+ * before top) still estimates size.
+ *
+ * The array slots are read and written using the emulation of
+ * volatiles/atomics provided by Unsafe. Insertions must in
+ * general use putOrderedObject as a form of releasing store to
+ * ensure that all writes to the task object are ordered before
+ * its publication in the queue. (Although we can avoid one case
+ * of this when locked in trySharedPush.) All removals entail a
+ * CAS to null. The array is always a power of two. To ensure
+ * safety of Unsafe array operations, all accesses perform
+ * explicit null checks and implicit bounds checks via
+ * power-of-two masking.
+ *
+ * In addition to basic queuing support, this class contains
+ * fields described elsewhere to control execution. It turns out
+ * to work better memory-layout-wise to include them in this
+ * class rather than a separate class.
+ *
+ * Performance on most platforms is very sensitive to placement of
+ * instances of both WorkQueues and their arrays -- we absolutely
+ * do not want multiple WorkQueue instances or multiple queue
+ * arrays sharing cache lines. (It would be best for queue objects
+ * and their arrays to share, but there is nothing available to
+ * help arrange that). Unfortunately, because they are recorded
+ * in a common array, WorkQueue instances are often moved to be
+ * adjacent by garbage collectors. To reduce impact, we use field
+ * padding that works OK on common platforms; this effectively
+ * trades off slightly slower average field access for the sake of
+ * avoiding really bad worst-case access. (Until better JVM
+ * support is in place, this padding is dependent on transient
+ * properties of JVM field layout rules.) We also take care in
+ * allocating and sizing and resizing the array. Non-shared queue
+ * arrays are initialized (via method growArray) by workers before
+ * use. Others are allocated on first use.
*/
- volatile ForkJoinWorkerThread[] workers;
+ static final class WorkQueue {
+ /**
+ * Capacity of work-stealing queue array upon initialization.
+ * Must be a power of two; at least 4, but set larger to
+ * reduce cacheline sharing among queues.
+ */
+ static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
- /**
- * Queue for external submissions.
- */
- private final LinkedTransferQueue> submissionQueue;
+ /**
+ * Maximum size for queue arrays. Must be a power of two less
+ * than or equal to 1 << (31 - width of array entry) to ensure
+ * lack of wraparound of index calculations, but defined to a
+ * value a bit less than this to help users trap runaway
+ * programs before saturating systems.
+ */
+ static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
- /**
- * Lock protecting updates to workers array.
- */
- private final ReentrantLock workerLock;
+ volatile long totalSteals; // cumulative number of steals
+ int seed; // for random scanning; initialize nonzero
+ volatile int eventCount; // encoded inactivation count; < 0 if inactive
+ int nextWait; // encoded record of next event waiter
+ int rescans; // remaining scans until block
+ int nsteals; // top-level task executions since last idle
+ final int mode; // lifo, fifo, or shared
+ int poolIndex; // index of this queue in pool (or 0)
+ int stealHint; // index of most recent known stealer
+ volatile int runState; // 1: locked, -1: terminate; else 0
+ volatile int base; // index of next slot for poll
+ int top; // index of next slot for push
+ ForkJoinTask>[] array; // the elements (initially unallocated)
+ final ForkJoinWorkerThread owner; // owning thread or null if shared
+ volatile Thread parker; // == owner during call to park; else null
+ ForkJoinTask> currentJoin; // task being joined in awaitJoin
+ ForkJoinTask> currentSteal; // current non-local task being executed
+ // Heuristic padding to ameliorate unfortunate memory placements
+ Object p00, p01, p02, p03, p04, p05, p06, p07, p08, p09, p0a;
+
+ WorkQueue(ForkJoinWorkerThread owner, int mode) {
+ this.owner = owner;
+ this.mode = mode;
+ // Place indices in the center of array (that is not yet allocated)
+ base = top = INITIAL_QUEUE_CAPACITY >>> 1;
+ }
- /**
- * Latch released upon termination.
- */
- private final CountDownLatch terminationLatch;
+ /**
+ * Returns number of tasks in the queue
+ */
+ final int queueSize() {
+ int n = base - top; // non-owner callers must read base first
+ return (n >= 0) ? 0 : -n;
+ }
- /**
- * Creation factory for worker threads.
- */
- private final ForkJoinWorkerThreadFactory factory;
+ /**
+ * Pushes a task. Call only by owner in unshared queues.
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @param p, if non-null, pool to signal if necessary
+ * @throw RejectedExecutionException if array cannot
+ * be resized
+ */
+ final void push(ForkJoinTask> task, ForkJoinPool p) {
+ ForkJoinTask>[] a;
+ int s = top, m, n;
+ if ((a = array) != null) { // ignore if queue removed
+ U.putOrderedObject
+ (a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
+ if ((n = (top = s + 1) - base) <= 2) {
+ if (p != null)
+ p.signalWork();
+ }
+ else if (n >= m)
+ growArray(true);
+ }
+ }
- /**
- * Sum of per-thread steal counts, updated only when threads are
- * idle or terminating.
- */
- private volatile long stealCount;
+ /**
+ * Pushes a task if lock is free and array is either big
+ * enough or can be resized to be big enough.
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @return true if submitted
+ */
+ final boolean trySharedPush(ForkJoinTask> task) {
+ boolean submitted = false;
+ if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
+ ForkJoinTask>[] a = array;
+ int s = top, n = s - base;
+ try {
+ if ((a != null && n < a.length - 1) ||
+ (a = growArray(false)) != null) { // must presize
+ int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
+ U.putObject(a, (long)j, task); // don't need "ordered"
+ top = s + 1;
+ submitted = true;
+ }
+ } finally {
+ runState = 0; // unlock
+ }
+ }
+ return submitted;
+ }
- /**
- * Encoded record of top of treiber stack of threads waiting for
- * events. The top 32 bits contain the count being waited for. The
- * bottom word contains one plus the pool index of waiting worker
- * thread.
- */
- private volatile long eventWaiters;
+ /**
+ * Takes next task, if one exists, in FIFO order.
+ */
+ final ForkJoinTask> poll() {
+ ForkJoinTask>[] a; int b, i;
+ while ((b = base) - top < 0 && (a = array) != null &&
+ (i = (a.length - 1) & b) >= 0) {
+ int j = (i << ASHIFT) + ABASE;
+ ForkJoinTask> t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t != null && base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ return t;
+ }
+ }
+ return null;
+ }
- private static final int EVENT_COUNT_SHIFT = 32;
- private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
+ /**
+ * Takes next task, if one exists, in LIFO order.
+ * Call only by owner in unshared queues.
+ */
+ final ForkJoinTask> pop() {
+ ForkJoinTask> t; int m;
+ ForkJoinTask>[] a = array;
+ if (a != null && (m = a.length - 1) >= 0) {
+ for (int s; (s = top - 1) - base >= 0;) {
+ int j = ((m & s) << ASHIFT) + ABASE;
+ if ((t = (ForkJoinTask>)U.getObjectVolatile(a, j)) == null)
+ break;
+ if (U.compareAndSwapObject(a, j, t, null)) {
+ top = s;
+ return t;
+ }
+ }
+ }
+ return null;
+ }
- /**
- * A counter for events that may wake up worker threads:
- * - Submission of a new task to the pool
- * - A worker pushing a task on an empty queue
- * - termination and reconfiguration
- */
- private volatile int eventCount;
+ /**
+ * Takes next task, if one exists, in order specified by mode.
+ */
+ final ForkJoinTask> nextLocalTask() {
+ return mode == 0 ? pop() : poll();
+ }
- /**
- * Lifecycle control. The 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. Bits 16-18
- * contain runLevel value. When all are zero, the pool is
- * running. Level transitions are monotonic (running -> shutdown
- * -> terminating -> terminated) so each transition adds a bit.
- * These are bundled together to ensure consistent read for
- * termination checks (i.e., that runLevel is at least SHUTDOWN
- * and active threads is zero).
- */
- private volatile int runState;
+ /**
+ * 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);
+ }
- // Note: The order among run level values matters.
- private static final int RUNLEVEL_SHIFT = 16;
- private static final int SHUTDOWN = 1 << RUNLEVEL_SHIFT;
- private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1);
- private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2);
- private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1;
- private static final int ONE_ACTIVE = 1; // active update delta
+ /**
+ * Returns task at index b if b is current base of queue.
+ */
+ final ForkJoinTask> pollAt(int b) {
+ ForkJoinTask>[] a; int i;
+ ForkJoinTask> task = null;
+ if ((a = array) != null && (i = ((a.length - 1) & b)) >= 0) {
+ int j = (i << ASHIFT) + ABASE;
+ ForkJoinTask> t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t != null && base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ task = t;
+ }
+ }
+ return task;
+ }
- /**
- * 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 together to ensure consistent snapshot when
- * making decisions about creating and suspending spare
- * threads. Updated only by CAS. Note that adding a new worker
- * requires incrementing both counts, since workers start off in
- * running state. This field is also used for memory-fencing
- * configuration parameters.
- */
- private volatile int workerCounts;
+ /**
+ * Pops the given task only if it is at the current top.
+ */
+ final boolean tryUnpush(ForkJoinTask> t) {
+ ForkJoinTask>[] a; int s;
+ if ((a = array) != null && (s = top) != base &&
+ U.compareAndSwapObject
+ (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
+ top = s;
+ return true;
+ }
+ return false;
+ }
- private static final int TOTAL_COUNT_SHIFT = 16;
- private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1;
- private static final int ONE_RUNNING = 1;
- private static final int ONE_TOTAL = 1 << TOTAL_COUNT_SHIFT;
+ /**
+ * Polls the given task only if it is at the current base.
+ */
+ final boolean pollFor(ForkJoinTask> task) {
+ ForkJoinTask>[] a; int b, i;
+ if ((b = base) - top < 0 && (a = array) != null &&
+ (i = (a.length - 1) & b) >= 0) {
+ int j = (i << ASHIFT) + ABASE;
+ if (U.getObjectVolatile(a, j) == task && base == b &&
+ U.compareAndSwapObject(a, j, task, null)) {
+ base = b + 1;
+ return true;
+ }
+ }
+ return false;
+ }
- /*
- * Fields parallelism. maxPoolSize, locallyFifo,
- * maintainsParallelism, and ueh are non-volatile, but external
- * reads/writes use workerCount fences to ensure visability.
- */
+ /**
+ * If present, removes from queue and executes the given task, or
+ * any other cancelled task. Returns (true) immediately on any CAS
+ * or consistency check failure so caller can retry.
+ *
+ * @return false if no progress can be made
+ */
+ final boolean tryRemoveAndExec(ForkJoinTask> task) {
+ boolean removed = false, empty = true, progress = true;
+ ForkJoinTask>[] a; int m, s, b, n;
+ if ((a = array) != null && (m = a.length - 1) >= 0 &&
+ (n = (s = top) - (b = base)) > 0) {
+ for (ForkJoinTask> t;;) { // traverse from s to b
+ int j = ((--s & m) << ASHIFT) + ABASE;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t == null) // inconsistent length
+ break;
+ else if (t == task) {
+ if (s + 1 == top) { // pop
+ if (!U.compareAndSwapObject(a, j, task, null))
+ break;
+ top = s;
+ removed = true;
+ }
+ else if (base == b) // replace with proxy
+ removed = U.compareAndSwapObject(a, j, task,
+ new EmptyTask());
+ break;
+ }
+ else if (t.status >= 0)
+ empty = false;
+ else if (s + 1 == top) { // pop and throw away
+ if (U.compareAndSwapObject(a, j, t, null))
+ top = s;
+ break;
+ }
+ if (--n == 0) {
+ if (!empty && base == b)
+ progress = false;
+ break;
+ }
+ }
+ }
+ if (removed)
+ task.doExec();
+ return progress;
+ }
- /**
- * The target parallelism level.
- */
- private int parallelism;
+ /**
+ * Initializes or doubles the capacity of array. Call either
+ * by owner or with lock held -- it is OK for base, but not
+ * top, to move while resizings are in progress.
+ *
+ * @param rejectOnFailure if true, throw exception if capacity
+ * exceeded (relayed ultimately to user); else return null.
+ */
+ final ForkJoinTask>[] growArray(boolean rejectOnFailure) {
+ ForkJoinTask>[] oldA = array;
+ int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
+ if (size <= MAXIMUM_QUEUE_CAPACITY) {
+ int oldMask, t, b;
+ ForkJoinTask>[] a = array = new ForkJoinTask>[size];
+ if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
+ (t = top) - (b = base) > 0) {
+ int mask = size - 1;
+ do {
+ ForkJoinTask> x;
+ int oldj = ((b & oldMask) << ASHIFT) + ABASE;
+ int j = ((b & mask) << ASHIFT) + ABASE;
+ x = (ForkJoinTask>)U.getObjectVolatile(oldA, oldj);
+ if (x != null &&
+ U.compareAndSwapObject(oldA, oldj, x, null))
+ U.putObjectVolatile(a, j, x);
+ } while (++b != t);
+ }
+ return a;
+ }
+ else if (!rejectOnFailure)
+ return null;
+ else
+ throw new RejectedExecutionException("Queue capacity exceeded");
+ }
- /**
- * The maximum allowed pool size.
- */
- private int maxPoolSize;
+ /**
+ * 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);
+ }
- /**
- * True if use local fifo, not default lifo, for local polling
- * Replicated by ForkJoinWorkerThreads
- */
- private boolean locallyFifo;
+ // Execution methods
- /**
- * Controls whether to add spares to maintain parallelism
- */
- private boolean maintainsParallelism;
+ /**
+ * Removes and runs tasks until empty, using local mode
+ * ordering.
+ */
+ final void runLocalTasks() {
+ if (base - top < 0) {
+ for (ForkJoinTask> t; (t = nextLocalTask()) != null; )
+ t.doExec();
+ }
+ }
- /**
- * The uncaught exception handler used when any worker
- * abruptly terminates
- */
- private Thread.UncaughtExceptionHandler ueh;
+ /**
+ * Executes a top-level task and any local tasks remaining
+ * after execution.
+ *
+ * @return true unless terminating
+ */
+ final boolean runTask(ForkJoinTask> t) {
+ boolean alive = true;
+ if (t != null) {
+ currentSteal = t;
+ t.doExec();
+ runLocalTasks();
+ ++nsteals;
+ currentSteal = null;
+ }
+ else if (runState < 0) // terminating
+ alive = false;
+ return alive;
+ }
- /**
- * Pool number, just for assigning useful names to worker threads
- */
- private final int poolNumber;
+ /**
+ * Executes a non-top-level (stolen) task
+ */
+ final void runSubtask(ForkJoinTask> t) {
+ if (t != null) {
+ ForkJoinTask> ps = currentSteal;
+ currentSteal = t;
+ t.doExec();
+ currentSteal = ps;
+ }
+ }
- // utilities for updating fields
+ /**
+ * Computes next value for random probes. Scans don't require
+ * a very high quality generator, but also not a crummy one.
+ * Marsaglia xor-shift is cheap and works well enough. Note:
+ * This is manually inlined in several usages in ForkJoinPool
+ * to avoid writes inside busy scan loops.
+ */
+ final int nextSeed() {
+ int r = seed;
+ r ^= r << 13;
+ r ^= r >>> 17;
+ r ^= r << 5;
+ return seed = r;
+ }
- /**
- * Adds delta to running count. Used mainly by ForkJoinTask.
- *
- * @param delta the number to add
- */
- final void updateRunningCount(int delta) {
- int wc;
- do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc = workerCounts,
- wc + delta));
+ // Unsafe mechanics
+ private static final sun.misc.Unsafe U;
+ private static final long RUNSTATE;
+ private static final int ABASE;
+ private static final int ASHIFT;
+ static {
+ int s;
+ try {
+ U = getUnsafe();
+ Class> k = WorkQueue.class;
+ Class> ak = ForkJoinTask[].class;
+ RUNSTATE = U.objectFieldOffset
+ (k.getDeclaredField("runState"));
+ ABASE = U.arrayBaseOffset(ak);
+ s = U.arrayIndexScale(ak);
+ } catch (Exception e) {
+ throw new Error(e);
+ }
+ if ((s & (s-1)) != 0)
+ throw new Error("data type scale not a power of two");
+ ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
+ }
}
/**
- * Write fence for user modifications of pool parameters
- * (parallelism. etc). Note that it doesn't matter if CAS fails.
+ * Class for artificial tasks that are used to replace the target
+ * of local joins if they are removed from an interior queue slot
+ * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
+ * actually do anything beyond having a unique identity.
*/
- private void workerCountWriteFence() {
- int wc;
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc = workerCounts, wc);
+ static final class EmptyTask extends ForkJoinTask {
+ EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
+ public Void getRawResult() { return null; }
+ public void setRawResult(Void x) {}
+ public boolean exec() { return true; }
}
/**
- * Read fence for external reads of pool parameters
- * (parallelism. maxPoolSize, etc).
+<<<<<<< ForkJoinPool.java
+ * Per-thread records for (typically non-FJ) threads that submit
+ * to pools. Cureently holds only psuedo-random seed / index that
+ * is used to chose submission queues in method doSubmit. In the
+ * future, this may incorporate a means to implement different
+ * task rejection and resubmission policies.
*/
- private void workerCountReadFence() {
- int ignore = workerCounts;
- }
+ static final class Submitter {
+ int seed; // seed for random submission queue selection
- /**
- * Tries incrementing active count; fails on contention.
- * Called by workers before executing tasks.
- *
- * @return true on success
- */
- final boolean tryIncrementActiveCount() {
- int c;
- return UNSAFE.compareAndSwapInt(this, runStateOffset,
- c = runState, c + ONE_ACTIVE);
- }
+ // Heuristic padding to ameliorate unfortunate memory placements
+ int p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
- /**
- * Tries decrementing active count; fails on contention.
- * Called when workers cannot find tasks to run.
- */
- final boolean tryDecrementActiveCount() {
- int c;
- return UNSAFE.compareAndSwapInt(this, runStateOffset,
- c = runState, c - ONE_ACTIVE);
- }
+ Submitter() {
+ // Use identityHashCode, forced negative, for seed
+ seed = System.identityHashCode(Thread.currentThread()) | (1 << 31);
+ }
- /**
- * Advances to at least the given level. Returns true if not
- * already in at least the given level.
- */
- private boolean advanceRunLevel(int level) {
- for (;;) {
- int s = runState;
- if ((s & level) != 0)
- return false;
- if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level))
- return true;
+ /**
+ * Computes next value for random probes. Like method
+ * WorkQueue.nextSeed, this is manually inlined in several
+ * usages to avoid writes inside busy loops.
+ */
+ final int nextSeed() {
+ int r = seed;
+ r ^= r << 13;
+ r ^= r >>> 17;
+ return seed = r ^= r << 5;
}
}
- // workers array maintenance
+ /** ThreadLocal class for Submitters */
+ static final class ThreadSubmitter extends ThreadLocal {
+ public Submitter initialValue() { return new Submitter(); }
+ }
/**
- * Records and returns a workers array index for new worker.
+ * Per-thread submission bookeeping. Shared across all pools
+ * to reduce ThreadLocal pollution and because random motion
+ * to avoid contention in one pool is likely to hold for others.
*/
- private int recordWorker(ForkJoinWorkerThread w) {
- // Try using slot totalCount-1. If not available, scan and/or resize
- int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1;
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- int len = ws.length;
- if (k < 0 || k >= len || ws[k] != null) {
- for (k = 0; k < len && ws[k] != null; ++k)
- ;
- if (k == len)
- ws = Arrays.copyOf(ws, len << 1);
- }
- ws[k] = w;
- workers = ws; // volatile array write ensures slot visibility
- } finally {
- lock.unlock();
- }
- return k;
- }
+ static final ThreadSubmitter submitters = new ThreadSubmitter();
/**
- * Nulls out record of worker in workers array
+ * Top-level runloop for workers
*/
- private void forgetWorker(ForkJoinWorkerThread w) {
- int idx = w.poolIndex;
- // Locking helps method recordWorker avoid unecessary expansion
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
- ws[idx] = null;
- } finally {
- lock.unlock();
- }
+ final void runWorker(ForkJoinWorkerThread wt) {
+ // Initialize queue array and seed in this thread
+ WorkQueue w = wt.workQueue;
+ w.growArray(false);
+ // Same initial hash as Submitters
+ w.seed = System.identityHashCode(Thread.currentThread()) | (1 << 31);
+
+ do {} while (w.runTask(scan(w)));
}
- // adding and removing workers
+ // Creating, registering and deregistering workers
/**
- * Tries to create and add new worker. Assumes that worker counts
- * are already updated to accommodate the worker, so adjusts on
- * failure.
- *
- * @return new worker or null if creation failed
+ * Tries to create and start a worker
*/
- private ForkJoinWorkerThread addWorker() {
+ private void addWorker() {
+ Throwable ex = null;
ForkJoinWorkerThread w = null;
try {
- w = factory.newThread(this);
- } finally { // Adjust on either null or exceptional factory return
- if (w == null) {
- onWorkerCreationFailure();
- return null;
+ if ((w = factory.newThread(this)) != null) {
+ w.start();
+ return;
}
+ } catch (Throwable e) {
+ ex = e;
}
- w.start(recordWorker(w), locallyFifo, ueh);
- return w;
+ deregisterWorker(w, ex);
}
/**
- * Adjusts counts upon failure to create worker
+ * Callback from ForkJoinWorkerThread constructor to assign a
+ * public name. This must be separate from registerWorker because
+ * it is called during the "super" constructor call in
+ * ForkJoinWorkerThread.
*/
- private void onWorkerCreationFailure() {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- c = workerCounts,
- c - (ONE_RUNNING|ONE_TOTAL)));
- tryTerminate(false); // in case of failure during shutdown
+ final String nextWorkerName() {
+ return workerNamePrefix.concat
+ (Integer.toString(nextWorkerNumber.addAndGet(1)));
}
/**
- * Create enough total workers to establish target parallelism,
- * giving up if terminating or addWorker fails
+ * Callback from ForkJoinWorkerThread constructor to establish and
+ * record its WorkQueue
+ *
+ * @param wt the worker thread
*/
- private void ensureEnoughTotalWorkers() {
- int wc;
- while (runState < TERMINATING &&
- ((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism) {
- if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
- addWorker() == null))
- break;
+ final void registerWorker(ForkJoinWorkerThread wt) {
+ WorkQueue w = wt.workQueue;
+ ReentrantLock lock = this.lock;
+ lock.lock();
+ try {
+ int k = nextPoolIndex;
+ WorkQueue[] ws = workQueues;
+ if (ws != null) { // ignore on shutdown
+ int n = ws.length;
+ if (k < 0 || (k & 1) == 0 || k >= n || ws[k] != null) {
+ for (k = 1; k < n && ws[k] != null; k += 2)
+ ; // workers are at odd indices
+ if (k >= n) // resize
+ workQueues = ws = Arrays.copyOf(ws, n << 1);
+ }
+ w.poolIndex = k;
+ w.eventCount = ~(k >>> 1) & SMASK; // Set up wait count
+ ws[k] = w; // record worker
+ nextPoolIndex = k + 2;
+ int rs = runState;
+ int m = rs & SMASK; // recalculate runState mask
+ if (k > m)
+ m = (m << 1) + 1;
+ runState = (rs & SHUTDOWN) | ((rs + RS_SEQ) & RS_SEQ_MASK) | m;
+ }
+ } finally {
+ lock.unlock();
}
}
/**
- * Final callback from terminating worker. Removes record of
+ * Final callback from terminating worker, as well as failure to
+ * construct or start a worker in addWorker. Removes record of
* worker from array, and adjusts counts. If pool is shutting
- * down, tries to complete terminatation, else possibly replaces
- * the worker.
+ * down, tries to complete termination.
*
- * @param w the worker
+ * @param wt the worker thread or null if addWorker failed
+ * @param ex the exception causing failure, or null if none
*/
- final void workerTerminated(ForkJoinWorkerThread w) {
- if (w.active) { // force inactive
- w.active = false;
- do {} while (!tryDecrementActiveCount());
+ final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
+ WorkQueue w = null;
+ if (wt != null && (w = wt.workQueue) != null) {
+ w.runState = -1; // ensure runState is set
+ stealCount.getAndAdd(w.totalSteals + w.nsteals);
+ int idx = w.poolIndex;
+ ReentrantLock lock = this.lock;
+ lock.lock();
+ try { // remove record from array
+ WorkQueue[] ws = workQueues;
+ if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
+ ws[nextPoolIndex = idx] = null;
+ } finally {
+ lock.unlock();
+ }
}
- forgetWorker(w);
-
- // decrement total count, and if was running, running count
- int unit = w.isTrimmed()? ONE_TOTAL : (ONE_RUNNING|ONE_TOTAL);
- int wc;
- do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc = workerCounts, wc - unit));
- accumulateStealCount(w); // collect final count
- if (!tryTerminate(false))
- ensureEnoughTotalWorkers();
+ 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) && w != null) {
+ w.cancelAll(); // cancel remaining tasks
+ if (w.array != null) // suppress signal if never ran
+ signalWork(); // wake up or create replacement
+ }
+
+ if (ex != null) // rethrow
+ U.throwException(ex);
}
- // Waiting for and signalling events
-
/**
- * Ensures eventCount on exit is different (mod 2^32) than on
- * entry. CAS failures are OK -- any change in count suffices.
- */
- private void advanceEventCount() {
- int c;
- UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
+ * Tries to add and register a new queue at the given index.
+ *
+ * @param idx the workQueues array index to register the queue
+ * @return the queue, or null if could not add because could
+ * not acquire lock or idx is unusable
+ */
+ private WorkQueue tryAddSharedQueue(int idx) {
+ WorkQueue q = null;
+ ReentrantLock lock = this.lock;
+ if (idx >= 0 && (idx & 1) == 0 && !lock.isLocked()) {
+ // create queue outside of lock but only if apparently free
+ WorkQueue nq = new WorkQueue(null, SHARED_QUEUE);
+ if (lock.tryLock()) {
+ try {
+ WorkQueue[] ws = workQueues;
+ if (ws != null && idx < ws.length) {
+ if ((q = ws[idx]) == null) {
+ int rs; // update runState seq
+ ws[idx] = q = nq;
+ runState = (((rs = runState) & SHUTDOWN) |
+ ((rs + RS_SEQ) & ~SHUTDOWN));
+ }
+ }
+ } finally {
+ lock.unlock();
+ }
+ }
+ }
+ return q;
}
+ // Maintaining ctl counts
+
/**
- * Releases workers blocked on a count not equal to current count.
+ * Increments active count; mainly called upon return from blocking
*/
- final void releaseWaiters() {
- long top;
- int id;
- while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 &&
- (int)(top >>> EVENT_COUNT_SHIFT) != eventCount) {
- ForkJoinWorkerThread[] ws = workers;
- ForkJoinWorkerThread w;
- if (ws.length >= id && (w = ws[id - 1]) != null &&
- UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
- top, w.nextWaiter))
- LockSupport.unpark(w);
- }
+ final void incrementActiveCount() {
+ long c;
+ do {} while (!U.compareAndSwapLong(this, CTL, c = ctl, c + AC_UNIT));
}
/**
- * Advances eventCount and releases waiters until interference by
- * other releasing threads is detected.
+ * Activates or creates a worker
*/
final void signalWork() {
- int ec;
- UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1);
- outer:for (;;) {
- long top = eventWaiters;
- ec = eventCount;
- for (;;) {
- ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
- int id = (int)(top & WAITER_INDEX_MASK);
- if (id <= 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec)
- return;
- if ((ws = workers).length < id || (w = ws[id - 1]) == null ||
- !UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
- top, top = w.nextWaiter))
- continue outer; // possibly stale; reread
- LockSupport.unpark(w);
- if (top != eventWaiters) // let someone else take over
- return;
- }
- }
- }
-
- /**
- * If worker is inactive, blocks until terminating or event count
- * advances from last value held by worker; in any case helps
- * release others.
- *
- * @param w the calling worker thread
- */
- private void eventSync(ForkJoinWorkerThread w) {
- if (!w.active) {
- int prev = w.lastEventCount;
- long nextTop = (((long)prev << EVENT_COUNT_SHIFT) |
- ((long)(w.poolIndex + 1)));
- long top;
- while ((runState < SHUTDOWN || !tryTerminate(false)) &&
- (((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 ||
- (int)(top >>> EVENT_COUNT_SHIFT) == prev) &&
- eventCount == prev) {
- if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
- w.nextWaiter = top, nextTop)) {
- accumulateStealCount(w); // transfer steals while idle
- Thread.interrupted(); // clear/ignore interrupt
- while (eventCount == prev)
- w.doPark();
+ /*
+ * The while condition is true if: (there is are too few total
+ * workers OR there is at least one waiter) AND (there are too
+ * few active workers OR the pool is terminating). The value
+ * of e distinguishes the remaining cases: zero (no waiters)
+ * for create, negative if terminating (in which case do
+ * nothing), else release a waiter. The secondary checks for
+ * release (non-null array etc) can fail if the pool begins
+ * terminating after the test, and don't impose any added cost
+ * because JVMs must perform null and bounds checks anyway.
+ */
+ long c; int e, u;
+ while ((((e = (int)(c = ctl)) | (u = (int)(c >>> 32))) &
+ (INT_SIGN|SHORT_SIGN)) == (INT_SIGN|SHORT_SIGN)) {
+ WorkQueue[] ws = workQueues; int i; WorkQueue w; Thread p;
+ if (e == 0) { // add a new worker
+ if (U.compareAndSwapLong
+ (this, CTL, c, (long)(((u + UTC_UNIT) & UTC_MASK) |
+ ((u + UAC_UNIT) & UAC_MASK)) << 32)) {
+ addWorker();
break;
}
}
- w.lastEventCount = eventCount;
- }
- releaseWaiters();
- }
-
- /**
- * Callback from workers invoked upon each top-level action (i.e.,
- * stealing a task or taking a submission and running
- * it). Performs one or both of the following:
- *
- * * If the worker cannot find work, updates its active status to
- * inactive and updates activeCount unless there is contention, in
- * which case it may try again (either in this or a subsequent
- * call). Additionally, awaits the next task event and/or helps
- * wake up other releasable waiters.
- *
- * * If there are too many running threads, suspends this worker
- * (first forcing inactivation if necessary). If it is not
- * resumed before a keepAlive elapses, the worker may be "trimmed"
- * -- killed while suspended within suspendAsSpare. Otherwise,
- * upon resume it rechecks to make sure that it is still needed.
- *
- * @param w the worker
- * @param worked false if the worker scanned for work but didn't
- * find any (in which case it may block waiting for work).
- */
- final void preStep(ForkJoinWorkerThread w, boolean worked) {
- boolean active = w.active;
- boolean inactivate = !worked & active;
- for (;;) {
- if (inactivate) {
- int c = runState;
- if (UNSAFE.compareAndSwapInt(this, runStateOffset,
- c, c - ONE_ACTIVE))
- inactivate = active = w.active = false;
- }
- int wc = workerCounts;
- if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
- if (!worked)
- eventSync(w);
- return;
+ else if (e > 0 && ws != null &&
+ (i = ((~e << 1) | 1) & SMASK) < ws.length &&
+ (w = ws[i]) != null &&
+ w.eventCount == (e | INT_SIGN)) {
+ if (U.compareAndSwapLong
+ (this, CTL, c, (((long)(w.nextWait & E_MASK)) |
+ ((long)(u + UAC_UNIT) << 32)))) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p); // release a waiting worker
+ break;
+ }
}
- if (!(inactivate |= active) && // must inactivate to suspend
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - ONE_RUNNING) &&
- !w.suspendAsSpare()) // false if trimmed
- return;
+ else
+ break;
}
}
/**
- * Adjusts counts and creates or resumes compensating threads for
- * a worker about to block on task joinMe, returning early if
- * joinMe becomes ready. First tries resuming an existing spare
- * (which usually also avoids any count adjustment), but must then
- * decrement running count to determine whether a new thread is
- * needed. See above for fuller explanation.
- */
- final void preJoin(ForkJoinTask> joinMe) {
- boolean dec = false; // true when running count decremented
- for (;;) {
- releaseWaiters(); // help other threads progress
-
- if (joinMe.status < 0) // surround spare search with done checks
- return;
- ForkJoinWorkerThread spare = null;
- for (ForkJoinWorkerThread w : workers) {
- if (w != null && w.isSuspended()) {
- spare = w;
- break;
+ * Tries to decrement active count (sometimes implicitly) and
+ * possibly release or create a compensating worker in preparation
+ * for blocking. Fails on contention or termination.
+ *
+ * @return true if the caller can block, else should recheck and retry
+ */
+ final boolean tryCompensate() {
+ WorkQueue[] ws; WorkQueue w; Thread p;
+ int pc = parallelism, e, u, ac, tc, i;
+ long c = ctl;
+
+ if ((e = (int)c) >= 0) {
+ if ((ac = ((u = (int)(c >>> 32)) >> UAC_SHIFT)) <= 0 &&
+ e != 0 && (ws = workQueues) != null &&
+ (i = ((~e << 1) | 1) & SMASK) < ws.length &&
+ (w = ws[i]) != null) {
+ if (w.eventCount == (e | INT_SIGN) &&
+ U.compareAndSwapLong
+ (this, CTL, c, ((long)(w.nextWait & E_MASK) |
+ (c & (AC_MASK|TC_MASK))))) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ return true; // release an idle worker
}
}
- if (joinMe.status < 0)
- return;
-
- if (spare != null && spare.tryUnsuspend()) {
- if (dec || joinMe.requestSignal() < 0) {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this,
- workerCountsOffset,
- c = workerCounts,
- c + ONE_RUNNING));
- } // else no net count change
- LockSupport.unpark(spare);
- return;
- }
-
- int wc = workerCounts; // decrement running count
- if (!dec && (wc & RUNNING_COUNT_MASK) != 0 &&
- (dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc -= ONE_RUNNING)) &&
- joinMe.requestSignal() < 0) { // cannot block
- int c; // back out
- do {} while (!UNSAFE.compareAndSwapInt(this,
- workerCountsOffset,
- c = workerCounts,
- c + ONE_RUNNING));
- return;
- }
-
- if (dec) {
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int pc = parallelism;
- int dc = pc - (wc & RUNNING_COUNT_MASK); // deficit count
- if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
- !maintainsParallelism)) ||
- tc >= maxPoolSize) // cannot add
- return;
- if (spare == null &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
- wc + (ONE_RUNNING|ONE_TOTAL))) {
+ else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
+ long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc))
+ return true; // no compensation needed
+ }
+ else if (tc + pc < MAX_ID) {
+ long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
addWorker();
- return;
+ return true; // create replacement
}
}
}
+ return false;
}
+ // Submissions
+
/**
- * Same idea as preJoin but with too many differing details to
- * integrate: There are no task-based signal counts, and only one
- * way to do the actual blocking. So for simplicity it is directly
- * incorporated into this method.
+ * Unless shutting down, adds the given task to a submission queue
+ * at submitter's current queue index. If no queue exists at the
+ * index, one is created unless pool lock is busy. If the queue
+ * and/or lock are busy, another index is randomly chosen.
*/
- final void doBlock(ManagedBlocker blocker, boolean maintainPar)
- throws InterruptedException {
- maintainPar &= maintainsParallelism; // override
- boolean dec = false;
- boolean done = false;
- for (;;) {
- releaseWaiters();
- if (done = blocker.isReleasable())
- break;
- ForkJoinWorkerThread spare = null;
- for (ForkJoinWorkerThread w : workers) {
- if (w != null && w.isSuspended()) {
- spare = w;
+ private void doSubmit(ForkJoinTask> task) {
+ if (task == null)
+ throw new NullPointerException();
+ Submitter s = submitters.get();
+ for (int r = s.seed;;) {
+ WorkQueue q; int k;
+ int rs = runState, m = rs & SMASK;
+ WorkQueue[] ws = workQueues;
+ if (rs < 0 || ws == null) // shutting down
+ throw new RejectedExecutionException();
+ if (ws.length > m && // k must be at index
+ ((q = ws[k = (r << 1) & m]) != null ||
+ (q = tryAddSharedQueue(k)) != null) &&
+ q.trySharedPush(task)) {
+ signalWork();
+ return;
+ }
+ r ^= r << 13; // xorshift seed to new position
+ r ^= r >>> 17;
+ if (((s.seed = r ^= r << 5) & m) == 0)
+ Thread.yield(); // occasionally yield if busy
+ }
+ }
+
+
+ // Scanning for tasks
+
+ /**
+ * Scans for and, if found, returns one task, else possibly
+ * inactivates the worker. This method operates on single reads of
+ * volatile state and is designed to be re-invoked continuously in
+ * part because it returns upon detecting inconsistencies,
+ * contention, or state changes that indicate possible success on
+ * re-invocation.
+ *
+ * The scan searches for tasks across queues, randomly selecting
+ * the first #queues probes, favoring steals 2:1 over submissions
+ * (by exploiting even/odd indexing), and then performing a
+ * circular sweep of all queues. The scan terminates upon either
+ * finding a non-empty queue, or completing a full sweep. If the
+ * worker is not inactivated, it takes and returns a task from
+ * this queue. On failure to find a task, we take one of the
+ * following actions, after which the caller will retry calling
+ * this method unless terminated.
+ *
+ * * If not a complete sweep, try to release a waiting worker. If
+ * the scan terminated because the worker is inactivated, then the
+ * released worker will often be the calling worker, and it can
+ * succeed obtaining a task on the next call. Or maybe it is
+ * another worker, but with same net effect. Releasing in other
+ * cases as well ensures that we have enough workers running.
+ *
+ * * If the caller has run a task since the the last empty scan,
+ * return (to allow rescan) if other workers are not also yet
+ * enqueued. Field WorkQueue.rescans counts down on each scan to
+ * ensure eventual inactivation, and occasional calls to
+ * Thread.yield to help avoid interference with more useful
+ * activities on the system.
+ *
+ * * If pool is terminating, terminate the worker
+ *
+ * * If not already enqueued, try to inactivate and enqueue the
+ * worker on wait queue.
+ *
+ * * If already enqueued and none of the above apply, either park
+ * awaiting signal, or if this is the most recent waiter and pool
+ * is quiescent, relay to idleAwaitWork to check for termination
+ * and possibly shrink pool.
+ *
+ * @param w the worker (via its WorkQueue)
+ * @return a task or null of none found
+ */
+ private final ForkJoinTask> scan(WorkQueue w) {
+ boolean swept = false; // true after full empty scan
+ WorkQueue[] ws; // volatile read order matters
+ int r = w.seed, ec = w.eventCount; // ec is negative if inactive
+ int rs = runState, m = rs & SMASK;
+ if ((ws = workQueues) != null && ws.length > m) {
+ ForkJoinTask> task = null;
+ for (int k = 0, j = -2 - m; ; ++j) {
+ WorkQueue q; int b;
+ if (j < 0) { // random probes while j negative
+ r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) | (j & 1);
+ } // worker (not submit) for odd j
+ else // cyclic scan when j >= 0
+ k += (m >>> 1) | 1; // step by half to reduce bias
+
+ if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
+ if (ec >= 0)
+ task = q.pollAt(b); // steal
break;
}
- }
- if (done = blocker.isReleasable())
- break;
- if (spare != null && spare.tryUnsuspend()) {
- if (dec) {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this,
- workerCountsOffset,
- c = workerCounts,
- c + ONE_RUNNING));
+ else if (j > m) {
+ if (rs == runState) // staleness check
+ swept = true;
+ break;
}
- LockSupport.unpark(spare);
- break;
}
- int wc = workerCounts;
- if (!dec && (wc & RUNNING_COUNT_MASK) != 0)
- dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc -= ONE_RUNNING);
- if (dec) {
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int pc = parallelism;
- int dc = pc - (wc & RUNNING_COUNT_MASK);
- if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
- !maintainPar)) ||
- tc >= maxPoolSize)
+ w.seed = r; // save seed for next scan
+ if (task != null)
+ return task;
+ }
+
+ // Decode ctl on empty scan
+ long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
+ if (!swept) { // try to release a waiter
+ WorkQueue v; Thread p;
+ if (e > 0 && a < 0 && ws != null &&
+ (v = ws[((~e << 1) | 1) & m]) != null &&
+ v.eventCount == (e | INT_SIGN) && U.compareAndSwapLong
+ (this, CTL, c, ((long)(v.nextWait & E_MASK) |
+ ((c + AC_UNIT) & (AC_MASK|TC_MASK))))) {
+ v.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = v.parker) != null)
+ U.unpark(p);
+ }
+ }
+ else if ((nr = w.rescans) > 0) { // continue rescanning
+ int ac = a + parallelism;
+ if ((w.rescans = (ac < nr) ? ac : nr - 1) > 0 && w.seed < 0 &&
+ w.eventCount == ec)
+ Thread.yield(); // 1 bit randomness for yield call
+ }
+ else if (e < 0) // pool is terminating
+ w.runState = -1;
+ else if (ec >= 0) { // try to enqueue
+ long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
+ w.nextWait = e;
+ w.eventCount = ec | INT_SIGN; // mark as inactive
+ if (!U.compareAndSwapLong(this, CTL, c, nc))
+ w.eventCount = ec; // back out on CAS failure
+ else if ((ns = w.nsteals) != 0) { // set rescans if ran task
+ if (a <= 0) // ... unless too many active
+ w.rescans = a + parallelism;
+ w.nsteals = 0;
+ w.totalSteals += ns;
+ }
+ }
+ else{ // already queued
+ if (parallelism == -a)
+ idleAwaitWork(w); // quiescent
+ if (w.eventCount == ec) {
+ Thread.interrupted(); // clear status
+ ForkJoinWorkerThread wt = w.owner;
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt; // emulate LockSupport.park
+ if (w.eventCount == ec) // recheck
+ U.park(false, 0L); // block
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ }
+ }
+ return null;
+ }
+
+ /**
+ * If inactivating worker w has caused pool to become quiescent,
+ * check for pool termination, and, so long as this is not the
+ * only worker, wait for event for up to SHRINK_RATE nanosecs On
+ * timeout, if ctl has not changed, terminate the worker, which
+ * will in turn wake up another worker to possibly repeat this
+ * process.
+ *
+ * @param w the calling worker
+ */
+ private void idleAwaitWork(WorkQueue w) {
+ long c; int nw, ec;
+ if (!tryTerminate(false) &&
+ (int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
+ (ec = w.eventCount) == ((int)c | INT_SIGN) &&
+ (nw = w.nextWait) != 0) {
+ long nc = ((long)(nw & E_MASK) | // ctl to restore on timeout
+ ((c + AC_UNIT) & AC_MASK) | (c & TC_MASK));
+ ForkJoinTask.helpExpungeStaleExceptions(); // help clean
+ ForkJoinWorkerThread wt = w.owner;
+ while (ctl == c) {
+ long startTime = System.nanoTime();
+ Thread.interrupted(); // timed variant of version in scan()
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt;
+ if (ctl == c)
+ U.park(false, SHRINK_RATE);
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ if (ctl != c)
break;
- if (spare == null &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
- wc + (ONE_RUNNING|ONE_TOTAL))){
- addWorker();
+ if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
+ U.compareAndSwapLong(this, CTL, c, nc)) {
+ w.runState = -1; // shrink
+ w.eventCount = (ec + E_SEQ) | E_MASK;
break;
}
}
}
-
- try {
- if (!done)
- do {} while (!blocker.isReleasable() && !blocker.block());
- } finally {
- if (dec) {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this,
- workerCountsOffset,
- c = workerCounts,
- c + ONE_RUNNING));
- }
- }
}
/**
- * Unless there are not enough other running threads, adjusts
- * counts for a a worker in performing helpJoin that cannot find
- * any work, so that this worker can now block.
- *
- * @return true if worker may block
- */
- final boolean preBlockHelpingJoin(ForkJoinTask> joinMe) {
- while (joinMe.status >= 0) {
- releaseWaiters(); // help other threads progress
-
- // if a spare exists, resume it to maintain parallelism level
- if ((workerCounts & RUNNING_COUNT_MASK) <= parallelism) {
- ForkJoinWorkerThread spare = null;
- for (ForkJoinWorkerThread w : workers) {
- if (w != null && w.isSuspended()) {
- spare = w;
- break;
+ * Tries to locate and execute tasks for a stealer of the given
+ * task, or in turn one of its stealers, Traces currentSteal ->
+ * currentJoin links looking for a thread working on a descendant
+ * of the given task and with a non-empty queue to steal back and
+ * execute tasks from. The first call to this method upon a
+ * waiting join will often entail scanning/search, (which is OK
+ * because the joiner has nothing better to do), but this method
+ * leaves hints in workers to speed up subsequent calls. The
+ * implementation is very branchy to cope with potential
+ * inconsistencies or loops encountering chains that are stale,
+ * unknown, or of length greater than MAX_HELP_DEPTH links. All
+ * of these cases are dealt with by just retrying by caller.
+ *
+ * @param joiner the joining worker
+ * @param task the task to join
+ * @return true if found or ran a task (and so is immediately retryable)
+ */
+ final boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask> task) {
+ ForkJoinTask> subtask; // current target
+ boolean progress = false;
+ int depth = 0; // current chain depth
+ int m = runState & SMASK;
+ WorkQueue[] ws = workQueues;
+
+ if (ws != null && ws.length > m && (subtask = task).status >= 0) {
+ outer:for (WorkQueue j = joiner;;) {
+ // Try to find the stealer of subtask, by first using hint
+ WorkQueue stealer = null;
+ WorkQueue v = ws[j.stealHint & m];
+ if (v != null && v.currentSteal == subtask)
+ stealer = v;
+ else {
+ for (int i = 1; i <= m; i += 2) {
+ if ((v = ws[i]) != null && v.currentSteal == subtask) {
+ stealer = v;
+ j.stealHint = i; // save hint
+ break;
+ }
}
+ if (stealer == null)
+ break;
}
- if (joinMe.status < 0)
- break;
- if (spare != null) {
- if (spare.tryUnsuspend()) {
- boolean canBlock = true;
- if (joinMe.requestSignal() < 0) {
- canBlock = false; // already done
- int c;
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
+
+ for (WorkQueue q = stealer;;) { // Try to help stealer
+ ForkJoinTask> t; int b;
+ if (task.status < 0)
+ break outer;
+ if ((b = q.base) - q.top < 0) {
+ progress = true;
+ if (subtask.status < 0)
+ break outer; // stale
+ if ((t = q.pollAt(b)) != null) {
+ stealer.stealHint = joiner.poolIndex;
+ joiner.runSubtask(t);
}
- LockSupport.unpark(spare);
- return canBlock;
}
- continue; // recheck -- another spare may exist
+ else { // empty - try to descend to find stealer's stealer
+ ForkJoinTask> next = stealer.currentJoin;
+ if (++depth == MAX_HELP_DEPTH || subtask.status < 0 ||
+ next == null || next == subtask)
+ break outer; // max depth, stale, dead-end, cyclic
+ subtask = next;
+ j = stealer;
+ break;
+ }
}
}
-
- int wc = workerCounts; // reread to shorten CAS window
- int rc = wc & RUNNING_COUNT_MASK;
- if (rc <= 2) // keep this and at most one other thread alive
- break;
-
- if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - ONE_RUNNING)) {
- if (joinMe.requestSignal() >= 0)
- return true;
- int c; // back out
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
- break;
- }
}
- return false;
+ return progress;
}
/**
- * Possibly initiates and/or completes termination.
+ * If task is at base of some steal queue, steals and executes it.
*
- * @param now if true, unconditionally terminate, else only
- * if shutdown and empty queue and no active workers
- * @return true if now terminating or terminated
+ * @param joiner the joining worker
+ * @param task the task
*/
- private boolean tryTerminate(boolean now) {
- if (now)
- advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN
- else if (runState < SHUTDOWN ||
- !submissionQueue.isEmpty() ||
- (runState & ACTIVE_COUNT_MASK) != 0)
- return false;
+ final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask> task) {
+ WorkQueue[] ws;
+ int m = runState & SMASK;
+ if ((ws = workQueues) != null && ws.length > m) {
+ for (int j = 1; j <= m && task.status >= 0; j += 2) {
+ WorkQueue q = ws[j];
+ if (q != null && q.pollFor(task)) {
+ joiner.runSubtask(task);
+ break;
+ }
+ }
+ }
+ }
- if (advanceRunLevel(TERMINATING))
- startTerminating();
+ /**
+ * Returns a non-empty steal queue, if one is found during a random,
+ * then cyclic scan, else null. This method must be retried by
+ * caller if, by the time it tries to use the queue, it is empty.
+ */
+ private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
+ int r = w.seed; // Same idea as scan(), but ignoring submissions
+ for (WorkQueue[] ws;;) {
+ int m = runState & SMASK;
+ if ((ws = workQueues) == null)
+ return null;
+ if (ws.length > m) {
+ WorkQueue q;
+ for (int n = m << 2, k = r, j = -n;;) {
+ r ^= r << 13; r ^= r >>> 17; r ^= r << 5;
+ if ((q = ws[(k | 1) & m]) != null && q.base - q.top < 0) {
+ w.seed = r;
+ return q;
+ }
+ else if (j > n)
+ return null;
+ else
+ k = (j++ < 0) ? r : k + ((m >>> 1) | 1);
- // Finish now if all threads terminated; else in some subsequent call
- if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
- advanceRunLevel(TERMINATED);
- terminationLatch.countDown();
+ }
+ }
}
- return true;
}
/**
- * Actions on transition to TERMINATING
- */
- private void startTerminating() {
- // Clear out and cancel submissions, ignoring exceptions
- ForkJoinTask> task;
- while ((task = submissionQueue.poll()) != null) {
- try {
- task.cancel(false);
- } catch (Throwable ignore) {
+ * Runs tasks until {@code isQuiescent()}. We piggyback on
+ * active count ctl maintenance, but rather than blocking
+ * when tasks cannot be found, we rescan until all others cannot
+ * find tasks either.
+ */
+ final void helpQuiescePool(WorkQueue w) {
+ for (boolean active = true;;) {
+ w.runLocalTasks(); // exhaust local queue
+ WorkQueue q = findNonEmptyStealQueue(w);
+ if (q != null) {
+ ForkJoinTask> t;
+ if (!active) { // re-establish active count
+ long c;
+ active = true;
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ }
+ if ((t = q.poll()) != null)
+ w.runSubtask(t);
}
- }
- // Propagate run level
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- w.shutdown(); // also resumes suspended workers
- }
- // Ensure no straggling local tasks
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- w.cancelTasks();
- }
- // Wake up idle workers
- advanceEventCount();
- releaseWaiters();
- // Unstick pending joins
- for (ForkJoinWorkerThread w : workers) {
- if (w != null && !w.isTerminated()) {
- try {
- w.interrupt();
- } catch (SecurityException ignore) {
+ 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;
}
}
}
}
- // misc support for ForkJoinWorkerThread
+ /**
+ * Gets and removes a local or stolen task for the given worker
+ *
+ * @return a task, if available
+ */
+ final ForkJoinTask> nextTaskFor(WorkQueue w) {
+ for (ForkJoinTask> t;;) {
+ WorkQueue q;
+ if ((t = w.nextLocalTask()) != null)
+ return t;
+ if ((q = findNonEmptyStealQueue(w)) == null)
+ return null;
+ if ((t = q.poll()) != null)
+ return t;
+ }
+ }
/**
- * Returns pool number
+ * Returns the approximate (non-atomic) number of idle threads per
+ * active thread to offset steal queue size for method
+ * ForkJoinTask.getSurplusQueuedTaskCount().
*/
- final int getPoolNumber() {
- return poolNumber;
+ final int idlePerActive() {
+ // Approximate at powers of two for small values, saturate past 4
+ int p = parallelism;
+ int a = p + (int)(ctl >> AC_SHIFT);
+ return (a > (p >>>= 1) ? 0 :
+ a > (p >>>= 1) ? 1 :
+ a > (p >>>= 1) ? 2 :
+ a > (p >>>= 1) ? 4 :
+ 8);
+ }
+
+ // Termination
+
+ /**
+ * Sets SHUTDOWN bit of runState under lock
+ */
+ private void enableShutdown() {
+ ReentrantLock lock = this.lock;
+ if (runState >= 0) {
+ lock.lock(); // don't need try/finally
+ runState |= SHUTDOWN;
+ lock.unlock();
+ }
}
/**
- * Accumulates steal count from a worker, clearing
- * the worker's value
+ * Possibly initiates and/or completes termination. Upon
+ * termination, cancels all queued tasks and then
+ *
+ * @param now if true, unconditionally terminate, else only
+ * if no work and no active workers
+ * @return true if now terminating or terminated
*/
- final void accumulateStealCount(ForkJoinWorkerThread w) {
- int sc = w.stealCount;
- if (sc != 0) {
- long c;
- w.stealCount = 0;
- do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
- c = stealCount, c + sc));
+ private boolean tryTerminate(boolean now) {
+ for (long c;;) {
+ if (((c = ctl) & STOP_BIT) != 0) { // already terminating
+ if ((short)(c >>> TC_SHIFT) == -parallelism) {
+ ReentrantLock lock = this.lock; // signal when no workers
+ lock.lock(); // don't need try/finally
+ termination.signalAll(); // signal when 0 workers
+ lock.unlock();
+ }
+ return true;
+ }
+ if (!now) {
+ if ((int)(c >> AC_SHIFT) != -parallelism || runState >= 0 ||
+ hasQueuedSubmissions())
+ return false;
+ // Check for unqueued inactive workers. One pass suffices.
+ WorkQueue[] ws = workQueues; WorkQueue w;
+ if (ws != null) {
+ int n = ws.length;
+ for (int i = 1; i < n; i += 2) {
+ if ((w = ws[i]) != null && w.eventCount >= 0)
+ return false;
+ }
+ }
+ }
+ if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT))
+ startTerminating();
}
}
/**
- * Returns the approximate (non-atomic) number of idle threads per
- * active thread.
+ * Initiates 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.
*/
- final int idlePerActive() {
- int ac = runState; // no mask -- artifically boosts during shutdown
- int pc = parallelism; // use targeted parallelism, not rc
- // Use exact results for small values, saturate past 4
- return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
+ private void startTerminating() {
+ for (int pass = 0; pass < 3; ++pass) {
+ WorkQueue[] ws = workQueues;
+ if (ws != null) {
+ WorkQueue w; Thread wt;
+ int n = ws.length;
+ for (int j = 0; j < n; ++j) {
+ if ((w = ws[j]) != null) {
+ w.runState = -1;
+ if (pass > 0) {
+ w.cancelAll();
+ if (pass > 1 && (wt = w.owner) != null &&
+ !wt.isInterrupted()) {
+ try {
+ wt.interrupt();
+ } catch (SecurityException ignore) {
+ }
+ }
+ }
+ }
+ }
+ // Wake up workers parked on event queue
+ int i, e; long c; Thread p;
+ while ((i = ((~(e = (int)(c = ctl)) << 1) | 1) & SMASK) < n &&
+ (w = ws[i]) != null &&
+ w.eventCount == (e | INT_SIGN)) {
+ long nc = ((long)(w.nextWait & E_MASK) |
+ ((c + AC_UNIT) & AC_MASK) |
+ (c & (TC_MASK|STOP_BIT)));
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ }
+ }
+ }
+ }
}
- // Public and protected methods
+ // Exported methods
// Constructors
/**
* Creates a {@code ForkJoinPool} with parallelism equal to {@link
- * java.lang.Runtime#availableProcessors}, and using the {@linkplain
- * #defaultForkJoinWorkerThreadFactory default thread factory}.
+ * 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
@@ -1147,13 +1894,14 @@ public class ForkJoinPool extends Abstra
*/
public ForkJoinPool() {
this(Runtime.getRuntime().availableProcessors(),
- defaultForkJoinWorkerThreadFactory);
+ defaultForkJoinWorkerThreadFactory, null, false);
}
/**
* Creates a {@code ForkJoinPool} with the indicated parallelism
- * level and using the {@linkplain
- * #defaultForkJoinWorkerThreadFactory default thread factory}.
+ * level, the {@linkplain
+ * #defaultForkJoinWorkerThreadFactory default thread factory},
+ * no UncaughtExceptionHandler, and non-async LIFO processing mode.
*
* @param parallelism the parallelism level
* @throws IllegalArgumentException if parallelism less than or
@@ -1164,31 +1912,25 @@ public class ForkJoinPool extends Abstra
* java.lang.RuntimePermission}{@code ("modifyThread")}
*/
public ForkJoinPool(int parallelism) {
- this(parallelism, defaultForkJoinWorkerThreadFactory);
- }
-
- /**
- * Creates a {@code ForkJoinPool} with parallelism equal to {@link
- * java.lang.Runtime#availableProcessors}, and using the given
- * thread factory.
- *
- * @param factory the factory for creating new threads
- * @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(ForkJoinWorkerThreadFactory factory) {
- this(Runtime.getRuntime().availableProcessors(), factory);
+ this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
}
/**
- * Creates a {@code ForkJoinPool} with the given parallelism and
- * thread factory.
+ * Creates a {@code ForkJoinPool} with the given parameters.
*
- * @param parallelism the parallelism level
- * @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 the factory is null
@@ -1197,60 +1939,55 @@ public class ForkJoinPool extends Abstra
* because it does not hold {@link
* java.lang.RuntimePermission}{@code ("modifyThread")}
*/
- public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
+ public ForkJoinPool(int parallelism,
+ ForkJoinWorkerThreadFactory factory,
+ Thread.UncaughtExceptionHandler handler,
+ boolean asyncMode) {
checkPermission();
if (factory == null)
throw new NullPointerException();
- if (parallelism <= 0 || parallelism > MAX_THREADS)
+ if (parallelism <= 0 || parallelism > MAX_ID)
throw new IllegalArgumentException();
- this.poolNumber = poolNumberGenerator.incrementAndGet();
- int arraySize = initialArraySizeFor(parallelism);
this.parallelism = parallelism;
this.factory = factory;
- this.maxPoolSize = MAX_THREADS;
- this.maintainsParallelism = true;
- this.workers = new ForkJoinWorkerThread[arraySize];
- this.submissionQueue = new LinkedTransferQueue>();
- this.workerLock = new ReentrantLock();
- this.terminationLatch = new CountDownLatch(1);
- // Start first worker; remaining workers added upon first submission
- workerCounts = ONE_RUNNING | ONE_TOTAL;
- addWorker();
- }
-
- /**
- * Returns initial power of two size for workers array.
- * @param pc the initial parallelism level
- */
- private static int initialArraySizeFor(int pc) {
- // See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
- int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
- size |= size >>> 1;
- size |= size >>> 2;
- size |= size >>> 4;
- size |= size >>> 8;
- return size + 1;
+ this.ueh = handler;
+ this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
+ this.nextPoolIndex = 1;
+ long np = (long)(-parallelism); // offset ctl counts
+ this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
+ // initialize workQueues array with room for 2*parallelism if possible
+ int n = parallelism << 1;
+ if (n >= MAX_ID)
+ n = MAX_ID;
+ else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
+ n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8;
+ }
+ this.workQueues = new WorkQueue[(n + 1) << 1];
+ ReentrantLock lck = this.lock = new ReentrantLock();
+ this.termination = lck.newCondition();
+ this.stealCount = new AtomicLong();
+ this.nextWorkerNumber = new AtomicInteger();
+ StringBuilder sb = new StringBuilder("ForkJoinPool-");
+ sb.append(poolNumberGenerator.incrementAndGet());
+ sb.append("-worker-");
+ this.workerNamePrefix = sb.toString();
+ // Create initial submission queue
+ WorkQueue sq = tryAddSharedQueue(0);
+ if (sq != null)
+ sq.growArray(false);
}
// Execution methods
/**
- * Common code for execute, invoke and submit
- */
- private void doSubmit(ForkJoinTask task) {
- if (task == null)
- throw new NullPointerException();
- if (runState >= SHUTDOWN)
- throw new RejectedExecutionException();
- submissionQueue.offer(task);
- advanceEventCount();
- releaseWaiters();
- if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
- ensureEnoughTotalWorkers();
- }
-
- /**
* 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
@@ -1283,6 +2020,8 @@ public class ForkJoinPool extends Abstra
* 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;
@@ -1292,11 +2031,27 @@ public class ForkJoinPool extends Abstra
}
/**
+ * Submits a ForkJoinTask for execution.
+ *
+ * @param task the task to submit
+ * @return the task
+ * @throws NullPointerException if the task is null
+ * @throws RejectedExecutionException if the task cannot be
+ * scheduled for execution
+ */
+ public ForkJoinTask submit(ForkJoinTask task) {
+ doSubmit(task);
+ return task;
+ }
+
+ /**
* @throws NullPointerException if the task is null
* @throws RejectedExecutionException if the task cannot be
* scheduled for execution
*/
public ForkJoinTask submit(Callable task) {
+ if (task == null)
+ throw new NullPointerException();
ForkJoinTask job = ForkJoinTask.adapt(task);
doSubmit(job);
return job;
@@ -1308,6 +2063,8 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public ForkJoinTask submit(Runnable task, T result) {
+ if (task == null)
+ throw new NullPointerException();
ForkJoinTask job = ForkJoinTask.adapt(task, result);
doSubmit(job);
return job;
@@ -1319,6 +2076,8 @@ public class ForkJoinPool extends Abstra
* 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;
@@ -1329,20 +2088,6 @@ public class ForkJoinPool extends Abstra
}
/**
- * Submits a ForkJoinTask for execution.
- *
- * @param task the task to submit
- * @return the task
- * @throws NullPointerException if the task is null
- * @throws RejectedExecutionException if the task cannot be
- * scheduled for execution
- */
- public ForkJoinTask submit(ForkJoinTask task) {
- doSubmit(task);
- return task;
- }
-
- /**
* @throws NullPointerException {@inheritDoc}
* @throws RejectedExecutionException {@inheritDoc}
*/
@@ -1384,178 +2129,28 @@ public class ForkJoinPool extends Abstra
* @return the handler, or {@code null} if none
*/
public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
- workerCountReadFence();
return ueh;
}
/**
- * 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 {@code 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();
- workerCountReadFence();
- Thread.UncaughtExceptionHandler old = ueh;
- if (h != old) {
- ueh = h;
- workerCountWriteFence();
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- w.setUncaughtExceptionHandler(h);
- }
- }
- return old;
- }
-
- /**
- * Sets the target 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")}
- */
- public void setParallelism(int parallelism) {
- checkPermission();
- if (parallelism <= 0 || parallelism > maxPoolSize)
- throw new IllegalArgumentException();
- workerCountReadFence();
- int pc = this.parallelism;
- if (pc != parallelism) {
- this.parallelism = parallelism;
- workerCountWriteFence();
- // Release spares. If too many, some will die after re-suspend
- for (ForkJoinWorkerThread w : workers) {
- if (w != null && w.tryUnsuspend()) {
- updateRunningCount(1);
- LockSupport.unpark(w);
- }
- }
- ensureEnoughTotalWorkers();
- advanceEventCount();
- releaseWaiters(); // force config recheck by existing workers
- }
- }
-
- /**
* Returns the targeted parallelism level of this pool.
*
* @return the targeted parallelism level of this pool
*/
public int getParallelism() {
- // workerCountReadFence(); // inlined below
- int ignore = workerCounts;
return parallelism;
}
/**
* Returns the number of worker threads that have started but not
- * yet terminated. This result returned by this method may differ
+ * 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 workerCounts >>> TOTAL_COUNT_SHIFT;
- }
-
- /**
- * Returns the maximum number of threads allowed to exist in the
- * pool. Unless set using {@link #setMaximumPoolSize}, the
- * maximum is an implementation-defined value designed only to
- * prevent runaway growth.
- *
- * @return the maximum
- */
- public int getMaximumPoolSize() {
- workerCountReadFence();
- return maxPoolSize;
- }
-
- /**
- * Sets the maximum number of threads allowed to exist in the
- * pool. The given value should normally be greater than or equal
- * to the {@link #getParallelism parallelism} level. Setting this
- * value has no effect on current pool size. It controls
- * construction of new threads. The use of this method may cause
- * tasks that intrinsically require extra threads for dependent
- * computations to indefinitely stall. If you are instead trying
- * to minimize internal thread creation, consider setting {@link
- * #setMaintainsParallelism} as false.
- *
- * @throws IllegalArgumentException if negative or greater than
- * internal implementation limit
- */
- public void setMaximumPoolSize(int newMax) {
- if (newMax < 0 || newMax > MAX_THREADS)
- throw new IllegalArgumentException();
- maxPoolSize = newMax;
- workerCountWriteFence();
- }
-
- /**
- * Returns {@code 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 {@code true} if maintains parallelism
- */
- public boolean getMaintainsParallelism() {
- workerCountReadFence();
- 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 {@code true} to maintain parallelism
- */
- public void setMaintainsParallelism(boolean enable) {
- maintainsParallelism = enable;
- workerCountWriteFence();
- }
-
- /**
- * 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 the pool is quiescent, and
- * typically only before any tasks are submitted. The effects of
- * invocations at other times may be unpredictable.
- *
- * @param async if {@code true}, use locally FIFO scheduling
- * @return the previous mode
- * @see #getAsyncMode
- */
- public boolean setAsyncMode(boolean async) {
- workerCountReadFence();
- boolean oldMode = locallyFifo;
- if (oldMode != async) {
- locallyFifo = async;
- workerCountWriteFence();
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- w.setAsyncMode(async);
- }
- }
- return oldMode;
+ return parallelism + (short)(ctl >>> TC_SHIFT);
}
/**
@@ -1563,11 +2158,9 @@ public class ForkJoinPool extends Abstra
* scheduling mode for forked tasks that are never joined.
*
* @return {@code true} if this pool uses async mode
- * @see #setAsyncMode
*/
public boolean getAsyncMode() {
- workerCountReadFence();
- return locallyFifo;
+ return localMode != 0;
}
/**
@@ -1579,7 +2172,21 @@ public class ForkJoinPool extends Abstra
* @return the number of worker threads
*/
public int getRunningThreadCount() {
- return workerCounts & RUNNING_COUNT_MASK;
+ int rc = 0;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ int n = ws.length;
+ for (int i = 1; i < n; i += 2) {
+ Thread.State s; ForkJoinWorkerThread wt;
+ if ((w = ws[i]) != null && (wt = w.owner) != null &&
+ w.eventCount >= 0 &&
+ (s = wt.getState()) != Thread.State.BLOCKED &&
+ s != Thread.State.WAITING &&
+ s != Thread.State.TIMED_WAITING)
+ ++rc;
+ }
+ }
+ return rc;
}
/**
@@ -1590,7 +2197,8 @@ public class ForkJoinPool extends Abstra
* @return the number of active threads
*/
public int getActiveThreadCount() {
- return runState & ACTIVE_COUNT_MASK;
+ int r = parallelism + (int)(ctl >> AC_SHIFT);
+ return (r <= 0) ? 0 : r; // suppress momentarily negative values
}
/**
@@ -1605,7 +2213,7 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if all threads are currently idle
*/
public boolean isQuiescent() {
- return (runState & ACTIVE_COUNT_MASK) == 0;
+ return (int)(ctl >> AC_SHIFT) + parallelism == 0;
}
/**
@@ -1620,7 +2228,16 @@ public class ForkJoinPool extends Abstra
* @return the number of steals
*/
public long getStealCount() {
- return stealCount;
+ long count = stealCount.get();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ int n = ws.length;
+ for (int i = 1; i < n; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.totalSteals;
+ }
+ }
+ return count;
}
/**
@@ -1635,22 +2252,35 @@ public class ForkJoinPool extends Abstra
*/
public long getQueuedTaskCount() {
long count = 0;
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- count += w.getQueueSize();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ int n = ws.length;
+ for (int i = 1; i < n; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
+ }
}
return count;
}
/**
* Returns an estimate of the number of tasks submitted to this
- * pool that have not yet begun executing. This method takes time
- * proportional to the number of submissions.
+ * 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) {
+ int n = ws.length;
+ for (int i = 0; i < n; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
+ }
+ }
+ return count;
}
/**
@@ -1660,7 +2290,15 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if there are any queued submissions
*/
public boolean hasQueuedSubmissions() {
- return !submissionQueue.isEmpty();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ int n = ws.length;
+ for (int i = 0; i < n; i += 2) {
+ if ((w = ws[i]) != null && w.queueSize() != 0)
+ return true;
+ }
+ }
+ return false;
}
/**
@@ -1671,7 +2309,15 @@ public class ForkJoinPool extends Abstra
* @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) {
+ int n = ws.length;
+ for (int i = 0; i < n; i += 2) {
+ if ((w = ws[i]) != null && (t = w.poll()) != null)
+ return t;
+ }
+ }
+ return null;
}
/**
@@ -1692,12 +2338,20 @@ public class ForkJoinPool extends Abstra
* @return the number of elements transferred
*/
protected int drainTasksTo(Collection super ForkJoinTask>> c) {
- int n = submissionQueue.drainTo(c);
- for (ForkJoinWorkerThread w : workers) {
- if (w != null)
- n += w.drainTasksTo(c);
+ int count = 0;
+ WorkQueue[] ws; WorkQueue w; ForkJoinTask> t;
+ if ((ws = workQueues) != null) {
+ int n = ws.length;
+ for (int i = 0; i < n; ++i) {
+ if ((w = ws[i]) != null) {
+ while ((t = w.poll()) != null) {
+ c.add(t);
+ ++count;
+ }
+ }
+ }
}
- return n;
+ return count;
}
/**
@@ -1711,14 +2365,20 @@ public class ForkJoinPool extends Abstra
long st = getStealCount();
long qt = getQueuedTaskCount();
long qs = getQueuedSubmissionCount();
- int wc = workerCounts;
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int rc = wc & RUNNING_COUNT_MASK;
+ int rc = getRunningThreadCount();
int pc = parallelism;
- int rs = runState;
- int ac = rs & ACTIVE_COUNT_MASK;
+ long c = ctl;
+ int tc = pc + (short)(c >>> TC_SHIFT);
+ int ac = pc + (int)(c >> AC_SHIFT);
+ if (ac < 0) // ignore transient negative
+ ac = 0;
+ String level;
+ if ((c & STOP_BIT) != 0)
+ level = (tc == 0) ? "Terminated" : "Terminating";
+ else
+ level = runState < 0 ? "Shutting down" : "Running";
return super.toString() +
- "[" + runLevelToString(rs) +
+ "[" + level +
", parallelism = " + pc +
", size = " + tc +
", active = " + ac +
@@ -1729,13 +2389,6 @@ public class ForkJoinPool extends Abstra
"]";
}
- private static String runLevelToString(int s) {
- return ((s & TERMINATED) != 0 ? "Terminated" :
- ((s & TERMINATING) != 0 ? "Terminating" :
- ((s & SHUTDOWN) != 0 ? "Shutting down" :
- "Running")));
- }
-
/**
* Initiates an orderly shutdown in which previously submitted
* tasks are executed, but no new tasks will be accepted.
@@ -1750,7 +2403,7 @@ public class ForkJoinPool extends Abstra
*/
public void shutdown() {
checkPermission();
- advanceRunLevel(SHUTDOWN);
+ enableShutdown();
tryTerminate(false);
}
@@ -1772,6 +2425,7 @@ public class ForkJoinPool extends Abstra
*/
public List shutdownNow() {
checkPermission();
+ enableShutdown();
tryTerminate(true);
return Collections.emptyList();
}
@@ -1782,7 +2436,9 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if all tasks have completed following shut down
*/
public boolean isTerminated() {
- return runState >= TERMINATED;
+ long c = ctl;
+ return ((c & STOP_BIT) != 0L &&
+ (short)(c >>> TC_SHIFT) == -parallelism);
}
/**
@@ -1790,13 +2446,18 @@ public class ForkJoinPool extends Abstra
* 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, causing this executor not
- * to properly terminate.
+ * 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 but not yet terminated
*/
public boolean isTerminating() {
- return (runState & (TERMINATING|TERMINATED)) == TERMINATING;
+ long c = ctl;
+ return ((c & STOP_BIT) != 0L &&
+ (short)(c >>> TC_SHIFT) != -parallelism);
}
/**
@@ -1805,7 +2466,7 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if this pool has been shut down
*/
public boolean isShutdown() {
- return runState >= SHUTDOWN;
+ return runState < 0;
}
/**
@@ -1821,18 +2482,39 @@ public class ForkJoinPool extends Abstra
*/
public boolean awaitTermination(long timeout, TimeUnit unit)
throws InterruptedException {
- return terminationLatch.await(timeout, unit);
+ long nanos = unit.toNanos(timeout);
+ final ReentrantLock lock = this.lock;
+ lock.lock();
+ try {
+ for (;;) {
+ if (isTerminated())
+ return true;
+ if (nanos <= 0)
+ return false;
+ nanos = termination.awaitNanos(nanos);
+ }
+ } finally {
+ lock.unlock();
+ }
}
/**
* Interface for extending managed parallelism for tasks running
* 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).
+ *
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:
@@ -1850,6 +2532,26 @@ 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 {
/**
@@ -1873,14 +2575,7 @@ public class ForkJoinPool extends Abstra
* Blocks in accord with the given blocker. If the current thread
* is a {@link 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 {@code 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.
+ * ensure sufficient parallelism while the current thread is blocked.
*
*
If the caller is not a {@link ForkJoinTask}, this method is
* behaviorally equivalent to
@@ -1894,29 +2589,24 @@ public class ForkJoinPool extends Abstra
* first be expanded to ensure parallelism, and later adjusted.
*
* @param blocker the blocker
- * @param maintainParallelism if {@code 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();
- if (t instanceof ForkJoinWorkerThread)
- ((ForkJoinWorkerThread) t).pool.
- doBlock(blocker, maintainParallelism);
- else
- awaitBlocker(blocker);
- }
-
- /**
- * Performs Non-FJ blocking
- */
- private static void awaitBlocker(ManagedBlocker blocker)
- throws InterruptedException {
- do {} while (!blocker.isReleasable() && !blocker.block());
+ ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
+ ((ForkJoinWorkerThread)t).pool : null);
+ while (!blocker.isReleasable()) {
+ if (p == null || p.tryCompensate()) {
+ try {
+ do {} while (!blocker.isReleasable() && !blocker.block());
+ } finally {
+ if (p != null)
+ p.incrementActiveCount();
+ }
+ break;
+ }
+ }
}
// AbstractExecutorService overrides. These rely on undocumented
@@ -1932,28 +2622,29 @@ public class ForkJoinPool extends Abstra
}
// Unsafe mechanics
-
- private static final sun.misc.Unsafe UNSAFE = getUnsafe();
- private static final long workerCountsOffset =
- objectFieldOffset("workerCounts", ForkJoinPool.class);
- private static final long runStateOffset =
- objectFieldOffset("runState", ForkJoinPool.class);
- private static final long eventCountOffset =
- objectFieldOffset("eventCount", ForkJoinPool.class);
- private static final long eventWaitersOffset =
- objectFieldOffset("eventWaiters",ForkJoinPool.class);
- private static final long stealCountOffset =
- objectFieldOffset("stealCount",ForkJoinPool.class);
-
-
- private static long objectFieldOffset(String field, Class> klazz) {
+ private static final sun.misc.Unsafe U;
+ private static final long CTL;
+ private static final long RUNSTATE;
+ private static final long PARKBLOCKER;
+
+ static {
+ poolNumberGenerator = new AtomicInteger();
+ modifyThreadPermission = new RuntimePermission("modifyThread");
+ defaultForkJoinWorkerThreadFactory =
+ new DefaultForkJoinWorkerThreadFactory();
+ int s;
try {
- return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
- } catch (NoSuchFieldException e) {
- // Convert Exception to corresponding Error
- NoSuchFieldError error = new NoSuchFieldError(field);
- error.initCause(e);
- throw error;
+ U = getUnsafe();
+ Class> k = ForkJoinPool.class;
+ Class> tk = Thread.class;
+ CTL = U.objectFieldOffset
+ (k.getDeclaredField("ctl"));
+ RUNSTATE = U.objectFieldOffset
+ (k.getDeclaredField("runState"));
+ PARKBLOCKER = U.objectFieldOffset
+ (tk.getDeclaredField("parkBlocker"));
+ } catch (Exception e) {
+ throw new Error(e);
}
}
@@ -1984,4 +2675,5 @@ public class ForkJoinPool extends Abstra
}
}
}
+
}