Implementation notes: This implementation restricts the + *
Implementation notes: This implementation restricts the
* maximum number of running threads to 32767. Attempts to create
* pools with greater than the maximum number result in
* {@code IllegalArgumentException}.
@@ -230,75 +233,92 @@
* in a circular buffer:
* q.array[q.top++ % length] = task;
*
- * (The actual code needs to null-check and size-check the array,
+ * The actual code needs to null-check and size-check the array,
* uses masking, not mod, for indexing a power-of-two-sized array,
- * adds a release fence for publication, and possibly signals
- * waiting workers to start scanning -- see below.) Both a
- * successful pop and poll mainly entail a CAS of a slot from
- * non-null to null.
- *
- * The pop operation (always performed by owner) is:
- * if ((the task at top slot is not null) and
- * (CAS slot to null))
- * decrement top and return task;
- *
- * And the poll operation (usually by a stealer) is
- * if ((the task at base slot is not null) and
- * (CAS slot to null))
- * increment base and return task;
- *
- * There are several variants of each of these. Most uses occur
- * within operations that also interleave contention or emptiness
- * tracking or inspection of elements before extracting them, so
- * must interleave these with the above code. When performed by
- * owner, getAndSet is used instead of CAS (see for example method
- * nextLocalTask) which is usually more efficient, and possible
- * because the top index cannot independently change during the
- * operation.
+ * enforces memory ordering, supports resizing, and possibly
+ * signals waiting workers to start scanning -- see below.
+ *
+ * The pop operation (always performed by owner) is of the form:
+ * if ((task = getAndSet(q.array, (q.top-1) % length, null)) != null)
+ * decrement top and return task;
+ * If this fails, the queue is empty.
+ *
+ * The poll operation by another stealer thread is, basically:
+ * if (CAS nonnull task at q.array[q.base % length] to null)
+ * increment base and return task;
+ *
+ * This may fail due to contention, and may be retried.
+ * Implementations must ensure a consistent snapshot of the base
+ * index and the task (by looping or trying elsewhere) before
+ * trying CAS. There isn't actually a method of this form,
+ * because failure due to inconsistency or contention is handled
+ * in different ways in different contexts, normally by first
+ * trying other queues. (For the most straightforward example, see
+ * method pollScan.) There are further variants for cases
+ * requiring inspection of elements before extracting them, so
+ * must interleave these with variants of this code. Also, a more
+ * efficient version (nextLocalTask) is used for polls by owners.
+ * It avoids some overhead because the queue cannot be growing
+ * during call.
*
* Memory ordering. See "Correct and Efficient Work-Stealing for
* Weak Memory Models" by Le, Pop, Cohen, and Nardelli, PPoPP 2013
* (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an
* analysis of memory ordering requirements in work-stealing
- * algorithms similar to (but different than) the one used here.
- * Extracting tasks in array slots via (fully fenced) CAS provides
- * primary synchronization. The base and top indices imprecisely
- * guide where to extract from. We do not usually require strict
- * orderings of array and index updates. Many index accesses use
- * plain mode, with ordering constrained by surrounding context
- * (usually with respect to element CASes or the two WorkQueue
- * volatile fields source and phase). When not otherwise already
- * constrained, reads of "base" by queue owners use acquire-mode,
- * and some externally callable methods preface accesses with
- * acquire fences. Additionally, to ensure that index update
- * writes are not coalesced or postponed in loops etc, "opaque"
- * mode is used in a few cases where timely writes are not
- * otherwise ensured. The "locked" versions of push- and pop-
- * based methods for shared queues differ from owned versions
- * because locking already forces some of the ordering.
+ * algorithms similar to the one used here. Inserting and
+ * extracting tasks in array slots via volatile or atomic accesses
+ * or explicit fences provides primary synchronization.
+ *
+ * Operations on deque elements require reads and writes of both
+ * indices and slots. When possible, we allow these to occur in
+ * any order. Because the base and top indices (along with other
+ * pool or array fields accessed in many methods) only imprecisely
+ * guide where to extract from, we let accesses other than the
+ * element getAndSet/CAS/setVolatile appear in any order, using
+ * plain mode. But we must still preface some methods (mainly
+ * those that may be accessed externally) with an acquireFence to
+ * avoid unbounded staleness. This is equivalent to acting as if
+ * callers use an acquiring read of the reference to the pool or
+ * queue when invoking the method, even when they do not. We use
+ * explicit acquiring reads (getSlot) rather than plain array
+ * access when acquire mode is required but not otherwise ensured
+ * by context. To reduce stalls by other stealers, we encourage
+ * timely writes to the base index by immediately following
+ * updates with a write of a volatile field that must be updated
+ * anyway, or an Opaque-mode write if there is no such
+ * opportunity.
*
* Because indices and slot contents cannot always be consistent,
- * a check that base == top indicates (momentary) emptiness, but
- * otherwise may err on the side of possibly making the queue
- * appear nonempty when a push, pop, or poll have not fully
- * committed, or making it appear empty when an update of top has
- * not yet been visibly written. (Method isEmpty() checks the
- * case of a partially completed removal of the last element.)
- * Because of this, 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)
- * visibly completes. This can stall threads when required to
- * consume from a given queue (see method poll()). However, in
- * the aggregate, we ensure at least probabilistic
- * non-blockingness. If an attempted steal fails, a scanning
- * thief 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.
+ * the emptiness check base == top is only quiescently accurate
+ * (and so used where this suffices). Otherwise, it may err on the
+ * side of possibly making the queue appear nonempty when a push,
+ * pop, or poll have not fully committed, or making it appear
+ * empty when an update of top or base has not yet been seen.
+ * Similarly, the check in push for the queue array being full may
+ * trigger when not completely full, causing a resize earlier than
+ * required.
+ *
+ * Mainly because of these potential inconsistencies among slots
+ * vs indices, 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) visibly
+ * completes. This can stall threads when required to consume
+ * from a given queue (which may spin). However, in the
+ * aggregate, we ensure probabilistic non-blockingness at least
+ * until checking quiescence (which is intrinsically blocking):
+ * If an attempted steal fails, a scanning thief chooses a
+ * different 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. The worst cases occur when many
+ * threads are looking for tasks being produced by a stalled
+ * producer.
*
* 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.
+ * message-passing frameworks in which tasks are never joined,
+ * although with increased contention among task producers and
+ * consumers.
*
* WorkQueues are also used in a similar way for tasks submitted
* to the pool. We cannot mix these tasks in the same queues used
@@ -308,13 +328,12 @@
* choosing existing queues, and 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. Insertion of tasks in shared mode
- * requires a lock but we use only a simple spinlock (using field
- * phase), because submitters encountering a busy queue move to a
- * different position to use or create other queues -- they block
- * only when creating and registering new queues. Because it is
- * used only as a spinlock, unlocking requires only a "releasing"
- * store (using setRelease) unless otherwise signalling.
+ * tasks that they submitted (or when known, subtasks thereof).
+ * Insertion of tasks in shared mode requires a lock. We use only
+ * a simple spinlock (using field "source"), because submitters
+ * encountering a busy queue move to a different position to use
+ * or create other queues. They block only when registering new
+ * queues.
*
* Management
* ==========
@@ -322,16 +341,17 @@
* The main throughput advantages of work-stealing stem from
* decentralized control -- workers mostly take tasks from
* themselves or each other, at rates that can exceed a billion
- * per second. The pool itself creates, activates (enables
- * scanning for and running tasks), deactivates, blocks, and
- * terminates threads, all with minimal central information.
- * There are only a few properties that we can globally track or
- * maintain, so we pack them into a small number of variables,
- * often maintaining atomicity without blocking or locking.
- * Nearly all essentially atomic control state is held in a few
- * volatile variables that are by far most often read (not
- * written) as status and consistency checks. We pack as much
- * information into them as we can.
+ * per second. Most non-atomic control is performed by some form
+ * of scanning across or within queues. The pool itself creates,
+ * activates (enables scanning for and running tasks),
+ * deactivates, blocks, and terminates threads, all with minimal
+ * central information. There are only a few properties that we
+ * can globally track or maintain, so we pack them into a small
+ * number of variables, often maintaining atomicity without
+ * blocking or locking. Nearly all essentially atomic control
+ * state is held in a few volatile variables that are by far most
+ * often read (not written) as status and consistency checks. We
+ * pack as much information into them as we can.
*
* Field "ctl" contains 64 bits holding information needed to
* atomically decide to add, enqueue (on an event queue), and
@@ -343,30 +363,28 @@
*
* Field "mode" holds configuration parameters as well as lifetime
* status, atomically and monotonically setting SHUTDOWN, STOP,
- * and finally TERMINATED bits.
+ * and finally TERMINATED bits. It is updated only via bitwise
+ * atomics (getAndBitwiseOr).
*
- * Field "workQueues" holds references to WorkQueues. It is
- * updated (only during worker creation and termination) under
- * lock (using field workerNamePrefix as lock), but is otherwise
- * concurrently readable, and accessed directly. We also ensure
- * that uses of the array reference itself never become too stale
- * in case of resizing, by arranging that (re-)reads are separated
- * by at least one acquiring read access. 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 the array needs
- * to expand to add more workers. Grouping them together in this
- * way simplifies and speeds up task scanning.
+ * Array "queues" holds references to WorkQueues. It is updated
+ * (only during worker creation and termination) under the
+ * registrationLock, but is otherwise concurrently readable, and
+ * accessed directly (although always prefaced by acquireFences or
+ * other acquiring reads). 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. Worker
+ * ids masked with SMASK match their index. Shared (submission)
+ * queues are at even indices. 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
+ * do not hold on to worker or task references that would prevent
+ * GC, all accesses to workQueues are via indices into the
+ * queues array (which is one source of some of the messy code
+ * constructions here). In essence, the queues array serves as
* a weak reference mechanism. Thus for example the stack top
* subfield of ctl stores indices, not references.
*
@@ -375,31 +393,26 @@
* 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
+ * tasks are submitted or generated. These latencies are mainly a
+ * function of JVM park/unpark (and underlying OS) performance,
+ * which can be slow and variable. In many usages, ramp-up time
* is the main limiting factor in overall performance, which is
* compounded at program start-up by JIT compilation and
- * allocation. So we streamline this as much as possible.
+ * allocation. On the other hand, throughput degrades when too
+ * many threads poll for too few tasks.
*
- * The "ctl" field atomically maintains total worker and
- * "released" worker counts, plus the head of the available worker
- * queue (actually stack, represented by the lower 32bit subfield
- * of ctl). Released workers are those known to be scanning for
+ * The "ctl" field atomically maintains total and "released"
+ * worker counts, plus the head of the available worker queue
+ * (actually stack, represented by the lower 32bit subfield of
+ * ctl). Released workers are those known to be scanning for
* and/or running tasks. Unreleased ("available") workers are
* recorded in the ctl stack. These workers are made available for
- * signalling by enqueuing in ctl (see method runWorker). The
+ * signalling by enqueuing in ctl (see method awaitWork). The
* "queue" is a form of Treiber stack. This is ideal for
* activating threads in most-recently used order, and improves
* performance and locality, outweighing the disadvantages of
* being prone to contention and inability to release a worker
- * unless it is topmost on stack. To avoid missed signal problems
- * inherent in any wait/signal design, available workers rescan
- * for (and if found run) tasks after enqueuing. Normally their
- * release status will be updated while doing so, but the released
- * worker ctl count may underestimate the number of active
- * threads. (However, it is still possible to determine quiescence
- * via a validation traversal -- see isQuiescent). After an
- * unsuccessful rescan, available workers are blocked until
- * signalled (see signalWork). The top stack state holds the
+ * unless it is topmost on stack. The top stack state holds the
* value of the "phase" field of the worker: its index and status,
* plus a version counter that, in addition to the count subfields
* (also serving as version stamps) provide protection against
@@ -407,109 +420,115 @@
*
* Creating workers. To create a worker, we pre-increment counts
* (serving as a reservation), and attempt to construct a
- * ForkJoinWorkerThread via its factory. Upon construction, the
- * new thread invokes registerWorker, where it constructs a
- * WorkQueue and is assigned an index in the workQueues array
- * (expanding the array if necessary). The thread is then started.
- * Upon any exception across these steps, or null return from
- * factory, deregisterWorker adjusts counts and records
- * accordingly. If a null return, the pool continues running with
- * fewer than the target number workers. If exceptional, the
- * exception is propagated, generally to some external caller.
- * Worker index assignment avoids the bias in scanning that would
- * occur if entries were sequentially packed starting at the front
- * of the workQueues array. We treat the array as a simple
- * power-of-two hash table, expanding as needed. The seedIndex
- * increment ensures no collisions until a resize is needed or a
- * worker is deregistered and replaced, and thereafter keeps
- * probability of collision low. We cannot use
- * ThreadLocalRandom.getProbe() for similar purposes here because
- * the thread has not started yet, but do so for creating
- * submission queues for existing external threads (see
- * externalPush).
+ * ForkJoinWorkerThread via its factory. On starting, the new
+ * thread first invokes registerWorker, where it constructs a
+ * WorkQueue and is assigned an index in the queues array
+ * (expanding the array if necessary). Upon any exception across
+ * these steps, or null return from factory, deregisterWorker
+ * adjusts counts and records accordingly. If a null return, the
+ * pool continues running with fewer than the target number
+ * workers. If exceptional, the exception is propagated, generally
+ * to some external caller.
*
* WorkQueue field "phase" is used by both workers and the pool to
* manage and track whether a worker is UNSIGNALLED (possibly
* blocked waiting for a signal). When a worker is enqueued its
- * phase field is set. Note that phase field updates lag queue CAS
- * releases so usage requires care -- seeing a negative phase does
- * not guarantee that the worker is available. When queued, the
- * lower 16 bits of scanState must hold its pool index. So we
- * place the index there upon initialization and otherwise keep it
- * there or restore it when necessary.
+ * phase field is set negative. Note that phase field updates lag
+ * queue CAS releases; seeing a negative phase does not guarantee
+ * that the worker is available. When queued, the lower 16 bits of
+ * its phase must hold its pool index. So we place the index there
+ * upon initialization and never modify these bits.
*
* The ctl field also serves as the basis for memory
* synchronization surrounding activation. This uses a more
* efficient version of a Dekker-like rule that task producers and
* consumers sync with each other by both writing/CASing ctl (even
- * if to its current value). This would be extremely costly. So
- * we relax it in several ways: (1) Producers only signal when
- * their queue is possibly empty at some point during a push
- * operation. (2) Other workers propagate this signal
- * when they find tasks in a queue with size greater than one. (3)
- * Workers only enqueue after scanning (see below) and not finding
- * any tasks. (4) Rather than CASing ctl to its current value in
- * the common case where no action is required, we reduce write
- * contention by equivalently prefacing signalWork when called by
- * an external task producer using a memory access with
- * full-volatile semantics or a "fullFence".
- *
- * Almost always, too many signals are issued, in part because a
- * task producer cannot tell if some existing worker is in the
- * midst of finishing one task (or already scanning) and ready to
- * take another without being signalled. So the producer might
- * instead activate a different worker that does not find any
- * work, and then inactivates. This scarcely matters in
- * steady-state computations involving all workers, but can create
- * contention and bookkeeping bottlenecks during ramp-up,
+ * if to its current value). However, rather than CASing ctl to
+ * its current value in the common case where no action is
+ * required, we reduce write contention by ensuring that
+ * signalWork invocations are prefaced with a full-volatile memory
+ * access (which is usually needed anyway).
+ *
+ * Signalling. Signals (in signalWork) cause new or reactivated
+ * workers to scan for tasks. Method signalWork and its callers
+ * try to approximate the unattainable goal of having the right
+ * number of workers activated for the tasks at hand, but must err
+ * on the side of too many workers vs too few to avoid stalls. If
+ * computations are purely tree structured, it suffices for every
+ * worker to activate another when it pushes a task into an empty
+ * queue, resulting in O(log(#threads)) steps to full activation.
+ * If instead, tasks come in serially from only a single producer,
+ * each worker taking its first (since the last quiescence) task
+ * from a queue should signal another if there are more tasks in
+ * that queue. This is equivalent to, but generally faster than,
+ * arranging the stealer take two tasks, re-pushing one on its own
+ * queue, and signalling (because its queue is empty), also
+ * resulting in logarithmic full activation time. Because we don't
+ * know about usage patterns (or most commonly, mixtures), we use
+ * both approaches. We approximate the second rule by arranging
+ * that workers in scan() do not repeat signals when repeatedly
+ * taking tasks from any given queue, by remembering the previous
+ * one. There are narrow windows in which both rules may apply,
+ * leading to duplicate or unnecessary signals. Despite such
+ * limitations, these rules usually avoid slowdowns that otherwise
+ * occur when too many workers contend to take too few tasks, or
+ * when producers waste most of their time resignalling. However,
+ * contention and overhead effects may still occur during ramp-up,
* ramp-down, and small computations involving only a few workers.
*
- * Scanning. Method scan (from runWorker) performs top-level
- * scanning for tasks. (Similar scans appear in helpQuiesce and
- * pollScan.) Each scan traverses and tries to poll from each
- * queue starting at a random index. Scans are not performed in
- * ideal random permutation order, to reduce cacheline
- * contention. The pseudorandom generator need not have
- * high-quality statistical properties in the long term, but just
- * within computations; We use Marsaglia XorShifts (often via
- * ThreadLocalRandom.nextSecondarySeed), which are cheap and
- * suffice. Scanning also includes contention reduction: When
- * scanning workers fail to extract an apparently existing task,
- * they soon restart at a different pseudorandom index. This form
- * of backoff improves throughput when many threads are trying to
- * take tasks from few queues, which can be common in some usages.
- * Scans do not otherwise explicitly take into account core
- * affinities, loads, cache localities, etc, However, they do
+ * Scanning. Method scan performs top-level scanning for (and
+ * execution of) tasks. Scans by different workers and/or at
+ * different times are unlikely to poll queues in the same
+ * order. Each scan traverses and tries to poll from each queue in
+ * a pseudorandom permutation order by starting at a random index,
+ * and using a constant cyclically exhaustive stride; restarting
+ * upon contention. (Non-top-level scans; for example in
+ * helpJoin, use simpler linear probes because they do not
+ * systematically contend with top-level scans.) The pseudorandom
+ * generator need not have high-quality statistical properties in
+ * the long term. We use Marsaglia XorShifts, seeded with the Weyl
+ * sequence from ThreadLocalRandom probes, which are cheap and
+ * suffice. Scans do not otherwise explicitly take into account
+ * core affinities, loads, cache localities, etc, However, they do
* exploit temporal locality (which usually approximates these) by
* preferring to re-poll from the same queue after a successful
- * poll before trying others (see method topLevelExec). However
- * this preference is bounded (see TOP_BOUND_SHIFT) as a safeguard
- * against infinitely unfair looping under unbounded user task
- * recursion, and also to reduce long-term contention when many
- * threads poll few queues holding many small tasks. The bound is
- * high enough to avoid much impact on locality and scheduling
- * overhead.
+ * poll before trying others (see method topLevelExec). This
+ * reduces fairness, which is partially counteracted by using a
+ * one-shot form of poll (tryPoll) that may lose to other workers.
+ *
+ * Deactivation. Method scan returns a sentinel when no tasks are
+ * found, leading to deactivation (see awaitWork). The count
+ * fields in ctl allow accurate discovery of quiescent states
+ * (i.e., when all workers are idle) after deactivation. However,
+ * this may also race with new (external) submissions, so a
+ * recheck is also needed to determine quiescence. Upon apparently
+ * triggering quiescence, awaitWork re-scans and self-signals if
+ * it may have missed a signal. In other cases, a missed signal
+ * may transiently lower parallelism because deactivation does not
+ * necessarily mean that there is no more work, only that that
+ * there were no tasks not taken by other workers. But more
+ * signals are generated (see above) to eventually reactivate if
+ * needed.
*
* 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 (see method runWorker) if the pool has
- * remained quiescent for period given by field keepAlive.
+ * time out and terminate if the pool has remained quiescent for
+ * period given by field keepAlive.
*
* Shutdown and Termination. A call to shutdownNow invokes
- * tryTerminate to atomically set a runState bit. The calling
- * thread, as well as every other worker thereafter terminating,
- * helps terminate others by cancelling their unprocessed tasks,
- * and waking them up, doing so repeatedly until stable. Calls to
- * non-abrupt shutdown() preface this by checking whether
- * termination should commence by sweeping through queues (until
- * stable) to ensure lack of in-flight submissions and workers
- * about to process them before triggering the "STOP" phase of
+ * tryTerminate to atomically set a mode bit. The calling thread,
+ * as well as every other worker thereafter terminating, helps
+ * terminate others by cancelling their unprocessed tasks, and
+ * waking them up. Calls to non-abrupt shutdown() preface this by
+ * checking isQuiescent before triggering the "STOP" phase of
* termination.
*
* Joining Tasks
* =============
*
- * Any of several actions may be taken when one worker is waiting
+ * Normally, the first option when joining a task that is not done
+ * is to try to unfork it from local queue and run it. Otherwise,
+ * 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
* always just let them block (as in Thread.join). We also cannot
@@ -520,40 +539,62 @@
* 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.
+ * could 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
+ * A third form (implemented via tryRemove) 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.
+ * without the need for a compensation thread; although with a
+ * (rare) possibility of reduced parallelism because of a
+ * transient gap in the queue array.
+ *
+ * Other intermediate forms available for specific task types (for
+ * example helpAsyncBlocker) often avoid or postpone the need for
+ * blocking or compensation.
*
* The ManagedBlocker extension API can't use helping so relies
* only on compensation in method awaitBlocker.
*
- * The algorithm in awaitJoin entails a form of "linear helping".
- * Each worker records (in field source) the id of the queue from
- * which it last stole a task. The scan in method awaitJoin 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. Thus, the joiner executes a task that
- * would be on its own local deque if the to-be-joined task had
- * not been stolen. This is a conservative variant of the approach
- * described in Wagner & Calder "Leapfrogging: a portable
+ * The algorithm in helpJoin entails a form of "linear helping".
+ * Each worker records (in field "source") the id of the queue
+ * from which it last stole a task. The scan in method helpJoin
+ * 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. Thus, the joiner executes a task
+ * that would be on its own local deque if the to-be-joined task
+ * had not been stolen. This is a conservative variant of the
+ * approach described in Wagner & Calder "Leapfrogging: a portable
* technique for implementing efficient futures" SIGPLAN Notices,
* 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs
* mainly in that we only record queue ids, not full dependency
- * links. This requires a linear scan of the workQueues array to
+ * links. This requires a linear scan of the queues array to
* locate stealers, but isolates cost to when it is needed, rather
- * than adding to per-task overhead. Searches can fail to locate
- * stealers GC stalls and the like delay recording sources.
- * Further, even when accurately identified, stealers might not
- * ever produce a task that the joiner can in turn help with. So,
- * compensation is tried upon failure to find tasks to run.
+ * than adding to per-task overhead. Also, searches are limited to
+ * direct and at most two levels of indirect stealers, after which
+ * there are rapidly diminishing returns on increased overhead.
+ * Searches can fail to locate stealers when stalls delay
+ * recording sources. Further, even when accurately identified,
+ * stealers might not ever produce a task that the joiner can in
+ * turn help with. So, compensation is tried upon failure to find
+ * tasks to run.
+ *
+ * Joining CountedCompleters (see helpComplete) differs from (and
+ * is generally more efficient than) other cases because task
+ * eligibility is determined by checking completion chains rather
+ * than tracking stealers.
+ *
+ * Joining under timeouts (ForkJoinTask timed get) uses a
+ * constrained mixture of helping and compensating in part because
+ * pools (actually, only the common pool) may not have any
+ * available threads: If the pool is saturated (all available
+ * workers are busy), the caller tries to remove and otherwise
+ * help; else it blocks under compensation so that it may time out
+ * independently of any tasks.
*
* Compensation does not by default aim to keep exactly the target
* parallelism number of unblocked threads running at any given
@@ -578,8 +619,8 @@
* footprint to the setup of about a dozen fields.
*
* When external threads submit to the common pool, they can
- * perform subtask processing (see externalHelpComplete and
- * related methods) upon joins. This caller-helps policy makes it
+ * perform subtask processing (see helpComplete and related
+ * methods) upon joins. This caller-helps policy makes it
* sensible to set common pool parallelism level to one (or more)
* less than the total number of available cores, or even zero for
* pure caller-runs. We do not need to record whether external
@@ -595,39 +636,61 @@
* InnocuousForkJoinWorkerThread when there is a SecurityManager
* present. These workers have no permissions set, do not belong
* to any user-defined ThreadGroup, and erase all ThreadLocals
- * after executing any top-level task (see
- * WorkQueue.afterTopLevelExec). The associated mechanics (mainly
- * in ForkJoinWorkerThread) may be JVM-dependent and must access
- * particular Thread class fields to achieve this effect.
+ * after executing any top-level task. The associated mechanics
+ * may be JVM-dependent and must access particular Thread class
+ * fields to achieve this effect.
+ *
+ * Interrupt handling
+ * ==================
+ *
+ * The framework is designed to manage task cancellation
+ * (ForkJoinTask.cancel) independently from the interrupt status
+ * of threads running tasks. (See the public ForkJoinTask
+ * documentation for rationale.) Interrupts are issued only in
+ * tryTerminate, when workers should be terminating and tasks
+ * should be cancelled anyway. Interrupts are cleared only when
+ * necessary to ensure that calls to LockSupport.park do not loop
+ * indefinitely (park returns immediately if the current thread is
+ * interrupted). If so, interruption is reinstated after blocking
+ * if status could be visible during the scope of any task. For
+ * cases in which task bodies are specified or desired to
+ * interrupt upon cancellation, ForkJoinTask.cancel can be
+ * overridden to do so (as is done for invoke{Any,All}).
*
* Memory placement
* ================
*
* Performance can be very sensitive to placement of instances of
* ForkJoinPool and WorkQueues and their queue arrays. To reduce
- * false-sharing impact, the @Contended annotation isolates
- * adjacent WorkQueue instances, as well as the ForkJoinPool.ctl
- * field. WorkQueue arrays are allocated (by their threads) with
- * larger initial sizes than most ever need, mostly to reduce
- * false sharing with current garbage collectors that use cardmark
- * tables.
+ * false-sharing impact, the @Contended annotation isolates the
+ * ForkJoinPool.ctl field as well as the most heavily written
+ * WorkQueue fields. These mainly reduce cache traffic by scanners.
+ * WorkQueue arrays are presized large enough to avoid resizing
+ * (which transiently reduces throughput) in most tree-like
+ * computations, although not in some streaming usages. Initial
+ * sizes are not large enough to avoid secondary contention
+ * effects (especially for GC cardmarks) when queues are placed
+ * near each other in memory. This is common, but has different
+ * impact in different collectors and remains incompletely
+ * addressed.
*
* Style notes
* ===========
*
- * Memory ordering relies mainly on VarHandles. This can be
+ * Memory ordering relies mainly on atomic operations (CAS,
+ * getAndSet, getAndAdd) along with explicit fences. This can be
* awkward and ugly, but also reflects the need to control
* outcomes across the unusual cases that arise in very racy code
* with very few invariants. All fields are read into locals
- * before use, and null-checked if they are references. Array
- * accesses using masked indices include checks (that are always
- * true) that the array length is non-zero to avoid compilers
- * inserting more expensive traps. This is usually done in a
- * "C"-like style of listing declarations at the heads of methods
- * or blocks, and using inline assignments on first encounter.
- * Nearly all explicit checks lead to bypass/return, not exception
- * throws, because they may legitimately arise due to
- * cancellation/revocation during shutdown.
+ * before use, and null-checked if they are references, even if
+ * they can never be null under current usages. Array accesses
+ * using masked indices include checks (that are always true) that
+ * the array length is non-zero to avoid compilers inserting more
+ * expensive traps. This is usually done in a "C"-like style of
+ * listing declarations at the heads of methods or blocks, and
+ * using inline assignments on first encounter. Nearly all
+ * explicit checks lead to bypass/return, not exception throws,
+ * because they may legitimately arise during shutdown.
*
* There is a lot of representation-level coupling among classes
* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
@@ -652,6 +715,22 @@
* (6) Callbacks and other support for ForkJoinTask methods
* (7) Exported methods
* (8) Static block initializing statics in minimally dependent order
+ *
+ * Revision notes
+ * ==============
+ *
+ * The main sources of differences of January 2020 ForkJoin
+ * classes from previous version are:
+ *
+ * * ForkJoinTask now uses field "aux" to support blocking joins
+ * and/or record exceptions, replacing reliance on builtin
+ * monitors and side tables.
+ * * Scans probe slots (vs compare indices), along with related
+ * changes that reduce performance differences across most
+ * garbage collectors, and reduce contention.
+ * * Refactoring for better integration of special task types and
+ * other capabilities that had been incrementally tacked on. Plus
+ * many minor reworkings to improve consistency.
*/
// Static utilities
@@ -666,6 +745,14 @@
security.checkPermission(modifyThreadPermission);
}
+ static AccessControlContext contextWithPermissions(Permission ... perms) {
+ Permissions permissions = new Permissions();
+ for (Permission perm : perms)
+ permissions.add(perm);
+ return new AccessControlContext(
+ new ProtectionDomain[] { new ProtectionDomain(null, permissions) });
+ }
+
// Nested classes
/**
@@ -693,379 +780,389 @@
public ForkJoinWorkerThread newThread(ForkJoinPool pool);
}
- static AccessControlContext contextWithPermissions(Permission ... perms) {
- Permissions permissions = new Permissions();
- for (Permission perm : perms)
- permissions.add(perm);
- return new AccessControlContext(
- new ProtectionDomain[] { new ProtectionDomain(null, permissions) });
- }
-
/**
* Default ForkJoinWorkerThreadFactory implementation; creates a
* new ForkJoinWorkerThread using the system class loader as the
* thread context class loader.
*/
- private static final class DefaultForkJoinWorkerThreadFactory
+ static final class DefaultForkJoinWorkerThreadFactory
implements ForkJoinWorkerThreadFactory {
+ // ACC for access to the factory
private static final AccessControlContext ACC = contextWithPermissions(
new RuntimePermission("getClassLoader"),
new RuntimePermission("setContextClassLoader"));
-
public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
return AccessController.doPrivileged(
new PrivilegedAction<>() {
public ForkJoinWorkerThread run() {
- return new ForkJoinWorkerThread(
- pool, ClassLoader.getSystemClassLoader()); }},
+ return new ForkJoinWorkerThread(null, pool, true, false);
+ }},
ACC);
}
}
+ /**
+ * Factory for CommonPool unless overridden by System property.
+ * Creates InnocuousForkJoinWorkerThreads if a security manager is
+ * present at time of invocation. Support requires that we break
+ * quite a lot of encapsulation (some via helper methods in
+ * ThreadLocalRandom) to access and set Thread fields.
+ */
+ static final class DefaultCommonPoolForkJoinWorkerThreadFactory
+ implements ForkJoinWorkerThreadFactory {
+ private static final AccessControlContext ACC = contextWithPermissions(
+ modifyThreadPermission,
+ new RuntimePermission("enableContextClassLoaderOverride"),
+ new RuntimePermission("modifyThreadGroup"),
+ new RuntimePermission("getClassLoader"),
+ new RuntimePermission("setContextClassLoader"));
+
+ public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
+ return AccessController.doPrivileged(
+ new PrivilegedAction<>() {
+ public ForkJoinWorkerThread run() {
+ return System.getSecurityManager() == null ?
+ new ForkJoinWorkerThread(null, pool, true, true):
+ new ForkJoinWorkerThread.
+ InnocuousForkJoinWorkerThread(pool); }},
+ ACC);
+ }
+ }
+
// Constants shared across ForkJoinPool and WorkQueue
// Bounds
static final int SWIDTH = 16; // width of short
static final int SMASK = 0xffff; // short bits == max index
static final int MAX_CAP = 0x7fff; // max #workers - 1
- static final int SQMASK = 0x007e; // max 64 (even) slots
// Masks and units for WorkQueue.phase and ctl sp subfield
static final int UNSIGNALLED = 1 << 31; // must be negative
static final int SS_SEQ = 1 << 16; // version count
- static final int QLOCK = 1; // must be 1
- // Mode bits and sentinels, some also used in WorkQueue id and.source fields
- static final int OWNED = 1; // queue has owner thread
+ // Mode bits and sentinels, some also used in WorkQueue fields
static final int FIFO = 1 << 16; // fifo queue or access mode
- static final int SHUTDOWN = 1 << 18;
- static final int TERMINATED = 1 << 19;
+ static final int SRC = 1 << 17; // set for valid queue ids
+ static final int INNOCUOUS = 1 << 18; // set for Innocuous workers
+ static final int QUIET = 1 << 19; // quiescing phase or source
+ static final int SHUTDOWN = 1 << 24;
+ static final int TERMINATED = 1 << 25;
static final int STOP = 1 << 31; // must be negative
- static final int QUIET = 1 << 30; // not scanning or working
- static final int DORMANT = QUIET | UNSIGNALLED;
-
- /**
- * Initial capacity of work-stealing queue array.
- * Must be a power of two, at least 2.
- */
- static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
+ static final int UNCOMPENSATE = 1 << 16; // tryCompensate return
/**
- * Maximum capacity 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.
+ * Initial capacity of work-stealing queue array. Must be a power
+ * of two, at least 2. See above.
*/
- static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
-
- /**
- * The maximum number of top-level polls per worker before
- * checking other queues, expressed as a bit shift. See above for
- * rationale.
- */
- static final int TOP_BOUND_SHIFT = 10;
+ static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
/**
* Queues supporting work-stealing as well as external task
* submission. See above for descriptions and algorithms.
*/
- @jdk.internal.vm.annotation.Contended
static final class WorkQueue {
- volatile int source; // source queue id, or sentinel
- int id; // pool index, mode, tag
- int base; // index of next slot for poll
- int top; // index of next slot for push
- volatile int phase; // versioned, negative: queued, 1: locked
+ volatile int phase; // versioned, negative if inactive
int stackPred; // pool stack (ctl) predecessor link
- int nsteals; // number of steals
+ int config; // index, mode, ORed with SRC after init
+ int base; // index of next slot for poll
ForkJoinTask[] array; // the queued tasks; power of 2 size
- final ForkJoinPool pool; // the containing pool (may be null)
final ForkJoinWorkerThread owner; // owning thread or null if shared
- WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner) {
- this.pool = pool;
- this.owner = owner;
- // Place indices in the center of array (that is not yet allocated)
- base = top = INITIAL_QUEUE_CAPACITY >>> 1;
+ // segregate fields frequently updated but not read by scans or steals
+ @jdk.internal.vm.annotation.Contended("w")
+ int top; // index of next slot for push
+ @jdk.internal.vm.annotation.Contended("w")
+ volatile int source; // source queue id, lock, or sentinel
+ @jdk.internal.vm.annotation.Contended("w")
+ int nsteals; // number of steals from other queues
+
+ // Support for atomic operations
+ private static final VarHandle QA; // for array slots
+ private static final VarHandle SOURCE;
+ private static final VarHandle BASE;
+ static final ForkJoinTask getSlot(ForkJoinTask[] a, int i) {
+ return (ForkJoinTask)QA.getAcquire(a, i);
+ }
+ static final ForkJoinTask getAndClearSlot(ForkJoinTask[] a,
+ int i) {
+ return (ForkJoinTask)QA.getAndSet(a, i, null);
+ }
+ static final void setSlotVolatile(ForkJoinTask[] a, int i,
+ ForkJoinTask v) {
+ QA.setVolatile(a, i, v);
+ }
+ static final boolean casSlotToNull(ForkJoinTask[] a, int i,
+ ForkJoinTask c) {
+ return QA.weakCompareAndSet(a, i, c, null);
+ }
+ final boolean tryLock() {
+ return SOURCE.compareAndSet(this, 0, 1);
+ }
+ final void setBaseOpaque(int b) {
+ BASE.setOpaque(this, b);
}
/**
- * Tries to lock shared queue by CASing phase field.
+ * Constructor used by ForkJoinWorkerThreads. Most fields
+ * are initialized upon thread start, in pool.registerWorker.
*/
- final boolean tryLockPhase() {
- return PHASE.compareAndSet(this, 0, 1);
+ WorkQueue(ForkJoinWorkerThread owner, boolean isInnocuous) {
+ this.config = (isInnocuous) ? INNOCUOUS : 0;
+ this.owner = owner;
}
- final void releasePhaseLock() {
- PHASE.setRelease(this, 0);
+ /**
+ * Constructor used for external queues.
+ */
+ WorkQueue(int config) {
+ array = new ForkJoinTask[INITIAL_QUEUE_CAPACITY];
+ this.config = config;
+ owner = null;
+ phase = -1;
}
/**
* Returns an exportable index (used by ForkJoinWorkerThread).
*/
final int getPoolIndex() {
- return (id & 0xffff) >>> 1; // ignore odd/even tag bit
+ return (config & 0xffff) >>> 1; // ignore odd/even tag bit
}
/**
* Returns the approximate number of tasks in the queue.
*/
final int queueSize() {
- int n = (int)BASE.getAcquire(this) - top;
- return (n >= 0) ? 0 : -n; // ignore transient negative
+ VarHandle.acquireFence(); // ensure fresh reads by external callers
+ int n = top - base;
+ return (n < 0) ? 0 : n; // ignore transient negative
}
/**
- * Provides a more accurate estimate of whether this queue has
- * any tasks than does queueSize, by checking whether a
- * near-empty queue has at least one unclaimed task.
+ * Provides a more conservative estimate of whether this queue
+ * has any tasks than does queueSize.
*/
final boolean isEmpty() {
- ForkJoinTask[] a; int n, cap, b;
- VarHandle.acquireFence(); // needed by external callers
- return ((n = (b = base) - top) >= 0 || // possibly one task
- (n == -1 && ((a = array) == null ||
- (cap = a.length) == 0 ||
- a[(cap - 1) & b] == null)));
+ return !((source != 0 && owner == null) || top - base > 0);
}
/**
* Pushes a task. Call only by owner in unshared queues.
*
* @param task the task. Caller must ensure non-null.
+ * @param pool (no-op if null)
* @throws RejectedExecutionException if array cannot be resized
*/
- final void push(ForkJoinTask task) {
- ForkJoinTask[] a;
- int s = top, d = s - base, cap, m;
- ForkJoinPool p = pool;
- if ((a = array) != null && (cap = a.length) > 0) {
- QA.setRelease(a, (m = cap - 1) & s, task);
- top = s + 1;
+ final void push(ForkJoinTask task, ForkJoinPool pool) {
+ ForkJoinTask[] a = array;
+ int s = top++, d = s - base, cap, m; // skip insert if disabled
+ if (a != null && pool != null && (cap = a.length) > 0) {
+ setSlotVolatile(a, (m = cap - 1) & s, task);
if (d == m)
- growArray(false);
- else if (QA.getAcquire(a, m & (s - 1)) == null && p != null) {
- VarHandle.fullFence(); // was empty
- p.signalWork(null);
- }
+ growArray();
+ if (d == m || a[m & (s - 1)] == null)
+ pool.signalWork(); // signal if was empty or resized
}
}
/**
- * Version of push for shared queues. Call only with phase lock held.
- * @return true if should signal work
+ * Pushes task to a shared queue with lock already held, and unlocks.
+ *
+ * @return true if caller should signal work
*/
final boolean lockedPush(ForkJoinTask task) {
- ForkJoinTask[] a;
- boolean signal = false;
- int s = top, d = s - base, cap, m;
- if ((a = array) != null && (cap = a.length) > 0) {
- a[(m = (cap - 1)) & s] = task;
- top = s + 1;
+ ForkJoinTask[] a = array;
+ int s = top++, d = s - base, cap, m;
+ if (a != null && (cap = a.length) > 0) {
+ a[(m = cap - 1) & s] = task;
if (d == m)
- growArray(true);
- else {
- phase = 0; // full volatile unlock
- if (((s - base) & ~1) == 0) // size 0 or 1
- signal = true;
- }
+ growArray();
+ source = 0; // unlock
+ if (d == m || a[m & (s - 1)] == null)
+ return true;
}
- return signal;
+ return false;
}
/**
- * Doubles the capacity of array. Call either by owner or with
- * lock held -- it is OK for base, but not top, to move while
- * resizings are in progress.
- */
- final void growArray(boolean locked) {
- ForkJoinTask[] newA = null;
- try {
- ForkJoinTask[] oldA; int oldSize, newSize;
- if ((oldA = array) != null && (oldSize = oldA.length) > 0 &&
- (newSize = oldSize << 1) <= MAXIMUM_QUEUE_CAPACITY &&
- newSize > 0) {
- try {
- newA = new ForkJoinTask[newSize];
- } catch (OutOfMemoryError ex) {
- }
- if (newA != null) { // poll from old array, push to new
- int oldMask = oldSize - 1, newMask = newSize - 1;
- for (int s = top - 1, k = oldMask; k >= 0; --k) {
- ForkJoinTask x = (ForkJoinTask)
- QA.getAndSet(oldA, s & oldMask, null);
- if (x != null)
- newA[s-- & newMask] = x;
- else
- break;
- }
- array = newA;
- VarHandle.releaseFence();
- }
+ * Doubles the capacity of array. Called by owner or with lock
+ * held after pre-incrementing top, which is reverted on
+ * allocation failure.
+ */
+ final void growArray() {
+ ForkJoinTask[] oldArray = array, newArray;
+ int s = top - 1, oldCap, newCap;
+ if (oldArray != null && (oldCap = oldArray.length) > 0 &&
+ (newCap = oldCap << 1) > 0) { // skip if disabled
+ try {
+ newArray = new ForkJoinTask[newCap];
+ } catch (Throwable ex) {
+ top = s;
+ if (owner == null)
+ source = 0; // unlock
+ throw new RejectedExecutionException(
+ "Queue capacity exceeded");
}
- } finally {
- if (locked)
- phase = 0;
- }
- if (newA == null)
- throw new RejectedExecutionException("Queue capacity exceeded");
- }
-
- /**
- * Takes next task, if one exists, in FIFO order.
- */
- final ForkJoinTask poll() {
- int b, k, cap; ForkJoinTask[] a;
- while ((a = array) != null && (cap = a.length) > 0 &&
- top - (b = base) > 0) {
- ForkJoinTask t = (ForkJoinTask)
- QA.getAcquire(a, k = (cap - 1) & b);
- if (base == b++) {
- if (t == null)
- Thread.yield(); // await index advance
- else if (QA.compareAndSet(a, k, t, null)) {
- BASE.setOpaque(this, b);
- return t;
- }
+ int newMask = newCap - 1, oldMask = oldCap - 1;
+ for (int k = oldCap; k > 0; --k, --s) {
+ ForkJoinTask x; // poll old, push to new
+ if ((x = getAndClearSlot(oldArray, s & oldMask)) == null)
+ break; // others already taken
+ newArray[s & newMask] = x;
}
+ VarHandle.releaseFence(); // fill before publish
+ array = newArray;
}
- return null;
}
+ // Variants of pop
+
/**
- * Takes next task, if one exists, in order specified by mode.
+ * Pops and returns task, or null if empty. Called only by owner.
*/
- final ForkJoinTask nextLocalTask() {
+ private ForkJoinTask pop() {
ForkJoinTask t = null;
- int md = id, b, s, d, cap; ForkJoinTask[] a;
- if ((a = array) != null && (cap = a.length) > 0 &&
- (d = (s = top) - (b = base)) > 0) {
- if ((md & FIFO) == 0 || d == 1) {
- if ((t = (ForkJoinTask)
- QA.getAndSet(a, (cap - 1) & --s, null)) != null)
- TOP.setOpaque(this, s);
- }
- else if ((t = (ForkJoinTask)
- QA.getAndSet(a, (cap - 1) & b++, null)) != null) {
- BASE.setOpaque(this, b);
- }
- else // on contention in FIFO mode, use regular poll
- t = poll();
- }
+ int s = top, cap; ForkJoinTask[] a;
+ if ((a = array) != null && (cap = a.length) > 0 && base != s-- &&
+ (t = getAndClearSlot(a, (cap - 1) & s)) != null)
+ top = s;
return t;
}
/**
- * Returns next task, if one exists, in order specified by mode.
+ * Pops the given task for owner only if it is at the current top.
*/
- final ForkJoinTask peek() {
- int cap; ForkJoinTask[] a;
- return ((a = array) != null && (cap = a.length) > 0) ?
- a[(cap - 1) & ((id & FIFO) != 0 ? base : top - 1)] : null;
+ final boolean tryUnpush(ForkJoinTask task) {
+ int s = top, cap; ForkJoinTask[] a;
+ if ((a = array) != null && (cap = a.length) > 0 && base != s-- &&
+ casSlotToNull(a, (cap - 1) & s, task)) {
+ top = s;
+ return true;
+ }
+ return false;
}
/**
- * Pops the given task only if it is at the current top.
+ * Locking version of tryUnpush.
*/
- final boolean tryUnpush(ForkJoinTask task) {
- boolean popped = false;
- int s, cap; ForkJoinTask[] a;
+ final boolean externalTryUnpush(ForkJoinTask task) {
+ boolean taken = false;
+ int s = top, cap, k; ForkJoinTask[] a;
if ((a = array) != null && (cap = a.length) > 0 &&
- (s = top) != base &&
- (popped = QA.compareAndSet(a, (cap - 1) & --s, task, null)))
- TOP.setOpaque(this, s);
- return popped;
+ a[k = (cap - 1) & (s - 1)] == task && tryLock()) {
+ if (top == s && array == a &&
+ (taken = casSlotToNull(a, k, task)))
+ top = s - 1;
+ source = 0; // release lock
+ }
+ return taken;
}
/**
- * Shared version of tryUnpush.
- */
- final boolean tryLockedUnpush(ForkJoinTask task) {
- boolean popped = false;
- int s = top - 1, k, cap; ForkJoinTask[] a;
- if ((a = array) != null && (cap = a.length) > 0 &&
- a[k = (cap - 1) & s] == task && tryLockPhase()) {
- if (top == s + 1 && array == a &&
- (popped = QA.compareAndSet(a, k, task, null)))
- top = s;
- releasePhaseLock();
+ * Deep form of tryUnpush: Traverses from top and removes task if
+ * present, shifting others to fill gap.
+ */
+ final boolean tryRemove(ForkJoinTask task, boolean owned) {
+ boolean taken = false;
+ int p = top, cap; ForkJoinTask[] a; ForkJoinTask t;
+ if ((a = array) != null && task != null && (cap = a.length) > 0) {
+ int m = cap - 1, s = p - 1, d = p - base;
+ for (int i = s, k; d > 0; --i, --d) {
+ if ((t = a[k = i & m]) == task) {
+ if (owned || tryLock()) {
+ if ((owned || (array == a && top == p)) &&
+ (taken = casSlotToNull(a, k, t))) {
+ for (int j = i; j != s; ) // shift down
+ a[j & m] = getAndClearSlot(a, ++j & m);
+ top = s;
+ }
+ if (!owned)
+ source = 0;
+ }
+ break;
+ }
+ }
}
- return popped;
+ return taken;
}
+ // variants of poll
+
/**
- * Removes and cancels all known tasks, ignoring any exceptions.
+ * Tries once to poll next task in FIFO order, failing on
+ * inconsistency or contention.
*/
- final void cancelAll() {
- for (ForkJoinTask t; (t = poll()) != null; )
- ForkJoinTask.cancelIgnoringExceptions(t);
+ final ForkJoinTask tryPoll() {
+ int cap, b, k; ForkJoinTask[] a;
+ if ((a = array) != null && (cap = a.length) > 0) {
+ ForkJoinTask t = getSlot(a, k = (cap - 1) & (b = base));
+ if (base == b++ && t != null && casSlotToNull(a, k, t)) {
+ setBaseOpaque(b);
+ return t;
+ }
+ }
+ return null;
}
- // Specialized execution methods
-
/**
- * Runs the given (stolen) task if nonnull, as well as
- * remaining local tasks and others available from the given
- * queue, up to bound n (to avoid infinite unfairness).
+ * Takes next task, if one exists, in order specified by mode.
*/
- final void topLevelExec(ForkJoinTask t, WorkQueue q, int n) {
- int nstolen = 1;
- for (int j = 0;;) {
- if (t != null)
- t.doExec();
- if (j++ <= n)
- t = nextLocalTask();
- else {
- j = 0;
- t = null;
- }
- if (t == null) {
- if (q != null && (t = q.poll()) != null) {
- ++nstolen;
- j = 0;
+ final ForkJoinTask nextLocalTask(int cfg) {
+ ForkJoinTask t = null;
+ int s = top, cap; ForkJoinTask[] a;
+ if ((a = array) != null && (cap = a.length) > 0) {
+ for (int b, d;;) {
+ if ((d = s - (b = base)) <= 0)
+ break;
+ if (d == 1 || (cfg & FIFO) == 0) {
+ if ((t = getAndClearSlot(a, --s & (cap - 1))) != null)
+ top = s;
+ break;
}
- else if (j != 0)
+ if ((t = getAndClearSlot(a, b++ & (cap - 1))) != null) {
+ setBaseOpaque(b);
break;
+ }
}
}
- ForkJoinWorkerThread thread = owner;
- nsteals += nstolen;
- source = 0;
- if (thread != null)
- thread.afterTopLevelExec();
+ return t;
}
/**
- * If present, removes task from queue and executes it.
+ * Takes next task, if one exists, using configured mode.
*/
- final void tryRemoveAndExec(ForkJoinTask task) {
- ForkJoinTask[] a; int s, cap;
- if ((a = array) != null && (cap = a.length) > 0 &&
- (s = top) - base > 0) { // traverse from top
- for (int m = cap - 1, ns = s - 1, i = ns; ; --i) {
- int index = i & m;
- ForkJoinTask t = (ForkJoinTask)QA.get(a, index);
- if (t == null)
- break;
- else if (t == task) {
- if (QA.compareAndSet(a, index, t, null)) {
- top = ns; // safely shift down
- for (int j = i; j != ns; ++j) {
- ForkJoinTask f;
- int pindex = (j + 1) & m;
- f = (ForkJoinTask)QA.get(a, pindex);
- QA.setVolatile(a, pindex, null);
- int jindex = j & m;
- QA.setRelease(a, jindex, f);
- }
- VarHandle.releaseFence();
- t.doExec();
- }
- break;
- }
- }
+ final ForkJoinTask nextLocalTask() {
+ return nextLocalTask(config);
+ }
+
+ /**
+ * Returns next task, if one exists, in order specified by mode.
+ */
+ final ForkJoinTask peek() {
+ VarHandle.acquireFence();
+ int cap; ForkJoinTask[] a;
+ return ((a = array) != null && (cap = a.length) > 0) ?
+ a[(cap - 1) & ((config & FIFO) != 0 ? base : top - 1)] : null;
+ }
+
+ // specialized execution methods
+
+ /**
+ * Runs the given (stolen) task if nonnull, as well as
+ * remaining local tasks and/or others available from the
+ * given queue.
+ */
+ final void topLevelExec(ForkJoinTask task, WorkQueue q) {
+ int cfg = config, nstolen = 1;
+ while (task != null) {
+ task.doExec();
+ if ((task = nextLocalTask(cfg)) == null &&
+ q != null && (task = q.tryPoll()) != null)
+ ++nstolen;
}
+ nsteals += nstolen;
+ source = 0;
+ if ((cfg & INNOCUOUS) != 0)
+ ThreadLocalRandom.eraseThreadLocals(Thread.currentThread());
}
/**
@@ -1073,84 +1170,85 @@
* until done, not found, or limit exceeded.
*
* @param task root of CountedCompleter computation
+ * @param owned true if owned by a ForkJoinWorkerThread
* @param limit max runs, or zero for no limit
- * @param shared true if must lock to extract task
* @return task status on exit
*/
- final int helpCC(CountedCompleter task, int limit, boolean shared) {
- int status = 0;
- if (task != null && (status = task.status) >= 0) {
- int s, k, cap; ForkJoinTask[] a;
- while ((a = array) != null && (cap = a.length) > 0 &&
- (s = top) - base > 0) {
- CountedCompleter v = null;
- ForkJoinTask o = a[k = (cap - 1) & (s - 1)];
- if (o instanceof CountedCompleter) {
- CountedCompleter t = (CountedCompleter)o;
- for (CountedCompleter f = t;;) {
- if (f != task) {
- if ((f = f.completer) == null)
- break;
- }
- else if (shared) {
- if (tryLockPhase()) {
- if (top == s && array == a &&
- QA.compareAndSet(a, k, t, null)) {
- top = s - 1;
- v = t;
- }
- releasePhaseLock();
- }
- break;
- }
- else {
- if (QA.compareAndSet(a, k, t, null)) {
- top = s - 1;
- v = t;
- }
- break;
- }
+ final int helpComplete(ForkJoinTask task, boolean owned, int limit) {
+ int status = 0, cap, k, p, s; ForkJoinTask[] a; ForkJoinTask t;
+ while (task != null && (status = task.status) >= 0 &&
+ (a = array) != null && (cap = a.length) > 0 &&
+ (t = a[k = (cap - 1) & (s = (p = top) - 1)])
+ instanceof CountedCompleter) {
+ CountedCompleter f = (CountedCompleter)t;
+ boolean taken = false;
+ for (;;) { // exec if root task is a completer of t
+ if (f == task) {
+ if (owned) {
+ if ((taken = casSlotToNull(a, k, t)))
+ top = s;
}
+ else if (tryLock()) {
+ if (top == p && array == a &&
+ (taken = casSlotToNull(a, k, t)))
+ top = s;
+ source = 0;
+ }
+ break;
}
- if (v != null)
- v.doExec();
- if ((status = task.status) < 0 || v == null ||
- (limit != 0 && --limit == 0))
+ else if ((f = f.completer) == null)
break;
}
+ if (!taken)
+ break;
+ t.doExec();
+ if (limit != 0 && --limit == 0)
+ break;
}
return status;
}
/**
* Tries to poll and run AsynchronousCompletionTasks until
- * none found or blocker is released
+ * none found or blocker is released.
*
* @param blocker the blocker
*/
final void helpAsyncBlocker(ManagedBlocker blocker) {
- if (blocker != null) {
- int b, k, cap; ForkJoinTask[] a; ForkJoinTask t;
- while ((a = array) != null && (cap = a.length) > 0 &&
- top - (b = base) > 0) {
- t = (ForkJoinTask)QA.getAcquire(a, k = (cap - 1) & b);
- if (blocker.isReleasable())
- break;
- else if (base == b++ && t != null) {
- if (!(t instanceof CompletableFuture.
- AsynchronousCompletionTask))
- break;
- else if (QA.compareAndSet(a, k, t, null)) {
- BASE.setOpaque(this, b);
- t.doExec();
- }
- }
+ int cap, b, d, k; ForkJoinTask[] a; ForkJoinTask t;
+ while (blocker != null && (d = top - (b = base)) > 0 &&
+ (a = array) != null && (cap = a.length) > 0 &&
+ (((t = getSlot(a, k = (cap - 1) & b)) == null && d > 1) ||
+ t instanceof
+ CompletableFuture.AsynchronousCompletionTask) &&
+ !blocker.isReleasable()) {
+ if (t != null && base == b++ && casSlotToNull(a, k, t)) {
+ setBaseOpaque(b);
+ t.doExec();
}
}
}
+ // misc
+
+ /** AccessControlContext for innocuous workers, created on 1st use. */
+ private static AccessControlContext INNOCUOUS_ACC;
+
+ /**
+ * Initializes (upon registration) InnocuousForkJoinWorkerThreads.
+ */
+ final void initializeInnocuousWorker() {
+ AccessControlContext acc; // racy construction OK
+ if ((acc = INNOCUOUS_ACC) == null)
+ INNOCUOUS_ACC = acc = new AccessControlContext(
+ new ProtectionDomain[] { new ProtectionDomain(null, null) });
+ Thread t = Thread.currentThread();
+ ThreadLocalRandom.setInheritedAccessControlContext(t, acc);
+ ThreadLocalRandom.eraseThreadLocals(t);
+ }
+
/**
- * Returns true if owned and not known to be blocked.
+ * Returns true if owned by a worker thread and not known to be blocked.
*/
final boolean isApparentlyUnblocked() {
Thread wt; Thread.State s;
@@ -1160,16 +1258,12 @@
s != Thread.State.TIMED_WAITING);
}
- // VarHandle mechanics.
- static final VarHandle PHASE;
- static final VarHandle BASE;
- static final VarHandle TOP;
static {
try {
+ QA = MethodHandles.arrayElementVarHandle(ForkJoinTask[].class);
MethodHandles.Lookup l = MethodHandles.lookup();
- PHASE = l.findVarHandle(WorkQueue.class, "phase", int.class);
+ SOURCE = l.findVarHandle(WorkQueue.class, "source", int.class);
BASE = l.findVarHandle(WorkQueue.class, "base", int.class);
- TOP = l.findVarHandle(WorkQueue.class, "top", int.class);
} catch (ReflectiveOperationException e) {
throw new ExceptionInInitializerError(e);
}
@@ -1213,17 +1307,9 @@
private static final int COMMON_MAX_SPARES;
/**
- * Sequence number for creating workerNamePrefix.
+ * Sequence number for creating worker names
*/
- private static int poolNumberSequence;
-
- /**
- * Returns the next sequence number. We don't expect this to
- * ever contend, so use simple builtin sync.
- */
- private static final synchronized int nextPoolId() {
- return ++poolNumberSequence;
- }
+ private static volatile int poolIds;
// static configuration constants
@@ -1248,12 +1334,6 @@
*/
private static final int DEFAULT_COMMON_MAX_SPARES = 256;
- /**
- * Increment for seed generators. See class ThreadLocal for
- * explanation.
- */
- private static final int SEED_INCREMENT = 0x9e3779b9;
-
/*
* Bits and masks for field ctl, packed with 4 16 bit subfields:
* RC: Number of released (unqueued) workers minus target parallelism
@@ -1271,10 +1351,10 @@
* deal with possibly negative fields, we use casts in and out of
* "short" and/or signed shifts to maintain signedness.
*
- * Because it occupies uppermost bits, we can add one release count
- * using getAndAddLong of RC_UNIT, rather than CAS, when returning
- * from a blocked join. Other updates entail multiple subfields
- * and masking, requiring CAS.
+ * Because it occupies uppermost bits, we can add one release
+ * count using getAndAdd of RC_UNIT, rather than CAS, when
+ * returning from a blocked join. Other updates entail multiple
+ * subfields and masking, requiring CAS.
*
* The limits packed in field "bounds" are also offset by the
* parallelism level to make them comparable to the ctl rc and tc
@@ -1298,13 +1378,16 @@
// Instance fields
- volatile long stealCount; // collects worker nsteals
final long keepAlive; // milliseconds before dropping if idle
- int indexSeed; // next worker index
+ volatile long stealCount; // collects worker nsteals
+ int scanRover; // advances across pollScan calls
+ volatile int threadIds; // for worker thread names
final int bounds; // min, max threads packed as shorts
volatile int mode; // parallelism, runstate, queue mode
- WorkQueue[] workQueues; // main registry
- final String workerNamePrefix; // for worker thread string; sync lock
+ WorkQueue[] queues; // main registry
+ final ReentrantLock registrationLock;
+ Condition termination; // lazily constructed
+ final String workerNamePrefix; // null for common pool
final ForkJoinWorkerThreadFactory factory;
final UncaughtExceptionHandler ueh; // per-worker UEH
final Predicate saturate;
@@ -1312,6 +1395,30 @@
@jdk.internal.vm.annotation.Contended("fjpctl") // segregate
volatile long ctl; // main pool control
+ // Support for atomic operations
+ private static final VarHandle CTL;
+ private static final VarHandle MODE;
+ private static final VarHandle THREADIDS;
+ private static final VarHandle POOLIDS;
+ private boolean compareAndSetCtl(long c, long v) {
+ return CTL.compareAndSet(this, c, v);
+ }
+ private long compareAndExchangeCtl(long c, long v) {
+ return (long)CTL.compareAndExchange(this, c, v);
+ }
+ private long getAndAddCtl(long v) {
+ return (long)CTL.getAndAdd(this, v);
+ }
+ private int getAndBitwiseOrMode(int v) {
+ return (int)MODE.getAndBitwiseOr(this, v);
+ }
+ private int getAndAddThreadIds(int x) {
+ return (int)THREADIDS.getAndAdd(this, x);
+ }
+ private static int getAndAddPoolIds(int x) {
+ return (int)POOLIDS.getAndAdd(x);
+ }
+
// Creating, registering and deregistering workers
/**
@@ -1338,83 +1445,63 @@
}
/**
- * Tries to add one worker, incrementing ctl counts before doing
- * so, relying on createWorker to back out on failure.
- *
- * @param c incoming ctl value, with total count negative and no
- * idle workers. On CAS failure, c is refreshed and retried if
- * this holds (otherwise, a new worker is not needed).
+ * Provides a name for ForkJoinWorkerThread constructor.
*/
- private void tryAddWorker(long c) {
- do {
- long nc = ((RC_MASK & (c + RC_UNIT)) |
- (TC_MASK & (c + TC_UNIT)));
- if (ctl == c && CTL.compareAndSet(this, c, nc)) {
- createWorker();
- break;
- }
- } while (((c = ctl) & ADD_WORKER) != 0L && (int)c == 0);
+ final String nextWorkerThreadName() {
+ String prefix = workerNamePrefix;
+ int tid = getAndAddThreadIds(1) + 1;
+ if (prefix == null) // commonPool has no prefix
+ prefix = "ForkJoinPool.commonPool-worker-";
+ return prefix.concat(Integer.toString(tid));
}
/**
- * Callback from ForkJoinWorkerThread constructor to establish and
- * record its WorkQueue.
- *
- * @param wt the worker thread
- * @return the worker's queue
- */
- final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
- UncaughtExceptionHandler handler;
- wt.setDaemon(true); // configure thread
- if ((handler = ueh) != null)
- wt.setUncaughtExceptionHandler(handler);
- int tid = 0; // for thread name
- int idbits = mode & FIFO;
- String prefix = workerNamePrefix;
- WorkQueue w = new WorkQueue(this, wt);
- if (prefix != null) {
- synchronized (prefix) {
- WorkQueue[] ws = workQueues; int n;
- int s = indexSeed += SEED_INCREMENT;
- idbits |= (s & ~(SMASK | FIFO | DORMANT));
- if (ws != null && (n = ws.length) > 1) {
- int m = n - 1;
- tid = m & ((s << 1) | 1); // odd-numbered indices
- for (int probes = n >>> 1;;) { // find empty slot
- WorkQueue q;
- if ((q = ws[tid]) == null || q.phase == QUIET)
- break;
- else if (--probes == 0) {
- tid = n | 1; // resize below
- break;
- }
- else
- tid = (tid + 2) & m;
- }
- w.phase = w.id = tid | idbits; // now publishable
+ * Finishes initializing and records owned queue.
+ *
+ * @param w caller's WorkQueue
+ */
+ final void registerWorker(WorkQueue w) {
+ ReentrantLock lock = registrationLock;
+ ThreadLocalRandom.localInit();
+ int seed = ThreadLocalRandom.getProbe();
+ if (w != null && lock != null) {
+ int modebits = (mode & FIFO) | w.config;
+ w.array = new ForkJoinTask[INITIAL_QUEUE_CAPACITY];
+ w.stackPred = seed; // stash for runWorker
+ if ((modebits & INNOCUOUS) != 0)
+ w.initializeInnocuousWorker();
+ int id = (seed << 1) | 1; // initial index guess
+ lock.lock();
+ try {
+ WorkQueue[] qs; int n; // find queue index
+ if ((qs = queues) != null && (n = qs.length) > 0) {
+ int k = n, m = n - 1;
+ for (; qs[id &= m] != null && k > 0; id -= 2, k -= 2);
+ if (k == 0)
+ id = n | 1; // resize below
+ w.phase = w.config = id | modebits; // now publishable
- if (tid < n)
- ws[tid] = w;
+ if (id < n)
+ qs[id] = w;
else { // expand array
- int an = n << 1;
+ int an = n << 1, am = an - 1;
WorkQueue[] as = new WorkQueue[an];
- as[tid] = w;
- int am = an - 1;
- for (int j = 0; j < n; ++j) {
- WorkQueue v; // copy external queue
- if ((v = ws[j]) != null) // position may change
- as[v.id & am & SQMASK] = v;
- if (++j >= n)
- break;
- as[j] = ws[j]; // copy worker
+ as[id & am] = w;
+ for (int j = 1; j < n; j += 2)
+ as[j] = qs[j];
+ for (int j = 0; j < n; j += 2) {
+ WorkQueue q;
+ if ((q = qs[j]) != null) // shared queues may move
+ as[q.config & am] = q;
}
- workQueues = as;
+ VarHandle.releaseFence(); // fill before publish
+ queues = as;
}
}
+ } finally {
+ lock.unlock();
}
- wt.setName(prefix.concat(Integer.toString(tid)));
}
- return w;
}
/**
@@ -1427,319 +1514,431 @@
* @param ex the exception causing failure, or null if none
*/
final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
+ ReentrantLock lock = registrationLock;
WorkQueue w = null;
- int phase = 0;
- if (wt != null && (w = wt.workQueue) != null) {
- Object lock = workerNamePrefix;
- int wid = w.id;
- long ns = (long)w.nsteals & 0xffffffffL;
- if (lock != null) {
- synchronized (lock) {
- WorkQueue[] ws; int n, i; // remove index from array
- if ((ws = workQueues) != null && (n = ws.length) > 0 &&
- ws[i = wid & (n - 1)] == w)
- ws[i] = null;
- stealCount += ns;
- }
- }
- phase = w.phase;
- }
- if (phase != QUIET) { // else pre-adjusted
- long c; // decrement counts
- do {} while (!CTL.weakCompareAndSet
- (this, c = ctl, ((RC_MASK & (c - RC_UNIT)) |
- (TC_MASK & (c - TC_UNIT)) |
- (SP_MASK & c))));
+ int cfg = 0;
+ if (wt != null && (w = wt.workQueue) != null && lock != null) {
+ WorkQueue[] qs; int n, i;
+ cfg = w.config;
+ long ns = w.nsteals & 0xffffffffL;
+ lock.lock(); // remove index from array
+ if ((qs = queues) != null && (n = qs.length) > 0 &&
+ qs[i = cfg & (n - 1)] == w)
+ qs[i] = null;
+ stealCount += ns; // accumulate steals
+ lock.unlock();
+ long c = ctl;
+ if ((cfg & QUIET) == 0) // unless self-signalled, decrement counts
+ do {} while (c != (c = compareAndExchangeCtl(
+ c, ((RC_MASK & (c - RC_UNIT)) |
+ (TC_MASK & (c - TC_UNIT)) |
+ (SP_MASK & c)))));
+ else if ((int)c == 0) // was dropped on timeout
+ cfg = 0; // suppress signal if last
+ for (ForkJoinTask t; (t = w.pop()) != null; )
+ ForkJoinTask.cancelIgnoringExceptions(t); // cancel tasks
}
- if (w != null)
- w.cancelAll(); // cancel remaining tasks
- if (!tryTerminate(false, false) && // possibly replace worker
- w != null && w.array != null) // avoid repeated failures
- signalWork(null);
-
- if (ex == null) // help clean on way out
- ForkJoinTask.helpExpungeStaleExceptions();
- else // rethrow
+ if (!tryTerminate(false, false) && w != null && (cfg & SRC) != 0)
+ signalWork(); // possibly replace worker
+ if (ex != null)
ForkJoinTask.rethrow(ex);
}
- /**
+ /*
* Tries to create or release a worker if too few are running.
- * @param q if non-null recheck if empty on CAS failure
*/
- final void signalWork(WorkQueue q) {
- for (;;) {
- long c; int sp; WorkQueue[] ws; int i; WorkQueue v;
- if ((c = ctl) >= 0L) // enough workers
- break;
- else if ((sp = (int)c) == 0) { // no idle workers
- if ((c & ADD_WORKER) != 0L) // too few workers
- tryAddWorker(c);
- break;
+ final void signalWork() {
+ for (long c = ctl; c < 0L;) {
+ int sp, i; WorkQueue[] qs; WorkQueue v;
+ if ((sp = (int)c & ~UNSIGNALLED) == 0) { // no idle workers
+ if ((c & ADD_WORKER) == 0L) // enough total workers
+ break;
+ if (c == (c = compareAndExchangeCtl(
+ c, ((RC_MASK & (c + RC_UNIT)) |
+ (TC_MASK & (c + TC_UNIT)))))) {
+ createWorker();
+ break;
+ }
}
- else if ((ws = workQueues) == null)
+ else if ((qs = queues) == null)
break; // unstarted/terminated
- else if (ws.length <= (i = sp & SMASK))
+ else if (qs.length <= (i = sp & SMASK))
break; // terminated
- else if ((v = ws[i]) == null)
+ else if ((v = qs[i]) == null)
break; // terminating
else {
- int np = sp & ~UNSIGNALLED;
- int vp = v.phase;
long nc = (v.stackPred & SP_MASK) | (UC_MASK & (c + RC_UNIT));
Thread vt = v.owner;
- if (sp == vp && CTL.compareAndSet(this, c, nc)) {
- v.phase = np;
- if (vt != null && v.source < 0)
- LockSupport.unpark(vt);
+ if (c == (c = compareAndExchangeCtl(c, nc))) {
+ v.phase = sp;
+ LockSupport.unpark(vt); // release idle worker
break;
}
- else if (q != null && q.isEmpty()) // no need to retry
- break;
}
}
}
/**
- * Tries to decrement counts (sometimes implicitly) and possibly
- * arrange for a compensating worker in preparation for blocking:
- * If not all core workers yet exist, creates one, else if any are
- * unreleased (possibly including caller) releases one, else if
- * fewer than the minimum allowed number of workers running,
- * checks to see that they are all active, and if so creates an
- * extra worker unless over maximum limit and policy is to
- * saturate. Most of these steps can fail due to interference, in
- * which case 0 is returned so caller will retry. A negative
- * return value indicates that the caller doesn't need to
- * re-adjust counts when later unblocked.
+ * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
+ * See above for explanation.
*
- * @return 1: block then adjust, -1: block without adjust, 0 : retry
+ * @param w caller's WorkQueue (may be null on failed initialization)
*/
- private int tryCompensate(WorkQueue w) {
- int t, n, sp;
- long c = ctl;
- WorkQueue[] ws = workQueues;
- if ((t = (short)(c >>> TC_SHIFT)) >= 0) {
- if (ws == null || (n = ws.length) <= 0 || w == null)
- return 0; // disabled
- else if ((sp = (int)c) != 0) { // replace or release
- WorkQueue v = ws[sp & (n - 1)];
- int wp = w.phase;
- long uc = UC_MASK & ((wp < 0) ? c + RC_UNIT : c);
- int np = sp & ~UNSIGNALLED;
- if (v != null) {
- int vp = v.phase;
- Thread vt = v.owner;
- long nc = ((long)v.stackPred & SP_MASK) | uc;
- if (vp == sp && CTL.compareAndSet(this, c, nc)) {
- v.phase = np;
- if (vt != null && v.source < 0)
- LockSupport.unpark(vt);
- return (wp < 0) ? -1 : 1;
- }
- }
- return 0;
- }
- else if ((int)(c >> RC_SHIFT) - // reduce parallelism
- (short)(bounds & SMASK) > 0) {
- long nc = ((RC_MASK & (c - RC_UNIT)) | (~RC_MASK & c));
- return CTL.compareAndSet(this, c, nc) ? 1 : 0;
- }
- else { // validate
- int md = mode, pc = md & SMASK, tc = pc + t, bc = 0;
- boolean unstable = false;
- for (int i = 1; i < n; i += 2) {
- WorkQueue q; Thread wt; Thread.State ts;
- if ((q = ws[i]) != null) {
- if (q.source == 0) {
- unstable = true;
- break;
- }
- else {
- --tc;
- if ((wt = q.owner) != null &&
- ((ts = wt.getState()) == Thread.State.BLOCKED ||
- ts == Thread.State.WAITING))
- ++bc; // worker is blocking
- }
- }
+ final void runWorker(WorkQueue w) {
+ if (w != null) { // skip on failed init
+ w.config |= SRC; // mark as valid source
+ int r = w.stackPred, src = 0; // use seed from registerWorker
+ do {
+ r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
+ } while ((src = scan(w, src, r)) >= 0 ||
+ (src = awaitWork(w)) == 0);
+ }
+ }
+
+ /**
+ * Scans for and if found executes top-level tasks: Tries to poll
+ * each queue starting at a random index with random stride,
+ * returning source id or retry indicator if contended or
+ * inconsistent.
+ *
+ * @param w caller's WorkQueue
+ * @param prevSrc the previous queue stolen from in current phase, or 0
+ * @param r random seed
+ * @return id of queue if taken, negative if none found, prevSrc for retry
+ */
+ private int scan(WorkQueue w, int prevSrc, int r) {
+ WorkQueue[] qs = queues;
+ int n = (w == null || qs == null) ? 0 : qs.length;
+ for (int step = (r >>> 16) | 1, i = n; i > 0; --i, r += step) {
+ int j, cap, b; WorkQueue q; ForkJoinTask[] a;
+ if ((q = qs[j = r & (n - 1)]) != null && // poll at qs[j].array[k]
+ (a = q.array) != null && (cap = a.length) > 0) {
+ int k = (cap - 1) & (b = q.base), nextBase = b + 1;
+ int nextIndex = (cap - 1) & nextBase, src = j | SRC;
+ ForkJoinTask t = WorkQueue.getSlot(a, k);
+ if (q.base != b) // inconsistent
+ return prevSrc;
+ else if (t != null && WorkQueue.casSlotToNull(a, k, t)) {
+ q.base = nextBase;
+ ForkJoinTask next = a[nextIndex];
+ if ((w.source = src) != prevSrc && next != null)
+ signalWork(); // propagate
+ w.topLevelExec(t, q);
+ return src;
+ }
+ else if (a[nextIndex] != null) // revisit
+ return prevSrc;
+ }
+ }
+ return (queues != qs) ? prevSrc: -1; // possibly resized
+ }
+
+ /**
+ * Advances worker phase, pushes onto ctl stack, and awaits signal
+ * or reports termination.
+ *
+ * @return negative if terminated, else 0
+ */
+ private int awaitWork(WorkQueue w) {
+ if (w == null)
+ return -1; // already terminated
+ int phase = (w.phase + SS_SEQ) & ~UNSIGNALLED;
+ w.phase = phase | UNSIGNALLED; // advance phase
+ long prevCtl = ctl, c; // enqueue
+ do {
+ w.stackPred = (int)prevCtl;
+ c = ((prevCtl - RC_UNIT) & UC_MASK) | (phase & SP_MASK);
+ } while (prevCtl != (prevCtl = compareAndExchangeCtl(prevCtl, c)));
+
+ Thread.interrupted(); // clear status
+ LockSupport.setCurrentBlocker(this); // prepare to block (exit also OK)
+ long deadline = 0L; // nonzero if possibly quiescent
+ int ac = (int)(c >> RC_SHIFT), md;
+ if ((md = mode) < 0) // pool is terminating
+ return -1;
+ else if ((md & SMASK) + ac <= 0) {
+ boolean checkTermination = (md & SHUTDOWN) != 0;
+ if ((deadline = System.currentTimeMillis() + keepAlive) == 0L)
+ deadline = 1L; // avoid zero
+ WorkQueue[] qs = queues; // check for racing submission
+ int n = (qs == null) ? 0 : qs.length;
+ for (int i = 0; i < n; i += 2) {
+ WorkQueue q; ForkJoinTask[] a; int cap, b;
+ if (ctl != c) { // already signalled
+ checkTermination = false;
+ break;
}
- if (unstable || tc != 0 || ctl != c)
- return 0; // inconsistent
- else if (t + pc >= MAX_CAP || t >= (bounds >>> SWIDTH)) {
- Predicate sat;
- if ((sat = saturate) != null && sat.test(this))
- return -1;
- else if (bc < pc) { // lagging
- Thread.yield(); // for retry spins
- return 0;
- }
- else
- throw new RejectedExecutionException(
- "Thread limit exceeded replacing blocked worker");
+ else if ((q = qs[i]) != null &&
+ (a = q.array) != null && (cap = a.length) > 0 &&
+ ((b = q.base) != q.top || a[(cap - 1) & b] != null ||
+ q.source != 0)) {
+ if (compareAndSetCtl(c, prevCtl))
+ w.phase = phase; // self-signal
+ checkTermination = false;
+ break;
}
}
+ if (checkTermination && tryTerminate(false, false))
+ return -1; // trigger quiescent termination
}
- long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); // expand pool
- return CTL.compareAndSet(this, c, nc) && createWorker() ? 1 : 0;
+ for (boolean alt = false;;) { // await activation or termination
+ if (w.phase >= 0)
+ break;
+ else if (mode < 0)
+ return -1;
+ else if ((c = ctl) == prevCtl)
+ Thread.onSpinWait(); // signal in progress
+ else if (!(alt = !alt)) // check between park calls
+ Thread.interrupted();
+ else if (deadline == 0L)
+ LockSupport.park();
+ else if (deadline - System.currentTimeMillis() > TIMEOUT_SLOP)
+ LockSupport.parkUntil(deadline);
+ else if (((int)c & SMASK) == (w.config & SMASK) &&
+ compareAndSetCtl(c, ((UC_MASK & (c - TC_UNIT)) |
+ (prevCtl & SP_MASK)))) {
+ w.config |= QUIET; // sentinel for deregisterWorker
+ return -1; // drop on timeout
+ }
+ else if ((deadline += keepAlive) == 0L)
+ deadline = 1L; // not at head; restart timer
+ }
+ return 0;
}
+ // Utilities used by ForkJoinTask
+
/**
- * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
- * See above for explanation.
+ * Returns true if all workers are busy, possibly creating one if allowed
*/
- final void runWorker(WorkQueue w) {
- int r = (w.id ^ ThreadLocalRandom.nextSecondarySeed()) | FIFO; // rng
- w.array = new ForkJoinTask[INITIAL_QUEUE_CAPACITY]; // initialize
- for (;;) {
- int phase;
- if (scan(w, r)) { // scan until apparently empty
- r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // move (xorshift)
- }
- else if ((phase = w.phase) >= 0) { // enqueue, then rescan
- long np = (w.phase = (phase + SS_SEQ) | UNSIGNALLED) & SP_MASK;
- long c, nc;
- do {
- w.stackPred = (int)(c = ctl);
- nc = ((c - RC_UNIT) & UC_MASK) | np;
- } while (!CTL.weakCompareAndSet(this, c, nc));
- }
- else { // already queued
- int pred = w.stackPred;
- Thread.interrupted(); // clear before park
- w.source = DORMANT; // enable signal
- long c = ctl;
- int md = mode, rc = (md & SMASK) + (int)(c >> RC_SHIFT);
- if (md < 0) // terminating
- break;
- else if (rc <= 0 && (md & SHUTDOWN) != 0 &&
- tryTerminate(false, false))
- break; // quiescent shutdown
- else if (w.phase < 0) {
- if (rc <= 0 && pred != 0 && phase == (int)c) {
- long nc = (UC_MASK & (c - TC_UNIT)) | (SP_MASK & pred);
- long d = keepAlive + System.currentTimeMillis();
- LockSupport.parkUntil(this, d);
- if (ctl == c && // drop on timeout if all idle
- d - System.currentTimeMillis() <= TIMEOUT_SLOP &&
- CTL.compareAndSet(this, c, nc)) {
- w.phase = QUIET;
- break;
- }
- }
- else {
- LockSupport.park(this);
- if (w.phase < 0) // one spurious wakeup check
- LockSupport.park(this);
+ final boolean isSaturated() {
+ int maxTotal = bounds >>> SWIDTH;
+ for (long c;;) {
+ if (((int)(c = ctl) & ~UNSIGNALLED) != 0)
+ return false;
+ if ((short)(c >>> TC_SHIFT) >= maxTotal)
+ return true;
+ long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
+ if (compareAndSetCtl(c, nc))
+ return !createWorker();
+ }
+ }
+
+ /**
+ * Returns true if can start terminating if enabled, or already terminated
+ */
+ final boolean canStop() {
+ outer: for (long oldSum = 0L;;) { // repeat until stable
+ int md; WorkQueue[] qs; long c;
+ if ((qs = queues) == null || ((md = mode) & STOP) != 0)
+ return true;
+ if ((md & SMASK) + (int)((c = ctl) >> RC_SHIFT) > 0)
+ break;
+ long checkSum = c;
+ for (int i = 1; i < qs.length; i += 2) { // scan submitters
+ WorkQueue q; ForkJoinTask[] a; int s = 0, cap;
+ if ((q = qs[i]) != null && (a = q.array) != null &&
+ (cap = a.length) > 0 &&
+ ((s = q.top) != q.base || a[(cap - 1) & s] != null ||
+ q.source != 0))
+ break outer;
+ checkSum += (((long)i) << 32) ^ s;
+ }
+ if (oldSum == (oldSum = checkSum) && queues == qs)
+ return true;
+ }
+ return (mode & STOP) != 0; // recheck mode on false return
+ }
+
+ /**
+ * Tries to decrement counts (sometimes implicitly) and possibly
+ * arrange for a compensating worker in preparation for
+ * blocking. May fail due to interference, in which case -1 is
+ * returned so caller may retry. A zero return value indicates
+ * that the caller doesn't need to re-adjust counts when later
+ * unblocked.
+ *
+ * @param c incoming ctl value
+ * @return UNCOMPENSATE: block then adjust, 0: block, -1 : retry
+ */
+ private int tryCompensate(long c) {
+ Predicate sat;
+ int b = bounds; // counts are signed; centered at parallelism level == 0
+ int minActive = (short)(b & SMASK),
+ maxTotal = b >>> SWIDTH,
+ active = (int)(c >> RC_SHIFT),
+ total = (short)(c >>> TC_SHIFT),
+ sp = (int)c & ~UNSIGNALLED;
+ if (total >= 0) {
+ if (sp != 0) { // activate idle worker
+ WorkQueue[] qs; int n; WorkQueue v;
+ if ((qs = queues) != null && (n = qs.length) > 0 &&
+ (v = qs[sp & (n - 1)]) != null) {
+ Thread vt = v.owner;
+ long nc = ((long)v.stackPred & SP_MASK) | (UC_MASK & c);
+ if (compareAndSetCtl(c, nc)) {
+ v.phase = sp;
+ LockSupport.unpark(vt);
+ return UNCOMPENSATE;
}
}
- w.source = 0; // disable signal
+ return -1; // retry
+ }
+ else if (active > minActive) { // reduce parallelism
+ long nc = ((RC_MASK & (c - RC_UNIT)) | (~RC_MASK & c));
+ return compareAndSetCtl(c, nc) ? UNCOMPENSATE : -1;
}
}
+ if (total < maxTotal) { // expand pool
+ long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
+ return (!compareAndSetCtl(c, nc) ? -1 :
+ !createWorker() ? 0 : UNCOMPENSATE);
+ }
+ else if (!compareAndSetCtl(c, c)) // validate
+ return -1;
+ else if ((sat = saturate) != null && sat.test(this))
+ return 0;
+ else
+ throw new RejectedExecutionException(
+ "Thread limit exceeded replacing blocked worker");
}
/**
- * Scans for and if found executes one or more top-level tasks from a queue.
+ * Readjusts RC count; called from ForkJoinTask after blocking.
+ */
+ final void uncompensate() {
+ getAndAddCtl(RC_UNIT);
+ }
+
+ /**
+ * Helps if possible until the given task is done. Scans other
+ * queues for a task produced by one of w's stealers; returning
+ * compensated blocking sentinel if none are found.
*
- * @return true if found an apparently non-empty queue, and
- * possibly ran task(s).
- */
- private boolean scan(WorkQueue w, int r) {
- WorkQueue[] ws; int n;
- if ((ws = workQueues) != null && (n = ws.length) > 0 && w != null) {
- for (int m = n - 1, j = r & m;;) {
- WorkQueue q; int b;
- if ((q = ws[j]) != null && q.top != (b = q.base)) {
- int qid = q.id;
- ForkJoinTask[] a; int cap, k; ForkJoinTask t;
- if ((a = q.array) != null && (cap = a.length) > 0) {
- t = (ForkJoinTask)QA.getAcquire(a, k = (cap - 1) & b);
- if (q.base == b++ && t != null &&
- QA.compareAndSet(a, k, t, null)) {
- q.base = b;
- w.source = qid;
- if (a[(cap - 1) & b] != null)
- signalWork(q); // help signal if more tasks
- w.topLevelExec(t, q, // random fairness bound
- (r | (1 << TOP_BOUND_SHIFT)) & SMASK);
+ * @param task the task
+ * @param w caller's WorkQueue
+ * @return task status on exit, or UNCOMPENSATE for compensated blocking
+ */
+ final int helpJoin(ForkJoinTask task, WorkQueue w) {
+ int s = 0;
+ if (task != null && w != null) {
+ int wsrc = w.source, wid = w.config & SMASK, r = wid + 2;
+ boolean scan = true;
+ long c = 0L; // track ctl stability
+ outer: for (;;) {
+ if ((s = task.status) < 0)
+ break;
+ else if (scan = !scan) { // previous scan was empty
+ if (mode < 0)
+ ForkJoinTask.cancelIgnoringExceptions(task);
+ else if (c == (c = ctl) && (s = tryCompensate(c)) >= 0)
+ break; // block
+ }
+ else { // scan for subtasks
+ WorkQueue[] qs = queues;
+ int n = (qs == null) ? 0 : qs.length, m = n - 1;
+ for (int i = n; i > 0; i -= 2, r += 2) {
+ int j; WorkQueue q, x, y; ForkJoinTask[] a;
+ if ((q = qs[j = r & m]) != null) {
+ int sq = q.source & SMASK, cap, b;
+ if ((a = q.array) != null && (cap = a.length) > 0) {
+ int k = (cap - 1) & (b = q.base);
+ int nextBase = b + 1, src = j | SRC, sx;
+ ForkJoinTask t = WorkQueue.getSlot(a, k);
+ boolean eligible = sq == wid ||
+ ((x = qs[sq & m]) != null && // indirect
+ ((sx = (x.source & SMASK)) == wid ||
+ ((y = qs[sx & m]) != null && // 2-indirect
+ (y.source & SMASK) == wid)));
+ if ((s = task.status) < 0)
+ break outer;
+ else if ((q.source & SMASK) != sq ||
+ q.base != b)
+ scan = true; // inconsistent
+ else if (t == null)
+ scan |= (a[nextBase & (cap - 1)] != null ||
+ q.top != b); // lagging
+ else if (eligible) {
+ if (WorkQueue.casSlotToNull(a, k, t)) {
+ q.base = nextBase;
+ w.source = src;
+ t.doExec();
+ w.source = wsrc;
+ }
+ scan = true;
+ break;
+ }
+ }
}
}
- return true;
}
- else if (--n > 0)
- j = (j + 1) & m;
- else
- break;
}
}
- return false;
+ return s;
}
/**
- * Helps and/or blocks until the given task is done or timeout.
- * First tries locally helping, then scans other queues for a task
- * produced by one of w's stealers; compensating and blocking if
- * none are found (rescanning if tryCompensate fails).
+ * Extra helpJoin steps for CountedCompleters. Scans for and runs
+ * subtasks of the given root task, returning if none are found.
*
- * @param w caller
- * @param task the task
- * @param deadline for timed waits, if nonzero
+ * @param task root of CountedCompleter computation
+ * @param w caller's WorkQueue
+ * @param owned true if owned by a ForkJoinWorkerThread
* @return task status on exit
*/
- final int awaitJoin(WorkQueue w, ForkJoinTask task, long deadline) {
+ final int helpComplete(ForkJoinTask task, WorkQueue w, boolean owned) {
int s = 0;
- int seed = ThreadLocalRandom.nextSecondarySeed();
- if (w != null && task != null &&
- (!(task instanceof CountedCompleter) ||
- (s = w.helpCC((CountedCompleter)task, 0, false)) >= 0)) {
- w.tryRemoveAndExec(task);
- int src = w.source, id = w.id;
- int r = (seed >>> 16) | 1, step = (seed & ~1) | 2;
- s = task.status;
- while (s >= 0) {
- WorkQueue[] ws;
- int n = (ws = workQueues) == null ? 0 : ws.length, m = n - 1;
- while (n > 0) {
- WorkQueue q; int b;
- if ((q = ws[r & m]) != null && q.source == id &&
- q.top != (b = q.base)) {
- ForkJoinTask[] a; int cap, k;
- int qid = q.id;
- if ((a = q.array) != null && (cap = a.length) > 0) {
- ForkJoinTask t = (ForkJoinTask)
- QA.getAcquire(a, k = (cap - 1) & b);
- if (q.source == id && q.base == b++ &&
- t != null && QA.compareAndSet(a, k, t, null)) {
- q.base = b;
- w.source = qid;
- t.doExec();
- w.source = src;
- }
- }
+ if (task != null && w != null) {
+ int r = w.config;
+ boolean scan = true, locals = true;
+ long c = 0L;
+ outer: for (;;) {
+ if (locals) { // try locals before scanning
+ if ((s = w.helpComplete(task, owned, 0)) < 0)
break;
- }
- else {
- r += step;
- --n;
- }
+ locals = false;
}
- if ((s = task.status) < 0)
+ else if ((s = task.status) < 0)
break;
- else if (n == 0) { // empty scan
- long ms, ns; int block;
- if (deadline == 0L)
- ms = 0L; // untimed
- else if ((ns = deadline - System.nanoTime()) <= 0L)
- break; // timeout
- else if ((ms = TimeUnit.NANOSECONDS.toMillis(ns)) <= 0L)
- ms = 1L; // avoid 0 for timed wait
- if ((block = tryCompensate(w)) != 0) {
- task.internalWait(ms);
- CTL.getAndAdd(this, (block > 0) ? RC_UNIT : 0L);
+ else if (scan = !scan) {
+ if (c == (c = ctl))
+ break;
+ }
+ else { // scan for subtasks
+ WorkQueue[] qs = queues;
+ int n = (qs == null) ? 0 : qs.length;
+ for (int i = n; i > 0; --i, ++r) {
+ int j, cap, b; WorkQueue q; ForkJoinTask[] a;
+ boolean eligible = false;
+ if ((q = qs[j = r & (n - 1)]) != null &&
+ (a = q.array) != null && (cap = a.length) > 0) {
+ int k = (cap - 1) & (b = q.base), nextBase = b + 1;
+ ForkJoinTask t = WorkQueue.getSlot(a, k);
+ if (t instanceof CountedCompleter) {
+ CountedCompleter f = (CountedCompleter)t;
+ do {} while (!(eligible = (f == task)) &&
+ (f = f.completer) != null);
+ }
+ if ((s = task.status) < 0)
+ break outer;
+ else if (q.base != b)
+ scan = true; // inconsistent
+ else if (t == null)
+ scan |= (a[nextBase & (cap - 1)] != null ||
+ q.top != b);
+ else if (eligible) {
+ if (WorkQueue.casSlotToNull(a, k, t)) {
+ q.setBaseOpaque(nextBase);
+ t.doExec();
+ locals = true;
+ }
+ scan = true;
+ break;
+ }
+ }
}
- s = task.status;
}
}
}
@@ -1747,112 +1946,164 @@
}
/**
+ * Scans for and returns a polled task, if available. Used only
+ * for untracked polls. Begins scan at an index (scanRover)
+ * advanced on each call, to avoid systematic unfairness.
+ *
+ * @param submissionsOnly if true, only scan submission queues
+ */
+ private ForkJoinTask pollScan(boolean submissionsOnly) {
+ VarHandle.acquireFence();
+ int r = scanRover += 0x61c88647; // Weyl increment; raciness OK
+ if (submissionsOnly) // even indices only
+ r &= ~1;
+ int step = (submissionsOnly) ? 2 : 1;
+ WorkQueue[] qs; int n;
+ while ((qs = queues) != null && (n = qs.length) > 0) {
+ boolean scan = false;
+ for (int i = 0; i < n; i += step) {
+ int j, cap, b; WorkQueue q; ForkJoinTask[] a;
+ if ((q = qs[j = (n - 1) & (r + i)]) != null &&
+ (a = q.array) != null && (cap = a.length) > 0) {
+ int k = (cap - 1) & (b = q.base), nextBase = b + 1;
+ ForkJoinTask t = WorkQueue.getSlot(a, k);
+ if (q.base != b)
+ scan = true;
+ else if (t == null)
+ scan |= (q.top != b || a[nextBase & (cap - 1)] != null);
+ else if (!WorkQueue.casSlotToNull(a, k, t))
+ scan = true;
+ else {
+ q.setBaseOpaque(nextBase);
+ return t;
+ }
+ }
+ }
+ if (!scan && queues == qs)
+ break;
+ }
+ return null;
+ }
+
+ /**
* Runs tasks until {@code isQuiescent()}. Rather than blocking
* when tasks cannot be found, rescans until all others cannot
* find tasks either.
- */
- final void helpQuiescePool(WorkQueue w) {
- int prevSrc = w.source;
- int seed = ThreadLocalRandom.nextSecondarySeed();
- int r = seed >>> 16, step = r | 1;
- for (int source = prevSrc, released = -1;;) { // -1 until known
- ForkJoinTask localTask; WorkQueue[] ws;
- while ((localTask = w.nextLocalTask()) != null)
- localTask.doExec();
- if (w.phase >= 0 && released == -1)
- released = 1;
- boolean quiet = true, empty = true;
- int n = (ws = workQueues) == null ? 0 : ws.length;
- for (int m = n - 1; n > 0; r += step, --n) {
- WorkQueue q; int b;
- if ((q = ws[r & m]) != null) {
- int qs = q.source;
- if (q.top != (b = q.base)) {
- quiet = empty = false;
- ForkJoinTask[] a; int cap, k;
- int qid = q.id;
- if ((a = q.array) != null && (cap = a.length) > 0) {
- if (released == 0) { // increment
- released = 1;
- CTL.getAndAdd(this, RC_UNIT);
- }
- ForkJoinTask t = (ForkJoinTask)
- QA.getAcquire(a, k = (cap - 1) & b);
- if (q.base == b++ && t != null &&
- QA.compareAndSet(a, k, t, null)) {
- q.base = b;
- w.source = qid;
- t.doExec();
- w.source = source = prevSrc;
- }
+ *
+ * @param nanos max wait time (Long.MAX_VALUE if effectively untimed)
+ * @param interruptible true if return on interrupt
+ * @return positive if quiescent, negative if interrupted, else 0
+ */
+ final int helpQuiescePool(WorkQueue w, long nanos, boolean interruptible) {
+ if (w == null)
+ return 0;
+ long startTime = System.nanoTime(), parkTime = 0L;
+ int prevSrc = w.source, wsrc = prevSrc, cfg = w.config, r = cfg + 1;
+ for (boolean active = true, locals = true;;) {
+ boolean busy = false, scan = false;
+ if (locals) { // run local tasks before (re)polling
+ locals = false;
+ for (ForkJoinTask u; (u = w.nextLocalTask(cfg)) != null;)
+ u.doExec();
+ }
+ WorkQueue[] qs = queues;
+ int n = (qs == null) ? 0 : qs.length;
+ for (int i = n; i > 0; --i, ++r) {
+ int j, b, cap; WorkQueue q; ForkJoinTask[] a;
+ if ((q = qs[j = (n - 1) & r]) != null && q != w &&
+ (a = q.array) != null && (cap = a.length) > 0) {
+ int k = (cap - 1) & (b = q.base);
+ int nextBase = b + 1, src = j | SRC;
+ ForkJoinTask t = WorkQueue.getSlot(a, k);
+ if (q.base != b)
+ busy = scan = true;
+ else if (t != null) {
+ busy = scan = true;
+ if (!active) { // increment before taking
+ active = true;
+ getAndAddCtl(RC_UNIT);
+ }
+ if (WorkQueue.casSlotToNull(a, k, t)) {
+ q.base = nextBase;
+ w.source = src;
+ t.doExec();
+ w.source = wsrc = prevSrc;
+ locals = true;
}
break;
}
- else if ((qs & QUIET) == 0)
- quiet = false;
+ else if (!busy) {
+ if (q.top != b || a[nextBase & (cap - 1)] != null)
+ busy = scan = true;
+ else if (q.source != QUIET && q.phase >= 0)
+ busy = true;
+ }
}
}
- if (quiet) {
- if (released == 0)
- CTL.getAndAdd(this, RC_UNIT);
- w.source = prevSrc;
- break;
- }
- else if (empty) {
- if (source != QUIET)
- w.source = source = QUIET;
- if (released == 1) { // decrement
- released = 0;
- CTL.getAndAdd(this, RC_MASK & -RC_UNIT);
+ VarHandle.acquireFence();
+ if (!scan && queues == qs) {
+ boolean interrupted;
+ if (!busy) {
+ w.source = prevSrc;
+ if (!active)
+ getAndAddCtl(RC_UNIT);
+ return 1;
+ }
+ if (wsrc != QUIET)
+ w.source = wsrc = QUIET;
+ if (active) { // decrement
+ active = false;
+ parkTime = 0L;
+ getAndAddCtl(RC_MASK & -RC_UNIT);
+ }
+ else if (parkTime == 0L) {
+ parkTime = 1L << 10; // initially about 1 usec
+ Thread.yield();
+ }
+ else if ((interrupted = interruptible && Thread.interrupted()) ||
+ System.nanoTime() - startTime > nanos) {
+ getAndAddCtl(RC_UNIT);
+ return interrupted ? -1 : 0;
+ }
+ else {
+ LockSupport.parkNanos(this, parkTime);
+ if (parkTime < nanos >>> 8 && parkTime < 1L << 20)
+ parkTime <<= 1; // max sleep approx 1 sec or 1% nanos
}
}
}
}
/**
- * Scans for and returns a polled task, if available.
- * Used only for untracked polls.
+ * Helps quiesce from external caller until done, interrupted, or timeout
*
- * @param submissionsOnly if true, only scan submission queues
+ * @param nanos max wait time (Long.MAX_VALUE if effectively untimed)
+ * @param interruptible true if return on interrupt
+ * @return positive if quiescent, negative if interrupted, else 0
*/
- private ForkJoinTask pollScan(boolean submissionsOnly) {
- WorkQueue[] ws; int n;
- rescan: while ((mode & STOP) == 0 && (ws = workQueues) != null &&
- (n = ws.length) > 0) {
- int m = n - 1;
- int r = ThreadLocalRandom.nextSecondarySeed();
- int h = r >>> 16;
- int origin, step;
- if (submissionsOnly) {
- origin = (r & ~1) & m; // even indices and steps
- step = (h & ~1) | 2;
+ final int externalHelpQuiescePool(long nanos, boolean interruptible) {
+ for (long startTime = System.nanoTime(), parkTime = 0L;;) {
+ ForkJoinTask t;
+ if ((t = pollScan(false)) != null) {
+ t.doExec();
+ parkTime = 0L;
}
- else {
- origin = r & m;
- step = h | 1;
+ else if (canStop())
+ return 1;
+ else if (parkTime == 0L) {
+ parkTime = 1L << 10;
+ Thread.yield();
}
- boolean nonempty = false;
- for (int i = origin, oldSum = 0, checkSum = 0;;) {
- WorkQueue q;
- if ((q = ws[i]) != null) {
- int b; ForkJoinTask t;
- if (q.top - (b = q.base) > 0) {
- nonempty = true;
- if ((t = q.poll()) != null)
- return t;
- }
- else
- checkSum += b + q.id;
- }
- if ((i = (i + step) & m) == origin) {
- if (!nonempty && oldSum == (oldSum = checkSum))
- break rescan;
- checkSum = 0;
- nonempty = false;
- }
+ else if ((System.nanoTime() - startTime) > nanos)
+ return 0;
+ else if (interruptible && Thread.interrupted())
+ return -1;
+ else {
+ LockSupport.parkNanos(this, parkTime);
+ if (parkTime < nanos >>> 8 && parkTime < 1L << 20)
+ parkTime <<= 1;
}
}
- return null;
}
/**
@@ -1862,7 +2113,7 @@
*/
final ForkJoinTask nextTaskFor(WorkQueue w) {
ForkJoinTask t;
- if (w == null || (t = w.nextLocalTask()) == null)
+ if (w == null || (t = w.nextLocalTask(w.config)) == null)
t = pollScan(false);
return t;
}
@@ -1870,115 +2121,94 @@
// External operations
/**
- * Adds the given task to a submission queue at submitter's
- * current queue, creating one if null or contended.
- *
- * @param task the task. Caller must ensure non-null.
+ * Finds and locks a WorkQueue for an external submitter, or
+ * returns null if shutdown or terminating.
*/
- final void externalPush(ForkJoinTask task) {
- int r; // initialize caller's probe
+ final WorkQueue submissionQueue() {
+ int r;
if ((r = ThreadLocalRandom.getProbe()) == 0) {
- ThreadLocalRandom.localInit();
+ ThreadLocalRandom.localInit(); // initialize caller's probe
r = ThreadLocalRandom.getProbe();
}
- for (;;) {
- WorkQueue q;
- int md = mode, n;
- WorkQueue[] ws = workQueues;
- if ((md & SHUTDOWN) != 0 || ws == null || (n = ws.length) <= 0)
- throw new RejectedExecutionException();
- else if ((q = ws[(n - 1) & r & SQMASK]) == null) { // add queue
- int qid = (r | QUIET) & ~(FIFO | OWNED);
- Object lock = workerNamePrefix;
- ForkJoinTask[] qa =
- new ForkJoinTask[INITIAL_QUEUE_CAPACITY];
- q = new WorkQueue(this, null);
- q.array = qa;
- q.id = qid;
- q.source = QUIET;
- if (lock != null) { // unless disabled, lock pool to install
- synchronized (lock) {
- WorkQueue[] vs; int i, vn;
- if ((vs = workQueues) != null && (vn = vs.length) > 0 &&
- vs[i = qid & (vn - 1) & SQMASK] == null)
- vs[i] = q; // else another thread already installed
- }
+ for (int id = r << 1;;) { // even indices only
+ int md = mode, n, i; WorkQueue q; ReentrantLock lock;
+ WorkQueue[] qs = queues;
+ if ((md & SHUTDOWN) != 0 || qs == null || (n = qs.length) <= 0)
+ return null;
+ else if ((q = qs[i = (n - 1) & id]) == null) {
+ if ((lock = registrationLock) != null) {
+ WorkQueue w = new WorkQueue(id | SRC);
+ lock.lock(); // install under lock
+ if (qs[i] == null)
+ qs[i] = w; // else lost race; discard
+ lock.unlock();
}
}
- else if (!q.tryLockPhase()) // move if busy
- r = ThreadLocalRandom.advanceProbe(r);
- else {
- if (q.lockedPush(task))
- signalWork(null);
- return;
- }
+ else if (!q.tryLock()) // move and restart
+ id = (r = ThreadLocalRandom.advanceProbe(r)) << 1;
+ else
+ return q;
}
}
/**
+ * Adds the given task to an external submission queue, or throws
+ * exception if shutdown or terminating.
+ *
+ * @param task the task. Caller must ensure non-null.
+ */
+ final void externalPush(ForkJoinTask task) {
+ WorkQueue q;
+ if ((q = submissionQueue()) == null)
+ throw new RejectedExecutionException(); // shutdown or disabled
+ else if (q.lockedPush(task))
+ signalWork();
+ }
+
+ /**
* Pushes a possibly-external submission.
*/
private For example, here is a ManagedBlocker based on a
* ReentrantLock:
@@ -3111,79 +3382,64 @@
*/
public static void managedBlock(ManagedBlocker blocker)
throws InterruptedException {
+ Thread t; ForkJoinPool p;
+ if ((t = Thread.currentThread()) instanceof ForkJoinWorkerThread &&
+ (p = ((ForkJoinWorkerThread)t).pool) != null)
+ p.compensatedBlock(blocker);
+ else
+ unmanagedBlock(blocker);
+ }
+
+ /** ManagedBlock for ForkJoinWorkerThreads */
+ private void compensatedBlock(ManagedBlocker blocker)
+ throws InterruptedException {
if (blocker == null) throw new NullPointerException();
- ForkJoinPool p;
- ForkJoinWorkerThread wt;
- WorkQueue w;
- Thread t = Thread.currentThread();
- if ((t instanceof ForkJoinWorkerThread) &&
- (p = (wt = (ForkJoinWorkerThread)t).pool) != null &&
- (w = wt.workQueue) != null) {
- int block;
- while (!blocker.isReleasable()) {
- if ((block = p.tryCompensate(w)) != 0) {
- try {
- do {} while (!blocker.isReleasable() &&
- !blocker.block());
- } finally {
- CTL.getAndAdd(p, (block > 0) ? RC_UNIT : 0L);
- }
- break;
+ for (;;) {
+ int comp; boolean done;
+ long c = ctl;
+ if (blocker.isReleasable())
+ break;
+ if ((comp = tryCompensate(c)) >= 0) {
+ long post = (comp == 0) ? 0L : RC_UNIT;
+ try {
+ done = blocker.block();
+ } finally {
+ getAndAddCtl(post);
}
+ if (done)
+ break;
}
}
- else {
- do {} while (!blocker.isReleasable() &&
- !blocker.block());
- }
}
- /**
- * If the given executor is a ForkJoinPool, poll and execute
- * AsynchronousCompletionTasks from worker's queue until none are
- * available or blocker is released.
- */
- static void helpAsyncBlocker(Executor e, ManagedBlocker blocker) {
- if (e instanceof ForkJoinPool) {
- WorkQueue w; ForkJoinWorkerThread wt; WorkQueue[] ws; int r, n;
- ForkJoinPool p = (ForkJoinPool)e;
- Thread thread = Thread.currentThread();
- if (thread instanceof ForkJoinWorkerThread &&
- (wt = (ForkJoinWorkerThread)thread).pool == p)
- w = wt.workQueue;
- else if ((r = ThreadLocalRandom.getProbe()) != 0 &&
- (ws = p.workQueues) != null && (n = ws.length) > 0)
- w = ws[(n - 1) & r & SQMASK];
- else
- w = null;
- if (w != null)
- w.helpAsyncBlocker(blocker);
- }
+ /** ManagedBlock for external threads */
+ private static void unmanagedBlock(ManagedBlocker blocker)
+ throws InterruptedException {
+ if (blocker == null) throw new NullPointerException();
+ do {} while (!blocker.isReleasable() && !blocker.block());
}
- // AbstractExecutorService overrides. These rely on undocumented
- // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
- // implement RunnableFuture.
+ // AbstractExecutorService.newTaskFor overrides rely on
+ // undocumented fact that ForkJoinTask.adapt returns ForkJoinTasks
+ // that also implement RunnableFuture.
+ @Override
protected