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 1.28 by dl, Mon Aug 3 13:40:07 2009 UTC vs.
Revision 1.33 by dl, Thu May 27 16:46:49 2010 UTC

#	Line 8 | Line 8 | 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
15		/**
16		* A thread managed by a {@link ForkJoinPool}.  This class is
#	Line 25 | Line 27 | import java.util.Collection;
27		*/
28		public class ForkJoinWorkerThread extends Thread {
29		/*
30	<	* Algorithm overview:
30	>	* Overview:
31		*
32	<	* 1. Work-Stealing: Work-stealing queues are special forms of
33	<	* Deques that support only three of the four possible
34	<	* end-operations -- push, pop, and deq (aka steal), and only do
35	<	* so under the constraints that push and pop are called only from
36	<	* the owning thread, while deq may be called from other threads.
37	<	* (If you are unfamiliar with them, you probably want to read
38	<	* Herlihy and Shavit's book "The Art of Multiprocessor
39	<	* programming", chapter 16 describing these in more detail before
40	<	* proceeding.)  The main work-stealing queue design is roughly
41	<	* similar to "Dynamic Circular Work-Stealing Deque" by David
42	<	* Chase and Yossi Lev, SPAA 2005
43	<	* (http://research.sun.com/scalable/pubs/index.html).  The main
44	<	* difference ultimately stems from gc requirements that we null
45	<	* out taken slots as soon as we can, to maintain as small a
46	<	* footprint as possible even in programs generating huge numbers
47	<	* of tasks. To accomplish this, we shift the CAS arbitrating pop
48	<	* vs deq (steal) from being on the indices ("base" and "sp") to
49	<	* the slots themselves (mainly via method "casSlotNull()"). So,
50	<	* both a successful pop and deq mainly entail CAS'ing a non-null
51	<	* slot to null.  Because we rely on CASes of references, we do
52	<	* not need tag bits on base or sp.  They are simple ints as used
53	<	* in any circular array-based queue (see for example ArrayDeque).
54	<	* Updates to the indices must still be ordered in a way that
55	<	* guarantees that (sp - base) > 0 means the queue is empty, but
56	<	* otherwise may err on the side of possibly making the queue
57	<	* appear nonempty when a push, pop, or deq have not fully
58	<	* committed. Note that this means that the deq operation,
59	<	* considered individually, is not wait-free. One thief cannot
60	<	* successfully continue until another in-progress one (or, if
61	<	* previously empty, a push) completes.  However, in the
62	<	* aggregate, we ensure at least probabilistic
63	<	* non-blockingness. If an attempted steal fails, a thief always
64	<	* chooses a different random victim target to try next. So, in
65	<	* order for one thief to progress, it suffices for any
66	<	* in-progress deq or new push on any empty queue to complete. One
67	<	* reason this works well here is that apparently-nonempty often
68	<	* means soon-to-be-stealable, which gives threads a chance to
69	<	* activate if necessary before stealing (see below).
32	>	* ForkJoinWorkerThreads are managed by ForkJoinPools and perform
33	>	* ForkJoinTasks. This class includes bookkeeping in support of
34	>	* 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
#	Line 75 | Line 86 | public class ForkJoinWorkerThread extend
86		* Efficient implementation of this approach currently relies on
87		* an uncomfortable amount of "Unsafe" mechanics. To maintain
88		* correct orderings, reads and writes of variable base require
89	<	* volatile ordering.  Variable sp does not require volatile write
90	<	* but needs cheaper store-ordering on writes.  Because they are
91	<	* protected by volatile base reads, reads of the queue array and
92	<	* its slots do not need volatile load semantics, but writes (in
93	<	* push) require store order and CASes (in pop and deq) require
94	<	* (volatile) CAS semantics.  (See "Idempotent work stealing" by
95	<	* Michael, Saraswat, and Vechev, PPoPP 2009
96	<	* http://portal.acm.org/citation.cfm?id=1504186 for an algorithm
97	<	* with similar properties, but without support for nulling
98	<	* slots.)  Since these combinations aren't supported using
99	<	* ordinary volatiles, the only way to accomplish these
89	>	* volatile ordering.  Variable sp does not require volatile
90	>	* writes but still needs store-ordering, which we accomplish by
91	>	* pre-incrementing sp before filling the slot with an ordered
92	>	* store.  (Pre-incrementing also enables backouts used in
93	>	* scanWhileJoining.)  Because they are protected by volatile base
94	>	* reads, reads of the queue array and its slots by other threads
95	>	* do not need volatile load semantics, but writes (in push)
96	>	* require store order and CASes (in pop and deq) require
97	>	* (volatile) CAS semantics.  (Michael, Saraswat, and Vechev's
98	>	* algorithm has similar properties, but without support for
99	>	* nulling slots.)  Since these combinations aren't supported
100	>	* using ordinary volatiles, the only way to accomplish these
101		* efficiently is to use direct Unsafe calls. (Using external
102		* AtomicIntegers and AtomicReferenceArrays for the indices and
103		* array is significantly slower because of memory locality and
104		* indirection effects.)
105	<	*
105	>	*
106		* Further, performance on most platforms is very sensitive to
107		* placement and sizing of the (resizable) queue array.  Even
108		* though these queues don't usually become all that big, the
109		* initial size must be large enough to counteract cache
110		* contention effects across multiple queues (especially in the
111		* presence of GC cardmarking). Also, to improve thread-locality,
112	<	* queues are currently initialized immediately after the thread
113	<	* gets the initial signal to start processing tasks.  However,
114	<	* all queue-related methods except pushTask are written in a way
115	<	* that allows them to instead be lazily allocated and/or disposed
116	<	* of when empty. All together, these low-level implementation
117	<	* choices produce as much as a factor of 4 performance
118	<	* improvement compared to naive implementations, and enable the
119	<	* processing of billions of tasks per second, sometimes at the
120	<	* expense of ugliness.
121	<	*
122	<	* 2. Run control: The primary run control is based on a global
123	<	* counter (activeCount) held by the pool. It uses an algorithm
124	<	* similar to that in Herlihy and Shavit section 17.6 to cause
113	<	* threads to eventually block when all threads declare they are
114	<	* inactive. For this to work, threads must be declared active
115	<	* when executing tasks, and before stealing a task. They must be
116	<	* inactive before blocking on the Pool Barrier (awaiting a new
117	<	* submission or other Pool event). In between, there is some free
118	<	* play which we take advantage of to avoid contention and rapid
119	<	* flickering of the global activeCount: If inactive, we activate
120	<	* only if a victim queue appears to be nonempty (see above).
121	<	* Similarly, a thread tries to inactivate only after a full scan
122	<	* of other threads.  The net effect is that contention on
123	<	* activeCount is rarely a measurable performance issue. (There
124	<	* are also a few other cases where we scan for work rather than
125	<	* retry/block upon contention.)
126	<	*
127	<	* 3. Selection control. We maintain policy of always choosing to
128	<	* run local tasks rather than stealing, and always trying to
129	<	* steal tasks before trying to run a new submission. All steals
130	<	* are currently performed in randomly-chosen deq-order. It may be
131	<	* worthwhile to bias these with locality / anti-locality
132	<	* information, but doing this well probably requires more
133	<	* lower-level information from JVMs than currently provided.
112	>	* queues are initialized after starting.  All together, these
113	>	* low-level implementation choices produce as much as a factor of
114	>	* 4 performance improvement compared to naive implementations,
115	>	* and enable the processing of billions of tasks per second,
116	>	* sometimes at the expense of ugliness.
117	>	*/
118	>
119	>	/**
120	>	* Generator for initial random seeds for random victim
121	>	* selection. This is used only to create initial seeds. Random
122	>	* steals use a cheaper xorshift generator per steal attempt. We
123	>	* expect only rare contention on seedGenerator, so just use a
124	>	* plain Random.
125		*/
126	+	private static final Random seedGenerator = new Random();
127	+
128	+	/**
129	+	* The timeout value for suspending spares. Spare workers that
130	+	* remain unsignalled for more than this time may be trimmed
131	+	* (killed and removed from pool).  Since our goal is to avoid
132	+	* long-term thread buildup, the exact value of timeout does not
133	+	* matter too much so long as it avoids most false-alarm timeouts
134	+	* under GC stalls or momentarily high system load.
135	+	*/
136	+	private static final long SPARE_KEEPALIVE_NANOS =
137	+	5L * 1000L * 1000L * 1000L; // 5 secs
138
139		/**
140		* Capacity of work-stealing queue array upon initialization.
#	Line 155 | Line 158 | public class ForkJoinWorkerThread extend
158
159		/**
160		* The work-stealing queue array. Size must be a power of two.
161	<	* Initialized when thread starts, to improve memory locality.
161	>	* Initialized in onStart, to improve memory locality.
162		*/
163		private ForkJoinTask<?>[] queue;
164
165		/**
163	–	* Index (mod queue.length) of next queue slot to push to or pop
164	–	* from. It is written only by owner thread, via ordered store.
165	–	* Both sp and base are allowed to wrap around on overflow, but
166	–	* (sp - base) still estimates size.
167	–	*/
168	–	private volatile int sp;
169	–
170	–	/**
166		* Index (mod queue.length) of least valid queue slot, which is
167		* always the next position to steal from if nonempty.
168		*/
169		private volatile int base;
170
171		/**
172	<	* Activity status. When true, this worker is considered active.
173	<	* Must be false upon construction. It must be true when executing
174	<	* tasks, and BEFORE stealing a task. It must be false before
175	<	* calling pool.sync.
176	<	*/
177	<	private boolean active;
172	>	* Index (mod queue.length) of next queue slot to push to or pop
173	>	* from. It is written only by owner thread, and accessed by other
174	>	* threads only after reading (volatile) base.  Both sp and base
175	>	* are allowed to wrap around on overflow, but (sp - base) still
176	>	* estimates size.
177	>	*/
178	>	private int sp;
179
180		/**
181	<	* Run state of this worker. Supports simple versions of the usual
182	<	* shutdown/shutdownNow control.
181	>	* Run state of this worker. In addition to the usual run levels,
182	>	* tracks if this worker is suspended as a spare, and if it was
183	>	* killed (trimmed) while suspended. However, "active" status is
184	>	* maintained separately.
185		*/
186		private volatile int runState;
187
188	+	private static final int TERMINATING = 0x01;
189	+	private static final int TERMINATED  = 0x02;
190	+	private static final int SUSPENDED   = 0x04; // inactive spare
191	+	private static final int TRIMMED     = 0x08; // killed while suspended
192	+
193	+	/**
194	+	* Number of LockSupport.park calls to block this thread for
195	+	* 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.
204	+	*/
205	+	int stealCount;
206	+
207		/**
208		* Seed for random number generator for choosing steal victims.
209	<	* Uses Marsaglia xorshift. Must be nonzero upon initialization.
209	>	* Uses Marsaglia xorshift. Must be initialized as nonzero.
210		*/
211		private int seed;
212
213		/**
214	<	* Number of steals, transferred to pool when idle
214	>	* Activity status. When true, this worker is considered active.
215	>	* Accessed directly by pool.  Must be false upon construction.
216		*/
217	<	private int stealCount;
217	>	boolean active;
218	>
219	>	/**
220	>	* True if use local fifo, not default lifo, for local polling.
221	>	* Shadows value from ForkJoinPool, which resets it if changed
222	>	* pool-wide.
223	>	*/
224	>	private boolean locallyFifo;
225
226		/**
227		* Index of this worker in pool array. Set once by pool before
228	<	* running, and accessed directly by pool during cleanup etc.
228	>	* running, and accessed directly by pool to locate this worker in
229	>	* its workers array.
230		*/
231		int poolIndex;
232
233		/**
234	<	* The last barrier event waited for. Accessed in pool callback
235	<	* methods, but only by current thread.
234	>	* The last pool event waited for. Accessed only by pool in
235	>	* callback methods invoked within this thread.
236		*/
237	<	long lastEventCount;
237	>	int lastEventCount;
238
239		/**
240	<	* True if use local fifo, not default lifo, for local polling
240	>	* Encoded index and event count of next event waiter. Used only
241	>	* by ForkJoinPool for managing event waiters.
242		*/
243	<	private boolean locallyFifo;
243	>	volatile long nextWaiter;
244
245		/**
246		* Creates a ForkJoinWorkerThread operating in the given pool.
#	Line 224 | Line 251 | public class ForkJoinWorkerThread extend
251		protected ForkJoinWorkerThread(ForkJoinPool pool) {
252		if (pool == null) throw new NullPointerException();
253		this.pool = pool;
254	<	// Note: poolIndex is set by pool during construction
255	<	// Remaining initialization is deferred to onStart
254	>	// To avoid exposing construction details to subclasses,
255	>	// remaining initialization is in start() and onStart()
256	>	}
257	>
258	>	/**
259	>	* Performs additional initialization and starts this thread
260	>	*/
261	>	final void start(int poolIndex, boolean locallyFifo,
262	>	UncaughtExceptionHandler ueh) {
263	>	this.poolIndex = poolIndex;
264	>	this.locallyFifo = locallyFifo;
265	>	if (ueh != null)
266	>	setUncaughtExceptionHandler(ueh);
267	>	setDaemon(true);
268	>	start();
269		}
270
271	<	// Public access methods
271	>	// Public/protected methods
272
273		/**
274		* Returns the pool hosting this thread.
#	Line 253 | Line 293 | public class ForkJoinWorkerThread extend
293		}
294
295		/**
296	<	* Establishes local first-in-first-out scheduling mode for forked
297	<	* tasks that are never joined.
298	<	*
299	<	* @param async if true, use locally FIFO scheduling
296	>	* Initializes internal state after construction but before
297	>	* processing any tasks. If you override this method, you must
298	>	* invoke super.onStart() at the beginning of the method.
299	>	* Initialization requires care: Most fields must have legal
300	>	* default values, to ensure that attempted accesses from other
301	>	* threads work correctly even before this thread starts
302	>	* processing tasks.
303		*/
304	<	void setAsyncMode(boolean async) {
305	<	locallyFifo = async;
306	<	}
264	<
265	<	// Runstate management
266	<
267	<	// Runstate values. Order matters
268	<	private static final int RUNNING     = 0;
269	<	private static final int SHUTDOWN    = 1;
270	<	private static final int TERMINATING = 2;
271	<	private static final int TERMINATED  = 3;
272	<
273	<	final boolean isShutdown()    { return runState >= SHUTDOWN;  }
274	<	final boolean isTerminating() { return runState >= TERMINATING;  }
275	<	final boolean isTerminated()  { return runState == TERMINATED; }
276	<	final boolean shutdown()      { return transitionRunStateTo(SHUTDOWN); }
277	<	final boolean shutdownNow()   { return transitionRunStateTo(TERMINATING); }
304	>	protected void onStart() {
305	>	int rs = seedGenerator.nextInt();
306	>	seed = rs == 0? 1 : rs; // seed must be nonzero
307
308	<	/**
309	<	* Transitions to at least the given state.
310	<	*
311	<	* @return {@code true} if not already at least at given state
283	<	*/
284	<	private boolean transitionRunStateTo(int state) {
285	<	for (;;) {
286	<	int s = runState;
287	<	if (s >= state)
288	<	return false;
289	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, state))
290	<	return true;
291	<	}
292	<	}
308	>	// Allocate name string and queue array in this thread
309	>	String pid = Integer.toString(pool.getPoolNumber());
310	>	String wid = Integer.toString(poolIndex);
311	>	setName("ForkJoinPool-" + pid + "-worker-" + wid);
312
313	<	/**
295	<	* Tries to set status to active; fails on contention.
296	<	*/
297	<	private boolean tryActivate() {
298	<	if (!active) {
299	<	if (!pool.tryIncrementActiveCount())
300	<	return false;
301	<	active = true;
302	<	}
303	<	return true;
313	>	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
314		}
315
316		/**
317	<	* Tries to set status to inactive; fails on contention.
317	>	* Performs cleanup associated with termination of this worker
318	>	* thread.  If you override this method, you must invoke
319	>	* {@code super.onTermination} at the end of the overridden method.
320	>	*
321	>	* @param exception the exception causing this thread to abort due
322	>	* to an unrecoverable error, or {@code null} if completed normally
323		*/
324	<	private boolean tryInactivate() {
325	<	if (active) {
326	<	if (!pool.tryDecrementActiveCount())
327	<	return false;
328	<	active = false;
324	>	protected void onTermination(Throwable exception) {
325	>	try {
326	>	cancelTasks();
327	>	setTerminated();
328	>	pool.workerTerminated(this);
329	>	} catch (Throwable ex) {        // Shouldn't ever happen
330	>	if (exception == null)      // but if so, at least rethrown
331	>	exception = ex;
332	>	} finally {
333	>	if (exception != null)
334	>	UNSAFE.throwException(exception);
335		}
315	–	return true;
316	–	}
317	–
318	–	/**
319	–	* Computes next value for random victim probe.  Scans don't
320	–	* require a very high quality generator, but also not a crummy
321	–	* one.  Marsaglia xor-shift is cheap and works well.
322	–	*/
323	–	private static int xorShift(int r) {
324	–	r ^= (r << 13);
325	–	r ^= (r >>> 17);
326	–	return r ^ (r << 5);
336		}
337
329	–	// Lifecycle methods
330	–
338		/**
339		* This method is required to be public, but should never be
340		* called explicitly. It performs the main run loop to execute
#	Line 337 | Line 344 | public class ForkJoinWorkerThread extend
344		Throwable exception = null;
345		try {
346		onStart();
340	–	pool.sync(this); // await first pool event
347		mainLoop();
348		} catch (Throwable ex) {
349		exception = ex;
#	Line 346 | Line 352 | public class ForkJoinWorkerThread extend
352		}
353		}
354
355	+	// helpers for run()
356	+
357		/**
358	<	* Executes tasks until shut down.
358	>	* Find and execute tasks and check status while running
359		*/
360		private void mainLoop() {
361	<	while (!isShutdown()) {
362	<	ForkJoinTask<?> t = pollTask();
363	<	if (t != null || (t = pollSubmission()) != null)
364	<	t.quietlyExec();
365	<	else if (tryInactivate())
366	<	pool.sync(this);
361	>	boolean ran = false; // true if ran task on previous step
362	>	ForkJoinPool p = pool;
363	>	for (;;) {
364	>	p.preStep(this, ran);
365	>	if (runState != 0)
366	>	return;
367	>	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	>	}
373		}
374		}
375
376		/**
377	<	* Initializes internal state after construction but before
378	<	* processing any tasks. If you override this method, you must
365	<	* invoke super.onStart() at the beginning of the method.
366	<	* Initialization requires care: Most fields must have legal
367	<	* default values, to ensure that attempted accesses from other
368	<	* threads work correctly even before this thread starts
369	<	* processing tasks.
377	>	* Runs local tasks until queue is empty or shut down.  Call only
378	>	* while active.
379		*/
380	<	protected void onStart() {
381	<	// Allocate while starting to improve chances of thread-local
382	<	// isolation
383	<	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
384	<	// Initial value of seed need not be especially random but
385	<	// should differ across workers and must be nonzero
386	<	int p = poolIndex + 1;
387	<	seed = p + (p << 8) + (p << 16) + (p << 24); // spread bits
380	>	private void runLocalTasks() {
381	>	while (runState == 0) {
382	>	ForkJoinTask<?> t = locallyFifo? locallyDeqTask() : popTask();
383	>	if (t != null)
384	>	t.tryExec();
385	>	else if (base == sp)
386	>	break;
387	>	}
388		}
389
390		/**
391	<	* Performs cleanup associated with termination of this worker
383	<	* thread.  If you override this method, you must invoke
384	<	* {@code super.onTermination} at the end of the overridden method.
391	>	* If a submission exists, try to activate and take it
392		*
393	<	* @param exception the exception causing this thread to abort due
387	<	* to an unrecoverable error, or {@code null} if completed normally
393	>	* @return a task, if available
394		*/
395	<	protected void onTermination(Throwable exception) {
396	<	// Execute remaining local tasks unless aborting or terminating
397	<	while (exception == null && pool.isProcessingTasks() && base != sp) {
398	<	try {
399	<	ForkJoinTask<?> t = popTask();
400	<	if (t != null)
395	<	t.quietlyExec();
396	<	} catch (Throwable ex) {
397	<	exception = ex;
395	>	private ForkJoinTask<?> pollSubmission() {
396	>	ForkJoinPool p = pool;
397	>	while (p.hasQueuedSubmissions()) {
398	>	if (active || (active = p.tryIncrementActiveCount())) {
399	>	ForkJoinTask<?> t = p.pollSubmission();
400	>	return t != null ? t : scan(); // if missed, rescan
401		}
402		}
403	<	// Cancel other tasks, transition status, notify pool, and
401	<	// propagate exception to uncaught exception handler
402	<	try {
403	<	do {} while (!tryInactivate()); // ensure inactive
404	<	cancelTasks();
405	<	runState = TERMINATED;
406	<	pool.workerTerminated(this);
407	<	} catch (Throwable ex) {        // Shouldn't ever happen
408	<	if (exception == null)      // but if so, at least rethrown
409	<	exception = ex;
410	<	} finally {
411	<	if (exception != null)
412	<	ForkJoinTask.rethrowException(exception);
413	<	}
403	>	return null;
404		}
405
406	<	// Intrinsics-based support for queue operations.
407	<
408	<	/**
409	<	* Adds in store-order the given task at given slot of q to null.
410	<	* Caller must ensure q is non-null and index is in range.
406	>	/*
407	>	* Intrinsics-based atomic writes for queue slots. These are
408	>	* basically the same as methods in AtomicObjectArray, but
409	>	* specialized for (1) ForkJoinTask elements (2) requirement that
410	>	* nullness and bounds checks have already been performed by
411	>	* callers and (3) effective offsets are known not to overflow
412	>	* from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't
413	>	* need corresponding version for reads: plain array reads are OK
414	>	* because they protected by other volatile reads and are
415	>	* confirmed by CASes.
416	>	*
417	>	* Most uses don't actually call these methods, but instead contain
418	>	* inlined forms that enable more predictable optimization.  We
419	>	* don't define the version of write used in pushTask at all, but
420	>	* instead inline there a store-fenced array slot write.
421		*/
422	–	private static void setSlot(ForkJoinTask<?>[] q, int i,
423	–	ForkJoinTask<?> t) {
424	–	UNSAFE.putOrderedObject(q, (i << qShift) + qBase, t);
425	–	}
422
423		/**
424	<	* CAS given slot of q to null. Caller must ensure q is non-null
425	<	* and index is in range.
424	>	* CASes slot i of array q from t to null. Caller must ensure q is
425	>	* non-null and index is in range.
426		*/
427	<	private static boolean casSlotNull(ForkJoinTask<?>[] q, int i,
428	<	ForkJoinTask<?> t) {
427	>	private static final boolean casSlotNull(ForkJoinTask<?>[] q, int i,
428	>	ForkJoinTask<?> t) {
429		return UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
430		}
431
432		/**
433	<	* Sets sp in store-order.
433	>	* Performs a volatile write of the given task at given slot of
434	>	* array q.  Caller must ensure q is non-null and index is in
435	>	* range. This method is used only during resets and backouts.
436		*/
437	<	private void storeSp(int s) {
438	<	UNSAFE.putOrderedInt(this, spOffset, s);
437	>	private static final void writeSlot(ForkJoinTask<?>[] q, int i,
438	>	ForkJoinTask<?> t) {
439	>	UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
440		}
441
442	<	// Main queue methods
442	>	// queue methods
443
444		/**
445	<	* Pushes a task. Called only by current thread.
445	>	* Pushes a task. Call only from this thread.
446		*
447		* @param t the task. Caller must ensure non-null.
448		*/
449		final void pushTask(ForkJoinTask<?> t) {
450	+	int s;
451		ForkJoinTask<?>[] q = queue;
452	<	int mask = q.length - 1;
453	<	int s = sp;
454	<	setSlot(q, s & mask, t);
455	<	storeSp(++s);
456	<	if ((s -= base) == 1)
452	>	int mask = q.length - 1; // implicit assert q != null
453	>	UNSAFE.putOrderedObject(q, (((s = sp++) & mask) << qShift) + qBase, t);
454	>	if ((s -= base) <= 0)
455		pool.signalWork();
456	<	else if (s >= mask)
456	>	else if (s + 1 >= mask)
457		growQueue();
458		}
459
460		/**
461		* Tries to take a task from the base of the queue, failing if
462	<	* either empty or contended.
462	>	* empty or contended. Note: Specializations of this code appear
463	>	* in scan and scanWhileJoining.
464		*
465		* @return a task, or null if none or contended
466		*/
467		final ForkJoinTask<?> deqTask() {
468		ForkJoinTask<?> t;
469		ForkJoinTask<?>[] q;
470	<	int i;
471	<	int b;
473	<	if (sp != (b = base) &&
470	>	int b, i;
471	>	if ((b = base) != sp &&
472		(q = queue) != null && // must read q after b
473		(t = q[i = (q.length - 1) & b]) != null &&
474	<	casSlotNull(q, i, t)) {
474	>	UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) {
475		base = b + 1;
476		return t;
477		}
#	Line 481 | Line 479 | public class ForkJoinWorkerThread extend
479		}
480
481		/**
482	<	* Tries to take a task from the base of own queue, activating if
483	<	* necessary, failing only if empty. Called only by current thread.
482	>	* Tries to take a task from the base of own queue. Assumes active
483	>	* status.  Called only by current thread.
484		*
485		* @return a task, or null if none
486		*/
487		final ForkJoinTask<?> locallyDeqTask() {
488	<	int b;
489	<	while (sp != (b = base)) {
490	<	if (tryActivate()) {
491	<	ForkJoinTask<?>[] q = queue;
492	<	int i = (q.length - 1) & b;
493	<	ForkJoinTask<?> t = q[i];
494	<	if (t != null && casSlotNull(q, i, t)) {
488	>	ForkJoinTask<?>[] q = queue;
489	>	if (q != null) {
490	>	ForkJoinTask<?> t;
491	>	int b, i;
492	>	while (sp != (b = base)) {
493	>	if ((t = q[i = (q.length - 1) & b]) != null &&
494	>	UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase,
495	>	t, null)) {
496		base = b + 1;
497		return t;
498		}
#	Line 503 | Line 502 | public class ForkJoinWorkerThread extend
502		}
503
504		/**
505	<	* Returns a popped task, or null if empty. Ensures active status
506	<	* if non-null. Called only by current thread.
505	>	* Returns a popped task, or null if empty. Assumes active status.
506	>	* Called only by current thread. (Note: a specialization of this
507	>	* code appears in popWhileJoining.)
508		*/
509		final ForkJoinTask<?> popTask() {
510	<	int s = sp;
511	<	while (s != base) {
512	<	if (tryActivate()) {
513	<	ForkJoinTask<?>[] q = queue;
514	<	int mask = q.length - 1;
515	<	int i = (s - 1) & mask;
516	<	ForkJoinTask<?> t = q[i];
517	<	if (t == null || !casSlotNull(q, i, t))
518	<	break;
519	<	storeSp(s - 1);
510	>	int s;
511	>	ForkJoinTask<?>[] q;
512	>	if (base != (s = sp) && (q = queue) != null) {
513	>	int i = (q.length - 1) & --s;
514	>	ForkJoinTask<?> t = q[i];
515	>	if (t != null && UNSAFE.compareAndSwapObject
516	>	(q, (i << qShift) + qBase, t, null)) {
517	>	sp = s;
518		return t;
519		}
520		}
#	Line 524 | Line 522 | public class ForkJoinWorkerThread extend
522		}
523
524		/**
525	<	* Specialized version of popTask to pop only if
526	<	* topmost element is the given task. Called only
527	<	* by current thread while active.
525	>	* Specialized version of popTask to pop only if topmost element
526	>	* is the given task. Called only by current thread while
527	>	* active.
528		*
529		* @param t the task. Caller must ensure non-null.
530		*/
531		final boolean unpushTask(ForkJoinTask<?> t) {
532	<	ForkJoinTask<?>[] q = queue;
533	<	int mask = q.length - 1;
534	<	int s = sp - 1;
535	<	if (casSlotNull(q, s & mask, t)) {
536	<	storeSp(s);
532	>	int s;
533	>	ForkJoinTask<?>[] q;
534	>	if (base != (s = sp) && (q = queue) != null &&
535	>	UNSAFE.compareAndSwapObject
536	>	(q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) {
537	>	sp = s;
538		return true;
539		}
540		return false;
#	Line 575 | Line 574 | public class ForkJoinWorkerThread extend
574		ForkJoinTask<?> t = oldQ[oldIndex];
575		if (t != null && !casSlotNull(oldQ, oldIndex, t))
576		t = null;
577	<	setSlot(newQ, b & newMask, t);
577	>	writeSlot(newQ, b & newMask, t);
578		} while (++b != bf);
579		pool.signalWork();
580		}
581
582		/**
583	+	* Computes next value for random victim probe in scan().  Scans
584	+	* don't require a very high quality generator, but also not a
585	+	* crummy one.  Marsaglia xor-shift is cheap and works well enough.
586	+	* Note: This is manually inlined in scan()
587	+	*/
588	+	private static final int xorShift(int r) {
589	+	r ^= r << 13;
590	+	r ^= r >>> 17;
591	+	return r ^ (r << 5);
592	+	}
593	+
594	+	/**
595		* Tries to steal a task from another worker. Starts at a random
596		* index of workers array, and probes workers until finding one
597		* with non-empty queue or finding that all are empty.  It
598		* randomly selects the first n probes. If these are empty, it
599	<	* resorts to a full circular traversal, which is necessary to
600	<	* accurately set active status by caller. Also restarts if pool
601	<	* events occurred since last scan, which forces refresh of
602	<	* workers array, in case barrier was associated with resize.
599	>	* resorts to a circular sweep, which is necessary to accurately
600	>	* set active status. (The circular sweep uses steps of
601	>	* approximately half the array size plus 1, to avoid bias
602	>	* stemming from leftmost packing of the array in ForkJoinPool.)
603		*
604		* This method must be both fast and quiet -- usually avoiding
605		* memory accesses that could disrupt cache sharing etc other than
606	<	* those needed to check for and take tasks. This accounts for,
607	<	* among other things, updating random seed in place without
608	<	* storing it until exit.
606	>	* those needed to check for and take tasks (or to activate if not
607	>	* already active). This accounts for, among other things,
608	>	* updating random seed in place without storing it until exit.
609		*
610		* @return a task, or null if none found
611		*/
612		private ForkJoinTask<?> scan() {
613	<	ForkJoinTask<?> t = null;
614	<	int r = seed;                    // extract once to keep scan quiet
615	<	ForkJoinWorkerThread[] ws;       // refreshed on outer loop
616	<	int mask;                        // must be power 2 minus 1 and > 0
617	<	outer:do {
618	<	if ((ws = pool.workers) != null && (mask = ws.length - 1) > 0) {
619	<	int idx = r;
620	<	int probes = ~mask;      // use random index while negative
621	<	for (;;) {
622	<	r = xorShift(r);     // update random seed
623	<	ForkJoinWorkerThread v = ws[mask & idx];
624	<	if (v == null || v.sp == v.base) {
625	<	if (probes <= mask)
626	<	idx = (probes++ < 0) ? r : (idx + 1);
627	<	else
628	<	break;
613	>	ForkJoinPool p = pool;
614	>	ForkJoinWorkerThread[] ws;        // worker array
615	>	int n;                            // upper bound of #workers
616	>	if ((ws = p.workers) != null && (n = ws.length) > 1) {
617	>	boolean canSteal = active;    // shadow active status
618	>	int r = seed;                 // extract seed once
619	>	int mask = n - 1;
620	>	int j = -n;                   // loop counter
621	>	int k = r;                    // worker index, random if j < 0
622	>	for (;;) {
623	>	ForkJoinWorkerThread v = ws[k & mask];
624	>	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift
625	>	if (v != null && v.base != v.sp) {
626	>	int b, i;             // inline specialized deqTask
627	>	ForkJoinTask<?>[] q;
628	>	ForkJoinTask<?> t;
629	>	if ((canSteal ||      // ensure active status
630	>	(canSteal = active = p.tryIncrementActiveCount())) &&
631	>	(q = v.queue) != null &&
632	>	(t = q[i = (q.length - 1) & (b = v.base)]) != null &&
633	>	UNSAFE.compareAndSwapObject
634	>	(q, (i << qShift) + qBase, t, null)) {
635	>	v.base = b + 1;
636	>	seed = r;
637	>	++stealCount;
638	>	return t;
639		}
640	<	else if (!tryActivate() || (t = v.deqTask()) == null)
641	<	continue outer;  // restart on contention
621	<	else
622	<	break outer;
640	>	j = -n;
641	>	k = r;                // restart on contention
642		}
643	+	else if (++j <= 0)
644	+	k = r;
645	+	else if (j <= n)
646	+	k += (n >>> 1) | 1;
647	+	else
648	+	break;
649		}
650	<	} while (pool.hasNewSyncEvent(this)); // retry on pool events
651	<	seed = r;
627	<	return t;
650	>	}
651	>	return null;
652		}
653
654	+	// Run State management
655	+
656	+	// status check methods used mainly by ForkJoinPool
657	+	final boolean isTerminating() { return (runState & TERMINATING) != 0; }
658	+	final boolean isTerminated()  { return (runState & TERMINATED) != 0; }
659	+	final boolean isSuspended()   { return (runState & SUSPENDED) != 0; }
660	+	final boolean isTrimmed()     { return (runState & TRIMMED) != 0; }
661	+
662		/**
663	<	* Gets and removes a local or stolen task.
632	<	*
633	<	* @return a task, if available
663	>	* Sets state to TERMINATING, also resuming if suspended.
664		*/
665	<	final ForkJoinTask<?> pollTask() {
666	<	ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask();
667	<	if (t == null && (t = scan()) != null)
668	<	++stealCount;
669	<	return t;
665	>	final void shutdown() {
666	>	for (;;) {
667	>	int s = runState;
668	>	if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended
669	>	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
670	>	(s & ~SUSPENDED) |
671	>	(TRIMMED|TERMINATING))) {
672	>	LockSupport.unpark(this);
673	>	break;
674	>	}
675	>	}
676	>	else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
677	>	s | TERMINATING))
678	>	break;
679	>	}
680		}
681
682		/**
683	<	* Gets a local task.
683	>	* Sets state to TERMINATED. Called only by this thread.
684	>	*/
685	>	private void setTerminated() {
686	>	int s;
687	>	do {} while (!UNSAFE.compareAndSwapInt(this, runStateOffset,
688	>	s = runState,
689	>	s | (TERMINATING|TERMINATED)));
690	>	}
691	>
692	>	/**
693	>	* Instrumented version of park. Also used by ForkJoinPool.awaitEvent
694	>	*/
695	>	final void doPark() {
696	>	++parkCount;
697	>	LockSupport.park(this);
698	>	}
699	>
700	>	/**
701	>	* If suspended, tries to set status to unsuspended.
702	>	* Caller must unpark to actually resume
703		*
704	<	* @return a task, if available
704	>	* @return true if successful
705		*/
706	<	final ForkJoinTask<?> pollLocalTask() {
707	<	return locallyFifo ? locallyDeqTask() : popTask();
706	>	final boolean tryUnsuspend() {
707	>	int s;
708	>	return (((s = runState) & SUSPENDED) != 0 &&
709	>	UNSAFE.compareAndSwapInt(this, runStateOffset, s,
710	>	s & ~SUSPENDED));
711		}
712
713		/**
714	<	* Returns a pool submission, if one exists, activating first.
714	>	* Sets suspended status and blocks as spare until resumed,
715	>	* shutdown, or timed out.
716		*
717	<	* @return a submission, if available
717	>	* @return false if trimmed
718		*/
719	<	private ForkJoinTask<?> pollSubmission() {
719	>	final boolean suspendAsSpare() {
720	>	for (;;) {               // set suspended unless terminating
721	>	int s = runState;
722	>	if ((s & TERMINATING) != 0) { // must kill
723	>	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
724	>	s | (TRIMMED | TERMINATING)))
725	>	return false;
726	>	}
727	>	else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
728	>	s | SUSPENDED))
729	>	break;
730	>	}
731	>	lastEventCount = 0;      // reset upon resume
732		ForkJoinPool p = pool;
733	<	while (p.hasQueuedSubmissions()) {
734	<	ForkJoinTask<?> t;
735	<	if (tryActivate() && (t = p.pollSubmission()) != null)
736	<	return t;
733	>	p.releaseWaiters();      // help others progress
734	>	p.accumulateStealCount(this);
735	>	interrupted();           // clear/ignore interrupts
736	>	if (poolIndex < p.getParallelism()) { // untimed wait
737	>	while ((runState & SUSPENDED) != 0)
738	>	doPark();
739	>	return true;
740		}
741	<	return null;
741	>	return timedSuspend();   // timed wait if apparently non-core
742	>	}
743	>
744	>	/**
745	>	* Blocks as spare until resumed or timed out
746	>	* @return false if trimmed
747	>	*/
748	>	private boolean timedSuspend() {
749	>	long nanos = SPARE_KEEPALIVE_NANOS;
750	>	long startTime = System.nanoTime();
751	>	while ((runState & SUSPENDED) != 0) {
752	>	++parkCount;
753	>	if ((nanos -= (System.nanoTime() - startTime)) > 0)
754	>	LockSupport.parkNanos(this, nanos);
755	>	else { // try to trim on timeout
756	>	int s = runState;
757	>	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
758	>	(s & ~SUSPENDED) |
759	>	(TRIMMED|TERMINATING)))
760	>	return false;
761	>	}
762	>	}
763	>	return true;
764	>	}
765	>
766	>	// Misc support methods for ForkJoinPool
767	>
768	>	/**
769	>	* Returns an estimate of the number of tasks in the queue.  Also
770	>	* used by ForkJoinTask.
771	>	*/
772	>	final int getQueueSize() {
773	>	return -base + sp;
774		}
775
776	<	// Methods accessed only by Pool
776	>	/**
777	>	* Set locallyFifo mode. Called only by ForkJoinPool
778	>	*/
779	>	final void setAsyncMode(boolean async) {
780	>	locallyFifo = async;
781	>	}
782
783		/**
784		* Removes and cancels all tasks in queue.  Can be called from any
785		* thread.
786		*/
787		final void cancelTasks() {
788	<	ForkJoinTask<?> t;
789	<	while (base != sp && (t = deqTask()) != null)
790	<	t.cancelIgnoringExceptions();
788	>	while (base != sp) {
789	>	ForkJoinTask<?> t = deqTask();
790	>	if (t != null)
791	>	t.cancelIgnoringExceptions();
792	>	}
793		}
794
795		/**
#	Line 682 | Line 799 | public class ForkJoinWorkerThread extend
799		*/
800		final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
801		int n = 0;
802	<	ForkJoinTask<?> t;
803	<	while (base != sp && (t = deqTask()) != null) {
804	<	c.add(t);
805	<	++n;
802	>	while (base != sp) {
803	>	ForkJoinTask<?> t = deqTask();
804	>	if (t != null) {
805	>	c.add(t);
806	>	++n;
807	>	}
808		}
809		return n;
810		}
811
812	+	// Support methods for ForkJoinTask
813	+
814		/**
815	<	* Gets and clears steal count for accumulation by pool.  Called
816	<	* only when known to be idle (in pool.sync and termination).
815	>	* Returns an estimate of the number of tasks, offset by a
816	>	* function of number of idle workers.
817	>	*
818	>	* This method provides a cheap heuristic guide for task
819	>	* partitioning when programmers, frameworks, tools, or languages
820	>	* have little or no idea about task granularity.  In essence by
821	>	* offering this method, we ask users only about tradeoffs in
822	>	* overhead vs expected throughput and its variance, rather than
823	>	* how finely to partition tasks.
824	>	*
825	>	* In a steady state strict (tree-structured) computation, each
826	>	* thread makes available for stealing enough tasks for other
827	>	* threads to remain active. Inductively, if all threads play by
828	>	* the same rules, each thread should make available only a
829	>	* constant number of tasks.
830	>	*
831	>	* The minimum useful constant is just 1. But using a value of 1
832	>	* would require immediate replenishment upon each steal to
833	>	* maintain enough tasks, which is infeasible.  Further,
834	>	* partitionings/granularities of offered tasks should minimize
835	>	* steal rates, which in general means that threads nearer the top
836	>	* of computation tree should generate more than those nearer the
837	>	* bottom. In perfect steady state, each thread is at
838	>	* approximately the same level of computation tree. However,
839	>	* producing extra tasks amortizes the uncertainty of progress and
840	>	* diffusion assumptions.
841	>	*
842	>	* So, users will want to use values larger, but not much larger
843	>	* than 1 to both smooth over transient shortages and hedge
844	>	* against uneven progress; as traded off against the cost of
845	>	* extra task overhead. We leave the user to pick a threshold
846	>	* value to compare with the results of this call to guide
847	>	* decisions, but recommend values such as 3.
848	>	*
849	>	* When all threads are active, it is on average OK to estimate
850	>	* surplus strictly locally. In steady-state, if one thread is
851	>	* maintaining say 2 surplus tasks, then so are others. So we can
852	>	* just use estimated queue length (although note that (sp - base)
853	>	* can be an overestimate because of stealers lagging increments
854	>	* of base).  However, this strategy alone leads to serious
855	>	* mis-estimates in some non-steady-state conditions (ramp-up,
856	>	* ramp-down, other stalls). We can detect many of these by
857	>	* further considering the number of "idle" threads, that are
858	>	* known to have zero queued tasks, so compensate by a factor of
859	>	* (#idle/#active) threads.
860		*/
861	<	final int getAndClearStealCount() {
862	<	int sc = stealCount;
699	<	stealCount = 0;
700	<	return sc;
861	>	final int getEstimatedSurplusTaskCount() {
862	>	return sp - base - pool.idlePerActive();
863		}
864
865		/**
866	<	* Returns {@code true} if at least one worker in the given array
705	<	* appears to have at least one queued task.
866	>	* Gets and removes a local task.
867		*
868	<	* @param ws array of workers
868	>	* @return a task, if available
869		*/
870	<	static boolean hasQueuedTasks(ForkJoinWorkerThread[] ws) {
871	<	if (ws != null) {
872	<	int len = ws.length;
873	<	for (int j = 0; j < 2; ++j) { // need two passes for clean sweep
713	<	for (int i = 0; i < len; ++i) {
714	<	ForkJoinWorkerThread w = ws[i];
715	<	if (w != null && w.sp != w.base)
716	<	return true;
717	<	}
718	<	}
870	>	final ForkJoinTask<?> pollLocalTask() {
871	>	while (base != sp) {
872	>	if (active || (active = pool.tryIncrementActiveCount()))
873	>	return locallyFifo? locallyDeqTask() : popTask();
874		}
875	<	return false;
875	>	return null;
876		}
877
878	<	// Support methods for ForkJoinTask
878	>	/**
879	>	* Gets and removes a local or stolen task.
880	>	*
881	>	* @return a task, if available
882	>	*/
883	>	final ForkJoinTask<?> pollTask() {
884	>	ForkJoinTask<?> t;
885	>	return (t = pollLocalTask()) != null ? t : scan();
886	>	}
887
888		/**
889	<	* Returns an estimate of the number of tasks in the queue.
889	>	* Executes or processes other tasks awaiting the given task
890	>	* @return task completion status
891		*/
892	<	final int getQueueSize() {
893	<	// suppress momentarily negative values
894	<	return Math.max(0, sp - base);
892	>	final int execWhileJoining(ForkJoinTask<?> joinMe) {
893	>	int s;
894	>	while ((s = joinMe.status) >= 0) {
895	>	ForkJoinTask<?> t = base != sp?
896	>	popWhileJoining(joinMe) :
897	>	scanWhileJoining(joinMe);
898	>	if (t != null)
899	>	t.tryExec();
900	>	}
901	>	return s;
902		}
903
904		/**
905	<	* Returns an estimate of the number of tasks, offset by a
906	<	* function of number of idle workers.
905	>	* Returns or stolen task, if available, unless joinMe is done
906	>	*
907	>	* This method is intrinsically nonmodular. To maintain the
908	>	* property that tasks are never stolen if the awaited task is
909	>	* ready, we must interleave mechanics of scan with status
910	>	* checks. We rely here on the commit points of deq that allow us
911	>	* to cancel a steal even after CASing slot to null, but before
912	>	* adjusting base index: If, after the CAS, we see that joinMe is
913	>	* ready, we can back out by placing the task back into the slot,
914	>	* without adjusting index. The loop is otherwise a variant of the
915	>	* one in scan().
916	>	*
917		*/
918	<	final int getEstimatedSurplusTaskCount() {
919	<	// The halving approximates weighting idle vs non-idle workers
920	<	return (sp - base) - (pool.getIdleThreadCount() >>> 1);
918	>	private ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
919	>	int r = seed;
920	>	ForkJoinPool p = pool;
921	>	ForkJoinWorkerThread[] ws;
922	>	int n;
923	>	outer:while ((ws = p.workers) != null && (n = ws.length) > 1) {
924	>	int mask = n - 1;
925	>	int k = r;
926	>	boolean contended = false; // to retry loop if deq contends
927	>	for (int j = -n; j <= n; ++j) {
928	>	if (joinMe.status < 0)
929	>	break outer;
930	>	int b;
931	>	ForkJoinTask<?>[] q;
932	>	ForkJoinWorkerThread v = ws[k & mask];
933	>	r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
934	>	if (v != null && (b=v.base) != v.sp && (q=v.queue) != null) {
935	>	int i = (q.length - 1) & b;
936	>	ForkJoinTask<?> t = q[i];
937	>	if (t != null && UNSAFE.compareAndSwapObject
938	>	(q, (i << qShift) + qBase, t, null)) {
939	>	if (joinMe.status >= 0) {
940	>	v.base = b + 1;
941	>	seed = r;
942	>	++stealCount;
943	>	return t;
944	>	}
945	>	UNSAFE.putObjectVolatile(q, (i<<qShift)+qBase, t);
946	>	break outer; // back out
947	>	}
948	>	contended = true;
949	>	}
950	>	k = j < 0 ? r : (k + ((n >>> 1) | 1));
951	>	}
952	>	if (!contended && p.tryAwaitBusyJoin(joinMe))
953	>	break;
954	>	}
955	>	return null;
956		}
957
958		/**
959	<	* Scans, returning early if joinMe done.
959	>	* Version of popTask with join checks surrounding extraction.
960	>	* Uses the same backout strategy as helpJoinTask. Note that
961	>	* we ignore locallyFifo flag for local tasks here since helping
962	>	* joins only make sense in LIFO mode.
963	>	*
964	>	* @return a popped task, if available, unless joinMe is done
965		*/
966	<	final ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
967	<	ForkJoinTask<?> t = pollTask();
968	<	if (t != null && joinMe.status < 0 && sp == base) {
969	<	pushTask(t); // unsteal if done and this task would be stealable
970	<	t = null;
966	>	private ForkJoinTask<?> popWhileJoining(ForkJoinTask<?> joinMe) {
967	>	int s;
968	>	ForkJoinTask<?>[] q;
969	>	while ((s = sp) != base && (q = queue) != null && joinMe.status >= 0) {
970	>	int i = (q.length - 1) & --s;
971	>	ForkJoinTask<?> t = q[i];
972	>	if (t != null && UNSAFE.compareAndSwapObject
973	>	(q, (i << qShift) + qBase, t, null)) {
974	>	if (joinMe.status >= 0) {
975	>	sp = s;
976	>	return t;
977	>	}
978	>	UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
979	>	break;  // back out
980	>	}
981		}
982	<	return t;
982	>	return null;
983		}
984
985		/**
#	Line 756 | Line 987 | public class ForkJoinWorkerThread extend
987		*/
988		final void helpQuiescePool() {
989		for (;;) {
990	<	ForkJoinTask<?> t = pollTask();
991	<	if (t != null)
992	<	t.quietlyExec();
993	<	else if (tryInactivate() && pool.isQuiescent())
994	<	break;
990	>	ForkJoinTask<?> t = pollLocalTask();
991	>	if (t != null || (t = scan()) != null)
992	>	t.tryExec();
993	>	else {
994	>	ForkJoinPool p = pool;
995	>	if (active) {
996	>	active = false; // inactivate
997	>	do {} while (!p.tryDecrementActiveCount());
998	>	}
999	>	if (p.isQuiescent()) {
1000	>	active = true; // re-activate
1001	>	do {} while (!p.tryIncrementActiveCount());
1002	>	return;
1003	>	}
1004	>	}
1005		}
765	–	do {} while (!tryActivate()); // re-activate on exit
1006		}
1007
1008		// Unsafe mechanics
1009
1010		private static final sun.misc.Unsafe UNSAFE = getUnsafe();
771	–	private static final long spOffset =
772	–	objectFieldOffset("sp", ForkJoinWorkerThread.class);
1011		private static final long runStateOffset =
1012		objectFieldOffset("runState", ForkJoinWorkerThread.class);
1013	<	private static final long qBase;
1013	>	private static final long qBase =
1014	>	UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1015		private static final int qShift;
1016
1017		static {
779	–	qBase = UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1018		int s = UNSAFE.arrayIndexScale(ForkJoinTask[].class);
1019		if ((s & (s-1)) != 0)
1020		throw new Error("data type scale not a power of two");

Diff Legend

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