ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166e/ForkJoinPool.java
(Generate patch)

Comparing jsr166/src/jsr166e/ForkJoinPool.java (file contents):
Revision 1.13 by jsr166, Wed Nov 14 18:45:53 2012 UTC vs.
Revision 1.56 by jsr166, Wed Feb 13 18:30:47 2013 UTC

# Line 17 | Line 17 | import java.util.concurrent.ExecutorServ
17   import java.util.concurrent.Future;
18   import java.util.concurrent.RejectedExecutionException;
19   import java.util.concurrent.RunnableFuture;
20 import java.util.concurrent.ThreadLocalRandom;
20   import java.util.concurrent.TimeUnit;
21  
22   /**
# Line 38 | Line 37 | import java.util.concurrent.TimeUnit;
37   * ForkJoinPool}s may also be appropriate for use with event-style
38   * tasks that are never joined.
39   *
40 < * <p>A static {@link #commonPool} is available and appropriate for
40 > * <p>A static {@link #commonPool()} is available and appropriate for
41   * most applications. The common pool is used by any ForkJoinTask that
42   * is not explicitly submitted to a specified pool. Using the common
43   * pool normally reduces resource usage (its threads are slowly
# Line 52 | Line 51 | import java.util.concurrent.TimeUnit;
51   * dynamically adding, suspending, or resuming internal worker
52   * threads, even if some tasks are stalled waiting to join
53   * others. However, no such adjustments are guaranteed in the face of
54 < * blocked IO or other unmanaged synchronization. The nested {@link
54 > * blocked I/O or other unmanaged synchronization. The nested {@link
55   * ManagedBlocker} interface enables extension of the kinds of
56   * synchronization accommodated.
57   *
# Line 63 | Line 62 | import java.util.concurrent.TimeUnit;
62   * {@link #toString} returns indications of pool state in a
63   * convenient form for informal monitoring.
64   *
65 < * <p> As is the case with other ExecutorServices, there are three
65 > * <p>As is the case with other ExecutorServices, there are three
66   * main task execution methods summarized in the following table.
67   * These are designed to be used primarily by clients not already
68   * engaged in fork/join computations in the current pool.  The main
# Line 100 | Line 99 | import java.util.concurrent.TimeUnit;
99   *
100   * <p>The common pool is by default constructed with default
101   * parameters, but these may be controlled by setting three {@link
102 < * System#getProperty properties} with prefix {@code
102 > * System#getProperty system properties} with prefix {@code
103   * java.util.concurrent.ForkJoinPool.common}: {@code parallelism} --
104   * an integer greater than zero, {@code threadFactory} -- the class
105   * name of a {@link ForkJoinWorkerThreadFactory}, and {@code
106   * exceptionHandler} -- the class name of a {@link
107 + * java.lang.Thread.UncaughtExceptionHandler
108   * Thread.UncaughtExceptionHandler}. Upon any error in establishing
109   * these settings, default parameters are used.
110   *
# Line 239 | Line 239 | public class ForkJoinPool extends Abstra
239       * enable shutdown.  When used as a lock, it is normally only very
240       * briefly held, so is nearly always available after at most a
241       * brief spin, but we use a monitor-based backup strategy to
242 <     * blocking when needed.
242 >     * block when needed.
243       *
244       * Recording WorkQueues.  WorkQueues are recorded in the
245       * "workQueues" array that is created upon first use and expanded
# Line 248 | Line 248 | public class ForkJoinPool extends Abstra
248       * by a lock but the array is otherwise concurrently readable, and
249       * accessed directly.  To simplify index-based operations, the
250       * array size is always a power of two, and all readers must
251 <     * tolerate null slots. Worker queues are at odd indices Shared
251 >     * tolerate null slots. Worker queues are at odd indices. Shared
252       * (submission) queues are at even indices, up to a maximum of 64
253       * slots, to limit growth even if array needs to expand to add
254       * more workers. Grouping them together in this way simplifies and
# Line 316 | Line 316 | public class ForkJoinPool extends Abstra
316       * execute. However, many other threads may notice the same task
317       * and each signal to wake up a thread that might take it. So in
318       * general, pools will be over-signalled.  When a submission is
319 <     * added or another worker adds a task to a queue that is
320 <     * apparently empty, they signal waiting workers (or trigger
319 >     * added or another worker adds a task to a queue that has fewer
320 >     * than two tasks, they signal waiting workers (or trigger
321       * creation of new ones if fewer than the given parallelism level
322 <     * -- see signalWork).  These primary signals are buttressed by
323 <     * signals whenever other threads scan for work or do not have a
324 <     * task to process. On most platforms, signalling (unpark)
325 <     * overhead time is noticeably long, and the time between
326 <     * signalling a thread and it actually making progress can be very
327 <     * noticeably long, so it is worth offloading these delays from
328 <     * critical paths as much as possible.
322 >     * -- signalWork), and may leave a hint to the unparked worker to
323 >     * help signal others upon wakeup).  These primary signals are
324 >     * buttressed by others (see method helpSignal) whenever other
325 >     * threads scan for work or do not have a task to process.  On
326 >     * most platforms, signalling (unpark) overhead time is noticeably
327 >     * long, and the time between signalling a thread and it actually
328 >     * making progress can be very noticeably long, so it is worth
329 >     * offloading these delays from critical paths as much as
330 >     * possible.
331       *
332       * Trimming workers. To release resources after periods of lack of
333       * use, a worker starting to wait when the pool is quiescent will
# Line 393 | Line 395 | public class ForkJoinPool extends Abstra
395       * steals, rather than use per-task bookkeeping.  This sometimes
396       * requires a linear scan of workQueues array to locate stealers,
397       * but often doesn't because stealers leave hints (that may become
398 <     * stale/wrong) of where to locate them.  A stealHint is only a
399 <     * hint because a worker might have had multiple steals and the
400 <     * hint records only one of them (usually the most current).
401 <     * Hinting isolates cost to when it is needed, rather than adding
402 <     * to per-task overhead.  (2) It is "shallow", ignoring nesting
403 <     * and potentially cyclic mutual steals.  (3) It is intentionally
398 >     * stale/wrong) of where to locate them.  It is only a hint
399 >     * because a worker might have had multiple steals and the hint
400 >     * records only one of them (usually the most current).  Hinting
401 >     * isolates cost to when it is needed, rather than adding to
402 >     * per-task overhead.  (2) It is "shallow", ignoring nesting and
403 >     * potentially cyclic mutual steals.  (3) It is intentionally
404       * racy: field currentJoin is updated only while actively joining,
405       * which means that we miss links in the chain during long-lived
406       * tasks, GC stalls etc (which is OK since blocking in such cases
# Line 438 | Line 440 | public class ForkJoinPool extends Abstra
440       * Common Pool
441       * ===========
442       *
443 <     * The static commonPool always exists after static
443 >     * The static common Pool always exists after static
444       * initialization.  Since it (or any other created pool) need
445       * never be used, we minimize initial construction overhead and
446       * footprint to the setup of about a dozen fields, with no nested
# Line 449 | Line 451 | public class ForkJoinPool extends Abstra
451       * perform some subtask processing (see externalHelpJoin and
452       * related methods).  We do not need to record whether these
453       * submissions are to the common pool -- if not, externalHelpJoin
454 <     * returns quicky (at the most helping to signal some common pool
454 >     * returns quickly (at the most helping to signal some common pool
455       * workers). These submitters would otherwise be blocked waiting
456       * for completion, so the extra effort (with liberally sprinkled
457       * task status checks) in inapplicable cases amounts to an odd
# Line 525 | Line 527 | public class ForkJoinPool extends Abstra
527       * Default ForkJoinWorkerThreadFactory implementation; creates a
528       * new ForkJoinWorkerThread.
529       */
530 <    static class DefaultForkJoinWorkerThreadFactory
530 >    static final class DefaultForkJoinWorkerThreadFactory
531          implements ForkJoinWorkerThreadFactory {
532 <        public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
532 >        public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
533              return new ForkJoinWorkerThread(pool);
534          }
535      }
536  
537      /**
538 +     * Per-thread records for threads that submit to pools. Currently
539 +     * holds only pseudo-random seed / index that is used to choose
540 +     * submission queues in method externalPush. In the future, this may
541 +     * also incorporate a means to implement different task rejection
542 +     * and resubmission policies.
543 +     *
544 +     * Seeds for submitters and workers/workQueues work in basically
545 +     * the same way but are initialized and updated using slightly
546 +     * different mechanics. Both are initialized using the same
547 +     * approach as in class ThreadLocal, where successive values are
548 +     * unlikely to collide with previous values. Seeds are then
549 +     * randomly modified upon collisions using xorshifts, which
550 +     * requires a non-zero seed.
551 +     */
552 +    static final class Submitter {
553 +        int seed;
554 +        Submitter(int s) { seed = s; }
555 +    }
556 +
557 +    /**
558       * Class for artificial tasks that are used to replace the target
559       * of local joins if they are removed from an interior queue slot
560       * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
# Line 598 | Line 620 | public class ForkJoinPool extends Abstra
620       * trades off slightly slower average field access for the sake of
621       * avoiding really bad worst-case access. (Until better JVM
622       * support is in place, this padding is dependent on transient
623 <     * properties of JVM field layout rules.)  We also take care in
623 >     * properties of JVM field layout rules.) We also take care in
624       * allocating, sizing and resizing the array. Non-shared queue
625       * arrays are initialized by workers before use. Others are
626       * allocated on first use.
# Line 624 | Line 646 | public class ForkJoinPool extends Abstra
646           */
647          static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
648  
649 +        // Heuristic padding to ameliorate unfortunate memory placements
650 +        volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
651 +
652          int seed;                  // for random scanning; initialize nonzero
653          volatile int eventCount;   // encoded inactivation count; < 0 if inactive
654          int nextWait;              // encoded record of next event waiter
655 <        final int mode;            // lifo, fifo, or shared
631 <        int nsteals;               // cumulative number of steals
655 >        int hint;                  // steal or signal hint (index)
656          int poolIndex;             // index of this queue in pool (or 0)
657 <        int stealHint;             // index of most recent known stealer
657 >        final int mode;            // 0: lifo, > 0: fifo, < 0: shared
658 >        int nsteals;               // number of steals
659          volatile int qlock;        // 1: locked, -1: terminate; else 0
660          volatile int base;         // index of next slot for poll
661          int top;                   // index of next slot for push
# Line 640 | Line 665 | public class ForkJoinPool extends Abstra
665          volatile Thread parker;    // == owner during call to park; else null
666          volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
667          ForkJoinTask<?> currentSteal; // current non-local task being executed
643        // Heuristic padding to ameliorate unfortunate memory placements
644        Object p00, p01, p02, p03, p04, p05, p06, p07;
645        Object p08, p09, p0a, p0b, p0c, p0d, p0e;
668  
669 <        WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
670 <            this.mode = mode;
669 >        volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
670 >        volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d;
671 >
672 >        WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
673 >                  int seed) {
674              this.pool = pool;
675              this.owner = owner;
676 +            this.mode = mode;
677 +            this.seed = seed;
678              // Place indices in the center of array (that is not yet allocated)
679              base = top = INITIAL_QUEUE_CAPACITY >>> 1;
680          }
681  
682          /**
683 <         * Pushes a task. Call only by owner in unshared queues.
657 <         * Cases needing resizing or rejection are relyaed to fullPush
658 <         * (that also handles shared queues).
659 <         *
660 <         * @param task the task. Caller must ensure non-null.
661 <         * @throw RejectedExecutionException if array cannot be resized
683 >         * Returns the approximate number of tasks in the queue.
684           */
685 <        final void push(ForkJoinTask<?> task) {
686 <            ForkJoinPool p; ForkJoinTask<?>[] a;
687 <            int s = top, n;
688 <            if ((a = array) != null && a.length > (n = s + 1 - base)) {
689 <                U.putOrderedObject
690 <                    (a, (((a.length - 1) & s) << ASHIFT) + ABASE, task);
691 <                top = s + 1;
692 <                if (n <= 1 && (p = pool) != null)
693 <                    p.signalWork(this, 1);
694 <            }
695 <            else
696 <                fullPush(task, true);
685 >        final int queueSize() {
686 >            int n = base - top;       // non-owner callers must read base first
687 >            return (n >= 0) ? 0 : -n; // ignore transient negative
688 >        }
689 >
690 >       /**
691 >         * Provides a more accurate estimate of whether this queue has
692 >         * any tasks than does queueSize, by checking whether a
693 >         * near-empty queue has at least one unclaimed task.
694 >         */
695 >        final boolean isEmpty() {
696 >            ForkJoinTask<?>[] a; int m, s;
697 >            int n = base - (s = top);
698 >            return (n >= 0 ||
699 >                    (n == -1 &&
700 >                     ((a = array) == null ||
701 >                      (m = a.length - 1) < 0 ||
702 >                      U.getObject
703 >                      (a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null)));
704          }
705  
706          /**
707 <         * Pushes a task if lock is free and array is either big
708 <         * enough or can be resized to be big enough. Note: a
680 <         * specialization of a common fast path of this method is in
681 <         * ForkJoinPool.externalPush. When called from a FJWT queue,
682 <         * this can fail only if the pool has been shut down or
683 <         * an out of memory error.
707 >         * Pushes a task. Call only by owner in unshared queues.  (The
708 >         * shared-queue version is embedded in method externalPush.)
709           *
710           * @param task the task. Caller must ensure non-null.
711 <         * @param owned if true, throw RJE on failure
711 >         * @throws RejectedExecutionException if array cannot be resized
712           */
713 <        final boolean fullPush(ForkJoinTask<?> task, boolean owned) {
714 <            ForkJoinPool p; ForkJoinTask<?>[] a;
715 <            if (owned) {
716 <                if (qlock < 0) // must be shutting down
717 <                    throw new RejectedExecutionException();
718 <            }
719 <            else if (!U.compareAndSwapInt(this, QLOCK, 0, 1))
720 <                return false;
721 <            try {
697 <                int s = top, oldLen, len;
698 <                if ((a = array) == null)
699 <                    a = array = new ForkJoinTask<?>[len=INITIAL_QUEUE_CAPACITY];
700 <                else if ((oldLen = a.length) > s + 1 - base)
701 <                    len = oldLen;
702 <                else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY)
703 <                    throw new RejectedExecutionException("Capacity exceeded");
704 <                else {
705 <                    int oldMask, b;
706 <                    ForkJoinTask<?>[] oldA = a;
707 <                    a = array = new ForkJoinTask<?>[len];
708 <                    if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) {
709 <                        int mask = len - 1;
710 <                        do {
711 <                            ForkJoinTask<?> x;
712 <                            int oldj = ((b & oldMask) << ASHIFT) + ABASE;
713 <                            int j    = ((b &    mask) << ASHIFT) + ABASE;
714 <                            x = (ForkJoinTask<?>)
715 <                                U.getObjectVolatile(oldA, oldj);
716 <                            if (x != null &&
717 <                                U.compareAndSwapObject(oldA, oldj, x, null))
718 <                                U.putObjectVolatile(a, j, x);
719 <                        } while (++b != s);
720 <                    }
713 >        final void push(ForkJoinTask<?> task) {
714 >            ForkJoinTask<?>[] a; ForkJoinPool p;
715 >            int s = top, m, n;
716 >            if ((a = array) != null) {    // ignore if queue removed
717 >                int j = (((m = a.length - 1) & s) << ASHIFT) + ABASE;
718 >                U.putOrderedObject(a, j, task);
719 >                if ((n = (top = s + 1) - base) <= 2) {
720 >                    if ((p = pool) != null)
721 >                        p.signalWork(this);
722                  }
723 <                U.putOrderedObject
724 <                    (a, (((len - 1) & s) << ASHIFT) + ABASE, task);
724 <                top = s + 1;
725 <            } finally {
726 <                if (!owned)
727 <                    qlock = 0;
723 >                else if (n >= m)
724 >                    growArray();
725              }
726 <            if ((p = pool) != null)
727 <                p.signalWork(this, 1);
728 <            return true;
726 >        }
727 >
728 >       /**
729 >         * Initializes or doubles the capacity of array. Call either
730 >         * by owner or with lock held -- it is OK for base, but not
731 >         * top, to move while resizings are in progress.
732 >         */
733 >        final ForkJoinTask<?>[] growArray() {
734 >            ForkJoinTask<?>[] oldA = array;
735 >            int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
736 >            if (size > MAXIMUM_QUEUE_CAPACITY)
737 >                throw new RejectedExecutionException("Queue capacity exceeded");
738 >            int oldMask, t, b;
739 >            ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
740 >            if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
741 >                (t = top) - (b = base) > 0) {
742 >                int mask = size - 1;
743 >                do {
744 >                    ForkJoinTask<?> x;
745 >                    int oldj = ((b & oldMask) << ASHIFT) + ABASE;
746 >                    int j    = ((b &    mask) << ASHIFT) + ABASE;
747 >                    x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
748 >                    if (x != null &&
749 >                        U.compareAndSwapObject(oldA, oldj, x, null))
750 >                        U.putObjectVolatile(a, j, x);
751 >                } while (++b != t);
752 >            }
753 >            return a;
754          }
755  
756          /**
# Line 852 | Line 874 | public class ForkJoinPool extends Abstra
874              return seed = r ^= r << 5;
875          }
876  
855        /**
856         * Provides a more accurate estimate of size than (top - base)
857         * by ordering reads and checking whether a near-empty queue
858         * has at least one unclaimed task.
859         */
860        final int queueSize() {
861            ForkJoinTask<?>[] a; int k, s, n;
862            return ((n = base - (s = top)) < 0 &&
863                    (n != -1 ||
864                     ((a = array) != null && (k = a.length) > 0 &&
865                      U.getObject
866                      (a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ?
867                -n : 0;
868        }
869
877          // Specialized execution methods
878  
879          /**
# Line 900 | Line 907 | public class ForkJoinPool extends Abstra
907           * or any other cancelled task. Returns (true) on any CAS
908           * or consistency check failure so caller can retry.
909           *
910 <         * @return false if no progress can be made, else true;
910 >         * @return false if no progress can be made, else true
911           */
912          final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
913              boolean stat = true, removed = false, empty = true;
# Line 945 | Line 952 | public class ForkJoinPool extends Abstra
952  
953          /**
954           * Polls for and executes the given task or any other task in
955 <         * its CountedCompleter computation
955 >         * its CountedCompleter computation.
956           */
957          final boolean pollAndExecCC(ForkJoinTask<?> root) {
958              ForkJoinTask<?>[] a; int b; Object o;
# Line 980 | Line 987 | public class ForkJoinPool extends Abstra
987              if (t != null) {
988                  (currentSteal = t).doExec();
989                  currentSteal = null;
990 <                if (++nsteals < 0) {     // spill on overflow
991 <                    ForkJoinPool p;
985 <                    if ((p = pool) != null)
986 <                        p.collectStealCount(this);
987 <                }
988 <                if (top != base) {       // process remaining local tasks
990 >                ++nsteals;
991 >                if (base - top < 0) {       // process remaining local tasks
992                      if (mode == 0)
993                          popAndExecAll();
994                      else
# Line 1017 | Line 1020 | public class ForkJoinPool extends Abstra
1020                      s != Thread.State.TIMED_WAITING);
1021          }
1022  
1020        /**
1021         * If this owned and is not already interrupted, try to
1022         * interrupt and/or unpark, ignoring exceptions.
1023         */
1024        final void interruptOwner() {
1025            Thread wt, p;
1026            if ((wt = owner) != null && !wt.isInterrupted()) {
1027                try {
1028                    wt.interrupt();
1029                } catch (SecurityException ignore) {
1030                }
1031            }
1032            if ((p = parker) != null)
1033                U.unpark(p);
1034        }
1035
1023          // Unsafe mechanics
1024          private static final sun.misc.Unsafe U;
1025          private static final long QLOCK;
1026          private static final int ABASE;
1027          private static final int ASHIFT;
1028          static {
1042            int s;
1029              try {
1030                  U = getUnsafe();
1031                  Class<?> k = WorkQueue.class;
# Line 1047 | Line 1033 | public class ForkJoinPool extends Abstra
1033                  QLOCK = U.objectFieldOffset
1034                      (k.getDeclaredField("qlock"));
1035                  ABASE = U.arrayBaseOffset(ak);
1036 <                s = U.arrayIndexScale(ak);
1036 >                int scale = U.arrayIndexScale(ak);
1037 >                if ((scale & (scale - 1)) != 0)
1038 >                    throw new Error("data type scale not a power of two");
1039 >                ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
1040              } catch (Exception e) {
1041                  throw new Error(e);
1042              }
1054            if ((s & (s-1)) != 0)
1055                throw new Error("data type scale not a power of two");
1056            ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
1043          }
1044      }
1045  
1060    /**
1061     * Per-thread records for threads that submit to pools. Currently
1062     * holds only pseudo-random seed / index that is used to choose
1063     * submission queues in method externalPush. In the future, this may
1064     * also incorporate a means to implement different task rejection
1065     * and resubmission policies.
1066     *
1067     * Seeds for submitters and workers/workQueues work in basically
1068     * the same way but are initialized and updated using slightly
1069     * different mechanics. Both are initialized using the same
1070     * approach as in class ThreadLocal, where successive values are
1071     * unlikely to collide with previous values. Seeds are then
1072     * randomly modified upon collisions using xorshifts, which
1073     * requires a non-zero seed.
1074     */
1075    static final class Submitter {
1076        int seed;
1077        Submitter(int s) { seed = s; }
1078    }
1079
1080    /** Property prefix for constructing common pool */
1081    private static final String propPrefix =
1082        "java.util.concurrent.ForkJoinPool.common.";
1083
1046      // static fields (initialized in static initializer below)
1047  
1048      /**
# Line 1091 | Line 1053 | public class ForkJoinPool extends Abstra
1053          defaultForkJoinWorkerThreadFactory;
1054  
1055      /**
1056 <     * Common (static) pool. Non-null for public use unless a static
1057 <     * construction exception, but internal usages null-check on use
1058 <     * to paranoically avoid potential initialization circularities
1059 <     * as well as to simplify generated code.
1056 >     * Per-thread submission bookkeeping. Shared across all pools
1057 >     * to reduce ThreadLocal pollution and because random motion
1058 >     * to avoid contention in one pool is likely to hold for others.
1059 >     * Lazily initialized on first submission (but null-checked
1060 >     * in other contexts to avoid unnecessary initialization).
1061       */
1062 <    static final ForkJoinPool commonPool;
1062 >    static final ThreadLocal<Submitter> submitters;
1063  
1064      /**
1065       * Permission required for callers of methods that may start or
# Line 1105 | Line 1068 | public class ForkJoinPool extends Abstra
1068      private static final RuntimePermission modifyThreadPermission;
1069  
1070      /**
1071 <     * Per-thread submission bookkeeping. Shared across all pools
1072 <     * to reduce ThreadLocal pollution and because random motion
1073 <     * to avoid contention in one pool is likely to hold for others.
1074 <     * Lazily initialized on first submission (but null-checked
1112 <     * in other contexts to avoid unnecessary initialization).
1071 >     * Common (static) pool. Non-null for public use unless a static
1072 >     * construction exception, but internal usages null-check on use
1073 >     * to paranoically avoid potential initialization circularities
1074 >     * as well as to simplify generated code.
1075       */
1076 <    static final ThreadLocal<Submitter> submitters;
1076 >    static final ForkJoinPool common;
1077  
1078      /**
1079 <     * Common pool parallelism. Must equal commonPool.parallelism.
1079 >     * Common pool parallelism. Must equal common.parallelism.
1080       */
1081 <    static final int commonPoolParallelism;
1081 >    static final int commonParallelism;
1082  
1083      /**
1084       * Sequence number for creating workerNamePrefix.
# Line 1124 | Line 1086 | public class ForkJoinPool extends Abstra
1086      private static int poolNumberSequence;
1087  
1088      /**
1089 <     * Return the next sequence number. We don't expect this to
1090 <     * ever contend so use simple builtin sync.
1089 >     * Returns the next sequence number. We don't expect this to
1090 >     * ever contend, so use simple builtin sync.
1091       */
1092      private static final synchronized int nextPoolId() {
1093          return ++poolNumberSequence;
# Line 1149 | Line 1111 | public class ForkJoinPool extends Abstra
1111      private static final long FAST_IDLE_TIMEOUT =  200L * 1000L * 1000L;
1112  
1113      /**
1114 +     * Tolerance for idle timeouts, to cope with timer undershoots
1115 +     */
1116 +    private static final long TIMEOUT_SLOP = 2000000L;
1117 +
1118 +    /**
1119       * The maximum stolen->joining link depth allowed in method
1120       * tryHelpStealer.  Must be a power of two.  Depths for legitimate
1121       * chains are unbounded, but we use a fixed constant to avoid
# Line 1248 | Line 1215 | public class ForkJoinPool extends Abstra
1215      static final int FIFO_QUEUE          =  1;
1216      static final int SHARED_QUEUE        = -1;
1217  
1218 +    // bounds for #steps in scan loop -- must be power 2 minus 1
1219 +    private static final int MIN_SCAN    = 0x1ff;   // cover estimation slop
1220 +    private static final int MAX_SCAN    = 0x1ffff; // 4 * max workers
1221 +
1222      // Instance fields
1223  
1224      /*
1225 <     * Field layout order in this class tends to matter more than one
1226 <     * would like. Runtime layout order is only loosely related to
1225 >     * Field layout of this class tends to matter more than one would
1226 >     * like. Runtime layout order is only loosely related to
1227       * declaration order and may differ across JVMs, but the following
1228       * empirically works OK on current JVMs.
1229       */
1230 +
1231 +    // Heuristic padding to ameliorate unfortunate memory placements
1232 +    volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06;
1233 +
1234      volatile long stealCount;                  // collects worker counts
1235      volatile long ctl;                         // main pool control
1261    final int parallelism;                     // parallelism level
1262    final int localMode;                       // per-worker scheduling mode
1263    volatile int indexSeed;                    // worker/submitter index seed
1236      volatile int plock;                        // shutdown status and seqLock
1237 +    volatile int indexSeed;                    // worker/submitter index seed
1238 +    final int config;                          // mode and parallelism level
1239      WorkQueue[] workQueues;                    // main registry
1240 <    final ForkJoinWorkerThreadFactory factory; // factory for new workers
1240 >    final ForkJoinWorkerThreadFactory factory;
1241      final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1242      final String workerNamePrefix;             // to create worker name string
1243  
1244 <    /*
1244 >    volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17;
1245 >    volatile Object pad18, pad19, pad1a, pad1b;
1246 >
1247 >    /**
1248       * Acquires the plock lock to protect worker array and related
1249       * updates. This method is called only if an initial CAS on plock
1250 <     * fails. This acts as a spinLock for normal cases, but falls back
1250 >     * fails. This acts as a spinlock for normal cases, but falls back
1251       * to builtin monitor to block when (rarely) needed. This would be
1252       * a terrible idea for a highly contended lock, but works fine as
1253 <     * a more conservative alternative to a pure spinlock.  See
1277 <     * internal ConcurrentHashMap documentation for further
1278 <     * explanation of nearly the same construction.
1253 >     * a more conservative alternative to a pure spinlock.
1254       */
1255      private int acquirePlock() {
1256          int spins = PL_SPINS, r = 0, ps, nps;
# Line 1283 | Line 1258 | public class ForkJoinPool extends Abstra
1258              if (((ps = plock) & PL_LOCK) == 0 &&
1259                  U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
1260                  return nps;
1261 <            else if (r == 0)
1262 <                r = ThreadLocalRandom.current().nextInt(); // randomize spins
1261 >            else if (r == 0) { // randomize spins if possible
1262 >                Thread t = Thread.currentThread(); WorkQueue w; Submitter z;
1263 >                if ((t instanceof ForkJoinWorkerThread) &&
1264 >                    (w = ((ForkJoinWorkerThread)t).workQueue) != null)
1265 >                    r = w.seed;
1266 >                else if ((z = submitters.get()) != null)
1267 >                    r = z.seed;
1268 >                else
1269 >                    r = 1;
1270 >            }
1271              else if (spins >= 0) {
1272                  r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift
1273                  if (r >= 0)
# Line 1318 | Line 1301 | public class ForkJoinPool extends Abstra
1301          synchronized (this) { notifyAll(); }
1302      }
1303  
1304 +    /**
1305 +     * Tries to create and start one worker if fewer than target
1306 +     * parallelism level exist. Adjusts counts etc on failure.
1307 +     */
1308 +    private void tryAddWorker() {
1309 +        long c; int u;
1310 +        while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1311 +               (u & SHORT_SIGN) != 0 && (int)c == 0) {
1312 +            long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1313 +                             ((u + UAC_UNIT) & UAC_MASK)) << 32;
1314 +            if (U.compareAndSwapLong(this, CTL, c, nc)) {
1315 +                ForkJoinWorkerThreadFactory fac;
1316 +                Throwable ex = null;
1317 +                ForkJoinWorkerThread wt = null;
1318 +                try {
1319 +                    if ((fac = factory) != null &&
1320 +                        (wt = fac.newThread(this)) != null) {
1321 +                        wt.start();
1322 +                        break;
1323 +                    }
1324 +                } catch (Throwable e) {
1325 +                    ex = e;
1326 +                }
1327 +                deregisterWorker(wt, ex);
1328 +                break;
1329 +            }
1330 +        }
1331 +    }
1332 +
1333      //  Registering and deregistering workers
1334  
1335      /**
1336 <     * Callback from ForkJoinWorkerThread constructor to establish its
1337 <     * poolIndex and record its WorkQueue. To avoid scanning bias due
1338 <     * to packing entries in front of the workQueues array, we treat
1339 <     * the array as a simple power-of-two hash table using per-thread
1340 <     * seed as hash, expanding as needed.
1341 <     *
1342 <     * @param w the worker's queue
1343 <     */
1344 <    final void registerWorker(WorkQueue w) {
1345 <        int s, ps; // generate a rarely colliding candidate index seed
1346 <        do {} while (!U.compareAndSwapInt(this, INDEXSEED,
1347 <                                          s = indexSeed, s += SEED_INCREMENT) ||
1336 >     * Callback from ForkJoinWorkerThread to establish and record its
1337 >     * WorkQueue. To avoid scanning bias due to packing entries in
1338 >     * front of the workQueues array, we treat the array as a simple
1339 >     * power-of-two hash table using per-thread seed as hash,
1340 >     * expanding as needed.
1341 >     *
1342 >     * @param wt the worker thread
1343 >     * @return the worker's queue
1344 >     */
1345 >    final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1346 >        Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1347 >        wt.setDaemon(true);
1348 >        if ((handler = ueh) != null)
1349 >            wt.setUncaughtExceptionHandler(handler);
1350 >        do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
1351 >                                          s += SEED_INCREMENT) ||
1352                       s == 0); // skip 0
1353 +        WorkQueue w = new WorkQueue(this, wt, config >>> 16, s);
1354          if (((ps = plock) & PL_LOCK) != 0 ||
1355              !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1356              ps = acquirePlock();
1357          int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1358          try {
1359 <            WorkQueue[] ws;
1343 <            if (w != null && (ws = workQueues) != null) {
1344 <                w.seed = s;
1359 >            if ((ws = workQueues) != null) {    // skip if shutting down
1360                  int n = ws.length, m = n - 1;
1361 <                int r = (s << 1) | 1;               // use odd-numbered indices
1362 <                if (ws[r &= m] != null) {           // collision
1363 <                    int probes = 0;                 // step by approx half size
1361 >                int r = (s << 1) | 1;           // use odd-numbered indices
1362 >                if (ws[r &= m] != null) {       // collision
1363 >                    int probes = 0;             // step by approx half size
1364                      int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
1365                      while (ws[r = (r + step) & m] != null) {
1366                          if (++probes >= n) {
# Line 1355 | Line 1370 | public class ForkJoinPool extends Abstra
1370                          }
1371                      }
1372                  }
1373 <                w.eventCount = w.poolIndex = r;     // establish before recording
1373 >                w.eventCount = w.poolIndex = r; // volatile write orders
1374                  ws[r] = w;
1375              }
1376          } finally {
1377              if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1378                  releasePlock(nps);
1379          }
1380 +        wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex)));
1381 +        return w;
1382      }
1383  
1384      /**
# Line 1377 | Line 1394 | public class ForkJoinPool extends Abstra
1394          WorkQueue w = null;
1395          if (wt != null && (w = wt.workQueue) != null) {
1396              int ps;
1380            collectStealCount(w);
1397              w.qlock = -1;                // ensure set
1398 +            long ns = w.nsteals, sc;     // collect steal count
1399 +            do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1400 +                                               sc = stealCount, sc + ns));
1401              if (((ps = plock) & PL_LOCK) != 0 ||
1402                  !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1403                  ps = acquirePlock();
# Line 1394 | Line 1413 | public class ForkJoinPool extends Abstra
1413              }
1414          }
1415  
1416 <        long c;                             // adjust ctl counts
1416 >        long c;                          // adjust ctl counts
1417          do {} while (!U.compareAndSwapLong
1418                       (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
1419                                             ((c - TC_UNIT) & TC_MASK) |
1420                                             (c & ~(AC_MASK|TC_MASK)))));
1421  
1422 <        if (!tryTerminate(false, false) && w != null) {
1423 <            w.cancelAll();                  // cancel remaining tasks
1424 <            if (w.array != null)            // suppress signal if never ran
1425 <                signalWork(null, 1);        // wake up or create replacement
1426 <            if (ex == null)                 // help clean refs on way out
1427 <                ForkJoinTask.helpExpungeStaleExceptions();
1422 >        if (!tryTerminate(false, false) && w != null && w.array != null) {
1423 >            w.cancelAll();               // cancel remaining tasks
1424 >            WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e;
1425 >            while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) {
1426 >                if (e > 0) {             // activate or create replacement
1427 >                    if ((ws = workQueues) == null ||
1428 >                        (i = e & SMASK) >= ws.length ||
1429 >                        (v = ws[i]) == null)
1430 >                        break;
1431 >                    long nc = (((long)(v.nextWait & E_MASK)) |
1432 >                               ((long)(u + UAC_UNIT) << 32));
1433 >                    if (v.eventCount != (e | INT_SIGN))
1434 >                        break;
1435 >                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
1436 >                        v.eventCount = (e + E_SEQ) & E_MASK;
1437 >                        if ((p = v.parker) != null)
1438 >                            U.unpark(p);
1439 >                        break;
1440 >                    }
1441 >                }
1442 >                else {
1443 >                    if ((short)u < 0)
1444 >                        tryAddWorker();
1445 >                    break;
1446 >                }
1447 >            }
1448          }
1449 <
1450 <        if (ex != null)                     // rethrow
1449 >        if (ex == null)                     // help clean refs on way out
1450 >            ForkJoinTask.helpExpungeStaleExceptions();
1451 >        else                                // rethrow
1452              ForkJoinTask.rethrow(ex);
1453      }
1454  
1415    /**
1416     * Collect worker steal count into total. Called on termination
1417     * and upon int overflow of local count. (There is a possible race
1418     * in the latter case vs any caller of getStealCount, which can
1419     * make its results less accurate than usual.)
1420     */
1421    final void collectStealCount(WorkQueue w) {
1422        if (w != null) {
1423            long sc;
1424            int ns = w.nsteals;
1425            w.nsteals = 0; // handle overflow
1426            long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE);
1427            do {} while (!U.compareAndSwapLong(this, STEALCOUNT,
1428                                               sc = stealCount, sc + steals));
1429        }
1430    }
1431
1455      // Submissions
1456  
1457      /**
# Line 1445 | Line 1468 | public class ForkJoinPool extends Abstra
1468              (ws = workQueues) != null && (m = (ws.length - 1)) >= 0 &&
1469              (q = ws[m & z.seed & SQMASK]) != null &&
1470              U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1471 <            int s = q.top, n;
1472 <            if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) {
1473 <                U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE,
1474 <                            task);
1471 >            int b = q.base, s = q.top, n, an;
1472 >            if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) {
1473 >                int j = (((an - 1) & s) << ASHIFT) + ABASE;
1474 >                U.putOrderedObject(a, j, task);
1475                  q.top = s + 1;                     // push on to deque
1476                  q.qlock = 0;
1477 <                if (n <= 1)
1478 <                    signalWork(q, 1);
1477 >                if (n <= 2)
1478 >                    signalWork(q);
1479                  return;
1480              }
1481              q.qlock = 0;
# Line 1463 | Line 1486 | public class ForkJoinPool extends Abstra
1486      /**
1487       * Full version of externalPush. This method is called, among
1488       * other times, upon the first submission of the first task to the
1489 <     * pool, so must perform secondary initialization: creating
1467 <     * workQueue array and setting plock to a valid value. It also
1489 >     * pool, so must perform secondary initialization.  It also
1490       * detects first submission by an external thread by looking up
1491       * its ThreadLocal, and creates a new shared queue if the one at
1492 <     * index if empty or contended. The lock bodies must be
1492 >     * index if empty or contended. The plock lock body must be
1493       * exception-free (so no try/finally) so we optimistically
1494 <     * allocate new queues/arrays outside the locks and throw them
1495 <     * away if (very rarely) not needed. Note that the plock seq value
1496 <     * can eventually wrap around zero, but if so harmlessly fails to
1497 <     * reinitialize.
1494 >     * allocate new queues outside the lock and throw them away if
1495 >     * (very rarely) not needed.
1496 >     *
1497 >     * Secondary initialization occurs when plock is zero, to create
1498 >     * workQueue array and set plock to a valid value.  This lock body
1499 >     * must also be exception-free. Because the plock seq value can
1500 >     * eventually wrap around zero, this method harmlessly fails to
1501 >     * reinitialize if workQueues exists, while still advancing plock.
1502       */
1503      private void fullExternalPush(ForkJoinTask<?> task) {
1504 <        for (Submitter z = null;;) {
1505 <            WorkQueue[] ws; WorkQueue q; int ps, m, r, s;
1506 <            if ((ps = plock) < 0)
1504 >        int r = 0; // random index seed
1505 >        for (Submitter z = submitters.get();;) {
1506 >            WorkQueue[] ws; WorkQueue q; int ps, m, k;
1507 >            if (z == null) {
1508 >                if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed,
1509 >                                        r += SEED_INCREMENT) && r != 0)
1510 >                    submitters.set(z = new Submitter(r));
1511 >            }
1512 >            else if (r == 0) {                  // move to a different index
1513 >                r = z.seed;
1514 >                r ^= r << 13;                   // same xorshift as WorkQueues
1515 >                r ^= r >>> 17;
1516 >                z.seed = r ^ (r << 5);
1517 >            }
1518 >            else if ((ps = plock) < 0)
1519                  throw new RejectedExecutionException();
1520 <            else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) {
1521 <                int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2;
1522 <                n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
1523 <                WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two
1524 <                if ((ps & PL_LOCK) != 0 ||
1520 >            else if (ps == 0 || (ws = workQueues) == null ||
1521 >                     (m = ws.length - 1) < 0) { // initialize workQueues
1522 >                int p = config & SMASK;         // find power of two table size
1523 >                int n = (p > 1) ? p - 1 : 1;    // ensure at least 2 slots
1524 >                n |= n >>> 1; n |= n >>> 2;  n |= n >>> 4;
1525 >                n |= n >>> 8; n |= n >>> 16; n = (n + 1) << 1;
1526 >                WorkQueue[] nws = ((ws = workQueues) == null || ws.length == 0 ?
1527 >                                   new WorkQueue[n] : null);
1528 >                if (((ps = plock) & PL_LOCK) != 0 ||
1529                      !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1530                      ps = acquirePlock();
1531 <                if ((ws = workQueues) == null)
1531 >                if (((ws = workQueues) == null || ws.length == 0) && nws != null)
1532                      workQueues = nws;
1533                  int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1534                  if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1535                      releasePlock(nps);
1536              }
1537 <            else if (z == null && (z = submitters.get()) == null) {
1538 <                if (U.compareAndSwapInt(this, INDEXSEED,
1539 <                                        s = indexSeed, s += SEED_INCREMENT) &&
1540 <                    s != 0) // skip 0
1541 <                    submitters.set(z = new Submitter(s));
1537 >            else if ((q = ws[k = r & m & SQMASK]) != null) {
1538 >                if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
1539 >                    ForkJoinTask<?>[] a = q.array;
1540 >                    int s = q.top;
1541 >                    boolean submitted = false;
1542 >                    try {                      // locked version of push
1543 >                        if ((a != null && a.length > s + 1 - q.base) ||
1544 >                            (a = q.growArray()) != null) {   // must presize
1545 >                            int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
1546 >                            U.putOrderedObject(a, j, task);
1547 >                            q.top = s + 1;
1548 >                            submitted = true;
1549 >                        }
1550 >                    } finally {
1551 >                        q.qlock = 0;  // unlock
1552 >                    }
1553 >                    if (submitted) {
1554 >                        signalWork(q);
1555 >                        return;
1556 >                    }
1557 >                }
1558 >                r = 0; // move on failure
1559              }
1560 <            else {
1561 <                int k = (r = z.seed) & m & SQMASK;
1562 <                if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) {
1563 <                    (q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k;
1564 <                    if (((ps = plock) & PL_LOCK) != 0 ||
1565 <                        !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1566 <                        ps = acquirePlock();
1567 <                    WorkQueue w = null;
1568 <                    if ((ws = workQueues) != null && k < ws.length &&
1569 <                        (w = ws[k]) == null)
1511 <                        ws[k] = q;
1512 <                    else
1513 <                        q = w;
1514 <                    int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1515 <                    if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1516 <                        releasePlock(nps);
1517 <                }
1518 <                if (q != null && q.qlock == 0 && q.fullPush(task, false))
1519 <                    return;
1520 <                r ^= r << 13;                // same xorshift as WorkQueues
1521 <                r ^= r >>> 17;
1522 <                z.seed = r ^= r << 5;        // move to a different index
1560 >            else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
1561 >                q = new WorkQueue(this, null, SHARED_QUEUE, r);
1562 >                if (((ps = plock) & PL_LOCK) != 0 ||
1563 >                    !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
1564 >                    ps = acquirePlock();
1565 >                if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
1566 >                    ws[k] = q;
1567 >                int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
1568 >                if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
1569 >                    releasePlock(nps);
1570              }
1571 +            else
1572 +                r = 0; // try elsewhere while lock held
1573          }
1574      }
1575  
# Line 1535 | Line 1584 | public class ForkJoinPool extends Abstra
1584      }
1585  
1586      /**
1587 <     * Tries to create (at most one) or activate (possibly several)
1539 <     * workers if too few are active. On contention failure, continues
1540 <     * until at least one worker is signalled or the given queue is
1541 <     * empty or all workers are active.
1587 >     * Tries to create or activate a worker if too few are active.
1588       *
1589 <     * @param q if non-null, the queue holding tasks to be signalled
1544 <     * @param signals the target number of signals.
1589 >     * @param q the (non-null) queue holding tasks to be signalled
1590       */
1591 <    final void signalWork(WorkQueue q, int signals) {
1592 <        long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p;
1591 >    final void signalWork(WorkQueue q) {
1592 >        int hint = q.poolIndex;
1593 >        long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p;
1594          while ((u = (int)((c = ctl) >>> 32)) < 0) {
1595              if ((e = (int)c) > 0) {
1596                  if ((ws = workQueues) != null && ws.length > (i = e & SMASK) &&
# Line 1552 | Line 1598 | public class ForkJoinPool extends Abstra
1598                      long nc = (((long)(w.nextWait & E_MASK)) |
1599                                 ((long)(u + UAC_UNIT) << 32));
1600                      if (U.compareAndSwapLong(this, CTL, c, nc)) {
1601 +                        w.hint = hint;
1602                          w.eventCount = (e + E_SEQ) & E_MASK;
1603                          if ((p = w.parker) != null)
1604                              U.unpark(p);
1605 <                        if (--signals <= 0)
1559 <                            break;
1605 >                        break;
1606                      }
1607 <                    else
1562 <                        signals = 1;
1563 <                    if ((q != null && q.queueSize() == 0))
1607 >                    if (q.top - q.base <= 0)
1608                          break;
1609                  }
1610                  else
1611                      break;
1612              }
1613 <            else if (e == 0 && (u & SHORT_SIGN) != 0) {
1614 <                long nc = (long)(((u + UTC_UNIT) & UTC_MASK) |
1615 <                                 ((u + UAC_UNIT) & UAC_MASK)) << 32;
1572 <                if (U.compareAndSwapLong(this, CTL, c, nc)) {
1573 <                    ForkJoinWorkerThread wt = null;
1574 <                    Throwable ex = null;
1575 <                    boolean started = false;
1576 <                    try {
1577 <                        ForkJoinWorkerThreadFactory fac;
1578 <                        if ((fac = factory) != null &&
1579 <                            (wt = fac.newThread(this)) != null) {
1580 <                            wt.start();
1581 <                            started = true;
1582 <                        }
1583 <                    } catch (Throwable rex) {
1584 <                        ex = rex;
1585 <                    }
1586 <                    if (!started)
1587 <                        deregisterWorker(wt, ex); // adjust counts on failure
1588 <                    break;
1589 <                }
1590 <            }
1591 <            else
1613 >            else {
1614 >                if ((short)u < 0)
1615 >                    tryAddWorker();
1616                  break;
1617 +            }
1618          }
1619      }
1620  
# Line 1599 | Line 1624 | public class ForkJoinPool extends Abstra
1624       * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1625       */
1626      final void runWorker(WorkQueue w) {
1627 <        // initialize queue array in this thread
1603 <        w.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY];
1627 >        w.growArray(); // allocate queue
1628          do { w.runTask(scan(w)); } while (w.qlock >= 0);
1629      }
1630  
# Line 1612 | Line 1636 | public class ForkJoinPool extends Abstra
1636       * contention, or state changes that indicate possible success on
1637       * re-invocation.
1638       *
1639 <     * The scan searches for tasks across a random permutation of
1640 <     * queues (starting at a random index and stepping by a random
1641 <     * relative prime, checking each at least once).  The scan
1642 <     * terminates upon either finding a non-empty queue, or completing
1643 <     * the sweep. If the worker is not inactivated, it takes and
1644 <     * returns a task from this queue. Otherwise, if not activated, it
1645 <     * signals workers (that may include itself) and returns so caller
1646 <     * can retry. Also returns for trtry if the worker array may have
1647 <     * changed during an empty scan.  On failure to find a task, we
1648 <     * take one of the following actions, after which the caller will
1649 <     * retry calling this method unless terminated.
1639 >     * The scan searches for tasks across queues (starting at a random
1640 >     * index, and relying on registerWorker to irregularly scatter
1641 >     * them within array to avoid bias), checking each at least twice.
1642 >     * The scan terminates upon either finding a non-empty queue, or
1643 >     * completing the sweep. If the worker is not inactivated, it
1644 >     * takes and returns a task from this queue. Otherwise, if not
1645 >     * activated, it signals workers (that may include itself) and
1646 >     * returns so caller can retry. Also returns for true if the
1647 >     * worker array may have changed during an empty scan.  On failure
1648 >     * to find a task, we take one of the following actions, after
1649 >     * which the caller will retry calling this method unless
1650 >     * terminated.
1651       *
1652       * * If pool is terminating, terminate the worker.
1653       *
1654       * * If not already enqueued, try to inactivate and enqueue the
1655       * worker on wait queue. Or, if inactivating has caused the pool
1656 <     * to be quiescent, relay to idleAwaitWork to check for
1657 <     * termination and possibly shrink pool.
1656 >     * to be quiescent, relay to idleAwaitWork to possibly shrink
1657 >     * pool.
1658       *
1659       * * If already enqueued and none of the above apply, possibly
1660 <     * (with 1/2 probablility) park awaiting signal, else lingering to
1661 <     * help scan and signal.
1660 >     * park awaiting signal, else lingering to help scan and signal.
1661 >     *
1662 >     * * If a non-empty queue discovered or left as a hint,
1663 >     * help wake up other workers before return.
1664       *
1665       * @param w the worker (via its WorkQueue)
1666       * @return a task or null if none found
1667       */
1668      private final ForkJoinTask<?> scan(WorkQueue w) {
1669 <        WorkQueue[] ws; WorkQueue q;           // first update random seed
1670 <        int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1671 <        int ps = plock, m;                     // volatile read order matters
1672 <        if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1673 <            int ec = w.eventCount;             // ec is negative if inactive
1674 <            int step = (r >>> 16) | 1;         // relatively prime
1675 <            for (int j = (m + 1) << 2;  ; --j, r += step) {
1676 <                ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b, n;
1677 <                if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1678 <                    (a = q.array) != null) {   // probably nonempty
1669 >        WorkQueue[] ws; int m;
1670 >        int ps = plock;                          // read plock before ws
1671 >        if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1672 >            int ec = w.eventCount;               // ec is negative if inactive
1673 >            int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1674 >            w.hint = -1;                         // update seed and clear hint
1675 >            int j = ((m + m + 1) | MIN_SCAN) & MAX_SCAN;
1676 >            do {
1677 >                WorkQueue q; ForkJoinTask<?>[] a; int b;
1678 >                if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 &&
1679 >                    (a = q.array) != null) {     // probably nonempty
1680                      int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1681 <                    t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1681 >                    ForkJoinTask<?> t = (ForkJoinTask<?>)
1682 >                        U.getObjectVolatile(a, i);
1683                      if (q.base == b && ec >= 0 && t != null &&
1684                          U.compareAndSwapObject(a, i, t, null)) {
1685 <                        if ((n = q.top - (q.base = b + 1)) > 0)
1686 <                            signalWork(q, n);
1687 <                        return t;              // taken
1688 <                    }
1689 <                    if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) {
1690 <                        if ((n = q.queueSize() - 1) > 0)
1691 <                            signalWork(q, n);
1692 <                        break;                 // let caller retry after signal
1693 <                    }
1694 <                }
1695 <                else if (j < 0) {              // end of scan
1696 <                    long c = ctl; int e;
1697 <                    if (plock != ps)           // incomplete sweep
1698 <                        break;
1699 <                    if ((e = (int)c) < 0)      // pool is terminating
1700 <                        w.qlock = -1;
1701 <                    else if (ec >= 0) {        // try to enqueue/inactivate
1702 <                        long nc = ((long)ec |
1703 <                                   ((c - AC_UNIT) & (AC_MASK|TC_MASK)));
1704 <                        w.nextWait = e;
1705 <                        w.eventCount = ec | INT_SIGN; // mark as inactive
1706 <                        if (ctl != c ||
1707 <                            !U.compareAndSwapLong(this, CTL, c, nc))
1708 <                            w.eventCount = ec; // unmark on CAS failure
1709 <                        else if ((int)(c >> AC_SHIFT) == 1 - parallelism)
1710 <                            idleAwaitWork(w, nc, c);  // quiescent
1685 >                        if ((q.base = b + 1) - q.top < 0)
1686 >                            signalWork(q);
1687 >                        return t;                // taken
1688 >                    }
1689 >                    else if ((ec < 0 || j < m) && (int)(ctl >> AC_SHIFT) <= 0) {
1690 >                        w.hint = (r + j) & m;    // help signal below
1691 >                        break;                   // cannot take
1692 >                    }
1693 >                }
1694 >            } while (--j >= 0);
1695 >
1696 >            int h, e, ns; long c, sc; WorkQueue q;
1697 >            if ((ns = w.nsteals) != 0) {
1698 >                if (U.compareAndSwapLong(this, STEALCOUNT,
1699 >                                         sc = stealCount, sc + ns))
1700 >                    w.nsteals = 0;               // collect steals and rescan
1701 >            }
1702 >            else if (plock != ps)                // consistency check
1703 >                ;                                // skip
1704 >            else if ((e = (int)(c = ctl)) < 0)
1705 >                w.qlock = -1;                    // pool is terminating
1706 >            else {
1707 >                if ((h = w.hint) < 0) {
1708 >                    if (ec >= 0) {               // try to enqueue/inactivate
1709 >                        long nc = (((long)ec |
1710 >                                    ((c - AC_UNIT) & (AC_MASK|TC_MASK))));
1711 >                        w.nextWait = e;          // link and mark inactive
1712 >                        w.eventCount = ec | INT_SIGN;
1713 >                        if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc))
1714 >                            w.eventCount = ec;   // unmark on CAS failure
1715 >                        else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK))
1716 >                            idleAwaitWork(w, nc, c);
1717                      }
1718 <                    else if (w.seed >= 0 && w.eventCount < 0) {
1718 >                    else if (w.eventCount < 0 && ctl == c) {
1719                          Thread wt = Thread.currentThread();
1720 <                        Thread.interrupted();  // clear status
1720 >                        Thread.interrupted();    // clear status
1721                          U.putObject(wt, PARKBLOCKER, this);
1722 <                        w.parker = wt;         // emulate LockSupport.park
1723 <                        if (w.eventCount < 0)  // recheck
1724 <                            U.park(false, 0L);
1722 >                        w.parker = wt;           // emulate LockSupport.park
1723 >                        if (w.eventCount < 0)    // recheck
1724 >                            U.park(false, 0L);   // block
1725                          w.parker = null;
1726                          U.putObject(wt, PARKBLOCKER, null);
1727                      }
1728 <                    break;
1728 >                }
1729 >                if ((h >= 0 || (h = w.hint) >= 0) &&
1730 >                    (ws = workQueues) != null && h < ws.length &&
1731 >                    (q = ws[h]) != null) {      // signal others before retry
1732 >                    WorkQueue v; Thread p; int u, i, s;
1733 >                    for (int n = (config & SMASK) - 1;;) {
1734 >                        int idleCount = (w.eventCount < 0) ? 0 : -1;
1735 >                        if (((s = idleCount - q.base + q.top) <= n &&
1736 >                             (n = s) <= 0) ||
1737 >                            (u = (int)((c = ctl) >>> 32)) >= 0 ||
1738 >                            (e = (int)c) <= 0 || m < (i = e & SMASK) ||
1739 >                            (v = ws[i]) == null)
1740 >                            break;
1741 >                        long nc = (((long)(v.nextWait & E_MASK)) |
1742 >                                   ((long)(u + UAC_UNIT) << 32));
1743 >                        if (v.eventCount != (e | INT_SIGN) ||
1744 >                            !U.compareAndSwapLong(this, CTL, c, nc))
1745 >                            break;
1746 >                        v.hint = h;
1747 >                        v.eventCount = (e + E_SEQ) & E_MASK;
1748 >                        if ((p = v.parker) != null)
1749 >                            U.unpark(p);
1750 >                        if (--n <= 0)
1751 >                            break;
1752 >                    }
1753                  }
1754              }
1755          }
# Line 1710 | Line 1769 | public class ForkJoinPool extends Abstra
1769       * @param prevCtl the ctl value to restore if thread is terminated
1770       */
1771      private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1772 <        if (w.eventCount < 0 &&
1773 <            (this == commonPool || !tryTerminate(false, false)) &&
1774 <            (int)prevCtl != 0) {
1772 >        if (w != null && w.eventCount < 0 &&
1773 >            !tryTerminate(false, false) && (int)prevCtl != 0 &&
1774 >            ctl == currentCtl) {
1775              int dc = -(short)(currentCtl >>> TC_SHIFT);
1776              long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT;
1777 <            long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop
1777 >            long deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP;
1778              Thread wt = Thread.currentThread();
1779              while (ctl == currentCtl) {
1780                  Thread.interrupted();  // timed variant of version in scan()
# Line 1730 | Line 1789 | public class ForkJoinPool extends Abstra
1789                  if (deadline - System.nanoTime() <= 0L &&
1790                      U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1791                      w.eventCount = (w.eventCount + E_SEQ) | E_MASK;
1792 +                    w.hint = -1;
1793                      w.qlock = -1;   // shrink
1794                      break;
1795                  }
# Line 1738 | Line 1798 | public class ForkJoinPool extends Abstra
1798      }
1799  
1800      /**
1801 <     * Scans through queues looking for work while joining a task;
1802 <     * if any are present, signals.
1801 >     * Scans through queues looking for work while joining a task; if
1802 >     * any present, signals. May return early if more signalling is
1803 >     * detectably unneeded.
1804       *
1805 <     * @param task to return early if done
1805 >     * @param task return early if done
1806       * @param origin an index to start scan
1807       */
1808 <    final int helpSignal(ForkJoinTask<?> task, int origin) {
1809 <        WorkQueue[] ws; WorkQueue q; int m, n, s;
1810 <        if (task != null && (ws = workQueues) != null &&
1811 <            (m = ws.length - 1) >= 0) {
1812 <            for (int i = 0; i <= m; ++i) {
1813 <                if ((s = task.status) < 0)
1814 <                    return s;
1815 <                if ((q = ws[(i + origin) & m]) != null &&
1816 <                    (n = q.queueSize()) > 0) {
1817 <                    signalWork(q, n);
1818 <                    if ((int)(ctl >> AC_SHIFT) >= 0)
1808 >    private void helpSignal(ForkJoinTask<?> task, int origin) {
1809 >        WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s;
1810 >        if (task != null && task.status >= 0 &&
1811 >            (u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 &&
1812 >            (ws = workQueues) != null && (m = ws.length - 1) >= 0) {
1813 >            outer: for (int k = origin, j = m; j >= 0; --j) {
1814 >                WorkQueue q = ws[k++ & m];
1815 >                for (int n = m;;) { // limit to at most m signals
1816 >                    if (task.status < 0)
1817 >                        break outer;
1818 >                    if (q == null ||
1819 >                        ((s = -q.base + q.top) <= n && (n = s) <= 0))
1820                          break;
1821 +                    if ((u = (int)((c = ctl) >>> 32)) >= 0 ||
1822 +                        (e = (int)c) <= 0 || m < (i = e & SMASK) ||
1823 +                        (w = ws[i]) == null)
1824 +                        break outer;
1825 +                    long nc = (((long)(w.nextWait & E_MASK)) |
1826 +                               ((long)(u + UAC_UNIT) << 32));
1827 +                    if (w.eventCount != (e | INT_SIGN))
1828 +                        break outer;
1829 +                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
1830 +                        w.eventCount = (e + E_SEQ) & E_MASK;
1831 +                        if ((p = w.parker) != null)
1832 +                            U.unpark(p);
1833 +                        if (--n <= 0)
1834 +                            break;
1835 +                    }
1836                  }
1837              }
1838          }
1762        return 0;
1839      }
1840  
1841      /**
# Line 1793 | Line 1869 | public class ForkJoinPool extends Abstra
1869                      }
1870                      if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
1871                          break restart;              // shutting down
1872 <                    if ((v = ws[h = (j.stealHint | 1) & m]) == null ||
1872 >                    if ((v = ws[h = (j.hint | 1) & m]) == null ||
1873                          v.currentSteal != subtask) {
1874                          for (int origin = h;;) {    // find stealer
1875                              if (((h = (h + 2) & m) & 15) == 1 &&
# Line 1801 | Line 1877 | public class ForkJoinPool extends Abstra
1877                                  continue restart;   // occasional staleness check
1878                              if ((v = ws[h]) != null &&
1879                                  v.currentSteal == subtask) {
1880 <                                j.stealHint = h;    // save hint
1880 >                                j.hint = h;        // save hint
1881                                  break;
1882                              }
1883                              if (h == origin)
# Line 1850 | Line 1926 | public class ForkJoinPool extends Abstra
1926  
1927      /**
1928       * Analog of tryHelpStealer for CountedCompleters. Tries to steal
1929 <     * and run tasks within the target's computation
1929 >     * and run tasks within the target's computation.
1930       *
1931       * @param task the task to join
1932       * @param mode if shared, exit upon completing any task
1933       * if all workers are active
1858     *
1934       */
1935      private int helpComplete(ForkJoinTask<?> task, int mode) {
1936 <        WorkQueue[] ws; WorkQueue q; int m, n, s;
1936 >        WorkQueue[] ws; WorkQueue q; int m, n, s, u;
1937          if (task != null && (ws = workQueues) != null &&
1938              (m = ws.length - 1) >= 0) {
1939              for (int j = 1, origin = j;;) {
# Line 1866 | Line 1941 | public class ForkJoinPool extends Abstra
1941                      return s;
1942                  if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
1943                      origin = j;
1944 <                    if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0)
1944 >                    if (mode == SHARED_QUEUE &&
1945 >                        ((u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0))
1946                          break;
1947                  }
1948                  else if ((j = (j + 2) & m) == origin)
# Line 1884 | Line 1960 | public class ForkJoinPool extends Abstra
1960       * may become starved.
1961       */
1962      final boolean tryCompensate() {
1963 <        int pc = parallelism, e, u, i, tc; long c;
1963 >        int pc = config & SMASK, e, i, tc; long c;
1964          WorkQueue[] ws; WorkQueue w; Thread p;
1965 <        if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) {
1965 >        if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) {
1966              if (e != 0 && (i = e & SMASK) < ws.length &&
1967                  (w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) {
1968                  long nc = ((long)(w.nextWait & E_MASK) |
# Line 1898 | Line 1974 | public class ForkJoinPool extends Abstra
1974                      return true;   // replace with idle worker
1975                  }
1976              }
1977 <            else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 &&
1978 <                     (u >> UAC_SHIFT) + pc > 1) {
1977 >            else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
1978 >                     (int)(c >> AC_SHIFT) + pc > 1) {
1979                  long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
1980                  if (U.compareAndSwapLong(this, CTL, c, nc))
1981 <                    return true;    // no compensation
1981 >                    return true;   // no compensation
1982              }
1983 <            else if ((tc = u + pc) < MAX_CAP) {
1983 >            else if (tc + pc < MAX_CAP) {
1984                  long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
1985                  if (U.compareAndSwapLong(this, CTL, c, nc)) {
1986 +                    ForkJoinWorkerThreadFactory fac;
1987                      Throwable ex = null;
1988                      ForkJoinWorkerThread wt = null;
1989                      try {
1913                        ForkJoinWorkerThreadFactory fac;
1990                          if ((fac = factory) != null &&
1991                              (wt = fac.newThread(this)) != null) {
1992                              wt.start();
# Line 1919 | Line 1995 | public class ForkJoinPool extends Abstra
1995                      } catch (Throwable rex) {
1996                          ex = rex;
1997                      }
1998 <                    deregisterWorker(wt, ex); // adjust counts etc
1998 >                    deregisterWorker(wt, ex); // clean up and return false
1999                  }
2000              }
2001          }
# Line 1938 | Line 2014 | public class ForkJoinPool extends Abstra
2014          if (joiner != null && task != null && (s = task.status) >= 0) {
2015              ForkJoinTask<?> prevJoin = joiner.currentJoin;
2016              joiner.currentJoin = task;
2017 <            do {} while ((s = task.status) >= 0 &&
1942 <                         joiner.queueSize() > 0 &&
2017 >            do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2018                           joiner.tryRemoveAndExec(task)); // process local tasks
2019 <            if (s >= 0 && (s = task.status) >= 0 &&
2020 <                (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
2021 <                (task instanceof CountedCompleter))
2022 <                s = helpComplete(task, LIFO_QUEUE);
2019 >            if (s >= 0 && (s = task.status) >= 0) {
2020 >                helpSignal(task, joiner.poolIndex);
2021 >                if ((s = task.status) >= 0 &&
2022 >                    (task instanceof CountedCompleter))
2023 >                    s = helpComplete(task, LIFO_QUEUE);
2024 >            }
2025              while (s >= 0 && (s = task.status) >= 0) {
2026 <                if ((joiner.queueSize() > 0 ||           // try helping
2026 >                if ((!joiner.isEmpty() ||           // try helping
2027                       (s = tryHelpStealer(joiner, task)) == 0) &&
2028 <                    (s = task.status) >= 0 && tryCompensate()) {
2029 <                    if (task.trySetSignal() && (s = task.status) >= 0) {
2030 <                        synchronized (task) {
2031 <                            if (task.status >= 0) {
2032 <                                try {                // see ForkJoinTask
2033 <                                    task.wait();     //  for explanation
2034 <                                } catch (InterruptedException ie) {
2028 >                    (s = task.status) >= 0) {
2029 >                    helpSignal(task, joiner.poolIndex);
2030 >                    if ((s = task.status) >= 0 && tryCompensate()) {
2031 >                        if (task.trySetSignal() && (s = task.status) >= 0) {
2032 >                            synchronized (task) {
2033 >                                if (task.status >= 0) {
2034 >                                    try {                // see ForkJoinTask
2035 >                                        task.wait();     //  for explanation
2036 >                                    } catch (InterruptedException ie) {
2037 >                                    }
2038                                  }
2039 +                                else
2040 +                                    task.notifyAll();
2041                              }
1960                            else
1961                                task.notifyAll();
2042                          }
2043 +                        long c;                          // re-activate
2044 +                        do {} while (!U.compareAndSwapLong
2045 +                                     (this, CTL, c = ctl, c + AC_UNIT));
2046                      }
1964                    long c;                          // re-activate
1965                    do {} while (!U.compareAndSwapLong
1966                                 (this, CTL, c = ctl, c + AC_UNIT));
2047                  }
2048              }
2049              joiner.currentJoin = prevJoin;
# Line 1984 | Line 2064 | public class ForkJoinPool extends Abstra
2064          if (joiner != null && task != null && (s = task.status) >= 0) {
2065              ForkJoinTask<?> prevJoin = joiner.currentJoin;
2066              joiner.currentJoin = task;
2067 <            do {} while ((s = task.status) >= 0 &&
1988 <                         joiner.queueSize() > 0 &&
2067 >            do {} while ((s = task.status) >= 0 && !joiner.isEmpty() &&
2068                           joiner.tryRemoveAndExec(task));
2069 <            if (s >= 0 && (s = task.status) >= 0 &&
2070 <                (s = helpSignal(task, joiner.poolIndex)) >= 0 &&
2071 <                (task instanceof CountedCompleter))
2072 <                s = helpComplete(task, LIFO_QUEUE);
2073 <            if (s >= 0 && joiner.queueSize() == 0) {
2069 >            if (s >= 0 && (s = task.status) >= 0) {
2070 >                helpSignal(task, joiner.poolIndex);
2071 >                if ((s = task.status) >= 0 &&
2072 >                    (task instanceof CountedCompleter))
2073 >                    s = helpComplete(task, LIFO_QUEUE);
2074 >            }
2075 >            if (s >= 0 && joiner.isEmpty()) {
2076                  do {} while (task.status >= 0 &&
2077                               tryHelpStealer(joiner, task) > 0);
2078              }
# Line 2001 | Line 2082 | public class ForkJoinPool extends Abstra
2082  
2083      /**
2084       * Returns a (probably) non-empty steal queue, if one is found
2085 <     * during a random, then cyclic scan, else null.  This method must
2086 <     * be retried by caller if, by the time it tries to use the queue,
2006 <     * it is empty.
2085 >     * during a scan, else null.  This method must be retried by
2086 >     * caller if, by the time it tries to use the queue, it is empty.
2087       * @param r a (random) seed for scanning
2088       */
2089      private WorkQueue findNonEmptyStealQueue(int r) {
2090 <        int step = (r >>> 16) | 1;
2091 <        for (WorkQueue[] ws;;) {
2092 <            int ps = plock, m;
2093 <            if ((ws = workQueues) == null || (m = ws.length - 1) < 1)
2094 <                return null;
2095 <            for (int j = (m + 1) << 2; ; r += step) {
2096 <                WorkQueue q = ws[((r << 1) | 1) & m];
2017 <                if (q != null && q.queueSize() > 0)
2018 <                    return q;
2019 <                else if (--j < 0) {
2020 <                    if (plock == ps)
2021 <                        return null;
2022 <                    break;
2090 >        for (;;) {
2091 >            int ps = plock, m; WorkQueue[] ws; WorkQueue q;
2092 >            if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) {
2093 >                for (int j = (m + 1) << 2; j >= 0; --j) {
2094 >                    if ((q = ws[(((r + j) << 1) | 1) & m]) != null &&
2095 >                        q.base - q.top < 0)
2096 >                        return q;
2097                  }
2098              }
2099 +            if (plock == ps)
2100 +                return null;
2101          }
2102      }
2103  
# Line 2033 | Line 2109 | public class ForkJoinPool extends Abstra
2109       */
2110      final void helpQuiescePool(WorkQueue w) {
2111          for (boolean active = true;;) {
2112 <            ForkJoinTask<?> localTask; // exhaust local queue
2113 <            while ((localTask = w.nextLocalTask()) != null)
2114 <                localTask.doExec();
2115 <            // Similar to loop in scan(), but ignoring submissions
2116 <            WorkQueue q = findNonEmptyStealQueue(w.nextSeed());
2117 <            if (q != null) {
2118 <                ForkJoinTask<?> t; int b;
2112 >            long c; WorkQueue q; ForkJoinTask<?> t; int b;
2113 >            while ((t = w.nextLocalTask()) != null) {
2114 >                if (w.base - w.top < 0)
2115 >                    signalWork(w);
2116 >                t.doExec();
2117 >            }
2118 >            if ((q = findNonEmptyStealQueue(w.nextSeed())) != null) {
2119                  if (!active) {      // re-establish active count
2044                    long c;
2120                      active = true;
2121                      do {} while (!U.compareAndSwapLong
2122                                   (this, CTL, c = ctl, c + AC_UNIT));
2123                  }
2124 <                if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2124 >                if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2125 >                    if (q.base - q.top < 0)
2126 >                        signalWork(q);
2127                      w.runSubtask(t);
2128 +                }
2129              }
2130 <            else {
2131 <                long c;
2132 <                if (active) {       // decrement active count without queuing
2130 >            else if (active) {       // decrement active count without queuing
2131 >                long nc = (c = ctl) - AC_UNIT;
2132 >                if ((int)(nc >> AC_SHIFT) + (config & SMASK) == 0)
2133 >                    return;          // bypass decrement-then-increment
2134 >                if (U.compareAndSwapLong(this, CTL, c, nc))
2135                      active = false;
2056                    do {} while (!U.compareAndSwapLong
2057                                 (this, CTL, c = ctl, c -= AC_UNIT));
2058                }
2059                else
2060                    c = ctl;        // re-increment on exit
2061                if ((int)(c >> AC_SHIFT) + parallelism == 0) {
2062                    do {} while (!U.compareAndSwapLong
2063                                 (this, CTL, c = ctl, c + AC_UNIT));
2064                    break;
2065                }
2136              }
2137 +            else if ((int)((c = ctl) >> AC_SHIFT) + (config & SMASK) == 0 &&
2138 +                     U.compareAndSwapLong(this, CTL, c, c + AC_UNIT))
2139 +                return;
2140          }
2141      }
2142  
# Line 2079 | Line 2152 | public class ForkJoinPool extends Abstra
2152                  return t;
2153              if ((q = findNonEmptyStealQueue(w.nextSeed())) == null)
2154                  return null;
2155 <            if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2155 >            if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2156 >                if (q.base - q.top < 0)
2157 >                    signalWork(q);
2158                  return t;
2159 +            }
2160          }
2161      }
2162  
# Line 2109 | Line 2185 | public class ForkJoinPool extends Abstra
2185       * producing extra tasks amortizes the uncertainty of progress and
2186       * diffusion assumptions.
2187       *
2188 <     * So, users will want to use values larger, but not much larger
2188 >     * So, users will want to use values larger (but not much larger)
2189       * than 1 to both smooth over transient shortages and hedge
2190       * against uneven progress; as traded off against the cost of
2191       * extra task overhead. We leave the user to pick a threshold
# Line 2133 | Line 2209 | public class ForkJoinPool extends Abstra
2209      static int getSurplusQueuedTaskCount() {
2210          Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2211          if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2212 <            int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base;
2213 <            int p = (pool = wt.pool).parallelism;
2212 >            int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK;
2213 >            int n = (q = wt.workQueue).top - q.base;
2214              int a = (int)(pool.ctl >> AC_SHIFT) + p;
2215 <            return q.top - b - (a > (p >>>= 1) ? 0 :
2216 <                                a > (p >>>= 1) ? 1 :
2217 <                                a > (p >>>= 1) ? 2 :
2218 <                                a > (p >>>= 1) ? 4 :
2219 <                                8);
2215 >            return n - (a > (p >>>= 1) ? 0 :
2216 >                        a > (p >>>= 1) ? 1 :
2217 >                        a > (p >>>= 1) ? 2 :
2218 >                        a > (p >>>= 1) ? 4 :
2219 >                        8);
2220          }
2221          return 0;
2222      }
# Line 2162 | Line 2238 | public class ForkJoinPool extends Abstra
2238       * @return true if now terminating or terminated
2239       */
2240      private boolean tryTerminate(boolean now, boolean enable) {
2241 <        if (this == commonPool)                     // cannot shut down
2241 >        int ps;
2242 >        if (this == common)                    // cannot shut down
2243              return false;
2244 +        if ((ps = plock) >= 0) {                   // enable by setting plock
2245 +            if (!enable)
2246 +                return false;
2247 +            if ((ps & PL_LOCK) != 0 ||
2248 +                !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2249 +                ps = acquirePlock();
2250 +            int nps = ((ps + PL_LOCK) & ~SHUTDOWN) | SHUTDOWN;
2251 +            if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2252 +                releasePlock(nps);
2253 +        }
2254          for (long c;;) {
2255 <            if (((c = ctl) & STOP_BIT) != 0) {      // already terminating
2256 <                if ((short)(c >>> TC_SHIFT) == -parallelism) {
2255 >            if (((c = ctl) & STOP_BIT) != 0) {     // already terminating
2256 >                if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) {
2257                      synchronized (this) {
2258 <                        notifyAll();                // signal when 0 workers
2258 >                        notifyAll();               // signal when 0 workers
2259                      }
2260                  }
2261                  return true;
2262              }
2263 <            if (plock >= 0) {                       // not yet enabled
2264 <                int ps;
2265 <                if (!enable)
2263 >            if (!now) {                            // check if idle & no tasks
2264 >                WorkQueue[] ws; WorkQueue w;
2265 >                if ((int)(c >> AC_SHIFT) != -(config & SMASK))
2266                      return false;
2267 <                if (((ps = plock) & PL_LOCK) != 0 ||
2268 <                    !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
2269 <                    ps = acquirePlock();
2270 <                int nps = SHUTDOWN;
2271 <                if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
2272 <                    releasePlock(nps);
2273 <            }
2274 <            if (!now) {                             // check if idle & no tasks
2275 <                if ((int)(c >> AC_SHIFT) != -parallelism ||
2276 <                    hasQueuedSubmissions())
2190 <                    return false;
2191 <                // Check for unqueued inactive workers. One pass suffices.
2192 <                WorkQueue[] ws = workQueues; WorkQueue w;
2193 <                if (ws != null) {
2194 <                    for (int i = 1; i < ws.length; i += 2) {
2195 <                        if ((w = ws[i]) != null && w.eventCount >= 0)
2196 <                            return false;
2267 >                if ((ws = workQueues) != null) {
2268 >                    for (int i = 0; i < ws.length; ++i) {
2269 >                        if ((w = ws[i]) != null) {
2270 >                            if (!w.isEmpty()) {    // signal unprocessed tasks
2271 >                                signalWork(w);
2272 >                                return false;
2273 >                            }
2274 >                            if ((i & 1) != 0 && w.eventCount >= 0)
2275 >                                return false;      // unqueued inactive worker
2276 >                        }
2277                      }
2278                  }
2279              }
2280              if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
2281                  for (int pass = 0; pass < 3; ++pass) {
2282 <                    WorkQueue[] ws = workQueues;
2283 <                    if (ws != null) {
2204 <                        WorkQueue w;
2282 >                    WorkQueue[] ws; WorkQueue w; Thread wt;
2283 >                    if ((ws = workQueues) != null) {
2284                          int n = ws.length;
2285                          for (int i = 0; i < n; ++i) {
2286                              if ((w = ws[i]) != null) {
2287                                  w.qlock = -1;
2288                                  if (pass > 0) {
2289                                      w.cancelAll();
2290 <                                    if (pass > 1)
2291 <                                        w.interruptOwner();
2290 >                                    if (pass > 1 && (wt = w.owner) != null) {
2291 >                                        if (!wt.isInterrupted()) {
2292 >                                            try {
2293 >                                                wt.interrupt();
2294 >                                            } catch (Throwable ignore) {
2295 >                                            }
2296 >                                        }
2297 >                                        U.unpark(wt);
2298 >                                    }
2299                                  }
2300                              }
2301                          }
2302                          // Wake up workers parked on event queue
2303                          int i, e; long cc; Thread p;
2304                          while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2305 <                               (i = e & SMASK) < n &&
2305 >                               (i = e & SMASK) < n && i >= 0 &&
2306                                 (w = ws[i]) != null) {
2307                              long nc = ((long)(w.nextWait & E_MASK) |
2308                                         ((cc + AC_UNIT) & AC_MASK) |
# Line 2244 | Line 2330 | public class ForkJoinPool extends Abstra
2330      static WorkQueue commonSubmitterQueue() {
2331          ForkJoinPool p; WorkQueue[] ws; int m; Submitter z;
2332          return ((z = submitters.get()) != null &&
2333 <                (p = commonPool) != null &&
2333 >                (p = common) != null &&
2334                  (ws = p.workQueues) != null &&
2335                  (m = ws.length - 1) >= 0) ?
2336              ws[m & z.seed & SQMASK] : null;
# Line 2255 | Line 2341 | public class ForkJoinPool extends Abstra
2341       */
2342      static boolean tryExternalUnpush(ForkJoinTask<?> t) {
2343          ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z;
2344 <        ForkJoinTask<?>[] a;  int m, s; long j;
2345 <        if ((z = submitters.get()) != null &&
2346 <            (p = commonPool) != null &&
2344 >        ForkJoinTask<?>[] a;  int m, s;
2345 >        if (t != null &&
2346 >            (z = submitters.get()) != null &&
2347 >            (p = common) != null &&
2348              (ws = p.workQueues) != null &&
2349              (m = ws.length - 1) >= 0 &&
2350              (q = ws[m & z.seed & SQMASK]) != null &&
2351              (s = q.top) != q.base &&
2352 <            (a = q.array) != null &&
2353 <            U.getObjectVolatile
2354 <            (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2355 <            U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2356 <            if (q.array == a && q.top == s && // recheck
2357 <                U.compareAndSwapObject(a, j, t, null)) {
2358 <                q.top = s - 1;
2352 >            (a = q.array) != null) {
2353 >            long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2354 >            if (U.getObject(a, j) == t &&
2355 >                U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2356 >                if (q.array == a && q.top == s && // recheck
2357 >                    U.compareAndSwapObject(a, j, t, null)) {
2358 >                    q.top = s - 1;
2359 >                    q.qlock = 0;
2360 >                    return true;
2361 >                }
2362                  q.qlock = 0;
2273                return true;
2363              }
2275            q.qlock = 0;
2364          }
2365          return false;
2366      }
# Line 2287 | Line 2375 | public class ForkJoinPool extends Abstra
2375          if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 &&
2376              root != null && root.status >= 0) {
2377              for (;;) {
2378 <                int s; Object o; CountedCompleter<?> task = null;
2378 >                int s, u; Object o; CountedCompleter<?> task = null;
2379                  if ((s = q.top) - q.base > 0) {
2380                      long j = ((m & (s - 1)) << ASHIFT) + ABASE;
2381                      if ((o = U.getObject(a, j)) != null &&
# Line 2310 | Line 2398 | public class ForkJoinPool extends Abstra
2398                  }
2399                  if (task != null)
2400                      task.doExec();
2401 <                if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0)
2401 >                if (root.status < 0 ||
2402 >                    (u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0)
2403                      break;
2404                  if (task == null) {
2405 <                    if (helpSignal(root, q.poolIndex) >= 0)
2405 >                    helpSignal(root, q.poolIndex);
2406 >                    if (root.status >= 0)
2407                          helpComplete(root, SHARED_QUEUE);
2408                      break;
2409                  }
# Line 2328 | Line 2418 | public class ForkJoinPool extends Abstra
2418      static void externalHelpJoin(ForkJoinTask<?> t) {
2419          // Some hard-to-avoid overlap with tryExternalUnpush
2420          ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z;
2421 <        ForkJoinTask<?>[] a;  int m, s, n; long j;
2422 <        if (t != null && t.status >= 0 &&
2421 >        ForkJoinTask<?>[] a;  int m, s, n;
2422 >        if (t != null &&
2423              (z = submitters.get()) != null &&
2424 <            (p = commonPool) != null &&
2424 >            (p = common) != null &&
2425              (ws = p.workQueues) != null &&
2426              (m = ws.length - 1) >= 0 &&
2427              (q = ws[m & z.seed & SQMASK]) != null &&
2428              (a = q.array) != null) {
2429 <            if ((s = q.top) != q.base &&
2430 <                U.getObjectVolatile
2431 <                (a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t &&
2432 <                U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2433 <                if (q.array == a && q.top == s &&
2434 <                    U.compareAndSwapObject(a, j, t, null)) {
2435 <                    q.top = s - 1;
2436 <                    q.qlock = 0;
2437 <                    t.doExec();
2429 >            int am = a.length - 1;
2430 >            if ((s = q.top) != q.base) {
2431 >                long j = ((am & (s - 1)) << ASHIFT) + ABASE;
2432 >                if (U.getObject(a, j) == t &&
2433 >                    U.compareAndSwapInt(q, QLOCK, 0, 1)) {
2434 >                    if (q.array == a && q.top == s &&
2435 >                        U.compareAndSwapObject(a, j, t, null)) {
2436 >                        q.top = s - 1;
2437 >                        q.qlock = 0;
2438 >                        t.doExec();
2439 >                    }
2440 >                    else
2441 >                        q.qlock = 0;
2442                  }
2349                else
2350                    q.qlock = 0;
2443              }
2444              if (t.status >= 0) {
2445                  if (t instanceof CountedCompleter)
# Line 2358 | Line 2450 | public class ForkJoinPool extends Abstra
2450          }
2451      }
2452  
2361    /**
2362     * Restricted version of helpQuiescePool for external callers
2363     */
2364    static void externalHelpQuiescePool() {
2365        ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b;
2366        int r = ThreadLocalRandom.current().nextInt();
2367        if ((p = commonPool) != null &&
2368            (q = p.findNonEmptyStealQueue(r)) != null &&
2369            (b = q.base) - q.top < 0 &&
2370            (t = q.pollAt(b)) != null)
2371            t.doExec();
2372    }
2373
2453      // Exported methods
2454  
2455      // Constructors
# Line 2387 | Line 2466 | public class ForkJoinPool extends Abstra
2466       *         java.lang.RuntimePermission}{@code ("modifyThread")}
2467       */
2468      public ForkJoinPool() {
2469 <        this(Runtime.getRuntime().availableProcessors(),
2469 >        this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
2470               defaultForkJoinWorkerThreadFactory, null, false);
2471      }
2472  
# Line 2442 | Line 2521 | public class ForkJoinPool extends Abstra
2521              throw new NullPointerException();
2522          if (parallelism <= 0 || parallelism > MAX_CAP)
2523              throw new IllegalArgumentException();
2445        this.parallelism = parallelism;
2524          this.factory = factory;
2525          this.ueh = handler;
2526 <        this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
2526 >        this.config = parallelism | (asyncMode ? (FIFO_QUEUE << 16) : 0);
2527          long np = (long)(-parallelism); // offset ctl counts
2528          this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2529          int pn = nextPoolId();
# Line 2462 | Line 2540 | public class ForkJoinPool extends Abstra
2540      ForkJoinPool(int parallelism, long ctl,
2541                   ForkJoinWorkerThreadFactory factory,
2542                   Thread.UncaughtExceptionHandler handler) {
2543 <        this.parallelism = parallelism;
2543 >        this.config = parallelism;
2544          this.ctl = ctl;
2545          this.factory = factory;
2546          this.ueh = handler;
2469        this.localMode = LIFO_QUEUE;
2547          this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2548      }
2549  
2550      /**
2551 <     * Returns the common pool instance.
2551 >     * Returns the common pool instance. This pool is statically
2552 >     * constructed; its run state is unaffected by attempts to {@link
2553 >     * #shutdown} or {@link #shutdownNow}. However this pool and any
2554 >     * ongoing processing are automatically terminated upon program
2555 >     * {@link System#exit}.  Any program that relies on asynchronous
2556 >     * task processing to complete before program termination should
2557 >     * invoke {@code commonPool().}{@link #awaitQuiescence}, before
2558 >     * exit.
2559       *
2560       * @return the common pool instance
2561 +     * @since 1.8
2562       */
2563      public static ForkJoinPool commonPool() {
2564 <        return commonPool; // cannot be null (if so, a static init error)
2564 >        // assert common != null : "static init error";
2565 >        return common;
2566      }
2567  
2568      // Execution methods
# Line 2599 | Line 2685 | public class ForkJoinPool extends Abstra
2685          // In previous versions of this class, this method constructed
2686          // a task to run ForkJoinTask.invokeAll, but now external
2687          // invocation of multiple tasks is at least as efficient.
2688 <        List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2603 <        // Workaround needed because method wasn't declared with
2604 <        // wildcards in return type but should have been.
2605 <        @SuppressWarnings({"unchecked", "rawtypes"})
2606 <            List<Future<T>> futures = (List<Future<T>>) (List) fs;
2688 >        ArrayList<Future<T>> futures = new ArrayList<Future<T>>(tasks.size());
2689  
2690          boolean done = false;
2691          try {
2692              for (Callable<T> t : tasks) {
2693                  ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2694 +                futures.add(f);
2695                  externalPush(f);
2613                fs.add(f);
2696              }
2697 <            for (ForkJoinTask<T> f : fs)
2698 <                f.quietlyJoin();
2697 >            for (int i = 0, size = futures.size(); i < size; i++)
2698 >                ((ForkJoinTask<?>)futures.get(i)).quietlyJoin();
2699              done = true;
2700              return futures;
2701          } finally {
2702              if (!done)
2703 <                for (ForkJoinTask<T> f : fs)
2704 <                    f.cancel(false);
2703 >                for (int i = 0, size = futures.size(); i < size; i++)
2704 >                    futures.get(i).cancel(false);
2705          }
2706      }
2707  
# Line 2648 | Line 2730 | public class ForkJoinPool extends Abstra
2730       * @return the targeted parallelism level of this pool
2731       */
2732      public int getParallelism() {
2733 <        return parallelism;
2733 >        return config & SMASK;
2734      }
2735  
2736      /**
2737       * Returns the targeted parallelism level of the common pool.
2738       *
2739       * @return the targeted parallelism level of the common pool
2740 +     * @since 1.8
2741       */
2742      public static int getCommonPoolParallelism() {
2743 <        return commonPoolParallelism;
2743 >        return commonParallelism;
2744      }
2745  
2746      /**
# Line 2669 | Line 2752 | public class ForkJoinPool extends Abstra
2752       * @return the number of worker threads
2753       */
2754      public int getPoolSize() {
2755 <        return parallelism + (short)(ctl >>> TC_SHIFT);
2755 >        return (config & SMASK) + (short)(ctl >>> TC_SHIFT);
2756      }
2757  
2758      /**
# Line 2679 | Line 2762 | public class ForkJoinPool extends Abstra
2762       * @return {@code true} if this pool uses async mode
2763       */
2764      public boolean getAsyncMode() {
2765 <        return localMode != 0;
2765 >        return (config >>> 16) == FIFO_QUEUE;
2766      }
2767  
2768      /**
# Line 2710 | Line 2793 | public class ForkJoinPool extends Abstra
2793       * @return the number of active threads
2794       */
2795      public int getActiveThreadCount() {
2796 <        int r = parallelism + (int)(ctl >> AC_SHIFT);
2796 >        int r = (config & SMASK) + (int)(ctl >> AC_SHIFT);
2797          return (r <= 0) ? 0 : r; // suppress momentarily negative values
2798      }
2799  
# Line 2726 | Line 2809 | public class ForkJoinPool extends Abstra
2809       * @return {@code true} if all threads are currently idle
2810       */
2811      public boolean isQuiescent() {
2812 <        return (int)(ctl >> AC_SHIFT) + parallelism == 0;
2812 >        return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0;
2813      }
2814  
2815      /**
# Line 2803 | Line 2886 | public class ForkJoinPool extends Abstra
2886          WorkQueue[] ws; WorkQueue w;
2887          if ((ws = workQueues) != null) {
2888              for (int i = 0; i < ws.length; i += 2) {
2889 <                if ((w = ws[i]) != null && w.queueSize() != 0)
2889 >                if ((w = ws[i]) != null && !w.isEmpty())
2890                      return true;
2891              }
2892          }
# Line 2889 | Line 2972 | public class ForkJoinPool extends Abstra
2972                  }
2973              }
2974          }
2975 <        int pc = parallelism;
2975 >        int pc = (config & SMASK);
2976          int tc = pc + (short)(c >>> TC_SHIFT);
2977          int ac = pc + (int)(c >> AC_SHIFT);
2978          if (ac < 0) // ignore transient negative
# Line 2915 | Line 2998 | public class ForkJoinPool extends Abstra
2998       * Possibly initiates an orderly shutdown in which previously
2999       * submitted tasks are executed, but no new tasks will be
3000       * accepted. Invocation has no effect on execution state if this
3001 <     * is the {@link #commonPool}, and no additional effect if
3001 >     * is the {@link #commonPool()}, and no additional effect if
3002       * already shut down.  Tasks that are in the process of being
3003       * submitted concurrently during the course of this method may or
3004       * may not be rejected.
# Line 2933 | Line 3016 | public class ForkJoinPool extends Abstra
3016      /**
3017       * Possibly attempts to cancel and/or stop all tasks, and reject
3018       * all subsequently submitted tasks.  Invocation has no effect on
3019 <     * execution state if this is the {@link #commonPool}, and no
3019 >     * execution state if this is the {@link #commonPool()}, and no
3020       * additional effect if already shut down. Otherwise, tasks that
3021       * are in the process of being submitted or executed concurrently
3022       * during the course of this method may or may not be
# Line 2962 | Line 3045 | public class ForkJoinPool extends Abstra
3045      public boolean isTerminated() {
3046          long c = ctl;
3047          return ((c & STOP_BIT) != 0L &&
3048 <                (short)(c >>> TC_SHIFT) == -parallelism);
3048 >                (short)(c >>> TC_SHIFT) == -(config & SMASK));
3049      }
3050  
3051      /**
# Line 2970 | Line 3053 | public class ForkJoinPool extends Abstra
3053       * commenced but not yet completed.  This method may be useful for
3054       * debugging. A return of {@code true} reported a sufficient
3055       * period after shutdown may indicate that submitted tasks have
3056 <     * ignored or suppressed interruption, or are waiting for IO,
3056 >     * ignored or suppressed interruption, or are waiting for I/O,
3057       * causing this executor not to properly terminate. (See the
3058       * advisory notes for class {@link ForkJoinTask} stating that
3059       * tasks should not normally entail blocking operations.  But if
# Line 2981 | Line 3064 | public class ForkJoinPool extends Abstra
3064      public boolean isTerminating() {
3065          long c = ctl;
3066          return ((c & STOP_BIT) != 0L &&
3067 <                (short)(c >>> TC_SHIFT) != -parallelism);
3067 >                (short)(c >>> TC_SHIFT) != -(config & SMASK));
3068      }
3069  
3070      /**
# Line 2996 | Line 3079 | public class ForkJoinPool extends Abstra
3079      /**
3080       * Blocks until all tasks have completed execution after a
3081       * shutdown request, or the timeout occurs, or the current thread
3082 <     * is interrupted, whichever happens first. Note that the {@link
3083 <     * #commonPool()} never terminates until program shutdown so
3084 <     * this method will always time out.
3082 >     * is interrupted, whichever happens first. Because the {@link
3083 >     * #commonPool()} never terminates until program shutdown, when
3084 >     * applied to the common pool, this method is equivalent to {@link
3085 >     * #awaitQuiescence} but always returns {@code false}.
3086       *
3087       * @param timeout the maximum time to wait
3088       * @param unit the time unit of the timeout argument
# Line 3008 | Line 3092 | public class ForkJoinPool extends Abstra
3092       */
3093      public boolean awaitTermination(long timeout, TimeUnit unit)
3094          throws InterruptedException {
3095 +        if (Thread.interrupted())
3096 +            throw new InterruptedException();
3097 +        if (this == common) {
3098 +            awaitQuiescence(timeout, unit);
3099 +            return false;
3100 +        }
3101          long nanos = unit.toNanos(timeout);
3102          if (isTerminated())
3103              return true;
# Line 3027 | Line 3117 | public class ForkJoinPool extends Abstra
3117      }
3118  
3119      /**
3120 +     * If called by a ForkJoinTask operating in this pool, equivalent
3121 +     * in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise,
3122 +     * waits and/or attempts to assist performing tasks until this
3123 +     * pool {@link #isQuiescent} or the indicated timeout elapses.
3124 +     *
3125 +     * @param timeout the maximum time to wait
3126 +     * @param unit the time unit of the timeout argument
3127 +     * @return {@code true} if quiescent; {@code false} if the
3128 +     * timeout elapsed.
3129 +     */
3130 +    public boolean awaitQuiescence(long timeout, TimeUnit unit) {
3131 +        long nanos = unit.toNanos(timeout);
3132 +        ForkJoinWorkerThread wt;
3133 +        Thread thread = Thread.currentThread();
3134 +        if ((thread instanceof ForkJoinWorkerThread) &&
3135 +            (wt = (ForkJoinWorkerThread)thread).pool == this) {
3136 +            helpQuiescePool(wt.workQueue);
3137 +            return true;
3138 +        }
3139 +        long startTime = System.nanoTime();
3140 +        WorkQueue[] ws;
3141 +        int r = 0, m;
3142 +        boolean found = true;
3143 +        while (!isQuiescent() && (ws = workQueues) != null &&
3144 +               (m = ws.length - 1) >= 0) {
3145 +            if (!found) {
3146 +                if ((System.nanoTime() - startTime) > nanos)
3147 +                    return false;
3148 +                Thread.yield(); // cannot block
3149 +            }
3150 +            found = false;
3151 +            for (int j = (m + 1) << 2; j >= 0; --j) {
3152 +                ForkJoinTask<?> t; WorkQueue q; int b;
3153 +                if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) {
3154 +                    found = true;
3155 +                    if ((t = q.pollAt(b)) != null) {
3156 +                        if (q.base - q.top < 0)
3157 +                            signalWork(q);
3158 +                        t.doExec();
3159 +                    }
3160 +                    break;
3161 +                }
3162 +            }
3163 +        }
3164 +        return true;
3165 +    }
3166 +
3167 +    /**
3168 +     * Waits and/or attempts to assist performing tasks indefinitely
3169 +     * until the {@link #commonPool()} {@link #isQuiescent}.
3170 +     */
3171 +    static void quiesceCommonPool() {
3172 +        common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
3173 +    }
3174 +
3175 +    /**
3176       * Interface for extending managed parallelism for tasks running
3177       * in {@link ForkJoinPool}s.
3178       *
# Line 3125 | Line 3271 | public class ForkJoinPool extends Abstra
3271          if (t instanceof ForkJoinWorkerThread) {
3272              ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3273              while (!blocker.isReleasable()) { // variant of helpSignal
3274 <                WorkQueue[] ws; WorkQueue q; int m, n;
3274 >                WorkQueue[] ws; WorkQueue q; int m, u;
3275                  if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) {
3276                      for (int i = 0; i <= m; ++i) {
3277                          if (blocker.isReleasable())
3278                              return;
3279 <                        if ((q = ws[i]) != null && (n = q.queueSize()) > 0) {
3280 <                            p.signalWork(q, n);
3281 <                            if ((int)(p.ctl >> AC_SHIFT) >= 0)
3279 >                        if ((q = ws[i]) != null && q.base - q.top < 0) {
3280 >                            p.signalWork(q);
3281 >                            if ((u = (int)(p.ctl >>> 32)) >= 0 ||
3282 >                                (u >> UAC_SHIFT) >= 0)
3283                                  break;
3284                          }
3285                      }
# Line 3178 | Line 3325 | public class ForkJoinPool extends Abstra
3325      private static final long QLOCK;
3326  
3327      static {
3328 <        // Establish common pool parameters
3182 <        // TBD: limit or report ignored exceptions?
3183 <
3184 <        int par = 0;
3185 <        ForkJoinWorkerThreadFactory fac = null;
3186 <        Thread.UncaughtExceptionHandler handler = null;
3187 <        try {
3188 <            String pp = System.getProperty(propPrefix + "parallelism");
3189 <            String hp = System.getProperty(propPrefix + "exceptionHandler");
3190 <            String fp = System.getProperty(propPrefix + "threadFactory");
3191 <            if (fp != null)
3192 <                fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3193 <                       getSystemClassLoader().loadClass(fp).newInstance());
3194 <            if (hp != null)
3195 <                handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3196 <                           getSystemClassLoader().loadClass(hp).newInstance());
3197 <            if (pp != null)
3198 <                par = Integer.parseInt(pp);
3199 <        } catch (Exception ignore) {
3200 <        }
3201 <
3202 <        int s; // initialize field offsets for CAS etc
3328 >        // initialize field offsets for CAS etc
3329          try {
3330              U = getUnsafe();
3331              Class<?> k = ForkJoinPool.class;
# Line 3219 | Line 3345 | public class ForkJoinPool extends Abstra
3345                  (wk.getDeclaredField("qlock"));
3346              Class<?> ak = ForkJoinTask[].class;
3347              ABASE = U.arrayBaseOffset(ak);
3348 <            s = U.arrayIndexScale(ak);
3349 <            ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
3348 >            int scale = U.arrayIndexScale(ak);
3349 >            if ((scale & (scale - 1)) != 0)
3350 >                throw new Error("data type scale not a power of two");
3351 >            ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
3352          } catch (Exception e) {
3353              throw new Error(e);
3354          }
3355 <        if ((s & (s-1)) != 0)
3356 <            throw new Error("data type scale not a power of two");
3355 >
3356 >        submitters = new ThreadLocal<Submitter>();
3357 >        ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory =
3358 >            new DefaultForkJoinWorkerThreadFactory();
3359 >        modifyThreadPermission = new RuntimePermission("modifyThread");
3360  
3361          /*
3362 <         * For extra caution, computations to set up pool state are
3363 <         * here; the constructor just assigns these values to fields.
3362 >         * Establish common pool parameters.  For extra caution,
3363 >         * computations to set up common pool state are here; the
3364 >         * constructor just assigns these values to fields.
3365           */
3366 <        ForkJoinWorkerThreadFactory defaultFac =
3367 <            defaultForkJoinWorkerThreadFactory =
3368 <            new DefaultForkJoinWorkerThreadFactory();
3369 <        if (fac == null)
3370 <            fac = defaultFac;
3366 >
3367 >        int par = 0;
3368 >        Thread.UncaughtExceptionHandler handler = null;
3369 >        try {  // TBD: limit or report ignored exceptions?
3370 >            String pp = System.getProperty
3371 >                ("java.util.concurrent.ForkJoinPool.common.parallelism");
3372 >            String hp = System.getProperty
3373 >                ("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
3374 >            String fp = System.getProperty
3375 >                ("java.util.concurrent.ForkJoinPool.common.threadFactory");
3376 >            if (fp != null)
3377 >                fac = ((ForkJoinWorkerThreadFactory)ClassLoader.
3378 >                       getSystemClassLoader().loadClass(fp).newInstance());
3379 >            if (hp != null)
3380 >                handler = ((Thread.UncaughtExceptionHandler)ClassLoader.
3381 >                           getSystemClassLoader().loadClass(hp).newInstance());
3382 >            if (pp != null)
3383 >                par = Integer.parseInt(pp);
3384 >        } catch (Exception ignore) {
3385 >        }
3386 >
3387          if (par <= 0)
3388              par = Runtime.getRuntime().availableProcessors();
3389          if (par > MAX_CAP)
3390              par = MAX_CAP;
3391 +        commonParallelism = par;
3392          long np = (long)(-par); // precompute initial ctl value
3393          long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
3394  
3395 <        commonPoolParallelism = par;
3247 <        commonPool = new ForkJoinPool(par, ct, fac, handler);
3248 <        modifyThreadPermission = new RuntimePermission("modifyThread");
3249 <        submitters = new ThreadLocal<Submitter>();
3395 >        common = new ForkJoinPool(par, ct, fac, handler);
3396      }
3397  
3398      /**
# Line 3259 | Line 3405 | public class ForkJoinPool extends Abstra
3405      private static sun.misc.Unsafe getUnsafe() {
3406          try {
3407              return sun.misc.Unsafe.getUnsafe();
3408 <        } catch (SecurityException se) {
3409 <            try {
3410 <                return java.security.AccessController.doPrivileged
3411 <                    (new java.security
3412 <                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
3413 <                        public sun.misc.Unsafe run() throws Exception {
3414 <                            java.lang.reflect.Field f = sun.misc
3415 <                                .Unsafe.class.getDeclaredField("theUnsafe");
3416 <                            f.setAccessible(true);
3417 <                            return (sun.misc.Unsafe) f.get(null);
3418 <                        }});
3419 <            } catch (java.security.PrivilegedActionException e) {
3420 <                throw new RuntimeException("Could not initialize intrinsics",
3421 <                                           e.getCause());
3422 <            }
3408 >        } catch (SecurityException tryReflectionInstead) {}
3409 >        try {
3410 >            return java.security.AccessController.doPrivileged
3411 >            (new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
3412 >                public sun.misc.Unsafe run() throws Exception {
3413 >                    Class<sun.misc.Unsafe> k = sun.misc.Unsafe.class;
3414 >                    for (java.lang.reflect.Field f : k.getDeclaredFields()) {
3415 >                        f.setAccessible(true);
3416 >                        Object x = f.get(null);
3417 >                        if (k.isInstance(x))
3418 >                            return k.cast(x);
3419 >                    }
3420 >                    throw new NoSuchFieldError("the Unsafe");
3421 >                }});
3422 >        } catch (java.security.PrivilegedActionException e) {
3423 >            throw new RuntimeException("Could not initialize intrinsics",
3424 >                                       e.getCause());
3425          }
3426      }
3279
3427   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines