--- jsr166/src/jsr166y/ForkJoinPool.java 2010/07/07 19:52:31 1.57
+++ jsr166/src/jsr166y/ForkJoinPool.java 2012/11/14 19:05:03 1.142
@@ -1,22 +1,23 @@
/*
* 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.concurrent.atomic.AtomicInteger;
-import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.AbstractExecutorService;
+import java.util.concurrent.Callable;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Future;
+import java.util.concurrent.RejectedExecutionException;
+import java.util.concurrent.RunnableFuture;
+import java.util.concurrent.TimeUnit;
/**
* An {@link ExecutorService} for running {@link ForkJoinTask}s.
@@ -27,21 +28,31 @@ import java.util.concurrent.CountDownLat
*
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). When setting asyncMode to true in
- * constructors, {@code ForkJoinPool}s may also be appropriate for use
- * with event-style tasks that are never joined.
+ * 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 static {@link #commonPool} is available and appropriate for
+ * most applications. The common pool is used by any ForkJoinTask that
+ * is not explicitly submitted to a specified pool. Using the common
+ * pool normally reduces resource usage (its threads are slowly
+ * reclaimed during periods of non-use, and reinstated upon subsequent
+ * use).
*
- *
A {@code ForkJoinPool} is constructed with a given target
- * parallelism level; by default, equal to the number of available
- * processors. The pool attempts to maintain enough active (or
- * available) threads by dynamically adding, suspending, or resuming
- * internal worker threads, even if some tasks are stalled waiting to
- * join others. However, no such adjustments are guaranteed in the
- * face of blocked IO or other unmanaged synchronization. The nested
- * {@link ManagedBlocker} interface enables extension of the kinds of
+ *
For applications that require separate or custom pools, a {@code
+ * ForkJoinPool} may be constructed with a given target parallelism
+ * level; by default, equal to the number of available processors. The
+ * pool attempts to maintain enough active (or available) threads by
+ * dynamically adding, suspending, or resuming internal worker
+ * threads, even if some tasks are stalled waiting to join
+ * others. However, no such adjustments are guaranteed in the face of
+ * blocked IO or other unmanaged synchronization. The nested {@link
+ * ManagedBlocker} interface enables extension of the kinds of
* synchronization accommodated.
*
*
In addition to execution and lifecycle control methods, this
@@ -52,19 +63,16 @@ import java.util.concurrent.CountDownLat
* convenient form for informal monitoring.
*
*
As is the case with other ExecutorServices, there are three
- * main task execution methods summarized in the follwoing
- * table. These are designed to be used 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
+ * 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
- * NOT use these pool execution methods, but instead use the
- * within-computation forms listed in the table. To avoid inadvertant
- * cyclic task dependencies and to improve performance, task
- * submissions to the current pool by an ongoing fork/join
- * computations may be implicitly translated to the corresponding
- * ForkJoinTask forms.
+ * 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.
*
*
*
@@ -73,7 +81,7 @@ import java.util.concurrent.CountDownLat
* Call from within fork/join computations |
*
*
- * Arange async execution |
+ * Arrange async execution |
* {@link #execute(ForkJoinTask)} |
* {@link ForkJoinTask#fork} |
*
@@ -88,24 +96,16 @@ import java.util.concurrent.CountDownLat
* {@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
- * bookkeeping overhead of creating a large set of threads. For
- * example, a common pool could be used for the {@code SortTasks}
- * illustrated in {@link RecursiveAction}. Because {@code
- * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
- * daemon} mode, there is typically no need to explicitly {@link
- * #shutdown} such a pool upon program exit.
*
- *
- * static final ForkJoinPool mainPool = new ForkJoinPool();
- * ...
- * public void sort(long[] array) {
- * mainPool.invoke(new SortTask(array, 0, array.length));
- * }
- *
+ * The common pool is by default constructed with default
+ * parameters, but these may be controlled by setting three {@link
+ * System#getProperty properties} with prefix {@code
+ * java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
+ * an integer greater than zero, {@code threadFactory} -- the class
+ * name of a {@link ForkJoinWorkerThreadFactory}, and {@code
+ * exceptionHandler} -- the class name of a {@link
+ * Thread.UncaughtExceptionHandler}. Upon any error in establishing
+ * these settings, default parameters are used.
*
*
Implementation notes: This implementation restricts the
* maximum number of running threads to 32767. Attempts to create
@@ -113,7 +113,8 @@ 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
@@ -123,204 +124,386 @@ 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 is why we normally use
+ * method pollAt and its variants that try once at the apparent
+ * base index, else consider alternative actions, rather than
+ * method poll.)
+ *
+ * This approach also enables support of a user mode in which local
+ * task processing is in FIFO, not LIFO order, simply by using
+ * poll rather than pop. This can be useful in message-passing
+ * frameworks in which tasks are never joined. However neither
+ * mode considers affinities, loads, cache localities, etc, so
+ * rarely provide the best possible performance on a given
+ * machine, but portably provide good throughput by averaging over
+ * these factors. (Further, even if we did try to use such
+ * information, we do not usually have a basis for exploiting it.
+ * For example, some sets of tasks profit from cache affinities,
+ * but others are harmed by cache pollution effects.)
+ *
+ * WorkQueues are also used in a similar way for tasks submitted
+ * to the pool. We cannot mix these tasks in the same queues used
+ * for work-stealing (this would contaminate lifo/fifo
+ * processing). Instead, we randomly associate submission queues
+ * with submitting threads, using a form of hashing. The
+ * ThreadLocal Submitter class contains a value initially used as
+ * a hash code for choosing existing queues, but may be randomly
+ * repositioned upon contention with other submitters. In
+ * essence, submitters act like workers except that they are
+ * restricted to executing local tasks that they submitted (or in
+ * the case of CountedCompleters, others with the same root task).
+ * However, because most shared/external queue operations are more
+ * expensive than internal, and because, at steady state, external
+ * submitters will compete for CPU with workers, ForkJoinTask.join
+ * and related methods disable them from repeatedly helping to
+ * process tasks if all workers are active. Insertion of tasks in
+ * shared mode requires a lock (mainly to protect in the case of
+ * resizing) but we use only a simple spinlock (using bits in
+ * field qlock), because submitters encountering a busy queue move
+ * on to try or create other queues -- they block only when
+ * creating and registering new queues.
+ *
+ * Management
+ * ==========
*
* The main throughput advantages of work-stealing stem from
- * decentralized control -- workers mostly 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
- * operations, the array size is always a power of two, and all
- * readers must tolerate null slots. Currently, 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.
- *
- * 2. Bookkeeping for dynamically adding and removing workers. We
- * aim to approximately maintain the given level of parallelism.
- * 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 to spare counts must be prospective (not
- * retrospective). For example, the running count is decremented
- * before blocking by a thread about to block as a spare, but
- * incremented by the thread about to unblock it. Updates upon
- * resumption ofr threads blocking in awaitJoin or awaitBlocker
- * cannot usually be prospective, so the running count is in
- * general an upper bound of the number of productively running
- * threads 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 performing an alternative
- * action on contention like releasing waiters or locating spares.
- *
- * 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 awaitJoin and awaitBlocker 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 awaitJoin and awaitBlocker. 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. Instead, we
- * use a combination of heuristics that adds threads only when the
- * pool appears to be approaching starvation. These effectively
- * reduce 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.
- *
- * Beware that 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
+ * decentralized control -- workers mostly take tasks from
+ * themselves or each other. We cannot negate this in the
+ * implementation of other management responsibilities. The main
+ * tactic for avoiding bottlenecks is packing nearly all
+ * essentially atomic control state into two volatile variables
+ * that are by far most often read (not written) as status and
+ * consistency checks.
+ *
+ * Field "ctl" contains 64 bits holding all the information needed
+ * to atomically decide to add, inactivate, enqueue (on an event
+ * queue), dequeue, and/or re-activate workers. To enable this
+ * packing, we restrict maximum parallelism to (1<<15)-1 (which is
+ * far in excess of normal operating range) to allow ids, counts,
+ * and their negations (used for thresholding) to fit into 16bit
+ * fields.
+ *
+ * Field "plock" is a form of sequence lock with a saturating
+ * shutdown bit (similarly for per-queue "qlocks"), mainly
+ * protecting updates to the workQueues array, as well as to
+ * enable shutdown. When used as a lock, it is normally only very
+ * briefly held, so is nearly always available after at most a
+ * brief spin, but we use a monitor-based backup strategy to
+ * blocking when needed.
+ *
+ * Recording WorkQueues. WorkQueues are recorded in the
+ * "workQueues" array that is created upon first use and expanded
+ * if necessary. Updates to the array while recording new workers
+ * and unrecording terminated ones are protected from each other
+ * by a lock but the array is otherwise concurrently readable, and
+ * accessed directly. To simplify index-based operations, the
+ * array size is always a power of two, and all readers must
+ * tolerate null slots. Worker queues are at odd indices Shared
+ * (submission) queues are at even indices, up to a maximum of 64
+ * slots, to limit growth even if array needs to expand to add
+ * more workers. Grouping them together in this way simplifies and
+ * speeds up task scanning.
+ *
+ * All worker thread creation is on-demand, triggered by task
+ * submissions, replacement of terminated workers, and/or
+ * compensation for blocked workers. However, all other support
+ * code is set up to work with other policies. To ensure that we
+ * do not hold on to worker references that would prevent GC, ALL
+ * accesses to workQueues are via indices into the workQueues
+ * array (which is one source of some of the messy code
+ * constructions here). In essence, the workQueues array serves as
+ * a weak reference mechanism. Thus for example the wait queue
+ * field of ctl stores indices, not references. Access to the
+ * workQueues in associated methods (for example signalWork) must
+ * both index-check and null-check the IDs. All such accesses
+ * ignore bad IDs by returning out early from what they are doing,
+ * since this can only be associated with termination, in which
+ * case it is OK to give up. All uses of the workQueues array
+ * also check that it is non-null (even if previously
+ * non-null). This allows nulling during termination, which is
+ * currently not necessary, but remains an option for
+ * resource-revocation-based shutdown schemes. It also helps
+ * reduce JIT issuance of uncommon-trap code, which tends to
+ * unnecessarily complicate control flow in some methods.
+ *
+ * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
+ * let workers spin indefinitely scanning for tasks when none can
+ * be found immediately, and we cannot start/resume workers unless
+ * there appear to be tasks available. On the other hand, we must
+ * quickly prod them into action when new tasks are submitted or
+ * generated. In many usages, ramp-up time to activate workers is
+ * the main limiting factor in overall performance (this is
+ * compounded at program start-up by JIT compilation and
+ * allocation). So we try to streamline this as much as possible.
+ * We park/unpark workers after placing in an event wait queue
+ * when they cannot find work. This "queue" is actually a simple
+ * Treiber stack, headed by the "id" field of ctl, plus a 15bit
+ * counter value (that reflects the number of times a worker has
+ * been inactivated) to avoid ABA effects (we need only as many
+ * version numbers as worker threads). Successors are held in
+ * field WorkQueue.nextWait. Queuing deals with several intrinsic
+ * races, mainly that a task-producing thread can miss seeing (and
+ * signalling) another thread that gave up looking for work but
+ * has not yet entered the wait queue. We solve this by requiring
+ * a full sweep of all workers (via repeated calls to method
+ * scan()) both before and after a newly waiting worker is added
+ * to the wait queue. During a rescan, the worker might release
+ * some other queued worker rather than itself, which has the same
+ * net effect. Because enqueued workers may actually be rescanning
+ * rather than waiting, we set and clear the "parker" field of
+ * WorkQueues to reduce unnecessary calls to unpark. (This
+ * requires a secondary recheck to avoid missed signals.) Note
+ * the unusual conventions about Thread.interrupts surrounding
+ * parking and other blocking: Because interrupts are used solely
+ * to alert threads to check termination, which is checked anyway
+ * upon blocking, we clear status (using Thread.interrupted)
+ * before any call to park, so that park does not immediately
+ * return due to status being set via some other unrelated call to
+ * interrupt in user code.
+ *
+ * Signalling. We create or wake up workers only when there
+ * appears to be at least one task they might be able to find and
+ * execute. However, many other threads may notice the same task
+ * and each signal to wake up a thread that might take it. So in
+ * general, pools will be over-signalled. When a submission is
+ * added or another worker adds a task to a queue that is
+ * apparently empty, they signal waiting workers (or trigger
+ * creation of new ones if fewer than the given parallelism level
+ * -- see signalWork). These primary signals are buttressed by
+ * signals whenever other threads scan for work or do not have a
+ * task to process. On most platforms, signalling (unpark)
+ * overhead time is noticeably long, and the time between
+ * signalling a thread and it actually making progress can be very
+ * noticeably long, so it is worth offloading these delays from
+ * critical paths as much as possible.
+ *
+ * Trimming workers. To release resources after periods of lack of
+ * use, a worker starting to wait when the pool is quiescent will
+ * time out and terminate if the pool has remained quiescent for a
+ * given period -- a short period if there are more threads than
+ * parallelism, longer as the number of threads decreases. This
+ * will slowly propagate, eventually terminating all workers after
+ * periods of non-use.
+ *
+ * Shutdown and Termination. A call to shutdownNow atomically sets
+ * a plock bit and then (non-atomically) sets each worker's
+ * qlock status, cancels all unprocessed tasks, and wakes up
+ * all waiting workers. Detecting whether termination should
+ * commence after a non-abrupt shutdown() call requires more work
+ * and bookkeeping. We need consensus about quiescence (i.e., that
+ * there is no more work). The active count provides a primary
+ * indication but non-abrupt shutdown still requires a rechecking
+ * scan for any workers that are inactive but not queued.
+ *
+ * Joining Tasks
+ * =============
+ *
+ * Any of several actions may be taken when one worker is waiting
+ * to join a task stolen (or always held) by another. Because we
+ * are multiplexing many tasks on to a pool of workers, we can't
+ * just let them block (as in Thread.join). We also cannot just
+ * reassign the joiner's run-time stack with another and replace
+ * it later, which would be a form of "continuation", that even if
+ * possible is not necessarily a good idea since we sometimes need
+ * both an unblocked task and its continuation to progress.
+ * Instead we combine two tactics:
+ *
+ * Helping: Arranging for the joiner to execute some task that it
+ * would be running if the steal had not occurred.
+ *
+ * Compensating: Unless there are already enough live threads,
+ * method tryCompensate() may create or re-activate a spare
+ * thread to compensate for blocked joiners until they unblock.
+ *
+ * A third form (implemented in tryRemoveAndExec) amounts to
+ * helping a hypothetical compensator: If we can readily tell that
+ * a possible action of a compensator is to steal and execute the
+ * task being joined, the joining thread can do so directly,
+ * without the need for a compensation thread (although at the
+ * expense of larger run-time stacks, but the tradeoff is
+ * typically worthwhile).
+ *
+ * The ManagedBlocker extension API can't use helping so relies
+ * only on compensation in method awaitBlocker.
+ *
+ * The algorithm in tryHelpStealer entails a form of "linear"
+ * helping: Each worker records (in field currentSteal) the most
+ * recent task it stole from some other worker. Plus, it records
+ * (in field currentJoin) the task it is currently actively
+ * joining. Method tryHelpStealer uses these markers to try to
+ * find a worker to help (i.e., steal back a task from and execute
+ * it) that could hasten completion of the actively joined task.
+ * In essence, the joiner executes a task that would be on its own
+ * local deque had the to-be-joined task not been stolen. This may
+ * be seen as a conservative variant of the approach in Wagner &
+ * Calder "Leapfrogging: a portable technique for implementing
+ * efficient futures" SIGPLAN Notices, 1993
+ * (http://portal.acm.org/citation.cfm?id=155354). It differs in
+ * that: (1) We only maintain dependency links across workers upon
+ * steals, rather than use per-task bookkeeping. This sometimes
+ * requires a linear scan of workQueues array to locate stealers,
+ * but often doesn't because stealers leave hints (that may become
+ * stale/wrong) of where to locate them. A stealHint is only a
+ * hint because a worker might have had multiple steals and the
+ * hint records only one of them (usually the most current).
+ * Hinting isolates cost to when it is needed, rather than adding
+ * to per-task overhead. (2) It is "shallow", ignoring nesting
+ * and potentially cyclic mutual steals. (3) It is intentionally
+ * racy: field currentJoin is updated only while actively joining,
+ * which means that we miss links in the chain during long-lived
+ * tasks, GC stalls etc (which is OK since blocking in such cases
+ * is usually a good idea). (4) We bound the number of attempts
+ * to find work (see MAX_HELP) and fall back to suspending the
+ * worker and if necessary replacing it with another.
+ *
+ * Helping actions for CountedCompleters are much simpler: Method
+ * helpComplete can take and execute any task with the same root
+ * as the task being waited on. However, this still entails some
+ * traversal of completer chains, so is less efficient than using
+ * CountedCompleters without explicit joins.
+ *
+ * It is impossible to keep exactly the target parallelism number
+ * of threads running at any given time. Determining the
+ * existence of conservatively safe helping targets, the
+ * availability of already-created spares, and the apparent need
+ * to create new spares are all racy, so we rely on multiple
+ * retries of each. Compensation in the apparent absence of
+ * helping opportunities is challenging to control on JVMs, where
+ * GC and other activities can stall progress of tasks that in
+ * turn stall out many other dependent tasks, without us being
+ * able to determine whether they will ever require compensation.
+ * Even though work-stealing otherwise encounters little
+ * degradation in the presence of more threads than cores,
+ * aggressively adding new threads in such cases entails risk of
+ * unwanted positive feedback control loops in which more threads
+ * cause more dependent stalls (as well as delayed progress of
+ * unblocked threads to the point that we know they are available)
+ * leading to more situations requiring more threads, and so
+ * on. This aspect of control can be seen as an (analytically
+ * intractable) game with an opponent that may choose the worst
+ * (for us) active thread to stall at any time. We take several
+ * precautions to bound losses (and thus bound gains), mainly in
+ * methods tryCompensate and awaitJoin.
+ *
+ * Common Pool
+ * ===========
+ *
+ * The static commonPool always exists after static
+ * initialization. Since it (or any other created pool) need
+ * never be used, we minimize initial construction overhead and
+ * footprint to the setup of about a dozen fields, with no nested
+ * allocation. Most bootstrapping occurs within method
+ * fullExternalPush during the first submission to the pool.
+ *
+ * When external threads submit to the common pool, they can
+ * perform some subtask processing (see externalHelpJoin and
+ * related methods). We do not need to record whether these
+ * submissions are to the common pool -- if not, externalHelpJoin
+ * returns quickly (at the most helping to signal some common pool
+ * workers). These submitters would otherwise be blocked waiting
+ * for completion, so the extra effort (with liberally sprinkled
+ * task status checks) in inapplicable cases amounts to an odd
+ * form of limited spin-wait before blocking in ForkJoinTask.join.
+ *
+ * Style notes
+ * ===========
+ *
+ * There is a lot of representation-level coupling among classes
+ * ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
+ * fields of WorkQueue maintain data structures managed by
+ * ForkJoinPool, so are directly accessed. There is little point
* trying to reduce this, since any associated future changes in
* representations will need to be accompanied by algorithmic
- * changes anyway.
- *
- * 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).
+ * changes anyway. Several methods intrinsically sprawl because
+ * they must accumulate sets of consistent reads of volatiles held
+ * in local variables. Methods signalWork() and scan() are the
+ * main bottlenecks, so are especially heavily
+ * micro-optimized/mangled. There are lots of inline assignments
+ * (of form "while ((local = field) != 0)") which are usually the
+ * simplest way to ensure the required read orderings (which are
+ * sometimes critical). This leads to a "C"-like style of listing
+ * declarations of these locals at the heads of methods or blocks.
+ * There are several occurrences of the unusual "do {} while
+ * (!cas...)" which is the simplest way to force an update of a
+ * CAS'ed variable. There are also other coding oddities (including
+ * several unnecessary-looking hoisted null checks) that help
+ * some methods perform reasonably even when interpreted (not
+ * compiled).
+ *
+ * The order of declarations in this file is:
+ * (1) Static utility functions
+ * (2) Nested (static) classes
+ * (3) Static fields
+ * (4) Fields, along with constants used when unpacking some of them
+ * (5) Internal control methods
+ * (6) Callbacks and other support for ForkJoinTask methods
+ * (7) Exported methods
+ * (8) Static block initializing statics in minimally dependent order
*/
+ // Static utilities
+
+ /**
+ * If there is a security manager, makes sure caller has
+ * permission to modify threads.
+ */
+ private static void checkPermission() {
+ SecurityManager security = System.getSecurityManager();
+ if (security != null)
+ security.checkPermission(modifyThreadPermission);
+ }
+
+ // Nested classes
+
/**
* Factory for creating new {@link ForkJoinWorkerThread}s.
* A {@code ForkJoinWorkerThreadFactory} must be defined and used
@@ -349,767 +532,1845 @@ public class ForkJoinPool extends Abstra
}
/**
- * Creates a new ForkJoinWorkerThread. This factory is used unless
- * overridden in ForkJoinPool constructors.
+ * Class for artificial tasks that are used to replace the target
+ * of local joins if they are removed from an interior queue slot
+ * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
+ * actually do anything beyond having a unique identity.
+ */
+ static final class EmptyTask extends ForkJoinTask {
+ private static final long serialVersionUID = -7721805057305804111L;
+ EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
+ public final Void getRawResult() { return null; }
+ public final void setRawResult(Void x) {}
+ public final boolean exec() { return true; }
+ }
+
+ /**
+ * Queues supporting work-stealing as well as external task
+ * submission. See above for main rationale and algorithms.
+ * Implementation relies heavily on "Unsafe" intrinsics
+ * and selective use of "volatile":
+ *
+ * Field "base" is the index (mod array.length) of the least valid
+ * queue slot, which is always the next position to steal (poll)
+ * from if nonempty. Reads and writes require volatile orderings
+ * but not CAS, because updates are only performed after slot
+ * CASes.
+ *
+ * Field "top" is the index (mod array.length) of the next queue
+ * slot to push to or pop from. It is written only by owner thread
+ * for push, or under lock for external/shared push, and accessed
+ * by other threads only after reading (volatile) base. Both top
+ * and base are allowed to wrap around on overflow, but (top -
+ * base) (or more commonly -(base - top) to force volatile read of
+ * base before top) still estimates size. The lock ("qlock") is
+ * forced to -1 on termination, causing all further lock attempts
+ * to fail. (Note: we don't need CAS for termination state because
+ * upon pool shutdown, all shared-queues will stop being used
+ * anyway.) Nearly all lock bodies are set up so that exceptions
+ * within lock bodies are "impossible" (modulo JVM errors that
+ * would cause failure anyway.)
+ *
+ * The array slots are read and written using the emulation of
+ * volatiles/atomics provided by Unsafe. Insertions must in
+ * general use putOrderedObject as a form of releasing store to
+ * ensure that all writes to the task object are ordered before
+ * its publication in the queue. All removals entail a CAS to
+ * null. The array is always a power of two. To ensure safety of
+ * Unsafe array operations, all accesses perform explicit null
+ * checks and implicit bounds checks via power-of-two masking.
+ *
+ * In addition to basic queuing support, this class contains
+ * fields described elsewhere to control execution. It turns out
+ * to work better memory-layout-wise to include them in this class
+ * rather than a separate class.
+ *
+ * Performance on most platforms is very sensitive to placement of
+ * instances of both WorkQueues and their arrays -- we absolutely
+ * do not want multiple WorkQueue instances or multiple queue
+ * arrays sharing cache lines. (It would be best for queue objects
+ * and their arrays to share, but there is nothing available to
+ * help arrange that). Unfortunately, because they are recorded
+ * in a common array, WorkQueue instances are often moved to be
+ * adjacent by garbage collectors. To reduce impact, we use field
+ * padding that works OK on common platforms; this effectively
+ * trades off slightly slower average field access for the sake of
+ * avoiding really bad worst-case access. (Until better JVM
+ * support is in place, this padding is dependent on transient
+ * properties of JVM field layout rules.) We also take care in
+ * allocating, sizing and resizing the array. Non-shared queue
+ * arrays are initialized by workers before use. Others are
+ * allocated on first use.
*/
- public static final ForkJoinWorkerThreadFactory
- defaultForkJoinWorkerThreadFactory =
- new DefaultForkJoinWorkerThreadFactory();
+ static final class WorkQueue {
+ /**
+ * Capacity of work-stealing queue array upon initialization.
+ * Must be a power of two; at least 4, but should be larger to
+ * reduce or eliminate cacheline sharing among queues.
+ * Currently, it is much larger, as a partial workaround for
+ * the fact that JVMs often place arrays in locations that
+ * share GC bookkeeping (especially cardmarks) such that
+ * per-write accesses encounter serious memory contention.
+ */
+ static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
- /**
- * Permission required for callers of methods that may start or
- * kill threads.
- */
- private static final RuntimePermission modifyThreadPermission =
- new RuntimePermission("modifyThread");
+ /**
+ * 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
- /**
- * If there is a security manager, makes sure caller has
- * permission to modify threads.
- */
- private static void checkPermission() {
- SecurityManager security = System.getSecurityManager();
- if (security != null)
- security.checkPermission(modifyThreadPermission);
+ int seed; // for random scanning; initialize nonzero
+ volatile int eventCount; // encoded inactivation count; < 0 if inactive
+ int nextWait; // encoded record of next event waiter
+ final int mode; // lifo, fifo, or shared
+ int nsteals; // cumulative number of steals
+ int poolIndex; // index of this queue in pool (or 0)
+ int stealHint; // index of most recent known stealer
+ volatile int qlock; // 1: locked, -1: terminate; else 0
+ volatile int base; // index of next slot for poll
+ int top; // index of next slot for push
+ ForkJoinTask>[] array; // the elements (initially unallocated)
+ final ForkJoinPool pool; // the containing pool (may be null)
+ final ForkJoinWorkerThread owner; // owning thread or null if shared
+ volatile Thread parker; // == owner during call to park; else null
+ volatile ForkJoinTask> currentJoin; // task being joined in awaitJoin
+ ForkJoinTask> currentSteal; // current non-local task being executed
+ // Heuristic padding to ameliorate unfortunate memory placements
+ Object p00, p01, p02, p03, p04, p05, p06, p07;
+ Object p08, p09, p0a, p0b, p0c, p0d, p0e;
+
+ WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
+ this.mode = mode;
+ this.pool = pool;
+ this.owner = owner;
+ // Place indices in the center of array (that is not yet allocated)
+ base = top = INITIAL_QUEUE_CAPACITY >>> 1;
+ }
+
+ /**
+ * Pushes a task. Call only by owner in unshared queues.
+ * Cases needing resizing or rejection are relayed to fullPush
+ * (that also handles shared queues).
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @throw RejectedExecutionException if array cannot be resized
+ */
+ final void push(ForkJoinTask> task) {
+ ForkJoinPool p; ForkJoinTask>[] a;
+ int s = top, n;
+ if ((a = array) != null && a.length > (n = s + 1 - base)) {
+ U.putOrderedObject
+ (a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
+ top = s + 1;
+ if (n <= 1 && (p = pool) != null)
+ p.signalWork(this, 1);
+ }
+ else
+ fullPush(task, true);
+ }
+
+ /**
+ * Pushes a task if lock is free and array is either big
+ * enough or can be resized to be big enough. Note: a
+ * specialization of a common fast path of this method is in
+ * ForkJoinPool.externalPush. When called from a FJWT queue,
+ * this can fail only if the pool has been shut down or
+ * an out of memory error.
+ *
+ * @param task the task. Caller must ensure non-null.
+ * @param owned if true, throw RJE on failure
+ */
+ final boolean fullPush(ForkJoinTask> task, boolean owned) {
+ ForkJoinPool p; ForkJoinTask>[] a;
+ if (owned) {
+ if (qlock < 0) // must be shutting down
+ throw new RejectedExecutionException();
+ }
+ else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
+ return false;
+ try {
+ int s = top, oldLen, len;
+ if ((a = array) == null)
+ a = array = new ForkJoinTask>[len=INITIAL_QUEUE_CAPACITY];
+ else if ((oldLen = a.length) > s + 1 - base)
+ len = oldLen;
+ else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
+ throw new RejectedExecutionException("Capacity exceeded");
+ else {
+ int oldMask, b;
+ ForkJoinTask>[] oldA = a;
+ a = array = new ForkJoinTask>[len];
+ if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
+ int mask = len - 1;
+ do {
+ ForkJoinTask> x;
+ int oldj = ((b & oldMask) << ASHIFT) + ABASE;
+ int j = ((b & mask) << ASHIFT) + ABASE;
+ x = (ForkJoinTask>)
+ U.getObjectVolatile(oldA, oldj);
+ if (x != null &&
+ U.compareAndSwapObject(oldA, oldj, x, null))
+ U.putObjectVolatile(a, j, x);
+ } while (++b != s);
+ }
+ }
+ U.putOrderedObject
+ (a, (((len - 1) & s) << ASHIFT) + ABASE, task);
+ top = s + 1;
+ } finally {
+ if (!owned)
+ qlock = 0;
+ }
+ if ((p = pool) != null)
+ p.signalWork(this, 1);
+ return true;
+ }
+
+ /**
+ * Takes next task, if one exists, in LIFO order. Call only
+ * by owner in unshared queues.
+ */
+ final ForkJoinTask> pop() {
+ ForkJoinTask>[] a; ForkJoinTask> t; int m;
+ if ((a = array) != null && (m = a.length - 1) >= 0) {
+ for (int s; (s = top - 1) - base >= 0;) {
+ long j = ((m & s) << ASHIFT) + ABASE;
+ if ((t = (ForkJoinTask>)U.getObject(a, j)) == null)
+ break;
+ if (U.compareAndSwapObject(a, j, t, null)) {
+ top = s;
+ return t;
+ }
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Takes a task in FIFO order if b is base of queue and a task
+ * can be claimed without contention. Specialized versions
+ * appear in ForkJoinPool methods scan and tryHelpStealer.
+ */
+ final ForkJoinTask> pollAt(int b) {
+ ForkJoinTask> t; ForkJoinTask>[] a;
+ if ((a = array) != null) {
+ int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ if ((t = (ForkJoinTask>)U.getObjectVolatile(a, j)) != null &&
+ base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ return t;
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Takes next task, if one exists, in FIFO order.
+ */
+ final ForkJoinTask> poll() {
+ ForkJoinTask>[] a; int b; ForkJoinTask> t;
+ while ((b = base) - top < 0 && (a = array) != null) {
+ int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t != null) {
+ if (base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ return t;
+ }
+ }
+ else if (base == b) {
+ if (b + 1 == top)
+ break;
+ Thread.yield(); // wait for lagging update (very rare)
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Takes next task, if one exists, in order specified by mode.
+ */
+ final ForkJoinTask> nextLocalTask() {
+ return mode == 0 ? pop() : poll();
+ }
+
+ /**
+ * Returns next task, if one exists, in order specified by mode.
+ */
+ final ForkJoinTask> peek() {
+ ForkJoinTask>[] a = array; int m;
+ if (a == null || (m = a.length - 1) < 0)
+ return null;
+ int i = mode == 0 ? top - 1 : base;
+ int j = ((i & m) << ASHIFT) + ABASE;
+ return (ForkJoinTask>)U.getObjectVolatile(a, j);
+ }
+
+ /**
+ * Pops the given task only if it is at the current top.
+ * (A shared version is available only via FJP.tryExternalUnpush)
+ */
+ final boolean tryUnpush(ForkJoinTask> t) {
+ ForkJoinTask>[] a; int s;
+ if ((a = array) != null && (s = top) != base &&
+ U.compareAndSwapObject
+ (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
+ top = s;
+ return true;
+ }
+ return false;
+ }
+
+ /**
+ * Removes and cancels all known tasks, ignoring any exceptions.
+ */
+ final void cancelAll() {
+ ForkJoinTask.cancelIgnoringExceptions(currentJoin);
+ ForkJoinTask.cancelIgnoringExceptions(currentSteal);
+ for (ForkJoinTask> t; (t = poll()) != null; )
+ ForkJoinTask.cancelIgnoringExceptions(t);
+ }
+
+ /**
+ * Computes next value for random probes. Scans don't require
+ * a very high quality generator, but also not a crummy one.
+ * Marsaglia xor-shift is cheap and works well enough. Note:
+ * This is manually inlined in its usages in ForkJoinPool to
+ * avoid writes inside busy scan loops.
+ */
+ final int nextSeed() {
+ int r = seed;
+ r ^= r << 13;
+ r ^= r >>> 17;
+ return seed = r ^= r << 5;
+ }
+
+ /**
+ * Provides a more accurate estimate of size than (top - base)
+ * by ordering reads and checking whether a near-empty queue
+ * has at least one unclaimed task.
+ */
+ final int queueSize() {
+ ForkJoinTask>[] a; int k, s, n;
+ return ((n = base - (s = top)) < 0 &&
+ (n != -1 ||
+ ((a = array) != null && (k = a.length) > 0 &&
+ U.getObject
+ (a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
+ -n : 0;
+ }
+
+ // Specialized execution methods
+
+ /**
+ * Pops and runs tasks until empty.
+ */
+ private void popAndExecAll() {
+ // A bit faster than repeated pop calls
+ ForkJoinTask>[] a; int m, s; long j; ForkJoinTask> t;
+ while ((a = array) != null && (m = a.length - 1) >= 0 &&
+ (s = top - 1) - base >= 0 &&
+ (t = ((ForkJoinTask>)
+ U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
+ != null) {
+ if (U.compareAndSwapObject(a, j, t, null)) {
+ top = s;
+ t.doExec();
+ }
+ }
+ }
+
+ /**
+ * Polls and runs tasks until empty.
+ */
+ private void pollAndExecAll() {
+ for (ForkJoinTask> t; (t = poll()) != null;)
+ t.doExec();
+ }
+
+ /**
+ * If present, removes from queue and executes the given task,
+ * or any other cancelled task. Returns (true) on any CAS
+ * or consistency check failure so caller can retry.
+ *
+ * @return false if no progress can be made, else true;
+ */
+ final boolean tryRemoveAndExec(ForkJoinTask> task) {
+ boolean stat = true, removed = false, empty = true;
+ ForkJoinTask>[] a; int m, s, b, n;
+ if ((a = array) != null && (m = a.length - 1) >= 0 &&
+ (n = (s = top) - (b = base)) > 0) {
+ for (ForkJoinTask> t;;) { // traverse from s to b
+ int j = ((--s & m) << ASHIFT) + ABASE;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, j);
+ if (t == null) // inconsistent length
+ break;
+ else if (t == task) {
+ if (s + 1 == top) { // pop
+ if (!U.compareAndSwapObject(a, j, task, null))
+ break;
+ top = s;
+ removed = true;
+ }
+ else if (base == b) // replace with proxy
+ removed = U.compareAndSwapObject(a, j, task,
+ new EmptyTask());
+ break;
+ }
+ else if (t.status >= 0)
+ empty = false;
+ else if (s + 1 == top) { // pop and throw away
+ if (U.compareAndSwapObject(a, j, t, null))
+ top = s;
+ break;
+ }
+ if (--n == 0) {
+ if (!empty && base == b)
+ stat = false;
+ break;
+ }
+ }
+ }
+ if (removed)
+ task.doExec();
+ return stat;
+ }
+
+ /**
+ * Polls for and executes the given task or any other task in
+ * its CountedCompleter computation
+ */
+ final boolean pollAndExecCC(ForkJoinTask> root) {
+ ForkJoinTask>[] a; int b; Object o;
+ outer: while ((b = base) - top < 0 && (a = array) != null) {
+ long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ if ((o = U.getObject(a, j)) == null ||
+ !(o instanceof CountedCompleter))
+ break;
+ for (CountedCompleter> t = (CountedCompleter>)o, r = t;;) {
+ if (r == root) {
+ if (base == b &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ base = b + 1;
+ t.doExec();
+ return true;
+ }
+ else
+ break; // restart
+ }
+ if ((r = r.completer) == null)
+ break outer; // not part of root computation
+ }
+ }
+ return false;
+ }
+
+ /**
+ * Executes a top-level task and any local tasks remaining
+ * after execution.
+ */
+ final void runTask(ForkJoinTask> t) {
+ if (t != null) {
+ (currentSteal = t).doExec();
+ currentSteal = null;
+ if (++nsteals < 0) { // spill on overflow
+ ForkJoinPool p;
+ if ((p = pool) != null)
+ p.collectStealCount(this);
+ }
+ if (top != base) { // process remaining local tasks
+ if (mode == 0)
+ popAndExecAll();
+ else
+ pollAndExecAll();
+ }
+ }
+ }
+
+ /**
+ * Executes a non-top-level (stolen) task.
+ */
+ final void runSubtask(ForkJoinTask> t) {
+ if (t != null) {
+ ForkJoinTask> ps = currentSteal;
+ (currentSteal = t).doExec();
+ currentSteal = ps;
+ }
+ }
+
+ /**
+ * Returns true if owned and not known to be blocked.
+ */
+ final boolean isApparentlyUnblocked() {
+ Thread wt; Thread.State s;
+ return (eventCount >= 0 &&
+ (wt = owner) != null &&
+ (s = wt.getState()) != Thread.State.BLOCKED &&
+ s != Thread.State.WAITING &&
+ s != Thread.State.TIMED_WAITING);
+ }
+
+ /**
+ * If this owned and is not already interrupted, try to
+ * interrupt and/or unpark, ignoring exceptions.
+ */
+ final void interruptOwner() {
+ Thread wt, p;
+ if ((wt = owner) != null && !wt.isInterrupted()) {
+ try {
+ wt.interrupt();
+ } catch (SecurityException ignore) {
+ }
+ }
+ if ((p = parker) != null)
+ U.unpark(p);
+ }
+
+ // Unsafe mechanics
+ private static final sun.misc.Unsafe U;
+ private static final long QLOCK;
+ private static final int ABASE;
+ private static final int ASHIFT;
+ static {
+ int s;
+ try {
+ U = getUnsafe();
+ Class> k = WorkQueue.class;
+ Class> ak = ForkJoinTask[].class;
+ QLOCK = U.objectFieldOffset
+ (k.getDeclaredField("qlock"));
+ ABASE = U.arrayBaseOffset(ak);
+ s = U.arrayIndexScale(ak);
+ } catch (Exception e) {
+ throw new Error(e);
+ }
+ if ((s & (s-1)) != 0)
+ throw new Error("data type scale not a power of two");
+ ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
+ }
}
/**
- * Generator for assigning sequence numbers as pool names.
+ * Per-thread records for threads that submit to pools. Currently
+ * holds only pseudo-random seed / index that is used to choose
+ * submission queues in method externalPush. In the future, this may
+ * also incorporate a means to implement different task rejection
+ * and resubmission policies.
+ *
+ * Seeds for submitters and workers/workQueues work in basically
+ * the same way but are initialized and updated using slightly
+ * different mechanics. Both are initialized using the same
+ * approach as in class ThreadLocal, where successive values are
+ * unlikely to collide with previous values. Seeds are then
+ * randomly modified upon collisions using xorshifts, which
+ * requires a non-zero seed.
*/
- private static final AtomicInteger poolNumberGenerator =
- new AtomicInteger();
+ static final class Submitter {
+ int seed;
+ Submitter(int s) { seed = s; }
+ }
- /**
- * Absolute bound for parallelism level. Twice this number must
- * fit into a 16bit field to enable word-packing for some counts.
- */
- private static final int MAX_THREADS = 0x7fff;
+ /** Property prefix for constructing common pool */
+ private static final String propPrefix =
+ "java.util.concurrent.ForkJoinPool.common.";
+
+ // static fields (initialized in static initializer below)
/**
- * 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.
+ * Creates a new ForkJoinWorkerThread. This factory is used unless
+ * overridden in ForkJoinPool constructors.
*/
- volatile ForkJoinWorkerThread[] workers;
+ public static final ForkJoinWorkerThreadFactory
+ defaultForkJoinWorkerThreadFactory;
/**
- * Queue for external submissions.
+ * Common (static) pool. Non-null for public use unless a static
+ * construction exception, but internal usages null-check on use
+ * to paranoically avoid potential initialization circularities
+ * as well as to simplify generated code.
*/
- private final LinkedTransferQueue> submissionQueue;
+ static final ForkJoinPool commonPool;
/**
- * Lock protecting updates to workers array.
+ * Permission required for callers of methods that may start or
+ * kill threads.
*/
- private final ReentrantLock workerLock;
+ private static final RuntimePermission modifyThreadPermission;
/**
- * Latch released upon termination.
+ * Per-thread submission bookkeeping. Shared across all pools
+ * to reduce ThreadLocal pollution and because random motion
+ * to avoid contention in one pool is likely to hold for others.
+ * Lazily initialized on first submission (but null-checked
+ * in other contexts to avoid unnecessary initialization).
*/
- private final Phaser termination;
+ static final ThreadLocal submitters;
/**
- * Creation factory for worker threads.
+ * Common pool parallelism. Must equal commonPool.parallelism.
*/
- private final ForkJoinWorkerThreadFactory factory;
+ static final int commonPoolParallelism;
/**
- * Sum of per-thread steal counts, updated only when threads are
- * idle or terminating.
+ * Sequence number for creating workerNamePrefix.
*/
- private volatile long stealCount;
+ private static int poolNumberSequence;
/**
- * 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.
+ * Return the next sequence number. We don't expect this to
+ * ever contend so use simple builtin sync.
*/
- private volatile long eventWaiters;
+ private static final synchronized int nextPoolId() {
+ return ++poolNumberSequence;
+ }
- private static final int EVENT_COUNT_SHIFT = 32;
- private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
+ // static constants
/**
- * 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
+ * Initial timeout value (in nanoseconds) for the thread
+ * triggering quiescence to park waiting for new work. On timeout,
+ * the thread will instead try to shrink the number of
+ * workers. The value should be large enough to avoid overly
+ * aggressive shrinkage during most transient stalls (long GCs
+ * etc).
*/
- private volatile int eventCount;
+ private static final long IDLE_TIMEOUT = 2000L * 1000L * 1000L; // 2sec
/**
- * 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).
+ * Timeout value when there are more threads than parallelism level
*/
- private volatile int runState;
-
- // 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
+ private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L;
/**
- * 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.
+ * The maximum stolen->joining link depth allowed in method
+ * tryHelpStealer. Must be a power of two. Depths for legitimate
+ * chains are unbounded, but we use a fixed constant to avoid
+ * (otherwise unchecked) cycles and to bound staleness of
+ * traversal parameters at the expense of sometimes blocking when
+ * we could be helping.
*/
- private volatile int workerCounts;
-
- 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;
+ private static final int MAX_HELP = 64;
/**
- * The target parallelism level.
- * Accessed directly by ForkJoinWorkerThreads.
+ * Increment for seed generators. See class ThreadLocal for
+ * explanation.
*/
- final int parallelism;
+ private static final int SEED_INCREMENT = 0x61c88647;
/**
- * True if use local fifo, not default lifo, for local polling
- * Read by, and replicated by ForkJoinWorkerThreads
+ * Bits and masks for control variables
+ *
+ * Field ctl is a long packed with:
+ * AC: Number of active running workers minus target parallelism (16 bits)
+ * TC: Number of total workers minus target parallelism (16 bits)
+ * ST: true if pool is terminating (1 bit)
+ * EC: the wait count of top waiting thread (15 bits)
+ * ID: poolIndex of top of Treiber stack of waiters (16 bits)
+ *
+ * When convenient, we can extract the upper 32 bits of counts and
+ * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
+ * (int)ctl. The ec field is never accessed alone, but always
+ * together with id and st. The offsets of counts by the target
+ * parallelism and the positionings of fields makes it possible to
+ * perform the most common checks via sign tests of fields: When
+ * ac is negative, there are not enough active workers, when tc is
+ * negative, there are not enough total workers, and when e is
+ * negative, the pool is terminating. To deal with these possibly
+ * negative fields, we use casts in and out of "short" and/or
+ * signed shifts to maintain signedness.
+ *
+ * When a thread is queued (inactivated), its eventCount field is
+ * set negative, which is the only way to tell if a worker is
+ * prevented from executing tasks, even though it must continue to
+ * scan for them to avoid queuing races. Note however that
+ * eventCount updates lag releases so usage requires care.
+ *
+ * Field plock is an int packed with:
+ * SHUTDOWN: true if shutdown is enabled (1 bit)
+ * SEQ: a sequence lock, with PL_LOCK bit set if locked (30 bits)
+ * SIGNAL: set when threads may be waiting on the lock (1 bit)
+ *
+ * The sequence number enables simple consistency checks:
+ * Staleness of read-only operations on the workQueues array can
+ * be checked by comparing plock before vs after the reads.
*/
- final boolean locallyFifo;
- /**
- * The uncaught exception handler used when any worker abruptly
- * terminates.
- */
- private final Thread.UncaughtExceptionHandler ueh;
+ // 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;
- /**
- * Pool number, just for assigning useful names to worker threads
- */
- private final int poolNumber;
+ // bounds
+ private static final int SMASK = 0xffff; // short bits
+ private static final int MAX_CAP = 0x7fff; // max #workers - 1
+ private static final int EVENMASK = 0xfffe; // even short bits
+ private static final int SQMASK = 0x007e; // max 64 (even) slots
+ private static final int SHORT_SIGN = 1 << 15;
+ private static final int INT_SIGN = 1 << 31;
- // utilities for updating fields
+ // 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;
- /**
- * Increments running count. Also used by ForkJoinTask.
- */
- final void incrementRunningCount() {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- c = workerCounts,
- c + ONE_RUNNING));
- }
-
- /**
- * Tries to decrement running count unless already zero
- */
- final boolean tryDecrementRunningCount() {
- int wc = workerCounts;
- if ((wc & RUNNING_COUNT_MASK) == 0)
- return false;
- return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - ONE_RUNNING);
- }
+ // units for incrementing and decrementing
+ private static final long TC_UNIT = 1L << TC_SHIFT;
+ private static final long AC_UNIT = 1L << AC_SHIFT;
- /**
- * 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);
- }
+ // 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;
- /**
- * 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);
- }
+ // 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;
- /**
- * 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;
- }
- }
+ // plock bits
+ private static final int SHUTDOWN = 1 << 31;
+ private static final int PL_LOCK = 2;
+ private static final int PL_SIGNAL = 1;
+ private static final int PL_SPINS = 1 << 8;
- // workers array maintenance
+ // access mode for WorkQueue
+ static final int LIFO_QUEUE = 0;
+ static final int FIFO_QUEUE = 1;
+ static final int SHARED_QUEUE = -1;
- /**
- * Records and returns a workers array index for new worker.
+ // Instance fields
+
+ /*
+ * Field layout order in this class tends to matter more than one
+ * would like. Runtime layout order is only loosely related to
+ * declaration order and may differ across JVMs, but the following
+ * empirically works OK on current JVMs.
+ */
+ volatile long stealCount; // collects worker counts
+ volatile long ctl; // main pool control
+ final int parallelism; // parallelism level
+ final int localMode; // per-worker scheduling mode
+ volatile int indexSeed; // worker/submitter index seed
+ volatile int plock; // shutdown status and seqLock
+ WorkQueue[] workQueues; // main registry
+ final ForkJoinWorkerThreadFactory factory; // factory for new workers
+ final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
+ final String workerNamePrefix; // to create worker name string
+
+ /*
+ * Acquires the plock lock to protect worker array and related
+ * updates. This method is called only if an initial CAS on plock
+ * fails. This acts as a spinLock for normal cases, but falls back
+ * to builtin monitor to block when (rarely) needed. This would be
+ * a terrible idea for a highly contended lock, but works fine as
+ * a more conservative alternative to a pure spinlock. See
+ * internal ConcurrentHashMap documentation for further
+ * explanation of nearly the same construction.
*/
- private 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 nws = ws.length;
- if (k < 0 || k >= nws || ws[k] != null) {
- for (k = 0; k < nws && ws[k] != null; ++k)
- ;
- if (k == nws)
- ws = Arrays.copyOf(ws, nws << 1);
+ private int acquirePlock() {
+ int spins = PL_SPINS, r = 0, ps, nps;
+ for (;;) {
+ if (((ps = plock) & PL_LOCK) == 0 &&
+ U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
+ return nps;
+ else if (r == 0)
+ r = ThreadLocalRandom.current().nextInt(); // randomize spins
+ else if (spins >= 0) {
+ r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
+ if (r >= 0)
+ --spins;
+ }
+ else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
+ synchronized (this) {
+ if ((plock & PL_SIGNAL) != 0) {
+ try {
+ wait();
+ } catch (InterruptedException ie) {
+ try {
+ Thread.currentThread().interrupt();
+ } catch (SecurityException ignore) {
+ }
+ }
+ }
+ else
+ notifyAll();
+ }
}
- ws[k] = w;
- workers = ws; // volatile array write ensures slot visibility
- } finally {
- lock.unlock();
}
- return k;
}
/**
- * Nulls out record of worker in workers array
+ * Unlocks and signals any thread waiting for plock. Called only
+ * when CAS of seq value for unlock fails.
*/
- private void forgetWorker(ForkJoinWorkerThread w) {
- int idx = w.poolIndex;
- // Locking helps method recordWorker avoid unecessary expansion
- final ReentrantLock lock = this.workerLock;
- lock.lock();
+ private void releasePlock(int ps) {
+ plock = ps;
+ synchronized (this) { notifyAll(); }
+ }
+
+ // Registering and deregistering workers
+
+ /**
+ * Callback from ForkJoinWorkerThread constructor to establish its
+ * poolIndex and record its WorkQueue. To avoid scanning bias due
+ * to packing entries in front of the workQueues array, we treat
+ * the array as a simple power-of-two hash table using per-thread
+ * seed as hash, expanding as needed.
+ *
+ * @param w the worker's queue
+ */
+ final void registerWorker(WorkQueue w) {
+ int s, ps; // generate a rarely colliding candidate index seed
+ do {} while (!U.compareAndSwapInt(this, INDEXSEED,
+ s = indexSeed, s += SEED_INCREMENT) ||
+ s == 0); // skip 0
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
try {
- ForkJoinWorkerThread[] ws = workers;
- if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify
- ws[idx] = null;
+ WorkQueue[] ws;
+ if (w != null && (ws = workQueues) != null) {
+ w.seed = s;
+ int n = ws.length, m = n - 1;
+ int r = (s << 1) | 1; // use odd-numbered indices
+ if (ws[r &= m] != null) { // collision
+ int probes = 0; // step by approx half size
+ int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
+ while (ws[r = (r + step) & m] != null) {
+ if (++probes >= n) {
+ workQueues = ws = Arrays.copyOf(ws, n <<= 1);
+ m = n - 1;
+ probes = 0;
+ }
+ }
+ }
+ w.eventCount = w.poolIndex = r; // establish before recording
+ ws[r] = w;
+ }
} finally {
- lock.unlock();
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
}
}
- // adding and removing 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
- */
- private ForkJoinWorkerThread addWorker() {
- ForkJoinWorkerThread w = null;
- try {
- w = factory.newThread(this);
- } finally { // Adjust on either null or exceptional factory return
- if (w == null) {
- onWorkerCreationFailure();
- return null;
+ * Final callback from terminating worker, as well as upon failure
+ * to construct or start a worker. Removes record of worker from
+ * array, and adjusts counts. If pool is shutting down, tries to
+ * complete termination.
+ *
+ * @param wt the worker thread or null if construction failed
+ * @param ex the exception causing failure, or null if none
+ */
+ final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
+ WorkQueue w = null;
+ if (wt != null && (w = wt.workQueue) != null) {
+ int ps;
+ collectStealCount(w);
+ w.qlock = -1; // ensure set
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ try {
+ int idx = w.poolIndex;
+ WorkQueue[] ws = workQueues;
+ if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
+ ws[idx] = null;
+ } finally {
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
}
}
- w.start(recordWorker(w), ueh);
- return w;
+
+ long c; // adjust ctl counts
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
+ ((c - TC_UNIT) & TC_MASK) |
+ (c & ~(AC_MASK|TC_MASK)))));
+
+ if (!tryTerminate(false, false) && w != null) {
+ w.cancelAll(); // cancel remaining tasks
+ if (w.array != null) // suppress signal if never ran
+ signalWork(null, 1); // wake up or create replacement
+ if (ex == null) // help clean refs on way out
+ ForkJoinTask.helpExpungeStaleExceptions();
+ }
+
+ if (ex != null) // rethrow
+ ForkJoinTask.rethrow(ex);
}
/**
- * Adjusts counts upon failure to create worker
- */
- private void onWorkerCreationFailure() {
- for (;;) {
- int wc = workerCounts;
- if ((wc >>> TOTAL_COUNT_SHIFT) > 0 &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - (ONE_RUNNING|ONE_TOTAL)))
- break;
+ * Collect worker steal count into total. Called on termination
+ * and upon int overflow of local count. (There is a possible race
+ * in the latter case vs any caller of getStealCount, which can
+ * make its results less accurate than usual.)
+ */
+ final void collectStealCount(WorkQueue w) {
+ if (w != null) {
+ long sc;
+ int ns = w.nsteals;
+ w.nsteals = 0; // handle overflow
+ long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
+ do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
+ sc = stealCount, sc + steals));
}
- tryTerminate(false); // in case of failure during shutdown
}
- /**
- * Create enough total workers to establish target parallelism,
- * giving up if terminating or addWorker fails
- */
- private void ensureEnoughTotalWorkers() {
- int wc;
- while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
- runState < TERMINATING) {
- if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
- addWorker() == null))
- break;
+ // Submissions
+
+ /**
+ * Unless shutting down, adds the given task to a submission queue
+ * at submitter's current queue index (modulo submission
+ * range). Only the most common path is directly handled in this
+ * method. All others are relayed to fullExternalPush.
+ *
+ * @param task the task. Caller must ensure non-null.
+ */
+ final void externalPush(ForkJoinTask> task) {
+ WorkQueue[] ws; WorkQueue q; Submitter z; int m; ForkJoinTask>[] a;
+ if ((z = submitters.get()) != null && plock > 0 &&
+ (ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
+ int s = q.top, n;
+ if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
+ U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
+ task);
+ q.top = s + 1; // push on to deque
+ q.qlock = 0;
+ if (n <= 1)
+ signalWork(q, 1);
+ return;
+ }
+ q.qlock = 0;
}
+ fullExternalPush(task);
}
/**
- * Final callback from terminating worker. Removes record of
- * worker from array, and adjusts counts. If pool is shutting
- * down, tries to complete terminatation, else possibly replaces
- * the worker.
- *
- * @param w the worker
- */
- final void workerTerminated(ForkJoinWorkerThread w) {
- if (w.active) { // force inactive
- w.active = false;
- do {} while (!tryDecrementActiveCount());
- }
- forgetWorker(w);
-
- // Decrement total count, and if was running, running count
- // Spin (waiting for other updates) if either would be negative
- int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
- int unit = ONE_TOTAL + nr;
- for (;;) {
- int wc = workerCounts;
- int rc = wc & RUNNING_COUNT_MASK;
- if (rc - nr < 0 || (wc >>> TOTAL_COUNT_SHIFT) == 0)
- Thread.yield(); // back off if waiting for other updates
- else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - unit))
- break;
+ * Full version of externalPush. This method is called, among
+ * other times, upon the first submission of the first task to the
+ * pool, so must perform secondary initialization: creating
+ * workQueue array and setting plock to a valid value. It also
+ * detects first submission by an external thread by looking up
+ * its ThreadLocal, and creates a new shared queue if the one at
+ * index if empty or contended. The lock bodies must be
+ * exception-free (so no try/finally) so we optimistically
+ * allocate new queues/arrays outside the locks and throw them
+ * away if (very rarely) not needed. Note that the plock seq value
+ * can eventually wrap around zero, but if so harmlessly fails to
+ * reinitialize.
+ */
+ private void fullExternalPush(ForkJoinTask> task) {
+ for (Submitter z = null;;) {
+ WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
+ if ((ps = plock) < 0)
+ throw new RejectedExecutionException();
+ else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) {
+ int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2;
+ n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
+ WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
+ if ((ps & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ if ((ws = workQueues) == null)
+ workQueues = nws;
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ else if (z == null && (z = submitters.get()) == null) {
+ if (U.compareAndSwapInt(this, INDEXSEED,
+ s = indexSeed, s += SEED_INCREMENT) &&
+ s != 0) // skip 0
+ submitters.set(z = new Submitter(s));
+ }
+ else {
+ int k = (r = z.seed) & m & SQMASK;
+ if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
+ (q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ WorkQueue w = null;
+ if ((ws = workQueues) != null && k < ws.length &&
+ (w = ws[k]) == null)
+ ws[k] = q;
+ else
+ q = w;
+ int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ if (q != null && q.qlock == 0 && q.fullPush(task, false))
+ return;
+ r ^= r << 13; // same xorshift as WorkQueues
+ r ^= r >>> 17;
+ z.seed = r ^= r << 5; // move to a different index
+ }
}
-
- accumulateStealCount(w); // collect final count
- if (!tryTerminate(false))
- ensureEnoughTotalWorkers();
}
- // Waiting for and signalling events
+ // Maintaining ctl counts
/**
- * Releases workers blocked on a count not equal to current count.
+ * Increments active count; mainly called upon return from blocking.
*/
- private 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));
+ }
+
+ /**
+ * Tries to create (at most one) or activate (possibly several)
+ * workers if too few are active. On contention failure, continues
+ * until at least one worker is signalled or the given queue is
+ * empty or all workers are active.
+ *
+ * @param q if non-null, the queue holding tasks to be signalled
+ * @param signals the target number of signals.
+ */
+ final void signalWork(WorkQueue q, int signals) {
+ long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
+ while ((u = (int)((c = ctl) >>> 32)) < 0) {
+ if ((e = (int)c) > 0) {
+ if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
+ (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
+ long nc = (((long)(w.nextWait & E_MASK)) |
+ ((long)(u + UAC_UNIT) << 32));
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ if (--signals <= 0)
+ break;
+ }
+ else
+ signals = 1;
+ if ((q != null && q.queueSize() == 0))
+ break;
+ }
+ else
+ break;
+ }
+ else if (e == 0 && (u & SHORT_SIGN) != 0) {
+ long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
+ ((u + UAC_UNIT) & UAC_MASK)) << 32;
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ ForkJoinWorkerThread wt = null;
+ Throwable ex = null;
+ boolean started = false;
+ try {
+ ForkJoinWorkerThreadFactory fac;
+ if ((fac = factory) != null &&
+ (wt = fac.newThread(this)) != null) {
+ wt.start();
+ started = true;
+ }
+ } catch (Throwable rex) {
+ ex = rex;
+ }
+ if (!started)
+ deregisterWorker(wt, ex); // adjust counts on failure
+ break;
+ }
+ }
+ else
+ break;
}
}
+ // Scanning for tasks
+
/**
- * Ensures eventCount on exit is different (mod 2^32) than on
- * entry and wakes up all waiters
+ * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
*/
- private void signalEvent() {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
- c = eventCount, c+1));
- releaseWaiters();
+ final void runWorker(WorkQueue w) {
+ // initialize queue array in this thread
+ w.array = new ForkJoinTask>[WorkQueue.INITIAL_QUEUE_CAPACITY];
+ do { w.runTask(scan(w)); } while (w.qlock >= 0);
}
/**
- * Advances eventCount and releases waiters until interference by
- * other releasing threads is detected.
- */
- final void signalWork() {
- // EventCount CAS failures are OK -- any change in count suffices.
- 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;
+ * Scans for and, if found, returns one task, else possibly
+ * inactivates the worker. This method operates on single reads of
+ * volatile state and is designed to be re-invoked continuously,
+ * in part because it returns upon detecting inconsistencies,
+ * contention, or state changes that indicate possible success on
+ * re-invocation.
+ *
+ * The scan searches for tasks across a random permutation of
+ * queues (starting at a random index and stepping by a random
+ * relative prime, checking each at least once). The scan
+ * terminates upon either finding a non-empty queue, or completing
+ * the sweep. If the worker is not inactivated, it takes and
+ * returns a task from this queue. Otherwise, if not activated, it
+ * signals workers (that may include itself) and returns so caller
+ * can retry. Also returns for trtry if the worker array may have
+ * changed during an empty scan. On failure to find a task, we
+ * take one of the following actions, after which the caller will
+ * retry calling this method unless terminated.
+ *
+ * * If pool is terminating, terminate the worker.
+ *
+ * * If not already enqueued, try to inactivate and enqueue the
+ * worker on wait queue. Or, if inactivating has caused the pool
+ * to be quiescent, relay to idleAwaitWork to check for
+ * termination and possibly shrink pool.
+ *
+ * * If already enqueued and none of the above apply, possibly
+ * (with 1/2 probability) park awaiting signal, else lingering to
+ * help scan and signal.
+ *
+ * @param w the worker (via its WorkQueue)
+ * @return a task or null if none found
+ */
+ private final ForkJoinTask> scan(WorkQueue w) {
+ WorkQueue[] ws; WorkQueue q; // first update random seed
+ int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
+ int ps = plock, m; // volatile read order matters
+ if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
+ int ec = w.eventCount; // ec is negative if inactive
+ int step = (r >>> 16) | 1; // relatively prime
+ for (int j = (m + 1) << 2; ; --j, r += step) {
+ ForkJoinTask> t; ForkJoinTask>[] a; int b, n;
+ if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
+ (a = q.array) != null) { // probably nonempty
+ int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ t = (ForkJoinTask>)U.getObjectVolatile(a, i);
+ if (q.base == b && ec >= 0 && t != null &&
+ U.compareAndSwapObject(a, i, t, null)) {
+ if ((n = q.top - (q.base = b + 1)) > 0)
+ signalWork(q, n);
+ return t; // taken
+ }
+ if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) {
+ if ((n = q.queueSize() - 1) > 0)
+ signalWork(q, n);
+ break; // let caller retry after signal
+ }
+ }
+ else if (j < 0) { // end of scan
+ long c = ctl; int e;
+ if (plock != ps) // incomplete sweep
+ break;
+ if ((e = (int)c) < 0) // pool is terminating
+ w.qlock = -1;
+ else if (ec >= 0) { // try to enqueue/inactivate
+ long nc = ((long)ec |
+ ((c - AC_UNIT) & (AC_MASK|TC_MASK)));
+ w.nextWait = e;
+ w.eventCount = ec | INT_SIGN; // mark as inactive
+ if (ctl != c ||
+ !U.compareAndSwapLong(this, CTL, c, nc))
+ w.eventCount = ec; // unmark on CAS failure
+ else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
+ idleAwaitWork(w, nc, c); // quiescent
+ }
+ else if (w.seed >= 0 && w.eventCount < 0) {
+ Thread wt = Thread.currentThread();
+ Thread.interrupted(); // clear status
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt; // emulate LockSupport.park
+ if (w.eventCount < 0) // recheck
+ U.park(false, 0L);
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ }
+ break;
+ }
}
}
+ return null;
}
/**
- * If 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();
+ * If inactivating worker w has caused the pool to become
+ * quiescent, checks for pool termination, and, so long as this is
+ * not the only worker, waits for event for up to a given
+ * duration. On timeout, if ctl has not changed, terminates the
+ * worker, which will in turn wake up another worker to possibly
+ * repeat this process.
+ *
+ * @param w the calling worker
+ * @param currentCtl the ctl value triggering possible quiescence
+ * @param prevCtl the ctl value to restore if thread is terminated
+ */
+ private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
+ if (w.eventCount < 0 &&
+ (this == commonPool || !tryTerminate(false, false)) &&
+ (int)prevCtl != 0) {
+ int dc = -(short)(currentCtl >>> TC_SHIFT);
+ long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
+ long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
+ Thread wt = Thread.currentThread();
+ while (ctl == currentCtl) {
+ Thread.interrupted(); // timed variant of version in scan()
+ U.putObject(wt, PARKBLOCKER, this);
+ w.parker = wt;
+ if (ctl == currentCtl)
+ U.park(false, parkTime);
+ w.parker = null;
+ U.putObject(wt, PARKBLOCKER, null);
+ if (ctl != currentCtl)
+ break;
+ if (deadline - System.nanoTime() <= 0L &&
+ U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
+ w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
+ w.qlock = -1; // shrink
break;
}
}
- 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.
+ * Scans through queues looking for work while joining a task;
+ * if any are present, signals.
*
- * * 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).
+ * @param task to return early if done
+ * @param origin an index to start scan
*/
- final void preStep(ForkJoinWorkerThread w, boolean worked) {
- boolean active = w.active;
- boolean inactivate = !worked & active;
- for (;;) {
- if (inactivate) {
- int rs = runState;
- if (UNSAFE.compareAndSwapInt(this, runStateOffset,
- rs, rs - ONE_ACTIVE))
- inactivate = active = w.active = false;
- }
- int wc = workerCounts;
- if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
- if (!worked)
- eventSync(w);
- return;
+ final int helpSignal(ForkJoinTask> task, int origin) {
+ WorkQueue[] ws; WorkQueue q; int m, n, s;
+ if (task != null && (ws = workQueues) != null &&
+ (m = ws.length - 1) >= 0) {
+ for (int i = 0; i <= m; ++i) {
+ if ((s = task.status) < 0)
+ return s;
+ if ((q = ws[(i + origin) & m]) != null &&
+ (n = q.queueSize()) > 0) {
+ signalWork(q, n);
+ if ((int)(ctl >> AC_SHIFT) >= 0)
+ break;
+ }
}
- if (!(inactivate |= active) && // must inactivate to suspend
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- wc, wc - ONE_RUNNING) &&
- !w.suspendAsSpare()) // false if trimmed
- return;
}
+ return 0;
}
/**
- * Tries to decrement running count, and if so, possibly creates
- * or resumes compensating threads before blocking on task joinMe.
- * This code is sprawled out with manual inlining to evade some
- * JIT oddities.
- *
- * @param joinMe the task to join
- * @return task status on exit
- */
- final int tryAwaitJoin(ForkJoinTask> joinMe) {
- int cw = workerCounts; // read now to spoil CAS if counts change as ...
- releaseWaiters(); // ... a byproduct of releaseWaiters
- int stat = joinMe.status;
- if (stat >= 0 && // inline variant of tryDecrementRunningCount
- (cw & RUNNING_COUNT_MASK) > 0 &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- cw, cw - ONE_RUNNING)) {
- int pc = parallelism;
- int scans = 0; // to require confirming passes to add threads
- outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) {
- if ((stat = joinMe.status) < 0)
- break;
- ForkJoinWorkerThread spare = null;
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- 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 so long that they are likely cyclic.
+ *
+ * @param joiner the joining worker
+ * @param task the task to join
+ * @return 0 if no progress can be made, negative if task
+ * known complete, else positive
+ */
+ private int tryHelpStealer(WorkQueue joiner, ForkJoinTask> task) {
+ int stat = 0, steps = 0; // bound to avoid cycles
+ if (joiner != null && task != null) { // hoist null checks
+ restart: for (;;) {
+ ForkJoinTask> subtask = task; // current target
+ for (WorkQueue j = joiner, v;;) { // v is stealer of subtask
+ WorkQueue[] ws; int m, s, h;
+ if ((s = task.status) < 0) {
+ stat = s;
+ break restart;
}
- }
- if ((stat = joinMe.status) < 0) // recheck to narrow race
- break;
- int wc = workerCounts;
- int rc = wc & RUNNING_COUNT_MASK;
- if (rc >= pc)
- break;
- if (spare != null) {
- if (spare.tryUnsuspend()) {
- int c; // inline incrementRunningCount
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
- LockSupport.unpark(spare);
- break;
+ if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
+ break restart; // shutting down
+ if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
+ v.currentSteal != subtask) {
+ for (int origin = h;;) { // find stealer
+ if (((h = (h + 2) & m) & 15) == 1 &&
+ (subtask.status < 0 || j.currentJoin != subtask))
+ continue restart; // occasional staleness check
+ if ((v = ws[h]) != null &&
+ v.currentSteal == subtask) {
+ j.stealHint = h; // save hint
+ break;
+ }
+ if (h == origin)
+ break restart; // cannot find stealer
+ }
+ }
+ for (;;) { // help stealer or descend to its stealer
+ ForkJoinTask[] a; int b;
+ if (subtask.status < 0) // surround probes with
+ continue restart; // consistency checks
+ if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
+ int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
+ ForkJoinTask> t =
+ (ForkJoinTask>)U.getObjectVolatile(a, i);
+ if (subtask.status < 0 || j.currentJoin != subtask ||
+ v.currentSteal != subtask)
+ continue restart; // stale
+ stat = 1; // apparent progress
+ if (t != null && v.base == b &&
+ U.compareAndSwapObject(a, i, t, null)) {
+ v.base = b + 1; // help stealer
+ joiner.runSubtask(t);
+ }
+ else if (v.base == b && ++steps == MAX_HELP)
+ break restart; // v apparently stalled
+ }
+ else { // empty -- try to descend
+ ForkJoinTask> next = v.currentJoin;
+ if (subtask.status < 0 || j.currentJoin != subtask ||
+ v.currentSteal != subtask)
+ continue restart; // stale
+ else if (next == null || ++steps == MAX_HELP)
+ break restart; // dead-end or maybe cyclic
+ else {
+ subtask = next;
+ j = v;
+ break;
+ }
+ }
}
- continue;
- }
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int sc = tc - pc;
- if (rc > 0) {
- int p = pc;
- int s = sc;
- while (s-- >= 0) { // try keeping 3/4 live
- if (rc > (p -= (p >>> 2) + 1))
- break outer;
- }
- }
- if (scans++ > sc && tc < MAX_THREADS &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
- wc + (ONE_RUNNING|ONE_TOTAL))) {
- addWorker();
- break;
}
}
- if (stat >= 0)
- stat = joinMe.internalAwaitDone();
- int c; // inline incrementRunningCount
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
}
return stat;
}
/**
- * Same idea as (and mostly pasted from) tryAwaitJoin, but
- * self-contained
- */
- final void awaitBlocker(ManagedBlocker blocker)
- throws InterruptedException {
- for (;;) {
- if (blocker.isReleasable())
- return;
- int cw = workerCounts;
- releaseWaiters();
- if ((cw & RUNNING_COUNT_MASK) > 0 &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset,
- cw, cw - ONE_RUNNING))
- break;
- }
- boolean done = false;
- int pc = parallelism;
- int scans = 0;
- outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) {
- if (done = blocker.isReleasable())
- break;
- ForkJoinWorkerThread spare = null;
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null && w.isSuspended()) {
- spare = w;
- break;
+ * Analog of tryHelpStealer for CountedCompleters. Tries to steal
+ * and run tasks within the target's computation
+ *
+ * @param task the task to join
+ * @param mode if shared, exit upon completing any task
+ * if all workers are active
+ *
+ */
+ private int helpComplete(ForkJoinTask> task, int mode) {
+ WorkQueue[] ws; WorkQueue q; int m, n, s;
+ if (task != null && (ws = workQueues) != null &&
+ (m = ws.length - 1) >= 0) {
+ for (int j = 1, origin = j;;) {
+ if ((s = task.status) < 0)
+ return s;
+ if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
+ origin = j;
+ if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
+ break;
}
- }
- if (done = blocker.isReleasable())
- break;
- int wc = workerCounts;
- int rc = wc & RUNNING_COUNT_MASK;
- if (rc >= pc)
- break;
- if (spare != null) {
- if (spare.tryUnsuspend()) {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
- LockSupport.unpark(spare);
+ else if ((j = (j + 2) & m) == origin)
break;
+ }
+ }
+ return 0;
+ }
+
+ /**
+ * Tries to decrement active count (sometimes implicitly) and
+ * possibly release or create a compensating worker in preparation
+ * for blocking. Fails on contention or termination. Otherwise,
+ * adds a new thread if no idle workers are available and pool
+ * may become starved.
+ */
+ final boolean tryCompensate() {
+ int pc = parallelism, e, u, i, tc; long c;
+ WorkQueue[] ws; WorkQueue w; Thread p;
+ if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
+ if (e != 0 && (i = e & SMASK) < ws.length &&
+ (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
+ long nc = ((long)(w.nextWait & E_MASK) |
+ (c & (AC_MASK|TC_MASK)));
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ return true; // replace with idle worker
}
- continue;
}
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int sc = tc - pc;
- if (rc > 0) {
- int p = pc;
- int s = sc;
- while (s-- >= 0) {
- if (rc > (p -= (p >>> 2) + 1))
- break outer;
- }
- }
- if (scans++ > sc && tc < MAX_THREADS &&
- UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
- wc + (ONE_RUNNING|ONE_TOTAL))) {
- addWorker();
- break;
+ else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
+ (u >> UAC_SHIFT) + pc > 1) {
+ long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc))
+ return true; // no compensation
+ }
+ else if ((tc = u + pc) < MAX_CAP) {
+ long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
+ if (U.compareAndSwapLong(this, CTL, c, nc)) {
+ Throwable ex = null;
+ ForkJoinWorkerThread wt = null;
+ try {
+ ForkJoinWorkerThreadFactory fac;
+ if ((fac = factory) != null &&
+ (wt = fac.newThread(this)) != null) {
+ wt.start();
+ return true;
+ }
+ } catch (Throwable rex) {
+ ex = rex;
+ }
+ deregisterWorker(wt, ex); // adjust counts etc
+ }
}
}
- try {
- if (!done)
- do {} while (!blocker.isReleasable() &&
- !blocker.block());
- } finally {
- int c;
- do {} while (!UNSAFE.compareAndSwapInt
- (this, workerCountsOffset,
- c = workerCounts, c + ONE_RUNNING));
- }
- }
+ return false;
+ }
/**
- * Possibly initiates and/or completes termination.
+ * Helps and/or blocks until the given task is done.
*
- * @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
+ * @return task status on exit
*/
- 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;
-
- if (advanceRunLevel(TERMINATING))
- startTerminating();
-
- // Finish now if all threads terminated; else in some subsequent call
- if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
- advanceRunLevel(TERMINATED);
- termination.arrive();
+ final int awaitJoin(WorkQueue joiner, ForkJoinTask> task) {
+ int s = 0;
+ if (joiner != null && task != null && (s = task.status) >= 0) {
+ ForkJoinTask> prevJoin = joiner.currentJoin;
+ joiner.currentJoin = task;
+ do {} while ((s = task.status) >= 0 &&
+ joiner.queueSize() > 0 &&
+ joiner.tryRemoveAndExec(task)); // process local tasks
+ if (s >= 0 && (s = task.status) >= 0 &&
+ (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
+ (task instanceof CountedCompleter))
+ s = helpComplete(task, LIFO_QUEUE);
+ while (s >= 0 && (s = task.status) >= 0) {
+ if ((joiner.queueSize() > 0 || // try helping
+ (s = tryHelpStealer(joiner, task)) == 0) &&
+ (s = task.status) >= 0 && tryCompensate()) {
+ if (task.trySetSignal() && (s = task.status) >= 0) {
+ synchronized (task) {
+ if (task.status >= 0) {
+ try { // see ForkJoinTask
+ task.wait(); // for explanation
+ } catch (InterruptedException ie) {
+ }
+ }
+ else
+ task.notifyAll();
+ }
+ }
+ long c; // re-activate
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ }
+ }
+ joiner.currentJoin = prevJoin;
}
- return true;
+ return s;
}
/**
- * Actions on transition to TERMINATING
+ * Stripped-down variant of awaitJoin used by timed joins. Tries
+ * to help join only while there is continuous progress. (Caller
+ * will then enter a timed wait.)
+ *
+ * @param joiner the joining worker
+ * @param task the task
*/
- private void startTerminating() {
- for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
- cancelSubmissions();
- shutdownWorkers();
- cancelWorkerTasks();
- signalEvent();
- interruptWorkers();
+ final void helpJoinOnce(WorkQueue joiner, ForkJoinTask> task) {
+ int s;
+ if (joiner != null && task != null && (s = task.status) >= 0) {
+ ForkJoinTask> prevJoin = joiner.currentJoin;
+ joiner.currentJoin = task;
+ do {} while ((s = task.status) >= 0 &&
+ joiner.queueSize() > 0 &&
+ joiner.tryRemoveAndExec(task));
+ if (s >= 0 && (s = task.status) >= 0 &&
+ (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
+ (task instanceof CountedCompleter))
+ s = helpComplete(task, LIFO_QUEUE);
+ if (s >= 0 && joiner.queueSize() == 0) {
+ do {} while (task.status >= 0 &&
+ tryHelpStealer(joiner, task) > 0);
+ }
+ joiner.currentJoin = prevJoin;
}
}
/**
- * Clear out and cancel submissions, ignoring exceptions
+ * Returns a (probably) non-empty steal queue, if one is found
+ * during a random, then cyclic scan, else null. This method must
+ * be retried by caller if, by the time it tries to use the queue,
+ * it is empty.
+ * @param r a (random) seed for scanning
*/
- private void cancelSubmissions() {
- ForkJoinTask> task;
- while ((task = submissionQueue.poll()) != null) {
- try {
- task.cancel(false);
- } catch (Throwable ignore) {
+ private WorkQueue findNonEmptyStealQueue(int r) {
+ int step = (r >>> 16) | 1;
+ for (WorkQueue[] ws;;) {
+ int ps = plock, m;
+ if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
+ return null;
+ for (int j = (m + 1) << 2; ; r += step) {
+ WorkQueue q = ws[((r << 1) | 1) & m];
+ if (q != null && q.queueSize() > 0)
+ return q;
+ else if (--j < 0) {
+ if (plock == ps)
+ return null;
+ break;
+ }
}
}
}
/**
- * Sets all worker run states to at least shutdown,
- * also resuming suspended workers
- */
- private void shutdownWorkers() {
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- w.shutdown();
+ * Runs tasks until {@code isQuiescent()}. We piggyback on
+ * active count ctl maintenance, but rather than blocking
+ * when tasks cannot be found, we rescan until all others cannot
+ * find tasks either.
+ */
+ final void helpQuiescePool(WorkQueue w) {
+ for (boolean active = true;;) {
+ ForkJoinTask> localTask; // exhaust local queue
+ while ((localTask = w.nextLocalTask()) != null)
+ localTask.doExec();
+ // Similar to loop in scan(), but ignoring submissions
+ WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
+ if (q != null) {
+ ForkJoinTask> t; int b;
+ if (!active) { // re-establish active count
+ long c;
+ active = true;
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ }
+ if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
+ w.runSubtask(t);
+ }
+ else {
+ long c;
+ if (active) { // decrement active count without queuing
+ active = false;
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c -= AC_UNIT));
+ }
+ else
+ c = ctl; // re-increment on exit
+ if ((int)(c >> AC_SHIFT) + parallelism == 0) {
+ do {} while (!U.compareAndSwapLong
+ (this, CTL, c = ctl, c + AC_UNIT));
+ break;
+ }
+ }
}
}
/**
- * Clears out and cancels all locally queued tasks
+ * Gets and removes a local or stolen task for the given worker.
+ *
+ * @return a task, if available
*/
- private void cancelWorkerTasks() {
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- w.cancelTasks();
+ final ForkJoinTask> nextTaskFor(WorkQueue w) {
+ for (ForkJoinTask> t;;) {
+ WorkQueue q; int b;
+ if ((t = w.nextLocalTask()) != null)
+ return t;
+ if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
+ return null;
+ if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
+ return t;
}
}
/**
- * Unsticks all workers blocked on joins etc
+ * Returns a cheap heuristic guide for task partitioning when
+ * programmers, frameworks, tools, or languages have little or no
+ * idea about task granularity. In essence by offering this
+ * method, we ask users only about tradeoffs in overhead vs
+ * expected throughput and its variance, rather than how finely to
+ * partition tasks.
+ *
+ * In a steady state strict (tree-structured) computation, each
+ * thread makes available for stealing enough tasks for other
+ * threads to remain active. Inductively, if all threads play by
+ * the same rules, each thread should make available only a
+ * constant number of tasks.
+ *
+ * The minimum useful constant is just 1. But using a value of 1
+ * would require immediate replenishment upon each steal to
+ * maintain enough tasks, which is infeasible. Further,
+ * partitionings/granularities of offered tasks should minimize
+ * steal rates, which in general means that threads nearer the top
+ * of computation tree should generate more than those nearer the
+ * bottom. In perfect steady state, each thread is at
+ * approximately the same level of computation tree. However,
+ * producing extra tasks amortizes the uncertainty of progress and
+ * diffusion assumptions.
+ *
+ * So, users will want to use values larger, but not much larger
+ * than 1 to both smooth over transient shortages and hedge
+ * against uneven progress; as traded off against the cost of
+ * extra task overhead. We leave the user to pick a threshold
+ * value to compare with the results of this call to guide
+ * decisions, but recommend values such as 3.
+ *
+ * When all threads are active, it is on average OK to estimate
+ * surplus strictly locally. In steady-state, if one thread is
+ * maintaining say 2 surplus tasks, then so are others. So we can
+ * just use estimated queue length. However, this strategy alone
+ * leads to serious mis-estimates in some non-steady-state
+ * conditions (ramp-up, ramp-down, other stalls). We can detect
+ * many of these by further considering the number of "idle"
+ * threads, that are known to have zero queued tasks, so
+ * compensate by a factor of (#idle/#active) threads.
+ *
+ * Note: The approximation of #busy workers as #active workers is
+ * not very good under current signalling scheme, and should be
+ * improved.
+ */
+ static int getSurplusQueuedTaskCount() {
+ Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
+ if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
+ int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
+ int p = (pool = wt.pool).parallelism;
+ int a = (int)(pool.ctl >> AC_SHIFT) + p;
+ return q.top - b - (a > (p >>>= 1) ? 0 :
+ a > (p >>>= 1) ? 1 :
+ a > (p >>>= 1) ? 2 :
+ a > (p >>>= 1) ? 4 :
+ 8);
+ }
+ return 0;
+ }
+
+ // Termination
+
+ /**
+ * Possibly initiates and/or completes termination. The caller
+ * triggering termination runs three passes through workQueues:
+ * (0) Setting termination status, followed by wakeups of queued
+ * workers; (1) cancelling all tasks; (2) interrupting lagging
+ * threads (likely in external tasks, but possibly also blocked in
+ * joins). Each pass repeats previous steps because of potential
+ * lagging thread creation.
+ *
+ * @param now if true, unconditionally terminate, else only
+ * if no work and no active workers
+ * @param enable if true, enable shutdown when next possible
+ * @return true if now terminating or terminated
*/
- private void interruptWorkers() {
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null && !w.isTerminated()) {
- try {
- w.interrupt();
- } catch (SecurityException ignore) {
+ private boolean tryTerminate(boolean now, boolean enable) {
+ if (this == commonPool) // cannot shut down
+ return false;
+ for (long c;;) {
+ if (((c = ctl) & STOP_BIT) != 0) { // already terminating
+ if ((short)(c >>> TC_SHIFT) == -parallelism) {
+ synchronized (this) {
+ notifyAll(); // signal when 0 workers
+ }
+ }
+ return true;
+ }
+ if (plock >= 0) { // not yet enabled
+ int ps;
+ if (!enable)
+ return false;
+ if (((ps = plock) & PL_LOCK) != 0 ||
+ !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
+ ps = acquirePlock();
+ int nps = SHUTDOWN;
+ if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
+ releasePlock(nps);
+ }
+ if (!now) { // check if idle & no tasks
+ if ((int)(c >> AC_SHIFT) != -parallelism ||
+ hasQueuedSubmissions())
+ return false;
+ // Check for unqueued inactive workers. One pass suffices.
+ WorkQueue[] ws = workQueues; WorkQueue w;
+ if (ws != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && w.eventCount >= 0)
+ return false;
+ }
+ }
+ }
+ if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
+ for (int pass = 0; pass < 3; ++pass) {
+ WorkQueue[] ws = workQueues;
+ if (ws != null) {
+ WorkQueue w;
+ int n = ws.length;
+ for (int i = 0; i < n; ++i) {
+ if ((w = ws[i]) != null) {
+ w.qlock = -1;
+ if (pass > 0) {
+ w.cancelAll();
+ if (pass > 1)
+ w.interruptOwner();
+ }
+ }
+ }
+ // Wake up workers parked on event queue
+ int i, e; long cc; Thread p;
+ while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
+ (i = e & SMASK) < n &&
+ (w = ws[i]) != null) {
+ long nc = ((long)(w.nextWait & E_MASK) |
+ ((cc + AC_UNIT) & AC_MASK) |
+ (cc & (TC_MASK|STOP_BIT)));
+ if (w.eventCount == (e | INT_SIGN) &&
+ U.compareAndSwapLong(this, CTL, cc, nc)) {
+ w.eventCount = (e + E_SEQ) & E_MASK;
+ w.qlock = -1;
+ if ((p = w.parker) != null)
+ U.unpark(p);
+ }
+ }
+ }
}
}
}
}
- // misc support for ForkJoinWorkerThread
+ // external operations on common pool
+
+ /**
+ * Returns common pool queue for a thread that has submitted at
+ * least one task.
+ */
+ static WorkQueue commonSubmitterQueue() {
+ ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
+ return ((z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0) ?
+ ws[m & z.seed & SQMASK] : null;
+ }
+
+ /**
+ * Tries to pop the given task from submitter's queue in common pool.
+ */
+ static boolean tryExternalUnpush(ForkJoinTask> t) {
+ ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
+ ForkJoinTask>[] a; int m, s; long j;
+ if ((z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ (s = q.top) != q.base &&
+ (a = q.array) != null &&
+ U.getObjectVolatile
+ (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s && // recheck
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ q.qlock = 0;
+ return true;
+ }
+ q.qlock = 0;
+ }
+ return false;
+ }
/**
- * Returns pool number
+ * Tries to pop and run local tasks within the same computation
+ * as the given root. On failure, tries to help complete from
+ * other queues via helpComplete.
*/
- final int getPoolNumber() {
- return poolNumber;
+ private void externalHelpComplete(WorkQueue q, ForkJoinTask> root) {
+ ForkJoinTask>[] a; int m;
+ if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
+ root != null && root.status >= 0) {
+ for (;;) {
+ int s; Object o; CountedCompleter> task = null;
+ if ((s = q.top) - q.base > 0) {
+ long j = ((m & (s - 1)) << ASHIFT) + ABASE;
+ if ((o = U.getObject(a, j)) != null &&
+ (o instanceof CountedCompleter)) {
+ CountedCompleter> t = (CountedCompleter>)o, r = t;
+ do {
+ if (r == root) {
+ if (U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ task = t;
+ }
+ q.qlock = 0;
+ }
+ break;
+ }
+ } while ((r = r.completer) != null);
+ }
+ }
+ if (task != null)
+ task.doExec();
+ if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0)
+ break;
+ if (task == null) {
+ if (helpSignal(root, q.poolIndex) >= 0)
+ helpComplete(root, SHARED_QUEUE);
+ break;
+ }
+ }
+ }
}
/**
- * Accumulates steal count from a worker, clearing
- * the worker's value
+ * Tries to help execute or signal availability of the given task
+ * from submitter's queue in common pool.
*/
- 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));
+ static void externalHelpJoin(ForkJoinTask> t) {
+ // Some hard-to-avoid overlap with tryExternalUnpush
+ ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
+ ForkJoinTask>[] a; int m, s, n; long j;
+ if (t != null && t.status >= 0 &&
+ (z = submitters.get()) != null &&
+ (p = commonPool) != null &&
+ (ws = p.workQueues) != null &&
+ (m = ws.length - 1) >= 0 &&
+ (q = ws[m & z.seed & SQMASK]) != null &&
+ (a = q.array) != null) {
+ if ((s = q.top) != q.base &&
+ U.getObjectVolatile
+ (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
+ U.compareAndSwapInt(q, QLOCK, 0, 1)) {
+ if (q.array == a && q.top == s &&
+ U.compareAndSwapObject(a, j, t, null)) {
+ q.top = s - 1;
+ q.qlock = 0;
+ t.doExec();
+ }
+ else
+ q.qlock = 0;
+ }
+ if (t.status >= 0) {
+ if (t instanceof CountedCompleter)
+ p.externalHelpComplete(q, t);
+ else
+ p.helpSignal(t, q.poolIndex);
+ }
}
}
/**
- * Returns the approximate (non-atomic) number of idle threads per
- * active thread.
+ * Restricted version of helpQuiescePool for external callers
*/
- final int idlePerActive() {
- int pc = parallelism; // use targeted parallelism, not rc
- int ac = runState; // no mask -- artifically boosts during shutdown
- // Use exact results for small values, saturate past 4
- return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
+ static void externalHelpQuiescePool() {
+ ForkJoinPool p; ForkJoinTask> t; WorkQueue q; int b;
+ int r = ThreadLocalRandom.current().nextInt();
+ if ((p = commonPool) != null &&
+ (q = p.findNonEmptyStealQueue(r)) != null &&
+ (b = q.base) - q.top < 0 &&
+ (t = q.pollAt(b)) != null)
+ t.doExec();
}
- // Public and protected methods
+ // Exported methods
// Constructors
@@ -1154,15 +2415,15 @@ public class ForkJoinPool extends Abstra
* 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 null
.
- * @param asyncMode if true,
+ * @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 false
.
+ * 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
@@ -1171,69 +2432,63 @@ public class ForkJoinPool extends Abstra
* because it does not hold {@link
* java.lang.RuntimePermission}{@code ("modifyThread")}
*/
- public ForkJoinPool(int parallelism,
+ 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_CAP)
throw new IllegalArgumentException();
this.parallelism = parallelism;
this.factory = factory;
this.ueh = handler;
- this.locallyFifo = asyncMode;
- int arraySize = initialArraySizeFor(parallelism);
- this.workers = new ForkJoinWorkerThread[arraySize];
- this.submissionQueue = new LinkedTransferQueue>();
- this.workerLock = new ReentrantLock();
- this.termination = new Phaser(1);
- this.poolNumber = poolNumberGenerator.incrementAndGet();
+ this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
+ long np = (long)(-parallelism); // offset ctl counts
+ this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
+ int pn = nextPoolId();
+ StringBuilder sb = new StringBuilder("ForkJoinPool-");
+ sb.append(Integer.toString(pn));
+ sb.append("-worker-");
+ this.workerNamePrefix = sb.toString();
}
/**
- * Returns 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;
+ * Constructor for common pool, suitable only for static initialization.
+ * Basically the same as above, but uses smallest possible initial footprint.
+ */
+ ForkJoinPool(int parallelism, long ctl,
+ ForkJoinWorkerThreadFactory factory,
+ Thread.UncaughtExceptionHandler handler) {
+ this.parallelism = parallelism;
+ this.ctl = ctl;
+ this.factory = factory;
+ this.ueh = handler;
+ this.localMode = LIFO_QUEUE;
+ this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
}
- // Execution methods
-
/**
- * Common code for execute, invoke and submit
+ * Returns the common pool instance.
+ *
+ * @return the common pool instance
*/
- private void doSubmit(ForkJoinTask task) {
- if (task == null)
- throw new NullPointerException();
- if (runState >= SHUTDOWN)
- throw new RejectedExecutionException();
- // Convert submissions to current pool into forks
- Thread t = Thread.currentThread();
- ForkJoinWorkerThread w;
- if ((t instanceof ForkJoinWorkerThread) &&
- (w = (ForkJoinWorkerThread) t).pool == this)
- w.pushTask(task);
- else {
- submissionQueue.offer(task);
- signalEvent();
- ensureEnoughTotalWorkers();
- }
+ public static ForkJoinPool commonPool() {
+ return commonPool; // cannot be null (if so, a static init error)
}
+ // Execution methods
+
/**
* Performs the given task, returning its result upon completion.
- * If the caller is already engaged in a fork/join computation in
- * the current pool, this method is equivalent in effect to
- * {@link ForkJoinTask#invoke}.
+ * 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
@@ -1242,15 +2497,14 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public T invoke(ForkJoinTask task) {
- doSubmit(task);
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(task);
return task.join();
}
/**
* Arranges for (asynchronous) execution of the given task.
- * If the caller is already engaged in a fork/join computation in
- * the current pool, this method is equivalent in effect to
- * {@link ForkJoinTask#fork}.
*
* @param task the task
* @throws NullPointerException if the task is null
@@ -1258,7 +2512,9 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public void execute(ForkJoinTask> task) {
- doSubmit(task);
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(task);
}
// AbstractExecutorService methods
@@ -1269,19 +2525,18 @@ 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;
else
- job = ForkJoinTask.adapt(task, null);
- doSubmit(job);
+ job = new ForkJoinTask.AdaptedRunnableAction(task);
+ externalPush(job);
}
/**
* Submits a ForkJoinTask for execution.
- * If the caller is already engaged in a fork/join computation in
- * the current pool, this method is equivalent in effect to
- * {@link ForkJoinTask#fork}.
*
* @param task the task to submit
* @return the task
@@ -1290,7 +2545,9 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public ForkJoinTask submit(ForkJoinTask task) {
- doSubmit(task);
+ if (task == null)
+ throw new NullPointerException();
+ externalPush(task);
return task;
}
@@ -1300,8 +2557,8 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public ForkJoinTask submit(Callable task) {
- ForkJoinTask job = ForkJoinTask.adapt(task);
- doSubmit(job);
+ ForkJoinTask job = new ForkJoinTask.AdaptedCallable(task);
+ externalPush(job);
return job;
}
@@ -1311,8 +2568,8 @@ public class ForkJoinPool extends Abstra
* scheduled for execution
*/
public ForkJoinTask submit(Runnable task, T result) {
- ForkJoinTask job = ForkJoinTask.adapt(task, result);
- doSubmit(job);
+ ForkJoinTask job = new ForkJoinTask.AdaptedRunnable(task, result);
+ externalPush(job);
return job;
}
@@ -1322,12 +2579,14 @@ 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;
else
- job = ForkJoinTask.adapt(task, null);
- doSubmit(job);
+ job = new ForkJoinTask.AdaptedRunnableAction(task);
+ externalPush(job);
return job;
}
@@ -1336,25 +2595,31 @@ public class ForkJoinPool extends Abstra
* @throws RejectedExecutionException {@inheritDoc}
*/
public List> invokeAll(Collection extends Callable> tasks) {
- ArrayList> forkJoinTasks =
- new ArrayList>(tasks.size());
- for (Callable task : tasks)
- forkJoinTasks.add(ForkJoinTask.adapt(task));
- invoke(new InvokeAll(forkJoinTasks));
-
+ // In previous versions of this class, this method constructed
+ // a task to run ForkJoinTask.invokeAll, but now external
+ // invocation of multiple tasks is at least as efficient.
+ List> fs = new ArrayList>(tasks.size());
+ // Workaround needed because method wasn't declared with
+ // wildcards in return type but should have been.
@SuppressWarnings({"unchecked", "rawtypes"})
- List> futures = (List>) (List) forkJoinTasks;
- return futures;
- }
+ List> futures = (List>) (List) fs;
- static final class InvokeAll extends RecursiveAction {
- final ArrayList> tasks;
- InvokeAll(ArrayList> tasks) { this.tasks = tasks; }
- public void compute() {
- try { invokeAll(tasks); }
- catch (Exception ignore) {}
+ boolean done = false;
+ try {
+ for (Callable t : tasks) {
+ ForkJoinTask f = new ForkJoinTask.AdaptedCallable(t);
+ externalPush(f);
+ fs.add(f);
+ }
+ for (ForkJoinTask f : fs)
+ f.quietlyJoin();
+ done = true;
+ return futures;
+ } finally {
+ if (!done)
+ for (ForkJoinTask f : fs)
+ f.cancel(false);
}
- private static final long serialVersionUID = -7914297376763021607L;
}
/**
@@ -1386,15 +2651,24 @@ public class ForkJoinPool extends Abstra
}
/**
+ * Returns the targeted parallelism level of the common pool.
+ *
+ * @return the targeted parallelism level of the common pool
+ */
+ public static int getCommonPoolParallelism() {
+ return commonPoolParallelism;
+ }
+
+ /**
* Returns the number of worker threads that have started but not
- * yet terminated. This result returned by this method may differ
+ * 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;
+ return parallelism + (short)(ctl >>> TC_SHIFT);
}
/**
@@ -1404,7 +2678,7 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if this pool uses async mode
*/
public boolean getAsyncMode() {
- return locallyFifo;
+ return localMode != 0;
}
/**
@@ -1416,7 +2690,15 @@ 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) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && w.isApparentlyUnblocked())
+ ++rc;
+ }
+ }
+ return rc;
}
/**
@@ -1427,7 +2709,8 @@ 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
}
/**
@@ -1442,7 +2725,7 @@ 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;
}
/**
@@ -1457,7 +2740,15 @@ public class ForkJoinPool extends Abstra
* @return the number of steals
*/
public long getStealCount() {
- return stealCount;
+ long count = stealCount;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.nsteals;
+ }
+ }
+ return count;
}
/**
@@ -1472,25 +2763,33 @@ public class ForkJoinPool extends Abstra
*/
public long getQueuedTaskCount() {
long count = 0;
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- count += w.getQueueSize();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
+ }
}
return count;
}
/**
* Returns an estimate of the number 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) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
+ }
+ }
+ return count;
}
/**
@@ -1500,7 +2799,14 @@ 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) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && w.queueSize() != 0)
+ return true;
+ }
+ }
+ return false;
}
/**
@@ -1511,7 +2817,14 @@ 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) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && (t = w.poll()) != null)
+ return t;
+ }
+ }
+ return null;
}
/**
@@ -1532,29 +2845,17 @@ public class ForkJoinPool extends Abstra
* @return the number of elements transferred
*/
protected int drainTasksTo(Collection super ForkJoinTask>> c) {
- int n = submissionQueue.drainTo(c);
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- n += w.drainTasksTo(c);
- }
- return n;
- }
-
- /**
- * Returns count of total parks by existing workers.
- * Used during development only since not meaningful to users.
- */
- private int collectParkCount() {
int count = 0;
- ForkJoinWorkerThread[] ws = workers;
- int nws = ws.length;
- for (int i = 0; i < nws; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- count += w.parkCount;
+ WorkQueue[] ws; WorkQueue w; ForkJoinTask> t;
+ if ((ws = workQueues) != null) {
+ for (int i = 0; i < ws.length; ++i) {
+ if ((w = ws[i]) != null) {
+ while ((t = w.poll()) != null) {
+ c.add(t);
+ ++count;
+ }
+ }
+ }
}
return count;
}
@@ -1567,18 +2868,38 @@ public class ForkJoinPool extends Abstra
* @return a string identifying this pool, as well as its state
*/
public String toString() {
- long st = getStealCount();
- long qt = getQueuedTaskCount();
- long qs = getQueuedSubmissionCount();
- int wc = workerCounts;
- int tc = wc >>> TOTAL_COUNT_SHIFT;
- int rc = wc & RUNNING_COUNT_MASK;
+ // Use a single pass through workQueues to collect counts
+ long qt = 0L, qs = 0L; int rc = 0;
+ long st = stealCount;
+ long c = ctl;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 0; i < ws.length; ++i) {
+ if ((w = ws[i]) != null) {
+ int size = w.queueSize();
+ if ((i & 1) == 0)
+ qs += size;
+ else {
+ qt += size;
+ st += w.nsteals;
+ if (w.isApparentlyUnblocked())
+ ++rc;
+ }
+ }
+ }
+ }
int pc = parallelism;
- int rs = runState;
- int ac = rs & ACTIVE_COUNT_MASK;
- // int pk = collectParkCount();
+ int tc = pc + (short)(c >>> TC_SHIFT);
+ int ac = pc + (int)(c >> AC_SHIFT);
+ if (ac < 0) // ignore transient negative
+ ac = 0;
+ String level;
+ if ((c & STOP_BIT) != 0)
+ level = (tc == 0) ? "Terminated" : "Terminating";
+ else
+ level = plock < 0 ? "Shutting down" : "Running";
return super.toString() +
- "[" + runLevelToString(rs) +
+ "[" + level +
", parallelism = " + pc +
", size = " + tc +
", active = " + ac +
@@ -1586,23 +2907,17 @@ public class ForkJoinPool extends Abstra
", steals = " + st +
", tasks = " + qt +
", submissions = " + qs +
- // ", parks = " + pk +
"]";
}
- 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.
- * Invocation has no additional effect if already shut down.
- * Tasks that are in the process of being submitted concurrently
- * during the course of this method may or may not be rejected.
+ * Possibly initiates an orderly shutdown in which previously
+ * submitted tasks are executed, but no new tasks will be
+ * accepted. Invocation has no effect on execution state if this
+ * is the {@link #commonPool}, and no additional effect if
+ * already shut down. Tasks that are in the process of being
+ * submitted concurrently during the course of this method may or
+ * may not be rejected.
*
* @throws SecurityException if a security manager exists and
* the caller is not permitted to modify threads
@@ -1611,19 +2926,20 @@ public class ForkJoinPool extends Abstra
*/
public void shutdown() {
checkPermission();
- advanceRunLevel(SHUTDOWN);
- tryTerminate(false);
+ tryTerminate(false, true);
}
/**
- * Attempts to cancel and/or stop all tasks, and reject all
- * subsequently submitted tasks. Tasks that are in the process of
- * being submitted or executed concurrently during the course of
- * this method may or may not be rejected. This method cancels
- * both existing and unexecuted tasks, in order to permit
- * termination in the presence of task dependencies. So the method
- * always returns an empty list (unlike the case for some other
- * Executors).
+ * Possibly attempts to cancel and/or stop all tasks, and reject
+ * all subsequently submitted tasks. Invocation has no effect on
+ * execution state if this is the {@link #commonPool}, and no
+ * additional effect if already shut down. Otherwise, tasks that
+ * are in the process of being submitted or executed concurrently
+ * during the course of this method may or may not be
+ * rejected. This method cancels both existing and unexecuted
+ * tasks, in order to permit termination in the presence of task
+ * dependencies. So the method always returns an empty list
+ * (unlike the case for some other Executors).
*
* @return an empty list
* @throws SecurityException if a security manager exists and
@@ -1633,7 +2949,7 @@ public class ForkJoinPool extends Abstra
*/
public List shutdownNow() {
checkPermission();
- tryTerminate(true);
+ tryTerminate(true, true);
return Collections.emptyList();
}
@@ -1643,7 +2959,9 @@ 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);
}
/**
@@ -1651,13 +2969,18 @@ 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);
}
/**
@@ -1666,13 +2989,15 @@ public class ForkJoinPool extends Abstra
* @return {@code true} if this pool has been shut down
*/
public boolean isShutdown() {
- return runState >= SHUTDOWN;
+ return plock < 0;
}
/**
- * Blocks until all tasks have completed execution after a shutdown
- * request, or the timeout occurs, or the current thread is
- * interrupted, whichever happens first.
+ * Blocks until all tasks have completed execution after a
+ * shutdown request, or the timeout occurs, or the current thread
+ * is interrupted, whichever happens first. Note that the {@link
+ * #commonPool()} never terminates until program shutdown so
+ * this method will always time out.
*
* @param timeout the maximum time to wait
* @param unit the time unit of the timeout argument
@@ -1682,22 +3007,41 @@ public class ForkJoinPool extends Abstra
*/
public boolean awaitTermination(long timeout, TimeUnit unit)
throws InterruptedException {
- try {
- return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0;
- } catch(TimeoutException ex) {
- return false;
+ long nanos = unit.toNanos(timeout);
+ if (isTerminated())
+ return true;
+ long startTime = System.nanoTime();
+ boolean terminated = false;
+ synchronized (this) {
+ for (long waitTime = nanos, millis = 0L;;) {
+ if (terminated = isTerminated() ||
+ waitTime <= 0L ||
+ (millis = unit.toMillis(waitTime)) <= 0L)
+ break;
+ wait(millis);
+ waitTime = nanos - (System.nanoTime() - startTime);
+ }
}
+ return terminated;
}
/**
* 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:
@@ -1715,6 +3059,26 @@ 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 {
/**
@@ -1757,10 +3121,35 @@ public class ForkJoinPool extends Abstra
public static void managedBlock(ManagedBlocker blocker)
throws InterruptedException {
Thread t = Thread.currentThread();
- if (t instanceof ForkJoinWorkerThread)
- ((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
+ if (t instanceof ForkJoinWorkerThread) {
+ ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
+ while (!blocker.isReleasable()) { // variant of helpSignal
+ WorkQueue[] ws; WorkQueue q; int m, n;
+ if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
+ for (int i = 0; i <= m; ++i) {
+ if (blocker.isReleasable())
+ return;
+ if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
+ p.signalWork(q, n);
+ if ((int)(p.ctl >> AC_SHIFT) >= 0)
+ break;
+ }
+ }
+ }
+ if (p.tryCompensate()) {
+ try {
+ do {} while (!blocker.isReleasable() &&
+ !blocker.block());
+ } finally {
+ p.incrementActiveCount();
+ }
+ break;
+ }
+ }
+ }
else {
- do {} while (!blocker.isReleasable() && !blocker.block());
+ do {} while (!blocker.isReleasable() &&
+ !blocker.block());
}
}
@@ -1769,36 +3158,94 @@ public class ForkJoinPool extends Abstra
// implement RunnableFuture.
protected RunnableFuture newTaskFor(Runnable runnable, T value) {
- return (RunnableFuture) ForkJoinTask.adapt(runnable, value);
+ return new ForkJoinTask.AdaptedRunnable(runnable, value);
}
protected RunnableFuture newTaskFor(Callable callable) {
- return (RunnableFuture) ForkJoinTask.adapt(callable);
+ return new ForkJoinTask.AdaptedCallable(callable);
}
// 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 PARKBLOCKER;
+ private static final int ABASE;
+ private static final int ASHIFT;
+ private static final long STEALCOUNT;
+ private static final long PLOCK;
+ private static final long INDEXSEED;
+ private static final long QLOCK;
+
+ static {
+ // Establish common pool parameters
+ // TBD: limit or report ignored exceptions?
+
+ int par = 0;
+ ForkJoinWorkerThreadFactory fac = null;
+ Thread.UncaughtExceptionHandler handler = null;
try {
- return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
- } catch (NoSuchFieldException e) {
- // Convert Exception to corresponding Error
- NoSuchFieldError error = new NoSuchFieldError(field);
- error.initCause(e);
- throw error;
+ String pp = System.getProperty(propPrefix + "parallelism");
+ String hp = System.getProperty(propPrefix + "exceptionHandler");
+ String fp = System.getProperty(propPrefix + "threadFactory");
+ if (fp != null)
+ fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
+ getSystemClassLoader().loadClass(fp).newInstance());
+ if (hp != null)
+ handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
+ getSystemClassLoader().loadClass(hp).newInstance());
+ if (pp != null)
+ par = Integer.parseInt(pp);
+ } catch (Exception ignore) {
}
+
+ int s; // initialize field offsets for CAS etc
+ try {
+ U = getUnsafe();
+ Class> k = ForkJoinPool.class;
+ CTL = U.objectFieldOffset
+ (k.getDeclaredField("ctl"));
+ STEALCOUNT = U.objectFieldOffset
+ (k.getDeclaredField("stealCount"));
+ PLOCK = U.objectFieldOffset
+ (k.getDeclaredField("plock"));
+ INDEXSEED = U.objectFieldOffset
+ (k.getDeclaredField("indexSeed"));
+ Class> tk = Thread.class;
+ PARKBLOCKER = U.objectFieldOffset
+ (tk.getDeclaredField("parkBlocker"));
+ Class> wk = WorkQueue.class;
+ QLOCK = U.objectFieldOffset
+ (wk.getDeclaredField("qlock"));
+ Class> ak = ForkJoinTask[].class;
+ ABASE = U.arrayBaseOffset(ak);
+ s = U.arrayIndexScale(ak);
+ ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
+ } catch (Exception e) {
+ throw new Error(e);
+ }
+ if ((s & (s-1)) != 0)
+ throw new Error("data type scale not a power of two");
+
+ /*
+ * For extra caution, computations to set up pool state are
+ * here; the constructor just assigns these values to fields.
+ */
+ ForkJoinWorkerThreadFactory defaultFac =
+ defaultForkJoinWorkerThreadFactory =
+ new DefaultForkJoinWorkerThreadFactory();
+ if (fac == null)
+ fac = defaultFac;
+ if (par <= 0)
+ par = Runtime.getRuntime().availableProcessors();
+ if (par > MAX_CAP)
+ par = MAX_CAP;
+ long np = (long)(-par); // precompute initial ctl value
+ long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
+
+ commonPoolParallelism = par;
+ commonPool = new ForkJoinPool(par, ct, fac, handler);
+ modifyThreadPermission = new RuntimePermission("modifyThread");
+ submitters = new ThreadLocal();
}
/**
@@ -1828,4 +3275,5 @@ public class ForkJoinPool extends Abstra
}
}
}
+
}