ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/jsr166/jsr166/src/jsr166y/ForkJoinPool.java

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.65 by dl, Wed Aug 18 14:05:27 2010 UTC vs.
Revision 1.66 by dl, Sun Aug 29 23:34:46 2010 UTC

#	Line 161 | Line 161 | public class ForkJoinPool extends Abstra
161		*      re-activate a spare thread to compensate for blocked
162		*      joiners until they unblock.
163		*
164	<	* Because the determining existence of conservatively safe
165	<	* helping targets, the availability of already-created spares,
166	<	* and the apparent need to create new spares are all racy and
167	<	* require heuristic guidance, we rely on multiple retries of
168	<	* each. Further, because it is impossible to keep exactly the
169	<	* target (parallelism) number of threads running at any given
170	<	* time, we allow compensation during joins to fail, and enlist
171	<	* all other threads to help out whenever they are not otherwise
172	<	* occupied (i.e., mainly in method preStep).
164	>	* It is impossible to keep exactly the target (parallelism)
165	>	* number of threads running at any given time.  Determining
166	>	* existence of conservatively safe helping targets, the
167	>	* availability of already-created spares, and the apparent need
168	>	* to create new spares are all racy and require heuristic
169	>	* guidance, so we rely on multiple retries of each.  Compensation
170	>	* occurs in slow-motion. It is triggered only upon timeouts of
171	>	* Object.wait used for joins. This reduces poor decisions that
172	>	* would otherwise be made when threads are waiting for others
173	>	* that are stalled because of unrelated activities such as
174	>	* garbage collection.
175		*
176		* The ManagedBlocker extension API can't use helping so relies
177		* only on compensation in method awaitBlocker.
#	Line 271 | Line 273 | public class ForkJoinPool extends Abstra
273		* In addition to allowing simpler decisions about need for
274		* wakeup, the event count bits in eventWaiters serve the role of
275		* tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
276	<	* released threads also try to release others (but give up upon
277	<	* contention to reduce useless flailing).  The net effect is a
278	<	* tree-like diffusion of signals, where released threads (and
279	<	* possibly others) help with unparks.  To further reduce
280	<	* contention effects a bit, failed CASes to increment field
279	<	* eventCount are tolerated without retries in signalWork.
276	>	* released threads also try to release at most two others.  The
277	>	* net effect is a tree-like diffusion of signals, where released
278	>	* threads (and possibly others) help with unparks.  To further
279	>	* reduce contention effects a bit, failed CASes to increment
280	>	* field eventCount are tolerated without retries in signalWork.
281		* Conceptually they are merged into the same event, which is OK
282		* when their only purpose is to enable workers to scan for work.
283		*
284	<	* 5. Managing suspension of extra workers. When a worker is about
285	<	* to block waiting for a join (or via ManagedBlockers), we may
286	<	* create a new thread to maintain parallelism level, or at least
287	<	* avoid starvation. Usually, extra threads are needed for only
288	<	* very short periods, yet join dependencies are such that we
289	<	* sometimes need them in bursts. Rather than create new threads
290	<	* each time this happens, we suspend no-longer-needed extra ones
291	<	* as "spares". For most purposes, we don't distinguish "extra"
292	<	* spare threads from normal "core" threads: On each call to
293	<	* preStep (the only point at which we can do this) a worker
294	<	* checks to see if there are now too many running workers, and if
295	<	* so, suspends itself.  Method helpMaintainParallelism looks for
296	<	* suspended threads to resume before considering creating a new
297	<	* replacement. The spares themselves are encoded on another
298	<	* variant of a Treiber Stack, headed at field "spareWaiters".
299	<	* Note that the use of spares is intrinsically racy.  One thread
300	<	* may become a spare at about the same time as another is
301	<	* needlessly being created. We counteract this and related slop
302	<	* in part by requiring resumed spares to immediately recheck (in
303	<	* preStep) to see whether they they should re-suspend.
304	<	*
305	<	* 6. Killing off unneeded workers. The Spare and Event queues use
306	<	* similar mechanisms to shed unused workers: The oldest (first)
307	<	* waiter uses a timed rather than hard wait. When this wait times
308	<	* out without a normal wakeup, it tries to shutdown any one (for
309	<	* convenience the newest) other waiter via tryShutdownSpare or
310	<	* tryShutdownWaiter, respectively. The wakeup rates for spares
311	<	* are much shorter than for waiters. Together, they will
312	<	* eventually reduce the number of worker threads to a minimum of
313	<	* one after a long enough period without use.
284	>	* 5. Managing suspension of extra workers. When a worker notices
285	>	* (usually upon timeout of a wait()) that there are too few
286	>	* running threads, we may create a new thread to maintain
287	>	* parallelism level, or at least avoid starvation. Usually, extra
288	>	* threads are needed for only very short periods, yet join
289	>	* dependencies are such that we sometimes need them in
290	>	* bursts. Rather than create new threads each time this happens,
291	>	* we suspend no-longer-needed extra ones as "spares". For most
292	>	* purposes, we don't distinguish "extra" spare threads from
293	>	* normal "core" threads: On each call to preStep (the only point
294	>	* at which we can do this) a worker checks to see if there are
295	>	* now too many running workers, and if so, suspends itself.
296	>	* Method helpMaintainParallelism looks for suspended threads to
297	>	* resume before considering creating a new replacement. The
298	>	* spares themselves are encoded on another variant of a Treiber
299	>	* Stack, headed at field "spareWaiters".  Note that the use of
300	>	* spares is intrinsically racy.  One thread may become a spare at
301	>	* about the same time as another is needlessly being created. We
302	>	* counteract this and related slop in part by requiring resumed
303	>	* spares to immediately recheck (in preStep) to see whether they
304	>	* they should re-suspend.
305	>	*
306	>	* 6. Killing off unneeded workers. A timeout mechanism is used to
307	>	* shed unused workers: The oldest (first) event queue waiter uses
308	>	* a timed rather than hard wait. When this wait times out without
309	>	* a normal wakeup, it tries to shutdown any one (for convenience
310	>	* the newest) other spare or event waiter via
311	>	* tryShutdownUnusedWorker. This eventually reduces the number of
312	>	* worker threads to a minimum of one after a long enough period
313	>	* without use.
314		*
315		* 7. Deciding when to create new workers. The main dynamic
316		* control in this class is deciding when to create extra threads
#	Line 322 | Line 323 | public class ForkJoinPool extends Abstra
323		* compilation, and wake-up lags. These transients are extremely
324		* common -- we are normally trying to fully saturate the CPUs on
325		* a machine, so almost any activity other than running tasks
326	<	* impedes accuracy. Our main defense is to allow some slack in
327	<	* creation thresholds, using rules that reflect the fact that the
328	<	* more threads we have running, the more likely that we are
329	<	* underestimating the number running threads. (We also include
330	<	* some heuristic use of Thread.yield when all workers appear to
331	<	* be busy, to improve likelihood of counts settling.) The rules
332	<	* also better cope with the fact that some of the methods in this
333	<	* class tend to never become compiled (but are interpreted), so
334	<	* some components of the entire set of controls might execute 100
334	<	* times faster than others. And similarly for cases where the
335	<	* apparent lack of work is just due to GC stalls and other
336	<	* transient system activity.
326	>	* impedes accuracy. Our main defense is to allow parallelism to
327	>	* lapse for a while during joins, and use a timeout to see if,
328	>	* after the resulting settling, there is still a need for
329	>	* additional workers.  This also better copes with the fact that
330	>	* some of the methods in this class tend to never become compiled
331	>	* (but are interpreted), so some components of the entire set of
332	>	* controls might execute 100 times faster than others. And
333	>	* similarly for cases where the apparent lack of work is just due
334	>	* to GC stalls and other transient system activity.
335		*
336		* Beware that there is a lot of representation-level coupling
337		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 420 | Line 418 | public class ForkJoinPool extends Abstra
418		new AtomicInteger();
419
420		/**
421	+	* The time to block in a join (see awaitJoin) before checking if
422	+	* a new worker should be (re)started to maintain parallelism
423	+	* level. The value should be short enough to maintain gloabal
424	+	* responsiveness and progress but long enough to avoid
425	+	* counterproductive firings during GC stalls or unrelated system
426	+	* activity, and to not bog down systems with continual re-firings
427	+	* on GCs or legitimately long waits.
428	+	*/
429	+	private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
430	+
431	+	/**
432		* The wakeup interval (in nanoseconds) for the oldest worker
433	<	* worker waiting for an event invokes tryShutdownWaiter to shrink
433	>	* worker waiting for an event invokes tryShutdownUnusedWorker to shrink
434		* the number of workers.  The exact value does not matter too
435		* much, but should be long enough to slowly release resources
436		* during long periods without use without disrupting normal use.
437		*/
438		private static final long SHRINK_RATE_NANOS =
439	<	60L * 1000L * 1000L * 1000L; // one minute
439	>	30L * 1000L * 1000L * 1000L; // 2 per minute
440
441		/**
442		* Absolute bound for parallelism level. Twice this number plus
#	Line 567 | Line 576 | public class ForkJoinPool extends Abstra
576		*/
577		private final int poolNumber;
578
570	–
579		// Utilities for CASing fields. Note that most of these
580		// are usually manually inlined by callers
581
#	Line 615 | Line 623 | public class ForkJoinPool extends Abstra
623		}
624
625		/**
618	–	* Increments event count
619	–	*/
620	–	private void advanceEventCount() {
621	–	int c;
622	–	do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset,
623	–	c = eventCount, c+1));
624	–	}
625	–
626	–	/**
627	–	* Tries incrementing active count; fails on contention.
628	–	* Called by workers before executing tasks.
629	–	*
630	–	* @return true on success
631	–	*/
632	–	final boolean tryIncrementActiveCount() {
633	–	int c;
634	–	return UNSAFE.compareAndSwapInt(this, runStateOffset,
635	–	c = runState, c + 1);
636	–	}
637	–
638	–	/**
626		* Tries decrementing active count; fails on contention.
627		* Called when workers cannot find tasks to run.
628		*/
#	Line 703 | Line 690 | public class ForkJoinPool extends Abstra
690		}
691		}
692
706	–	// adding and removing workers
707	–
708	–	/**
709	–	* Tries to create and add new worker. Assumes that worker counts
710	–	* are already updated to accommodate the worker, so adjusts on
711	–	* failure.
712	–	*
713	–	* @return the worker, or null on failure
714	–	*/
715	–	private ForkJoinWorkerThread addWorker() {
716	–	ForkJoinWorkerThread w = null;
717	–	try {
718	–	w = factory.newThread(this);
719	–	} finally { // Adjust on either null or exceptional factory return
720	–	if (w == null) {
721	–	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
722	–	tryTerminate(false); // in case of failure during shutdown
723	–	}
724	–	}
725	–	if (w != null) {
726	–	w.start(recordWorker(w), ueh);
727	–	advanceEventCount();
728	–	}
729	–	return w;
730	–	}
731	–
693		/**
694		* Final callback from terminating worker.  Removes record of
695		* worker from array, and adjusts counts. If pool is shutting
#	Line 750 | Line 711 | public class ForkJoinPool extends Abstra
711		* Releases workers blocked on a count not equal to current count.
712		* Normally called after precheck that eventWaiters isn't zero to
713		* avoid wasted array checks. Gives up upon a change in count or
714	<	* contention, letting other workers take over.
714	>	* upon releasing two workers, letting others take over.
715		*/
716		private void releaseEventWaiters() {
717		ForkJoinWorkerThread[] ws = workers;
718		int n = ws.length;
719		long h = eventWaiters;
720		int ec = eventCount;
721	+	boolean releasedOne = false;
722		ForkJoinWorkerThread w; int id;
723	<	while ((int)(h >>> EVENT_COUNT_SHIFT) != ec &&
724	<	(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
725	<	id < n && (w = ws[id]) != null &&
726	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
727	<	h,  h = w.nextWaiter)) {
728	<	LockSupport.unpark(w);
729	<	if (eventWaiters != h || eventCount != ec)
723	>	while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
724	>	(int)(h >>> EVENT_COUNT_SHIFT) != ec &&
725	>	id < n && (w = ws[id]) != null) {
726	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
727	>	h,  w.nextWaiter)) {
728	>	LockSupport.unpark(w);
729	>	if (releasedOne) // exit on second release
730	>	break;
731	>	releasedOne = true;
732	>	}
733	>	if (eventCount != ec)
734		break;
735	+	h = eventWaiters;
736		}
737		}
738
#	Line 782 | Line 749 | public class ForkJoinPool extends Abstra
749
750		/**
751		* Adds the given worker to event queue and blocks until
752	<	* terminating or event count advances from the workers
786	<	* lastEventCount value
752	>	* terminating or event count advances from the given value
753		*
754		* @param w the calling worker thread
755	+	* @param ec the count
756		*/
757	<	private void eventSync(ForkJoinWorkerThread w) {
791	<	int ec = w.lastEventCount;
757	>	private void eventSync(ForkJoinWorkerThread w, int ec) {
758		long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
759		long h;
760		while ((runState < SHUTDOWN || !tryTerminate(false)) &&
#	Line 808 | Line 774 | public class ForkJoinPool extends Abstra
774		* event waiter) until terminating or event count advances from
775		* the given value. The oldest (first) waiter uses a timed wait to
776		* occasionally one-by-one shrink the number of workers (to a
777	<	* minumum of one) if the pool has not been used for extended
777	>	* minimum of one) if the pool has not been used for extended
778		* periods.
779		*
780		* @param w the calling worker thread
#	Line 821 | Line 787 | public class ForkJoinPool extends Abstra
787		(workerCounts & RUNNING_COUNT_MASK) <= 1);
788		long startTime = untimed? 0 : System.nanoTime();
789		Thread.interrupted();         // clear/ignore interrupt
790	<	if (eventCount != ec || !w.isRunning() ||
790	>	if (eventCount != ec || w.runState != 0 ||
791		runState >= TERMINATING)  // recheck after clear
792		break;
793		if (untimed)
794		LockSupport.park(w);
795		else {
796		LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
797	<	if (eventCount != ec || !w.isRunning() ||
797	>	if (eventCount != ec || w.runState != 0 ||
798		runState >= TERMINATING)
799		break;
800		if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
801	<	tryShutdownWaiter(ec);
801	>	tryShutdownUnusedWorker(ec);
802		}
803		}
804		}
805		}
806
807	<	/**
842	<	* Callback from the oldest waiter in awaitEvent waking up after a
843	<	* period of non-use. Tries (once) to shutdown an event waiter (or
844	<	* a spare, if one exists). Note that we don't need CAS or locks
845	<	* here because the method is called only from one thread
846	<	* occasionally waking (and even misfires are OK). Note that
847	<	* until the shutdown worker fully terminates, workerCounts
848	<	* will overestimate total count, which is tolerable.
849	<	*
850	<	* @param ec the event count waited on by caller (to abort
851	<	* attempt if count has since changed).
852	<	*/
853	<	private void tryShutdownWaiter(int ec) {
854	<	if (spareWaiters != 0) { // prefer killing spares
855	<	tryShutdownSpare();
856	<	return;
857	<	}
858	<	ForkJoinWorkerThread[] ws = workers;
859	<	int n = ws.length;
860	<	long h = eventWaiters;
861	<	ForkJoinWorkerThread w; int id; long nh;
862	<	if (runState == 0 &&
863	<	submissionQueue.isEmpty() &&
864	<	eventCount == ec &&
865	<	(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
866	<	id < n && (w = ws[id]) != null &&
867	<	(nh = w.nextWaiter) != 0L && // keep at least one worker
868	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) {
869	<	w.shutdown();
870	<	LockSupport.unpark(w);
871	<	}
872	<	releaseEventWaiters();
873	<	}
874	<
875	<	// Maintaining spares
807	>	// Maintaining parallelism
808
809		/**
810		* Pushes worker onto the spare stack
#	Line 884 | Line 816 | public class ForkJoinPool extends Abstra
816		}
817
818		/**
819	<	* Callback from oldest spare occasionally waking up.  Tries
820	<	* (once) to shutdown a spare. Same idea as tryShutdownWaiter.
819	>	* Tries (once) to resume a spare if the number of running
820	>	* threads is less than target.
821		*/
822	<	final void tryShutdownSpare() {
822	>	private void tryResumeSpare() {
823		int sw, id;
892	–	ForkJoinWorkerThread w;
893	–	ForkJoinWorkerThread[] ws;
894	–	if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
895	–	id < (ws = workers).length && (w = ws[id]) != null &&
896	–	(workerCounts & RUNNING_COUNT_MASK) >= parallelism &&
897	–	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
898	–	sw, w.nextSpare)) {
899	–	w.shutdown();
900	–	LockSupport.unpark(w);
901	–	advanceEventCount();
902	–	}
903	–	}
904	–
905	–	/**
906	–	* Tries (once) to resume a spare if worker counts match
907	–	* the given count.
908	–	*
909	–	* @param wc workerCounts value on invocation of this method
910	–	*/
911	–	private void tryResumeSpare(int wc) {
824		ForkJoinWorkerThread[] ws = workers;
825		int n = ws.length;
826	<	int sw, id, rs;  ForkJoinWorkerThread w;
827	<	if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
826	>	ForkJoinWorkerThread w;
827	>	if ((sw = spareWaiters) != 0 &&
828	>	(id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
829		id < n && (w = ws[id]) != null &&
830	<	(rs = runState) < TERMINATING &&
831	<	eventWaiters == 0L && workerCounts == wc) {
832	<	// In case all workers busy, heuristically back off to let settle
833	<	Thread.yield();
834	<	if (eventWaiters == 0L && runState == rs && // recheck
835	<	workerCounts == wc && spareWaiters == sw &&
836	<	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
837	<	sw, w.nextSpare)) {
838	<	int c;              // increment running count before resume
839	<	do {} while(!UNSAFE.compareAndSwapInt
840	<	(this, workerCountsOffset,
841	<	c = workerCounts, c + ONE_RUNNING));
929	<	if (w.tryUnsuspend())
930	<	LockSupport.unpark(w);
931	<	else               // back out if w was shutdown
932	<	decrementWorkerCounts(ONE_RUNNING, 0);
933	<	}
830	>	(workerCounts & RUNNING_COUNT_MASK) < parallelism &&
831	>	spareWaiters == sw &&
832	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
833	>	sw, w.nextSpare)) {
834	>	int c; // increment running count before resume
835	>	do {} while(!UNSAFE.compareAndSwapInt
836	>	(this, workerCountsOffset,
837	>	c = workerCounts, c + ONE_RUNNING));
838	>	if (w.tryUnsuspend())
839	>	LockSupport.unpark(w);
840	>	else   // back out if w was shutdown
841	>	decrementWorkerCounts(ONE_RUNNING, 0);
842		}
843		}
844
937	–	// adding workers on demand
938	–
845		/**
846	<	* Adds one or more workers if needed to establish target parallelism.
847	<	* Retries upon contention.
846	>	* Tries to increase the number of running workers if below target
847	>	* parallelism: If a spare exists tries to resume it via
848	>	* tryResumeSpare.  Otherwise, if not enough total workers or all
849	>	* existing workers are busy, adds a new worker. In all casses also
850	>	* helps wake up releasable workers waiting for work.
851		*/
852	<	private void addWorkerIfBelowTarget() {
852	>	private void helpMaintainParallelism() {
853		int pc = parallelism;
854	<	int wc;
855	<	while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < pc &&
856	<	runState < TERMINATING) {
857	<	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
858	<	wc + (ONE_RUNNING|ONE_TOTAL))) {
859	<	if (addWorker() == null)
854	>	int wc, rs, tc;
855	>	while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
856	>	(rs = runState) < TERMINATING) {
857	>	if (spareWaiters != 0)
858	>	tryResumeSpare();
859	>	else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
860	>	(tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
861	>	break;   // enough total
862	>	else if (runState == rs && workerCounts == wc &&
863	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
864	>	wc + (ONE_RUNNING|ONE_TOTAL))) {
865	>	ForkJoinWorkerThread w = null;
866	>	try {
867	>	w = factory.newThread(this);
868	>	} finally { // adjust on null or exceptional factory return
869	>	if (w == null) {
870	>	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
871	>	tryTerminate(false); // handle failure during shutdown
872	>	}
873	>	}
874	>	if (w == null)
875		break;
876	+	w.start(recordWorker(w), ueh);
877	+	if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) {
878	+	int c; // advance event count
879	+	UNSAFE.compareAndSwapInt(this, eventCountOffset,
880	+	c = eventCount, c+1);
881	+	break; // add at most one unless total below target
882	+	}
883		}
884		}
885	+	if (eventWaiters != 0L)
886	+	releaseEventWaiters();
887		}
888
889		/**
890	<	* Tries (once) to add a new worker if all existing workers are
891	<	* busy, and there are either no running workers or the deficit is
892	<	* at least twice the surplus.
893	<	*
894	<	* @param wc workerCounts value on invocation of this method
895	<	*/
896	<	private void tryAddWorkerIfBusy(int wc) {
964	<	int tc, rc, rs;
965	<	int pc = parallelism;
966	<	if ((tc = wc >>> TOTAL_COUNT_SHIFT) < MAX_WORKERS &&
967	<	((rc = wc & RUNNING_COUNT_MASK) == 0 ||
968	<	rc < pc - ((tc - pc) << 1)) &&
969	<	(rs = runState) < TERMINATING &&
970	<	(rs & ACTIVE_COUNT_MASK) == tc) {
971	<	// Since all workers busy, heuristically back off to let settle
972	<	Thread.yield();
973	<	if (eventWaiters == 0L && spareWaiters == 0 && // recheck
974	<	runState == rs && workerCounts == wc &&
975	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
976	<	wc + (ONE_RUNNING|ONE_TOTAL)))
977	<	addWorker();
978	<	}
979	<	}
980	<
981	<	/**
982	<	* Does at most one of:
983	<	*
984	<	* 1. Help wake up existing workers waiting for work via
985	<	*    releaseEventWaiters. (If any exist, then it doesn't
986	<	*    matter right now if under target parallelism level.)
987	<	*
988	<	* 2. If a spare exists, try (once) to resume it via tryResumeSpare.
989	<	*
990	<	* 3. If there are not enough total workers, add some
991	<	*    via addWorkerIfBelowTarget;
890	>	* Callback from the oldest waiter in awaitEvent waking up after a
891	>	* period of non-use. If all workers are idle, tries (once) to
892	>	* shutdown an event waiter or a spare, if one exists. Note that
893	>	* we don't need CAS or locks here because the method is called
894	>	* only from one thread occasionally waking (and even misfires are
895	>	* OK). Note that until the shutdown worker fully terminates,
896	>	* workerCounts will overestimate total count, which is tolerable.
897		*
898	<	* 4. Try (once) to add a new worker if all existing workers
899	<	*     are busy, via tryAddWorkerIfBusy
898	>	* @param ec the event count waited on by caller (to abort
899	>	* attempt if count has since changed).
900		*/
901	<	private void helpMaintainParallelism() {
902	<	long h; int pc, wc;
903	<	if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0) {
904	<	if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
905	<	releaseEventWaiters(); // avoid useless call
906	<	}
907	<	else if ((pc = parallelism) >
908	<	((wc = workerCounts) & RUNNING_COUNT_MASK)) {
909	<	if (spareWaiters != 0)
910	<	tryResumeSpare(wc);
911	<	else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
912	<	addWorkerIfBelowTarget();
913	<	else
914	<	tryAddWorkerIfBusy(wc);
901	>	private void tryShutdownUnusedWorker(int ec) {
902	>	if (runState == 0 && eventCount == ec) { // only trigger if all idle
903	>	ForkJoinWorkerThread[] ws = workers;
904	>	int n = ws.length;
905	>	ForkJoinWorkerThread w = null;
906	>	boolean shutdown = false;
907	>	int sw;
908	>	long h;
909	>	if ((sw = spareWaiters) != 0) { // prefer killing spares
910	>	int id = (sw & SPARE_ID_MASK) - 1;
911	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
912	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
913	>	sw, w.nextSpare))
914	>	shutdown = true;
915	>	}
916	>	else if ((h = eventWaiters) != 0L) {
917	>	long nh;
918	>	int id = ((int)(h & WAITER_ID_MASK)) - 1;
919	>	if (id >= 0 && id < n && (w = ws[id]) != null &&
920	>	(nh = w.nextWaiter) != 0L && // keep at least one worker
921	>	UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
922	>	shutdown = true;
923	>	}
924	>	if (w != null && shutdown) {
925	>	w.shutdown();
926	>	LockSupport.unpark(w);
927	>	}
928		}
929	+	releaseEventWaiters(); // in case of interference
930		}
931
932		/**
#	Line 1015 | Line 934 | public class ForkJoinPool extends Abstra
934		* stealing a task or taking a submission and running it).
935		* Performs one or more of the following:
936		*
937	<	* 1. If the worker is active, try to set its active status to
938	<	*    inactive and update activeCount. On contention, we may try
939	<	*    again on this or subsequent call.
940	<	*
941	<	* 2. Release any existing event waiters that are now relesable
942	<	*
943	<	* 3. If there are too many running threads, suspend this worker
944	<	*    (first forcing inactive if necessary).  If it is not
945	<	*    needed, it may be killed while suspended via
946	<	*    tryShutdownSpare. Otherwise, upon resume it rechecks to make
947	<	*    sure that it is still needed.
948	<	*
949	<	* 4. If more than 1 miss, await the next task event via
950	<	*    eventSync (first forcing inactivation if necessary), upon
951	<	*    which worker may also be killed, via tryShutdownWaiter.
952	<	*
953	<	* 5. Help reactivate other workers via helpMaintainParallelism
937	>	* 1. If the worker is active and either did not run a task
938	>	*    or there are too many workers, try to set its active status
939	>	*    to inactive and update activeCount. On contention, we may
940	>	*    try again in this or a subsequent call.
941	>	*
942	>	* 2. If not enough total workers, help create some.
943	>	*
944	>	* 3. If there are too many running workers, suspend this worker
945	>	*    (first forcing inactive if necessary).  If it is not needed,
946	>	*    it may be shutdown while suspended (via
947	>	*    tryShutdownUnusedWorker).  Otherwise, upon resume it
948	>	*    rechecks running thread count and need for event sync.
949	>	*
950	>	* 4. If worker did not run a task, await the next task event via
951	>	*    eventSync if necessary (first forcing inactivation), upon
952	>	*    which the worker may be shutdown via
953	>	*    tryShutdownUnusedWorker.  Otherwise, help release any
954	>	*    existing event waiters that are now releasable,
955		*
956		* @param w the worker
957	<	* @param misses the number of scans by caller failing to find work
1038	<	* (saturating at 2 to avoid wraparound)
957	>	* @param ran true if worker ran a task since last call to this method
958		*/
959	<	final void preStep(ForkJoinWorkerThread w, int misses) {
959	>	final void preStep(ForkJoinWorkerThread w, boolean ran) {
960	>	int wec = w.lastEventCount;
961		boolean active = w.active;
962	+	boolean inactivate = false;
963		int pc = parallelism;
964	<	for (;;) {
965	<	int rs, wc, rc, ec; long h;
966	<	if (active && UNSAFE.compareAndSwapInt(this, runStateOffset,
967	<	rs = runState, rs - 1))
968	<	active = w.active = false;
969	<	if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0 &&
970	<	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) {
971	<	releaseEventWaiters();
1051	<	if (misses > 1)
1052	<	continue;                  // clear before sync below
1053	<	}
1054	<	if ((rc = ((wc = workerCounts) & RUNNING_COUNT_MASK)) > pc) {
1055	<	if (!active &&                 // must inactivate to suspend
964	>	int rs;
965	>	while (w.runState == 0 && (rs = runState) < TERMINATING) {
966	>	if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
967	>	UNSAFE.compareAndSwapInt(this, runStateOffset, rs, rs - 1))
968	>	inactivate = active = w.active = false;
969	>	int wc = workerCounts;
970	>	if ((wc & RUNNING_COUNT_MASK) > pc) {
971	>	if (!(inactivate |= active) && // must inactivate to suspend
972		workerCounts == wc &&      // try to suspend as spare
973		UNSAFE.compareAndSwapInt(this, workerCountsOffset,
974	<	wc, wc - ONE_RUNNING)) {
974	>	wc, wc - ONE_RUNNING))
975		w.suspendAsSpare();
1060	–	if (!w.isRunning())
1061	–	break;                 // was killed while spare
1062	–	}
1063	–	continue;
976		}
977	<	if (misses > 0) {
978	<	if ((ec = eventCount) == w.lastEventCount && misses > 1) {
979	<	if (!active) {             // must inactivate to sync
980	<	eventSync(w);
981	<	if (w.isRunning())
982	<	misses = 1;        // don't re-sync
983	<	else
984	<	break;             // was killed while waiting
985	<	}
986	<	continue;
977	>	else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
978	>	helpMaintainParallelism();     // not enough workers
979	>	else if (!ran) {
980	>	long h = eventWaiters;
981	>	int ec = eventCount;
982	>	if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
983	>	releaseEventWaiters();     // release others before waiting
984	>	else if (ec != wec) {
985	>	w.lastEventCount = ec;     // no need to wait
986	>	break;
987		}
988	<	w.lastEventCount = ec;
988	>	else if (!(inactivate |= active))
989	>	eventSync(w, wec);         // must inactivate before sync
990		}
991	<	if (rc < pc)
992	<	helpMaintainParallelism();
1080	<	break;
991	>	else
992	>	break;
993		}
994		}
995
996		/**
997		* Helps and/or blocks awaiting join of the given task.
998	<	* Alternates between helpJoinTask() and helpMaintainParallelism()
1087	<	* as many times as there is a deficit in running count (or longer
1088	<	* if running count would become zero), then blocks if task still
1089	<	* not done.
998	>	* See above for explanation.
999		*
1000		* @param joinMe the task to join
1001	+	* @param worker the current worker thread
1002		*/
1003		final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
1004	<	int threshold = parallelism;         // descend blocking thresholds
1004	>	int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1005		while (joinMe.status >= 0) {
1006	<	boolean block; int wc;
1006	>	int wc;
1007		worker.helpJoinTask(joinMe);
1008		if (joinMe.status < 0)
1009		break;
1010	<	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1011	<	if (threshold > 0)
1012	<	--threshold;
1013	<	else
1014	<	advanceEventCount(); // force release
1015	<	block = false;
1016	<	}
1017	<	else
1018	<	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1019	<	wc, wc - ONE_RUNNING);
1020	<	helpMaintainParallelism();
1021	<	if (block) {
1022	<	int c;
1023	<	joinMe.internalAwaitDone();
1010	>	else if (retries > 0)
1011	>	--retries;
1012	>	else if (((wc = workerCounts) & RUNNING_COUNT_MASK) != 0 &&
1013	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1014	>	wc, wc - ONE_RUNNING)) {
1015	>	int stat, c; long h;
1016	>	while ((stat = joinMe.status) >= 0 &&
1017	>	(h = eventWaiters) != 0L && // help release others
1018	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1019	>	releaseEventWaiters();
1020	>	if (stat >= 0 &&
1021	>	((workerCounts & RUNNING_COUNT_MASK) == 0 ||
1022	>	(stat =
1023	>	joinMe.internalAwaitDone(JOIN_TIMEOUT_MILLIS)) >= 0))
1024	>	helpMaintainParallelism(); // timeout or no running workers
1025		do {} while (!UNSAFE.compareAndSwapInt
1026		(this, workerCountsOffset,
1027		c = workerCounts, c + ONE_RUNNING));
1028	<	break;
1028	>	if (stat < 0)
1029	>	break;   // else restart
1030		}
1031		}
1032		}
1033
1034		/**
1035	<	* Same idea as awaitJoin, but no helping
1035	>	* Same idea as awaitJoin, but no helping, retries, or timeouts.
1036		*/
1037		final void awaitBlocker(ManagedBlocker blocker)
1038		throws InterruptedException {
1127	–	int threshold = parallelism;
1039		while (!blocker.isReleasable()) {
1040	<	boolean block; int wc;
1041	<	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1042	<	if (threshold > 0)
1043	<	--threshold;
1133	<	else
1134	<	advanceEventCount();
1135	<	block = false;
1136	<	}
1137	<	else
1138	<	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1139	<	wc, wc - ONE_RUNNING);
1140	<	helpMaintainParallelism();
1141	<	if (block) {
1040	>	int wc = workerCounts;
1041	>	if ((wc & RUNNING_COUNT_MASK) != 0 &&
1042	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1043	>	wc, wc - ONE_RUNNING)) {
1044		try {
1045	<	do {} while (!blocker.isReleasable() && !blocker.block());
1045	>	while (!blocker.isReleasable()) {
1046	>	long h = eventWaiters;
1047	>	if (h != 0L &&
1048	>	(int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1049	>	releaseEventWaiters();
1050	>	else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1051	>	runState < TERMINATING)
1052	>	helpMaintainParallelism();
1053	>	else if (blocker.block())
1054	>	break;
1055	>	}
1056		} finally {
1057		int c;
1058		do {} while (!UNSAFE.compareAndSwapInt
#	Line 1190 | Line 1102 | public class ForkJoinPool extends Abstra
1102		private void startTerminating() {
1103		cancelSubmissions();
1104		for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1105	<	advanceEventCount();
1105	>	int c; // advance event count
1106	>	UNSAFE.compareAndSwapInt(this, eventCountOffset,
1107	>	c = eventCount, c+1);
1108		eventWaiters = 0L; // clobber lists
1109		spareWaiters = 0;
1110		ForkJoinWorkerThread[] ws = workers;
#	Line 1353 | Line 1267 | public class ForkJoinPool extends Abstra
1267		* @param pc the initial parallelism level
1268		*/
1269		private static int initialArraySizeFor(int pc) {
1270	<	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1270	>	// If possible, initially allocate enough space for one spare
1271		int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1272	+	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1273		size |= size >>> 1;
1274		size |= size >>> 2;
1275		size |= size >>> 4;
#	Line 1373 | Line 1288 | public class ForkJoinPool extends Abstra
1288		if (runState >= SHUTDOWN)
1289		throw new RejectedExecutionException();
1290		submissionQueue.offer(task);
1291	<	advanceEventCount();
1292	<	if (eventWaiters != 0L)
1293	<	releaseEventWaiters();
1379	<	if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
1380	<	addWorkerIfBelowTarget();
1291	>	int c; // try to increment event count -- CAS failure OK
1292	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1293	>	helpMaintainParallelism(); // create, start, or resume some workers
1294		}
1295
1296		/**

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines