ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/ForkJoinWorkerThread.java
(Generate patch)

Comparing jsr166/src/jsr166y/ForkJoinWorkerThread.java (file contents):
Revision 1.38 by dl, Fri Jul 23 16:49:11 2010 UTC vs.
Revision 1.56 by jsr166, Thu Nov 18 00:39:15 2010 UTC

# Line 6 | Line 6
6  
7   package jsr166y;
8  
9 import java.util.concurrent.*;
10
9   import java.util.Random;
10   import java.util.Collection;
11   import java.util.concurrent.locks.LockSupport;
12 + import java.util.concurrent.RejectedExecutionException;
13  
14   /**
15   * A thread managed by a {@link ForkJoinPool}.  This class is
# Line 97 | Line 96 | public class ForkJoinWorkerThread extend
96       * technique for implementing efficient futures" SIGPLAN Notices,
97       * 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs
98       * in that: (1) We only maintain dependency links across workers
99 <     * upon steals, rather than maintain per-task bookkeeping.  This
100 <     * may require a linear scan of workers array to locate stealers,
101 <     * but usually doesn't because stealers leave hints (that may
102 <     * become stale/wrong) of where to locate the kathem. This
103 <     * isolates cost to when it is needed, rather than adding to
104 <     * per-task overhead.  (2) It is "shallow", ignoring nesting and
105 <     * potentially cyclic mutual steals.  (3) It is intentionally
106 <     * racy: field currentJoin is updated only while actively joining,
107 <     * which means that we could miss links in the chain during
108 <     * long-lived tasks, GC stalls etc.  (4) We bound the number of
109 <     * attempts to find work (see MAX_HELP_DEPTH) and fall back to
110 <     * suspending the worker and if necessary replacing it with a
111 <     * spare (see ForkJoinPool.tryAwaitJoin).
99 >     * upon steals, rather than use per-task bookkeeping.  This may
100 >     * require a linear scan of workers array to locate stealers, but
101 >     * usually doesn't because stealers leave hints (that may become
102 >     * stale/wrong) of where to locate them. This isolates cost to
103 >     * when it is needed, rather than adding to per-task overhead.
104 >     * (2) It is "shallow", ignoring nesting and potentially cyclic
105 >     * mutual steals.  (3) It is intentionally racy: field currentJoin
106 >     * is updated only while actively joining, which means that we
107 >     * miss links in the chain during long-lived tasks, GC stalls etc
108 >     * (which is OK since blocking in such cases is usually a good
109 >     * idea).  (4) We bound the number of attempts to find work (see
110 >     * MAX_HELP_DEPTH) and fall back to suspending the worker and if
111 >     * necessary replacing it with a spare (see
112 >     * ForkJoinPool.awaitJoin).
113       *
114       * Efficient implementation of these algorithms currently relies
115       * on an uncomfortable amount of "Unsafe" mechanics. To maintain
# Line 154 | Line 154 | public class ForkJoinWorkerThread extend
154      private static final Random seedGenerator = new Random();
155  
156      /**
157     * The timeout value for suspending spares. Spare workers that
158     * remain unsignalled for more than this time may be trimmed
159     * (killed and removed from pool).  Since our goal is to avoid
160     * long-term thread buildup, the exact value of timeout does not
161     * matter too much so long as it avoids most false-alarm timeouts
162     * under GC stalls or momentarily high system load.
163     */
164    private static final long SPARE_KEEPALIVE_NANOS =
165        5L * 1000L * 1000L * 1000L; // 5 secs
166
167    /**
157       * The maximum stolen->joining link depth allowed in helpJoinTask.
158       * Depths for legitimate chains are unbounded, but we use a fixed
159       * constant to avoid (otherwise unchecked) cycles and bound
# Line 182 | Line 171 | public class ForkJoinWorkerThread extend
171  
172      /**
173       * Maximum work-stealing queue array size.  Must be less than or
174 <     * equal to 1 << 28 to ensure lack of index wraparound. (This
175 <     * is less than usual bounds, because we need leftshift by 3
176 <     * to be in int range).
174 >     * equal to 1 << (31 - width of array entry) to ensure lack of
175 >     * index wraparound. The value is set in the static block
176 >     * at the end of this file after obtaining width.
177       */
178 <    private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
178 >    private static final int MAXIMUM_QUEUE_CAPACITY;
179  
180      /**
181       * The pool this thread works in. Accessed directly by ForkJoinTask.
# Line 227 | Line 216 | public class ForkJoinWorkerThread extend
216       * Run state of this worker. In addition to the usual run levels,
217       * tracks if this worker is suspended as a spare, and if it was
218       * killed (trimmed) while suspended. However, "active" status is
219 <     * maintained separately.
219 >     * maintained separately and modified only in conjunction with
220 >     * CASes of the pool's runState (which are currently sadly
221 >     * manually inlined for performance.)  Accessed directly by pool
222 >     * to simplify checks for normal (zero) status.
223       */
224 <    private volatile int runState;
224 >    volatile int runState;
225  
226      private static final int TERMINATING = 0x01;
227      private static final int TERMINATED  = 0x02;
# Line 237 | Line 229 | public class ForkJoinWorkerThread extend
229      private static final int TRIMMED     = 0x08; // killed while suspended
230  
231      /**
232 <     * Number of LockSupport.park calls to block this thread for
233 <     * suspension or event waits. Used for internal instrumention;
242 <     * currently not exported but included because volatile write upon
243 <     * park also provides a workaround for a JVM bug.
244 <     */
245 <    volatile int parkCount;
246 <
247 <    /**
248 <     * Number of steals, transferred and reset in pool callbacks pool
249 <     * when idle Accessed directly by pool.
232 >     * Number of steals. Directly accessed (and reset) by
233 >     * pool.tryAccumulateStealCount when idle.
234       */
235      int stealCount;
236  
# Line 256 | Line 240 | public class ForkJoinWorkerThread extend
240       */
241      private int seed;
242  
259
243      /**
244       * Activity status. When true, this worker is considered active.
245       * Accessed directly by pool.  Must be false upon construction.
# Line 265 | Line 248 | public class ForkJoinWorkerThread extend
248  
249      /**
250       * True if use local fifo, not default lifo, for local polling.
251 <     * Shadows value from ForkJoinPool, which resets it if changed
269 <     * pool-wide.
251 >     * Shadows value from ForkJoinPool.
252       */
253      private final boolean locallyFifo;
254  
# Line 284 | Line 266 | public class ForkJoinWorkerThread extend
266      int lastEventCount;
267  
268      /**
269 <     * Encoded index and event count of next event waiter. Used only
270 <     * by ForkJoinPool for managing event waiters.
269 >     * Encoded index and event count of next event waiter. Accessed
270 >     * only by ForkJoinPool for managing event waiters.
271       */
272      volatile long nextWaiter;
273  
274      /**
275 +     * Number of times this thread suspended as spare. Accessed only
276 +     * by pool.
277 +     */
278 +    int spareCount;
279 +
280 +    /**
281 +     * Encoded index and count of next spare waiter. Accessed only
282 +     * by ForkJoinPool for managing spares.
283 +     */
284 +    volatile int nextSpare;
285 +
286 +    /**
287       * The task currently being joined, set only when actively trying
288 <     * to helpStealer. Written only by current thread, but read by
289 <     * others.
288 >     * to help other stealers in helpJoinTask. Written only by this
289 >     * thread, but read by others.
290       */
291      private volatile ForkJoinTask<?> currentJoin;
292  
293      /**
294       * The task most recently stolen from another worker (or
295 <     * submission queue).  Not volatile because always read/written in
296 <     * presence of related volatiles in those cases where it matters.
295 >     * submission queue).  Written only by this thread, but read by
296 >     * others.
297       */
298 <    private ForkJoinTask<?> currentSteal;
298 >    private volatile ForkJoinTask<?> currentSteal;
299  
300      /**
301       * Creates a ForkJoinWorkerThread operating in the given pool.
# Line 312 | Line 306 | public class ForkJoinWorkerThread extend
306      protected ForkJoinWorkerThread(ForkJoinPool pool) {
307          this.pool = pool;
308          this.locallyFifo = pool.locallyFifo;
309 +        setDaemon(true);
310          // To avoid exposing construction details to subclasses,
311          // remaining initialization is in start() and onStart()
312      }
313  
314      /**
315 <     * Performs additional initialization and starts this thread
315 >     * Performs additional initialization and starts this thread.
316       */
317      final void start(int poolIndex, UncaughtExceptionHandler ueh) {
318          this.poolIndex = poolIndex;
319          if (ueh != null)
320              setUncaughtExceptionHandler(ueh);
326        setDaemon(true);
321          start();
322      }
323  
# Line 354 | Line 348 | public class ForkJoinWorkerThread extend
348      /**
349       * Initializes internal state after construction but before
350       * processing any tasks. If you override this method, you must
351 <     * invoke super.onStart() at the beginning of the method.
351 >     * invoke @code{super.onStart()} at the beginning of the method.
352       * Initialization requires care: Most fields must have legal
353       * default values, to ensure that attempted accesses from other
354       * threads work correctly even before this thread starts
# Line 382 | Line 376 | public class ForkJoinWorkerThread extend
376       */
377      protected void onTermination(Throwable exception) {
378          try {
379 +            ForkJoinPool p = pool;
380 +            if (active) {
381 +                int a; // inline p.tryDecrementActiveCount
382 +                active = false;
383 +                do {} while (!UNSAFE.compareAndSwapInt
384 +                             (p, poolRunStateOffset, a = p.runState, a - 1));
385 +            }
386              cancelTasks();
387              setTerminated();
388 <            pool.workerTerminated(this);
388 >            p.workerTerminated(this);
389          } catch (Throwable ex) {        // Shouldn't ever happen
390              if (exception == null)      // but if so, at least rethrown
391                  exception = ex;
# Line 414 | Line 415 | public class ForkJoinWorkerThread extend
415      // helpers for run()
416  
417      /**
418 <     * Find and execute tasks and check status while running
418 >     * Finds and executes tasks, and checks status while running.
419       */
420      private void mainLoop() {
421 <        int emptyScans = 0; // consecutive times failed to find work
421 >        boolean ran = false; // true if ran a task on last step
422          ForkJoinPool p = pool;
423          for (;;) {
424 <            p.preStep(this, emptyScans);
424 >            p.preStep(this, ran);
425              if (runState != 0)
426 <                return;
427 <            ForkJoinTask<?> t; // try to get and run stolen or submitted task
427 <            if ((t = scan()) != null || (t = pollSubmission()) != null) {
428 <                t.tryExec();
429 <                if (base != sp)
430 <                    runLocalTasks();
431 <                currentSteal = null;
432 <                emptyScans = 0;
433 <            }
434 <            else
435 <                ++emptyScans;
426 >                break;
427 >            ran = tryExecSteal() || tryExecSubmission();
428          }
429      }
430  
431      /**
432 <     * Runs local tasks until queue is empty or shut down.  Call only
433 <     * while active.
432 >     * Tries to steal a task and execute it.
433 >     *
434 >     * @return true if ran a task
435       */
436 <    private void runLocalTasks() {
437 <        while (runState == 0) {
438 <            ForkJoinTask<?> t = locallyFifo? locallyDeqTask() : popTask();
439 <            if (t != null)
440 <                t.tryExec();
441 <            else if (base == sp)
442 <                break;
436 >    private boolean tryExecSteal() {
437 >        ForkJoinTask<?> t;
438 >        if ((t = scan()) != null) {
439 >            t.quietlyExec();
440 >            UNSAFE.putOrderedObject(this, currentStealOffset, null);
441 >            if (sp != base)
442 >                execLocalTasks();
443 >            return true;
444          }
445 +        return false;
446      }
447  
448      /**
449 <     * If a submission exists, try to activate and take it
449 >     * If a submission exists, try to activate and run it.
450       *
451 <     * @return a task, if available
451 >     * @return true if ran a task
452       */
453 <    private ForkJoinTask<?> pollSubmission() {
453 >    private boolean tryExecSubmission() {
454          ForkJoinPool p = pool;
455 +        // This loop is needed in case attempt to activate fails, in
456 +        // which case we only retry if there still appears to be a
457 +        // submission.
458          while (p.hasQueuedSubmissions()) {
459 <            if (active || (active = p.tryIncrementActiveCount())) {
460 <                ForkJoinTask<?> t = p.pollSubmission();
461 <                if (t != null) {
462 <                    currentSteal = t;
463 <                    return t;
459 >            ForkJoinTask<?> t; int a;
460 >            if (active || // inline p.tryIncrementActiveCount
461 >                (active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
462 >                                                   a = p.runState, a + 1))) {
463 >                if ((t = p.pollSubmission()) != null) {
464 >                    UNSAFE.putOrderedObject(this, currentStealOffset, t);
465 >                    t.quietlyExec();
466 >                    UNSAFE.putOrderedObject(this, currentStealOffset, null);
467 >                    if (sp != base)
468 >                        execLocalTasks();
469 >                    return true;
470                  }
467                return scan(); // if missed, rescan
471              }
472          }
473 <        return null;
473 >        return false;
474 >    }
475 >
476 >    /**
477 >     * Runs local tasks until queue is empty or shut down.  Call only
478 >     * while active.
479 >     */
480 >    private void execLocalTasks() {
481 >        while (runState == 0) {
482 >            ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask();
483 >            if (t != null)
484 >                t.quietlyExec();
485 >            else if (sp == base)
486 >                break;
487 >        }
488      }
489  
490      /*
491       * Intrinsics-based atomic writes for queue slots. These are
492 <     * basically the same as methods in AtomicObjectArray, but
492 >     * basically the same as methods in AtomicReferenceArray, but
493       * specialized for (1) ForkJoinTask elements (2) requirement that
494       * nullness and bounds checks have already been performed by
495       * callers and (3) effective offsets are known not to overflow
496       * from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't
497       * need corresponding version for reads: plain array reads are OK
498 <     * because they protected by other volatile reads and are
498 >     * because they are protected by other volatile reads and are
499       * confirmed by CASes.
500       *
501       * Most uses don't actually call these methods, but instead contain
# Line 502 | Line 519 | public class ForkJoinWorkerThread extend
519       * range. This method is used only during resets and backouts.
520       */
521      private static final void writeSlot(ForkJoinTask<?>[] q, int i,
522 <                                              ForkJoinTask<?> t) {
522 >                                        ForkJoinTask<?> t) {
523          UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
524      }
525  
# Line 535 | Line 552 | public class ForkJoinWorkerThread extend
552          ForkJoinTask<?> t;
553          ForkJoinTask<?>[] q;
554          int b, i;
555 <        if ((b = base) != sp &&
555 >        if (sp != (b = base) &&
556              (q = queue) != null && // must read q after b
557              (t = q[i = (q.length - 1) & b]) != null && base == b &&
558              UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) {
# Line 547 | Line 564 | public class ForkJoinWorkerThread extend
564  
565      /**
566       * Tries to take a task from the base of own queue. Assumes active
567 <     * status.  Called only by current thread.
567 >     * status.  Called only by this thread.
568       *
569       * @return a task, or null if none
570       */
# Line 570 | Line 587 | public class ForkJoinWorkerThread extend
587  
588      /**
589       * Returns a popped task, or null if empty. Assumes active status.
590 <     * Called only by current thread.
590 >     * Called only by this thread.
591       */
592 <    final ForkJoinTask<?> popTask() {
593 <        int s;
594 <        ForkJoinTask<?>[] q;
595 <        if (base != (s = sp) && (q = queue) != null) {
596 <            int i = (q.length - 1) & --s;
597 <            ForkJoinTask<?> t = q[i];
598 <            if (t != null && UNSAFE.compareAndSwapObject
599 <                (q, (i << qShift) + qBase, t, null)) {
600 <                sp = s;
601 <                return t;
592 >    private ForkJoinTask<?> popTask() {
593 >        ForkJoinTask<?>[] q = queue;
594 >        if (q != null) {
595 >            int s;
596 >            while ((s = sp) != base) {
597 >                int i = (q.length - 1) & --s;
598 >                long u = (i << qShift) + qBase; // raw offset
599 >                ForkJoinTask<?> t = q[i];
600 >                if (t == null)   // lost to stealer
601 >                    break;
602 >                if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
603 >                    sp = s; // putOrderedInt may encourage more timely write
604 >                    // UNSAFE.putOrderedInt(this, spOffset, s);
605 >                    return t;
606 >                }
607              }
608          }
609          return null;
# Line 589 | Line 611 | public class ForkJoinWorkerThread extend
611  
612      /**
613       * Specialized version of popTask to pop only if topmost element
614 <     * is the given task. Called only by current thread while
593 <     * active.
614 >     * is the given task. Called only by this thread while active.
615       *
616       * @param t the task. Caller must ensure non-null.
617       */
618      final boolean unpushTask(ForkJoinTask<?> t) {
619          int s;
620 <        ForkJoinTask<?>[] q;
621 <        if (base != (s = sp) && (q = queue) != null &&
620 >        ForkJoinTask<?>[] q = queue;
621 >        if ((s = sp) != base && q != null &&
622              UNSAFE.compareAndSwapObject
623              (q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) {
624 <            sp = s;
624 >            sp = s; // putOrderedInt may encourage more timely write
625 >            // UNSAFE.putOrderedInt(this, spOffset, s);
626              return true;
627          }
628          return false;
629      }
630  
631      /**
632 <     * Returns next task or null if empty or contended
632 >     * Returns next task, or null if empty or contended.
633       */
634      final ForkJoinTask<?> peekTask() {
635          ForkJoinTask<?>[] q = queue;
# Line 649 | Line 671 | public class ForkJoinWorkerThread extend
671       * Computes next value for random victim probe in scan().  Scans
672       * don't require a very high quality generator, but also not a
673       * crummy one.  Marsaglia xor-shift is cheap and works well enough.
674 <     * Note: This is manually inlined in scan()
674 >     * Note: This is manually inlined in scan().
675       */
676      private static final int xorShift(int r) {
677          r ^= r << 13;
# Line 688 | Line 710 | public class ForkJoinWorkerThread extend
710              for (;;) {
711                  ForkJoinWorkerThread v = ws[k & mask];
712                  r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift
713 <                if (v != null && v.base != v.sp) {
714 <                    if (canSteal ||       // ensure active status
715 <                        (canSteal = active = p.tryIncrementActiveCount())) {
716 <                        int b = v.base;   // inline specialized deqTask
717 <                        ForkJoinTask<?>[] q;
718 <                        if (b != v.sp && (q = v.queue) != null) {
719 <                            ForkJoinTask<?> t;
720 <                            int i = (q.length - 1) & b;
721 <                            long u = (i << qShift) + qBase; // raw offset
722 <                            if ((t = q[i]) != null && v.base == b &&
723 <                                UNSAFE.compareAndSwapObject(q, u, t, null)) {
724 <                                currentSteal = t;
725 <                                v.stealHint = poolIndex;
726 <                                v.base = b + 1;
727 <                                seed = r;
728 <                                ++stealCount;
729 <                                return t;
730 <                            }
713 >                ForkJoinTask<?>[] q; ForkJoinTask<?> t; int b, a;
714 >                if (v != null && (b = v.base) != v.sp &&
715 >                    (q = v.queue) != null) {
716 >                    int i = (q.length - 1) & b;
717 >                    long u = (i << qShift) + qBase; // raw offset
718 >                    int pid = poolIndex;
719 >                    if ((t = q[i]) != null) {
720 >                        if (!canSteal &&  // inline p.tryIncrementActiveCount
721 >                            UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
722 >                                                     a = p.runState, a + 1))
723 >                            canSteal = active = true;
724 >                        if (canSteal && v.base == b++ &&
725 >                            UNSAFE.compareAndSwapObject(q, u, t, null)) {
726 >                            v.base = b;
727 >                            v.stealHint = pid;
728 >                            UNSAFE.putOrderedObject(this,
729 >                                                    currentStealOffset, t);
730 >                            seed = r;
731 >                            ++stealCount;
732 >                            return t;
733                          }
734                      }
735                      j = -n;
# Line 725 | Line 749 | public class ForkJoinWorkerThread extend
749      // Run State management
750  
751      // status check methods used mainly by ForkJoinPool
752 <    final boolean isTerminating() { return (runState & TERMINATING) != 0; }
753 <    final boolean isTerminated()  { return (runState & TERMINATED) != 0; }
754 <    final boolean isSuspended()   { return (runState & SUSPENDED) != 0; }
755 <    final boolean isTrimmed()     { return (runState & TRIMMED) != 0; }
752 >    final boolean isRunning()    { return runState == 0; }
753 >    final boolean isTerminated() { return (runState & TERMINATED) != 0; }
754 >    final boolean isSuspended()  { return (runState & SUSPENDED) != 0; }
755 >    final boolean isTrimmed()    { return (runState & TRIMMED) != 0; }
756 >
757 >    final boolean isTerminating() {
758 >        if ((runState & TERMINATING) != 0)
759 >            return true;
760 >        if (pool.isAtLeastTerminating()) { // propagate pool state
761 >            shutdown();
762 >            return true;
763 >        }
764 >        return false;
765 >    }
766  
767      /**
768 <     * Sets state to TERMINATING, also resuming if suspended.
768 >     * Sets state to TERMINATING. Does NOT unpark or interrupt
769 >     * to wake up if currently blocked. Callers must do so if desired.
770       */
771      final void shutdown() {
772          for (;;) {
773              int s = runState;
774 +            if ((s & (TERMINATING|TERMINATED)) != 0)
775 +                break;
776              if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended
777                  if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
778                                               (s & ~SUSPENDED) |
779 <                                             (TRIMMED|TERMINATING))) {
743 <                    LockSupport.unpark(this);
779 >                                             (TRIMMED|TERMINATING)))
780                      break;
745                }
781              }
782              else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
783                                                s | TERMINATING))
# Line 751 | Line 786 | public class ForkJoinWorkerThread extend
786      }
787  
788      /**
789 <     * Sets state to TERMINATED. Called only by this thread.
789 >     * Sets state to TERMINATED. Called only by onTermination().
790       */
791      private void setTerminated() {
792          int s;
# Line 761 | Line 796 | public class ForkJoinWorkerThread extend
796      }
797  
798      /**
799 <     * Instrumented version of park used by ForkJoinPool.awaitEvent
800 <     */
766 <    final void doPark() {
767 <        ++parkCount;
768 <        LockSupport.park(this);
769 <    }
770 <
771 <    /**
772 <     * If suspended, tries to set status to unsuspended and unparks.
799 >     * If suspended, tries to set status to unsuspended.
800 >     * Does NOT wake up if blocked.
801       *
802       * @return true if successful
803       */
804 <    final boolean tryResumeSpare() {
805 <        int s = runState;
806 <        if ((s & SUSPENDED) != 0 &&
807 <            UNSAFE.compareAndSwapInt(this, runStateOffset, s,
808 <                                     s & ~SUSPENDED)) {
809 <            LockSupport.unpark(this);
782 <            return true;
804 >    final boolean tryUnsuspend() {
805 >        int s;
806 >        while (((s = runState) & SUSPENDED) != 0) {
807 >            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
808 >                                         s & ~SUSPENDED))
809 >                return true;
810          }
811          return false;
812      }
813  
814      /**
815 <     * Sets suspended status and blocks as spare until resumed,
816 <     * shutdown, or timed out.
790 <     *
791 <     * @return false if trimmed
815 >     * Sets suspended status and blocks as spare until resumed
816 >     * or shutdown.
817       */
818 <    final boolean suspendAsSpare() {
819 <        for (;;) {               // set suspended unless terminating
818 >    final void suspendAsSpare() {
819 >        for (;;) {                  // set suspended unless terminating
820              int s = runState;
821              if ((s & TERMINATING) != 0) { // must kill
822                  if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
823                                               s | (TRIMMED | TERMINATING)))
824 <                    return false;
824 >                    return;
825              }
826              else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
827                                                s | SUSPENDED))
828                  break;
829          }
830 <        boolean timed;
831 <        long nanos;
807 <        long startTime;
808 <        if (poolIndex < pool.parallelism) {
809 <            timed = false;
810 <            nanos = 0L;
811 <            startTime = 0L;
812 <        }
813 <        else {
814 <            timed = true;
815 <            nanos = SPARE_KEEPALIVE_NANOS;
816 <            startTime = System.nanoTime();
817 <        }
818 <        pool.accumulateStealCount(this);
819 <        lastEventCount = 0;      // reset upon resume
820 <        interrupted();           // clear/ignore interrupts
830 >        ForkJoinPool p = pool;
831 >        p.pushSpare(this);
832          while ((runState & SUSPENDED) != 0) {
833 <            ++parkCount;
834 <            if (!timed)
833 >            if (p.tryAccumulateStealCount(this)) {
834 >                interrupted();          // clear/ignore interrupts
835 >                if ((runState & SUSPENDED) == 0)
836 >                    break;
837                  LockSupport.park(this);
825            else if ((nanos -= (System.nanoTime() - startTime)) > 0)
826                LockSupport.parkNanos(this, nanos);
827            else { // try to trim on timeout
828                int s = runState;
829                if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
830                                             (s & ~SUSPENDED) |
831                                             (TRIMMED|TERMINATING)))
832                    return false;
838              }
839          }
835        return true;
840      }
841  
842      // Misc support methods for ForkJoinPool
# Line 842 | Line 846 | public class ForkJoinWorkerThread extend
846       * used by ForkJoinTask.
847       */
848      final int getQueueSize() {
849 <        return -base + sp;
849 >        int n; // external calls must read base first
850 >        return (n = -base + sp) <= 0 ? 0 : n;
851      }
852  
853      /**
# Line 850 | Line 855 | public class ForkJoinWorkerThread extend
855       * thread.
856       */
857      final void cancelTasks() {
858 <        ForkJoinTask<?> cj = currentJoin; // try to kill live tasks
858 >        ForkJoinTask<?> cj = currentJoin; // try to cancel ongoing tasks
859          if (cj != null) {
860              currentJoin = null;
861              cj.cancelIgnoringExceptions();
862 +            try {
863 +                this.interrupt(); // awaken wait
864 +            } catch (SecurityException ignore) {
865 +            }
866          }
867          ForkJoinTask<?> cs = currentSteal;
868          if (cs != null) {
# Line 892 | Line 901 | public class ForkJoinWorkerThread extend
901       * @return a task, if available
902       */
903      final ForkJoinTask<?> pollLocalTask() {
904 +        ForkJoinPool p = pool;
905          while (sp != base) {
906 <            if (active || (active = pool.tryIncrementActiveCount()))
907 <                return locallyFifo? locallyDeqTask() : popTask();
906 >            int a; // inline p.tryIncrementActiveCount
907 >            if (active ||
908 >                (active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
909 >                                                   a = p.runState, a + 1)))
910 >                return locallyFifo ? locallyDeqTask() : popTask();
911          }
912          return null;
913      }
# Line 905 | Line 918 | public class ForkJoinWorkerThread extend
918       * @return a task, if available
919       */
920      final ForkJoinTask<?> pollTask() {
921 <        ForkJoinTask<?> t;
922 <        return (t = pollLocalTask()) != null ? t : scan();
921 >        ForkJoinTask<?> t = pollLocalTask();
922 >        if (t == null) {
923 >            t = scan();
924 >            // cannot retain/track/help steal
925 >            UNSAFE.putOrderedObject(this, currentStealOffset, null);
926 >        }
927 >        return t;
928      }
929  
930      /**
931       * Possibly runs some tasks and/or blocks, until task is done.
914     * The main body is basically a big spinloop, alternating between
915     * calls to helpJoinTask and pool.tryAwaitJoin with increased
916     * patience parameters until either the task is done without
917     * waiting, or we have, if necessary, created or resumed a
918     * replacement for this thread while it blocks.
932       *
933       * @param joinMe the task to join
934 <     * @return task status on exit
934 >     * @param timed true if use timed wait
935 >     * @param nanos wait time if timed
936       */
937 <    final int joinTask(ForkJoinTask<?> joinMe) {
938 <        int stat;
937 >    final void joinTask(ForkJoinTask<?> joinMe, boolean timed, long nanos) {
938 >        // currentJoin only written by this thread; only need ordered store
939          ForkJoinTask<?> prevJoin = currentJoin;
940 <        currentJoin = joinMe;
941 <        if ((stat = joinMe.status) >= 0 &&
942 <            (sp == base || (stat = localHelpJoinTask(joinMe)) >= 0)) {
943 <            ForkJoinPool p = pool;
944 <            int helpRetries = 2;     // initial patience values
945 <            int awaitRetries = -1;   // -1 is sentinel for replace-check only
932 <            do {
933 <                helpJoinTask(joinMe, helpRetries);
934 <                if ((stat = joinMe.status) < 0)
935 <                    break;
936 <                boolean busy = p.tryAwaitJoin(joinMe, awaitRetries);
937 <                if ((stat = joinMe.status) < 0)
938 <                    break;
939 <                if (awaitRetries == -1)
940 <                    awaitRetries = 0;
941 <                else if (busy)
942 <                    ++awaitRetries;
943 <                if (helpRetries < p.parallelism)
944 <                    helpRetries <<= 1;
945 <                Thread.yield(); // tame unbounded loop
946 <            } while (joinMe.status >= 0);
947 <        }
948 <        currentJoin = prevJoin;
949 <        return stat;
940 >        UNSAFE.putOrderedObject(this, currentJoinOffset, joinMe);
941 >        if (isTerminating())                // cancel if shutting down
942 >            joinMe.cancelIgnoringExceptions();
943 >        else
944 >            pool.awaitJoin(joinMe, this, timed, nanos);
945 >        UNSAFE.putOrderedObject(this, currentJoinOffset, prevJoin);
946      }
947  
948      /**
949       * Run tasks in local queue until given task is done.
950 +     * Not currently used because it complicates semantics.
951       *
952       * @param joinMe the task to join
956     * @return task status on exit
953       */
954 <    private int localHelpJoinTask(ForkJoinTask<?> joinMe) {
955 <        int stat, s;
954 >    private void localHelpJoinTask(ForkJoinTask<?> joinMe) {
955 >        int s;
956          ForkJoinTask<?>[] q;
957 <        while ((stat = joinMe.status) >= 0 &&
962 <               base != (s = sp) && (q = queue) != null) {
963 <            ForkJoinTask<?> t;
957 >        while (joinMe.status >= 0 && (s = sp) != base && (q = queue) != null) {
958              int i = (q.length - 1) & --s;
959              long u = (i << qShift) + qBase; // raw offset
960 <            if ((t = q[i]) != null &&
961 <                UNSAFE.compareAndSwapObject(q, u, t, null)) {
960 >            ForkJoinTask<?> t = q[i];
961 >            if (t == null)  // lost to a stealer
962 >                break;
963 >            if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
964                  /*
965                   * This recheck (and similarly in helpJoinTask)
966                   * handles cases where joinMe is independently
# Line 972 | Line 968 | public class ForkJoinWorkerThread extend
968                   * available. Back out of the pop by putting t back
969                   * into slot before we commit by writing sp.
970                   */
971 <                if ((stat = joinMe.status) < 0) {
971 >                if (joinMe.status < 0) {
972                      UNSAFE.putObjectVolatile(q, u, t);
973                      break;
974                  }
975                  sp = s;
976 <                t.tryExec();
976 >                // UNSAFE.putOrderedInt(this, spOffset, s);
977 >                t.quietlyExec();
978              }
979          }
983        return stat;
980      }
981  
982      /**
# Line 988 | Line 984 | public class ForkJoinWorkerThread extend
984       * given task, or in turn one of its stealers.  Traces
985       * currentSteal->currentJoin links looking for a thread working on
986       * a descendant of the given task and with a non-empty queue to
987 <     * steal back and execute tasks from. Restarts search upon
988 <     * encountering chains that are stale, unknown, or of length
989 <     * greater than MAX_HELP_DEPTH links, to avoid unbounded cycles.
990 <     *
991 <     * The implementation is very branchy to cope with the restart
992 <     * cases.  Returns void, not task status (which must be reread by
993 <     * caller anyway) to slightly simplify control paths.
987 >     * steal back and execute tasks from.
988 >     *
989 >     * The implementation is very branchy to cope with potential
990 >     * inconsistencies or loops encountering chains that are stale,
991 >     * unknown, or of length greater than MAX_HELP_DEPTH links.  All
992 >     * of these cases are dealt with by just returning back to the
993 >     * caller, who is expected to retry if other join mechanisms also
994 >     * don't work out.
995       *
996       * @param joinMe the task to join
997       */
998 <    final void helpJoinTask(ForkJoinTask<?> joinMe, int retries) {
999 <        ForkJoinWorkerThread[] ws = pool.workers;
998 >    final void helpJoinTask(ForkJoinTask<?> joinMe) {
999 >        ForkJoinWorkerThread[] ws;
1000          int n;
1001 <        if (ws == null || (n = ws.length) <= 1)
1002 <            return;                   // need at least 2 workers
1003 <
1004 <        restart:while (joinMe.status >= 0 && --retries >= 0) {
1005 <            ForkJoinTask<?> task = joinMe;        // base of chain
1006 <            ForkJoinWorkerThread thread = this;   // thread with stolen task
1007 <            for (int depth = 0; depth < MAX_HELP_DEPTH; ++depth) {
1008 <                // Try to find v, the stealer of task, by first using hint
1009 <                ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)];
1010 <                if (v == null || v.currentSteal != task) {
1011 <                    for (int j = 0; ; ++j) {      // search array
1012 <                        if (task.status < 0 || j == n)
1013 <                            continue restart;     // stale or no stealer
1014 <                        if ((v = ws[j]) != null && v.currentSteal == task) {
1015 <                            thread.stealHint = j; // save for next time
1016 <                            break;
1017 <                        }
1018 <                    }
1019 <                }
1020 <                // Try to help v, using specialized form of deqTask
1021 <                int b;
1025 <                ForkJoinTask<?>[] q;
1026 <                while ((b = v.base) != v.sp && (q = v.queue) != null) {
1027 <                    int i = (q.length - 1) & b;
1028 <                    long u = (i << qShift) + qBase;
1029 <                    ForkJoinTask<?> t = q[i];
1030 <                    if (task.status < 0)          // stale
1031 <                        continue restart;
1032 <                    if (v.base == b) {            // recheck after reading t
1033 <                        if (t == null)            // producer stalled
1034 <                            continue restart;     // retry via restart
1035 <                        if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
1036 <                            if (joinMe.status < 0) {
1037 <                                UNSAFE.putObjectVolatile(q, u, t);
1038 <                                return;           // back out on cancel
1001 >        if (joinMe.status < 0)                // already done
1002 >            return;
1003 >        if ((ws = pool.workers) == null || (n = ws.length) <= 1)
1004 >            return;                           // need at least 2 workers
1005 >
1006 >        ForkJoinTask<?> task = joinMe;        // base of chain
1007 >        ForkJoinWorkerThread thread = this;   // thread with stolen task
1008 >        for (int d = 0; d < MAX_HELP_DEPTH; ++d) { // chain length
1009 >            // Try to find v, the stealer of task, by first using hint
1010 >            ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)];
1011 >            if (v == null || v.currentSteal != task) {
1012 >                for (int j = 0; ; ++j) {      // search array
1013 >                    if (j < n) {
1014 >                        ForkJoinTask<?> vs;
1015 >                        if ((v = ws[j]) != null && v != this &&
1016 >                            (vs = v.currentSteal) != null) {
1017 >                            if (joinMe.status < 0 || task.status < 0)
1018 >                                return;       // stale or done
1019 >                            if (vs == task) {
1020 >                                thread.stealHint = j;
1021 >                                break;        // save hint for next time
1022                              }
1040                            ForkJoinTask<?> prevSteal = currentSteal;
1041                            currentSteal = t;
1042                            v.stealHint = poolIndex;
1043                            v.base = b + 1;
1044                            t.tryExec();
1045                            currentSteal = prevSteal;
1023                          }
1024                      }
1025 <                    if (joinMe.status < 0)
1026 <                        return;
1025 >                    else
1026 >                        return;               // no stealer
1027                  }
1028 <                // Try to descend to find v's stealer
1029 <                ForkJoinTask<?> next = v.currentJoin;
1053 <                if (next == null || task.status < 0)
1054 <                    continue restart;             // no descendent or stale
1028 >            }
1029 >            for (;;) { // Try to help v, using specialized form of deqTask
1030                  if (joinMe.status < 0)
1031                      return;
1032 <                task = next;
1033 <                thread = v;
1032 >                int b = v.base;
1033 >                ForkJoinTask<?>[] q = v.queue;
1034 >                if (b == v.sp || q == null)
1035 >                    break;
1036 >                int i = (q.length - 1) & b;
1037 >                long u = (i << qShift) + qBase;
1038 >                ForkJoinTask<?> t = q[i];
1039 >                int pid = poolIndex;
1040 >                ForkJoinTask<?> ps = currentSteal;
1041 >                if (task.status < 0)
1042 >                    return;                   // stale or done
1043 >                if (t != null && v.base == b++ &&
1044 >                    UNSAFE.compareAndSwapObject(q, u, t, null)) {
1045 >                    if (joinMe.status < 0) {
1046 >                        UNSAFE.putObjectVolatile(q, u, t);
1047 >                        return;               // back out on cancel
1048 >                    }
1049 >                    v.base = b;
1050 >                    v.stealHint = pid;
1051 >                    UNSAFE.putOrderedObject(this, currentStealOffset, t);
1052 >                    t.quietlyExec();
1053 >                    UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1054 >                }
1055              }
1056 +            // Try to descend to find v's stealer
1057 +            ForkJoinTask<?> next = v.currentJoin;
1058 +            if (task.status < 0 || next == null || next == task ||
1059 +                joinMe.status < 0)
1060 +                return;
1061 +            task = next;
1062 +            thread = v;
1063          }
1064      }
1065  
1066      /**
1067 +     * Implements ForkJoinTask.getSurplusQueuedTaskCount().
1068       * Returns an estimate of the number of tasks, offset by a
1069       * function of number of idle workers.
1070       *
# Line 1115 | Line 1119 | public class ForkJoinWorkerThread extend
1119       * Runs tasks until {@code pool.isQuiescent()}.
1120       */
1121      final void helpQuiescePool() {
1122 +        ForkJoinTask<?> ps = currentSteal; // to restore below
1123          for (;;) {
1124              ForkJoinTask<?> t = pollLocalTask();
1125 <            if (t != null || (t = scan()) != null) {
1126 <                t.tryExec();
1122 <                currentSteal = null;
1123 <            }
1125 >            if (t != null || (t = scan()) != null)
1126 >                t.quietlyExec();
1127              else {
1128                  ForkJoinPool p = pool;
1129 +                int a; // to inline CASes
1130                  if (active) {
1131 +                    if (!UNSAFE.compareAndSwapInt
1132 +                        (p, poolRunStateOffset, a = p.runState, a - 1))
1133 +                        continue;   // retry later
1134                      active = false; // inactivate
1135 <                    do {} while (!p.tryDecrementActiveCount());
1135 >                    UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1136                  }
1137                  if (p.isQuiescent()) {
1138                      active = true; // re-activate
1139 <                    do {} while (!p.tryIncrementActiveCount());
1139 >                    do {} while (!UNSAFE.compareAndSwapInt
1140 >                                 (p, poolRunStateOffset, a = p.runState, a+1));
1141                      return;
1142                  }
1143              }
# Line 1139 | Line 1147 | public class ForkJoinWorkerThread extend
1147      // Unsafe mechanics
1148  
1149      private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1150 +    private static final long spOffset =
1151 +        objectFieldOffset("sp", ForkJoinWorkerThread.class);
1152      private static final long runStateOffset =
1153          objectFieldOffset("runState", ForkJoinWorkerThread.class);
1154 +    private static final long currentJoinOffset =
1155 +        objectFieldOffset("currentJoin", ForkJoinWorkerThread.class);
1156 +    private static final long currentStealOffset =
1157 +        objectFieldOffset("currentSteal", ForkJoinWorkerThread.class);
1158      private static final long qBase =
1159          UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1160 +    private static final long poolRunStateOffset = // to inline CAS
1161 +        objectFieldOffset("runState", ForkJoinPool.class);
1162 +
1163      private static final int qShift;
1164  
1165      static {
# Line 1150 | Line 1167 | public class ForkJoinWorkerThread extend
1167          if ((s & (s-1)) != 0)
1168              throw new Error("data type scale not a power of two");
1169          qShift = 31 - Integer.numberOfLeadingZeros(s);
1170 +        MAXIMUM_QUEUE_CAPACITY = 1 << (31 - qShift);
1171      }
1172  
1173      private static long objectFieldOffset(String field, Class<?> klazz) {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines