< prev index next > src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java
Print this page
* required, we reduce write contention by ensuring that
* signalWork invocations are prefaced with a fully fenced 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. Emptiness must be conservatively approximated,
- * which may result in unnecessary signals. Also, to reduce
- * resource usages in some cases, at the expense of slower
- * startup in others, activation of an idle thread is preferred
- * over creating a new one, here and elsewhere.
- *
- * * At the other extreme, if "flat" tasks (those that do not in
- * turn generate others) come in serially from only a single
- * producer, each worker taking a task from a queue should
- * propagate a signal if there are more tasks in that
- * queue. This is equivalent to, but generally faster than,
- * arranging the stealer take multiple tasks, re-pushing one or
- * more on its own queue, and signalling (because its queue is
- * empty), also resulting in logarithmic full activation
- * time. If tasks do not not engage in unbounded loops based on
- * the actions of other workers with unknown dependencies loop,
- * this form of proagation can be limited to one signal per
- * activation (phase change). We distinguish the cases by
- * further signalling only if the task is an InterruptibleTask
- * (see below), which are the only supported forms of task that
- * may do so.
- *
- * * Because we don't know about usage patterns (or most commonly,
- * mixtures), we use both approaches, which present even more
- * opportunities to over-signal. (Failure to distinguish these
- * cases in terms of submission methods was arguably an early
- * design mistake.) Note that in either of these contexts,
- * signals may be (and often are) unnecessary because active
- * workers continue scanning after running tasks without the
- * need to be signalled (which is one reason work stealing is
- * often faster than alternatives), so additional workers
- * aren't needed.
- *
- * * For rapidly branching tasks that require full pool resources,
- * oversignalling is OK, because signalWork will soon have no
- * more workers to create or reactivate. But for others (mainly
- * externally submitted tasks), overprovisioning may cause very
- * noticeable slowdowns due to contention and resource
- * wastage. We reduce impact by deactivating workers when
- * queues don't have accessible tasks, but reactivating and
- * rescanning if other tasks remain.
- *
- * * Despite these, signal contention and overhead effects still
- * occur during ramp-up and ramp-down of small computations.
+ * workers to scan for tasks. SignalWork is invoked in two cases:
+ * (1) When a task is pushed onto an empty queue, and (2) When a
+ * worker takes a top-level task from a queue that has additional
+ * tasks. Together, these suffice in O(log(#threads)) steps to
+ * fully activate with at least enough workers, and ideally no
+ * more than required. This ideal is unobtainable: Callers do not
+ * know whether another worker will finish its current task and
+ * poll for others without need of a signal (which is otherwise an
+ * advantage of work-stealing vs other schemes), and also must
+ * conservatively estimate the triggering conditions of emptiness
+ * or non-emptiness; all of which usually cause more activations
+ * than necessary (see below). (Method signalWork is also used as
+ * failsafe in case of Thread failures in deregisterWorker, to
+ * activate or create a new worker to replace them).
+ *
+ * Top-Level scheduling
+ * ====================
*
* Scanning. Method runWorker performs top-level scanning for (and
* execution of) tasks by polling a pseudo-random permutation of
* the array (by starting at a given index, and using a constant
* cyclically exhaustive stride.) It uses the same basic polling
* method as WorkQueue.poll(), but restarts with a different
- * permutation on each invocation. The pseudorandom generator
- * need not have high-quality statistical properties in the long
+ * permutation on each rescan. 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. Each queue's polling attempts to avoid becoming stuck
- * when other scanners/pollers stall. 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, which also reduces bookkeeping, cache traffic, and
- * scanning overhead. But it also reduces fairness, which is
- * partially counteracted by giving up on detected interference
- * (which also reduces contention when too many workers try to
- * take small tasks from the same queue).
+ * from ThreadLocalRandom probes, which are cheap and suffice.
*
* Deactivation. When no tasks are found by a worker in runWorker,
- * it tries to deactivate()), giving up (and rescanning) on "ctl"
- * contention. To avoid missed signals during deactivation, the
- * method rescans and reactivates if there may have been a missed
- * signal during deactivation. To reduce false-alarm reactivations
- * while doing so, we scan multiple times (analogously to method
- * quiescent()) before trying to reactivate. Because idle workers
- * are often not yet blocked (parked), we use a WorkQueue field to
- * advertise that a waiter actually needs unparking upon signal.
+ * it invokes awaitWork, that first deactivates (to an IDLE
+ * phase). Avoiding missed signals during deactivation requires a
+ * (conservative) rescan, reactivating if there may be tasks to
+ * poll. Because idle workers are often not yet blocked (parked),
+ * we use a WorkQueue field to advertise that a waiter actually
+ * needs unparking upon signal.
+ *
+ * When tasks are constructed as (recursive) DAGs, top-level
+ * scanning is usually infrequent, and doesn't encounter most
+ * of the following problems addressed by runWorker and awaitWork:
+ *
+ * Locality. Polls are organized into "runs", continuing until
+ * empty or contended, while also minimizing interference by
+ * postponing bookeeping to ends of runs. This may reduce
+ * fairness.
+ *
+ * Contention. When many workers try to poll few queues, they
+ * often collide, generating CAS failures and disrupting locality
+ * of workers already running their tasks. This also leads to
+ * stalls when tasks cannot be taken because other workers have
+ * not finished poll operations, which is detected by reading
+ * ahead in queue arrays. In both cases, workers restart scans in a
+ * way that approximates randomized backoff.
+ *
+ * Oversignalling. When many short top-level tasks are present in
+ * a small number of queues, the above signalling strategy may
+ * activate many more workers than needed, worsening locality and
+ * contention problems, while also generating more global
+ * contention (field ctl is CASed on every activation and
+ * deactivation). We filter out (both in runWorker and
+ * signalWork) attempted signals that are surely not needed
+ * because the signalled tasks are already taken.
+ *
+ * Shutdown and Quiescence
+ * =======================
*
* Quiescence. Workers scan looking for work, giving up when they
* don't find any, without being sure that none are available.
* However, some required functionality relies on consensus about
* quiescence (also termination, discussed below). The count
* InterruptibleTasks include a "runner" field (implemented
* similarly to FutureTask) to support cancel(true). Upon pool
* shutdown, runners are interrupted so they can cancel. Since
* external joining callers never run these tasks, they must await
* cancellation by others, which can occur along several different
- * paths. The inability to rely on caller-runs may also require
- * extra signalling (resulting in scanning and contention) so is
- * done only conditionally in methods push and runworker.
+ * paths.
*
* Across these APIs, rules for reporting exceptions for tasks
* with results accessed via join() differ from those via get(),
* which differ from those invoked using pool submit methods by
* non-workers (which comply with Future.get() specs). Internal
* would be, while also reducing total footprint vs using
* multiple @Contended regions, which tends to slow down
* less-contended applications. To help arrange this, some
* non-reference fields are declared as "long" even when ints or
* shorts would suffice. For class WorkQueue, an
- * embedded @Contended region segregates fields most heavily
- * updated by owners from those most commonly read by stealers or
- * other management.
+ * embedded @Contended isolates the very busy top index, and
+ * another segregates status and bookkeeping fields written
+ * (mostly) by owners, that otherwise interfere with reading
+ * array, top, and base fields. There are other variables commonly
+ * contributing to false-sharing-related performance issues
+ * (including fields of class Thread), but we can't do much about
+ * this except try to minimize access.
*
* Initial sizing and resizing of WorkQueue arrays is an even more
* delicate tradeoff because the best strategy systematically
* varies across garbage collectors. Small arrays are better for
* locality and reduce GC scan time, but large arrays reduce both
* direct false-sharing and indirect cases due to GC bookkeeping
* (cardmarks etc), and reduce the number of resizes, which are
* not especially fast because they require atomic transfers.
- * Currently, arrays for workers are initialized to be just large
- * enough to avoid resizing in most tree-structured tasks, but
- * larger for external queues where both false-sharing problems
- * and the need for resizing are more common. (Maintenance note:
- * any changes in fields, queues, or their uses, or JVM layout
- * policies, must be accompanied by re-evaluation of these
- * placement and sizing decisions.)
+ * Currently, arrays are initialized to be just large enough to
+ * avoid resizing in most tree-structured tasks, but grow rapidly
+ * until large. (Maintenance note: any changes in fields, queues,
+ * or their uses, or JVM layout policies, must be accompanied by
+ * re-evaluation of these placement and sizing decisions.)
*
* Style notes
* ===========
*
* Memory ordering relies mainly on atomic operations (CAS,
* running out of resources needed to do so.
*/
static final int DEFAULT_COMMON_MAX_SPARES = 256;
/**
- * Initial capacity of work-stealing queue array for workers.
+ * Initial capacity of work-stealing queue array.
* Must be a power of two, at least 2. See above.
*/
static final int INITIAL_QUEUE_CAPACITY = 1 << 6;
- /**
- * Initial capacity of work-stealing queue array for external queues.
- * Must be a power of two, at least 2. See above.
- */
- static final int INITIAL_EXTERNAL_QUEUE_CAPACITY = 1 << 9;
-
// conversions among short, int, long
static final int SMASK = 0xffff; // (unsigned) short bits
static final long LMASK = 0xffffffffL; // lower 32 bits of long
static final long UMASK = ~LMASK; // upper 32 bits
final ForkJoinWorkerThread owner; // null if shared
ForkJoinTask<?>[] array; // the queued tasks; power of 2 size
int base; // index of next slot for poll
final int config; // mode bits
- // fields otherwise causing more unnecessary false-sharing cache misses
- @jdk.internal.vm.annotation.Contended("w")
+ @jdk.internal.vm.annotation.Contended("t") // segregate
int top; // index of next slot for push
+
+ // fields otherwise causing more unnecessary false-sharing cache misses
@jdk.internal.vm.annotation.Contended("w")
volatile int phase; // versioned active status
@jdk.internal.vm.annotation.Contended("w")
int stackPred; // pool stack (ctl) predecessor link
@jdk.internal.vm.annotation.Contended("w")
+ volatile int parking; // nonzero if parked in awaitWork
+ @jdk.internal.vm.annotation.Contended("w")
volatile int source; // source queue id (or DROPPED)
@jdk.internal.vm.annotation.Contended("w")
int nsteals; // number of steals from other queues
- @jdk.internal.vm.annotation.Contended("w")
- volatile int parking; // nonzero if parked in awaitWork
// Support for atomic operations
private static final Unsafe U;
private static final long PHASE;
private static final long BASE;
* Constructor. For internal queues, most fields are initialized
* upon thread start in pool.registerWorker.
*/
WorkQueue(ForkJoinWorkerThread owner, int id, int cfg,
boolean clearThreadLocals) {
- array = new ForkJoinTask<?>[owner == null ?
- INITIAL_EXTERNAL_QUEUE_CAPACITY :
- INITIAL_QUEUE_CAPACITY];
- this.owner = owner;
this.config = (clearThreadLocals) ? cfg | CLEAR_TLS : cfg;
+ if ((this.owner = owner) == null) {
+ array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
+ phase = id | IDLE;
+ }
}
/**
* Returns an exportable index (used by ForkJoinWorkerThread).
*/
* @param internal if caller owns this queue
* @throws RejectedExecutionException if array could not be resized
*/
final void push(ForkJoinTask<?> task, ForkJoinPool pool, boolean internal) {
int s = top, b = base, m, cap, room; ForkJoinTask<?>[] a;
- if ((a = array) != null && (cap = a.length) > 0 && // else disabled
- task != null) {
- int pk = task.noUserHelp() + 1; // prev slot offset
+ if ((a = array) != null && (cap = a.length) > 0) { // else disabled
if ((room = (m = cap - 1) - (s - b)) >= 0) {
top = s + 1;
long pos = slotOffset(m & s);
if (!internal)
U.putReference(a, pos, task); // inside lock
else
U.getAndSetReference(a, pos, task); // fully fenced
- if (room == 0) // resize
- growArray(a, cap, s);
+ if (room == 0 && (a = growArray(a, cap, s)) != null)
+ m = a.length - 1; // resize
}
if (!internal)
unlockPhase();
if (room < 0)
throw new RejectedExecutionException("Queue capacity exceeded");
- if ((room == 0 || a[m & (s - pk)] == null) &&
- pool != null)
- pool.signalWork(); // may have appeared empty
+ if (pool != null && a != null &&
+ U.getReferenceAcquire(a, slotOffset(m & (s - 1))) == null)
+ pool.signalWork(a, m & s); // may have appeared empty
}
}
/**
* Resizes the queue array unless out of memory.
* @param a old array
* @param cap old array capacity
* @param s current top
+ * @return new array, or null on failure
*/
- private void growArray(ForkJoinTask<?>[] a, int cap, int s) {
- int newCap = cap << 1;
+ private ForkJoinTask<?>[] growArray(ForkJoinTask<?>[] a, int cap, int s) {
+ int newCap = (cap >= 1 << 16) ? cap << 1 : cap << 2;
+ ForkJoinTask<?>[] newArray = null;
if (a != null && a.length == cap && cap > 0 && newCap > 0) {
- ForkJoinTask<?>[] newArray = null;
try {
newArray = new ForkJoinTask<?>[newCap];
} catch (OutOfMemoryError ex) {
}
if (newArray != null) { // else throw on next push
newArray[k & newMask] = u;
}
updateArray(newArray); // fully fenced
}
}
+ return newArray;
}
/**
- * Takes next task, if one exists, in order specified by mode,
- * so acts as either local-pop or local-poll. Called only by owner.
- * @param fifo nonzero if FIFO mode
+ * Takes next task, if one exists, in lifo order.
*/
- private ForkJoinTask<?> nextLocalTask(int fifo) {
+ private ForkJoinTask<?> localPop() {
ForkJoinTask<?> t = null;
- ForkJoinTask<?>[] a = array;
- int b = base, p = top, cap;
- if (p - b > 0 && a != null && (cap = a.length) > 0) {
- for (int m = cap - 1, s, nb;;) {
- if (fifo == 0 || (nb = b + 1) == p) {
- if ((t = (ForkJoinTask<?>)U.getAndSetReference(
- a, slotOffset(m & (s = p - 1)), null)) != null)
- updateTop(s); // else lost race for only task
- break;
+ int s = top - 1, cap; long k; ForkJoinTask<?>[] a;
+ if ((a = array) != null && (cap = a.length) > 0 &&
+ U.getReference(a, k = slotOffset((cap - 1) & s)) != null &&
+ (t = (ForkJoinTask<?>)U.getAndSetReference(a, k, null)) != null)
+ updateTop(s);
+ return t;
+ }
+
+ /**
+ * Takes next task, if one exists, in fifo order.
+ */
+ private ForkJoinTask<?> localPoll() {
+ ForkJoinTask<?> t = null;
+ int p = top, cap; ForkJoinTask<?>[] a;
+ if ((a = array) != null && (cap = a.length) > 0) {
+ for (int b = base; p - b > 0; ) {
+ int nb = b + 1;
+ long k = slotOffset((cap - 1) & b);
+ if (U.getReference(a, k) == null) {
+ if (nb == p)
+ break; // else base is lagging
+ while (b == (b = U.getIntAcquire(this, BASE)))
+ Thread.onSpinWait(); // spin to reduce memory traffic
}
- if ((t = (ForkJoinTask<?>)U.getAndSetReference(
- a, slotOffset(m & b), null)) != null) {
+ else if ((t = (ForkJoinTask<?>)
+ U.getAndSetReference(a, k, null)) != null) {
updateBase(nb);
break;
}
- while (b == (b = U.getIntAcquire(this, BASE)))
- Thread.onSpinWait(); // spin to reduce memory traffic
- if (p - b <= 0)
- break;
+ else
+ b = base;
}
}
return t;
}
/**
* Takes next task, if one exists, using configured mode.
- * (Always internal, never called for Common pool.)
*/
final ForkJoinTask<?> nextLocalTask() {
- return nextLocalTask(config & FIFO);
+ return (config & FIFO) == 0 ? localPop() : localPoll();
}
/**
* Pops the given task only if it is at the current top.
* @param task the task. Caller must ensure non-null.
* Runs the given task, as well as remaining local tasks.
*/
final void topLevelExec(ForkJoinTask<?> task, int fifo) {
while (task != null) {
task.doExec();
- task = nextLocalTask(fifo);
+ task = (fifo == 0) ? localPop() : localPoll();
}
}
/**
* Deep form of tryUnpush: Traverses from top and removes and
/**
* Cancels all local tasks. Called only by owner.
*/
final void cancelTasks() {
- for (ForkJoinTask<?> t; (t = nextLocalTask(0)) != null; ) {
+ for (ForkJoinTask<?> t; (t = localPop()) != null; ) {
try {
t.cancel(false);
} catch (Throwable ignore) {
}
}
* Finishes initializing and records internal queue.
*
* @param w caller's WorkQueue
*/
final void registerWorker(WorkQueue w) {
- if (w != null && (runState & STOP) == 0L) {
+ if (w != null) {
+ w.array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
ThreadLocalRandom.localInit();
int seed = w.stackPred = ThreadLocalRandom.getProbe();
int phaseSeq = seed & ~((IDLE << 1) - 1); // initial phase tag
int id = ((seed << 1) | 1) & SMASK; // base of linear-probe-like scan
long stop = lockRunState() & STOP;
unlockRunState();
}
}
if ((tryTerminate(false, false) & STOP) == 0L &&
phase != 0 && w != null && w.source != DROPPED) {
- signalWork(); // possibly replace
w.cancelTasks(); // clean queue
+ signalWork(null, 0); // possibly replace
}
if (ex != null)
ForkJoinTask.rethrow(ex);
}
/**
- * Releases an idle worker, or creates one if not enough exist.
+ * Releases an idle worker, or creates one if not enough exist,
+ * giving up if array a is nonnull and task at a[k] already taken.
*/
- final void signalWork() {
+ final void signalWork(ForkJoinTask<?>[] a, int k) {
int pc = parallelism;
for (long c = ctl;;) {
WorkQueue[] qs = queues;
long ac = (c + RC_UNIT) & RC_MASK, nc;
int sp = (int)c, i = sp & SMASK;
break;
WorkQueue w = qs[i], v = null;
if (sp == 0) {
if ((short)(c >>> TC_SHIFT) >= pc)
break;
- nc = ((c + TC_UNIT) & TC_MASK);
+ nc = ((c + TC_UNIT) & TC_MASK) | ac;
}
else if ((v = w) == null)
break;
else
- nc = (v.stackPred & LMASK) | (c & TC_MASK);
- if (c == (c = compareAndExchangeCtl(c, nc | ac))) {
+ nc = (v.stackPred & LMASK) | (c & TC_MASK) | ac;
+ if (a != null && k < a.length && k >= 0 && a[k] == null)
+ break;
+ if (c == (c = ctl) && c == (c = compareAndExchangeCtl(c, nc))) {
if (v == null)
createWorker();
else {
v.phase = sp;
if (v.parking != 0)
*
* @param w caller's WorkQueue (may be null on failed initialization)
*/
final void runWorker(WorkQueue w) {
if (w != null) {
- int phase = w.phase, r = w.stackPred; // seed from registerWorker
- int fifo = w.config & FIFO, nsteals = 0, src = -1;
- for (;;) {
- WorkQueue[] qs;
+ int r = w.stackPred; // seed from registerWorker
+ int fifo = (int)config & FIFO;
+ int nsteals = 0; // shadow w.nsteals
+ boolean rescan = true;
+ WorkQueue[] qs; int n;
+ while ((rescan || deactivate(w) == 0) && (runState & STOP) == 0L &&
+ (qs = queues) != null && (n = qs.length) > 0) {
+ rescan = false;
+ int i = r, step = (r >>> 16) | 1;
r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
- if ((runState & STOP) != 0L || (qs = queues) == null)
- break;
- int n = qs.length, i = r, step = (r >>> 16) | 1;
- boolean rescan = false;
- scan: for (int l = n; l > 0; --l, i += step) { // scan queues
- int j, cap; WorkQueue q; ForkJoinTask<?>[] a;
- if ((q = qs[j = i & (n - 1)]) != null &&
- (a = q.array) != null && (cap = a.length) > 0) {
- for (int m = cap - 1, pb = -1, b = q.base;;) {
- ForkJoinTask<?> t; long k;
- t = (ForkJoinTask<?>)U.getReferenceAcquire(
- a, k = slotOffset(m & b));
- if (b != (b = q.base) || t == null ||
- !U.compareAndSetReference(a, k, t, null)) {
- if (a[b & m] == null) {
- if (rescan) // end of run
- break scan;
- if (a[(b + 1) & m] == null &&
- a[(b + 2) & m] == null) {
- break; // probably empty
- }
- if (pb == (pb = b)) { // track progress
- rescan = true; // stalled; reorder scan
- break scan;
- }
+ scan: for (int j = n << 1; j != 0; --j, i += step) { // 2 sweeps
+ WorkQueue q; int qid;
+ if ((q = qs[qid = i & (n - 1)]) != null) {
+ for (;;) { // poll queue q
+ ForkJoinTask<?>[] a; int cap, b, m, nb, nk;
+ if ((a = q.array) == null || (cap = a.length) <= 0)
+ break;
+ long bp = slotOffset((m = cap - 1) & (b = q.base));
+ long np = slotOffset(nk = m & (nb = b + 1));
+ ForkJoinTask<?> t = (ForkJoinTask<?>)
+ U.getReferenceAcquire(a, bp);
+ if (q.array != a || q.base != b ||
+ U.getReference(a, bp) != t)
+ continue; // inconsistent
+ if (t == null) {
+ if (rescan) { // end of run
+ w.nsteals = nsteals;
+ break scan;
+ }
+ if (U.getReference(a, np) != null) {
+ rescan = true; // stalled; reorder scan
+ break scan;
}
+ break; // probably empty
}
- else {
- boolean propagate;
- int nb = q.base = b + 1, prevSrc = src;
- w.nsteals = ++nsteals;
- w.source = src = j; // volatile
- rescan = true;
- int nh = t.noUserHelp();
- if (propagate =
- (prevSrc != src || nh != 0) && a[nb & m] != null)
- signalWork();
- w.topLevelExec(t, fifo);
- if ((b = q.base) != nb && !propagate)
- break scan; // reduce interference
+ if (U.compareAndSetReference(a, bp, t, null)) {
+ q.base = nb;
+ Object nt = U.getReferenceAcquire(a, np);
+ if (!rescan) { // begin run
+ rescan = true;
+ w.source = qid;
+ }
+ ++nsteals;
+ if (nt != null && // confirm a[nk]
+ U.getReference(a, np) == nt)
+ signalWork(a, nk); // propagate
+ w.topLevelExec(t, fifo); // run t & its subtasks
}
}
}
}
- if (!rescan) {
- if (((phase = deactivate(w, phase)) & IDLE) != 0)
- break;
- src = -1; // re-enable propagation
- }
}
}
}
/**
- * Deactivates and if necessary awaits signal or termination.
+ * Deactivates and awaits signal or termination.
*
- * @param w the worker
- * @param phase current phase
- * @return current phase, with IDLE set if worker should exit
+ * @param w the work queue
+ * @return zero if now active
*/
- private int deactivate(WorkQueue w, int phase) {
- if (w == null) // currently impossible
- return IDLE;
- int p = phase | IDLE, activePhase = phase + (IDLE << 1);
- long pc = ctl, qc = (activePhase & LMASK) | ((pc - RC_UNIT) & UMASK);
- int sp = w.stackPred = (int)pc; // set ctl stack link
- w.phase = p;
- if (!compareAndSetCtl(pc, qc)) // try to enqueue
- return w.phase = phase; // back out on possible signal
- int ac = (short)(qc >>> RC_SHIFT), n; long e; WorkQueue[] qs;
- if (((e = runState) & STOP) != 0L ||
- ((e & SHUTDOWN) != 0L && ac == 0 && quiescent() > 0) ||
- (qs = queues) == null || (n = qs.length) <= 0)
- return IDLE; // terminating
-
- for (int prechecks = Math.min(ac, 2), // reactivation threshold
- k = Math.max(n + (n << 1), SPIN_WAITS << 1);;) {
- WorkQueue q; int cap; ForkJoinTask<?>[] a; long c;
- if (w.phase == activePhase)
- return activePhase;
- if (--k < 0)
- return awaitWork(w, p); // block, drop, or exit
- if ((q = qs[k & (n - 1)]) == null)
- Thread.onSpinWait();
- else if ((a = q.array) != null && (cap = a.length) > 0 &&
- a[q.base & (cap - 1)] != null && --prechecks < 0 &&
- (int)(c = ctl) == activePhase &&
- compareAndSetCtl(c, (sp & LMASK) | ((c + RC_UNIT) & UMASK)))
- return w.phase = activePhase; // reactivate
+ private int deactivate(WorkQueue w) {
+ int idle = 1;
+ if (w != null) { // always true; hoist checks
+ int inactive = w.phase |= IDLE; // set status
+ int activePhase = inactive + IDLE; // phase value when reactivated
+ long ap = activePhase & LMASK, pc = ctl, qc;
+ do { // enqueue
+ qc = ap | ((pc - RC_UNIT) & UMASK);
+ w.stackPred = (int)pc; // set ctl stack link
+ } while (pc != (pc = compareAndExchangeCtl(pc, qc)));
+
+ WorkQueue[] qs; int n; long e;
+ if (((e = runState) & STOP) == 0 && // quiescence checks
+ ((e & SHUTDOWN) == 0L || (qc & RC_MASK) > 0L || quiescent() <= 0) &&
+ (qs = queues) != null && (n = qs.length) > 1) {
+ long psp = pc & LMASK; // ctl predecessor prefix
+ for (int i = 1; i < n; ++i) { // scan; stagger origins
+ WorkQueue q; long c; // missed signal check
+ if ((q = qs[(activePhase + i) & (n - 1)]) != null &&
+ q.top - q.base > 0) {
+ if ((idle = w.phase - activePhase) != 0 &&
+ (int)(c = ctl) == activePhase &&
+ compareAndSetCtl(c, psp | ((c + RC_UNIT) & UMASK))) {
+ w.phase = activePhase;
+ idle = 0; // reactivated
+ } // else ineligible or lost race
+ break;
+ }
+ }
+ if (idle != 0 && (idle = w.phase - activePhase) != 0)
+ idle = awaitWork(w, activePhase, n);
+ }
}
+ return idle;
}
/**
* Awaits signal or termination.
*
* @param w the work queue
- * @param p current phase (known to be idle)
- * @return current phase, with IDLE set if worker should exit
+ * @param activePhase w's next active phase
+ * @param qsize current size of queues array
+ * @return zero if now active
*/
- private int awaitWork(WorkQueue w, int p) {
- if (w != null) {
- ForkJoinWorkerThread t; long deadline;
- if ((w.config & CLEAR_TLS) != 0 && (t = w.owner) != null)
- t.resetThreadLocals(); // clear before reactivate
- if ((ctl & RC_MASK) > 0L)
+ private int awaitWork(WorkQueue w, int activePhase, int qsize) {
+ int idle = 1;
+ int spins = qsize | (qsize - 1); // approx traversal cost
+ if (w != null) { // always true; hoist checks
+ boolean trimmable; long deadline, c;
+ long trimTime = (w.source == INVALID_ID) ? TIMEOUT_SLOP : keepAlive;
+ if ((w.config & CLEAR_TLS) != 0 && // instanceof check always true
+ Thread.currentThread() instanceof ForkJoinWorkerThread f)
+ f.resetThreadLocals(); // clear while accessing thread state
+ LockSupport.setCurrentBlocker(this);
+ if (trimmable = (((c = ctl) & RC_MASK) == 0L && (int)c == activePhase))
+ deadline = trimTime + System.currentTimeMillis();
+ else
deadline = 0L;
- else if ((deadline =
- (((w.source != INVALID_ID) ? keepAlive : TIMEOUT_SLOP)) +
- System.currentTimeMillis()) == 0L)
- deadline = 1L; // avoid zero
- int activePhase = p + IDLE;
- if ((p = w.phase) != activePhase && (runState & STOP) == 0L) {
- LockSupport.setCurrentBlocker(this);
- w.parking = 1; // enable unpark
- while ((p = w.phase) != activePhase) {
- boolean trimmable = false; int trim;
- Thread.interrupted(); // clear status
- if ((runState & STOP) != 0L)
+ for (;;) {
+ int s = spins, trim;
+ Thread.interrupted(); // clear status
+ if ((runState & STOP) != 0L)
+ break;
+ while ((idle = w.phase - activePhase) != 0 && --s != 0)
+ Thread.onSpinWait(); // spin before blocking
+ if (idle == 0)
+ break;
+ if (trimmable &&
+ (trim = tryTrim(w, activePhase, deadline)) != 0) {
+ if (trim > 0)
break;
- if (deadline != 0L) {
- if ((trim = tryTrim(w, p, deadline)) > 0)
- break;
- else if (trim < 0)
- deadline = 0L;
- else
- trimmable = true;
- }
- U.park(trimmable, deadline);
+ trimmable = false;
+ deadline = 0L;
}
- w.parking = 0;
- LockSupport.setCurrentBlocker(null);
+ w.parking = 1; // enable unpark and recheck
+ if ((idle = w.phase - activePhase) != 0)
+ U.park(trimmable, deadline);
+ w.parking = 0; // close unpark window
+ if (idle == 0 || (idle = w.phase - activePhase) == 0)
+ break;
}
+ LockSupport.setCurrentBlocker(null);
}
- return p;
+ return idle;
}
/**
* Tries to remove and deregister worker after timeout, and release
* another to do the same.
* @return > 0: trimmed, < 0 : not trimmable, else 0
*/
- private int tryTrim(WorkQueue w, int phase, long deadline) {
- long c, nc; int stat, activePhase, vp, i; WorkQueue[] vs; WorkQueue v;
- if ((activePhase = phase + IDLE) != (int)(c = ctl) || w == null)
+ private int tryTrim(WorkQueue w, int activePhase, long deadline) {
+ long c, nc; int stat, vp, i; WorkQueue[] vs; WorkQueue v;
+ long waitTime = deadline - System.currentTimeMillis();
+ if ((int)(c = ctl) != activePhase || w == null)
stat = -1; // no longer ctl top
- else if (deadline - System.currentTimeMillis() >= TIMEOUT_SLOP)
+ else if (waitTime > TIMEOUT_SLOP)
stat = 0; // spurious wakeup
else if (!compareAndSetCtl(
c, nc = ((w.stackPred & LMASK) | (RC_MASK & c) |
(TC_MASK & (c - TC_UNIT)))))
stat = -1; // lost race to signaller
// External operations
/**
* Finds and locks a WorkQueue for an external submitter, or
- * throws RejectedExecutionException if shutdown or terminating.
- * @param r current ThreadLocalRandom.getProbe() value
+ * throws RejectedExecutionException if shutdown
* @param rejectOnShutdown true if RejectedExecutionException
- * should be thrown when shutdown (else only if terminating)
+ * should be thrown when shutdown
*/
- private WorkQueue submissionQueue(int r, boolean rejectOnShutdown) {
- int reuse; // nonzero if prefer create
- if ((reuse = r) == 0) {
- ThreadLocalRandom.localInit(); // initialize caller's probe
+ final WorkQueue externalSubmissionQueue(boolean rejectOnShutdown) {
+ int r;
+ if ((r = ThreadLocalRandom.getProbe()) == 0) {
+ ThreadLocalRandom.localInit(); // initialize caller's probe
r = ThreadLocalRandom.getProbe();
}
- for (int probes = 0; ; ++probes) {
- int n, i, id; WorkQueue[] qs; WorkQueue q;
- if ((qs = queues) == null)
- break;
- if ((n = qs.length) <= 0)
+ for (;;) {
+ WorkQueue q; WorkQueue[] qs; int n, id, i;
+ if ((qs = queues) == null || (n = qs.length) <= 0)
break;
if ((q = qs[i = (id = r & EXTERNAL_ID_MASK) & (n - 1)]) == null) {
- WorkQueue w = new WorkQueue(null, id, 0, false);
- w.phase = id;
- boolean reject = ((lockRunState() & SHUTDOWN) != 0 &&
- rejectOnShutdown);
- if (!reject && queues == qs && qs[i] == null)
- q = qs[i] = w; // else lost race to install
+ WorkQueue newq = new WorkQueue(null, id, 0, false);
+ lockRunState();
+ if (qs[i] == null && queues == qs)
+ q = qs[i] = newq; // else lost race to install
unlockRunState();
- if (q != null)
- return q;
- if (reject)
- break;
- reuse = 0;
}
- if (reuse == 0 || !q.tryLockPhase()) { // move index
- if (reuse == 0) {
- if (probes >= n >> 1)
- reuse = r; // stop prefering free slot
+ if (q != null && q.tryLockPhase()) {
+ if (rejectOnShutdown && (runState & SHUTDOWN) != 0L) {
+ q.unlockPhase(); // check while q lock held
+ break;
}
- else if (q != null)
- reuse = 0; // probe on collision
- r = ThreadLocalRandom.advanceProbe(r);
- }
- else if (rejectOnShutdown && (runState & SHUTDOWN) != 0L) {
- q.unlockPhase(); // check while q lock held
- break;
- }
- else
return q;
+ }
+ r = ThreadLocalRandom.advanceProbe(r); // move
}
throw new RejectedExecutionException();
}
private <T> ForkJoinTask<T> poolSubmit(boolean signalIfEmpty, ForkJoinTask<T> task) {
internal = true;
q = wt.workQueue;
}
else { // find and lock queue
internal = false;
- q = submissionQueue(ThreadLocalRandom.getProbe(), true);
+ q = externalSubmissionQueue(true);
}
q.push(task, signalIfEmpty ? this : null, internal);
return task;
}
- /**
- * Returns queue for an external submission, bypassing call to
- * submissionQueue if already established and unlocked.
- */
- final WorkQueue externalSubmissionQueue(boolean rejectOnShutdown) {
- WorkQueue[] qs; WorkQueue q; int n;
- int r = ThreadLocalRandom.getProbe();
- return (((qs = queues) != null && (n = qs.length) > 0 &&
- (q = qs[r & EXTERNAL_ID_MASK & (n - 1)]) != null && r != 0 &&
- q.tryLockPhase()) ? q : submissionQueue(r, rejectOnShutdown));
- }
-
/**
* Returns queue for an external thread, if one exists that has
* possibly ever submitted to the given pool (nonzero probe), or
* null if none.
*/
< prev index next >