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 1.40 by dl, Wed Aug 11 18:45:12 2010 UTC vs.
Revision 1.68 by dl, Thu Jan 26 00:08:13 2012 UTC

#	Line 1 | Line 1
1		/*
2		* Written by Doug Lea with assistance from members of JCP JSR-166
3		* Expert Group and released to the public domain, as explained at
4	<	* http://creativecommons.org/licenses/publicdomain
4	>	* http://creativecommons.org/publicdomain/zero/1.0/
5		*/
6
7		package jsr166y;
8
9	–	import java.util.concurrent.*;
10	–
11	–	import java.util.Random;
12	–	import java.util.Collection;
13	–	import java.util.concurrent.locks.LockSupport;
14	–
9		/**
10	<	* A thread managed by a {@link ForkJoinPool}.  This class is
11	<	* subclassable solely for the sake of adding functionality -- there
12	<	* are no overridable methods dealing with scheduling or execution.
13	<	* However, you can override initialization and termination methods
14	<	* surrounding the main task processing loop.  If you do create such a
15	<	* subclass, you will also need to supply a custom {@link
16	<	* ForkJoinPool.ForkJoinWorkerThreadFactory} to use it in a {@code
17	<	* ForkJoinPool}.
10	>	* A thread managed by a {@link ForkJoinPool}, which executes
11	>	* {@link ForkJoinTask}s.
12	>	* This class is subclassable solely for the sake of adding
13	>	* functionality -- there are no overridable methods dealing with
14	>	* scheduling or execution.  However, you can override initialization
15	>	* and termination methods surrounding the main task processing loop.
16	>	* If you do create such a subclass, you will also need to supply a
17	>	* custom {@link ForkJoinPool.ForkJoinWorkerThreadFactory} to use it
18	>	* in a {@code ForkJoinPool}.
19		*
20		* @since 1.7
21		* @author Doug Lea
22		*/
23		public class ForkJoinWorkerThread extends Thread {
24		/*
30	–	* Overview:
31	–	*
25		* ForkJoinWorkerThreads are managed by ForkJoinPools and perform
26	<	* ForkJoinTasks. This class includes bookkeeping in support of
27	<	* worker activation, suspension, and lifecycle control described
35	<	* in more detail in the internal documentation of class
36	<	* ForkJoinPool. And as described further below, this class also
37	<	* includes special-cased support for some ForkJoinTask
38	<	* methods. But the main mechanics involve work-stealing:
39	<	*
40	<	* Work-stealing queues are special forms of Deques that support
41	<	* only three of the four possible end-operations -- push, pop,
42	<	* and deq (aka steal), under the further constraints that push
43	<	* and pop are called only from the owning thread, while deq may
44	<	* be called from other threads.  (If you are unfamiliar with
45	<	* them, you probably want to read Herlihy and Shavit's book "The
46	<	* Art of Multiprocessor programming", chapter 16 describing these
47	<	* in more detail before proceeding.)  The main work-stealing
48	<	* queue design is roughly similar to those in the papers "Dynamic
49	<	* Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
50	<	* (http://research.sun.com/scalable/pubs/index.html) and
51	<	* "Idempotent work stealing" by Michael, Saraswat, and Vechev,
52	<	* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
53	<	* The main differences ultimately stem from gc requirements that
54	<	* we null out taken slots as soon as we can, to maintain as small
55	<	* a footprint as possible even in programs generating huge
56	<	* numbers of tasks. To accomplish this, we shift the CAS
57	<	* arbitrating pop vs deq (steal) from being on the indices
58	<	* ("base" and "sp") to the slots themselves (mainly via method
59	<	* "casSlotNull()"). So, both a successful pop and deq mainly
60	<	* entail a CAS of a slot from non-null to null.  Because we rely
61	<	* on CASes of references, we do not need tag bits on base or sp.
62	<	* They are simple ints as used in any circular array-based queue
63	<	* (see for example ArrayDeque).  Updates to the indices must
64	<	* still be ordered in a way that guarantees that sp == base means
65	<	* the queue is empty, but otherwise may err on the side of
66	<	* possibly making the queue appear nonempty when a push, pop, or
67	<	* deq have not fully committed. Note that this means that the deq
68	<	* operation, considered individually, is not wait-free. One thief
69	<	* cannot successfully continue until another in-progress one (or,
70	<	* if previously empty, a push) completes.  However, in the
71	<	* aggregate, we ensure at least probabilistic non-blockingness.
72	<	* If an attempted steal fails, a thief always chooses a different
73	<	* random victim target to try next. So, in order for one thief to
74	<	* progress, it suffices for any in-progress deq or new push on
75	<	* any empty queue to complete. One reason this works well here is
76	<	* that apparently-nonempty often means soon-to-be-stealable,
77	<	* which gives threads a chance to set activation status if
78	<	* necessary before stealing.
79	<	*
80	<	* This approach also enables support for "async mode" where local
81	<	* task processing is in FIFO, not LIFO order; simply by using a
82	<	* version of deq rather than pop when locallyFifo is true (as set
83	<	* by the ForkJoinPool).  This allows use in message-passing
84	<	* frameworks in which tasks are never joined.
85	<	*
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.tryAwaitJoin).
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	<	* 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
137	<	* though these queues don't usually become all that big, the
138	<	* initial size must be large enough to counteract cache
139	<	* contention effects across multiple queues (especially in the
140	<	* presence of GC cardmarking). Also, to improve thread-locality,
141	<	* queues are initialized after starting.  All together, these
142	<	* low-level implementation choices produce as much as a factor of
143	<	* 4 performance improvement compared to naive implementations,
144	<	* and enable the processing of billions of tasks per second,
145	<	* sometimes at the expense of ugliness.
146	<	*/
147	<
148	<	/**
149	<	* Generator for initial random seeds for random victim
150	<	* selection. This is used only to create initial seeds. Random
151	<	* steals use a cheaper xorshift generator per steal attempt. We
152	<	* expect only rare contention on seedGenerator, so just use a
153	<	* plain Random.
154	<	*/
155	<	private static final Random seedGenerator = new Random();
156	<
157	<	/**
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 int MAX_HELP_DEPTH = 8;
165	<
166	<	/**
167	<	* The wakeup interval (in nanoseconds) for the first worker
168	<	* suspended as spare.  On each wakeup not signalled by a
169	<	* resumption, it may ask the pool to reduce the number of spares.
170	<	*/
171	<	private static final long TRIM_RATE_NANOS = 200L * 1000L * 1000L;
172	<
173	<	/**
174	<	* Capacity of work-stealing queue array upon initialization.
175	<	* Must be a power of two. Initial size must be at least 4, but is
176	<	* padded to minimize cache effects.
177	<	*/
178	<	private static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
179	<
180	<	/**
181	<	* Maximum work-stealing queue array size.  Must be less than or
182	<	* equal to 1 << 28 to ensure lack of index wraparound. (This
183	<	* is less than usual bounds, because we need leftshift by 3
184	<	* to be in int range).
185	<	*/
186	<	private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
187	<
188	<	/**
189	<	* The pool this thread works in. Accessed directly by ForkJoinTask.
190	<	*/
191	<	final ForkJoinPool pool;
192	<
193	<	/**
194	<	* The work-stealing queue array. Size must be a power of two.
195	<	* Initialized in onStart, to improve memory locality.
196	<	*/
197	<	private ForkJoinTask<?>[] queue;
198	<
199	<	/**
200	<	* Index (mod queue.length) of least valid queue slot, which is
201	<	* always the next position to steal from if nonempty.
202	<	*/
203	<	private volatile int base;
204	<
205	<	/**
206	<	* Index (mod queue.length) of next queue slot to push to or pop
207	<	* from. It is written only by owner thread, and accessed by other
208	<	* threads only after reading (volatile) base.  Both sp and base
209	<	* are allowed to wrap around on overflow, but (sp - base) still
210	<	* estimates size.
211	<	*/
212	<	private int sp;
213	<
214	<	/**
215	<	* The index of most recent stealer, used as a hint to avoid
216	<	* traversal in method helpJoinTask. This is only a hint because a
217	<	* worker might have had multiple steals and this only holds one
218	<	* of them (usually the most current). Declared non-volatile,
219	<	* relying on other prevailing sync to keep reasonably current.
220	<	*/
221	<	private int stealHint;
222	<
223	<	/**
224	<	* Run state of this worker. In addition to the usual run levels,
225	<	* tracks if this worker is suspended as a spare, and if it was
226	<	* killed (trimmed) while suspended. However, "active" status is
227	<	* maintained separately.
228	<	*/
229	<	private volatile int runState;
230	<
231	<	private static final int TERMINATING = 0x01;
232	<	private static final int TERMINATED  = 0x02;
233	<	private static final int SUSPENDED   = 0x04; // inactive spare
234	<	private static final int TRIMMED     = 0x08; // killed while suspended
235	<
236	<	/**
237	<	* Number of steals, transferred and reset in pool callbacks pool
238	<	* when idle Accessed directly by pool.
239	<	*/
240	<	int stealCount;
241	<
242	<	/**
243	<	* Seed for random number generator for choosing steal victims.
244	<	* Uses Marsaglia xorshift. Must be initialized as nonzero.
245	<	*/
246	<	private int seed;
247	<
248	<	/**
249	<	* Activity status. When true, this worker is considered active.
250	<	* Accessed directly by pool.  Must be false upon construction.
251	<	*/
252	<	boolean active;
253	<
254	<	/**
255	<	* True if use local fifo, not default lifo, for local polling.
256	<	* Shadows value from ForkJoinPool.
257	<	*/
258	<	private final boolean locallyFifo;
259	<
260	<	/**
261	<	* Index of this worker in pool array. Set once by pool before
262	<	* running, and accessed directly by pool to locate this worker in
263	<	* its workers array.
264	<	*/
265	<	int poolIndex;
266	<
267	<	/**
268	<	* The last pool event waited for. Accessed only by pool in
269	<	* callback methods invoked within this thread.
270	<	*/
271	<	int lastEventCount;
272	<
273	<	/**
274	<	* Encoded index and event count of next event waiter. Used only
275	<	* by ForkJoinPool for managing event waiters.
276	<	*/
277	<	volatile long nextWaiter;
278	<
279	<	/**
280	<	* Number of times this thread suspended as spare
281	<	*/
282	<	int spareCount;
283	<
284	<	/**
285	<	* Encoded index and count of next spare waiter. Used only
286	<	* by ForkJoinPool for managing spares.
287	<	*/
288	<	volatile int nextSpare;
289	<
290	<	/**
291	<	* The task currently being joined, set only when actively trying
292	<	* to helpStealer. Written only by current thread, but read by
293	<	* others.
26	>	* ForkJoinTasks. For explanation, see the internal documentation
27	>	* of class ForkJoinPool.
28		*/
295	–	private volatile ForkJoinTask<?> currentJoin;
29
30	<	/**
31	<	* The task most recently stolen from another worker (or
299	<	* submission queue).  Not volatile because always read/written in
300	<	* presence of related volatiles in those cases where it matters.
301	<	*/
302	<	private ForkJoinTask<?> currentSteal;
30	>	final ForkJoinPool.WorkQueue workQueue; // Work-stealing mechanics
31	>	final ForkJoinPool pool;                // the pool this thread works in
32
33		/**
34		* Creates a ForkJoinWorkerThread operating in the given pool.
#	Line 308 | Line 37 | public class ForkJoinWorkerThread extend
37		* @throws NullPointerException if pool is null
38		*/
39		protected ForkJoinWorkerThread(ForkJoinPool pool) {
40	<	this.pool = pool;
312	<	this.locallyFifo = pool.locallyFifo;
40	>	super(pool.nextWorkerName());
41		setDaemon(true);
42	<	// To avoid exposing construction details to subclasses,
315	<	// remaining initialization is in start() and onStart()
316	<	}
317	<
318	<	/**
319	<	* Performs additional initialization and starts this thread
320	<	*/
321	<	final void start(int poolIndex, UncaughtExceptionHandler ueh) {
322	<	this.poolIndex = poolIndex;
42	>	Thread.UncaughtExceptionHandler ueh = pool.ueh;
43		if (ueh != null)
44		setUncaughtExceptionHandler(ueh);
45	<	start();
45	>	this.pool = pool;
46	>	this.workQueue = new ForkJoinPool.WorkQueue(this, pool.localMode);
47	>	pool.registerWorker(this);
48		}
49
328	–	// Public/protected methods
329	–
50		/**
51		* Returns the pool hosting this thread.
52		*
#	Line 346 | Line 66 | public class ForkJoinWorkerThread extend
66		* @return the index number
67		*/
68		public int getPoolIndex() {
69	<	return poolIndex;
69	>	return workQueue.poolIndex;
70		}
71
72		/**
73		* Initializes internal state after construction but before
74		* processing any tasks. If you override this method, you must
75	<	* invoke super.onStart() at the beginning of the method.
75	>	* invoke {@code super.onStart()} at the beginning of the method.
76		* Initialization requires care: Most fields must have legal
77		* default values, to ensure that attempted accesses from other
78		* threads work correctly even before this thread starts
79		* processing tasks.
80		*/
81		protected void onStart() {
362	–	int rs = seedGenerator.nextInt();
363	–	seed = rs == 0? 1 : rs; // seed must be nonzero
364	–
365	–	// Allocate name string and arrays in this thread
366	–	String pid = Integer.toString(pool.getPoolNumber());
367	–	String wid = Integer.toString(poolIndex);
368	–	setName("ForkJoinPool-" + pid + "-worker-" + wid);
369	–
370	–	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
82		}
83
84		/**
#	Line 379 | Line 90 | public class ForkJoinWorkerThread extend
90		* to an unrecoverable error, or {@code null} if completed normally
91		*/
92		protected void onTermination(Throwable exception) {
382	–	try {
383	–	cancelTasks();
384	–	while (active)              // force inactive
385	–	active = !pool.tryDecrementActiveCount();
386	–	setTerminated();
387	–	pool.workerTerminated(this);
388	–	} catch (Throwable ex) {        // Shouldn't ever happen
389	–	if (exception == null)      // but if so, at least rethrown
390	–	exception = ex;
391	–	} finally {
392	–	if (exception != null)
393	–	UNSAFE.throwException(exception);
394	–	}
93		}
94
95		/**
96		* This method is required to be public, but should never be
97		* called explicitly. It performs the main run loop to execute
98	<	* ForkJoinTasks.
98	>	* {@link ForkJoinTask}s.
99		*/
100		public void run() {
101		Throwable exception = null;
102		try {
103		onStart();
104	<	mainLoop();
104	>	pool.runWorker(this);
105		} catch (Throwable ex) {
106		exception = ex;
107		} finally {
410	–	onTermination(exception);
411	–	}
412	–	}
413	–
414	–	// helpers for run()
415	–
416	–	/**
417	–	* Find and execute tasks and check status while running
418	–	*/
419	–	private void mainLoop() {
420	–	int misses = 0; // track consecutive times failed to find work; max 2
421	–	ForkJoinPool p = pool;
422	–	for (;;) {
423	–	p.preStep(this, misses);
424	–	if (runState != 0)
425	–	break;
426	–	misses = ((tryExecSteal() || tryExecSubmission()) ? 0 :
427	–	(misses < 2 ? misses + 1 : 2));
428	–	}
429	–	}
430	–
431	–	/**
432	–	* Try to steal a task and execute it
433	–	*
434	–	* @return true if ran a task
435	–	*/
436	–	private boolean tryExecSteal() {
437	–	ForkJoinTask<?> t;
438	–	if ((t  = scan()) != null) {
439	–	t.quietlyExec();
440	–	currentSteal = null;
441	–	if (sp != base)
442	–	execLocalTasks();
443	–	return true;
444	–	}
445	–	return false;
446	–	}
447	–
448	–	/**
449	–	* If a submission exists, try to activate and run it;
450	–	*
451	–	* @return true if ran a task
452	–	*/
453	–	private boolean tryExecSubmission() {
454	–	ForkJoinPool p = pool;
455	–	while (p.hasQueuedSubmissions()) {
456	–	ForkJoinTask<?> t;
457	–	if (active || (active = p.tryIncrementActiveCount())) {
458	–	if ((t = p.pollSubmission()) != null) {
459	–	currentSteal = t;
460	–	t.quietlyExec();
461	–	currentSteal = null;
462	–	if (sp != base)
463	–	execLocalTasks();
464	–	return true;
465	–	}
466	–	}
467	–	}
468	–	return false;
469	–	}
470	–
471	–	/**
472	–	* Runs local tasks until queue is empty or shut down.  Call only
473	–	* while active.
474	–	*/
475	–	private void execLocalTasks() {
476	–	while (runState == 0) {
477	–	ForkJoinTask<?> t = locallyFifo? locallyDeqTask() : popTask();
478	–	if (t != null)
479	–	t.quietlyExec();
480	–	else if (sp == base)
481	–	break;
482	–	}
483	–	}
484	–
485	–	/*
486	–	* Intrinsics-based atomic writes for queue slots. These are
487	–	* basically the same as methods in AtomicObjectArray, but
488	–	* specialized for (1) ForkJoinTask elements (2) requirement that
489	–	* nullness and bounds checks have already been performed by
490	–	* callers and (3) effective offsets are known not to overflow
491	–	* from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't
492	–	* need corresponding version for reads: plain array reads are OK
493	–	* because they protected by other volatile reads and are
494	–	* confirmed by CASes.
495	–	*
496	–	* Most uses don't actually call these methods, but instead contain
497	–	* inlined forms that enable more predictable optimization.  We
498	–	* don't define the version of write used in pushTask at all, but
499	–	* instead inline there a store-fenced array slot write.
500	–	*/
501	–
502	–	/**
503	–	* CASes slot i of array q from t to null. Caller must ensure q is
504	–	* non-null and index is in range.
505	–	*/
506	–	private static final boolean casSlotNull(ForkJoinTask<?>[] q, int i,
507	–	ForkJoinTask<?> t) {
508	–	return UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
509	–	}
510	–
511	–	/**
512	–	* Performs a volatile write of the given task at given slot of
513	–	* array q.  Caller must ensure q is non-null and index is in
514	–	* range. This method is used only during resets and backouts.
515	–	*/
516	–	private static final void writeSlot(ForkJoinTask<?>[] q, int i,
517	–	ForkJoinTask<?> t) {
518	–	UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
519	–	}
520	–
521	–	// queue methods
522	–
523	–	/**
524	–	* Pushes a task. Call only from this thread.
525	–	*
526	–	* @param t the task. Caller must ensure non-null.
527	–	*/
528	–	final void pushTask(ForkJoinTask<?> t) {
529	–	ForkJoinTask<?>[] q = queue;
530	–	int mask = q.length - 1; // implicit assert q != null
531	–	int s = sp++;            // ok to increment sp before slot write
532	–	UNSAFE.putOrderedObject(q, ((s & mask) << qShift) + qBase, t);
533	–	if ((s -= base) == 0)
534	–	pool.signalWork();   // was empty
535	–	else if (s == mask)
536	–	growQueue();         // is full
537	–	}
538	–
539	–	/**
540	–	* Tries to take a task from the base of the queue, failing if
541	–	* empty or contended. Note: Specializations of this code appear
542	–	* in locallyDeqTask and elsewhere.
543	–	*
544	–	* @return a task, or null if none or contended
545	–	*/
546	–	final ForkJoinTask<?> deqTask() {
547	–	ForkJoinTask<?> t;
548	–	ForkJoinTask<?>[] q;
549	–	int b, i;
550	–	if (sp != (b = base) &&
551	–	(q = queue) != null && // must read q after b
552	–	(t = q[i = (q.length - 1) & b]) != null && base == b &&
553	–	UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) {
554	–	base = b + 1;
555	–	return t;
556	–	}
557	–	return null;
558	–	}
559	–
560	–	/**
561	–	* Tries to take a task from the base of own queue. Assumes active
562	–	* status.  Called only by current thread.
563	–	*
564	–	* @return a task, or null if none
565	–	*/
566	–	final ForkJoinTask<?> locallyDeqTask() {
567	–	ForkJoinTask<?>[] q = queue;
568	–	if (q != null) {
569	–	ForkJoinTask<?> t;
570	–	int b, i;
571	–	while (sp != (b = base)) {
572	–	if ((t = q[i = (q.length - 1) & b]) != null && base == b &&
573	–	UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase,
574	–	t, null)) {
575	–	base = b + 1;
576	–	return t;
577	–	}
578	–	}
579	–	}
580	–	return null;
581	–	}
582	–
583	–	/**
584	–	* Returns a popped task, or null if empty. Assumes active status.
585	–	* Called only by current thread.
586	–	*/
587	–	private ForkJoinTask<?> popTask() {
588	–	ForkJoinTask<?>[] q = queue;
589	–	if (q != null) {
590	–	int s;
591	–	while ((s = sp) != base) {
592	–	int i = (q.length - 1) & --s;
593	–	long u = (i << qShift) + qBase; // raw offset
594	–	ForkJoinTask<?> t = q[i];
595	–	if (t == null)   // lost to stealer
596	–	break;
597	–	if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
598	–	sp = s; // putOrderedInt may encourage more timely write
599	–	// UNSAFE.putOrderedInt(this, spOffset, s);
600	–	return t;
601	–	}
602	–	}
603	–	}
604	–	return null;
605	–	}
606	–
607	–	/**
608	–	* Specialized version of popTask to pop only if topmost element
609	–	* is the given task. Called only by current thread while
610	–	* active.
611	–	*
612	–	* @param t the task. Caller must ensure non-null.
613	–	*/
614	–	final boolean unpushTask(ForkJoinTask<?> t) {
615	–	int s;
616	–	ForkJoinTask<?>[] q = queue;
617	–	if ((s = sp) != base && q != null &&
618	–	UNSAFE.compareAndSwapObject
619	–	(q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) {
620	–	sp = s;
621	–	// UNSAFE.putOrderedInt(this, spOffset, s);
622	–	return true;
623	–	}
624	–	return false;
625	–	}
626	–
627	–	/**
628	–	* Returns next task or null if empty or contended
629	–	*/
630	–	final ForkJoinTask<?> peekTask() {
631	–	ForkJoinTask<?>[] q = queue;
632	–	if (q == null)
633	–	return null;
634	–	int mask = q.length - 1;
635	–	int i = locallyFifo ? base : (sp - 1);
636	–	return q[i & mask];
637	–	}
638	–
639	–	/**
640	–	* Doubles queue array size. Transfers elements by emulating
641	–	* steals (deqs) from old array and placing, oldest first, into
642	–	* new array.
643	–	*/
644	–	private void growQueue() {
645	–	ForkJoinTask<?>[] oldQ = queue;
646	–	int oldSize = oldQ.length;
647	–	int newSize = oldSize << 1;
648	–	if (newSize > MAXIMUM_QUEUE_CAPACITY)
649	–	throw new RejectedExecutionException("Queue capacity exceeded");
650	–	ForkJoinTask<?>[] newQ = queue = new ForkJoinTask<?>[newSize];
651	–
652	–	int b = base;
653	–	int bf = b + oldSize;
654	–	int oldMask = oldSize - 1;
655	–	int newMask = newSize - 1;
656	–	do {
657	–	int oldIndex = b & oldMask;
658	–	ForkJoinTask<?> t = oldQ[oldIndex];
659	–	if (t != null && !casSlotNull(oldQ, oldIndex, t))
660	–	t = null;
661	–	writeSlot(newQ, b & newMask, t);
662	–	} while (++b != bf);
663	–	pool.signalWork();
664	–	}
665	–
666	–	/**
667	–	* Computes next value for random victim probe in scan().  Scans
668	–	* don't require a very high quality generator, but also not a
669	–	* crummy one.  Marsaglia xor-shift is cheap and works well enough.
670	–	* Note: This is manually inlined in scan()
671	–	*/
672	–	private static final int xorShift(int r) {
673	–	r ^= r << 13;
674	–	r ^= r >>> 17;
675	–	return r ^ (r << 5);
676	–	}
677	–
678	–	/**
679	–	* Tries to steal a task from another worker. Starts at a random
680	–	* index of workers array, and probes workers until finding one
681	–	* with non-empty queue or finding that all are empty.  It
682	–	* randomly selects the first n probes. If these are empty, it
683	–	* resorts to a circular sweep, which is necessary to accurately
684	–	* set active status. (The circular sweep uses steps of
685	–	* approximately half the array size plus 1, to avoid bias
686	–	* stemming from leftmost packing of the array in ForkJoinPool.)
687	–	*
688	–	* This method must be both fast and quiet -- usually avoiding
689	–	* memory accesses that could disrupt cache sharing etc other than
690	–	* those needed to check for and take tasks (or to activate if not
691	–	* already active). This accounts for, among other things,
692	–	* updating random seed in place without storing it until exit.
693	–	*
694	–	* @return a task, or null if none found
695	–	*/
696	–	private ForkJoinTask<?> scan() {
697	–	ForkJoinPool p = pool;
698	–	ForkJoinWorkerThread[] ws;        // worker array
699	–	int n;                            // upper bound of #workers
700	–	if ((ws = p.workers) != null && (n = ws.length) > 1) {
701	–	boolean canSteal = active;    // shadow active status
702	–	int r = seed;                 // extract seed once
703	–	int mask = n - 1;
704	–	int j = -n;                   // loop counter
705	–	int k = r;                    // worker index, random if j < 0
706	–	for (;;) {
707	–	ForkJoinWorkerThread v = ws[k & mask];
708	–	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift
709	–	if (v != null && v.base != v.sp) {
710	–	ForkJoinTask<?>[] q; int b;
711	–	if ((canSteal ||       // ensure active status
712	–	(canSteal = active = p.tryIncrementActiveCount())) &&
713	–	(q = v.queue) != null && (b = v.base) != v.sp) {
714	–	int i = (q.length - 1) & b;
715	–	long u = (i << qShift) + qBase; // raw offset
716	–	ForkJoinTask<?> t = q[i];
717	–	if (v.base == b && t != null &&
718	–	UNSAFE.compareAndSwapObject(q, u, t, null)) {
719	–	int pid = poolIndex;
720	–	currentSteal = t;
721	–	v.stealHint = pid;
722	–	v.base = b + 1;
723	–	seed = r;
724	–	++stealCount;
725	–	return t;
726	–	}
727	–	}
728	–	j = -n;
729	–	k = r;                // restart on contention
730	–	}
731	–	else if (++j <= 0)
732	–	k = r;
733	–	else if (j <= n)
734	–	k += (n >>> 1) | 1;
735	–	else
736	–	break;
737	–	}
738	–	}
739	–	return null;
740	–	}
741	–
742	–	// Run State management
743	–
744	–	// status check methods used mainly by ForkJoinPool
745	–	final boolean isRunning()     { return runState == 0; }
746	–	final boolean isTerminating() { return (runState & TERMINATING) != 0; }
747	–	final boolean isTerminated()  { return (runState & TERMINATED) != 0; }
748	–	final boolean isSuspended()   { return (runState & SUSPENDED) != 0; }
749	–	final boolean isTrimmed()     { return (runState & TRIMMED) != 0; }
750	–
751	–	/**
752	–	* Sets state to TERMINATING, also, unless "quiet", unparking if
753	–	* not already terminated
754	–	*
755	–	* @param quiet don't unpark (used for faster status updates on
756	–	* pool termination)
757	–	*/
758	–	final void shutdown(boolean quiet) {
759	–	for (;;) {
760	–	int s = runState;
761	–	if ((s & (TERMINATING|TERMINATED)) != 0)
762	–	break;
763	–	if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended
764	–	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
765	–	(s & ~SUSPENDED) |
766	–	(TRIMMED|TERMINATING)))
767	–	break;
768	–	}
769	–	else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
770	–	s | TERMINATING))
771	–	break;
772	–	}
773	–	if (!quiet && (runState & TERMINATED) != 0)
774	–	LockSupport.unpark(this);
775	–	}
776	–
777	–	/**
778	–	* Sets state to TERMINATED. Called only by onTermination()
779	–	*/
780	–	private void setTerminated() {
781	–	int s;
782	–	do {} while (!UNSAFE.compareAndSwapInt(this, runStateOffset,
783	–	s = runState,
784	–	s | (TERMINATING|TERMINATED)));
785	–	}
786	–
787	–	/**
788	–	* If suspended, tries to set status to unsuspended and unparks.
789	–	*
790	–	* @return true if successful
791	–	*/
792	–	final boolean tryUnsuspend() {
793	–	int s;
794	–	while (((s = runState) & SUSPENDED) != 0) {
795	–	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
796	–	s & ~SUSPENDED))
797	–	return true;
798	–	}
799	–	return false;
800	–	}
801	–
802	–	/**
803	–	* Sets suspended status and blocks as spare until resumed
804	–	* or shutdown.
805	–	* @returns true if still running on exit
806	–	*/
807	–	final boolean suspendAsSpare() {
808	–	lastEventCount = 0;         // reset upon resume
809	–	for (;;) {                  // set suspended unless terminating
810	–	int s = runState;
811	–	if ((s & TERMINATING) != 0) { // must kill
812	–	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
813	–	s | (TRIMMED | TERMINATING)))
814	–	return false;
815	–	}
816	–	else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
817	–	s | SUSPENDED))
818	–	break;
819	–	}
820	–	ForkJoinPool p = pool;
821	–	p.pushSpare(this);
822	–	while ((runState & SUSPENDED) != 0) {
823	–	if (!p.tryAccumulateStealCount(this))
824	–	continue;
825	–	interrupted();          // clear/ignore interrupts
826	–	if ((runState & SUSPENDED) == 0)
827	–	break;
828	–	if (nextSpare != 0)     // untimed
829	–	LockSupport.park(this);
830	–	else {
831	–	long startTime = System.nanoTime();
832	–	LockSupport.parkNanos(this, TRIM_RATE_NANOS);
833	–	if ((runState & SUSPENDED) == 0)
834	–	break;
835	–	long now = System.nanoTime();
836	–	if (now - startTime >= TRIM_RATE_NANOS)
837	–	pool.tryTrimSpare(now);
838	–	}
839	–	}
840	–	return runState == 0;
841	–	}
842	–
843	–	// Misc support methods for ForkJoinPool
844	–
845	–	/**
846	–	* Returns an estimate of the number of tasks in the queue.  Also
847	–	* used by ForkJoinTask.
848	–	*/
849	–	final int getQueueSize() {
850	–	int n; // external calls must read base first
851	–	return (n = -base + sp) <= 0 ? 0 : n;
852	–	}
853	–
854	–	/**
855	–	* Removes and cancels all tasks in queue.  Can be called from any
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)
876	–	t.cancelIgnoringExceptions();
877	–	}
878	–	}
879	–
880	–	/**
881	–	* Drains tasks to given collection c.
882	–	*
883	–	* @return the number of tasks drained
884	–	*/
885	–	final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
886	–	int n = 0;
887	–	while (base != sp) {
888	–	ForkJoinTask<?> t = deqTask();
889	–	if (t != null) {
890	–	c.add(t);
891	–	++n;
892	–	}
893	–	}
894	–	return n;
895	–	}
896	–
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	–	while (sp != base) {
906	–	if (active || (active = pool.tryIncrementActiveCount()))
907	–	return locallyFifo? locallyDeqTask() : popTask();
908	–	}
909	–	return null;
910	–	}
911	–
912	–	/**
913	–	* Gets and removes a local or stolen task.
914	–	*
915	–	* @return a task, if available
916	–	*/
917	–	final ForkJoinTask<?> pollTask() {
918	–	ForkJoinTask<?> t = pollLocalTask();
919	–	if (t == null) {
920	–	t = scan();
921	–	currentSteal = null; // cannot retain/track/help
922	–	}
923	–	return t;
924	–	}
925	–
926	–	/**
927	–	* Possibly runs some tasks and/or blocks, until task is done.
928	–	*
929	–	* @param joinMe the task to join
930	–	*/
931	–	final void joinTask(ForkJoinTask<?> joinMe) {
932	–	// currentJoin only written by this thread; only need ordered store
933	–	ForkJoinTask<?> prevJoin = currentJoin;
934	–	UNSAFE.putOrderedObject(this, currentJoinOffset, joinMe);
935	–	if (sp != base)
936	–	localHelpJoinTask(joinMe);
937	–	if (joinMe.status >= 0)
938	–	pool.awaitJoin(joinMe, this);
939	–	UNSAFE.putOrderedObject(this, currentJoinOffset, prevJoin);
940	–	}
941	–
942	–	/**
943	–	* Run tasks in local queue until given task is done.
944	–	*
945	–	* @param joinMe the task to join
946	–	*/
947	–	private void localHelpJoinTask(ForkJoinTask<?> joinMe) {
948	–	int s;
949	–	ForkJoinTask<?>[] q;
950	–	while (joinMe.status >= 0 && (s = sp) != base && (q = queue) != null) {
951	–	int i = (q.length - 1) & --s;
952	–	long u = (i << qShift) + qBase; // raw offset
953	–	ForkJoinTask<?> t = q[i];
954	–	if (t == null)  // lost to a stealer
955	–	break;
956	–	if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
957	–	/*
958	–	* This recheck (and similarly in helpJoinTask)
959	–	* handles cases where joinMe is independently
960	–	* cancelled or forced even though there is other work
961	–	* available. Back out of the pop by putting t back
962	–	* into slot before we commit by writing sp.
963	–	*/
964	–	if (joinMe.status < 0) {
965	–	UNSAFE.putObjectVolatile(q, u, t);
966	–	break;
967	–	}
968	–	sp = s;
969	–	// UNSAFE.putOrderedInt(this, spOffset, s);
970	–	t.quietlyExec();
971	–	}
972	–	}
973	–	}
974	–
975	–	/**
976	–	* Tries to locate and help perform tasks for a stealer of the
977	–	* given task, or in turn one of its stealers.  Traces
978	–	* currentSteal->currentJoin links looking for a thread working on
979	–	* a descendant of the given task and with a non-empty queue to
980	–	* steal back and execute tasks from.
981	–	*
982	–	* The implementation is very branchy to cope with the potential
983	–	* inconsistencies or loops encountering chains that are stale,
984	–	* unknown, or of length greater than MAX_HELP_DEPTH links.  All
985	–	* of these cases are dealt with by just returning back to the
986	–	* caller, who is expected to retry if other join mechanisms also
987	–	* don't work out.
988	–	*
989	–	* @param joinMe the task to join
990	–	*/
991	–	final void helpJoinTask(ForkJoinTask<?> joinMe) {
992	–	ForkJoinWorkerThread[] ws = pool.workers;
993	–	int n; // need at least 2 workers
994	–	if (ws != null && (n = ws.length) > 1 && joinMe.status >= 0) {
995	–	ForkJoinTask<?> task = joinMe;        // base of chain
996	–	ForkJoinWorkerThread thread = this;   // thread with stolen task
997	–	for (int d = 0; d < MAX_HELP_DEPTH; ++d) { // chain length
998	–	// Try to find v, the stealer of task, by first using hint
999	–	ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)];
1000	–	if (v == null || v.currentSteal != task) {
1001	–	for (int j = 0; ; ++j) {      // search array
1002	–	if (j < n) {
1003	–	if ((v = ws[j]) != null) {
1004	–	if (task.status < 0)
1005	–	return;       // stale or done
1006	–	if (v.currentSteal == task) {
1007	–	thread.stealHint = j;
1008	–	break;        // save hint for next time
1009	–	}
1010	–	}
1011	–	}
1012	–	else
1013	–	return;               // no stealer
1014	–	}
1015	–	}
1016	–	// Try to help v, using specialized form of deqTask
1017	–	int b;
1018	–	ForkJoinTask<?>[] q;
1019	–	while ((b = v.base) != v.sp && (q = v.queue) != null) {
1020	–	int i = (q.length - 1) & b;
1021	–	long u = (i << qShift) + qBase;
1022	–	ForkJoinTask<?> t = q[i];
1023	–	if (task.status < 0)
1024	–	return;                   // stale or done
1025	–	if (v.base == b) {
1026	–	if (t == null)
1027	–	return;               // producer stalled
1028	–	if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
1029	–	if (joinMe.status < 0) {
1030	–	UNSAFE.putObjectVolatile(q, u, t);
1031	–	return;           // back out on cancel
1032	–	}
1033	–	int pid = poolIndex;
1034	–	ForkJoinTask<?> prevSteal = currentSteal;
1035	–	currentSteal = t;
1036	–	v.stealHint = pid;
1037	–	v.base = b + 1;
1038	–	t.quietlyExec();
1039	–	currentSteal = prevSteal;
1040	–	}
1041	–	}
1042	–	if (joinMe.status < 0)
1043	–	return;
1044	–	}
1045	–	// Try to descend to find v's stealer
1046	–	ForkJoinTask<?> next = v.currentJoin;
1047	–	if (task.status < 0 || next == null || next == task ||
1048	–	joinMe.status < 0)
1049	–	return;
1050	–	task = next;
1051	–	thread = v;
1052	–	}
1053	–	}
1054	–	}
1055	–
1056	–	/**
1057	–	* Returns an estimate of the number of tasks, offset by a
1058	–	* function of number of idle workers.
1059	–	*
1060	–	* This method provides a cheap heuristic guide for task
1061	–	* partitioning when programmers, frameworks, tools, or languages
1062	–	* have little or no idea about task granularity.  In essence by
1063	–	* offering this method, we ask users only about tradeoffs in
1064	–	* overhead vs expected throughput and its variance, rather than
1065	–	* how finely to partition tasks.
1066	–	*
1067	–	* In a steady state strict (tree-structured) computation, each
1068	–	* thread makes available for stealing enough tasks for other
1069	–	* threads to remain active. Inductively, if all threads play by
1070	–	* the same rules, each thread should make available only a
1071	–	* constant number of tasks.
1072	–	*
1073	–	* The minimum useful constant is just 1. But using a value of 1
1074	–	* would require immediate replenishment upon each steal to
1075	–	* maintain enough tasks, which is infeasible.  Further,
1076	–	* partitionings/granularities of offered tasks should minimize
1077	–	* steal rates, which in general means that threads nearer the top
1078	–	* of computation tree should generate more than those nearer the
1079	–	* bottom. In perfect steady state, each thread is at
1080	–	* approximately the same level of computation tree. However,
1081	–	* producing extra tasks amortizes the uncertainty of progress and
1082	–	* diffusion assumptions.
1083	–	*
1084	–	* So, users will want to use values larger, but not much larger
1085	–	* than 1 to both smooth over transient shortages and hedge
1086	–	* against uneven progress; as traded off against the cost of
1087	–	* extra task overhead. We leave the user to pick a threshold
1088	–	* value to compare with the results of this call to guide
1089	–	* decisions, but recommend values such as 3.
1090	–	*
1091	–	* When all threads are active, it is on average OK to estimate
1092	–	* surplus strictly locally. In steady-state, if one thread is
1093	–	* maintaining say 2 surplus tasks, then so are others. So we can
1094	–	* just use estimated queue length (although note that (sp - base)
1095	–	* can be an overestimate because of stealers lagging increments
1096	–	* of base).  However, this strategy alone leads to serious
1097	–	* mis-estimates in some non-steady-state conditions (ramp-up,
1098	–	* ramp-down, other stalls). We can detect many of these by
1099	–	* further considering the number of "idle" threads, that are
1100	–	* known to have zero queued tasks, so compensate by a factor of
1101	–	* (#idle/#active) threads.
1102	–	*/
1103	–	final int getEstimatedSurplusTaskCount() {
1104	–	return sp - base - pool.idlePerActive();
1105	–	}
1106	–
1107	–	/**
1108	–	* Runs tasks until {@code pool.isQuiescent()}.
1109	–	*/
1110	–	final void helpQuiescePool() {
1111	–	for (;;) {
1112	–	ForkJoinTask<?> t = pollLocalTask();
1113	–	if (t != null || (t = scan()) != null) {
1114	–	t.quietlyExec();
1115	–	currentSteal = null;
1116	–	}
1117	–	else {
1118	–	ForkJoinPool p = pool;
1119	–	if (active) {
1120	–	if (!p.tryDecrementActiveCount())
1121	–	continue;   // retry later
1122	–	active = false; // inactivate
1123	–	}
1124	–	if (p.isQuiescent()) {
1125	–	active = true; // re-activate
1126	–	do {} while (!p.tryIncrementActiveCount());
1127	–	return;
1128	–	}
1129	–	}
1130	–	}
1131	–	}
1132	–
1133	–	// Unsafe mechanics
1134	–
1135	–	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1136	–	private static final long spOffset =
1137	–	objectFieldOffset("sp", ForkJoinWorkerThread.class);
1138	–	private static final long runStateOffset =
1139	–	objectFieldOffset("runState", ForkJoinWorkerThread.class);
1140	–	private static final long currentJoinOffset =
1141	–	objectFieldOffset("currentJoin", ForkJoinWorkerThread.class);
1142	–	private static final long currentStealOffset =
1143	–	objectFieldOffset("currentSteal", ForkJoinWorkerThread.class);
1144	–	private static final long qBase =
1145	–	UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1146	–
1147	–	private static final int qShift;
1148	–
1149	–	static {
1150	–	int s = UNSAFE.arrayIndexScale(ForkJoinTask[].class);
1151	–	if ((s & (s-1)) != 0)
1152	–	throw new Error("data type scale not a power of two");
1153	–	qShift = 31 - Integer.numberOfLeadingZeros(s);
1154	–	}
1155	–
1156	–	private static long objectFieldOffset(String field, Class<?> klazz) {
1157	–	try {
1158	–	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1159	–	} catch (NoSuchFieldException e) {
1160	–	// Convert Exception to corresponding Error
1161	–	NoSuchFieldError error = new NoSuchFieldError(field);
1162	–	error.initCause(e);
1163	–	throw error;
1164	–	}
1165	–	}
1166	–
1167	–	/**
1168	–	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1169	–	* Replace with a simple call to Unsafe.getUnsafe when integrating
1170	–	* into a jdk.
1171	–	*
1172	–	* @return a sun.misc.Unsafe
1173	–	*/
1174	–	private static sun.misc.Unsafe getUnsafe() {
1175	–	try {
1176	–	return sun.misc.Unsafe.getUnsafe();
1177	–	} catch (SecurityException se) {
108		try {
109	<	return java.security.AccessController.doPrivileged
110	<	(new java.security
111	<	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
112	<	public sun.misc.Unsafe run() throws Exception {
113	<	java.lang.reflect.Field f = sun.misc
114	<	.Unsafe.class.getDeclaredField("theUnsafe");
1185	<	f.setAccessible(true);
1186	<	return (sun.misc.Unsafe) f.get(null);
1187	<	}});
1188	<	} catch (java.security.PrivilegedActionException e) {
1189	<	throw new RuntimeException("Could not initialize intrinsics",
1190	<	e.getCause());
109	>	onTermination(exception);
110	>	} catch (Throwable ex) {
111	>	if (exception == null)
112	>	exception = ex;
113	>	} finally {
114	>	pool.deregisterWorker(this, exception);
115		}
116		}
117		}
118		}
119	+

Diff Legend

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