--- jsr166/src/jsr166e/ForkJoinPool.java 2012/11/19 18:12:28 1.18 +++ jsr166/src/jsr166e/ForkJoinPool.java 2013/01/28 19:01:38 1.51 @@ -51,7 +51,7 @@ import java.util.concurrent.TimeUnit; * dynamically adding, suspending, or resuming internal worker * threads, even if some tasks are stalled waiting to join * others. However, no such adjustments are guaranteed in the face of - * blocked IO or other unmanaged synchronization. The nested {@link + * blocked I/O or other unmanaged synchronization. The nested {@link * ManagedBlocker} interface enables extension of the kinds of * synchronization accommodated. * @@ -99,16 +99,12 @@ import java.util.concurrent.TimeUnit; * *

The common pool is by default constructed with default * parameters, but these may be controlled by setting three {@link - * System#getProperty properties} with prefix {@code + * System#getProperty system properties} with prefix {@code * java.util.concurrent.ForkJoinPool.common}: {@code parallelism} -- * an integer greater than zero, {@code threadFactory} -- the class * name of a {@link ForkJoinWorkerThreadFactory}, and {@code -<<<<<<< ForkJoinPool.java - * exceptionHandler} -- the class name of a {@code -======= * exceptionHandler} -- the class name of a {@link * java.lang.Thread.UncaughtExceptionHandler ->>>>>>> 1.111 * Thread.UncaughtExceptionHandler}. Upon any error in establishing * these settings, default parameters are used. * @@ -320,17 +316,18 @@ public class ForkJoinPool extends Abstra * execute. However, many other threads may notice the same task * and each signal to wake up a thread that might take it. So in * general, pools will be over-signalled. When a submission is - * added or another worker adds a task to a queue that is - * apparently empty, they signal waiting workers (or trigger - * creation of new ones if fewer than the given parallelism - * level). These primary signals are buttressed by signals - * whenever other threads scan for work or do not have a task to - * process (including the case of leaving a hint to unparked - * threads to help signal others upon wakeup). On most platforms, - * signalling (unpark) overhead time is noticeably long, and the - * time between signalling a thread and it actually making - * progress can be very noticeably long, so it is worth offloading - * these delays from critical paths as much as possible. + * added or another worker adds a task to a queue that has fewer + * than two tasks, they signal waiting workers (or trigger + * creation of new ones if fewer than the given parallelism level + * -- signalWork), and may leave a hint to the unparked worker to + * help signal others upon wakeup). These primary signals are + * buttressed by others (see method helpSignal) whenever other + * threads scan for work or do not have a task to process. On + * most platforms, signalling (unpark) overhead time is noticeably + * long, and the time between signalling a thread and it actually + * making progress can be very noticeably long, so it is worth + * offloading these delays from critical paths as much as + * possible. * * Trimming workers. To release resources after periods of lack of * use, a worker starting to wait when the pool is quiescent will @@ -443,7 +440,7 @@ public class ForkJoinPool extends Abstra * Common Pool * =========== * - * The static commonPool always exists after static + * The static common Pool always exists after static * initialization. Since it (or any other created pool) need * never be used, we minimize initial construction overhead and * footprint to the setup of about a dozen fields, with no nested @@ -538,6 +535,26 @@ public class ForkJoinPool extends Abstra } /** + * Per-thread records for threads that submit to pools. Currently + * holds only pseudo-random seed / index that is used to choose + * submission queues in method externalPush. In the future, this may + * also incorporate a means to implement different task rejection + * and resubmission policies. + * + * Seeds for submitters and workers/workQueues work in basically + * the same way but are initialized and updated using slightly + * different mechanics. Both are initialized using the same + * approach as in class ThreadLocal, where successive values are + * unlikely to collide with previous values. Seeds are then + * randomly modified upon collisions using xorshifts, which + * requires a non-zero seed. + */ + static final class Submitter { + int seed; + Submitter(int s) { seed = s; } + } + + /** * Class for artificial tasks that are used to replace the target * of local joins if they are removed from an interior queue slot * in WorkQueue.tryRemoveAndExec. We don't need the proxy to @@ -603,7 +620,10 @@ public class ForkJoinPool extends Abstra * trades off slightly slower average field access for the sake of * avoiding really bad worst-case access. (Until better JVM * support is in place, this padding is dependent on transient - * properties of JVM field layout rules.) + * properties of JVM field layout rules.) We also take care in + * allocating, sizing and resizing the array. Non-shared queue + * arrays are initialized by workers before use. Others are + * allocated on first use. */ static final class WorkQueue { /** @@ -626,6 +646,9 @@ public class ForkJoinPool extends Abstra */ static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M + // Heuristic padding to ameliorate unfortunate memory placements + volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06; + int seed; // for random scanning; initialize nonzero volatile int eventCount; // encoded inactivation count; < 0 if inactive int nextWait; // encoded record of next event waiter @@ -643,73 +666,65 @@ public class ForkJoinPool extends Abstra volatile ForkJoinTask currentJoin; // task being joined in awaitJoin ForkJoinTask currentSteal; // current non-local task being executed - // Heuristic padding to ameliorate unfortunate memory placements - Object p00, p01, p02, p03, p04, p05, p06, p07; - Object p08, p09, p0a, p0b, p0c; + volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17; + volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d; WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode, int seed) { - this.array = new ForkJoinTask[WorkQueue.INITIAL_QUEUE_CAPACITY]; this.pool = pool; this.owner = owner; this.mode = mode; this.seed = seed; - // Place indices in the center of array + // Place indices in the center of array (that is not yet allocated) base = top = INITIAL_QUEUE_CAPACITY >>> 1; } /** - * Pushes a task. Call only by owner in unshared queues. - * Cases needing resizing or rejection are relayed to fullPush - * (that also handles shared queues). + * Returns the approximate number of tasks in the queue. + */ + final int queueSize() { + int n = base - top; // non-owner callers must read base first + return (n >= 0) ? 0 : -n; // 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. + */ + final boolean isEmpty() { + ForkJoinTask[] a; int m, s; + int n = base - (s = top); + return (n >= 0 || + (n == -1 && + ((a = array) == null || + (m = a.length - 1) < 0 || + U.getObject + (a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null))); + } + + /** + * Pushes a task. Call only by owner in unshared queues. (The + * shared-queue version is embedded in method externalPush.) * * @param task the task. Caller must ensure non-null. - * @throw RejectedExecutionException if array cannot be resized + * @throws RejectedExecutionException if array cannot be resized */ final void push(ForkJoinTask task) { ForkJoinTask[] a; ForkJoinPool p; int s = top, m, n; if ((a = array) != null) { // ignore if queue removed - U.putOrderedObject - (a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task); - if ((n = (top = s + 1) - base) <= 1) { + int j = (((m = a.length - 1) & s) << ASHIFT) + ABASE; + U.putOrderedObject(a, j, task); + if ((n = (top = s + 1) - base) <= 2) { if ((p = pool) != null) - p.signalWork(this, 0); + p.signalWork(this); } else if (n >= m) growArray(); } } - /** - * Pushes a task if lock is free and array is either big - * enough or can be resized to be big enough. - * - * @param task the task. Caller must ensure non-null. - * @return true if submitted - */ - final boolean trySharedPush(ForkJoinTask task) { - boolean submitted = false; - if (qlock == 0 && U.compareAndSwapInt(this, QLOCK, 0, 1)) { - ForkJoinTask[] a = array; ForkJoinPool p; - int s = top; - try { - if ((a != null && a.length > s + 1 - base) || - (a = growArray()) != null) { // must presize - int j = (((a.length - 1) & s) << ASHIFT) + ABASE; - U.putOrderedObject(a, j, task); - top = s + 1; - submitted = true; - } - } finally { - qlock = 0; // unlock - } - if (submitted && (p = pool) != null) - p.signalWork(this, 0); - } - return submitted; - } - /** * Initializes or doubles the capacity of array. Call either * by owner or with lock held -- it is OK for base, but not @@ -859,21 +874,6 @@ public class ForkJoinPool extends Abstra return seed = r ^= r << 5; } - /** - * Provides a more accurate estimate of size than (top - base) - * by ordering reads and checking whether a near-empty queue - * has at least one unclaimed task. - */ - final int queueSize() { - ForkJoinTask[] a; int k, s, n; - return ((n = base - (s = top)) < 0 && - (n != -1 || - ((a = array) != null && (k = a.length) > 0 && - U.getObject - (a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ? - -n : 0; - } - // Specialized execution methods /** @@ -988,7 +988,7 @@ public class ForkJoinPool extends Abstra (currentSteal = t).doExec(); currentSteal = null; ++nsteals; - if (top != base) { // process remaining local tasks + if (base - top < 0) { // process remaining local tasks if (mode == 0) popAndExecAll(); else @@ -1020,29 +1020,12 @@ public class ForkJoinPool extends Abstra s != Thread.State.TIMED_WAITING); } - /** - * If this owned and is not already interrupted, try to - * interrupt and/or unpark, ignoring exceptions. - */ - final void interruptOwner() { - Thread wt, p; - if ((wt = owner) != null && !wt.isInterrupted()) { - try { - wt.interrupt(); - } catch (SecurityException ignore) { - } - } - if ((p = parker) != null) - U.unpark(p); - } - // Unsafe mechanics private static final sun.misc.Unsafe U; private static final long QLOCK; private static final int ABASE; private static final int ASHIFT; static { - int s; try { U = getUnsafe(); Class k = WorkQueue.class; @@ -1050,13 +1033,13 @@ public class ForkJoinPool extends Abstra QLOCK = U.objectFieldOffset (k.getDeclaredField("qlock")); ABASE = U.arrayBaseOffset(ak); - s = U.arrayIndexScale(ak); + int scale = U.arrayIndexScale(ak); + if ((scale & (scale - 1)) != 0) + throw new Error("data type scale not a power of two"); + ASHIFT = 31 - Integer.numberOfLeadingZeros(scale); } catch (Exception e) { throw new Error(e); } - if ((s & (s-1)) != 0) - throw new Error("data type scale not a power of two"); - ASHIFT = 31 - Integer.numberOfLeadingZeros(s); } } @@ -1070,26 +1053,6 @@ public class ForkJoinPool extends Abstra defaultForkJoinWorkerThreadFactory; /** - * Per-thread records for threads that submit to pools. Currently - * holds only pseudo-random seed / index that is used to choose - * submission queues in method externalPush. In the future, this may - * also incorporate a means to implement different task rejection - * and resubmission policies. - * - * Seeds for submitters and workers/workQueues work in basically - * the same way but are initialized and updated using slightly - * different mechanics. Both are initialized using the same - * approach as in class ThreadLocal, where successive values are - * unlikely to collide with previous values. Seeds are then - * randomly modified upon collisions using xorshifts, which - * requires a non-zero seed. - */ - static final class Submitter { - int seed; - Submitter(int s) { seed = s; } - } - - /** * Per-thread submission bookkeeping. Shared across all pools * to reduce ThreadLocal pollution and because random motion * to avoid contention in one pool is likely to hold for others. @@ -1099,23 +1062,23 @@ public class ForkJoinPool extends Abstra static final ThreadLocal submitters; /** + * Permission required for callers of methods that may start or + * kill threads. + */ + private static final RuntimePermission modifyThreadPermission; + + /** * Common (static) pool. Non-null for public use unless a static * construction exception, but internal usages null-check on use * to paranoically avoid potential initialization circularities * as well as to simplify generated code. */ - static final ForkJoinPool commonPool; + static final ForkJoinPool common; /** - * Permission required for callers of methods that may start or - * kill threads. + * Common pool parallelism. Must equal common.parallelism. */ - private static final RuntimePermission modifyThreadPermission; - - /** - * Common pool parallelism. Must equal commonPool.parallelism. - */ - static final int commonPoolParallelism; + static final int commonParallelism; /** * Sequence number for creating workerNamePrefix. @@ -1123,8 +1086,8 @@ public class ForkJoinPool extends Abstra private static int poolNumberSequence; /** - * Return the next sequence number. We don't expect this to - * ever contend so use simple builtin sync. + * 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; @@ -1148,6 +1111,11 @@ public class ForkJoinPool extends Abstra private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L; /** + * Tolerance for idle timeouts, to cope with timer undershoots + */ + private static final long TIMEOUT_SLOP = 2000000L; + + /** * The maximum stolen->joining link depth allowed in method * tryHelpStealer. Must be a power of two. Depths for legitimate * chains are unbounded, but we use a fixed constant to avoid @@ -1259,6 +1227,10 @@ public class ForkJoinPool extends Abstra * declaration order and may differ across JVMs, but the following * empirically works OK on current JVMs. */ + + // Heuristic padding to ameliorate unfortunate memory placements + volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06; + volatile long stealCount; // collects worker counts volatile long ctl; // main pool control volatile int plock; // shutdown status and seqLock @@ -1269,15 +1241,16 @@ public class ForkJoinPool extends Abstra final Thread.UncaughtExceptionHandler ueh; // per-worker UEH final String workerNamePrefix; // to create worker name string - /* + volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17; + volatile Object pad18, pad19, pad1a, pad1b; + + /** * Acquires the plock lock to protect worker array and related * updates. This method is called only if an initial CAS on plock - * fails. This acts as a spinLock for normal cases, but falls back + * fails. This acts as a spinlock for normal cases, but falls back * to builtin monitor to block when (rarely) needed. This would be * a terrible idea for a highly contended lock, but works fine as - * a more conservative alternative to a pure spinlock. See - * internal ConcurrentHashMap documentation for further - * explanation of nearly the same construction. + * a more conservative alternative to a pure spinlock. */ private int acquirePlock() { int spins = PL_SPINS, r = 0, ps, nps; @@ -1329,49 +1302,32 @@ public class ForkJoinPool extends Abstra } /** - * Tries to create and start a worker; adjusts counts etc on failure + * Tries to create and start one worker if fewer than target + * parallelism level exist. Adjusts counts etc on failure. */ - private void addWorker() { - ForkJoinWorkerThread wt = null; - try { - (wt = factory.newThread(this)).start(); - } catch (Throwable ex) { - deregisterWorker(wt, ex); // adjust on failure - } - } - - /** - * Performs secondary initialization, called when plock is zero. - * Creates workQueue array and sets plock to a valid value. The - * lock body must be exception-free (so no try/finally) so we - * optimistically allocate new array outside the lock and throw - * away if (very rarely) not needed. (A similar tactic is used in - * fullExternalPush.) Because the plock seq value can eventually - * wrap around zero, this method harmlessly fails to reinitialize - * if workQueues exists, while still advancing plock. - */ - private void initWorkQueuesArray() { - WorkQueue[] ws; int ps; - int p = config & SMASK; // find power of two table size - int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots - n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16; - WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; - if (((ps = plock) & PL_LOCK) != 0 || - !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) - ps = acquirePlock(); - if ((ws = workQueues) == null || ws.length == 0) - workQueues = nws; - int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); - if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) - releasePlock(nps); + private void tryAddWorker() { long c; int u; - if ((u = (int)((c = ctl) >>> 32)) < 0 && (int)c == 0) { + while ((u = (int)((c = ctl) >>> 32)) < 0 && + (u & SHORT_SIGN) != 0 && (int)c == 0) { long nc = (long)(((u + UTC_UNIT) & UTC_MASK) | ((u + UAC_UNIT) & UAC_MASK)) << 32; - if (U.compareAndSwapLong(this, CTL, c, nc)) - addWorker(); + if (U.compareAndSwapLong(this, CTL, c, nc)) { + ForkJoinWorkerThreadFactory fac; + Throwable ex = null; + ForkJoinWorkerThread wt = null; + try { + if ((fac = factory) != null && + (wt = fac.newThread(this)) != null) { + wt.start(); + break; + } + } catch (Throwable e) { + ex = e; + } + deregisterWorker(wt, ex); + break; + } } - } // Registering and deregistering workers @@ -1384,42 +1340,45 @@ public class ForkJoinPool extends Abstra * expanding as needed. * * @param wt the worker thread + * @return the worker's queue */ - final void registerWorker(ForkJoinWorkerThread wt) { - if (wt != null && wt.workQueue == null) { - int s, ps; // generate a rarely colliding candidate index seed - do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed, - s += SEED_INCREMENT) || - s == 0); // skip 0 - WorkQueue w = new WorkQueue(this, wt, config >>> 16, s); - if (((ps = plock) & PL_LOCK) != 0 || - !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) - ps = acquirePlock(); - int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); - try { - WorkQueue[] ws; - if ((ws = workQueues) != null && wt.workQueue == null) { - int n = ws.length, m = n - 1; - int r = (s << 1) | 1; // use odd-numbered indices - if (ws[r &= m] != null) { // collision - int probes = 0; // step by approx half size - int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2; - while (ws[r = (r + step) & m] != null) { - if (++probes >= n) { - workQueues = ws = Arrays.copyOf(ws, n <<= 1); - m = n - 1; - probes = 0; - } + final WorkQueue registerWorker(ForkJoinWorkerThread wt) { + Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps; + wt.setDaemon(true); + if ((handler = ueh) != null) + wt.setUncaughtExceptionHandler(handler); + do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed, + s += SEED_INCREMENT) || + s == 0); // skip 0 + WorkQueue w = new WorkQueue(this, wt, config >>> 16, s); + if (((ps = plock) & PL_LOCK) != 0 || + !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) + ps = acquirePlock(); + int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); + try { + if ((ws = workQueues) != null) { // skip if shutting down + int n = ws.length, m = n - 1; + int r = (s << 1) | 1; // use odd-numbered indices + if (ws[r &= m] != null) { // collision + int probes = 0; // step by approx half size + int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2; + while (ws[r = (r + step) & m] != null) { + if (++probes >= n) { + workQueues = ws = Arrays.copyOf(ws, n <<= 1); + m = n - 1; + probes = 0; } } - w.eventCount = w.poolIndex = r; // volatile write orders - wt.workQueue = ws[r] = w; } - } finally { - if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) - releasePlock(nps); + w.eventCount = w.poolIndex = r; // volatile write orders + ws[r] = w; } + } finally { + if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) + releasePlock(nps); } + wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex))); + return w; } /** @@ -1454,21 +1413,42 @@ public class ForkJoinPool extends Abstra } } - long c; // adjust ctl counts + long c; // adjust ctl counts do {} while (!U.compareAndSwapLong (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) | ((c - TC_UNIT) & TC_MASK) | (c & ~(AC_MASK|TC_MASK))))); - if (!tryTerminate(false, false) && w != null) { - w.cancelAll(); // cancel remaining tasks - if (w.array != null) // suppress signal if never ran - helpSignal(null, 0); // wake up or create replacement - if (ex == null) // help clean refs on way out - ForkJoinTask.helpExpungeStaleExceptions(); + if (!tryTerminate(false, false) && w != null && w.array != null) { + w.cancelAll(); // cancel remaining tasks + WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e; + while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) { + if (e > 0) { // activate or create replacement + if ((ws = workQueues) == null || + (i = e & SMASK) >= ws.length || + (v = ws[i]) == null) + break; + long nc = (((long)(v.nextWait & E_MASK)) | + ((long)(u + UAC_UNIT) << 32)); + if (v.eventCount != (e | INT_SIGN)) + break; + if (U.compareAndSwapLong(this, CTL, c, nc)) { + v.eventCount = (e + E_SEQ) & E_MASK; + if ((p = v.parker) != null) + U.unpark(p); + break; + } + } + else { + if ((short)u < 0) + tryAddWorker(); + break; + } + } } - - if (ex != null) // rethrow + if (ex == null) // help clean refs on way out + ForkJoinTask.helpExpungeStaleExceptions(); + else // rethrow ForkJoinTask.rethrow(ex); } @@ -1490,11 +1470,12 @@ public class ForkJoinPool extends Abstra U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock int b = q.base, s = q.top, n, an; if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) { - U.putObject(a, (long)(((an - 1) & s) << ASHIFT) + ABASE, task); + int j = (((an - 1) & s) << ASHIFT) + ABASE; + U.putOrderedObject(a, j, task); q.top = s + 1; // push on to deque q.qlock = 0; if (n <= 2) - signalWork(q, 0); + signalWork(q); return; } q.qlock = 0; @@ -1505,16 +1486,22 @@ public class ForkJoinPool extends Abstra /** * Full version of externalPush. This method is called, among * other times, upon the first submission of the first task to the - * pool, so must perform secondary initialization (via - * initWorkQueuesArray). It also detects first submission by an - * external thread by looking up its ThreadLocal, and creates a - * new shared queue if the one at index if empty or contended. The - * lock body must be exception-free (so no try/finally) so we - * optimistically allocate new queues outside the lock and throw - * them away if (very rarely) not needed. + * pool, so must perform secondary initialization. It also + * detects first submission by an external thread by looking up + * its ThreadLocal, and creates a new shared queue if the one at + * index if empty or contended. The plock lock body must be + * exception-free (so no try/finally) so we optimistically + * allocate new queues outside the lock and throw them away if + * (very rarely) not needed. + * + * Secondary initialization occurs when plock is zero, to create + * workQueue array and set plock to a valid value. This lock body + * must also be exception-free. Because the plock seq value can + * eventually wrap around zero, this method harmlessly fails to + * reinitialize if workQueues exists, while still advancing plock. */ private void fullExternalPush(ForkJoinTask task) { - int r = 0; + int r = 0; // random index seed for (Submitter z = submitters.get();;) { WorkQueue[] ws; WorkQueue q; int ps, m, k; if (z == null) { @@ -1522,22 +1509,53 @@ public class ForkJoinPool extends Abstra r += SEED_INCREMENT) && r != 0) submitters.set(z = new Submitter(r)); } - else if (r == 0) { // move to a different index + else if (r == 0) { // move to a different index r = z.seed; - r ^= r << 13; // same xorshift as WorkQueues + r ^= r << 13; // same xorshift as WorkQueues r ^= r >>> 17; z.seed = r ^ (r << 5); } else if ((ps = plock) < 0) throw new RejectedExecutionException(); else if (ps == 0 || (ws = workQueues) == null || - (m = ws.length - 1) < 0) - initWorkQueuesArray(); + (m = ws.length - 1) < 0) { // initialize workQueues + int p = config & SMASK; // find power of two table size + int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots + n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; + n |= n >>> 8; n |= n >>> 16; n = (n + 1) << 1; + WorkQueue[] nws = ((ws = workQueues) == null || ws.length == 0 ? + new WorkQueue[n] : null); + if (((ps = plock) & PL_LOCK) != 0 || + !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) + ps = acquirePlock(); + if (((ws = workQueues) == null || ws.length == 0) && nws != null) + workQueues = nws; + int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); + if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) + releasePlock(nps); + } else if ((q = ws[k = r & m & SQMASK]) != null) { - if (q.trySharedPush(task)) - return; - else - r = 0; // move on contention + if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) { + ForkJoinTask[] a = q.array; + int s = q.top; + boolean submitted = false; + try { // locked version of push + if ((a != null && a.length > s + 1 - q.base) || + (a = q.growArray()) != null) { // must presize + int j = (((a.length - 1) & s) << ASHIFT) + ABASE; + U.putOrderedObject(a, j, task); + q.top = s + 1; + submitted = true; + } + } finally { + q.qlock = 0; // unlock + } + if (submitted) { + signalWork(q); + return; + } + } + r = 0; // move on failure } else if (((ps = plock) & PL_LOCK) == 0) { // create new queue q = new WorkQueue(this, null, SHARED_QUEUE, r); @@ -1566,17 +1584,13 @@ public class ForkJoinPool extends Abstra } /** - * Tries to create (at most one) or activate (possibly several) - * workers if too few are active. On contention failure, continues - * until at least one worker is signalled or the given queue is - * empty or all workers are active. - * - * @param q if non-null, the queue holding tasks to be signalled - * @param signals the target number of signals (at least one -- - * if argument is zero also sets signallee hint if parked). + * Tries to create or activate a worker if too few are active. + * + * @param q the (non-null) queue holding tasks to be signalled */ - final void signalWork(WorkQueue q, int signals) { - long c; int e, u, i, s; WorkQueue[] ws; WorkQueue w; Thread p; + final void signalWork(WorkQueue q) { + int hint = q.poolIndex; + long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p; while ((u = (int)((c = ctl) >>> 32)) < 0) { if ((e = (int)c) > 0) { if ((ws = workQueues) != null && ws.length > (i = e & SMASK) && @@ -1584,32 +1598,23 @@ public class ForkJoinPool extends Abstra long nc = (((long)(w.nextWait & E_MASK)) | ((long)(u + UAC_UNIT) << 32)); if (U.compareAndSwapLong(this, CTL, c, nc)) { + w.hint = hint; w.eventCount = (e + E_SEQ) & E_MASK; - if ((p = w.parker) != null) { - if (q != null && signals == 0) - w.hint = q.poolIndex; + if ((p = w.parker) != null) U.unpark(p); - } - if (--signals <= 0) - break; + break; } - if (q != null && (s = q.queueSize()) <= signals && - (signals = s) <= 0) + if (q.top - q.base <= 0) break; } else break; } - else if (e == 0 && (u & SHORT_SIGN) != 0) { - long nc = (long)(((u + UTC_UNIT) & UTC_MASK) | - ((u + UAC_UNIT) & UAC_MASK)) << 32; - if (U.compareAndSwapLong(this, CTL, c, nc)) { - addWorker(); - break; - } - } - else + else { + if ((short)u < 0) + tryAddWorker(); break; + } } } @@ -1619,8 +1624,8 @@ public class ForkJoinPool extends Abstra * Top-level runloop for workers, called by ForkJoinWorkerThread.run. */ final void runWorker(WorkQueue w) { - if (w != null) // skip on initialization failure - do { w.runTask(scan(w)); } while (w.qlock >= 0); + w.growArray(); // allocate queue + do { w.runTask(scan(w)); } while (w.qlock >= 0); } /** @@ -1648,23 +1653,27 @@ public class ForkJoinPool extends Abstra * * * If not already enqueued, try to inactivate and enqueue the * worker on wait queue. Or, if inactivating has caused the pool - * to be quiescent, relay to idleAwaitWork to check for - * termination and possibly shrink pool. + * to be quiescent, relay to idleAwaitWork to possibly shrink + * pool. * * * If already enqueued and none of the above apply, possibly - * (with 1/2 probability) park awaiting signal, else lingering to - * help scan and signal. + * park awaiting signal, else lingering to help scan and signal. + * + * * If a non-empty queue discovered or left as a hint, + * help wake up other workers before return. * * @param w the worker (via its WorkQueue) * @return a task or null if none found */ private final ForkJoinTask scan(WorkQueue w) { - WorkQueue[] ws; int m, hint; + WorkQueue[] ws; int m; int ps = plock; // read plock before ws if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) { int ec = w.eventCount; // ec is negative if inactive int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5; - for (int j = ((m + m + 1) | MIN_SCAN) & MAX_SCAN; ; --j) { + w.hint = -1; // update seed and clear hint + int j = ((m + m + 1) | MIN_SCAN) & MAX_SCAN; + do { WorkQueue q; ForkJoinTask[] a; int b; if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 && (a = q.array) != null) { // probably nonempty @@ -1674,56 +1683,74 @@ public class ForkJoinPool extends Abstra if (q.base == b && ec >= 0 && t != null && U.compareAndSwapObject(a, i, t, null)) { if ((q.base = b + 1) - q.top < 0) - signalWork(q, 0); + signalWork(q); return t; // taken } - else if (ec < 0 || j < m) { // cannot take or cannot rescan - w.hint = q.poolIndex; // use hint below - break; // let caller retry after signal - } - } - else if (j < 0) { // end of scan; in loop to simplify code - long c, sc; int e, ns; - if ((ns = w.nsteals) != 0) { - if (U.compareAndSwapLong(this, STEALCOUNT, - sc = stealCount, sc + ns)) - w.nsteals = 0; // collect steals - } - else if (plock != ps) // ws may have changed - break; - else if ((e = (int)(c = ctl)) < 0) - w.qlock = -1; // pool is terminating - else if (ec >= 0) { // try to enqueue/inactivate - long nc = ((long)ec | - ((c - AC_UNIT) & (AC_MASK|TC_MASK))); + else if ((ec < 0 || j < m) && (int)(ctl >> AC_SHIFT) <= 0) { + w.hint = (r + j) & m; // help signal below + break; // cannot take + } + } + } while (--j >= 0); + + int h, e, ns; long c, sc; WorkQueue q; + if ((ns = w.nsteals) != 0) { + if (U.compareAndSwapLong(this, STEALCOUNT, + sc = stealCount, sc + ns)) + w.nsteals = 0; // collect steals and rescan + } + else if (plock != ps) // consistency check + ; // skip + else if ((e = (int)(c = ctl)) < 0) + w.qlock = -1; // pool is terminating + else { + if ((h = w.hint) < 0) { + if (ec >= 0) { // try to enqueue/inactivate + long nc = (((long)ec | + ((c - AC_UNIT) & (AC_MASK|TC_MASK)))); w.nextWait = e; // link and mark inactive - w.hint = -1; // use hint if set while parked w.eventCount = ec | INT_SIGN; - if (ctl != c || - !U.compareAndSwapLong(this, CTL, c, nc)) - w.eventCount = ec; // unmark on CAS failure + if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc)) + w.eventCount = ec; // unmark on CAS failure else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK)) idleAwaitWork(w, nc, c); } - else if (w.eventCount < 0) { // block + else if (w.eventCount < 0 && ctl == c) { Thread wt = Thread.currentThread(); Thread.interrupted(); // clear status U.putObject(wt, PARKBLOCKER, this); w.parker = wt; // emulate LockSupport.park if (w.eventCount < 0) // recheck - U.park(false, 0L); + U.park(false, 0L); // block w.parker = null; U.putObject(wt, PARKBLOCKER, null); } - break; } - } - if ((hint = w.hint) >= 0) { // help signal - WorkQueue[] vs; WorkQueue v; int k; - w.hint = -1; // suppress resignal - if ((vs = workQueues) != null && hint < vs.length && - (v = vs[hint]) != null && (k = v.base - v.top) < -1) - signalWork(v, 1 - k); + if ((h >= 0 || (h = w.hint) >= 0) && + (ws = workQueues) != null && h < ws.length && + (q = ws[h]) != null) { // signal others before retry + WorkQueue v; Thread p; int u, i, s; + for (int n = (config & SMASK) - 1;;) { + int idleCount = (w.eventCount < 0) ? 0 : -1; + if (((s = idleCount - q.base + q.top) <= n && + (n = s) <= 0) || + (u = (int)((c = ctl) >>> 32)) >= 0 || + (e = (int)c) <= 0 || m < (i = e & SMASK) || + (v = ws[i]) == null) + break; + long nc = (((long)(v.nextWait & E_MASK)) | + ((long)(u + UAC_UNIT) << 32)); + if (v.eventCount != (e | INT_SIGN) || + !U.compareAndSwapLong(this, CTL, c, nc)) + break; + v.hint = h; + v.eventCount = (e + E_SEQ) & E_MASK; + if ((p = v.parker) != null) + U.unpark(p); + if (--n <= 0) + break; + } + } } } return null; @@ -1743,10 +1770,11 @@ public class ForkJoinPool extends Abstra */ private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) { if (w != null && w.eventCount < 0 && - !tryTerminate(false, false) && (int)prevCtl != 0) { + !tryTerminate(false, false) && (int)prevCtl != 0 && + ctl == currentCtl) { int dc = -(short)(currentCtl >>> TC_SHIFT); long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT; - long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop + long deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP; Thread wt = Thread.currentThread(); while (ctl == currentCtl) { Thread.interrupted(); // timed variant of version in scan() @@ -1761,8 +1789,8 @@ public class ForkJoinPool extends Abstra if (deadline - System.nanoTime() <= 0L && U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) { w.eventCount = (w.eventCount + E_SEQ) | E_MASK; + w.hint = -1; w.qlock = -1; // shrink - w.hint = -1; // suppress helping break; } } @@ -1770,28 +1798,44 @@ public class ForkJoinPool extends Abstra } /** - * Scans through queues looking for work (optionally, while - * joining a task); if any are present, signals. May return early - * if more signalling is detectably unneeded. + * Scans through queues looking for work while joining a task; if + * any present, signals. May return early if more signalling is + * detectably unneeded. * - * @param task if non-null, return early if done + * @param task return early if done * @param origin an index to start scan */ - final int helpSignal(ForkJoinTask task, int origin) { - WorkQueue[] ws; WorkQueue q; int m, n, s, u; - if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) { - for (int i = 0; i <= m; ++i) { - if (task != null && (s = task.status) < 0) - return s; - if ((q = ws[(i + origin) & m]) != null && - (n = q.queueSize()) > 0) { - signalWork(q, n); - if ((u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0) + private void helpSignal(ForkJoinTask task, int origin) { + WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s; + if (task != null && task.status >= 0 && + (u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 && + (ws = workQueues) != null && (m = ws.length - 1) >= 0) { + outer: for (int k = origin, j = m; j >= 0; --j) { + WorkQueue q = ws[k++ & m]; + for (int n = m;;) { // limit to at most m signals + if (task.status < 0) + break outer; + if (q == null || + ((s = -q.base + q.top) <= n && (n = s) <= 0)) break; + if ((u = (int)((c = ctl) >>> 32)) >= 0 || + (e = (int)c) <= 0 || m < (i = e & SMASK) || + (w = ws[i]) == null) + break outer; + long nc = (((long)(w.nextWait & E_MASK)) | + ((long)(u + UAC_UNIT) << 32)); + if (w.eventCount != (e | INT_SIGN)) + break outer; + if (U.compareAndSwapLong(this, CTL, c, nc)) { + w.eventCount = (e + E_SEQ) & E_MASK; + if ((p = w.parker) != null) + U.unpark(p); + if (--n <= 0) + break; + } } } } - return 0; } /** @@ -1887,7 +1931,6 @@ public class ForkJoinPool extends Abstra * @param task the task to join * @param mode if shared, exit upon completing any task * if all workers are active - * */ private int helpComplete(ForkJoinTask task, int mode) { WorkQueue[] ws; WorkQueue q; int m, n, s, u; @@ -1940,8 +1983,19 @@ public class ForkJoinPool extends Abstra else if (tc + pc < MAX_CAP) { long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); if (U.compareAndSwapLong(this, CTL, c, nc)) { - addWorker(); - return true; + ForkJoinWorkerThreadFactory fac; + Throwable ex = null; + ForkJoinWorkerThread wt = null; + try { + if ((fac = factory) != null && + (wt = fac.newThread(this)) != null) { + wt.start(); + return true; + } + } catch (Throwable rex) { + ex = rex; + } + deregisterWorker(wt, ex); // clean up and return false } } } @@ -1960,27 +2014,20 @@ public class ForkJoinPool extends Abstra if (joiner != null && task != null && (s = task.status) >= 0) { ForkJoinTask prevJoin = joiner.currentJoin; joiner.currentJoin = task; - do {} while ((s = task.status) >= 0 && - joiner.queueSize() > 0 && + do {} while ((s = task.status) >= 0 && !joiner.isEmpty() && joiner.tryRemoveAndExec(task)); // process local tasks - if (s >= 0 && (s = task.status) >= 0 && - (s = helpSignal(task, joiner.poolIndex)) >= 0 && - (task instanceof CountedCompleter)) - s = helpComplete(task, LIFO_QUEUE); - int k = 0; // to perform pre-block yield for politeness + if (s >= 0 && (s = task.status) >= 0) { + helpSignal(task, joiner.poolIndex); + if ((s = task.status) >= 0 && + (task instanceof CountedCompleter)) + s = helpComplete(task, LIFO_QUEUE); + } while (s >= 0 && (s = task.status) >= 0) { - if ((joiner.queueSize() > 0 || // try helping + if ((!joiner.isEmpty() || // try helping (s = tryHelpStealer(joiner, task)) == 0) && (s = task.status) >= 0) { - if (k < 3) { - if (++k < 3) - s = helpSignal(task, joiner.poolIndex); - else - Thread.yield(); - } - else if (!tryCompensate()) - k = 0; - else { + helpSignal(task, joiner.poolIndex); + if ((s = task.status) >= 0 && tryCompensate()) { if (task.trySetSignal() && (s = task.status) >= 0) { synchronized (task) { if (task.status >= 0) { @@ -2017,14 +2064,15 @@ public class ForkJoinPool extends Abstra if (joiner != null && task != null && (s = task.status) >= 0) { ForkJoinTask prevJoin = joiner.currentJoin; joiner.currentJoin = task; - do {} while ((s = task.status) >= 0 && - joiner.queueSize() > 0 && + do {} while ((s = task.status) >= 0 && !joiner.isEmpty() && joiner.tryRemoveAndExec(task)); - if (s >= 0 && (s = task.status) >= 0 && - (s = helpSignal(task, joiner.poolIndex)) >= 0 && - (task instanceof CountedCompleter)) - s = helpComplete(task, LIFO_QUEUE); - if (s >= 0 && joiner.queueSize() == 0) { + if (s >= 0 && (s = task.status) >= 0) { + helpSignal(task, joiner.poolIndex); + if ((s = task.status) >= 0 && + (task instanceof CountedCompleter)) + s = helpComplete(task, LIFO_QUEUE); + } + if (s >= 0 && joiner.isEmpty()) { do {} while (task.status >= 0 && tryHelpStealer(joiner, task) > 0); } @@ -2034,29 +2082,22 @@ public class ForkJoinPool extends Abstra /** * Returns a (probably) non-empty steal queue, if one is found - * during a random, then cyclic scan, else null. This method must - * be retried by caller if, by the time it tries to use the queue, - * it is empty. + * during a scan, else null. This method must be retried by + * caller if, by the time it tries to use the queue, it is empty. * @param r a (random) seed for scanning */ private WorkQueue findNonEmptyStealQueue(int r) { - for (WorkQueue[] ws;;) { - int ps = plock, m, n; - if ((ws = workQueues) == null || (m = ws.length - 1) < 1) - return null; - for (int j = (m + 1) << 2; ;) { - WorkQueue q = ws[(((r + j) << 1) | 1) & m]; - if (q != null && (n = q.queueSize()) > 0) { - if (n > 1) - signalWork(q, 0); - return q; - } - else if (--j < 0) { - if (plock == ps) - return null; - break; + for (;;) { + int ps = plock, m; WorkQueue[] ws; WorkQueue q; + if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) { + for (int j = (m + 1) << 2; j >= 0; --j) { + if ((q = ws[(((r + j) << 1) | 1) & m]) != null && + q.base - q.top < 0) + return q; } } + if (plock == ps) + return null; } } @@ -2068,37 +2109,34 @@ public class ForkJoinPool extends Abstra */ final void helpQuiescePool(WorkQueue w) { for (boolean active = true;;) { - ForkJoinTask localTask; // exhaust local queue - while ((localTask = w.nextLocalTask()) != null) - localTask.doExec(); - // Similar to loop in scan(), but ignoring submissions - WorkQueue q = findNonEmptyStealQueue(w.nextSeed()); - if (q != null) { - ForkJoinTask t; int b; + long c; WorkQueue q; ForkJoinTask t; int b; + while ((t = w.nextLocalTask()) != null) { + if (w.base - w.top < 0) + signalWork(w); + t.doExec(); + } + if ((q = findNonEmptyStealQueue(w.nextSeed())) != null) { if (!active) { // re-establish active count - long c; active = true; do {} while (!U.compareAndSwapLong (this, CTL, c = ctl, c + AC_UNIT)); } - if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) + if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) { + if (q.base - q.top < 0) + signalWork(q); w.runSubtask(t); + } } - else { - long c; - if (active) { // decrement active count without queuing + else if (active) { // decrement active count without queuing + long nc = (c = ctl) - AC_UNIT; + if ((int)(nc >> AC_SHIFT) + (config & SMASK) == 0) + return; // bypass decrement-then-increment + if (U.compareAndSwapLong(this, CTL, c, nc)) active = false; - do {} while (!U.compareAndSwapLong - (this, CTL, c = ctl, c -= AC_UNIT)); - } - else - c = ctl; // re-increment on exit - if ((int)(c >> AC_SHIFT) + (config & SMASK) == 0) { - do {} while (!U.compareAndSwapLong - (this, CTL, c = ctl, c + AC_UNIT)); - break; - } } + else if ((int)((c = ctl) >> AC_SHIFT) + (config & SMASK) == 0 && + U.compareAndSwapLong(this, CTL, c, c + AC_UNIT)) + return; } } @@ -2114,8 +2152,11 @@ public class ForkJoinPool extends Abstra return t; if ((q = findNonEmptyStealQueue(w.nextSeed())) == null) return null; - if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) + if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) { + if (q.base - q.top < 0) + signalWork(q); return t; + } } } @@ -2197,61 +2238,71 @@ public class ForkJoinPool extends Abstra * @return true if now terminating or terminated */ private boolean tryTerminate(boolean now, boolean enable) { - if (this == commonPool) // cannot shut down + int ps; + if (this == common) // cannot shut down return false; + if ((ps = plock) >= 0) { // enable by setting plock + if (!enable) + return false; + if ((ps & PL_LOCK) != 0 || + !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) + ps = acquirePlock(); + int nps = ((ps + PL_LOCK) & ~SHUTDOWN) | SHUTDOWN; + if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) + releasePlock(nps); + } for (long c;;) { - if (((c = ctl) & STOP_BIT) != 0) { // already terminating + if (((c = ctl) & STOP_BIT) != 0) { // already terminating if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) { synchronized (this) { - notifyAll(); // signal when 0 workers + notifyAll(); // signal when 0 workers } } return true; } - if (plock >= 0) { // not yet enabled - int ps; - if (!enable) - return false; - if (((ps = plock) & PL_LOCK) != 0 || - !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) - ps = acquirePlock(); - int nps = SHUTDOWN; - if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) - releasePlock(nps); - } - if (!now) { // check if idle & no tasks - if ((int)(c >> AC_SHIFT) != -(config & SMASK) || - hasQueuedSubmissions()) + if (!now) { // check if idle & no tasks + WorkQueue[] ws; WorkQueue w; + if ((int)(c >> AC_SHIFT) != -(config & SMASK)) return false; - // Check for unqueued inactive workers. One pass suffices. - WorkQueue[] ws = workQueues; WorkQueue w; - if (ws != null) { - for (int i = 1; i < ws.length; i += 2) { - if ((w = ws[i]) != null && w.eventCount >= 0) - return false; + if ((ws = workQueues) != null) { + for (int i = 0; i < ws.length; ++i) { + if ((w = ws[i]) != null) { + if (!w.isEmpty()) { // signal unprocessed tasks + signalWork(w); + return false; + } + if ((i & 1) != 0 && w.eventCount >= 0) + return false; // unqueued inactive worker + } } } } if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) { for (int pass = 0; pass < 3; ++pass) { - WorkQueue[] ws = workQueues; - if (ws != null) { - WorkQueue w; + WorkQueue[] ws; WorkQueue w; Thread wt; + if ((ws = workQueues) != null) { int n = ws.length; for (int i = 0; i < n; ++i) { if ((w = ws[i]) != null) { w.qlock = -1; if (pass > 0) { w.cancelAll(); - if (pass > 1) - w.interruptOwner(); + if (pass > 1 && (wt = w.owner) != null) { + if (!wt.isInterrupted()) { + try { + wt.interrupt(); + } catch (Throwable ignore) { + } + } + U.unpark(wt); + } } } } // Wake up workers parked on event queue int i, e; long cc; Thread p; while ((e = (int)(cc = ctl) & E_MASK) != 0 && - (i = e & SMASK) < n && + (i = e & SMASK) < n && i >= 0 && (w = ws[i]) != null) { long nc = ((long)(w.nextWait & E_MASK) | ((cc + AC_UNIT) & AC_MASK) | @@ -2279,7 +2330,7 @@ public class ForkJoinPool extends Abstra static WorkQueue commonSubmitterQueue() { ForkJoinPool p; WorkQueue[] ws; int m; Submitter z; return ((z = submitters.get()) != null && - (p = commonPool) != null && + (p = common) != null && (ws = p.workQueues) != null && (m = ws.length - 1) >= 0) ? ws[m & z.seed & SQMASK] : null; @@ -2290,24 +2341,26 @@ public class ForkJoinPool extends Abstra */ static boolean tryExternalUnpush(ForkJoinTask t) { ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z; - ForkJoinTask[] a; int m, s; long j; - if ((z = submitters.get()) != null && - (p = commonPool) != null && + ForkJoinTask[] a; int m, s; + if (t != null && + (z = submitters.get()) != null && + (p = common) != null && (ws = p.workQueues) != null && (m = ws.length - 1) >= 0 && (q = ws[m & z.seed & SQMASK]) != null && (s = q.top) != q.base && - (a = q.array) != null && - U.getObjectVolatile - (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t && - U.compareAndSwapInt(q, QLOCK, 0, 1)) { - if (q.array == a && q.top == s && // recheck - U.compareAndSwapObject(a, j, t, null)) { - q.top = s - 1; + (a = q.array) != null) { + long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE; + if (U.getObject(a, j) == t && + U.compareAndSwapInt(q, QLOCK, 0, 1)) { + if (q.array == a && q.top == s && // recheck + U.compareAndSwapObject(a, j, t, null)) { + q.top = s - 1; + q.qlock = 0; + return true; + } q.qlock = 0; - return true; } - q.qlock = 0; } return false; } @@ -2349,7 +2402,8 @@ public class ForkJoinPool extends Abstra (u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0) break; if (task == null) { - if (helpSignal(root, q.poolIndex) >= 0) + helpSignal(root, q.poolIndex); + if (root.status >= 0) helpComplete(root, SHARED_QUEUE); break; } @@ -2364,27 +2418,28 @@ public class ForkJoinPool extends Abstra static void externalHelpJoin(ForkJoinTask t) { // Some hard-to-avoid overlap with tryExternalUnpush ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z; - ForkJoinTask[] a; int m, s, n; long j; + ForkJoinTask[] a; int m, s, n; if (t != null && (z = submitters.get()) != null && - (p = commonPool) != null && + (p = common) != null && (ws = p.workQueues) != null && (m = ws.length - 1) >= 0 && (q = ws[m & z.seed & SQMASK]) != null && - (a = q.array) != null && - t.status >= 0) { - if ((s = q.top) != q.base && - U.getObjectVolatile - (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t && - U.compareAndSwapInt(q, QLOCK, 0, 1)) { - if (q.array == a && q.top == s && - U.compareAndSwapObject(a, j, t, null)) { - q.top = s - 1; - q.qlock = 0; - t.doExec(); + (a = q.array) != null) { + int am = a.length - 1; + if ((s = q.top) != q.base) { + long j = ((am & (s - 1)) << ASHIFT) + ABASE; + if (U.getObject(a, j) == t && + U.compareAndSwapInt(q, QLOCK, 0, 1)) { + if (q.array == a && q.top == s && + U.compareAndSwapObject(a, j, t, null)) { + q.top = s - 1; + q.qlock = 0; + t.doExec(); + } + else + q.qlock = 0; } - else - q.qlock = 0; } if (t.status >= 0) { if (t instanceof CountedCompleter) @@ -2395,18 +2450,6 @@ public class ForkJoinPool extends Abstra } } - /** - * Restricted version of helpQuiescePool for external callers - */ - static void externalHelpQuiescePool() { - ForkJoinPool p; ForkJoinTask t; WorkQueue q; int b; - if ((p = commonPool) != null && - (q = p.findNonEmptyStealQueue(1)) != null && - (b = q.base) - q.top < 0 && - (t = q.pollAt(b)) != null) - t.doExec(); - } - // Exported methods // Constructors @@ -2480,7 +2523,7 @@ public class ForkJoinPool extends Abstra throw new IllegalArgumentException(); this.factory = factory; this.ueh = handler; - this.config = parallelism | (asyncMode? (FIFO_QUEUE << 16) : 0); + this.config = parallelism | (asyncMode ? (FIFO_QUEUE << 16) : 0); long np = (long)(-parallelism); // offset ctl counts this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK); int pn = nextPoolId(); @@ -2505,13 +2548,21 @@ public class ForkJoinPool extends Abstra } /** - * Returns the common pool instance. + * Returns the common pool instance. This pool is statically + * constructed; its run state is unaffected by attempts to {@link + * #shutdown} or {@link #shutdownNow}. However this pool and any + * ongoing processing are automatically terminated upon program + * {@link System#exit}. Any program that relies on asynchronous + * task processing to complete before program termination should + * invoke {@code commonPool().}{@link #awaitQuiescence}, before + * exit. * * @return the common pool instance + * @since 1.8 */ public static ForkJoinPool commonPool() { - // assert commonPool != null : "static init error"; - return commonPool; + // assert common != null : "static init error"; + return common; } // Execution methods @@ -2634,27 +2685,23 @@ public class ForkJoinPool extends Abstra // In previous versions of this class, this method constructed // a task to run ForkJoinTask.invokeAll, but now external // invocation of multiple tasks is at least as efficient. - List> fs = new ArrayList>(tasks.size()); - // Workaround needed because method wasn't declared with - // wildcards in return type but should have been. - @SuppressWarnings({"unchecked", "rawtypes"}) - List> futures = (List>) (List) fs; + ArrayList> futures = new ArrayList>(tasks.size()); boolean done = false; try { for (Callable t : tasks) { ForkJoinTask f = new ForkJoinTask.AdaptedCallable(t); + futures.add(f); externalPush(f); - fs.add(f); } - for (ForkJoinTask f : fs) - f.quietlyJoin(); + for (int i = 0, size = futures.size(); i < size; i++) + ((ForkJoinTask)futures.get(i)).quietlyJoin(); done = true; return futures; } finally { if (!done) - for (ForkJoinTask f : fs) - f.cancel(false); + for (int i = 0, size = futures.size(); i < size; i++) + futures.get(i).cancel(false); } } @@ -2690,9 +2737,10 @@ public class ForkJoinPool extends Abstra * Returns the targeted parallelism level of the common pool. * * @return the targeted parallelism level of the common pool + * @since 1.8 */ public static int getCommonPoolParallelism() { - return commonPoolParallelism; + return commonParallelism; } /** @@ -2838,7 +2886,7 @@ public class ForkJoinPool extends Abstra WorkQueue[] ws; WorkQueue w; if ((ws = workQueues) != null) { for (int i = 0; i < ws.length; i += 2) { - if ((w = ws[i]) != null && w.queueSize() != 0) + if ((w = ws[i]) != null && !w.isEmpty()) return true; } } @@ -2950,7 +2998,7 @@ public class ForkJoinPool extends Abstra * Possibly initiates an orderly shutdown in which previously * submitted tasks are executed, but no new tasks will be * accepted. Invocation has no effect on execution state if this - * is the {@link #commonPool}, and no additional effect if + * is the {@link #commonPool()}, and no additional effect if * already shut down. Tasks that are in the process of being * submitted concurrently during the course of this method may or * may not be rejected. @@ -2968,7 +3016,7 @@ public class ForkJoinPool extends Abstra /** * Possibly attempts to cancel and/or stop all tasks, and reject * all subsequently submitted tasks. Invocation has no effect on - * execution state if this is the {@link #commonPool}, and no + * execution state if this is the {@link #commonPool()}, and no * additional effect if already shut down. Otherwise, tasks that * are in the process of being submitted or executed concurrently * during the course of this method may or may not be @@ -3005,7 +3053,7 @@ public class ForkJoinPool extends Abstra * commenced but not yet completed. This method may be useful for * debugging. A return of {@code true} reported a sufficient * period after shutdown may indicate that submitted tasks have - * ignored or suppressed interruption, or are waiting for IO, + * ignored or suppressed interruption, or are waiting for I/O, * causing this executor not to properly terminate. (See the * advisory notes for class {@link ForkJoinTask} stating that * tasks should not normally entail blocking operations. But if @@ -3031,9 +3079,10 @@ public class ForkJoinPool extends Abstra /** * Blocks until all tasks have completed execution after a * shutdown request, or the timeout occurs, or the current thread - * is interrupted, whichever happens first. Note that the {@link - * #commonPool()} never terminates until program shutdown so - * this method will always time out. + * is interrupted, whichever happens first. Because the {@link + * #commonPool()} never terminates until program shutdown, when + * applied to the common pool, this method is equivalent to {@link + * #awaitQuiescence} but always returns {@code false}. * * @param timeout the maximum time to wait * @param unit the time unit of the timeout argument @@ -3043,6 +3092,12 @@ public class ForkJoinPool extends Abstra */ public boolean awaitTermination(long timeout, TimeUnit unit) throws InterruptedException { + if (Thread.interrupted()) + throw new InterruptedException(); + if (this == common) { + awaitQuiescence(timeout, unit); + return false; + } long nanos = unit.toNanos(timeout); if (isTerminated()) return true; @@ -3062,6 +3117,62 @@ public class ForkJoinPool extends Abstra } /** + * If called by a ForkJoinTask operating in this pool, equivalent + * in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise, + * waits and/or attempts to assist performing tasks until this + * pool {@link #isQuiescent} or the indicated timeout elapses. + * + * @param timeout the maximum time to wait + * @param unit the time unit of the timeout argument + * @return {@code true} if quiescent; {@code false} if the + * timeout elapsed. + */ + public boolean awaitQuiescence(long timeout, TimeUnit unit) { + long nanos = unit.toNanos(timeout); + ForkJoinWorkerThread wt; + Thread thread = Thread.currentThread(); + if ((thread instanceof ForkJoinWorkerThread) && + (wt = (ForkJoinWorkerThread)thread).pool == this) { + helpQuiescePool(wt.workQueue); + return true; + } + long startTime = System.nanoTime(); + WorkQueue[] ws; + int r = 0, m; + boolean found = true; + while (!isQuiescent() && (ws = workQueues) != null && + (m = ws.length - 1) >= 0) { + if (!found) { + if ((System.nanoTime() - startTime) > nanos) + return false; + Thread.yield(); // cannot block + } + found = false; + for (int j = (m + 1) << 2; j >= 0; --j) { + ForkJoinTask t; WorkQueue q; int b; + if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) { + found = true; + if ((t = q.pollAt(b)) != null) { + if (q.base - q.top < 0) + signalWork(q); + t.doExec(); + } + break; + } + } + } + return true; + } + + /** + * Waits and/or attempts to assist performing tasks indefinitely + * until the {@link #commonPool()} {@link #isQuiescent} + */ + static void quiesceCommonPool() { + common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS); + } + + /** * Interface for extending managed parallelism for tasks running * in {@link ForkJoinPool}s. * @@ -3160,13 +3271,13 @@ public class ForkJoinPool extends Abstra if (t instanceof ForkJoinWorkerThread) { ForkJoinPool p = ((ForkJoinWorkerThread)t).pool; while (!blocker.isReleasable()) { // variant of helpSignal - WorkQueue[] ws; WorkQueue q; int m, n, u; + WorkQueue[] ws; WorkQueue q; int m, u; if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) { for (int i = 0; i <= m; ++i) { if (blocker.isReleasable()) return; - if ((q = ws[i]) != null && (n = q.queueSize()) > 0) { - p.signalWork(q, n); + if ((q = ws[i]) != null && q.base - q.top < 0) { + p.signalWork(q); if ((u = (int)(p.ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0) break; @@ -3214,7 +3325,7 @@ public class ForkJoinPool extends Abstra private static final long QLOCK; static { - int s; // initialize field offsets for CAS etc + // initialize field offsets for CAS etc try { U = getUnsafe(); Class k = ForkJoinPool.class; @@ -3234,17 +3345,19 @@ public class ForkJoinPool extends Abstra (wk.getDeclaredField("qlock")); Class ak = ForkJoinTask[].class; ABASE = U.arrayBaseOffset(ak); - s = U.arrayIndexScale(ak); - ASHIFT = 31 - Integer.numberOfLeadingZeros(s); + int scale = U.arrayIndexScale(ak); + if ((scale & (scale - 1)) != 0) + throw new Error("data type scale not a power of two"); + ASHIFT = 31 - Integer.numberOfLeadingZeros(scale); } catch (Exception e) { throw new Error(e); } - if ((s & (s-1)) != 0) - throw new Error("data type scale not a power of two"); submitters = new ThreadLocal(); ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory = new DefaultForkJoinWorkerThreadFactory(); + modifyThreadPermission = new RuntimePermission("modifyThread"); + /* * Establish common pool parameters. For extra caution, * computations to set up common pool state are here; the @@ -3275,12 +3388,11 @@ public class ForkJoinPool extends Abstra par = Runtime.getRuntime().availableProcessors(); if (par > MAX_CAP) par = MAX_CAP; - commonPoolParallelism = par; + commonParallelism = par; long np = (long)(-par); // precompute initial ctl value long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK); - commonPool = new ForkJoinPool(par, ct, fac, handler); - modifyThreadPermission = new RuntimePermission("modifyThread"); + common = new ForkJoinPool(par, ct, fac, handler); } /** @@ -3293,22 +3405,23 @@ public class ForkJoinPool extends Abstra private static sun.misc.Unsafe getUnsafe() { try { return sun.misc.Unsafe.getUnsafe(); - } catch (SecurityException se) { - try { - return java.security.AccessController.doPrivileged - (new java.security - .PrivilegedExceptionAction() { - public sun.misc.Unsafe run() throws Exception { - java.lang.reflect.Field f = sun.misc - .Unsafe.class.getDeclaredField("theUnsafe"); - f.setAccessible(true); - return (sun.misc.Unsafe) f.get(null); - }}); - } catch (java.security.PrivilegedActionException e) { - throw new RuntimeException("Could not initialize intrinsics", - e.getCause()); - } + } catch (SecurityException tryReflectionInstead) {} + try { + return java.security.AccessController.doPrivileged + (new java.security.PrivilegedExceptionAction() { + public sun.misc.Unsafe run() throws Exception { + Class k = sun.misc.Unsafe.class; + for (java.lang.reflect.Field f : k.getDeclaredFields()) { + f.setAccessible(true); + Object x = f.get(null); + if (k.isInstance(x)) + return k.cast(x); + } + throw new NoSuchFieldError("the Unsafe"); + }}); + } catch (java.security.PrivilegedActionException e) { + throw new RuntimeException("Could not initialize intrinsics", + e.getCause()); } } - }