ViewVC Help

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

root/jsr166/jsr166/src/jsr166y/ForkJoinWorkerThread.java

Comparing jsr166/src/jsr166y/ForkJoinWorkerThread.java (file contents):
Revision 1.31 by dl, Mon Apr 5 15:52:26 2010 UTC vs.
Revision 1.51 by jsr166, Mon Sep 20 20:42:37 2010 UTC

#	Line 83 | Line 83 | public class ForkJoinWorkerThread extend
83		* by the ForkJoinPool).  This allows use in message-passing
84		* frameworks in which tasks are never joined.
85		*
86	<	* Efficient implementation of this approach currently relies on
87	<	* an uncomfortable amount of "Unsafe" mechanics. To maintain
86	>	* When a worker would otherwise be blocked waiting to join a
87	>	* task, it first tries a form of linear helping: Each worker
88	>	* records (in field currentSteal) the most recent task it stole
89	>	* from some other worker. Plus, it records (in field currentJoin)
90	>	* the task it is currently actively joining. Method joinTask uses
91	>	* these markers to try to find a worker to help (i.e., steal back
92	>	* a task from and execute it) that could hasten completion of the
93	>	* actively joined task. In essence, the joiner executes a task
94	>	* that would be on its own local deque had the to-be-joined task
95	>	* not been stolen. This may be seen as a conservative variant of
96	>	* the approach in Wagner & Calder "Leapfrogging: a portable
97	>	* technique for implementing efficient futures" SIGPLAN Notices,
98	>	* 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs
99	>	* in that: (1) We only maintain dependency links across workers
100	>	* upon steals, rather than use per-task bookkeeping.  This may
101	>	* require a linear scan of workers array to locate stealers, but
102	>	* usually doesn't because stealers leave hints (that may become
103	>	* stale/wrong) of where to locate them. This isolates cost to
104	>	* when it is needed, rather than adding to per-task overhead.
105	>	* (2) It is "shallow", ignoring nesting and potentially cyclic
106	>	* mutual steals.  (3) It is intentionally racy: field currentJoin
107	>	* is updated only while actively joining, which means that we
108	>	* miss links in the chain during long-lived tasks, GC stalls etc
109	>	* (which is OK since blocking in such cases is usually a good
110	>	* idea).  (4) We bound the number of attempts to find work (see
111	>	* MAX_HELP_DEPTH) and fall back to suspending the worker and if
112	>	* necessary replacing it with a spare (see
113	>	* ForkJoinPool.awaitJoin).
114	>	*
115	>	* Efficient implementation of these algorithms currently relies
116	>	* on an uncomfortable amount of "Unsafe" mechanics. To maintain
117		* correct orderings, reads and writes of variable base require
118		* volatile ordering.  Variable sp does not require volatile
119		* writes but still needs store-ordering, which we accomplish by
120		* pre-incrementing sp before filling the slot with an ordered
121		* store.  (Pre-incrementing also enables backouts used in
122	<	* scanWhileJoining.)  Because they are protected by volatile base
123	<	* reads, reads of the queue array and its slots by other threads
124	<	* do not need volatile load semantics, but writes (in push)
125	<	* require store order and CASes (in pop and deq) require
126	<	* (volatile) CAS semantics.  (Michael, Saraswat, and Vechev's
127	<	* algorithm has similar properties, but without support for
128	<	* nulling slots.)  Since these combinations aren't supported
129	<	* using ordinary volatiles, the only way to accomplish these
130	<	* efficiently is to use direct Unsafe calls. (Using external
131	<	* AtomicIntegers and AtomicReferenceArrays for the indices and
132	<	* array is significantly slower because of memory locality and
133	<	* indirection effects.)
122	>	* joinTask.)  Because they are protected by volatile base reads,
123	>	* reads of the queue array and its slots by other threads do not
124	>	* need volatile load semantics, but writes (in push) require
125	>	* store order and CASes (in pop and deq) require (volatile) CAS
126	>	* semantics.  (Michael, Saraswat, and Vechev's algorithm has
127	>	* similar properties, but without support for nulling slots.)
128	>	* Since these combinations aren't supported using ordinary
129	>	* volatiles, the only way to accomplish these efficiently is to
130	>	* use direct Unsafe calls. (Using external AtomicIntegers and
131	>	* AtomicReferenceArrays for the indices and array is
132	>	* significantly slower because of memory locality and indirection
133	>	* effects.)
134		*
135		* Further, performance on most platforms is very sensitive to
136		* placement and sizing of the (resizable) queue array.  Even
#	Line 126 | Line 155 | public class ForkJoinWorkerThread extend
155		private static final Random seedGenerator = new Random();
156
157		/**
158	<	* The timeout value for suspending spares. Spare workers that
159	<	* remain unsignalled for more than this time may be trimmed
160	<	* (killed and removed from pool).  Since our goal is to avoid
161	<	* long-term thread buildup, the exact value of timeout does not
162	<	* matter too much so long as it avoids most false-alarm timeouts
134	<	* under GC stalls or momentarily high system load.
158	>	* The maximum stolen->joining link depth allowed in helpJoinTask.
159	>	* Depths for legitimate chains are unbounded, but we use a fixed
160	>	* constant to avoid (otherwise unchecked) cycles and bound
161	>	* staleness of traversal parameters at the expense of sometimes
162	>	* blocking when we could be helping.
163		*/
164	<	private static final long SPARE_KEEPALIVE_NANOS =
137	<	5L * 1000L * 1000L * 1000L; // 5 secs
164	>	private static final int MAX_HELP_DEPTH = 8;
165
166		/**
167		* Capacity of work-stealing queue array upon initialization.
168	<	* Must be a power of two. Initial size must be at least 2, but is
168	>	* Must be a power of two. Initial size must be at least 4, but is
169		* padded to minimize cache effects.
170		*/
171		private static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
172
173		/**
174		* Maximum work-stealing queue array size.  Must be less than or
175	<	* equal to 1 << 28 to ensure lack of index wraparound. (This
176	<	* is less than usual bounds, because we need leftshift by 3
177	<	* to be in int range).
175	>	* equal to 1 << (31 - width of array entry) to ensure lack of
176	>	* index wraparound. The value is set in the static block
177	>	* at the end of this file after obtaining width.
178		*/
179	<	private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
179	>	private static final int MAXIMUM_QUEUE_CAPACITY;
180
181		/**
182		* The pool this thread works in. Accessed directly by ForkJoinTask.
#	Line 178 | Line 205 | public class ForkJoinWorkerThread extend
205		private int sp;
206
207		/**
208	+	* The index of most recent stealer, used as a hint to avoid
209	+	* traversal in method helpJoinTask. This is only a hint because a
210	+	* worker might have had multiple steals and this only holds one
211	+	* of them (usually the most current). Declared non-volatile,
212	+	* relying on other prevailing sync to keep reasonably current.
213	+	*/
214	+	private int stealHint;
215	+
216	+	/**
217		* Run state of this worker. In addition to the usual run levels,
218		* tracks if this worker is suspended as a spare, and if it was
219		* killed (trimmed) while suspended. However, "active" status is
220	<	* maintained separately.
220	>	* maintained separately and modified only in conjunction with
221	>	* CASes of the pool's runState (which are currently sadly
222	>	* manually inlined for performance.)  Accessed directly by pool
223	>	* to simplify checks for normal (zero) status.
224		*/
225	<	private volatile int runState;
225	>	volatile int runState;
226
227		private static final int TERMINATING = 0x01;
228		private static final int TERMINATED  = 0x02;
#	Line 191 | Line 230 | public class ForkJoinWorkerThread extend
230		private static final int TRIMMED     = 0x08; // killed while suspended
231
232		/**
233	<	* Number of LockSupport.park calls to block this thread for
234	<	* suspension or event waits. Used for internal instrumention;
196	<	* currently not exported but included because volatile write upon
197	<	* park also provides a workaround for a JVM bug.
198	<	*/
199	<	private volatile int parkCount;
200	<
201	<	/**
202	<	* Number of steals, transferred and reset in pool callbacks pool
203	<	* when idle Accessed directly by pool.
233	>	* Number of steals. Directly accessed (and reset) by
234	>	* pool.tryAccumulateStealCount when idle.
235		*/
236		int stealCount;
237
#	Line 218 | Line 249 | public class ForkJoinWorkerThread extend
249
250		/**
251		* True if use local fifo, not default lifo, for local polling.
252	<	* Shadows value from ForkJoinPool, which resets it if changed
222	<	* pool-wide.
252	>	* Shadows value from ForkJoinPool.
253		*/
254	<	private boolean locallyFifo;
254	>	private final boolean locallyFifo;
255
256		/**
257		* Index of this worker in pool array. Set once by pool before
#	Line 237 | Line 267 | public class ForkJoinWorkerThread extend
267		int lastEventCount;
268
269		/**
270	<	* Encoded index and event count of next event waiter. Used only
271	<	* by ForkJoinPool for managing event waiters.
270	>	* Encoded index and event count of next event waiter. Accessed
271	>	* only by ForkJoinPool for managing event waiters.
272		*/
273		volatile long nextWaiter;
274
275		/**
276	+	* Number of times this thread suspended as spare. Accessed only
277	+	* by pool.
278	+	*/
279	+	int spareCount;
280	+
281	+	/**
282	+	* Encoded index and count of next spare waiter. Accessed only
283	+	* by ForkJoinPool for managing spares.
284	+	*/
285	+	volatile int nextSpare;
286	+
287	+	/**
288	+	* The task currently being joined, set only when actively trying
289	+	* to help other stealers in helpJoinTask. Written only by this
290	+	* thread, but read by others.
291	+	*/
292	+	private volatile ForkJoinTask<?> currentJoin;
293	+
294	+	/**
295	+	* The task most recently stolen from another worker (or
296	+	* submission queue).  Written only by this thread, but read by
297	+	* others.
298	+	*/
299	+	private volatile ForkJoinTask<?> currentSteal;
300	+
301	+	/**
302		* Creates a ForkJoinWorkerThread operating in the given pool.
303		*
304		* @param pool the pool this thread works in
305		* @throws NullPointerException if pool is null
306		*/
307		protected ForkJoinWorkerThread(ForkJoinPool pool) {
252	–	if (pool == null) throw new NullPointerException();
308		this.pool = pool;
309	+	this.locallyFifo = pool.locallyFifo;
310	+	setDaemon(true);
311		// To avoid exposing construction details to subclasses,
312		// remaining initialization is in start() and onStart()
313		}
314
315		/**
316	<	* Performs additional initialization and starts this thread
316	>	* Performs additional initialization and starts this thread.
317		*/
318	<	final void start(int poolIndex, boolean locallyFifo,
262	<	UncaughtExceptionHandler ueh) {
318	>	final void start(int poolIndex, UncaughtExceptionHandler ueh) {
319		this.poolIndex = poolIndex;
264	–	this.locallyFifo = locallyFifo;
320		if (ueh != null)
321		setUncaughtExceptionHandler(ueh);
267	–	setDaemon(true);
322		start();
323		}
324
#	Line 295 | Line 349 | public class ForkJoinWorkerThread extend
349		/**
350		* Initializes internal state after construction but before
351		* processing any tasks. If you override this method, you must
352	<	* invoke super.onStart() at the beginning of the method.
352	>	* invoke @code{super.onStart()} at the beginning of the method.
353		* Initialization requires care: Most fields must have legal
354		* default values, to ensure that attempted accesses from other
355		* threads work correctly even before this thread starts
#	Line 305 | Line 359 | public class ForkJoinWorkerThread extend
359		int rs = seedGenerator.nextInt();
360		seed = rs == 0? 1 : rs; // seed must be nonzero
361
362	<	// Allocate name string and queue array in this thread
362	>	// Allocate name string and arrays in this thread
363		String pid = Integer.toString(pool.getPoolNumber());
364		String wid = Integer.toString(poolIndex);
365		setName("ForkJoinPool-" + pid + "-worker-" + wid);
#	Line 323 | Line 377 | public class ForkJoinWorkerThread extend
377		*/
378		protected void onTermination(Throwable exception) {
379		try {
380	+	ForkJoinPool p = pool;
381	+	if (active) {
382	+	int a; // inline p.tryDecrementActiveCount
383	+	active = false;
384	+	do {} while (!UNSAFE.compareAndSwapInt
385	+	(p, poolRunStateOffset, a = p.runState, a - 1));
386	+	}
387		cancelTasks();
388		setTerminated();
389	<	pool.workerTerminated(this);
389	>	p.workerTerminated(this);
390		} catch (Throwable ex) {        // Shouldn't ever happen
391		if (exception == null)      // but if so, at least rethrown
392		exception = ex;
#	Line 355 | Line 416 | public class ForkJoinWorkerThread extend
416		// helpers for run()
417
418		/**
419	<	* Find and execute tasks and check status while running
419	>	* Finds and executes tasks, and checks status while running.
420		*/
421		private void mainLoop() {
422	<	boolean ran = false; // true if ran task on previous step
422	>	boolean ran = false; // true if ran a task on last step
423		ForkJoinPool p = pool;
424		for (;;) {
425		p.preStep(this, ran);
426		if (runState != 0)
427	<	return;
428	<	ForkJoinTask<?> t; // try to get and run stolen or submitted task
368	<	if (ran = (t = scan()) != null || (t = pollSubmission()) != null) {
369	<	t.tryExec();
370	<	if (base != sp)
371	<	runLocalTasks();
372	<	}
427	>	break;
428	>	ran = tryExecSteal() || tryExecSubmission();
429		}
430		}
431
432		/**
433	<	* Runs local tasks until queue is empty or shut down.  Call only
434	<	* while active.
433	>	* Tries to steal a task and execute it.
434	>	*
435	>	* @return true if ran a task
436		*/
437	<	private void runLocalTasks() {
438	<	while (runState == 0) {
439	<	ForkJoinTask<?> t = locallyFifo? locallyDeqTask() : popTask();
440	<	if (t != null)
441	<	t.tryExec();
442	<	else if (base == sp)
443	<	break;
437	>	private boolean tryExecSteal() {
438	>	ForkJoinTask<?> t;
439	>	if ((t = scan()) != null) {
440	>	t.quietlyExec();
441	>	UNSAFE.putOrderedObject(this, currentStealOffset, null);
442	>	if (sp != base)
443	>	execLocalTasks();
444	>	return true;
445		}
446	+	return false;
447		}
448
449		/**
450	<	* If a submission exists, try to activate and take it
450	>	* If a submission exists, try to activate and run it.
451		*
452	<	* @return a task, if available
452	>	* @return true if ran a task
453		*/
454	<	private ForkJoinTask<?> pollSubmission() {
454	>	private boolean tryExecSubmission() {
455		ForkJoinPool p = pool;
456	+	// This loop is needed in case attempt to activate fails, in
457	+	// which case we only retry if there still appears to be a
458	+	// submission.
459		while (p.hasQueuedSubmissions()) {
460	<	if (active || (active = p.tryIncrementActiveCount())) {
461	<	ForkJoinTask<?> t = p.pollSubmission();
462	<	return t != null ? t : scan(); // if missed, rescan
460	>	ForkJoinTask<?> t; int a;
461	>	if (active || // inline p.tryIncrementActiveCount
462	>	(active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
463	>	a = p.runState, a + 1))) {
464	>	if ((t = p.pollSubmission()) != null) {
465	>	UNSAFE.putOrderedObject(this, currentStealOffset, t);
466	>	t.quietlyExec();
467	>	UNSAFE.putOrderedObject(this, currentStealOffset, null);
468	>	if (sp != base)
469	>	execLocalTasks();
470	>	return true;
471	>	}
472		}
473		}
474	<	return null;
474	>	return false;
475	>	}
476	>
477	>	/**
478	>	* Runs local tasks until queue is empty or shut down.  Call only
479	>	* while active.
480	>	*/
481	>	private void execLocalTasks() {
482	>	while (runState == 0) {
483	>	ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask();
484	>	if (t != null)
485	>	t.quietlyExec();
486	>	else if (sp == base)
487	>	break;
488	>	}
489		}
490
491		/*
492		* Intrinsics-based atomic writes for queue slots. These are
493	<	* basically the same as methods in AtomicObjectArray, but
493	>	* basically the same as methods in AtomicReferenceArray, but
494		* specialized for (1) ForkJoinTask elements (2) requirement that
495		* nullness and bounds checks have already been performed by
496		* callers and (3) effective offsets are known not to overflow
497		* from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't
498		* need corresponding version for reads: plain array reads are OK
499	<	* because they protected by other volatile reads and are
499	>	* because they are protected by other volatile reads and are
500		* confirmed by CASes.
501		*
502		* Most uses don't actually call these methods, but instead contain
#	Line 435 | Line 520 | public class ForkJoinWorkerThread extend
520		* range. This method is used only during resets and backouts.
521		*/
522		private static final void writeSlot(ForkJoinTask<?>[] q, int i,
523	<	ForkJoinTask<?> t) {
523	>	ForkJoinTask<?> t) {
524		UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
525		}
526
#	Line 447 | Line 532 | public class ForkJoinWorkerThread extend
532		* @param t the task. Caller must ensure non-null.
533		*/
534		final void pushTask(ForkJoinTask<?> t) {
450	–	int s;
535		ForkJoinTask<?>[] q = queue;
536		int mask = q.length - 1; // implicit assert q != null
537	<	UNSAFE.putOrderedObject(q, (((s = sp++) & mask) << qShift) + qBase, t);
538	<	if ((s -= base) <= 0)
539	<	pool.signalWork();
540	<	else if (s + 1 >= mask)
541	<	growQueue();
537	>	int s = sp++;            // ok to increment sp before slot write
538	>	UNSAFE.putOrderedObject(q, ((s & mask) << qShift) + qBase, t);
539	>	if ((s -= base) == 0)
540	>	pool.signalWork();   // was empty
541	>	else if (s == mask)
542	>	growQueue();         // is full
543		}
544
545		/**
546		* Tries to take a task from the base of the queue, failing if
547		* empty or contended. Note: Specializations of this code appear
548	<	* in scan and scanWhileJoining.
548	>	* in locallyDeqTask and elsewhere.
549		*
550		* @return a task, or null if none or contended
551		*/
#	Line 468 | Line 553 | public class ForkJoinWorkerThread extend
553		ForkJoinTask<?> t;
554		ForkJoinTask<?>[] q;
555		int b, i;
556	<	if ((b = base) != sp &&
556	>	if (sp != (b = base) &&
557		(q = queue) != null && // must read q after b
558	<	(t = q[i = (q.length - 1) & b]) != null &&
558	>	(t = q[i = (q.length - 1) & b]) != null && base == b &&
559		UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) {
560		base = b + 1;
561		return t;
#	Line 480 | Line 565 | public class ForkJoinWorkerThread extend
565
566		/**
567		* Tries to take a task from the base of own queue. Assumes active
568	<	* status.  Called only by current thread.
568	>	* status.  Called only by this thread.
569		*
570		* @return a task, or null if none
571		*/
#	Line 490 | Line 575 | public class ForkJoinWorkerThread extend
575		ForkJoinTask<?> t;
576		int b, i;
577		while (sp != (b = base)) {
578	<	if ((t = q[i = (q.length - 1) & b]) != null &&
578	>	if ((t = q[i = (q.length - 1) & b]) != null && base == b &&
579		UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase,
580		t, null)) {
581		base = b + 1;
#	Line 503 | Line 588 | public class ForkJoinWorkerThread extend
588
589		/**
590		* Returns a popped task, or null if empty. Assumes active status.
591	<	* Called only by current thread. (Note: a specialization of this
507	<	* code appears in scanWhileJoining.)
591	>	* Called only by this thread.
592		*/
593	<	final ForkJoinTask<?> popTask() {
510	<	int s;
593	>	private ForkJoinTask<?> popTask() {
594		ForkJoinTask<?>[] q = queue;
595	<	if (q != null && (s = sp) != base) {
596	<	int i = (q.length - 1) & --s;
597	<	ForkJoinTask<?> t = q[i];
598	<	if (t != null && UNSAFE.compareAndSwapObject
599	<	(q, (i << qShift) + qBase, t, null)) {
600	<	sp = s;
601	<	return t;
595	>	if (q != null) {
596	>	int s;
597	>	while ((s = sp) != base) {
598	>	int i = (q.length - 1) & --s;
599	>	long u = (i << qShift) + qBase; // raw offset
600	>	ForkJoinTask<?> t = q[i];
601	>	if (t == null)   // lost to stealer
602	>	break;
603	>	if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
604	>	sp = s; // putOrderedInt may encourage more timely write
605	>	// UNSAFE.putOrderedInt(this, spOffset, s);
606	>	return t;
607	>	}
608		}
609		}
610		return null;
611		}
612
613		/**
614	<	* Specialized version of popTask to pop only if
615	<	* topmost element is the given task. Called only
527	<	* by current thread while active.
614	>	* Specialized version of popTask to pop only if topmost element
615	>	* is the given task. Called only by this thread while active.
616		*
617		* @param t the task. Caller must ensure non-null.
618		*/
619		final boolean unpushTask(ForkJoinTask<?> t) {
620		int s;
621		ForkJoinTask<?>[] q = queue;
622	<	if (q != null && UNSAFE.compareAndSwapObject
623	<	(q, (((q.length - 1) & (s = sp - 1)) << qShift) + qBase, t, null)){
624	<	sp = s;
622	>	if ((s = sp) != base && q != null &&
623	>	UNSAFE.compareAndSwapObject
624	>	(q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) {
625	>	sp = s; // putOrderedInt may encourage more timely write
626	>	// UNSAFE.putOrderedInt(this, spOffset, s);
627		return true;
628		}
629		return false;
630		}
631
632		/**
633	<	* Returns next task or null if empty or contended
633	>	* Returns next task, or null if empty or contended.
634		*/
635		final ForkJoinTask<?> peekTask() {
636		ForkJoinTask<?>[] q = queue;
#	Line 582 | Line 672 | public class ForkJoinWorkerThread extend
672		* Computes next value for random victim probe in scan().  Scans
673		* don't require a very high quality generator, but also not a
674		* crummy one.  Marsaglia xor-shift is cheap and works well enough.
675	<	* Note: This is manually inlined in scan()
675	>	* Note: This is manually inlined in scan().
676		*/
677		private static final int xorShift(int r) {
678		r ^= r << 13;
#	Line 610 | Line 700 | public class ForkJoinWorkerThread extend
700		*/
701		private ForkJoinTask<?> scan() {
702		ForkJoinPool p = pool;
703	<	ForkJoinWorkerThread[] ws = p.workers;
704	<	int n = ws.length;            // upper bound of #workers
705	<	boolean canSteal = active;    // shadow active status
706	<	int r = seed;                 // extract seed once
707	<	int k = r;                    // index: random if j<0 else step
708	<	for (int j = -n; j < n; ++j) {
709	<	ForkJoinWorkerThread v = ws[k & (n - 1)];
710	<	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
711	<	if (v != null && v.base != v.sp) {
712	<	if (canSteal ||       // ensure active status
713	<	(canSteal = active = p.tryIncrementActiveCount())) {
714	<	int b, i;         // inlined specialization of deqTask
715	<	ForkJoinTask<?> t;
716	<	ForkJoinTask<?>[] q;
717	<	if ((b = v.base) != v.sp &&  // recheck
718	<	(q = v.queue) != null &&
719	<	(t = q[i = (q.length - 1) & b]) != null &&
720	<	UNSAFE.compareAndSwapObject
721	<	(q, (i << qShift) + qBase, t, null)) {
722	<	v.base = b + 1;
723	<	seed = r;
724	<	++stealCount;
725	<	return t;
703	>	ForkJoinWorkerThread[] ws;        // worker array
704	>	int n;                            // upper bound of #workers
705	>	if ((ws = p.workers) != null && (n = ws.length) > 1) {
706	>	boolean canSteal = active;    // shadow active status
707	>	int r = seed;                 // extract seed once
708	>	int mask = n - 1;
709	>	int j = -n;                   // loop counter
710	>	int k = r;                    // worker index, random if j < 0
711	>	for (;;) {
712	>	ForkJoinWorkerThread v = ws[k & mask];
713	>	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift
714	>	ForkJoinTask<?>[] q; ForkJoinTask<?> t; int b, a;
715	>	if (v != null && (b = v.base) != v.sp &&
716	>	(q = v.queue) != null) {
717	>	int i = (q.length - 1) & b;
718	>	long u = (i << qShift) + qBase; // raw offset
719	>	int pid = poolIndex;
720	>	if ((t = q[i]) != null) {
721	>	if (!canSteal &&  // inline p.tryIncrementActiveCount
722	>	UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
723	>	a = p.runState, a + 1))
724	>	canSteal = active = true;
725	>	if (canSteal && v.base == b++ &&
726	>	UNSAFE.compareAndSwapObject(q, u, t, null)) {
727	>	v.base = b;
728	>	v.stealHint = pid;
729	>	UNSAFE.putOrderedObject(this,
730	>	currentStealOffset, t);
731	>	seed = r;
732	>	++stealCount;
733	>	return t;
734	>	}
735		}
736	+	j = -n;
737	+	k = r;                // restart on contention
738		}
739	<	j = -n;               // reset on contention
739	>	else if (++j <= 0)
740	>	k = r;
741	>	else if (j <= n)
742	>	k += (n >>> 1) | 1;
743	>	else
744	>	break;
745		}
640	–	k = j >= 0? k + ((n >>> 1) | 1) : r;
746		}
747		return null;
748		}
#	Line 645 | Line 750 | public class ForkJoinWorkerThread extend
750		// Run State management
751
752		// status check methods used mainly by ForkJoinPool
753	<	final boolean isTerminating() { return (runState & TERMINATING) != 0; }
753	>	final boolean isRunning()     { return runState == 0; }
754		final boolean isTerminated()  { return (runState & TERMINATED) != 0; }
755		final boolean isSuspended()   { return (runState & SUSPENDED) != 0; }
756		final boolean isTrimmed()     { return (runState & TRIMMED) != 0; }
757
758	+	final boolean isTerminating() {
759	+	if ((runState & TERMINATING) != 0)
760	+	return true;
761	+	if (pool.isAtLeastTerminating()) { // propagate pool state
762	+	shutdown();
763	+	return true;
764	+	}
765	+	return false;
766	+	}
767	+
768		/**
769	<	* Sets state to TERMINATING, also resuming if suspended.
769	>	* Sets state to TERMINATING. Does NOT unpark or interrupt
770	>	* to wake up if currently blocked. Callers must do so if desired.
771		*/
772		final void shutdown() {
773		for (;;) {
774		int s = runState;
775	+	if ((s & (TERMINATING|TERMINATED)) != 0)
776	+	break;
777		if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended
778		if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
779		(s & ~SUSPENDED) |
780	<	(TRIMMED|TERMINATING))) {
663	<	LockSupport.unpark(this);
780	>	(TRIMMED|TERMINATING)))
781		break;
665	–	}
782		}
783		else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
784		s | TERMINATING))
#	Line 671 | Line 787 | public class ForkJoinWorkerThread extend
787		}
788
789		/**
790	<	* Sets state to TERMINATED. Called only by this thread.
790	>	* Sets state to TERMINATED. Called only by onTermination().
791		*/
792		private void setTerminated() {
793		int s;
#	Line 681 | Line 797 | public class ForkJoinWorkerThread extend
797		}
798
799		/**
684	–	* Instrumented version of park. Also used by ForkJoinPool.awaitEvent
685	–	*/
686	–	final void doPark() {
687	–	++parkCount;
688	–	LockSupport.park(this);
689	–	}
690	–
691	–	/**
800		* If suspended, tries to set status to unsuspended.
801	<	* Caller must unpark to actually resume
801	>	* Does NOT wake up if blocked.
802		*
803		* @return true if successful
804		*/
805		final boolean tryUnsuspend() {
806		int s;
807	<	return (((s = runState) & SUSPENDED) != 0 &&
808	<	UNSAFE.compareAndSwapInt(this, runStateOffset, s,
809	<	s & ~SUSPENDED));
807	>	while (((s = runState) & SUSPENDED) != 0) {
808	>	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
809	>	s & ~SUSPENDED))
810	>	return true;
811	>	}
812	>	return false;
813		}
814
815		/**
816	<	* Sets suspended status and blocks as spare until resumed,
817	<	* shutdown, or timed out.
707	<	*
708	<	* @return false if trimmed
816	>	* Sets suspended status and blocks as spare until resumed
817	>	* or shutdown.
818		*/
819	<	final boolean suspendAsSpare() {
820	<	for (;;) {               // set suspended unless terminating
819	>	final void suspendAsSpare() {
820	>	for (;;) {                  // set suspended unless terminating
821		int s = runState;
822		if ((s & TERMINATING) != 0) { // must kill
823		if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
824		s | (TRIMMED | TERMINATING)))
825	<	return false;
825	>	return;
826		}
827		else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
828		s | SUSPENDED))
829		break;
830		}
722	–	lastEventCount = 0;      // reset upon resume
831		ForkJoinPool p = pool;
832	<	p.releaseWaiters();      // help others progress
725	<	p.accumulateStealCount(this);
726	<	interrupted();           // clear/ignore interrupts
727	<	if (poolIndex < p.getParallelism()) { // untimed wait
728	<	while ((runState & SUSPENDED) != 0)
729	<	doPark();
730	<	return true;
731	<	}
732	<	return timedSuspend();   // timed wait if apparently non-core
733	<	}
734	<
735	<	/**
736	<	* Blocks as spare until resumed or timed out
737	<	* @return false if trimmed
738	<	*/
739	<	private boolean timedSuspend() {
740	<	long nanos = SPARE_KEEPALIVE_NANOS;
741	<	long startTime = System.nanoTime();
832	>	p.pushSpare(this);
833		while ((runState & SUSPENDED) != 0) {
834	<	++parkCount;
835	<	if ((nanos -= (System.nanoTime() - startTime)) > 0)
836	<	LockSupport.parkNanos(this, nanos);
837	<	else { // try to trim on timeout
838	<	int s = runState;
748	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
749	<	(s & ~SUSPENDED) |
750	<	(TRIMMED|TERMINATING)))
751	<	return false;
834	>	if (p.tryAccumulateStealCount(this)) {
835	>	interrupted();          // clear/ignore interrupts
836	>	if ((runState & SUSPENDED) == 0)
837	>	break;
838	>	LockSupport.park(this);
839		}
840		}
754	–	return true;
841		}
842
843		// Misc support methods for ForkJoinPool
#	Line 761 | Line 847 | public class ForkJoinWorkerThread extend
847		* used by ForkJoinTask.
848		*/
849		final int getQueueSize() {
850	<	return -base + sp;
851	<	}
766	<
767	<	/**
768	<	* Set locallyFifo mode. Called only by ForkJoinPool
769	<	*/
770	<	final void setAsyncMode(boolean async) {
771	<	locallyFifo = async;
850	>	int n; // external calls must read base first
851	>	return (n = -base + sp) <= 0 ? 0 : n;
852		}
853
854		/**
#	Line 776 | Line 856 | public class ForkJoinWorkerThread extend
856		* thread.
857		*/
858		final void cancelTasks() {
859	+	ForkJoinTask<?> cj = currentJoin; // try to cancel ongoing tasks
860	+	if (cj != null) {
861	+	currentJoin = null;
862	+	cj.cancelIgnoringExceptions();
863	+	try {
864	+	this.interrupt(); // awaken wait
865	+	} catch (SecurityException ignore) {
866	+	}
867	+	}
868	+	ForkJoinTask<?> cs = currentSteal;
869	+	if (cs != null) {
870	+	currentSteal = null;
871	+	cs.cancelIgnoringExceptions();
872	+	}
873		while (base != sp) {
874		ForkJoinTask<?> t = deqTask();
875		if (t != null)
#	Line 803 | Line 897 | public class ForkJoinWorkerThread extend
897		// Support methods for ForkJoinTask
898
899		/**
900	+	* Gets and removes a local task.
901	+	*
902	+	* @return a task, if available
903	+	*/
904	+	final ForkJoinTask<?> pollLocalTask() {
905	+	ForkJoinPool p = pool;
906	+	while (sp != base) {
907	+	int a; // inline p.tryIncrementActiveCount
908	+	if (active ||
909	+	(active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
910	+	a = p.runState, a + 1)))
911	+	return locallyFifo ? locallyDeqTask() : popTask();
912	+	}
913	+	return null;
914	+	}
915	+
916	+	/**
917	+	* Gets and removes a local or stolen task.
918	+	*
919	+	* @return a task, if available
920	+	*/
921	+	final ForkJoinTask<?> pollTask() {
922	+	ForkJoinTask<?> t = pollLocalTask();
923	+	if (t == null) {
924	+	t = scan();
925	+	// cannot retain/track/help steal
926	+	UNSAFE.putOrderedObject(this, currentStealOffset, null);
927	+	}
928	+	return t;
929	+	}
930	+
931	+	/**
932	+	* Possibly runs some tasks and/or blocks, until task is done.
933	+	*
934	+	* @param joinMe the task to join
935	+	*/
936	+	final void joinTask(ForkJoinTask<?> joinMe) {
937	+	// currentJoin only written by this thread; only need ordered store
938	+	ForkJoinTask<?> prevJoin = currentJoin;
939	+	UNSAFE.putOrderedObject(this, currentJoinOffset, joinMe);
940	+	if (sp != base)
941	+	localHelpJoinTask(joinMe);
942	+	if (joinMe.status >= 0)
943	+	pool.awaitJoin(joinMe, this);
944	+	UNSAFE.putOrderedObject(this, currentJoinOffset, prevJoin);
945	+	}
946	+
947	+	/**
948	+	* Run tasks in local queue until given task is done.
949	+	*
950	+	* @param joinMe the task to join
951	+	*/
952	+	private void localHelpJoinTask(ForkJoinTask<?> joinMe) {
953	+	int s;
954	+	ForkJoinTask<?>[] q;
955	+	while (joinMe.status >= 0 && (s = sp) != base && (q = queue) != null) {
956	+	int i = (q.length - 1) & --s;
957	+	long u = (i << qShift) + qBase; // raw offset
958	+	ForkJoinTask<?> t = q[i];
959	+	if (t == null)  // lost to a stealer
960	+	break;
961	+	if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
962	+	/*
963	+	* This recheck (and similarly in helpJoinTask)
964	+	* handles cases where joinMe is independently
965	+	* cancelled or forced even though there is other work
966	+	* available. Back out of the pop by putting t back
967	+	* into slot before we commit by writing sp.
968	+	*/
969	+	if (joinMe.status < 0) {
970	+	UNSAFE.putObjectVolatile(q, u, t);
971	+	break;
972	+	}
973	+	sp = s;
974	+	// UNSAFE.putOrderedInt(this, spOffset, s);
975	+	t.quietlyExec();
976	+	}
977	+	}
978	+	}
979	+
980	+	/**
981	+	* Unless terminating, tries to locate and help perform tasks for
982	+	* a stealer of the given task, or in turn one of its stealers.
983	+	* Traces currentSteal->currentJoin links looking for a thread
984	+	* working on a descendant of the given task and with a non-empty
985	+	* queue to steal back and execute tasks from.
986	+	*
987	+	* The implementation is very branchy to cope with potential
988	+	* inconsistencies or loops encountering chains that are stale,
989	+	* unknown, or of length greater than MAX_HELP_DEPTH links.  All
990	+	* of these cases are dealt with by just returning back to the
991	+	* caller, who is expected to retry if other join mechanisms also
992	+	* don't work out.
993	+	*
994	+	* @param joinMe the task to join
995	+	*/
996	+	final void helpJoinTask(ForkJoinTask<?> joinMe) {
997	+	ForkJoinWorkerThread[] ws;
998	+	int n;
999	+	if (joinMe.status < 0)                // already done
1000	+	return;
1001	+	if ((runState & TERMINATING) != 0) {  // cancel if shutting down
1002	+	joinMe.cancelIgnoringExceptions();
1003	+	return;
1004	+	}
1005	+	if ((ws = pool.workers) == null || (n = ws.length) <= 1)
1006	+	return;                           // need at least 2 workers
1007	+
1008	+	ForkJoinTask<?> task = joinMe;        // base of chain
1009	+	ForkJoinWorkerThread thread = this;   // thread with stolen task
1010	+	for (int d = 0; d < MAX_HELP_DEPTH; ++d) { // chain length
1011	+	// Try to find v, the stealer of task, by first using hint
1012	+	ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)];
1013	+	if (v == null || v.currentSteal != task) {
1014	+	for (int j = 0; ; ++j) {      // search array
1015	+	if (j < n) {
1016	+	ForkJoinTask<?> vs;
1017	+	if ((v = ws[j]) != null &&
1018	+	(vs = v.currentSteal) != null) {
1019	+	if (joinMe.status < 0 || task.status < 0)
1020	+	return;       // stale or done
1021	+	if (vs == task) {
1022	+	thread.stealHint = j;
1023	+	break;        // save hint for next time
1024	+	}
1025	+	}
1026	+	}
1027	+	else
1028	+	return;               // no stealer
1029	+	}
1030	+	}
1031	+	for (;;) { // Try to help v, using specialized form of deqTask
1032	+	if (joinMe.status < 0)
1033	+	return;
1034	+	int b = v.base;
1035	+	ForkJoinTask<?>[] q = v.queue;
1036	+	if (b == v.sp || q == null)
1037	+	break;
1038	+	int i = (q.length - 1) & b;
1039	+	long u = (i << qShift) + qBase;
1040	+	ForkJoinTask<?> t = q[i];
1041	+	int pid = poolIndex;
1042	+	ForkJoinTask<?> ps = currentSteal;
1043	+	if (task.status < 0)
1044	+	return;                   // stale or done
1045	+	if (t != null && v.base == b++ &&
1046	+	UNSAFE.compareAndSwapObject(q, u, t, null)) {
1047	+	if (joinMe.status < 0) {
1048	+	UNSAFE.putObjectVolatile(q, u, t);
1049	+	return;               // back out on cancel
1050	+	}
1051	+	v.base = b;
1052	+	v.stealHint = pid;
1053	+	UNSAFE.putOrderedObject(this, currentStealOffset, t);
1054	+	t.quietlyExec();
1055	+	UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1056	+	}
1057	+	}
1058	+	// Try to descend to find v's stealer
1059	+	ForkJoinTask<?> next = v.currentJoin;
1060	+	if (task.status < 0 || next == null || next == task ||
1061	+	joinMe.status < 0)
1062	+	return;
1063	+	task = next;
1064	+	thread = v;
1065	+	}
1066	+	}
1067	+
1068	+	/**
1069	+	* Implements ForkJoinTask.getSurplusQueuedTaskCount().
1070		* Returns an estimate of the number of tasks, offset by a
1071		* function of number of idle workers.
1072		*
#	Line 854 | Line 1118 | public class ForkJoinWorkerThread extend
1118		}
1119
1120		/**
857	–	* Gets and removes a local task.
858	–	*
859	–	* @return a task, if available
860	–	*/
861	–	final ForkJoinTask<?> pollLocalTask() {
862	–	while (base != sp) {
863	–	if (active || (active = pool.tryIncrementActiveCount()))
864	–	return locallyFifo? locallyDeqTask() : popTask();
865	–	}
866	–	return null;
867	–	}
868	–
869	–	/**
870	–	* Gets and removes a local or stolen task.
871	–	*
872	–	* @return a task, if available
873	–	*/
874	–	final ForkJoinTask<?> pollTask() {
875	–	ForkJoinTask<?> t;
876	–	return (t = pollLocalTask()) != null ? t : scan();
877	–	}
878	–
879	–	/**
880	–	* Returns a stolen task, if available, unless joinMe is done
881	–	*
882	–	* This method is intrinsically nonmodular. To maintain the
883	–	* property that tasks are never stolen if the awaited task is
884	–	* ready, we must interleave mechanics of scan with status
885	–	* checks. We rely here on the commit points of deq that allow us
886	–	* to cancel a steal even after CASing slot to null, but before
887	–	* adjusting base index: If, after the CAS, we see that joinMe is
888	–	* ready, we can back out by placing the task back into the slot,
889	–	* without adjusting index. The scan loop is otherwise the same as
890	–	* in scan.
891	–	*
892	–	* The outer loop cannot be allowed to run forever, because it
893	–	* could lead to a form of deadlock if all threads are executing
894	–	* this method. However, we must also be patient before giving up,
895	–	* to cope with GC stalls, transient high loads, etc. The loop
896	–	* terminates (causing caller to possibly block this thread and
897	–	* create a replacement) only after #workers clean sweeps during
898	–	* which all running threads are active.
899	–	*/
900	–	final ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
901	–	int sweeps = 0;
902	–	int r = seed;
903	–	ForkJoinPool p = pool;
904	–	p.releaseWaiters(); // help other threads progress
905	–	while (joinMe.status >= 0) {
906	–	ForkJoinWorkerThread[] ws = p.workers;
907	–	int n = ws.length;
908	–	int k = r;
909	–	for (int j = -n; j < n; ++j) {
910	–	ForkJoinWorkerThread v = ws[k & (n - 1)];
911	–	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
912	–	if (v != null) {
913	–	int b = v.base;
914	–	ForkJoinTask<?>[] q;
915	–	if (b != v.sp && (q = v.queue) != null) {
916	–	int i = (q.length - 1) & b;
917	–	ForkJoinTask<?> t = q[i];
918	–	if (t != null) {
919	–	if (joinMe.status < 0)
920	–	return null;
921	–	if (UNSAFE.compareAndSwapObject
922	–	(q, (i << qShift) + qBase, t, null)) {
923	–	if (joinMe.status < 0) {
924	–	writeSlot(q, i, t); // back out
925	–	return null;
926	–	}
927	–	v.base = b + 1;
928	–	seed = r;
929	–	++stealCount;
930	–	return t;
931	–	}
932	–	}
933	–	sweeps = 0; // ensure rescan on contention
934	–	}
935	–	}
936	–	k = j >= 0? k + ((n >>> 1) | 1) : r;
937	–	if ((j & 7) == 0 && joinMe.status < 0) // periodically recheck
938	–	return null;
939	–	}
940	–	if ((sweeps = p.inactiveCount() == 0 ? sweeps + 1 : 0) > n)
941	–	return null;
942	–	}
943	–	return null;
944	–	}
945	–
946	–	/**
1121		* Runs tasks until {@code pool.isQuiescent()}.
1122		*/
1123		final void helpQuiescePool() {
1124	+	ForkJoinTask<?> ps = currentSteal; // to restore below
1125		for (;;) {
1126		ForkJoinTask<?> t = pollLocalTask();
1127		if (t != null || (t = scan()) != null)
1128	<	t.tryExec();
1128	>	t.quietlyExec();
1129		else {
1130		ForkJoinPool p = pool;
1131	+	int a; // to inline CASes
1132		if (active) {
1133	+	if (!UNSAFE.compareAndSwapInt
1134	+	(p, poolRunStateOffset, a = p.runState, a - 1))
1135	+	continue;   // retry later
1136		active = false; // inactivate
1137	<	do {} while (!p.tryDecrementActiveCount());
1137	>	UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1138		}
1139		if (p.isQuiescent()) {
1140		active = true; // re-activate
1141	<	do {} while (!p.tryIncrementActiveCount());
1141	>	do {} while (!UNSAFE.compareAndSwapInt
1142	>	(p, poolRunStateOffset, a = p.runState, a+1));
1143		return;
1144		}
1145		}
#	Line 969 | Line 1149 | public class ForkJoinWorkerThread extend
1149		// Unsafe mechanics
1150
1151		private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1152	+	private static final long spOffset =
1153	+	objectFieldOffset("sp", ForkJoinWorkerThread.class);
1154		private static final long runStateOffset =
1155		objectFieldOffset("runState", ForkJoinWorkerThread.class);
1156	+	private static final long currentJoinOffset =
1157	+	objectFieldOffset("currentJoin", ForkJoinWorkerThread.class);
1158	+	private static final long currentStealOffset =
1159	+	objectFieldOffset("currentSteal", ForkJoinWorkerThread.class);
1160		private static final long qBase =
1161		UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1162	+	private static final long poolRunStateOffset = // to inline CAS
1163	+	objectFieldOffset("runState", ForkJoinPool.class);
1164	+
1165		private static final int qShift;
1166
1167		static {
#	Line 980 | Line 1169 | public class ForkJoinWorkerThread extend
1169		if ((s & (s-1)) != 0)
1170		throw new Error("data type scale not a power of two");
1171		qShift = 31 - Integer.numberOfLeadingZeros(s);
1172	+	MAXIMUM_QUEUE_CAPACITY = 1 << (31 - qShift);
1173		}
1174
1175		private static long objectFieldOffset(String field, Class<?> klazz) {

Diff Legend

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