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 1.57 by dl, Wed Jul 7 19:52:31 2010 UTC vs.
Revision 1.62 by dl, Wed Aug 11 20:28:22 2010 UTC

#	Line 52 | Line 52 | import java.util.concurrent.CountDownLat
52		* convenient form for informal monitoring.
53		*
54		* <p> As is the case with other ExecutorServices, there are three
55	<	* main task execution methods summarized in the follwoing
55	>	* main task execution methods summarized in the following
56		* table. These are designed to be used by clients not already engaged
57		* in fork/join computations in the current pool.  The main forms of
58		* these methods accept instances of {@code ForkJoinTask}, but
#	Line 60 | Line 60 | import java.util.concurrent.CountDownLat
60		* Runnable}- or {@code Callable}- based activities as well.  However,
61		* tasks that are already executing in a pool should normally
62		* <em>NOT</em> use these pool execution methods, but instead use the
63	<	* within-computation forms listed in the table. To avoid inadvertant
64	<	* cyclic task dependencies and to improve performance, task
65	<	* submissions to the current pool by an ongoing fork/join
66	<	* computations may be implicitly translated to the corresponding
67	<	* ForkJoinTask forms.
63	>	* within-computation forms listed in the table.
64		*
65		* <table BORDER CELLPADDING=3 CELLSPACING=1>
66		*  <tr>
#	Line 88 | Line 84 | import java.util.concurrent.CountDownLat
84		*    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
85		*  </tr>
86		* </table>
87	<	*
87	>	*
88		* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
89		* used for all parallel task execution in a program or subsystem.
90		* Otherwise, use would not usually outweigh the construction and
#	Line 113 | Line 109 | import java.util.concurrent.CountDownLat
109		* {@code IllegalArgumentException}.
110		*
111		* <p>This implementation rejects submitted tasks (that is, by throwing
112	<	* {@link RejectedExecutionException}) only when the pool is shut down.
112	>	* {@link RejectedExecutionException}) only when the pool is shut down
113	>	* or internal resources have been exhausted.
114		*
115		* @since 1.7
116		* @author Doug Lea
#	Line 140 | Line 137 | public class ForkJoinPool extends Abstra
137		* of tasks profit from cache affinities, but others are harmed by
138		* cache pollution effects.)
139		*
140	+	* Beyond work-stealing support and essential bookkeeping, the
141	+	* main responsibility of this framework is to take actions when
142	+	* one worker is waiting to join a task stolen (or always held by)
143	+	* another.  Becauae we are multiplexing many tasks on to a pool
144	+	* of workers, we can't just let them block (as in Thread.join).
145	+	* We also cannot just reassign the joiner's run-time stack with
146	+	* another and replace it later, which would be a form of
147	+	* "continuation", that even if possible is not necessarily a good
148	+	* idea. Given that the creation costs of most threads on most
149	+	* systems mainly surrounds setting up runtime stacks, thread
150	+	* creation and switching is usually not much more expensive than
151	+	* stack creation and switching, and is more flexible). Instead we
152	+	* combine two tactics:
153	+	*
154	+	*   Helping: Arranging for the joiner to execute some task that it
155	+	*      would be running if the steal had not occurred.  Method
156	+	*      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
157	+	*      links to try to find such a task.
158	+	*
159	+	*   Compensating: Unless there are already enough live threads,
160	+	*      method helpMaintainParallelism() may create or or
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).
173	+	*
174	+	* The ManagedBlocker extension API can't use helping so relies
175	+	* only on compensation in method awaitBlocker.
176	+	*
177		* The main throughput advantages of work-stealing stem from
178		* decentralized control -- workers mostly steal tasks from each
179		* other. We do not want to negate this by creating bottlenecks
180	<	* implementing the management responsibilities of this class. So
181	<	* we use a collection of techniques that avoid, reduce, or cope
182	<	* well with contention. These entail several instances of
183	<	* bit-packing into CASable fields to maintain only the minimally
184	<	* required atomicity. To enable such packing, we restrict maximum
185	<	* parallelism to (1<<15)-1 (enabling twice this to fit into a 16
186	<	* bit field), which is far in excess of normal operating range.
187	<	* Even though updates to some of these bookkeeping fields do
188	<	* sometimes contend with each other, they don't normally
189	<	* cache-contend with updates to others enough to warrant memory
190	<	* padding or isolation. So they are all held as fields of
191	<	* ForkJoinPool objects.  The main capabilities are as follows:
180	>	* implementing other management responsibilities. So we use a
181	>	* collection of techniques that avoid, reduce, or cope well with
182	>	* contention. These entail several instances of bit-packing into
183	>	* CASable fields to maintain only the minimally required
184	>	* atomicity. To enable such packing, we restrict maximum
185	>	* parallelism to (1<<15)-1 (enabling twice this (to accommodate
186	>	* unbalanced increments and decrements) to fit into a 16 bit
187	>	* field, which is far in excess of normal operating range.  Even
188	>	* though updates to some of these bookkeeping fields do sometimes
189	>	* contend with each other, they don't normally cache-contend with
190	>	* updates to others enough to warrant memory padding or
191	>	* isolation. So they are all held as fields of ForkJoinPool
192	>	* objects.  The main capabilities are as follows:
193		*
194		* 1. Creating and removing workers. Workers are recorded in the
195		* "workers" array. This is an array as opposed to some other data
#	Line 170 | Line 205 | public class ForkJoinPool extends Abstra
205		* blocked workers. However, all other support code is set up to
206		* work with other policies.
207		*
208	+	* To ensure that we do not hold on to worker references that
209	+	* would prevent GC, ALL accesses to workers are via indices into
210	+	* the workers array (which is one source of some of the unusual
211	+	* code constructions here). In essence, the workers array serves
212	+	* as a WeakReference mechanism. Thus for example the event queue
213	+	* stores worker indices, not worker references. Access to the
214	+	* workers in associated methods (for example releaseEventWaiters)
215	+	* must both index-check and null-check the IDs. All such accesses
216	+	* ignore bad IDs by returning out early from what they are doing,
217	+	* since this can only be associated with shutdown, in which case
218	+	* it is OK to give up. On termination, we just clobber these
219	+	* data structures without trying to use them.
220	+	*
221		* 2. Bookkeeping for dynamically adding and removing workers. We
222		* aim to approximately maintain the given level of parallelism.
223		* When some workers are known to be blocked (on joins or via
#	Line 179 | Line 227 | public class ForkJoinPool extends Abstra
227		* that are neither blocked nor artifically suspended) as well as
228		* the total number.  These two values are packed into one field,
229		* "workerCounts" because we need accurate snapshots when deciding
230	<	* to create, resume or suspend.  To support these decisions,
231	<	* updates to spare counts must be prospective (not
232	<	* retrospective).  For example, the running count is decremented
233	<	* before blocking by a thread about to block as a spare, but
186	<	* incremented by the thread about to unblock it. Updates upon
187	<	* resumption ofr threads blocking in awaitJoin or awaitBlocker
188	<	* cannot usually be prospective, so the running count is in
189	<	* general an upper bound of the number of productively running
190	<	* threads Updates to the workerCounts field sometimes transiently
191	<	* encounter a fair amount of contention when join dependencies
192	<	* are such that many threads block or unblock at about the same
193	<	* time. We alleviate this by sometimes performing an alternative
194	<	* action on contention like releasing waiters or locating spares.
230	>	* to create, resume or suspend.  Note however that the
231	>	* correspondance of these counts to reality is not guaranteed. In
232	>	* particular updates for unblocked threads may lag until they
233	>	* actually wake up.
234		*
235		* 3. Maintaining global run state. The run state of the pool
236		* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
#	Line 220 | Line 259 | public class ForkJoinPool extends Abstra
259		* workers that previously could not find a task to now find one:
260		* Submission of a new task to the pool, or another worker pushing
261		* a task onto a previously empty queue.  (We also use this
262	<	* mechanism for termination and reconfiguration actions that
263	<	* require wakeups of idle workers).  Each worker maintains its
264	<	* last known event count, and blocks when a scan for work did not
265	<	* find a task AND its lastEventCount matches the current
266	<	* eventCount. Waiting idle workers are recorded in a variant of
267	<	* Treiber stack headed by field eventWaiters which, when nonzero,
268	<	* encodes the thread index and count awaited for by the worker
269	<	* thread most recently calling eventSync. This thread in turn has
270	<	* a record (field nextEventWaiter) for the next waiting worker.
271	<	* In addition to allowing simpler decisions about need for
272	<	* wakeup, the event count bits in eventWaiters serve the role of
273	<	* tags to avoid ABA errors in Treiber stacks.  To reduce delays
274	<	* in task diffusion, workers not otherwise occupied may invoke
275	<	* method releaseWaiters, that removes and signals (unparks)
276	<	* workers not waiting on current count. To minimize task
277	<	* production stalls associate with signalling, any worker pushing
278	<	* a task on an empty queue invokes the weaker method signalWork,
279	<	* that only releases idle workers until it detects interference
280	<	* by other threads trying to release, and lets them take
281	<	* over. The net effect is a tree-like diffusion of signals, where
282	<	* released threads (and possibly others) help with unparks.  To
283	<	* further reduce contention effects a bit, failed CASes to
262	>	* mechanism for termination actions that require wakeups of idle
263	>	* workers).  Each worker maintains its last known event count,
264	>	* and blocks when a scan for work did not find a task AND its
265	>	* lastEventCount matches the current eventCount. Waiting idle
266	>	* workers are recorded in a variant of Treiber stack headed by
267	>	* field eventWaiters which, when nonzero, encodes the thread
268	>	* index and count awaited for by the worker thread most recently
269	>	* calling eventSync. This thread in turn has a record (field
270	>	* nextEventWaiter) for the next waiting worker.  In addition to
271	>	* allowing simpler decisions about need for wakeup, the event
272	>	* count bits in eventWaiters serve the role of tags to avoid ABA
273	>	* errors in Treiber stacks.  To reduce delays in task diffusion,
274	>	* workers not otherwise occupied may invoke method
275	>	* releaseEventWaiters, that removes and signals (unparks) workers
276	>	* not waiting on current count. To reduce stalls, To minimize
277	>	* task production stalls associate with signalling, any worker
278	>	* pushing a task on an empty queue invokes the weaker method
279	>	* signalWork, that only releases idle workers until it detects
280	>	* interference by other threads trying to release, and lets them
281	>	* take over.  The net effect is a tree-like diffusion of signals,
282	>	* where released threads (and possibly others) help with unparks.
283	>	* To further reduce contention effects a bit, failed CASes to
284		* increment field eventCount are tolerated without retries.
285		* Conceptually they are merged into the same event, which is OK
286		* when their only purpose is to enable workers to scan for work.
#	Line 249 | Line 288 | public class ForkJoinPool extends Abstra
288		* 5. Managing suspension of extra workers. When a worker is about
289		* to block waiting for a join (or via ManagedBlockers), we may
290		* create a new thread to maintain parallelism level, or at least
291	<	* avoid starvation (see below). Usually, extra threads are needed
292	<	* for only very short periods, yet join dependencies are such
293	<	* that we sometimes need them in bursts. Rather than create new
294	<	* threads each time this happens, we suspend no-longer-needed
295	<	* extra ones as "spares". For most purposes, we don't distinguish
296	<	* "extra" spare threads from normal "core" threads: On each call
297	<	* to preStep (the only point at which we can do this) a worker
291	>	* avoid starvation. Usually, extra threads are needed for only
292	>	* very short periods, yet join dependencies are such that we
293	>	* sometimes need them in bursts. Rather than create new threads
294	>	* each time this happens, we suspend no-longer-needed extra ones
295	>	* as "spares". For most purposes, we don't distinguish "extra"
296	>	* spare threads from normal "core" threads: On each call to
297	>	* preStep (the only point at which we can do this) a worker
298		* checks to see if there are now too many running workers, and if
299	<	* so, suspends itself.  Methods awaitJoin and awaitBlocker look
300	<	* for suspended threads to resume before considering creating a
301	<	* new replacement. We don't need a special data structure to
302	<	* maintain spares; simply scanning the workers array looking for
303	<	* worker.isSuspended() is fine because the calling thread is
265	<	* otherwise not doing anything useful anyway; we are at least as
266	<	* happy if after locating a spare, the caller doesn't actually
267	<	* block because the join is ready before we try to adjust and
268	<	* compensate.  Note that this is intrinsically racy.  One thread
299	>	* so, suspends itself.  Method helpMaintainParallelism looks for
300	>	* suspended threads to resume before considering creating a new
301	>	* replacement. The spares themselves are encoded on another
302	>	* variant of a Treiber Stack, headed at field "spareWaiters".
303	>	* Note that the use of spares is intrinsically racy.  One thread
304		* may become a spare at about the same time as another is
305		* needlessly being created. We counteract this and related slop
306		* in part by requiring resumed spares to immediately recheck (in
307	<	* preStep) to see whether they they should re-suspend. The only
308	<	* effective difference between "extra" and "core" threads is that
309	<	* we allow the "extra" ones to time out and die if they are not
310	<	* resumed within a keep-alive interval of a few seconds. This is
311	<	* implemented mainly within ForkJoinWorkerThread, but requires
312	<	* some coordination (isTrimmed() -- meaning killed while
313	<	* suspended) to correctly maintain pool counts.
307	>	* preStep) to see whether they they should re-suspend.  To avoid
308	>	* long-term build-up of spares, the oldest spare (see
309	>	* ForkJoinWorkerThread.suspendAsSpare) occasionally wakes up if
310	>	* not signalled and calls tryTrimSpare, which uses two different
311	>	* thresholds: Always killing if the number of spares is greater
312	>	* that 25% of total, and killing others only at a slower rate
313	>	* (UNUSED_SPARE_TRIM_RATE_NANOS).
314		*
315		* 6. Deciding when to create new workers. The main dynamic
316	<	* control in this class is deciding when to create extra threads,
317	<	* in methods awaitJoin and awaitBlocker. We always need to create
318	<	* one when the number of running threads becomes zero. But
319	<	* because blocked joins are typically dependent, we don't
320	<	* necessarily need or want one-to-one replacement. Instead, we
321	<	* use a combination of heuristics that adds threads only when the
322	<	* pool appears to be approaching starvation.  These effectively
323	<	* reduce churn at the price of systematically undershooting
324	<	* target parallelism when many threads are blocked.  However,
325	<	* biasing toward undeshooting partially compensates for the above
326	<	* mechanics to suspend extra threads, that normally lead to
327	<	* overshoot because we can only suspend workers in-between
328	<	* top-level actions. It also better copes with the fact that some
329	<	* of the methods in this class tend to never become compiled (but
330	<	* are interpreted), so some components of the entire set of
331	<	* controls might execute many times faster than others. And
332	<	* similarly for cases where the apparent lack of work is just due
333	<	* to GC stalls and other transient system activity.
316	>	* control in this class is deciding when to create extra threads
317	>	* in method helpMaintainParallelism. We would like to keep
318	>	* exactly #parallelism threads running, which is an impossble
319	>	* task. We always need to create one when the number of running
320	>	* threads would become zero and all workers are busy. Beyond
321	>	* this, we must rely on heuristics that work well in the the
322	>	* presence of transients phenomena such as GC stalls, dynamic
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. The rules also
330	>	* better cope with the fact that some of the methods in this
331	>	* class tend to never become compiled (but are interpreted), so
332	>	* some components of the entire set of controls might execute 100
333	>	* times faster than others. And similarly for cases where the
334	>	* apparent lack of work is just due to GC stalls and other
335	>	* transient system activity.
336		*
337		* Beware that there is a lot of representation-level coupling
338		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 308 | Line 345 | public class ForkJoinPool extends Abstra
345		*
346		* Style notes: There are lots of inline assignments (of form
347		* "while ((local = field) != 0)") which are usually the simplest
348	<	* way to ensure read orderings. Also several occurrences of the
349	<	* unusual "do {} while(!cas...)" which is the simplest way to
350	<	* force an update of a CAS'ed variable. There are also a few
351	<	* other coding oddities that help some methods perform reasonably
352	<	* even when interpreted (not compiled).
348	>	* way to ensure the required read orderings (which are sometimes
349	>	* critical). Also several occurrences of the unusual "do {}
350	>	* while(!cas...)" which is the simplest way to force an update of
351	>	* a CAS'ed variable. There are also other coding oddities that
352	>	* help some methods perform reasonably even when interpreted (not
353	>	* compiled), at the expense of some messy constructions that
354	>	* reduce byte code counts.
355		*
356		* The order of declarations in this file is: (1) statics (2)
357		* fields (along with constants used when unpacking some of them)
#	Line 380 | Line 419 | public class ForkJoinPool extends Abstra
419		new AtomicInteger();
420
421		/**
422	<	* Absolute bound for parallelism level. Twice this number must
423	<	* fit into a 16bit field to enable word-packing for some counts.
422	>	* Absolute bound for parallelism level. Twice this number plus
423	>	* one (i.e., 0xfff) must fit into a 16bit field to enable
424	>	* word-packing for some counts and indices.
425		*/
426	<	private static final int MAX_THREADS = 0x7fff;
426	>	private static final int MAX_WORKERS   = 0x7fff;
427
428		/**
429		* Array holding all worker threads in the pool.  Array size must
#	Line 423 | Line 463 | public class ForkJoinPool extends Abstra
463		private volatile long stealCount;
464
465		/**
466	+	* The last nanoTime that a spare thread was trimmed
467	+	*/
468	+	private volatile long trimTime;
469	+
470	+	/**
471	+	* The rate at which to trim unused spares
472	+	*/
473	+	static final long UNUSED_SPARE_TRIM_RATE_NANOS =
474	+	1000L * 1000L * 1000L; // 1 sec
475	+
476	+	/**
477		* Encoded record of top of treiber stack of threads waiting for
478		* events. The top 32 bits contain the count being waited for. The
479	<	* bottom word contains one plus the pool index of waiting worker
480	<	* thread.
479	>	* bottom 16 bits contains one plus the pool index of waiting
480	>	* worker thread. (Bits 16-31 are unused.)
481		*/
482		private volatile long eventWaiters;
483
484		private static final int  EVENT_COUNT_SHIFT = 32;
485	<	private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
485	>	private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
486
487		/**
488		* A counter for events that may wake up worker threads:
489		*   - Submission of a new task to the pool
490		*   - A worker pushing a task on an empty queue
491	<	*   - termination and reconfiguration
491	>	*   - termination
492		*/
493		private volatile int eventCount;
494
495		/**
496	+	* Encoded record of top of treiber stack of spare threads waiting
497	+	* for resumption. The top 16 bits contain an arbitrary count to
498	+	* avoid ABA effects. The bottom 16bits contains one plus the pool
499	+	* index of waiting worker thread.
500	+	*/
501	+	private volatile int spareWaiters;
502	+
503	+	private static final int SPARE_COUNT_SHIFT = 16;
504	+	private static final int SPARE_ID_MASK     = (1 << 16) - 1;
505	+
506	+	/**
507		* Lifecycle control. The low word contains the number of workers
508		* that are (probably) executing tasks. This value is atomically
509		* incremented before a worker gets a task to run, and decremented
#	Line 470 | Line 532 | public class ForkJoinPool extends Abstra
532		* making decisions about creating and suspending spare
533		* threads. Updated only by CAS. Note that adding a new worker
534		* requires incrementing both counts, since workers start off in
535	<	* running state.  This field is also used for memory-fencing
474	<	* configuration parameters.
535	>	* running state.
536		*/
537		private volatile int workerCounts;
538
#	Line 503 | Line 564 | public class ForkJoinPool extends Abstra
564		*/
565		private final int poolNumber;
566
567	<	// utilities for updating fields
567	>
568	>	// Utilities for CASing fields. Note that several of these
569	>	// are manually inlined by callers
570
571		/**
572	<	* Increments running count.  Also used by ForkJoinTask.
572	>	* Increments running count part of workerCounts
573		*/
574		final void incrementRunningCount() {
575		int c;
576		do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
577	<	c = workerCounts,
577	>	c = workerCounts,
578		c + ONE_RUNNING));
579		}
580	<
580	>
581		/**
582		* Tries to decrement running count unless already zero
583		*/
#	Line 527 | Line 590 | public class ForkJoinPool extends Abstra
590		}
591
592		/**
593	+	* Forces decrement of encoded workerCounts, awaiting nonzero if
594	+	* (rarely) necessary when other count updates lag.
595	+	*
596	+	* @param dr -- either zero or ONE_RUNNING
597	+	* @param dt == either zero or ONE_TOTAL
598	+	*/
599	+	private void decrementWorkerCounts(int dr, int dt) {
600	+	for (;;) {
601	+	int wc = workerCounts;
602	+	if (wc == 0 && (runState & TERMINATED) != 0)
603	+	return; // lagging termination on a backout
604	+	if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
605	+	(wc >>> TOTAL_COUNT_SHIFT) - dt < 0)
606	+	Thread.yield();
607	+	if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
608	+	wc, wc - (dr + dt)))
609	+	return;
610	+	}
611	+	}
612	+
613	+	/**
614	+	* Increments event count
615	+	*/
616	+	private void advanceEventCount() {
617	+	int c;
618	+	do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset,
619	+	c = eventCount, c+1));
620	+	}
621	+
622	+	/**
623		* Tries incrementing active count; fails on contention.
624		* Called by workers before executing tasks.
625		*
#	Line 574 | Line 667 | public class ForkJoinPool extends Abstra
667		lock.lock();
668		try {
669		ForkJoinWorkerThread[] ws = workers;
670	<	int nws = ws.length;
671	<	if (k < 0 || k >= nws || ws[k] != null) {
672	<	for (k = 0; k < nws && ws[k] != null; ++k)
670	>	int n = ws.length;
671	>	if (k < 0 || k >= n || ws[k] != null) {
672	>	for (k = 0; k < n && ws[k] != null; ++k)
673		;
674	<	if (k == nws)
675	<	ws = Arrays.copyOf(ws, nws << 1);
674	>	if (k == n)
675	>	ws = Arrays.copyOf(ws, n << 1);
676		}
677		ws[k] = w;
678		workers = ws; // volatile array write ensures slot visibility
#	Line 612 | Line 705 | public class ForkJoinPool extends Abstra
705		* Tries to create and add new worker. Assumes that worker counts
706		* are already updated to accommodate the worker, so adjusts on
707		* failure.
615	–	*
616	–	* @return new worker or null if creation failed
708		*/
709	<	private ForkJoinWorkerThread addWorker() {
709	>	private void addWorker() {
710		ForkJoinWorkerThread w = null;
711		try {
712		w = factory.newThread(this);
713		} finally { // Adjust on either null or exceptional factory return
714		if (w == null) {
715	<	onWorkerCreationFailure();
716	<	return null;
715	>	decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
716	>	tryTerminate(false); // in case of failure during shutdown
717		}
718		}
719	<	w.start(recordWorker(w), ueh);
720	<	return w;
630	<	}
631	<
632	<	/**
633	<	* Adjusts counts upon failure to create worker
634	<	*/
635	<	private void onWorkerCreationFailure() {
636	<	for (;;) {
637	<	int wc = workerCounts;
638	<	if ((wc >>> TOTAL_COUNT_SHIFT) > 0 &&
639	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
640	<	wc, wc - (ONE_RUNNING|ONE_TOTAL)))
641	<	break;
642	<	}
643	<	tryTerminate(false); // in case of failure during shutdown
644	<	}
645	<
646	<	/**
647	<	* Create enough total workers to establish target parallelism,
648	<	* giving up if terminating or addWorker fails
649	<	*/
650	<	private void ensureEnoughTotalWorkers() {
651	<	int wc;
652	<	while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism &&
653	<	runState < TERMINATING) {
654	<	if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
655	<	wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
656	<	addWorker() == null))
657	<	break;
658	<	}
719	>	if (w != null)
720	>	w.start(recordWorker(w), ueh);
721		}
722
723		/**
724		* Final callback from terminating worker.  Removes record of
725		* worker from array, and adjusts counts. If pool is shutting
726	<	* down, tries to complete terminatation, else possibly replaces
665	<	* the worker.
726	>	* down, tries to complete terminatation.
727		*
728		* @param w the worker
729		*/
730		final void workerTerminated(ForkJoinWorkerThread w) {
670	–	if (w.active) { // force inactive
671	–	w.active = false;
672	–	do {} while (!tryDecrementActiveCount());
673	–	}
731		forgetWorker(w);
732	<
733	<	// Decrement total count, and if was running, running count
734	<	// Spin (waiting for other updates) if either would be negative
735	<	int nr = w.isTrimmed() ? 0 : ONE_RUNNING;
679	<	int unit = ONE_TOTAL + nr;
680	<	for (;;) {
681	<	int wc = workerCounts;
682	<	int rc = wc & RUNNING_COUNT_MASK;
683	<	if (rc - nr < 0 || (wc >>> TOTAL_COUNT_SHIFT) == 0)
684	<	Thread.yield(); // back off if waiting for other updates
685	<	else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
686	<	wc, wc - unit))
687	<	break;
688	<	}
689	<
690	<	accumulateStealCount(w); // collect final count
691	<	if (!tryTerminate(false))
692	<	ensureEnoughTotalWorkers();
732	>	decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL);
733	>	while (w.stealCount != 0) // collect final count
734	>	tryAccumulateStealCount(w);
735	>	tryTerminate(false);
736		}
737
738		// Waiting for and signalling events
739
740		/**
741		* Releases workers blocked on a count not equal to current count.
742	+	* Normally called after precheck that eventWaiters isn't zero to
743	+	* avoid wasted array checks.
744	+	*
745	+	* @param signalling true if caller is a signalling worker so can
746	+	* exit upon (conservatively) detected contention by other threads
747	+	* who will continue to release
748		*/
749	<	private void releaseWaiters() {
750	<	long top;
751	<	int id;
752	<	while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 &&
753	<	(int)(top >>> EVENT_COUNT_SHIFT) != eventCount) {
754	<	ForkJoinWorkerThread[] ws = workers;
755	<	ForkJoinWorkerThread w;
756	<	if (ws.length >= id && (w = ws[id - 1]) != null &&
757	<	UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
758	<	top, w.nextWaiter))
749	>	private void releaseEventWaiters(boolean signalling) {
750	>	ForkJoinWorkerThread[] ws = workers;
751	>	int n = ws.length;
752	>	long h; // head of stack
753	>	ForkJoinWorkerThread w; int id, ec;
754	>	while ((id = ((int)((h = eventWaiters) & WAITER_ID_MASK)) - 1) >= 0 &&
755	>	(int)(h >>> EVENT_COUNT_SHIFT) != (ec = eventCount) &&
756	>	id < n && (w = ws[id]) != null) {
757	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
758	>	h, h = w.nextWaiter))
759		LockSupport.unpark(w);
760	+	if (signalling && (eventCount != ec || eventWaiters != h))
761	+	break;
762		}
763		}
764
765		/**
766	<	* Ensures eventCount on exit is different (mod 2^32) than on
767	<	* entry and wakes up all waiters
766	>	* Tries to advance eventCount and releases waiters. Called only
767	>	* from workers.
768		*/
769	<	private void signalEvent() {
770	<	int c;
771	<	do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset,
772	<	c = eventCount, c+1));
773	<	releaseWaiters();
769	>	final void signalWork() {
770	>	int c; // try to increment event count -- CAS failure OK
771	>	UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
772	>	if (eventWaiters != 0L)
773	>	releaseEventWaiters(true);
774		}
775
776		/**
777	<	* Advances eventCount and releases waiters until interference by
778	<	* other releasing threads is detected.
777	>	* Blocks worker until terminating or event count
778	>	* advances from last value held by worker
779	>	*
780	>	* @param w the calling worker thread
781		*/
782	<	final void signalWork() {
783	<	// EventCount CAS failures are OK -- any change in count suffices.
784	<	int ec;
785	<	UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1);
786	<	outer:for (;;) {
787	<	long top = eventWaiters;
788	<	ec = eventCount;
789	<	for (;;) {
790	<	ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
791	<	int id = (int)(top & WAITER_INDEX_MASK);
792	<	if (id <= 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec)
793	<	return;
794	<	if ((ws = workers).length < id || (w = ws[id - 1]) == null ||
795	<	!UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
796	<	top, top = w.nextWaiter))
797	<	continue outer;      // possibly stale; reread
798	<	LockSupport.unpark(w);
799	<	if (top != eventWaiters) // let someone else take over
800	<	return;
782	>	private void eventSync(ForkJoinWorkerThread w) {
783	>	int wec = w.lastEventCount;
784	>	long nh = (((long)wec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
785	>	long h;
786	>	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
787	>	((h = eventWaiters) == 0L ||
788	>	(int)(h >>> EVENT_COUNT_SHIFT) == wec) &&
789	>	eventCount == wec) {
790	>	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
791	>	w.nextWaiter = h, nh)) {
792	>	while (runState < TERMINATING && eventCount == wec) {
793	>	if (!tryAccumulateStealCount(w))  // transfer while idle
794	>	continue;
795	>	Thread.interrupted();             // clear/ignore interrupt
796	>	if (eventCount != wec)
797	>	break;
798	>	LockSupport.park(w);
799	>	}
800	>	break;
801		}
802		}
803	+	w.lastEventCount = eventCount;
804		}
805
806	+	// Maintaining spares
807	+
808		/**
809	<	* If worker is inactive, blocks until terminating or event count
754	<	* advances from last value held by worker; in any case helps
755	<	* release others.
756	<	*
757	<	* @param w the calling worker thread
809	>	* Pushes worker onto the spare stack
810		*/
811	<	private void eventSync(ForkJoinWorkerThread w) {
812	<	if (!w.active) {
813	<	int prev = w.lastEventCount;
814	<	long nextTop = (((long)prev << EVENT_COUNT_SHIFT) |
815	<	((long)(w.poolIndex + 1)));
816	<	long top;
817	<	while ((runState < SHUTDOWN || !tryTerminate(false)) &&
818	<	(((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 ||
819	<	(int)(top >>> EVENT_COUNT_SHIFT) == prev) &&
820	<	eventCount == prev) {
821	<	if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
822	<	w.nextWaiter = top, nextTop)) {
823	<	accumulateStealCount(w); // transfer steals while idle
824	<	Thread.interrupted();    // clear/ignore interrupt
825	<	while (eventCount == prev)
826	<	w.doPark();
827	<	break;
828	<	}
811	>	final void pushSpare(ForkJoinWorkerThread w) {
812	>	int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex+1);
813	>	do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
814	>	w.nextSpare = spareWaiters,ns));
815	>	}
816	>
817	>	/**
818	>	* Tries (once) to resume a spare if running count is less than
819	>	* target parallelism. Fails on contention or stale workers.
820	>	*/
821	>	private void tryResumeSpare() {
822	>	int sw, id;
823	>	ForkJoinWorkerThread w;
824	>	ForkJoinWorkerThread[] ws;
825	>	if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
826	>	id < (ws = workers).length && (w = ws[id]) != null &&
827	>	(workerCounts & RUNNING_COUNT_MASK) < parallelism &&
828	>	eventWaiters == 0L &&
829	>	spareWaiters == sw &&
830	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
831	>	sw, w.nextSpare) &&
832	>	w.tryUnsuspend()) {
833	>	int c; // try increment; if contended, finish after unpark
834	>	boolean inc = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
835	>	c = workerCounts,
836	>	c + ONE_RUNNING);
837	>	LockSupport.unpark(w);
838	>	if (!inc) {
839	>	do {} while(!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
840	>	c = workerCounts,
841	>	c + ONE_RUNNING));
842		}
778	–	w.lastEventCount = eventCount;
843		}
844	<	releaseWaiters();
844	>	}
845	>
846	>	/**
847	>	* Callback from oldest spare occasionally waking up.  Tries
848	>	* (once) to shutdown a spare if more than 25% spare overage, or
849	>	* if UNUSED_SPARE_TRIM_RATE_NANOS have elapsed and there are at
850	>	* least #parallelism running threads. Note that we don't need CAS
851	>	* or locks here because the method is called only from the oldest
852	>	* suspended spare occasionally waking (and even misfires are OK).
853	>	*
854	>	* @param now the wake up nanoTime of caller
855	>	*/
856	>	final void tryTrimSpare(long now) {
857	>	long lastTrim = trimTime;
858	>	trimTime = now;
859	>	helpMaintainParallelism(); // first, help wake up any needed spares
860	>	int sw, id;
861	>	ForkJoinWorkerThread w;
862	>	ForkJoinWorkerThread[] ws;
863	>	int pc = parallelism;
864	>	int wc = workerCounts;
865	>	if ((wc & RUNNING_COUNT_MASK) >= pc &&
866	>	(((wc >>> TOTAL_COUNT_SHIFT) - pc) > (pc >>> 2) + 1 ||// approx 25%
867	>	now - lastTrim >= UNUSED_SPARE_TRIM_RATE_NANOS) &&
868	>	(id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 &&
869	>	id < (ws = workers).length && (w = ws[id]) != null &&
870	>	UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
871	>	sw, w.nextSpare))
872	>	w.shutdown(false);
873	>	}
874	>
875	>	/**
876	>	* Does at most one of:
877	>	*
878	>	* 1. Help wake up existing workers waiting for work via
879	>	*    releaseEventWaiters. (If any exist, then it probably doesn't
880	>	*    matter right now if under target parallelism level.)
881	>	*
882	>	* 2. If below parallelism level and a spare exists, try (once)
883	>	*    to resume it via tryResumeSpare.
884	>	*
885	>	* 3. If neither of the above, tries (once) to add a new
886	>	*    worker if either there are not enough total, or if all
887	>	*    existing workers are busy, there are either no running
888	>	*    workers or the deficit is at least twice the surplus.
889	>	*/
890	>	private void helpMaintainParallelism() {
891	>	// uglified to work better when not compiled
892	>	int pc, wc, rc, tc, rs; long h;
893	>	if ((h = eventWaiters) != 0L) {
894	>	if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
895	>	releaseEventWaiters(false); // avoid useless call
896	>	}
897	>	else if ((pc = parallelism) >
898	>	(rc = ((wc = workerCounts) & RUNNING_COUNT_MASK))) {
899	>	if (spareWaiters != 0)
900	>	tryResumeSpare();
901	>	else if ((rs = runState) < TERMINATING &&
902	>	((tc = wc >>> TOTAL_COUNT_SHIFT) < pc ||
903	>	(tc == (rs & ACTIVE_COUNT_MASK) && // all busy
904	>	(rc == 0 ||                       // must add
905	>	rc < pc - ((tc - pc) << 1)) &&   // within slack
906	>	tc < MAX_WORKERS && runState == rs)) && // recheck busy
907	>	workerCounts == wc &&
908	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
909	>	wc + (ONE_RUNNING|ONE_TOTAL)))
910	>	addWorker();
911	>	}
912		}
913
914		/**
915		* Callback from workers invoked upon each top-level action (i.e.,
916		* stealing a task or taking a submission and running
917	<	* it). Performs one or both of the following:
917	>	* it). Performs one or more of the following:
918	>	*
919	>	* 1. If the worker cannot find work (misses > 0), updates its
920	>	*    active status to inactive and updates activeCount unless
921	>	*    this is the first miss and there is contention, in which
922	>	*    case it may try again (either in this or a subsequent
923	>	*    call).
924	>	*
925	>	* 2. If there are at least 2 misses, awaits the next task event
926	>	*    via eventSync
927	>	*
928	>	* 3. If there are too many running threads, suspends this worker
929	>	*    (first forcing inactivation if necessary).  If it is not
930	>	*    needed, it may be killed while suspended via
931	>	*    tryTrimSpare. Otherwise, upon resume it rechecks to make
932	>	*    sure that it is still needed.
933		*
934	<	* * If the worker cannot find work, updates its active status to
935	<	* inactive and updates activeCount unless there is contention, in
790	<	* which case it may try again (either in this or a subsequent
791	<	* call).  Additionally, awaits the next task event and/or helps
792	<	* wake up other releasable waiters.
793	<	*
794	<	* * If there are too many running threads, suspends this worker
795	<	* (first forcing inactivation if necessary).  If it is not
796	<	* resumed before a keepAlive elapses, the worker may be "trimmed"
797	<	* -- killed while suspended within suspendAsSpare. Otherwise,
798	<	* upon resume it rechecks to make sure that it is still needed.
934	>	* 4. Helps release and/or reactivate other workers via
935	>	*    helpMaintainParallelism
936		*
937		* @param w the worker
938	<	* @param worked false if the worker scanned for work but didn't
939	<	* find any (in which case it may block waiting for work).
938	>	* @param misses the number of scans by caller failing to find work
939	>	* (saturating at 2 just to avoid wraparound)
940		*/
941	<	final void preStep(ForkJoinWorkerThread w, boolean worked) {
941	>	final void preStep(ForkJoinWorkerThread w, int misses) {
942		boolean active = w.active;
943	<	boolean inactivate = !worked & active;
943	>	int pc = parallelism;
944		for (;;) {
945	<	if (inactivate) {
946	<	int rs = runState;
945	>	int wc = workerCounts;
946	>	int rc = wc & RUNNING_COUNT_MASK;
947	>	if (active && (misses > 0 || rc > pc)) {
948	>	int rs;                      // try inactivate
949		if (UNSAFE.compareAndSwapInt(this, runStateOffset,
950	<	rs, rs - ONE_ACTIVE))
951	<	inactivate = active = w.active = false;
950	>	rs = runState, rs - ONE_ACTIVE))
951	>	active = w.active = false;
952	>	else if (misses > 1 || rc > pc ||
953	>	(rs & ACTIVE_COUNT_MASK) >= pc)
954	>	continue;                // force inactivate
955		}
956	<	int wc = workerCounts;
957	<	if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
958	<	if (!worked)
959	<	eventSync(w);
960	<	return;
956	>	if (misses > 1) {
957	>	misses = 0;                  // don't re-sync
958	>	eventSync(w);                // continue loop to recheck rc
959	>	}
960	>	else if (rc > pc) {
961	>	if (workerCounts == wc &&   // try to suspend as spare
962	>	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
963	>	wc, wc - ONE_RUNNING) &&
964	>	!w.suspendAsSpare())    // false if killed
965	>	break;
966	>	}
967	>	else {
968	>	if (rc < pc || eventWaiters != 0L)
969	>	helpMaintainParallelism();
970	>	break;
971		}
820	–	if (!(inactivate |= active) &&  // must inactivate to suspend
821	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
822	–	wc, wc - ONE_RUNNING) &&
823	–	!w.suspendAsSpare())        // false if trimmed
824	–	return;
972		}
973		}
974
975		/**
976	<	* Tries to decrement running count, and if so, possibly creates
977	<	* or resumes compensating threads before blocking on task joinMe.
978	<	* This code is sprawled out with manual inlining to evade some
979	<	* JIT oddities.
976	>	* Helps and/or blocks awaiting join of the given task.
977	>	* Alternates between helpJoinTask() and helpMaintainParallelism()
978	>	* as many times as there is a deficit in running count (or longer
979	>	* if running count would become zero), then blocks if task still
980	>	* not done.
981		*
982		* @param joinMe the task to join
835	–	* @return task status on exit
983		*/
984	<	final int tryAwaitJoin(ForkJoinTask<?> joinMe) {
985	<	int cw = workerCounts; // read now to spoil CAS if counts change as ...
986	<	releaseWaiters();      // ... a byproduct of releaseWaiters
987	<	int stat = joinMe.status;
988	<	if (stat >= 0 && // inline variant of tryDecrementRunningCount
989	<	(cw & RUNNING_COUNT_MASK) > 0 &&
990	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
991	<	cw, cw - ONE_RUNNING)) {
992	<	int pc = parallelism;
993	<	int scans = 0;  // to require confirming passes to add threads
994	<	outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) {
995	<	if ((stat = joinMe.status) < 0)
996	<	break;
997	<	ForkJoinWorkerThread spare = null;
998	<	ForkJoinWorkerThread[] ws = workers;
999	<	int nws = ws.length;
1000	<	for (int i = 0; i < nws; ++i) {
1001	<	ForkJoinWorkerThread w = ws[i];
1002	<	if (w != null && w.isSuspended()) {
1003	<	spare = w;
1004	<	break;
1005	<	}
1006	<	}
1007	<	if ((stat = joinMe.status) < 0) // recheck to narrow race
1008	<	break;
862	<	int wc = workerCounts;
863	<	int rc = wc & RUNNING_COUNT_MASK;
864	<	if (rc >= pc)
865	<	break;
866	<	if (spare != null) {
867	<	if (spare.tryUnsuspend()) {
868	<	int c; // inline incrementRunningCount
869	<	do {} while (!UNSAFE.compareAndSwapInt
870	<	(this, workerCountsOffset,
871	<	c = workerCounts, c + ONE_RUNNING));
872	<	LockSupport.unpark(spare);
873	<	break;
874	<	}
875	<	continue;
876	<	}
877	<	int tc = wc >>> TOTAL_COUNT_SHIFT;
878	<	int sc = tc - pc;
879	<	if (rc > 0) {
880	<	int p = pc;
881	<	int s = sc;
882	<	while (s-- >= 0) { // try keeping 3/4 live
883	<	if (rc > (p -= (p >>> 2) + 1))
884	<	break outer;
885	<	}
886	<	}
887	<	if (scans++ > sc && tc < MAX_THREADS &&
888	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
889	<	wc + (ONE_RUNNING|ONE_TOTAL))) {
890	<	addWorker();
891	<	break;
892	<	}
984	>	final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) {
985	>	int threshold = parallelism;         // descend blocking thresholds
986	>	while (joinMe.status >= 0) {
987	>	boolean block; int wc;
988	>	worker.helpJoinTask(joinMe);
989	>	if (joinMe.status < 0)
990	>	break;
991	>	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
992	>	if (threshold > 0)
993	>	--threshold;
994	>	else
995	>	advanceEventCount(); // force release
996	>	block = false;
997	>	}
998	>	else
999	>	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1000	>	wc, wc - ONE_RUNNING);
1001	>	helpMaintainParallelism();
1002	>	if (block) {
1003	>	int c;
1004	>	joinMe.internalAwaitDone();
1005	>	do {} while (!UNSAFE.compareAndSwapInt
1006	>	(this, workerCountsOffset,
1007	>	c = workerCounts, c + ONE_RUNNING));
1008	>	break;
1009		}
894	–	if (stat >= 0)
895	–	stat = joinMe.internalAwaitDone();
896	–	int c; // inline incrementRunningCount
897	–	do {} while (!UNSAFE.compareAndSwapInt
898	–	(this, workerCountsOffset,
899	–	c = workerCounts, c + ONE_RUNNING));
1010		}
901	–	return stat;
1011		}
1012
1013		/**
1014	<	* Same idea as (and mostly pasted from) tryAwaitJoin, but
906	<	* self-contained
1014	>	* Same idea as awaitJoin, but no helping
1015		*/
1016		final void awaitBlocker(ManagedBlocker blocker)
1017		throws InterruptedException {
1018	<	for (;;) {
1019	<	if (blocker.isReleasable())
1020	<	return;
1021	<	int cw = workerCounts;
1022	<	releaseWaiters();
1023	<	if ((cw & RUNNING_COUNT_MASK) > 0 &&
1024	<	UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1025	<	cw, cw - ONE_RUNNING))
1026	<	break;
919	<	}
920	<	boolean done = false;
921	<	int pc = parallelism;
922	<	int scans = 0;
923	<	outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) {
924	<	if (done = blocker.isReleasable())
925	<	break;
926	<	ForkJoinWorkerThread spare = null;
927	<	ForkJoinWorkerThread[] ws = workers;
928	<	int nws = ws.length;
929	<	for (int i = 0; i < nws; ++i) {
930	<	ForkJoinWorkerThread w = ws[i];
931	<	if (w != null && w.isSuspended()) {
932	<	spare = w;
933	<	break;
934	<	}
1018	>	int threshold = parallelism;
1019	>	while (!blocker.isReleasable()) {
1020	>	boolean block; int wc;
1021	>	if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) {
1022	>	if (threshold > 0)
1023	>	--threshold;
1024	>	else
1025	>	advanceEventCount();
1026	>	block = false;
1027		}
1028	<	if (done = blocker.isReleasable())
1029	<	break;
1030	<	int wc = workerCounts;
1031	<	int rc = wc & RUNNING_COUNT_MASK;
1032	<	if (rc >= pc)
1033	<	break;
1034	<	if (spare != null) {
1035	<	if (spare.tryUnsuspend()) {
1028	>	else
1029	>	block = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1030	>	wc, wc - ONE_RUNNING);
1031	>	helpMaintainParallelism();
1032	>	if (block) {
1033	>	try {
1034	>	do {} while (!blocker.isReleasable() && !blocker.block());
1035	>	} finally {
1036		int c;
1037		do {} while (!UNSAFE.compareAndSwapInt
1038		(this, workerCountsOffset,
1039		c = workerCounts, c + ONE_RUNNING));
948	–	LockSupport.unpark(spare);
949	–	break;
950	–	}
951	–	continue;
952	–	}
953	–	int tc = wc >>> TOTAL_COUNT_SHIFT;
954	–	int sc = tc - pc;
955	–	if (rc > 0) {
956	–	int p = pc;
957	–	int s = sc;
958	–	while (s-- >= 0) {
959	–	if (rc > (p -= (p >>> 2) + 1))
960	–	break outer;
1040		}
962	–	}
963	–	if (scans++ > sc && tc < MAX_THREADS &&
964	–	UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
965	–	wc + (ONE_RUNNING|ONE_TOTAL))) {
966	–	addWorker();
1041		break;
1042		}
1043		}
1044	<	try {
971	<	if (!done)
972	<	do {} while (!blocker.isReleasable() &&
973	<	!blocker.block());
974	<	} finally {
975	<	int c;
976	<	do {} while (!UNSAFE.compareAndSwapInt
977	<	(this, workerCountsOffset,
978	<	c = workerCounts, c + ONE_RUNNING));
979	<	}
980	<	}
1044	>	}
1045
1046		/**
1047		* Possibly initiates and/or completes termination.
#	Line 1007 | Line 1071 | public class ForkJoinPool extends Abstra
1071
1072		/**
1073		* Actions on transition to TERMINATING
1074	+	*
1075	+	* Runs up to four passes through workers: (0) shutting down each
1076	+	* quietly (without waking up if parked) to quickly spread
1077	+	* notifications without unnecessary bouncing around event queues
1078	+	* etc (1) wake up and help cancel tasks (2) interrupt (3) mop up
1079	+	* races with interrupted workers
1080		*/
1081		private void startTerminating() {
1082	<	for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers
1083	<	cancelSubmissions();
1084	<	shutdownWorkers();
1085	<	cancelWorkerTasks();
1086	<	signalEvent();
1087	<	interruptWorkers();
1082	>	cancelSubmissions();
1083	>	for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1084	>	advanceEventCount();
1085	>	eventWaiters = 0L; // clobber lists
1086	>	spareWaiters = 0;
1087	>	ForkJoinWorkerThread[] ws = workers;
1088	>	int n = ws.length;
1089	>	for (int i = 0; i < n; ++i) {
1090	>	ForkJoinWorkerThread w = ws[i];
1091	>	if (w != null) {
1092	>	w.shutdown(true);
1093	>	if (passes > 0 && !w.isTerminated()) {
1094	>	w.cancelTasks();
1095	>	LockSupport.unpark(w);
1096	>	if (passes > 1) {
1097	>	try {
1098	>	w.interrupt();
1099	>	} catch (SecurityException ignore) {
1100	>	}
1101	>	}
1102	>	}
1103	>	}
1104	>	}
1105		}
1106		}
1107
#	Line 1031 | Line 1118 | public class ForkJoinPool extends Abstra
1118		}
1119		}
1120
1034	–	/**
1035	–	* Sets all worker run states to at least shutdown,
1036	–	* also resuming suspended workers
1037	–	*/
1038	–	private void shutdownWorkers() {
1039	–	ForkJoinWorkerThread[] ws = workers;
1040	–	int nws = ws.length;
1041	–	for (int i = 0; i < nws; ++i) {
1042	–	ForkJoinWorkerThread w = ws[i];
1043	–	if (w != null)
1044	–	w.shutdown();
1045	–	}
1046	–	}
1047	–
1048	–	/**
1049	–	* Clears out and cancels all locally queued tasks
1050	–	*/
1051	–	private void cancelWorkerTasks() {
1052	–	ForkJoinWorkerThread[] ws = workers;
1053	–	int nws = ws.length;
1054	–	for (int i = 0; i < nws; ++i) {
1055	–	ForkJoinWorkerThread w = ws[i];
1056	–	if (w != null)
1057	–	w.cancelTasks();
1058	–	}
1059	–	}
1060	–
1061	–	/**
1062	–	* Unsticks all workers blocked on joins etc
1063	–	*/
1064	–	private void interruptWorkers() {
1065	–	ForkJoinWorkerThread[] ws = workers;
1066	–	int nws = ws.length;
1067	–	for (int i = 0; i < nws; ++i) {
1068	–	ForkJoinWorkerThread w = ws[i];
1069	–	if (w != null && !w.isTerminated()) {
1070	–	try {
1071	–	w.interrupt();
1072	–	} catch (SecurityException ignore) {
1073	–	}
1074	–	}
1075	–	}
1076	–	}
1077	–
1121		// misc support for ForkJoinWorkerThread
1122
1123		/**
#	Line 1085 | Line 1128 | public class ForkJoinPool extends Abstra
1128		}
1129
1130		/**
1131	<	* Accumulates steal count from a worker, clearing
1132	<	* the worker's value
1131	>	* Tries to accumulates steal count from a worker, clearing
1132	>	* the worker's value.
1133	>	*
1134	>	* @return true if worker steal count now zero
1135		*/
1136	<	final void accumulateStealCount(ForkJoinWorkerThread w) {
1136	>	final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1137		int sc = w.stealCount;
1138	<	if (sc != 0) {
1139	<	long c;
1140	<	w.stealCount = 0;
1141	<	do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1142	<	c = stealCount, c + sc));
1138	>	long c = stealCount;
1139	>	// CAS even if zero, for fence effects
1140	>	if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1141	>	if (sc != 0)
1142	>	w.stealCount = 0;
1143	>	return true;
1144		}
1145	+	return sc == 0;
1146		}
1147
1148		/**
#	Line 1103 | Line 1150 | public class ForkJoinPool extends Abstra
1150		* active thread.
1151		*/
1152		final int idlePerActive() {
1153	<	int pc = parallelism; // use targeted parallelism, not rc
1153	>	int pc = parallelism; // use parallelism, not rc
1154		int ac = runState;    // no mask -- artifically boosts during shutdown
1155		// Use exact results for small values, saturate past 4
1156		return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
#	Line 1154 | Line 1201 | public class ForkJoinPool extends Abstra
1201		* use {@link java.lang.Runtime#availableProcessors}.
1202		* @param factory the factory for creating new threads. For default value,
1203		* use {@link #defaultForkJoinWorkerThreadFactory}.
1204	<	* @param handler the handler for internal worker threads that
1205	<	* terminate due to unrecoverable errors encountered while executing
1204	>	* @param handler the handler for internal worker threads that
1205	>	* terminate due to unrecoverable errors encountered while executing
1206		* tasks. For default value, use <code>null</code>.
1207	<	* @param asyncMode if true,
1207	>	* @param asyncMode if true,
1208		* establishes local first-in-first-out scheduling mode for forked
1209		* tasks that are never joined. This mode may be more appropriate
1210		* than default locally stack-based mode in applications in which
#	Line 1171 | Line 1218 | public class ForkJoinPool extends Abstra
1218		*         because it does not hold {@link
1219		*         java.lang.RuntimePermission}{@code ("modifyThread")}
1220		*/
1221	<	public ForkJoinPool(int parallelism,
1221	>	public ForkJoinPool(int parallelism,
1222		ForkJoinWorkerThreadFactory factory,
1223		Thread.UncaughtExceptionHandler handler,
1224		boolean asyncMode) {
1225		checkPermission();
1226		if (factory == null)
1227		throw new NullPointerException();
1228	<	if (parallelism <= 0 || parallelism > MAX_THREADS)
1228	>	if (parallelism <= 0 || parallelism > MAX_WORKERS)
1229		throw new IllegalArgumentException();
1230		this.parallelism = parallelism;
1231		this.factory = factory;
#	Line 1190 | Line 1237 | public class ForkJoinPool extends Abstra
1237		this.workerLock = new ReentrantLock();
1238		this.termination = new Phaser(1);
1239		this.poolNumber = poolNumberGenerator.incrementAndGet();
1240	+	this.trimTime = System.nanoTime();
1241		}
1242
1243		/**
#	Line 1197 | Line 1245 | public class ForkJoinPool extends Abstra
1245		* @param pc the initial parallelism level
1246		*/
1247		private static int initialArraySizeFor(int pc) {
1248	<	// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1249	<	int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1248	>	// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1249	>	int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1250		size |= size >>> 1;
1251		size |= size >>> 2;
1252		size |= size >>> 4;
#	Line 1216 | Line 1264 | public class ForkJoinPool extends Abstra
1264		throw new NullPointerException();
1265		if (runState >= SHUTDOWN)
1266		throw new RejectedExecutionException();
1267	<	// Convert submissions to current pool into forks
1268	<	Thread t = Thread.currentThread();
1269	<	ForkJoinWorkerThread w;
1222	<	if ((t instanceof ForkJoinWorkerThread) &&
1223	<	(w = (ForkJoinWorkerThread) t).pool == this)
1224	<	w.pushTask(task);
1225	<	else {
1226	<	submissionQueue.offer(task);
1227	<	signalEvent();
1228	<	ensureEnoughTotalWorkers();
1229	<	}
1267	>	submissionQueue.offer(task);
1268	>	advanceEventCount();
1269	>	helpMaintainParallelism();         // start or wake up workers
1270		}
1271
1272		/**
1273		* Performs the given task, returning its result upon completion.
1274		* If the caller is already engaged in a fork/join computation in
1275	<	* the current pool, this method is equivalent in effect to
1275	>	* the current pool, this method is equivalent in effect to
1276		* {@link ForkJoinTask#invoke}.
1277		*
1278		* @param task the task
#	Line 1249 | Line 1289 | public class ForkJoinPool extends Abstra
1289		/**
1290		* Arranges for (asynchronous) execution of the given task.
1291		* If the caller is already engaged in a fork/join computation in
1292	<	* the current pool, this method is equivalent in effect to
1292	>	* the current pool, this method is equivalent in effect to
1293		* {@link ForkJoinTask#fork}.
1294		*
1295		* @param task the task
#	Line 1280 | Line 1320 | public class ForkJoinPool extends Abstra
1320		/**
1321		* Submits a ForkJoinTask for execution.
1322		* If the caller is already engaged in a fork/join computation in
1323	<	* the current pool, this method is equivalent in effect to
1323	>	* the current pool, this method is equivalent in effect to
1324		* {@link ForkJoinTask#fork}.
1325		*
1326		* @param task the task to submit
#	Line 1473 | Line 1513 | public class ForkJoinPool extends Abstra
1513		public long getQueuedTaskCount() {
1514		long count = 0;
1515		ForkJoinWorkerThread[] ws = workers;
1516	<	int nws = ws.length;
1517	<	for (int i = 0; i < nws; ++i) {
1516	>	int n = ws.length;
1517	>	for (int i = 0; i < n; ++i) {
1518		ForkJoinWorkerThread w = ws[i];
1519		if (w != null)
1520		count += w.getQueueSize();
#	Line 1532 | Line 1572 | public class ForkJoinPool extends Abstra
1572		* @return the number of elements transferred
1573		*/
1574		protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1575	<	int n = submissionQueue.drainTo(c);
1536	<	ForkJoinWorkerThread[] ws = workers;
1537	<	int nws = ws.length;
1538	<	for (int i = 0; i < nws; ++i) {
1539	<	ForkJoinWorkerThread w = ws[i];
1540	<	if (w != null)
1541	<	n += w.drainTasksTo(c);
1542	<	}
1543	<	return n;
1544	<	}
1545	<
1546	<	/**
1547	<	* Returns count of total parks by existing workers.
1548	<	* Used during development only since not meaningful to users.
1549	<	*/
1550	<	private int collectParkCount() {
1551	<	int count = 0;
1575	>	int count = submissionQueue.drainTo(c);
1576		ForkJoinWorkerThread[] ws = workers;
1577	<	int nws = ws.length;
1578	<	for (int i = 0; i < nws; ++i) {
1577	>	int n = ws.length;
1578	>	for (int i = 0; i < n; ++i) {
1579		ForkJoinWorkerThread w = ws[i];
1580		if (w != null)
1581	<	count += w.parkCount;
1581	>	count += w.drainTasksTo(c);
1582		}
1583		return count;
1584		}
#	Line 1576 | Line 1600 | public class ForkJoinPool extends Abstra
1600		int pc = parallelism;
1601		int rs = runState;
1602		int ac = rs & ACTIVE_COUNT_MASK;
1579	–	//        int pk = collectParkCount();
1603		return super.toString() +
1604		"[" + runLevelToString(rs) +
1605		", parallelism = " + pc +
#	Line 1586 | Line 1609 | public class ForkJoinPool extends Abstra
1609		", steals = " + st +
1610		", tasks = " + qt +
1611		", submissions = " + qs +
1589	–	//            ", parks = " + pk +
1612		"]";
1613		}
1614
#	Line 1693 | Line 1715 | public class ForkJoinPool extends Abstra
1715		* Interface for extending managed parallelism for tasks running
1716		* in {@link ForkJoinPool}s.
1717		*
1718	<	* <p>A {@code ManagedBlocker} provides two methods.
1719	<	* Method {@code isReleasable} must return {@code true} if
1720	<	* blocking is not necessary. Method {@code block} blocks the
1721	<	* current thread if necessary (perhaps internally invoking
1722	<	* {@code isReleasable} before actually blocking).
1718	>	* <p>A {@code ManagedBlocker} provides two methods.  Method
1719	>	* {@code isReleasable} must return {@code true} if blocking is
1720	>	* not necessary. Method {@code block} blocks the current thread
1721	>	* if necessary (perhaps internally invoking {@code isReleasable}
1722	>	* before actually blocking). The unusual methods in this API
1723	>	* accommodate synchronizers that may, but don't usually, block
1724	>	* for long periods. Similarly, they allow more efficient internal
1725	>	* handling of cases in which additional workers may be, but
1726	>	* usually are not, needed to ensure sufficient parallelism.
1727	>	* Toward this end, implementations of method {@code isReleasable}
1728	>	* must be amenable to repeated invocation.
1729		*
1730		* <p>For example, here is a ManagedBlocker based on a
1731		* ReentrantLock:
#	Line 1715 | Line 1743 | public class ForkJoinPool extends Abstra
1743		*     return hasLock || (hasLock = lock.tryLock());
1744		*   }
1745		* }}</pre>
1746	+	*
1747	+	* <p>Here is a class that possibly blocks waiting for an
1748	+	* item on a given queue:
1749	+	*  <pre> {@code
1750	+	* class QueueTaker<E> implements ManagedBlocker {
1751	+	*   final BlockingQueue<E> queue;
1752	+	*   volatile E item = null;
1753	+	*   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1754	+	*   public boolean block() throws InterruptedException {
1755	+	*     if (item == null)
1756	+	*       item = queue.take
1757	+	*     return true;
1758	+	*   }
1759	+	*   public boolean isReleasable() {
1760	+	*     return item != null || (item = queue.poll) != null;
1761	+	*   }
1762	+	*   public E getItem() { // call after pool.managedBlock completes
1763	+	*     return item;
1764	+	*   }
1765	+	* }}</pre>
1766		*/
1767		public static interface ManagedBlocker {
1768		/**
#	Line 1757 | Line 1805 | public class ForkJoinPool extends Abstra
1805		public static void managedBlock(ManagedBlocker blocker)
1806		throws InterruptedException {
1807		Thread t = Thread.currentThread();
1808	<	if (t instanceof ForkJoinWorkerThread)
1809	<	((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker);
1808	>	if (t instanceof ForkJoinWorkerThread) {
1809	>	ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1810	>	w.pool.awaitBlocker(blocker);
1811	>	}
1812		else {
1813		do {} while (!blocker.isReleasable() && !blocker.block());
1814		}
#	Line 1789 | Line 1839 | public class ForkJoinPool extends Abstra
1839		objectFieldOffset("eventWaiters",ForkJoinPool.class);
1840		private static final long stealCountOffset =
1841		objectFieldOffset("stealCount",ForkJoinPool.class);
1842	+	private static final long spareWaitersOffset =
1843	+	objectFieldOffset("spareWaiters",ForkJoinPool.class);
1844
1845		private static long objectFieldOffset(String field, Class<?> klazz) {
1846		try {

Diff Legend

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